content
stringlengths
1
15.9M
\section{Introduction} Primordial black holes (PBHs) arise naturally in most cosmologies. Perhaps the least speculative mechanism for PBH formation comes from the collapse of overdense regions in the primordial density fluctuations that gave rise to structure in the universe \cite{carrhawking}. Thus PBHs carry information of an epoch about which we know comparatively little, and are a very useful tool for restricting theories of the very early universe, especially within the context of an inflationary scenario. In this paradigm the density spectrum is a consequence of the quantum fluctuations of the inflaton field, and can in principle be calculated given an underlying model. Thus, the non-observation of the by-products or of the effects of the energy density of these PBHs constrains the underlying microscopic theory. The simplest bound that can be extracted from PBH formation is generated by insisting that $\Omega_{PBH}\leq 1$. Other bounds may be derived by studying the consequences of their evaporation. Given that black holes evaporate at a rate proportional to their inverse mass squared~\cite{hawking}, the phenomenological relevance of a PBH will depend upon its initial mass. Smaller mass PBHs ($10^9 < M_{bh} < 10^{13}$~g) will alter the heavy elements abundances \cite{nuc} as well a distort the microwave background \cite{micro}, whereas PBHs with larger masses will affect the diffuse gamma-ray background \cite{diffuse,mac2,macreport}. The net evaporation spectrum from a collection of PBHs will depend on the initial mass distribution, which in turn depends upon the probability distribution for the density fluctuations. Prior to the emergence of the COBE data, bounds from PBHs were calculated assuming a Harrison-Zel'dovich spectrum $|\delta_k|^2 \propto k$, with an unknown normalization. This lead to the famous ``Page-Hawking'' bound and all its subsequent improvements \cite{diffuse,mac2}. However, assuming that the distribution is Gaussian, we can now relate the mass variance at the time of formation with the mass variance at large scales today, if we know the (model-dependent) power spectrum. None the less, bounds on $n$ have been derived within the class of models where the power spectrum is given by a power law $|\delta_k|^2 \propto k^n$ over the scales of interest \cite{carrlidsey,cgl,green,kim1,kim2}. Indeed, in this way it follows that for the scale-invariant Harrison-Zel'dovich spectrum PBHs have too small a number density to be of any astrophysical significance. However, observations (see references in \cite{cgl}) now seem to favor a tilted blue spectrum with $n>1$ (in CDM models) with more power at smaller scales, and thus the bounds derived from black hole evaporation can be used to constrain the tilt of the spectrum. In this paper we revisit the aforementioned bounds in light of recent calculations which indicate that a spectrum of primordial black hole masses are produced through near critical gravitational collapse. As was pointed out by Jedamzik and Niemeyer \cite{JZI}, if PBH formation is a result of a critical phenomena, then the initial mass function will be quite different then what was expected from the classic calculation of Carr \cite{carr}. In particular, the PBH mass formed at a given epoch is no longer necessarily proportional to the horizon mass. The resulting difference in the initial mass function leads to new bounds, which is the main thrust of this paper. In particular, we revisit the density bounds and the bounds derived from the diffuse gamma-ray observations. We will first derive bounds on $n$ in the class of models where the power spectrum is a power law over the scales of interest. We then relax this assumption and instead place bounds on the mass variance at the formation epoch.\footnote{As will be discussed later, PBH formation is dominated by the earliest formation epoch if they form via critical collapse.} These bounds can then be applied to a chosen model by extrapolating the variance today to the formation epoch according to the appropriate power spectrum. Before continuing to the body of this work, we would like to point out that primordial black holes have also played a large role in attempting to explain various data. PBHs can serve as significant cosmological flux sources for all particle species via Hawking radiation\cite{hawking}. Thus it is very tempting to postulate that present-day observed particle fluxes of unknown origin are a consequence of PBH evaporation. However, to predict these fluxes, or model them realistically, we need to know the mass distribution of black holes, since their emission spectra are determined by their temperature (or inverse mass). If we assume that the black holes of interest were formed from initial density inhomogeneities generated in an inflationary scenario (which is usually assumed), then the black holes are either tremendously over-abundant or completely negligible. To get a phenomenologically interesting quantity of PBHs thus requires an extreme fine-tuning, as will be demonstrated below. Succinctly, this fine-tuning arises because the PBH number density is an extremely rapidly varying function of the spectral index $n$. Thus, without even analyzing the details of the spectral profile, explaining unknown fluxes via PBH evaporation is far from compelling. \section {The Initial Mass Function} Carr \cite{carr} first calculated the PBH spectrum resulting from a scale-invariant Harrison-Zel'dovich spectrum up to an overall normalization. Subsequently Page and Hawking calculated a bound on the normalization by calculating the expected diffuse gamma-ray spectrum from these PBHs \cite{diffuse,mac2}. However, using the COBE measurements of the temperature anisotropies translated into density fluctuations (within a CDM), the overall normalization can be determined. For a scale-invariant spectrum, no significant number density of PBHs is generated. However, for a tilted blue power spectrum with more power on smaller scales, a larger number density of PBHs is expected. Given an initially overdense region with density contrast $\delta_i$ and radius $R$ at time $t_i$ (using the usual comoving coordinates), analytic arguments predict \cite{carrhawking,carr} a black hole will form if \begin{equation} 1/3\leq \delta_i \leq 1 \; . \end{equation} The lower bound on the density contrast comes from insisting that the size of the region at the time of collapse be greater than the Jeans length, while the upper bounds come from the consistency of the initial data with the assumption of a connected topology. The first calculations of the PBH mass distribution assumed the relation \begin{equation} \label{bhPropMh} M_{bh}\simeq\gamma^{3/2}M_h, \end{equation} where $\gamma$ determines the equation of state $p = \gamma \rho$ and $M_h$ is the horizon mass when the scale of interest crossed the horizon. Recently, numerical evidence suggests that near the threshold of black hole formation, gravitational collapse behaves as a critical phenomena with scaling and self-similarity \cite{scaling}. A scaling relation of the following form was found \begin{equation} \label{scaling} M_{bh}(\delta)=kM_h(\delta-\delta_c)^{\rho}, \end{equation} where $\rho$ is a universal scaling exponent which is independent of the initial shape of the density fluctuation. It was later shown \cite{JZI} that such scaling should be relevant for PBH formation. Indeed, in Ref.~\cite{JZII} the authors found relation (\ref{scaling}) to hold for PBH formation when the initial conditions are adjusted to be nearly critical. They found the exponent to be $\rho\approx 0.37$. They also found that for several different initial density shapes, $\delta_c\approx 0.7$, which is significantly larger than the analytic prediction of $1/3$ found by requiring that the initial overdensity be larger than the Jeans mass. Given Eq.~(\ref{scaling}), calculating the initial PBH mass distribution becomes analytically cumbersome, since in principle one needs to sum over all epochs of PBH formation. However, we would expect that the initial mass function would be dominated by the earliest epoch of formation if we assume a Gaussian distribution with a blue power spectrum, since for larger scales the formation probability should be suppressed. This expectation was tested in Ref.~\cite{green2}, where the authors used the excursion set formalism \cite{bond} to calculate the initial mass function allowing for PBH formation at all epochs. The authors found that it was approximately true the that earliest epoch dominates, for the conditions of interest to us. This simplification allows us to derive quite easily the initial mass distribution \cite{JZI}. We assume Gaussian fluctuations (the effects of non-Gaussianity will be briefly discussed in Sec.~\ref{robust-sec}) and define the usual smoothed density contrast \begin{equation} \delta_R(x)=\int d^3y\,\delta (x+y) W_R(y) \; , \end{equation} where $\delta(x)=(\rho(x)-\rho_b)/\rho_b$, and $\rho_b$ is background energy density. $W_R$ is the window function with support in a region of size $R$. The probability that a region of size $R$ has density contrast between $\delta$ and $\delta+d\delta$ is given by \begin{equation} P(R,\delta)d\delta =\frac{1}{\sqrt{2 \pi}\sigma_R}\exp\left(-\frac{\delta^2} {2\sigma_R^2}\right) d\delta \; , \end{equation} where $\sigma_R$ is the mass variance for a region of size $R$, $\sigma^2_R=\langle \delta_R^2(x) \rangle/R^3$. Then using Eq.~(\ref{scaling}), the physical number density of PBHs within the horizon per logarithmic mass interval, at the formation epoch, can be written as \begin{equation} \label{IMF1} \frac{d n_{bh}}{d\log M_{bh}} = V_h^{-1}\,P[\delta(M_{bh})]\,\frac{d\delta}{d\log M_{bh}} = \frac{V_h^{-1}}{\sqrt{2 \pi}\,\sigma\,\rho} \left( \frac{M_{bh}}{k M_H} \right)^{1/\rho}\, \exp\left[-\frac{1}{2\sigma^2} \left[ \delta_c + \left(\frac{M_{bh}}{k M_H} \right)^{1/\rho} \right]^2 \right], \end{equation} where $V_{h}$ is the horizon volume the the epoch of PBH formation. We assume prompt reheating, and therefore take the formation epoch to be the time of reheating, which corresponds to the minimum horizon mass~\cite{green2}. Assuming that the power law spectrum holds down to the scales of the reheat temperature, then $\sigma_H^2 \propto R^{-(n+3)}$. We can then relate the mass variance today $\sigma_0$ to the mass variance at the epoch of PBH formation $\sigma(M_H)$, using \cite{green,green2} \begin{equation} \label{dis1} \sigma^2(M_H)=\sigma_0^2 \left( \frac{M_{eq}}{M_0} \right)^{(1-n)/3} \left( \frac{M_{H}}{M_{eq}} \right)^{(1-n)/2}, \end{equation} where \begin{equation} M_H = M_0\left(\frac{T_{eq}}{T_{RH}}\right)^2 \left(\frac{T_0}{T_{eq}}\right)^{3/2}, \end{equation} $M_0$ is the mass inside the horizon today, $T_{RH}$ is the reheat temperature, and $T_{eq}$ is the temperature at radiation-matter equality. This relation is essential to connect present-day density fluctuations to those of much earlier times. Two important assumptions underly this useful relation: First, $n$ is taken to be constant over all the scales of interest. Second, the universe was assumed to be radiation dominated until the temperature dropped below $T_{eq} \sim 5$~eV (matter-radiation equality), and then matter dominated thereafter. From the COBE anisotropy data, the mass variance can be calculated \cite{bunn,green} \begin{equation} \label{COBE} \sigma_0=9.5 \times 10^{-5} . \end{equation} Using this result we can then calculate the physical number density per unit mass interval at $T = T_{RH}$ \begin{equation} \label{IMF} \frac{d n_{bh}}{dM_{bh}}= \frac{V_h^{-1}}{\sqrt{2\pi}\sigma(M_H)M_{bh}\rho}y^{1/\rho}\exp\left[ -\frac{\left(\delta_c+y^{1/\rho}\right)^2}{2\sigma^2(M_H)}\right], \label{phys-num-density-eq} \end{equation} where \begin{equation} y=\frac{M_{bh}}{kM_H}=\frac{M_{bh}T_{RH}^2} {0.301 k g_\star^{-1/2} M_{Pl}^3} \; , \end{equation} and $M_{Pl}$ is the Planck mass. The physical number density at time $t$ is simply Eq.~(\ref{phys-num-density-eq}) rescaled by a ratio of scale factors, that can be written as \begin{equation} \frac{d n_{bh}}{d M_{bh}}(t) = \left( \frac{T(t)}{T_{RH}} \right)^3 \frac{d n_{bh}}{d M_{bh}} \; . \label{scaled-density-eq} \end{equation} Finally, we should note that relation (\ref{scaling}) is only valid for $\delta \approx \delta_c$. Thus we expect that we may integrate over $\delta$ with small errors, as long as the width of the Gaussian is sufficiently small. In particular, our results should be trustworthy provided $\sigma \lesssim 1$, which implies $n$ should not exceed the maximum \begin{equation} n_{max}\simeq 1+\frac{2 \log(\sigma_0^2)}{\log(T_{eq}T_0/T_{RH}^2)} \; . \end{equation} We will see that $n$ does not exceed this maximum value for all of the bounds we consider. \section{Bounds from $\Omega \leq 1$} Let us now calculate the total energy density in PBHs. We assume the ``standard cosmology'' where the universe began in an inflationary phase, reheated, was radiation dominated from the reheating period until matter-radiation equality, and then has been matter dominated. The contribution of a PBH with a given initial mass, $M_{bh}$, to the energy density today will depend upon its lifetime. The time-dependent PBH mass $M(t)$ is given by~\cite{hawking} \begin{equation} \label{massevol} M(t) = M_\star\left[\left(\frac{M_{bh}}{M_\star}\right)^3 -\frac{t}{t_0}\right]^{1/3}, \end{equation} where $M_\star$ is the initial mass of a PBH which would be decaying today, $M_\star \simeq 5 \times 10^{14}$~g. It is a good approximation to assume that the black hole decays instantaneously at a fixed decay time, $t_d$, which we use in the following. There are two components to the PBH density bounds that we can calculate. The first is the total energy density of the PBHs that have not decayed by a given time $t$. The second is the total energy density of the products of PBH evaporation. The sum of these components must be less than the critical density $\Omega_{pbh,evap} + \Omega_{pbh} < 1$, at any time. The evaporated products of PBHs, in particular photons, could break up elements during nucleosynthesis, disrupting the well-measured elemental abundances. This and other processes during nucleosynthesis provide additional bounds on the density of PBHs~\cite{nuc} that we do not discuss here. The simplest bound comes from the density of PBHs that have not decayed by time $t$, \begin{equation} \rho_{pbh}(t) = \left( \frac{T(t)}{T_{RH}} \right)^3 \rho_{tot}(t_{RH}) \> I_{M_\star(t)}^{M_{max}}(0) \end{equation} where $T(t)$ is the temperature of the universe at time $t$, and $M_\star(t)$ is the initial PBH mass that has just completely evaporated by time $t$, \begin{equation} M_\star(t) \approx M_\star \left( \frac{t}{t_0} \right)^{1/3} \; . \end{equation} $I_{M_1}^{M_2}(\xi)$ is a dimensionless weighted integral over the PBH mass spectrum between $M_1$ to $M_2$, normalized to the total density $\rho_{tot}(t_{RH})$, \begin{equation} I_{M_1}^{M_2}(\xi) = \frac{1}{\rho_{tot}(t_{RH})} \int_{M_1}^{M_2} d M_{bh} M_{bh} \frac{d n_{bh}}{d M_{bh}} \left( \frac{M_{bh}}{M_\star} \right)^\xi \; , \end{equation} where $d n_{bh}/d M_{bh}$ is given by Eq.~(\ref{phys-num-density-eq}). We can use the above to trivially compute the ratio of the PBH density to the critical density, $\Omega(t)$. In particular, we need only compute the density ratio at three relevant epochs: immediately after reheating $t=t_{RH}$, at matter-radiation equality $t=t_{eq}$, and present-day $t=t_0$. The density ratios are\footnote{Note that $\Omega(t_{RH}) = I_0^{M_{max}}(0)$ is often denoted by $\beta(t_{RH})$.} \begin{eqnarray} \Omega(t_{RH}) &=& I_0^{M_{max}}(0) \\ \Omega(t_{eq}) &=& \frac{T_{RH}}{T_{eq}} I_{M_{eq}}^{M_{max}}(0) \\ \Omega(t_{0}) &=& \frac{T_{RH}}{T_{eq}} I_{M_\star}^{M_{max}}(0) \; . \end{eqnarray} where $M_{max}$ corresponds to $\delta = 1$, which is of order the horizon mass at reheating. The integral should be independent of the upper limit $M_{max}$ if we are to trust our results. Making the conservative approximation that all the PBH decay products are relativistic, the contribution to the density ratio of the products of PBH evaporation that has occurred up until today can be written as \begin{equation} \Omega_{pbh,evap}(t_0) = \frac{T_{RH}}{T_{eq}} \left( \frac{T_0}{T_{eq}} \right)^{1/4} I_{0}^{M_{eq}}(3/2) + \frac{T_{RH}}{T_{eq}} I_{M_{eq}}^{M_\star}(2) \; . \end{equation} In Fig.~\ref{bound-fig}, we show the upper limit on $n$ as a function of $T_{RH}$ coming from bounding $\Omega_{pbh,evap}(t_0) + \Omega_{pbh}(t_0) < 1$ (solid line). For larger values of the reheat temperature we get a more stringent bound by imposing the constraint $\Omega_{pbh}(t_{RH}) \leq 1$, simply because as $T_{RH}$ is increased more of the black holes will have decayed at an earlier epoch. Given that most of the energy of the decay products resides in radiation, the effect on $\Omega_{pbh}(t_0)$ is diminished due to the redshifting. This new bound is given by the dotted line in Fig.~\ref{bound-fig}. If we assume that the PBH leaves behind a Planck mass remnant, then we have additional bounds which become important for very large reheat temperature\cite{mac,barrow,cgl}. The best bound in this case comes from calculating $\Omega_{remnant}(t_{eq})$ which is given by \begin{equation} \Omega_{remnant}(t_{eq}) = \frac{T_{RH}}{T_{eq}} \frac{M_{Pl}}{M_\star} I_0^{M_{eq}}(-1) \; . \end{equation} The bound in this case is shown as the dashed line in Fig.~\ref{bound-fig}, and is the best bound at the largest values of the reheat temperature. \section{Bounds from Diffuse Gamma-Rays} For a certain range of $T_{RH}$ we can improve our bounds on $n$ from diffuse gamma-ray constraints. The present day flux is determined by convoluting the initial mass function with the black hole emission spectrum \begin{equation} \label{hawkingspectrum} f(x)=\frac1{2\pi}\frac{\Gamma_s(x)} {\exp(8\pi x)-(-1)^{2 s}}, \end{equation} where $s$ is the spin of the emitted particle, $x=\omega(t) M(t)/M_{Pl}^2$, $\omega(t)$ is the frequency and $M(t)$ is the PBH mass at the time $t$ of emission. $\Gamma_s(x)$ is the absorption coefficient and may be written as $[\omega(t)]^2 \sigma_s/\pi$. $\sigma_s$ is the absorption cross section and is calculated using the principle of detailed balance. The values for $\sigma_s$ were calculated some time ago by Page\cite{page1,page2}. Let us consider how $\sigma_s$ behaves for massless particles. At large values of $x$, $\sigma_s$ performs small oscillations about the geometric optics limit of $\sigma_g=27 \pi M^2/M_{Pl}^4$. As $x$ approaches zero, $\sigma_s$ goes to zero for $s=1/2,1$ but goes to a constant value for $s=0$. We will use the approximation \begin{equation} \Gamma_s(x)= (56.7,~20.4) x^2/\pi~~ $\rm for$~~ s=(\frac12,~1). \end{equation} This approximation is poor at low energies, as it is in error by $50\%$ at $x=0.05$. However, as we shall see, the contribution to the spectrum of interest is greatly peaked at $x\simeq0.2$. The case of strongly interacting particles is complicated by the hadronization process. There is a large contribution coming from pion decay, however, given the extreme sensitivity of the flux to the value $n$, the effect on the bound is negligible. It has been recently suggested \cite{heckler1,heckler2} that the self-interactions of the emitted particles will induce a photosphere, thus distorting the spectrum considerably from Eq.~(\ref{hawkingspectrum}). It was suggested that two types of photospheres should form. A QCD photosphere\footnote{In the case of QCD what is meant by ``photosphere'' is a quark-gluon cloud.} generated by parton-parton interactions as well as a QED photosphere generated by electron-positron-photon interactions. This is idea has been tested more quantitatively via a numerical solution of the Boltzmann equation~\cite{mcgill}. Again, while this effect may change the spectrum, especially at higher energies, it is irrelevant as far as the bound on the spectral index is concerned. The flux measured today is given by \begin{equation} \label{cosmflux} \frac{dJ}{d\omega_0} = \frac{1}{4\pi} \int_{t_i}^{t_0} dt (1+z) \int d M_{bh} \, \frac{d n_{bh}}{dM_{bh}}(t) \, f(x) \; , \end{equation} where $d n_{bh}/d M_{bh}$ is evaluated at time $t$ using Eq.~(\ref{scaled-density-eq}), $t_0$ is the age of the universe, $t_i$ is the time of last scatter, and $f(x)$ is the instantaneous emission spectrum given above with \begin{equation} x = \frac{\omega(t) M(t)}{M_{Pl}^2} = \frac{\omega_0 (1+z)}{M_{Pl}^2} M_\star \left[ \left( \frac{M_{bh}}{M_\star} \right)^3 - \frac{t}{t_0} \right]^{1/3} . \end{equation} The integral over $t$ is cut off at early times, since at redshifts above $z = z_0 \simeq 700$ the optical depth will be larger than unity due to either pair production off of matter or ionized matter \cite{ZS}. Those processes will degrade the energy below the window we are interested in. This integral may be rewritten in the more illuminating form \begin{equation} \label{cosmoSpectrum} \frac{dJ}{d\omega_0} = \frac{1}{4\pi}\frac{M_{Pl}^6}{(\omega_0M_\star)^3} \int_{0}^{z_0}\frac{dz}{ H_0 (1+z)^{5/2}} \int_{0}^{\infty} dx \> x^2 \alpha^{-2}\,f(x)\frac{d n_{bh}(x,z)}{dM_{bh}}, \end{equation} where \begin{equation} \label{mbh} \alpha=\frac{M(t)}{M_\star}=\left\{(1+z)^{-3/2} + \frac{x^3M_{Pl}^6}{[(1+z)\omega_0M_\star]^3}\right\}^{1/3}. \end{equation} Let us study the qualitative behavior of the above integral as a function of $\omega$ at fixed $n$ and $T_{RH}$. The $x$ integration is controlled by the Boltzmann factor in $f(x)$. Indeed, a little manipulation shows the the integrand is highly peaked near $x \simeq 0.2$. Furthermore, the $\omega$ dependence in $\alpha^{-2}$ is almost completely canceled by the $\omega$ dependence in the factor $M_{bh}^{-1}y^{1/\rho}\propto\alpha^{(1/\rho-1)}$ in $d n_{bh}/dM_{bh}$. Thus the $\omega$ dependent part of the integrand may be written as \begin{equation} \frac{dJ}{d\omega_0}\propto \omega_0^{-3} \exp\left[-\frac{\left(\delta_c+aT_{RH}^{2/\rho}\alpha^{1/\rho}\right)^2} {2\sigma^2(M_H)}\right], \end{equation} where $a^\rho = M_\star g_\star^{1/2}/(0.301 k M_{Pl}^3)$, and the only $\omega$ dependence in the exponential is through $\alpha$. If for now we assume that the dominant contribution the higher energy photons comes from recent decays ($z\sim 0$), and most of the support for the $x$ integral comes with $x \sim 0.2$, then $\alpha$ simplifies to \begin{equation} \alpha \approx \left[ 1 + \left( \frac{0.2 M_{Pl}^2}{w_0 M_\star} \right)^3 \right]^{1/3}. \end{equation} As $\omega_0$ gets larger than $0.2 M_{Pl}^2/M_\star \sim 100\ \rm{ MeV} $, $\alpha$ becomes independent of $\omega_0$ and therefore the flux behaves as $dJ/d\omega_0 \propto \omega_0^{-3}$. For lower energies we can make the approximation $\alpha\sim 0.2 M_{Pl}^2/(w_0M_\star)$, and we would expect that at some point the $\omega$ dependence in the exponential will begin to dominate such that the flux should begin to rapidly decrease as we go to lower photon energies. The energy at which the flux turns over is determined by the competition between the two terms $\delta_c$ and $a T_{RH}^{2/\rho}\alpha^{1/\rho}$ in the exponential, which is set by the reheat temperature. As we lower the reheat temperature the position of the kink moves to lower energies. If the reheat temperature is higher than $T_{RH}\sim 10^9\ \rm{ GeV} $, however, the peak will stay around 100 MeV, since at these temperatures the second term in the exponential will always dominate. Indeed, we expect the position of the fall off to be near \begin{equation} \label{kink} \omega_{\rm kink}\simeq \min\left(100{\ \rm{ MeV} }\,,\, \frac{0.2g_\star^{1/2}\,T_{RH}^2} {0.301\,k\,M_{Pl}\,\delta_c^{\rho}}\right). \end{equation} Figure~\ref{examples-fig} shows the flux for fixed $n$ for a few different reheat temperatures. The position of the kink is well tracked by Eq.~(\ref{kink}). Note however that the flux does not fall off exponentially at energies below the kink. This is because as we go to lower energies we pick up more of a contribution from higher redshifts. It is interesting to contrast this behavior with the flux calculated assuming that the mass of a PBH formed at a given epoch is proportional to the horizon mass at the time of collapse. In Refs.~\cite{kim1,kim2} the authors calculated an initial mass function following the Press-Schecter formalism, summing over all epochs and assuming the relation $M_{bh}\simeq\gamma^{3/2} M_H$ at each epoch. They found \begin{equation} \label{KLM} \frac{d n_{bh}}{dM_{bh}} = \frac{n+3}{4} \sqrt{\frac{2}{\pi}} \gamma^{7/4} \rho_i M^{1/2}_{H_i}M_{bh}^{-5/2}\sigma_H^{-1} \exp\left(-\frac{\gamma^2}{2\sigma_H^2}\right), \end{equation} where $\rho_i$ and $M_{H_i}$ are the energy density and horizon mass at $T_{RH}$ and \begin{equation} \sigma_H=\sigma_0\left(\frac{M_{bh}}{\gamma^{3/2}M_0}\right)^{(1-n)/4}. \end{equation} This result reduces to the initial mass function first computed by Page \cite{carr} for the Harrison-Zel'dovich spectrum with $n=1$ and $d n_{bh}/dM_{bh} \propto M_{bh}^{-5/2}$. The $\omega_0$ dependence of this result arises only through $M(t)$ given by Eq.~(\ref{mbh}). Using this initial mass distribution in Eq.~(\ref{cosmflux}), we expect, as in the previous case, $dJ/d\omega_0 \propto \omega_0^{-3}$ for larger energies, and exponential decay into the lower energies (which will again be mollified from photons descending from higher redshifts). However, for this case the position of the kink will be fixed at around 100 MeV, independent of the reheat temperature. Let us compare the above prediction with the recent COMPTEL and EGRET data. The EGRET collaboration found that the flux in the energy range $30\ \rm{ MeV} -100\ \rm{ GeV} $ is well fit by the single power law \cite{EGRET} \begin{equation} \label{EGRET} \frac{dJ}{d\omega_0}=7.32\pm 0.34\times 10^{-9} \left(\frac{\omega_0}{451\ \rm{ MeV} }\right)^{-2.10\pm 0.03}~ ({\rm cm^2~sec~sr~MeV})^{-1}, \end{equation} while the COMPTEL data \cite{COMPTEL} in the range $0.8$--$30\ \rm{ MeV} $ can be fit \cite{kribs} to the power law \begin{equation} \frac{dJ}{d\omega_0}=6.40\times 10^{-3} \left(\frac{\omega_0}{1\ \rm{ MeV} }\right)^{-2.38}~ ({\rm cm^2~sec~sr~MeV})^{-1}. \end{equation} Below $0.8\ \rm{ MeV} $ there is large increase in the measured flux. Thus, the best bounds are found by comparing the measured flux to predicted flux at $\omega_{\rm kink}$ or at $0.8\ \rm{ MeV} $, whichever is larger. Because of the rapid rise of the predicted spectrum relative to the measured spectrum, a change in the kink position can change the bound on $n$ on the order of $0.01$, which we consider within the accuracy of our calculation. The bounds on $n$ from the diffuse gamma-rays are specified by the dot-dashed line in Fig.~\ref{bound-fig}. The bound terminates when all but the exponential tail of the PBHs decay prior to a redshift of $700$, since the optical depth at such early times exceeds unity, as discussed above. We may compare our results to those derived by Yokoyama \cite{yokoyama},\footnote{After this work was completed, we became aware of Ref.~\cite{greensusy} that also utilized the critical collapse initial mass function to derive bounds on the PBH mass density by requiring that LSPs (in supersymmetric models) not be overproduced.} where the author placed bounds on mass fraction of PBHs at $t_{RH}$, $\beta(t_{RH}) = \Omega(t_{RH})$, using the initial mass function, Eq.~(\ref{IMF}). He found that the bounds on $\beta$ did not differ significantly from the previous bounds derived using the standard initial mass functions, except for the bounds coming from diffuse gamma-rays. In the latter case, applicable for horizon masses in the range $M_H\geq 5\times 10^{14}$~g, he found more stringent constraints. Our bounds on $n$, translated into bounds on $\beta$, agree with his bounds coming from energy density constraints except for the case of larger reheat temperature, since we included the proper scaling of the energy density of photons emitted after PBH decay. Thus our bounds on $\beta$ can differ by many orders of magnitude. Our bounds coming from diffuse gamma-rays can also differ by orders of magnitude, but in this case it is for a different reason. Yokoyama determined his bound on $\beta$ by imposing the constraint on $\Omega_{pbh}(t_0)$ derived in Ref.~\cite{mac2}. However, when we change the initial mass function we also change the diffuse gamma-ray spectrum significantly in both shape and normalization, as discussed above. Thus, it is inappropriate to directly take the bounds from Ref.~\cite{mac2} and apply them to the case with the new initial mass function, Eq.~(\ref{IMF}). We find that our bounds on $\beta$ from diffuse gamma-rays are more stringent than those determined in Ref.~\cite{yokoyama} in the range $M_H>5 \times 10^{15}$~g by several orders of magnitude. \section{Robustness of the Bounds} \label{robust-sec} Let us consider the robustness of the bounds. We might worry that the bounds are highly sensitive to the choice of parameters given the sharpness of the initial mass function. Indeed, in the case where it is assumed that the PBH mass is given by Eq.~(\ref{bhPropMh}), the bounds are $n$ are exceptionally sensitive to the exactness of this relation. This is clear from the exponential factor in Eq.~(\ref{KLM}). Given the initial mass function calculated by Jedamzik and Niemeyer, Eq.~(\ref{IMF}), we must check the sensitivity to the parameters $\delta_c,~k$ and $\sigma_0$. In Ref.~\cite{JZII}, the authors tested the scaling relation (\ref{scaling}) using several different initial shapes density perturbations shapes. They found $(\delta_c=0.70, ~k=11.9)$, $(\delta_c=0.67, ~k=2.85)$, $(\delta_c=0.71, ~k=2.39)$, for Gaussian, Mexican Hat and fourth order polynomial fluctuations, respectively. We varied the value of $\delta_c$ between $0.60-0.80$ and found that the bounds changed by at most $0.01$. The sensitivity to the variation being maximal at the smaller values of $T_{RH}$. Given that the initial mass function is peaked at a number smaller than $k M_h$, the sensitivity is increased at smaller $T_{RH}$ because $\sigma$ is an decreasing function of $T_{RH}$. Variations in $k$ are equivalent to a scaling in $T_{RH}$. Thus, varying $k$ by an order of magnitude has essentially no effect on the bound. Lastly, let us consider the sensitivity to the parameter $\sigma_0$. The value we used for $\sigma_0$ in Eq.~(\ref{COBE}) was calculated in Ref.~\cite{green2} using the result \cite{bunn}\footnote{It should be emphasized that this result assumed the spectra can be approximated as a power law over the range of $k$ that COBE probes. We are then making the further assumption that $n$ is constant down to the mass scales of relevance for PBHs.} \begin{equation} \label{fit} \delta_0=1.91\times 10^{-5} \frac{\exp[1.01(1-n)]}{\sqrt{1+0.75 r}} \; , \end{equation} where $r$ is a measure of the size of the tensor perturbations. The $1\sigma$ observational error being $7\%$. The fit, Eq.~(\ref{fit}), is good to within $1.5\%$ everywhere within the region $0.7 \leq n \leq 1.3$ and $0\leq r \leq 2$. The authors of Ref.~\cite{bunn} quote a $9\%$ uncertainty in Eq.~(\ref{fit}) at $1\sigma$, once uncertainties in the systematics and variations in the cosmological parameters are taken into account. The value in Eq.~(\ref{COBE}) was determined ignoring tensor perturbations. Given that $\sigma_0$ scales with $\delta_0$ we find that varying $\sigma_0$ at the $2\sigma$ level has no effect on our bound at the level of $0.01$. On the other hand, including some contribution from tensor perturbation will weaken the bound. We found that taking $r=2$ weakened the bound by $0.01-0.02$ throughout the range in the reheat temperature. We can also consider the effects of a non-vanishing $\Omega_\Lambda$. Bunn et~al.~\cite{bunn} extended their results to this case and found \begin{equation} \label{labmda} \delta_0|_{\Omega_\Lambda}=1.91\times 10^{-5} \frac{\exp[1.01(1-n)]} {\sqrt{1+(0.75-0.13\,\Omega_\Lambda^2)r}}\Omega_0^{-0.80-0.05\log{\Omega_0}} \, \left(1+0.18\,(n-1)-0.03\,r\,\Omega_\Lambda \right) \> . \end{equation} If $0 \le r \le 2$, we can express $\delta_0|_{\Omega_\Lambda}$ extracted assuming a nonzero cosmological constant to a very good approximation by a scaling of $\delta_0$ extracted without a cosmological constant \begin{equation} \delta_0|_{\Omega_\Lambda} \approx \Omega_0^{-0.80-0.05\log{\Omega_0}} \> \delta_0 \; , \end{equation} (where $\Omega_0 + \Omega_\Lambda = 1$) and thus $\sigma_0$ also acquires a correction. Consequently, the bound on $n$ is shifted by \begin{equation} \Delta n \equiv n - n|_{\Omega_\Lambda} = \frac{2 \> (-0.8 - 0.05 \ln \Omega_0) \, \ln \Omega_0}{42.9 + \ln (T_{RH}/10^8 \; \mathrm{GeV})} \; . \end{equation} In Fig.~\ref{delta-n-fig} we show the above correction as a function of $T_{RH}$ for several choices of $\Omega_\Lambda$. If we take $\Omega_\Lambda \approx 0.7$ as recent observational data suggests, our bounds on $n$ strengthen by about $0.03-0.06$ for $T_{RH}$ between $10^{16}-10^{3}$ GeV respectively, as shown in Fig.~\ref{bound-lambda-fig}. We can also calculate bounds on the mass variance at reheating \cite{liddle-grqc} which is essentially model-independent. If the relation Eq.~(\ref{dis1}) is violated by, for example, a power spectrum with a spectral index that depends on scale, then our previous bounds on $n$ cannot be applied. However, given a inflationary model one could in principle calculate the power spectrum, normalize to the COBE data at our present epoch, and then match onto the mass variance at reheating. In Fig.~\ref{sigma-fig} we show the bounds on the mass variance from both the density bounds as well as the bounds from the diffuse gamma-ray observations. Notice that the diffuse gamma-ray observation bounds on $\sigma(M_H)$ are a significant improvement over the density bounds in the applicable range of reheating temperatures. Finally, we must address the issue of non-Gaussianity. It has been pointed out \cite{bullock} that skewness could very well be important for PBH formation given that its effects are amplified in the tail of the distribution $P[\delta]$, which contributes to PBH formation. In general, the amount of non-Gaussianity expected is highly model dependent. Bullock and Primack investigated several inflationary models to study the amount of non-Gaussianity one would expect at larger values of $\delta$. They calculated $P[\delta]$ for three toy models, and found in one case no deviation from Gaussianity and in the other two found a significant suppression in the probability of of large perturbations. However, as was pointed out in Ref.~\cite{green}, while these effects can drastically effect the PBH mass fraction $\beta$, we expect that, even in the most extreme case considered in Ref.~\cite{bullock}, the effect on the bound on $n$ is only at the level of $0.05$. For hybrid inflation, where the approximation that $n$ is constant actually holds, the perturbations are in fact Gaussian due to the linear dynamics of the inflaton field \cite{Yi}. Therefore, these bounds should be applied to specific models, with the roughness of the bound determined by the deviations away from Gaussianity. \section{Conclusions} We have calculated the density of primordial black holes using the the near critical collapse mass function that results in a spectrum of PBH masses for a given horizon mass. The normalization of the PBH mass spectrum was determined using the COBE anisotropy data that allowed us to set bounds on the spectral index $n$ as a function on the reheat temperature. We find that restricting the density of PBHs to be less than the critical density corresponds to the restriction that the spectral index $n$ be less than about $1.45$ to $1.2$, throughout the range of reheating temperatures resulting after inflation, $10^{3}$ to $10^{16}$ GeV respectively. (The precise limits are shown in Fig.~\ref{bound-fig}.) For a smaller range of reheating temperatures, between about $10^{7}$ to $10^{10}$ GeV, significant PBH evaporation occurs when the optical depth of the universe is less than one. Hence, we found a slightly stronger bound on the spectral index by restricting the cosmological PBH evaporation into photons to be less than the present-day observed diffuse gamma-ray flux. Due to the extreme sensitivity of the PBH mass density to the spectral index, effects such as the indirect photon flux from PBH evaporation into quarks and gluons which fragment into pions or the formation of a QCD photosphere are completely negligible when calculating the bound on $n$. We should also remark that slightly stronger bounds on $n$ for larger reheating temperatures $\gtrsim 10^{10}$ GeV are expected from PBHs that decay during the epoch of nucleosynthesis. If the universe is vacuum-energy dominated, there are corrections to our bounds on $n$ that can be substantial. We calculated these corrections for a range of $\Omega_\Lambda$ and applied them to our bounds on $n$ for the case of $\Omega_\Lambda = 0.7$. The improvement is apparent by contrasting Fig.~\ref{bound-fig} with Fig.~\ref{bound-lambda-fig}. Finally, we calculated bounds on the mass variance at reheating. These bounds in principle could be used to constrain any given inflationary model, once the power spectrum is calculated. \acknowledgments This work was supported in part by the Department of Energy under grant number DOE-ER-40682-143. We thank Rich Holman and Jane MacGibbon for useful discussions. We also thank Andrew Liddle useful discussions and comments on the manuscript. {\tighten
\section{#1}} {\setcounter{equation}{0}}} \newtheorem{th}{Theorem}[section] \newtheorem{lm}{Lemma}[section] \newtheorem{prop}{Proposition}[section] \newtheorem{de}{Definition}[section] \newtheorem{co}{Corollary}[section] \newtheorem{re}{Remark}[section] \newtheorem{con}{Conjecture}[section] \def{K\"{a}hler }{{K\"{a}hler }} \def{K\"{a}hler-Einstein }{{K\"{a}hler-Einstein }} \def{Calabi-Yau }{{Calabi-Yau }} \def{Lagrangian }{{Lagrangian }} \def{special Lagrangian }{{special Lagrangian }} \input epsf \begin{document} \hbadness=10000 \title{{\bf Lagrangian torus fibration of quintic {Calabi-Yau } hypersurfaces I:}\\ {\Large {\bf Fermat quintic case}}} \author{Wei-Dong Ruan\\ Department of mathematics\\ Columbia University\\ New York, NY 10027\\ <EMAIL>} \date{Revised December 1999} \footnotetext{Partially supported by NSF Grant DMS-9703870.} \maketitle \begin{abstract} In this paper we give a construction of Lagrangian torus fibration for Fermat type quintic {Calabi-Yau } hypersurfaces via the method of gradient flow. We also compute the monodromy of the expected {special Lagrangian } torus fibration and discuss structures of singular fibers. \end{abstract} \tableofcontents \se{Introduction and background} In this paper we give a construction of Lagrangian torus fibration for Fermat type quintic {Calabi-Yau } manifolds via the method of gradient flow. This method will produce Lagrangian torus fibration for {Calabi-Yau } hypersurfaces in general toric varieties. The results for general quintic hypersurfaces appeared in \cite{lag3}; for {Calabi-Yau } hypersurfaces in general toric varieties are written in \cite{tor}.\\\\ The motivation of our work comes from the study of Mirror Symmetry. Mirror Symmetry conjecture originated from physicists' work in conformal field theory and string theory. It proposes that for a {Calabi-Yau } 3-fold $M$ there exists a {Calabi-Yau } 3-fold $W$ as its mirror manifold. The quantum geometry of $M$ and $W$ are closely related. In particular one can compute the number of rational curves in $M$ by solving the Picard-Fuchs equation coming from variation of Hodge structure of $W$. \\\\ The first example of mirror symmetry was the Fermat type quintic given in the calculations by Candelas et al (\cite{Can}). Numerous examples of such calculations were worked out after that. A more general construction via toric varieties was given by Batyrev (\cite{Bat}). For a more complete history and reference the readers can consult \cite{mirrorbook}.\\\\ In 1996 Strominger, Yau and Zaslow (\cite{SYZ}) proposed a geometric construction of mirror manifold via special Lagrangian torus fibration. According to their program (we will call it SYZ construction), a {Calabi-Yau } 3-fold should admit a special Lagrangian torus fibration. The mirror manifold can be obtained by dualizing the fibers. Or equivalently, the mirror manifold of $M$ is the moduli space of special Lagrangian 3-torus in $M$ with a flat $U(1)$ connection. This conjectural construction was the first to give the mirror manifold directly from a {Calabi-Yau } manifold itself. \\\\ The notion of special Lagrangian submanifolds was first given by Harvey and Lawson in their celebrated paper \cite{HL}. For a {Calabi-Yau } manifold $(X, \Omega, \omega_g)$, where $\Omega$ is a holomorphic $(n,0)$ form and $\omega_g$ is the {K\"{a}hler } form of the {Calabi-Yau } metric, a {Lagrangian } submanifold $L$ in $X$ is called {special Lagrangian } if $L$ is area minimizing, or equivalently $\Omega$ restrict to $L$ is a constant multiple of the volume form of $L$.\\\\ According to the SYZ construction, {special Lagrangian } submanifold and {special Lagrangian } fibration for {Calabi-Yau } manifolds seem to play very important roles in understanding mirror symmetry. However, despite its great potential in solving the mirror symmetry conjecture, our understanding in special Lagrangian submanifolds is very limited. The known examples are mostly explicit local examples or examples coming from $n=2$. There are very few examples of {special Lagrangian } submanifold or {special Lagrangian } fibration for dimension higher than two. M. Gross, P.M.H. Wilson, N. Hitchin, P. Lu and R. Bryant(\cite{Gross1}\cite{Gross2}\cite{GW}\cite{H}) did some work in this area in recent years. On the other extreme, in \cite{Z}, Zharkov constructed some non-{Lagrangian } torus fibration of {Calabi-Yau } hypersurfaces in toric variety.\\\\ When dimension $n=2$, {Calabi-Yau } manifold is hyper{K\"{a}hler }. Therefore for any {Calabi-Yau } metric, there is an $S^2$ family of complex structures that are compatible with the given {Calabi-Yau } metric. Special Lagrangian submanifold for one compatible complex structure exactly corresponds to complex curves for another compatible complex structure. Therefore {special Lagrangian } theory is reduced to the theory of complex curves in $X$, which is fairly well understood. For $n\geq 3$, there are no nice interpretations like this. \\\\ Given our lack of knowledge for {special Lagrangian }, one may consider relaxing the requirement to consider {Lagrangian } fibration. Special Lagrangians are very rigid and hard to find. On the other hand, {Lagrangian } submanifolds are more flexible and can be modified locally by Hamiltonian deformation. This is a reasonable first step to take. For many applications to mirror symmetry, especially those concerning (symplectic) topological structure of fibration, {Lagrangian } fibration will provide quite sufficient information. In this paper we mainly consider Lagrangian torus fibration of {Calabi-Yau } hypersurfaces in toric varieties. \\\\ Our idea is a very natural one. We try to use gradient flow to get Lagrangian torus fibration from a known Lagrangian torus fibration at the "Large Complex Limit". It will in principle be able to produce {Lagrangian } fibration in general {Calabi-Yau } hypersurfaces. For simplicity, it is helpful to explore the case of Fermat type quintic {Calabi-Yau } threefold in ${\bf CP^4}$ in detail first, which is the case studied in \cite{Can}. Most of the essential features for {Calabi-Yau } hypersurfaces in general toric varieties already show up here.\\\\ The paper is organized as follows: In Section 2 we will first describe a {Lagrangian } torus fibration for Fermat type quintic {Calabi-Yau } familly $\{X_\psi\}$ in ${\bf P^4}$\\ \[ p_{\psi}=\sum_1^5 z_k^5 - 5\psi \prod_{k=1}^5 z_k=0 \]\\ at the large complex limit $X_{\infty}$\\ \[ p_{\infty}=\prod_{k=1}^5 z_k=0. \]\\ In Section 3 we construct an appropriate vector field whose gradient flow will give {Lagrangian } fibration to nearby {Calabi-Yau } hypersurfaces. In Section 4, we will discuss the expected fibration structure by explicitly computing the monodromy transformations of the expected fibration if the fibration is actually {special Lagrangian } fibration. In Section 5 we will describe the expected structures of singular fibers implied by monodromy information. In Section 6 we will compare our {Lagrangian } fibration with the expected {special Lagrangian } fibration---their topological structure are not exactly the same; we will discuss the cause for their differences. Finally in Section 7, we will discuss the relevance to mirror construction for toric {Calabi-Yau } manifolds through the dual polyhedra construction.\\\\ This paper is the first part of several papers in this subject. In \cite{lag3} we discuss general construction of Lagrangian torus fibration for general quintic (non-Fermat) {Calabi-Yau } hypersurfaces. In \cite{lag2} we address the technical problem of modifying our construction into a {Lagrangian } torus fibration with the topological type of the expected one. \\\\ \se{{Lagrangian } torus fibration for large complex limit} Consider the well studied Fermat type quintic {Calabi-Yau } familly $\{X_\psi\}$ in ${\bf P^4}$ defined by\\ \[ p_{\psi}=\sum_1^5 z_k^5 - 5\psi \prod_{k=1}^5 z_k=0. \]\\ When $\psi$ approaches $\infty$, The familly approach its ``large complex limit'' $X_{\infty}$\\ \[ p_{\infty}=\prod_{k=1}^5 z_k=0. \]\\ $X_{\infty}$ is a union of five ${\bf P^3}$. There is a natural degenerate $T^3$ fibration structure for $X_{\infty}$. Let $\{P_i\}_{i=1}^5$ be five points in ${\bf R^4}$ that are in general position. Consider the natural map $F: \ P^4\longrightarrow R^4$.\\ \[ F([z])=\sum_{k=1}^5 \frac{|z_k|^2}{\sum_{i=1}^5 |z_i|^2}P_k. \]\\ Then $\Delta={\rm Image}(F)$ is a 4-simplex. $X_{\infty}$ is naturally fibered over $\partial \Delta$ via this map $F$ with general fiber being $T^3$. This is precisely the $T^3$ special Lagrangian fibration for $X_{\infty}$ as indicated by SYZ construction.\\\\ To see this fact, we need to construct a suitable {Calabi-Yau } metric on $X_{\infty}$ such that $F$ is a {special Lagrangian } torus fibration with respect to the {K\"{a}hler } form of the {Calabi-Yau } metric. Since $X_{\infty}$ is just a union of several ${\bf CP^3}$'s, we will concentrate on one of these ${\bf CP^3}$. We will actually carry out the discussion for general ${\bf CP^n}$.\\\\ In general, on $ {\bf C^{n+1}} $ there is a natural flat {K\"{a}hler } metric with {K\"{a}hler } form\\ \[ \omega_0 = \sum_{i=1}^{n+1} \frac{dz_i\wedge d\bar{z}_i}{|z_i|^2} \]\\ which is singular along the coordinate hyperplanes. Consider the natural projection $\pi: \ {\bf C^{n+1}}-\{0\}\rightarrow {\bf CP^{n}} $. The restriction of $\omega_0$ to the hypersurface $\prod_{i=1}^{n+1}z_i =1$ in $\bf C^{n+1}$ naturally push forward via $\pi$ to a flat {K\"{a}hler } metric $\omega$ on ${\bf CP^{n}}$ that is singular along the union of projective coordinate hyperplanes, which is exactly our large complex limit $Y_{\infty}$. (Here we use $Y_{\infty}$ to denote dimention $n-1$ version of the large complex limit in ${\bf CP^{n}}$ to distinguish from $X_\infty$ that corresponding to $n=4$ case.) Take local coordinate $x_i=\frac{z_i}{z_{n+1}}$ for $i=1,\cdots, n$. Then\\ \[ \omega = \sum_{i=1}^{n} \frac{dx_i\wedge d\bar{x}_i}{|x_i|^2} - \frac{1}{n+1} \left(\sum_{i=1}^{n} \frac{dx_i}{x_i}\right)\left(\sum_{i=1}^{n} \frac{d\bar{x}_i}{\bar{x}_i}\right) \]\\ It is nice to compute their {K\"{a}hler } potentials:\\ \[ \omega_0 = \partial \bar{\partial} \left( \sum_{i=1}^{n+1} (\log|z_i|^2)^2 \right) \]\\ \[ \omega = \partial \bar{\partial} \left( \sum_{i=1}^{n} (\log|x_i|^2)^2 - \frac{1}{n+1} (\sum_{i=1}^{n}\log|x_i|^2)^2\right) \]\\ Clearly\\ \[ \omega^n = (-1)^{\frac{n(n-1)}{2}} \frac{n!}{n+1}\Omega \wedge \bar{\Omega} \]\\ where\\ \[ \Omega = \bigwedge_{i=1}^n \frac{dx_i}{x_i} \]\\ is the holomorphic n-form on ${\bf CP^{n}}\backslash Y_{\infty}$. Notice that $\omega$ corresponds to a complete {Calabi-Yau } metric on ${\bf CP^{n}}\backslash Y_{\infty}$. Consider the n-dimension real torus $T^n$ represented by\\ \[ e^{i\theta} = (e^{i\theta_1}, e^{i\theta_2}, \cdots, e^{i\theta_n}) \]\\ as a real abelian Lie group. Then $T^n$ act naturally on ${\bf CP^{n}}$\\ \[ e^{i\theta}(x) = (e^{i\theta_1}x_1, e^{i\theta_2}x_2, \cdots, e^{i\theta_n}x_n) \]\\ as symplectomorphisms. The moment map is exactly $F: {\bf CP^{n}}\backslash Y_{\infty} \rightarrow {\bf R^n}$\\ \[ F(x)= ( \log |x_1|^2, \log |x_2|^2, \cdots, \log |x_n|^2) \]\\ It is easy to see that when $\Omega$ is restricted to the fibre, we have\\ \[ \Omega|_{T^n(x)} = i^n\bigwedge_{i=1}^n d\theta_i \]\\ Therefore $F$ naturally gives us the {special Lagrangian } fibration of ${\bf CP^{n}}\backslash Y_{\infty}$. Specialize to $n=3$ case, this construction gives us the {special Lagrangian } torus fibration with respect to a complete (flat) {Calabi-Yau } metric on $X_\infty\backslash {\rm Sing}(X_\infty)$.\\\\ When $\psi$ is large, according to SYZ conjecture, we {\it expect} that $X_{\psi}$ will also possess a special largrangian $T^3$ fibration with base identified with $\partial \Delta$ which is topologically an $S^3$. There are still serious analysis and geometric works to be done to totally justify the special Lagrangian fibration. We will instead give a natural construction of Lagrangian torus fibration.\\\\ For our purpose, we will consider the Fubini-Study metric\\ \[ \omega_{FS} = \partial \bar{\partial} \log(1+|x|^2). \]\\ The nice thing about the Fubini-Study metric is that the $T^n$ action is also a symplectomorphism with respect to $\omega_{FS}$. The corresponding moment map is\\ \[ F(x)= \left( \frac{|x_1|^2}{1+|x|^2}, \frac{|x_2|^2}{1+|x|^2}, \cdots, \frac{|x_n|^2}{1+|x|^2}\right) \]\\ which is easy to see if we write $\omega_{FS}$ in polar coordinates.\\ \[ \omega_{FS} = \partial \bar{\partial} \log(1+|x|^2)=i\sum_{j=1}^n d\theta_i \wedge d\left(\frac{|x_i|^2}{1+|x|^2}\right). \]\\ The most symmetric expression of the {K\"{a}hler } potential for the Fubini-Study metric is\\ \[ h= \frac{1 + |x|^2}{\left(\prod_{i=1}^n|x_i|^2\right)^{\frac{1}{n+1}}} = \frac{|z|^2}{\left( \prod_{i=1}^{n+1}|z_i|^2 \right)^{\frac{1}{n+1}}} \]\\ This expression can also be derived by restricting $|z|^2$ to the hypersurface $\prod_{i=1}^{n+1}z_i=1$ and then push down by $\pi$.\\\\ In general, we have the following lemma:\\ \begin{lm} For a {K\"{a}hler } metric\\ \[ \omega = \partial \bar{\partial} h(x) \]\\ $T^n$ acts as symplectomorphism if and only if $h$ can be chosen to depend only on $|x_i|^2$. \end{lm} {\bf Proof:} Assume that this is the case, then\\ \[ \omega = \partial \bar{\partial} h = i\sum_{j=1}^n d\theta_i \wedge d h_i. \]\\ where\\ \[ h_i=|x_i|^2\frac{\partial h}{\partial |x_i|^2}. \]\\ From above expression of $\omega$, clearly $T^n$ acts as symplectomorphism and\\ \[ F_h(x) = (h_1, h_2, \cdots, h_n) \]\\ is the moment map. \begin{flushright} $\Box$ \end{flushright} For the purpose of this paper, any $T^n$-invariant {K\"{a}hler } metric as in Lemma 2.1 is as good. We will mainly use the Fubini-Study metric.\\\\ \se{Lagrangian torus fibration via a gradient flow} \subsection{The gradient vector field} Consider the meromorphic function\\ \[ s= \frac{\prod_{k=1}^5 z_k}{\sum_1^5 z_k^5} \]\\ defined on ${\bf P^4}$. Let $\omega$ denote the {K\"{a}hler } form of a {K\"{a}hler } metric $g$ on ${\bf P^4}$, and $\nabla f$ denote the gradient vector field of real function $f=Re(s)$ with respect to the {K\"{a}hler } metric $g$. To describe the construction, we need the following facts.\\ \begin{lm} The gradient flow of $f$ leaves the set $\{Im(s)=0\}$ invariant and deforms {Lagrangian } submanifolds in $X_{\infty}$ to {Lagrangian } submanifolds in $X_{\psi}$.\\ \end{lm} {\bf Proof:} Clearly $\nabla f$ is always perpendicular to level sets of $f$. Therefore,\\ \[ J\nabla f (f)=0. \]\\ Let $s = f+ ih$. Since $s$ is holomorphic, we have\\ \[ \nabla f (h) = - J\nabla f (f)=0. \]\\ Therefore $h$ is constant along the gradient flow, or in another word, $\{Im(s)={\rm constant}\}$ is invariant under the gradient flow of $f$. In particular, $\{Im(s)=0\}$ is invariant under the gradient flow of $f$.\\\\ Notice the fact that\\ \[ \{Im(s)=0\} = \bigcup_{\psi ^{-1}\in {\bf R}} X_{\psi} \]\\ Let $L$ be a {Lagrangian } submanifold of $X_{\infty}$. To prove the lemma, it is sufficient to show that the 4-dimension submanifold $S$ swept out by $L$ under the gradient flow of $f$ is {Lagrangian } submanifold of ${\bf P^4}$. \\\\ First, for a vector field $v$ on $L$, since $L\subset X_{\infty}$ and $\nabla f$ is perpendicular to level sets of $f$ (in particular, perpendicular to $X_{\infty}$), we see that $Jv$ is along $X_{\infty}$ and $\nabla f$ is perpendicular to $Jv$ along $L$. Therefore, $\omega(v,\nabla f)=0$. \\\\ Let $V$, $W$ denote vector fields on $S$ that are invariant under the gradient flow $\phi_t$. $\nabla f$ is a vector field of this type. Using the fact that $f$ is pluri-harmonic and $\omega_g$ is {K\"{a}hler } form, it is easy to derive that\\ \begin{eqnarray*} \frac{d}{dt}\omega_g(V,W) &=& -{\cal L}_{\nabla f} \omega_g(V,W) = -(d i(\nabla f)\omega_g)(V,W)\\ &=& (dJdf)(V,W) = -2i\partial \bar{\partial} f(V,W)=0. \end{eqnarray*} Therefore $\omega_g(V,W)$ is constant along the flow. Since initial value of invariant vector fields are spanned by vector fields along $L$ and $\nabla f$, by the {Lagrangian } property of $L$ and the fact that $\omega(v,\nabla f)=0$ for vector field $v$ on $L$, we have $\omega_g(V,W)=0$, therefore $S$ is {Lagrangian }. \begin{flushright} $\Box$ \end{flushright} {\bf Remark:}\\ (i) The lemma can also be understood roughly by the fact: $\nabla f = H_h$ (the hamiltonian vector field generated by $h=Im(s)$). At the smooth part of the vector field, this fact essentially implies the lemma, although at singular part of the vector field, additional argument as in the proof is needed.\\\\ (ii) The proof of the lemma actually implies something more. Lower dimension torus in $X_{\infty}$ form critical set of $f$. The proof of the lemma also implies that the stable manifold of a lower dimensional torus with respect to flow of $\nabla f$ is {Lagrangian } in ${\bf P^4}$, therefore intersect with $X_{\psi}$ at a {Lagrangian } submanifold. \begin{flushright} $\Box$ \end{flushright} With this lemma in mind, the construction of {Lagrangian } torus fibration of $X_{\psi}$ for $\psi$ large is immediate. Recall that in last section we had a canonical Lagrangian torus fibration of $X_{\infty}$ over $\partial \Delta$. Deform along gradient flow of $f$ will naturally induce a {Lagrangian } torus fibration of $X_{\psi}$ over $\partial \Delta$ for $\psi$ large and real.\\\\ However $X_{\infty}$ is singular and $\nabla f$ is also singular where $s$ is singular. To get a really honest {Lagrangian } torus fibration, we need to discuss how to deal with these singularities.\\\\ Along the gradient flow of $f$ we have\\ \[ \frac{df}{dt} = \nabla f \cdot \frac{dx}{dt} = \nabla f \cdot \nabla f = |\nabla f|^2 \]\\ We see the gradient flow of $f$ does not exactly move $X_{\infty}$ to $X_{\psi}$. To ensure this property, the flow has to satisfy $\frac{df}{dt} = {\rm constant}$. This will be true if we scale the vector field $\nabla f$ to\\ \[ V=\frac{\nabla f}{|\nabla f|^2}. \]\\ \subsection{Flow from the smooth part of $X_{\infty}$} To understand the flow of $\nabla f$ and $V$, it is helpful to express them in local coordinate. Recall that $X_{\infty}=\cup_{k=1}^5 D_k$, where $D_k=\{z_k=0\}$. Let's choose coordinate $x_i=z_i/z_5$, for $i=1,2,3,4$ and consider $D_4$, where $x_4=0$. Under this coordinate, we have\\ \[ s=\frac{\prod_{i=1}^4x_i}{\sum_{i=1}^4x_i^5 +1} \]\\ \[ ds =\frac{\prod_{i=1}^4x_i}{\left(\sum_{i=1}^4x_i^5 +1\right)^2}\sum_{i=1}^4 \left(\sum_{j=1}^4x_j^5 +1-5x^5_i\right)\frac{dx_i}{x_i} \]\\ \[ ds|_{D_4} =\frac{\prod_{i=1}^3x_i}{\sum_{i=1}^3x_i^5 +1}dx_4 \]\\ For simplicity, we choose {K\"{a}hler } metric to be\\ \[ g=\sum_{i=1}^4 dx_i\wedge d\bar{x}_i \]\\ Then when restricted to $D_4$, we have\\ \[ |ds|^2 = \left|\frac{\prod_{i=1}^3x_i}{\sum_{i=1}^3x_i^5 +1}\right|^2 \]\\ \[ \nabla s = \frac{\prod_{i=1}^3x_i}{\sum_{i=1}^3x_i^5 +1} \frac{\partial}{\partial \bar{x}_4} \]\\ These give us\\ \[ V=\frac{\nabla f}{|\nabla f|^2} = Re\left(\frac{\sum_{i=1}^3x_i^5 +1}{\prod_{i=1}^3x_i} \frac{\partial}{\partial x_4}\right) \]\\ So we see that {\bf the flow of $V$ is smooth when restricted to the smooth part of $X_{\infty}$.}\\\\ Another way to understand $V$ is to realize that $ds$ is a holomorphic section of $N^*_X$, $(ds)^{-1}$ is a natural holomorphic section of $N_X$. Notice the exact sequence\\ \[ 0 \rightarrow T_X \rightarrow T_{\bf P^4}|_X \rightarrow N_X \rightarrow 0 \]\\ With respect to the {K\"{a}hler } metric $g$ on $T_{\bf P^4}$, the exact sequence has a natural (non-holomorphic) splitting\\ \[ T_{\bf P^4}|_X =T_X \oplus N_X \]\\ $V$ is just real part of the natural lift of $(ds)^{-1}$ via this splitting. $V$ is singular exactly when $(ds)^{-1}$ is singular or equivalently, when $ds=0$, which corresponds to singular part of $X_{\infty}$. On the other hand, the union of the smooth three-dimensional {Lagrangian } torus fibers of $X_{\infty}$ is exactly the smooth part of $X_{\infty}$. All these 3-torus fibers will be carried to $X_{\psi}$ nicely by the flow of $V$.\\ \subsection{Flow from the singular part of $X_{\infty}$} Now we will try to understand how the gradient flow of $f$ behaves at singularities of $X_{\infty}$. For this purpose, it is helpful to understand the following example.\\\\ {\bf Example:} Consider holomorphic function $s(z)=e^{i\theta}z_1z_2\cdots z_n$ on ${\bf C^n}$. $X=\{s=0\}$ is a variety with normal crossing singularities. There is a natural map \\ \[ F:\ {\bf C^n} \rightarrow {\bf R^n_+},\ \ F(z_1,\cdots,z_n) = (|z_1|,\cdots,|z_n|). \]\\ For $c\in {\bf R^n_+}$, the fiber\\ \[ F^{-1}(c) = \{z:|z_i|=c_i,\ {\rm for}\ i=1,\cdots,n \} \]\\ is $n$-torus for generic $c$. Let $f=Re(s)$, then with respect to the flat metric on ${\bf C^n}$, we have\\ \[ \nabla f = {\rm Re}\left( s\sum_{i=1}^n \frac{1}{z_i}\frac{\partial}{\partial \bar{z}_i}\right) \]\\ By our previous argument, $h(z) ={\rm Im}(s)=Im(e^{i\theta}z_1z_2\cdots z_n)$ is invariant under gradient flow of $f$. It is also easy to observe that $\nabla f$ will leave $h_{ij}(z)=|z_i|^2 -|z_j|^2$ invariant. For $c=(c_1,\cdots, c_n)$ such that $\sum c_i =0$, let\\ \[ S_c = \{z: h(z)=0,\ h_{ij} =c_{ij}=c_i-c_j,\ {\rm for}\ i,j=1,\cdots,n \}. \]\\ It is easy to see that $S_c$ are special Lagrangians in ${\bf C^n}$. Actually this is an example mentioned in Harvey and Lawson's paper. Let $X_r = \{z: s(z)=r\}$, then $L_{c,r} = S_c\cap X_r$ give us a smooth {Lagrangian } $n$-torus fibration of $X_r$ for $r$ real.\\\\ Define\\ \[ \theta_{ij}= \theta_i-\theta_j = \frac{i}{2}\left( \log\frac{\bar{z}_i}{z_i} -\log\frac{\bar{z}_j}{z_j}\right), \]\\ where $\theta_i$ is the argument of $z_i$. Then\\ \[ \nabla f (\theta_{ij}) =-\frac{1}{2} {\rm Im} \left(\frac{s}{|z_i|^2} - \frac{s}{|z_j|^2}\right)=-\frac{1}{2} \left(\frac{1}{|z_i|^2} - \frac{1}{|z_j|^2}\right)h(z). \]\\ Recall that the real hypersurface $\{h(z)=0\}$ is invariant under the flow. Above equation implies that when restricted to $\{h(z)=0\}$ $\theta_{ij}$ is invariant. For $c=(c_1,\cdots, c_n)$ such that $\sum c_i =0$, let\\ \[ T_c = \{z: h(z)=0,\ \theta_{ij} =c_{ij}=c_i-c_j,\ {\rm for}\ i,j =1,\cdots,n \}. \]\\ There is another very illustrative way to write $T_c$.\\ \[ T_c = \{z: \theta_i-\theta_j=c_{ij}=c_i-c_j, \theta+\sum_{i=1}^n\theta_i = 0\}=\{z:\theta_i=c_i-\frac{\theta}{n}\} \]\\ It is easy to see that $T_c$ are special Lagrangians in ${\bf C^n}$ that is invariant under the flow. $T_{c,r} = T_c\cap X_r$ give us a smooth {Lagrangian } fibration on the horizontal directions of $X_r$ for $r$ real. \begin{flushright} $\Box$ \end{flushright} As we know, $X_{\infty}$ has only normal crossing singularities. The above example gave us a rough picture of the local behavior of the gradient flow of $f$ around singularities of $X_{\infty}$ {\bf when denominator of $s$ is non-zero}. For detailed discussion and proof, please refer to \cite{lag2}.\\\\ Finally it remains to analyze the case when denominator of $s$ is zero and $X_{\infty}$ is singular, which is the following set $\Sigma$. Let\\ \[ \Sigma_{ijk} =\{[z]\in {\bf CP^4}| z_i^5 + z_j^5 + z_k^5 =0, z_l=0\ \ {\rm for}\ l\in\{1,2,3,4,5\}\backslash\{i,j,k\}\}. \] $\Sigma_{ijk}$ is a genus 6 curve. Let\\ \[ \Sigma = \bigcup_{\{i,j,k\}\in \{1,2,3,4,5\}} \Sigma_{ijk} \] We see that $\Sigma = {\rm Sing}(X_{\infty})\cap X_{\psi}$ for any $\psi$.\\\\ In general, points in $X_{\infty}\cap X_{\psi}$ are fixed under the flow. In particular, $\Sigma \subset X_{\infty}\cap X_{\psi}$ is fixed under the flow. Since the vector field $V$ is singular along $\Sigma$, more argument is needed to ensure the flow behave as expected near $\Sigma$. For the argument to work, it is actually necessary to deform the {K\"{a}hler } form slightly near $\Sigma$. For details, please also refer to \cite{lag2}.\\ \subsection{Lagrangian torus fibration structure} To understand the {Lagrangian } torus fibration of $X_{\psi}$, it is helpful to first review the torus fibration of $X_{\infty}$. For any subset $I \subset \{1,2,3,4,5\}$, Let\\ \[ D_I = \{z:\ z_i=0,z_j\not= 0,\ {\rm for}\ i\in I,j\in\{1,2,3,4,5\}\backslash I\}, \]\\ $\Delta_I = F(D_I)\subset \partial\Delta$. Let $|I|$ denote the cardinality of $I$. We have\\ \[ \partial\Delta = \bigcup_{\tiny{\begin{array}{c}I \subset \{1,2,3,4,5\}\\ 0<|I|<5\end{array}}} \Delta_I \]\\ The fibers over $\Delta_I$ are $4-|I|$ dimensional torus.\\\\ The flow of $V$ moves $X_{\infty}$ to $X_{\psi}$. When $|I|=1$, the flow is diffeomorphism on $D_I$ and moves the corresponding smooth 3-torus fibration in $D_I$ to 3-torus fibration in $X_{\psi}$.\\\\ When $|I|>1$, by above discussion after the example, each point of $D_I\backslash \Sigma$ will be deform to a $|I|-1$ torus under the flow of $V$. Therefore a torus fiber in $D_I$ not intersecting $\Sigma$ will be deformed to a smooth 3-torus {Lagrangian } fiber of $X_{\psi}$ under the flow of $V$.\\\\ Now let's try to understand the singular fibers. Notice that $D_I\cap\Sigma\not= \emptyset$ if and only if $|I|=2,3$. Let $\tilde{\Gamma}_{ijk} = F(\Sigma_{ijk})$. There are 3 types of singular fiber over different portion of\\ \[ \tilde{\Gamma} = \bigcup_{\{i,j,k\}\in \{1,2,3,4,5\}} \tilde{\Gamma}_{ijk} \]\\ Let $\tilde{\Gamma}^2$ denote the interior of $\tilde{\Gamma}$, and\\ \[ \tilde{\Gamma}^1 = \partial \tilde{\Gamma} \cap \left(\bigcup_{|I|=2}\Delta_I\right), \] \[ \tilde{\Gamma}^0 = \partial \tilde{\Gamma} \cap \left(\bigcup_{|I|=3}\Delta_I\right). \]\\ Then\\ \[ \tilde{\Gamma} = \tilde{\Gamma}^0 \cup \tilde{\Gamma}^1 \cup \tilde{\Gamma}^2. \]\\ Without loss of generality, let us concentrate on\\ \[ \overline{D}_{45} = \{z: z_4=z_5=0\} \]\\ with the natural coordinate $x_i=z_i/z_3$ for $i=1,2$. Under this coordinate\\ \[ \Sigma_{123} =\{x=(x_1,x_2)| x_1^5 + x_2^5 + 1 =0\}. \]\\ $r=(r_1,r_2)=(|x_1|,|x_2|)$ can be thought of as coordinate on $\overline{\Delta}_{45}$. Under this coordinate\\ \[ \tilde{\Gamma}_{123} = \{r=(r_1,r_2)|r_1^5 + r_2^5 \geq 1, r_1^5 +1 \geq r_2^5, r_2^5 +1 \geq r_1^5\} \]\\ For simplicity, we will omit the index and denote $\Sigma_{123}$ by $\Sigma$ and $\tilde{\Gamma}_{123}$ by $\tilde{\Gamma}$. Then\\ \[ \tilde{\Gamma}^2 = \{r=(r_1,r_2)|r_1^5 + r_2^5 > 1, r_1^5 +1 > r_2^5, r_2^5 +1 > r_1^5 \}, \] \[ \tilde{\Gamma}^0 = \{(1,0),(0,1)\}, \] \[ \tilde{\Gamma}^1 = \{r=(r_1,r_2)|r_1^5 + r_2^5 = 1\ {\rm or}\ r_1^5 = r_2^5 + 1\ {\rm or}\ r_2^5 = r_1^5 + 1\}\backslash \tilde{\Gamma}^0. \]\\ $(x_1,x_2)\rightarrow (e^{\frac{2\pi i}{5}k_1}x_1,e^{\frac{2\pi i}{5}k_2}x_2)$ give us a natural action of ${\bf Z_5\times Z_5}$ on $\Sigma$. Combine with the ${\bf Z_2}$ action $(x_1,x_2)\rightarrow (\bar{x}_1,\bar{x}_2)$ we have\\ \[ \begin{array}{ccc} {\bf Z_2\times Z_5\times Z_5}&\rightarrow& \Sigma\\ &&\\ &&\downarrow^F\\ &&\\ &&\tilde{\Gamma}\\ \end{array} \]\\ $F$ is a $50$-sheet covering map over $ \tilde{\Gamma}^2$. ${\bf Z_2}$ is the stablizer of the group action on $F^{-1}(\tilde{\Gamma}^1)\cap\Sigma$. For $p\in \tilde{\Gamma}^1$, $F^{-1}(p)$ is a 2-torus that intersect $\Sigma$ at $25$ points. ${\bf Z_2\times Z_5}$ is the stablizer of the group action on $F^{-1}(\tilde{\Gamma}^0)\cap\Sigma$. For $p\in \tilde{\Gamma}^0$, $F^{-1}(p)$ is a circle that intersect $\Sigma$ at $5$ points.\\\\ From our previous discussion, when $|I|>1$, each point of $D_I\backslash \Sigma$ will be deform to a $|I|-1$ torus under the flow of $V$ and points in $\Sigma$ will not move under the flow of $V$. Let $L_{\infty}\subset D_I \subset X_{\infty}$ be the original fiber of $F$ that move to {Lagrangian } fiber $L\subset X_{\psi}$ under the flow of $V$. Then follow the flow backward we will have a fibration $\pi: L \rightarrow L_{\infty}$. Over $L_{\infty}\backslash \Sigma$, $\pi: L\backslash \Sigma \rightarrow L_{\infty}\backslash \Sigma$ is a $|I|-1$ torus smooth fibration. Over $L_{\infty}\cap \Sigma$, $\pi: L\cap \Sigma \cong L_{\infty}\cap \Sigma$ is an identification.\\\\ It is now easy to see that for $p\in \tilde{\Gamma}^2$, $F^{-1}(p)$ under the flow of $V$ will deform to {Lagrangian } 3-torus with $50$ circles collapsed to $50$ singular points. For $p\in \tilde{\Gamma}^1$, $F^{-1}(p)$ under the flow of $V$ will deform to {Lagrangian } 3-torus with $25$ circles collapsed to $25$ singular points. For $p\in \tilde{\Gamma}^0$, $F^{-1}(p)$ under the flow of $V$ will deform to {Lagrangian } 3-torus with $5$ two-torus collapsed to $5$ singular points. Now we have finished the discussion of all {Lagrangian } fibers of our {Lagrangian } fibration of $X_{\psi}$.\\ \begin{th} Flow of $V$ will produce a {Lagrangian } fibration $F: X_{\psi} \rightarrow \partial\Delta$. There are 4 types of fibers.\\ (i). For $p\in \partial\Delta\backslash \tilde{\Gamma}$, $F^{-1}(p)$ is a smooth {Lagrangian } 3-torus.\\ (ii). For $p\in \tilde{\Gamma}^2$, $F^{-1}(p)$ is a {Lagrangian } 3-torus with $50$ circles collapsed to $50$ singular points.\\ (iii). For $p\in \tilde{\Gamma}^1$, $F^{-1}(p)$ is a {Lagrangian } 3-torus with $25$ circles collapsed to $25$ singular points.\\ (iv). For $p\in \tilde{\Gamma}^0$, $F^{-1}(p)$ is a {Lagrangian } 3-torus with $5$ two-torus collapsed to $5$ singular points. \end{th} \begin{flushright} $\Box$ \end{flushright} \se{Monodromy of expected {special Lagrangian } fibration} \subsection{Introduction and assumptions} In this section we will explore what the SYZ {special Lagrangian } fibration of $X_{\psi}$ should look like if it exists. Recall from Section 2 that $X_{\infty}$ is a union of five ${\bf P^3}$. There is a natural degenerate $T^3$ fibration structure for $X_{\infty}$. Let $\{P_i\}_{i=1}^5$ be five points in ${\bf R^4}$ that are in general position. Consider the natural map $F: \ {\bf P^4}\longrightarrow {\bf R^4}$.\\ \[ F([z])=\sum_{k=1}^5 \frac{|z_k|^2}{\sum_{i=1}^5 |z_i|^2}P_k \]\\ $\Delta={\rm Image}(F)$ is a 4-simplex. $X_{\infty}$ is naturally fibered over $\partial \Delta$ via this map $F$ with general fiber being $T^3$. This is precisely the $T^3$ special Lagrangian fibration for $X_{\infty}$ as indicated by SYZ construction.\\\\ When $\psi$ is large, we expect that $X_{\psi}$ will also possess a special Largrangian $T^3$ fibration with base identified with $\partial \Delta$ which topologically is an $S^3$. There are still serious analysis and geometric works to be done to totally justify the special Lagrangian fibration. We will discuss in this section that {\bf IF} such special Lagrangian fibration exist on $X_{\psi}$, what should be its expected topological and geometrical structures.\\\\ Our discussion is based on two principle assumptions:\\ (1) {special Lagrangian } fibration on $X_{\psi}$ is a deformation of {special Lagrangian } fibration on $X_{\infty}$. \\ (2) Singular locus of the {special Lagrangian } fibration on $X_{\psi}$ is of codimension 2. \\\\ (1) is very natural, because we expect the {Calabi-Yau } metric on $X_{\psi}$ suitably rescaled (keeping fiber class constant volume) will approach the standard {Calabi-Yau } metric on $X_{\infty}$ when $\psi$ approach infinity. (2) is also very reasonable given the structure of elliptic fibration of complex surface that correspond to dimension $n=2$ situation.\\\\ For $\psi$ large, $X_{\psi}$ will approach $X_{\infty}=\cup_{k=1}^5 D_k$, where $D_k=\{z_k=0\}$. Let us consider the part of the $X_{\psi}$ that is close to $D_5$. (We denote it as $U_5$, for example we may take $U_5$ as inverse image in $X_{\psi}$ of an open set in $D_5$ that stay away from $D_k$ for $k\not=5$ by the following map $\pi_5$.) Then the restriction of the projection\\ \[ \pi_5:\ U_5 \longrightarrow D_5,\ \ \pi_5([z_1,z_2,z_3,z_4,z_5])=[z_1,z_2,z_3,z_4,0] \] will identify $U_5$ with an open set in $D_5$, which at the same time will carry over the $T^3$ fibration to $U_5$ from $D_5$. When $\psi$ is large, this fibration should be very close to the special Lagrangian $T^3$ fibration for $U_5$. When we discuss the topology aspect, it will be sufficient to use the induced fibration instead of the special Lagrangian fibration when staying away from intersections of $D_k$'s, where the special Lagrangian fibration of $X_{\infty}$ degenerate. We will use these identifications to compute monodromy of the special Lagrangian fibration in the following.\\ \subsection{One forms and one cycles} We first fix some notation and introduce some construction. On $D_5$, let $\gamma^1_{52}$ be the circles determined by $\{z_5=0,\ |\frac{z_1}{z_2}| = c_1,\ \frac{z_3}{z_2}=c_2,\ \frac{z_4}{z_2}=c_3\}$. We will also use it to denote circles carried over to $U_5$. In general, we have $\gamma_{ik}^j$ for $\{i,j,k\} \subset \{1,2,3,4,5\}$. Understanding monodromy is equivalent to understanding transformations among $\{\gamma_{ik}^j\}$. It is easy to check that\\ \begin{eqnarray} \gamma_{ik}^j &=& \gamma_{il}^j,\ \ \ \ \ {\rm for}\ \{i,j,k,l\}\subset \{1,2,3,4,5\}.\label{ba}\\ \gamma_{ik}^j &=& -\sum_{l(\not=i,j)=1}^5\gamma_{ij}^l,\ \ {\rm for}\ \{i,j,k\}\subset \{1,2,3,4,5\}.\nonumber \end{eqnarray} On ${\bf P^4}$ we can introduce meromorphic 1-forms\\ \[ \alpha_{ij}= d(\log \frac{z_i}{z_j}), \ \ i \not=j. \] They have the simple relations\\ \[ \alpha_{ij}= \alpha_{ik}+ \alpha_{kj}\ {\rm for}\ \{i,j,k\}\subset \{1,2,3,4,5\}. \] We also use the same notation for the restriction to $X_{\psi}$. $\alpha_{ij}$ have pole along $D_j\cap X_{\psi}$ on $X_{\psi}$, where\\ \[ D_j\cap X_{\psi}=\left\{z\left| z_j=0,\ \sum_{i(\not=j)=1}^5 z_i^5 = 0\right. \right\} \] For this reason we introduce\\ \[ U_j^i = \left\{z\in U_j \left|\ \sum_{k(\not= i,j)=1}^5\left|\frac{z_k}{z_i}\right|^5<1\right. \right\}. \] On $U_j^i$, all $\alpha_{kl}$ are regular. On $U_i^j$ we have circles \\ \[ \gamma_{ij}^k, \ {\rm for}\ k\in \{1,2,3,4,5\}\backslash \{i,j\} \] and 1-forms\\ \[ \alpha_{kj},\ {\rm for}\ k\in \{1,2,3,4,5\}\backslash \{j\} \] It is easy to check that\\ \[ <\gamma_{ij}^k, \alpha_{lj}> = \delta_{kl}\ {\rm for}\ \{i,j,k,l\}\subset \{1,2,3,4,5\} \] and\\ \[ <\gamma_{ij}^k, \alpha_{ij}> = -1\ {\rm for}\ \{i,j,k\}\subset \{1,2,3,4,5\} \] Use these relations for $U_i^j$ and $U_k^j$, we will get\\ \begin{eqnarray} \gamma_{ij}^k&=&-\gamma_{kj}^i,\ \ \ \ {\rm for}\ \{i,j,k\}\subset \{1,2,3,4,5\}\label{bb}\\ \gamma_{ij}^l&=&-\gamma_{kj}^i+\gamma_{kj}^l,\ \ {\rm for}\ \{i,j,k,l\}\subset \{1,2,3,4,5\}\nonumber \end{eqnarray} From (\ref{ba}) we can see that without confusion, we may denote $\gamma_{ik}^j$ by $\gamma_i^j$. then the relations (\ref{ba}) and (\ref{bb}) can be rewrite together as\\ \begin{eqnarray} \sum_{j(\not=i)=1}^5\gamma_i^j&=&0,\ \ {\rm for}\ \{i\}\subset \{1,2,3,4,5\}\nonumber\\ \gamma_i^j+\gamma_j^i&=&0,\ \ \ {\rm for}\ \{i,j\}\subset \{1,2,3,4,5\}\label{bc}\\ \gamma_i^j+\gamma_j^k+\gamma_k^i&=&0,\ \ \ {\rm for}\ \{i,j,k\}\subset \{1,2,3,4,5\}\nonumber \end{eqnarray} Let\\ \[ U^i = \left\{z\in X_{\psi} \left|\ \sum_{k(\not= i)=1}^5\left|\frac{z_k}{z_i}\right|^5<1\right. \right\}. \] Then $\alpha_{ki}$, for $k\in \{1,2,3,4,5\}\backslash \{i\}$ are regular on $U^i$. So there are no monodromy among $U^i_j\subset U^i$ for $j\in \{1,2,3,4,5\}\backslash \{i\}$.\\\\ On the other hand, in $U_j$ the $T^3$ fibration is regular. So there are also no monodromy among $U^i_j\subset U_j$ for $j\in \{1,2,3,4,5\}\backslash \{i\}$. From these discussions we can see that the discriminant locus of the fibration (where the $T^3$ fibration is singular) is topologically a graph $\Gamma$ in $\partial \Delta$. Vertices of $\Gamma$ are $P_{ij}$ (baricenter for 2-simplex $\Delta_{ij}$) for $\{i,j\}\subset \{1,2,3,4,5\}$ and $P_{ijk}$ (baricenter for 3-simplex $\Delta_{ijk}$) for $\{i,j,k\}\subset \{1,2,3,4,5\}$. Legs of $\Gamma$ are $\Gamma_{ij}^k$ which connects $P_{ijk}$ and $P_{ij}$ for $\{i,j,k\}\subset \{1,2,3,4,5\}$.\\\\ \begin{figure}[h] \begin{center} \leavevmode \hbox{% \epsfxsize=4in \epsffile{p2.ps}} \end{center} \caption{} \end{figure}\\ \subsection{Monodromy around the legs of $\Gamma$} We would like to compute monodromy $T_{ij}^k$ around $\Gamma_{ij}^k$. To simplify the notation without loss of generality, we will compute $T_{24}^3$. This amounts to compute the transformation:\\ \[ U_5^4\rightarrow U_5^2\rightarrow U_1^2\rightarrow U_1^4\rightarrow U_5^4 \] Let $S_i^{jk}$ denote the transformation: $U_i^j\rightarrow U_i^k$ and $S^i_{jk}$ denote the transformation: $U^i_j\rightarrow U^i_k$ for $\{i,j,k\}\subset \{1,2,3,4,5\}$. Then\\ \[ T_{24}^3= S_{15}^4 S_1^{24} S_{51}^2 S_5^{42} \] For $U_5^4\rightarrow U_5^2$ we have\\ \[ S_5^{42}\left( \begin{array}{ccc} \gamma_{54}^1 & \gamma_{54}^2 & \gamma_{54}^3 \end{array} \right) = \left( \begin{array}{ccc} \gamma_{52}^1 & \gamma_{52}^3 & \gamma_{52}^4 \end{array} \right) \left( \begin{array}{ccc} 1 & -1 & 0\\ 0 & -1 & 1\\ 0 & -1 & 0 \end{array} \right) \] For $U_5^2\rightarrow U_1^2$ we have\\ \[ S_{51}^2\left( \begin{array}{ccc} \gamma_{52}^1 & \gamma_{52}^3 & \gamma_{52}^4 \end{array} \right) = \left( \begin{array}{ccc} \gamma_{12}^3 & \gamma_{12}^4 & \gamma_{12}^5 \end{array} \right) \left( \begin{array}{ccc} 0 & 1 & 0\\ 0 & 0 & 1\\ -1& -1& -1 \end{array} \right) \] For $U_1^2\rightarrow U_1^4$ we have\\ \[ S_1^{24}\left( \begin{array}{ccc} \gamma_{12}^3 & \gamma_{12}^4 & \gamma_{12}^5 \end{array} \right) = \left( \begin{array}{ccc} \gamma_{14}^2 & \gamma_{14}^3 & \gamma_{14}^5 \end{array} \right) \left( \begin{array}{ccc} 0 & -1 & 0\\ 1 & -1 & 0\\ 0 & -1 & 1 \end{array} \right) \] For $U_1^4\rightarrow U_5^4$ we have\\ \[ S_{15}^2\left( \begin{array}{ccc} \gamma_{14}^2 & \gamma_{14}^3 & \gamma_{14}^5 \end{array} \right) = \left( \begin{array}{ccc} \gamma_{54}^1 & \gamma_{54}^2 & \gamma_{54}^3 \end{array} \right) \left( \begin{array}{ccc} -1&-1 &-1\\ 1 & 0 & 0\\ 0 & 1 & 0 \end{array} \right) \] Therefore\\ \[ T_{24}^3= S_{15}^4 S_1^{24} S_{51}^2 S_5^{42} \] \[ =\left( \begin{array}{ccc} -1&-1 &-1\\ 1 & 0 & 0\\ 0 & 1 & 0 \end{array} \right) \left( \begin{array}{ccc} 0 & -1 & 0\\ 1 & -1 & 0\\ 0 & -1 & 1 \end{array} \right) \left( \begin{array}{ccc} 0 & 1 & 0\\ 0 & 0 & 1\\ -1& -1& -1 \end{array} \right) \left( \begin{array}{ccc} 1 & -1 & 0\\ 0 & -1 & 1\\ 0 & -1 & 0 \end{array} \right) \] \[ =\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right) \] and\\ \[ T_{24}^3 \left( \begin{array}{ccc} \gamma_{54}^1 & \gamma_{54}^2 & \gamma_{54}^3 \end{array} \right)=\left( \begin{array}{ccc} \gamma_{54}^1 & \gamma_{54}^2 & \gamma_{54}^3 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right) \] In general for $\{i,j,k,l,m\}=\{1,2,3,4,5\}$ representing the same orientation with the written order\\ \[ T_{jl}^k \left( \begin{array}{ccc} \gamma_{m}^i & \gamma_{m}^j & \gamma_{m}^k \end{array} \right)=\left( \begin{array}{ccc} \gamma_{m}^i & \gamma_{m}^j & \gamma_{m}^k \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right), \]\\ where the monodromy is computed along the path\\ \[ U_m^l\rightarrow U_m^j\rightarrow U_i^j\rightarrow U_i^l\rightarrow U_m^l. \]\\ If orientation is different, then $-5$ should be replaced by $5$.\\ \subsection{Monodromy around baricenter of 2-cell} Now we compute the monodromy around a baricenter of a 2-cell.\\\\ Let $S_m^{kl}$ denote the transformation: $U_m^k\rightarrow U_m^l$\\ \[ S_m^{kl}\left( \begin{array}{ccc} \gamma_m^i & \gamma_m^j & \gamma_m^l \end{array} \right) = \left( \begin{array}{ccc} \gamma_m^i & \gamma_m^j & \gamma_m^k \end{array} \right) \left( \begin{array}{ccc} 1 & 0 & -1\\ 0 & 1 & -1\\ 0 & 0 & -1 \end{array} \right) \]\\ and $S^k_{ml}$ denote the transformation: $U^k_m\rightarrow U^k_l$\\ \[ S_{ml}^k\left( \begin{array}{ccc} \gamma_m^i & \gamma_m^j & \gamma_m^l \end{array} \right) = \left( \begin{array}{ccc} \gamma_l^i & \gamma_l^j & \gamma_l^m \end{array} \right) \left( \begin{array}{ccc} 1 & 0 & 0\\ 0 & 1 & 0\\ -1& -1& -1 \end{array} \right) \]\\ For $(i,j,k,l,m)=(1,4,2,3,5)$\\ \[ T_{43}^2 \left( \begin{array}{ccc} \gamma_5^1 & \gamma_5^4 & \gamma_5^2 \end{array} \right)=\left( \begin{array}{ccc} \gamma_5^1 & \gamma_5^4 & \gamma_5^2 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right) \]\\ For $(i,j,k,l,m)=(1,3,4,2,5)$\\ \[ T_{32}^4 \left( \begin{array}{ccc} \gamma_5^1 & \gamma_5^3 & \gamma_5^4 \end{array} \right)=\left( \begin{array}{ccc} \gamma_5^1 & \gamma_5^3 & \gamma_5^4 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right) \]\\ Write under the same basis\\ \[ T_{43}^2 \left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)= T_{43}^2 \left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^4 \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & -1\\ 0 & 1 & -1\\ 0 & 0 & -1 \end{array} \right) \]\\ \[ =\left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^4 \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & -5\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & -1\\ 0 & 1 & -1\\ 0 & 0 & -1 \end{array} \right) \]\\ \[ =\left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & -1\\ 0 & 1 & -1\\ 0 & 0 & -1 \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & -5\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & -1\\ 0 & 1 & -1\\ 0 & 0 & -1 \end{array} \right) \]\\ which gives\\ \[ T_{43}^2 \left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)=\left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & 5\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right) \]\\ \[ T_{32}^4 \left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)= T_{32}^4 \left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^3 & \gamma_{5}^4 \end{array} \right)\left( \begin{array}{ccc} 1 & -1 & 0\\ 0 & -1 & 1\\ 0 & -1 & 0 \end{array} \right) \]\\ \[ =\left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^3 & \gamma_{5}^4 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right)\left( \begin{array}{ccc} 1 & -1 & 0\\ 0 & -1 & 1\\ 0 & -1 & 0 \end{array} \right) \]\\ \[ =\left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & -1\\ 0 & 0 & -1\\ 0 & 1 & -1 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right)\left( \begin{array}{ccc} 1 & -1 & 0\\ 0 & -1 & 1\\ 0 & -1 & 0 \end{array} \right) \]\\ which gives\\ \[ T_{32}^4 \left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)=\left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)\left( \begin{array}{ccc} 1 & 5 & -5\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right) \]\\ From these expressions, it is clear that $T_{32}^4$, $T_{24}^3$ and $T_{43}^2$ commute with each other and\\ \[ T_{32}^4 T_{24}^3 T_{43}^2= Id. \]\\ They determine a natural filtration:\\ \[ {\cal W}^0 \subset {\cal W}^1 = H_1(T^3, {\bf Z}) \]\\ with ${\cal W}^0$ generated by the vanishing circle $\gamma_{5}^1$, which is the common vanishing cycle of $T_{32}^4$, $T_{24}^3$ and $T_{43}^2$. Recall that the $T^3$ fibration of $X_{\infty}$ over $\partial \Delta$ degenerate to be a $T^2$ fibration over $\Delta _{15}$. An interesting fact is that this $T^2$ is exactly the quotient of $T^3$ by $\gamma_{5}^1$.\\ \subsection{Monodromy around baricenter of 1-cell} We can similarly compute the monodromies around a midpoint of a 1-edge, for instance, $P_{24}$. We have for $(i,j,k,l,m)=(5,2,1,4,3)$\\ \[ T_{24}^1 \left( \begin{array}{ccc} \gamma_{3}^5 & \gamma_{3}^2 & \gamma_{3}^1 \end{array} \right)=\left( \begin{array}{ccc} \gamma_{3}^5 & \gamma_{3}^2 & \gamma_{3}^1 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right) \]\\ and for $(i,j,k,l,m)=(3,2,5,4,1)$\\ \[ T_{24}^5 \left( \begin{array}{ccc} \gamma_{1}^3 & \gamma_{1}^2 & \gamma_{1}^5 \end{array} \right)=\left( \begin{array}{ccc} \gamma_{1}^3 & \gamma_{1}^2 & \gamma_{1}^5 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right) \]\\ write under the same basis $(\gamma_{5}^1, \gamma_{5}^2, \gamma_{5}^3)$.\\ \[ T_{24}^1 \left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)= T_{24}^1 \left( \begin{array}{ccc} \gamma_{3}^5 & \gamma_{3}^2 & \gamma_{3}^1 \end{array} \right)\left( \begin{array}{ccc} -1 & -1 & -1\\ 0 & 1 & 0\\ 1 & 0 & 0 \end{array} \right) \]\\ \[ =\left( \begin{array}{ccc} \gamma_{3}^5 & \gamma_{3}^2 & \gamma_{3}^1 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right)\left( \begin{array}{ccc} -1&-1 & -1\\ 0 & 1 & 0\\ 1 & 0 & 0 \end{array} \right) \]\\ \[ =\left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)\left( \begin{array}{ccc} 0 & 0 & 1\\ 0 & 1 & 0\\ -1& -1 &-1 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right)\left( \begin{array}{ccc} -1&-1 & -1\\ 0 & 1 & 0\\ 1 & 0 & 0 \end{array} \right) \]\\ which gives\\ \[ T_{24}^1 \left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)=\left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & 0\\ 0 & 1 & 0\\ 0 & 5 & 1 \end{array} \right) \]\\ \[ T_{24}^5 \left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)= T_{24}^5 \left( \begin{array}{ccc} \gamma_{1}^3 & \gamma_{14}^2 & \gamma_{1}^5 \end{array} \right)\left( \begin{array}{ccc} 0 & 0 & 1\\ 0 & 1 & 0\\ -1 & -1 & -1 \end{array} \right) \]\\ \[ =\left( \begin{array}{ccc} \gamma_{1}^3 & \gamma_{1}^2 & \gamma_{1}^5 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right)\left( \begin{array}{ccc} 0 & 0 & 1\\ 0 & 1 & 0\\ -1 & -1 & -1 \end{array} \right) \]\\ \[ =\left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)\left( \begin{array}{ccc} -1& -1 &-1\\ 0& 1 & 0\\ 1& 0 & 0 \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right)\left( \begin{array}{ccc} 0 & 0 & 1\\ 0 & 1 & 0\\ -1 & -1 & -1 \end{array} \right) \]\\ which gives\\ \[ T_{24}^5 \left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)=\left( \begin{array}{ccc} \gamma_{5}^1 & \gamma_{5}^2 & \gamma_{5}^3 \end{array} \right)\left( \begin{array}{ccc} 1 & 5 & 0\\ 0 & 1 & 0\\ 0 & -5 & 1 \end{array} \right) \]\\ From these expressions, it is clear that $T_{24}^1$, $T_{24}^3$ and $T_{24}^5$ commute with each other and\\ \[ T_{24}^1 T_{24}^3 T_{24}^5= Id. \]\\ They determine a natural filtration:\\ \[ {\cal W}^0 \subset {\cal W}^1 = H_1(T^3, {\bf Z}) \]\\ with ${\cal W}^0$ generated by the vanishing circle $\gamma_{5}^1$ and $\gamma_{5}^3$. Recall that the $T^3$ fibration of $X_{\infty}$ over $\partial \Delta$ degenerate to be a $T^1$ fibration over $\Delta _{135}$. An interesting fact is that this $T^1$ is exactly the quotient of $T^3$ by $\gamma_{5}^1\times \gamma_{5}^3$.\\\\ \se{Geometry of the singular fibers} With monodromy information in mind, we would like to discuss possible structure of singular special Lagrangian fibers. We will start by describing a possible model for the structure of {special Lagrangian } fibration, especially the singular fibers, that is of conjectural nature. Then we will use our construction of {Lagrangian } fibration to give some partial justification.\\\\ Suppose that we have a {special Lagrangian } fibration $F: X_{\psi} \rightarrow \partial \Delta$. Then the monodromy information will suggest that it is likely that $F$ is smooth fibration over $\partial \Delta\backslash \Gamma$, where the singular locus $\Gamma$ is topologically a graph in $\partial \Delta$. Vertices of $\Gamma$ are $P_{ij}$ (baricenter for 2-simplex $\Delta_{ij}$) for $\{i,j\}\subset \{1,2,3,4,5\}$ and $P_{ijk}$ (baricenter for 3-simplex $\Delta_{ijk}$) for $\{i,j,k\}\subset \{1,2,3,4,5\}$. Legs of $\Gamma$ are $\Gamma_{ij}^k$ which connects $P_{ijk}$ and $P_{ij}$ for $\{i,j,k\}\subset \{1,2,3,4,5\}$.\\\\ In general for $\{i,j,k,l,m\}=\{1,2,3,4,5\}$ representing the same orientation with the written order, the monodromy around $\Gamma_{ij}^k$ computed along the path\\ \[ U_m^j\rightarrow U_m^i\rightarrow U_l^i\rightarrow U_l^j\rightarrow U_m^j. \]\\ can be written as\\ \[ T_{ij}^k \left( \begin{array}{ccc} \gamma_{m}^l & \gamma_{m}^i & \gamma_{m}^k \end{array} \right)=\left( \begin{array}{ccc} \gamma_{m}^l & \gamma_{m}^i & \gamma_{m}^k \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right). \]\\ This suggests that singular fibers on $\Gamma_{ij}^k$ should be of $I_5 \times S^1$ type, where $I_5$ is a Kodaira type elliptic singular fiber as degeneration of $\gamma_{m}^l \times \gamma_{m}^i$ with vanishing cycle $\gamma_{m}^l$, and the $S^1$ corresponds to $\gamma_{m}^k$. This is a general type of singular fibers. Let's call it type $I_5$ singular fiber. Since Type $I_5$ singular fibers have $S^1$ factor, they give no contribution to Euler number.\\\\ On the other hand, around $P_{ijk}$, monodromies are\\ \[ T_{ij}^k \left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)=\left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right). \] \[ T_{jk}^i \left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)=\left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & 5\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right). \] \[ T_{ki}^j \left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)=\left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)\left( \begin{array}{ccc} 1 & 5 &-5\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right). \] They determine the natural filtration:\\ \[ {\cal W}^0 \subset {\cal W}^1 = H_1(T^3, {\bf Z}) \] with ${\cal W}^0$ generated by the common vanishing cycle $\gamma_m^l$. By the symmetric property, the singular fiber over $P_{ijk}$ can be described as follows. Let $\Gamma_{ijk}$ be the graph in $T^2\cong T^3/\gamma_{m}^l$ as discribed by the following picture, where $\{lm\}=\overline{\{ijk\}}$. Then the singular fiber over $P_{ijk}$ can be identified topologically as $T^3$ collapsing circles $(\gamma_{m}^l)$ over $\Gamma_{ijk}$ to points. We call this kind of fibers type $II_{5\times 5}$. It is easy to see that the Euler number of a type $II_{5\times 5}$ fiber is $-25$.\\ \begin{figure}[h] \begin{center} \leavevmode \hbox{% \epsfxsize=3in \epsffile{six2.eps}} \end{center} \caption{Type $II_{5\times 5}$ fiber} \end{figure}\\ Finally we are left to determine the singular fiber over $P_{ij}$. Around $P_{ij}$ the monodromies are\\ \[ T_{ij}^k \left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)=\left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)\left( \begin{array}{ccc} 1 & -5 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1 \end{array} \right). \] \[ T_{ij}^l \left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)=\left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)\left( \begin{array}{ccc} 1 & 0 & 0\\ 0 & 1 & 0\\ 0 & 5 & 1 \end{array} \right). \] \[ T_{ij}^l \left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)=\left( \begin{array}{ccc} \gamma_m^l & \gamma_m^i & \gamma_m^k \end{array} \right)\left( \begin{array}{ccc} 1 & 5 & 0\\ 0 & 1 & 0\\ 0 & -5 & 1 \end{array} \right). \] They determine the natural filtration:\\ \[ {\cal W}^0 \subset {\cal W}^1 = H_1(T^3, {\bf Z}) \] with ${\cal W}^0$ generated by the vanishing circles $\gamma_m^l$ and $\gamma_m^k$. The singular fiber over $P_{ij}$ is sort of generalization of $I_5$ singularity to $3$-dimension. Topologically equivalent to $T^3$ collapsing five of sub-$T^2$ equivalent to $\gamma_m^l\times \gamma_m^k$. It looks like five $3$-dimensional pseudomanifolds linked into a ``necklace'' via their singular points. Each $3$-dimensional pseudomanifold is a $T^2\times [0,1]$ collapsing the two boundaries to two singular points. An illustration of this singular fiber is in the following picture. We will call these type $III_5$ singular fibers. A type $III_5$ singular fiber has Euler number $5$.\\ \begin{figure}[h] \begin{center} \leavevmode \hbox{% \epsfxsize=4in \epsffile{p3.eps}} \end{center} \caption{Type $III_5$ fiber} \end{figure}\\ All together we have $10$ type $II_{5\times 5}$ and $10$ type $III_5$ singular fibers, which give total Euler number\\ \[ \chi(X)=10\times (-25)+10\times 5 = -200. \] This is exactly the Euler number of a smooth quintic {Calabi-Yau } threefold.\\\\ It is instructive to understand the singular point set of the fibration map $f$, namely the points in $X_{\psi}$ where tangent maps of $f$ are not surjective. The singular point set of a type $I_5$ fiber is five $S^1$'s. The singular point set of a type $II_{5\times 5}$ fiber is the graph in figure 2. The singular point set of a type $III_5$ fiber is five points. Together they form a Riemann surface. The Riemann surface has $10$ irreducable components $\{\Sigma_{ijk}\}$. $\Sigma_{ijk}$ is fibered over $\Gamma_{ij}^k\cup \Gamma_{ki}^j\cup \Gamma_{jk}^i$ and centered over $P_{ijk}$. Singular fibers are three of $5$-point sets and $\Gamma_{ijk}$. Total Euler number is \\ \[ \chi(\Sigma_{ijk})=3\times 5 - 25 = -10 = 2-2g(\Sigma_{ijk}). \] So $g(\Sigma_{ijk})= 6$.\\\\ Actually there is a very explicit description of the local structure of the map $f$ arround a singular point in a type $III_5$ fiber. It was given in the celebrated paper of R. Harvey and H. B. Lawson (\cite{HL}). The construction is as follows.\\ \begin{th} Let $M_c=f^{-1}(c)$, where $f:\ {\bf C^3}\longrightarrow {\bf R^3}$\\ \[ f_i(z)= |z_1|^2 - |z_i|^2,\ \ i=2, 3;\ \ \ f_1= {\rm Im}(z_1 z_2 z_3). \] Then $M_c$ (with the correct orientation) is a special Lagrangian submanifold of ${\bf C^3}$. \end{th} \begin{flushright} $\Box$ \end{flushright} It is pretty obvious why this characterize the singular points in type $III_5$ singular fiber. Since each $M_c$ is invariant under the action of the group\\ \[ T^2=\left( \begin{array}{ccc} e^{i\theta_1} & 0 & 0\\ 0 & e^{i\theta_2} & 0\\ 0 & 0 & e^{i\theta_3} \end{array} \right)\ \ \theta_1+\theta_2+\theta_3=0. \] An easy computation will show that the singular point set around origin is exactly the union of the three coordinate axes. This is where three of the irreducible components of the singular point set meet. The singular locus in ${\bf R^3}$ is\\ \[ \{c\in{\bf R^3}| c_2\leq 0, c_3=c_1=0\}\cup \{c\in{\bf R^3}| c_3\leq 0, c_2=c_1=0\} \] \[ \cup \{c\in{\bf R^3}| c_2+c_3\geq 0, c_2-c_3=c_1=0\}. \] Fiber over origin is of type $III_5$ singularity and fibers over other points in the singular locus are Type $I_5$. An important observation is that $\Sigma_{ijk}$ is exactly equivalent to the genus 6 curve\\ \[ \{[z]\in {\bf CP^4}| z_i^5 + z_j^5 + z_k^5 =0, z_l=0\ \ {\rm for}\ l\in\{1,2,3,4,5\}\backslash\{i,j,k\}\} \] Associate to the special lagrangian torus fiberation $f$ there is the Leray spectral sequence, which abuts to $H^\cdot(X_{\psi}, {\bf Q})$ and in which\\ \[ E_2^{p,q} = {\bf H}^p(\partial \Delta; R^qf_*{\bf Q}) \] This spectral sequence degenerates at $E_2$ term. We can use it to compute cohomology of $X_{\psi}$.\\\\ This spectral sequence was discussed and computed in \cite{GW} for expected SYZ {special Lagrangian } fiberation of generic {Calabi-Yau } manifolds. Our situation is far from generic. Fermat type quintics have a lot of symmetries and actually represent singular (orbifold) points in the moduli space of {Calabi-Yau }. It is interesting to compute the spectral sequence of the fiberation and compare with the generic situation.\\\\ Since $R^0f_*{\bf Q} ={\bf Q}$, $\partial \Delta \cong S^3$. It is easy to see that the bottom row of the spectral sequence is\\ \[ E_2^{3,q}\cong H^q(S^3,{\bf Q}):\ \ \ {\bf Q},\ 0,\ 0,\ {\bf Q}. \] In general, we have the following exact sequence\\ \[ 0\longrightarrow K_q \longrightarrow R^qf_*{\bf Q} \longrightarrow i_* E_q \longrightarrow 0 \] with the second non-trivial map being the attachment map with $E_q=i^* R^qf_*{\bf Q}$ for $i:\ \partial\Delta\backslash \Gamma \longrightarrow \partial\Delta$. Space $B=\partial\Delta$ has a natural filtration\\ \[ B=B_3\supset B_2\supset B_1\supset B_0\supset B_0 = \phi \] where $B_2=B_1=\Gamma$, $B_0=\{P_{ij}|i,j\in \{1,2,3,4,5\}\}\cup \{P_{ijk}|i,j,k\in \{1,2,3,4,5\}\}$. Each stratum $S_i=B_i\backslash B_{i-1}$ is a $i$-manifold. We denote $V_i=B\backslash B_i$. $E_q$ is a local system on $V_1=\partial\Delta\backslash \Gamma$ and $K_q$ is supported on $B_1=\Gamma$. Each $R^qf_*{\bf Q}$ is a locally constant sheaf. Its germ over $P\in B$ is\\ \[ (R^qf_*{\bf Q})_P\cong H^q(f^{-1}(P), {\bf Q}) \cong (H_q(f^{-1}(P), {\bf Q}))^\vee \] $E_3$ is obviously a constant sheaf, so $i_*E_3\cong {\bf Q}$. According to our geometric description of the singular fibers, $R^3f_*{\bf Q}$ is of dimension $1$ over $S_3$, dimension $5$ over $S_1$, dimension $5$ over $P_{ij}$ and dimension $25$ over $P_{ijk}$. Therefore $K_3$ is supported in $B_1=\Gamma$ and is of dimension $4$ over $S_1$, dimension $4$ over $P_{ij}$ and dimension $24$ over $P_{ijk}$. Clearly only possible non-trivial cohomologies for $K_3$ are $H^0(K_3,{\bf Q})$ and $H^1(K_3,{\bf Q})$.\\\\ \begin{prop} \[ H^1(K_3,{\bf Q})=0 \] \end{prop} {\bf Proof:} We will compute the {\v C}ech cohomology. There is a natural map\\ \[ \pi_{ijk}:\ (K_3)_{P_{ijk}} \longrightarrow (K_3)_{\Gamma^k_{ij}}\oplus (K_3)_{\Gamma^i_{jk}}\oplus (K_3)_{\Gamma^j_{ki}} \] We will show that $\pi_{ijk}$ is surjective, which easily implies that $H^1(K_3,{\bf Q})=0$. Observe that $\pi_{ijk}$ is induced from\\ \[ {\tilde \pi}_{ijk}: (R^3f_*{\bf Q})_{P_{ijk}} \longrightarrow (R^3f_*{\bf Q})_{\Gamma^k_{ij}}\oplus (R^3f_*{\bf Q})_{\Gamma^i_{jk}}\oplus (K_3)_{\Gamma^j_{ki}} \] Choose generators $\{x_{ij}|i,j\in \{1,2,3,4,5\}\}$ of $(R^3f_*{\bf Q})_{P_{ijk}}$, $\{u_i\}_{i=1}^5$ of $(R^3f_*{\bf Q})_{\Gamma^k_{ij}}$, $\{v_i\}_{i=1}^5$ of $(R^3f_*{\bf Q})_{\Gamma^j_{ki}}$ and $\{w_i\}_{i=1}^5$ of $(R^3f_*{\bf Q})_{\Gamma^k_{ij}}$ (representing geometric cycles). If we choose correctly, we should have\\ \[ {\tilde \pi}_{ijk}(x_{lm})=(u_l, v_m, w_n),\ \ n \equiv -l-m\ (mod\ 5). \] By simple linear algebra, we can see that\\ \[ \left( \sum^5_{i=1}a_l u_l,\ \sum^5_{i=1}b_l v_l,\ \sum^5_{i=1}c_l w_l\right) \in {\rm Im}({\tilde \pi}_{ijk}) \] if and only if \[ \sum^5_{i=1}a_l =\sum^5_{i=1}b_l =\sum^5_{i=1}c_l; \] and \[ \left( \sum^5_{i=1}a_l u_l,\ \sum^5_{i=1}b_l v_l,\ \sum^5_{i=1}c_l w_l\right) \in {\rm Im}(\pi_{ijk}) \] if and only if \[ \sum^5_{i=1}a_l =\sum^5_{i=1}b_l =\sum^5_{i=1}c_l =0. \] Therefore $\pi_{ijk}$ is surjective. \begin{flushright} $\Box$ \end{flushright} \begin{prop} \[ h^0(K_3)=160. \] \end{prop} {\bf Proof:} We will compute Euler number $e(K_3)=h^0(K_3)-h^1(K_3)$. We compute in the chain level. Let $c^i=\dim(C^i)$ be the dimension of the space of dimension $i$ cochain. It is easy to see\\ \[ c^0=10\times 24 + 10 \times4 =280,\ \ c^1=30\times4=120,\ \ e(K_3)=c^0-c^1=160. \] Then $h^1(K_3)=0$ from last proposition gives us that $h^0(K_3)=160.$ \begin{flushright} $\Box$ \end{flushright} Now use the long exact sequence associated to\\ \[ 0\longrightarrow K_3 \longrightarrow R^3f_*{\bf Q} \longrightarrow i_* E_3 \longrightarrow 0 \] we get \begin{prop} \[ h^0(R^3f_*{\bf Q})=161,\ h^1(R^3f_*{\bf Q})=h^2(R^3f_*{\bf Q})=0,\ h^3(R^3f_*{\bf Q})=1. \] \end{prop} \begin{flushright} $\Box$ \end{flushright} To compute the cohomology for $R^1f_*{\bf Q}$, notice that $R^1f_*{\bf Q}=i_*E_1$ for local system $E_1$. It is easy to see that the monodromy of the local system $E_1$ generate $SL_3({\bf Z})$. So $h^0(R^1f_*{\bf Q})=0$. From this we will have\\ \begin{prop} \[ h^3(R^1f_*{\bf Q})=h^3(R^2f_*{\bf Q})=0. \] \end{prop} {\bf Proof:} By Poincare duality for general fibers, $E_2=E_1^\vee$. So\\ \[ R^2f_*{\bf Q}\cong {\cal D}i_*E_1 \ \ \ {\rm in}\ V_2 \] where ${\cal D}E$ is the dual of $E$ in the sense of Verdier. Then by Verdier duality\\ \[ H^3(R^2f_*{\bf Q})\cong H^3({\cal D}i_*E_1)\cong H^0(i_*E_1)^\vee=0. \] The space $B$ have a natural symmetry $s$ which respects the filtration. It is a piecewise linear map satisfying\\ \begin{equation} \label{ca} s^{-1}=s:\ B \longrightarrow B;\ s(P_i)=P_{\overline{i}},\ s(P_{ij})=P_{\overline{ij}} \end{equation} where $\overline{S}$ indicate compliment of $S\in \{1,2,3,4,5\}$. It is easy to check by looking at the monodromy that $s^*E_1\cong E_2$. (Later we will see that this fact is related to the mirror symmetry construction of Fermat type mirror construction.) Therefore\\ \[ i_*E_1\cong s^*R^2f_*{\bf Q}\ \ \ {\rm in}\ V_2 \] and \[ H^3(R^1f_*{\bf Q})=H^3(i_*E_1)\cong H^3(R^2f_*{\bf Q})=0. \] \begin{flushright} $\Box$ \end{flushright} To compute $H^1(R^1f_*{\bf Q})$, we need to use intersection cohomology. Since $R^1f_*{\bf Q}=i_*E_1$, We have $H^1(R^1f_*{\bf Q})=I_0 H^1(B,E_1)$. Where $I_p H^i(B,E)$ denote the $i$-th intersection cohomology of $B$ with coefficient in the local system $E$ and perversity $p$.\\ \begin{prop} \label{bd} \[ h^1(R^1f_*{\bf Q})=1. \] \end{prop} {\bf Proof:} By Poincare duality and $E_2=E_1^\vee$, we have\\ \[ H^1(R^1f_*{\bf Q})=I_0 H^1(B,E_1)=I_t H^2(B,E_2)^\vee \] where $t$ denote the top perversity. Intersection homology is related to intersection cohomology as\\ \[ I_t H^2(B,E_2)=I_t H_1(B,E_2). \] So we only need to compute $I_t H_1(B,E_2)$, where for $P\in V_1$ we have $(E_2)_P\cong H_1(f^{-1}(P),{\bf Q})$. Consider the first barycentric subdivision of the standard triangulation of $B$. The 1-simpleces allowed to compute $I_t H_1(B,E_2)$ are the 1-simpleces $\Gamma^i_{jkl}$ connecting $P_i$ and $P_{ijkl}$.\\ \begin{figure}[h] \begin{center} \leavevmode \hbox{% \epsfxsize=5.5in \epsffile{p4a.eps}} \end{center} \caption{} \end{figure}\\ We can introduce the following 1-chain\\ \[ L=\gamma_j^i \cdot \Gamma^i_{\overline{ij}} \] By (\ref{bc}), it is easy to see that $\partial L=0$. Actually this $L$ generate $I_t H_1(B,E_2)$, namely\\ \[ h^1(R^1f_*{\bf Q}) = \dim (I_t H_1(B,E_2))=1 \] \begin{flushright} $\Box$ \end{flushright} \begin{prop} \label{be} \[ h^2(R^1f_*{\bf Q})=1. \] \end{prop} {\bf Proof:} We will show this by proving that $\chi (R^1f_*{\bf Q})=0$, which clearly implies this proposition by results in proposition \ref{bd} and \ref{be}. The Euler number can be computed in two ways, either by straightforward {\v C}ech cohomology computation or by the identification\\ \[ H^i(R^1f_*{\bf Q}) \cong I_0 H^i(B,E_1)\cong I_0 H_{3-i}(B,E_1). \] We will use the latter approach, which is more elegant. To get the number right, the key point is to choose the right triangulation, which respects the filtration of $B$. The previous trangulation as indicated in figure 4 is not good enough. The problem is that the 0-simplices $P_{ij},P_{ijk}\in B_0$ will span a 1-simplex $\Gamma^k_{ij}$ which is in $B_1$ but not in $B_0$. The right triangulation which respects the filtration of $B$ can be achieved by taking the baricentric subdivision of $\Gamma$ and then naturally extend to a subdivision of the previous triangulation of $B$. Practically, each old 3-simplex is divided into two new 3-simleces by a new 2-simplex as indicated in the following picture.\\ \begin{figure}[h] \begin{center} \leavevmode \hbox{% \epsfxsize=5.5in \epsffile{p5b.eps}} \end{center} \caption{} \end{figure}\\ It is easy to see that\\ \[ I^0C_0=\bigoplus_{i=1}^5(E_1)_{P_{i}}\cdot P_{i} \oplus (E_1)_{P_{\overline{i}}}\cdot P_{\overline{i}} \] \[ I^0C_1=\bigoplus_{\tiny{\begin{array}{c} i,j=1\\ (i\not= j)\end{array}}}^5(E_1)_{<P_{i},P_{\overline{j}}>}\cdot <P_{i},P_{\overline{j}}>, \] and $c_0=10\times 3=30$, $c_1=20\times 3=60$.\\\\ For the $I^0C_2$, it is easy to see that only the newly added 2-simplices $<P_{i}, P_{\{ij\}\{ijk\}}, P_{ijkl}>$ will be relavent. Since the boundary of the intersection 2-chain should not intersect $\Gamma$. An intersection 2-chain containing $<P_{i}, P_{\{ij\}\{ijk\}}, P_{ijkl}>$ should also contain all other newly added 2-simplices that contain $P_{\{ij\}\{ijk\}}$, namely should contain\\ \[ \Delta^k_{ij}=\bigcup\{{\rm 2-simplex\ with}\ P_{\{ij\}\{ijk\}}\ {\rm as\ one\ vertex}\}. \] Therefore\\ \[ I^0C_2=\bigoplus_{\Gamma_{ij}^k}(E_1)^{T^k_{ij}}_{\Delta^k_{ij}}\cdot \Delta^k_{ij} \] where the upper index ${T^k_{ij}}$ indicates the ${T^k_{ij}}$ invariant piece of $E_1$. So $c_2=30\times2=60$.\\\\ Since an intersection 2-chain is not supposed to contain $P_{ij}$ or $P_{ijk}$. The boundary of an intersection 3-chain also should not contain $P_{ij}$ or $P_{ijk}$. Hence an intersection 3-chain containing $<P_{i}, P_{ij}, P_{\{ij\}\{ijk\}}, P_{ijkl}>$ has to contain\\ \[ \Delta_{ij}=\bigcup\{{\rm 3-simplex\ with}\ P_{ij}\ {\rm as\ one\ vertex}\}. \] An intersection 3-chain containing $<P_{i}, P_{ijk}, P_{\{ij\}\{ijk\}}, P_{ijkl}>$ has to contain\\ \[ \Delta_{ij}=\bigcup\{{\rm 3-simplex\ with}\ P_{ijk}\ {\rm as\ one\ vertex}\}. \] And\\ \[ I^0C_3=\left(\bigoplus_{ij} (E_1)^{G_{ij}}_{\Delta_{ij}}\cdot \Delta_{ij}\right) \bigoplus \left(\bigoplus_{ijk} (E_1)^{G_{ijk}}_{\Delta_{ijk}}\cdot \Delta_{ijk}\right), \] where $G_{ij}$ is the monodromy group around $P_{ij}$ and $G_{ijk}$ is the monodromy group around $P_{ijk}$. And $c_3=10\times 1+10\times2=30$. Therefore\\ \[ \chi(R^1f_*{\bf Q})=-\chi(I^0C.)=-\sum_{i=0}^3 (-1)^i c_i = -(30-60+60-30)=0. \] \begin{flushright} $\Box$ \end{flushright} Now the only thing we do not yet know is $h^1(R^2f_*{\bf Q})$.\\ \begin{prop} \[ h^1(R^2f_*{\bf Q})=41. \] \end{prop} {\bf Proof:} Clearly we only need to compute $h^1(K_2)$. Since the germ of $K_2$ at $P_{ij}$ only have zero section. It is clear that $h^0(K_2)=0$. It is now reduced to compute $\chi(K_2)$. We will simply count the {\v C}ech cochains.\\ \[ c^1=(5-1)\times 30=120,\ \ c^0=(10-2)\times 10 = 80. \] \[ \chi(K_2)=c^0-c^1 = 80-120=-40. \] So $h^1(K_2)=40$ and $h^1(R^2f_*{\bf Q})=41$. \begin{flushright} $\Box$ \end{flushright} Now we have the whole $E_2$ term of the spectral sequence.\\ \begin{th} $E_2$ term of the Leray spectral sequence of fibration $F$ is\\ \[ \begin{array}{cccc} 161 & 0 & 0 & 1\\ 0 & 41 & 1 & 0\\ 0 & 1 & 1 &0\\ 1 & 0 & 0 & 1 \end{array} \] \end{th} \begin{flushright} $\Box$ \end{flushright} One can see that the $E_2$ term of our situation is quite different from the $E_2$ term of generic {Calabi-Yau } situation as indicated in \cite{GW}.\\\\ Note: Our discussion so far is of conjectural nature. Since we can not construct {special Lagrangian } fibration, we can only guess its possible structure based on our monodromy discussion.\\ \se{Comparison} The {Lagrangian } fibration we constructed in Section 3 is different from the expected {special Lagrangian } fibration structure described in the previous two sections. Yet they are very closely related. The singular locus $\Gamma$ of the expected {special Lagrangian } fibration is of codimension 2 (an one dimensional graph) and each singular fiber has singularity of codimension 2. The singular locus $\tilde{\Gamma}$ of the {Lagrangian } fibration we constructed is of codimension 1 and most singular fibers have singularity of codimension 3. On the other hand, $\tilde{\Gamma}$ is just a fattened version of $\Gamma$. $S^3\backslash\tilde{\Gamma}$ and $S^3\backslash\Gamma$ have the identical fundamental group. We can compare monodromies of the two fibrations.\\\\ Since our {Lagrangian } fibration on $X_{\psi}$ is a deformation of the standard {special Lagrangian } fibration on $X_{\infty}$, it satisfies the principle assumption (1) that we based to compute monodromy. For this reason, it is not hard to see that above computation of monodromy naturally apply to our {Lagrangian } fibration. Therefore\\ \begin{th} Our {Lagrangian } fibration have the same monodromy as the monodromy computed for the expected {special Lagrangian } fibration. \end{th} \begin{flushright} $\Box$ \end{flushright} The singularities of the two fibrations are also closely related. For one thing, the singular point set of our construction in $X_{\psi}$ is $\Sigma$ (the union of 10 genus 6 curves), and the singular point set of the expected {special Lagrangian } fiberation is naturally equivalent to $\Sigma$. One way to see this is to notice that we can easily construct a homotopy contraction $p: \tilde{\Gamma} \rightarrow \Gamma$ that is the homotopy inverse of the natural injection $i: \Gamma \rightarrow \tilde{\Gamma}$ that satisfy $p\cdot i= id$ (for instance map point in $\tilde{\Gamma}$ to the nearest point in $\Gamma$). Let $q$ denote the composition of projection of $\Sigma$ to $\tilde{\Gamma}$ and $p$. Then $q$ map $\Sigma$ to $\Gamma$. It is easy to observe that the inverse image of a point $P$ in $\Gamma$ with respect to $q$ is exactly corresponding to singular set of the expected singular {special Lagrangian } fiber over $P$ as described in the previous section. With this fact in mind, we intend to modify our {Lagrangian } torus fibration into the expected topological shape.\\\\ If one is willing to sacrifies {Lagrangian } property, it is not hard to deform the fibration we constructed into a non-{Lagrangian } smooth fiberation with the same topological shape as we expected of {special Lagrangian } fibration. One way to do this is to notice that there is a natural horizontal foliation in ${\bf P^2}$ where one of the component of $\Sigma$ is located. Then conceivably, one can smoothly deform part of 2-torus above $\tilde{\Gamma}$ that intersect $\Sigma$ along horizontal direction away from $\Gamma$ in direction indicated by map $p$. So that eventually only those 2-torus above $\Gamma$ will intersect $\Sigma$ at one dimensional graphs that is the inverse image under $q$ of the corresponding point. Then we run the flow of $V$, we will get a non-{Lagrangian } torus fibration with topological structure that is the same as expected {special Lagrangian } fibration. This will give an alternative simple proof of the result in \cite{Z} that gave a smooth torus non-{Lagrangian } fibration. Since our major concern is to construct smooth {Lagrangian } torus fiberation, we will not get into great details in this direction.\\\\ However to get a {Lagrangian } torus fibration of the right shape is much trickier. We will address this problem in the sequel to this paper(\cite{lag2}).\\\\ Another important observation is that for an $n$-torus {Lagrangian } fibration such that the fibration map is $C^{\infty}$ (actually $C^2$ is enough), there is a natural action of ${\bf R^n}$ on each fiber (smooth or singular). Look at our construction, one can see that there are no action of ${\bf R^3}$ on most of our singular fibers, which seems to be a contradiction. It turns out that fibration map of our construction is not $C^{\infty}$---it is typically only piecewise smooth and globally merely $C^{0,1}$ (Lipschitz). This reveals a major difference between a non-{Lagrangian } fibration and a {Lagrangian } fibration. If one does not care keeping the {Lagrangian } property, the map usually can be smoothed by a small perturbation. Above ovservation shows that in general a {Lagrangian } fiberation can not be deformed to a smooth {Lagrangian } fiberation by small perturbation. In (\cite{lag2}) we will explore this issue further and try to get a smooth {Lagrangian } fibration. We will also discuss general construction of {Lagrangian } torus fibration for general quintic {Calabi-Yau } hypersurfaces in \cite{lag3} and {Calabi-Yau } hypersurfaces in more general toric varieties in \cite{tor}.\\\\ \se{The Mirror construction and dual polyhedra} It is well known that the mirror familly $\{Y_\psi\}$ of the family $\{X_\psi\}$ can be given by $Y_\psi=X_\psi/G$ with the induced complex structure and the {Calabi-Yau } orbifold metric. $G=<g_0, g_1, g_2, g_3>$ is the symmetry group of $X_\psi$. $G$ acts on ${\bf CP^4}$ as follows\\ \begin{eqnarray*} g_0&=&(\xi, 1, 1, 1, \xi^{-1}),\\ g_1&=&(1, \xi, 1, 1, \xi^{-1}),\\ g_2&=&(1, 1, \xi, 1, \xi^{-1}),\\ g_3&=&(1, 1, 1, \xi, \xi^{-1}), \end{eqnarray*} where $\xi$ is a fifth root of unity. By analyzing the action of $G$ on $X_\psi$, it should be clear that $G\cong {\bf Z}_5^3$ will map every special Lagrangian fiber to itself, and is conjugate to the standard action of ${\bf Z}_5^3$ on $T^3$. The special Lagrangian fibration $f:\ X_\psi\rightarrow B$ will naturally induce special Lagrangian fibration $\hat{f}:\ Y_\psi\rightarrow B$. For any $p\in B$, $\hat{f}^{-1}(p)=f^{-1}(p)/G$, especially $\hat{f}^{-1}$ can be identified with $f^{-1}(p)$ quite canonically.\\\\ SYZ construction predict that the $T^3$ fibrations $f$ and $\hat{f}$ when restrict to $V_1\subset B$ should be dual to each other. But the above discussion implies that the two fibrations can be natually identified to each other instead of dual, especially the monodromy for $\hat{f}$ is the same as the monodromy for $f$. The key point is that the mirror map induce nontrivial identification $s:\ B\rightarrow B$ of base $B$ as defined in (\ref{ca}). Fibrations $\hat{f}$ and $B\circ f$ are actually dual to each other when restricted to $V_1\subset B$. This corresponds to the fact that the monodromy around $P_{ij}$ is dual to the monodromy around $P_{\overline{ij}}$. The reason why this is the correct interpretation is that the two fibrations should actually be written as\\ \[ f:\ X_\psi\longrightarrow \Delta,\ \ \ \hat{f}:\ Y_\psi\longrightarrow \hat{\Delta} \] where $\hat{\Delta}$ is the dual polyhedron of $\Delta$. The only canonical map from $\Delta$ to $\hat{\Delta}$ is the one corresponding to map $s$.\\\\ Now let us analyze the singular fibers of $\hat{f}$. For the type $I_5$ singular fiber $I_5\times S^1$, one ${\bf Z}_5$ will rotate $S^1$, another ${\bf Z}_5$ will rotate each $S^2$ in $I_5$ while keeping the nodal points fixed, the ${\bf Z}_5$ left will permute the five $S^2$'s in $I_5$. The quotient will natually be $I_1\times S^1$, we call it type $I$. A type $I$ fiber has Euler number $0$.\\\\ For the type $II_{5\times 5}$ singular fiber over $P_{ijk}$, two ${\bf Z}_5$'s will act on $T^2\cong T^3/{\gamma^m_l}$ as indicated by the two arrows in the following picture, which move a sixgon along the arrows. The ${\bf Z}_5$ left will act on $\gamma^m_l$ in standard way, which do not affect topology. Upon the action, edges of a fundamental reigon in the $T^2$ as indicated as darker area in the picture will be identified as indicated in the picture and turn into a $T^2$, with image of the graph $\Gamma_{ijk}$ turn into $\hat{\Gamma}_{ijk}$, topologically, a contractor of ``a pair of pants''. The resulting singular fiber can be identified as $S^1\times T^2$ with $S^1$'s over $\hat{\Gamma}_{ijk}$ collapsed to points. We will call it type $II$ singular fiber. A type $II$ fiber has Euler number $1$.\\ \begin{figure}[h] \begin{center} \leavevmode \hbox{% \epsfxsize=5in \epsffile{six3c.eps}} \end{center} \caption{Type $II$ fiber} \end{figure}\\ The quotients of the type $III_5$ singular fibers are very easy to describe. Two ${\bf Z}_5$'s will act on $T^2$ in standard way, topologically cause no change. The ${\bf Z}_5$ left will rotate among the chain of the five suspensions of $T^2$. The resulting singular fiber will simply be a suspension of $T^2$ with the two pole points identified as indicated in the following picture. We will call it type $III$ fiber. A type $III$ fiber has Euler number $-1$.\\ \begin{figure}[h] \begin{center} \leavevmode \hbox{% \epsfxsize=5in \epsffile{p6.eps}} \end{center} \caption{Type $III$ fiber} \end{figure}\\ It is interesting to see what is the image of the singular point sets\\ \[ \Sigma_{ijk}=\{[z]\in {\bf CP^4}| z_i^5 + z_j^5 + z_k^5 =0, z_l=0\ \ {\rm for}\ l\in\{1,2,3,4,5\}\backslash\{i,j,k\}\} \] under the quotient map. It is easy to see that $G$ map $\Sigma_{ijk}$ to itself. Two of the ${\bf Z}_5$'s act as identity, and the ${\bf Z}_5$ left acts freely on $\Sigma_{ijk}$. So $\Sigma_{ijk}$ is a 5-sheet cover of the quotient $\hat{\Sigma}_{ijk}$. Since\\ \[ 5\cdot \chi(\hat{\Sigma}_{ijk})=\chi(\Sigma_{ijk})=-10 \] we have\\ \[ \chi(\hat{\Sigma}_{ijk})=-2,\ \ g(\hat{\Sigma}_{ijk})=0. \] Namely $\hat{\Sigma}_{ijk}$ is a Riemann sphere. Their union is exactly the singular point set of the quotient $Y_{\psi}$.\\\\ Now we can try to compute the Leray sqectral sequence of the fibration $\hat{f}$. By analyzing the singular fibers described at above, we can esaily see that\\ \[ R^0 \hat{f}_*{\bf Q}\cong R^3 \hat{f}_*{\bf Q}\cong {\bf Q} \] and \[ R^1 \hat{f}_*{\bf Q}\cong (R^2 \hat{f}_*{\bf Q})^\vee \cong i_* E_1 \] These combine with computation from last section, give us\\ \begin{prop} $E_2$ term of the Leray spectral sequence of fibration $\hat{f}$ is\\ \[ \begin{array}{cccc} 1 & 0 & 0 & 1\\ 0 & 1 & 1 & 0\\ 0 & 1 & 1 & 0\\ 1 & 0 & 0 & 1 \end{array} \] \end{prop} \begin{flushright} $\Box$ \end{flushright} What follows is a discussion of the well known crepant resolution of $Y_{\psi}$. For completeness of mirror picture, we include it here. We need some degression on toric varieties. Let $X=Spec({\bf C}[C])$ be an affine toric variety, where \[ C=<v^1, v^2, \cdots, v^n>_{\bf Q^+}\cap M \subset M=<e^1, e^2, \cdots, e^n>_{\bf Z} \] is a cone in lattice $M$. Following convention, let $N$ be the dual lattice of $M$ and ${\check C}$ be the dual cone of $C$. Suppose \[ {\check C}=<v_1, v_2, \cdots, v_n>_{\bf Q^+}\cap N \subset N=<e_1, e_2, \cdots, e_n>_{\bf Z}. \] It is well known that a birational modification $X'$ of $X$ corresponds to subdividing ${\check C}$ into union of smaller cones (we require each new cone to be generated by $n$ elements), one dimensional edges of these cones correspond to toric divisors in $X'$. Let $D_v$ denote the toric divisor corresponding to the 1-dimensional cone generated by $v$. We will use $X_0$ to denote the smooth part of $X$. It is obvious that the singular part of $X$ is of codimension greater or equal to two. We need the following result.\\ \begin{prop} There is a canonical nowhere vanishing holomorphic n-form $\Omega$ unique up to constant on $X_0$. Assume $X'$ is a birational modification of $X$, then $\Omega$ can be extend to $X_0$ as a nowhere vanishing holomorphic n-form if and only if the generater of any of the 1-dimension subcone coming from the subdivision is a convex combination of $ v_1, v_2, \cdots, v_n$. \end{prop} {\bf Proof:} It is well know that there is a canonical holomorphic n-form $\Omega_0$ on the $n$-torus defined as \[ \Omega_0=\frac{{\rm d}e^1}{e^1}\wedge\frac{{\rm d}e^2}{e^2}\wedge\cdots \wedge\frac{{\rm d}e^n}{e^n} \] For any primary $v\in N$, we can get $D_v=Spec({\bf C}[v^\bot])$. $\Omega_0$ will have pole of order 1 around $D_v$. To cancel the pole, we need to multiply a function, which vanish to order one at $D_v$. These kind of function will be in\\ \[ v^\vee =\{w\in M|<v, w>=1\} \] Suppose that\\ \[ \bigcap _{i=1}^n v_i^\vee =\{w\} \] then $\Omega=w\cdot\Omega_0$ is exactly what we need.\\\\ Suppose that $v$ is the generater of an 1-dimension subcone coming from the subdivision corresponding to a birational modification $X'$. Then in order to extend $\Omega=w\cdot\Omega_0$ to $D_v\subset X'$, $v$ must satisfy $<v, w>=1$, which is the same as that $v$ is a convex combination of $ v_1, v_2, \cdots, v_n$. \begin{flushright} $\Box$ \end{flushright} {\bf Remark:} The above proposition must be quite well known in toric geometry. But we do not know exactly where it was located.\\\\ $Y_{\psi}$ is a singular {Calabi-Yau } manifold. It has singularity at $\tilde{P}_{ij}$ and $\hat{\Sigma}_{ijk}$, where $\tilde{P}_{ij}$ is the singular point of the fiber over $P_{ij}$. Around $\tilde{P}_{ij}$, $Y_{\psi}$ has quotient singularity. It can be modeled by affine toric variety $X=Spec({\bf C}[C])$, where\\ \[ C=<5e^1, 5e^2, 5e^3>_{\bf Q^+}\cap M \subset M=<5e^1, 5e^2, 5e^3, \sum_{i=1}^3 e^i>_{\bf Z} \] Here $\{e^i\}$ denote the standard base of ${\bf Z}^3$. We would like to get a crepent resolution of the toric variety $X=Spec({\bf C}[C])$, for which we have to subdivide the dual cone\\ \[ {\check C}=<e_1, e_2, e_3>_{\bf Q^+}\cap N \subset N=M^\vee, \] here $\{e_i\}$ denote the standard base of the dual ${\bf Z}^3$. By above discussion, we only need to look at the plane passing through the three points $e_1, e_2, e_3$. ${\check C}$ will cut off a triangle with this three points as vertices in the plane. All the lattice points in $N$ that is in this triangle are\\ \[ v_{ijk}=\frac{i}{5}\cdot e_1+\frac{j}{5}\cdot e_2 +\frac{k}{5}\cdot e_3,\ \ i,j,k\geq 0\ \ i+j+k=5 \] We will divide this triangle as indicated in following picture. This division will result in a subdivision of ${\check C}$ which gives us a desired crepent resolution.\\ \begin{figure}[h] \begin{center} \leavevmode \hbox{% \epsfxsize=5in \epsffile{tri.eps}} \end{center} \caption{} \end{figure}\\ There are exactly three curves of singularity $\hat{\Sigma}_{ijk}$ for $k\in \overline{\{ij\}}$ that pass through the singular point $\tilde{P}_{ij}$. In the picture, they correspond to the interiors of the three edges of the first triangle, while the singular point correspond to the interior of the triangle. After the birational modification, we will have newly added exceptional divisors that correspond to dots other than $e_1,e_2,e_3$. The ones in the interiors of the three edges correspond to exceptional divisors coming from the three curves of singularity. The ones in the interior of the first triangle correspond to exceptional divisors coming from the singular point. So each curve of singularity will give rise to 4 exceptional divisors and each singular point will give rise to 6 exceptional divisors. There are 10 curves of singularity and 10 singular points in $Y_\psi$. So all together we will have $4\times 10 + 6\times 10= 100$ of exceptional divisors in the smooth {Calabi-Yau } $\tilde{Y}_\psi$ we get as the crepant resolution of $Y_\psi$. This verify the fact\\ \begin{prop} \[ h^0(\tilde{Y}_\psi)=h^6(\tilde{Y}_\psi)=1, \ h^1(\tilde{Y}_\psi)=h^5(\tilde{Y}_\psi)=0,\ h^2(\tilde{Y}_\psi)=h^4(\tilde{Y}_\psi)=101,\ h^3(\tilde{Y}_\psi)=4. \] \end{prop} \begin{flushright} $\Box$ \end{flushright} {\bf Acknowledgement:} I would like to thank Qin Jing for many very stimulating discussions during the course of my work, and helpful suggestions while carefully reading my early draft. I would also like to thank Prof. Yau for his constant encouragement.\\\\
\section*{Tables} \begin{table}[hbt] \begin{center} \begin{tabular}{|c|c|c|c|c|c|c|c|} \hline $f$ & $N$ & $\delta t/\tau$ & $t_{max}/\delta t$ & $R_{G}/\sigma_{LJ}$ & $(R_{c}/\sigma_{LJ}) + 1/2$ & $R_{d}/\sigma_{LJ}$ & $\lambda$\\ \hline 5 & 100 & 0.004 & 500000 & 13.53 & 0.65 & 1.39 & 0.61\\ 10 & 50 & 0.003 & 400000 & 10.37 & 1.1 & 1.21 & 0.66\\ 10 & 100 & 0.003 & 400000 & 16.18 & 1.1 & 0.89 & 0.64\\ 10 & 150 & 0.003 & 400000 & 19.71 & 1.1 & 1.31 & 0.60\\ 10 & 200 & 0.003 & 400000 & 24.52 & 1.1 & 1.42 & 0.67\\ 18 & 50 & 0.002 & 350000 & 11.19 & 1.25 & 1.38 & 0.68\\ 18 & 100 & 0.002 & 350000 & 17.10 & 1.25 & 1.64 & 0.65\\ 30 & 50 & 0.002 & 350000 & 12.22 & 1.6 & 1.89 & 0.66\\ 50 & 50 & 0.002 & 350000 & 13.35 & 1.8 & 2.40 & 0.69\\ \hline \end{tabular} \end{center} \caption{ \label{parameters.table} List of the simulation parameters and the corresponding results for $R_{G}$ and $\lambda=\sigma/2R_{G}$. } \end{table} \section*{Figure captions} \begin{figure}[hbt] \caption{ \label{snapshot.plot} Typical configuration for two stars with $f=10$ and $N=50$. The distance between the central core particles, which are shown as big black spheres, is $r=5.2\sigma_{LJ}$. The gray and light gray monomers are belonging to the first and second star respectively. } \end{figure} \begin{figure}[hbt] \caption{ \label{invforce.plot} Reduced inverse force $k_{B}T/(FR_{G})$ between the centers of two star polymers (for $f=10$ and $N=50$) versus reduced distance $r/R_{G}$. The error bars were obtained by averaging over the results of $8$ independent simulations. } \end{figure} \begin{figure}[hbt] \caption{ \label{results.plot1} Simulation results (symbols) and theoretical results (lines) for the reduced effective force $FR_{G}/k_{B}T$ versus reduced distance $(r-2R_{d})/R_{G}$. a) for $f=5, 10, 18$ and $N=100$, b) for $f=18, 30, 50$ and $N=50$. } \end{figure} \begin{figure}[hbt] \caption{ \label{lnforce.plot} Logarithm of the reduced force $\ln\left(FR_{G}/k_{B}T\right)$ versus reduced distance $(r-2R_{d})/R_{G}$ for $f=10$ and $N=50$. The error bars were obtained by averaging over the results of $8$ independent simulations. } \end{figure} \vfill \end{document}
\section{Introduction} A few years ago we have proposed a quark-parton model of cumulative phenomena in the interactions with nuclei \cite{NPB94,YF97} based on perturbative QCD calculations of the corresponding quark diagrams near the thresholds, at which other quarks ("donors") in the nuclear flucton transfer all their longitudinal momenta to the distinguished active quark and become soft. Consider the scattering of a hadronic projectile off a nucleus with the momentum $P$ in the c.m. system. At high energies the momentum $K$ of the produced pion belongs to the cumulative region if $K_z>P_z/A$. As a reasonable first approximation, we treat the nucleus as a collection of $N=3A$ valence quarks, which, on the average, carry each longitudinal momentum $x_{0}P_{z}/A$ with $x_{0}=1/3$. In our approach the cumulative pion production proceeds in two steps. First a valence quark with a scaling variable $x>1$ is created. Afterwards it decays into the observed hadron with its scaling variable $x$ close to the initial cumulative quark's one. This second step is described by the well-known quark fragmentation functions \cite{CapellaT81} and will not be discussed here. The produced cumulative ("active") quark acquires the momentum much greater than $x_{0}P_{z}/A$ only if this quark has interacted by means of gluon exchanges with other $p$ quarks of flucton ("donors") and has taken some of their longitudinal momenta (see Fig.1). If this active quark accumulates all longitudinal momentum of these $p$ quarks then $K_z=(p+1)x_0P_z/A$ and the donors become soft. It is well-known that interactions which make the longitudinal momentum of one of the quark equal to zero may be treated by perturbation theory \cite{Brodsky92}. This allows to calculate the part of Fig. 1 responsible for the creation of a cumulative quark explicitly. This was done in \cite{NPB94,YF97}, to which papers we refer the reader for all the details. As a result we were able to explain the exponential fall-off of the production rate in the cumulative region. Since with the rise of $x$ the active quark has to interact with a greater number of donors, one expects that its average transverse momentum also grows with $x$. Roughly one expects that $\langle K_{\perp}^2 \rangle$ is proportional to the number of interactions, that is, to $x$. In \cite{NPB94,YF97} this point was not studied: we have limited ourselves with the inclusive cross-section integrated over the transverse momenta, which lead to some simplifications. The aim of the present paper is to find the pion production rate dependence on the transverse momentum and the mean value of the latter as a function of $x$ in the cumulative region. This dependence and also the magnitude of $\langle K_{\perp}^2 \rangle$ have been studied experimentally. The comparison of our predictions with the data allows to obtain further support for our model and fix one of the two its parameters (the infrared cutoff). \section{The $K_{\perp}$ dependence} Repeating the calculations of the diagram in Fig.1 described in \cite{NPB94,YF97} but not limiting ourselves with the inclusive cross-section integrated over the transverse momentum, we readily find that all dependence upon the transverse momentum $K_{\perp}$ of the produced particle is concentrated in a factor: \begin{equation} J (K_{\perp})= \int \rho_A(\underbrace{r,...,r}_{p+1}|\underbrace{\ol{r},...,\ol{r}}_{p+1}) G(c_1,...,c_p) \prod_{i=1}^{p} \lambda(c_i-r)\lambda(c_i-\ol{r})d^2 c_i e^{i(\ol{r}-r)K_{\perp}} d^2 r d^2 \ol{r} \end{equation} Here $\rho_A$ is the (translationally invariant) quark density matrix of the nucleus: \begin{equation} \rho_A(r_i|\ol{r}_i)\equiv \int \psi_{\perp A}(r_i,r_m) \psi^*_{\perp A}(\ol{r}_i,r_m) \prod_{m=p+2}^{N} d^2 r_m \end{equation} where $\psi_{\perp A}$ is the transverse part of the nuclear quark wave function. The propgation of soft donor quarks is decribed by \begin{equation} \lambda(c)=\frac{K_0(m|c|)}{2\pi} \label{lam} \end{equation} where $m$ is the constituent quark mass and $K_0$ is the modified Bessel function (the Mac-Donald function). The interaction with the projectile contributes a factor \begin{equation} G(c_1,...,c_p)= \int \prod_{i=1}^{p} \sigma_{qq}(c_i-b_i) \eta_H(b_1,...,b_p) d^2 b_i \end{equation} where $\sigma_{qq}(c)$ is the quark-quark cross-section at a given value of impact parameter $c$ and \begin{equation} \eta_H(b_1,...,b_p)= \sum_{L\geq p}\frac{L!}{(L-p)!} \int|\psi_{\perp H}(b_i)|^2 \delta^{(2)}(\frac{1}{L}\sum_{i=1}^{L} b_i) \ d^2 b_{p+1}...d^2 b_L \end{equation} is a multiparton distribution in the projectile, expressed via the transverse part of its partonic wave function $\psi_{\perp H}$ . If we integrate $J (K_{\perp})$ over $K_{\perp}$ we come back to our old result (Eq. (33) in \cite{NPB94}): $$ \int J (K_{\perp}) \frac{d^2 K_{\perp}}{(2\pi)^2}= \rho_A(\underbrace{0,...,0}_{p+1}|\underbrace{0,...,0}_{p+1}) \int G(c_1,...,c_p) \prod_{i=1}^{p} \lambda^2(c_i-r) d^2 c_i d^2 r $$ If one assumes factorization of the multiparton distribution $\eta_H(b_1,...,b_p)$ then\\ $G(c_1,...,c_p)$ also factorizes: \begin{equation} G(c_1,...,c_p)= \prod_{i=1}^{p} G_0(c_i) \label{facG} \end{equation} Following \cite{YF97} we use the quasi-eikonal approximation for $\eta_H$: $$ \eta_H(b_1,...,b_p)= \xi^{(p-1)/2}\nu_H^{p}\prod_{i=1}^{p}\eta_H(b_i) $$ where $\xi$ is the quasi-eikonal diffraction factor, $\nu_H^{}$ is the mean number of partons in the projectile hadron and the single parton distribution $\eta_H(b)$ is normalized to unity. In a Gaussian approximation for $\sigma(c)$ and $\eta_H(b)$ we find: $$ G_0(c)= \xi^{\frac{1}{2}-\frac{1}{2p}} \frac{\nu_H\sigma_{qq}}{\pi r_{0H}^2} e^{-\frac{c^2}{r_{0H}^2}} $$ where $\sigma_{qq}$ is the total quark-quark cross-section, $r_{0H}^2=r_0^2+r_H^2$, $r_0$ and $r_H$ are the widths of $\sigma(c)$ and $\eta_H(b)$ respectively. With the factorised $G(c_1,...,c_p)$ (\ref{facG}) we have $$ J (K_{\perp})= \int \rho_A(\underbrace{0,...,0}_{p+1}|\underbrace{\ol{r}-r,...,\ol{r}-r}_{p+1}) j^p(r,\ol{r}) e^{i(\ol{r}-r)K_{\perp}} d^2 r d^2 \ol{r} $$ where $$ j(r,\ol{r})= \int d^2 c G_0(c) \lambda(c-r) \lambda(c-\ol{r}) $$ We also have used the translational invariance of the $\rho$-matrix. Note that near the real threshold we have no spectators and $$ \rho_A(\underbrace{0,...,0}_{p+1}|\underbrace{\ol{r}-r,...,\ol{r}-r}_{p+1}) =\rho_A(\underbrace{0,...,0}_{p+1}|\underbrace{0,...,0}_{p+1}) $$ In any case large $K_{\perp}$ corresponds to small $\ol{r}-r$ so we factor $\rho_A$ out of the integral in zero point. In the rest integral we pass to the variables $$ B=\frac{r+\ol{r}}{2}, \hs 1 b=\ol{r}-r $$ and shift the integration variable $c$, then \begin{equation} J (K_{\perp})= \rho_A(\underbrace{0,...,0}_{p+1}|\underbrace{0,...,0}_{p+1}) \int j^p(B,b) e^{ibK_{\perp}} d^2 b d^2 B \end{equation} where \begin{equation} j(B,b)= \int G_0(B+c) \lambda(\frac{b}{2}-c) \lambda(\frac{b}{2}+c) d^2 c \label{j} \end{equation} \section{The calculation of $\langle |K_{\perp}|\rangle $} Now we would like to find the width of the distribution on $K_{\perp}$ as a function of $p$ or what is the same of the cumulative number $x=(p+1)/3$. From the mathematical point of view it is simpler to calculate the mean squared width of the distribution $\langle K_{\perp}^2\rangle $. Unfortunately in our case this quantity is logarithmically divergent at large $K_{\perp}$. This divergency results from the behavior of $j(B,b)$ at small $b$. This behavior is determined by the behavior of the $ \lambda(b)=K_0(m|b|)/(2\pi) $ (\ref{lam}), which has a logarithmical singularity at $|b|=0$. Smooth $G_0(B+c)$ does not affect this behavior. For this reason we shall rather calculate $\langle |K_{\perp}|\rangle $: \begin{equation} \langle |K_{\perp}|\rangle =\frac{1}{J_N} \int j^p(B,b) |K_{\perp}|e^{ibK_{\perp}} d^2 b d^2 B \frac{d^2 K_{\perp}}{(2\pi)^2} \end{equation} where $J_N$ is the same integral as in the numerator but without $|K_{\perp}|$. Presenting $|K_{\perp}|$ as $K_{\perp}^2/|K_{\perp}|$ and $K_{\perp}^2$ as the Laplacian $\Delta_b$ applied to the exponent we find $$ \langle |K_{\perp}|\rangle =-\frac{1}{J_N} \int j^p(B,b) \Delta_b e^{ibK_{\perp}} d^2 b d^2 B \frac{d^2 K_{\perp}}{|K_{\perp}|(2\pi)^2} $$ Twice integrating by parts and using the formula $$ \int \frac{d^2 K_{\perp}}{|K_{\perp}|}e^{ibK_{\perp}} = \frac{2\pi}{|b|} $$ we find $$ \langle |K_{\perp}|\rangle =-\frac{1}{2\pi J_N} \int \frac{1}{|b|} \Delta_b j^p(B,b) d^2 b d^2 B $$ Now we again integrate by parts once to find $$ \langle |K_{\perp}|\rangle =-\frac{1}{2\pi J_N} \int d^2 B \frac{d^2 b}{|b|^2} (n_b \nabla_b) j^p(B,b) $$ where $n_b=b/|b|$. This leads to our final formula \begin{equation} \langle |K_{\perp}|\rangle =-\frac{p}{2\pi J_N} \int d^2 B \frac{d^2 b}{|b|^2} j^{p-1}(B,b) (n_b \nabla_b) j(B,b) \label{Kperp} \end{equation} where $j(B,b)$ is given by (\ref{j}), $ \lambda(b) $ is given by (\ref{lam}) and $$ J_N=\int d^2 B j^{p}(B,b=0) $$ \section{Approximations} To simplify numerical calculations we make some additional approximations, which are not very essential but are well supported by the comparison with exact calulations at a few sample points. As follows from the the asymptotics of $K_0(z)$ at large $z$ $$ K_0(z)\simeq\sqrt{\frac{\pi}{2z}}e^{-z} $$ the width of $\lambda(b)$ (\ref{lam}) is of the order $m^{-1}$. The function $G_0$ is smooth in the vicinity of the origin and its width $r_{0H}=\sqrt{r_0^2+r_H^2}$ is substancially larger than the width of $\lambda$. For this reason we factor $G_0(B+c)$ out of the integral (\ref{j}) over $c$ at the point $B$: \begin{equation} j(B,b)= G_0(B)\Lambda(b), \hs 1 \Lambda(b) \equiv \int \lambda(c) \lambda(c-b) d^2 c= \frac{|b|}{4\pi m}K_1(m|b|) \label{Lam} \end{equation} Then we find that the integrals over $B$ and $b$ decouple \begin{equation} J (K_{\perp})= \rho_A(\underbrace{0,...,0}_{p+1}|\underbrace{0,...,0}_{p+1}) \int G^p_0(B) d^2 B \int \Lambda^p(b) e^{ibK_{\perp}} d^2 b \end{equation} In this approximation we find that $\langle |K_{\perp}|\rangle $ depends only on one parameter - the constituent quark mass $m$, which in our approach plays the role of an infrared cutoff: \begin{equation} \langle |K_{\perp}|\rangle =pm \int_0^{\infty} dz K_0(z) (zK_1(z))^{p-1} \label{apprKperp} \end{equation} This allows to relate $m$ directly to the experimental data on the transverse momentum dependence. \section{Comparison with the data and discussion} The integral in (\ref{apprKperp}) can be easy calculated numerically. For values of $p=1,...,12$ it is very well approximated by a power dependence (see Fig.2), so that we obtain \begin{equation} \langle |K_{\perp}|\rangle/m =1.594\, p^{0.625} \end{equation} As we observe, the rise of $\langle |K_{\perp}|\rangle$ turns out to be even faster than expected on naive physical grounds mentioned in the Introduction ($\sim\sqrt{p}$). The resulting plots for $\langle |K_{\perp}|\rangle^2$ as a function of the cumulative number $x=(p+1)/3$ at different values of parameter $m$ are shown in Fig.3 together with avaiable experimental data from \cite{Boyarinov94} on $\langle K_{\perp}^2\rangle$ for pion production obtained in experiments \cite{Boyarinov94}-\cite{Boyarinov87} with 10 $GeV$ protons and \cite{Baldin82, Baldin83} with 8.94 $GeV$ protons. Note that earlier publications of the first group \cite{Boyarinov92,Boyarinov87} reported a much stronger increase of $\langle K_{\perp}^2\rangle$ with $x$, up to value 2 $(GeV/c)^2$ at $x=3$ for pion production. In our approach such an increase would require the quark mass to be as high as $m \simeq 225 MeV$. In a more recent publication \cite{Boyarinov94} the rise of $\langle K_{\perp}^2\rangle$ is substancially weaker (it corresponds to $m \simeq 175 MeV$ in our approach). The authors of \cite{Boyarinov94} explain this by new experimental data obtained and by a cutoff $K_{\perp max}$ introduced in calculations of $\langle K_{\perp}^2\rangle$ in \cite{Boyarinov94}. The introduction of this cutoff considerably (approximatly two times) decreases the experimental value of $\langle K_{\perp}^2\rangle$ at $x=3$. In our opinion this is a confirmation that the cumulative pion production rate only weakly decreases with $K_{\perp}$ in the cumulative region so that the the integral over $K_{\perp}^2$ which enters the definition of $\langle K_{\perp}^2\rangle$ is weakly convergent or even divergent, as in our approach. Undoubtedly presentation of the experimental data in terms of the mean value $\langle |K_{\perp}| \rangle^2$, rather than $\langle K_{\perp}^2\rangle$ should reduce the dependence on the cutoff $K_{\perp max}$ and make the results more informative. One of the ideas behind the investigations of the cumulative phenomena is that they may be a manifestation of a cold quark-gluon plasma formed when several nucleons overlap in the nuclear matter. In \cite{NPB94} we pointed out that our model does not correspond to this picture. It implies coherent interactions of the active quark with donors and, as a result, strong correlations between the longitudinal and transverse motion. Predictions for the dependence of $\langle |K_{\perp}| \rangle$ on $x$ are also different. From the cold quark-gluon plasma model one expects $\langle |K_{\perp}| \rangle$ to behave as $x^{1/3}$, since the Fermi momentum of the quarks inside the overlap volume is proportional to the cubic root of the quark density. Our model predicts a much faster increase, with a power twice larger. The experimental data seem to support our predictions. \section{Acknowledgments} The authors are greatly thankful to Prof. P.Hoyer who attracted their attention to the problem. This work is supported by the Russian Foundation for Fundamental Research, Grant No. 97-02-18123. \newpage
\section{Introduction} When the core of a large star ($M \ge 8 M_{\odot}$) runs out of nuclear fuel, it collapses and forms a proto-neutron star. The total energy released in the collapse, i.e., the gravitational binding energy of the core ($E_B \sim G_N M_ {\odot}^2/R$ with $R \sim$ 10 km), is about $3 \times 10^{53}$ ergs; $\sim$ 99\% of that is carried away by neutrinos and antineutrinos, the particles with the longest mean free path. It is believed that neutrinos of all three flavors are emitted with approximately equal luminosities over a timescale of several seconds. Those flavors which interact the most with the matter will decouple at the largest radius and thus the lowest temperature. The $\nu_\mu$ and $\nu_\tau$ neutrinos and their antiparticles have only neutral-current interactions with the matter, and therefore leave with the highest temperature, about 8 MeV (or $\langle E \rangle \simeq$ 25 MeV). The $\bar{\nu}_e$ and $\nu_e$ neutrinos have also charged-current interactions, and so leave with lower temperatures, about 5 MeV ($\langle E \rangle \simeq$ 16 MeV) and 3.5 MeV ($\langle E \rangle \simeq$ 11 MeV), respectively. The $\nu_e$ temperature is lower because the material is neutron-rich and thus the $\nu_e$ interact more than the $\bar{\nu}_e$. The observation of supernova $\nu_\mu$ and $\nu_\tau$ neutrinos and their antiparticles would allow the details of the picture above to be tested. In this talk I concentrate on two aspects of the neutrino signal: \begin{itemize} \item The possibility of measuring or constraining the mass of the $\nu_\tau$ and/or $\nu_\mu$. \item The possibility of locating the supernova by its neutrino signal, independently of or prior to the optical observation. \end{itemize} The physics of these task is straightforward, but there are complications due to: \begin{itemize} \item The finite statistics of the neutrino signal. \item The finite time duration of the signal. \end{itemize} The details of the work reported here can be found in the joint work with John Beacom of Caltech \cite{SK,SNO,point}. One can find a much more complete list of the relevant earlier references there. Numerical supernova models suggest that the neutrino luminosity rises quickly over a time of order 0.1 s, and then falls over a time of order several seconds. The rise is so fast that the details of its shape are largely irrelevant for our task. We model the luminosity fall by an exponential with time constant $\tau$ = 3 s. The luminosity then has a width of about 10 s, consistent with the SN 1987A observations. Later, I will show how the conclusions depend on the luminosity decay time constant $\tau$. \section{Neutrino mass determination} The requirement that neutrinos do not overclose the universe gives a bound for the sum of masses of stable neutrinos (see e.g., \cite{Raffelt}): \begin{equation} \sum_{i=1}^3 m_{\nu_i} \le 100 {\rm\ eV}\,. \label{eq:cosmo} \end{equation} However, laboratory kinematic tests of neutrino mass currently give limits for the masses compatible with the above cosmological bound only for the electron neutrino, $m_{\bar{\nu}_e} \le 5$ eV \cite{Belesev}. For the $\nu_\mu$ and $\nu_\tau$ they far exceed the cosmological bound: $m_{\nu_\mu} < 170$ keV\cite{RPP}, and $m_{\nu_\tau} < 18$ MeV \cite{RPP}. It is very unlikely that these mass limits can improve by the necessary orders of magnitude any time soon. When neutrinos are emitted by a supernova, even a tiny mass will make the velocity less than for a massless particle, and will cause a measurable delay in the arrival time. A neutrino with a mass $m$ (in eV) and energy $E$ (in MeV) will experience an energy-dependent delay (in s) relative to a massless neutrino in traveling over a distance D (in 10 kpc) of \begin{equation} \Delta t(E) = 0.515 \left(\frac{m}{E}\right)^2 D\,. \label{eq:delay} \end{equation} For a supernova at 10 kpc distance (approximately at the center of the galaxy), the delay for $\nu_e$ and $\bar{\nu}_e$ will be negligible, and their signal can be used as a reference clock. The $\nu_\mu$ and $\nu_\tau$ neutrinos and their antiparticles will interact only by the neutral current. Thus, in order to determine the $\nu_\tau$ and/or $\nu_\mu$ mass, we should find ways of separating the neutral and charged current signals, and of determining the possible time delay of the former with respect to the latter. There are three neutral current reactions that give rise to potentially measurable signals: a) neutrino-electron scattering (we will show below that it is difficult to separate the charged and neutral current events in that case), b) neutral current excitation of $^{16}$O nuclei in water, followed by the $\gamma$ emission as suggested in \cite{LVK}, and c) the neutral current deuteron disintegration (relevant for SNO). In Table \ref{tab:rate} I show the corresponding numbers of events (see \cite{SK,SNO} for details how the table was made and refs. \cite{SKD,SNOD} for description of the detectors) for the individual reactions, calculated for the ``standard'' supernova defined above. \begin{table}[h] \caption{Calculated numbers of events expected in SK and SNO. In SNO events in 1 kton of D$_2$O and in 1.4 kton of H$_2$O are added. By $\nu_x$ we denote the combined effect of $\nu_\mu$ and $\nu_\tau$, each accounts for half of the events. In all except the top row, the events caused by $\nu$ and $\bar{\nu}$ are added.} \label{tab:rate} \vspace{5 mm} \begin{center} \small \begin{tabular}{|l|l|l|} \hline reaction & events in SK & events in SNO \\ \hline\hline $\bar{\nu}_e + p \rightarrow e^+ + n$ & 8300 & 365 \\ \hline $\nu_e + d \rightarrow e^- + p + p $ & - & 160 \\ $\bar{\nu}_e + d \rightarrow e^+ + n + n $ & & \\ \hline $\nu_x + d \rightarrow \nu_x + n + p $ & - & 400 \\ \hline $\nu_x+ ^{16}{\rm O} \rightarrow \nu_x + \gamma + X$ & 710 & 50 \\ \hline $\nu_x+ ^{16}{\rm O} \rightarrow \nu_x + n + ^{15}{\rm O} $ & - & 15 \\ \hline $\nu_e + e^- \rightarrow \nu_e + e^-$ & 200 & 15 \\ \hline $\nu_x + e^- \rightarrow \nu_x + e^-$ & 120 & 10 \\ \hline \hline \end{tabular} \end{center} \end{table} Given the assumed known time dependence of the supernova luminosity $L(t)$, and assuming that all flavors develop in time the same way and keep their temperatures constant, the arrival time of massive neutrinos is then described by $L(t - \Delta t(E_{\nu}))$. Since in the neutral current scattering one cannot determine the incoming neutrino energy $E_{\nu}$, we have at our disposal only the {\it time distribution} of the events \begin{equation} \frac{dN}{dt} = C \int dE_{\nu} f(E_{\nu}) \sigma(E_{\nu}) L(t - \Delta t(E_{\nu})) \,, \label{eq:rate} \end{equation} where $C$ is a constant proportional to $1/(D^2 \times \langle E_{\nu} \rangle)$ and $f(E_{\nu})$ is the thermal neutrino spectrum. The only way one can decide whether there is a time delay or not is to compare the neutral current time distribution (called ``Signal'') with the time distribution of the charged current events (called ``Reference''). Since $\nu_\tau$ and $\nu_{\mu}$ neutrinos and their antiparticles have higher energies than $\nu_e$ and $\bar{\nu}_e$, the neutral current events will contain a substantial fraction of possibly delayed events, while the charged current events will have no delay. It turns out, see \cite{SK}, that the most efficient way to accomplish this is also the simplest one, i.e., to use the diffence in the {\it mean arrival time}: \begin{equation} \langle t \rangle_S = \sum_k t_k /N_S \,, ~ \langle t \rangle_R = \sum_k t_k /N_R \,, \end{equation} where $N_S (N_R)$ is the total number of the Signal (Reference) events, and $t_k$ are the arrival times of the individual events. The signature of neutrino mass is then the inequality \begin{equation} \langle t \rangle_S > \langle t \rangle_R \,, \end{equation} valid with significance beyond statistical fluctuations. \begin{figure}[h] \vspace*{13pt} \begin{center} \mbox{\epsfig{figure=moments1m3s.eps,width=8cm}} \vspace*{13pt} \fcaption{The results of the $\langle t \rangle$ analysis for a massive $\nu_\tau$ in SK using the $\gamma$ following $^{16}$O excitation. In the upper panel, the relative frequencies of various $\langle t \rangle_S - \langle t \rangle_R$ values are shown for a few example masses. In the lower panel, the range of masses corresponding to a given $\langle t \rangle_S - \langle t \rangle_R$ is shown. The solid line is the 50\% confidence level, and the upper and lower dashed lines are the 10\% and 90\% confidence levels, respectively.} \label{fig:mom} \end{center} \end{figure} The analysis below is based on the assumption that only one of the neutrino flavors is massive, say $\nu_{\tau}$, and the other one, $\nu_{\mu}$ in this case, is either massless or has so much smaller mass that the corresponding time delay is negligible. The ``Signal'' then consist of part that is delayed and another part that is not because it is either caused by the massless $\nu_{\mu}$ or belongs to background that cannot be separated from the signal since it has the same energy and angle, etc. With these assumptions, the neutrino-electron scattering signal in SK will contain 60 delayed events and 700 background events, since a rather large number of the charged current $\bar{\nu}_e + p \rightarrow e^+ + n$ events will be present in the forward cone. For the $\gamma$ signal from the $^{16}$O excitation, the ratio delayed/background is a more favorable 355/885. And in SNO the neutral current deuteron disintegration, with a single neutron and no charged lepton, is characterized by the ratio 219/316. The last two ratios above also show that events which look like the neutral current (i.e., the true neutral current plus background with similar characteristics) are dominated by the response to $\nu_{\tau}$ and $\nu_{\mu}$ neutrinos. Indeed, since the cross sections are the same for these two flavors, one can simply multiply the numerators by a factor of two and make the corresponding adjustment in the denominator, to obtain the fraction of events caused by the $\nu_{\tau}$ and $\nu_{\mu}$ neutrinos. So, for the $\gamma$ signal from the $^{16}$O excitation, that contribution is about 57\%, and for the deuteron disintegration it is about 82\%. On the other hand, for the neutrino-electron scattering it is only about 16\%. By measuring the total number of the Signal events one can determine the temperatures of the $\nu_{\tau}$ and $\nu_{\mu}$ neutrinos (see \cite{SK,SNO}) with reasonable accuracy. To judge the statistical significance of the delay, we used the Monte Carlo simulation of a large number of supernovae for each mass value. We then histogram the differences $ \langle t \rangle_S - \langle t \rangle_R$, and find the 10\%, 50\%, and 90\% confidence levels. Representative cases are plotted in Fig. 1 for the $\gamma$ from $^{16}$O excitation in SK. In the upper panel one can see that if, e.g., $m_{\nu}$ = 75 eV the most probable difference in average arrival times is about 0.2 s, and the 10 - 90 \% CL band is 0.1 - 0.3 s, clearly separated from the massless case. In fact, the smallest recognizable mass is about 50 eV. In SNO, using the deuteron disintegration, the smallest recognizable mass is even smaller, about 30 eV. How does the mass sensitivity depend on the assumptions we made above? Much of the analysis can be made analytically. We shall concentrate on the dependence on the assumed temperatures $T$, distance $D$, and the constant which characterizes the time duration of the neutrino signal, $\tau$. The time delay and its error depend on \begin{equation} \langle t \rangle_S - \langle t \rangle_R \sim (m/T)^2 D ~;~ \delta( \langle t \rangle_S - \langle t \rangle_R) \sim \tau D/\sqrt{T} ~. \end{equation} Since the significance of the result, and thus the smallest recognizable mass $m_{lim}$, is the ratio of these two quantities, we conclude that this neutrino mass limit is, remarkably, independent on the distance $D$, and \begin{equation} m_{lim} \sim \sqrt{\tau} T^{3/4} ~. \end{equation} Clearly, the shorter the duration of the neutrino pulse (i.e., smaller $\tau$), the better the ability to determine the neutrino mass. (We verified that detailed numerical simulation closely follows the $\sqrt{\tau}$ scaling above) Also, naturally, if e.g. the $\nu_{\mu}$ and $\nu_{\tau}$ masses are close to each other, the mass limit is improved, roughly by $\sqrt{2}$. \section{ Supernova localization with neutrinos} A future core-collapse supernova in our Galaxy will be detected by several neutrino detectors around the world. The neutrinos escape from the supernova core over several seconds from the time of collapse, unlike the electromagnetic radiation, emitted from the envelope, which is delayed by a time of order hours. In addition, the electromagnetic radiation can be obscured by dust in the intervening interstellar space. The question therefore arises whether a supernova can be located by its neutrinos alone. The early warning of a supernova and its location might allow greatly improved astronomical observations. There are two types of techniques to locate a supernova by its neutrinos. The first one is based on angular distributions of the neutrino reaction products, which can be correlated with the neutrino direction. In this case, a single experiment can independently announce a direction and its error. However, to suppress false alarms one can demand coincidence with other experiments. The second method of supernova location is based on triangulation using two or more widely-separated detectors. This technique would require significant and immediate data sharing among the different experiments. The theme of this section (for more details and more complete reference list, see \cite{point}) is a careful and realistic assessment of this question, taking into account the statistical significance of the various neutrino signals. \subsection{ Reactions with angular dependence} Neutrino-electron scattering occurs for all flavors of neutrinos and antineutrinos, and is detected by observing the recoil electrons with kinetic energy $T$ . The scattering angle is dictated by the kinematics and is given by \begin{equation} \cos\alpha = \frac{E_{\nu} + m_e}{E_{\nu}} \left( \frac{ T}{T + 2m_e} \right)^{1/2}\,. \end{equation} With threshold of about $T_{\rm min} = 5$ MeV, the recoil electrons will be sharply forward scattered, i.e., pointing away from the supernova, with the combined average $\langle \cos \alpha \rangle$ = 0.98, corresponding to about $11^{\circ}$. However, multiple scattering will smear the \v{C}erenkov cone, resulting in a one-sigma width of $\sim 25^\circ$. In order to evaluate the pointing ability of this signal we have to take into account the finite statistics, the two-dimensional form of the resulting distribution, and the presence of the unavoidable background. The background worsens the pointing ability from the simple expectation by a factor \cite{point} $C(R)$ \begin{equation} \delta x = \frac{\sigma}{\sqrt{N_S}} \times C(R); ~{\rm with} ~ C(R) \approx \sqrt{1 + 4 R} ~, \end{equation} where $R$ is the ratio (at the peak) of the flat background and the signal with $N_S$ events. For SK and SNO the background reduction factor is $C(R) \simeq 2 - 3$, and with our standard supernova parameters we find that the one-sigma error based on the neutrino-electron scattering will be about $5^{\circ}$ in SK and about $20^{\circ}$ in SNO. This is by far the most accurate pointing ability at our disposal. The reaction with the most events is $\bar{\nu}_e + p \rightarrow e^+ + n$, with $\simeq 10^4$ events expected in SK, and $\simeq 400$ events expected in the light water of SNO. In \v{C}erenkov detectors one can determine the direction of the positrons, whose angular distributions with respect to the direction of the neutrino beam is of the form \begin{equation} \frac{d N}{d \cos\alpha} = \frac{N}{2} \left(1 + a \cos\alpha \right)\,. \label{eq:acos} \end{equation} It is relatively easy to show that the error in the pointing ability for $N$ observed events in this case is given by \begin{equation} \delta (\cos\alpha) = \frac{2}{|a|} \frac{1}{\sqrt{N}}\,. \end{equation} Since, in general $a = a(E_\nu)$, we have to investigate further the neutrino energy dependence of this coefficient, and perform the necessary energy averaging. \begin{figure}[h] \vspace*{13pt} \begin{center} \mbox{\epsfig{figure=pos60.eps,width=8cm}} \vspace*{13pt} \fcaption{Upper panel: total cross section for $\bar{\nu}_e + p \rightarrow e^+ + n$; bottom panel: $\langle \cos\theta \rangle$ for the same reaction; both as a function of the antineutrino energy. The solid line is the ${\cal O}(1/M)$ result and the short-dashed line is the ${\cal O}(1)$ result. The long-dashed line is the result of Eq.(3.18) of Ref.~\protect\cite{LS}, and the dot-dashed line contains our threshold modifications to the same.} \label{fig:cos} \end{center} \end{figure} In the limit where the nucleon mass $M$ is taken to be infinite, i.e., zeroth order in $1/M$ (${\cal O}(1)$), the asymmetry coefficient $a$ is independent of $E_\nu$ and is given simply by the competition of the non-spin-flip (Fermi) and spin-flip (Gamow-Teller) contributions, and is \begin{equation} a^{(0)} = \frac{f^2 - g^2}{f^2 + 3g^2} \simeq -0.10 ~~;~~ f = 1, g = 1.26 ~, \label{eq:a0} \end{equation} and thus the angular distribution of the positrons is weakly backward. However, $a(E_\nu)$ is substantially modified when weak magnetism and recoil corrections of ${\cal O}(1/M)$ are included. It turns out \cite{angul} that the inclusion to this order gives a very accurate formula for $\langle \cos \theta \rangle$. This quantity and the total cross section are shown in Fig. \ref{fig:cos}, evaluated in various approximations (see \cite{angul}). At high energies, the formula (3.18) of \cite{LS}, valid to all orders in $1/M$, but neglecting the threshold effects, is applicable. One can see in Fig. \ref{fig:cos} that the dot-dashed line smoothly interpolates between the correct low energy and high energy behaviour. As far as the pointing ability of the $\bar{\nu}_e + p \rightarrow e^+ + n$ reaction is concerned, due to the rather small angular asymmetry (and its energy dependence) we estimate that the uncertainty $\delta (\cos\alpha) \simeq 0.2$ even for the high statistics detector like SK. Nevertheless, it would be important and useful to use this additional information constraining the supernova direction. \subsection{ Triangulation} For two detectors separated by a distance $d$, there will be a delay between the arrival times of the neutrino pulse. The magnitude of the delay $\Delta t$ depends upon the angle $\theta$ between the supernova direction and the axis connecting the two detectors. Given a measured time delay $\Delta t$, the unknown angle $\theta$ and its error are then: \begin{equation} \cos\theta = \frac{\Delta t}{d} ~~;~~ \delta(\cos\theta) = \frac{\delta(\Delta t)}{d}\,. \end{equation} Thus two detectors define a cone along their axis with opening $\cos\theta$ and thickness $2 \times \delta(\cos\theta)$ in which the supernova can lie. Obviously, in order to have a reasonable pointing accuracy from triangulation, one will need $\delta(\Delta t) \ll d$. (The Earth diameter is $d \approx 40$ ms.) Following \cite{point} I discuss whether an appropriate time delay can be defined, and what its error would likely be. Basically, the question can be reduced to the following problem in statistics: given $N$ events of duration $\tau$, is the uncertainty $\delta(\Delta t)$ equal to $\tau/N$ (i.e., the interval between events) or the much larger $\tau/\sqrt{N}$? The answer, of obvious practical significance, requires a degree of subtlety. Let us model, as before, the time dependence of the neutrino pulse (i.e. the supernova luminosity $L(t)$) by two exponentials, the increasing sharp rise with time constant $\tau_1$ and the slow decay with the time constant $\tau_2$ ($\tau_1 \ll \tau_2$). Now take the limit (unrealistic) of zero risetime ($\tau_1 \rightarrow 0$). Then, in fact, the first answer is applicable, i.e., $\delta(\Delta t) \rightarrow \tau_2/N$. One would then simply determine the arrival time of the first event in each detector, and the triangulation would be feasible, though still not very accurate. But any finite leading edge, or background, would invalidate this picture. Moreover, we know that the leading edge has a finite duration related to the shock propagation time in the supernova. The best strategy then is to try to determine the rather sharp point of maximum rate $t_0$. The error in its determination depends on the duration of the leading edge $\tau_1$, which can be measured in the largest detector, and on the number of events $N_1$ in the leading edge for the given detector, \begin{equation} (\delta t_0)_{min} ~\approx ~ \frac{\tau_1}{\sqrt{N_1}} ~. \end{equation} At the same time, the number of events $N_1$ in the leading edge depends somewhat indirectly on the total duration of the pulse, since $N_1 \approx N \tau_1/\tau_2$. For the existing detectors, this leads to rather large uncertainty, $\delta(\cos\theta) \approx 0.5$. Nevertheless, if there will be several large detectors available in not too distant future, triangulation would offer another handle to the supernova localization, besides the obvious benefit of the false alarm elimination by the coincidence requirement. \section{Conclusions} In this talk, which is based on the results of Refs. \cite{SK,SNO,point,angul}, I have shown that: \begin{itemize} \item The supernova signal caused by $\nu_{\tau} + \nu_{\mu}$ and their antiparticles can be isolated. \item By measuring the average arrival time difference of the neutral and charged current events, one will be able to (conservatively) determine the upper limit for $m_{\nu_{\tau}}$ of 30-50 eV, representing an improvement by $10^6$ when compared to the existing limits. \item Neutrino electron scattering can be used for pointing with accuracy of about $5^{\circ}$. \item $\bar{\nu}_e + p \rightarrow e^+ + n$ can be also used for crude pointing, provided the correct differential cross section is used. (Remembering that the naive formula suggests that the positron are slightly backward while in reality they should be slightly forward.) \item Triangulation appears to be difficult if the supernova signal is going to last more than one second. But it would be useful if more than two detectors (and even better if they are going to be large) will participate in the warning network. \end{itemize} \section{Acknowledgements} The collaboration with John Beacom is gratefully acknowledged. This work was supported in part by the US Department of Energy under Grant No. DE-FG03-88ER-40397.
\section{Introduction} The angular size - redshift relation, $\Theta (z)$, is a kinematic test which potentially may discriminate the several cosmological models proposed in the literature. As widely known, because of the spacetime curvature, the expanding universe acts gravitationally as a lens of large focal length. Though nearby objects are not affected, a fixed angular size of an extragalactic source is initially seen decreasing up to a minimal value, say, at a critical redshift ($z_m$), after which increasing for higher redshifts. The precise determination of $z_m$, or equivalently, the corresponding minimal angular size value $\Theta(z_m)$, may constitute a powerful tool in the search for deciding which are the more realistic world models. This lensing effect was first predicted by Hoyle, originally aiming to distinguish between the steady-state and Einstein-de Sitter cosmologies \cite{hoyle}. Later on, the accumulated evidences against the steady state (mainly from CMBR) have put it aside, and more recently, the same is occurring with the theoretically favoured critical density FRW model \cite{krauss2,Riess,Perlmutter,Alcaniz,Roos}. The data concerning the angular size - redshift relation are until nowadays somewhat controversial, specially because they envolve at least two kinds of observational dificulties. First, any large redshift object may have a wide range of proper sizes, and, second, evolutionary and selection effects probably are not negligible. The $\Theta(z)$ relation for some extended sources samples seems to be quite imcompatible with the predictions of the standard FRW model when the latter effects are not taken into account \cite{Sand,Kapa1,Kapa2}. There have also been some claims that the best fit model for the observed distribution of high redshifts extended objects is provided by the standard Einstein-de Sitter universe ($q_o={1 \over 2}$, $\Omega_\Lambda=0$) with no significant evolution \cite{Bucha}. Parenthetically, these results are in contradiction with recent observations from type Ia supernovae, which seems to ruled out world models filled only by baryonic matter, and more generally, any model with positive deceleration parameter \cite{Riess,Perlmutter}. The same happens with the corresponding bounds using the ages of old high redshift galaxies \cite{Alcaniz,Dunlop,Krauss3}. The case for compact radio sources is also of great interest. These objects are apparently less sensitive to evolutionary effects since they are short-lived ($\sim 10^{3} yr$) and much smaller than their host galaxy. Initially, the data from a sample of 82 objects gave remarkable suport for the Einstein-de Sitter Universe \cite{Kell}. However, some analysis suggest that Kellerman has not really detected a significant increasing beyond the minimum \cite{Dabro,Step,Cool}. Some authors have also argued that models where $\Theta(z)$ diminishes and after a given $z$ remains constant may also provide a good fit to Kellerman's data. In particular, by analysing a subset of 59 compact sources within the same sample, Dabrowski et al. (1995) found that no useful bounds on the value of the deceleration parameter $q_o$ can be derived. Further, even considering that Euclidean angular sizes ($\Theta \sim z^{-1}$) are excluded at 99$\%$ confidence level, and that the data are consistent with $q_o=1/2$, they apparently do not rule out extreme values of the deceleration parameter as $q_{o} \sim 5$ \cite{Step}. More recently, based in a more complete sample of data, which include the ones originally obtained by Kellermann, it was argued that the $\Theta(z)$ relation may be consistent with any model of the FRW class with deceleration parameter $\leq 0.5$ \cite{Gurv}. In this context, we discuss here how the critical redshift giving the turn-up in angular sizes is determined for any expanding cosmology based on the FRW geometry. An analytical expression quite convenient for numerical evaluation is derived. The approach is exemplified for three different models of current cosmological interest: (i) open matter dominated FRW universe (OCDM), (ii) flat FRW type models with cosmological constant ($\Lambda$CDM), (iii) the class of scalar field cosmologies (SF) proposed by Ratra and Peebles \cite{Ratra}. Hopefully, the results derived here may be useful near future, when more accurate data become available. \section{The method} Let us now consider the FRW line element $(c=1)$ \begin{equation} ds^2 = dt^2 - R^{2}(t) [d\chi^{2} + S^{2}_{k}(\chi) (d \theta^2 + \rm{sin}^{2} \theta d \phi^{2})] \quad , \end{equation} where $\chi$, $\theta$, and $\phi$ are dimensionless comoving coordinates, $R(t)$ is the scale factor, and $S_{k}(\chi)$ depends on the curvature parameter ($k=0$, $\pm 1$). The later function is defined by one of the following forms: $S_k (\chi) = \rm{sinh} (\chi)$, $\chi$, $\rm{sin} \chi$, respectively, for open, flat and closed Universes. In this background, the angular size-redshift relation for a rod of intrinsic length $D$ is easily obtained by integrating the spatial part of the above expression for $\chi$ and $\phi$ fixed. One finds \begin{equation} \theta(z) = {D (1 + z) \over R_{o}S_{k}(\chi)} \quad . \end{equation} The dimensionless coordinate $\chi$ is given by \begin{equation} \chi(z) = {1 \over H_o R_o} \int_{(1 + z)^{-1}}^{1} {dx \over x E(x)} \quad , \end{equation} where $x = {R(t) \over R_o} = (1 + z)^{-1}$ is a convenient integration variable. For the three kinds of cosmological models considered here (OCDM, $\Lambda$CDM and SF) the dimensionless function $E(x)$ assume one of the following forms: \begin{equation} E_{FRW}(x) = \left[1 - \Omega_{M} + \Omega_{M} x^{-1}\right]^{{1 \over 2}} \quad , \end{equation} \begin{equation} E_{\Lambda}(x) = \left[(1 - \Omega_{\Lambda}) x^{-1} + \Omega_{\Lambda}x^{2}\right]^{{1 \over 2}} \quad , \end{equation} \begin{equation} E_{SF}(x) = \left[(1 - \Omega_{\phi})x^{-1} + \Omega_{\phi}x^{{4-\alpha} \over {2 + \alpha}}\right]^{{1}\over{2}} \quad , \end{equation} where $\Omega_{M} = {{8 \pi G\rho_{M}} \over 3 H_{o}^{2}}$, $\Omega_{\Lambda} = {\Lambda \over 3H_{o}^{2}}$ and $\Omega_{\phi} = {8 \pi G\rho_{\phi} \over 3 H_{o}^{2}}$, are the present day density parameters associated with the matter component, cosmological constant and the scalar field $\phi$, respectively. Notice that equations (5) and (6) become identical if one takes $\alpha = 0$ in the later, thereby showing that the scalar field model proposed by Ratra and Peebles may kinematically be equivalent to a flat $\Lambda$CDM cosmology. The redshift $z_{m}$ at which the angular size takes the minimum value is the one cancelling out the derivative of $\Theta$ with respect to $z$. Hence, from (2) we have the condition \begin{equation} S_k (\chi_m) = (1 + z_m)S'_k (\chi_m) \quad , \end{equation} where $S'_k (\chi) = {\partial S_{k} \over \partial \chi}{\partial \chi \over \partial z}$, a prime denotes differentiation with respect to $z$ and by definition $\chi_{m}= \chi(z_{m})$. To proceed further, observe that (3) can readily be differentiated yielding, respectively, for the standard FRW (matter dominated), $\Lambda$CDM and scalar field cosmologies \begin{equation} (1 + z_{m})\chi'_{m} = {(R_o H_o)^{-1} \over \left[1 - \Omega_M + \Omega_M (1 + z_m)\right]^{{1 \over 2}}} = (R_o H_o)^{-1} F(\Omega_{M}, z_m) \quad , \end{equation} \begin{equation} (1 + z_{m})\chi'_{m} = {(R_o H_o)^{-1} \over \left[(1 - \Omega_{\Lambda})(1 + z_m) + \Omega_{\Lambda}(1 + z_m)^{-2}\right]^{{1 \over 2}}} = (R_o H_o)^{-1} L(\Omega_{\Lambda}, z_m) \quad , \end{equation} \begin{equation} (1 + z_{m})\chi'_{m} = {(R_o H_o)^{-1} \over \left[(1 - \Omega_{\phi})(1 + z_m) + \Omega_{\phi}(1 + z_m)^{{\alpha - 4 \over \alpha + 2}}\right]^{{1 \over 2}}} = (R_o H_o)^{-1} S(\Omega_{\phi}, \alpha, z_m) \quad . \end{equation} Now, inserting the above equations into (7) we find for the cases above considered \begin{equation} {1 \over (1 - \Omega_{M})^{1 \over 2}}{\rm{tanh}}\left[(1 - \Omega_{M})^{1 \over 2}\int_{(1 + z_m)^{-1}}^{1} {dx \over x E_{FRW}(x)}\right] = F(\Omega_{M}, z_m) \quad , \end{equation} \begin{equation} \int_{(1 + z_m)^{-1}}^{1} {dx \over x E_{\Lambda}(x)} = L(\Omega_{\Lambda}, z_m) \quad , \end{equation} \begin{equation} \int_{(1 + z_m)^{-1}}^{1} {dx \over x E_{SF}(x)} = S(\Omega_{\phi}, \alpha, z_m) \quad . \end{equation} The meaning of equations (11)-(13) is self evident. Each one represents an integro-algebraic equation for the critical redshift $z_m$ as a function of the physically meaningful parameters of the models. In general, these equations cannot be solved in closed analytical form for $z_m$. However, as one may check, if we take the limit $\Omega_M \rightarrow 1$ in (11), the value $z_m = {5 \over 4}$ is readily achieved, which corresponds to the well known standard result for the dust filled FRW flat universe. The interesting point is that expressions (11)-(13) are quite convenient for numerical evaluations. As a matter of fact, their solutions can straightforwardly be obtained, for instance, by programming the integrations using simple numerical recipes in FORTRAN. In Fig.~1 we show the diagrams of $z_m$ as a function of the density parameter for each kind of model. As expected, in the standard FRW model, the critical redshift starts at $z_m=1.25$ when $\Omega_M$ goes to unity. This value is pushed to the right direction, that is, it is displaced to higher redshifts as the $\Omega_M$ parameter is decreased. For instance, for $\Omega_M=0.5$ and $\Omega_M=0.2$, we find $z_m=1.58$ and $z_m=2.20$, respectively. In the limiting case, $\Omega_M \rightarrow 0$, there is no minimum at all since $z_{m} \to \infty$. This means that the angular size decreases monotonically as a function of the redshift. For the scalar field case, one needs to fix the value of $\alpha$ in order to have a bidimensional plot. Given a value of $\Omega_{\phi}$, the minimum is also displaced for higher redshifts when the $\alpha$ parameter diminishes. Conversely, for a fixed value of $\alpha$, the minimum moves for lower redshifts when $\Omega_\phi$ is decreased. The limiting case ($\alpha=0$) is fully equivalent to a $\Lambda$CDM model. As happens in the limiting case $\Omega_M \rightarrow 0$ ($\Omega_{\Lambda} = 0$), the minimal value for $\Theta(z)$ disappears when the cosmological constant contributes all the energy density of the Universe, that is, $z_m \rightarrow \infty$ if $\Omega_M \rightarrow 0$ and $\Omega_{\Lambda} \rightarrow 1$ (in this connection see also \cite{krauss1}). For the class of models considered in this paper, the redshifts having the minimal angular size are displayed for several values of $\Omega_M$ and $\alpha$ in Table 1. As can be seen there, the critical redshift at which the angular size is a minimal cannot alone discriminate between world models since different scenarios may provide the same $z_m$ value. However, when combinated with other tests, some interesting constraints on the cosmological models can be obtained. For example, when $\Omega_\phi$ is bigger than $0.55$, the model proposed by Ratra and Peebles yields a $z_m$ between the standard FRW flat model and the $\Lambda$CDM cosmology. Then, suposing that the universe is really accelerating today ($q_{o} < 0$), as indicated recently by measurements using type Ia supernovae \cite{Riess,Perlmutter}, and by considering the results by Gurvits et al. \cite{Gurv}, i.e., that the data are compatible with $q_{o} \leq 0.5$, the Ratra and Peebles models with $0 < \alpha \leq 4$ seems to be more in accordance with the angular size data for compact radio sources than the $\Lambda$CDM model. \begin{table}[t] \begin{center} \begin{tabular}{rrlll} \hline \multicolumn{1}{c}{$\Omega_{m}$ ($z_{m}$)}& \multicolumn{1}{c}{$\Omega_{\Lambda}$ ($z_{m}$)}& \multicolumn{1}{c}{$\Omega_{\phi} (\alpha = 2)$ ($z_{m}$)}& \multicolumn{1}{c}{$\Omega_{\phi} (\alpha = 4)$ ($z_{m}$)}& \multicolumn{1}{c}{$\Omega_{\phi} (\alpha = 6)$ ($z_{m}$)}\\ \\ \hline 1.0 (1.25)& 1.0 ($\infty$)& 1.0 (2.16)& 1.0 (1.72)& 1.0 (1.57)\\ \\ 0.8 (1.35)& 0.8 (1.76)& 0.8 (1.65)& 0.8 (1.53)& 0.8 (1.46)\\ \\ 0.7 (1.41)& 0.7 (1.60)& 0.7 (1.55)& 0.7 (1.47)& 0.7 (1.42)\\ \\ 0.5 (1.58)& 0.5 (1.44)& 0.5 (1.42)& 0.5 (1.38)& 0.5 (1.36)\\ \\ 0.2 (2.20)& 0.2 (1.31)& 0.2 (1.30)& 0.2 (1.30)& 0.2 (1.29)\\ \hline \end{tabular} \caption{Critical redshift $Z_m$ in OCDM, $\Lambda$CDM, and scalar field cosmologies for some selected values of the density parameters.} \end{center} \end{table} \pagebreak \begin{figure} \vspace{.2in} \centerline{\psfig{figure=angsn.ps,width=3truein,height=3truein} \hskip 0.1in} \caption{Critical redshift $Z_m$ as a function of the density parameter in open, ${\Lambda}\rm{CDM}$ and scalar field cosmologies. Solid curve is the prediction for a model with nonnull cosmological constant. The same curve is also obtained as a limiting case ($\alpha \rightarrow 0$) of the scalar field cosmology proposed by Ratra and Peebles} \end{figure} It is worth notice that the same procedure may be applied when evolutionary and/or selection effects due to a linear size-redshift or to a linear size-luminosity dependence are taken into account. As widely believed, a plausible way of standing for such effects is to consider that the intrinsic linear size has a similar dependence on the redshift as the coordinate dependence, i.e., $D = D_o (1 + z)^{c}$, being $c < 0$ (see, for instance, \cite{Bucha} and Refs. therein). In this case, equations (11)-(13) are still valid but the functions $F(\Omega_{M}, z_m)$, $L(\Omega_{\Lambda}, z_m)$, and $S(\Omega_{\phi}, \alpha, z_m)$ must be divided by a factor $(1 + c)$. The displacement of $z_{m}$ relative to the case with no evolution ($c = 0$) due to the effects cited above may be unexpectedly large. For example, if one takes $c = -0.8$ as found by Buchalter et al. \cite{Bucha}, the redshift of the minimum angular size for the Einstein-de Sitter case ($\Omega_{M} = 1$) moves from $z_{m} = 1.25$ to $z_{m} = 11.25$. In particular, this explains why the data of Gurvits et al. \cite{Gurv}, although apparently in agreement with the Einstein-de Sitter universe, do not show clear evidence for a minimal angular size close to $z = 1.25$, as should be expected for this model. \hspace{0.3cm} {\bf Acknowledgments:}This work was partially supported by the project Pronex/FINEP (No. 41.96.0908.00) and Conselho Nacional de Desenvolvimento Cient\'{\i}fico e Tecnol\'ogico - CNPq (Brazilian Research Agency).
\section{Introduction: Wide-angle scattering in QCD} Although attempts to apply perturbative QCD to wide-angle elastic hadron scatterings have been undertaken in a number of papers [1-10], explicit predictions have been available only for elastic processes involving external photons, such as $\gamma + \gamma\rightarrow$ hadrons, Compton scattering of hadrons, etc. Predictions based on perturbative QCD rest on three premises: 1) hadronic interactions become weak at small invariant separation $r\ll \Lambda_{QCD}^{-1}$; 2) the perturbative expansion in $\alpha_s(Q)$ is well-defined; and 3) factorization, implying that all effects of collinear singularities, confinement, non-perturbative interactions and bound state dynamics can be isolated at large momentum transfer in terms of the (process independent) structure functions $G_{i/H}(x,Q),$ fragmentation functions $D_{H/i}(z,Q)$ or, in the case of exclusive processes, distribution amplitudes $\Phi_H(s_i,Q).$ Consequently the hadronic scattering amplitude takes the form \begin{equation} A=\int\prod_H\phi_H(x_i,Q) T(x_i,p_H;Q)[dx_i]~, \label{eq1} \end{equation} where $\Phi(x_i,Q)$ is a universal distribution amplitude which gives the probability amplitude for finding the valence $q\bar q$ or $qqq$ in the hadronic wave function collinear up to the scale $Q=\sqrt{s\over 2}$, and $T$ is the hard scattering amplitude for valence quark collisions. The technical complication which has made particularly difficult to compute the behavior of hadron-hadron amplitudes is the possibility of multiple scatterings. The standard factorized form for the elastic scattering of hadrons $\{i\}$ is \begin{equation} A_1(s,t)=\int^1_0\prod_{i=1}^4[dx_i]\phi_i(x_i)T(\{x_i\},s,t)~, \label{eq2} \end{equation} where $x_i$ represents collectively the fractional momenta of hadron $i$ carried by its valence partons. According to this concept, all of the partons collide in a small region of the space-time of typical dimension $Q^{-1}$. The relevant contribution to the amplitude behaves according to the dimensional counting \cite{BF73, mtm73}, i.e. \begin{equation} A_1(s,t)\sim({\mu^2\over s})^{n/2-2}f_1(s/t) \label{eq3} \end{equation} for $n$ partons participating in the hard scattering, $\mu$ representing hadronic mass scales, which make the amplitude dimensionless. An extension of this "single-scattering" scenario is the (double) "independent- scattering" picture, due to P. Landshoff \cite{Lan74}, in which two pairs of partons scatter independently off two scattering centers. According to this picture, the lowest order diagrams contribute with \begin{equation} A_m(s,t)\sim\biggl({\mu^2\over s}\biggr)^{(n-m+1)/2-2}f_m(s/t)~, \label{eq4} \end{equation} where $m$ is the number of independent scatterings. If so, the multiple scattering should dominate in the case of wide angle scattering. A solution to this problem was pointed out in Refs.~\cite{LB80} and \cite{LP81}, where it was shown that the Sudakov logarithms associated with the rescattering diagrams do not cancel. In the leading logarithmic approximation they exponentiate to suppress the typical double scattering contribution by a factor \begin{displaymath} exp(-const\ln Q^2\ln(\ln Q^2))~, \end{displaymath} characteristic of the Sudakov suppression in QCD. More quantitatively \cite{M81}, \begin{equation} A_2\sim{1\over{Q^4}}\biggl({Q\over \Lambda_{QCD}}\biggr)^{1-2c\ln (1/r)}\,, \label{eq5} \end{equation} where \begin{displaymath} r=2c/(1+2c)~,~~~\ \ \ \ c=32/(33-2n_f)~, \end{displaymath} and $n_f$ is the number of flavors. Interestingly, for $n_f=3$ the power turns out to be $Q^{-3.8},$ nearly the same as the dimensional counting power $Q^{-4}$ in the single-scattering scenario. Higher order diagrams were calculated e.g. in Ref.~\cite{FSZ89}, however soon it became evident that even the first order QCD correction involves an immense number of Feynman diagrams, so further attempts to go beyond the simple quark counting rule were abandoned. It may be that perturbative QCD is not the relevant (or not the only physically interesting) expansion of the wide-angle scattering amplitude. Recent developments in M-branes (see e.g. Ref.~cite{polchinski}) may open new prospects in the realization of a hypothetical duality between small and large distances (or, equivalently, large- and small-angle scattering). The search for a relevant expansion parameter is of crucial importance on this way. In this paper we are solving an "inverse problem": we use the known explicit expression of the dual amplitude with Mandelstam analyticity (DAMA), that has correct wide angle scaling behavior. By identifying it with that resulting from the quark counting rules, we then calculate two sub-leading terms in the expansion of the known full dual amplitude and study the behavior of the resulting series. \section{ Wide angle behavior of the dual amplitude with Mandelstam analyticity} \label{sec1} Wide-angle scaling behavior within the $S-$matrix approach was discussed in Ref.~\cite{coon78}, where by means of a logarithmic Regge trajectory an interpolation from the ``soft`` Regge behavior to the ``hard`` scaling regime was suggested. The motivation of the logarithmic trajectory came from earlier papers \cite{alik75}, where a class of dual models requiring a logarithmic trajectory was suggested. The logarithmic asymptotic behavior of the trajectory and the large angle scaling behavior are uniquely related also in a different class of dual models, called dual amplitudes with Mandelstam analytics (DAMA) \cite{jenk79,bcj80}. The link between this class of models in the scaling limit and the parton model in the infinite momentum frame was studied in Ref.~\cite{sch73}. In all those papers only the leading asymptotic ($s,|t|\rightarrow \infty,\ s/t = const$) term was treated. The results of different approaches vary in such details as the form of the scaling violation (normally, logarithmic), the form of the angular dependence $f(\theta )$ and the way active quarks are counted. In this paper we calculate the sub-leading terms in the pre-asymptotic (larger $s$ and $|t|$) behavior of DAMA. Since the model is realistic enough in the sense that it satisfies the general requirements of the theory (see Refs.~\cite{jenk79,bcj80}), we believe that our result is universal and thus it may be used as a guide e. g. in QCD calculations. Apart from the leading term, we have explicitly calculated two more sub-leading terms. Our technique allows further calculations of still higher orders, but the obtained first three terms of the series already show a regular trend that may be interpreted as the expansion in the running coupling constant $g(s) \sim 1/\ln s,$ valid at large $s$ and $|t|$. This situation takes place for anyone trajectory with the logarithmic asymptotic. The aim of the present paper is two-fold. First, by identifying the leading term of the asymptotic (wide-angle) expansion of DAMA with that derived from perturbative QCD \cite{LB80} we tentatively assume that the DAMA in the wide angle asymptotic region is equivalent to the asymptotically free regime in QCD. With this identification in mind, we calculate within DAMA corrections to the leading term in the hope that their form may give some insight into the relevant corrections in perturbative QCD that are known to be very complicated. Clearly, the above identity has the chance to be true only in the vicinity of the wide angle region (small distances), where perturbative calculations are assumed to be still valid. The second aspect is purely phenomenological. Since, however, the experimental situation in the wide-angle region did not change for almost two decades, we are left with the earlier fits to the data. Let us now calculate the ``perturbative`` expansion of DAMA. We write the elastic scattering amplitude for spinless particles in the following symmetric form \cite{bcj80}: \begin{equation} A(s,t,u)=C(s-u)\lbrack D(s,t)-D(u,t)\rbrack~, \label{eq6} \end{equation} where $C$ is a constant and \begin{equation} D(s,t)=\int\limits_{0}^{1}dx\left( \frac{x}{g} \right) ^{-\alpha (s' )} \left( \frac{1-x}{g} \right) ^{-\alpha (t' )}~. \label{eq7} \end{equation} Here $s' =s(1-x),\ \ \ t' =tx$ and $g$ is a dimensionless parameter, $g>1$. Only one, leading trajectory was included and it was chosen in a simple, but representative form: \begin{equation} \alpha (s)=\alpha _{0} -\gamma \ln \left( \frac{1+\beta \sqrt{s_{0} -s}} {1+\beta \sqrt{s_{0} } } \right)~, \label{eq8} \end{equation} that account both for the threshold and the asymptotic behavior and is nearly linear for very small $|s|,|s|<<s_{0}\,.$ For simplicity we have included only the leading trajectories in both channels: the Pomeron trajectory in the $t$-channel and the exotic trajectory in the $s$-channel. While the parameters of the Pomeron trajectory are well known, only a little is known about the exotic trajectory. Fortunately, this has no substantial effect on our results, since our goal is the functional form of the series and its individual terms rather than fits to the data. Given the scarcity of the data and the freedom available in the model, the wide-angle behavior of DAMA cannot be determined completely. Let us consider the asymptotic behavior of Eq.~(\ref{eq7}) in the limit $s,\left| t\right| \rightarrow \infty ,\ s/t=const.$ For the Regge trajectories we have \begin{equation} \alpha(s)=\alpha (0)-\frac{\gamma }{2} \ln \delta ^{2}+ i\pi \frac{\gamma} {2}- \frac{\gamma }{2}\ln s=-a-\lambda~, \label{eq9} \end{equation} \begin{equation} \alpha (t)=\alpha (0)-\frac{\gamma }{2} \ln \delta ^{2}- \frac{\gamma } {2}\ln\left(-\frac{t}{s}\right)-\frac{\gamma }{2}\ln s= -b-\lambda~, \label{eq10} \end{equation} with \begin{displaymath} a=-\alpha (0)+\frac{\gamma }{2} \ln \delta ^{2} -i\pi \frac{\gamma }{2}~,~~~~ b=-\alpha (0)+\frac{\gamma }{2} \ln \delta ^{2} +\frac{\gamma }{2} \ln(-\frac{t}{s})~, \end{displaymath} \begin{equation} \delta=\frac{\beta \sqrt{s_{0}}}{1+\beta \sqrt{s_{0} } }~,~~~~~~ \lambda = \frac{\gamma }{2} \ln s~. \label{eq11} \end{equation} From here on, $s, t, u$ will be dimensionless variables, measured in units of $s_0$. In this domain the saddle point method can be used to calculate the integral in Eq.~(\ref{eq7}) \cite{mag97}. To do this we can rewrite Eq.~(\ref{eq7}) in the following form \begin{equation} D(s,t)=(2g)^{-a-b-2\lambda} g^{\gamma \ln 2} {1\over 2} \int\limits_{-1}^{1}g(u)~e^{\lambda f(u)} du~, \label{eq12} \end{equation} where we have changed the variable $x$ to $u$, $x=(1-u)/2$, and introduced new functions: \begin{equation} g(u)=(1-u)^{\tilde{a}}(1+u)^{\tilde{b}}e^{\gamma \ln{\frac{1-u}{2}} \ln{\frac{1+u}{2}}}~, \label{eq13} \end{equation} \begin{equation} f(u)=\ln (1-u^2) ~, \label{eq14} \end{equation} \begin{equation} \tilde{a}=a-\frac{\gamma}{2}\ln g ~,\ \ \ \ \ \tilde{b}=b-\frac{\gamma}{2}\ln g~. \label{eq15} \end{equation} We see now that $f(u)$ has a sharp maximum at the saddle point $u_0=0$. We quote the explicit expression for the saddle point expansion in the \ref{ap_coef}. Using formulas from this Appendix we obtain the power series for $D(s,t)$ in \ref{powerD}. It reads \begin{equation} D(s,t)\approx \frac{A_1 s^{-\gamma \ln 2g}}{\sqrt{\gamma \ln s}} \left(-\frac{t}{s}\right)^{-\frac{\gamma }{2} \ln 2g}\left\{1+ \frac{h_1(\tilde{a},\tilde{b})}{\gamma \ln s}+\frac{h_2(\tilde{a},\tilde{b})}{(\gamma \ln s)^2}\right\}~, \label{eq16} \end{equation} where $A_1,\ h_1, \ h_2$ are given by the expressions (\ref{b5}, \ref{b8}, \ref{b9}). The expression for $D(u,t)$ can be calculated in a similar way (see Eq.~(\ref{b10}) in \ref{powerD}). In the kinematical region $s, \left| t\right| \rightarrow \infty,\ t/s=const$ we can use the substitutions \begin{equation} t\approx -s\cdot \sin ^{2} \left( \theta/2 \right)~,~~~~~~ u\approx -s\cdot \cos ^{2} \left( \theta/2 \right)~. \label{eq17} \end{equation} Substituting the results for $D(s,t)$ and $D(u,t)$ into Eq.~(\ref{eq6}) and changing the variables we get the expression for the full amplitude as a function of the $s$ and $\theta$ variables (see Eq.~(\ref{b14}) in \ref{powerD}): \begin{equation} A(s,\theta)\approx \frac{C A s^{-N}}{\sqrt{\gamma \ln s}} f(\theta) I(s,\theta )~, \label{eq18} \end{equation} where $A,\ N,\ f(\theta),\ I(s,\theta )$ are given by the expressions (\ref{b13}, \ref{b15} - \ref{b17}). To summarize, we have expanded the wide-angle scattering amplitude in a power series of $1/\ln{s}$ and have evaluated explicitly the coefficients of the first two terms (beyond the leading one). \section{Comparison with the data and discussion of the results} \label{sec2} New experimental data on wide-angle scatterings are not likely to appear any more because of the simple reason that as energy increases more particles tend to fly in the forward direction and there is no chance to detect e.g. the proton-proton differential cross section at $90^{\circ}$ for, say, $\sqrt s> 10\ GeV$. ``Wide angles'', of course, extend beyond $90^{\circ}$. Still the complication due to the huge number of Born diagrams contributing to large angle exclusive reactions \cite{LB80}, overwhelming the contribution due to the Landshoff pinch singularity \cite{Lan74}, will remain for long topical in this field. We use the data given in the compilation of \cite{LP73} to fix the scale. The errors, quoted in the original papers (see Ref.~\cite{LP73} and references therein), are typically about 10 overall normalization factor, the "quark counting power" in the cross section being set equal to $N=4$ in the case of proton-proton cross section, in agreement with the data \cite{LP73, LB80} (see Fig.~\ref{fig1}). \begin{figure}[htb] \centerline{\psfig{figure={pryamye_0.eps},height=11.0cm}} \caption{\footnotesize Cross section $\frac{d\sigma }{dt}$ for {\bf pp} $\rightarrow$ {\bf pp} scattering at various center of mass scattering angles. Both axes are in logarithmic scale. Stars denote the experimental points from Ref.~\cite{LP73}. The straight lines correspond to a falloff of $\sim 1/s^{10}$. They are calculated according to the power series for the scattering amplitude, discussed above ($\frac {d\sigma}{dt}= \frac{4\pi}{(s s_0)^2}|A(s,\theta)|^2$), with the following set of parameters: $\alpha_0 = 1$, $N=4,\ \gamma = 2.84 \ (g=2.9),$ $\beta=0.05 \ GeV^{-1}$, $C=2.7\cdot10^{-14} GeV^{-2}$ and $s_0 = 4m_\pi^2$. } \label{fig1 \end{figure} Our main goal is the behavior of the scaling-violating corrections to the leading term obeying quark counting rules. Fig. \ref{fig2} shows the relative contribution of these terms. We draw the correction power series: \begin{displaymath} J(s,\theta)=|I(s,\theta)|^2 \approx 1+ 2\frac{Re(f_1(\theta)/Z(\theta))} {\gamma \ln s}+ 2\frac{Re(f_2(\theta)/Z(\theta))}{(\gamma \ln s)^2}+\frac{|f_1(\theta) /Z(\theta)|^2}{(\gamma \ln s)^2} \end{displaymath} \begin{equation} +O({1\over \lambda^{3}})~, \label{eq19} \end{equation} where $f_1(\theta),~f_2(\theta),~Z(\theta)$ are given by expressions (\ref{b18} - \ref{b20}). We can see that the corrections are quite large for small $s$, especially for angles close to $90^0$. That is not a surprise, since the lowest order of our expansion is valid for large $s$ ($\gamma \ln s /2 >> 1$). In the experimental energy interval the corrections give factor $4-6$ to the cross sections and should not be neglected. This was missed in the references \cite{jenk79, bcj80}. Moreover we find that the corrections are very sensitive to variations of $\beta$ and $\gamma$. \begin{figure}[htb] \centerline{\psfig{figure={popravki_0.eps},height=11.0cm}} \caption{\footnotesize The corrections $J(s,\theta)$, given by Eq.~(\ref{eq19}), to the differential cross section $\frac{d\sigma }{dt}$ for {\bf pp} $\rightarrow$ {\bf pp} scattering . We have used the same values of parameters as in Fig.~\ref{fig1}: $\alpha_0 = 1$, $N=4,\ \gamma = 2.84\ (g=2.9),$ $\beta=0.05 \ GeV^{-1}$ and $s_0 = 4m_\pi^2$, coming from the comparison with the data.} \label{fig2} \end{figure} \vskip 1cm {\bf Acknowledgment} V.K. Magas is thankful for the hospitality extended to him by the Bogolyubov Institute for Theoretical Physics in Kiev where part of this work was done.
\section*{Introduction} Every representation $\alpha:G\to\Aut B({\mathcal H})$ of a discrete group $G$ as automorphisms of the algebra $B({\mathcal H})$ of bounded linear operators on a complex Hilbert space ${\mathcal H}$ determines a unique element of $H^2(G,\field{T})$: each $\alpha_s$ is of the form $\Ad U_s$ for some unitary operator $U_s$, and $U_sU_t = \omega(s,t)U_{st}$ determines a $2$\nobreakdash- cocycle $\omega$ whose cohomology class is independent of the choice of implementing unitaries. If instead one starts with an action of a semigroup $P$ as endomorphisms of $B({\mathcal H})$, then, as described in \cite[Remark~2.3]{fowrae}, the cohomological obstruction one encounters is a {\em product system\/} over $P$: roughly speaking, a collection $E = \{E_s:s\in P\}$ of complex Hilbert spaces together with an associative multiplication which implements unitary isomorphisms $E_s\otimes E_t \to E_{st}$. Product systems were first defined by Arveson \cite{arv} in his study of one-parameter semigroups of endomorphisms of $B({\mathcal H})$. In this note we continue the investigations of \cite{dinhjfa, dinhjot, laca, fowrae, fowler} into the $C^*$\nobreakdash- algebras associated with discrete product systems. Each of the algebras studied in these previous papers can be regarded as a Toeplitz algebra: unit vectors of the product system correspond to isometries in the universal $C^*$\nobreakdash- algebra, and finite orthonormal sets of vectors from a fiber $E_s$ map to isometries whose range projections are orthogonal and whose sum is strictly less than the identity. Here we consider only product systems $E$ whose fibers are finite-dimensional, and in our universal $C^*$\nobreakdash- algebra ${\mathcal O}_E$ the range projections of the isometries associated with an orthonormal basis for any fiber sum to the identity. Hence ${\mathcal O}_E$ is generated by a collection of Cuntz algebras, one for each element of the underlying semigroup $P$; the relations between these Cuntz algebras are determined by the multiplication in $E$. We think of ${\mathcal O}_E$ as a semigroup version of the twisted group algebra $C^*(G,\omega)$. We begin in Section~\ref{section:OE} by establishing the existence of ${\mathcal O}_E$ and discussing a few examples. In particular one can recover as ${\mathcal O}_E$ the Cuntz algebras (Remark~\ref{remark:cuntz algebra}) and the irrational rotation algebras (Remark~\ref{remark:irrational}). In Section~\ref{section:abelian} we develop some basic results regarding the structure of ${\mathcal O}_E$ when the underlying semigroup $P$ embeds in an abelian group; this is a key hypothesis in most of our results. In Section~\ref{section:simplicity} we prove our main result, Theorem~\ref{theorem:main}: if no two fibers of $E$ have the same dimension, then ${\mathcal O}_E$ is simple and purely infinite. For the lexicographic product systems of \cite{fowrae} our Theorem~\ref{theorem:lexicographic} gives a sharper simplicity result: ${\mathcal O}_E$ is simple if and only if the dimension function is injective. This shows in particular that ${\mathcal O}_E$ need not be simple even if the only one-dimensional fiber of $E$ is the fiber over the identity of $P$. For product systems over $\field{N}^2$ we show in Theorem~\ref{theorem:N2} that ${\mathcal O}_E$ is either simple or isomorphic to ${\mathcal O}_l\otimes C(\field{T})$, where $l$ is determined by the dimensions of the fibers over $(1,0)$ and $(0,1)$. Finally, in Theorem~\ref{theorem:nuclearity} we generalize a technique of Laca \cite{laca} to show that ${\mathcal O}_E$ is nuclear if $E$ admits a twisted unit. \section{The Cuntz algebra of a product system}\label{section:OE} Suppose $P$ is a countable discrete semigroup with identity $e$. A {\em product system\/} over $P$ is a family $p:E\to P$ of nontrivial separable complex Hilbert spaces $E_t:= p^{-1}(t)$ which is endowed with an associative multiplication $E\times E\to E$ in such a way that $p$ is a semigroup homomorphism, and such that for every $s,t\in P$ the map $x\otimes y\in E_s\otimes E_t \mapsto xy\in E_{st}$ extends to a unitary isomorphism. We also insist that $\dim E_e = 1$, so that $E$ has an identity $1\in E_e$ \cite[Lemma~1.3]{fowrae}. A {\em representation\/} of $E$ in a $C^*$\nobreakdash- algebra $B$ is a map $\phi:E\to B$ which satisfies \begin{itemize} \item[(i)] $\phi(xy) = \phi(x)\phi(y)$ for every $x,y\in E$; and \item[(ii)] $\phi(y)^*\phi(x) = (x\mid y) \phi(1)$ if $p(x) = p(y)$. \end{itemize} This definition is slightly different than that found in \cite{fowrae}, where representations were on Hilbert space and condition (ii) was that $\phi(y)^*\phi(x) = (x\mid y) 1$ if $p(x) = p(y)$. The main advantage to our definition is that it allows us to consider the trivial map $E\to\{0\}$ as a representation, without regarding $\{0\}$ as a unital $C^*$\nobreakdash- algebra. If $\phi \ne 0$, then $\phi(1)$ is a nontrivial projection which serves as an identity for the $C^*$\nobreakdash- subalgebra $C^*(\phi)\subseteq B$ generated by $\phi(E)$; moreover, $\phi$ restricts to an isometric linear map on each of the fibers of $E$ \cite[p.8]{arv}. For nontrivial $\phi$, condition (ii) implies that each $\phi(x)$ is a multiple of an isometry in $C^*(\phi)$; indeed, if ${\mathcal B}_t$ is an orthonormal basis for $E_t$, then $\{\phi(f): f\in {\mathcal B}_t\}$ is a family of isometries with mutually orthogonal range projections. In this note we will consider only product systems whose fibers are finite-dimensional, and we are primarily interested in representations $\phi:E\to B$ which satisfy \begin{equation}\label{eq:cuntz rep} \sum_{f\in {\mathcal B}_t} \phi(f)\phi(f)^* = \phi(1) \qquad\text{for all $t\in P$.} \end{equation} We call a representation $\phi$ which satisfies \eqref{eq:cuntz rep} a {\em Cuntz representation\/}. While nontrivial representations of $E$ always exist \cite[Lemma~1.10]{fowrae}, it is not clear that every product system admits a nontrivial Cuntz representation. \begin{prop}\label{prop:OE} Let $E$ be a product system over $P$ of finite-dimensional Hilbert spaces. There is a pair $({\mathcal O}_E,i_E)$ consisting of a $C^*$\nobreakdash- algebra ${\mathcal O}_E$ and a Cuntz representation $i_E:E\to{\mathcal O}_E$ with the following properties: \textup{(a)} for every Cuntz representation $\phi$ of $E$, there is a homomorphism $\phi_*$ of ${\mathcal O}_E$, called the {\em integrated form of $\phi$\/}, such that $\phi_*\circ i_E = \phi$; and \textup{(b)} ${\mathcal O}_E$ is generated as a $C^*$\nobreakdash- algebra by $i_E(E)$. \noindent The pair $({\mathcal O}_E,i_E)$ is unique up to canonical isomorphism. \end{prop} \begin{remark}\label{remark:trivial} We emphasize that ${\mathcal O}_E$ is trivial if $E$ does not admit a nontrivial Cuntz representation. In all other cases ${\mathcal O}_E$ is unital and $i_E$ is isometric. \end{remark} \begin{proof}[Proof of Proposition~\ref{prop:OE}] The Proposition can be proved by modifying the standard argument of, for example, \cite[Proposition~1.3]{fowrae2}. Briefly, let $S$ be a collection of Cuntz representations on Hilbert space which are cyclic (i.e. generate a $C^*$\nobreakdash- algebra which admits a cyclic vector), and such that every cyclic Cuntz representation is unitarily equivalent to an element of $S$. Define $i_E := \bigoplus_{\phi\in S} \phi$ and ${\mathcal O}_E := C^*(i_E)$. Condition (a) holds because every Cuntz representation of $E$ on a Hilbert space decomposes as the direct sum of a zero representation and a collection of cyclic representations, and the uniqueness assertion follows from a standard argument. \end{proof} \begin{example}\label{example:lexicographic} {\em (Lexicographic product systems.)\/} For any product system $E$ whose fibers are finite-dimensional, $s \mapsto \dim E_s$ is a semigroup homomorphism from $P$ to the multiplicative positive integers $\field{N}^*$. We call this homomorphism the {\em dimension function\/} of $E$. Let $d:P\to\field{N}^*$ be an arbitrary homomorphism. In \cite[Examples~1.4(E2)]{fowrae} it was shown how one can construct a product system $E(d)$, called the {\em lexicographic product system determined by $d$\/}, whose dimension function is $d$: for each $n$ let $\{\delta_0,\dots,\delta_{n-1}\}$ be the canonical basis for $\field{C}^n$, take \[ E(d) := \bigsqcup_{s\in P} \{s\}\times \field{C}^{d(s)}, \] and define multiplication on basis vectors using the lexicographic order on $\{0,\dots, d(r) - 1\} \times \{0, \dots, d(s) - 1\}$; that is, \[ (r,\delta_j)(s,\delta_k) := (rs,\delta_{jd(s) + k}) \qquad\text{for $0 \le j \le d(r) - 1$ and $0 \le k \le d(s) - 1$.} \] Multiplication in $E(d)$ is defined by extending bilinearly. There is a distinguished Cuntz representation of $E(d)$ on $L^2(\field{T})$. Suppose $r\in P$ and $0 \le j \le d(r) - 1$. Let $S(r,\delta_j)$ be the operator on $L^2(\field{T})$ whose value on a vector $\xi\in L^2(\field{T})$ is given by \[ S(r,\delta_j)\xi(e^{2\pi it}) := \begin{cases} d(r)^{1/2} \xi(e^{2\pi i(td(r) - j)}) & \text{if $t\in \bigl[ \frac j{d(r)}, \frac{j+1}{d(r)} \bigr)$} \\ 0 & \text{otherwise.} \end{cases} \] It is easy to see that $S(r,\delta_j)$ is an isometry, and that \[ \sum_{j=0}^{d(r) - 1} S(r,\delta_j)S(r,\delta_j)^* = 1 = S(1). \] Since \begin{align*} & S(r,\delta_j)S(s,\delta_k)\xi(e^{2\pi it}) \\ & \qquad = \begin{cases} d(r)^{1/2} S(s,\delta_k)\xi(e^{2\pi i(td(r) - j)}) & \text{if $t\in \bigl[ \frac j{d(r)}, \frac{j+1}{d(r)} \bigr)$} \\ 0 & \text{otherwise} \end{cases} \\ & \qquad = \begin{cases} d(r)^{1/2} d(s)^{1/2} \xi(e^{2\pi i((td(r) - j)d(s) - k)}) & \text{if $td(r) - j\in \bigl[ \frac k{d(s)}, \frac{k+1}{d(s)} \bigr)$} \\ 0 & \text{otherwise} \end{cases} \\ & \qquad = \begin{cases} d(rs)^{1/2} \xi(e^{2\pi i(td(rs) - (jd(s) + k))}) & \text{if $t \in \bigl[ \frac {jd(s) + k}{d(rs)}, \frac{jd(s) + k + 1}{d(rs)} \bigr)$} \\ 0 & \text{otherwise} \end{cases} \\ & \qquad = S(rs,\delta_{jd(s)+k})\xi(e^{2\pi it}), \end{align*} we have $S(r,\delta_j)S(s,\delta_k) = S(rs,\delta_{jd(s)+k}) = S((r,\delta_j)(s,\delta_k))$. Hence defining \[ S(r,x) := \sum_{j=0}^{d(r)-1} (x \mid \delta_j) S(r,\delta_j) \qquad\text{for $(r,x)\in E(d)$} \] gives a Cuntz representation $S:E(d)\to B(L^2(\field{T}))$. \end{example} \begin{remark}\label{remark:cuntz algebra} Fix $n \ge 2$ and let $d$ be the homomorphism $a\in\field{N} \mapsto n^a\in\field{N}^*$. Then ${\mathcal O}_{E(d)}$ is the Cuntz algebra ${\mathcal O}_n$. \end{remark} \begin{example} {\em (One-dimensional product systems.)\/} Suppose $\omega$ is a multiplier of $P$; that is, a function $\omega:P\times P\to\field{T}$ such that $\omega(e,e) = 1$ and \[ \omega(r,s)\omega(rs,t) = \omega(r,st)\omega(s,t) \qquad\text{for all $r,s,t\in P$.} \] Then $(r,z)(s,w) := (rs,\omega(r,s)zw)$ for $r,s\in P$ and $z,w\in\field{C}$ defines an associative multiplication which gives $P\times\field{C}$ the structure of a product system; we write $(P\times\field{C})^\omega$ for this product system. It is easy to see that every product system whose fibers are one-dimensional is isomorphic to $(P\times\field{C})^\omega$ for some multiplier $\omega$. Suppose $P$ is an Ore semigroup; that is, suppose $P$ is cancellative and satisfies $Pr\cap Ps \ne\emptyset$ for every pair $r,s\in P$. (For example, every cancellative commutative semigroup has this property.) By \cite[Theorem~1.1.2]{laca2}, $P$ can be embedded in a group $G$ with $P^{-1}P = G$, and by \cite[Theorem~1.2.2]{laca2} there is a multiplier $\omega'$ of $G$ which extends $\omega$. Define $l^\omega:(P\times\field{C})^\omega\to B(\ell^2(G))$ by \[ l^\omega(r,z)\xi(s) := z\omega'(r,r^{-1}s)\xi(r^{-1}s) \qquad\text{for $\xi\in\ell^2(G)$ and $s\in G$.} \] It is routine to check that each $l^\omega(r,1)$ is unitary and that $l^\omega$ is multiplicative. Hence $l^\omega$ is a Cuntz representation of $(P\times\field{C})^\omega$. \end{example} \begin{remark}\label{remark:irrational} Let $\theta\in(0,1)$ be irrational, let $\omega:\field{N}^2\times\field{N}^2\to\field{T}$ be the multiplier $\omega((a,b),(c,d)) := e^{2\pi i\theta bc}$, and let $E = (\field{N}^2\times\field{C})^\omega$. Then $U := i_E((0,1),1)$ and $V := i_E((1,0),1)$ are unitaries which generate ${\mathcal O}_E$ and satisfy $UV = e^{2\pi i\theta} VU$, so ${\mathcal O}_E$ is the irrational rotatation algebra ${\mathcal A}_\theta$. \end{remark} \begin{example} {\em (Twisting.)\/} If $E$ is a product system over $P$ and $\omega$ is a multiplier of $P$, then $(x,y)\in E\times E \mapsto \omega(p(x),p(y))xy$ defines a multiplication on $E$ which also gives $E$ the structure of a product system; we write $E^\omega$ for this new system, and say $E$ has been {\em twisted by $\omega$\/}. If $\phi$ is a Cuntz representation of $E$ on a Hilbert space ${\mathcal H}$, then \[ \phi^\omega(x) := \phi(x)\otimes l^\omega(p(x),1) \qquad\text{for $x\in E^\omega$} \] defines a Cuntz representation $\phi^\omega$ of $E^\omega$ on ${\mathcal H}\otimes\ell^2(G)$. \end{example} The results obtained in the previous examples allow us to state: \begin{prop}\label{prop:nontrivial} Every twisted lexicographic product system over an Ore semigroup admits a nontrivial Cuntz representation. \end{prop} \section{Product systems over abelian semigroups} \label{section:abelian} The following Proposition collects some results concerning the structure of ${\mathcal O}_E$ when $P$ embeds in an abelian group. \begin{prop}\label{prop:abelian} Suppose $P$ is a subsemigroup of a countable abelian group $G$ and $E$ is a product system over $P$ of finite-dimensional Hilbert spaces. \textup{(1)} Let $\phi$ be a Cuntz representation of $E$, let $s,t\in P$, and let ${\mathcal B}_s$ and ${\mathcal B}_t$ be orthonormal bases for $E_s$ and $E_t$, respectively. Then for any $x'\in E_s$ and $y'\in E_t$ we have \[ \phi(y')^*\phi(x') = \sum_{x\in {\mathcal B}_s} \sum_{y\in {\mathcal B}_t} (x'y \mid y'x) \phi(x)\phi(y)^*. \] \textup{(2)} ${\mathcal O}_E = \clsp\{i_E(x)i_E(y)^*: x,y\in E\}$. \textup{(3)} There is a strongly continuous action $\gamma:\widehat G\to\Aut{\mathcal O}_E$, called the {\em gauge action\/}, such that \[ \gamma_\lambda(i_E(x)) = \lambda(p(x))i_E(x) \qquad\text{for all $\lambda\in\widehat G$ and $x\in E$.} \] \textup{(4)} Let $m$ be Haar measure on $\widehat G$. Then \[ \Phi(b) := \int_{\widehat G} \gamma_\lambda(b)\,dm(\lambda) \qquad\text{for $b\in{\mathcal O}_E$} \] defines a faithful conditional expectation $\Phi$ of ${\mathcal O}_E$ onto ${\mathcal O}_E^\gamma$, the fixed-point algebra of the gauge action. \textup{(5)} ${\mathcal O}_E^\gamma = \clsp\{i_E(x)i_E(y)^*: x,y\in E,\ p(x) = p(y)\}$. \end{prop} \begin{proof} (1) We use \eqref{eq:cuntz rep} to calculate \begin{align*} \phi(y')^*\phi(x') & = \Bigl( \sum_{x\in {\mathcal B}_s} \phi(x)\phi(x)^* \Bigr) \phi(y')^*\phi(x') \Bigl( \sum_{y\in {\mathcal B}_t} \phi(y)\phi(y)^* \Bigr) \\ & = \sum_{x\in {\mathcal B}_s} \sum_{y\in {\mathcal B}_t} \phi(x)\phi(y'x)^*\phi(x'y)\phi(y)^* \\ & = \sum_{x\in {\mathcal B}_s} \sum_{y\in {\mathcal B}_t} (x'y \mid y'x)\phi(x)\phi(y)^*, \end{align*} noting that $p(y'x) = p(x'y)$ since $P$ is abelian. (2) Take $\phi = i_E$ in (1) to see that the linear span of monomials of the form $i_E(x)i_E(y)^*$ is closed under multiplication, and hence a ${}^*$\nobreakdash- algebra. (3) For each $\lambda\in\widehat G$ the map $x\in E \mapsto \lambda(p(x))i_E(x)$ is a Cuntz representation, and hence integrates to an endomorphism $\gamma_\lambda$ of ${\mathcal O}_E$. Since $\gamma_\lambda \circ \gamma_{\lambda^{-1}}$ and $\gamma_{\lambda^{-1}} \circ \gamma_\lambda$ are both the identity on ${\mathcal O}_E$, $\gamma_\lambda$ is an automorphism. Obviously $\gamma$ is a group homomorphism, and its continuity follows from a straightforward $\epsilon/3$ argument. (4) This is a standard result about automorphic actions of discrete abelian groups. (5) Since $\Phi(i_E(x)i_E(y)^*) = \delta_{p(x),p(y)}i_E(x)i_E(y)^*$, (5) follows from (2) and the continuity of $\Phi$. \end{proof} Our next goal is to give an abstract characterization of the fixed-point algebra ${\mathcal O}_E^\gamma$. First some notation. If $s,t\in P$, $S\in{\mathcal K}(E_s)$ and $T\in{\mathcal K}(E_t)$, write $S\otimes T$ for the operator on $E_{st}$ which satisfies \[ S\otimes T(xy) = (Sx)(Ty) \qquad\text{for $x\in E_s$ and $y\in E_t$.} \] Write $1^t$ for the identity operator on $E_t$. Define a relation $\preceq$ on $P$ by $s \preceq t$ if and only if $t\in sP$, and observe that $\preceq$ is a preorder on $P$; that is, it is reflexive and transitive. Since $P$ is commutative, for every pair $s,t\in P$ we have $st \in sP\cap tP$, and hence $(P,\preceq)$ is upwardly directed. Now $S\mapsto S\otimes 1^t$ is a unital embedding of ${\mathcal K}(E_s)$ in ${\mathcal K}(E_{st})$, and since multiplication in $E$ is associative we have $S\otimes 1^{tr} = (S\otimes 1^t)\otimes 1^r$. Hence $S\in{\mathcal K}(E_s) \mapsto S\otimes 1^t\in{\mathcal K}(E_{st})$ is a directed system of $C^*$\nobreakdash- algebras, and we can define \[ {\mathcal F}_E := \varinjlim {\mathcal K}(E_s). \] Since each ${\mathcal K}(E_s)$ is simple, so is ${\mathcal F}_E$. (It is not hard to see that ${\mathcal F}_E$ is a UHF algebra.) Let $\iota_s$ be the canonical embedding of ${\mathcal K}(E_s)$ in ${\mathcal F}_E$. \begin{prop}\label{prop:FE} There is a unique homomorphism $i_{\mathcal F}:{\mathcal F}_E\to{\mathcal O}_E$ which satisfies \[ i_{\mathcal F}(\iota_{p(x)}(\rankone xy)) = i_E(x)i_E(y)^* \qquad\text{whenever $p(x) = p(y)$,} \] where $\rankone xy$ is the rank-one operator $z\in E_{p(x)} \mapsto (z\mid y)x$. The image of ${\mathcal F}_E$ is precisely ${\mathcal O}_E^\gamma$. If $E$ admits a nontrivial Cuntz representation, then $i_{\mathcal F}$ is injective. \end{prop} \begin{proof} For each $s\in P$, there is a unique homomorphism $\sigma_s:{\mathcal K}(E_s)\to{\mathcal O}_E$ such that $\sigma_s(\rankone xy) = i_E(x)i_E(y)^*$ for all $x,y\in E_s$. Let ${\mathcal B}_t$ be an orthonormal basis for $E_t$. Since \begin{align*} \sigma_{st}((\rankone xy)\otimes 1^t) & = \sigma_{st}\Bigl( \sum_{f\in {\mathcal B}_t} (\rankone xy) \otimes (\rankone ff) \Bigr) = \sigma_{st}\Bigl( \sum_{f\in {\mathcal B}_t} (\rankone {xf}{yf} \Bigr) \\ & = \sum_{f\in {\mathcal B}_t} i_E(xf)i_E(yf)^* = i_E(x)\Bigl( \sum_{f\in {\mathcal B}_t} i_E(f)i_E(f)^* \Bigr) i_E(y)^* \\ & = i_E(x)i_E(y)^* = \sigma_s(\rankone xy), \end{align*} we deduce that $\sigma_{st}(S\otimes 1^t) = \sigma_s(S)$ for all $S\in{\mathcal K}(E_s)$, and hence there is a homomorphism $i_{\mathcal F}:{\mathcal F}_E\to{\mathcal O}_E$ such that $i_{\mathcal F}\circ\iota_s = \sigma_s$ for every $s\in P$. From Proposition~\ref{prop:abelian}(5) it is obvious that $i_{\mathcal F}$ maps ${\mathcal F}_E$ onto ${\mathcal O}_E^\gamma$. If $E$ admits a nontrivial Cuntz representation then $i_E$ is nonzero (see Remark~\ref{remark:trivial}), hence each $\sigma_s$ is nonzero, and finally $i_{\mathcal F}$ is nonzero. Since ${\mathcal F}_E$ is simple, we deduce that $i_{\mathcal F}$ is injective. \end{proof} \section{Simplicity and pure infiniteness} \label{section:simplicity} \begin{theorem}\label{theorem:main} Let $G$ be a countable abelian group, let $P$ be a subsemigroup of $G$ which contains the identity, and let $E$ be a product system over $P$ of finite-dimensional Hilbert spaces which admits a nontrivial Cuntz representation. If the dimension function $s \mapsto \dim E_s$ is injective, then ${\mathcal O}_E$ is simple and purely infinite. \end{theorem} To prove the Theorem we require a technical lemma. For the Proposition which precedes it, see \cite[Proposition~2.7]{arv} and \cite[Proposition~1.11 and Lemma~3.6]{fowrae}. \begin{prop}\label{prop:alpha} Let $E$ be a product system over $P$ of finite-dimensional Hilbert spaces, and let $\phi$ be a Cuntz representation of $E$ in a unital $C^*$\nobreakdash- algebra $B$ such that $\phi(1) = 1$. For each $s\in P$ there is a unital endomorphism $\alpha^\phi_s$ of $B$ which satisfies \[ \alpha^\phi_s(b) = \sum_{f\in {\mathcal B}_s} \phi(f)b\phi(f)^* \qquad\text{for all $b\in B$} \] whenever ${\mathcal B}_s$ is an orthonormal basis for $E_s$. Moreover, $\alpha_e$ is the identity, and \[ \alpha^\phi_{st}(b)\phi(x) = \phi(x)\alpha^\phi_t(b) \qquad\text{for all $x\in E_s$.} \] \end{prop} \begin{lemma}\label{lemma:kill} Let $E$ be a product system satisfying the hypotheses of Theorem~\ref{theorem:main}, and let $\phi$ be a Cuntz representation of $E$ in a unital $C^*$\nobreakdash- algebra $B$ such that $\phi(1) = 1$. Suppose $x_1$, \dots, $x_n$, $y_1$, \dots, $y_n\in E$ satisfy $p(x_i) \ne p(y_i)$ for $1 \le i \le n$. Let $c\in P$ be such that $p(x_i)^{-1}c \in P$ and $p(y_i)^{-1}c\in P$ for every $i$. Then there is a unit vector $w\in E$ such that \begin{equation}\label{eq:w} \alpha^\phi_c(\phi(w)\phi(w)^*)\phi(x_i)\phi(y_i)^* \alpha^\phi_c(\phi(w)\phi(w)^*) = 0 \qquad\text{for $1 \le i \le n$.} \end{equation} \end{lemma} \begin{proof} Define $s_i := p(x_i)^{-1}c$, $t_i := p(y_i)^{-1}c$, and for each $r\in P$ let ${\mathcal B}_r$ be an orthonormal basis for $E_r$. Suppose $Q\in \phi_*({\mathcal O}_E)$; we will eventually take $Q = \phi(w)\phi(w)^*$. By Proposition~\ref{prop:alpha} we have \begin{multline*} \alpha^\phi_c(Q)\phi(x_i)\phi(y_i)^*\alpha^\phi_c(Q) = \phi(x_i)\alpha^\phi_{s_i}(Q)\alpha^\phi_{t_i}(Q)\phi(y_i)^* \\ = \sum_{f'\in {\mathcal B}_{s_i}} \sum_{g'\in {\mathcal B}_{t_i}} \phi(x_i)\phi(f')Q\phi(f')^*\phi(g')Q\phi(g')^*\phi(y_i)^*, \end{multline*} and by Proposition~\ref{prop:abelian}(1) \[ Q\phi(f')^*\phi(g')Q = \sum_{g\in {\mathcal B}_{t_i}} \sum_{f\in {\mathcal B}_{s_i}} (g'f \mid f'g) Q\phi(g)\phi(f)^*Q, \] so it suffices to find a unit vector $w\in E$ such that \begin{equation}\label{eq:sufficient} \phi(w)^*\phi(g)\phi(f)^*\phi(w) = 0 \qquad\text{whenever $(f,g)\in {\mathcal B}_{s_i}\times {\mathcal B}_{t_i}$ for some $i$.} \end{equation} Let $(f_l, g_l)_{l=1}^m$ be an enumeration of these pairs. We claim that for each $j\in \{0, \dots, m\}$ there is a unit vector $v_j \in E$ such that \begin{equation}\label{eq:v} \phi(v_j)^*\phi(g_l)\phi(f_l)^*\phi(v_j) = 0 \qquad\text{for $1 \le l \le j$.} \end{equation} Given the claim, $w := v_m$ satisfies \eqref{eq:sufficient}, and hence \eqref{eq:w}, completing the proof. The claim is vacuous when $j = 0$: taking any unit vector $v_0\in E$ works. Suppose inductively that there exists $k \le m-1$ such that \eqref{eq:v} holds when $j = k$. Let $r := p(v_k)$, $s := p(f_{k+1})$, and $t := p(g_{k+1})$. Since $s \ne t$, by hypothesis $\dim E_s \ne \dim E_t$, and we can assume without loss of generality that $\dim E_s < \dim E_t$. We have \begin{align*} & \phi(v_k)^*\phi(g_{k+1})\phi(f_{k+1})^*\phi(v_k) \\ & \quad = \sum_{f\in {\mathcal B}_s} \sum_{g\in {\mathcal B}_t} \sum_{v\in {\mathcal B}_r} \phi(g)\phi(g)^*\phi(v_k)^*\phi(g_{k+1}) \phi(v)\phi(v)^*\phi(f_{k+1})^*\phi(v_k) \phi(f)\phi(f)^* \\ & \quad = \sum_{f\in {\mathcal B}_s} \sum_{g\in {\mathcal B}_t} \sum_{v\in {\mathcal B}_r} (g_{k+1}v \mid v_kg)(v_kf \mid f_{k+1}v) \phi(g)\phi(f)^* \\ & \quad = \sum_{f\in {\mathcal B}_s} \phi(u_f)\phi(f)^*, \end{align*} where \[ u_f := \sum_{g\in {\mathcal B}_t} \sum_{v\in {\mathcal B}_r} (g_{k+1}v \mid v_kg)(v_kf \mid f_{k+1}v) g \in E_t. \] Since $\dim E_s < \dim E_t$, there is a unit vector $v'\in E_t$ which is orthogonal to each $u_f$, and taking $v_{k+1} := v_kv'$ gives \eqref{eq:v} for $j = k+1$. \end{proof} \begin{proof}[Proof of Theorem~\ref{theorem:main}] Suppose $B$ is a $C^*$\nobreakdash- algebra and $\pi:{\mathcal O}_E\to B$ is a nonzero homomorphism. We will show that $\pi$ is injective, thus establishing the simplicity of ${\mathcal O}_E$. It does not harm to assume that $\pi$ is surjective, so that $B$ is unital and $\pi(1) = 1$. Recall from Proposition~\ref{prop:abelian}(4) that there is a faithful expectation $\Phi$ of ${\mathcal O}_E$ onto ${\mathcal O}_E^\gamma$, given by averaging over the orbits of the gauge action. We will show that there is an expectation $\Phi_\pi$ of $\pi({\mathcal O}_E)$ onto $\pi({\mathcal O}_E^\gamma)$ such that \begin{equation}\label{eq:spatial Phi} \Phi_\pi\circ\pi = \pi\circ\Phi. \end{equation} To see that this implies that $\pi$ is injective, suppose $b\in\ker\pi$. Then $\pi\circ\Phi(b^*b) = \Phi_\pi\circ\pi(b^*b) = 0$, so $\Phi(b^*b)\in {\mathcal O}_E^\gamma\cap\ker\pi$. But ${\mathcal O}_E^\gamma$ is simple and contains the identity of ${\mathcal O}_E$, so ${\mathcal O}_E^\gamma\cap\ker\pi = \{0\}$. Thus $\Phi(b^*b) = 0$, and we deduce from the faithfulness of $\Phi$ that $b = 0$. By Proposition~\ref{prop:abelian}(2), finite sums of the form $b = \sum i_E(x_i)i_E(y_i)^*$ are dense in ${\mathcal O}_E$. Hence to prove the existence of an expectation $\Phi_\pi$ satisfying \eqref{eq:spatial Phi}, it suffices to fix such an element $b$ and show that $\lVert \pi(b) \rVert \ge \lVert \pi(\Phi(b)) \rVert$. Thus with $\phi := \pi\circ i_E$, we must show that \begin{equation}\label{eq:inequality} \lVert \sum \phi(x_i)\phi(y_i)^* \rVert \ge \Bigl\lVert \sum_{p(x_i) = p(y_i)} \phi(x_i)\phi(y_i)^* \Bigr\rVert. \end{equation} Let $c := \prod_i p(x_i)p(y_i)$. Since $\phi$ is a Cuntz representation of $E$, the relations \eqref{eq:cuntz rep} allow us to assume that $p(x_i) = p(y_i) = c$ whenever $p(x_i) = p(y_i)$. Since $\phi(1) = 1$, Lemma~\ref{lemma:kill} applies and provides a unit vector $w\in E$ such that \[ \alpha^\phi_c(\phi(w)\phi(w)^*) \phi(x_i)\phi(y_i) \alpha^\phi_c(\phi(w)\phi(w)^*) = 0 \qquad\text{whenever $p(x_i) \ne p(y_i)$.} \] Let $Q := \phi(w)\phi(w)^*$. Then \begin{equation}\label{eq:Q inequality} \begin{split} \lVert \sum \phi(x_i)\phi(y_i)^* \rVert & \ge \lVert \alpha_c(Q) \sum \phi(x_i)\phi(y_i)^* \alpha_c(Q) \rVert \\ & = \Bigl\lVert \alpha_c(Q)\sum_{p(x_i) = p(y_i)} \phi(x_i)\phi(y_i)^*\alpha_c(Q) \Bigr\rVert \\ & = \Bigl\lVert \sum_{p(x_i) = p(y_i)} \phi(x_i)Q\phi(y_i)^* \Bigr\rVert. \end{split} \end{equation} Since $\rankone xy \mapsto \phi(x)Q\phi(y)^*$ and $\rankone xy \mapsto \phi(x)\phi(y)^*$ both extend linearly to nontrivial homomorphisms of the simple algebra ${\mathcal K}(E_c)$, we have \[ \Bigl\lVert \sum_{p(x_i) = p(y_i)} \phi(x_i)Q\phi(y_i)^* \Bigr\rVert = \Bigl\lVert \sum_{p(x_i) = p(y_i)} \phi(x_i)\phi(y_i)^* \Bigr\rVert. \] Combining this with \eqref{eq:Q inequality} gives \eqref{eq:inequality}, completing the proof of simplicity. Our proof that ${\mathcal O}_E$ is purely infinite is an adaptation of the proof of \cite[Proposition~5.3]{bprs}. Let $A$ be a hereditary subalgebra of ${\mathcal O}_E$; we will show that $A$ has an infinite projection. Fix a positive element $a\in A$, scaled so that $\lVert\Phi(a)\rVert = 1$. Choose a finite sum $b = \sum i_E(x_i)i_E(y_i)^* \in{\mathcal O}_E$ such that $b \ge 0$ and $\lVert a - b \rVert < 1/4$. Then $b_0 := \Phi(b)$ is also positive and satisfies $\lVert b_0 \rVert \ge 3/4$. By applying the relations \eqref{eq:cuntz rep} we can assume that there exists $c\in P$ such that $p(x_i) = p(y_i) = c$ whenever $p(x_i) = p(y_i)$, and such that $p(x_i)^{-1}c \in P$ and $p(y_i)^{-1}c \in P$ for all $i$. Then $b_0$ is a positive element of the algebra \[ {\mathcal F}_c := \lsp\{i_E(x)i_E(y)^*: p(x) = p(y) = c\}, \] and hence its image under the canonical isomorphism ${\mathcal F}_c \cong {\mathcal K}(E_c)$ has a unit eigenvector $f\in E_c$ with eigenvalue $\lVert b_0 \rVert$. It follows that $b_0i_E(f) = \lVert b_0 \rVert i_E(f)$, and hence the projection $r := i_E(f)i_E(f)^*$ satisfies $rb_0r = \lVert b_0 \rVert r$. By Lemma~\ref{lemma:kill}, there is a unit vector $w\in E$ such that \[ \alpha^{i_E}_c(i_E(w)i_E(w)^*) i_E(x_i)i_E(y_i) \alpha^{i_E}_c(i_E(w)i_E(w)^*) = 0 \] whenever $p(x_i) \ne p(y_i)$. Let $q := i_E(fw)i_E(fw)^*$, noting that $q \le r$ and $q \le \alpha^{i_E}_c(i_E(w)i_E(w)^*)$. Then \[ qbq = qb_0q = qrb_0rq = \lVert b_0 \rVert qrq = \lVert b_0 \rVert q \ge \textstyle\frac 34 q, \] and since $\lVert a - b \rVert < 1/4$ we thus have $qaq \ge qbq - \frac 14q \ge \frac 12q$. It follows that $qaq$ is invertible in $q{\mathcal O}_Eq$. Let $c$ be its inverse, and define $v := c^{1/2}qa^{1/2}$. Then $v$ is a partial isometry since $vv^* = c^{1/2}qaqc^{1/2} = q$, and its initial projection belongs to $A$ since $v^*v = a^{1/2}qcqa^{1/2} \le \lVert c\rVert a$ and $A$ is hereditary. Since $v^*v \sim q = i_E(fw)i_E(fw)^* \sim i_E(fw)^*i_E(fw) = 1$ and ${\mathcal O}_E$ contains nonunitary isometries, we deduce that $v^*v$ is an infinite projection in $A$. \end{proof} \section{Simplicity for lexicographic product systems} For the lexicographic product systems of Example~\ref{example:lexicographic}, we have the following partial converse to Theorem~\ref{theorem:main}. \begin{theorem}\label{theorem:lexicographic} Let $G$ be a countable abelian group, let $P$ be a subsemigroup of $G$ which contains the identity, let $d:P\to\field{N}^*$ be a semigroup homomorphism, and let $E$ be the lexicographic product system determined by $d$. Then ${\mathcal O}_E$ is simple if and only if $d$ is injective. \end{theorem} \begin{proof} By Proposition~\ref{prop:nontrivial}, $E$ admits a nontrivial Cuntz representation, so one direction follows from Theorem~\ref{theorem:main}. For the converse, suppose $d$ is not injective. Fix $s,t\in P$ such that $s\ne t$ and $d(s) = d(t)$. We claim that $b := i_E(s,\delta_0) - i_E(t,\delta_0)$ generates a proper ideal ${\mathcal I}\triangleleft{\mathcal O}_E$, so that ${\mathcal O}_E$ is not simple. Let $S:E\to B(L^2(\field{T}))$ be the distinguished Cuntz representation defined in Example~\ref{example:lexicographic}. Since $S_*(b) = S(s,\delta_0) - S(t,\delta_0) = 0$ we have $b\in\ker S_*$, and hence ${\mathcal I}$ is not all of ${\mathcal O}_E$. To see that ${\mathcal I}$ is nonzero, choose $\lambda\in\widehat G$ such that $\lambda(s) \ne \lambda(t)$, and define $T(r,x) := \lambda(r)S(r,x)$ for all $(r,x)\in E$. Then $T$ is a Cuntz representation of $E$ such that \[ T_*(b) = T(s,\delta_0) - T(t,\delta_0) = (\lambda(s) - \lambda(t))S(s,\delta_0) \ne 0, \] and we deduce that $b$, and hence ${\mathcal I}$, is nonzero. \end{proof} When $P = \field{N}^2$ we can say a bit more. For every $m,n\in\field{N}^*$, write $E(m,n)$ for lexicographic product system over $\field{N}^2$ determined by the homomorphism $(a,b)\in\field{N}^2 \mapsto m^an^b \in\field{N}^*$. Note that $E(m,n) \cong E(n,m)$. \begin{theorem}\label{theorem:N2} If $m = 1$, then ${\mathcal O}_{E(m,n)} \cong {\mathcal O}_n\otimes C(\field{T})$. If $m,n \ge 2$ and $\log_m n$ is irrational, then ${\mathcal O}_{E(m,n)}$ is simple. If $m,n \ge 2$ and $\log_m n$ is rational, then there exists a unique positive integer $l$ such that $m = l^a$ and $n = l^b$ for some relatively prime positive integers $a$ and $b$; for this $l$ we have ${\mathcal O}_{E(m,n)} \cong {\mathcal O}_l\otimes C(\field{T})$. \end{theorem} The proof of this Theorem is preceded by two Propositions. The first deals with constructing representations of lexicographic product systems over $\field{N}^k$, allowing for $k = \infty$ by defining $\field{N}^\infty := \bigoplus_{i=1}^\infty \field{N}$. Let $\{e_a: 1 \le a \le k\}$ be the canonical basis for $\field{N}^k$. \begin{prop}\label{prop:construct reps} Let $d:\field{N}^k\to\field{N}^*$ be a semigroup homomorphism, and let $B$ be a unital $C^*$\nobreakdash- algebra. Suppose $\{U_{a,i}: 1 \le a \le k,\ 0 \le i \le d(e_a) - 1 \}$ is a set of isometries in $B$ which satisfies \[ U_{a,i}^*U_{a,j} = 0 \qquad\text{whenever $i \ne j$, and} \] \begin{equation}\label{eq:commutation} U_{a,i}U_{b,j} = U_{b,p}U_{a,q} \qquad\text{whenever $id(e_b) + j = pd(e_a) + q$.} \end{equation} Then there is a unique representation $\phi:E(d) \to B$ such that $\phi(a,\delta_i) = U_{a,i}$, and $\phi$ is a Cuntz representation if \begin{equation}\label{eq:cuntz} \sum_{i=0}^{d(e_a) - 1} U_{a,i}U_{a,i}^* = 1 \qquad\text{for every $a$.} \end{equation} \end{prop} \begin{proof} Fix $s\in\field{N}^k$ and $m\in\{0,\dots, d(s) - 1\}$. Our first goal is to define $\phi(s,\delta_m)$. In any expression of $s$ as an ordered sum $e_{a_1} + \dotsb + e_{a_l}$ of basis elements we have $l = \lVert s \rVert_1$. For each of these finitely-many ordered $l$\nobreakdash- tuples $(a_1, \dots, a_l)$, it is easy to see that there is a unique way of factoring $(s,\delta_m)$ as a product $(e_{a_1}, \delta_{i_1})\dotsb (e_{a_l}, \delta_{i_l})$. These factorizations can be obtained from one another by using the commutation relations \[ (e_a,\delta_i)(e_b, \delta_j) = (e_b, \delta_p)(e_a, \delta_q) \qquad\text{whenever $id(e_b) + j = pd(e_a) + q$} \] on adjacent factors a finite number of times, and hence \eqref{eq:commutation} implies that the corresponding product $U_{a_1,i_1}\dotsb U_{a_l,i_l}$ is independent of the factorization; we define $\phi(s,\delta_m)$ to be this common element of $B$. It is obvious that $\phi(s,\delta_m)\phi(t,\delta_n) = \phi((s,\delta_m)(t,\delta_n))$, and defining \[ \phi(s,x) := \sum_{i=0}^{d(s) - 1} (x \mid \delta_i) \phi(s, \delta_i) \qquad\text{for $(s,x)\in E(d)$} \] gives the desired representation $\phi$. If \eqref{eq:cuntz} holds, then \[ \sum_{m=0}^{d(s) - 1} \phi(s,\delta_m)\phi(s,\delta_m)^* = \sum_{i_1 = 0}^{d(e_{a_1}) - 1} \dotsb \sum_{i_l = 0}^{d(e_{a_l}) - 1} U_{a_1,i_1}\dotsb U_{a_l,i_l}U_{a_l,i_l}^*\dotsb U_{a_1,i_1}^* = 1, \] so $\phi$ is a Cuntz representation. \end{proof} \begin{prop}\label{prop:factor} ${\mathcal O}_{E(m,n)} \cong {\mathcal O}_{E(m, mn)}$. \end{prop} \begin{proof} Let $E = E(m,n)$ and $F = E(m,mn)$. Since $\dim F_{(a,b)} = m^a(mn)^b = m^{a+b}n^b = \dim E_{(a+b,b)}$, we can define $\psi:F\to{\mathcal O}_E$ by \[ \psi((a,b),x) = i_E((a+b,b),x) \qquad\text{for all $((a,b),x)\in F$.} \] It is easy to see that $\psi$ is a Cuntz representation of $F$, so there is a homomorphism $\psi_*:{\mathcal O}_F \to {\mathcal O}_E$ such that $\psi_*\circ i_F = \psi$. Note that for any $(a,b)\in\field{N}^2$ and $j\in\{0,\dots, a^mb^n-1\}$ we have \begin{align*} i_E((a,b),\delta_j) & = i_E((b,0),\delta_i)^*i_E((a+b,b),\delta_{im^an^b + j}) \\ & = \psi((b,0),\delta_i)^*\psi((a,b),\delta_{im^an^b + j}); \end{align*} since elements of the form $i_E((a,b),\delta_j)$ generate ${\mathcal O}_E$, $\psi_*$ is surjective. To see that $\psi_*$ is injective, we construct its inverse. First define \[ V_{1,i} := i_F((1,0),\delta_i) \in {\mathcal O}_F \qquad\text{for $0 \le i \le m-1$, and} \] \[ V_{2,k} := i_F((0,1),\delta_k) \in {\mathcal O}_F \qquad\text{for $0 \le k \le mn-1$.} \] Note that the $V_{1,i}$'s and $V_{2,k}$'s are isometries which generate ${\mathcal O}_F$, that $\sum V_{1,i}V_{1,i}^* = 1 = \sum V_{2,k}V_{2,k}^*$, and that \[ V_{1,i}V_{2,k} = V_{2,k'}V_{1,i'} \qquad\text{whenever $imn + k = k'm + i'$.} \] Now define $U_{1,i} := V_{1,i}$ for $0 \le i \le m-1$ and \[ U_{2,j} := \sum_{l=0}^{m-1} V_{2,jm+l}V_{1,l}^* \qquad\text{for $0 \le j \le n-1$.} \] It routine to check that the $U_{2,j}$'s are isometries whose range projections sum to the identity. Suppose $in + j = pm + q$, where $0 \le i,q \le m-1$ and $0 \le j,p \le n-1$. Then \begin{align*} U_{1,i}U_{2,j} & = \sum_{l=0}^{m-1} V_{1,i}V_{2,jm + l}V_{1,l}^* = \sum_{l=0}^{m-1} V_{2,in + j}V_{1,l}V_{1,l}^* \\ & = V_{2,in + m} = V_{2,pm + q} = \sum_{l=0}^{m-1} V_{2,pm + l}V_{1,l}^*V_{1,q} = U_{2,p}U_{1,q}, \end{align*} so by Proposition~\ref{prop:construct reps} there is a Cuntz representation $\phi:E\to{\mathcal O}_F$ such that $\phi((1,0),\delta_i) = U_{1,i}$ and $\phi((0,1),\delta_j) = U_{2,j}$. We check that $\phi_*\circ \psi_*$ is the identity on ${\mathcal O}_F$, from which it follows that $\psi_*$ is injective, and hence an isomorphism: if $0 \le i \le m-1$, then \[ \phi_*\circ\psi_*(V_{1,i}) = \phi_*(\psi((1,0),\delta_i)) = \phi_*(i_E((1,0),\delta_i)) = U_{1,i} = V_{1,i}; \] if also $0 \le j \le n-1$, then \begin{align*} \phi_*\circ\psi_*(V_{2,in + j}) & = \phi_*(\psi((0,1),\delta_{in + j})) \\ & = \phi_*(i_E((1,1),\delta_{in + j})) = U_{1,i}U_{2,j} = V_{2,in + j}. \end{align*} \end{proof} \begin{proof}[Proof of Theorem~\ref{theorem:N2}] By Proposition~\ref{prop:construct reps}, ${\mathcal O}_{E(1,n)}$ is the universal $C^*$\nobreakdash- al\-ge\-bra for collections $\{U_{1,0}\} \cup \{U_{2,0}, \dots, U_{2,n-1}\}$ of isometries satisfying \[ U_{1,0}U_{1,0}^* = 1, \quad \sum_{i=0}^{n-1} U_{2,i}U_{2,i}^* = 1, \quad\text{and}\quad U_{1,0}U_{2,i} = U_{2,i}U_{1,0}; \] i.e., ${\mathcal O}_{E(1,n)} \cong {\mathcal O}_n\otimes C(\field{T})$. Suppose $m,n\ge 2$. If $\log_m n$ is irrational, then $d: (a,b)\in\field{N}^2 \mapsto m^an^b$ is injective, and ${\mathcal O}_{E(m,n)}$ is simple by Theorem~\ref{theorem:lexicographic}. The existence and uniqueness of $l$ is elementary when $\log_m n$ is rational, and repeated applications of Proposition~\ref{prop:factor} give ${\mathcal O}_{E(m,n)} \cong {\mathcal O}_{E(l,1)}$. \end{proof} \section{Nuclearity} Following \cite{dinhjfa,laca}, we call a cross section $u:s\in P\mapsto u_s\in E_s\setminus\{0\}$ a {\em twisted unit\/} of $E$ if $u_su_t \in \field{C} u_{st}$ for every $s,t\in P$. \begin{theorem}\label{theorem:nuclearity} Let $G$ be a countable abelian group, let $P$ be a subsemigroup of $G$ which contains the identity, and let $E$ be a product system over $P$ of finite-dimensional Hilbert spaces. If $E$ admits a twisted unit, then ${\mathcal O}_E$ is nuclear. \end{theorem} Our proof is modelled on the one given by Laca in \cite[Section~3]{laca}: we realize ${\mathcal O}_E$ as a twisted semigroup crossed product of an AF algebra by $P$, and deduce nuclearity from a theorem of Murphy \cite[Theorem~3.1]{murphy}. We begin by recalling the definition of a twisted semigroup crossed product. Let $P$ be as in the theorem, let $\beta$ be an action of $P$ as endomorphisms of a unital $C^*$\nobreakdash- algebra $A$, and let $\omega$ be a multiplier of $P$. We call $(A,P,\beta,\omega)$ a {\em twisted semigroup dynamical system\/}. Suppose $B$ is a unital $C^*$\nobreakdash- algebra. An {\em isometric $\omega$\nobreakdash- representation\/} of $P$ in $B$ is a map $V:P\to B$ such that each $V_s$ is an isometry and $V_sV_t = \omega(s,t)V_{st}$ for all $s,t\in P$. A {\em covariant representation\/} of $(A,P,\beta,\omega)$ in $B$ is a pair $(\pi,V)$ consisting of a unital homomorphism $\pi:A\to B$ and an isometric $\omega$\nobreakdash- representation $V:P\to B$ such that \[ \pi(\beta_s(a)) = V_s\pi(a)V_s^* \qquad\text{for all $s\in P$ and $a\in A$.} \] A {\em crossed product\/} for $(A,P,\beta,\omega)$ is a triple $(C,i_A,i_P)$ consisting of a unital $C^*$\nobreakdash- algebra $C$ and a covariant representation $(i_A,i_P)$ of $(A,P,\beta,\omega)$ in $C$ such that (a) for every covariant representation $(\pi,V)$ of $(A,P,\beta,\omega)$ in a unital $C^*$\nobreakdash- algebra $B$, there is a homomorphism $\pi\times V:C\to B$ such that $(\pi\times V)\circ i_A = \pi$ and $(\pi\times V)\circ i_P = V$; and (b) $C$ is generated as a $C^*$\nobreakdash- algebra by $i_A(A)\cup i_P(P)$. \noindent The triple $(C,i_A,i_P)$ is unique up to canonical isomorphism. \begin{proof}[Proof of Theorem~\ref{theorem:nuclearity}] Let $u$ be a twisted unit for $E$. By replacing $u_s$ with $\lVert u_s \rVert^{-1}u_s$, we can assume that each $u_s$ is a unit vector. Then $u_su_t = \omega(s,t)u_{st}$ determines a multiplier $\omega$ of $P$. As in Section~\ref{section:abelian}, let ${\mathcal F}_E$ be the inductive limit $\varinjlim{\mathcal K}(E_s)$ under the embeddings $S\in{\mathcal K}(E_s) \mapsto S\otimes 1^t\in{\mathcal K}(E_{st})$, and let $\iota_s$ be the canonical embedding of ${\mathcal K}(E_s)$ in ${\mathcal F}_E$. Since tensoring on the left by the rank-one projection $\rankone{u_r}{u_r}$ commutes with tensoring on the right by the identity, for each $r\in P$ there is an endomorphism $\beta_r$ of ${\mathcal F}_E$ which satisfies \[ \beta_r(\iota_s(S)) = \iota_{rs}((\rankone{u_r}{u_r})\otimes S) \qquad\text{for all $s\in P$ and $S\in{\mathcal K}(E_s)$.} \] Note that $\beta_r$ is injective. Moreover, for any $a\in {\mathcal F}_E$ we have \begin{align*} \beta_r\circ\beta_s(a) & = (\rankone{u_r}{u_r}) \otimes (\rankone{u_s}{u_s}) \otimes a = (\rankone{u_ru_s}{u_ru_s}) \otimes a \\ & = (\rankone{\omega(r,s)u_{rs}}{\omega(r,s)u_{rs}}) \otimes a = (\rankone{u_{rs}}{u_{rs}}) \otimes a = \beta_{rs}(a), \end{align*} and hence $\beta:P\to\End {\mathcal F}_E$ is a semigroup homomorphism. Let $i_{\mathcal F}:{\mathcal F}_E\to{\mathcal O}_E$ be the embedding of Proposition~\ref{prop:FE}, and define $i_P:P\to{\mathcal O}_E$ by $i_P(s) := i_E(u_s)$. We claim that $({\mathcal O}_E,i_{\mathcal F},i_P)$ is a crossed product for the twisted semigroup dynamical system $({\mathcal F}_E,P,\beta,\omega)$. Each $i_P(s)$ is an isometry, and $i_P(s)i_P(t) = i_E(u_su_t) = \omega(s,t)i_E(u_{st}) = \omega(s,t)i_P(st)$, so $i_P$ is an isometric $\omega$-representation of $P$ in ${\mathcal O}_E$. If $x,y\in E_s$, then \begin{align*} i_{\mathcal F}(\beta_r(\iota_s(\rankone xy))) & = i_{\mathcal F}(\iota_{rs}((\rankone{u_r}{u_r})\otimes(\rankone xy))) = i_{\mathcal F}(\iota_{rs}(\rankone{u_rx}{u_ry})) \\ & = i_E(u_rx)i_E(u_ry)^* = i_P(r)i_{\mathcal F}(\iota_s(\rankone xy))i_P(r)^*, \end{align*} and since elements of the form $\iota_s(\rankone xy)$ have dense linear span in ${\mathcal F}_E$, we deduce that $i_{\mathcal F}(\beta_r(a)) = i_P(r)i_{\mathcal F}(a)i_P(r)^*$ for all $r\in P$ and $a\in {\mathcal F}_E$. Thus $(i_{\mathcal F},i_P)$ is a covariant representation of $({\mathcal F}_E,P,\beta,\omega)$ in ${\mathcal O}_E$. We now verify condition (a) of a crossed product. Suppose $(\pi,V)$ is a covariant representation of $({\mathcal F}_E,P,\beta,\omega)$ in a unital $C^*$\nobreakdash- algebra $B$. Define $\phi:E\to B$ by \[ \phi(x) := \pi(\iota_{p(x)}(\rankone x{u_{p(x)}}))V_{p(x)} \qquad\text{for $x\in E$.} \] We claim that $\phi$ is a Cuntz representation of $E$ in $B$. If $x,y\in E_s$, then \begin{align*} \phi(y)^*\phi(x) & = V_s^*\pi(\iota_s(\rankone y{u_s}))^*\pi(\iota_s(\rankone x{u_s}))V_s \\ & = V_s^*\pi(\iota_s((\rankone {u_s}y)(\rankone x{u_s})))V_s \\ & = (x\mid y)V_s^*\pi(\iota_s(\rankone{u_s}{u_s}))V_s \\ & = (x\mid y)V_s^*\pi(\beta_s(1))V_s = (x\mid y)1, \end{align*} and \begin{equation}\label{eq:xystar} \begin{split} \phi(x)\phi(y)^* & = \pi(\iota_s(\rankone x{u_s}))V_sV_s^*\pi(\iota_s(\rankone y{u_s}))^* \\ & = \pi(\iota_s((\rankone x{u_s})(\rankone{u_s}{u_s})(\rankone {u_s}y))) = \pi(\iota_s(\rankone xy)). \end{split} \end{equation} With ${\mathcal B}_s$ an orthonormal basis for $E_s$, \eqref{eq:xystar} gives \[ \sum_{f\in {\mathcal B}_s} \phi(f)\phi(f)^* = \sum_{f\in {\mathcal B}_s} \pi(\iota_s(\rankone ff)) = \pi(\iota_s(1)) = 1 = \phi(1). \] To see that $\phi$ is multiplicative, suppose $x\in E_s$ and $y\in E_t$. Then \begin{align*} \phi(x)\phi(y) & = \pi(\iota_s(\rankone x{u_s}))V_s\pi(\iota_t(\rankone y{u_t}))V_t \\ & = \pi(\iota_s(\rankone x{u_s}))\pi(\beta_s(\iota_t(\rankone y{u_t})))V_sV_t \\ & = \pi(\iota_{st}((\rankone x{u_s})\otimes 1^t) ((\rankone{u_s}{u_s})\otimes(\rankone y{u_t})))V_sV_t \\ & = \pi(\iota_{st}((\rankone x{u_s})\otimes(\rankone y{u_t})))V_sV_t \\ & = \pi(\iota_{st}(\rankone{xy}{u_su_t}))V_sV_t \\ & = \pi(\iota_{st}(\rankone{xy}{u_{st}}))\overline{\omega(s,t)}V_sV_t \\ & = \pi(\iota_{st}(\rankone{xy}{u_{st}}))V_{st} = \phi(xy). \end{align*} Thus $\phi$ is a Cuntz representation of $E$. Let $\pi\times V:{\mathcal O}_E\to B$ be the integrated form of $\phi$; that is, $\pi\times V$ satisfies $(\pi\times V)\circ i_E = \phi$. We claim that $(\pi\times V)\circ i_{\mathcal F} = \pi$ and $(\pi\times V)\circ i_P = V$, giving condition (a) of a crossed product. If $x,y\in E_s$, then by \eqref{eq:xystar} we have \[ (\pi\times V)\circ i_{\mathcal F}(\iota_s(\rankone xy)) = \pi\times V(i_E(x)i_E(y)^*) = \phi(x)\phi(y)^* = \pi(\iota_s(\rankone xy)), \] and since elements of the form $\iota_s(\rankone xy)$ have dense linear span in ${\mathcal F}_E$, we deduce that $(\pi\times V)\circ i_{\mathcal F} = \pi$. For the second part of the claim, we calculate \begin{align*} (\pi\times V)\circ i_P(s) & = \pi\times V(i_E(u_s)) = \phi(u_s) \\ & = \pi(\iota_s(\rankone{u_s}{u_s}))V_s = \pi(\beta_s(1))V_s = V_sV_s^*V_s = V_s. \end{align*} To verify condition (b) of a crossed product, suppose $x\in E_s$. Then \[ i_E(x) = i_E(x)i_E(u_s)^*i_E(u_s) = i_{\mathcal F}(\iota_s(\rankone x{u_s}))i_P(s) \in C^*(i_{\mathcal F}({\mathcal F}_E)\cup i_P(P)), \] and since ${\mathcal O}_E$ is generated as a $C^*$\nobreakdash- algebra by $i_E(E)$, this shows that it is also generated by $i_{\mathcal F}({\mathcal F}_E)\cup i_P(P)$, as required. We have shown that $({\mathcal O}_E,i_{\mathcal F},i_P)$ is a crossed product for $({\mathcal F}_E,P,\beta,\omega)$. Since ${\mathcal O}_E$ is the twisted semigroup crossed product of a nuclear $C^*$\nobreakdash- algebra by an abelian semigroup of injective endomorphisms, it is nuclear by \cite[Theorem~3.1]{murphy}. \end{proof} \begin{cor}\label{cor:nuclearity} If $E$ is a twisted lexicographic product system over a subsemigroup of a countable abelian group, then ${\mathcal O}_E$ is nuclear. \end{cor} \begin{proof} Since $u_s := (s,\delta_0)$ is a twisted unit for $E$, Theorem~\ref{theorem:nuclearity} applies. \end{proof} \begin{remark} Suppose $P$ is a subsemigroup of a countable abelian group, $d:P\to\field{N}^*$ is an injective homomorphism, and $\omega$ is a multiplier of $P$. Let $E = E(d)^\omega$ be the lexicographic product system $E(d)$ twisted by $\omega$. Then ${\mathcal O}_E$ is a separable, unital $C^*$\nobreakdash- algebra which is simple and purely infinite (Theorem~\ref{theorem:main}) and nuclear (Corollary~\ref{cor:nuclearity}). If $P = \field{N}^k$ and $\omega$ is trivial (i.e., $E = E(d)$), then by the proof of Theorem~\ref{theorem:nuclearity} ${\mathcal O}_E$ is the crossed product of an AF algebra by $\field{N}^k$, hence a full corner of the crossed product of an AF algebra by $\field{Z}^k$, and hence belongs to the bootstrap class of algebras which satisfy the Universal Coefficient Theorem \cite[Theorem~23.1.1]{blackadar}. \end{remark}
\section{Introduction} The bra-ket calculus invented by Dirac \cite{D} for unitary or Hilbert spaces is one of the most coherent, efficient and elegant formalisms. This is based on the natural relation among vectors and functionals represented by ket and respectively bra vectors. In this way, one obtains simple calculation rules which include the axioms of the scalar product as well as the consequences of the Frechet-Riesz theorem. Grace of these qualities, the Dirac formalism offers us the opportunity of working with general operator relations, independent on the concrete representations that can have very different features in the general case of the rigged Hilbert spaces used in quantum theories. However, in many problems the vector spaces can not be organized as unitary, Euclidian or Hilbert spaces. We refer especially to the spaces of finite-dimensional linear representations of the non-compact Lie groups \cite{G,BR} where the bilinear forms we need to construct invariants are not positive definite \cite{W}. In this case one uses the tensor calculus with indices in upper and lower positions and bilinear (or inner) forms defined with the help of a metric tensor. For unitary, Euclidian or Hilbert spaces the tensor calculus is equivalent with the Dirac formalism in a representation given by an orthonormal basis but for the spaces with indefinite metric \cite{B} we have not yet a suitable Dirac formalism which should reproduce all the mechanisms of calculus with covariant and contravariant indices. For this reason we would like to propose here a generalization of the Dirac formalism to vector spaces with indefinite metric. We restrict ourselves only to the finite-dimensional case when these spaces are called semi-unitary or semi-Euclidian \cite{ON}. The main problem here is the generalization of the mutual bra-ket relation to spaces with indefinite metric. In our opinion, this problem can not be solved using only one vector space and its dual space. We mean that starting with a space considered as being of covariant vectors (with contravariant components) and with its dual space of contravariant functionals, we can not relate covariant vectors with contravariant functionals in a satisfactory bra-ket formalism. This could be achieved only by introducing new ingredients, namely contravariant vectors and covariant functionals, which should allow one to relate vectors and functionals of same kind (covariant or contravariant). Therefore, the generalization of the Dirac formalism requires to doubly the number of vector spaces like in the theory of general tensors (with simple and dotted indices) where one uses four vector spaces associated with the four unequivalent fundamental representations of the group of general linear transformations \cite{W}. This means that we have already the framework we need. It remains only to correctly define mutual bra-ket relations in accordance with the mentioned exigency of a coherent bra-ket mechanism. We show that this can be done with the help of two anti-linear mappings that conserve the covariance. These will relate the spaces of covariant and contravariant ket vectors with their corresponding spaces of bra vectors. Moreover, we {\em couple} between themselves the spaces of ket vectors as well as those of bra vectors through an isometry compatible with the considered anti-linear mappings. In these conditions we can use the operator of this isometry as {\em metric operator}. This will replace the metric tensor taking over its role in defining scalar products. Our objective is to generalize the Dirac bra-ket formalism in this mathematical context which combines the framework of the general tensor calculus with the theory of bilinear forms given by metric operators \cite{N}. We establish the basic calculation rules in suitable notations and we verify that these lead to the common tensor calculus in any representation given by a system of dual bases. Actually, we shall see that our formalism recovers all the main results of the theory of the general linear transformations but expressed only in terms of components with usual covariant or contravariant indices. This is because in the bra-ket formalism, where the complex conjugated components appear naturally as the components of bra vectors, the artifice of dotted indices is no more needed. On the other hand, we show that in our approach we have all the technical advantages of the standard bra-ket formalism. One of them is that we can work directly with general relations involving operators instead of their matrix elements in particular representations. Moreover, we can manipulate simultaneously different spectral representations of these operators allowing us to easily study the theory of finite-dimensional representation of the non-compact Lie groups, including those of the symmetry group of the metric operator. The first step is to precise the mathematical framework of our attempt. This is presented in the second section where we define the mutually related pairs of coupled vector spaces which allows us to introduce the metric operator and correct relations among vectors and functionals. The next section is devoted to the specific symbols and notations we propose for vectors and operators. In section 4 we construct two kind of compatible hermitian forms, called dual forms and scalar products respectively, while in section 5 we define the hermitian and Dirac conjugations for all the linear operators we work. Section 6 is devoted to the theory of orthogonal projection operators which help us to define pairs of coupled subspaces with hermitian metric operators. The matrix representations in systems of dual bases are studied in section 7 pointing out the advantages of the orthonormal ones. The theory of basis transformations is briefly treated in section 8 where we give the form of the general linear transformations and we study the symmetry transformations that leave invariant the form of the scalar product. The example we propose in section 9 is the theory of finite-dimensional representations of the $SL(2, \Comp)$ group with invariant scalar products. \section{Coupled vector spaces} Our construction is based on a pair of complex vector spaces, ${\cal V}$ and $\hat{\cal V}$, of {\em covariant} vectors, $x,y,... \in {\cal V}$, and {\em contravariant} vectors, $\hat x, \hat y,...\in \hat{\cal V}$, respectively. The dual of ${\cal V}$, denoted by $\overline{\cal V}$, is the space of contravariant functionals, $\bar x,...$, while the dual of $\hat{\cal V}$, denoted by $\overline{\hat{\cal V}}$, is the space of covariant functionals, $\bar{\hat x}...$. The values of these functionals are $\bar x(y)$ or $\bar{\hat x}(\hat y)$. The basis vectors and the vector components are labeled by Latin indices $i,j,k,...$ while the first Latin ones, $a,b,...$ are held for current needs. We suppose that all the spaces we work are finite-dimensional of dimension $N$ such that $i,j,...=1,2,...,N$. Furthermore, we consider the covariant basis $\{e_{i}\}\subset {\cal V}$, the contravariant basis $\{\hat e^{i}\}\subset \hat{\cal V}$, and the {\em canonical dual} bases, from which $\{\bar{e}^{i}\}\subset \overline{\cal V}$ is the contravariant one while $\{\bar{\hat e_{i}}\}\subset \overline{\hat{\cal V}}$ is the covariant one. Since we assume that these bases satisfy the usual duality conditions, \begin{equation}\label{(dual)} \bar{e}^{i}(e_{j})=\delta^{i}_{j}\,,\qquad \bar{\hat e}_{i}(\hat e^{j})=\delta^{j}_{i}\,, \end{equation} we say that these form a {\em system of dual bases}. A coherent bra-ket mechanism requires to {\em mutually relate} among themselves the vectors and functionals. To this end, we introduce the anti-linear mappings which conserve the covariance, $\phi : {\cal V}\to \overline{\hat{\cal V}}$ and $\hat\phi : \hat{\cal V}\to \overline{\cal V}$, defined by \begin{equation}\label{(phi)} \phi[e_{i}]=\bar{\hat e}_{i}\,,\quad \hat \phi[\hat e^{i}]= \bar{e}^{i}\,, \end{equation} such that \begin{equation}\label{(phixy1)} \phi[x]=(x^{i})^{*}\bar{\hat e}_{i}\,,\quad \hat \phi[\hat y]= (\hat y_{i})^{*}\bar{e}^{i}\,, \end{equation} for any $x=e_{i}x^{i}\in {\cal V}$ and $\hat y=\hat e^{i}\hat y_{i} \in \hat{\cal V}$. Then from (\ref{(dual)}) we find \begin{equation}\label{(phixy)} \phi[x](\hat y)=(x^{i})^{*}\hat y_{i}=\left(\hat\phi[\hat y](x)\right)^{*}\,. \end{equation} In other respects, equations (\ref{(phixy1)}) show that $\overline{\hat{\cal V}}\sim {\cal V}^{*}$ and $\hat{\cal V}\sim \overline{{\cal V}^{*}}$ where ${\cal V}^{*}$ is the complex conjugate vector space of ${\cal V}$. Therefore, our system of mutually related vector spaces is similar to that of the theory of general tensors, $({\cal V}, {\cal V}^{*}, \overline{{\cal V}}, \overline{{\cal V}^{*}})$. Let us consider now the isomorphism $\eta : {\cal V}\to \hat{\cal V}$, defined by the {\em invertible} matrix $|\eta|$ as \begin{equation}\label{(etae)} \eta\,e_{i}=\hat e^{j}\eta_{ji}. \end{equation} This {\em couples} each covariant vector $x=e_{i}x^{i}$ with the contravariant vector $\hat x=\eta\, x =\hat e^{i} x_{i}$ of components $x_{i}= \eta_{ij} x^{j}$. The corresponding isomorphism of the dual spaces, $\bar\eta: \overline{\cal V}\to \overline{\hat{\cal V}}$, is given by \begin{equation} \bar{\eta}\,\bar{e}^{k}=(\eta^{-1})^{kj}\bar{\hat e}_{j} \end{equation} so that $\bar\eta\,\bar y(\eta\,x)= y(x)$. Moreover, it is not difficult to verify that the isomorphism $\eta$ is compatible with the mappings $\phi$ and $\hat\phi$ (closing the diagram) only if \begin{equation}\label{(comp)} \phi[x]=\bar\eta\,\hat\phi[\eta\,x], \quad \forall x\in {\cal V}. \end{equation} \begin{theor} The condition {\rm (\ref{(comp)})} is accomplished if and only if the matrix $|\eta|$ is hermitian, i.e. $\eta_{ij}=(\eta_{ji})^{*}$. \end{theor} \begin{demo} Let us take $x=e_{i}$ and calculate $\bar{\hat e}_{i}= \bar\eta\,\hat\phi[\eta\,e_{i}]$. According to (\ref{(etae)}) and (\ref{(phi)}), this gives $\bar{\hat e}_{i}=(\eta_{ji})^{*}\bar\eta\,\hat\phi[\hat e^{j}] =(\eta^{-1})^{jk}(\eta_{ji})^{*}\bar{\hat e}_{k}$ from which we obtain the desired result. \end{demo} In what follows we consider only invertible operators $\eta$ with hermitian matrices in a given system of dual bases. They will be used as the {\em metric operators} that define the {\em bilinear forms} of the spaces ${\cal V}$ and $\hat{\cal V}$, \begin{eqnarray}\label{(bilf)} h(x,y)&=&\phi[x](\eta\,y),\quad ~~~ x,y \in {\cal V},\\ \hat h(\hat x,\hat y)&=&\hat\phi[\hat x](\eta^{-1}\,\hat y),\quad \hat x,\, \hat y \in \hat{\cal V}. \end{eqnarray} A little calculation giving us the bilinear forms in terms of vector components points out that these are {\em hermitian} since \begin{equation}\label{(herm)} h(x,y)=h(y,x)^{*}\,,\quad \hat h(\hat x,\hat y)=\hat h(\hat y,\hat x)^{*}. \end{equation} In addition, we can verify that \begin{equation}\label{(izom)} \hat h(\eta\,x,\, \eta\,y)=h(x,y) \end{equation} which means that the isomorphism $\eta$ is in fact an {\em isometry} when $|\eta|=|\eta|^{+}$. \begin{defin} The spaces ${\cal V}$ and $\hat{\cal V}$ isometric through the metric operator $\eta$ represent a pair of coupled vector spaces {\rm (cvs)} denoted by $({\cal V},\hat{\cal V},\eta)$. The corresponding dual cvs are $(\overline{\cal V},\overline{\hat{\cal V}},\bar\eta)$. We say that the mappings $\phi$ and $\hat\phi$ mutually relate these pairs of cvs. \end{defin} \section{Notations} The above defined related pairs of cvs represent the appropriate framework of our generalized Dirac formalism. Let us start with the notations of the basic elements. \begin{defin} The spaces ${\cal K}\equiv {\cal V}$ and $\hat{\cal K}\equiv \hat{\cal V}$ are the spaces of {\em ket-down} vectors, $\kd{~} \in {\cal K}$, or of {\em ket-up} vectors, $\ku{~}\in \hat{\cal K}$. \end{defin} Thus in our formalism the covariant vectors appear as ket-down vectors while the contravariant ones as ket-up vectors. The bra vectors mutually related with these ket vectors can be defined with the help of the mappings $\phi$ and $\hat \phi$. \begin{defin} The bra vector related to the ket-down vector $\kd{x}$ is the {\em bra-down} vector $\bd{x}= \phi[\kd{x}] \in {\cal B}\equiv \overline{\hat{\cal V}}$ while the bra vector related to the ket-up vector $\ku{\hat x}$ is the {\em bra-up} vector $\bu{\hat x} = \hat\phi[\ku{\hat x}] \in \hat{\cal B} \equiv \overline{\cal V}$. \end{defin} In this manner we have related the covariant ket vectors of ${\cal K}$ with the covariant bra vectors of ${\cal B}$, which are just the covariant functionals defined on the other ket space, $\hat{\cal K}$. Similarly, the spaces of contravariant vectors, $\hat{\cal K}$ and $\hat{\cal B}$, are also related between themselves even though the contravariant functionals of $\hat{\cal B}$ are defined on ${\cal K}$. Of course, the mutually related ket and bra vectors will be denoted systematically with the same symbol as in the usual bra-ket formalism. Apparently these crossed bra-ket relations seem to be forced but this is the unique way to obtain well-defined hermitian bilinear forms compatible with the natural duality. In other respects, our bra-ket relations are correct in the sense that the bra vector related with a linear combination of ket vectors is the corresponding anti-linear combination of bra vectors. Obviously, this is because the mappings $\phi$ and $\hat\phi$ are anti-linear (e.g. $\phi[\alpha\kd{x}+\beta\kd{y}]=\alpha^{*}\bd{x}+\beta^{*}\bd{y}$). Let us consider now the linear operators defined on our cvs. In general, we denote by $L({\cal V},{\cal V}')$ the set of the linear operators which map ${\cal V}$ onto ${\cal V}'$. In our case, the main pieces are the algebras $L({\cal K},{\cal K}))$ and $L(\hat{\cal K},\hat{\cal K})$, but we are also interested by the operators which map the spaces ${\cal K}$ and $\hat{\cal K}$ to each other. We start with the observation that it is natural to denote by $/\!\!/\in L({\cal K},{\cal K})$ and $\backslash \!\!\backslash\in L(\hat{\cal K},\hat{\cal K})$ the {\em identity} operators of these algebras since they act upon the ket-down and respectively ket-up vectors. Moreover, this notation helps us to indicate the action of the operators of different kind by writing them between identity operators. Thus, the operators of $L({\cal K},{\cal K})$ can be denoted either simply by $A,B,...$ or by $\opd{A}, \opd{B},...$ in order to avoid possible confusions with the operators of the other algebra, $\opu{\hat A},\opu{\hat B},...\in L(\hat{\cal K},\hat{\cal K})$, or with those from $L({\cal K},\hat{\cal K})$ or $L(\hat{\cal K},{\cal K})$ that have to be delimited by both identity operators in a suitable order. On the other hand, the notation we propose has the advantage of indicating the allowed algebraic operations. For example, it is clear that the operators $\backslash A/\in L({\cal K},\hat{\cal K})$ and $/ B\backslash\in L(\hat{\cal K},{\cal K})$ can be multiplied to each other while their sum does not make sense. A special case is that of the metric operators, $\eta\in L({\cal K}, \hat{\cal K})$ and $\eta^{-1}\in L(\hat{\cal K},{\cal K})$, which play the central role in our construction. They will be represented by \begin{equation} \eta\equiv \backslash\!/ \,,\quad \eta^{-1}\equiv /\!\backslash \,, \end{equation} in order to obtain the intuitive calculation rules \begin{equation} /\!\backslash \backslash\!/ = /\!\!/ \,,\quad \backslash\!//\!\backslash=\backslash \!\!\backslash\,. \end{equation} With these symbols, the pair of cvs of ket vectors can be denoted now by $({\cal K}, \hat{\cal K}, \backslash\!/)$ while the related pair of bra cvs is $(\hat{\cal B},{\cal B},/\!\backslash)$. The isometry $\backslash\!/ : {\cal K}\to \hat{\cal K}$ couples not only the vectors $\kd{x}$ and $\ku{\hat x}=\ku{\eta x}\equiv \backslash\!/\!\!\kd{x}$, but also couples the operators $\opd{A}\in L({\cal K},{\cal K})$ and $\opu{\hat A}\in L(\hat{\cal K},\hat{\cal K})$ through \begin{equation}\label{(asop)} \opu{\hat A}= \backslash\!/ A /\!\backslash \,. \end{equation} \section{Hermitian forms} Here we can define two kind of brackets. The first one represents the values of covariant or contravariant functionals. We denote by $\bkud{\hat x}{y}\equiv \hat\phi[\hat x](y)$ the value of the contravariant functional $\bu{\hat x}$ calculated for the vector $\kd{y}$, and by $\bkdu{y}{\hat x}\equiv\phi[y](\hat x)$, the value of the covariant functional $\bd{y}$ calculated for $\ku{\hat x}$. We say that the mappings $\bkud{~}{~}: \hat{\cal K}\times {\cal K}\to \Comp$ and $\bkdu{~}{~}: {\cal K}\times \hat{\cal K}\to \Comp$ are {\em dual forms}. From (\ref{(phixy)}) it results that the dual forms are {\em hermitian}, i.e. \begin{equation}\label{(bk1)} \bkdu{y}{\hat x}=\bkud{\hat x}{y}^{*}\,, \quad \forall\, \kd{y}\in {\cal K},\,\ku{\hat x}\in \hat{\cal K}. \end{equation} The second kind of brackets are just the hermitian bilinear forms defined by (\ref{(bilf)}). Since the metric operator is invertible, these bilinear forms are nondegenerate and, therefore, can be called {\em scalar products} \cite{ON}. In our new notation the scalar product of ${\cal K}$, $\bkdd{~}{~}: {\cal K}\times {\cal K}\to \Comp$, has the values \begin{equation} \bkdd{x}{y}= \bd{x}\!\!\backslash\!/\!\!\kd{y}\equiv h(x,y)\,, \quad \kd{x},\kd{y}\in {\cal K}, \end{equation} while that of $\hat{\cal K}$, $\bkuu{~}{~}: \hat{\cal K}\times\hat{\cal K}\to \Comp$, gives \begin{equation} \bkuu{\hat x}{\hat y}= \bu{\hat x}\!\!/\!\backslash\!\!\ku{\hat y} \equiv \hat h(\hat x,\hat y)\,, \quad \ku{\hat x},\ku{\hat y}\in \hat{\cal K}\,. \end{equation} The equations (\ref{(herm)}) which show that these scalar products are hermitian take the form \begin{equation} \bkdd{x}{y}=\bkdd{y}{x}^{*}\,,\quad \bkuu{\hat x}{\hat y}= \bkuu{\hat y}{\hat x}^{*}\,. \end{equation}\label{(izom1)} Other useful relations can be written starting with the coupled vectors $\ku{\hat x}=\backslash\!/\!\!\kd{x}$ and $\ku{\hat y}=\backslash\!/\!\!\kd{y}$. Thus we can find equivalences among the values of the dual forms and those of the scalar products (e.g. $\bkdu{x}{\hat y}=\bkdd{x}{y}$, $\bkud{\hat x}{y}=\bkuu{\hat x} {\hat y}$, etc.) or to recover equation (\ref{(izom)}) giving us the isometry of cvs, \begin{equation}\label{(xyxy)} \bkdd{x}{y}=\bkuu{\hat x}{\hat y}\,. \end{equation} All these brackets can be imagined as resulting from the traditional ``juxtaposition" of the Dirac formalism \cite{D}. For example, we can write $\bkdu{y}{\hat x}=(\bd{y})(\ku{\hat x})$, $\bkdd{x}{y}=\bd{x} (\backslash\!/\!\!\kd{y})=(\bd{x}\!\!\backslash\!/)\kd{y}$, and so on. Hence the conclusion is that we can combine ket and bra vectors of any kind in order to write brackets. If the slash-lines are parallel we obtain a dual form but when these are not parallel, leaving an empty angle, then we understand that therein is a metric operator giving us a scalar product. The orthogonality is defined by the scalar product which play the same role as those of unitary or Hilbert spaces with the difference that here the ``squared norm" $\bkdd{x}{x}$ can take any real value, including $0$ even for vectors $\kd{x}\not=0$. From (\ref{(xyxy)}) we see that if two ket-down vectors are orthogonal then their coupled ket-up vectors are also orthogonal. As mentioned before, the cvs are isometric in the sense that for two coupled ket vectors we have $\bkdd{x}{x}=\bkuu{\hat x}{\hat x}$. Consequently, the separation of the orbits for which this number is positive, negative or zero, can be done simultaneously for both cvs \cite{B}. \section{Hermitian and Dirac conjugations} Now we have all the elements for defining the hermitian conjugation that gives us the {\em hermitian adjoint} operators of our linear operators. First we consider the operators from $L({\cal K},\hat{\cal K})$ and $L(\hat{\cal K},{\cal K})$ and we define: \begin{defin} The hermitian adjoint operators of $\backslash A/ \in L({\cal K}, \hat{\cal K})$ and $/ B\backslash \in L(\hat{\cal K}, {\cal K})$ are $\backslash A^{+}/ \in L({\cal K}, \hat{\cal K})$ and $/ B^{+}\backslash \in L(\hat{\cal K}, {\cal K})$ which satisfy \begin{eqnarray} \bd{x}\!A^{+}\!\kd{y}=\bd{y}\!A\!\kd{x}^{*}\,,\quad \forall\, \kd{x}\,,\kd{y}\in {\cal K}\,,\\ \bu{\hat x}\!B^{+}\!\ku{\hat y}=\bu{\hat y}\!B\!\ku{\hat x}^{*} \,,\quad \forall\, \ku{\hat x}\,,\ku{\hat y}\in \hat{\cal K}\,. \end{eqnarray} If $A=A^{+}$ or $B=B^{+}$ we say that these operators are {\em hermitian}. \end{defin} Obviously, from (\ref{(herm)}) we see that the metric operators are hermitian, \begin{equation} \backslash\!/^{+}=\backslash\!/\,,\quad /\!\backslash^{+}=/\!\backslash\,. \end{equation} For other operators the situation is more complicated as it results from the following definitions. \begin{defin} The hermitian adjoint operator of $/A/\in L({\cal K},{\cal K})$ is the operator $(\opd{A})^{+}=\opu{A^{+}} \in L(\hat{\cal K},\hat{\cal K})$ which accomplishes \begin{equation}\label{(herma)} \bd{x}\!A^{+}\backslash\!/\!\! \kd{y}=\bd{y}\!\!\backslash\!/ A\!\kd{x}^{*}\,, \quad \forall \kd{x},\kd{y}\in {\cal K}\,. \end{equation} For $\opu{B}\in L(\hat{\cal K},\hat{\cal K})$ the hermitian conjugation is defined by \begin{equation}\label{(hermb)} \bu{\hat x}\!B^{+}/\!\backslash\!\! \ku{\hat y}= \bu{\hat y}\!\!/\!\backslash B\!\ku{\hat x}^{*}\,, \quad \forall \ku{\hat x},\ku{\hat y}\in \hat{\cal K}\,, \end{equation} where $/ B^{+}/=(\backslash B\backslash)^{+}\in L({\cal K},{\cal K})$. \end{defin} One can convince ourselves that the mutually related bra vectors with the ket vectors $/A\kd{x}$ and $\backslash B\ku{y}$ are $\phi[/A\kd{x}]= \bd{x}A^{+}\backslash$ and $\hat \phi[\backslash B\ku{y}]=\bu{y}B^{+}/$ respectively. \begin{defin} We say that the operators which satisfy \begin{equation} \opu{A^{+}}=\backslash\!/ A/\!\backslash \,, \quad \opd{B^{+}}=/\!\backslash B\backslash\!/\,, \end{equation} are hermitian with respect to the metric $\backslash\!/$, or simply {\em semi-hermitian}. \end{defin} In other words an operator $/A/\in L({\cal K},{\cal K})$ is semi-hermitian if its adjoint operator coincides with its coupled operator $\backslash \hat A\backslash$ defined by (\ref{(asop)}). We specify that these definitions can be formulated in terms of dual form. For example, if we take $\ku{\hat y}=\backslash\!/\!\!\kd{y}$ then (\ref{(herma)}) and (\ref{(hermb)}) can be rewritten as \begin{eqnarray} \bd{x}\!A^{+}\!\ku{\hat y}&=&\bu{\hat y}\!A\!\kd{x}^{*}\,,\quad \forall\, \kd{x}\in {\cal K},\, \ku{\hat y}\in \hat{\cal K}\,,\label{(hermab)}\\ \bu{\hat x}\!B^{+}\!\kd{y}&=&\bd{y}\!B\!\ku{\hat x}^{*}\,,\quad \forall\, \kd{y}\in {\cal K},\,\ku{\hat x}\in \hat{\cal K}. \end{eqnarray} Particularly, from (\ref{(bk1)}) we can draw the conclusion that the identity operators are semi-hermitian, $(/\!\!/)^{+}=\backslash \!\!\backslash$. In practice it is convenient to introduce another conjugation operation which should unify the above definitions. This is just the generalization of the familiar Dirac conjugation of the theory of four-component spinors. \begin{defin}\label{(adjd)} Given an operator $X$, the {\em Dirac adjoint} operator of $X$ is \begin{equation} \overline{X}=\left\{ \begin{array}{cll} /\!\backslash X^{+}\backslash\!/& if &X\in L({\cal K},{\cal K})\,,\\ \backslash\!/ X^{+}/\!\backslash& if &X\in L(\hat{\cal K},\hat{\cal K})\,,\\ X^{+}& if &X\in L({\cal K},\hat{\cal K})\, or\, L(\hat{\cal K},{\cal K}) \,. \end{array}\right. \end{equation} The operator $X$ is {\em self-adjoint} if $\,\overline{X}=X$. \end{defin} According to this definition all the hermitian and semi-hermitian operators are self-adjoint, including the metric and the identity operators, \begin{equation} \overline{\backslash\!/}=\backslash\!/\,,\quad \overline{/\!\backslash}=/\!\backslash\,,\quad \overline{/\!\!/}=/\!\!/\,,\quad \overline{\backslash \!\!\backslash}=\backslash \!\!\backslash\,. \end{equation} Furthermore, it is not difficult to demonstrate that \begin{equation} \overline{\left(\overline{X}\right)}=X\,, \end{equation} and that for two operators, $A$ and $B$, from the same linear space $L$ and $\alpha, \beta \in \Comp$ we have \begin{equation} \overline{(\alpha A+\beta B)}=\alpha^{*}\,\overline{A}+\beta^{*}\, \overline{B}\,. \end{equation} If the multiplication of two operators makes sense, e.g. $A,\,B\in L({\cal K},{\cal K})$ or $A\in L({\cal K},\hat{\cal K})$ and $B\in L(\hat{\cal K},{\cal K})$, etc, then we can show that \begin{equation} \overline{AB}=\overline{B}\,\overline{A}\,. \end{equation} Thus we obtain a simple and homogeneous set of calculation rules for all the linear operators we manipulate. However, despite of this advantage, we prefer to use here the hermitian conjugation rather than the Dirac one since we are interested to follow the coherence of the presented formalism in its all details. \section{Projection operators} In the case of our cvs, where the orthogonality is defined by the metric operator, the problem of the decomposition in orthogonal subspaces is more complicated than that of unitary spaces \cite{ON} and, therefore, the theory of projection operators needs some specifications. Let us consider the cvs $({\cal K},\hat{\cal K},\backslash\!/)$ and an idempotent operator $/P/$, satisfying $P^{2}=P$, coupled with $\backslash \hat P\backslash = \backslash\!/ P/\!\backslash$ . Then $P$ is the projection operator on the subspace $P{\cal K}\subset {\cal K}$ while $\opu{\hat P}$, is the projection operator on the subspace $\hat P\hat{\cal K}\subset \hat{\cal K}$. Two projection operators, $/P_{1}/$ and $/P_{2}/$, are {\em additive} if $P_{1}P_{2}= P_{2}P_{1}=0$ since then $P_{1}+P_{2}$ is a projection operator too. The operators $P_{a}, a=1,2,...,n$, which satisfy \begin{equation} P_{a}P_{b}=\delta_{a}^{b}P_{a}\,, \quad \not\!\!{\Sigma}\,, \end{equation} form a set of additive projection operators. This set is called {\em complete} if \begin{equation} \sum_{a}P_{a}=/\!\!/\,. \end{equation} However, these projection operators do not have good orthogonality properties since the additive ones generally are not orthogonal. For this reason one prefers the term perp instead of orthogonal \cite{ON}. \begin{defin} Two projection operators, $P_{1}$ and $P_{2}$, are perp if they satisfy \begin{equation} P_{1}^{+}\backslash\!/ P_{2}=0 \end{equation} Then the projection subspaces are called perp to each other. \end{defin} The projection subspaces of these projection operators can not be coupled anytime because there is the risk to find that the restriction of the metric operator to these subspaces is no more hermitian. Two projection subspaces, $P{\cal K}$ and $\hat P\hat{\cal K}$, can be coupled only if their metric operator $\hat P\backslash\!/ P$ is hermitian. This requires $\hat P=P^{+}$ which means that $P$ must be semi-hermitian (self-adjoint). Then this has the property \begin{equation}\label{(vpr)} P^{+}\backslash\!/ =\backslash\!/ P\,, \end{equation} which indicates that the subspaces $P{\cal K}$ and $P^{+}\hat{\cal K}$ are {\em invariant} subspaces of the metric operators $\backslash\!/$ and $/\!\backslash$. In these conditions the operators \begin{equation} \backslash\!/_{P}=P^{+}\backslash\!/ P= P^{+}\backslash\!/ = \backslash\!/ P\,,\quad /\!\backslash_{P}=P/\!\backslash P^{+}= P/\!\backslash = /\!\backslash P^{+}, \end{equation} can be considered as metric operators since they are hermitian and invertible in the sense that \begin{equation} \backslash\!/_{P}/\!\backslash_{P}=P^{+}\,,\quad /\!\backslash_{P}\backslash\!/_{P}=P\,. \end{equation} \begin{defin} The invariant subspaces $P{\cal K}$ and $P^{+}\hat{\cal K}$ isometric through $\backslash\!/_{P}$ represent a pair of coupled subspaces {\em (css)} denoted by $(P{\cal K},P^{+}\hat{\cal K}, \backslash\!/_{P}) \subset ({\cal K}, \hat{\cal K}, \backslash\!/)$. \end{defin} The related css, $(\hat {\cal B}P, {\cal B}P^{+}, /\!\backslash_{P})$, can be defined using the restrictions to $P{\cal K}$ and $P^{+}\hat{\cal K}$ of the mappings $\phi$ and $\hat \phi$. Now the theory of orthogonal decomposition can be done in terms of css determined by semi-hermitian projection operators. \begin{defin} If two semi-hermitian projection operators are perp then they as well as their projection css are called orthogonal. \end{defin} This definition is justified by the fact that the semi-hermitian projection operators have similar properties as the usual hermitian ones. \begin{theor} Two semi-hermitian projection operators are orthogonal if and only if they are additive. \end{theor} \begin{demo} Let us consider that $P_{1}$ and $P_{2}$ are semi-hermitian and additive, satisfying $P_{1}P_{2}=0$. Then, according to (\ref{(vpr)}), we can write $0=\backslash\!/ P_{1}P_{2}=P_{1}^{+}\backslash\!/ P_{2}$ which means that these projection operators are orthogonal. From the same relation it results that the orthogonal projection operators must be additive since, by hypothesis, $\backslash\!/$ is invertible. \end{demo}\\ The consequence is that the css $({\cal K}_{3},\hat {\cal K}_{3}, \backslash\!/_{3})$ defined by the projection operator $P_{3}=P_{1}+P_{2}$ is the {\em direct sum} \begin{equation} ({\cal K}_{3},\hat {\cal K}_{3}, \backslash\!/_{3})= ({\cal K}_{1},\hat {\cal K}_{1}, \backslash\!/_{1})\oplus ({\cal K}_{2},\hat {\cal K}_{2}, \backslash\!/_{2}) \end{equation} where \begin{equation} \backslash\!/_{3}=\backslash\!/ (P_{1}+P_{2})=\backslash\!/_{1}+\backslash\!/_{2}\,. \end{equation} If $P_{2}=/\!\!/-P_{1}$ then the css defined by $P_{1}$ and $P_{2}$ are {\em orthogonal complements}. In general, a complete set of additive projection operators is a complete set of orthogonal projection operators if all of of these operators are semi-hermitian. \section{Matrix representations} \subsection{Dual bases} According to the usual terminology, any system of dual bases defines a {\em matrix representation} of the cvs of ket and bra vectors. In our new notation the vectors of the dual bases we have introduced in section 2 are \begin{eqnarray} \kd{(i)}\equiv e_{i}\in {\cal K} \,, &\quad& \ku{(i)}\equiv \hat e^{i}\in \hat{\cal K}\,,\label{(b1)}\\ \bu{(i)}\equiv {\bar e}^{i} \in \hat{\cal B}\,, &\quad& \bd{(i)}\equiv \bar{\hat e}_{i}\in {\cal B}\,.\label{(b2)} \end{eqnarray} We consider that the indices $\backslash(i)$ or $(i)/$ are in {\em upper} position while $/(i)$ or $(i)\backslash$ are in {\em lower} position and we use the summation convention over dummy indices in opposite positions (e.g. we sum over $i$ in expressions where we find $... \backslash (i)...(i)\backslash...$ or $... / (i)...(i) /...$). In a given representation, the main tools of our formalism are the duality conditions (\ref{(dual)}), written now as \begin{equation}\label{(dc)} \bkud{(i)}{(j)}=\delta^{i}_{j}\,,\quad \bkdu{(i)}{(j)}=\delta_{i}^{j}\,, \end{equation} and the {\em completeness relations} \begin{equation}\label{(cr)} \kd{(i)}\bu{(i)}=/\!\!/\,, \quad \ku{(i)}\bd{(i)}=\backslash \!\!\backslash\,, \end{equation} which show that the sets of {\em elementary} projection operators \begin{equation}\label{(epr)} /P_{i}/=\kd{(i)}\bu{(i)} \quad \not\!\!{\Sigma}\,,\quad i=1,2,...,N, \end{equation} and respectively $P_{i}^{+}, i=1,2,...,N$, are complete sets of additive projection operators. These formulas contain all the information concerning the vector calculus with upper and lower indices, allowing us to express the final results in terms of vector components and matrix elements. A special role play the matrix elements of the metric operator, \begin{eqnarray} \bkdd{(i)}{(j)}&=& \bd{(i)}\!\!\backslash\!/\!\!\kd{(j)}\equiv\eta_{ij}\,,\\ \bkuu{(i)}{(j)}&=& \bu{(i)}\!\!/\!\backslash\!\!\ku{(j)}\equiv {(\eta^{-1})}^{ij}\,, \end{eqnarray} which change the positions of indices. The components of ket-down vectors are $\bkud{(i)}{x}\equiv x^{i}$ or $\bkdd{(i)}{x}\equiv x_{i}=\eta_{ij}x^{j}$, and similarly for the ket-up vectors. The components of the corresponding bra vectors have to be obtained through complex conjugation, according to the usual properties of the dual forms or scalar products. However, the most important is that now we are able to explicitly use spectral representations. For example, the operator $\opd{A}$ can be written as \begin{equation} \opd{A}=\kd{(i)}\bu{(i)}\!A\!\kd{(j)}\bu{(j)}=\kd{(i)}A^{i\,\cdot}_{\cdot\,j} \bu{(j)} \end{equation} where $A^{i\,\cdot}_{\cdot\,j}$ are the matrix elements in usual notation. The adjoint operator of $A$ is \begin{equation} \opu{A^{+}}=\ku{(i)}\bd{(i)}\!A^{+}\!\ku{(j)}\bd{(j)}=\ku{(i)} (A_{\cdot\,i}^{j\,\cdot})^{*}\bd{(j)} \end{equation}\label{(hermaa)} since from (\ref{(hermab)}) we have \begin{equation}\label{(hermaa)} (A^{+})_{i\,\cdot}^{\cdot\,j}\equiv \bd{(i)}\!A^{+}\!\ku{(j)}= \bu{(j)}\!A\!\kd{(i)}^{*}\equiv(A_{\cdot\,i}^{j\,\cdot})^{*}\,. \end{equation} If, in addition, $A$ is semi-hermitian then $\bd{(i)}\!A^{+}\!\ku{(j)}=\bd{(i)}\!\!\backslash\!/ A/\!\backslash\!\!\ku{(j)}$ and using (\ref{(hermaa)}) we recover the well-known property $\eta_{ik}A^{k\,\cdot}_{\cdot\,l}(\eta^{-1})^{lj}= (A_{\cdot\,i}^{j\,\cdot})^{*}$. In any system of dual bases the traces of the operators $A\in L({\cal K},{\cal K})$ and $B\in L(\hat{\cal K},\hat{\cal K})$ are defined by \begin{equation} {\rm Tr}(A)=\bu{(k)}\!A\!\kd{(k)}\equiv A^{k\,\cdot}_{\cdot \, k}\,,\quad {\rm Tr}(B)=\bd{(k)}\!B\!\ku{(k)}\equiv B_{k\,\cdot}^{\cdot \, k}\,. \end{equation} It is easy to show that the coupled operators (\ref{(asop)}) have the same trace. \subsection{Orthonormal bases} In applications one prefers the representations given by the {\em orthonormal} dual bases where the matrix of the metric operator is diagonal, \begin{equation}\label{(etad)} \eta_{ij}\equiv\bkdd{(i)}{(j)}=\eta_{i}\delta_{ij} \quad \not\!\!{\Sigma}\,,\quad \eta_{i}=\pm1\,, \end{equation} since there the ``squared norms" have the simplest expressions, \begin{equation} \bkdd{x}{x}=\sum_{i=1}^{N}\eta_{i}|\bkud{(i)}{x}|^{2}\,. \end{equation} The numbers $\eta_{i}$ may take $n_{+}$ times the value $1$ and $n_{-}=N-n_{+}$ times the value $-1$. Thus in orthonormal bases the metric operator is defined by its {\em signature} that can be given either explicitly as a sequence of signs or simply as $(n_{+},n_{-})$. We note that $n_{-}$ is called the {\em index} of the metric operator \cite{ON}. The main advantage of the orthonormal bases is that there the elementary projection operators (\ref{(epr)}) are semi-hermitian and, consequently, they are orthogonal to each other. Of a particular interest are the orthogonal projection operators \begin{equation} P_{+}=\sum_{\eta_{i}=1}P_{i}\,, \qquad P_{-}=\sum_{\eta_{i}=-1}P_{i}\,, \end{equation} which satisfy $/P_{+}/+/P_{-}/=/\!\!/$. They split the space ${\cal K}$ into the pair of unitary spaces ${\cal K}_{+}=P_{+}{\cal K}$ and ${\cal K}_{-}=P_{-}{\cal K}$, of dimensions ${\rm dim}\,{\cal K}_{+}=n_{+}$ and ${\rm dim}\,{\cal K}_{-}=n_{-}$. Since the coupled space can be split in the same manner we have \begin{equation}\label{(deco)} ({\cal K},\hat{\cal K},\backslash\!/)= ({\cal K}_{+},\hat{\cal K}_{+},\backslash\!/_{+})\oplus ({\cal K}_{-},\hat{\cal K}_{-},\backslash\!/_{-}) \end{equation} where \begin{equation} \backslash\!/=\backslash\!/_{+} +\backslash\!/_{-}\,;\quad \backslash\!/_{+}=\backslash\!/ P_{+}\,,\quad \backslash\!/_{-}=\backslash\!/ P_{-}\,. \end{equation} Obviously, the same decomposition can be done for the related cvs of bra vectors. Finally we specify that in the particular case of the metric operators with signature $(N,0)$ or $(0,N)$, the matrix $|\eta|$ in orthonormal bases coincides up to sign with the unit matrix. Then our scalar products become usual inner forms (i.e. scalar products in the sense of the theory of Hilbert spaces) and, therefore, ${\cal K}$ and $\hat{\cal K}$ will be {\em unitary} spaces. In this situations we have two options. The first one is to keep the cvs structure if this is appropriate for our problem. The second option is to consider only one ket space ${\cal K}\equiv \hat{\cal K}$ in usual Dirac formalism, with $\ku{~}=\kd{~}=\ket{~}$ and $\backslash\!/=/\!\backslash=/\!\!/=\backslash \!\!\backslash=I$, where $I$ is the identity operator on ${\cal K}$. \section{Basis transformations} \subsection{General linear transformations} The change of the system of dual bases of our related pairs of cvs can be done using for each basis an arbitrary linear transformation, but then it is possible to obtain new bases which do not satisfy the canonical duality conditions or giving a non-hermitian matrix for the metric operator. In order to avoid these unwanted situations, we consider only the transformations which preserve the cvs structure in the sense that (i) leave invariant the duality conditions and (ii) transform the hermitian matrix $|\eta|$ into another hermitian matrix, $|\eta'|$. \begin{theor} The general form of a transformation which satisfies the conditions (i) and (ii) is \begin{eqnarray} &&\kd{(i)}\to\kd{(i)'}=/ T\kd{(i)},\nonumber\\ &&\ku{(i)}\to\ku{(i)'}=\backslash (T^{-1})^{+}\ku{(i)},\label{(cant)}\\ &&\bu{(i)}\to \bu{(i)'}=\bu{(i)}T^{-1}/,\nonumber\\ &&\bd{(i)}\to\bd{(i)'}=\bd{(i)}T^{+}\backslash.\nonumber \end{eqnarray} where $T\in {\rm Aut}\,({\cal K})\subset L({\cal K},{\cal K})$. \end{theor} \begin{demo} Let us start with the following linear transformation of the ket vectors \begin{equation}\label{(trket)} \kd{(i)'}=/ T\kd{(i)}\,,\quad \ku{(i)'}=\backslash\tilde T\ku{(i)}\,, \end{equation} given by two operators $T\in {\rm Aut}({\cal K})$ and $\tilde T\in {\rm Aut} (\hat{\cal K})$ arbitrarily chosen. The transformations of the bra vectors which conserve the duality conditions are \begin{equation}\label{(trbra)} \bu{(i)'}=\bu{(i)}T^{-1}/\,,\quad \bd{(i)'}=\bd{(i)}{\tilde T}^{-1}\backslash\,. \end{equation} In these new bases the matrix of the metric operator remains hermitian only if $\tilde T=(T^{-1})^{+}$ since then we have \begin{equation}\label{(etap)} \eta_{ij}'\equiv\bkdd{(i)'}{(j)'}=\bd{(i)}\! T^{+}\backslash\!/ T\!\kd{(j)}= (T^{k\,\cdot}_{\cdot\,i})^{*}T^{l\,\cdot}_{\cdot\,j}\,\eta_{kl}\,. \end{equation} Consequently, the transformations (\ref{(trket)}) and (\ref{(trbra)}) take the form (\ref{(cant)}). \end{demo} The conclusion is that $T$ is the operator of a {\em general linear} transformation of the group $GL(N,\Comp)\subset L({\cal K},{\cal K})$. This means that our formalism is equivalent with the representation theory of general linear transformations in the vector spaces of the theory of general tensors. Indeed, in a given system of dual bases, the components of the ket-down and respectively ket-up vectors transform according to a pair of unequivalent fundamental representations of the $GL(N,\Comp)$ group while the bra components transform according to the corresponding complex conjugate representations. Moreover, we observe that the bra components of our formalism, $\bkdu{x}{(k)}\equiv (x^{k})^{*}$ and $\bkud{\hat y}{(k)}\equiv (\hat y_{k})^{*}$, are just those replaced in the theory of general tensors by components carrying dotted indices in opposite positions. With this specification, it is clear that the four transformations laws (\ref{(cant)}) correspond to the four unequivalent fundamental representations of the $GL(N,\Comp)$ group \cite{W}. The advantage of our formalism is that we work directly with the operators $T$ instead of the matrices of their four representations. In this way we can easily find the transformation laws of the matrices of all the operators we use. The matrices of the operators $/A/$ and $\backslash B\backslash$ transform as \begin{eqnarray} A^{i\,\cdot}_{\cdot\,j}\to A^{'i\,\cdot}_{\cdot\,j}\equiv \bu{(i)'}\!A\!\kd{(j)'}&=& \bu{(i)}\!T^{-1}AT\!\kd{(j)}\nonumber\\ &=&(T^{-1})^{i\,\cdot}_{\cdot\,k} A_{\cdot\,l}^{k\,\cdot}T^{l\,\cdot}_{\cdot\,j} \\ B_{i\,\cdot}^{\cdot\,j}\to B_{i\,\cdot}^{'\cdot\,j}\equiv \bd{(i)'}\!B\!\ku{(j)'}&=& \bd{(i)}\!T^{+}B(T^{-1})^{+}\!\ku{(j)}\nonumber\\ &=&(T_{\cdot\,i}^{k\,\cdot})^{*} B^{\cdot\,l}_{k\,\cdot}\left[(T^{-1})^{j\,\cdot}_{\cdot\,l}\right]^{*} \end{eqnarray} while those of the operators from $L({\cal K},\hat{\cal K})$ and $L(\hat{\cal K}, {\cal K})$ transform like the matrices of the metric operators, according to equation (\ref{(etap)}). These transformations leave invariant the traces of the operators from $L({\cal K},{\cal K})$ and $L(\hat{\cal K},\hat{\cal K})$. Moreover, one can show that any system of dual bases can be transformed at any time in a system of othonormal bases using a suitable general transformation. \subsection{Symmetry transformations} The general transformations change the form of all the operator matrices including that of the metric operator. However, there is a special case of some transformations which do not change the matrix of the metric operator. \begin{defin} The general transformations that leave invariant the matrix $|\eta|$ are called symmetry transformations. \end{defin} These transformations are of the form (\ref{(cant)}) but their operators have special properties. \begin{theor} The operators $U$ of the symmetry transformations must satisfy the {\em semi-unitarity} condition, \begin{equation}\label{(etaun)} \backslash U^{+}\backslash\!/ U/=\backslash\!/\,, \end{equation} which can be written as $\overline{U}=U^{-1}$. \end{theor} \begin{demo} A transformation (\ref{(cant)}) generally change the matrix $|\eta|$ according to (\ref{(etap)}). This is a symmetry transformation only if $\bkdd{(i)'}{(j)'}=\bkdd{(i)}{(j)}$. Hereby it results (\ref{(etaun)}). \end{demo}\\ Consequently, the symmetry transformations have the form \begin{eqnarray} &&\kd{(i)}\to\kd{(i)'}=/ U\kd{(i)},\nonumber\\ &&\ku{(i)}\to\ku{(i)'}=\backslash \hat U\ku{(i)},\label{(cantu)}\\ &&\bu{(i)}\to \bu{(i)'}=\bu{(i)} U^{-1}/,\nonumber\\ &&\bd{(i)}\to\bd{(i)'}=\bd{(i)}\hat U^{-1}\backslash.\nonumber \end{eqnarray} where we recall that $\hat U=\backslash\!/ U/\!\backslash$ is the operator coupled with $U$. The main virtue of these transformations is that they do not change the expressions of scalar product in terms of vector components. In other words these leave invariant not only the expressions of the dual forms but also those of the scalar products. Moreover, when we work with orthonormal bases the symmetry transformations conserve the orthogonality. The semi-unitary operators $U$ which accomplish the condition (\ref{(etaun)}) form a subgroup of $GL(N,\Comp)$, namely the {\em maximal symmetry group} or the {\em gauge group} of the metric operator $\backslash\!/$. Any system of dual bases defines a pair of {\em coupled} fundamental representations of this group and its algebra in the carrier spaces ${\cal K}$ and $\hat{\cal K}$. Obviously, these representations are {\em equivalent} through the metric operator. A given pair of ket cvs can be used as carrier spaces for the coupled {\em semi-unitary} representations of any subgroup of the gauge group. In general, these representations are reducible in usual sense but it is not sure that their subspaces can be correctly coupled. For this reason we consider here a modified definition of reducibility. \begin{defin}\label{(redu)} The coupled representations are irreducible if their generators and the metric operator have no common non-trivial invariant subspaces. Otherwise the representations are reducible. \end{defin} Since these representations are semi-unitary one can show that, like in the unitary case, the reducible representations are decomposable (i.e. fully reducible). Consequently, the original cvs can be written as a direct sum of css carrying irreducible representations. Thus, using this definition we preserve the important advantage of the css structure that guarantees the existence of invariant scalar products. \subsection{Group generators} The properties of the $GL(N,\Comp)$ group and its subgroups are well-studied but it is interesting to review few among them in our formalism where we can work directly with the spectral representations of the group generators. Let us consider the related pairs of cvs carrying the fundamental representations of the $GL(N,\Comp)$ group (\ref{(cant)}). In any system of dual bases (\ref{(b1)}) and (\ref{(b2)}) the usual parametrisation of the operators $T\in GL(N,\Comp)\subset L({\cal K},{\cal K})$ reads \begin{equation}\label{(param)} T(\omega)=e^{\omega^{ij}X_{ij}}\,, \quad X_{ij}=\kd{(i)}\bd{(j)}\!\!\backslash\!/, \end{equation} where $\omega^{ij}$ are arbitrary c-numbers and $X_{ij}$ are the ``real generators" (having real matrix elements in orthonormal bases) which satisfy \begin{equation}\label{(hermx)} (X_{ij})^{+}=\backslash\!/ X_{ji}/\!\backslash \qquad \left(\overline{X}_{ij}=X_{ji}\right)\,. \end{equation} Hereby we can separate the $SL(N,\Comp)$ generators \begin{equation}\label{(hij)} H_{ij}=X_{ij}-\frac{1}{N}\eta_{ji}\,/\!\!/\,, \end{equation} which have the properties \begin{equation} \bkuu{(i)}{(j)}H_{ij}=0\,,\quad {\rm Tr}(H_{ij})=0\,. \end{equation} If we consider only the gauge group of the metric operator $\backslash\!/$ then we have to use the same generators but with restrictions imposed upon the parameter values. From equations (\ref{(etaun)}) and (\ref{(hermx)}) we obtain that the parameters of the gauge group must satisfy the condition $\omega^{ij}+(\omega^{ji})^{*}=0$, which means that \begin{equation}\label{(roio)} \Re\, \omega^{ij}=-\Re\, \omega^{ji}\,,\quad \Im\,\omega^{ij}= \Im\,\omega^{ji}\,. \end{equation} There are $N^{2}$ real parameters as it was expected since the gauge group of the metric operator $\backslash\!/$ of signature $(n_{+},n_{-})$ is the {\em semi-unitary} group $U(n_{+},n_{-})=U(1)\otimes SU(n_{+},n_{-})$. When one uses the parameters $\omega ^{ij}$ then the operators (\ref{(hij)}) are considered as ``real" $SU(n_{+},n_{-})$ generators. However, the canonical parametrisation with the {\em real} parameters (\ref{(roio)}) reads $\omega^{ij}H_{ij}=-i(\Re\omega^{ij}A_{ij}+\Im\omega^{ij}S_{ij})$ involving the {\em semi-hermitian} (self-adjoint) $SU(n_{+},n_{-})$ generators defined as \begin{eqnarray} A_{ij}&=&\frac{i}{2}(X_{ij}-X_{ji})+\frac{1}{N}\Im\,\eta_{ji}\,/\!\!/\,,\nonumber \\ S_{ij}&=&-\frac{1}{2}(X_{ij}+X_{ji})+\frac{1}{N}\Re\,\eta_{ji}\,/\!\!/\,.\label{(AS)} \end{eqnarray} It is clear that the antisymmetric ones, $A_{ij}$, are the generators of the subgroup $SO(n_{+},n_{-})\subset SU(n_{+},n_{-})$ \cite{BR}. In the particular case of real cvs the gauge group reduces to $O(n_{+},n_{-})$. On the other hand, with the help of the generators (\ref{(AS)}) one can show that the group $SL(N,\Comp)$ is the complexification of the $SU(n_{+},n_{-})$ group in the sense that in a parametrisation with real numbers the $SL(N,\Comp)$ generators are $A_{ij}$, $S_{ij}$, $iA_{ij}$ and $iS_{ij}$. An interesting problem is how transform the generators $X_{ij}$ when we change the bases through a general linear transformation (\ref{(cant)}). In our formalism it is easy to show that, according to (\ref{(param)}), the transformed generators are \begin{equation} X'_{ij}=\kd{(i)'}\bd{(j)'}\!\!\backslash\!/=T X_{ij}/\!\backslash T^{+}\backslash\!/=TX_{ij}\overline{T}. \end{equation} Particularly, if we consider only symmetry transformations, $T=U$, then from (\ref{(etaun)}) we recover the usual transformation law of the $SU(n_{+},n_{-})$ generators, \begin{equation} H'_{ij}=U H_{ij}U^{-1}=U^{k\,\cdot}_{\cdot\,i} (U^{l\,\cdot}_{\cdot\,j})^{*}H_{kl}\,, \end{equation} which indicates that they transform according to the {\em adjoint} representation of $SU(n_{+},n_{-})$. \section{The semi-unitary representations of the $SL(2,\Comp)$ group} The application presented in order to illustrate how works our formalism is the problem of the finite-dimensional representations of the $SL(2,\Comp)$ group with invariant scalar products. It is well-known that the finite-dimensional irreducible representations of the $sl(2,\Comp)$ algebra can be constructed with the help of those of the $su(2)$ algebra \cite{W}. However, in general, these do not have invariant scalar products under $SL(2,\Comp)$ transformations. In practice these scalar products are defined in each particular case of physical interest separately starting with a suitable representation which is often reducible. In this section we would like to present the general theory of the semi-unitary and irreducible finite-dimensional representations of $sl(2,\Comp)$, in coupled carrier spaces where the invariant scalar products are well-defined. \subsection{The representations of $su(2)$ in cvs} The problem of the irreducible representations of $su(2)$ in cvs reduces to that of the canonical irreducible representations in unitary spaces ${\cal K}^{j}\sim \hat{\cal K}^{j}$ of weight $j$ \cite{AB}. Therefore the non-trivial cvs may be carrier spaces only for reducible coupled representations. These cvs have the general structure \begin{equation}\label{(cvsrr)} ({\cal K},\hat {\cal K}, \backslash\!/)=\sum_{j\in {\bf J}}\oplus ({\cal K}^{j},\hat {\cal K}^{j}, \backslash\!/^{j}) \end{equation} where \begin{equation} \backslash\!/=\sum_{j\in {\bf J}} \backslash\!/^{j} \end{equation} and ${\bf J}$ is an arbitrary set of weights. In each css we consider the system of {\em canonical} bases of ket vectors, $\{\kd{j,\lambda}\}\subset {\cal K}^{j}$ and $\{\ku{j,\lambda}\}\subset \hat{\cal K}^{j}$, $\lambda=-j,-j+1,...,j$, and the related bases of bra vectors satisfying \begin{equation} \bkud{j,\lambda}{j',\lambda'}=\delta^{j}_{j'}\delta^{\lambda}_{\lambda'}\,, \quad \bkdu{j,\lambda}{j',\lambda'}=\delta_{j}^{j'}\delta_{\lambda}^{\lambda'}\,. \end{equation} The metric operators of css, \begin{equation} \backslash\!/^{j}=\epsilon_{j}\sum_{\lambda}\ku{j,\lambda}\bu{j,\lambda}\,, \end{equation} are defined by the set of numbers $\epsilon_{j}=\pm 1$, $j\in {\bf J}$, that gives the signature of the whole metric operator $\backslash\!/$ of the cvs (\ref{(cvsrr)}). For each css the metric operator $\backslash\!/^{j}$ couples the subspaces ${\cal K}^{j}=P^{j}{\cal K}$ and $\hat{\cal K}^{j}={P^{j}}^{+}\hat{\cal K}$ given by the semi-hermitian projection operator \begin{equation} P^{j}=\sum_{\lambda}\kd{j,\lambda}\bu{j,\lambda}. \end{equation} The spectral representations of the projections of the operators $X\in su(2)$ are \begin{equation} X^{j}=P^{j}XP^{j}=\sum_{\lambda\lambda'}\kd{j,\lambda}X^{j}_{\lambda, \lambda'}\bu{j,\lambda'}, \end{equation} where $X^{j}_{\lambda,\lambda'}$ are the usual matrix elements of $X$ in the canonical basis of the unitary irreducible representation of weight $j$. Hereby it is easily to verify that the generators of the irreducible representations with values in $L({\cal K}^{j}, {\cal K}^{j})$, denoted by $/J^{j}_{a}/$, $a=1,2,3$, are semi-hermitian. Consequently, the generators of the coupled representation are $\hat J_{a}^{j}=(J_{a}^{j})^{+} \in L(\hat{\cal K}^{j}, \hat{\cal K}^{j})$. \subsection{Finite-dimensional representations of $sl(2,\Comp)$} The usual non-covariant generators of the $sl(2,\Comp)$ algebra are the rotation generators, $I_{a},\, a=1,2,3$, and the Lorentz boosts, $K_{a}$. From their well-known commutation rules it results that the algebras generated by \begin{equation} M_{a}=\frac{1}{2}\left(I_{a}+iK_{a}\right)\,,\quad N_{a}=\frac{1}{2}\left(I_{a}-iK_{a}\right) \end{equation} are two $su(2)$ algebras commuting with each other \cite{W}. Notice that these algebras can not be seen as ideals of $sl(2,\Comp)$ since their generators are complex linear combinations of the generators of a real algebra. Our aim is to construct the semi-unitary finite-dimensional representations of $sl(2,\Comp)$ using our formalism. This means to consider from the beginning that the generators of the coupled representations are semi-hermitian satisfying $I_{a}^{+}=\backslash\!/ I_{a}/\!\backslash=\hat I_{a}$ and $K_{a}^{+}=\backslash\!/ K_{a}/\!\backslash=\hat K_{a}$ (or $\overline{I_{a}}=I_{a}$ and $\overline{K_{a}}=K_{a}$). Consequently, we must have \begin{equation}\label{(mnv)} M_{a}^{+}=\backslash\!/ N_{a}/\!\backslash=\hat N_{a}\,,\quad N_{a}^{+}=\backslash\!/ M_{a}/\!\backslash=\hat M_{a}\,,\quad (\overline{M_{a}}=N_{a})\,. \end{equation} The general solution of this problem can be written starting with the space of the reducible representation $(j_{1},j_{2})\oplus (j_{2},j_{1})$ of the $sl(2,\Comp)$ algebra \cite{W}, ${\cal K}=({\cal K}^{j_{1}}\otimes {\cal K}^{j_{2}})\oplus ({\cal K}^{j_{2}}\otimes {\cal K}^{j_{1}})$, with $j_{1}\not=j_{2}$. We find that the generators \begin{eqnarray} M_{a}&=&J_{a}^{j_{1}}\otimes P^{j_{2}} + J_{a}^{j_{2}}\otimes P^{j_{1}}\,,\\ N_{a}&=&P^{j_{1}}\otimes J_{a}^{j_{2}} + P^{j_{2}}\otimes J_{a}^{j_{1}}\,, \end{eqnarray} and the metric operator \begin{eqnarray} \backslash\!/=\epsilon_{j_{1},j_{2}}\sum_{\lambda\lambda'}\left(\ku{j_{1},\lambda} \otimes \ku{j_{2},\lambda'}\bu{j_{2},\lambda'}\otimes \bu{j_{1},\lambda} \right.\nonumber\\ \left.+ \ku{j_{2},\lambda'}\otimes \ku{j_{1},\lambda}\bu{j_{1},\lambda}\otimes \bu{j_{2},\lambda'}\right) \end{eqnarray} with $\epsilon_{j_{1}j_{2}}=\pm 1$, satisfy equations (\ref{(mnv)}). Hereby it results that the vector space coupled with ${\cal K}$ must be $\hat {\cal K}=({\cal K}^{j_{2}}\otimes {\cal K}^{j_{1}}) \oplus ({\cal K}^{j_{1}}\otimes {\cal K}^{j_{2}})$. These semi-unitary representations in cvs $({\cal K},\hat{\cal K},\backslash\!/)$ will be denoted by $[j_{1},j_{2}]$. They can be considered {\em irreducible} in the sense of definition (\ref{(redu)}) since the metric operator $\backslash\!/$ and the generators $M_{a}$ and $N_{a}$ do not have common non-trivial invariant subspaces. We can convince that with the help of the {\em chiral} projection operators, $/P_{L}/= P^{j_{1}}\otimes P^{j_{2}}$ and $/P_{R}/=P^{j_{2}}\otimes P^{j_{1}}$, which generalize the familiar left and right-handed ones of the theory of Dirac spinors. These projection operators are just those of the invariant subspaces of the generators $M_{a}$ and $N_{a}$. They form a complete set of additive projection operators, \begin{equation} P_{L}P_{R}=0\,,\quad P_{L}+P_{R}=/\!\!/, \end{equation} but they are not semi-hermitian (or self-adjoint) operators since \begin{equation} {P_{L}}^{+}=\backslash\!/ P_{R}/\!\backslash\,,\quad {P_{R}}^{+}=\backslash\!/ P_{L}/\!\backslash \,, \quad (\overline{P_{L}}=P_{R})\,. \end{equation} Therefore, the subspaces $P_{L}{\cal K}$ and $P_{R}{\cal K}$ are not invariant subspaces of $\backslash\!/$ and the coupled representations $[j_{1},j_{2}]$ are irreducible from our point of view. In the particular case of $j_{1}=j_{2}=j$ the solution is simpler. The space ${\cal K}=\hat{\cal K}={\cal K}^{j}\otimes {\cal K}^{j}$ is just that of the irreducible representation $(j,j)$ of $sl(2,\Comp)$ \cite{W}. The generators have the form \begin{eqnarray} M_{a}&=&J_{a}^{j}\otimes P^{j}\\ N_{a}&=&P^{j}\otimes J_{a}^{j} \end{eqnarray} while the metric operator reads \begin{equation} \backslash\!/=\epsilon_{j}\sum_{\lambda\lambda'}\ku{j,\lambda}\otimes \ku{j,\lambda'}\bu{j,\lambda'}\otimes \bu{j,\lambda}\,. \end{equation} These irreducible representations will be denoted by $[j]$. \subsection{Rotation bases} Now we can introduce the system of rotation dual bases in which the operators $I^{2}=(I_{1})^{2}+(I_{2})^{2}+(I_{3})^{2}$ and $I_{3}$ as well as their coupling partners are diagonal. Since $I_{a}=M_{a}+N_{a}$, the vectors of these bases can be constructed with the help of the Clebsh-Gordan coefficients of the $SU(2)$ group \cite{AB}. The ket-down vectors of the rotation basis of the subspace ${\cal K}^{j_{1}}\otimes {\cal K}^{j_{2}}\subset {\cal K}$ are defined as \begin{equation} \kd{(j_{1},j_{2})\,s,\sigma}=\sum_{\lambda,\lambda'=\sigma-\lambda}\kd{j_{1},\lambda}\otimes \kd{j_{2},\lambda'}\,\bk{j_{1},\lambda;j_{2},\lambda'}{s,\sigma} \end{equation} such that \begin{eqnarray} I^{2}\kd{(j_{1},j_{2})\,s,\sigma}&=&s(s+1)\kd{(j_{1},j_{2})\,s,\sigma}\,,\\ I_{3}\kd{(j_{1},j_{2})\,s,\sigma}&=&\sigma\kd{(j_{1},j_{2})\,s,\sigma}\,. \end{eqnarray} The rotation bases of the other spaces of our cvs have to be introduced in the same manner. Then by taking into account that \cite{AB} \begin{equation} \bk{j_{1},\lambda;j_{2},\lambda'}{s,\sigma}= (-1)^{s-j_{1}-j_{2}}\bk{j_{2},\lambda';j_{1},\lambda}{s,\sigma} \end{equation} and using the orthogonality relations of these coefficients, we find the final forms of the metric operators in rotation bases. When $j_{1}\not=j_{2}$ this is \begin{eqnarray} \backslash\!/=\epsilon_{j_{1},j_{2}}(-1)^{-j_{1}-j_{2}}\sum_{s=|j_{1}-j_{2}|} ^{j_{1}+j_{2}}(-1)^{s}\sum_{\sigma=-s}^{s}( \ku{(j_{1},j_{2})\,s,\sigma}\bu{(j_{2},j_{1})\,s,\sigma}\nonumber\\ +\ku{(j_{2},j_{1})\,s,\sigma}\bu{(j_{1},j_{2})\,s,\sigma})\,, \end{eqnarray} while for $j_{1}=j_{2}=j$ we have \begin{equation}\label{(omj)} \backslash\!/=\epsilon_{j}(-1)^{2j}\sum_{s=0}^{2j}(-1)^{s}\sum_{\sigma=-s}^{s} \ku{(j,j)\,s,\sigma} \bu{(j,j)\,s,\sigma}\,. \end{equation} In our opinion a good choice of the factors $\epsilon$ could be \begin{equation}\label{(eee)} \epsilon_{j_{1},j_{2}}=(-1)^{j_{1}+j_{2}-|j_{1}-j_{2}|}\,,\quad \epsilon_{j}=(-1)^{2j}\,. \end{equation} Thus we obtain the spectral representations of the metric operators in rotation dual bases. We observe that these bases are orthonormal only for $j_{1}=j_{2}$. In the general case of $j_{1}\not=j_{2}$ the ket-down vectors of the orthonormal basis of ${\cal K}$ are given by the linear combinations \begin{equation} \kd{(\pm)\,s,\sigma}=\frac{1}{\sqrt{2}}(\kd{(j_{1},j_{2})\,s,\sigma}\pm \kd{(j_{2},j_{1})\,s,\sigma})\,. \end{equation} Similarly we get the ket or bra vectors of the other orthonormal bases where the metric operator can be represented as \begin{eqnarray} \backslash\!/=\epsilon_{j_{1},j_{2}}(-1)^{-j_{1}-j_{2}}\sum_{s=|j_{1}-j_{2}|} ^{j_{1}+j_{2}}(-1)^{s}\sum_{\sigma=-s}^{s}( \ku{(+)\,s,\sigma}\bu{(+)\,s,\sigma}\nonumber\\ -\ku{(-)\,s,\sigma}\bu{(-)\,s,\sigma})\,. \end{eqnarray} From this formula we see that for $j_{1}\not=j_{2}$ the metric operator has the symmetric signature $(n,n)$ with $n=(2j_{1}+1)(2j_{2}+1)$. The signatures of the metric operators of the representations $[j]$ result from (\ref{(omj)}) and (\ref{(eee)}) to be either $(n,m)$ for integer $j$ or $(m,n)$ if $j$ is a half-integer, where $n=(j+1)(2j+1)$ and $m=j(2j+1)$. Hence for each pair of coupled irreducible representations of $sl(2,\Comp)$ the metric operator has a well-determined signature, $(n_{+},n_{-})$, which shows us that the corresponding maximal symmetry group is just $U(n_{+},n_{-})$. This result can be important from the physical point of view since the transformations of this group are semi-unitary, leaving invariant the scalar product of the carrier spaces of the $sl(2,\Comp)$ representations. For example, in the theory of Dirac spinors the metric operator of the representation $[1/2,0]$ has the signature $(2,2)$ which explains why the whole algebra of $\gamma$ matrices and $sl(2,\Comp)$ generators is just the $u(2,2)$ algebra \cite{TY}. We note that the corresponding $U(2,2)$ group was recently considered as an extended gauge group of the theory of the Dirac field in curved spacetime, obtaining thus new interesting results \cite{F}. \section{Concluding remarks} The presented formalism is the natural generalization of the Dirac's bra-ket calculus to spaces with indefinite metric. The main point of our proposal is to organize the four different spaces of the theory of general tensors into a pair of coupled ket spaces mutually related with a pair of coupled bra one. Then the metric operator can be correctly introduced in accordance with the natural duality such that compatible scalar products and dual forms do co-exist. In this way we recover the results of the theory of the general linear transformations formulated simply in terms of operators, independent on the concrete representations given by systems of dual bases. For this reason, our approach helps one to precise the nature of the different mathematical objects, avoiding the risk to confuse among themselves those which accidently could have the same components or matrix elements in several representations (e.g. the components of second rank tensors and the operator matrix elements). Moreover, in this framework some special operations used up to now only in particular problems get a general meaning. We refer especially to the Dirac adjoint which in our formalism can be defined for all the types of involved operators, taking over the role of the Hermitian adjoint from the usual unitary case. The presented example points out few of these advantages. Using our general bra-ket calculus we are able to write the spectral representations in different basis of the $sl(2,\Comp)$ generators and the metric operators of the finite-dimensional irreducible representations with invariant scalar products. The study of basis transformations, the generalization of the chiral (left and right-handed) projection operators and the analyze of the gauge group of the metric operator can be easily done in this context. We tried here to remain in the spirit of the original Dirac's bra-ket calculus, imagining a formalism with simple calculation rules but having the ``memory" of the definitions and properties of its basic elements. We hope that this should be appropriate for different applications including algebraic programming on computers. \subsection*{Acknowledgments} I would like to thank Mircea Bundaru and Lucian Gligor for useful discussions concerning the construction of the proposed formalism.
\section{Introduction } It is well established that energy is supplied more or less continuously to the outer lobes by beams of plasma ejected from the nucleus at relativistic speeds. The radio jets, which are the signatures of these energy-carrying beams, expand rapidly with distance from the nucleus for the low-luminosity FRI sources (Fanaroff \& Riley 1974), but are highly collimated and less dissipative in the high-luminosity FRII objects (cf. Bridle \& Perley 1984). However, jets in both these classes show evidence of flaring and recollimation on different scales (cf. Hardee, Bridle \& Zensus 1996). The transverse widths of the knots, $\Phi$, tend to increase with distance, $\Theta$, from the central or nuclear component. However, the increase is not linear as would be expected in a freely expanding jet. The spreading rate, defined to be the ratio $\Phi/\Theta$, is usually larger close to the nucleus, which is followed by a slower spreading rate or recollimation. Some examples of well-studied jets in the high-luminosity FRII sources are the ones in the 3CR quasars studied by Bridle et al. (1994, hereinafter referred to as B94), the N galaxy 3C390.3 by Leahy \& Perley (1995), the radio galaxies Cygnus A (cf. Carilli et al. 1996) and 3C353 (Swain, Bridle \& Baum 1996). The transverse knot widths in the inner jet in the quasar 3C351 suggest a high spreading rate of 0.13, while the outer jet is better collimated. The jets in the quasars 3C204, 3C263 and 3C334 exhibit faster than average spreading rate when they are closest to the nucleus. The knots in the jets in 3C175 and 3C334 exhibit a decrease in their transverse widths before entering the hotspots, while the jets in the quasars 3C204 and 3C263 show evidence of flaring before entering the hotspots (B94). In the N galaxy 3C390.3, the width of the jet is almost constant from 30 to 120 kpc, it decreases till about 180 kpc and then increases again before entering the hotspot (Leahy and Perley 1995). The width of the extended jet in the radio galaxy 3C353 remains roughly constant beyond about 20 kpc (Swain et al. 1996). Evidence of recollimation are also seen in the small-scale nuclear jets. For example, in the compact steep-spectrum quasar 3C138, Fanti et al. (1989) have reported evidence of recollimation of the jet on scales of less than few hundred pc. The nuclear jet in Cygnus A has a width of about 2.2 mas over a distance ranging from 2 to 20 mas from the nucleus (Carilli et al. 1996). These observations suggest that radio jets are often confined on scales ranging from the nuclear jets on the scale of parsecs to the extended ones on scales of hundreds of kpc. The jets could be pressure confined by the external environment or by magnetic fields around the jets (cf. Sanders 1983; Bridle \& Perley 1984; Appl \& Camenzind 1992, 1993a,b; Appl 1996; Kaiser \& Alexander 1997). In this paper, we concentrate on the hotspots and radio knots in the jets and use their sizes over a large range of source sizes to investigate the collimation of jets in the compact steep spectrum and larger sources. The hotspots indicate the Mach disks where the jets terminate and subtend small angles in the radio cores suggesting the high degree of collimation of the jets in these sources (Bridle \& Perley 1984; B94; Fernini, Burns \& Perley 1997, hereinafter referred to as F97). Inclusion of hotspots enables us to study the collimation of jets for a large sample of sources independent of whether a jet has been detected and its transverse width determined. However, the jet momentum may be spread over a larger area than the cross-section of the jet itself due to the dentist-drill effect discussed by Scheuer (1982), where the end of the jet wanders about the leading contact surface drilling into the external medium at slightly different places at different times. Recent simulations of 3D supersonic jets suggest that cocoon turbulence drives the dentist-drill effect (Norman 1996). Over the last few years the sizes of hotspots have been determined for samples of compact steep-spectrum radio sources using largely VLBI and MERLIN observations, as well as for the larger objects using the VLA. Although there is no well-accepted definition of a hotspot (cf. Laing 1989; Perley 1989), we are interested in features which mark the collimation of jets and have used the following empirical definition. The hotspots are defined to be the brightest features in the lobes located further from the nucleus than the end of any jet, and in the presence of more extended diffuse emission these should be brighter by atleast a factor of about 4 (cf. B94). In the presence of multiple hotspots, only the primary hotspot has been considered. The jets which we consider have been well-mapped and follow the defining criteria suggested by Bridle \& Perley (1984). We have used the information on the sizes of the hotspots and the widths of the knots in the jets to study the collimation and expansion of the radio jets over a wide range of angular and linear scales. The overall linear sizes of our objects are spread over about 5 orders of magnitude ranging from about 50 pc to nearly 1 Mpc. \input{jet_collimation_table.tex} \section{Sample of sources} Our sample has been chosen from well-defined samples of compact and larger sources which have been observed with high angular resolution. We have compiled our sample of compact sources from the following samples of CSS and Gigahertz Peaked Spectrum or GPS sources: (i) the sample of 67 sources listed by O'Dea (1998) and O'Dea \& Baum (1997) which is based on the Fanti et al. (1990) CSSs and Stanghellini et al. (1990, 1996) GPS objects; (ii) the Sanghera et al. (1995) compilation of 62 objects, which consists of all sources from the complete samples of 3C (Fanti et al. 1990), PW (Peacock \& Wall 1982) and a Jodrell-Bank sample; and (iii) the 7 confirmed compact symmetric objects or CSOs listed by Taylor, Vermeulen \& Pearson (1995) and Taylor, Readhead \& Pearson (1996). Our resulting combined sample consists of 86 objects, because there are many sources which are common to the above lists, and have been counted only once in our combined list. Of these 86, we have considered those which have a well-defined double-lobed structure and where the hotspot sizes are available or could be determined. Sources with a complex and highly distorted structure have been excluded. Typical examples of such sources are 3C48 (Wilkinson et al. 1991) and 3C119 (Ren-dong et al. 1991a). In addition, we have excluded the source 0319+121 which has been listed as a GPS object by Stanghellini et al. (1990), but single-epoch observations of the nucleus from 1.4 to 15 GHz show that it has a flat radio spectrum (Saikia et al. 1998). A few of our sources, such as 0108+388 (Baum et al. 1990), have extended emission in addition to the compact double or triple structure closer to the nucleus. The extended emission must be due to earlier periods of activity, with the luminosity and size of the component being governed by ageing and expansion of the component. On the other hand, the compact structure closer to the nucleus represent more recent activity with the size of the hotspots indicating the degree of collimation of the jets. Hence, in such cases we have considered only the compact double or triple structure. The sample which consists of 58 objects at this stage has been further restricted to those which have been observed with at least 10 resolution elements, n$_b$, along the main axis. We also focus on the subsample which has been observed with n$_b \geq$ 20, for a more reliable estimation of the hotspot parameters. In our entire sample only 6 lobes have no hotspots which meet our defining criteria. Excluding these lobes does not affect any of the conclusions presented in this paper. Sources which were earlier classified as CSS objects but were later found to be of larger dimensions have not been excluded from the sample. Our final sample of largely CSS and GPS objects consists of 39 sources, 9 of which are $>$20 kpc. This sample is listed in Table 1, which is arranged as follows: columns 1 and 2: source name and an alternative name; column 3: optical identification where G denotes a galaxy, Q a quasar and EF an empty field; column 4: the sample where O, S and T denote O'Dea \& Baum (1997) and O'Dea (1998), Sanghera et al. (1995) and Taylor et al. (1995, 1996) respectively; column 5: redshift; column 6: the structural classification of the source where D denotes a double, T a triple with a radio core, and T? a triple with a possible core component; column 7: the angular separation of the hotspots on opposite sides of the nucleus, expressed in arcsec; column 8: the corresponding linear size in kpc in an Einstein-de Sitter Universe with H$_o$=50 km s$^{-1}$ Mpc$^{-1}$; column 9: the number of resolution elements, n$_b$, along the longest axis of the source, which is the largest angular size of the radio source divided by the size of the restoring beam along the axis of the source; columns 10 and 11: the major and minor axes of one hotspot in mas; columns 12 and 13: the major and minor axes for the other hotspot in mas; column 14 : ratio of the separation of the farther component from the radio core to the nearer one for those classified as T or T?; column 15: references for radio structure. For sources with a radio core, the size of the hotspot farther from the nucleus is listed in columns 10 and 11. The hotspot sizes refer to the full width at half maximum. For some of the images where the authors quote the size of the entire component from the lowest reliable contour, we have estimated the sizes of the hotspots from the images. These have been marked with an asterisk in the Table. For comparison with larger sources we consider those which have been observed with a similar number of resolution elements, n$_b$, along the main axis of the source. The comparison sample has been compiled from the complete sample of 3CR sources (Laing, Riley \& Longair 1983) and consists of the FRII sources which have been observed with n$_b$ between about 10 and 40. These limits were chosen so that the distribution of n$_b$ is similar to the CSS and GPS objects. For the 3CR sources observed with the Cambridge 5-km telescope we have confined ourselves to sources above a declination of 30$^\circ$ so that the beams are not very elliptical. This sample is listed in Table 2 which is arranged similarly to Table 1 except for the following differences. These are all triples from the 3CR sample, and hence the sample column and structual information have been omitted. In addition, we discuss briefly the collimation of jets using sources which have been observed with a larger number of resolution elements along their axes. These have been been observed with high resolution and sensitivity with the Very Large Array (VLA) by B94, Fernini et al. (1993), F97, Black et al. (1992), Leahy et al. (1997), and Hardcastle et al. (1997, 1998). These observations give us estimates of the sizes of hotspots in larger sources. B94 have reported observations of 13 quasars while F97 have listed hotspot sizes for 9 galaxies. Hardcastle et al. (1998, hereinafter referred to as H98) have summarized the properties of jets, cores and hotspots in the sample of FRII galaxies with redshift $<$0.3 which have been observed by the Cambridge group. In addition to the hotspots we have also considered the sizes of the knots in the jets listed in Table 9 of B94 and of the knots in the CSOs observed by Taylor et al. (1995, 1996). \begin{figure*} \vspace{-5.8in} \vbox{ \hbox{ \hspace{-0.5in} \psfig{figure=fig1a.ps,width=7.1in} \hspace{-3.6in} \psfig{figure=fig1b.ps,width=7.1in} } \vspace{-6.0in} \hbox{ \hspace{-0.5in} \psfig{figure=fig1c.ps,width=7.1in} \hspace{-3.6in} \psfig{figure=fig1d.ps,width=7.1in} } } \vspace{-0.2in} \caption{The size of each hotspot is plotted against the largest angular size of each source defined to be the separation of the oppositely-directed hotspots. Filled circles indicate quasars while open circles denote radio galaxies from the sample of largely CSS and GPS objects (Table 1). The knots in the jets of the compact symmetric objects are denoted by open squares. The hotspot sizes of the 3CR sources described in the text (Table 2) are indicated by open and filled triangles for the galaxies and quasars respectively. The arrows indicate upper limits. The upper panels indicate all sources observed with at least 10 resolution elements along the main axes while the lower panels show those which have been observed with at least 20 resolution elements along the main axes. The solid lines denote the linear least-squares fit to the average hotspot size for CSS and GPS sources, defined to be those below 20 kpc. These linear fits have been extended beyond 20 kpc by a dotted line to highlight the flattening of the relationship for larger sources. The dashed lines denote the parabolic fit to the average hotspot size of the sources in both the samples. For sources with upperlimits, the hotspot sizes have been estimated by assuming the true values to be close to the limits. } \end{figure*} \section{Collimation of radio jets} In the upper-left panel of Figure 1, we plot the angular sizes of the major and minor axes of each hotspot against the largest angular separation of each source in Tables 1 and 2, while on the upper-right panel we present the equivalent linear sizes for the same sources. The lower panels show the same plots but only for those sources which have been observed with at least about 20 resolution elements along the long axis of the source. Since we sometimes do not have information on the radio core, especially for the CSS and GPS objects we have plotted the hotspot sizes against the overall separation of the oppositely-directed hotspots. In this figure, we have also plotted the knots in the CSOs imaged by Taylor et al. (1995, 1996). In this paper, we have made the least-squares fits to the hotspots by taking the average hotspot size for each source. This is defined to be the geometric mean of each hotspot and, for those with hotspots on both sides, the average of the two oppositely-directed hotspots for each source. As mentioned earlier, a distance of 20 kpc was chosen as the canonical limit for the size of the CSS and GPS objects (cf. Fanti et al. 1990; O'Dea 1998). A linear least-squares fit to the hotspots for the CSS and GPS objects in Figure 1 observed with at least 10 resolution elements shows that if the jet width is expressed as $d_{jet} = k l^{n}$, then $n \sim 1.02\pm0.06$ and log$(k) \sim -1.23\pm0.05$. For those observed with at least 20 resolution elements the value of $n$ is $1.10\pm0.07$ and log$(k) \sim -1.4\pm0.05$. The sizes of the knots in the CSOs are consistent with this trend. There appears to be a well-defined relationship with the hotspot size increasing with the total size of the source upto a distance of about 20 kpc, the canonical limit for the sizes of CSSs. The sizes of the knots in the jets are similar to those of the hotspots. This relationship is consistent with self-similar models for the evolution of radio sources during the CSS and GPS phase (cf. Kaiser \& Alexander 1997). \subsection{Comparison with larger 3CR sources observed with similar resolution elements} To investigate this relationship for sources $>$ 20 kpc, we have compared the CSS and GPS sources (Table 1) with the sample of 3CR sources (Table 2) which have been observed with similar resolution relative to the largest angular size as used for the CSS sources. The number of resolution elements, n$_b$, for the CSS sources vary from about 10 to 77 with a median value of about 19 while for the 3CR sources n$_b$ ranges from about 10 to 39 with a median value of about 23. Only 5 of the CSSs have n$_b$ $>$40. Considering the sources which have been observed with n$_b$ between about 20 and 40, the median values of n$_b$ for CSS and GPS, and 3CR sources are 25 and 28 respectively, and a Kolmogorov-Smirnov test shows that the two distributions are the same with a probability of 66 per cent. The redshifts of the CSS and GPS sources range from about 0.1 to 2 with a median value of about 0.6 while the redshifts for the 3CR sample is somewhat larger ranging from about 0.16 to 2.0, with a median value of about 0.8. The extension of the linear least-squares fit to the CSS sources beyond 20 kpc shows that most of the larger sources observed with a similar value of n$_b$ lie below this line, suggesting a flattening of this relationship. This indicates recollimation of the jets beyond $\sim$20 kpc. Since most of the measured sizes are upper limits, it is difficult to determine the precise nature of the relationship from the present sample. Assuming that the sizes of the hotspots with upper limits are close to these limits, we have attempted a linear least-squares as well as a parabolic fit to the entire data. The parabolic fits are significantly better, suggesting again a flattening of the relationship. To confirm the trend and determine the degree of flattening, we need to measure more precisely the sizes of the hotspots, rather than have limits to their sizes. \subsection{3CR sources observed with larger values of n$_b$} Over the last few years several authors have determined the sizes of the hotspots of 3CR FRII sources from high-resolution VLA observations with n$_b$, the number of resolution elements along the axis of the source, in the range of about 35 to 1040 (e.g. B94; F97; Black et al. 1992; Leahy et al. 1997; Hardcastle et al. 1997). With higher values of n$_b$ one should expect to see smaller-scale structures within the hotspots, and hence a decrease in the size of the hotspots. A linear least-squares fit to the hotspot size $-$ n$_b$ diagram for the B94, F97 and H98 samples exhibits a weak trend for the hotspot size to decrease with n$_b$ with a slope of $-0.15\pm0.10$. Because of different definitions of hotspots and also difficulties in measuring their sizes, detailed comparisons are often difficult and contentious. However, the sizes of the hotspots from the B94, F97 and H98 have median values in the range of about 1.9 to 4.5 kpc and are consistent with a flattening of the hotspot size $-$ linear size relationship beyond about 20 kpc. The knots in the jets (Table 9 of B94) are also consistent with this flattening. \begin{figure} \vspace{-5.8in} \hbox{ \hspace{-0.5in} \psfig{figure=fig2.ps,width=7.1in} } \vspace{-0.2in} \caption{The ratio, r$_{hs}$, of the size of the hotspot farther from the radio core to that of the oppositely-directed one closer to the core, is plotted against the separation ratio, r$_d$, defined to be the ratio of the separations of the corresponding hotspots from the core. The open and filled circles denote galaxies and quasars from the sample of largely CSS and GPS objects (Table 1), the open and filled stars denote galaxies and quasars from B94 and F97, while the triangles represent radio galaxies from H98. Filled and open squares indicate quasars and radio galaxies from Table 2 which have not been already plotted. } \end{figure} \begin{figure} \vspace{-5.5in} \hbox{ \hspace{-0.5in} \psfig{figure=fig3.ps,width=7.1in} } \vspace{-0.2in} \caption{The ratio, r$_{hs}$, of the size of the hotspot farther from the radio core to that of the oppositely-directed one closer to the core, is plotted against the total luminosity at 5 GHz. The symbols are as defined in Figure 2. } \end{figure} \subsection{Opening angle of the jets} Using the geometric mean of the major and minor axes for each hotspot and the average of the hotspots for each source, we estimate the mean opening angle, ($\phi_{obs}$), for sources below 20 kpc to be about 7$^\circ$. For sources observed with atleast 20 beam widths along the source axes the mean opening angle is about 5$^\circ$. The estimates of the mean opening angle is roughly consistent with estimates for individual jets in different angular and linear scales. It is relevant to note that the opening angle estimated from the hotspots could, in principle, be larger than those estimated from knots in the jets because of the dentist drill effect (Scheuer 1982; Norman 1996). In the radio galaxy Cygnus A, the opening angle of the jet determined from the knots as well as the low surface brightness inter-knot regions is $\approx$1.6$^\circ$ (Carilli et al. 1996), although the morphology of the jet is consistent with regions of larger opening angle of $\approx$5$^\circ$ (Perley, Dreher \& Cowan 1984). The jet in M87 is well-collimated on all scales out to about 2 kpc. The innermost VLBI structure indicates an opening angle of 18$^\circ$, while VLA observations indicate that between the jet and knot A, the jet appears as a uniformly expanding cone with an opening angle of 7$^\circ$ (Biretta 1996; Biretta \& Junor 1995). Among the core-dominated radio sources, one of the best studied ones is the quasar 3C345 which has an opening angle of 26$^\circ$ (Zensus, Cohen \& Unwin 1995). Other examples are NRAO 140 with an opening angle of 12$^\circ$ (Marscher 1988; Unwin \& Wehrle 1992) and 0836+710 which has a mean opening angle of 4.4$^\circ$ (Hummel et al. 1992). \subsection{Asymmetry in collimation} We have also examined the dependence of the ratio of the sizes of the two oppositely-directed hotspots, r$_{hs}$, on their relative separations from the nucleus. In Figure 2 we plot the ratio, r$_{hs}$, defined to be the ratio of the size of the farther hotspot to that of the nearer one, against the separation ratio, r$_d$, for all the sources with a detected radio core and hotspots on both sides. The separation ratio, r$_d$, is defined to be $\geq$ 1, while r$_{hs}$ is the ratio of the geometric mean of the corresponding hotspots. We have used only those hotspots without an upperlimit to any of their axes and with a detected radio core from the sample of largely CSS and GPS objects (Table 1), the 3CR objects listed in Table 2 and the observations of B94, F97 and H98. For the 33 high-luminosity sources with a radio luminosity at 5 GHz $\geq 10^{26}$ W Hz$^{-1}$ sr$^{-1}$, 24 have r$_{hs}$ $>$ 1, implying that the farther hotspot is larger. This trend is also seen in the most asymmetric objects, defined to be those with r$_d >$ 2. Of the 12 sources with r$_d >$ 2, 9 have r$_{hs}$ $>$ 1, and almost all these objects are CSSs. Either light-travel time effects or an intrinsic or environmental asymmetry can produce the effect in the observed sense. For a source inclined at about 50$^\circ$ to the line-of-sight and a hotspot speed of about 0.5c, r$_d$ is close to about 2. There have been several estimates of hotspot advance speeds in VLBI-scale double sources. O'Dea (1998) lists 7 objects which have upper limits to component proper motion which are subluminal, ranging from 0.05-0.5c. Since O'Dea's review the upper limit for 1934$-$638 has been revised to $\sim$0.03$\pm$0.2c (Tzioumis et al. 1998), and an upper limit for 1607+26 (CTD93) has been reported to be less than 0.35c (Shaffer \& Kellermann 1998). Current estimates of hotspot speeds suggest that the effects of an asymmetry dominate. Almost all these objects with r$_d >$2 are compact steep spectrum radio sources, and the deficit of objects with r$_d >$2 and r$_{hs} <1$ is possibly a reflection of their evolution in an asymmetric environment. The tendency for r$_{hs}$ $>$ 1 is not seen in objects of lower luminosity (Figure 3). For example, in the 50 sources with radio luminosity at 5 GHz less than 10$^{26}$ W Hz$^{-1}$ sr$^{-1}$, only 24 have r$_{hs}$ $>$ 1. A number of these objects have r$_{hs}$ significantly smaller than 1, which could be due to an intrinsic asymmetry in the collimation of jets on opposite sides of the nucleus. \section{Discussion and concluding remarks} We summarise the principal trends reported in this paper. \begin{enumerate} \item The relationship between the hotspot sizes and the overall size of the CSS and GPS sources and studies of individual knots in the jets in these sources, suggest that they evolve in an approximately self-similar way. The hotspot size increases with distance from the core as $d_{jet} \propto l^{n}$ where $n \sim$ 1.0 for sources smaller than about 20 kpc. This is similar for sources which have been observed with at least either 10 or 20 resolution elements, n$_b$, along the main axes so that the hotspots could be identified reliably. \item For larger sources observed with at least either 10 or 20 resolution elements along the main axes and with a similar distribution of n$_b$ to the sample of largely CSS and GPS sources, there appears to be a flattening of the relationship. However, since most of the sources have upper limits to their hotspot sizes the precise slope could not be determined reliably. \item For samples of sources observed by Bridle et al. (1994) and Fernini et al. (1993, 1997) with n$_b$ in the range of about 40 to 200, and a sample of 3CR sources with z$<$0.3 summarized by Hardcastle et al. (1998) with n$_b$ in the range of about 35 to 1040, the dependence of hotspot size on n$_b$ is weak. The hotspot size varies by a factor of about 2 for an increase in n$_b$ by a factor of about 300. The hotspot sizes of these large sources are consistent with a flattening of the hotspot size-linear size relationship seen for CSS and GPS objects. \item There is a tendency for the farther hotspot to be larger in sources with a luminosity at 5 GHz $>$ 10$^{26}$ W Hz$^{-1}$ sr$^{-1}$. This could be caused by both light travel time effects as well as an asymmetric environment. No such trend is seen in objects of lower luminosity where the nearer hotspot is significantly larger in a number of sources. This could be due to an asymmetry in collimation on opposite sides of the nucleus. \item There is a trend for the farther hotspot to be larger in the most asymmetric objects, defined to be those with r$_d >$ 2. Almost all the objects with r$_d >$ 2 are CSS objects. Current estimates of hotspot advance speeds suggest that these trends are due to an intrinsic asymmetry in the environment rather than light travel time effects. \end{enumerate} In the CSS phase, the variation of the sizes of the hotspots and the widths of the knots in the jets with linear size could be due to the ambient pressure falling with distance from the nucleus, if the jet is pressure confined. The confinement of the jet by the ambient medium depends on whether the jet pressure is comparable to the external pressure. The equipartition pressure in the hotspots or knots varies from about 10$^{-6}$ N m$^{-2}$ on scales of a few tens of parsec to about $10^{-9}$ N m$^{-2}$ at about 10 kpc from the nucleus (Fanti et al. 1995; Readhead 1995; Readhead et al. 1996a,b). For large sources, the equipartition pressure in the knots of the jets is in the range of about $10^{-10}$ to $10^{-12}$ N m$^{-2}$ (Potash and Wardle 1980; Bridle et al. 1994). Although the nature of the confining medium within about 20 kpc, i.e. the CSS phase, needs to be better understood (cf. Fanti et al. 1995), the gaseous components seen in absorption at X-ray, UV and optical wavelengths (Mathur et al. 1994; Elvis et al. 1996; Netzer 1996) could play an important role in addition to the broad-line and narrow-line gas. For the larger FRII sources the flatter relationship seen in the present data suggests recollimation of the jets beyond the CSS phase. It is important to understand the physical process responsible for the recollimation of jets. Two interesting scenarios which have suggested for recollimation are hydrodynamic collimation involving shocks (Sanders 1983; Falle \& Wilson 1985; Komissarov \& Falle 1996) and magnetic collimation (Begelman, Blandford \& Rees 1984; Appl \& Camenzind 1992, 1993a,b; Appl 1996). The latter could provide a natural explanation of a constant width in large sources, which appears to be suggested by some of the observations such as those of B94 and F97, and in 3C353 (cf. Swain et al. 1996). \section*{Acknowledgments} We thank Robert Laing, Peter Scheuer and an anonymous referee for their critical comments and suggestions, and our colleagues at NCRA especially Gopal-Krishna and late Vijay Kapahi for their comments on this piece of work, and Judith Irwin, Ishwara-Chandra and Vasant Kulkarni for their detailed comments on the manuscript. This research has made use of the NASA/IPAC extragalactic database (NED) which is operated by the Jet Propulsion Laboratory, Caltech, under contract with the National Aeronautics and Space Administration.
\section{Introduction} Any dynamical evolution that occurs in nature is affected by noise. In a neuronal system the noise might be comparable in magnitude to purported underlying deterministic dynamics; in celestial mechanics the degrees of freedom omitted from a particular set of equations may be accounted for by very weak noise. Our task here and in two preceding papers\rf{noisy_Fred,conjug_Fred} is to systematically account for the effects of noise on measurable properties such as dynamical averages\cite{bene} in classical chaotic dynamical systems. The theory is also closely related to the semiclassical expansions based on Gutzwiller's formula for the trace in terms of classical periodic orbits\rf{gutbook} in that both are perturbative theories (in the noise strength or $\hbar$) derived from saddlepoint expansions of a path integral containing a Cantor set of unstable stationary points (typically periodic orbits). The analogy with quantum mechanics and field theory has been made explicit in\rf{noisy_Fred} where Feynman diagrams were used to find the lowest nontrivial noise corrections. Unfortunately like its quantum counterpart, the Feynman diagram method for stochastic dynamics quickly becomes unwieldy at higher orders; rather than applying it directly we turn the argument around and suggest that the more efficient recent approaches of\rf{conjug_Fred} and the present paper be applied to difficult perturbative problems of quantum mechanics and field theory. An elegant method, inspired by the classical perturbation theory of celestial mechanics, is that of smooth conjugations\rf{conjug_Fred}. In this approach the neighborhood of each saddlepoint is flattened by an appropriate coordinate transformation, so the focus shifts from the original dynamics to the properties of the transformations involved. An elementary example is the Ulam map $f(x)=4x(1-x)$ which is solved exactly by the transformation $x=\sin^2(\pi\theta/2)$ leading to the piecewise linear tent map $f(\theta)=1-|1-2\theta|$. In general there is no such explicit solution, but the expressions obtained for perturbative corrections are much simpler than those found from the equivalent Feynman diagrams. Using these techniques, we were able to extend the stochastic perturbation theory to the fourth order in the noise strength. Fourth order should be sufficient for most realistic calculations, but does not provide enough information to determine the convergence properties of the expansion, or determine eigenvalues beyond the first few. In this paper we develop a third approach, based on construction of an explicit matrix representation of the stochastic evolution operator. Numerical implementation requires a truncation to finite dimensional matrices, and is less elegant than the smooth conjugation method, but for high expansion orders (here eighth, but higher orders seem quite feasible) and many eigenvalues it is currently unsurpassed. As with the previous formulations, it retains the periodic orbit structure, thus inheriting valuable information about the dynamics. In the following sections we define the stochastic dynamics and show how to obtain matrix representations, both globally and located on the periodic orbits, as an expansion in terms of the noise strength $\sigma$. The matrix elements are obtained from derivatives of the dynamics computed around each periodic orbit. We give as a numerical example the quartic map considered in both previous papers, although the approach is very general and is by no means restricted to one dimension, to maps, or to Gaussian noise. We find that up to eighth order, the cumulants converge super-exponentially with the length of periodic orbit and the expansion is now shown to be accurate to larger values of $\sigma$. \section{The stochastic evolution operator and its spectrum} An individual trajectory in presence of additive noise is generated by iteration \beq x_{n+1}=f(x_{n})+\sigma\xi_{n} \,, \ee{mapf(x)-Diag} where $f(x)$ is a map, $\xi_n$ a random variable with the normalized distribution $p(\xi)$, and $\sigma$ parametrizes the noise strength. In what follows we shall assume that the mapping $f(x)$ is one-dimensional and expanding, and that the $\xi_n$ are uncorrelated. A density of trajectories $\phi(x)$ evolves with time as \beq \phi_{n+1}(y) = \left( \Lop \circ \phi_{n}\right)(y) = \int dx \, \Lop(y,x) \phi_{n}(x) \ee{DensEvol} where $\Lop$ is the {\evOper} \bea \Lop(y,x) &=& \delta_\sigma(y-f(x)) \continue \delta_\sigma(x) &=& \int \delta(x-\sigma \xi) p(\xi) d\xi \,=\, \frac{1}{\sigma} p\left( \frac{x}{\sigma} \right) \,. \label{oper-Diag} \eea For a repeller the leading eigenvalue of the {\evOper} yields a physically measurable property of the dynamical system, the escape rate from the repeller. In the case of deterministic flows, the periodic orbit theory yields explicit formulas for the spectrum of $\Lop$ as zeros of its {\fd}\rf{QCcourse}. Our goal here is to explore the extent to which such methods are applicable to systems with noise and to quantum systems. In particular, we are interested in exploring the dependence of the eigenvalues $\eigenvL(\sigma)$ of $\Lop$ on the noise strength parameter $\sigma$. The eigenvalues are determined by the eigenvalue condition \beq \eigCond(\sigma,\eigenvL(\sigma)) = \det(1-\Lop/\eigenvL(\sigma)) =0 \label{eigCond} \eeq where $ \eigCond(\sigma,1/z) = \det(1-z\Lop) $ is the {\fd} of the {\evOper} $\Lop$. Computation of such determinants commences with evaluation of the traces of powers of the {\evOper} \bea \tr {z\Lop \over 1-z\Lop} &=& \sum_{n=1}^{\infty} C_{n} z^n \,,\qquad C_{n} = \tr \Lop^n \,, \label{tr-L-ith-Diag} \eea which are then used to compute the cumulants $Q_{n}=Q_{n}(\Lop)$ in the cumulant expansion \beq \det(1-z\Lop) = 1- \sum_{n=1}^{\infty} Q_{n} z^n \,, \ee{Fred-cyc-exp-Diag} by means of the recursion formula \beq Q_n = {1 \over n}\left( C_n -C_{n-1} Q_1 - \cdots C_1 Q_{n-1}\right) \ee{Fd-cyc-exp-Diag} which follows from the relation \beq \det(1-z\Lop) \defeq \exp\left(-\sum_n^\infty {z^n \over n} \tr \Lop^n \right) \,. \ee{det-tr-Diag} Our task is to compute the cumulants $Q_n$. We start by introducing a matrix representation for $\Lop$. \section{Matrix representation of evolution operator} \label{ENTIRE} As the mapping $f(x)$ is expanding by assumption, the {\evOper} \refeq{DensEvol} smoothes the initial distribution $\phi(x)$. Hence it is natural to assume that the distribution $\phi_{n}(x)$ is analytic, and represent it as a Taylor series, intuition being that the action of $\Lop$ will smooth out fine detail in initial distributions and the expansion of $\phi_{n}(x)$ will be dominated by the leading terms in the series. An analytic function $g(x)$ has a Taylor series expansion \[ g(x) = \sum_{m=0}^\infty \frac{x^{m}}{m!} \left. \frac{\pde^{m}}{\pde y^{m}}g(y) \right|_{y=0} \,. \] Expanding $\Lop(y,x)$ in Taylor series in $y$ enables us to rewrite traces of $\Lop^n$ as \bea \tr \Lop^2 &=& \int dx dy\, \Lop(y,x) \Lop(x,y) \continue &=& \sum_{m,m'} \int dx dy\, \left(\frac{y^{m'}}{m'!} \left. \frac{\pde^{m'}}{\pde v^{m'}} \Lop(v,x) \right|_{v=0}\right) \ceq \qquad\qquad \left(\frac{x^{m}}{m!} \left. \frac{\pde^{m}}{\pde u^{m}} \Lop(u,y) \right|_{u=0}\right) \nnu \eea Following H.H.~Rugh\rf{Rugh92} we now define the matrix ($m,m'= 0,1,2, ...$) \beq \left(\Lmat{}\right)_{m'm} = \left. \frac{\pde^{m'}}{\pde y^{m'}} \int dx \, \Lop(y,x) \frac{x^{m}}{m!} \right|_{y=0} . \ee{Lmat-Diag} $ \Lmat{}$ is a matrix representation of $\Lop$ which maps the $x^m$ component of the density of trajectories $\phi_n(x)$ in \refeq{DensEvol} to the $y^{m'}$ component of the density $\phi_{n+1}(y)$, with $y=f(x)$. The desired traces can now be evaluated as traces of the matrix representation $ \Lmat{}$, $ \tr \Lop^n = \tr \Lmat{}^n \,. $ As $\Lmat{}$ is infinite dimensional, in actual computations we have to truncate it to a given finite order. The Feynman diagrammatic and the smooth conjugation methods developed in the preceding papers\rf{noisy_Fred,conjug_Fred} require no such approximations. However, as we shall see below, for expanding flows the structure of $ \Lmat{}$ is such that its finite truncations give very accurate spectra. Our next task is to evaluate the matrix elements of $ \Lmat{}$. \section{Weak noise expansion of the evolution operator} We have written the operator $\Lop$ in \refeq{oper-Diag} in terms of the Dirac delta function, $ \Lop(x',x)=\int \delta(x'-f(x)-\sigma\xi)p(\xi)d\xi $, in order to emphasize that in the weak noise limit the stochastic trajectories are concentrated along the deterministic trajectory $x' = f(x)$. Hence it is natural to expand the delta function in a Taylor series in $\sigma$ \bea \Lop(x',x) &=& \delta(x'-f(x)) \ceq +\, \sum_{n=2}^{\infty}\frac{(-\sigma)^n}{n!}\delta^{(n)}(x'-f(x)) \int \xi^n p(\xi)d\xi \,, \nnu \eea where $ \delta^{(n)}(y) = {\pde^n \over \pde y^n} \delta(y) \,. $ This yields a representation of the {\evOper} centered along the deterministic trajectory, with the {\FPoper} $\delta(x'-f(x))$, and corrections given by derivatives of delta functions weighted by moments of the noise distribution $a_n=\int p(\xi)\xi^nd\xi$, \begin{equation} \Lop(x',x) = \delta(x'-f(x)) \,+\, \sum_{n=2}^{\infty}\frac{(-\sigma)^n}{n!} a_n \delta^{(n)}(x'-f(x)). \label{opexp} \end{equation} In our numerical tests we find it convenient to assume that the noise is Gaussian, $ p(\xi)= e^{-\xi^2/2}/{\sqrt{2\pi}} \,. $ For the Gaussian noise all $a_n$ moments are known, and the weak noise expansion of $\Lop$ is \bea \Lop(x',x) &=& {1 \over \sqrt{2 \pi \sigma^2}} e^{-{(x'-f(x))^2/2\sigma^2} } \continue &=& \sum_{n=0}^{\infty} \frac{\sigma^{2n}}{n!2^n} \delta^{(2n)}(x'-f(x)) \continue &=& \delta(x'-f(x)) + {\sigma^2 \over 2} \delta^{(2)}(x'-f(x)) \ceq \qquad\qquad + {\sigma^4 \over 8} \delta^{(4)}(x'-f(x)) + \cdots \,. \label{delGaussExp} \eea The choice of Gaussian noise is not essential, as the methods that we develop here apply equally well to any other peaked smooth noise distribution, as well as space dependent noise distributions $p(x,\xi)$. In any case, as the neighborhood of any trajectory is nonlinearly distorted by the flow, the integrated noise is never Gaussian, but colored. \section{Local matrix representation of evolution operator} \label{LocMatr-Diag} Traces of powers of the {\evOper} $\Lop^n$ are now also a power series in $\sigma$, with contributions composed of $\delta^{(m)}(f(x_a)-x_{a+1})$ segments. The contribution is non-vanishing only if the sequence $x_1,x_2,...,x_n, x_{n+1} = x_{1}$ is a periodic orbit of the deterministic map $f(x)$. Thus the series expansion of $\tr \Lop^n$ has support on all periodic points $x_a = x_{a+n}$ of period $n$, $f^n(x_a)=x_a$; the skeleton of periodic points of the deterministic problem also serves to describe the weakly stochastic flows. The contribution of the $x_a$ neighborhood is best presented by introducing a coordinate system $\field_a$ centered on the cycle points, together with a notation for the map \refeq{mapf(x)-Diag} and the operator \refeq{oper-Diag} centered on the $a$-th cycle point \bea x_a &\to& x_a+\field_a \,, \qquad a=1,...,n_p \continue f_a(\field) &=& f(x_a+\field) \continue {\Lop}_a(\field_{a+1},\field_a) &=& \Lop(x_{a+1}+\field_{a+1},x_a+\field_a) \,. \eea The weak noise expansion \refeq{opexp} for the $a$-th segment operator is given by \[ {\Lop}_a(\field',\field)=\sum_{n=0}^{\infty}\frac{(- \sigma)^{n}}{n!} a_n \delta^{(n)}(\field'+x_{a+1}- f_a(\field)) \,. \] Repeating the steps that led to \refeq{Lmat-Diag} we construct the local matrix representation of $\Lop_{a}$ centered on the $x_a \to x_{a+1}$ segment of the deterministic trajectory \bea \left(\Lmat{a}\right)_{m'm} &=& \left. \frac{\pde^{m'}}{\pde \field'^{m'}} \int d\field \,\Lop_{a}(\field',\field) \frac{\field^{m}}{m!} \right|_{\field'=0} . \continue &=& \sum_{n=max(m-m',0)}^{\infty} \frac{(-\sigma)^n}{n!}a_n(\Bmat{a})_{m'+n,m} \,. \label{BtoL} \eea Due to its simple dependence on the Dirac delta function, $\Bmat{}$ can expressed in terms of derivatives of the inverse of $f_a(\field)$: \bea (\Bmat{a})_{n m} &=& \left. \frac{\pde^{n}}{\pde \field'^{n}} \int d\field \, \delta(\field' + x_{a+1} -f_a(\field)) \frac{\field^{m}}{m!} \right|_{\field'=0} \continue &=& \left. \frac{\pde^{n}}{\pde \field'^{n}} \frac{(f_a^{-1}(x_{a+1}+\field')-x_a)^{m}}{m!|f_a'(f_a^{-1}(x_{a+1}+\field'))|}\right|_{\field'=0} \continue &=& \frac{\mbox{sign}(f_a')}{(m+1)!}\left.\frac{\pde^{n+1}( {\cal F}_a(\field')^{m+1})}{\pde \field'^{n+1}}\right|_{\field'=0}, \label{Bmatrix} \eea where we introduced the shorthand notation ${\cal F}_a(\field')= f_a^{-1}(x_{a+1}+\field')-x_a$. If we expand ${\cal F}_a(\field')$ in a Taylor series, the constant term is zero, since $f_a^{-1}(x_{a+1})=x_a$. So we can write: \bea {\cal F}_a(\field')=\sum_{l=1}^{\infty}\frac{{\cal F}_a^{(l)}}{l!}\field'^l, \eea where $1/{\cal F}_a^{(1)}=f'_a$. The matrix elements can be calculated explicitly as a multinomial expansion \cite{abramo} \bea \left(\sum_{l=1}^{\infty}\frac{x_l}{l!}t^l\right)^m & = & m!\sum_{n=l}^{\infty} \frac{t^n}{n!} \continue \ceq \cdot \, \sum(n|a_1,...,a_n)'x_1^{a_1}...x_n^{a_n}, \label{Abram} \eea where the second sum $\left(\sum\right)$ goes over all non-negative integers such that: \bea a_1+2a_2+...+na_n=n, \mbox{\hspace{0.3cm}} a_1+a_2+...+a_n=m, \eea and the multinomial coefficient is: \bea (n|a_1,a_2,...,a_n)'=\frac{n!}{(1!)^{a_1}a_1!(2!)^{a_2}a_2!...(n!)^{a_n}a_n!}. \eea We apply the formula $\refeq{Abram}$ to ${\cal F}_a(\field')$ with power $m+1$: \bea ({\cal F}_a(\field'))^{m+1} &=& (m+1)! \sum_{l=m+1}^{\infty}\frac{\field'^n}{n!}\sum(l|a_1,a_2,...,a_l)' \continue \ceq \mbox{} \cdot \, ({{\cal F}_a}^{(1)})^{a_1}({{\cal F}_a}^{(2)})^{a_2}...({{\cal F}_a}^{(l)})^{a_l}. \eea For the $(n+1)$ -th derivative of this expression evaluated at $\field'=0$ only the $l=n+1$ term is non-vanishing. The matrix elements vanish for $n<m$, so $\Bmat{}$ is a lower triangular matrix: \bea (\Bmat{a})_{n m} &=& \sum(n+1|a_1,a_2,...,a_{n+1})' \continue \ceq \cdot \, ({{\cal F}_a}^{(1)})^{a_1}({\cal F}_a^{(2)})^{a_2}...({\cal F}_a^{(n+1)})^{a_{n+1}}. \eea The diagonal and the nearest off-diagonal matrix elements can easily be worked out. Here we show the first four expressed in terms of the derivatives of the original map: \bea (\Bmat{a})_{m m}~~~ &=& \frac{1}{|f_a'|f_a'^{m}} \continue (\Bmat{a})_{m+1,m} &=& - \frac{1}{2} \frac{(m+2)!}{m!} \frac{f_a''}{|f_a'|f_a'^{m+2}} \label{Bexplicit}\\ (\Bmat{a})_{m+2,m} &=& -\frac{(m+3)!}{24 m! |f_a'|f_a'^{m}} \left( \frac{f_a'''}{f_a'^{3}} -3 (m+4) \frac{(f_a'')^2}{f_a'^{4}} \right) \continue (\Bmat{a})_{m+3,m} &=& - \frac{(m+4)!}{48 m!}{|f_a'|f_a'^{m}} \left( 2 \frac{f_a''''}{f_a'^{4}} - 4(m+5) \frac{f_a'' f_a'''}{f_a'^{5}} \right. \ceq \qquad\qquad \left. + (m+5)(m+6) \frac{f_a''^{3}}{f_a'^{6}} \right) \continue & &\cdots \nnu \,, \eea where $f_a'$, $f_a''$, $\cdots$ refer to the derivatives of $f(x)$ evaluated at the periodic point $x_a$. By assumption the map is expanding, $|f_a'|>1$. Hence the diagonal terms drop off exponentially, as $1/|f_a'|^{m+1}$, the terms below the diagonal fall off even faster, and we are justified in truncating $\Bmat{a}$, as truncating the matrix to a finite one introduces only exponentially small errors. In the local matrix approximation the traces of \evOper s are approximated by \beq \left. \tr {\Lop}^n\right|_{\mbox{\tiny saddles}} =\sum_{p} n_p \sum_{r=1}^{\infty} \delta_{n, n_p r} \tr \Lmat{p}^r = \sum_{j=0}^{\infty}C_{n j}\sigma^j \,, \label{tracenp} \eeq where $\tr \Lmat{p} = \tr{ \Lmat{\cl{p}}\Lmat{2}\cdots \Lmat{1}}$ is the contribution of the $p$ cycle, and the power series in $\sigma^j$ follows from the expansion \refeq{BtoL} of $\Lmat{a}$ in terms of $\Bmat{a}$. The subscript {\tiny saddles} is a reminder that this is a saddle-point approximation to $\tr {\Lop}^n$ (see \refref{noisy_Fred} for a discussion), valid as an asymptotic series in the limit of weak noise. As a simple check of the above formulas, consider the noiseless case, for which the $(\Lmat{a})_{m'm} = (\Bmat{a})_{m'm}$ matrices are a representation of the deterministic \FPoper\ $\left. \Lop \right|_{\sigma=0}$. The $\Lmat{a}$ are triangular with diagonal elements $ (\Lmat{a})_{m m}= \frac{1}{|f_a'|f_a'^{m}} \,. $ The trace of the $\Lop$ on a periodic orbit $p$ is therefore \[ \tr \Lmat{p} = \tr{ \Lmat{\cl{p}}\Lmat{2}\cdots \Lmat{1}} =\sum_{m=0}^{\infty} \frac{1}{|\ExpaEig_p|\ExpaEig_p^{m}} =\frac{1}{|1-\ExpaEig_p|} \,, \] and we recover the standard deterministic trace formula\rf{QCcourse} for the {\FPoper} \beq \tr{\Lop}^n= \sum_p \cl{p} \sum_{r=1}^\infty \delta_{n,\cl{p} r} \frac{1}{|1-\ExpaEig_p^r|} \,. \eeq \section{Perturbative corrections to eigenvalues} \label{trtoeig} The eigenvalue condition \refeq{eigCond} is an implicit equation for the eigenvalue $\eigenvL=\eigenvL(\sigma)$ of form $\eigCond(\sigma,\eigenvL(\sigma)) = 0$. As the eigenvalue condition is satisfied for any $\sigma$, all total derivatives of the eigenvalue condition with respect to $\sigma$ vanish, leading to \bea 0 &=& {d \over d\sigma} \eigCond(\sigma,\eigenvL(\sigma )) = {d \eigenvL \over d\sigma} {\pde \eigCond \over \pde \eigenvL} \,+\, {\pde \eigCond \over \pde \sigma} \continue 0 &=& {d^2 \eigenvL \over d\sigma^2} {\pde \eigCond \over \pde \eigenvL} \,+\, \left({d \eigenvL \over d\sigma}\right)^2 {\pde^2 \eigCond \over \pde \eigenvL^2} \,+\, 2 {d \eigenvL \over d\sigma} {\pde^2 \eigCond \over \pde \sigma \pde \eigenvL} \,+\,{\pde^2 \eigCond \over \pde \sigma^2} \label{sec-derPert} \\ 0 &=& {d^3 \eigenvL \over d\sigma^3} {\pde \eigCond \over \pde \eigenvL} \,+\, 3{d^2 \eigenvL \over d\sigma^2} {d \eigenvL \over d\sigma} {\pde^2 \eigCond \over \pde \eigenvL^2} \,+\, \left({d \eigenvL \over d\sigma}\right)^3 {\pde^3 \eigCond \over \pde \eigenvL^3} \ceq \,+\, 3 {d^2 \eigenvL \over d\sigma^2} {\pde^2 \eigCond \over \pde \sigma \pde \eigenvL} \,+\, 3 \left({d \eigenvL \over d\sigma}\right)^2 {\pde^3 \eigCond \over \pde \sigma \pde \eigenvL^2} \continue \ceq \,+\, 3 {d \eigenvL \over d\sigma} {\pde^3 \eigCond \over \pde \sigma^2 \pde \eigenvL} \,+\,{\pde^3 \eigCond \over \pde \sigma^3} \,, \nnu \eea and so on. $\eigenvL(0)$ can be computed by cycle expansions for a deterministic, noiseless flow. $\sigma \neq 0$ then parametrizes a weak perturbation to the deterministic \FPoper\ $\left. \Lop \right|_{\sigma=0}$. The above formulas enable us to compute recursively, order by order in $\sigma^n$, the perturbative corrections to the eigenvalues of $\Lop$ \beq \eigenvL(\sigma) = \sum_{m=0}^\infty \eigenvL_m \sigma^m \,,\qquad \eigenvL_m = {1 \over m!} \left.{d^m~ \over d\sigma^m}\eigenvL(\sigma) \right|_{\sigma=0} \,, \ee{pertExp1} in terms of partial derivatives of the eigenvalue condition $\eigCond(\sigma,\eigenvL(\sigma))$ \beq F_{kl} = \left.{\pde^{k+l}~~~ \over \pde \eigenvL^k \pde \sigma^l} \eigCond(\sigma,\eigenvL) \right|_{\sigma=0, \eigenvL = \eigenvL(0)} \,. \eeq In this notation the formulas \refeq{sec-derPert} for $\eigenvL_m$ take the form \bea \eigenvL_1 &=& - {F_{01} \over F_{10}} \continue \eigenvL_2 &=& -{1 \over 2 \, F_{10}}\left( F_{02} + 2 \, F_{11} \, \eigenvL_1 + 2 \, F_{20} \, \eigenvL_1^{2} \right) \label{EigPert-Diag}\\ \eigenvL_3 &=& -{1 \over 3! \, F_{10}} \left( F_{01} + 3 \, F_{12} \, \eigenvL_1 + 6 \, F_{11} \, \eigenvL_2 \right. \ceq \qquad\qquad \left. + \, 3 \, F_{21} \, \eigenvL_1^{2} + 6 \, F_{20} \, \eigenvL_1 \, \eigenvL_2 + F_{30} \, \eigenvL_1^{3} \right) \,. \nnu \eea As shown in \refref{QCcourse}, $F_{kl}$ can be computed from explicit cycle expansions. However, in numerical calculations we find it more expedient to proceede by first expressing the {\fd} $F$ in terms of the cumulants. The traces of $\Lmat{}^n$ evaluated by \refeq{BtoL} yield a series in $\sigma^j$, and the $\sigma^j$ coefficients $Q_{n j}$ in the cumulant expansion \beq F=\det(1-z\Lop) = 1-\sum_{n=1}^{\infty}\sum_{j=0}^{\infty}Q_{n j}z^n\sigma^j \label{qum} \eeq are then obtained recursively from the traces, as in \refeq{Fd-cyc-exp-Diag}: \beq Q_{n j} = \frac{1}{n}\left(C_{n j} - \sum_{k=1}^{n-1}\sum_{l=0}^{j}Q_{k,j-l}C_{n-k,l} \right) \,. \ee{Q-Sig-expan} This gives $F = F(z = 1/\eigenvL \, , \sigma)$ and the partial derivatives $F_{kl}$ can be found. Substituted in \refeq{EigPert-Diag} they yield the perturbative corrections to the eigenvalues. The above calculations can be efficiently done by manipulating formal Taylor series. \section{Numerical tests} Here we continue the calculations of the eigenvalue corrections described in \refrefs{noisy_Fred,conjug_Fred}, where more details and discussion may be found. We test our perturbative expansion on the repeller of the 1-dimensional map \beq f(x)=20\left(\frac{1}{16}-\left(\frac{1}{2}-x\right)^4\right) \,. \ee{testQuartic} This repeller is a clean example of an ``Axiom~$A$'' expanding system of bounded nonlinearity and complete binary symbolic dynamics, for which the deterministic \evOper\ eigenvalues converge super-exponentially with the cycle length\rf{Rugh92}. We start the numerical calculations by determining all prime cycles up to a given length. For each prime cycle $p$ we compute the truncated evolution matrix $\Lmat{p}$ and its repetitions $\Lmat{p}^r$ to the given order in $\sigma$, and evaluate the traces \refeq{tracenp}. For the map at hand we find that truncations of size [$16\times 16$] suffice to achive double precision accuracy for most cycles. However, as the short orbits are less unstable, they require larger matrix truncations in order to attain the same precision, and we employ a [$28\times 28$] truncation for the $2$-cycles, and a [$34\times 34$] truncation for the fixed points. With the coefficients in the traces expansion \refeq{tracenp} evaluated numerically, the cumulants and the perturbative eigenvalue corrections follow from \refeq{Q-Sig-expan} and \refeq{EigPert-Diag}. In case at hand, a good first approximation is obtained already at $n=2$ level, using only 3 prime cycles, and $n=6$ (23 prime cycles in all) is in this example sufficient to exhaust the limits of double precision arithmetic. \begin{figure}[hbt] \centerline{\psfig{figure=qum1.eps,width=9cm}} \caption{The perturbative corrections \refeq{Q-Sig-expan} to the cumulants $Q_{n j}$ plotted as a function of cycle length $n$ (for perturbation orders $j=0,2,4,6,8$) all exhibit super-exponential convergence.} \label{qumfig} \end{figure} The size of the cumulants is indicated in \reffig{qumfig}, and the perturbative corrections to the leading eigenvalue of the weak-noise \evOper\ are given in \reftab{tabPert}. Encouragingly, the value of $\eigenvL_6 = 2076.47\ldots$ computed here is not wildly different to our previous numerical estimate\rf{conjug_Fred} of 2700. Both the cumulants and the eigenvalue corrections exhibit a super-exponential convergence with the truncation cycle length $n$. The super-exponential convergence has been proven for the deterministic, $\eigenvL_0$ part of the eigenvalue\rf{Rugh92}, but the proof has not been extended to the stochastic evolution operators. We have chosen to test the formalism on this simple example, as here we are in a fortunate situation that the escape rate for arbitrary noise strength $\sigma$ can be calculated numerically by other methods to a rather high accuracy. For example, one can discretize the stochastic kernel on a spatial lattice\rf{noisy_Fred} and determine numerically the leading eigenvalue. The perturbative result in terms of periodic orbits and the weak noise corrections is compared to the eigenvalue computed by the numerical lattice discretizationin \reffig{nufig2}, with the absolute difference between the numerical and the $m$th order perturbative results plotted. We see that the perturbative result $\eigenvL(m,\sigma)=\sum_{k=0}^{m/2} \eigenvL_{2k}\sigma^{2k}$ indeed improves as more perturbative terms are added. \begin{figure}[hbt] \centerline{\psfig{figure=nufig2.eps,width=9cm}} \caption{ The difference between the numerical and perturbative eigenvalue $|\eigenvL(\sigma) - \eigenvL(m,\sigma)|$. The plateau at $10^{-7}$ is a numerical artifact due to the limited accuracy of the lattice discretization calculation. \label{nufig2}} \end{figure} \section{Summary and outlook} In this paper we study evolution of a classical dynamical system with additive noise. In the limit of weak noise the traces of the corresponding {\evOper} are approximated by sums of local traces computed on periodic orbits. Here we present a new, computationally efficient technique for evaluation of these local traces based on a matrix representation of the {\evOper}, and show that method is powerful enough to enable us to compute 2 more orders of perturbation theory. The local matrix representation can be interpreted as follows. Substituting \refeq{tracenp} into \refeq{det-tr-Diag} we obtain \bea \left. \det(1 - z \Lop) \right|_{\mbox{\tiny saddles}} = \prod_{p} \det(1 - z^{n_p} \Lmat{p}) \,. \label{fredholmprod} \eea In other words, in the saddle-point approximation the spectrum of the {\em global} {\evOper} $\Lop$ is in this approach pieced together from the {\em local} spectra computed cycle-by-cycle on neighborhoods of individual prime cycles with periodic boundary conditions. Vattay\rf{VatBS} was first to formulate the $\hbar$ corrections to the semi-classical Gutzwiller theory in terms of local spectra. Here we have shown that also the stochastic flows can be suspended on the skeleton of classical periodic orbits in this way. With so many orders of perturbation theory, we are now poised to address the issues raised by the asymptotic series nature of perturbative expansions. We can now hope to resum the series to all orders, making use of techniques such as the Borel resummation, the asymptotic expansions of general integrals of saddlepoint type, and asymptotics beyond all orders\rf{Dingle}. All of this is beyond the scope of the present paper, and we defer a full discussion of asymptotics to a forthcoming paper\rf{asym_Fred}. \section{Acknowledgements} G.V. and G.P. gratefully acknowledges the financial support of the Hungarian Ministry of Education, FKFP 0159/1997, OMFB, OTKA T25866/F17166. G.V. thanks Bruno Eckhardt the cordial hospitalty at the Department of Physics of the Philipps-Universit\"at Marburg and the Humboldt Fundation for support. G.P. thanks the EU network ``Pattern formation, noise and spatio-temporal chaos in complex systems'', TMR contract ERBFMRXCT960085, for partial support. N.S. is supported by the Danish Research Academy Ph.D. fellowship.
\section{Overview of the Cosmic Microwave Background} In 1948, Alpher \& Hermann \cite{AlpHer49} realized that if light elements were produced in a hot big bang, as Gamow and others had suggested \cite{Gam46}, then the Universe today should have a temperature of about 5 K. When Penzias \& Wilson discovered an anomalous background in 1964, consistent with a blackbody spectrum at a temperature of $\sim3$ K \cite{PenWil65}, Dicke and his collaborators immediately recognized it as the radiation associated with this nonzero cosmological temperature \cite{Dicetal65}. Subsequent observations that confirm a remarkable degree of isotropy (apart from a dipole \cite{CorWil76,SmoGorMul77}, which can be interpreted as our motion of $627\pm22$ km~s$^{-1}$ with respect to the blackbody rest frame \cite{Smoetal91,Smoetal92,Kogetal93,Fixetal94}) suggest an extragalactic origin for this cosmic microwave background (CMB). Strong upper limits to any angular cross-correlation between the CMB temperature and the extragalactic X-ray background intensity \cite{Kneetal97,BouCriTur98} suggest that the CMB comes from redshifts greater than those ($z\simeq2-4$) probed by the active galactic nuclei and galaxy clusters that produce the X-ray background. This evidence, as well as the exquisite blackbody spectrum of the CMB \cite{Matetal94,Wrietal94,Fixetal96}, further supports the notion that this radiation is the cosmological blackbody postulated by Alpher \& Hermann. Although they have a Planck spectrum, CMB photons are not in thermal equilibrium. The mean free path for scattering of photons in the Universe must be huge, or else we would not see galaxies and quasars out to distances of thousands of Mpc.\footnote{Mpc$=3.3\times10^6$~light years$=3.09\times10^{24}$~cm.} So where did these photons come from? At early times ($t\la10^5$ y; redshifts $z\ga1000$), the temperature of the Universe exceeded an eV, so the Universe consisted of a plasma of free electrons and light nuclei. CMB photons were tightly coupled to this plasma via Thomson scattering from the free electrons. At a redshift of $z\simeq1000$, the temperature dropped below a few eV, and electrons and nuclei combined to form atoms. At this point, photons ceased interacting. A detailed analysis of ``recombination'' and the almost simultaneous (although slightly later) decoupling of photons shows that CMB photons last scattered near a redshift of $z\simeq1100$ \cite{JonWys76,KolTur90,Pee93}. When we look at these CMB photons coming to us from all directions in the sky, we are therefore looking directly at a spherical surface in the Universe that surrounds us at a distance of $\sim10^4$ Mpc, as it was when the Universe was only about 300,000 years old. The temperature of the CMB is found to be the same, to roughly one part in $10^5$, in every direction on the sky. This remarkable isotropy poses a fundamental conundrum for the standard big-bang theory. When these photons last scattered, the size of a causally connected region of the Universe was roughly 300,000 light years, and such a region subtends an angle of only one degree on the sky. Thus, when we look at the CMB, we are looking at roughly 40,000 causally disconnected regions of the Universe. How is it, then, that each of these has the same temperature to one part in $10^5$? This is the well-known isotropy, homogeneity, or horizon problem. Another fundamental question in cosmology today is the origin of the large-scale structure of the galaxy distribution. The simplest and most plausible explanation is that the observed inhomogeneities grew from tiny density perturbations in the early Universe via gravitational instability. Mass from underdense regions is drawn towards overdense regions, and in this way, small primordial perturbations are amplified into the structure we see in the Universe today. New support for this hypothesis was provided by the Cosmic Background Explorer (\COBE) detection of temperature differences in the CMB of roughly one part in $10^5$ \cite{Smoetal90}. Heuristically, density perturbations induce gravitational-potential perturbations at the surface of last scatter; photons that arrive from denser regions climb out of deeper potential wells and thus appear redder than those from underdense regions (the Sachs-Wolfe effect \cite{SacWol67}). Thus, the temperature fluctuations seen with \COBE\ provide a snapshot of the tiny primordial perturbations that gave rise to the large-scale structure we see in the Universe today. But this raises a second question: If large-scale structure grew via gravitational infall from tiny inhomogeneities in the early Universe, where did these primordial perturbations come from? Before \COBE, there was no shortage of ideas for the origin of large-scale structure, and, quite remarkably, all causal mechanisms for producing primordial perturbations have come from new ideas in particle theory: primordial adiabatic perturbations from inflation \cite{GutPi82,Haw82,Lin82b,Sta82,BarSteTur83}, late-time phase transitions \cite{Was86,HilSchFry89}, a loitering Universe \cite{SahFelSte92}, scalar-field ordering \cite{Vil82,Pre80}, topological defects \cite{ZelKozOku74,Kib76} (such as cosmic strings \cite{Zel80,Vil81,SilVil84,Tur85}, domain walls \cite{HilSchFry89,PreRydSpe89}, textures \cite{Tur89,Tur91}, or global monopoles \cite{BarVil89,BenRhi91}), superconducting cosmic strings \cite{Wit85,WitThoOst86}, isocurvature axion perturbations \cite{SecTur85,TurWil91,AxeBraTur83,SteTur83,KofLin87,Lyt90,Lin91}, etc. However, after \COBE, primordial adiabatic perturbations (perturbations to the total density with equal fractional number-density perturbations in each species in the Universe) seem to provide the only workable models. Such perturbations are produced naturally during inflation, a period of exponential expansion in the early Universe driven by the vacuum energy associated with some new scalar field \cite{Gut81,Lin82a,AlbSte82}. With adiabatic perturbations, hotter regions at the surface of last scatter are embedded in deeper potential wells, so the reddening due to the gravitational redshift of the photons from these regions partially cancels the higher intrinsic temperatures. When normalized to the amplitude of density perturbations indicated by galaxy surveys, alternative models generically produce a larger temperature fluctuation than that measured by \COBE\ \cite{KodSas86,EfsBon87,JafSteFri94}. Recently, more detailed calculations of the expected CMB-anisotropy amplitude have led proponents of topological defects, the primary alternative to inflation, to concede that these models have difficulty accounting for the origin of large-scale structure \cite{PenSelTur97,AlbBatRob97,Alletal97}. Although inflation now seems to provide the best candidate for the origin of large-scale structure, the primary attraction of inflation was originally that it provided (and still does) the best (if not only) solution of the horizon problem. For these reasons, inflation has taken center stage in cosmology. Although inflation was for a long time speculative physics beyond the realm of experimental tests, we are now entering a new era in which the predictions of inflation will be tested with unprecedented precision by CMB measurements. The primary focus of this article is therefore to review the predictions of inflation and how they will be tested with the CMB. Although inflation currently seems to provide the most promising paradigm for the origin of large-scale structure, it is not yet well established. Moreover, although the simplest topological-defect models seem to be ruled out, it is still certainly plausible that some more involved models may be able to account for large-scale structure. We therefore review the CMB predictions of topological-defect models. We also discuss a number of other promising links between the CMB and particle physics that do not necessarily have to do with the origin of structure, e.g.\ dark matter, neutrino properties, decaying particles, cosmological magnetic fields from early-Universe phase transitions, parity violation, gravity theories, time variation of fundamental parameters, and baryogenesis scenarios. We are unfortunately unable to cover the larger bodies of excellent work on the CMB in general, nor on the intersections between particle physics and cosmology more generally. Fortunately, a number of excellent reviews cover those subjects, to which we cannot do justice here. Lyth and Riotto \cite{LytRio99} review particle-physics models of inflation; Liddle and Lyth discuss structure formation in inflation-inspired cold-dark-matter models. Lidsey et al \cite{Lidetal97} review the production of density perturbations and reconstruction of the inflaton potential from the power spectra of density perturbations and gravitational waves. White, Scott, and Silk \cite{WhiScoSil94} review the CMB and structure formation, and Hu and White \cite{HuWhi97b} provide a brief review of the theory of CMB polarization. Finally, see References \cite{SunZel80,Rep95,Bir99} for reviews of the Sunyaev-Zeldovich effect (the scattering of CMB photons from hot gas in clusters of galaxies), an intriguing and potentially very important probe of the physics of clusters. \section{Cosmic Microwave Background Observables} \label{sec:cmbobservables} \subsection{The Frequency Spectrum} \label{sec:freqspectrum} Standard cosmology predicts the CMB frequency spectrum to be that of a perfect blackbody, \begin{equation} S(\nu;T) = {2 h c^2 \nu^3 \over e^x-1}, \label{eq:blackbody} \end{equation} where $x=hc\nu/kT$, $h$ is Planck's constant, $c$ is the velocity of light, $\nu$ is the frequency, $k$ is Boltzmann's constant, and $T$ is the temperature. Of the infinitude of possible distortions to this spectrum, two common forms often considered in the literature---Bose-Einstein and Compton distortions---could arise from basic physical processes before recombination. If photons are released into the Universe from some nonthermal process (e.g.\ decay of a massive particle) when the temperature of the Universe exceeds roughly 1 keV (redshifts $z\ga10^6$ when the age of the Universe is $t\la10^7$ sec), they will come into complete thermal equilibrium with the photons in the primordial plasma. More precisely, they attain kinetic equilibrium through Compton scattering, double Compton scattering, and bremsstrahlung, and they attain chemical equilibrium (chemical potential $\mu=0$) because the rate for photon-number-changing processes (e.g.\ $\gamma\gamma\rightarrow\gamma\gamma\gamma$) that maintain a chemical potential $\mu=0$ exceeds the expansion rate. Therefore, if any electromagnetic energy is released into the Universe at such early times, it will have no observable effect on the CMB. However, if photons are released at later times (but still before recombination), they can distort the CMB frequency spectrum \cite{SunZel80,DanDeZ77,Lig81,DanDeZ82,SarCoo83,FukKaw90,BurDanDeZ91a,BurDanDeZ91b,HuSil93b}. \subsubsection{Bose-Einstein distortion} Nonthermal photons produced in the redshift range $10^5 \mathrel{\mathpalette\fun <} z \mathrel{\mathpalette\fun <} 3\times10^6$ (temperatures $T\simeq 0.1-1$ keV and ages $t\simeq10^{7-9}$ sec) can still attain kinetic equilibrium, but they will not attain chemical equilibrium, as interactions that change the photon number occur less rapidly than the expansion rate. If electromagnetic energy is released at these times, the CMB frequency dependence will be that of a Bose-Einstein gas with a nonzero chemical potential, \begin{equation} S_\mu(\nu;T,\mu) = {2 h c^2 \nu^3 \over e^{x+\mu}-1}, \label{eq:boseeinstein} \end{equation} where $\mu$ is the (dimensionless) chemical potential. The Far Infrared Absolute Spectrophotometer (FIRAS) result for $\mu$ is $\mu=-1\pm10\times10^{-5}$ or a 95\% confidence-level upper limit of $|\mu| < 9\times10^{-5}$ \cite{Fixetal96}. It is possible that values of $\mu$ as small as $10^{-6}$ could be probed by a future satellite mission \cite{Shaetal95}. \subsubsection{Compton distortion} If photons are released at later times ($z\la10^5$) but still before recombination ($z\simeq1100$; temperatures $T\simeq 1-100$ eV and times $t\simeq10^{9-13}$ sec), they do not have enough time to come to either thermal or kinetic equilibrium and wind up producing a ``Compton distortion'' of the form \begin{equation} S_y(\nu;T,y)={2 h c^2 \nu^3 \over e^x-1}\left(1 + y x {1 \over 1-e^{-x}} \left[x\coth(x/2)-4\right]\right), \label{eq:comptony} \end{equation} to linear order in $y$ (the Kompaneets or ``Compton-$y$'' parameter). If some CMB photons were rescattered after recombination by a hot intergalactic gas, this would also produce a Compton-$y$ distortion. The FIRAS result for this type of distortion is $y=-1\pm6\times10^{-6}$ or an upper limit of $|y|<15\times10^{-6}$ at the 95\% confidence level \cite{Fixetal96}. The consensus among the experimentalists we have surveyed seems to be that it would be difficult to improve on this limit. \subsection{Temperature and Polarization Power Spectra} The primary aim of forthcoming CMB satellite experiments, such as NASA's Microwave Anisotropy Probe (MAP) \cite{MAP} and the European Space Agency's Planck Surveyor \cite{Planck}, will be to map the temperature $T({\bf \hat n})$ of the CMB and its linear polarization, described by Stokes parameters $Q({\bf \hat n})$ and $U({\bf \hat n})$, as functions of position ${\bf \hat n}$ on the sky. Several temperature-polarization angular correlation functions, or equivalently, power spectra, can be extracted from such maps. These quantities can be compared with detailed predictions from cosmological models. \subsubsection{Harmonic analysis for temperature anisotropies and polarization} {\it Temperature Anisotropies}\\ The temperature map can be expanded in spherical harmonics, \begin{equation} {T({\bf \hat n}) \over T_0} = 1 + \sum_{lm} a^{\rm T}_{(lm)} Y_{(lm)}({\bf \hat n}), \label{Texpansion} \end{equation} where the mode amplitudes are given by \begin{equation} a^{\rm T}_{(lm)}={1\over T_0}\int d{\bf \hat n}\,T({\bf \hat n})\,Y_{(lm)}^*({\bf \hat n}); \label{Talms} \end{equation} this follows from orthonormality of the spherical harmonics. Here, $T_0=2.728\pm0.002$ K is the cosmological mean CMB temperature \cite{Fixetal96}. {\it Linear polarization}\\ The Stokes parameters (where $Q$ and $U$ are measured with respect to the polar ${\bf \hat\theta}$ and azimuthal ${\bf \hat \phi}$ axes) are components of a $2\times2$ symmetric traceless tensor with two independent components, \begin{equation} {\cal P}_{ab}({\bf \hat n})={1 \over 2} \left( \begin{array}{cc} \vphantom{1\over 2}Q({\bf \hat n}) & -U({\bf \hat n}) \sin\theta \\ -U({\bf \hat n})\sin\theta & -Q({\bf \hat n})\sin^2\theta \\ \end{array} \right), \label{eq:whatPis} \end{equation} where the subscripts $ab$ are tensor indices, and $Q$ and $U$ are given in temperature units. Just as the temperature is expanded in terms of spherical harmonics, the polarization tensor can be expanded \cite{KamKosSte97a,KamKosSte97b,SelZal97,ZalSel97}, \begin{equation} {{\cal P}_{ab}({\bf \hat n})\over T_0} = \sum_{lm} \left[ a_{(lm)}^{{\rm G}}Y_{(lm)ab}^{{\rm G}}({\bf \hat n}) +a_{(lm)}^{{\rm C}}Y_{(lm)ab}^{{\rm C}}({\bf \hat n}) \right], \label{eq:Pexpansion} \end{equation} in terms of tensor spherical harmonics, $Y_{(lm)ab}^{\rm G}$ and $Y_{(lm)ab}^{\rm C}$. It is well known that a vector field can be decomposed into a curl (C) and a curl-free (gradient) (G) part. Similarly, a $2\times2$ symmetric traceless tensor field can be decomposed into a tensor analogue of a curl and a gradient part; the $Y_{(lm)ab}^{\rm G}$ and $Y_{(lm)ab}^{\rm C}$ form a complete orthonormal basis for the ``gradient'' (i.e.\ curl-free) and ``curl'' components of the tensor field, respectively.\footnote{Our G and C are sometimes referred to as the ``scalar'' and ``pseudo-scalar'' components \cite{Ste96}, respectively, or with slightly different normalization as E and B modes \cite{ZalSel97} (although these should not be confused with the radiation's electric- and magnetic-field vectors).} Lengthy but digestible expressions for the $Y_{(lm)ab}^{\rm G}$ and $Y_{(lm)ab}^{\rm C}$ are given in terms of derivatives of spherical harmonics and also in terms of Legendre functions in Reference \cite{KamKosSte97b}. The mode amplitudes in Equation \ref{eq:Pexpansion} are given by \begin{eqnarray} a^{\rm G}_{(lm)}&=&{1\over T_0}\int d{\bf \hat n}\,{\cal P}_{ab}({\bf \hat n})\, Y_{(lm)}^{{\rm G} \,ab\, *}({\bf \hat n}),\cr a^{\rm C}_{(lm)}&=&{1\over T_0}\int d{\bf \hat n}\,{\cal P}_{ab}({\bf \hat n})\, Y_{(lm)}^{{\rm C} \, ab\, *}({\bf \hat n}), \label{eq:Amplitudes} \end{eqnarray} which can be derived from the orthonormality properties of these tensor harmonics \cite{KamKosSte97b}. Thus, given a polarization map ${\cal P}_{ab}({\bf \hat n})$, the G and C components can be isolated by first carrying out the transformations in Equation \ref{eq:Amplitudes} to the $a^{\rm G}_{(lm)}$ and $a^{\rm C}_{(lm)}$ and then summing over the first term on the right-hand side of Equation \ref{eq:Pexpansion} to get the G component and over the second term to get the C component. \subsubsection{The power spectra} \label{sec:powerspectra} Theories for the origin of large-scale structure predict that the mass distribution in the Universe is a single realization of a statistically isotropic random field. In other words, the Fourier components $\tilde\delta({\vec k})$ of the fractional density perturbation $\delta({\vec x}) = [\rho({\vec x})-\bar\rho]/\bar\rho$ [where $\rho({\vec x})$ is the density at comoving position ${\vec x}$ and $\bar\rho$ is the universal mean density] are random variables that have expectation values $\VEV{\tilde\delta({\vec k})}=0$ and covariance given by \begin{equation} \VEV{\tilde \delta({\vec k}) \tilde \delta({\vec k}')} = (2\pi)^3 \, \delta_D({\vec k}+{\vec k}') \, P_s(k). \label{eq:scalarspectrum} \end{equation} Here $P_s(k)$ is the scalar power spectrum (so called because density perturbations produce scalar perturbations to the spacetime metric), or alternatively, the power spectrum for the spatial mass distribution. Statistical isotropy demands that the power spectrum depends only on the amplitude (rather than orientation) of ${\vec k}$. Because the temperature perturbation and polarization of the CMB are due to density perturbations, the $a_{(lm)}^{\rm X}$ must be random variables with zero mean, $\VEV{a_{(lm)}^{\rm X}}=0$, and covariance, \begin{equation} \VEV{\left(a_{(l'm')}^{\rm X'} \right)^* a_{(lm)}^{\rm X}} = C_l^{{\rm XX}'} \delta_{ll'}\delta_{mm'}, \label{eq:covariance} \end{equation} for ${\rm X},{\rm X}' = \{{\rm T,G,C}\}$. The statistical independence of each $lm$ mode (i.e.\ the presence of the Kronecker deltas) is a consequence of statistical isotropy. The scalar spherical harmonics $Y_{(lm)}$ and the gradient tensor spherical harmonics $Y_{(lm)}^{\rm G}$ have parity $(-1)^l$, whereas the curl tensor spherical harmonics $Y_{(lm)}^{\rm C}$ have the opposite parity, $(-1)^{l+1}$. Thus, $C_l^{\rm TC}=C_l^{\rm GC}=0$ if the physics that gives rise to temperature anisotropies and polarization is parity-invariant. In this case, the two-point statistics of the CMB temperature-polarization map are completely specified by the four sets of moments, $C_l^{\rm TT}$, $C_l^{\rm TG}$, $C_l^{\rm GG}$, and $C_l^{\rm CC}$. Nonzero $C_l^{\rm TC}$ or $C_l^{\rm GC}$ would provide a signature of cosmological parity breaking. \subsubsection{Angular correlation functions} The temperature two-point correlation function is \begin{equation} C^{\rm TT}(\alpha) = \VEV{ {\Delta T({\bf \hat m})\over T} {\Delta T({\bf \hat n}) \over T}}_{{\bf \hat m} \cdot {\bf \hat n}=\cos\alpha}, \label{eq:correlationfns} \end{equation} where the average is over all pairs of points on the sky separated by an angle $\alpha$. It can be written in terms of the temperature power spectrum (i.e.\ the $C_l^{\rm TT}$) as \begin{equation} C^{\rm TT}(\alpha) = \sum_l {2 l +1 \over 4 \pi} C^{\rm TT}_l P_l(\cos\alpha), \label{eq:corrpower} \end{equation} where $P_l(\cos\alpha)$ are Legendre polynomials. Likewise, orthonormality of Legendre polynomials guarantees that the multipole coefficients, $C_l^{\rm TT}$, can be written as integrals over the product of the correlation function and a Legendre polynomial. Thus, specification of the correlation function is equivalent to specification of the power spectrum, and vice versa. CMB theorists and experimentalists have now adopted the convention of showing model predictions and presenting experimental results as power spectra ($C_l$) rather than as correlation functions, and we subsequently stick to this convention. Auto- and cross-correlation functions for the Stokes parameters and temperature-polarization cross-correlation functions can also be defined and written in terms of the polarization and temperature-polarization power spectra \cite{KamKosSte97b}, but we do not list them here. In practice, the temperature intensity (or polarization) can never be determined at a given point on the sky; it can only be measured by a receiver of some finite angular resolution (referred to as a ``finite beamwidth''). Thus, the correlation function in Equation~\ref{eq:corrpower} cannot be measured; one can only measure a smoothed version. Likewise, a finite beamwidth $\theta_{\rm fwhm}$ (at full-width half maximum) limits determination of the power spectrum to multipole moments $l\mathrel{\mathpalette\fun <} 200\,(\theta_{\rm fwhm}/1^\ifmmode ^\circ\, \else $^\circ\,$ \fi)^{-1}$. The Differential Microwave Radiometer (DMR) experiment \cite{Smoetal92} aboard \COBE\ produced the first map of the temperature of the CMB. The receivers also provided some information on the polarization, but the sensitivity was not sufficient to detect the signal expected in most cosmological models. Measurements of the CMB intensity were made at three different frequencies (31.5, 53, and 90 GHz) near the blackbody peak to disentangle the possible contribution of foreground contaminants (e.g.\ dust or synchrotron emission) {}from the Galaxy, as these would have a frequency spectrum that differs from a blackbody. The DMR beamwidth was $7^\ifmmode ^\circ\, \else $^\circ\,$ \fi$, so the temperature power spectrum was recoverable only for $l\mathrel{\mathpalette\fun <} 15$. MAP, scheduled for launch in the year 2000, will map the sky with an angular resolution better than $0.3^\ifmmode ^\circ\, \else $^\circ\,$ \fi$ ($l\la700$). MAP should have sufficient sensitivity to see the polarization, although probably not enough to map the polarization power spectra precisely (the polarization is expected to be roughly an order of magnitude smaller than the temperature anisotropy). The Planck Surveyor, scheduled for launch around 2007, will map the temperature and polarization with even finer angular resolution (out to $l\la2000-3000$). Its sensitivity should be sufficient to map the polarization power spectra expected {}from density perturbations (discussed below) with good precision. \subsection{Gaussianity} Angular three-point and higher $n$-point temperature correlation functions can be constructed analogously to the two-point correlation functions in Equation~\ref{eq:correlationfns}. Fourier analogs of higher-order correlation functions can be defined. In particular, the temperature bispectrum $B(l_1,l_2,l_3)$ is the $l$-space version of the temperature three-point correlation function. It is defined by \begin{equation} \VEV{a^{\rm T}_{(l_1 m_1)} a^{\rm T}_{(l_2 m_2)} a^{\rm T}_{(l_3 m_3)}} = \left( \begin{array}{ccc} l_1 & l_2 & l_3 \\ m_1 & m_2 & m_3 \\ \end{array} \right) B(l_1,l_2,l_3), \label{eq:bispectrum} \end{equation} where the array is the Wigner $3j$ symbol. This particular $m$ and $l$ dependence follows from the assumption of statistical isotropy. Closely related statistics include the skewness and kurtosis (respectively, the three- and four-point correlation functions at zero lag) \cite{ScaVit91,LuoSch93,Luo94} and higher cumulants \cite{FerMagSil97}. As discussed further below, inflationary models predict the primordial distribution of perturbations to be perfectly (or very nearly) Gaussian. Gaussianity dictates that all the odd-$n$ $n$-point correlation functions vanish and that for even $n$, the higher $n$-point correlation functions can be given in terms of the two-point correlation function. Numerous other statistical tests of CMB Gaussianity have also been proposed, including (but not limited to) topology of temperature contours \cite{Col88,Gotetal90,Kogetal96} and the related Minkowski functionals \cite{WinKos97,SchGor98}, temperature peak statistics \cite{BonEfs87,VitJus87,Kogetal96}, Fourier space patterns \cite{FerMag97,LewAlbMag99}, and wavelet analysis \cite{PanValFan98,HobJonLas98}. \begin{figure} \centerline{\psfig{file=potentia.ps,width=32pc}} \caption{Two toy models for the inflationary potential.} \vskip-12pt \label{fig:potentials} \end{figure} \section{Predictions of Inflation} \label{sec:inflationpredictions} If the energy density of the Universe is dominated by matter or radiation, then the expansion of the Universe is decelerating. If so, the horizon grows more rapidly than the scale factor. In such a Universe, objects that are now beyond our horizon and therefore inaccessible to us will eventually enter the horizon and become visible. Thus, the observable Universe contains more information and is more complicated at later times. Inflation postulates the existence of a period of accelerated expansion in the early Universe. In such a Universe, the scale factor grows more rapidly than the horizon. Thus, objects currently visible to any given observer will eventually exit that observer's horizon (in much the same way that objects that fall into a black hole disappear when they pass through the black hole's event horizon). A period of accelerated expansion therefore makes the Universe simpler and smoother. As we now discuss, this accelerated expansion also generically drives the observable Universe to be flat and provides a mechanism for producing primordial density perturbations and gravitational waves. \subsection{Scalar-Field Dynamics} \label{sec:scalarfields} Inflation supposes the existence of some new scalar field $\phi$ (the ``inflaton''), with a potential $V(\phi)$ that roughly resembles either of those shown in Figure \ref{fig:potentials}. The shape is not particularly important. All we require is that, at some time in the history of the Universe, the field is displaced from the minimum of the potential, and then it rolls slowly How slowly is slowly enough? This is determined by the Friedmann equation, \begin{equation} H^2 \equiv \left( {\dot a \over a}\right)^2 ={8 \pi G\rho \over 3} - {k\over a^2} = {8 \pi G \over 3} \left({1\over2} \dot\phi^2 + V(\phi) \right) -{k \over a^2}, \label{eq:friedmann} \end{equation} which governs the time $t$ dependence of the scale factor $a(t)$ of the Universe (the dot denotes derivative with respect to time), as well as the scalar-field equation of motion, \begin{equation} \ddot \phi + 3H \dot\phi +V'(\phi)=0, \label{eq:eom} \end{equation} where $V'=dV/d\phi$. In Equation~\ref{eq:friedmann}, $\rho$ is the energy density of the Universe, which is assumed to be dominated by the inflaton potential-energy density $V(\phi)$ and kinetic-energy density $\dot\phi^2/2$. The term $k/a^2$ is the curvature term, and $k>0$, $k<0$, or $k=0$ for a closed, open, or flat Universe, respectively. Note that the expansion acts as a friction term for the scalar-field equation of motion in Equation~\ref{eq:eom}. If \begin{equation} \epsilon\equiv {m_{\rm Pl}^2 \over 16 \pi} \left({ V' \over V} \right)^2 \ll 1, \label{eq:epsilon} \end{equation} and \begin{equation} \eta \equiv {m_{\rm Pl}^2 \over 8\pi} \left[{ V'' \over V} - {1 \over 2} \left({V' \over V} \right)^2 \right] \ll 1, \label{eq:eta} \end{equation} where $m_{\rm Pl}=1.22 \times 10^{19}$ GeV is the Planck mass, then the field rolls slowly enough so that the requirement for acceleration [$p<-\rho/3$, where $p= (1/2)\dot\phi^2-V$ is the pressure and $\rho= (1/2)\dot\phi^2 +V$ is the energy density] is satisfied. (Note that the definitions of $\epsilon$ and especially of $\eta$ may differ in some papers.) The identity of the inflaton remains a mystery. It was originally hypothesized to be associated with a Higgs field in grand unified theories, but it may also have something to do with Peccei-Quinn symmetry breaking, a dilaton field, electroweak-symmetry breaking \cite{KnoTur93}, some new pseudo-Nambu-Goldstone symmetry \cite{FreFriOli90,Adaetal93}, supersymmetry \cite{RanSolGut96}, or some other new physics. As discussed below, the primary predictions of slow-roll inflation do not depend on the details of the physics responsible for inflation but rather on some gross features that are easily quantified. \subsection{The Geometry} Given any inflationary potential $V(\phi)$, the equations of motion in Equations \ref{eq:eom} and \ref{eq:epsilon} can be solved numerically, if not analytically. Heuristically, during inflation, the potential $V(\phi)$ is roughly constant, and $\dot\phi^2 \ll V(\phi)$. If the curvature term is appreciable initially, it rapidly decays relative to the potential term as the Universe expands, and the solution for the scale factor approaches an exponential, $a(t) \propto e^{-Ht}$. If $k$ is nonzero initially, the curvature term is then driven exponentially to zero during the inflationary epoch. In other words, if the duration of inflation is sufficiently long to place the observable Universe in a causally connected pre-inflationary patch, then the curvature radius is generically inflated to an exponentially (and unobservably) large value. In the language above, any initial nonzero curvature disappears beyond the horizon during accelerated expansion. Thus, the first prediction of slow-roll inflation is that the Universe should be flat today; i.e.\ the total density of all components of matter should sum to the critical density. \subsubsection{``Open inflation''} It is, of course, mathematically possible that inflation did occur but that the inflationary epoch was prematurely terminated \cite{LytSte90,KamSpe94} at just the right time so that the Universe today would be open with density $\Omega_0\simeq0.3$. Such a model requires some additional mechanism (e.g.\ another prior period of inflation and/or some arbitrary new ``feature'' in the inflaton potential) to solve the isotropy problem as well as to produce density perturbations. Several such open-inflation models have recently been constructed \cite{Got82,LytSte90,BucGolTur95,RatPee95,Lin95,CorSpeSta96,HawTur98}, motivated by observations that suggest $\Omega_0\simeq0.3$. The predictions of a scale-invariant spectrum and Gaussian perturbations (discussed below) are the same as in ordinary inflation, but the Universe would be open. We do not find these models even nearly as compelling as the ordinary slow-roll models that produce a flat Universe, although some theorists may disagree. Fortunately, the correct model will not be determined by debate; forthcoming CMB measurements, described below, should distinguish conclusively between these two classes---simple and elegant versus complicated and unappealing---of inflationary models. \subsection{Density Perturbations, Gravitational Waves, and the Inflationary Observables} \subsubsection{Production of density perturbations} Density perturbations are produced as a result of novel quantum-mechanical effects (analogous to the production of Hawking radiation from black holes) that occur in a Universe with accelerating expansion \cite{GutPi82,Haw82,Lin82b,Sta82,BarSteTur83}. This process has been reviewed in detail recently \cite{Lin90,Lidetal97}, so here we review the physics only heuristically. Consider perturbations $\delta\phi({\vec x},t)$ (as a function of comoving position ${\vec x}$) to the homogeneous slowly rolling field $\phi(t)$. These perturbations satisfy a massless Klein-Gordon equation, and the equation of motion for each Fourier mode $\widetilde{\delta\phi}({\vec k})$ is that of a simple harmonic oscillator in an expanding Universe. At sufficiently early times, when the wavelength of any given Fourier mode is less than the Hubble radius $H^{-1}$, it undergoes quantum-mechanical zero-point oscillations. However, if the expansion is accelerating, then the physical wavelength of this comoving scale grows faster than the Hubble radius and eventually becomes larger than $H^{-1}$. At this point, crests and troughs of a given Fourier mode can no longer communicate, and the zero-point fluctuation becomes frozen in as a classical perturbation $\delta\phi({\vec x})$ to the scalar field. Because the inflaton potential is not perfectly flat, this induces perturbations to the density $\delta\rho({\vec x}) =(\partial V/\partial \phi)\delta\phi({\vec x})$. \subsubsection{Production of gravitational waves} Tensor perturbations to the spacetime metric (i.e.\ gravitational waves) satisfy a massless Klein-Gordon equation. A stochastic background of gravitational waves are therefore produced in the same way as classical perturbations to the inflaton are produced \cite{AbbWis84}. Moreover, the power spectra for the inflaton-field perturbations and for the tensor metric perturbations should be identical. The power spectrum of density perturbations is a little different from that for gravitational waves because a density perturbation is related to a scalar-field perturbation by $\delta\rho = (\partial V/\partial\phi)\delta\phi$. The production of scalar and tensor perturbations depends only on the expansion rate during inflation. If the expansion rate were perfectly constant during inflation, it would produce flat scalar and tensor power spectra, $P_s \propto k$ (the ``Peebles-Harrison-Zeldovich'' \cite{PeeYu70,Har70,Zel72} spectrum) and $P_t(k) \propto {\rm constant}$. \subsubsection{Inflationary observables} A constant expansion rate is an oversimplification because the field must in fact be rolling slowly down the potential during inflation. Given any specific functional form for the potential, it is straightforward, using the tools of quantum field theory in curved spacetimes (see e.g.\ \cite{BirDav82}), to predict precisely the functional forms of $P_s(k)$ and $P_t(k)$. Measurement of these power spectra could then be used to reconstruct the inflaton potential \cite{Lidetal97}. Since the field must be rolling fairly slowly during inflation, a good approximation (in most models) can be obtained by expanding about a constant expansion rate. In this slow-roll approximation, the primordial scalar power spectrum is \footnote{Note that this is the spectrum for the primordial perturbations. After the Universe becomes matter dominated at a redshift $z\simeq10^4$, density perturbations grow via gravitational infall, and the growth factor depends on the wave number. Therefore, the power spectrum for matter today is different from the primordial spectrum (it becomes $k^{-4}$ times the primordial spectrum at large $k$), but it is straightforward to relate the primordial and current power spectra.} \begin{equation} P_s = A_s k^{n_s}, \label{eq:scalarPs} \end{equation} and the primordial power spectrum for tensor perturbations is \begin{equation} P_t = A_t k^{n_t}. \label{eq:tensorPt} \end{equation} The amplitudes $A_t$ and $A_s$ and power-law indices $n_s$ and $n_t$ have come to be known as the ``inflationary observables.'' These parameters can provide information on the inflaton potential. In the slow-roll approximation, the power-spectrum indices are roughly constant and given by \cite{Sta85,LidLyt92,Davetal92,LucMatMol92,LidCol92,Tur93a,Adaetal93,Tur93b,Copetal93a,Copetal93b,Copetal94,LidTur94,Lidetal97} \begin{equation} n_s=1-4\epsilon+2\eta, \qquad n_t= -2\epsilon, \label{eq:nsnt} \end{equation} where $\epsilon$ and $\eta$ are the slow-roll parameters given in Equations \ref{eq:epsilon} and \ref{eq:eta}. Strictly speaking, $\epsilon$ and $\eta$ may change (logarithmically with $k$) during inflation \cite{KosTur95,Lidetal97}, but, as the name implies, the field rolls slowly during slow-roll inflation, so the running of the spectral indices is, for all practical purposes, very small. The amplitudes $A_s$ and $A_t$ are similarly fixed by the inflaton potential, but their precise values depend on Fourier conventions and on how the scale factor today is chosen. However, $A_s$ and $A_t$ are proportional, respectively, to the amplitude of the scalar and tensor contributions to $C_2^{\rm TT}$, the quadrupole moment of the CMB temperature. In terms of the slow-roll parameter $\epsilon$ and height $V$ of the inflaton potential during inflation, these CMB observables are \begin{eqnarray} {\cal S} &\equiv & 6\, C_2^{{\rm TT},{\rm scalar}}= 0.66\, {V \over \epsilon m_{\rm Pl}^4} \nonumber\\ {\cal T} &\equiv & 6\, C_2^{{\rm TT},{\rm tensor}}= 9.2 {V \over m_{\rm Pl}^4}. \label{eq:amplitudes} \end{eqnarray} For nearly scale-invariant spectra, \COBE\ fixes $C_2^{\rm TT}=C_2^{\rm TT,scalar}+C_2^{\rm TT,tensor}=(1.0\pm0.1)\times10^{-10}$. In terms of the slow-roll parameters, the tensor-to-scalar ratio is usually defined to be \begin{equation} r \equiv { {\cal T} \over {\cal S}} = 13.7\,\epsilon. \label{eq:TtoS} \end{equation} Comparing Equation~\ref{eq:TtoS} with Equation~\ref{eq:nsnt}, we observer that the observables $n_t$ and $r$ must satisfy a consistency relation, $n_t=-0.145 r$, in slow-roll models. To summarize, slow-roll inflation models (which account for the overwhelming majority of inflation models that appear in the literature) are parameterized by (\textit{a}) the height $V$ of the inflaton potential (i.e.\ the energy scale of inflation), (\textit{b}) $\epsilon$, which depends on the first derivative $V'$ of the inflaton potential, and (\textit{c}) $\eta$, which depends additionally on the second derivative $V''$. The discussion above suggests that because the inflaton is always rolling down the potential, the scalar spectral index must be $n_s<1$. Although this may be true for simple single-field inflation models, more complicated models (e.g.\ with multiple fields or with different potentials) may produce ``blue'' spectra with $n_s>1$ \cite{MolMatLuc94}. \COBE\ alone already constrains $V^{1/4} \leq 2.3\times10^{16}$ GeV. With some additional (but reasonable) modeling, the \COBE\ constraint can be combined with current degree-scale CMB-anisotropy measurements and large-scale-structure observations to reduce this to $V^{1/4} \leq 1.7\times10^{16}$ GeV (e.g.\ \cite{ZibScoWhi99}). The \COBE\ anisotropy implies $n_s=1.1\pm0.3$ if it is attributed entirely to density perturbations \cite{Benetal96}, or $n_t=0.2\pm0.3$ if it is attributed entirely to gravitational waves. Therefore, barring strange coincidences, the \COBE\ spectral index and relations above seem to suggest that if slow-roll inflation is right, then the scalar and tensor spectra must both be nearly scale-invariant ($n_s\simeq 1$ and $n_t\simeq 0$). \subsection{Character of Primordial Perturbations} \subsubsection{Adiabatic versus isocurvature} The density perturbations produced by quantum fluctuations in the inflaton field are referred to as adiabatic, curvature, or isentropic perturbations. These are perturbations to the total density of the Universe, or equivalently, scalar perturbations to the spacetime metric. Adiabaticity further implies that the spatial distribution of each species in the Universe (e.g.\ baryons, photons, neutrinos, dark matter) is the same---that is, the ratio of number densities of any two of these species is everywhere the same. Adiabatic perturbations can be contrasted with primordial isocurvature, or equivalently, pressure or entropy perturbations, perturbations to the ratios between the various species in the Universe (usually in a Universe with a homogeneous total density). Such varying ratios would set up perturbations to the pressure or equivalently to the entropy. When two initially causally-disconnected regions with different pressures come into causal contact, the pressure perturbations push matter around, thus seeding large-scale structure. {\it Axion Inflation}\\ Although adiabatic perturbations are generically produced during inflation, it is also possible to obtain isocurvature perturbations. One example is isocurvature perturbations to an axion density from quantum fluctuations in the Peccei-Quinn field during inflation \cite{SecTur85,TurWil91,AxeBraTur83,SteTur83,KofLin87,Lyt90,Lin91}. As discussed below, comparison of the measured amplitude of CMB anisotropies with the amplitude of galaxy clustering essentially rules out these models. Inflation models that produce both adiabatic and isocurvature perturbations have also been considered \cite{MukZel91,PolSta92,PetPolSta94,StaYok95,GarWan96,SasSte96,KawSugYan96,LinMuk97,MukSte98,Kanetal98a,Kanetal98b}; future experiments should tightly constrain the relative contributions of the two types of perturbations. \subsubsection{Causal versus acausal} Perturbations produced by inflation are said to be ``super-horizon'' or ``acausal.'' This simply refers to the fact that inflation produces a primordial (meaning before matter-radiation equality, when gravitational amplification of perturbations can begin) spectrum of perturbations of all wavelengths, including those much larger than the Hubble length at any given time. This is to be contrasted, for example, with ``causal'' models of structure formation, in which perturbations are generated by causal physics on scales smaller than the horizon. Since inflation implies distance scales much larger than the Hubble length can be within a causally connected pre-inflationary patch, the term acausal is really a misnomer. \subsubsection{(Nearly) Gaussian distribution of perturbations} If the inflaton potential is flat enough for the slow-roll approximation to be valid, then each Fourier mode of the inflaton perturbation evolves independently; that is, the inflaton behaves essentially like an uncoupled massless scalar field. As a result, inflation predicts that the primordial density field is a realization of a Gaussian random field: each Fourier mode is decoupled from every other, and the probability distribution for each is Gaussian. Of course, Gaussianity is an approximation that becomes increasingly valid in the slow-roll limit, in which the inflaton perturbation can be treated as a noninteracting scalar field. Deviations from Gaussianity are generally accepted to be small, and most theorists have adopted a pure Gaussian distribution as a prediction of inflation. However, the deviations in some models might be observable, and if so, would shed light on the physics responsible for inflation \cite{AllGriWis87,SalBonBar89,Sal92,FalRanSre93,Ganetal94,Gan94}. This can be quantified more precisely with the three-point statistic \cite{FerMagGor98}, \begin{equation} I_l^3 \equiv {1 \over (2l+1)^{3/2} (C_l^{\rm TT})^{3/2}} \left( \begin{array}{ccc} l & l & l \\ 0 & 0 & 0 \\ \end{array} \right) B(l,l,l). \end{equation} In slow-roll models with smooth inflaton potentials, the prediction for this quantity is (L Wang, M Kamionkowski, manuscript in preparation) \begin{equation} \sqrt{l(l+1)}I_l^3 = {2 \over m_{\rm Pl}^2} \sqrt{ 3V \over \epsilon} (3\epsilon -2\eta). \label{eq:Il3prediction} \end{equation} Thus, in slow-roll models, one expects $I_0 \mathrel{\mathpalette\fun <} 10^{-6}$ (unless for some unforeseen reason $\epsilon$ is extremely small and $\eta$ is not), too small to be observed. A larger non-Gaussian signal may conceivably arise if there is a glitch in the inflaton potential, but even this non-Gaussianity would be extremely small (L Wang, M Kamionkowski, manuscript in preparation). Detection of nonzero $I_0$ would thus rule out the simplest slow-roll models. Note that the theory predicts that the \textit{primordial} distribution of perturbations is Gaussian. When the Universe becomes matter dominated, and density perturbations undergo gravitational amplification, an initially Gaussian distribution will become non-Gaussian \cite{Pee80}. Such departures from initial Gaussianity have a specific form and may be probed as consistency checks of inflation with galaxy surveys that probe the matter distribution today. \subsection{Brief Overview of Models} A huge literature is devoted to construction of inflationary models (for a comprehensive review, see \cite{LytRio99}). Here we follow the classification of Dodelson et al \cite{DodKinKol97}. Models can be regarded as either large-field, small-field, or hybrid models. Linear models live at the border of large- and small-field models. In large-field (small-field) models, the inflaton moves a distance $\Delta\phi$ that is large (small) compared with the Planck mass during inflation. Hybrid models introduce a second scalar field and allow a broader range of phenomenology. \begin{figure} \centerline{\psfig{file=regions.ps,width=32pc}} \caption{Regions in the $n_s$-$r$ plane occupied by the various classes of inflationary models. (From References \cite{DodKinKol97,Kin98}; their $n$ is our $n_s$.)} \vskip-12pt \label{fig:regions} \end{figure} The models can be distinguished experimentally by the values of $V$, $\epsilon$, and $\eta$ that they predict, or equivalently by the set of $V$, $r$, and $n_s$, as shown in Figure \ref{fig:regions}. Examples of large-field models are single-field models with polynomial potentials, $V(\phi) \propto (\phi/\mu)^p$ (with $p>1$), or in the $p\rightarrow \infty$ limit, exponential potentials, $V(\phi) \propto \exp(\phi/\mu)$.\footnote{Exponential potentials are sometimes referred to as ``power-law inflation,'' since the scale factor grows as a power law during inflation in these models.} The potentials in these models resemble qualitatively the potential shown in Figure \ref{fig:potentials}\textit{a}. These models have $V''>0$ and predict $\epsilon=[p/(p-2)]\eta>0$ and $r\simeq 7 [p/(p+2)](1-n_s)$. Thus, a large tensor amplitude is expected for large $p$ (and therefore for exponential potentials as well) and for a sufficiently large deviation of $n_s$ from unity. Figure \ref{fig:potentials}\textit{b} shows a potential typical of a small-field model. These are the types of potentials that often occur in spontaneous symmetry breaking and can be approximated by $V(\phi) \propto [1-(\phi/\mu)^p]$. These models have $V''<0$. Demanding that the field move a distance that is small compared with $m_{\rm Pl}$ requires that $(\phi/\mu) \ll 1$, and in this limit, $\epsilon=[p/2(p-1)]|\eta| (\phi/\mu)^p \ll \eta$, $\eta<0$, and $r\simeq7(1-n_s) \epsilon/|\eta|$. Note that the slow-roll condition $\phi \ll \mu$ implies that $\epsilon \ll 1$, so $\epsilon \ll \eta$. It thus follows that $n_s \simeq 1+2\eta$, and that the tensor amplitude in these models is expected to be very small. Note that both small- and large-field models predict $n_s<1$. Linear models live at the border of small- and large-field models. They have potentials $V(\phi) \propto \phi$ (i.e.\ they have $V''=0$) and predict $\epsilon\simeq-\eta>0$ and $r\simeq(7/3)(1-n_s)$. Although hybrid models generally involve multiple scalar fields, they can be parameterized by a single-field model with a potential $V\propto [1+(\phi/\mu)^p]$. These models have $\epsilon>0$ and \begin{equation} {\eta \over \epsilon} = {2 (p-1) \over p} \left({\phi \over\mu} \right)^{-p} \left[ 1 + {p-2 \over 2(p-1)} \left( {\phi \over \mu}\right)^p \right] >0. \label{eq:hybrid} \end{equation} Unlike small- or large-field models, hybrid models can (although are not required to) produce blue spectra, $n_s>1$. Although both $r$ and $n_t$ depend only on $\epsilon$ and are thus related, there is no general relation between $r$ and $n_s$ in hybrid models. The tensor amplitude is only constrained to be smaller than it is in exponential models. \section{Cosmic Microwave Background Tests of Inflation} \label{sec:cmbtests} Photons from overdense regions at the surface of last scatter are redder since they must climb out of deeper potential wells (the Sachs-Wolfe effect \cite{SacWol67}). However, this is really only one of a number of physical mechanisms that give rise to temperature perturbations. We have also mentioned that if primordial perturbations are adiabatic, then the gas in deeper potential wells is hotter, and this partially offsets the reddening due to the depth of the potential. Density perturbations induce peculiar velocities, and these also produce temperature perturbations via Doppler shifts. Growth of the gravitational potential near the CMB surface of last scatter can produce temperature anisotropies \cite{HuSug94a} [the early-time integrated Sachs-Wolfe (ISW) effect], and so can the growth of the gravitational potential at late times in a flat cosmological-constant \cite{KofSta86} or open \cite{KamSpe94} Universe (the late-time ISW effect). Modern calculations of the CMB power spectra (the $C_l$) take into account all of these effects. The cosmological perturbation theory underlying these calculations has been reviewed \cite{Bar80,KodSas84,MukFelBra92}, and solution of the Boltzmann equations for the observed angular distribution of CMB photons is discussed elsewhere \cite{MaBer95,Huetal95,SelZal96}. Such calculations for the CMB power spectra from density perturbations were developed in a series of pioneering papers {}from 1970 until the late 1980s \cite{PeeYu70,WilSil80,WilSil81,SilWil81,BonEfs84,VitSil84, BonEfs87,Hol89}, and these calculations have been refined extensively in the post-\COBE\ era. Similar calculations can also be carried out for the CMB power spectra from gravitational waves \cite{AbbWis84,MilVal86,HarZal93,CriDavSte93,FrePolCol94,NgNg95,Kos96, KamKosSte97b,ZalSel97,HuWhi97}. The calculations for both scalar and tensor power spectra require solution of a series of several thousand coupled differential equations for the perturbations to the spacetime metric, densities and velocities of baryons and cold dark matter, and the moments of the CMB photon and neutrino distributions. A code for carrying out these calculations (that required several hours for each model) was made publicly available \cite{Ber95}. Hu \& Sugiyama \cite{HuSug94} came up with useful semianalytic fits to the numerical calculations that provided some physical intuition into the numerical results. More recently, Seljak \& Zaldarriaga \cite{SelZal96,SelZal97} developed a line-of-sight approach that speeded up the numerical calculations by several orders of magnitude. A code (CMBFAST) was made publicly available and has become widely used. \begin{figure} \centerline{\psfig{file=colormod.ps,width=32pc}} \caption{Theoretical predictions for cosmic-microwave-background temperature angular power spectra as a function of multipole moment $l$ for models with primordial adiabatic perturbations. Each graph shows the effect of variation of one of these parameters. In the lower right panel, $\Omega\equiv\Omega_0+\Omega_\Lambda=1$. (From Reference \cite{Junetal96b}.)} \vskip-12pt \label{fig:models} \end{figure} Given the values of the classical cosmological parameters (e.g.\ the nonrelativistic matter density $\Omega_0$, cosmological constant $\Omega_\Lambda$, and baryon density $\Omega_b$, all in units of the critical density, and the Hubble parameter $h$ in units of 100~km~sec$^{-1}$~Mpc$^{-1}$), and primordial scalar and tensor power spectra, $P_s(k)$ and $P_t(k)$, it is straightforward to calculate the $C_l$ with the machinery described above. Figure~\ref{fig:models} shows results of such calculations for models with a Peebles-Harrison-Zeldovich (i.e.\ $n_s=1$) power spectrum of primordial adiabatic perturbations. Each panel shows the effect of independent variation of one of the cosmological parameters. As illustrated, the height, width, and spacing of the acoustic peaks in the angular spectrum depend on these (and other) cosmological parameters. The wiggles\footnote{These are sometimes referred to inaccurately as ``Doppler'' peaks, but are more accurately referred to as acoustic peaks. They are sometimes called Sakharov oscillations in honor of the scientist who first postulated the existence of photon-baryon oscillations in the primordial plasma \cite{Sak65}. The existence of these peaks in the CMB power spectrum was, to the best of our knowledge, first identified by Sunyaev \& Zeldovich \cite{SunZel70} and Peebles \& Yu \cite{PeeYu70}.} come from oscillations in the photon-baryon fluid at the surface of last scatter. Consider an individual Fourier mode of an initial adiabatic density perturbation. Because the density perturbation is nonzero initially, this mode begins at its maximum amplitude. The amplitude remains fixed initially when the wavelength of the mode is larger than the Hubble radius. When the Universe has expanded enough that the Hubble radius becomes larger than the wavelength of this particular mode, then causal physics can act, and the amplitude of this Fourier mode begins to oscillate as a standing acoustic wave \cite{Sak65}. Since modes with smaller wavelengths come within the horizon earlier and oscillate more rapidly, they have at any given time undergone more oscillations than longer-wavelength modes have. The peaks evident in Figure \ref{fig:models} arise because modes of different wavelength are at different points of their oscillation cycles \cite{SunZel70}. The first peak corresponds to the mode that has had just enough time to come within the horizon and compress once. The second peak corresponds to the mode that is at its maximum amplitude after the first compression, and so forth. \subsection{Determination of the Geometry} These acoustic peaks in the CMB temperature power spectrum can be used to determine the geometry of the Universe \cite{KamSpeSug94}. The angle subtended by the horizon at the surface of last scatter is $\theta_H \sim \Omega^{1/2} \;1^\ifmmode ^\circ\, \else $^\circ\,$ \fi$, where $\Omega=\Omega_0+\Omega_\Lambda$ is the total density (objects appear to be larger in a closed Universe than they would in a flat Universe, and smaller in an open Universe than they would in an flat Universe). Moreover, the peaks in the CMB spectrum are due to causal processes at the surface of last scatter. Therefore, the angles (or values of $l$) at which the peaks occur determine the geometry of the Universe. This is illustrated in the top left panel of Figure~\ref{fig:models}, where the CMB spectra for several values of $\Omega$ are shown. As illustrated in the other panels, the angular position of the first peak is relatively insensitive to the values of other undetermined (or still imprecisely determined) cosmological parameters such as the baryon density, the Hubble constant, and the cosmological constant. Small changes to the spectral index $n_s$ tilt the entire spectrum slightly to smaller (larger) $l$ for $n_s<1$ ($n_s>1$), and the location of the first peak is only weakly affected. Gravitational waves would only add to the temperature power spectrum at $l\ll 200$ (as discussed below in Section 4.4). Therefore, although gravitational waves could affect the height of the peaks relative to the normalization at small $l$, the locations would not be affected. It is plausible that an early generation of star formation released a sufficient flux of ionizing radiation to at least partially reionize the Universe, and if so, these ionized electrons would rescatter some fraction $\tau$ of the CMB photons. A variety of theoretical arguments suggest that a fraction $\tau = {\cal O}(0.1)$ of CMB photons were rescattered \cite{KamSpeSug94,TegSilBla94,HaiLoe97} (and the amplitude of anisotropy at degree scales observed already supports this). Although reionization would damp the peaks by a factor $e^{-2\tau}$, as indicated by the curve labeled ``reion'' in the top left panel of Figure \ref{fig:models} (but note that the figure assumes $\tau=1$), the location of the peaks would remain unchanged. Therefore, if peak structure is observed in the CMB power spectrum, determination of the location of the first peak will provide a robust determination of the geometry of the Universe \cite{KamSpeSug94}. \subsubsection{Open inflation} The most distinctive signature of an open-inflation model would be a low-$\Omega_0$ CMB power spectrum from adiabatic perturbations such as one of those shown in the top left panel of Figure \ref{fig:models}, in which the first peak is shifted to larger $l$. Open inflation would also produce an increase in large-angle anisotropy from the integrated Sachs-Wolfe effect \cite{LytSte90,KamSpe94}, but different open-inflation models make different predictions about the large-angle anisotropy. Moreover, determination of the CMB power spectrum at these large angular scales is cosmic-variance limited, so it is unlikely that large-angle CMB anisotropies alone will be able to provide a robust test of open-inflation models. The ISW effect may alternatively be identified by cross-correlation of the CMB with some tracer of the mass density along the line of sight \cite{CriTur96,Kam96}, such as the X-ray background \cite{BouCriTur98,KinKam98} or possibly weak-lensing maps \cite{ZalSel99} (as discussed further in Section \ref{sec:cosmologicalconstant}). \begin{figure} \centerline{\psfig{file=huwhite.ps,width=32pc}} \caption{The angular power spectrum for an inflationary model with primordial adiabatic perturbations and for another with primordial isocurvature perturbations. \textit{Solid line}, cold dark matter and inflation; \textit{dashed line}, axion isocurvature. (From Reference \cite{HuWhi96a}.)} \vskip-12pt \label{fig:huwhite} \end{figure} \subsection{Adiabatic Versus Isocurvature Modes} The physics described above yields a distinctive pattern in the peak structure of the CMB power spectrum, and this leads to an important test of inflation. If primordial perturbations are isocurvature rather than adiabatic, then when a given Fourier mode comes within the horizon and begins to oscillate, it begins to oscillate from its minimum (rather than maximum) amplitude. Thus, the phase of its oscillation differs by $\pi/2$ from what it would be if the perturbation were adiabatic. As a result, the locations of the peaks in the CMB power spectrum in isocurvature models differ in phase from what they would be in adiabatic models \cite{HuSug95,HuWhi96a,Kos98}, as shown in Figure \ref{fig:huwhite}. It has also been shown that the relative locations of the higher peaks differ in adiabatic and isocurvature models, independent of the shift in the locations of the peaks due to the geometry \cite{HuWhi96}. \subsubsection{Axion inflation} When the matter power spectrum is normalized to the amplitude of galaxy clustering, isocurvature models with nearly scale-invariant primordial power spectra (e.g.\ axion isocurvature or ``axion inflation'' models) produce roughly six times the CMB anisotropy seen by \COBE\ \cite{KodSas86,EfsBon87} (since there is no cancellation between the effects of the intrinsic temperature and the potential-well depth at the surface of last scatter) and are thus ruled out. \subsection{Coherent Perturbations and Polarization} Each Fourier component of the density field induces a Fourier component of the peculiar-velocity field, and the oscillations of these peculiar velocities are out of phase with the oscillations in the density perturbation (just as the velocity and position of a harmonic oscillator are out of phase). These peculiar velocities induce temperature anisotropies (via the Doppler effect) that are thus out of phase with those from density perturbations. This Doppler effect fills in the troughs in the $C_l^{\rm TT}$, which would otherwise fall to zero. The CMB polarization is related to the peculiar velocity at the surface of last scatter \cite{ZalHar95}, so the peaks in the polarization power spectrum (from density perturbations) are out of phase from those in the temperature power spectrum and fall close to zero (Figure \ref{fig:clsplot}). This relative positioning of the temperature and polarization peaks is a signature of coherent perturbations (rather than those produced, for example, by the action of topological defects, as discussed below) \cite{Kos98}. Zaldarriaga \& Spergel \cite{SpeZal97} argue that the location of the first peak in the polarization power spectrum provides a test of primordial perturbations as the origin of structure and thus of inflation. If some causal mechanism (such as topological defects) produced large-scale structure, the the first peak would have to occur at smaller angular scales in order to be within the horizon at the surface of last scatter (see below). A peak so close to the causal horizon could only occur with super-horizon-sized primordial perturbations, for which inflation is the only causal mechanism. \begin{figure}[htbp] \centerline{\psfig{file=clt0.ps,width=7in}} \bigskip \caption{ Theoretical predictions for the four nonzero cosmic-microwave-background temperature-polarization spectra as a function of multipole moment $l$. \textit{Solid curves} are the predictions for a \COBE-normalized inflationary model with no reionization and no gravitational waves for $h=0.65$, $\Omega_b h^2=0.024$, and $\Lambda=0$. \textit{Dotted curves} are the predictions that would be obtained if the \COBE\ anisotropy were due entirely to a stochastic gravity-wave background with a flat scale-invariant spectrum (with the same cosmological parameters). The panel for $C_l^{\rm CC}$ contains no dotted curve because scalar perturbations produce no ``C'' polarization component; instead, the \textit{dashed line} in the \textit{bottom right panel} shows a reionized model with optical depth $\tau=0.1$ to the surface of last scatter. (From Reference \cite{KamKos98}.) } \label{fig:clsplot} \end{figure} \subsection{Polarization and Gravitational Waves} Gravitational waves are usually detected by observation of the motion they induce in test masses. The photon-baryon fluid at the surface of last scatter acts as a set of test masses for detection of gravitational waves with wavelengths comparable to the horizon, such as those predicted by inflation. These motions are imprinted onto the temperature and polarization of the CMB. The top left panel of Figure \ref{fig:clsplot} (\textit{solid curve}) shows the temperature power spectrum for a \COBE-normalized flat scale-invariant ($n_t=0$) spectrum of gravitational waves. It is flat and relatively featureless for $l\la70$. The dropoff at $l \ga70$ is due to the fact that the amplitudes of gravitational-wave modes that enter the horizon before the epoch of last scatter have decayed with the expansion of the Universe. Unfortunately, cosmic variance from scalar perturbations provides a fundamental limit to the sensitivity of CMB temperature maps to tensor perturbations \cite{KnoTur93}. Even if all other cosmological parameters are somehow fixed, a perfect temperature map can never detect an inflaton-potential height smaller than one-tenth the upper limit provided by \COBE\ \cite{Junetal96b}. More realistically, the effects of gravitational waves and reionization on the temperature power spectrum are similar and difficult to disentangle, so improvements to the current \COBE\ sensitivity to gravitational waves is unlikely with a temperature map alone. However, with a polarization map of the CMB, the scalar and tensor contributions to CMB polarization can be geometrically decomposed in a model-independent fashion, and the cosmic-variance limit present in temperature maps can thereby be circumvented \cite{Ste96,KamKosSte97a,SelZal97}. Scalar perturbations have no handedness, so they cannot give rise to a curl component. On the other hand, tensor perturbations do have a handedness, so they induce a curl component. Therefore, if any curl coefficient, $a_{(lm)}^{\rm C}$, is found to be nonzero, it suggests the presence of gravitational waves.\footnote{A curl component may also be due to vector perturbations. Although topological-defect models may excite such modes, they do not arise in inflationary models.} To illustrate, Figure~\ref{fig:clsplot} shows the four nonzero temperature-polarization power spectra. The \textit{dotted curves} correspond to a \COBE-normalized inflationary model with no gravitational waves. The \textit{solid curves} show the spectra for a \COBE-normalized stochastic gravitational-wave background. \subsubsection{Detectability of gravitational waves: curl component only} Consider a mapping experiment that measures the polarization on the entire sky with a temperature sensitivity $s$ (which has units $\mu$K~$\sqrt{\rm sec}$) for a time $t_{\rm yr}$ years. If only the curl component of the polarization is used to detect tensor perturbations, then such an experiment can distinguish a tensor signal from a null result at the $2\sigma$ level if the inflaton potential height is \cite{KamKos98} \begin{equation} V \mathrel{\mathpalette\fun >} (4\times 10^{15}\, {\rm GeV})^4 \, t_{\rm yr}^{-1} \, (s/\mu{\rm K}\, \sqrt{\rm sec})^2. \label{eq:tensordetectable} \end{equation} Equation~\ref{eq:tensordetectable} indicates that to access an inflaton-potential height not already excluded by \COBE\ requires a detector sensitivity $s\la35\,t_{\rm yr}^{1/2}\,\mu$K$\sqrt{\rm sec}$. To compare this with realistic values, the effective sensitivity of MAP is $s\simeq150\,t_{\rm yr}^{1/2}\,\mu$K$\sqrt{\rm sec}$ and that for Planck is about $s\simeq35\,t_{\rm yr}^{1/2}\,\mu$K$\sqrt{\rm sec}$, and technological developments have improved the detector sensitivity roughly an order of magnitude per decade for the past several decades. Even better sensitivities may be possible with deep integration on a smaller region of the sky. \subsubsection{Reionization} In some sense, Equation~\ref{eq:tensordetectable} is conservative because even a small amount of reionization will significantly increase the polarization signal at low $l$ (indicated by the \textit{dashed curve} in the CC panel of Figure~\ref{fig:clsplot} \cite{Zal97}). If $\tau=0.1$, then the sensitivity to tensor modes is increased by roughly a factor of five \cite{KamKos98}. \subsubsection{Full polarization and temperature spectra} Although searching only for the curl component provides a model-independent probe of tensor modes, a stochastic gravitational-wave background leads to specific predictions for all four nonzero temperature-polarization power spectra (Figure \ref{fig:clsplot}). Fitting an inflationary model to the entire set of temperature and polarization power spectra can improve tensor detectability, especially at poorer sensitivities, albeit with the introduction of some model dependence. For detector sensitivities $s\mathrel{\mathpalette\fun >} 15\,t_{\rm yr}^{1/2}\,\mu$K$\sqrt{\rm sec}$, the sensitivity to a tensor signal is improved by factors of a few or so \cite{KamKos98}, depending on the cosmological model, whereas for detector sensitivities $s\mathrel{\mathpalette\fun <} 15\,t_{\rm yr}^{1/2}\,\mu$K$\sqrt{\rm sec}$, the sensitivity is attributable almost entirely to the CC power spectrum and approaches the limit in Equation~\ref{eq:tensordetectable}. \begin{figure}[htbp] \centerline{\psfig{file=newmodel.ps,width=5.5in}} \bigskip \caption{Simulated $2\sigma$ error ellipses that would be obtained by a cosmic-variance-limited temperature map, the Planck Surveyor (with polarization), and an experiment with three times the sensitivity of Planck. This assumes an inflationary model with $r=0.01$ and $n_s=0.95$ and an optical depth to the surface of last scatter of $\tau=0.05$. \textit{Shaded regions} indicate the predictions of various inflationary models. \textit{Solid horizontal curve} indicates the regions of this logarithmic parameter space that would be accessible with a putative polarization experiment with 30 times the Planck instrumental sensitivity \cite{KamKos98}. Even better sensitivities may be possible with deep integration on a smaller region of the sky. (From Reference \cite{Kin98}.) } \label{fig:Plancklog} \end{figure} \subsubsection{Measurement of inflationary observables} Several authors have addressed the question of how precisely the inflationary observables can be reconstructed in the case a positive detection of the stochastic gravitational-wave background with only a temperature map \cite{Kno95,Junetal96b,BonEfsTeg97,DodKinKol97} and with a polarization map as well \cite{ZalSelSpe97,Lidetal97b,Kin98}. We follow the discussion of Ref. \cite{Kin98}. Figure \ref{fig:Plancklog} shows the $2\sigma$ error ellipses that would be obtained by the Planck Surveyor using the temperature only (i.e. the cosmic-variance limit) and with the polarization, assuming a gravitational-wave background with $r\simeq0.01$ and $n_s\simeq0.9$. (A larger gravitational-wave amplitude would be detectable, as shown in Figures~3--6 in Reference \cite{Kin98}.) The $2\sigma$ cosmic-variance limit from a temperature map is shown as is the $2\sigma$ constraint to the parameter space expected for Planck (with polarization). Although such a tensor amplitude cannot be distinguished from a null result, the figure shows (the dark shaded region) that a hypothetical experiment with three times the Planck polarization sensitivity could discriminate between a such a tensor signal and a null result. It would also discriminate between a single small-field model and a hybrid model. Of course, the sensitivity to tensor modes can be improved as the instrumental sensitivity is improved, as indicated by Equation~\ref{eq:tensordetectable}. For example, the thin horizontal line at $r=0.001$ shows the smallest value of $r$ that could be distinguished from a null result by a hypothetical one-year experiment with an instrumental sensitivity $s=\mu$K$\sqrt{\rm sec}$, roughly 30 times that of Planck \cite{KamKos98}. A null result from such an experiment would suggest that if inflation occurred, it would have required a small-field model. \subsection{Gaussianity} The prediction of primordial Gaussianity or of some specific small deviations from Gaussianity can be probed with the CMB angular bispectrum or higher $n$-point correlation functions discussed above. A nonzero large-angle CMB bispectrum may arise from the integrated Sachs-Wolfe effect if there is a cosmological constant \cite{Ganetal94}. Such a bispectrum, as well as that probed by the Sunyaev-Zeldovich effect, may be discernible via cross-correlation between gravitational-lensing maps and the CMB \cite{GolSpe98}. A more powerful test of inflation models may arise from probing the bispectrum induced by nonlinear evolution at the surface of last scattering (S Winitzki, A Kosowsky, DN Spergel, manuscript in preparation). Ferreira et al and Pando et al \cite{FerMagGor98,PanValFan98} recently claim to have already found some signature of non-Gaussianity in the \COBE\ maps. In particular, Ferreira et al \cite{FerMagGor98} find $I_l^3 \sim1$ for $l\sim16$ (although it is still not clear if the effect is real \cite{KamJaf98}). If this result is correct, then the simplest slow-roll inflation models are not viable (see Equation~\ref{eq:Il3prediction}). \section{Topological-Defect Models} \label{sec:defects} The leading alternative to structure-formation models based on inflation have been those based on topological defects, particularly cosmic strings \cite{Zel80,Vil81,SilVil84,Tur85,HinKib95}, global monopoles \cite{BarVil89,BenRhi91}, domain walls \cite{HilSchFry89,PreRydSpe89}, and textures \cite{Tur89,Tur91} (for reviews, see \cite{Vil85,VilShe94}). Defect models postulate a phase transition in the early Universe that leads to a vacuum manifold with nontrivial topology; the type of defect depends on the specific topology (see \cite{HinKib95} for a review). Since defect formation is a process governed by causal physics, the vacuum state of the field must be uncorrelated on scales larger than the horizon at the time of the phase transition, guaranteeing the formation of defects with a characteristic length scale of the horizon (the ``Kibble mechanism'' \cite{Kib76}). The simplest defects are domain walls, which arise in theories with a discrete symmetry. Domain-wall models are not viable because their energy densities are large enough to produce larger CMB temperature fluctuations than those observed \cite{ZelKozOku74,SteTur89,TurWatWid91}. Cosmic strings are stable defects that arise in gauge models with a $U(1)$ symmetry. They produce density perturbations by their gravitational interactions with ordinary matter. Global-monopole and texture models are unstable defects that arise in models with a perfect global symmetry. They provide two mechanisms for structure formation: (\textit{a}) the energy-density provided by misalignment of scalar fields as causally disconnected regions of the Universe come into causal contact, and (\textit{b}) the explosive events that occur when the topological defects unwind. Non-topological texture models \cite{Vil82,Pre80,TurSpe91,Jaf94} postulate an even higher global symmetry and seed structure via scalar-field alignment even though no topological defects are formed. Generically, one expects quantum gravity to violate global symmetries to the level that would render global-monopole, texture, and scalar-field-alignment models unworkable \cite{KamMar92,Holetal92}. If it could be shown that such models do seed large-scale structure, valuable information on Planck-scale physics would thus be provided. \subsection{Cosmic-Microwave-Background Power Spectra in Defect Models} In contrast to inflationary models, which lay down an initial spectrum of density perturbations, defect models produce perturbations actively throughout the history of the Universe. This generally leads to loss of coherence in the perturbations and a corresponding smoothing of the acoustic peaks \cite{Albetal96,Magetal96a,HuWhi97c}. Defect-model perturbations are also causal, being generated by physical processes inside the horizon \cite{SteVee90,Tur96b,Tur96c,HuSpeWhi97,DurKun98}. And finally, primordial perturbations in defect models more closely resemble those in primordial-isocurvature rather than those in adiabatic models \cite{SteVee90,DurSak97,HuSpeWhi97}. Moreover, the action of topological defects generically produces vector and tensor perturbations which increase the anisotropy on small angular scales \cite{PenSelTur97,Alletal97}, further suppressing any peak structure (although it may produce some characteristic C polarization \cite{SelPenTur97}). \begin{figure}[htbp] \centerline{\psfig{file=defects.ps,width=5.5in}} \bigskip \caption{Cosmic-microwave-background power spectra from topological-defect models. \textit{Solid line}, total; \textit{dotted}, scalar; \textit{short-dashed}, vector; \textit{long-dashed}, tensor. (From Reference \cite{SelPenTur97}.)} \label{fig:defects} \end{figure} Until recently, different groups obtained different results about the extent to which acoustic peaks exist in defect models, and under what circumstances \cite{CriTur95,Tur96b,DurGanSak96,DurZho96,Tur96b,Tur96c}. Calculations of CMB power spectra based on large simulations of a variety of defect sources have now been performed \cite{Alletal97,PenSelTur97,ConHinMag99} (see Figure \ref{fig:defects} for some). The numerical results indicate that the acoustic peaks are washed out. At this point, it appears that the simplest defect models are inconsistent with the observed CMB fluctuations and the large-scale structure traced by galaxy surveys. Although this could have been inferred from the generic arguments discussed in the Introduction \cite{JafSteFri94}, it has been supported by these more recent precise calculations \cite{AlbBatRob97,Alletal97,PenSelTur97}. The question now is whether any more complicated (or ``sophisticated'') defect models be viable. Albrecht et al \cite{AlbBatRob98a,AlbBatRob98b} have suggested that a cosmological constant might help improve concordance with current data. However, suppose the CMB temperature power spectrum continues to look increasingly like that caused by inflation (i.e.\ with identifiable acoustic peaks), as new data seem to suggest. If so, can any defect model reproduce such a power spectrum? Turok \cite{Tur96c} produced a power spectrum with a phenomenological defect model that closely mimicked an inflation power spectrum, and Hu \cite{Hu99} has invented a similar isocurvature model. But it is hard to see how to position the acoustic peaks in isocurvature-like models at the same angular scales as in adiabatic models without some rather artificial initial conditions \cite{HuSpeWhi97,Magetal96b}. It is also difficult to simultaneously account for the fluctuation amplitude in the CMB and galaxy surveys, unless there is a breaking of scale invariance \cite{Pen98} (possibly from some finite breaking of the global symmetry \cite{KamMar92,Holetal92}). It may, in fact, be possible to construct some causal models that produce peaks in the CMB power spectrum \cite{DurKun98,DurSak97}, but it is unclear whether fluctuations that mimic a specific inflationary model can be produced, particularly when additional constraints from polarization are taken into account. Finally, it should be noted that hybrid models with both primordial adiabatic perturbations and defects have been entertained \cite{Jen96,LinRio97,AveCalMar98}. \subsection{Non-Gaussianity} Topological-defect models may also be distinguished by the non-Gaussian signatures they produce in the CMB. Because the evolution of topological defects is nonlinear, they generically produce non-Gaussian structures in the CMB. Put another way, the production of defects via the Kibble mechanism is a Poisson process; the number of defects within any volume in the Universe is Poisson distributed. The central-limit theorem guarantees that as the number density of defects becomes large, the distribution should become increasingly Gaussian. Thus cosmic-string models should look more like Gaussian perturbations than textures should, since the Kibble mechanism produces roughly one texture per 25 Hubble volumes as opposed to roughly one cosmic string per Hubble volume \cite{SchBer91,Per93b,BenRhi93,GilPer95,SchSch95}. In the large-$N$ limit of the $O(N)$ sigma model, the clearest signature of non-Gaussianity from scalar-field alignment is at large angular scales \cite{Jaf94}; on small distance scales, the theory looks roughly Gaussian. Constraints to non-Gaussianity from the galaxy distribution have already posed problems for scalar-field-alignment models for several years. Since defects are coherent structures, they can produce corresponding coherent structures in the CMB temperature anisotropy. For example, a cosmic string can produce a linear discontinuity in the CMB temperature \cite{KaiSte84}, which can be searched for most efficiently through statistics tailored to match this particular signal \cite{MoePerBra94,Per97,Per98}. Textures might form large hot spots \cite{TurSpe90,Tur96a}. \section{Dark Matter} \label{sec:darkmatter} The CMB can potentially provide a wealth of information about the dark matter known to dominate the mass of the Universe. The smallness of the amplitude of CMB temperature fluctuations has for a long time provided some of the strongest evidence for the existence of dark matter. In a low-density Universe, density perturbations grow when the Universe becomes matter-dominated and end when it becomes curvature-dominated. If the luminous matter ($\Omega_{\rm lum} \sim10^{-3}$) were all the mass in the Universe, then the epoch of structure formation would be too short to allow density perturbations to grow from their early-Universe amplitude, fixed by \COBE, to the amplitude observed today in galaxy surveys. More precise measurements of the CMB power spectrum hold the promise of providing much more detailed information about the properties and distribution of dark matter. There are currently several very plausible dark-matter candidates that arise from new particle physics, and some evidence has already been claimed for the existence of several of these. For example, some observational evidence points to the existence of a cosmological constant \cite{Peretal97,Peretal99,Rieetal98}, and the LSND experiment suggests that massive neutrinos may constitute a significant fraction of the mass of the Universe \cite{Athetal95,Athetal96}. Moreover, there are good arguments that a significant fraction of the mass in galactic halos is made of some type of cold-dark-matter particle, e.g.\ weakly interacting massive particles (WIMPs) \cite{JunKamGri96} or axions \cite{Tur90,Raf90}. \subsection{Cold Dark Matter} A number of dynamical measurements suggest that the nonrelativistic-matter density is $\Omega_0\ga0.1$, whereas big-bang nucleosynthesis suggests a baryon density of $\Omega_b\la0.1$. Observations of X-ray emission from galaxy clusters suggest that the nonrelativistic matter in clusters outweighs the baryonic matter by a factor of three or more \cite{Whietal93}, and weak lensing of background galaxies by clusters directly reveals large amounts of dark matter \cite{TysKocDel98}. This evidence strongly indicates the existence of some nonbaryonic dark matter. By fitting the power spectra from MAP and Planck to theoretical predictions, one should simultaneously be able to determine both $\Omega_0 h^2$ and $\Omega_b h^2$ to far better precision than that obtained by current observations \cite{Junetal96b,BonEfsTeg97,ZalSelSpe97}. If a substantial fraction of the mass in the Universe is in fact made of nonbaryonic dark matter (e.g.\ WIMPS or axions), then it will become evident after MAP and Planck. Unfortunately, there is no way to discriminate between WIMPs and axions with the CMB. \subsection{Neutrinos} One of the primary goals of experimental particle physics is pursuit of a nonzero neutrino mass. Some recent (still controversial) experimental results suggest that one of the neutrinos may have a mass of ${\cal O}(5\,{\rm eV})$ \cite{Athetal95,Athetal96}. There have been some arguments (again, still controversial) that such a neutrino mass is exactly what is required to explain apparent discrepancies between large-scale-structure observations and the simplest inflation-inspired standard-CDM model \cite{ShaSte84,DalSch92,DavSumSch92,Klyetal93,Prietal95,BonPie98}. If the neutrino does indeed have a mass of ${\cal O}(5 \, {\rm eV})$, then roughly 30\% of the mass in the Universe is in the form of light neutrinos. These neutrinos will affect the growth of gravitational-potential wells near the epoch of last scatter, thus leaving an imprint on the CMB angular power spectrum \cite{DodGatSte96,MaBer95,Lopetal98a}. The effect of a light neutrino on the power spectrum is small, so other cosmological parameters that might affect the shape of the power spectrum at larger $l$ must be known well. Eisenstein et al \cite{EisHuTeg99} argue that by combining measurements of the CMB power spectrum with those of the mass power spectrum measured by, for instance, the Sloan Digital Sky Survey, a neutrino mass of ${\cal O}(5 \, {\rm eV})$ can be identified. The CMB may constrain the number of noninteracting relativistic degrees of freedom in the early Universe \cite{Junetal96b}. Although weaker than the bound from big-bang nucleosynthesis \cite{SteSchGun77, CopSchTur97}, the CMB probes a different epoch ($T\sim$eV rather than $T\sim$MeV) and may thus be viewed as complementary. \subsection{Cosmological Constant} \label{sec:cosmologicalconstant} Some recent evidence seems to point to the existence of an accelerating expansion, possibly due to a nonzero cosmological constant (\cite{Peretal97,Peretal99}; for a review of the cosmological constant, see \cite{CarPreTur92}). The CMB may help probe the existence of a cosmological constant in a number of ways. As discussed above, if adiabatic perturbations are responsible for large-scale structure, then the position of the first acoustic peak in the CMB power spectrum provides a model-independent probe of the total density, $\Omega=\Omega_0 + \Omega_\Lambda$ \cite{KamSpeSug94}. In contrast, the supernova measurements of the Hubble diagram at large redshifts determine primarily the deceleration parameter $q_0 = \Omega_0/2 - \Omega_\Lambda$, so the two measurements together can give tight limits on both $\Omega_0$ and $\Omega_\Lambda$ individually \cite{Efsetal98,Garetal98,Teg98,TegEisHu98,Whi98,Lin98}. As the bottom panels of Figure \ref{fig:models} show, variations to $\Omega_0$ and $h$ affect the the height and width of the first acoustic peak; the dependence is more precisely on the quantity $\Omega_0 h^2$. Thus, if the Hubble constant is known, then the CMB can determine $\Omega_0$ and $\Omega$ (from the peak location) and therefore the cosmological constant $\Omega_\Lambda$. A cosmological constant may also be distinguished from the CMB via the additional large-angle anisotropy it produces via the ISW effect \cite{KofSta86} from density perturbations at redshifts $z\mathrel{\mathpalette\fun <}$few. If there is a cosmological constant, there should be a cross-correlation between the CMB temperature and some tracer of the mass distribution, e.g.\ the extragalactic X-ray background \cite{BouCriTur98} or weak lensing \cite{ZalSel99}, at these redshifts \cite{CriTur96} (the same also occurs in an open Universe \cite{Kam96,KinKam98}). An experimental upper limit to the amplitude of this cross-correlation \cite{BouCriTur98} can already be used to constrain $\Omega_0$, with some assumptions about the bias of sources that give rise to the extragalactic X-ray background. If $\Omega_0\simeq0.3$ (either in an open or a flat cosmological-constant model), then these X-ray sources can be no more than weakly biased tracers of the mass distribution \cite{KinKam98}. \subsection{Rolling Scalar Fields} The supernova evidence for an accelerating expansion has engendered a burst of theoretical activity on exotic forms of matter with an equation of state $p<-\rho/3$ (i.e.\ the equation of state needed for $q_0<0$). The simplest possibility is of course a cosmological constant. However, as explained in Section \ref{sec:scalarfields}, a rolling scalar field may also provide such an equation of state, provided the scalar field is not rolling too quickly. An almost endless variety of equations of state---and expansion histories---are possible in principle, given the freedom to choose the scalar-field potential and the initial conditions. This idea is variously referred to in the literature as rolling-scalar-field, variable-cosmological-constant, x-matter, generalized-dark-matter, loitering-Universe, and/or quintessence models \cite{RatPee88,SahFelSte92,SugSat92,Frietal95,CobDodFri97, SilWag97,TurWhi97,ChiSugNak97,CalDavSte98,ChiSugNak98}. Additional work has explored attractor solutions based on exponential potentials \cite{LucMat85,Wet88,WanCopLid93,FerJoy97,CopLidWan98,LidSch99} or ``tracker-field'' solutions \cite{ZlaWanSte99,SteWanZla99} that attempt to explain why the matter density would be comparable to a scalar-field energy density today. Because the expansion rate at decoupling in such models is the same as in cosmological-constant models with the same $\Omega_0$, the peak structure in the CMB is virtually indistinguishable from that in cosmological-constant models \cite{Hueetal99}. However, perturbations in the scalar field track perturbations to the matter density on large scales in such a way that the large-angle ISW effect that appears in cosmological-constant models is canceled by the effect of scalar-field perturbations \cite{CalDavSte98}. Data from cosmological observations, particularly supernova measurements of the expansion history and measurements of the power spectrum through large galaxy surveys, may in principle be used to break these degeneracies \cite{Huetal99,Wanetal99}. \section{Other Constraints on Particle Physics} \begin{figure}[htbp] \centerline{\psfig{file=huspectr.ps,width=5.5in,angle={-90}}} \bigskip \caption{Constraints to the mass-lifetime plane for particles decaying to photons from FIRAS constraints to distortions to the CMB blackbody spectrum. \textit{Solid curve} is the numerical result; \textit{dashed curves} show various approximations. The quantity $n_X/n_\gamma$ is the initial ratio of the particle number density to the photon number density. (From Reference \cite{HuSil93a}.)} \label{fig:decaylimits} \end{figure} \subsection{Decaying Particles} As discussed in Section \ref{sec:freqspectrum}, FIRAS limits to $\mu$ and $y$ distortions limit the injection of energy into the early Universe and can thus be used to constrain the mass-lifetime plane of particles that decay to electromagnetically interacting particles (as shown in Figure~\ref{fig:decaylimits}) \cite{HuSil93a,Elletal92}. The CMB power spectrum can also constrain decaying particles. For example, a neutrino of mass $\ga10$ eV that decays to relativistic particles with a lifetime $\tau \simeq 10^{13-17}$ alters the expansion rate of the Universe between recombination and today and thus produces large-angle anisotropy (via the ISW effect) in disagreement with observations \cite{Lopetal98b,Han98b,Han98c}. \subsection{Time Variation of Fundamental Constants} A number of ideas for new physics postulate that some of the fundamental constants of nature, such as the fine-structure constant $\alpha$, may actually be varying (for a review, see \cite{VarPot95}). Such a variation could be caused by the cosmological evolution of compact spatial dimensions in string theory or Kaluza-Klein theories \cite{Mar84,Bar87,DamPol94} or through scalar fields coupled to electromagnetism \cite{Car98}. Limits of $|\Delta\alpha/\alpha| \mathrel{\mathpalette\fun <} 10^{-7}$ were provided by the natural nuclear reactor at Oklo \cite{Shy76,DamDys96}, and observations of atomic- and molecular-line positions at high redshifts \cite{Sav56} provide limits of $|\Delta\alpha/\alpha| < 3\times 10^{-6}$ at redshifts less than 1 \cite{Drietal98} and $|\Delta\alpha/\alpha| \mathrel{\mathpalette\fun <} 3\times 10^{-4}$ at redshifts of 3 \cite{CowSon95,IvaPotVar98}. In fact, a detection of $\Delta\alpha/\alpha = -1.9\pm 0.5 \times 10^{-5}$ has been claimed on the basis of absorption lines at redshifts greater than 1 \cite{Webetal99}, but there are some potential problems with this result \cite{IvaPotVar98}. Primordial nucleosynthesis can also provide a less useful model-dependent limit \cite{KolPerWal86}. A change in $\alpha$ would affect the recombination rate of hydrogen and thus alter the redshift of last scatter. This effect on the CMB can potentially lead to upper limits on $|\Delta\alpha/\alpha|$ between 0.01 and 0.001 \cite{KapSchTur98,Han98a} out to redshifts $z\simeq1100$, much larger than those probed by quasar absorption spectra. \subsection{Topology of the Universe} The fundamental cosmological assumptions of homogeneity and isotropy require the Universe to be either the open, closed, or flat Friedmann-Robertson-Walker model. However, if the assumption of isotropy is incorrect, then the Universe may have some nontrivial topology (see~\cite{LacLum95} for a review). The open and flat FRW models have infinite volume, but a Universe with either zero or negative curvature can have finite volume if the Universe has nontrivial topology. A number of (somewhat imprecise) theoretical arguments suggest that a finite Universe is easier to explain than an open Universe \cite{ZelSta84} or could be used to explain the homogeneity of the Universe \cite{Got80,EllSch86}. If the volume of such a Universe is comparable to or less than that observable today, then there may be signatures in the CMB. Consider the simplest nontrivial topology (for a flat Universe), that of a toroid. If the Universe is a three-dimensional toroid, then two different directions on the sky will point to the same point in space, and there should be observable correlations between the CMB temperature at distant locations on the sky. Such models have essentially been ruled out by \COBE\ \cite{FanMo87,Sok93,Sta93,Fan93,SteScoSil93,JinFan94,CosSmo95,CosSmoSta96,ScaLevSil98,LevScaSil98}. Interest in negative-curvature models with nontrivial topology has reawakened recently because evidence seems to suggest $\Omega_0\simeq0.3<1$, and thus possibly an open Universe. If the Universe is negatively curved (hyperbolic), the spacetime volume element increases rapidly with distance, so that even if the volume of the Universe is close to the horizon volume, many copies of the Universe may still fit inside the horizon volume. Thus, none of the flat-Universe limits on topology apply to hyperbolic Universes \cite{CorSpeSta98b}. Furthermore, if the total density of the Universe is $\Omega\simeq0.3$, the curvature scale is small enough so that a huge number of topologies exist that have proper volumes significantly smaller than the proper Hubble volume \cite{CorSpeSta98c}. Because the surface of last scatter is spherical, matched pairs of temperature circles would appear in a negatively-curved Universe with nontrivial topology provided that the topology radius were smaller than the current horizon \cite{CorSpeSta96,Wee98,CorSpeSta98c}. Levin et al \cite{Levetal98} propose searching for specific correlations between a given pixel and all others in a map. A null search for such correlations in the \COBE\ maps ruled out a particular horn topology \cite{Levetal97}. Souradeep et al \cite{SouPogBon98} claim that the \COBE\ maps already rule out most hyperbolic Universes through this technique, although details have not been presented. \subsection{Primordial Magnetic Fields} Magnetic fields of strength $10^{-6}$ G are ubiquitous in our Galaxy and in distant clusters of galaxies. All mechanisms for the origin of these magnetic fields postulate that they grew via some mechanism (e.g.\ dynamo or adiabatic compression) from small primordial magnetic fields. However, the origin of these primordial seed fields remains a mystery. Many of the most intriguing hypotheses the origin of these fields come from new ideas in particle theory. Proposed generation mechanisms include inflation \cite{TurWid88,CarFie91,GarFieCar92,Rat92,Dol93,GasGioVen95a,GasGioVen95b}, the electroweak \cite{Vac91,EnqOle93} or QCD phase transitions \cite{QuaLoeSpe89,CheOli94}, a ferromagnetic Yang-Mills vacuum state \cite{EnqOle94}, charge asymmetry \cite{DolSil93}, and dilaton evolution \cite{Gio97}. Magnetic fields have several potentially measurable effects: Faraday rotation \cite{KosLoe96} (A Mack, A Kosowsky, manuscript in preparation) and and associated depolarization \cite{HarHayZal96} of the original CMB polarization; magnetosonic waves that modify the acoustic oscillation frequencies \cite{Adaetal96}; and Alfven waves, which can amplify vector perturbations and induce additional correlations \cite{DurKahYat98}, and for which diffusion damping is decreased, thereby increasing CMB power at small scales \cite{SubBar98}. The Faraday rotation signals can be detected through the CC, TC, and GC power spectra they induce \cite{ScaFer97} (although these power spectra are frequency dependent). A recent analysis of the \COBE\ maps has placed a limit on a homogeneous primordial field strength corresponding to $B_0< 3.4\times 10^{-9}(\Omega_0h_{50}^2)^{1/2}$ G \cite{BarFerSil97} by searching for the temperature pattern of a Bianchi type VII anisotropic spacetime \cite{Nov68,BarJusSon85}. \subsection{Large-Scale Parity Violation} It is usually assumed that gravity is parity-invariant. However, weak interactions are parity-violating \cite{LeeYan56,Wu57}, and we surmise that the electroweak interactions are united with gravity at the Planck scale by some fundamental unified theory. Are there any remnants of parity-violating new physics in the early Universe? As discussed in Section \ref{sec:powerspectra}, if either of the temperature-polarization cross-correlation moments $C_l^{\rm TC}$ or $C_l^{\rm TG}$ is nonzero, it signals cosmological parity breaking. Lue et al and Lepora \cite{LueWanKam98,Lep98} discuss how a parity-violating term, $\phi F_{\mu\nu}\tilde F^{\mu\nu}$ \cite{CarFie90,CarFie91,Car98}, that couples a scalar field $\phi$ to the pseudoscalar ${\vec E}\cdot {\vec B}$ of electromagnetism could yield nonzero $C_l^{\rm TC}$ and $C_l^{\rm GC}$. Lue et al \cite{LueWanKam98} also discuss a parity-violating term in the Lagrangian for gravitation that would yield an asymmetry between the density of right- and left-handed gravitational waves produced during inflation; such an asymmetry would also give rise to nonzero $C_l^{\rm TC}$ and $C_l^{\rm GC}$. These parity-breaking effects would produce frequency-independent $C_l^{\rm TC}$ and $C_l^{\rm GC}$, unlike the frequency-dependent effect of Faraday rotation. \subsection{Baryon Asymmetry} There are very good theoretical and observational reasons to believe that our entire observable Universe is made of baryons and no antibaryons. But suppose momentarily that the observable Universe consisted of some domains with antibaryons rather than baryons (see e.g.\ \cite{Ste81}). If so, then particle-antiparticle annihilations at the interfaces of the matter and antimatter regions would release a significant amount of energy in $\gamma$-rays, thus heating the region and causing a $y$-distortion of the CMB spectrum of order $y\simeq 10^{-6}$ \cite{KinKolTur97,CohRujGla98}. These distortions would appear in thin strips on the sky which could potentially be identified. However, the point is moot because limits on the diffuse extragalactic $\gamma$-ray background limit the size of our matter domain to be essentially as large as the horizon \cite{CohRujGla98}. \subsection{Alternative Gravity Models} We now know through a variety of experiments that general relativity provides an accurate accounting of observed gravitational phenomena. On the other hand, string theories generically predict at least some small deviation from general relativity, often in the form of scalar-tensor theories of gravity \cite{Ber68,Nor70,Wag70,Bek77,BekMei78}. The simplest of these is Jordan-Fierz-Brans-Dicke (more commonly, Brans-Dicke) theory \cite{Jor49,Fie56,Jor59,BraDic61,Dic62,Dic68}. An inflation theory (``extended inflation'') based on Brans-Dicke gravity \cite{LaSte89a,LaSte89b} was ruled out by the isotropy of the CMB \cite{Wei89,LaSteBer89}, although models based on more complicated scalar-tensor theories (``hyperextended inflation'') have also been considered \cite{SteAcc90,BarMae90,GarQui90,HolKolWan90}. Brans-Dicke theory includes a scalar field $\Phi$ and a new parameter $\omega$. As $\omega \rightarrow \infty$, the theory recovers general relativity (in some sense). Solar-system constraints from Viking spacecraft data limit $\omega \geq 500$ (for a review, see \cite{Wil93}) and recent Very-Long Baseline Interferometry measurements of time delays of millisecond pulsars may further raise this limit \cite{Wil98}. In cosmological models based on Brans-Dicke theories, general relativity is an attractor solution \cite{DamNor93a,DamNor93b}, so gravity could have conceivably differed from general relativity in the early Universe even if it resembles general relativity today. Because the expansion rate and growth of gravitational-potential perturbations are different in alternative-gravity theories, the precise predictions for CMB power spectra should be different in these models. The epoch of matter-radiation equality is altered in Brans-Dicke theories, and this may produce an observable signal in forthcoming precise CMB maps \cite{LidMazBar98}. Cosmological perturbation theory in scalar-tensor theories has been worked out \cite{Nar69,PeeYu70,BapFabGon96,ChiSugYok98} and the CMB power spectra calculated (X Chen, M Kamionkowski, manuscript in preparation). If the scalar-field time derivative $\dot\Phi$ is fixed to be small enough to be consistent with big-bang-nucleosynthesis constraints \cite{KamTur90,DamGun91,CasGarQui92,DamPic98} and $\omega>500$, then the differences between the general-relativistic and Brans-Dicke predictions is small, although conceivably detectable with the Planck Surveyor. Of course, the scalar-field evolution may be significantly different in more sophisticated scalar-tensor theories, but predictions for these models have yet to be carried out. \subsection{Cosmic Rays} We close this tour of the CMB/particle intersection with possibly the oldest and most venerable connection between these two topics. Soon after the initial discovery of the CMB, it was realized that cosmic rays with energy $E\ga5\times 10^{19}$ eV can scatter from CMB photons and produce pions. If a cosmic ray is produced with an energy above $5\times 10^{19}$ eV, repeated scatterings will reduce its energy to below this threshold within a distance of about 50 Mpc \cite{Gre66,ZatKuz66,Cro92,ElbSom95} (the Greisen-Zatsepin-Kuzmin bound). These constraint have become increasingly intriguing recently, as several cosmic rays with energies $>10^{20}$ eV have been observed \cite{Lin63,Biretal94,Hayetal94,Biretal95}, and they do not appear to be coming from any identifiable astrophysical sources (e.g.\ radio galaxies or quasars \cite{Hil84}) as near as 50 Mpc \cite{Hayetal94,Biretal95,ElbSom95,Bie97}. So where are these cosmic rays coming from? Some possibilities are exotic production mechanisms such as topological defects \cite{Hiletal86,AhaBhaSch92,BhaHilSch92,Chietal93, SigSchBha94} or supermassive unstable particles (\cite{KuzRub98,BerKac98,BirSar98}; see \cite{SigBha98} for a review). If a recently claimed alignment of the highest-energy events with very distant radio quasars \cite{FarBie98} is confirmed by larger numbers of events, then it may be that these cosmic rays are exotic particles that interact with baryons but not photons \cite{ChuFarKol98,AlbFarKol99}, e.g.\ supersymmetric $S_0$ baryons \cite{Far84,Far95,Far96}. In the absence of any compelling traditional astrophysical origin, it seems that the simultaneous existence of the CMB and these cosmic rays may be pointing to some intriguing new particle physics. \section{Summary, Current Results, and Future Prospects} The primary cosmological observables pursued by CMB experiments are the frequency spectrum of the CMB, parameterized by $\mu$ and $y$ distortions, and the angular temperature and polarization power spectra, $C_l^{\rm TT}$, $C_l^{\rm GG}$, $C_l^{\rm CC}$, $C_l^{\rm TG}$, $C_l^{\rm TC}$, and $C_l^{\rm GC}$. There are additional observables, such as higher-order correlation functions or cross-correlation of the CMB temperature/polarization with other diffuse extragalactic backgrounds. Rough estimates of the $C_l^{\rm TT}$ at degree angular scales were obtained \cite{WhiScoSil94} from the first generation of ground-based and balloon-borne CMB experiments. Forthcoming experiments will require far more sophisticated techniques for disentangling the CMB from foregrounds, and for recovering the power spectra from noisy data and from maps that cover only a fraction of the sky. A large literature is now devoted to these important issues, which we cannot review here. Progress in CMB experiments is so rapid at the time of writing that any current data we might review would almost certainly become obsolete by the time of publication. We therefore refrain from showing any experimental results in detail and instead describe the current observations qualitatively. First, there is the isotropy of the CMB, which has long been explained only by inflation. Among the numerous pre-\COBE\ models for the origin of large-scale structure, those based on a nearly scale-free spectrum of primordial adiabatic perturbations seem to account most easily for the amplitudes of both the large-angle CMB anisotropy measured by \COBE\ and the amplitude of clustering in galaxy surveys. The galaxy distribution seems to be consistent with primordial Gaussianity. Moreover, data from a large number of CMB experiments that probe the angular power spectrum at degree angular scales have now found (fairly convincingly) that there is significantly more power at degree angular separations ($l\sim200$) than at \COBE\ scales, as one would expect if the acoustic peaks do exist, but in apparent conflict with most theorists' expectations for the degree-scale anisotropy in topological-defect models. The existence of this small-scale anisotropy further suggests no more than a small level of reionization (i.e.\ $\tau\ll 1$). At the time of writing, the measurements are not precise enough to discern either the first or any higher peaks in the temperature power spectrum (some recent data are shown in Reference \cite{Kam98} and are usually updated at Reference \cite{Teg99}). Some experiments have claimed to see the outline of a first acoustic peak at $l\sim200$ \cite{Petetal99} (which would indicate a flat Universe). Moreover, some maximum-likelihood analyses of combined results of all experiments claim that the data indicate a flat Universe \cite{LinBar98,Hanetal98}. However, these results are not yet robust. Thus, although inflation is by no means yet in the clear, observations do seem to be pointing increasingly toward inflation. MAP and the Planck Surveyor will soon make far more precise tests of inflation (see \cite{Kam98} for simulated data from MAP and Planck). First of all, the predictions of primordial adiabatic perturbations will be tested with unprecedented precision by the peak structure in the CMB temperature power spectrum. If the peaks do appear, then MAP and the Planck Surveyor should be able to measure the total density $\Omega$ to a few percent or better \cite{Junetal96a} by determining the location of the first acoustic peak \cite{KamSpeSug94}. Moreover, by fitting MAP and Planck satellite data to theoretical curves, such as those shown in Figure \ref{fig:models}, precise information on the values of other classical cosmological parameters can also be obtained \cite{Junetal96b,BonEfsTeg97,ZalSelSpe97,DodKinKol97,BonEfs98}. If nonrelativistic matter outweighs baryons, then it should become evident with MAP and Planck. The existence of a cosmological constant will further be tested, and some of the tests of gravity, decaying particles, etc, that we have reviewed will become possible. If MAP and Planck confirm that the Universe is flat and that structure grew from primordial adiabatic perturbations, then the next step will be to search for the gravitational-wave background predicted by inflation. Such a gravitational-wave background could be isolated uniquely with the curl component of the polarization. If the inflaton-potential height is $V^{1/4}\ll 10^{15}$ GeV, then the gravitational-wave background will be unobservably small. However, if inflation had something to do with grand unification (i.e.\ $V^{1/4} \sim 10^{15-16}$~GeV, as many theorists surmise), then the curl component of the polarization is conceivably detectable with the Planck Surveyor or with a realistic next-generation dedicated polarization satellite experiment. If detected, the curl component would provide a ``smoking-gun'' signature of inflation and indicate unambiguously that inflation occurred at $T\sim10^{15-16}$ GeV. Although an observable signature is by no means guaranteed, even if inflation did occur, the prospects for peering directly back to $10^{-40}$ sec after the big bang are so tantalizing that a vigorous pursuit is certainly warranted. \bigskip \leftline{\textsc{Acknowledgments}} We thank R Caldwell, A Liddle, and L Wang for very useful comments. MK was supported by a DOE Outstanding Junior Investigator Award, DE-FG02-92ER40699, NASA Astrophysics Theory Program grant NAG5-3091, and the Alfred P. Sloan Foundation. AK was supported by NASA Astrophysics Theory Program grant NAG5-7015 and acknowledges the kind hospitality of the Institute for Advanced Study. {\twocolumn
\section{Introduction} According to the theory of strong interactions, Quantum Chromodynamics (QCD)~\cite{qcd}, the strong coupling constant \as\ is the same for all quark flavours. Therefore a precise measurement of \as\ for the individual quark flavours is an important test of this theory. The coupling constant \as\ for charm and bottom quarks can be compared to \as\ for light (uds) quarks by measuring the ratios \rascasu\ and \rasbasu\ using hadronic events of the type $\epem \rightarrow \qqbar g$. The coupling constants are measured using an event sample originating from a pair of light quarks (\uubar , \ssbar\ or \ddbar), c quarks (\ccbar) or b quarks (\bbbar), respectively. If either of the above ratios deviates significantly from unity then this may indicate physics beyond the Standard Model. In many QCD studies at LEP corrections due to quark mass effects can be safely ignored because they typically appear as powers of the ratio of the quark mass to the total energy. However, in those studies where \as\ is determined in event samples enriched in heavy quarks, mass effects become non-negligible. Gluon emission by bottom quarks, and to a lesser extent charm quarks, will be suppressed largely due to the reduced phase-space available. Observables sensitive to the three-jet rate measured in heavy quark events will be modified with respect to the same quantities measured in light quark events. Tests of the flavour independence of \as\ have previously been conducted at both LEP and SLC ~\cite{flvtstopal1,flvtstopal2,flvtstl3,flvtstdelphi,flvtstaleph,flvtstsld} using ratios of \as\ for one flavour over \as\ of either the complementary \footnote{For example, \as\ for c quarks was compared to \as\ for a mixture of u, d, s and b quarks.} or inclusive quark mixture. In all these tests flavour independence of \as\ was confirmed. These tests, however, used either massless QCD calculations or a leading order calculation with massive quarks~\cite{heavyQ0}. The leading order calculation includes the process $\epem \rightarrow \qqbar gg$, with mass effects, but virtual corrections to the process $\epem \rightarrow \qqbar g$ are not included. Complete next-to-leading-order (NLO) calculations of the heavy-flavour production cross section in \epem\ collisions, including quark mass effects, have been published recently~\cite{heavyQ1, heavyQ2, heavyQ3}. A comparison between these calculations has been made and they were found to be in agreement~\cite{heavyQ3}. In recent publications DELPHI~\cite{delphi_massive} and SLD~\cite{sld_massive} report on measurements of \rasbasu\ where NLO massive calculations~\cite{heavyQ1,heavyQ2} were used to account for mass effects in b quark events. These results are consistent with the flavour independence of \as . In this study we use the results of P.~Nason and C.~Oleari~\cite{heavyQ3} along with theoretical predictions assuming massless quarks in fits to global event shape distributions in order to determine \asu\ , \rascasu\ and \rasbasu. The results presented here are intended to update and supersede the corresponding OPAL results in~\cite{flvtstopal1,flvtstopal2} insofar as we now have greatly increased the charm event statistics and have used improved theoretical predictions. This paper is organised as follows. In Sect.~\ref{sec:dete} the parts of the OPAL detector most important to this analysis are described. In Sect.~\ref{sec:evt} the hadronic event sample and the Monte Carlo event sample are introduced. In Sect.~\ref{sec:buds} and \ref{sec:dstar} the flavour tagging methods are described. In Sect.~\ref{sec:shp} the event shape observables used in this study are introduced and the procedure for correcting the event shape distributions is explained. Next, in Sect.~\ref{sec:oas2}, the procedure for fitting the NLO QCD prediction to the corrected distributions is explained and in Sect.~\ref{sec:sys} the systematic uncertainties that have been taken into account are discussed. In Sect.~\ref{sec:res}, the results of the test of flavour independence of \as\ are presented. Finally, in Sect.~\ref{sec:con}, conclusions are drawn. \section{The OPAL detector} \label{sec:dete} The OPAL detector operates at the LEP \epem\ collider at CERN. A detailed description can be found in Refs.~\cite{opaldete,si}. The analysis presented here relies mainly on the reconstruction of charged particle trajectories and momenta in the central tracking chambers, on energy deposits (``clusters'') in the electromagnetic calorimeters and on information from the silicon micro-vertex detector. All tracking systems are located inside a solenoidal magnet which provides a uniform magnetic field of 0.435~T along the beam axis\footnote{In the OPAL coordinate system the $x$ axis points towards the centre of the LEP ring, the $y$ axis points upwards and the $z$ axis points in the direction of the electron beam. The polar angle $\theta$ and the azimuthal angle $\phi$ are defined with respect to the $z$- and $x$-axes, respectively, while $r$ is the distance from the $z$-axis.}. The magnet is surrounded by a lead glass electromagnetic calorimeter and a hadron calorimeter of the sampling type. Outside the hadron calorimeter, the detector is surrounded by a system of muon chambers. There are similar layers of detectors in the barrel ($|\rm{cos}\theta | < 0.82$) and endcap ($|\rm{cos}\theta | > 0.81$) regions. The central tracking detector consists of a silicon micro-vertex detector~\cite{si} and three drift chamber devices: the vertex detector, a large jet chamber, and surrounding $z$-chambers. The silicon micro-vertex detector, close to the beam pipe, consists of two layers of silicon strips with a single-hit resolution of about 7\,$\mu$m in the $r\phi$ plane. The vertex chamber is a cylindrical drift chamber covering a range of $|\cos\theta|<0.95$. Its single hit resolution is 50\,$\mu$m in the $r\phi$ plane and 700\,$\mu$m in the $z$ direction. The jet chamber is a cylindrical drift chamber with an inner radius of 25\,cm, an outer radius of 185\,cm, and a length of about 4\,m. Its spatial resolution is about 135\,$\mu$m in the $r\phi$ plane from drift time information and about 6\,cm in the $z$ direction from charge division. The $z$-chambers provide a more accurate $z$ measurement with a resolution of about 300\,$\mu$m. In combination, the three drift chambers yield a momentum resolution of $\sigma_{p_t}/p_t \approx \sqrt{0.02^2+(0.0015\cdot p_t)^2}$ for $|\cos(\theta)| < 0.7$, where $p_t$ is the transverse momentum in GeV/$c$. Electromagnetic energy is measured by lead glass calorimeters surrounding the solenoid magnet coil. They consist of a barrel and two endcap arrays with a total of 11704 lead glass blocks covering a range of $|\cos\theta|<0.98$. \section{Event sample and Monte Carlo simulation} \label{sec:evt} This analysis is based on a sample of 4.4 million hadronic decays of the \znull\ recorded with the OPAL detector between 1990 and 1995. Hadronic \znull\ decays were selected by placing requirements on the number of reconstructed tracks and the energy deposited in the calorimeter. A detailed description of the criteria is given in~\cite{bib-OPALmh}. The parts of the detector essential for the present analysis (central detector and electromagnetic calorimeter) were required to be fully operational. The track selection criteria were the same as presented in a previous OPAL study~\cite{opalresummed}. The number of accepted tracks was required to be at least five to reduce $\tau^+\tau^-$ background. Clusters of electromagnetic energy were used if their observed energy was greater than 0.25\,GeV, and known noisy channels in the detector were ignored. The event thrust axis~\cite{opalresummed} was determined using all accepted tracks and clusters, and its direction was required to fulfil the condition $|\cos\theta_\mathrm{Th}|<0.9$ in order that the event be well contained. Using these selection criteria, Monte Carlo studies indicate that, within the chosen range of $\cos\theta_\mathrm{Th}$, 99.86$\pm$0.07\% of hadronic \znull\ decays are accepted, with a contamination of about 0.14\% from $\tau^+\tau^-$ events, and around 0.07\% from two-photon interactions~\cite{opalresummed}. To correct the measured event shape distribitions (see Sect.~\ref{sec:cor}), 4 million hadronic decays of the \znull\ have been simulated using the JETSET 7.4 Monte Carlo model~\cite{bib-JETSET} with parameters tuned to represent OPAL data well \cite{bib-OPALtune}. For all simulated events heavy quark fragmentation has been implemented using the model of Peterson et al.~\cite{bib-PETERSON}. All events have been passed through a detailed simulation of the OPAL detector~\cite{bib-OPALGOPAL} before being analysed using the same programs as for data. \section{Selection of uds and b quark events} \label{sec:buds} Two event samples, one enriched in uds quark events and one enriched in b quark events, were selected by first determining the number of tracks in each event ($N_{sig}$) with a large impact parameter significance, $b/\sigma_b > 2.5$. Here $b$ is the distance of closest approach in the $xy$ plane of the track to the \epem\ interaction point (IP) and $\sigma_b$ its error. The sign of $b$ was determined with respect to the crossing point between the track and the jet axis. $b$ is positive if the track crosses the jet axis downstream of the IP and negative otherwise. An event was classified as having a high probability to have come from light quarks if $N_{sig}$ was zero and from b quarks if $N_{sig} \geq 5$. In the text below these will be referred to as the ``uds-tag'' and the ``b-tag'' event samples. For the uds and b quark event selection all data recorded during 1994 were used. This represents 1.4 million events. The analysis was restricted to only the 1994 data because of the uniform configuration of the silicon micro-vertex detector during this time period. In addition to the track selection criteria outlined above, the track was required to contain at least one silicon hit and an algorithm was applied to reject tracks which were consistent with arising from photon conversions~\cite{bib-idgcon}. In order for the event to be contained within the acceptance of the silicon micro-vertex detector the event's thrust axis was further restricted to lie within the range $|\cos\theta_\mathrm{Th}| < 0.7$. The distribution of $N_{sig}$ is shown in Fig.~\ref{fig:nsig} compared with the result of Monte Carlo simulation. The simulation is decomposed into the contributions from uds, c and b quark events. The general agreement between the sum of the three contributions from Monte Carlo and the data is good. This tagging procedure resulted in 325\,111 events selected for the uds-tag event sample and 71\,521 for the b-tag event sample. The flavour compositions of the uds-tag and the b-tag event samples as determined from the Monte Carlo, along with their combined statistical and systematic errors, are presented in Table~\ref{tab:flavor}. The efficiency for tagging uds events was about 35\% and for tagging b events 23\%. \subsection{Uncertainties in Flavour Composition} The systematic errors on the flavour compositions result from uncertainties in the detector modelling and imprecise knowledge of physics processes. For each source of uncertainty some aspect of the Monte Carlo was varied and the flavour composition of the uds-tag and b-tag event samples were recalculated. The difference between the flavour fractions calculated for each variation and the central value was taken as a systematic error on the determined flavour fractions. \subsubsection{Detector Modelling Uncertainties} Since the determination of the flavour fractions of the uds-tag and the b-tag event sample depend on Monte Carlo this requires an accurate simulation of the detector resolution for charged tracks measured with the silicon micro-vertex detector. The simulation has been tuned to reproduce the tracking resolutions seen in data by studying the impact parameter distributions of tracks, as functions of track momentum, polar angle and the different sub-detectors contributing hits. This tuning procedure was affected by uncertainties in the radial alignment within the silicon micro-vertex detector, the efficiency for associating silicon hits to tracks and the modelling of known inefficient regions, and the overall track reconstruction efficiency. These uncertainties were evaluated according to the procedure given in \cite{opalrb} and their effect on the flavour fractions determined from Monte Carlo was calculated. In addition, the agreement between data and Monte Carlo in Fig.~\ref{fig:nsig} was improved by degrading the resolution of impact parameters in the Monte Carlo simulation by 5\%. This was done by applying a single multiplicative factor $\beta$ to the difference between the reconstructed and true impact parameters. Fig.~\ref{fig:nsig} is shown with this smearing applied. To evaluate the sensitivity of the determined flavour fractions to the tracking resolution the Monte Carlo smearing was removed. \subsubsection{Physics Modelling Uncertainties} The tagging efficiencies of the uds and b tags for the various quark flavours are also sensitive to various physics input parameters in the Monte Carlo simulation. The rate of gluon splitting to \ccbar\ was varied in the range $(2.38\pm 0.48)\times 10^{-2}$, based on the OPAL measurement \cite{opal_glue_split}, and the rate of gluon splitting to \bbbar\ was varied in the range $(3.1\pm 1.1)\times 10^{-3}$, based on theoretical expectation \cite{glue_split_calc}. The error due to these two sources was negligible. The production fractions of the different weakly decaying b hadrons were varied according to the experimental uncertainties \cite{PDG}. The production fractions of the weakly decaying c hadrons, the fragmentation of b and c quarks, and the charged decay multiplicities and lifetimes of b and c hadrons were varied according to the prescription given in \cite{opalrb}. The largest effect on the tag flavour fractions comes from the b hadron charged decay multiplicity. All these uncertainties, along with the uncertainties due to the detector modelling, were added in quadrature and are given as the errors on the flavour fraction shown in Table~\ref{tab:flavor}. \section{Selection of c quark events} \label{sec:dstar} Events having a high probability to have originated from \ccbar\ were identified by the presence of a highly energetic D$^{*+}$ meson\footnote{Throughout this paper charged conjugate modes are always implicitly included.}. These events will be referred to in the text below as the ``c-tag'' event sample. For the D$^{*+}$ reconstruction and the determination of the flavour composition of the c-tag event sample the methods of a previous OPAL study have been used~\cite{dstarsel}. In brief, five D$^{*+}$ decay modes were searched for: \begin{center} \begin{tabbing} \hspace{5cm} \= \hspace{5cm} \= \kill \> ${\rm D^{*+}} \rightarrow {\rm D^0}\pi^+$ \\ \> $\phantom{D^{*+} \rightarrow }\hspace{4pt} \downto {\rm K^-}\pi^+$ \>``3 prong''\ ,\\ \> $\phantom{D^{*+} \rightarrow }\hspace{4pt} \downto {\rm K^-}\pi^+\pi^0$ \> ``satellite''\ ,\\ \> $\phantom{D^{*+} \rightarrow }\hspace{4pt} \downto {\rm K^-}\pi^+\pi^-\pi^+$ \>``5 prong''\ ,\\ \> $\phantom{D^{*+} \rightarrow }\hspace{4pt} \downto {\rm K^-}{\rm e}^+ \nu_{{\rm e}}$ \>``electron''\ ,\\ \> $\phantom{D^{*+} \rightarrow }\hspace{4pt} \downto {\rm K^-}\mu^+\nu_{\mu}$ \>``muon''\ .\ \end{tabbing} \label{eq-decaychannels} \end{center} \noindent No attempt was made to reconstruct the $\pi^0$ in the satellite channel, nor the neutrino direction or energy in the electron and muon channels. The last two channels are referred to as ``semileptonic'' channels in the following text. A number of tracks appropriate for the selected channel were combined to form a D$^0$ candidate and their invariant mass was calculated. Candidates were selected if the reconstructed mass lay within the expected range for that channel. After adding a further track as a possible pion from the D$^{*+}$ decay, the combined mass was calculated and the candidate was selected if the mass difference $\Delta M = M_{D^{*+}}-M_{D^0}$ was within certain limits. Some of the D$^{*+}$ selection criteria are given in Table~\ref{tab:dstarcuts}. To reduce the background in the c-tag event sample we required $x_{{\rm D^{*+}}} > 0.4$, where $x_{{\rm D^{*+}}}=E_{{\rm D^{*+}}}^{\mathrm calc}/E_{\rm beam}$ is the scaled energy\footnote{In this paper any reference to the scaled energy $x$ of a D$^{*+}$ candidate is to be understood as being the calculated energy of the D$^{*+}$, $E_{{\rm D^{*+}}}^{\mathrm calc}$, obtained from the reconstructed tracks, without correcting for missing or wrongly associated tracks, divided by the beam energy $E_{\mathrm beam}$.} of the D$^{*+}$ meason. For the 5-prong event selection we required $x_{{\rm D^{*+}}} > 0.5$. This cut is effective in rejecting D$^{*+}$ mesons which originate from cascade decays of B hadrons and from events where a gluon splits into a pair of charm quarks. The distribution of $\Delta M$ for all five decay channels is presented in Fig.~\ref{fig:delm}. The points with error bars are the signal candidates and the solid histograms are background estimator distributions constructed from data. The background estimator was constructed by choosing the candidate for the pion in the $ {\rm D^{*+}} \rightarrow {\rm D^0}\pi^+$ decay from the opposite hemisphere relative to the rest of the decay products, reflecting it through the orgin, and then using it in the calculation of the invariant mass. A significant fraction of the sample of D$^{*+}$ mesons were only partially reconstructed. These mesons produce an enhancement in the $\Delta M$ spectrum very similar to the true signal. Only a few of the events are present in the 3-prong sample. They are more important in the 5-prong tagged events where a clear tail is visible in the $\Delta M$ distribution for values above 0.145 GeV (see Fig.~\ref{fig:delm}d). Since such events originate from D$^{*+}$ decays, they can still be used in the analysis. In all, 27\,005 D$^{*+}$ candidate events were selected and the background was estimated to be $11\,366 \pm 107$ events, where the error on the number of background events is statistical only. All of the LEP-1 data recorded by OPAL, 4.4 million events, was used for the D$^{*+}$ event selection. The candidate events are composed of three components: genuine D$^{*+}$ mesons from b events, genuine D$^{*+}$ mesons from c events and combinatorial background which is a mixture of uds, c and b quark events. Genuine D$^{*+}$'s from uds quark events can only occur in events where a gluon splits into two heavy quarks. The possibility of D$^{*+}$'s from this source was neglected because these events are highly suppressed due to the high $x_{\rm D^{*+}}$ cut. Using Monte Carlo simulation the total contribution of this source to the number of D$^{*+}$ candidates was found to be $(0.2 \pm 0.1)$\%, where the error quoted is due to Monte Carlo statistics. In Ref.~\cite{dstarsel} the fraction of genuine D$^{*+}$ mesons originating from c events was determined by OPAL to be $f_{\rm c}^{{\rm D^{*+}}} = 0.774 \pm 0.023$, where the error is the combined statistical and systematic error. The fraction of genuine D$^{*+}$ mesons from b events is given by $f_{\rm b}^{{\rm D^{*+}}} = 1 - f_{\rm c}^{{\rm D^{*+}}}$. The fractions of uds, c and b quark events in the combinatorial background, as determined from Monte Carlo, were $0.584 \pm 0.009$, $0.238 \pm 0.009$ and $0.178 \pm 0.009$, respectively. The errors are a combination of statistical errors due to finite Monte Carlo statistics and a systematic error accounting for the overall quality of the background estimation procedure. Based on a Monte Carlo study~\cite{dstarsel} the background estimate was found to be accurate to within 1\%, and an additional 1\% error was therefore assigned to the number of background events. Since that study was flavour blind, a conservative approach was taken here where the entire error was assigned in turn to each flavour component in the combinatorial background. The overall flavour composition of the c-tag event sample is presented in Table~\ref{tab:flavor} along with the combined statistical and systematic errors. The efficiency for tagging c events was about 2.0\%. \section{Event shape observables} \label{sec:shp} For each flavour tagged event sample described above (uds-tag, c-tag and b-tag) the distributions of the event shape variables 1-Thrust ($1-T$), Heavy Jet Mass scaled by the centre-of-mass energy ($M_H/\sqrt{s}$), Wide Jet Broadening ($B_W$), the $y_{cut}$ at which an event changes from being a 2-jet event to being a 3-jet event ($y_{23}$) determined using the Durham jet finder and the C-parameter ($C$) were determined. The definitions for these observables are given in~\cite{opalresummed, as_global} and the references therein. These quantities were measured using all tracks and electromagnetic clusters which satisfied the selection criteria described in Sect.~\ref{sec:evt}. In order to extract the values of \rascasu\ and \rasbasu , the measured event shape distributions were fitted using a QCD analytic calculation~\cite{heavyQ3}. Since the QCD calculation is only valid for event shape distributions determined from final-state partons, hadronization effects caused by the transformation of final-state partons into hadrons, which are experimentally accessible, must be taken into account. This was done by applying correction factors to the analytic predictions. The measured event shape distributions must be corrrected for experimental effects which distort them. These effects include finite detector resolution, initial-state photon radiation and biases introduced by the flavour tagging methods. In the text below describing the correction procedure the term {\it detector level} is used to refer to distributions determined using the measured tracks and electromagnetic clusters and {\it hadron level} to refer to these distributions corrected for detector resolution, initial state radiation and biases introduced by the flavour tagging methods. The comparisons between the measured distributions and the QCD predictions were performed at the hadron level. The event shape variable Total Jet Broadening ($B_T$)~\cite{opalresummed} was initially considered to be included in this analysis. It was found, however, that for the distribution of $B_T$ measured with b quark events, the size of the hadronization corrections were greater that 20\% over the entire distribution. Therefore this variable was dropped from further consideration. \subsection{Correction procedure} \label{sec:cor} In the first step the selected events were corrected for distortions caused by finite detector resolution using an unfolding matrix. The unfolding matrix was constructed using a Monte Carlo data sample including initial-state radiation, full detector simulation and subjected to the same event selection criteria that were applied to the data. The Monte Carlo events which pass the selection criteria were used to calculate a correction matrix $M^{q-tag}$. The element $M^{q-tag}(y_i,y_j)$ gives the probability that an event shape observable $y$ measured at the detector level and located in bin $i$ of its corresponding distribution, has migrated from bin $j$ on the hadron level. A matrix was computed for each event shape observable and flavour-tagged event sample, ``q-tag'', where q-tag was either uds-tag, c-tag or b-tag. In the next step bin-by-bin corrections are used to correct the data for biases introduced by the flavour tagging methods, event acceptance and the effects of initial-state photon radiation. Defining $G^q(y_i)$ to be the number of events of flavour $q$ in the untagged Monte Carlo sample and $H^{q-tag}(y_i)$ the number of events with the tag applied, the correction factor $K^{q-tag}(y_i)$ for the $i$-th bin is given by \begin{eqnarray} K^{q-tag}(y_i) = \frac{f^{q-tag}_{\rm uds} G^{\rm{uds}}(y_i) + f^{q-tag}_{\rm c} G^{\rm{ c}}(y_i) + f^{q-tag}_{\rm b} G^{\rm{ b}}(y_i) } {H^{q-tag}(y_i)} \end{eqnarray} \noindent The factors $f_q^{q-tag}$, taken from Table~\ref{tab:flavor}, are the fractions of events of flavour $q$ in each flavour-tagged event sample. The terms $G^q(y_i)$ are normalized to the total number of events of flavour $q$. Fig.~\ref{fig:bwcor} shows, as an example, the size of the bin-by-bin corrections $K^{q-tag}$ for the observable $B_W$ measured using each flavour-tagged event sample. The size of the corrections for $B_W$ were typical of the other event shape observables studied. One can see for the distributions measured with the uds-tag and b-tag event samples that, within the chosen fit range, the size of the corrections are about 10\%. (An explanation of how the fit range was chosen is presented in the next section.) In contrast, the size of the corrections for the c-tag event sample were an order of magnitude larger. This is due to a kinematic bias introduced by requiring the events in the c-tag sample to contain a high-$x$ $D^{*+}$ meson. The number of events in bin $i$ corrected to the hadron level is given by \begin{eqnarray} N^{q-tag,cor}(y_i) = K^{q-tag}(y_i) \sum_{j} M^{q-tag}(y_i,y_j)N^{q-tag}(y_j)~~, \end{eqnarray} \noindent where $N^{q-tag}(y_j)$ is the uncorrected number of events in the $j$-th bin of the event shape distributions in question. This correction procedure does not depend on the values of \asu , \rascasu\ or \rasbasu\ in the Monte Carlo samples. The resulting distributions normalized to the total hadronic cross section are given by \begin{eqnarray} \left(\frac{1}{\sigma_{tot}}\frac{\rm{d}\sigma}{\rm{d}y}\right)^{q-tag,cor} = \frac{1}{\Delta y_i \cdot \sum_{j} N^{q-tag,cor}(y_j)} N^{q-tag,cor}(y_i)~~, \end{eqnarray} \noindent where $\Delta y_i$ is the width of the $i$-th bin and $y$ corresponds to one of the five event shapes studied. \section{Fit procedure} \label{sec:oas2} The value of \asu\ and the ratios \rascasu\ and \rasbasu\ were determined by simultaneously fitting theoretical predictions for a particular event shape observable to three hadron-level event shape distributions: one determined with the uds-tag event sample, one with the c-tag event sample and one with the b-tag event sample. The theoretical predictions were given by a linear combination of the theoretical predictions for each of the three tagged flavours: \begin{eqnarray} \label{eq:theo_shp} \left(\frac{1}{\sigma_{tot}}\frac{\rm{d}\sigma}{\rm{d}y}\right)^{q-tag,th} &=& \label{eqn:lincomb} f_{\rm{uds}}^{q-tag}R(y)^{\rm{uds}}\left(\frac{1}{\sigma_{tot}}\frac{\rm{d}\sigma}{\rm{d}y}\right)^{{\rm uds},th} \nonumber + f_{\rm{ c}}^{q-tag}R(y)^{\rm{c }}\left(\frac{1}{\sigma_{tot}}\frac{\rm{d}\sigma}{\rm{d}y}\right)^{{\rm c },th} \\ & & \mbox{} + f_{\rm{ b}}^{q-tag}R(y)^{\rm{b }}\left(\frac{1}{\sigma_{tot}}\frac{\rm{d}\sigma}{\rm{d}y}\right)^{{\rm b },th}~~. \end{eqnarray} \noindent The coefficients $f_q^{q-tag}$ are the flavour fractions given in Table~\ref{tab:flavor} and ($1/\sigma_{tot}\cdot {\rm d}\sigma/{\rm d}y)^{\rm{uds},th}$, ($1/\sigma_{tot}\cdot {\rm d}\sigma/{\rm d}y)^{\rm{c},th}$ and ($1/\sigma_{tot}\cdot {\rm d}\sigma/{\rm d}y)^{\rm{b},th}$ are the theoretical predictions for an event shape observable $y$ measured with a sample of uds, c and b quark events, respectively. The factors $R(y)^q$ correct the theoretical prediction for a particular tagged flavour $q$ for hadronization effects so that the theoretical predictions, valid for final state partons, can be compared directly with the measured distributions corrected to the hadron level. The hadronization correction factors $R(y)^q$ were computed from JETSET 7.4 using the parton shower option by taking the ratio of an event shape distribution at the hadron level to the same distribution at the parton level. Here the parton level is defined by the cut-off $Q_0$ of the QCD shower in JETSET which is set to 1.9\,GeV~\cite{bib-OPALtune}. After the termination of the parton shower the partons are transformed into hadrons using string hadronization~\cite{string-had}. For this process we choose a hybrid scheme for the longitudinal fragmentation function where light quarks are treated with the symmetric Lund fragmentation function and charm and bottom quarks according to the model of Peterson et al.~\cite{bib-PETERSON}. The c and b quark masses in JETSET were set to their default values of 1.35\,GeV and 5.0\,GeV, respectively. In Equation~(\ref{eqn:lincomb}), the differential cross section of a generic observable $y$, for massless quarks, normalized to the total hadronic cross section is given by~\cite{oas2} \begin{eqnarray} \left(\frac{1}{\sigma_{tot}}\frac{{\rm d}\sigma}{{\rm d}y}\right)^{{\rm uds},th} & = &\nonumber \label{eqn:udsth} \frac{{\rm d}A^{\rm uds}}{{\rm d}y}\left(\frac{\alpha_s^{uds}(\mu)}{2\pi}\right) \\ & & \mbox{} + \left(\left(2\pi \beta_0 \mathrm{log}(x_\mu^2) - 2 \right) \frac{{\rm d}A^{\rm uds}}{{\rm d}y} + \frac{{\rm d}B^{\rm uds}}{{\rm d}y}\right) \left(\frac{\alpha_s^{uds}(\mu)}{2\pi}\right)^2~~. \end{eqnarray} \noindent The coefficients ${\rm d}A^{\rm uds}/{\rm d}y$ and ${\rm d}B^{\rm uds}/{\rm d}y$ are the \oa\ and \oaa\ QCD coefficients, respectively, $\sigma_{tot}$ is the one-loop cross section for the process $\epem \rightarrow \mathrm{hadrons}$ and $\beta_0$ is the coefficient of the QCD beta function for one-loop~\cite{PDG}. The renormalization scale $\mu$ can be related to the \epem\ centre-of-mass energy by \begin{eqnarray} \mu = x_\mu \cdot E_{cm}~~, \end{eqnarray} \noindent where $x_\mu$ is the renormalization scale factor. The coefficients ${\rm d}A^{\rm uds}/{\rm d}y$ and ${\rm d}B^{\rm uds}/{\rm d}y$ were obtained for each event shape observable by integrating the \oaa\ matrix elements in~\cite{oas2}. In Equation~(\ref{eqn:udsth}), in addition to the approximation of massless quarks, the simplifying assumption was made that \as\ is the same for up, down and strange quarks. The terms ($1/\sigma_{tot}\cdot {\rm d}\sigma/{\rm d}y)^{\rm{c},th}$ and ($1/\sigma_{tot}\cdot {\rm d}\sigma/{\rm d}y)^{\rm{b},th}$ in Equation~(\ref{eqn:lincomb}) were both given by an \oaa\ expression for massive quarks. This expression has only recently been made available~\cite{heavyQ3}. The result of this calculation was implemented in a FORTRAN program named ZBB4~\cite{zbb4} analogous to the program EVENT~\cite{event} which was used to integrate the \oaa\ matrix elements of the massless calculation. ZBB4 was run separately for c quark and b quark events in order to calculate the \oa\ and \oaa\ coefficients for the massive calculation. This calculation was performed in the pole mass scheme and the c and b quark pole masses were set to 1.35\,GeV and 5.0\,GeV, respectively. The differential cross section of a generic observable $y$, for massive quarks, normalized to the total hadronic cross section is given by~\cite{zbb4} \begin{eqnarray} \left(\frac{1}{\sigma_{tot}}\frac{{\rm d}\sigma}{{\rm d}y}\right)^{Q,th} & = & \nonumber \label{eqn:Qth} \frac{{\rm d}A^{Q}}{{\rm d}y}\left(\frac{\alpha_s^{Q}(\mu)}{2\pi}\right) \\ & & \mbox{} + \left(\left(2\pi \beta_0 \mathrm{log}(x_\mu^2) - \frac{2}{3}{\mathrm{log}}\left(\frac{\mu}{m_{Q}}\right) - 2 \right) \frac{{\rm d}A^{Q}}{{\rm d}y} + \frac{{\rm d}B^{Q}}{{\rm d}y}\right) \left(\frac{\alpha_s^{Q}(\mu)}{2\pi}\right)^2 \end{eqnarray} \noindent where $Q$ is either c or b and $m_Q$ is the energy scale corresponding to the heavy quark pole mass. When performing the fit using Equation~(\ref{eqn:lincomb}) we make the substitution $\asc~=~\asu \cdot \rascasu$ and $\asb =~\asu \cdot \rasbasu$. This substitution enables \asu\ , \rascasu\ and \rasbasu\ to be determined directly as free parameters in the fit and allows correlations between these variables to be properly taken into account. In tests with Monte Carlo events it was verified that this fitting procedure was sensitive to changes in \as\ for c and b quarks with respect to uds quarks. The fit ranges for each observable were determined by the range of $y$ where the parton to hadron level corrections were below 10\% and the resulting $\chi^2/\mathrm{d.o.f.}$ remained small ($\sim 5$ for $B_W$ and $\sim 2-3$ for the rest). The remaining variations of the fit results due to the choice of the fit ranges were taken as systematic uncertainties. It was also checked that within the fit range the \oaa\ calculation and the JETSET parton shower model were in agreement with each other. The fit ranges used for this analysis are slightly smaller than the fit ranges used in previous OPAL studies~\cite{opalresummed, as_global} which used \oaa\ calculations in fits to global event shapes. This is because the hadronization corrections for distributions of event shape variables calculated with b quark events are generally larger than the corrections for distributions calculated with an inclusive event sample. Consequently, the range of the distribution in which the corrections are below 10\% is smaller. The fit ranges, and the results of the $\chi^2$ fits with $x_\mu = 1$ are shown in Table~\ref{tab:oas2}. The values for \asu\ obtained appear large when compared with \as\ measurements performed using resummed NLO calculations~\cite{opalresummed}. This is due to the fact that fixed order QCD calculations, valid only to \oaa , were used here. To estimate the error in our results due to the choice of the renormalization scale we adopted a method used in another OPAL analysis~\cite{as_global} which used \oaa\ fits to global event shapes. We performed fits with $x_\mu$ fixed equal one, and fits in which $x_\mu$ was an additional free parameter. The average of the two results is taken as our main result, and half their difference as the ``scale uncertainty''. The result of fits with $x_\mu$ as a free parameter are presented in Table~\ref{tab:oas2free}. One sees that the values of \asu\ decrease for all event shapes and that the values of $\chi^2/$d.o.f. show a significant improvement. This strong scale dependence is typical of determinations of \as\ where fixed order calculations are used. It was checked that the values of \asu\ determined here were consistent with previous OPAL measurements of \as~\cite{opalresummed,as_global}. The possibility of a different scale dependence for heavy quark events, due to mass effects present in higher order terms, was investigated by introducing a separate $x_\mu$ parameter for each quark flavour. The effect was found to be negligible within the statistical precision of the measurement. In Figs.~\ref{fig:d2th} through~\ref{fig:cp} the results of fits with $x_\mu = 1$ are plotted along with the measured distributions corrected to the hadron level. The fit ranges used are indicated by the arrow on each figure. It can be seen that within the fit ranges the agreement between the fit results and the data are good. In Fig.~\ref{fig:d2th} one sees a small disagreement between the fit results and the data for values of $1-T > 0.25$. When the fits to $1-T$ were repeated with $x_\mu$ as a free parameter, the agreement between the fit results and data were good over the entire fit range. \section{Systematic uncertainties} \label{sec:sys} The main result was obtained using the default selection and correction procedure described above. The systematic uncertainties were divided into two groups: experimental and theoretical uncertainties. Each uncertainty was estimated by modifying details of the event selection and correction procedure and repeating the analysis. The difference between the results obtained with the standard analysis and the results obtained with the analysis corresponding to each variation were taken (unless otherwise noted) as symmetric systematic errors. In the case where a parameter was varied above or below its nominal value, the largest deviation from the main result was taken as a symmetric systematic error associated with that parameter. Finally, the systematic errors for each type of variation were added in quadrature. The systematic uncertainties investigated are described below and the effects they had on the ratios \rascasu\ and \rasbasu\ are presented in Tables~\ref{tab:sys_ascasu} and \ref{tab:sys_asbasu}. \subsection{Experimental systematic uncertainties} \begin{itemize} \item The error due to the uncertainty on the flavour composition of the flavour-tagged event samples was evaluated by varying the flavour fractions in Table~\ref{tab:flavor} within their errors. \item For the central result event shape observables were measured using all charged tracks and electromagnetic clusters. To evaluate the relative response of the central tracking and the electromagnetic calorimeter the measurements were performed again using electromagnetic clusters only and charged tracks only. \item The homogeneity of the response of the detector in the endcap region was checked by restricting the analysis to the barrel region of the detector, requiring the thrust axis of accepted events to lie within the range $|\cos\theta_{\rm Th}| < 0.7$. This only modifies event shape distributions measured with the c-tag event sample because the uds-tag and b-tag event samples are already restricted to the barrel region. \item The minimum number of accepted charged tracks was increased from 5 to 7 in order to further suppress background from $\tau^+ \tau^-$ events and two-photon interactions. \item The dependency of the result on the chosen fit range was evaluated by adding or subtracting one bin from the lower and upper end of the fit range. Each new fit range was treated as a separate systematic variation. \item In the charm event selection the requirement that $x_{D^{*+}} > 0.4$ leads to a large kinematic bias in the c-tag event sample. The uncertainty due to this aspect of the $D^{*+}$ selection was evaluated by varying the $x_{D^{*+}}$ cut between 0.3 and 0.5. \end{itemize} \subsection{Theoretical systematic uncertainties} \begin{itemize} \item The fragmentation of the final-state partons into hadrons was modelled by JETSET where the longitudinal fragmentation function for light quarks was treated by the symmetric Lund model and for heavy quarks by the model of Peterson et al~\cite{bib-PETERSON}. The parameters for each model were determined from a fit to OPAL data on global event shapes~\cite{bib-OPALtune}. This fit yielded a value of $b = 0.52 \pm 0.04$ ({\tt PARJ(42)}) for the Lund model and $\epsilon_c = 0.031 \pm 0.011$ ({\tt PARJ(54)}) and $\epsilon_b = 0.0038 \pm 0.0010$ ({\tt PARJ(55)}) for the Peterson model. To evaluate the systematic error due to uncertainties in the JETSET hadronization model these parameters were varied independently within one standard deviation of their optimised values. \item In JETSET, the parameter $Q_0$ ({\tt PARJ(82)}) is the virtuality cut-off of partons. It determines the boundary between the pertubative QCD and hadronization phases and is essentially arbitrary. The optimum value to describe the OPAL data was determined to be $Q_0 = (1.90 \pm 0.50)$\,GeV~\cite{bib-OPALtune} . This parameter was varied within one standard deviation of its optimised value. \item The width of the transverse momentum distributions of quarks and antiquarks produced in the fragmentation process is determined by the parameter $\sigma_q = (0.40 \pm 0.03)$\,GeV ({\tt PARJ(21)}). This parameter was varied within one standard deviation of its optimum value. \item The renormalization scale uncertainty was estimated by performing fits with $x_\mu$ fixed equal one, and fits in which $x_\mu$ is an additional free parameter. The average of the two results is taken as our main result, and half their difference as the scale uncertainty. \item To evaluate the dependence of this analysis on the choice of using JETSET to calculate the hadronization corrections, the analysis was repeated using ARIADNE 4.08~\cite{ariadne} instead of JETSET. The ARIADNE parton shower is based upon a colour dipole model and provides an alternative to the Lund parton shower model in JETSET. ARIADNE employs the same fragmentation model as JETSET for the subsequent hadronization. The difference between the results using JETSET hadronization corrections and ARIADNE corrections were used as an estimate of this systematic error. \item The error due to uncertainties in the c and b quark masses was evaluated by varying the quark masses in the ranges given in~\cite{PDG} and repeating the integration of the matrix elements corresponding to massive quarks for each variation. The value for the b quark pole mass was varied from 4.5 to 5.5\,GeV/$c^2$; the c quark pole mass was varied from 1.2 to 1.9\,GeV/$c^2$. \end{itemize} \section{Results} \label{sec:res} The results of the $\chi^2$ fits to determine \rascasu\ and \rasbasu\ for each of the event shape observables studied are presented in Tables~\ref{tab:sys_ascasu} and \ref{tab:sys_asbasu} and summarised in Fig.~\ref{fig:oas2}. The values quoted are the average of fits with $x_\mu = 1$ and with $x_\mu$ as a free parameter. The top half of the Fig.~\ref{fig:oas2} shows the result of determining \rascasu\ and the lower half \rasbasu . The vertical bars on the error bars show the size of the statistical error and the full error bar is the total error, which is the sum of the statistical, experimental systematic and theoretical errors added in quadrature. Also shown is the weighted mean of the results obtained with the six observables. The weights were given by the reciprocal of the square of the total error on the ratios \rascasu\ and \rasbasu\ given in Tables~\ref{tab:sys_ascasu} and \ref{tab:sys_asbasu}. The statistical error on the mean was calculated taking into account correlations between the five observables. The correlation matrix was calculated from 100 Monte Carlo event samples. The systematic uncertainty on the weighted mean was determined from the change in the mean that occurred when each systematic check was applied to all the event shape distributions simultaneously. The mean values of \rascasu\ and \rasbasu\ were determined to be \begin{eqnarray} \rascasu &=& 0.997 \pm 0.038 ~(stat.) \pm 0.030 ~(syst.) \pm 0.012 ~(theory) \nonumber \\ \rasbasu &=& 0.993 \pm 0.008 ~(stat.) \pm 0.006 ~(syst.) \pm 0.011 ~(theory)~. \nonumber \end{eqnarray} \noindent In both cases the results are consistent with unity, indicating flavour independence of \as . As shown in Table~\ref{tab:sys_asbasu}, the largest experimental systematic error on the results obtained from the variable $y_{23}$ originated from moving the lower bound of the fit range from $y_{23} = 0.015$ to 0.025. It was found that this relatively large variation in the result was due to a small discrepancy between the data and the Monte Carlo, of the order of a few percent, for values of $y_{23} < 0.03$. If one excludes the results obtained from $y_{23}$ from the calculation of the weighted means, the mean value obtained for $\alpha_s^b/\alpha_s^{uds}$ remains unchanged within the experimental precision, while the mean value of $\alpha_s^c/\alpha_s^{uds}$ changes from 0.997 to 0.990. The size of the mass effect for c and b quarks and its relevance to the measurement of \as\ are also of interest. From phase space considerations one can estimate the size of the mass effects for 3-jet observables. The ratio of the phase space of two massive quarks and a gluon to the phase space for three massless particles is $1 + 8(\rm{M}_q/\rm{M}_Z)^2 \rm{log}(\rm{M}_q/\rm{M}_Z)$~\cite{rodrigo_thesis}. This represents a 7\% effect for M$_q = 5$\,GeV/$c^2$ and 0.7\% for M$_q = 1.35$\,GeV/$c^2$. Fig.~\ref{fig:masseff}a shows the ratios \rascasu\ and \rasbasu\ determined using \oaa\ massless calculations for both uds and c quarks and \oaa\ massive calculations for b quarks. Again we show the average of fits with \xmu\ = 1.0 and \xmu\ as a free parameter. One sees a small systematic shift of \rascasu\ with respect to the fits with the massive calculation, although the large statistical error on the measurements for c quarks makes a definitive statement on the exact size of the mass effect difficult. In Fig.~\ref{fig:masseff}b, \rascasu\ and \rasbasu\ were determined using massless calculations for uds and b quarks and massive calculations for c quarks. In this case \rasbasu\ shows a large systematic shift of order 5 to 7\%. It should be noted that the values of \asu\ and \rascasu\ remained unchanged within their statistical errors with respect to the results presented in Fig.~\ref{fig:oas2}, therefore the change in \rasbasu\ can be attributed entirely to changes in \asb . This sensitivity to the b quark pole mass ( \mb\ ) suggests that this effect could be exploited to measure \mb\ itself, by assuming flavour independence of \as\ and fitting for \mb . Monte Carlo studies show that the value of \mb\ determined in this way depends strongly on the input value of \mb\ used in JETSET to calculate hadronization corrections, thus making a determination of \mb\ in this way problematic. As expected, fits to $M_H$ do not exhibit the mass effect like the other variables do. Here the quark mass fixes a lower bound of the $M_H$ distribution since the invariant mass of the jet cannot be less than the mass of the quark that originates the jet. For \mb\ = 5.0\,GeV the lower bound lies at about $M_H/\sqrt{s} = 0.05$, which is well outside of the fit range. \section{Conclusion} \label{sec:con} We have presented a test of the flavour independence of the strong coupling constant for charm and bottom quarks with respect to light (uds) quarks. This analysis was based on a sample of hadronic decays of the \znull\ resonance recorded by the OPAL detector at LEP. The global event shapes $y_{23}$, $1-T$, $M_H$, $B_W$ and $C$ were used to measure \as\ in three flavour tagged event samples (uds, c and b). The event shape distributions were fitted by \oaa\ calculations of jet production taking into account mass effects for the c and b quarks. The ratios \rascasu\ and \rasbasu\ were both found to be consistent with unity, indicating the flavour independence of \as . The measurement of \rascasu\ achieved a precision of 5\%. The relatively large statistical error was due to the low efficiency for tagging c quark events ($\approx 2\%$) which relied on finding D$^{*+}$ mesons. The experimental error was dominated by varying the cut on the scaled energy of the D$^{*+}$ mesons which was required to reduce the background from c quark events coming from cascade decays of b hadrons. The measurement of \rasbasu\ achieved a 1.5\% precision, the error being dominated by the theoretical systematic error. The largest theoretical errors were due to uncertainties in the b quark mass and the renormalization scale factor $x_\mu$ . In addition, we have presented a study of the effect of heavy quark masses on global event shape variables. It was observed that the values of \as\ determined from $y_{23}$, $1-T$, $B_W$ and $C$ were reduced by 5 to 7\% when these event shapes were measured with a sample of b quark events and a massless QCD calculation was used. The shape variable $M_H$ was found to be insensitive to the b quark mass. \section{Acknowledgements} \par We particularly wish to thank the SL Division for the efficient operation of the LEP accelerator at all energies and for their continuing close cooperation with our experimental group. We thank our colleagues from CEA, DAPNIA/SPP, CE-Saclay for their efforts over the years on the time-of-flight and trigger systems which we continue to use. 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, \\ Minerva Gesellschaft, \\ Benoziyo Center for High Energy Physics,\\ Japanese Ministry of Education, Science and Culture (the Monbusho) and a grant under the Monbusho International Science Research Program,\\ Japanese Society for the Promotion of Science (JSPS),\\ German Israeli Bi-national Science Foundation (GIF), \\ Bundesministerium f\"ur Bildung, Wissenschaft, Forschung und Technologie, Germany, \\ National Research Council of Canada, \\ Research Corporation, USA,\\ Hungarian Foundation for Scientific Research, OTKA T-016660, T023793 and OTKA F-023259.\\
\section{Introduction} Recent developments in the physics of granular matter \cite{jae-nag96} have illustrated that the dissipative nature of the interactions between grains can result in a variety of different phenomena. Of particular interest in recent years has been the dynamics of vibrated granular materials \cite{warretal95,meloetal95}, which exhibit stationary states as well as waves and complex patterns. In order to describe these diverse states of the material, it is necessary to derive macroscopic descriptions by averaging over the microscopic details of the motion and interactions between individual grains. This goal has proved elusive, however, because a vibrated granular material is a driven dissipative system, and the interactions between the particles are characterised by a loss of energy due to inelastic collisions. The statistical mechanics framework developed for equilibrium or near equilibrium systems cannot be used in this case. Consequently, phenomenological models \cite{shrin97,tsim-aran97,venkat-ott98} have been used to describe the dynamics of granular materials. The kinetic theories developed for granular flows \cite{jen-sav83,kum98:vib} usually assume that the system is close to ``equilibrium'' and the velocity distribution function is close to the Maxwell-Boltzmann distribution. Experimental studies and computer simulations have reported the presence of a uniformly fluidised state in a vibrated bed of granular material. Luding, Herrmann and Blumen \cite{ludetal94} carried out `Event Driven' (ED) simulations of a two dimensional system of inelastic disks in a gravitational field vibrated from below, and obtained scaling laws for the density variations in the bed. An experimental study of a vibrated fluidised bed was carried out by Warr, Huntley and Jacques \cite{warretal95}. Their experimental set up consisted of steel spheres confined between two glass plates that are separated by a distance slightly larger than the diameter of the spheres. The particles were fluidised by a vibrating surface at the bottom of the bed, and the statistics of the velocity distribution of the particles were obtained using visualisation techniques. Profiles for the density and the mean square velocity were obtained, and the particle velocity distributions were also determined at certain positions in the bed. Both of these studies reported that there is an exponential dependence of the density on the height near the top of the bed, similar to the Boltzmann distribution for the density of a gas in a gravitational field. However, the dependence of the density deviates from the exponential behaviour near the bottom. The dependence of the mean square velocity on the vibration frequency and amplitude were found to be different in the two studies. A theoretical calculation of the distribution function in a vibro-fluidised bed was carried out by Kumaran \cite{kum98:vib,kum98:vibscal}. The limit of low dissipation, where the coefficient of restitution $e$ is close to $1$ was considered. In this limit, the mean square velocity of the particles is large compared to the mean square of the velocity of the vibrating surface, and the dissipation of energy during a binary collision is small compared to the energy of a particle. A perturbation approximation is used, where the energy dissipation is neglected in the leading order approximation, and the system resembles a gas at equilibrium in a gravitational field. The velocity distribution function is a Maxwell-Boltzmann distribution, and the density decreases exponentially from the vibrating surface. The first order correction to the distribution due to dissipative effects was calculated using the moment expansion method, and the results were found to be in qualitative agreement with the experiments of Warr et. al. \cite{warretal95}. The theoretical predictions \cite{kum98:vib,kum98:vibscal} were compared with previous experimental and simulation studies by McNamara and Luding \cite{mclud98}. They found that the theory was in good agreement with experiments for dilute beds, where the area fraction of the particles is low, but there were systematic deviations from the theoretical predictions as the area fraction increases. This is to be expected, since the analysis assumed that the density is small and the pair distribution function was set equal to $1$ and therefore the pressure is related to the density by the ideal gas law. These assumptions become inaccurate as the area fraction of the bed increases. An approximate method for including the correction to the pair distribution function was suggested by Huntley \cite{hunt98}. In the present analysis, the correction to the low density theory of Kumaran \cite{kum98:vib,kum98:vibscal} is determined for a vibro-fluidised bed where the coefficient of restitution is close to $1$. An asymptotic analysis is used, where the dissipation is neglected in the leading approximation. The leading order density and velocity profiles are determined using the momentum balance equation in the vertical direction. In contrast to the earlier theory \cite{kum98:vib,kum98:vibscal}, the virial equation of state for a non-ideal two dimensional gas is used to determine the leading order density profile. The density profile differs from the Boltzmann distribution, but the velocity distribution function is still a Maxwell-Boltzmann distribution. The leading order temperature is determined by a balance between the source and dissipation of energy as before. The complete equilibrium pair distribution function is used to determine the rate of dissipation of energy due to inelastic collisions. The results are compared with hard sphere MD simulations, and also with earlier theoretical and simulation studies. \section{Analysis} The system consists of a bed of circular disks (of diameter $\sigma$) in a gravitational field driven by a vibrating surface. The vibrating surface has a periodic amplitude function but no assumption is made regarding the form of the function. There is a source of energy at the vibrating surface due to particle collisions with the surface, and the dissipation is due to inelastic collisions. A balance between the two determines the ``temperature'', which is the mean square velocity of the particles. The limit of low dissipation, where the coefficient of restitution $e$ is close to $1$, is considered. In this limit, it can be shown that the mean square velocity of the particles is large compared to the mean square velocity of the vibrating surface. An asymptotic expansion in the parameter $\epsilon \equiv U_0^2/T_0$ is used \cite{kum98:vib}. If the source and dissipation of energy are neglected in the leading approximation, the system resembles a gas of hard disks at equilibrium in a gravitational field. The velocity distribution function is a Maxwell-Boltzmann distribution at equilibrium \begin{equation} F({\bf u}) = \frac{1}{2 \pi T_{0}} \exp{ \left( - \frac{u^{2}}{2 T_{0}} \right)}, \end{equation} where $T_{0}$ is the leading order temperature. The density profile is determined by solving the momentum balance equation in the vertical direction \begin{equation} \label{eq:mombal} \pder{p}{z} - \rho g = 0, \end{equation} where $p$ is the pressure, $\rho$ is the density (number of particles per area) and $g$ is the acceleration due to gravity. For a gas at equilibrium, the pressure is related to the density by the virial equation of state, which in the case of inelastic circular disks is \begin{equation} p = \rho T_{0} \left[\frac{1 + e}{2} + (1 + e) g_0(\nu) \, \nu \right], \end{equation} where $g_{0}(\nu)$ is the pair distribution function at contact, which for circular disks is given by \cite{verlet82} \begin{equation} g_{0}(\nu) = \frac{1}{16 (1 - \nu)^{2}} \left[ 16 - 7 \nu - \frac{\nu^{3}}{4 (1 - \nu)^{2}} \right], \end{equation} and $\nu$ is the area fraction corresponding to $\rho$. If the coefficient of restitution is set equal to $1$ in the leading approximation, the equation for the pressure reduces to the standard virial equation of state \begin{equation} p = \rho T_{0} \left[ 1 + 2 g_{0}(\nu) \, \nu\right]. \end{equation} The resulting equation from \Eq{eq:mombal} for the density profile is a first order ordinary differential equation, which can be solved using the mass conservation condition \begin{equation} \label{eq:masscons} \Int{0}{\infty}{z} \rho = N, \end{equation} where $N$ is the number of particles per unit width of the bed. Note that the leading order temperature $T_{0}$ is still unknown at this stage. This is determined using a balance between the source and dissipation of energy. The source of energy due to particle collisions with the vibrating surface is determined using an equilibrium average over the increase in energy due to particle collisions with the vibrating surface \cite{kum98:vib,kum98:vibscal} \begin{equation} \label{eq:s0} S_{0} = 2 \sqrt{\frac{2}{\pi}} \, T_{0}^{1/2} \braket{U^2} \, g_0(\nu) \, \rho \, \Big|_{z=0}. \end{equation} Here $\braket{U^2}$ represents the mean square velocity of the vibrating surface. The rate of dissipation of energy per unit width is calculated by averaging over the energy loss over all the collisions between particles and integrating over the height of the bed \cite{kum98:vib} \begin{equation} \label{eq:d0e} D_0 = \sqrt{\pi} \, \sigma (1-e^2)\, T_0^{3/2} \Int{0}{\infty}{z} g_0(\nu) \, \rho^2. \end{equation} Note that the $g_0$ appearing in $S_0$ and $D_0$ is the Enskog factor which accounts for the increase in the frequency of collision for hard disks at high densities. The temperature $T_{0}$ can now be determined from the relation \begin{equation} \label{eq:s0d0} S_0 = D_0 \end{equation} An analytical solution to the density variation \Eq{eq:mombal} can be determined in the low density limit using the equation of state for an ideal gas for the pressure \cite{kum98:vib}. \begin{equation} \label{eq:nulow} \rho = \frac{N g}{T_{0}} \exp{ \left( - \frac{g z}{T_{0}} \right)} \end{equation} where the leading order temperature is given by, \begin{equation} T_{0} = \frac{4 \sqrt{2}}{\pi} \frac{\braket{U^2}}{N \sigma (1 - e^{2})}. \end{equation} In the low-density limit the density decays exponentially from the bottom of the bed. At higher densities the solution to the density variation is no longer exponential throughout, and has to be obtained numerically by an iterative scheme. However, at large distances from the bottom, the bed is dilute and the ideal gas law holds good, hence the decay is exponential, even though near the bottom it is not. This gives a convenient starting point for the numerical integration from a \emph{finite} height, above which we assume the asymptotic solution ($z \rightarrow \infty$) to be given by an exponential decay known to within two undetermined constants. A value for the density and the temperature is assumed at this height and the integration is carried out up to the vibrating plate ($z=0$). The complete density profile is obtained by combining the numerical and the asymptotic solutions. If the conditions \Eq{eq:masscons} and \Eq{eq:s0d0} are not satisfied after one such integration, a new value is determined for the density and temperature using the Newton-Raphson method, and the iteration is repeated till convergence. In cases where the convergence is poor, the solution is obtained by \emph{continuing} a low density solution in a parameter such as $N \sigma$ or $U_0$. \textbf{Viscous dissipation:} The above analysis can be easily extended to the case of dissipation purely due to viscous drag. The expression for the source of energy remains the same as given by \Eq{eq:s0}. A drag law given by $a_i = -\mu u_i$ is assumed. The total leading order rate of dissipation per unit width will then be \begin{eqnarray} \label{eq:d0v} D_{D0} & = & \mu \Int{0}{\infty}{z} \, \rho \Int{}{}{{\bf u}} \, F({\bf u}) \,\, {\bf u}\cdot{\bf u} \nonumber\\ & = & 2 \mu N T_0 \end{eqnarray} Unlike \Eq{eq:d0e}, the leading order dissipation is the same for the low density and the high density cases. Nevertheless, the density profile has to be obtained numerically in the manner outlined above, with \Eq{eq:d0v} substituted for \Eq{eq:d0e} in \Eq{eq:s0d0}. \section{Simulation and Results} The hard sphere molecular dynamics (MD), also known as event driven (ED) method \cite{ludetal94} is used for the simulations of the vibro-fluidised bed. Periodic boundary conditions are used in the horizontal direction and the vibrating surface at the bottom has a sawtooth form for the amplitude function. The simulations are carried out only for the case of inelastic collisions, since the viscous drag requires a different treatment than the ED method. The density profiles obtained using the present analysis, as well as the earlier low density approximations of Kumaran \cite{kum98:vib}, are compared with the simulation results in \Figs{fig:ldlenu} and~\ref{fig:hdlenu}. It is seen that the density profiles of the present analysis are in good agreement with the simulation results even when the density near the bottom of the bed becomes large, while the profiles from the low density approximation have significant errors. \Fig{fig:hvis} shows the nature of the density profile in the high density limit in the case of dissipation due to viscous drag. Here too the present analysis gives reasonable values for packing fraction near the bottom, while the low density theory predicts physically incorrect values. \begin{figure}[tb] \begin{center} \xlabel{$z/\sigma$} \ylabel{$\nu$} \gpfig{ldlenu.eps} \caption{Exponential decay of packing fraction ($\nu$) with a normalised height ($z/\sigma$) at low densities. The predictions of the present analysis (solid line) and the low density theory (dotted line) of \protect\cite{kum98:vib} is compared with simulation (points). Both the predictions are nearly identical. Here, $\epsilon=0.3$, $N \sigma=3$, $g=1$, and $U_0=6$.} \label{fig:ldlenu} \end{center} \end{figure} \begin{figure}[tb] \begin{center} \xlabel{$z/\sigma$} \ylabel{$\nu$} \gpfig{hdlenu.eps} \caption{Deviation of the density profile from the exponential decay at high densities in the case of dissipation due to inelastic collisions. The simulation result (points) is captured by the present analysis (solid line) which is lower than the exponential decay (dotted line) of the low density theory of \protect\cite{kum98:vib} near the bottom of the bed. Here $\epsilon=0.3$, $N \sigma=3$, $g=1$, and $U_0=1$.} \label{fig:hdlenu} \end{center} \end{figure} \begin{figure}[tb] \begin{center} \xlabel{$z/\sigma$} \ylabel{$\nu$} \gpfig{hvis.eps} \caption{Deviation of the density profile from the exponential decay at high densities in the case of dissipation due to viscous drag. The present analysis (solid line) gives physically plausible values for the packing fraction near the bottom, while the low density theory (dotted line) of \protect\cite{kum98:vib} predicts values higher than the maximum closed packing. Here $\epsilon=0.2$, $N \sigma=20$, $g=20$, $\mu=0.1$, and $U_0=5$.} \label{fig:hvis} \end{center} \end{figure} In a recent work, McNamara and Luding \cite{mclud98} reported the scaling of dissipation with the center of mass obtained from simulations. The results agreed with the low density theory of \cite{kum98:vibscal} but a systematic deviation was observed at high densities in all the cases. This deviation is captured in the present analysis. The leading order dissipation at low densities in the bed is given by \cite{kum98:vib} \begin{equation} \label{eq:d0} D_0 = \frac{\sqrt{\pi}}{2}\, (1-e^2) \, N^2 \sigma g \sqrt{T_0}. \end{equation} In \cite{mclud98} the total dissipation obtained from the simulation was normalised by a factor taken out from this leading order dissipation and a non dimensional number was defined as \begin{equation} \label{e:cpp} C_{pp} \equiv \frac{D_0}{(1-e) N^2 \sigma g \sqrt{T_0/2}}. \end{equation} The scaling of this factor with the height of the center of mass ($h$) above the position at rest ($h_0$) was studied. This factor was found out for different parameter sets by varying the bottom wall velocity $U_0$ over several decades such that the bed is taken from a densely packed regime to a very low density regime. They chose a central data set and varied the parameters one at a time. It was found that in all the cases considered, the scaling relation collapsed to a single curve. The central parameter set has the following values $N = 3.2$, $\sigma=1$, $g=1$, $e=0.95$. The present analysis is valid when $\epsilon \equiv U_0^2/T_0 \ll 1$ and when the frequency of particle-particle collision is much greater than the frequency of particle-wall collisions. It can be shown that in the leading order the ratio of the frequency of particle-particle collisions to the frequency particle-wall collisions is $ \sqrt{2}\pi \,N \sigma$. Hence the present analysis will hold good when $N \sigma \gg 1/\sqrt{2}\pi$. The central set corresponds to $\epsilon = 0.35$, $N \sigma = 3.2$ and therefore we expect the present analysis to hold good for this case. Most of the parameter sets used in \cite{mclud98} also fall within the limits of the theory derived here. \begin{figure}[tb] \begin{center} \xlabel{$2(h-h_0)/\sigma$} \ylabel{$C_{pp}$} \gpfig{theo.eps} \caption{Theoretical scaling of the normalised dissipation ($C_{pp}$) against the center of mass ($h$) above the position at rest ($h_0$) for the different cases reported in \protect\cite{mclud98}. All except two---(N+) with $\epsilon = 1.73$ and (N--) with $N \sigma = 0.65$ collapse on to a single curve in the linear region. The parameters indicated correspond to $N=16 $ (N+), $N=0.65$ (N--), $g=25$ (g+), $g=0.04$ (g--), $e=0.99$ (e+), $e=0.75$ (e--), rest of the parameters being same as the one in the central set, which has the following values $N = 3.2$, $\sigma=1$, $g=1$, $e=0.95$.} \label{fig:theoscal} \end{center} \end{figure} \begin{figure}[tb] \begin{center} \xlabel{$2(h-h_0)/\sigma$} \ylabel{$C_{pp}$} \gpfig{simtheo.eps} \caption{Scaling of the normalised dissipation with the center of mass: Predictions from the present analysis is compared with the results from our simulations and the reported results in \protect\cite{mclud98}. The linear portion of all the curves from theory, except two, fall on the solid line denoted as `Theory'. The two exceptions are also shown. A set of points correspond to the simulation data with parameter values $N = 16$ (N+), $N=0.65$ (N--), $g=25$ (g+), $g=0.04$ (g--), $e=0.99$ (e+), $e=0.75$ (e--); rest of the parameters in a set being the same as the one in the central set, which has the following values $N = 3.2$, $\sigma=1$, $g=1$, $e=0.95$.} \label{fig:simtheo} \end{center} \end{figure} \Fig{fig:theoscal} shows the theoretical predictions of the total dissipation for the different cases reported in Fig.~2 in \cite{mclud98}. It is compared with the results of two simulations in \Fig{fig:simtheo}. It is seen that the present analysis correctly predicts the lowering of the coefficient $C_{pp}$ at high densities. This reduction in the dissipation from the constant value at low densities is the net result of two opposing factors: (i) decrease in the density from the exponential behaviour near the vibrating bottom (see \Fig{fig:hdlenu}), hence reducing the total value of the dissipation, and (ii) increase in frequency of collisions at high densities, increasing the dissipation. It is also seen that not all the theoretical predictions collapse on to a curve as is the case with the data from the simulation. In two of the cases the theory does not agree with the simulations because (i) in one the value of the perturbation parameter is high ($\epsilon = 1.73$) and the leading order theory is valid only for low $\epsilon$, and (ii) in the other case the value of $N \sigma = 0.65 $ is low. In \Fig{fig:theoscal}, the apparent mismatch with `e-' is not a discrepancy with the model, but has got to do with the formula chosen used in \cite{mclud98} for the normalisation of the dissipation factor $C_{pp}$. They had chosen to normalise the dissipation by a factor $(1-e)$. While this might have given a better fit for high densities (low center of mass), the correct factor for very low densities is $(1-e^2)$ as given by \Eq{eq:d0}. The difference is more pronounced in the case of $e\ll 1$, which, here, has a value $e=0.75$. A close inspection of the curves `e-' in \Fig{fig:theoscal} and \Fig{fig:simtheo} show that the theory and simulation do indeed agree with each other. We also note here that the data taken from the reported simulation \cite{mclud98} is for asymmetric sawtooth vibration, whereas our simulation is for the symmetric sawtooth. Both these give similar results for the scaling of $C_{pp}$. Also the theoretical predictions for the symmetric and the asymmetric sawtooth are identical, indicating that the form of the bottom wall vibration does not affect the scaling of the dissipation with the center of mass. \section{Conclusion} In summary, a theory to describe the state of a vibro-fluidised bed in the dense limit was derived. This is different from the earlier theory of Kumaran \cite{kum98:vib,kum98:vibscal}, which is valid for low densities where the ideal gas equation was used and the pair distribution function was set equal to $1$. We have made use of the virial equation of state to obtain a correction to the exponential density profile obtained in low densities and the pair distribution function is used to calculate the increased frequency of collisions in the source and the dissipation of energy. The theoretical predictions of density and temperature were compared with the results obtained from MD simulation of two dimensional disks. The theory correctly predicts the lowering of the density from the exponential value at high densities near the bottom. The theory also predicts the scaling relations of the total dissipation in the bed reported in \cite{mclud98}.
\section{Introduction} How many perfect matchings does a given graph $G$ have? That is, in how many ways can one choose a subset of the edges of $G$ so that each vertex of $G$ belongs to one and only one chosen edge? (See Figure 1(a) for an example of a perfect matching of a graph.) For general graphs $G$, it is computationally hard to obtain the answer \cite{PrV}, and even when we have the answer, it is not so clear that we are any the wiser for knowing this number. However, for many infinite families of special graphs the number of perfect matchings is given by compellingly simple formulas. Over the past ten years a great many families of this kind have been discovered, and while there is no single unified result that encompasses all of them, many of these families resemble one another, both in terms of the form of the results and in terms of the methods that have been useful in proving them. \begin{figure} \centerline{\psfig{file=\figdir/azgraph,scale=110}} \caption{The Aztec diamond of order 4.} \label{azgraph} \end{figure} The deeper significance of these formulas is not clear. Some of them are related to results in representation theory or the theory of symmetric functions, but others seem to be self-contained combinatorial puzzles. Much of the motivation for this branch of research lies in the fact that we are still unable to predict ahead of time which enumerative problems lead to beautiful formulas and which do not; each new positive result seems like an undeserved windfall. Hereafter, I will use the term ``matching'' to signify ``perfect matching''. (See the book of Lov\'asz and Plummer \citeyear{PrLP} for general background on the theory of matchings.) As far as I have been able to determine, problems involving enumeration of matchings were first examined by chemists and physicists in the 1930s, for two different (and unrelated) purposes: the study of aromatic hydrocarbons and the attempt to create a theory of the liquid state. Shortly after the advent of quantum chemistry, chemists turned their attentions to molecules like benzene composed of carbon rings with attached hydrogen atoms. For these researchers, matchings of a graph corresponded to ``Kekul\'e structures'', i.e., ways of assigning single and double bonds in the associated hydrocarbon (with carbon atoms at the vertices and tacit hydrogen atoms attached to carbon atoms with only two neighboring carbon atoms). See for example the article of Gordon and Davison \citeyear{PrGD}, whose use of nonintersecting lattice paths anticipates certain later work \cite{PrGV,PrSac1,PrJS}. There are strong connections between combinatorics and chemistry for such molecules; for instance, those edges which are present in comparatively few of the matchings of a graph turn out to correspond to the bonds that are least stable, and the more matchings a polyhex graph possesses the more stable is the corresponding benzenoid molecule. Since hexagonal rings are so predominant in the structure of hydrocarbons, chemists gave most of their attention to counting matchings of subgraphs of the infinite honeycomb grid. At approximately the same time, scientists were trying to understand the behavior of liquids. As an extension of a more basic model for liquids containing only molecules of one type, Fowler and Rushbrooke \citeyear{PrFR} devised a lattice-based model for liquids containing two types of molecules, one large and one small. In the case where the large molecule was roughly twice the size of the small molecule, it made sense to model the small molecules as occupying sites of a three-dimensional grid and the large molecules as occupying pairs of adjacent sites. In modern parlance, this is a monomer-dimer model. In later years, the two-dimensional version of the model was found to have applicability to the study of molecules adsorbed on films; if the adsorption sites are assumed to form a lattice, and an adsorbed molecule is assumed to occupy two such sites, then one can imagine fictitious molecules that occupy all the unoccupied sites (one each). Major progress was made when Temperley and Fisher \citeyear{PrTF} and Kasteleyn \citeyear{PrKa1} independently found ways to count pure dimer configurations on subgraphs of the infinite square grid, with no monomers present. Although the physical significance of this special case was (and remains) unclear, this result, along with Onsager's earlier exact solution of the two-dimensional Ising model \cite{PrO}, paved the way for other advances such as Lieb's exact solution of the six-vertex model \cite{PrL}, culminating in a new field at the intersection of physics and mathematics: exactly solved statistical mechanics models in two-dimensional lattices. (Intriguingly, virtually none of the three- and higher-dimensional analogues of these models have succumbed to researchers' efforts at obtaining exact solutions.) For background on lattice models in statistical mechanics, see the book by Baxter \citeyear{PrB}. An infinite two-dimensional grid has many finite subgraphs; in choosing which ones to study, physicists were guided by the idea that the shape of boundary should be chosen so as to minimize the effect of the boundary\emdash that is, to maximize the number of configurations, at least in the asymptotic sense. For example, Kasteleyn, in his study of the dimer model on the square grid, counted the matchings of the $m$-by-$n$ rectangle (see the double-product formula at the beginning of Section~5) and of the $m$-by-$n$ rectangular torus, and showed that the two numbers grow at the same rate as $m,n$ go to infinity, namely $C^{mn}$ for a known constant $C$. (Analytically, $C$ is $e^{G/\pi}$, where $G$ is Catalan's constant $1 - \frac{1}{9} + \frac{1}{25} - \frac{1}{49} + \frac{1}{81} - \cdots{}$; numerically, $C$ is approximately $1.34$.) Kasteleyn \citeyear{PrKa1} wrote: ``The effect of boundary conditions is, however, not entirely trivial and will be discussed in more detail in a subsequent paper.'' (See the article of Cohn, Kenyon and Propp \cite{PrCKP} for a rigorous mathematical treatment of boundary conditions.) Kasteleyn never wrote such a followup paper, but other physicists did give some attention to the issue of boundary shape, most notably Grensing, Carlsen and Zapp \cite{PrGCZ}. These authors considered a one-parameter family of graphs of the kind shown in Figure 1(a), and they asserted that every graph in this family has $2^{N/4}$ matchings, where $N$ is the number of vertices. They did not give a proof, nor did they indicate whether they had one. The result was rediscovered in the late 1980s by Elkies, Kuperberg, Larsen, and Propp \cite{PrEKLP}, who gave four proofs of the formula. This article led to a great deal of work among enumerative combinatorialists, who refer to graphs like the one shown in Figure 1 as ``Aztec diamond graphs'', or sometimes just Aztec diamonds for short. (It should be noted that Elkies et al.\ \citeyear{PrEKLP} used the term ``Aztec diamond'' to denote regions like the one shown in Figure 1(b). The two sorts of Aztec diamonds are dual to one another; matchings of Aztec diamond graphs correspond to domino tilings of Aztec diamond regions.) At about the same time, it became clear that there had been earlier work within the combinatorial community that was pertinent to the study of matchings, though its relevance had not hitherto been recognized. For instance, Mills, Robbins and Rumsey \cite{PrMRR}, in their work on alternating sign matrices, had counted pairs of ``compatible'' ASMs of consecutive size; these can be put into one-to-one correspondence with matchings of an associated Aztec diamond graph \cite{PrEKLP}. Looking into earlier mathematical literature, one can even see intimations of enumerative matching theory in the work of MacMahon \citeyear{PrM}, who nearly a century ago found a formula for the number of plane partitions whose solid Young diagram fits inside an $a$-by-$b$-by-$c$ box, as will be discussed in Section~2. (See the book by Andrews \citeyear{PrA} and the article by Stanley \citeyear{PrSt1} for background on plane partitions.) Such a Young diagram is nothing more than an assemblage of cubes, and it has long been known in the extra-mathematical world that such assemblages, viewed from a distant point, looks like tilings (consider Islamic art, for instance). Thus it was natural for mathematicians to interpret MacMahon's theorem on plane partitions as a result about tilings of a hexagon by rhombuses. This insight may have occurred to a number of people independently; the earliest chain of oral communication that I have followed leads back to Klarner (who did not publish his observation but relayed it to Stanley in the 1970s), and the earliest published statement I have found is in a paper by David and Tomei \citeyear{PrDT}.{\looseness=-1\par} In any case, each of the Young diagrams enumerated by MacMahon corresponds to a tiling of a hexagon by rhombuses, where the hexagon is semiregular (its opposite sides are parallel and of equal length, with all internal angles equal to 120 degrees) and has side-lengths $a,b,c,a,b,c$, and where the rhombuses have all side-lengths equal to 1. These tilings in turn correspond to matchings of the ``honeycomb'' graph that is dual to the dissection of the hexagon into unit equilateral triangles; see Figure 2, which shows a matching of the honeycomb graph and the associated tiling of a hexagon. Kuperberg \citeyear{PrKu1} was the first to exploit the connection between plane partitions and the dimer model. (Interestingly, some of the same graphs that Kuperberg studied had been investigated independently by chemists in their study of benzenoids hydrocarbons; Cyvin and Gutman \citeyear{PrCG} give a survey of this work.) \begin{figure} \centerline{\psfig{file=\figdir/hexgraph,scale=110}} \caption{A matching and its associated tiling.} \label{hexgraph} \end{figure} Similarly, variants of MacMahon's problem in which the plane partition is subjected to various symmetry constraints (considered by Macdonald, Stanley, and others [\citeNP{PrSt3}; \citeyearNP{PrSt4}]) correspond to the problem of enumerating matchings possessing corresponding kinds of symmetry. Kuperberg \citeyear{PrKu1} used this correspondence in solving one of Stanley's open problems, and this created further interest in matchings among combinatorialists. One of Kuperberg's chief tools was an old result of Kasteleyn, which showed that for any planar graph $G$, the number of matchings of $G$ is equal to the Pfaffian of a certain matrix of zeros and ones associated with $G$. A special case of this result, enunciated by Percus \citeyear{PrPe}, can be used when $G$ is bipartite; in this case, one can use a determinant instead of a Pfaffian. Percus' determinant is a modified version of the bipartite adjacency matrix of the graph, in which rows correspond to ``white'' vertices and columns correspond to ``black'' vertices (under a coloring scheme whereby white vertices have only black neighbors and vice versa); the $(i,j)$-th entry is $\pm 1$ if the $i$-th white vertex and $j$-th black vertex are adjacent, and 0 otherwise. For more details on how the signs of the entries are chosen, see the expositions of Kasteleyn \citeyear{PrKa3} and Percus \citeyear{PrPe}. Percus' theorem, incorporated into computer software, makes it easy to count the matchings of many planar graphs and look for patterns in the numbers that arise. Two such programs are {\tt vaxmaple}, written by Greg Kuperberg, David Wilson and myself, and {\tt vaxmacs}, written by David Wilson. Most of the patterns described below were discovered with the aid of this software, which is available from \url{http://math.wisc.edu/~propp/software.html}. Both programs treat subgraphs of the infinite square grid; this might seem restrictive, but it turns out that counting the matchings of an arbitrary bipartite planar graph can be fit into this framework, with a bit of tweaking. The mathematically interesting part of each program is the routine for choosing the signs of the nonzero entries. There are many choices that would work, but Wilson's sign-rule is far and away the simplest: If an edge is horizontal, we give it weight $+1$, and if an edge is vertical, joining a vertex in one row to a vertex in the row below it, we give the edge weight $(-1)^k$, where $k$ is the number of vertices in the upper row to the left of the vertical edge. The main difference between {\tt vaxmaple} and {\tt vaxmacs} is that the former creates \Maple/ code which, if sent to \Maple/, results in \Maple/ printing out the number of matchings of the graph; {\tt vaxmacs}, on the other hand, is a customized {\sc Emacs} environment that fully integrates text-editing operations (used for defining the graph one wishes to study) with the mathematical operations of interest. Both programs represent bipartite planar graphs in ``VAX-format'', where V's, A's, X's, and other letters denote vertices. (An example of VAX-format can be found on page~\pageref{vaxexample}; for a detailed explanation see \url{http://math.wisc.edu/~propp/vaxmaple.doc}.) Quite recently, the study of matchings of nonbipartite graphs has been expedited by the programs {\tt graph} and {\tt planemaple}, created by Matt Blum and Ben Wieland, respectively. These programs make it easy to define a planar graph by pointing and clicking, after which one can count its matchings using an efficient implementation of Kasteleyn's Pfaffian method. This makes it easy to try out new ideas and look for patterns, outside of the better-explored bipartite case.{\looseness=1\par} Interested readers with access to the World Wide Web can obtain copies of all of these programs via \url{http://math.wisc.edu/~propp/software.html}. Most of the formulas that have been discovered express the number of matchings of a graph as a product of many comparatively small factors. Even before one has conjectured (let alone proved) such a formula, one can frequently infer its existence from the fact that the number of matchings has only small prime factors. Numbers that are large compared to their largest prime factor are sometimes called ``smooth'' or ``round''; the latter term will be used here. The definition of roundness is not precise, since it is not intended for use as a technical term. Its vagueness is intended to capture the uncertainties and the suspense of formula-hunting, and the debatable issue of whether the occurrence of a single larger-than-expected prime factor rules out the existence of a product formula. (For an example of a number whose roundness lies in this gray area, see the table of numbers given in Problem 8.) It is worth noting that Kuperberg \citeyear[Section VII-A]{PrKu2} has shown that rigorous proofs of roundness need not always yield explicit product formulas. Christian Krattenthaler has written a Mathematica program called RATE that greatly expedites the process of guessing patterns in experimental data on enumeration of matchings; see \url{http://radon.mat.univie.ac.at/People/kratt/rate/rate.html}. A great source of the appeal of research on enumeration of matchings is the ease with which undergraduate research assistants can participate in the hunt for formulas and proofs; many members of the M.I.T.\ Tilings Research Group (composed mostly of undergraduates like Blum and Wieland) played a role in the developments that led to the writing of this article. Enumeration of matchings has turned out to be a rich avenue of combinatorial inquiry, and many more beautiful patterns undoubtedly await discovery. Updates on the status of these problems can be found on the Web at \url{http://math.wisc.edu/~propp/update.ps.gz}. \section{Lozenges} We begin with problems related to lozenge tilings of hexagons. A \textit{lozenge} is a rhombus of side-length 1 whose internal angles measure 60 and 120 degrees; all the hexagons we will consider will tacitly have integer side-lengths and internal angles of 120 degrees. Every such hexagon $H$ can be dissected into unit equilateral triangles in a unique way, and one can use this dissection to define a graph $G$ whose vertices correspond to the triangles and whose edges correspond to pairs of triangles that share an edge; this is the ``finite honeycomb graph'' dual to the dissection. It is easy to see that the tilings of $H$ by lozenges are in one-to-one correspondence with the matchings of $G$. The $a,b,c$ semiregular hexagon is the hexagon whose side lengths are, in cyclical order, $a,b,c,a,b,c$. Lozenge tilings of this region are in correspondence with plane partitions with at most $a$ rows, at most $b$ columns, and no part exceeding $c$. Such hexagons are represented in VAX-format by diagrams like $$ \label{vaxexample} \vcenter\brda{% AVAVAVAVA AVAVAVAVAVA AVAVAVAVAVAVA AVAVAVAVAVAVAVA VAVAVAVAVAVAVAV VAVAVAVAVAVAV VAVAVAVAVAV VAVAVAVAV } $$ where A's and V's represent upward-pointing and downward-pointing triangles, respectively. In this article we will use triangles instead: $$\vcenter\brd{ AVAVAVAVA AVAVAVAVAVA AVAVAVAVAVAVA AVAVAVAVAVAVAVA VAVAVAVAVAVAVAV VAVAVAVAVAVAV VAVAVAVAVAV VAVAVAVAV } $$ MacMahon \citeyear{PrM} showed that the number of such plane partitions is $$\prod_{i=0}^{a-1} \prod_{j=0}^{b-1} \prod_{k=0}^{c-1} \frac{i+j+k+2}{i+j+k+1}\,.$$ (This form of MacMahon's formula is due to Macdonald; a short, self-contained proof is given by Cohn et al. \citeyear[Section~2]{PrCLP}.) \begin{problem} \label{prob1} Show that in the $2n-1$, $2n$, $2n-1$ semiregular hexagon, the central location (consisting of the two innermost triangles) is covered by a lozenge in exactly one-third of the tilings. (Equivalently: Show that if one chooses a random matching of the dual graph, the probability that the central edge is contained in the matching is exactly $\frac13$.) \end{problem} \begin{progress} Two independent and very different solutions of this problem have been found; one by Mihai Ciucu and Christian Krattenthaler and the other by Harald Helfgott and Ira Gessel. Ciucu and Krattenthaler \citeyear{PrCK} compute more generally the number of rhombus tilings of a hexagon with sides $a,a,b,a,a,b$ that contain the central unit rhombus, where $a$ and $b$ must have opposite parity (the special case $a=2n-1$, $b=2n$ solves Problem 1). The same generalization was obtained (in a different but equivalent form) by Helfgott and Gessel \citeyear{PrHG}, using a completely different method. One might still try to look for a proof whose simplicity is comparable to that of the answer ``one-third''. Also worthy of note is the paper of Fulmek and Krattenthaler \citeyear{PrFK1}, which generalizes the result of Ciucu and Krattenthaler \citeyear{PrCK}. \end{progress} \hskip\parindent The hexagon of side-lengths $n$, $n+1$, $n$, $n+1$, $n$, $n+1$ cannot be tiled by lozenges at all, for in the dissection into unit triangles, the number of upward-pointing triangles differs from the number of downward-pointing triangles. However, if one removes the central triangle, one gets a region that can be tiled, and the sort of numbers one gets for small values of $n$ are striking. Here they are, in factored form: $$ \medmuskip 1mu \displaylines{ 2\cr 2\cdot3^3\cr 2^5\cdot3^3\cdot5\cr 2^5\cdot5^7\cr 2^2\cdot5^7\cdot7^5\cr 2^8\cdot3^3\cdot5\cdot7^{11}\cr 2^{13}\cdot3^9\cdot7^{11}\cdot11\cr 2^{13}\cdot3^{18}\cdot7^5\cdot11^7\cr 2^8\cdot3^{18}\cdot11^{13}\cdot13^5\cr 2^2\cdot3^9\cdot11^{19}\cdot13^{11}\cr 2^{10}\cdot3^3\cdot11^{19}\cdot13^{17}\cdot17\cr 2^{16}\cdot11^{13}\cdot13^{23}\cdot17^7\cr } $$ These are similar to the numbers one gets from counting lozenge tilings of an $n,n,n,n,n,n$ hexagon, in that the largest prime factor seems to be bounded by a linear function of $n$. \begin{problem} \label{prob2} Enumerate the lozenge tilings of the region obtained from the $n$, $n+1 $, $n $, $n+1 $, $n $, $n+1$ hexagon by removing the central triangle. \end{problem} \begin{progress} Mihai Ciucu has solved the more general problem of counting the rhombus tilings of an $(a$, $b+1 $, $b $, $a+1 $, $b $, $b+1)$-hexagon with the central triangle removed \cite{PrCi2}. Ira Gessel proved this result independently using the nonintersecting lattice-paths method \cite{PrHG}. Soichi Okada and Christian Krattenthaler have solved the even more general problem of counting the rhombus tilings of an $(a $, $b+1 $, $c $, $a+1 $, $b $, $c+1)$-hexagon with the central triangle removed \cite{PrOK}. \end{progress} \hskip\parindent One can also take a $2n $, $2n+3 $, $2n $, $2n+3 $, $2n $, $2n+3$ hexagon and make it lozenge-tilable by removing a triangle from the middle of each of its three long sides, as shown: $$\vcenter\brd{ AVAVAVAVAVAVAVAVA AVAVAVAVAVAVAVAVAVA AVAVAVAVAVAVAVAVAVAVA AVAVAVAVAVAVAVAVAVAVAVA AVAVAVAVAVAVAVAVAVAVAVAVA VAVAVAVAVAVAVAVAVAVAVAVAV AVAVAVAVAVAVAVAVAVAVAVAVAVAVA AVAVAVAVAVAVAVAVAVAVAVAVAVAVAVA AVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVA AVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVA AVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVA VAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAV VAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAV VAVAVAVAVAVAVAVAVAVAVAVAVAVAVAVAV VAVAVAVAVAVAVAVAVAVAVAVAVAVAVAV VAVAVAVAVAVAVAVAVAVAVAVAVAVAV VAVAVAVAVAVAVAVAVAVAVAVAVAV VAVAVAVAVAVAVAVAVAVAVAVAV VAVAVAVAVAV VAVAVAVAVAV }$$ Here one obtains an equally tantalizing sequence of factorizations: $$ \medmuskip 1mu \displaylines{ 1\cr 2^7\cdot 7^2\cr 2^2\cdot 7^4\cdot 11^4\cdot 13^2\cr 2^{10}\cdot 3^3\cdot 5^8\cdot 13^2\cdot 17^4 \cdot 19^2\cr 2^2\cdot 5^2\cdot 7^2\cdot 11^3\cdot 13^4\cdot 17^4\cdot 19^8\cdot 23^4 } $$ \begin{problem} \label{prob3} Enumerate the lozenge tilings of the region obtained from the $2n$, $2n+3 $, $2n $, $2n+3 $, $2n $, $2n+3$ hexagon by removing a triangle from the middle of each of its long sides. \end{problem} \begin{progress} Theresia Eisenk\"olbl solved this problem. What she does in fact is to compute the number of all rhombus tilings of a hexagon with sides $a$, $b+3$, $c$, $a+3$, $b$, $c+3$, where an arbitrary triangle is removed from each of the ``long'' sides of the hexagon (not necessarily the triangle in the middle). For the proof of her formula \cite{PrE1} she uses nonintersecting lattice paths, determinants, and the Jacobi determinant formula \cite{PrT}. However, I still know of no conceptual explanation for why these numbers are so close (in the multiplicative sense) to being perfect squares. \end{progress} \hskip\parindent We now return to ordinary $a,b,c$ semiregular hexagons. When $a=b=c$, there are not two but six central triangles. There are two geometrically distinct ways in which we can choose to remove an upward-pointing triangle and downward-pointing triangle from these six, according to whether the triangles are opposite or adjacent: $$\vcenter\brd{ AVAVAVA AVAVAVA AVAVAVAVA AVAVAVAVA AVAVA AVAVA AVAV VAVAVA VAVAV VAVAV VAVA AVAVAV VAVAVAVAV VAVAVAVAV VAVAVAV VAVAVAV } \vadjust{\vskip3pt} $$ Such regions may be called ``holey hexagons'' of two different kinds. Matt Blum tabulated the number of lozenge tilings of these regions, for small values of $a=b=c$. In the first (``opposite'') case, the number of tilings of the holey hexagon is a nice round number (its greatest prime factor appears to be bounded by a linear function of the size of the region). In the second (``adjacent'') case, the number of tilings is not round. Note, however, that in the second case, the number of tilings of the holey hexagon divided by the number of tilings of the unaltered hexagon (given to us by MacMahon's formula) is equal to the probability that a random lozenge tiling of the hexagon contains a lozenge that covers these two triangles; this probability tends to $\frac13$ for large $a$, at least on average \cite{PrCLP}. Following this clue, we examine the difference between the aforementioned probability (with its messy, un-round numerator) and the number $\frac13$. The result is a fraction in which the numerator is now a nice round number. So, in both cases, we have reason to think that there is an exact product formula. \begin{problem} \label{prob4} Determine the number of lozenge tilings of a regular hexagon from which two of its innermost unit triangles (one upward-pointing and one downward-pointing) have been removed. \end{problem} \begin{progress} Theresia Eisenk\"olbl solved the first case of Problem 4 and Markus Fulmek and Christian Krattenthaler solved the second case. Eisenk\"olbl \citeyear{PrE2} solves a generalization of the problem by applying Mihai Ciucu's matchings factorization theorem, nonintersecting lattice paths, and a nontrivial determinant evaluation. Fulmek and Krattenthaler \citeyear{PrFK2} compute the number of rhombus tilings of a hexagon with sides $a,b,a,a,b,a$ (with $a$ and $b$ having the same parity) that contain the rhombus that touches the center of the hexagon and lies symmetric with respect to the symmetry axis that runs parallel to the sides of length $b$. For the proof of their formula they compute Hankel determinants featuring Bernoulli numbers, which they do by using facts about continued fractions, orthogonal polynomials, and, in particular, continuous Hahn polynomials. The special case $a=b$ solves the second part of Problem 4. \end{progress} \hskip\parindent I mentioned earlier that Kasteleyn's method, as interpreted by Percus, allows one to write the number of matchings of a bipartite planar graph as the determinant of a signed version of the bipartite adjacency matrix. In the case of lozenge tilings of hexagons and the associated matchings, it turns out that there is no need to modify signs of entries; the ordinary bipartite adjacency matrix will do. Greg Kuperberg \citeyear{PrKu2} has noticed that when row-reduction and column-reduction are systematically applied to the Kasteleyn--Percus matrix of an $a,b,c$ semiregular hexagon, one can obtain the $b$-by-$b$ Carlitz matrix \cite{PrCS} whose $(i,j)$-th entry is $a+c \choose a+i-j$. (This matrix can also be recognized as the Gessel--Viennot matrix that arises from interpreting each tiling as a family of nonintersecting lattice paths \cite{PrGV}.) Such reductions do not affect the determinant, so we have a pleasing way of understanding the relationship between the Kasteleyn--Percus matrix method and the Gessel--Viennot lattice-path method. In fact, such reductions do not affect the \textit{cokernel} of the matrix (an abelian group whose order is the determinant). On the other hand, the cokernel of the Kasteleyn--Percus matrix for the $a,b,c$ hexagon is clearly invariant under permuting $a$, $b$, and $c$. This gives rise to three different Carlitz matrices that nontrivially have the same cokernel. For example, if $c=1$, then one gets an $a$-by-$a$ matrix and a $b$-by-$b$ matrix that both have the same cokernel, whose structure can be determined ``by inspection'' if one notices that the third Carlitz matrix of the trio is just a 1-by-1 matrix whose sole entry is (plus or minus) a binomial coefficient. In this special case, the cokernel is just a cyclic group. Greg Kuperberg poses this challenge: \begin{problem} \label{prob5} Determine the cokernel of the Carlitz matrix, or equivalently of the Kasteleyn--Percus matrix of the $a,b,c$ hexagon, and if possible find a way to interpret the cokernel in terms of the tilings. This combines Questions 1 and 2 of Kuperberg \citeyear{PrKu2}. As he points out in that article, in the case $a=b=c=2$, one gets the noncyclic group $\Z/2\Z \times \Z/10\Z$ as the cokernel. \end{problem} \hskip\parindent As was remarked above, one nice thing about the Kasteleyn--Percus matrices of honeycomb graphs is that it is not necessary to make any of the entries negative. For general graphs, however, there is no canonical way of defining $K$, in the sense that there may be many ways of modifying the signs of certain entries of the bipartite adjacency matrix of a graph so that all nonzero contributions to the determinant have the same sign. Thus, one should not expect the eigenvalues of $K$ to possess combinatorial significance. However, the spectrum of $K$ times its adjoint $K^*$ is independent of which Kasteleyn--Percus matrix $K$ one chooses (as was independently shown by David Wilson and Horst Sachs). Thus, digressing somewhat from the topic of lozenge tilings, we find it natural to ask: \begin{problem} \label{prob6} What is the significance of the spectrum of $K K^*$, where $K$ is any Kasteleyn--Percus matrix associated with a bipartite planar graph? \end{problem} \begin{progress} Nicolau Saldanha \citeyear{PrSal} has proposed a combinatorial interpretation of the spectrum of $K K^*$. Horst Sachs says (personal communication) that $K K^*$ may have some significance in the chemistry of polycyclic hydrocarbons (so-called benzenoids) and related compounds as a useful approximate measure of the ``degree of aromaticity''. \end{progress} \hskip\parindent Returning now to lozenge tilings, or equivalently, matchings of finite subgraphs of the infinite honeycomb, consider the hexagon graph with $a=b=c=2$:% $$ \vcenter{\rot{\psfig{file=\figdir/hex,width=1.4in}}{0}{0}} $$ This is the graph whose 20 matchings correspond to the 20 tilings of the regular hexagon of side 2 by rhombuses of side 1. If we look at the probability of each individual vertical edge belonging to a matching chosen uniformly at random (``edge-probabilities''), we get \vadjust{\vskip4pt} $$ \makeatletter \vcenter{\rot{\psfig{file=\figdir/hexa,width=1.4in}}{0}{0}} $$ Now look at this table of numbers as if it described a distribution of mass. If we assign the three rows $y$-coordinates $-1$ through 1, we find that the weighted sum of the squares of the $y$-coordinates is equal to $$(0.3+0.4+0.3)(-1)^2+(0.7+0.3+0.7+0.3)(0)^2+(0.3+0.4+0.3)(1)^2=2.$$ If we assign to the seven columns $x$-coordinates $-3$ through 3, we find that the weighted sum of the squares of the $x$-coordinates is equal to $(0.7)(-3)^2+(0.6)(-2)^2+(0.3)(-1)^2+(0.8)(0)^2 +(0.3)(1)^2+(0.6)(2)^2+(0.7)(3)^2=20$. \vadjust{\goodbreak}% You can do a similar (but even easier) calculation yourself for the case $a=b=c=1$, to see that the ``moments of inertia'' of the vertical edge-probabilities around the horizontal and vertical axes are 0 and 1, respectively. Using {\tt vaxmaple} to study the case $a=b=c=n$ for larger values of $n$, I find that the moment of inertia about the horizontal axis goes like $$0, 2, 12, 40, 100, \dots$$ and the moment of inertia about the vertical axis goes like $$1, 20, 93, 296, 725, \dots .$$ It is easy to show that the former moments of inertia are given in general by the polynomial $(n^4-n^2)/6$ (in fact, the number of vertical lozenges that have any particular $y$-coordinate does not depend on the tiling chosen). The latter moments of inertia are subtler; they are not given by a polynomial of degree 4, though it is noteworthy that the $n$-th term is an integer divisible by $n$, at least for the first few values of $n$. \begin{problem} \label{prob7} Find the ``moments of inertia'' for the mass on edges arising from edge-probabilities for random matchings of the $a,b,c$ honeycomb graph. \end{problem} \section{Dominoes} Now let us turn from lozenge-tiling problems to domino-tiling problems. A \textit{domino} is a 1-by-2 or 2-by-1 rectangle. Although lozenge tilings (in the guise of constrained plane partitions) were studied first, it was really the study of domino tilings in Aztec diamonds that gave current work on enumeration of matchings its current impetus. Here is the Aztec diamond of order 5: $$\vcenter\brd{ XX XXXX XXXXXX XXXXXXXX XXXXXXXXXX XXXXXXXXXX XXXXXXXX XXXXXX XXXX XX }$$ A tiling of such a region by dominos is equivalent to a matching of a certain (dual) subgraph of the infinite square graph. This grid is bipartite, and it is convenient to color its vertices alternately black and white; equivalently, it is convenient to color the 1-by-1 squares alternately black and white, so that every domino contains one 1-by-1 square of each color. Elkies, Kuperberg, Larsen, and Propp showed in \cite{PrEKLP} that the number of domino tilings of such a region is $2^{n(n+1)/2}$ (where $2n$ is the number of rows), and Gessel, Ionescu, and Propp proved in \cite{PrGIP} an exact formula (originally conjectured by Jockusch) for the number of tilings of regions like $$\vcenter\brd{ XX XXXX XXXXXX XXXXXXXX XXXX XXXXX XXXX XXXXX XXXXXXXX XXXXXX XXXX XX }$$ in which two innermost squares of opposite color have been removed. (For some values of $n$, the number of tilings is exactly $\frac14$ times $2^{n(n+1)/2}$; in the other cases, there is an exact product formula for the difference between the number of tilings and $\bigl(\frac14\bigr)2^{n(n+1)/2}$. It is this latter fact that motivated the idea of trying something similar in the case of lozenge tilings, as described in the paragraph preceding the statement of Problem 4.) Now suppose one removes two squares from the middle of an Aztec diamond of order $n$ in the following way: $$\vcenter\brd{ XX XXXX XXXXXX XXXX XXX XXXXXXXXXX XXXX XXXXX XXXXXXXX XXXXXX XXXX XX }$$ (The two squares removed are a knight's-move apart, and subject to that constraint, they are as close to being in the middle as they can be. Up to symmetries of the square, there is only one way of doing this.) The numbers of tilings one gets are as follows (for $n = 2$ through 10): $$ \medmuskip 1mu \displaylines{ 2\cr 2^3\cr 2^5\cdot 5\cr 2^9\cdot 3^2\cr 2^{17}\cdot 3\cr 2^{22}\cdot 3^2\cr 2^{24}\cdot 3^2\cdot73\cr 2^{31}\cdot 3^2\cdot5^2\cdot11\cr 2^{47}\cdot 3^2\cdot5\cr } $$ Only the presence of the large prime factor 73 makes one doubt that there is a general product formula; the other prime factors are reassuringly small. \begin{problem} \label{prob8} Count the domino tilings of an Aztec diamond from which two close-to-central squares, related by a knight's move, have been deleted. \end{problem} \begin{progress} Harald Helfgott has solved this problem; it follows from the main result in his thesis \citeyear{PrH}. The formula is somewhat complicated, as the prime factor 73 might have led us to expect. (One of the factors in Helfgott's product formula is a single-indexed sum; 73 arises as $128-60+5$.) \end{progress} \hskip\parindent One can also look at ``Aztec rectangles'' from which squares have been removed so as to restore the balance between black and white squares (a necessary condition for tileability). For instance, one can remove the central square from an $a$-by-$b$ Aztec rectangle in which $a$ and $b$ differ by 1, with the larger of $a,b$ odd: $$\vcenter\brd{ XX XXXX XXXXXX XXXXXXXX XXXX XXXX XXXXXXXX XXXXXX XXXX XX }$$ \begin{problem} \label{prob9} Find a formula for the number of domino tilings of a $2n$-by-$(2n+1)$ Aztec rectangle with its central square removed. \end{problem} \begin{progress} This had already been solved when I posed the problem; it is a special case of a result of Ciucu \citeyear[Theorem 4.1]{PrCi1}. Eric Kuo solved the problem independently. \end{progress} \hskip\parindent What about $(2n-1)$-by-$2n$ rectangles? For these regions, removing the central square does not make the region tilable. However, if one removes any one of the four squares adjacent to the middle square, one obtains a region that is tilable, and moreover, for this region the number of tilings appears to be a nice round number. \begin{problem} \label{prob10} Find a formula for the number of domino tilings of a $(2n-1)$-by-$2n$ Aztec rectangle with a square adjoining the central square removed. \end{problem} \begin{progress} This problem was solved independently three times:\,\ by Harald Helfgott and Ira Gessel \citeyear{PrHG}, by Christian Krattenthaler \citeyear{PrKr}, and by Eric Kuo (private communication). Gessel and Helfgott solve a more general problem than Problem 10. Krattenthaler's preprint gives several results concerning the enumeration of matchings of Aztec rectangles where (a suitable number of) collinear vertices are removed, of which Problem 10 is just a special case. There is some overlap between the results of Helfgott and Gessel and the results of Krattenthaler. \end{progress} \hskip\parindent At this point, some readers may be wondering why $m$-by-$n$ rectangles have not played a bigger part in the story. Indeed, one of the surprising facts of life in the study of enumeration of matchings is that Aztec diamonds and their kin have been much more fertile ground for exact combinatorics that the seemingly more natural rectangles. There are, however, a few cases I know of in which something rather nice turns up. One is the problem of Ira Gessel that appears as Problem 20 in this document. Another is the work done by Jockusch \citeyear{PrJ} and, later, Ciucu \citeyear{PrCi1} on why the number of domino tilings of the square is always either a perfect square or twice a perfect square. In the spirit of the work of Jockusch and Ciucu, I offer here a problem based on Lior Pachter's observation \cite{PrPK} that the region on the left below, obtained by removing 8 dominos from a 16-by-16 square, has exactly one tiling. What if we make the intrusion half as long, as in the region on the right? $$\vcenter\brd{ XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXX XXXXXXX XXXXXX XXXXXXXX XXXXX XXXXXXXXX XXXX XXXXXXXXXX XXX XXXXXXXXXXX XX XXXXXXXXXXXX X XXXXXXXXXXXXX XXXXXXXXXXXXXX }\hskip .25\hsize \vcenter\brd{ XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXX XXXXXXXXXXX XX XXXXXXXXXXXX X XXXXXXXXXXXXX XXXXXXXXXXXXXX }$$ That is, we take a $2n$-by-$2n$ square (with $n$ even) and remove $n/2$ dominos from it, in a partial zig-zag pattern that starts from the corner. Here are the numbers we get, in factored form, for $n=2,4,6,8,10$: $$ \medmuskip 1mu \displaylines{ 2\cdot3^2\cr 2^2\cdot3^6\cdot13^2\cr 2^3\cdot3^2\cdot5^4\cdot7^2\cdot3187^2\cr 2^4\cdot11771899^2\cdot27487^2\cr 2^5\cdot2534588575976069659^2 }$$ The factors are ugly, but the exponents are nice: we get $2^{n/2}$ times an odd square. Perhaps this is a special case of a two-parameter fact that says that you can take an intrusion of length $m$ in a $2n$-by-$2n$ square and the number of tilings of the resulting region will always be a square or twice a square. \begin{problem} \label{prob11} What is going on with ``intruded Aztec diamonds''? In particular, why is the number of tilings so square-ish? \end{problem} \hskip\parindent It should also be noted that the square root of the odd parts of these numbers (3, $3^3 \cdot 13$, etc.)\ alternate between 1 and 3 mod 4. Perhaps these quantities are continuous functions of $n$ in the 2-adic sense, as is the case for intact $2n$-by-$2n$ squares \cite{PrCo}; however, the presence of large prime factors means that no simple product formula is available, and that the analysis will require new techniques. \medbreak We now return to the Kasteleyn--Percus matrices discussed earlier. Work of Rick Kenyon and David Wilson \cite{PrKe} has shown that the \textit{inverses} of these matrices are loaded with combinatorial information, so it would be nice to get our hands on them. Unfortunately, there are many nonzero entries in the inverse-matrices. (Recall that the Kasteleyn--Percus matrices themselves, being nothing more than adjacency matrices in which some of the 1's have been strategically replaced by $-1$'s, are sparse; their inverses, however, tend to have most if not all of their entries nonzero.) Nonetheless, some exploratory ``numerology'' leaves room for hope that this is do-able. Consider the Kasteleyn--Percus matrix $K_n$ for the Aztec diamond of order $n$, in which every vertical domino with its white square on top (relative to some fixed checkerboard coloring) has its sign inverted\emdash that is, the corresponding 1 in the bipartite adjacency matrix is replaced by $-1$. \begin{problem} \label{prob12} Show that the sum of the entries of the matrix inverse of $K_n$ is $\frac12(n-1)(n+3) - 2^{n-1} + 2$. \end{problem} \hskip\parindent (This formula works for $n=1$ through $n=8$.) \begin{progress} Harald Helfgott has solved a similar problem using the main result of his thesis \citeyear{PrH}, and it is likely that the result asserted in Problem 12 can be proved similarly. (A slight technical hurdle arises from the fact that Helfgott's thesis uses a different sign-convention for the Kasteleyn--Percus matrix, which results in different signs, and a different sum, for the inverse matrix; however, Helfgott's methods are quite general, so there is no conceptual obstacle to applying them to Problem 12.) I should mention that my original reason for examining the sum of the entries of the inverse Kasteleyn--Percus matrix was to see whether there might be formulas governing the individual entries themselves. Helfgott's work provides such formulas. Also, in this connection, Greg Kuperberg and Douglas Zare have some high-tech ruminations on the inverses of Kasteleyn--Percus matrices, and there is a chance that representation-theory methods will give a different way of proving the result. \end{progress} \hskip\parindent Now we turn to a class of regions I call ``pillows''. Here are a ``0 mod 4'' pillow of ``order 5'' and a ``2 mod 4'' pillow of ``order 7'': $$\vcenter\brd{ XXXX XXXXXXXX XXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXX XXXXXXXX XXXX }\hskip.1\hsize \vcenter\brd{ XX XXXXXX XXXXXXXXXX XXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXX XXXXXXXXXX XXXXXX XX }$$ It turns out (empirically) that the number of tilings of the 0-mod-4 pillow of order $n$ is a perfect square times the coefficient of $x^n$ in the Taylor expansion of $(5+3x+x^2-x^3)/(1-2x-2x^2-2x^3+x^4)$. This fact came to light in several steps. First it was noticed that the number of tilings has a comparatively small square-free part. Then it was noticed that in the derived sequence of square-free parts, many terms were roughly three times the preceding term. Then it was noticed that, by judiciously including some of the square factors, one could obtain a sequence in which each term was roughly three times the preceding term. Finally it was noticed that this approximately geometric sequence satisfied a fourth-order linear recurrence relation. Similarly, it appears that the number of tilings of the 2-mod-4 pillow of order $n$ is a perfect square times the coefficient of $x^n$ in the Taylor expansion of $(5+6x+3x^2-2x^3)/(1-2x-2x^2-2x^3+x^4)$. (If you are wondering about ``odd pillows'', I should mention that there is a nice formula for the number of tilings, but this is not an interesting result, because an odd pillow splits up into many small noncommunicating sub-regions such that a tiling of the whole region corresponds to a choice of tiling on each of the sub-regions.) \begin{problem} \label{prob13} Find a general formula for the number of domino tilings of even pillows. \end{problem} \hskip\parindent Jockusch looked at the Aztec diamond of order $n$ with a 2-by-2 hole in the center, for small values of $n$; he came up with a conjecture for the number of domino tilings, subsequently proved by Gessel, Ionescu, and Propp \cite{PrGIP}. One way to generalize this is to make the hole larger, as was suggested by Douglas Zare and investigated by David Wilson. Here is an abridged and adapted version of the report David Wilson sent me on October 15, 1996: \medskip Define the Aztec window with outer order $y$ and inner order $x$ to be the Aztec diamond of order $y$ with an Aztec diamond of order $x$ deleted from its center. For example, this is the Aztec window with orders 8 and 2: $$\vcenter\brd{ XX XXXX XXXXXX XXXXXXXX XXXXXXXXXX XXXXXXXXXXXX XXXXXX XXXXXX XXXXXX XXXXXX XXXXXX XXXXXX XXXXXX XXXXXX XXXXXXXXXXXX XXXXXXXXXX XXXXXXXX XXXXXX XXXX XX }$$ There are a number of interesting patterns that show up when we count tilings of Aztec windows. For one thing, if $w$ is a fixed even number, and $y = x+w$, then for any $w$ the number of tilings appears to be a polynomial in $x$. (When $w$ is odd, and $x$ is large enough, there are no tilings.) For $w=6$, the polynomial is $$ \displaylines{ 8192 x^8 + 98304 x^7 + 573440 x^6 + 2064384 x^5 + 4988928 x^4 \hfill\cr \hfill + 8257536 x^3 + 9175040 x^2 + 6291456 x + 2097152. } $$ This can be written as $$ 2^{17} \bigl( \tfrac{1}{2} \bigl(x+\tfrac{3}{2}\bigr)^2 + \tfrac{7}{8} \bigr)^4 $$ or as $$ 2^{17}x^4 \ \ \circ \ \ \tfrac{1}{2}x+\tfrac{7}{8} \ \ \circ \ \ \bigl(x+\tfrac{3}{2}\bigr)^2, $$ where it is understood that these three polynomials get composed. More generally, all the polynomials in $x$ that arise in this fashion appear to ``factor'' in the sense of functional composition. Here are the factored forms of the polynomials for $n=2,4,6,8,10$: $$ \arraycolsep=4pt \begin{array}{rcccl} 2^{3}x^4&\circ&1&\circ&\bigl(x+\frac12\bigr)^2\\[2ex] 2^{8}x^2&\circ&x+1&\circ&\bigl(x+1\bigr)^2\\[2ex] 2^{17}x^4&\circ&\frac12x+\frac{7}{8}&\circ&\bigl(x+\frac32\bigr)^2\\[2ex] 2^{28}x^2&\circ&\frac{1}{144}x^4+\frac{7}{72}x^3 +\frac{41}{144}x^2+\frac{11}{18}x+1&\circ&\bigl(x+2\bigr)^2\\[2ex] 2^{43}x^4&\circ&\frac{1}{144}x^3+\frac{61}{576}x^2 +\frac{451}{2304}x+\frac{967}{1024}&\circ&\bigl(x+\frac52\bigr)^2 \end{array} $$ In general the rightmost polynomial is $(x+w/4)^2$, and the leftmost polynomial is either a perfect square, twice a fourth power, or half a fourth power, depending on $w$ mod 8. A pattern for the middle polynomial however is elusive. \begin{problem} \label{prob14} Find a general formula for the number of domino tilings of Aztec windows. \end{problem} \begin{progress} Constantin Chiscanu found a polynomial bound on the number of domino tilings of the Aztec window of inner order $x$ and outer order $x+w$ \cite{PrCh}. Douglas Zare used the transfer-matrix method to show that the number of tilings is not just bounded by a polynomial, but given by a polynomial, for each fixed $w$ \cite{PrZ}. \end{progress} \section{Miscellaneous} Now we come to some problems involving tiling that fit neither the domino-tiling nor the lozenge-tiling framework. Here the more general picture is that we have some periodic dissection of the plane by polygons, such that an even number of polygons meet at each vertex, allowing us to color the polygons alternately black or white. We then make a suitable choice of a finite region $R$ composed of equal numbers of black and white polygons, and we look at the number of ``diform'' tilings of the region, where a \textit{diform} is the union of two polygonal cells that share an edge. In the case of domino tilings, the underlying dissection of the infinite plane is the tiling by squares, 4 around each vertex; in the case of lozenge tilings, the underlying dissection of the infinite plane is the tiling by equilateral triangles, 6 around each vertex. Other sorts of periodic dissections have already played a role in the theory of enumeration of matchings. For instance, there is a tiling of the plane by isosceles right triangles associated with a discrete reflection group in the plane; in this case, the right choice of $R$ (see Figure 3) gives us a region that can be tiled in $5^{n^2/4}$ ways when $n$ is even and in $5^{(n^2-1)/2}$ or $2 \cdot 5^{(n^2-1)/2}$ ways when $n$ is odd \cite{PrY}. \begin{figure} \vskip1pt \centerline{\psfig{file=\figdir/fortress}} \vskip-1pt \caption{A fortress of order 5, with $2 \times 5^6$ diform tilings.} \label{fortress} \end{figure} \begin{figure}[b] \vskip1pt \centerline{\psfig{file=\figdir/azdungeon}} \vskip-1pt \caption{An Aztec dungeon of order 2, with $13^3$ diform tilings.} \label{azdungeon} \end{figure} Similarly, in the tiling of the plane by triangles that comes from a 30 degree, 60 degree, 90 degree right triangle by repeatedly reflecting it in its edges, a certain region called the ``Aztec dungeon'' (see Figure 4) gives rise to a tiling problem in which powers of 13 occur (as was proved in not-yet-published work of Mihai Ciucu). A key feature of these regions $R$ is revealed by looking at the colors of those polygons in the dissection that share an edge with the border of $R$. One sees that the border splits up into four long stretches such that along each stretch, all the polygons that touch the border have the same color. It is not clear why regions with this sort of property should be the ones that give rise to the nicest enumerations, but this appears to happen in practice. One interesting case arises from a rather symmetric dissection of the plane into equilateral triangles, squares, and regular hexagons, with 4 polygons meeting at each vertex and with no two squares sharing an edge. A typical diform tiling of this region (called a ``dragon'') is shown in Figure 5. \begin{figure} \centerline{\psfig{file=\figdir/dragon}} \caption{A dragon of order 10 (tiled).} \label{trisquex} \end{figure} Empirically, one finds that the number of diform tilings is $2^{n(n+1)}$. \begin{problem} \label{prob15} Prove that the number of diform tilings of the dragon of order $n$ is $2^{n(n+1)}$. \end{problem} \begin{progress} Ben Wieland solved this problem (private communication). \end{progress} \hskip\parindent Incidentally, the tiling shown in Figure 5 was generated using an algorithm that generates each of the possible diform tilings of the region with equal probability. It is no fluke that the tiling looks so orderly in the left and right corners of the region; this appears to be typical behavior in situations of this kind. This phenomenon has been analyzed rigorously for two tiling-models: lozenge tilings of hexagons \cite{PrCLP} and domino tilings of Aztec diamonds \cite{PrCEP}. One way to get a new dissection of the plane from an old one is to refine it. For instance, starting from the dissection of the plane into squares, one can draw in every $k$-th southwest-to-northeast diagonal. When $k$ is 1, this is just a distortion of the dissection of the plane into equilateral triangles. When $k$ is 2, this is a dissection that leads to finite regions for which the number of diform tilings is a known power of 2, thanks to a theorem of Chris Douglas \citeyear{PrD}. But what about $k=3$ and higher? For instance, we have the roughly hexagonal region shown in Figure 6; certain boundary vertices have been marked with a dot so as to bring out the large-scale $2,3,2,2,3,2$ hexagonal structure more clearly. \begin{figure}[b] \centerline{\psfig{file=\figdir/wieland,width=2in}} \caption{A region for Problem 16.} \label{wieland} \end{figure} The cells of this region are triangles and squares. The region has $17920 = 2^9 \cdot 5 \cdot 7$ diform tilings. \begin{problem} \label{prob16} Find a formula for the number of diform tilings in the $a,b,c$ quasihexagon in the dissection of the plane that arises from slicing the dissection into squares along every third upward-sloping diagonal. \end{problem} \hskip\parindent One reason for my special interest in Problem 16 is that it seems to be a genuine hybrid of domino tilings of Aztec diamonds and lozenge tilings of hexagons. \begin{progress} Ben Wieland solved this problem in the case $a=b=c$ (which, as it turns out, is also the solution to the case $a=b<c$ and the case $a=c<b$). In these cases the number of tilings is always a power of two. The general case does not yield round numbers, so there is no simple product formula. \end{progress} \hskip\parindent The approach underlying Ben Wieland's solutions to the last two problems is a method of subgraph substitution that has already been of great use in enumeration of matchings of graphs. I will not go into great detail here on this method [\citeNP{PrPr1}; \citeyearNP{PrPr2}], but here is an overview: One studies graphs with weights assigned to their edges, and one does weighted enumeration of matchings, where the weight of a matching is the product of the weights of the constituent edges. One then looks at local substitutions of subgraphs within a graph that preserve the sum of the weights of the matchings, or more generally, multiply the sum of the weights of the matchings by some predictable factor. Then the problem of weight-enumerating matchings of one graph reduces to the problem of weight-enumerating matchings of another graph. Iterating this procedure, one can often eventually reduce the graph to something easier to understand. \medskip Problems 15 and 16 are just two instances of a broad class of problems arising from periodic graphs in the plane. A unified understanding of this class of problems has begun to emerge, by way of subgraph substitution. The most important open problem connected with this class of results is the following: \begin{problem} \label{prob17} Characterize those local substitutions that have a predictable effect on the weighted sum of matchings of a graph. \end{problem} \begin{figure}[b] \centerline{\psfig{file=\figdir/urban,width=3in}} \caption{The ``urban renewal'' substitution.} \label{urban} \end{figure} \hskip\parindent The most useful local substitution so far has been the one shown in Figure~7, where unmarked edges have weight 1 and where $A,B,C,D$ are respectively obtained from $a,b,c,d$ by dividing by $ad+bc$; if $G$ and $G'$ denote the graph before and after the substitution, one can check that the sum of the weights of the matchings of $G'$ equals the sum of the weights of the matchings of $G$ divided by $ad+bc$. It is required that the four innermost vertices have no neighbors other than the four vertices shown; this constraint is indicated by circling them. Noncircled vertices may have any number of neighbors. \begin{figure}[h] \vskip2pt \centerline{\psfig{file=\figdir/kenyon,width=2in}} \vskip-2pt \caption{Rick Kenyon's substitution.} \label{kenyon} \end{figure} The substitution shown in Figure 8 (a straightforward generalization of a clever substitution due to Rick Kenyon) has also been of use. Here the new weights are not entirely determined by the old, but have a single degree of freedom; the relevant formulas can be written as $$ \def\hskip 10 pt minus 6pt{\hskip 10 pt minus 6pt} A = \frac{abc+aeg+cdf}{bc+eg} \,, \hskip 10 pt minus 6pt B = b \,, \hskip 10 pt minus 6pt D = \frac{dg}{bc+eg} E \,, \hskip 10 pt minus 6pt F = ef \frac{1}{E} \,,\hskip 10 pt minus 6pt G = (bc+eg) \frac{1}{E}\,, $$ with $E$ free. As before, the circled vertices must not have any neighbors other than the ones shown. In this case, the sum of the weights in the before-graph $G$ is exactly equal to the sum of the weights in the after-graph $G'$; there is no need for a correction factor like the $1/(ad+bc)$ that arises in urban renewal. The extremely powerful ``wye-delta'' substitution of Colbourn, Provan, and Vertigan \cite{PrCPV} should also be mentioned. \medskip Up till now we have been dealing exclusively with bipartite planar graphs. We now turn to the less well-explored nonbipartite case. For instance, one can look at the triangle graph of order $n$, shown in Figure~9 in the case $n=4$. (Here $n$ is the number of vertices in the longest row.) \begin{figure} \centerline{\psfig{file=\figdir/triangle,width=1.1in}} \caption{The triangle graph.} \label{triangle} \end{figure} Let $M(n)$ denote the number of matchings of the triangle graph of order $n$. When $n$ is 1 or 2 mod 4, the graph has an odd number of vertices and $M(n)$ is 0; hence let us only consider the cases in which $n$ is 0 or 3 mod 4. Here are the first few values of $M(n)$, expressed in factored form: $2$, $2 \cdot 3$, $2 \cdot 2 \cdot 3 \cdot 3 \cdot 61$, $2 \cdot 2 \cdot 11 \cdot 29 \cdot 29$, $2^3 \cdot 3^3 \cdot 5^2 \cdot 7^2 \cdot 19 \cdot 461$, $2^3 \cdot 5^2 \cdot 37^2 \cdot 41 \cdot 139^2$, $2^4 \cdot 73 \cdot 149 \cdot 757 \cdot 33721 \cdot 523657$, $2^4 \cdot 3^8 \cdot 17 \cdot 37^2 \cdot 703459^2$, \dots. It is interesting that $M(n)$ seems to be divisible by $2^{\lfloor (n+1)/4 \rfloor}$ but no higher power of 2; it is also interesting that when we divide by this power of 2, in the case where $n$ is a multiple of 4, the quotient we get, in addition to being odd, is a perfect square times a small number $(3, 11, 41, 17, \dots{})$. \begin{problem} \label{prob18} How many matchings does the triangle graph of order $n$ have? \end{problem} \begin{progress} Horst Sachs \citeyear{PrSac2} has responded to this problem. \end{progress} \hskip\parindent One can also look at graphs that are bipartite but not planar. A natural example is the $n$-cube (that is, the $n$-dimensional cube with $2^n$ vertices). It has been shown that the number of matchings of the $n$-cube goes like $1$, $2$, $9=3^2$, $272=16 \cdot 17$, $589185=3^2 \cdot 5 \cdot 13093$, \dots. \begin{problem} \label{prob19} Find a formula for the number of matchings of the $n$-cube. \end{problem} \hskip\parindent (This may be intractable; after all, the graph has exponentially many vertices.) \begin{progress} L\'aszl\'o Lov\'asz gave a simple proof of my (oral) conjecture that the number of matchings of the $n$-cube has the same parity as $n$ itself. Consider the orbit of a particular matching of the $n$-cube under the group generated by the $n$ standard reflections of the $n$-cube. If all the edges are parallel (which can happen in exactly $n$ ways), the orbit has size 1; otherwise the size of the orbit is of the form $2^k$ (with $k \geq 1$)\emdash an even number. The claim follows, and similar albeit more complex reasoning should allow one to compute the enumerating sequence modulo any power of 2. Meanwhile, L. H.\ Clark, J. C.\ George, and T. D.\ Porter have shown \cite{PrCGP} that if one lets $f(n)$ denote the number of 1-factors in the $n$-cube, then $$f(n)^{2^{1-n}} \sim n/e$$ as $n \rightarrow \infty$. It was subsequently pointed out by Bruce Sagan that the main result of Clark et al.\ \citeyear{PrCGP} is a special case of the theorem cited by Lov\'asz and Plummer \citeyear[top of page 312]{PrLP}. \end{progress} \hskip\parindent Finally, we turn to a problem involving domino tilings of rectangles, submitted by Ira Gessel (what follows are his words): \catcode `!=11 \newdimen\squaresize \newdimen\thickness \newdimen\Thickness \newdimen\ll! \newdimen \uu! \newdimen\dd! \newdimen \rr! \newdimen \temp! \def\sq!#1#2#3#4#5{% \ll!=#1 \uu!=#2 \dd!=#3 \rr!=#4 \setbox0=\hbox{% \temp!=\squaresize\advance\temp! by .5\uu! \rlap{\kern -.5\ll! \vbox{\hrule height \temp! width#1 depth .5\dd!}}% \temp!=\squaresize\advance\temp! by -.5\uu! \rlap{\raise\temp! \vbox{\hrule height #2 width \squaresize}}% \rlap{\raise -.5\dd! \vbox{\hrule height #3 width \squaresize}}% \temp!=\squaresize\advance\temp! by .5\uu! \rlap{\kern \squaresize \kern-.5\rr! \vbox{\hrule height \temp! width#4 depth .5\dd!}}% \rlap{\kern .5\squaresize\raise .5\squaresize \vbox to 0pt{\vss\hbox to 0pt{\hss $#5$\hss}\vss}}% \ht0=0pt \dp0=0pt \box0 \def\vsq!#1#2#3#4#5\endvsq!{\vbox to \squaresize{\hrule width\squaresize height 0pt% \vss\sq!{#1}{#2}{#3}{#4}{#5}}} \newdimen \LL! \newdimen \UU! \newdimen \DD! \newdimen \RR! \def\vvsq!{\futurelet\next\vvvsq!} \def\vvvsq!{\relax \ifx \next l\LL!=\Thickness \let\continue!=\skipnexttoken! \else\ifx\next u\UU!=\Thickness \let\continue!=\skipnexttoken! \else\ifx\next d\DD!=\Thickness \let\continue!=\skipnexttoken! \else\ifx\next r\RR!=\Thickness \let\continue!=\skipnexttoken! \else\ifx\next P\let\continue!=\place! \else\def\continue!{\vsq!\LL!\UU!\DD!\RR!}% \fi\fi\fi\fi\fi \continue!} \def\skipnexttoken!#1{\vvsq!} \def\place! P#1#2#3{% \rlap{\kern.5\squaresize\temp!=.5\squaresize\kern#1\temp! \temp!=\squaresize \advance\temp! by #2\squaresize \temp!=.5\temp! \raise\temp!\vbox to 0pt{\vss\hbox to 0pt{\hss$#3$\hss}\vss}}\vvsq!} \def\Young#1{\LL!=\thickness \UU!=\thickness \DD! = \thickness \RR! = \thickness \vbox{\smallskip\offinterlineskip \halign{&\vvsq! ## \endvsq!\cr #1}}} \def\Youngt#1{\LL!=\thickness \UU!= \thickness \DD! = \thickness \RR! = \thickness \vtop{\offinterlineskip \halign{&\vvsq! ## \endvsq!\cr #1}}} \def\omit\hskip\squaresize{\omit\hskip\squaresize} \catcode `!=12 \thickness=.4pt We consider dimer coverings of an $m\times n$ rectangle, with $m$ and $n$ even. We assign a vertical domino from row $i$ to row $i+1$ the weight $\sqrt {y_i}$ and a horizontal domino from column $j$ to column $j+1$ the weight $\sqrt {x_j}$. For example, the covering $$\Thickness=0pt \squaresize=30pt \Young{dP0{-1}{\sqrt{y_1}}&rP10 {\sqrt{x_2}}&l&dP0{-1} {\sqrt{y_1}}& rP10{\sqrt{x_5}}&l&rP10{\sqrt{x_7}} &l&dP0{-1}{\sqrt{y_1}}&dP0{-1}{\sqrt{y_1}}\cr u&rP10{\sqrt{x_2}}&l&u&rP10 {\sqrt{x_5}}&l&rP10{\sqrt{x_7}}&l&u&u\cr} $$ for $m=2$ and $n=10$ has weight $y_1^2 x_2x_5x_7$. (The weight will always be a product of integral powers of the $x_i$ and $y_j$.) Now I'll define what I call ``dimer tableaux.'' Take an $m/2$ by $n/2$ rectangle and split it into two parts by a path from the lower left corner to the upper right corner. For example (with $m=6$ and $n=10$) $$\Thickness=1pt \squaresize 15pt \Young{&&&&r\cr &&&&lu\cr d&dr&u&u&\cr} $$ Then fill in the upper left part with entries from 1, 2, \dots, $n-1$ so that for adjacent entries $\squaresize 10pt\lower 2pt\vbox{\Young{\mathstrut{\scriptstyle i}&\mathstrut{\scriptstyle j}\cr}}$ we have $i<j-1$ and for adjacent entries $\squaresize 10pt\lower 8pt\vbox{\Young{\mathstrut{\scriptstyle i}\cr\mathstrut{\scriptstyle j}\cr}}$ we have $i\le j+1$, and fill in the lower-right partition with entries from $1,2\ldots, m-1$ with the reverse inequalities ( $\squaresize 10pt\lower 2pt\vbox{\Young{\mathstrut{\scriptstyle i}&\mathstrut{\scriptstyle j}\cr}}$ implies $i\le j+1$ and $\squaresize 10pt\lower 8pt\vbox{\Young{\mathstrut{\scriptstyle i}\cr\mathstrut{\scriptstyle j}\cr}}$ implies $i<j-1$). We weight an $i$ in the upper-left part by $x_i$ and a $j$ in the lower-right part by $y_j$. \begin{theorem} The sum of the weights of the $m\times n$ dimer coverings is equal to the sum of the weights of the $m/2\times n/2$ dimer tableaux. \end{theorem} My proof is not very enlightening; it essentially involves showing that both of these are counted by the same formula. \begin{problem} \label{prob20} Is there an ``explanation'' for this equality? In particular, is there a reasonable bijective proof? Notes: \begin{enumerate} \item[(1)] The case $m=2$ is easy: the $2\times 10$ dimer covering above corresponds to the $1\times 5$ dimer tableau $$\squaresize 20pt\Thickness1pt\Young{dx_2&dx_5&drx_7&uy_1&uy_1\cr}$$ (there's only one possibility!). \item[(2)] If we set $x_i=y_i=0$ when $i$ is even (so that every two-by-two square of the dimer covering may be chosen independently), then the equality is equivalent to the identity $$\prod_{i,j}(x_i+y_j)=\sum_{\lambda}s_{\lambda}(x)s_{\tilde \lambda'}(y);$$ compare \cite[p.~37]{Macdonald}. This identity can be proved by a variant of Schensted's correspondence, so a bijective proof of the general equality would be essentially a generalization of Schensted. Several people have looked at the problem of a Schensted generalization corresponding to the case in which $y_i=0$ when $i$ is even. \item[(3)] The analogous results in which $m$ or $n$ is odd are included in the case in which $m$ and $n$ are both even. For example, if we take $m=4$ and set $y_3=0$, then the fourth row of a dimer covering must consist of $n/2$ horizontal dominoes, which contribute $\sqrt{x_1x_3\cdots x_{n-1}}$ to the weight, so we are essentially looking at dimer coverings with three rows. \end{enumerate} \end{problem} \begin{progress} A special case of the Robinson--Schensted algorithm given by Sundquist et al.\ \citeyear{PrSWW} can be used to get a bijection for a special case of the problem, in which one sets $y_i = 0$ for all $i$ even, so that we are looking at dimer coverings (or domino tilings) in which every vertical domino goes from row $2i+1$ to row $2i+2$ for some $i$. These tilings are not very interesting because they break up into tilings of 2-by-$n$ rectangles. But even so, the Robinson--Schensted bijection is nontrivial. \end{progress} \section{New Problems} Let $N(a,b)$ denote the number of matchings of the $a$-by-$b$ rectangular grid. Kasteleyn showed that $N(a,b)$ is equal to the square root of the absolute value of $$\prod_{j=1}^{a} \prod_{k=1}^{b} \left(2 \cos \frac{\pi j}{a+1} + 2i \cos \frac{\pi k}{b+1}\right).$$ Some number-theoretic properties of $N(a,b)$ follow from this representation (see, e.g., \cite{PrCo}) but lack a combinatorial explanation. The next two problems describe two such facts. \begin{problem} \label{prob21} Give a combinatorial proof of the fact that $N(a,b)$ divides $N(A,B)$ whenever $a+1$ divides $A+1$ and $b+1$ divides $B+1$. \end{problem} \begin{progress} Bruce Sagan has given an answer in the ``Fibonacci case'' $a=2$. A matching of a $2$-by-$(kn-1)$ grid either splits up as a matching of a $2$-by-$(n-1)$ grid on the left and a $2$-by-$(kn-n)$ grid on the right or it splits up as a matching of a $2$-by-$(n-2)$ grid on the left, a horizontal matching of a $2$-by-$2$ grid in the middle, and a matching of a $2$-by-$(kn-n-1)$ grid on the right. Hence $$N(2,@kn-1) = N(2,@n-1)@N(2,@kn-n) + N(2,@n-2)@N(2,@(k-1)n-1).$$ {}From this formula one can prove that $N(2,n-1)$ divides $N(2,kn-1)$ by induction on $k$. Volker Strehl has approached the problem in a different way; his ideas make it seem likely that a better combinatorial understanding of resultants, in combination with known interpretations of Chebyshev polynomials, would be helpful in approaching this problem. \end{progress} \begin{problem} \label{prob22} Give a combinatorial proof of the fact that $N(a,2a)$ is always congruent to 1 mod 4. \end{problem} \hskip\parindent (Pachter \citeyear{PrPa} has demonstrated the sort of combinatorial methods one can use in such problems.) \medbreak Even without Kasteleyn's formula, it is easy to show (e.g., via the transfer-matrix method) that for any fixed $a$, the sequence of numbers $N(a,b)$ (with $b$ varying) satisfies a linear recurrence relation with constant coefficients. Indeed, consider all $2^a$ different ways of removing some subset of the $a$ rightmost vertices in the $a$-by-$b$ grid; this gives us $2^a$ ``mutilated'' versions of the graph. We can set up recurrences that link matchings of mutilated graphs of width $b$ with matchings of mutilated graphs of width $b$ and $b-1$, and standard algebraic methods allow us to turn this system of joint mutual recurrences of low degree into a single recurrence of high degree governing the particular sequence of interest, which enumerates matchings of unmutilated rectangles. The recurrence obtained in this way is not, however, best possible, as one can see even in the simple case $a=2$. \begin{problem}[Stanley] Prove or disprove that the minimum degree of a linear recurrence governing the sequence $N(a,1),N(a,2),N(a,3),\dots$ is $2^{\lfloor (a+1)/2 \rfloor}$. \end{problem} \begin{progress} Observations made by Stanley \citeyear[p.~87]{PrSt2} imply that the conjecture is true when $a+1$ is an odd prime. \end{progress} \hskip\parindent The idea of mutilating a graph by removing some vertices along its boundary leads us to the next problem. It has been observed for small values of $n$ that if one removes equal numbers of black and white vertices from the boundary of a $2n$-by-$2n$ square grid, the number of matchings of the mutilated graph is less than the number of matchings of the original graph. In fact, it appears to be true that one can delete \textit{any} subset of the vertices of the square grid and obtain an induced graph with strictly fewer matchings than the original. It is worth pointing out that not every graph shares this property with the square grid. For instance, if $G$ is the Aztec diamond graph of order 5 and $G'$ is the graph obtained from $G$ by deleting the middle vertices along the northwest and northeast borders, then $G$ has 32768 matchings while $G'$ has 59493. \begin{problem} Prove or disprove that every subgraph of the $2n$-by-$2n$ grid graph has strictly fewer matchings. \end{problem} \hskip\parindent Next we come to a variant on the Aztec dungeon region shown in Figure 4. Figure 10 shows an ``hexagonal dungeon'' with sides $2,4,4,2,4,4$. Matt Blum's investigation of these shapes has led him to discover many patterns; the most striking of these patterns forms the basis of the next problem. \begin{figure} \centerline{\psfig{file=\figdir/hexdungeon,scale=78}} \caption{An hexagonal dungeon.} \label{hexdungeon} \end{figure} \begin{problem} Show that the hexagonal dungeon with sides $a,2a,b,a,2a,b$ has exactly $$13^{2a^2} 14^{\lfloor a^2/2 \rfloor}$$ diform tilings, for all $b \geq 2a$. \end{problem} \hskip\parindent Unmatchable bipartite graphs can sometimes give rise to interesting quasimatching problems, either by way of $K K^*$ (see Problem 6) or by systematic addition or deletion of vertices or edges. The former sort of problem simply asks for the determinant of $K K^*$ (where we may assume that $K$ has more columns than rows). When the underlying graph has equal numbers of black and white vertices, this is just the square of the number of matchings, but when $K$ is a rectangular matrix, $K K^*$ will in general have a nonzero determinant, even though the graph has no matchings. \setcounter{topnumber}{1} \begin{problem} Calculate the determinant of $K K^*$ where $K$ is the Kasteleyn--Percus matrix of the $a,b,c,d,e,f$ honeycomb graph. \end{problem} \hskip\parindent (Note that in this case we can simply take $K$ to be the bipartite adjacency matrix of the graph.) Cases of special interest are $a$, $b+1 $, $c $, $a+1 $, $b $, $c+1$ and $a,b,a,b,a,b$ hexagons. These two cases overlap in the one-parameter family of $a $, $a+1 $, $a $, $a+1 $, $a $, $a+1$ hexagons. For instance, in the case of the $3,4,3,4,3,4$ hexagon, $\det(K K^*)$ is $2^8\cdot 3^3\cdot 7^6$. \begin{problem} \label{prob27} Calculate the determinant of $K K^*$ where $K$ is the Kasteleyn--Percus matrix of an $m$-by-$n$ Aztec rectangle, or where $K$ is the Kasteleyn--Percus matrix of the ``fool's diamond'' of order $n$. (The fool's diamond of order 3 is the following region: $$\vcenter\brd{ X XXX XXXXX XXX X }$$ Fool's diamonds of higher orders are defined in a similar way.) \end{problem} \begin{progress} In the case of Aztec rectangles, Matt Blum has found general formulas for $\det(KK^*)$ when $m$ is 1, 2, or 3. For fool's diamonds, we get $$ \medmuskip 1mu \displaylines{ 1\cr 2\cr 3\cdot 5\cr 2^7\cdot 3\cr 3^2\cdot 5^3\cdot 29\cr 2^9\cdot 3\cdot 5 \cdot 7 \cdot 13^2\cr 7^3\cdot 13^4\cdot 29^2\cr 2^{25}\cdot 3\cdot 7^2\cdot 17^3\cr} $$ (One might also look at ``fool's rectangles''.) \end{progress} \hskip\parindent Another thing one can do with an unmatchable graph is add extra edges. Even when this ruins the bipartiteness of the graph, there can still be interesting combinatorics. For instance, consider the $2,4,2,4,2,4$ hexagon-graph; it has an even number of vertices, but it has a surplus of black vertices over white vertices. We therefore introduce edges between every black vertex and the six nearest black vertices. (That is, in each hexagon of the honeycomb, we draw a triangle connecting the three black vertices, as in Figure~\ref{hextwo}.) Then the graph has $5187 = 3\cdot7\cdot13\cdot19$ matchings. \begin{figure} \centerline{\psfig{file=\figdir/hextwo,scale=90}} \caption{A hexagon with extra edges.} \label{hextwo} \end{figure} \begin{problem} \label{prob28} Count the matchings of the $a,b,c,d,e,f$ hexagon-graph in which extra edges have been drawn connecting vertices of the majority color. \end{problem} \begin{figure} \centerline{\psfig{file=\figdir/rectwo,scale=96}} \caption{An Aztec rectangle with extra edges.} \label{rectwo} \end{figure} \begin{figure} \centerline{\psfig{file=\figdir/holtwo,scale=75}} \caption{A holey Aztec rectangle with extra edges.} \label{holtwo} \end{figure} \hskip\parindent What works for honeycomb graphs works (or seems to work) for square-grid graphs as well. If one adds edges joining each vertex of majority color to the four nearest like-colored vertices in the $n$ by $n+2$ Aztec rectangle graph as in Figure~\ref{rectwo}, one gets a graph for which the number of matchings grows like $2^2\cdot3$, $2^3\cdot3\cdot7$, $2^7\cdot 3\cdot 11$, $2^{17}\cdot 5\cdot 31$, etc. If one does the same for the holey $2n-1$ by $2n$ Aztec rectangle from which the central vertex has been removed, as in Figure~\ref{holtwo}, one gets the numbers $2^6\cdot 7$, $2^9\cdot 3^2\cdot 13\cdot 17$, $2^{23}\cdot 5^3\cdot 31$, etc. \begin{problem} \label{prob29} Count the matchings of the $a$ by $b$ Aztec rectangle (with $a+b$ even) in which extra edges have been drawn connecting vertices of the majority color. Do the same for the $2n-1$ by $2n$ holey Aztec rectangle. \end{problem} \hskip\parindent Other examples of nonbipartite graphs for which the number of matchings has only small prime factors arise when one takes the quotient of a symmetrical bipartite graph modulo a symmetry-group at least one element of which interchanges the two colors; Kuperberg \citeyear{PrKu1} gives some examples of this. In general, there seem to be fewer product-formula enumerations of matchings for nonbipartite graphs than for bipartite graphs. Nevertheless, even in cases where no product formula has been found, there can be patterns in need of explanation. Consider the one-parameter family of graphs illustrated in Figure 14 for the case $n=7$ (based on the same nonbipartite infinite graph \begin{figure} \centerline{\psfig{file=\figdir/diab,width=2.5in}} \caption{An isosceles right triangle graph with extra edges.} \label{diab} \end{figure} as Figures 12 and 13). Such a graph has an even number of vertices whenever $n$ is congruent to 0 or 3 modulo 4. Here are the data for the first few cases, courtesy of Matt Blum: $$ \medmuskip 1mu \tabskip0pt plus 1fil \vbox{ \halign to \hsize{\hfil$#$&\hfil#&$#$\hfil\cr n& number of matchings& $factorization$\cr 3& 3& 3\cr 4& 6& 2 \cdot 3\cr 7& 1065& 3 \cdot 5 \cdot 71\cr 8& 6276& 2^2 \cdot 3 \cdot 523\cr 11& 45949563& 3^2 \cdot 11 \cdot 464137\cr 12& 807343128& 2^3 \cdot 3^2 \cdot 1109 \cdot 10111\cr 15& 221797080594801& 3^2 \cdot 24644120066089\cr 16& 11812299253803024& 2^4 \cdot 3 \cdot 246089567787563\cr 19& 117066491250943949567763& 3 \cdot 89 \cdot 28289 \cdot 15499002371714201\cr 20& 19100803250397148607852640& 2^5 \cdot 3^2 \cdot 5 \cdot 41 \cdot 367 \cdot 881534305952328473\cr }}$$ The following problem describes some of Blum's conjectures: \begin{problem} \label{prob30} Show that for the isosceles right triangle graph with extra edges, the number of matchings is always a multiple of 3. Furthermore, show that the exact power of 2 dividing the number of matchings is $2^{n/4}$ when $n$ is 0 modulo 4, and $2^0(=1)$ when $n$ is 3 modulo 4. \end{problem} \hskip\parindent This property of divisibility by 3 pops up in another problem of a similar flavor. Consider the graph shown in Figure 15, which is just like the one shown in Figure 9, except that half of the triangular cells have an extra vertex in them, connected to the three nearest vertices. (Note also the resemblance to Figure~11.) \begin{figure} \centerline{\psfig{file=\figdir/newtri,width=1.4in}} \caption{An equilateral triangle graph with extra vertices and edges.} \label{newtri} \end{figure} \begin{problem} \label{prob31} Show that for the equilateral triangle graph with extra vertices and edges, the number of matchings is always a multiple of 3. \end{problem} \hskip\parindent (I refrain from making a conjecture about the exponent of 2, though the data contain patterns suggestive of a general rule.) \medskip It may be too soon to try to assemble into one coherent picture all the diverse phenomena discussed in the preceding 31 problems. But I have noticed a gratuitous symmetry that governs many of the exact formulas, and I will close by pointing it out. Consider, for example, the MacMahon--Macdonald product $$M_n = \prod_{i=0}^{n-1} \prod_{j=0}^{n-1} \prod_{k=0}^{n-1} \frac{i+j+k+2}{i+j+k+1}$$ that counts matchings of the $n,n,n$ semiregular honeycomb graph. We find that the ``second quotient'' $M_{n-1} M_{n+1} / M_n^2$ is the rational function $$\frac{27}{64} \,\frac{(3n-2)(3n-1)^2(3n+1)^2(3n+2)}{(2n-1)^3(2n+1)^3}$$ which is an even function of $n$. The right hand side in Bo-Yin Yang's theorem (giving the number of diabolo tilings of a fortress of order $n$) has a power of 5 whose exponent is $n^2/4$ when $n$ is even and $(n^2-1)/4$ when $n$ is odd; this too is an even function of $n$. Domino tilings of Aztec diamonds are enumerated by the formula $2^{n(n+1)}$. Here the symmetry is a bit different: replacing $n$ by $-1-n$ leaves the answer unaffected. The right hand side of Mihai Ciucu's theorem (giving the number of diform tilings of an Aztec dungeon of order $n$) has a power of 13 whose exponent is $(n+1)^2/3$ or $n(n+2)/3$ (according to whether or not $n$ is 2 mod 3). so that the symmetry corresponds to replacing $n$ by $-2-n$. There are other instances of this kind that arise, in which some base is raised to the power of some quadratic function of $n$; in each case, the quadratic function admits a symmetry that preserves the integrality of $n$ (unlike, say, the quadratic function $n(3n+1)/2$, which as a function from integers to integers does not possess such a symmetry). \begin{problem} \label{prob32} For many of our formulas, the ``algebraic'' (right hand) side is invariant under substitutions that make the ``combinatorial'' (left hand) side meaningless, insofar as one cannot speak of graphs with negative numbers of vertices or edges. Might this invariance nonetheless have some deeper significance? \end{problem} \hskip\parindent Cohn \citeyear{PrCo} has found another example of gratuitous symmetry related to tilings. \section*{Acknowledgements} This research was conducted with the support of the National Science Foundation, the National Security Agency, and the M.I.T. Class of 1922 Career Development fund. I am deeply indebted to the past and present members of the Tilings Research Group for their many forms of assistance: Pramod Achar, Karen Acquista, Josie Ammer, Federico Ardila, Rob Blau, Matt Blum, Carl Bosley, Ruth Britto-Pacumio, Constantin Chiscanu, Henry Cohn, Chris Douglas, Edward Early, Nicholas Eriksson, David Farris, Lukasz Fidkowski, Marisa Gioioso, David Gupta, Harald Helfgott, Sharon Hollander, Dan Ionescu, Sameera Iyengar, Julia Khodor, Neelakantan Krishnaswami, Eric Kuo, Yvonne Lai, Ching Law, Andrew Menard, Alyce Moy, Anne-Marie Oreskovich, Ben Raphael, Vis Taraz, Jordan Weitz, Ben Wieland, Lauren Williams, David Wilson, Jessica Wong, Jason Woolever, and Laurence Yogman. I also acknowledge the helpful comments on this manuscript given by Henry Cohn and Richard Stanley, and the information provided by Jerry Dias, Michael Fisher and Horst Sachs concerning the connections between matching theory and the physical sciences. \bibitemsep=5pt plus 1pt \def\citeauthoryear #1#2#3{#2\ #3} \bibliographystyle{msribib}
\section{Introduction} In recent years, there has been extensive experimental and theoretical work in the field of mesoscopic systems, including superconducting structures in proximity with normal metals \cite{review}. In particular, the paramagnetic reentrance phenomenon \cite{visaniprl} has received a wide interest, mainly because of being promoted by the recent understanding of the high-temperature diamagnetic response of rather clean normal-metal--superconductor (NS) proximity structures \cite{bmueller1} in the context of the quasiclassical Eilenberger theory including elastic scattering \cite{belzig}. In this Letter, we discuss the very low temperature reentrant behavior of two of the AgNb samples of Ref.\ \cite{bmueller1}, covering a larger mesoscopic regime with respect to previous measurements \cite{visaniprl}. Recently, two Letters \cite{bruder,fauchere99} have addressed the origin of paramagnetic currents in NS systems, which might lead to an understanding of the paramagnetic reentrance phenomenon. The work of Bruder and Imry \cite{bruder} is based on the presence of non-Andreev-reflecting semiclassical trajectories at the outer surface of a nonsingly connected proximity system (glancing states), which carry predominantly paramagnetic currents. This work has been subject to debate because of the small magnitude \cite{comment}. A different, more elaborate approach by Fauch\`ere \textit{et al}. \cite{fauchere99} assumes a net repulsive interaction in the noble metals. The $\pi$ shift of the order parameter at the NS interface then leads to a paramagnetic instability of Andreev pairs. The first work reflects the cylindrical geometry of our NS system, but it does not address the experimental signatures of the paramagnetic reentrance, namely absolute value of order 1, temperature dependence, nonlinearity, hysteresis, or dissipation \cite{visaniprl}. The latter three features might give evidence for a spontaneous magnetization in the samples, as proposed in Ref.\ \cite{fauchere99}. However, the second theoretical approach \cite{fauchere99} does not reach beyond qualitative accordance with our experiment. Here we discuss an investigation of the paramagnetic reentrant effect extending to the $\mu$K region. By covering five decades in temperature, we have been able to extract the correct temperature dependence of the NS proximity structure below $T_{c}$. Over a large mesoscopic regime, the cylindrical structure clearly displays different levels of coherence along integer multiples of the wire perimeter $L$. Unfortunately, neither of the two present theories \cite{bruder,fauchere99}, which discuss different sources of paramagnetism in NS structures, obtain the correct absolute value as well as the $T$ dependence of the paramagnetic reentrant effect. The two samples reported here are ensembles of cylindrical wires with a superconducting core of soft niobium ($RRR\approx 300$) concentrically embedded in a normal-metal matrix of 6N silver. Their total diameter was mechanically reduced by several steps of swagging and co-drawing \cite{flukiger} to final values $41\,\mu{\mathrm m}$ and $23\,\mu{\mathrm m}$, with normal layer thicknesses $d_{N}=5.5\,\mu{\mathrm m}$ and $3.3\,\mu{\mathrm m}$ (the ratio $d_{N}/L$ is approximately the same), respectively. The samples were \begin{figure} \includegraphics[width=0.96\linewidth]{XdclogT.eps} \caption{Magnetic susceptibility $\chi_{N}(T)$, between $150\,\mu{\mathrm K}$ and $9\,{\mathrm K}$. For both samples we show $\chi_{ac}(T)$ ($+$) and $\chi_{dc}(T)$ ($\bullet$). The arrows mark the direction of $T$--changes. } \protect\label{Xdc(logT)} \end{figure} \noindent annealed after the last drawing, and values of the mean free path $\ell_{N}\sim(0.5$--$0.8)d_{N}$ were obtained. For more details see Refs.\ \cite{journlowtempphys,bmueller1}. Extensions of the measurements to $\mu$K temperatures were performed at the ultralow temperature (ULT) facility at the University of Bayreuth. There, an experimental setup was installed for inductive measurements using an rf--SQUID sensor. Magnetic fields were applied along the axis of the wires. For the ULT experiments, we took parts of the wire bundle measured in our dilution refrigerator \cite{visaniprl,bmueller1}, and glued them with GE 7031 varnish to high purity gold foils tightly attached to a silver finger, in good electrical contact with the Cu demagnetization stage \cite{gloos88}. Thus, about 200 wires were mounted. Temperatures were measured with a Pt pulsed NMR thermometer \cite{gloos88}. In the following we report on the temperature dependent magnetic susceptibility of our relatively clean silver-niobium samples 3AgNb [$d_{N}=5.5\,\mu{\mathrm m}$] and 5AgNb [$d_{N}=3.3\,\mu{\mathrm m}$]. In Fig.\ \ref{Xdc(logT)} the total magnetic susceptibility is shown as a function of temperature. We show $\chi_{ac}(T)$ between $4\,{\mathrm mK}$ and $9\,{\mathrm K}$, measured in our dilution refrigerator with field amplitude $H_{ac}=33\,{\mathrm mOe}$ and frequency $\nu=80\,{\mathrm Hz}$, as well as $\chi_{dc}(T)$ at constant $H_{dc}$ at ULT and LT. At temperatures below the critical temperature of Nb $T_{c}=9.2\,{\mathrm K}$, the magnetic susceptibility of the N layer exhibits diamagnetism induced through Andreev reflection at the highly transparent NS interface. At lower temperatures, it develops almost total Meissner screening in the Ag layer \cite{bmueller1}. Below $T_{r}\sim 100\,{\mathrm mK}$ the signature of reentrance is observed both in $\chi_{ac}(T)$ \cite{visaniprl} and $\chi_{dc}(T)$, with the development of an additional paramagnetic susceptibility $\chi_{\mathrm para}(T)$, such that $\chi_{N}(T)=\chi_{\mathrm dia}(T)+\chi_{\mathrm para}(T)$. For sample 3AgNb, the susceptibility $\chi_{N}$ saturates below $T^{\mathrm sat}\approx 400\,\mu{\mathrm K}$, displaying a complete cancellation of the induced diamagnetic susceptibility in N, such that only the diamagnetism in S seems to remain. For sample 5AgNb the susceptibility $\chi_{N}$ shows saturation below $T^{\mathrm sat}\approx 800\,\mu{\mathrm K}$ at a paramagnetic value $4\pi\chi_{N}^{\mathrm sat}\approx 1$, indicating a complete cancellation of the \textit{total} diamagnetic susceptibility in N plus S. The anomalously large magnitude of the paramagnetic reentrant susceptibility $\chi_{\mathrm para}$ at ultralow temperatures is rather intriguing. This magnitude, and particularly its dependence on the sample size, is not explained by existing theories. At this moment, it is not clear, if paramagnetic reentrance is (i) an intrinsic effect of mesoscopic NS proximity structures in the very low temperature limit or (ii) the result of two independent phenomena. In the latter case the selection of the NS materials could be important. A detailed inspection $\chi_{N}$ around its minimum reveals that, the reentrance temperature $T_{r}$ is decreased under a field $H_{dc}\approx 0.2\,{\mathrm Oe}$. The ac and dc curves show reentrant temperatures $T_{r}=97\,{\mathrm mK}$ and $T_{r}^{H}=83\,{\mathrm mK}$ for sam- \begin{figure} \includegraphics[width=0.96\linewidth]{XdcTlog.eps} \caption{Reentrant paramagnetic susceptibility $\chi_{\mathrm para}(T)$ below $T_{r}$. For both samples two measurements of $\chi_{dc}(T)$ ($\bullet$) are shown, and $\chi_{ac}(T)$ ($+$). The thick arrows indicate the direction of temperature changes. The temperature scales are $\propto 1/L$. The vertical arrows indicate the temperatures at which the condition $\xi_{N}(T)=nL$ is met. } \protect\label{Xdc(T)log} \end{figure} \noindent ple 3AgNb ($T_{r}=149\,{\mathrm mK}$ and $T_{r}^{H}=113\,{\mathrm mK}$ for sample 5AgNb), respectively. Neglecting the weak-field effect on $T_{r}$, $\chi_{dc}(T)$ matches $\chi_{ac}(T)$ for sample 3AgNb (5AgNb) between $15\,{\mathrm mK}$ ($30\,{\mathrm mK}$) and $1\,{\mathrm K}$. This is rather noticeable, considering the strong nonlinearity in magnetic field discussed below. Above $T\approx1\,{\mathrm K}$, $\chi_{dc}(T)$ deviates from $\chi_{ac}(T,H_{ac}\approx 0)$, due to the depression of the weak induced Andreev pair potential by finite fields. Below $15\,{\mathrm mK}$ ($30\,{\mathrm mK}$), $\chi_{dc}(T)$ agrees with $\chi_{ac}(T)$ only on measurements done under cooling. Below $15\,{\mathrm mK}$ ($30\,{\mathrm mK}$) the warming up curve of $\chi_{dc}(T)$ lies above the cooling curves, corresponding to stronger paramagnetism adding to the full Meissner screening. In our arrangement, samples are in rather good thermal contact with the Cu demagnetization stage and the Pt NMR thermometer. Indeed, the measured thermal relaxation times at the lowest temperatures remain below $1000\,{\mathrm s}$. Nevertheless, the susceptibility shows hysteresis. We have found the highest levels of paramagnetic reentrance only after allowing the NS system to remain at much below their saturation temperatures for long periods of time (one week or longer). Fig.\ \ref{Xdc(T)log} shows the reentrant paramagnetic susceptibility $\chi_{\mathrm para}(T)$ below $T_{r}$. The data exponentially increases as $\chi_{\mathrm para}(T)=A\exp{(-T/T^\ast)}$. For sample 3AgNb [Fig.\ \ref{Xdc(T)log}(a)], the prefactor $A_{1}=0.24$ and characteristic temperature $T_{1}^\ast=13\,{\mathrm mK}$ were obtained from $\chi_{ac}(T)$. The cooling and warming $\chi_{dc}(T)$, above $\approx 13\,{\mathrm mK}$ reproduce well the behavior of $\chi_{\mathrm para}$ observed in $\chi_{ac}(T)$. Near $\approx 13\,{\mathrm mK}$, $\chi_{dc}$ displays a kink, leading to an approximate doubling of the logarithmic slope, with $T^\ast_{2}=5.5\,{\mathrm mK}\sim T^\ast_{1}/2$ and a prefactor $A_{2}=1.1$. For sample 5AgNb [Fig.\ \ref{Xdc(T)log}(b)], the prefactor $A_{1}=0.4$ and characteristic temperature $T_{1}^\ast=20\,{\mathrm mK}$ were obtained from $\chi_{ac}(T)$. Again the cooling and warming $\chi_{dc}(T)$ above $\approx 30\,{\mathrm mK}$ follow the $T$ behavior of $\chi_{ac}(T)$. The curves show a kink in the susceptibility, displaying a doubling of the logarithmic slope, with $T^\ast_{2}=11\,{\mathrm mK}\sim T^\ast_{1}/2$ for the second line. In addition, around $\approx14\,{\mathrm mK}$, a second doubling of the logarithmic slope of $\chi_{dc}(T)$ occurs, with $T^\ast_{3}=4.8\,{\mathrm mK}\sim T^\ast_{2}/2$ for the third line. The coherence length of the Andreev pairs $\xi_{N}$ in Ag, obtained from our breakdown field measurements, is in agreement with the clean limit theory, $\xi_{N}=\hbar v_{F}/2\pi k_{B}T=1.69\,\mu{\mathrm m}/T({\mathrm K})$ \cite{bmueller1}. At the temperature of the first kink in $\chi_{\mathrm para}$, $\xi_{N}(T)$ reaches approximately the value of a single wire's circumference $L=130\,\mu{\mathrm m}$ ($L=72\,\mu{\mathrm m}$). In Fig.\ \ref{Xdc(T)log} we have indicated by vertical arrows the temperature at which the equality $\xi_{N}(T)=L$ is met. For sample 5AgNb, at the temperature of the second kink it is $\xi_{N}(T)=2L$. The values of $T_{1}^\ast$, $T_{2}^\ast$, and $T_{3}^\ast$, as well as the position of the kinks, which are located approximately at $T_{1}^\ast$, $T_{2}^\ast$, give evidence for different levels of quantum coherence on the mesoscopic length scale $L$. The temperatures $T^\ast$, which characterize the different levels as well as the kinks can be written as $T^\ast\approx\hbar v_{F}/2\pi k_{B}nL$, or in the equivalent form $\xi_{N}(T^\ast)=nL$, with $n=1,2,4$. The reentrant paramagnetic susceptibility then is \begin{equation} \label{eq1} \chi_{\mathrm para}=A_{n} \exp{\left[-\frac{nL}{\xi_{N}(T)}\right]},\quad {\mathrm with}\, n=1,2,4\, . \end{equation} This characteristic behavior is not obtained by present theories. The correct theory of the paramagnetic reentrant effect should describe the suscptibility in accordance with Eq.\ \ref{eq1}. For a theoretical understanding of the origin of the reentrant effect, it is important to investigate the susceptibility under fields, as well as the magnetization. In the following we discuss isothermal ac susceptibility measurements of sample 5AgNb, as a function of magnetic field [Figs.\ \ref{X(H,[T])nh} and \ref{XacMdc(H)nh}(a)]. The isothermal susceptibility $\chi_{N}(H)$ shows nonlinear behavior in the entire field regime. At $7\,{\mathrm mK}$ two curves are shown, the first measurement directly after cooldown, and the second one \begin{figure} \includegraphics[width=0.96\linewidth]{XHT.eps} \caption{Isothermal ac-magnetic susceptibility $\chi_{N}(H)$, below $50\,{\mathrm mK}$. The arrows indicate the direction of field changes.} \protect\label{X(H,[T])nh} \end{figure} \begin{figure} \includegraphics[width=0.96\linewidth]{XacMdcH.eps} \caption{(a) Nonlinear ac susceptibility $\chi_{N}(H)$ and (b) isothermal dc magnetization $M_{dc}(H)$ in full field cycles, starting with positive values; the black lines guide the eye.} \protect\label{XacMdc(H)nh} \end{figure} \noindent $\sim 10^6\,{\mathrm s}$ after the first one. The lower of the two curves at $7\,{\mathrm mK}$ shows pronounced hysteresis of the reentrant part at fields below $20\,{\mathrm Oe}$. Near 20\,Oe the specimen screens the magnetic flux most effectively, before the well known magnetic breakdown transition occurs \cite{bmueller1,journlowtempphys,belzig:96,fauchere}. For this sample, the transition is nearly temperature independent for $T<50\,{\mathrm mK}$. However, the minimum susceptibility before the breakdown is far from complete screening. At $T=7\,{\mathrm mK}$, it reaches only 30\,$\%$ of $-1/4\pi$, due to the paramagnetic contribution. For very small dc magnetic fields, the magnetic susceptibility grows rather steeply. At higher fields, that increase slows down, the susceptibility reaching its maximum value at about $2.5\,{\mathrm Oe}$, before turning towards less paramagnetic values. As the temperature is increased, the curves show a less pronounced maximum at about $2.5\,{\mathrm Oe}$ and reduced hysteresis. The detailed behavior at very low fields differs from the results of similar measurements, performed on a bigger wire bundle, reported in Ref.\ \cite{visaniprl}. Possibly, the different geometrical arrangement of the wires in the bundles affect the average susceptibility. Indeed, measurements on a single wire are desirable. Cycling the magnetic field in both directions at $7\,{\mathrm mK}$ [Fig.\ \ref{XacMdc(H)nh}(a)], we observe two low-field peaks of $\chi_{N}(H)$, which are displaced symmetrically from zero (residual field $<2\,{\mathrm mOe}$). However, the curves are not symmetric below $20\,{\mathrm Oe}$, displaying a reduction of hysteresis after each half-cycle. Furthermore, after cooldown from above $50\,{\mathrm mK}$, cycling the field at $7\,{\mathrm mK}$, and waiting for $\sim 10^6\,\mathrm{s}$, the low-field peak at about $2.5\,{\mathrm Oe}$ grows up. This can be observed in the upper curves in Figs.\ \ref{X(H,[T])nh} and \ref{XacMdc(H)nh}(a). After cycling the field and waiting, the whole system crosses to a more stable state, with more pronounced paramagnetic susceptibility. Hysteresis and nonlinearity are also observed in dc magnetization curves, e.g. shown for a full cycle at $7\,{\mathrm mK}$ in Fig.\ \ref{XacMdc(H)nh}(b). For the second half-cycle, nonlinearity becomes less pronounced, in accordance with our findings in the susceptibility curves. The measured magnetization lies between two lines at fields below the breakdown transition. At low fields one observes clearly the deviation of the magnetization from the induced Meissner screening. At higher fields the curve asymptotically approaches the drawn line indicating linear Meissner-like behavior plus a constant \textit{paramagnetic} magnetization. The $H=0$ intersections of the lines suggest a field-independent magnetization $4\pi M_{0}\approx 1\,{\mathrm G}$. It is interesting to notice that a spontaneous magnetization ot the same order was found in the theoretical model of Fauch\'ere \textit{et al}. under the assumption of a negative average order parameter in N, $\Delta_{N}/k_{B}\approx-160\,{\mathrm mK}$ \cite{fauchere99}. Moreover, some features of the paramagnetic reentrant effect, such as nonlinearity, hysteresis, and dissipation, are in qualitative agreemnet with their results \cite{fauchere99}. At this point, a direct comparison of our experiment with theory \cite{bruder,fauchere99} is not possible. The magnitude of $\chi_{\mathrm para}$ of order 1, the characteristic dependence of the magnetic susceptibility on $T$ and $L$, as described by Eq.\ \ref{eq1}, as well as its field dependence remain to be obtained by theory. More theoretical work and also more experiments are necessary. In summary, the paramagnetic reentrance phenomenon in AgNb cylinders of high purity is a nonlinear effect of anomalously strong magnitude. It shows strong deviations from induced Meissner screening in the low-temperature--low-field corner of the $H$-$T$ phase diagram. It displays dissipation \cite{visaniprl}, hysteresis, and long time relaxation behavior. In the mesoscopic regime, the exponential temperature dependence of the magnetic susceptibility with characteristic temperature $T^\ast\approx\hbar v_{F}/2\pi k_{B}nL$ has the fingerprint of several levels of quantum coherence along integer multiples $n=1,2,4$ of the wire perimeter $L$. Paramagnetic reentrance has also been observed in other NS materials \cite{visaniprl}. In our most recent experiments we have found reentrant behavior in gold-niobium cylinders. This has to be viewed in the light of expected superconductivity in Au below $T_{c}\approx 200\,\mu{\mathrm K}$ \cite{hoyt}. In consideration of this, the origin of this puzzling paramagnetism in mesoscopic NS cylinders is still an open question. We wish to express our gratitude to R. Frassanito, M. Nider{\"o}st, and P. Visani for their invaluable contributions to earlier experimental work. We thank the group at the ULT facility in Bayreuth for their help and support. We acknowledge discussions with W. Belzig, G. Blatter, C. Bruder, A. Fauch\`ere, and Y. Imry. We acknowledge partial support from the ``Schweizerischer Nationalfonds zur F{\"o}rderung der Wissenschaftlichen Forschung'' and the ``Bundesamt f{\"u}r Bildung und Wissenschaft'' (EU Program ``Training and Mobility of Researchers'').
\section{Introduction} \newfont{\graf}{eufm10} \newcommand{\mbox{\graf h}}{\mbox{\graf h}} Let $X^3 (c)$ be a 3-dimensional space form of constant curvature $c=0$ or 1 and admitting a real Killing spinor with respect to some spin structure. Consider a compact, oriented and immersed surface $M^2 \subset X^3(c)$ with mean curvature $H$. The spin structure of $X^3 (c)$ induces a spin structure on $M^2$. Denote by $D$ the corresponding Dirac operator acting on spinor fields defined over the surface $M^2$. The first eigenvalue $\lambda_1^2 (D)$ of the operator $D^2$ and the first eigenvalue $\mu_1$ of the Schr\"odinger operator $\Delta + H^2 +c$ are related by the inequality \[ \lambda_1^2 (D) \le \mu_1 (\Delta + H^2 + c) . \] Equality holds if and only if the mean curvature $H$ is constant (see [1], [5]). Moreover, the Killing spinor defines a map $f \mapsto \Phi (f)$ of the space $L^2 (M^2)$ of functions into the space $L^2 (M^2; S)$ of spinors such that \[ ||D(\Phi (f)) ||^2_{L^2} = \langle \Delta f + H^2f+ c f,f \rangle_{L^2} . \] In particular, the mentioned inequality holds for all eigenvalues, i.e., \[ \lambda_k^2 (D) \le \mu_k (\Delta + H + c) . \] This inequality was used in order to estimate the first eigenvalue of the Dirac operator defined on special surfaces of Euclidean space (see [1]). On the other hand, in case we know $\lambda_1^2 (D)$, the inequality yields a lower bound for the spectrum of the Schr\"odinger operator $\Delta + H^2 + c$. For example, for any Riemannian metric $g$ on the 2-dimensional sphere $S^2$ the inequality \[ \lambda_1^2 (D) \ge \frac{4 \pi}{\mbox{vol} \, (S^2,g)} \] holds (see [2], [6]). Consequently, we obtain \[ \frac{4 \pi}{\mbox{vol} \, (M^2,g)} \le \mu_1 (\Delta + H^2) \] for any surface $M^2 \hookrightarrow {\Bbb R}^3$ of genus zero in Euclidean space ${\Bbb R}^3$. In this note we expose the described idea and, in particular, we estimate the spectrum of special periodic Schr\"odinger operators where the potential is given by the curvature $\kappa$ of a spherical curve. \\ \section{The one-dimensional case} First of all, let us consider the 1-dimensional case, i.e., a curve $\gamma$ of length $L$ in a two-dimensional space form $X^2 (c)$. Let $\Phi$ be a Killing spinor of length one on $X^2 (c)$: \[ \nabla_{T} {\Phi} = \frac{1}{2} c \cdot T \cdot \Phi . \] The restriction $\varphi = \Phi_{|\gamma}$ defines a pair of spinors on $\gamma$ and the spinor field $\psi= f \cdot \varphi$ satisfies the equation: \[ |D \psi |^2 = | \dot{f}|^2 +f^2 \left( \frac{c}{4} + \frac{1}{4} \kappa^2_g \right) , \] where $\kappa_g$ is the curvature of the curve $\gamma$ in $X^2(c)$. Therefore, we obtain \[ \lambda_k^2 (D) \le \mu_k \left( - \frac{d^2}{ds^2} + \frac{c}{4} + \frac{1}{4} \kappa^2_g \right) . \] Suppose now that the spin structure on $\gamma$ induced by the spin structure of $X^2 (c)$ is non-trivial. Then we have $\displaystyle \lambda_{k+1}^2 (D)= \frac{4 \pi^2}{L^2} (k + 1/2)^2$ (see [4]) and, in particular, we obtain \[ \frac{4 \pi^2}{L^2} \left( k + \frac{1}{2} \right)^2 \le \mu_{k+1} \left( - \frac{d^2}{ds^2} + \frac{c}{4} + \frac{1}{4} \kappa^2_g \right) . \] {\bf Theorem 1:} {\it Let $\gamma \subset {\Bbb R}^3$ be a plane or spherical curve and denote by $\kappa^2 = c + \kappa_g^2$ the square of its curvature. Suppose that the induced spin structure on $\gamma$ is non-trivial, i.e., the tangent vector field has an odd rotation number. Then the inequality \[ \frac{4 \pi^2}{L^2} \le \mu_1 \left( - 4 \frac{d^2}{ds^2} + \kappa^2 \right) \] holds, where $\mu_1$ is the first eigenvalue of the periodic Sturm-Liouville operator on the interval $[0,L]$. Moreover, equality occurs if and only if the curvature is constant.}\\ {\bf Remark:} No geometric lower bound for the Sturm-Liouville operator $ - 4 \frac{d^2}{ds^2} + \kappa^2$ with potential defined by the square of the curvature $\kappa (s)$ of a closed curve $\gamma$ in Euclidean space seems to be known. We conjecture that the estimate given in \mbox{Theorem 1} holds for any closed curve in ${\Bbb R}^3$. Let us compare this inequality with the well-known Fenchel-Milnor inequality \[ 2 \pi \le \oint\limits_{\gamma} \kappa . \] Thus, by the Cauchy-Schwarz inequality we obtain \[ \frac{4 \pi^2}{L^2} \le \frac{1}{L} \oint\limits_{\gamma} \kappa^2 . \] Moreover, using the test function $f \equiv 1$, we have \[ \mu_1 \left( - 4 \frac{d^2}{ds^2} + \kappa^2 \right) \le \frac{1}{L} \oint\limits_{\gamma} \kappa^2 . \] Suppose that $\gamma$ is a simple curve in ${\Bbb R}^3$ and denote by $ \rho$ the minimal number of generators of the fundamental group $\pi_1 ({\Bbb R}^3 \backslash \gamma)$. Then we have \[ 2 \pi \rho \le \oint\limits_{\gamma} \kappa . \] In the spirit of this remark one should be able to prove the stronger inequality \[ \frac{4 \pi^2}{L^2} \rho^2 \le \mu_1 \left( - 4 \frac{d^2}{ds^2} + \kappa^2 \right) \] in case of a simple curve in ${\Bbb R}^3$. \\ {\bf Examples:} We calculated the eigenvalue $\mu_1$ for some classical curves in ${\Bbb R}^3$: \begin{itemize} \item[a.)] {\it The lemniscate:} $x= \sin (t) , \quad y = \cos (t) \sin (t).$\\ \mbox{} \hspace{2.7cm} $4 \pi^2 /L^2 = 1.06193, \quad \mu_1 = 3.7315, \quad \displaystyle \frac{1}{L} \oint\limits_{\gamma} \kappa^2 = 4.36004.$\\ \item[b.)] {\it The trefoil:} $x= \sin (3t) \cos (t) , \quad y = \sin (3t) \sin (t).$\\ \mbox{} \hspace{1.8cm} $4 \pi^2 /L^2 =0.221, \quad \mu_1 = 5.21, \quad \displaystyle \frac{1}{L} \oint\limits_{\gamma} \kappa^2 = 8.16.$\\ \item[c.)] {\it Viviani's curve:} $x= 1+ \cos (t) , \quad y = \sin (2t) , \quad z= 2 \sin (t) .$\\ \mbox{} \hspace{2.6cm} $4 \pi^2 /L^2 =0.169071, \quad \mu_1 = 0.5335, \quad \displaystyle \frac{1}{L} \oint\limits_{\gamma} \kappa^2 = 0.567803.$\\ \item[d.)] {\it Torus knot:} $x= (8+3 \cos (5t)) \cos (2t) , \quad y = (8+ 3 \cos (5t)) \sin (2t), \\ \mbox{} \hspace{2.3cm} z= 5 \sin (5t).$\\ \mbox{} \hspace{2.0cm} $4 \pi^2 /L^2 =0.00146034, \quad \mu_1 = 0.03232, \quad \displaystyle \frac{1}{L} \oint\limits_{\gamma} \kappa^2 = 0.0333803.$\\ \item[e.)] {\it The spherical spiral:} $x= \cos (t) \cos (4t) , \quad y = \cos (t) \sin(4t), \quad z= \sin(t).$\\ \mbox{} \hspace{3.4cm} $4 \pi^2 /L^2 =0.127036, \quad \mu_1 = 1.744, \quad \displaystyle \frac{1}{L} \oint\limits_{\gamma} \kappa^2 = 4.93147.$\\ \end{itemize} \section{The two-dimensional Schr\"odinger operator} We generalize this inequality to the case of the two-dimensional periodic Schr\"odinger operator \[ P_{A,L} = - \left( 1+ \frac{A^2}{L^2} \right) \frac{\partial}{\partial t^2} - 4 \frac{\partial^2}{\partial s^2} - \frac{4 A}{L} \frac{\partial}{\partial t} \frac{\partial}{\partial s} + \kappa^2 (s) \] defined on $[0,2\pi] \times [0,L]$:\\ {\bf Theorem 2:} {\it Let $\gamma \subset S^2 \subset {\Bbb R}^3$ be a closed, simple curve of length $L$ bounding a region of area $A$, and denote by $\kappa$ its curvature. Then the spectrum of the two-dimensional periodic Schr\"odinger operator $P_{A,L}$ is bounded by } \[ \frac{4 \pi^2}{L^2} \le \mu_1 (P_{A,L}) . \] {\it Equality holds if and only if the curvature of $\gamma$ is constant.}\\ In general, let us consider a Riemannian manifold $(Y^n,g)$ of dimension $n$ as well as an $S^1$-principal fibre bundle $\pi : P \to Y^n$ over $Y^n$. Denote by $\vec{V}$ the vertical vector field on $P$ induced by the action of the group $S^1$ on the total space $P$, i.e., \[ \vec{V} (p) = \frac{d}{dt} \left( p \cdot e^{it} \right)_{t=0} \quad , \quad p \in P . \] A connection $Z$ in the bundle $P$ defines a decomposition of the tangent bundle $T(P)=T^v (P) \oplus T^h (P)$ into its vertical and horizontal subspace. We introduce a Riemannian metric $g^*$ on the total space $P$, requiring that \begin{itemize} \item[a)] $g^* (\vec{V} , \vec{V})=1$,\\ \item[b)] $ g^* (T^v, T^h)=0$,\\ \item[c)] the differential $d \pi$ maps $T^h (P)$ isometrically onto $T(Y^n)$. \end{itemize} \bigskip A closed curve $\gamma : [0,L] \to Y^n$ of length $L$ defines a torus $H(\gamma) := \pi^{-1} (\gamma) \subset P$ and we want to study the isometry class of this flat torus in $P$. Let $\alpha = e^{i \Theta} \in S^1 $ be the holonomy of the connection $Z$ along the closed curve $\gamma$. Consider a horizontal lift $\hat{\gamma} :[0,L] \to P$ of the curve $\gamma$. Then \[ \hat{\gamma} (L) = \hat{\gamma} (0) e^{i \Theta} \] holds. Consequently, the formula \[ \Phi (t,s) = \hat{\gamma} (s) e^{-i \Theta s / L } e^{it} \] defines a parametrization $\Phi : [0,2 \pi] \times [0,L] \to H(\gamma)$ of the torus $H(\gamma)$. Since \[ \frac{\partial \Phi}{\partial t} = \vec{V} \quad , \quad \frac{\partial \Phi}{\partial s} = dR_{e^{it} e^{-i \Theta s/ L}} (\dot{\hat{\gamma}} (s)) - \frac{\Theta}{L} \vec{V} , \] we obtain \[ g^* \left( \frac{\partial \phi}{\partial t} , \frac{\partial \phi}{\partial t} \right) = 1 , \quad g^* \left( \frac{\partial \phi}{\partial t} , \frac{\partial \phi}{\partial s} \right) = - \frac{\Theta}{L} , \quad g^* \left( \frac{\partial \phi}{\partial s} , \frac{\partial \phi}{\partial s} \right) = 1 + \frac{\Theta^2}{L^2} , \] i.e., the torus $H(\gamma)$ is isometric to the flat torus $({\Bbb R}^2 / \Gamma_o , g^*)$, where $\Gamma_o$ is the orthogonal lattice $\Gamma_o = 2 \pi \cdot {\Bbb Z} \oplus L \cdot {\Bbb Z}$ and the metric $g^*$ has the non-diagonal form \[ g^* = \left( \begin{array}{cc} 1 & - \frac{\Theta}{L} \\ \\- \frac{\Theta}{L} & 1+ \frac{\Theta^2}{L^2} \end{array} \right) . \] Using the transformation \[ x = - \frac{\Theta}{L} s + t \quad , \quad y=s , \] we see that $H(\gamma)$ is isometric to the flat torus $({\Bbb R}^2 / \Gamma , \, \, \, dx^2 + dy^2 )$, where the lattice $\Gamma$ is generated by the two vectors \[ v_1 = \left( \begin{array}{c} 2 \pi\\0 \end{array} \right) \quad , \quad v_2 = \left( \begin{array}{c} \Theta \\ L \end{array} \right) . \] In case the closed curve $\gamma : [0,L] \to Y^n$ is the oriented boundary of an oriented compact surface $M^2 \subset Y^n$, we can calculate the holonomy $\alpha = e^{i \Theta}$ along the curve $\gamma$. Indeed, let $\Omega^Z$ be the curvature form of the connection $Z$. $\Omega^Z$ is a 2-form defined on the manifold $Y^n$ with values in the Lie algebra of the group $S^1$, i.e., with values in $i \cdot {\Bbb R}^1$. The parameter $\Theta$ is given by the integral \[ \Theta = i \int\limits_{M^2} \Omega^Z . \] Let us consider the Hopf fibration $\pi: S^3 \to S^2$, where \[ S^3 = \{ (z_1, z_2) \in {\Bbb C}^2 : |z_1|^2 + |z_2|^2 =1 \} \] is the 3-dimensional sphere of radius 1. The connection $Z$ is given by the formula \[ Z= \frac{1}{2} \{ \bar{z}_1 dz_1 - z_1 d \bar{z}_1 + \bar{z}_2 dz_2 - z_2 d \bar{z}_2 \} \] and its curvature form ($\omega = z_1 / z_2$) \[ \Omega^Z = - \frac{d \omega \wedge d \bar{\omega}}{(1+|\omega|^2)^2} = - \frac{i}{2} dS^2 \] essentially coincides with one half of the volume form of the unit sphere $S^2$ of radius 1. However, the differential $d \pi : T^h (S^3) \to T(S^2)$ multiplies the length of a vector by two, i.e., the Hopf fibration is a Riemannian submersion in the sense described before if we fix the metric of the sphere $S^2 (\frac{1}{2})= \{ x \in {\Bbb R}^3 : |x| = \frac{1}{2} \}$ on $S^2$. Consequently, in case of a closed simple curve $\gamma \subset S^2$ bounding a region of area $A$, the Hopf torus $H (\gamma) \subset S^3$ is isometric to the flat torus ${\Bbb R}^2 / \Gamma$ and the lattice $\Gamma$ is generated by the two vectors \[ v_1 = \left( \begin{array}{c} 2 \pi\\0 \end{array} \right) \quad , \quad v_2 = \left( \begin{array}{c} {A}/{2} \\ {L}/{2} \end{array} \right) . \] The mean curvature $H$ of the torus $H(\gamma) \subset S^3$ coincides with the geodesic curvature $\kappa_g$ of the curve $\gamma \subset S^2 \subset {\Bbb R}^3$ (see [7], [8]). We apply now the inequality \[ \lambda_1^2 (D) \le \mu_1 ( \Delta + H^2 + 1) \] to the Hopf torus $H (\gamma) \subset S^3$. Then we obtain the estimate \[ \lambda_1^2 (D) \le \mu_1 \left( P_{A,L} \right) , \] where $D$ is the Dirac operator on the flat torus ${\Bbb R}^2 / \Gamma$ with respect to the induced spin structure. All spin structures of a 2-dimensional torus are classified by pairs $(\varepsilon_1, \varepsilon_2)$ of numbers $\varepsilon_i =0,1$. If $\gamma$ is a simple curve in $S^2$, the induced spin structure on the Hopf torus $H(\gamma)$ is non-trivial and given by the pairs $(\varepsilon_1, \varepsilon_2) =(0,1)$. The spectrum of the Dirac operator for all flat tori is well-known (see [4]): The dual lattice $\Gamma^*$ is generated by \[ v_1^* = \left( \begin{array}{c} \displaystyle \frac{1}{2 \pi} \\ \\ \displaystyle - \frac{A}{2 \pi L} \end{array} \right) \quad , \quad v_2^* = \left( \begin{array}{c} 0 \\ \\ \displaystyle \frac{2}{L} \end{array} \right) \] and the eigenvalues of $D^2$ are given by \begin{eqnarray*} \lambda^2 (k,l) &=& 4 \pi^2 \Big| \Big| \, \, k v_1^* + \left(l + \frac{1}{2}\right) v_2^* \Big| \Big|^2 =\\ &=& k^2 + \frac{4 \pi^2}{L^2} \left((2l+1) - k \frac{A}{2 \pi} \right)^2 . \end{eqnarray*} We minimize $\lambda^2 (k,l)$ on the integral lattice ${\Bbb Z}^2$. The isoperimetric inequality $4 \pi A - A^2 \le L^2$ and $A \le \mbox{vol} \, (S^2) = 4 \pi$ yield the result that $\lambda^2 (k,l)$ attends its minimum at $(k,l) = (0,1)$, i.e., \[ \frac{4 \pi^2}{L^2} \le \lambda^2 (k,l) . \] {\bf Remark 1:} We replace the Hopf fibration by the $S^1$-principal fibre bundle of Chern class $m \ge 0$. The corresponding total space is the Lens space $L(m,1)$ and we have the commutative diagram \[ \begin{diagram} \node{S^3} \arrow{se,b}{\pi} \arrow[2]{e} \node{} \node{L(m,1)} \arrow{sw,b}{\pi_m}\\ \node{} \node{S^2} \node{} \end{diagram} \] Let $H_m (\gamma) \subset L(m,1)$ be the Hopf torus. $H_m (\gamma)$ is isometric to ${\Bbb R}^2 / \Gamma_m$, where the lattice $\Gamma_m$ is generated by the vectors \[ v_1 = \left( \begin{array}{c} 2 \pi / m \\ 0 \end{array} \right) \quad , \quad v_2 = \left( \begin{array}{c} A/2 \\ L/2 \end{array} \right) . \] Moreover, the Lens space $L(m,1)$ admits a unique spin structure with a Killing spinor (see [3]). Even in case of $m \not= 1$, the induced spin structure on $H_m (\gamma)$ is described by the parameters $(\varepsilon_1 , \varepsilon_2)=(0,1)$. Since the local geometry of $H_m (\gamma)$ in $L(m,1)$ essentially coincides with the geometry of $H(\gamma)$ in $S^3$, we obtain the inequality \[ \frac{4 \pi^2}{L^2} = \min\limits_{(k,l)} \left\{ k^2m^2 + \frac{4 \pi^2}{L^2} \left( (2l+1) - k \frac{mA}{2 \pi} \right)^2 \right\} \le \mu_1 \left( - 4 \frac{d^2}{ds^2} + \kappa^2 \right) . \] Consequently, the investigation of the two-dimensional Schr\"odinger operator in case of $m \not= 1$ yields the same result for the Sturm-Liouville operator as above. \\ {\bf Remark 2:} Suppose now that equality holds for some curve $\gamma \subset S^2$. We consider the corres\-ponding Hopf torus $H(\gamma) \subset S^3$ and then we obtain \[ \lambda_1^2 (D) = \mu_1 ( \Delta + H^2 +1) . \] Therefore, the mean curvature $H= \kappa$ is constant, i.e., $\gamma$ is a curve on $S^2$ of constant curvature $\kappa$. Consequently, $\gamma$ is a circle in a 2-dimensional plane. Denote by $r$ its radius. Then \[ \kappa^2 = \frac{1}{r^2} \quad , \quad L= 2 \pi r \quad , \quad A=2 \pi (1 - \sqrt{1-r^2}) , \] and the inequality \[ \frac{4 \pi^2}{L^2} \le \kappa^2 \] is an equality for all $r \not= 0$. \\
\section{Introduction} If a body moves with a constant velocity, then, as is well known, the body is relativistically contracted in the direction of motion, whereas its length in the normal direction is unchanged. A naive generalization to a rotating disc leads to the conclusion that the circumference of the disc is contracted, whereas the radius of the disc is unchanged. This paradox is known as the Ehrenfest paradox. Obviously, the paradox is a consequence of the application of the constant-velocity result to a system with a nonconstant velocity. The standard resolution (see \cite{gron1}, \cite{gron2} and references therein) of the Ehrenfest paradox is as follows: One introduces the coordinates of the rotating frame $S'$ \begin{equation}\label{eq1} \varphi'=\varphi -\omega t \; , \;\;\;\; r'=r \; , \;\;\;\; z'=z \; , \;\;\;\; t'=t \; , \end{equation} where $\varphi$, $r$, $z$, $t$ are cylindrical coordinates of the inertial frame $S$ and $\omega$ is the angular velocity. The metric in $S'$ is given by \begin{equation}\label{eq2} ds^2=(c^2-\omega^2 r'^2)dt'^2 -2\omega r'^2 \, d\varphi' dt' -dr'^2 -r'^2 \, d\varphi'^2 -dz'^2 \; . \end{equation} It is generally accepted that the space line element should be calculated by the formula \cite{land} \begin{equation}\label{eq3} dl'^2=\gamma'_{ij}dx'^i dx'^j \; , \;\;\; i,j=1,2,3 \; , \end{equation} where \begin{equation}\label{eq4} \gamma'_{ij}=\frac{g'_{0i}g'_{0j}}{g'_{00}}-g'_{ij} \; . \end{equation} This leads to the circumference of the disc \begin{equation}\label{eq5} L'=\int_{0}^{2\pi} \frac{r' d\varphi'}{\sqrt{1-\omega^2 r'^2/c^2}} =\frac{2\pi r'}{\sqrt{1-\omega^2 r'^2/c^2}} \equiv\gamma(r')2\pi r' \; . \end{equation} The circumference of the same disc as seen from $S$ is $L=2\pi r=2\pi r'$. If the disc is constrained to have the same radius $r$ as the same disc when it does not rotate, then $L$ is not changed by the rotation, but the proper circumference $L'$ is larger than the proper circumference of the nonrotating disc. This implies that there are tensile stresses in the rotating disc. However, there is something wrong with this standard resolution of the Ehrenfest paradox. Consider a slightly simpler situation; a rotating ring in a rigid nonrotating circular gutter with the radius $r=r'$. The statement that (\ref{eq5}) represents the proper circumference implies that the {\em proper} frame of the rotating ring is given by (\ref{eq1}). This means that an observer on the ring sees that the circumference is $L'=\gamma L$. The circumference of the gutter seen by him cannot be different from the circumference of the ring seen by him, so the observer on the ring sees that the circumference of the relatively moving gutter is {\em larger} than the proper circumference of the gutter, whereas we expect that he should see that it is smaller. This leads to another paradox. It cannot be resolved by saying that the observer on the ring accelerates, because one can consider a limit $r\rightarrow\infty$, $\omega\rightarrow 0$, $r\omega\equiv u=$constant, which implies that the acceleration $a=r\omega^2$ becomes zero, whereas the paradox remains. Before explaining how we resolve this paradox, we give some general notes on the physical meaning of various coordinate frames in the theory of relativity. In practice, one usually uses the coordinates that simplify the technicalities of the physical problem considered. For example, when one describes physical effects in a rigid body, it may be convenient to use a comoving coordinate frame, i.e., a frame in which all particles of the rigid body have constant spatial coordinates. The coordinates of $S'$ in (\ref{eq1}) may be interpreted in this way. However, {\em the choice of the coordinate frame is more than a matter of convenience}. The main lesson we have learned from Lorentz coordinate frames is the fact that what an observer observes (time intervals, space intervals, components of a tensor, ...) depends on how the observer moves. The main purpose of theoretical physics is to predict what will be {\em observed} under given circumstances. Therefore, unless stated otherwise, in this paper {\em by a coordinate frame we understand a coordinate frame that is inherent to an observer}, not to a set of physical particles. Our criticism of some earlier treatments originates from such an interpretation of coordinate frames. To avoid a possible misunderstanding, we note that coordinate frames do not necessarily need to be interpreted in this way, in which case our criticism does not apply. We resolve the paradox by recognizing that, according to our interpretation, the frame defined by (\ref{eq1}) is the proper frame {\em only} of the observer at $r=r'=0$. This observer has no velocity relative to $S$, so the corresponding coordinate transformation (\ref{eq1}) does not depend on any velocity. As will become clear from the discussion of Section 2, the frame defined by (\ref{eq1}) is actually the Fermi frame of an observer who rotates, but has no velocity with respect to the frame $S$. Observers at different positions on the rotating disc have different velocities, so one has to use a different coordinate transformation for each of them. In other words, {\em although there is no relative motion among different points on a rotating disc, each point belongs to a different noninertial frame.} This is not strange to those who are familiar with the theory of Fermi coordinates \cite{mtw}, \cite{synge}, but it seems that many relativity-theorists are not. Note also that since we do not interpret the coordinates of $S'$ in (\ref{eq1}) as something inherent to the disc as a whole, $r'$ can be arbitrarily large in (\ref{eq2}), although there is a coordinate singularity at $r'=c/\omega$. It resembles the Schwarzschild singularity of a black hole, where the radial coordinate is not restricted to be larger than the Schwarzschild radius. However, to avoid a possible misunderstanding, note that the coordinate singularity in (\ref{eq2}) does not correspond to an event horizon, because a rotating observer at $r'=0$ {\em can} receive information from $r'\geq c/\omega$. There is also another paradox connected with the standard approach to rotating frames. Let us consider how the nonrotating gutter looks like to a rotating observer in the center. His proper frame {\em is} given by (\ref{eq1}). If (\ref{eq3}) is the correct definition of the space line element, then he should see that the circumference of the gutter is larger than the proper circumference of the gutter by a factor $\gamma(r')$. However, $\omega r'/c$ can be arbitrarily large, so $\gamma(r')$ can be not only arbitrarily large, but also even imaginary. On the other hand, we know from everyday experience that the apparent velocity $\omega r'$ of stars, due to our rotation, can exceed the velocity of light, but we see neither a contraction, nor an elongation of the stars observed. We resolve this paradox by examining the assumptions under which formula (\ref{eq3}) is obtained. We find that this formula should be used with great care and show that it is not applicable in our case. The correct definition of the space line element depends on how it is measured, and we find that, in our case, $\gamma'_{ij}$ should be replaced by $-g'_{ij}$ in (\ref{eq3}). It is fair to note that there are also some other ``nonstandard" approaches to the Ehrenfest paradox (see \cite{tart2}, \cite{kla}, and references therein), but none of these approaches is similar to ours. In particular, the crucial fact that each point of the rotating ring belongs to a different frame has not been taken into account in any of these approaches. Formula (\ref{eq3}) has already been criticized \cite{kla}, but our criticism of (\ref{eq3}) is quite different and more general. The paper is organized as follows: In Section 2 we find the correct coordinate transformation that leads to the frame of an observer moving in flat spacetime. In Section 3 we explain why (\ref{eq3}) is not always a correct definition of a space line element and show that in a frame that corresponds to an observer in flat spacetime it is more appropriate to calculate the space line element by $-g'_{ij}$. We also make some general remarks on the physical meaning of general coordinate transformations. In Section 4 we study the relativistic contraction as seen by various observers and resolve the Ehrenfest paradox. In Section 5 we study the rate of clocks at various positions, as seen by various observers. In Section 6 we discuss the velocity of light as seen by various observers. In Section 7 we discuss our results, resolve some additional physical problems, and give some generalizations. Section 8 is devoted to concluding remarks, where the relevance of our results to general relativity is emphasized. \section{The frame of an observer moving in flat spacetime} The generalized Lorentz transformations for a local Fermi frame of an observer that has arbitrary time-dependent velocity and angular velocity in flat spacetime are found in \cite{nels}. We present the final results, using slightly different notation. Let $S$ be an inertial frame and let $S'$ be the frame of the observer whose velocity and angular velocity are $u^i(t')$ and $\omega^i(t')$, respectively, as seen by an observer in $S$. The coordinate transformation between these two frames is given by \begin{equation}\label{er1} x^i =-A_{j}^{\; i}(t')x'^j +\int_{0}^{t'} \gamma(t')u^i(t')\, dt' + \frac{1}{\mbox{\bf{u}}^2(t')} [\gamma(t')-1][u^k(t')A_{jk}(t')x'^j]u^i(t') \; , \end{equation} \begin{equation}\label{er2} t=\int_{0}^{t'} \gamma(t')\, dt' + \frac{1}{c^2}\gamma(t')[u^k(t')A_{jk}(t')x'^j] \; , \end{equation} where $\gamma(t')=1/\sqrt{1-\mbox{\bf{u}}^2(t')/c^2}$ and $A_{ji}(t')=-A_{j}^{\; i}(t')$ is the rotation matrix evaluated at $\mbox{\bf{x}}'=0$. The rotation matrix satisfies the differential equation \begin{equation}\label{er4} \frac{d A_{ij}}{dt}=-A_{i}^{\; k}\omega_{kj} \; , \end{equation} where $\omega_{ik}=\varepsilon_{ikl}\omega^{l}$, $\varepsilon_{123}=1$. The metric tensor in $S'$ is \begin{eqnarray}\label{metric} & g'_{ij}=-\delta_{ij} \; , \;\;\;\;\; g'_{0j}=-(\mbox{\boldmath $\omega$}'\times\mbox{\bf{x}}')_j \; , & \nonumber \\ & g'_{00}=c^2 \left( 1+\displaystyle\frac{\mbox{\bf{a}}'\cdot\mbox{\bf{x}}'}{c^2} \right)^2 -(\mbox{\boldmath $\omega$}'\times\mbox{\bf{x}}')^2 \; , & \end{eqnarray} where \begin{equation} \omega'^i =\gamma (\omega^i -\Omega^i) \; , \;\;\;\;\; a'^i =\gamma^2 \left[a^i +\frac{1}{\mbox{\bf{u}}^2}(\gamma -1)(\mbox{\bf{u}}\cdot\mbox{\bf{a}})u^i\right] \; , \end{equation} $\Omega^{i}$ is the time-dependent Thomas precession frequency \begin{equation} \Omega_{i}=\frac{1}{2\mbox{\bf{u}}^2}(\gamma -1)\varepsilon_{ikj} (u^k a^j -u^j a^k) \; , \end{equation} and $a^i=du^i/dt$ is the time-dependent acceleration. The transformations (\ref{er1})-(\ref{er2}) are chosen such that the space origins of $S$ and $S'$ coincide for $t=t'=0$. If $\mbox{\bf{u}}$ is time independent and $\mbox{\boldmath $\omega$}=0$, then (\ref{er1})-(\ref{er2}) reduce to the well-known ordinary Lorentz boosts. If $\mbox{\bf{u}}=0$ and $\mbox{\boldmath $\omega$}$ is time independent, then (\ref{er1})-(\ref{er2}) reduce to (\ref{eq1}). It is important to emphasize that $\mbox{\bf{u}}(t')$ is the velocity of the {\em space origin} $\mbox{\bf{x}}'=0$ of $S'$. If $S'$ is a rotating frame, then other space points of $S'$ have a different velocity. (Remind that rotation is {\em not} a motion along a circle, but rather a change of orientation of the axes with respect to an inertial frame.) Therefore, in general, $S'$ is the proper frame {\em only} of the observer at $\mbox{\bf{x}}'=0$. Note also that $g'_{\mu\nu}=\eta_{\mu\nu}$ only at $\mbox{\bf{x}}'=0$, which is another confirmation that $S'$ is the frame of the observer at $\mbox{\bf{x}}'=0$ only. The metric (\ref{metric}) is also consistent with a more general theory of Fermi coordinates \cite{mtw}, which are coordinates of an observer arbitrarily moving in curved spacetime, and also have the property that $g_{\mu\nu}=\eta_{\mu\nu}$ at the space origin, i.e., at the position of the observer. Note also that if $\mbox{\bf{a}}'$ and $\mbox{\boldmath $\omega$}'$ vanish, then (\ref{metric}) is a metric of an inertial frame and is equal to $\eta_{\mu\nu}$ everywhere, so, in this case, $S'$ can be considered as a frame of an observer at {\em arbitrary} constant $\mbox{\bf{x}}'$. It is interesting to note that the geometrical construction of Fermi coordinates is well established \cite{mtw}, \cite{synge}, but no analog of (\ref{er1})-(\ref{er2}) is known for curved spacetime. The transformations (\ref{er1})-(\ref{er2}) are obtained by summation of infinitesimal Lorentz transformations (and rotations). It is not so easy to find an analog of Lorentz transformations in curved spacetime, because they correspond to the coordinate transformation between Fermi frames of two different free-falling observers. We can, however, write the transformations (\ref{er1})-(\ref{er2}) in a more elegant form, which could be illuminating for a generalization to curved spacetime. Let \begin{equation}\label{el1} x^{\mu}=f^{\mu}(t',\mbox{\bf{x}}';\mbox{\bf{u}}) \end{equation} denote the ordinary Lorentz transformations, i.e., the transformations between two inertial frames specified by the relative velocity $\mbox{\bf{u}}$, which can be considered as the relative velocity between two inertial (free-falling) observers at the instant when they have the same position. The differential of (\ref{el1}) is \begin{equation}\label{el2} dx^{\mu}=f^{\mu}_{\; ,\nu} (t',\mbox{\bf{x}}';\mbox{\bf{u}}) dx^{\nu} \; . \end{equation} The transition to a noninertial frame introduces a time-dependent velocity: $\mbox{\bf{u}} \rightarrow \mbox{\bf{u}}(t')$. The transformations (\ref{er1})-(\ref{er2}) may be obtained by integrating (\ref{el2}) in the following way: \begin{equation}\label{el3} x^{\mu}=\int_{0}^{t'}f^{\mu}_{\; ,0} (t',0;\mbox{\bf{u}}(t')) dt' + \int_{C} f^{\mu}_{\; ,i} (t',\mbox{\bf{x}}';\mbox{\bf{u}}(t')) dx'^{i} \; , \end{equation} where $C$ is an arbitrary curve with constant $t'$, starting from $0$ and ending at $-A_{j}^{\; i}(t')x'^{j}$. The subintegral function in the second term of (\ref{el3}) is a total derivative, so this term does not depend on the curve $C$ and can be easily integrated. The time derivative in the first term is taken with $\mbox{\bf{u}}(t')$ kept fixed, so $f^{\mu}_{\; ,0}$ in this term is not a total derivative. Let us now apply the general formalism described in this section to a uniformly rotating ring. We assume that the ring is put in a rigid nonrotating circular gutter with the radius $R$, which provides that the radius of the rotating ring is the same as the radius of the same ring when it does not rotate, and is equal to $R$, as seen by an observer in $S$. This allows us not to worry about the complicated dynamical forces that tend to change the radius of the ring as seen by the observer in $S$, and pay all our attention to the kinematic effects resulting from the transformations (\ref{er1})-(\ref{er2}). The ring can be considered as a series of independent short rods, uniformly distributed along the gutter. (By a short rod we understand a rod with a length much shorter than $R$.) We assume that the gutter is placed at the $z=0$ plane. We put the space origin of $S$ at a fixed point on the gutter, such that the $y$-axis is tangential to the gutter and the $x$-axis is perpendicular to the gutter at $\mbox{\bf{x}}=0$. (In the rest of this section, as well as in Sections 4 and 5, $\mbox{\bf{x}}\equiv ( x,y)$ and the $z$-coordinate is suppressed.) We study a single short rod initially placed at $\mbox{\bf{x}}=0$ and uniformly moving along the gutter in the counterclockwise direction. (This mimics a uniform motion of an electron in a synchrotron). The gutter causes a torque that provides that the rod is always directed tangentially to the gutter. Therefore, $\omega=u/R$, where $u=\sqrt{\mbox{\bf{u}}^2}$ is time independent. Now, $\gamma=1/\sqrt{1-\omega^2 R^2/c^2}$ is also time independent. Since a clock in $S'$ is at $\mbox{\bf{x}}'=0$, the clock rate between a clock in $S$ and a clock in $S'$ is given by $t=\gamma t'$, as seen by an observer in $S$. We assume that, initially, the axes $x'$, $y'$ are parallel to the axes $x$, $y$, respectively. Therefore the velocity \begin{equation} \mbox{\bf{u}}(t')=\omega R (-\sin \gamma\omega t', \cos \gamma\omega t') \end{equation} is always in the $y'$-direction and the solution of (\ref{er4}) is \begin{equation} A_{ij}(t')=\left( \begin{array}{cc} \cos \gamma\omega t' & \sin \gamma\omega t' \\ -\sin \gamma\omega t' & \cos \gamma\omega t' \end{array} \right) \; . \end{equation} The transformations (\ref{er1})-(\ref{er2}) become \begin{equation}\label{er1'} \left( \begin{array}{c} x \\ y \end{array} \right)= \left( \begin{array}{cc} \cos \gamma\omega t' & -\gamma\sin \gamma\omega t' \\ \sin \gamma\omega t' & \gamma\cos \gamma\omega t' \end{array} \right) \left( \begin{array}{c} x' \\ y' \end{array} \right) +R \left( \begin{array}{c} \cos \gamma\omega t' -1 \\ \sin \gamma\omega t' \end{array} \right) \; , \end{equation} \begin{equation}\label{er2'} t=\gamma t' +\frac{\gamma}{c^2}\omega R y' \; . \end{equation} In particular, at $t'=0$ these transformations become \begin{equation}\label{t=0} x=x' \; , \;\;\;\; y=\gamma y' \; , \;\;\;\; t=\frac{\gamma u}{c^2}y' \; , \end{equation} which coincide with the ordinary Lorentz boost at $t'=0$ for the velocity in the $y$-direction. \section{General coordinate transformations and the space line element in a non-time-orthogonal frame} A non-time-orthogonal frame is a frame in which $g'_{0i}$ is different from zero. It is generally accepted that the space line element in such a frame is given by (\ref{eq3}). However, if we assume that this formula can be applied to calculate the space distance as seen by a local observer, then, as we have found in Section 1, Eq. (\ref{eq3}) leads to an imaginary length of a distant unaccelerated object as seen by a rotating observer. In order to resolve this puzzle, we examine the assumptions under which formula (\ref{eq3}) is derived. In \cite{land}, formula (\ref{eq3}) is derived by assuming that the space distance between two points is measured by measuring the time $\Delta t'$ that light needs to travel from point $A$ to point $B$ and then back to point $A$. It is also assumed that the time is measured by a clock that does not change its position $x'^i$. The definition of the space distance $l'=c\,\Delta t'/2$ leads to (\ref{eq3}). In order to perform the described measurement in a rotating frame, the clock must be positioned at point $A$. However, according to our interpretation of (\ref{eq1}), this point can be faraway from the center of the rotation, so the required velocity of point $A$ can exceed $c$, as seen in $S$. Therefore, in general, such a measurement cannot be performed. In practice, we measure space distances between distant objects in a completely different way, namely, by measuring the angles under which we see the objects. (We assume that we know the radial distance of these objects from us. The radial distance is not problematic in the theoretical sense, because $g'_{0r}=0$ in (\ref{eq2})). Our rotation does not influence this angle. Therefore, the apparent velocity of distant objects can exceed the velocity of light owing to our rotation, but a pure rotation (without velocity) will not lead to relativistic contraction, nor to elongation. The effect is that, in a rotating frame, it is more appropriate to calculate the space line element as \begin{equation}\label{dl} dl'^2=-g'_{ij}dx'^{i}dx'^{j} \; , \end{equation} despite the fact that $g'_{0i}$ is different from zero. This formula should be used to calculate the space distance between two arbitrary points which have the same $t'$ coordinate, no matter how far these points are from the observer at $x'^i =0$. Of course, if these points are end points of a body, then, in general, the distance calculated in this way will not be equal to the proper length of the body, but merely to the length seen by the observer. Formula (\ref{dl}) is also correct for frames that are both accelerated and rotating, defined by (\ref{er1})-(\ref{er2}). To clarify the meaning of formula (\ref{eq3}) completely, note that in \cite{mol} this formula is derived in a completely different way, without referring to any particular method of measurement. However, what is actually derived in \cite{mol} is the fact that the quantity (\ref{eq3}) does not change under coordinate transformations of the form \begin{equation}\label{sameframe} t''=f^0 (t',x'^1,x'^2,x'^3) \; , \;\;\;\; x''^i=f^i (x'^1,x'^2,x'^3) \; . \end{equation} We refer to such transformations as {\em internal transformations}. Obviously, (\ref{eq1}) is not an internal transformation. Regular internal transformations form a subgroup of the group of all regular coordinate transformations. Note that the invariant quantity $ds^2=g_{\mu\nu}dx^{\mu}dx^{\nu}$ can always be written as \begin{equation}\label{logun} ds^2=d\eta^2 -\gamma_{ij}dx^i dx^j \; , \end{equation} where \begin{equation}\label{logun2} d\eta^2 =\left[ \frac{g_{0\mu}dx^{\mu}}{\sqrt{g_{00}}} \right]^2 \; , \end{equation} so $d\eta^2$ also does not change under internal transformations. The quantity $d\eta^2$ is nothing else but a time line element \cite{land}, defined by a measuring procedure similar to the measuring procedure used to define the space line element (\ref{eq3}). Let us illustrate the power of (\ref{eq3}), (\ref{sameframe}), and (\ref{logun2}) on the example that has already been discussed at some length in \cite{mol}. The Galilei transformation $t''=t$, $x''=x-ut$ can also serve as a correct coordinate transformation needed to describe the relativistic effects related to a frame moving with a constant velocity $u$. The metric in these coordinates is given by \begin{equation}\label{mol1} ds^2=c^2(1-u^2/c^2)dt''^2 -2udx''dt''-dx''^2 \; , \end{equation} where it has been assumed that the metric of $S$ is given by $ds^2=c^2\, dt^2 -dx^2$. From (\ref{eq3}) and $dt=0$ one can obtain the relativistic contraction $dl=dx=dl''/\gamma$, where $\gamma =1/\sqrt{1-u^2/c^2}$. Similarly, from (\ref{logun2}) and $dx''=0$ one can obtain $dt=\gamma d\eta''$. The frame $S''$ is physically equivalent to the frame $S'$ which would be obtained from $S$ by the ordinary Lorentz transformations, in the sense that $S''$ and $S'$ are connected by an internal coordinate transformation \begin{equation}\label{mol2} x'=\gamma x'' \; , \;\;\;\;\; t'=t''/\gamma -\gamma u x''/c^2 \; . \end{equation} Note that the non-time-orthogonal metric (\ref{mol1}), unlike (\ref{eq2}) and (\ref{metric}), can be transformed to a time-orthogonal metric by an {\em internal} transformation. Note also that the metric (\ref{mol1}), unlike (\ref{eq2}) and (\ref{metric}), is not a metric of a Fermi frame. In \cite{mol}, internal transformations are interpreted as transformations that correspond to a redefinition of the coordinates of the same {\em physical} observer. However, there is something unphysical about internal transformations; if $t'$ is a measure of the physical time for the observer in $S'$, then $t''$ is not, because it corresponds to a ``time" of the same observer which depends on the space point $x'^i$. Therefore, we introduce a more restrictive class of coordinate transformations, which could be better suited to interpret them as transformations that correspond to a redefinition of the coordinates of the same physical observer: \begin{equation}\label{veryweak} t''=f^0 (t') \; , \;\;\;\; x''^i=f^i (x'^1,x'^2,x'^3) \; . \end{equation} We refer to such transformations as {\em restricted internal transformations}. Regular restricted internal transformations form a subgroup of the group of all regular internal transformations. The quantities $g'_{00}dt'^2$ and (\ref{dl}) do not change under restricted internal transformations. Now we have two definitions of the space line element, (\ref{eq3}) and (\ref{dl}), and related to this, two types of restricted coordinate transformations, internal and restricted internal. The space line element (\ref{eq3}) reduces to (\ref{dl}) if $g_{0i}=0$. However, as we have shown in this section, (\ref{dl}) is more appropriate in some cases, even if $g_{0i}\neq 0$. How to know in general what is the suitable definition of the space line element? We can immediately formulate one rule which is certainly suitable: {\em If the metric of a frame can be transformed to a time-orthogonal frame by an internal transformation, then the space line element should be calculated by (\ref{eq3})}. According to the results of this section, we can also formulate another rule: {\em If the metric of a frame in flat spacetime can be obtained from $g_{\mu\nu}=\eta_{\mu\nu}$ by a transformation of the form of (\ref{er1})-(\ref{er2}) followed by an arbitrary restricted internal transformation, then the space line element should be calculated by (\ref{dl}).} Such coordinate transformations can be interpreted as the most general coordinate transformations in flat spacetime that correspond to a physical observer who has a positive mass. We still do not know a general rule. However, one can be satisfied to have a rule for Fermi frames only, or for Fermi frames modified by an arbitrary restricted internal transformation, because only such frames have a direct physical interpretation. One can be tempted to guess that for all such frames the space line element should be calculated by (\ref{dl}), but such a conjecture requires further investigation. For the sake of completeness, let us make a few remarks on general coordinate transformations in curved spacetime. The most general coordinate transformation that corresponds to a physical observer who has a positive mass is a transformation that leads to Fermi coordinates, followed by an arbitrary restricted internal transformation. Other coordinate frames may be useful for some physical calculations, for example, because it is easier to solve some covariant equations of motion in these coordinates. However, if one is interested in how the physical system looks like to a physical observer, one must transform the results to the coordinates specific for this observer. To summarize this section, we conclude that the correct definition of the space line element depends on how it is measured. Formula (\ref{eq3}) is not incorrect, but its applicability is limited and it should be used with great care. In our case of accelerated, rotating frames, it is more appropriate to calculate the space line element with $-g'_{ij}$ instead of with $\gamma'_{ij}$. \section{Relativistic contraction} In Section 2 we have found the coordinate transformation that describes the frame of a short rod uniformly moving along the circular gutter. Let as assume for a while that the length of the rod is infinitesimally small and that the rod is rigid (i.e., its proper length $dL'$ is equal to the proper length of the same rod when it does not accelerate). Let us determine the relativistic contraction of the rod, as seen by an observer in $S$. The observer in $S$ sees both ends of the rod at the same instant, so $dt=0$. From symmetry it is obvious that the relativistic contraction cannot depend on $t$, so, in order to simplify the calculations, we evaluate this at $t=0$. Since the rod is at $x'=y'=0$, (\ref{er2'}) implies that $t'=0$. Taking the differential of (\ref{er1'}) and (\ref{er2'}) with respect to space and time coordinates, and then putting $x'=y'=t'=dt=0$, we find that the observer in $S$ sees the length \begin{equation}\label{inf} dL=dy=\frac{dy'}{\gamma}=\frac{dL'}{\gamma} \; , \end{equation} which is the expected relativistic contraction. Let us now turn our attention to the concept of the proper length of a body. Traditionally, it is defined as a length of the body as seen from the proper frame of the body. However, as we have seen, in general, there is no such thing as a proper frame of the body as a whole. Such a thing exists only for a nonrotating, inertially moving body in flat spacetime. The concept of a proper length of a large body does not have any fundamental meaning, simply because a ``large body" is not actually one object, but a set of many interacting particles. However, the proper length of an infinitesimally small part of a body is well defined. Therefore, we can define the proper length of a whole body as the sum of the proper lengths of its infinitesimal parts. Applying this to (\ref{inf}), we see that the relativistic contraction of a short (but not infinitesimal) rigid rod uniformly moving along the circular gutter is given by $L=L_0/\gamma$, as seen by the observer in $S$. Here $L_0$ is the proper length defined as above. Now, as in Section 2, assume that the rotating ring is a series of independent short rods, uniformly distributed along the gutter. Each rod is relativistically contracted, but the ring is not. This means that the distances between the neighboring ends of the neighboring rods are larger than those for a nonrotating ring, so the proper length of the ring is also larger than that of a nonrotating ring. This is concluded also in \cite{gron2}. This situation mimics a more realistic ring made of elastic material, where atoms play the role of short rigid rods. Owing to the rotation the distances between neighboring atoms increase, so there are tensile stresses in the material. However, it is important to emphasize that the rotation is not essential for understanding of the origin of these tensile forces, because a similar effect also occurs in a linear relativistic motion \cite{dew}. The same relativistic contraction of short rods will be seen by a rotating observer in the center, because his frame is given by the Galilei transformation (\ref{eq1}) and the lengths are calculated by $g_{ij}$, as explained in Section 3. Let us now study how the nonrotating gutter looks like from the point of view of an observer on the rotating ring. Without losing on generality, we evaluate this at $t'=0$. We calculate the length of an infinitesimal part of the gutter lying near the observer, so $x=y=0$. Both ends are seen at the same instant, so $dt'=0$. Taking the differential of (\ref{er1'}) with respect to space coordinates, and then putting $t'=0$, we find that the observer in $S'$ sees the length \begin{equation}\label{inf2} dL'=dy'=\frac{dy}{\gamma}=\frac{dL}{\gamma} \; , \end{equation} which is the expected relativistic contraction. It is important to emphasize that (\ref{inf2}) is correct only in the infinitesimal form. The observer on the ring will not see other distant parts of the gutter contracted in the same way; for him, the gutter and the ring do not look azimuthally symmetric. In the following we study how other parts of the ring look like from the point of view of the observer on the ring. We introduce polar coordinates $r$, $\varphi$, defined by \begin{equation} y=r \sin \varphi \; , \;\;\;\; R+x=r \cos \varphi \; , \end{equation} which are new space coordinates for $S$, with the origin in the center of the circular gutter. The angle $\varphi$ is a good label of the position of any part of the ring even in $S'$. (To visualize this, one can draw angular marks on the gutter. The number of marks separating two points on the gutter or on the ring is a measure of the ``angular distance" in any frame.) Let $S''$ be the frame of another part of the ring. The position of that part of the ring is $x''=y''=0$. The relative position of the space origin of $S''$ with respect to that of $S'$ is given by the constant relative angle $\Delta\varphi_0$, as seen by an observer in $S$. In analogy with (\ref{er1'})-(\ref{er2'}), we find that $S''$ is determined by \begin{equation}\label{er1''} \left( \begin{array}{c} x \\ y \end{array} \right)= \left( \begin{array}{cc} \cos (\gamma\omega t''+\Delta\varphi_0) & -\gamma\sin (\gamma\omega t''+\Delta\varphi_0) \\ \sin (\gamma\omega t''+\Delta\varphi_0) & \gamma\cos (\gamma\omega t''+\Delta\varphi_0) \end{array} \right) \left( \begin{array}{c} x'' \\ y'' \end{array} \right) +R \left( \begin{array}{c} \cos (\gamma\omega t''+\Delta\varphi_0) -1 \\ \sin (\gamma\omega t''+\Delta\varphi_0) \end{array} \right) \; , \end{equation} \begin{equation}\label{er2''} t=\gamma t'' +\frac{\gamma}{c^2}\omega R y'' \; . \end{equation} The observer in $S'$ will see the other part of the ring at the relative ``angular distance" $\Delta\varphi$, which, owing to the relativistic effects, differs from $\Delta\varphi_0$. Let the labels $A$, $B$ denote the coordinates of the part of the ring that lie at $S'$ and $S''$, respectively. Since the rotation is uniform, the relative ``angular distance" \begin{equation}\label{Dphi} \Delta\varphi=\varphi_{B}(t''_{B})-\varphi_{A}(t''_{A}) = \Delta\varphi_0 +\gamma\omega t''_{B}-\gamma\omega t'_{A} \; , \end{equation} cannot depend on $t'$, so without losing on generality, we evaluate this at $t'=0$. Since the observer sees both parts of the ring at the same instant, we have $t'_{A}=t'_{B}=0$. Since $x''_{B}=y''_{B}=0$, from (\ref{er1''}) we find \begin{equation}\label{y2} y_{B}=R\sin (\gamma\omega t''_{B}+\Delta\varphi_0) \; , \end{equation} and from (\ref{er2''}) \begin{equation}\label{t2} t_{B}=\gamma t''_{B} \; . \end{equation} From $t'_{B}=0$ and (\ref{t=0}) it follows $t_{B}=\omega R y_{B}/c^2$, which, because of (\ref{t2}), can be written as $\gamma t''_{B}=\omega R y_{B}/c^2$. This, together with (\ref{y2}), leads to the equation that determines $t''_{B}$: \begin{equation}\label{eqgron1} \gamma\omega t''_{B}=\beta^2\sin (\gamma\omega t''_{B}+\Delta\varphi_0) \; , \end{equation} where $\beta^2 \equiv \omega^2 R^2 /c^2$. From $t'_{A}=0$ and (\ref{Dphi}) we see that $\Delta\varphi=\gamma\omega t''_{B}+\Delta\varphi_0$, so (\ref{eqgron1}) can be written as \begin{equation}\label{eqgron1'} \Delta\varphi -\Delta\varphi_0 =\beta^2\sin \Delta\varphi \; . \end{equation} Equation (\ref{eqgron1'}) determines the relative ``angular distance" $\Delta\varphi$ between two points on the ring as seen by the observer at one of the points, if the relative angle between these two points, as seen by the observer in $S$, is $\Delta\varphi_0$. In other words, (\ref{eqgron1'}) determines how the ring looks like to the observer on the ring. For an inertial observer whose instantaneous position and velocity are equal to that of the observer on the ring, the same equation (\ref{eqgron1'}) is found in \cite{gron1}, where the solution is graphically depicted. This means, contrary to the conclusion of \cite{gron1}, that the inertial and the noninertial observers see the ring in the same way. If the two points on the ring are very close to each other, then $\Delta\varphi_0$ and $\Delta\varphi$ are very small. By expanding equation (\ref{eqgron1'}) for small angles we find the approximative solution $\Delta\varphi=\gamma^2 \Delta\varphi_0$. The factor $\gamma^2$ is easy to understand; one factor of $\gamma$ appears because the part of the gutter close to the observer on the ring looks shorter for that observer than it really is, and the other factor of $\gamma$ appears because the part of the ring close to the observer on the ring is longer than that of the same ring when it does not rotate. \section{The rate of clocks} Assume that there are two clocks at different positions on the ring. Assume also that they show the same time, as seen by an observer in $S$. Then, as shown in Section 2, both clocks show the time $t'=t/\gamma$, as seen from $S$. These two clocks do not show the same time as seen by an observer on the ring. If the position of the observer coincides with the position of one of the clocks, then the time-shift of the other clock is given by (\ref{eqgron1}). Let us calculate the time-shift of the clock at the fixed position $(x,y)$, as seen by the observer in $S'$. From (\ref{er1'}) we express $y'$ as a function of $x$, $y$, and $t'$, and put this in (\ref{er2'}). The result is \begin{equation}\label{txy} t=\gamma t' + \frac{\omega R}{c^2}[y \cos \gamma \omega t' -(x+R) \sin \gamma \omega t'] \; . \end{equation} For comparison, if (\ref{er1'}) and (\ref{er2'}) are replaced by the ordinary Lorentz boosts for a constant velocity in the $y$-direction, then (\ref{txy}) should be replaced by \begin{equation}\label{txyo} t=\frac{t'}{\gamma}+\frac{u}{c^2}y \; . \end{equation} To understand the physical meaning of (\ref{txy}), we explore some special cases. If $\gamma \omega t'=2k\pi$, then $t=\gamma t'+\omega R y/c^2$. In this case, the rate of clocks $\Delta t/\Delta t'=\gamma$ is the same as that for the observer in $S$. This can also be understood as a time-averaged rate, because the oscillatory functions in (\ref{txy}) vanish when they are averaged over time. Therefore, the observer in $S'$ agrees with the observer in $S$ that the clock in $S'$ is slower, but only in a time-averaged sense. At some instants the observer in $S'$ sees that the clock in $S$ is slower than his clock. For example, by putting $x=0$ and expanding (\ref{txy}) for small $t'$, we recover formula (\ref{txyo}), with $u=\omega R$. If the clock in $S$ is in the center, which corresponds to $x=-R$, $y=0$, then (\ref{txy}) gives $t=\gamma t'$, so in this case there is no oscillatory behavior. \section{Velocity of light} Let us also make some comments on the velocity of light. The Sagnac effect is usually interpreted as a dependence of the velocity of light on the direction of light propagation in a rotating frame (see, for example, \cite{post}, \cite{tart} and references therein). However, such an interpretation is based on the interpretation of the frame $S'$ defined by (\ref{eq1}) as a proper frame of all observers on a rotating platform. Now we know that each observer belongs to a different local Fermi frame, and from (\ref{metric}) we see that in the {\em vicinity} of any observer the metric is equal to the Minkowski metric $\eta_{\mu\nu}$. This implies that {\em for any local observer the velocity of light is isotropic and is equal to $c$, providing that it is measured by propagating a light beam in a {\bf small} neighborhood of the observer, using Einstein synchronized clocks}. This is also true for an observer in curved spacetime, because his proper frame is given by the appropriate Fermi coordinates, which also have a property that $g_{\mu\nu}=\eta_{\mu\nu}$ at the position of the observer. The phrases ``local" and ``small" denote spatial dimensions inside which the metric tensor does not change significantly. Of course, the velocity of light does not have to be equal to $c$ for an observer which is not at the same position as the light. However, this is not only a property of non-time-orthogonal frames. For example, if the acceleration of an uniformly accelerated observer and the propagation of light are both in the $x'$-direction, then from (\ref{metric}) one can find that the accelerated observer sees the velocity of light as $|dx'/dt'|=c\sqrt{1+a'x'/c^2}$, being equal to $c$ only at $x'=0$. A similar effect occurs for a radial motion of light in the vicinity of the Schwarzschild radius of a black hole, as seen by a static observer faraway from the Schwarzschild radius. Concerning the Sagnac effect, we do not claim that the standard prediction for the phase shift is incorrect. It can also be derived by performing calculations in the nonrotating frame $S$ \cite{post}, and such a derivation, based on the well-understood Minkowski spacetime, is perfectly correct. We have nothing new to say about the phase shift, which appears when clockwise and counterclockwise propagated light beams finally meet. However, as seen by an observer on the rim of a rotating disc, the velocity of the light beam will be a complicated function of time $t'$, or equivalently, of the position $(x',y')$ of the beam. The trajectory of the light beam expressed in $S$-coordinates takes a simple form \begin{equation}\label{sag} y=R \sin \omega_L t \; , \;\;\;\;\; x=R(-1+\cos \omega_L t) \; , \end{equation} where $\omega_L =\pm c/R$. The plus and minus signs refer to the counterclockwise and clockwise propagated beams, respectively. Using (\ref{er1'}), (\ref{er2'}), and (\ref{sag}), one can eliminate $x,y,t$ and express $x',y'$ as functions of $t'$. The speed of light as seen by the observer in $S'$ is \begin{equation}\label{sag2} v'_L =\sqrt{\left( \frac{dx'}{dt'}\right)^2 + \left( \frac{dy'}{dt'}\right)^2 } \; . \end{equation} Expanding (\ref{er1'}) and (\ref{sag}) for small $t'$ and $t$, respectively, one can easily find $y'=\pm ct'+{\cal O}(t'^2)$, $x'={\cal O}(t'^2)$, which means that the observer sees the velocity of light equal to $c$ when the light is at the same position as the observer, just as expected. \section{Discussion} From the experience acquired by careful calculations in the preceding sections, we can generalize some of the results without much effort, using qualitative and intuitive arguments. If an observation in $S$ is performed at the instant $t$, then the solution of (\ref{er4}) can always be chosen such that at $t$ the axes $x'^i$ are parallel to the corresponding axes $x^i$. Therefore, for a small range of values of $t'$, the transformations (\ref{er1})-(\ref{er2}) can be approximated by the ordinary Lorentz boosts (see (\ref{t=0})). From this fact we conclude that if a moving rigid body is short enough, then its relativistic contraction in the direction of the instantaneous velocity, as seen from $S$, is simply given by $L(t)=L'/\gamma(t)$, i.e., it depends only on the instantaneous velocity, not on its acceleration and rotation. (``Short enough" means that $L'\ll c^2/a'_{\|}$, where $a'_{\|}$ is the component of the proper acceleration parallel to the direction of the velocity \cite{nikol}). By a similar argument we may conclude that an arbitrarily accelerated and rotating observer sees equal lengths of other differently moving objects as an inertial observer whose instantaneous position and velocity are equal to that of the arbitrarily accelerated and rotating observer. So far we have studied a rotating ring. A rotating disc is a more complicated object, with some additional dynamical effects related to elastic and inertial forces. However, a disc can be modeled as a series of concentric rings, each of them being constrained to have a fixed radius. In this case, the analysis of a rotating disc becomes essentially the same as that of a rotating ring. Let us also give some additional arguments why our resolution of the Ehrenfest paradox is correct. Our method, based on coordinate transformations (\ref{er1})-(\ref{er2}), is really a generalization of the well-known derivation of the Lorentz contraction for constant velocities. In our approach the origin of the relativistic contraction lies in the non-Galilean transformation, not in the nontrivial metric, whereas in the standard approach the transformation is Galilean and the contraction is due to the nontrivial metric (\ref{eq2}). Note finally that our approach allows a generalization to a more complicated motion, whereas the standard approach does not. Finally, let us make some comments on the observability of the relativistic contraction. In principle, it could be observed by photographing a rod with a very short exposition, such that both ends are observed at the same instant. Since the velocity of the incoming information (velocity of light) is finite, both ends of the rod should be positioned at the same distance from the observer. Therefore, the ideal setup for such a measurement is a rod in a uniform circular motion and a camera in the center, providing that we can achieve a short enough exposition. It is assumed that in this experiment the only object that moves circularly is a rod (with two ends); there is neither a rotating disc, nor a rotating ring. An indirect, but easier-to-perform experimental verification of the relativistic contraction could perhaps be obtained by measuring the velocity of a rotating ring in a rigid circular gutter, needed to achieve the break of the ring, and comparing it with the elongation needed to achieve the break of the ring caused by ordinary stretching. Of course, in both types of experiments the problem is to achieve a relativistic velocity of macroscopic objects, so these can be considered merely as {\it gedanken} experiments. \section{Conclusion} In this paper a new resolution of the Ehrenfest paradox has been provided by taking into consideration the fact that although there is no relative motion among different points on a rotating disc, each point belongs to a different noninertial local Fermi frame. If a rotating ring (or a disc) is constrained to have a fixed radius from the point of view of an inertial observer, it has been found that there are tensile stresses in the disc, in agreement with the prediction of the standard approach. However, contrary to the prediction of the standard approach, it has been found that an observer on the rim of the disc will see equal lengths of other differently moving objects as an inertial observer whose instantaneous position and velocity are equal to that of the observer on the rim, providing that the observations of different events are simultaneous. This also generalizes to observers arbitrarily moving in flat spacetime. The paper deals mainly with flat spacetime, with particular attention paid to circular motion. However, it gives several results which are of very general relevance, not only for arbitrary motion in flat spacetime, but also for general relativity and curved spacetime. First, it has been demonstrated that the generally accepted formula (\ref{eq3}) is not always correct. The correct definition of the space line element depends on how it is measured, so (\ref{eq3}) should be used with great care. In some cases, the ``naive" formula (\ref{dl}) is more appropriate. One such case is a metric of a frame in flat spacetime that can be obtained from $g_{\mu\nu}=\eta_{\mu\nu}$ by a transformation of the form of (\ref{er1})-(\ref{er2}), followed by an arbitrary restricted internal transformation. Further investigation is needed in order to generalize this result. Second, the paper demonstrates the importance of the use of Fermi coordinates. One of the consequences of their use is the result that for any local observer the velocity of light is isotropic and is equal to $c$, providing that it is measured by propagating a light beam in a small neighborhood of the observer. This fact should be used for a correct treatment of the Sagnac effect if one wants to explore the general relativistic corrections. Fermi coordinates should also be used in order to understand the physical effects related to a rotating black hole, to give a correct treatment of the Hawking radiation, as well as for any other physical effect, whenever intended to describe the world how it looks like to a particular observer. \section*{Acknowledgment} The author is grateful to Damir Stoi\'{c} for motivating discussions. This work was supported by the Ministry of Science and Technology of the Republic of Croatia under Contract No. 00980102.
\section{Introduction} The unified model for Seyfert galaxies proposes that orientation of a molecular torus determines the optical emission-line characteristics (e.g.\ Antonucci\markcite{1} 1993). When the molecular torus lies in our line of sight, it blocks our view of the broad optical emission lines, leading to a Seyfert 2 classification. The X-ray emission of Seyfert 2 galaxies is frequently absorbed (e.g.\ Turner et al.\markcite{25} 1998; Bassani et al.\markcite{23} 1999), a result which supports this unified model. In the most extreme case, the obscuring material presents such a high column density to the observer that no X-rays are transmitted. Such objects are termed ``Compton-thick'' Seyfert 2s, and the only X-rays observed from them are ones that have been scattered from surrounding material. (Diffuse thermal X-rays may also contribute). Such objects are important because they directly support unified models for Seyfert galaxies. The scattering is thought to originate in one or both of two types of material, each of which imparts characteristic signatures on the observed X-ray spectrum (for a review, Matt\markcite{15} 1997). Scattering can occur in the warm optically thin gas that is thought to produce the polarized broad lines seen in some Seyfert 2s. The resulting spectrum has the same slope as the intrinsic spectrum with superimposed emission lines from recombination. This is the process which appears to dominate in the archetype Seyfert 2 galaxy NGC~1068 (e.g.\ Netzer \& Turner\markcite{20} 1997). Scattering can also occur in optically thick cool material located on the surface of the molecular torus. In this case, the process is called Compton reflection (e.g.\ Lightman \& White\markcite{10} 1988), and the observed continuum spectrum is flat with superimposed K-shell fluorescence lines (e.g.\ Reynolds et al.\markcite{22} 1994). Circinus can be considered the prototype of a Compton-reflection dominated Seyfert 2 galaxy (Matt et al.\markcite{17} 1996). Compton-reflection dominated Seyfert 2 galaxies are important because they may comprise a significant fraction of the X-ray background (Fabian et al.\markcite{9} 1990). They were once thought to be rare (e.g. Matt\markcite{15} 1997); however, new observations of objects selected according to their [\ion{O}{III}] emission-line flux, a method which ideally does not discriminate against highly absorbed objects, show that they may be more common than previously thought (Maiolino et al.\markcite{13} 1998). However, {\it bright} examples of this class remain rare. This is not surprising, because the reflected X-rays are very much weaker than the primary continuum. We present the results of an {\it RXTE} observation of the nearby (18 Mpc) Seyfert 2 galaxy NGC~6300. This object has a flat hard X-ray spectrum and huge equivalent width iron line which suggests that it is a Compton-reflection dominated Seyfert 2 galaxy. If so, it is one of the brightest members of this class known, about half as bright as the prototype, Circinus, and far brighter than other examples. \section{Data Analysis} NGC~6300 was first detected in hard X-rays during a {\it Ginga} maneuver (Awaki\markcite{2} 1991). We proposed scanning and pointing observations of this galaxy using {\it RXTE} to confirm the {\it Ginga} detection. Another Seyfert 2 galaxy, NGC~6393, was detected during a {\it Ginga} scan and was also investigated as part of this proposal. The data show that NGC~6393 was very faint ($<0.5\rm \,counts\, s^{-1}$ in the top-layer for 5 PCUs). The scanning {\it RXTE} observation of NGC 6300 was performed on 1997 February 14--15. Four of 5 PCUs were on for the entire observation and analysis was confined to these detectors. A pointed observation followed on February 20, 1997, performed with all five PCUs on. The data were reduced using Ftools 4.1 and 4.2 and standard data selection criteria recommended for faint sources. The resulting exposure for the pointed observation was 24,896 seconds. NGC 6300 was detected in all three layers of the PCA, and the top and mid layers were used for spectral fitting. Background subtraction yielded net count rates for 5 PCUs of 4.4 counts s$^{-1}$ (12.5\% of the total) between 3 and 24 keV for the top layer, and 0.86 counts s$^{-1}$ (8.8\% of the total) between 9 and 24 keV for the mid layer. The current standard background model for the {\it RXTE} PCA is quite good; however, NGC~6300 is a relatively faint source and therefore we attempt to estimate systematic errors associated with the background subtraction. Above 30~keV, no signal should be detected; however, we found a positive signal which could be removed if the background normalization were increased by 1\%. Below about 7~keV, no signal should be detected in the mid or bottom layers. We observed a small deficit in signal which would be removed if the background normalization were decreased by 1\%. We consider this evidence that the systematic error on the background subtraction is less than 1\%. The scan observation consisted of 4 passes over the object with a total scan length of 6 degrees. The scans were performed keeping the declination constant during the first two passes and the right ascension constant during the second two passes. The resulting scan profiles when compared with the optical position clearly indicate that NGC~6300 is the X-ray source. The field of view of the PCA is less than two degrees in total width. Since the scan length was 6 degrees, and since there are apparently no other X-ray sources in the field of view, the ends of the scan paths can be used to check the quality of the background subtraction. This was of some concern since NGC~6300 is rather near the Galactic plane ($l=328$, $b=-14$) and thus there could be Galactic X-ray emission not accounted for in the background model. We accumulated spectra with offset from the source position $>1.5^\circ$. The exposure time was 1472 seconds. The count rate between 3 and 24 keV was $-0.28 \pm 0.19 \rm\, counts\,s^{-1}$, so there was no evidence for unmodeled Galactic emission. Thermal model residuals show no pattern; i.e., there is no evidence for a 6.7~keV iron emission line from the Galactic Ridge (Yamauchi \& Koyama\markcite{28} 1993). \begin{figure}[t] \vbox to3.0in{\rule{0pt}{3.0in}} \special{psfile=ngc6300_fig1.ps angle=270 hoffset=10 voffset=250 vscale=100 hscale=100} \caption{Ratio of data to a model consisting of power law plus Galactic absorption model. The excess near 6.4~keV clearly indicates the large equivalent width iron line. Solid points and crosses denote the top-layer and mid-layer data, respectively.} \end{figure} The spectrum from the pointed observation was first modeled using a power law plus Galactic absorption set equal to $9.38\times 10^{20}\rm\, cm^{-2}$ (Figure 1; Dickey \& Lockman\markcite{6} 1990). This model did not fit the data well ($\chi^2=609$ for 82 degrees of freedom (d.o.f.)). The photon index is very flat ($\Gamma=0.60$), there is clear evidence for an iron emission line and there are negative low-energy residuals. Addition of a narrow ($\sigma=0.05\rm \,keV$) line with energy fixed at $6.4\rm\, keV$ improves the fit substantially ($\Delta\chi^2=473$) but low energy residuals remain. Additional absorption in the galaxy rest frame improves the fit substantially ($\Delta\chi^2=45$). Freeing the line energy again improves the fit ($\Delta\chi^2=11$); the best fit rest-frame line energy is $6.26 \rm\,keV$. Freeing the line width marginally improves the fit ($\Delta\chi^2=5$). The final fit parameters are listed in Table 1 and fit results are shown in Figure 2. The absorbed power law plus iron line is an acceptable model. The notable properties of the fit are a very flat photon index ($\Gamma=0.68$) and very large equivalent width ($920\,\rm eV$). Such parameters suggest that the spectrum of NGC~6300 is dominated by Compton-reflection (Matt et al.\markcite{17} 1996; Malaguti et al.\markcite{12} 1998; Reynolds et al.\markcite{22} 1994). We next use the {\it pexrav} model in XSPEC to explore this possibility. This model calculates the expected X-ray spectrum when a point source of X-rays is incident on optically thick, predominately neutral (except hydrogen and helium) material. The parameter $R$ measures the solid angle $\Omega$ subtended by the optically thick material: $R=\Omega/2\pi$. The model that was fit includes a narrow iron line and a direct and reflected power law; additional absorption also appears to be necessary. The low resolution and limited band pass provided by the {\it RXTE} spectrum means that not all of the model parameters could be constrained by the data; thus, the energy of the exponential cutoff was fixed at 500~keV, approximately the value that has been found in {\it OSSE} data from Seyfert galaxies (Zdziarski et al.\markcite{29} 1995), and the inclination was initially fixed arbitrarily at $45^\circ$. This model provided a fairly good fit to the data ($\chi^2=112$ for 77 d.o.f.). However, the best fit value of $R$ is very large (1450) and not well constrained. This indicates that the spectrum can be modeled using reflection alone, and there is no significant contribution of direct emission (e.g.\ Matt et al.\markcite{17} 1996). Reflection alone gives the same $\chi^2$ as the model which includes a weak direct component, but the fit is still not completely satisfactory, and it could be improved in two ways. The first is to allow the iron abundance to vary. We set the iron abundance relative to solar in the {\it pexrav} model equal to the abundances of light elements and allow these parameters to vary together. The fit is improved significantly ($\Delta\chi^2=-36$ for $\Delta$d.o.f.=1) and gives a slightly subsolar abundance. The second is to allow the inclination to be free. The fit is not very sensitive to this parameter, as $\Delta\chi^2$ over the whole range is 9.2. The best fit value is $\cos(\Theta)=0.22$, corresponding to 77$^\circ$ from the normal. The parameters are listed in Table 1. \begin{deluxetable}{ll} \small \tablewidth{20pc} \tablenum{1} \tablecaption{Spectral Fitting Results} \tablehead{ \colhead{Parameter} & \colhead{Value} \\} \startdata \multicolumn{2}{c}{Power Law Model} \nl $\Gamma$ & $0.68_{-0.09,-0.13}^{+0.09,+0.16}$ \nl $\rm N_H$ ($10^{22}\rm \,cm^{-2}$) & $5.2_{-1.7,-2.7}^{+1.8,+3.5}$ \nl $\rm E_{Fe}$ (keV) & $6.26_{-0.06,-0.02}^{+0.06,+0.02}$ \nl $\rm \sigma_{Fe}$ (keV) & $0.32_{-0.16,+0.07}^{+0.13,-0.07}$ \nl $\rm F_{Fe}$ ($10^{-5}\,\rm \,cm^{-2}\,s^{-1}$) & $8.2_{-1.1,+0.5}^{+1.2,-0.4}$ \nl $\rm EW_{Fe}$ (eV) & $920_{-130,+100}^{+140,-90}$ \nl $\chi^2$/79 d.o.f. & $75.5_{\ldots,-2.2}^{\ldots,+6.7}$ \nl \tableline \multicolumn{2}{c}{Compton Reflection Model} \nl $\Gamma$ & $1.89_{-0.09,+0.08}^{+0.08,-0.13}$ \nl Abundance $^a$ & $0.61_{-0.11,+0.15}^{+0.11,-0.16}$ \nl $\cos(\Theta)$ & $0.22_{-0.22,+0.02}^{+0.16,-0.04}$ \nl $\rm E_{Fe}$ (keV) & $6.29_{-0.08,+0.02}^{+0.09,-0.02}$ \nl $\rm \sigma_{Fe}$ (keV) & $0.22_{-0.22,-0.13}^{+0.20,+0.06}$ \nl $\rm F_{Fe}$ ($10^{-5}\,\rm \,cm^{-2}\,s^{-1}$) & $4.7_{-1.0,-1.2}^{+1.2,+1.2}$ \nl $\rm EW_{Fe}$ (eV) & $470_{-100,-150}^{+120,+200}$ \nl $\chi^2$/77 d.o.f. & $70.3_{\ldots,+9.2}^{\ldots,-1.6}$ \nl \tableline \multicolumn{2}{c}{Dual Absorber Model} \nl $\Gamma$ & $1.71^{+0.21,+0.27}_{-0.19,-0.21}$ \nl $\rm N_{H(thin)}$ ($10^{22}\rm\,cm^{-2}$) & $7.7_{-5.2,-1.4}^{+4.0,+2.0}$ \nl $\rm N_{H(thick)}$ ($10^{22}\rm\,cm^{-2}$) & $58_{-22,-2}^{+23,+5}$ \nl $\rm A_{thick}/A_{thin}$$^b$ & $1.9_{-0.9,-0.7}^{+1.4,+1.0}$ \nl $\rm E_{Fe}$ (keV) & $6.26_{-0.12,+0.002}^{+0.09,-0.005}$ \nl $\rm \sigma_{Fe}$ (keV) & $0.23_{-0.23,+0.04}^{+0.25,-0.05}$ \nl $\rm F_{Fe}$ ($10^{-5}\,\rm \,cm^{-2}\,s^{-1}$) & $3.2_{-1.5,+0.8}^{+1.4,-0.7}$ \nl $\rm EW_{Fe}$ (eV) & $470_{-220,+50}^{+210,-60}$ \nl $\chi^2$/79 d.o.f. & $67.6_{\ldots,-1.2}^{\ldots,+2.4}$ \nl \enddata \tablecomments{Two kinds of errors are given for each parameter value. The first one is the statistical error which represents 90\% confidence for one parameter of interest ($\Delta\chi^2=2.71$). The second one is an estimate of the systematic error obtained by changing the normalization of the background. The results of background under- and oversubtraction by 1\% are given in the sub- and superscript, respectively.} \tablenotetext{a}{Fraction of solar abundance in iron and light elements.} \tablenotetext{b}{Ratio of power-law normalizations.} \end{deluxetable} We investigated the choice of fixed parameters in the {\it pexrav} model {\it a posteriori}. Increasing the cutoff energy did not change the fit. Decreasing the cutoff to 100~keV produced small differences in the parameters; namely, the photon index was smaller, the abundance was higher, and the line flux was lower, but the differences are within the statistical errors of the adopted model. We also investigated the situation when the iron abundance was allowed to vary, but the abundances of lighter metals were maintained at the solar value. A significantly larger photon index by $\Delta\Gamma \approx 0.15$ was required, due to the decreased reflectivity in soft X-rays. We also investigated the effect of a 1\% systematic error in the background normalization, and the results are listed in Table 1. This resulted in the largest change in the fit parameters, but the resulting estimated systematic errors are in the worst case less than a factor of two larger than the statistical errors. Because of the low energy resolution of the {\it RXTE} spectra, other models can be found which fit equally well. It is possible to describe the spectra using a sum of absorbed power laws (the ``dual absorber'' model; e.g.\ Weaver et al.\markcite{27} 1994). Specifically, the model consisted of two power laws, both absorbed by a moderate column, and one absorbed by a heavy column. The resulting photon index was very flat ($\Gamma=1.15$), and is therefore deemed unphysical. However, including reflection with $R=1$ in the dual absorber model gave an insignificant improvement in fit ($\Delta\chi^2=-0.6$) but a more plausible photon index ($\Gamma=1.7$). The fit parameters are given in Table 1. It is notable that the spectra cannot be described using a highly absorbed transmitted component and an unabsorbed Compton-reflection dominated component, as has been found to be appropriate for Mrk~3 (Cappi et al.\markcite{4} 1999). \begin{figure}[t] \vbox to4.0in{\rule{0pt}{4.0in}} \special{psfile=ngc6300_fig2.ps angle=270 hoffset=-40 voffset=420 vscale=70 hscale=70} \caption{Unfolded spectra and ratio of data to model for the three models considered (see Table 1).} \end{figure} \section{Discussion} \subsection{Compton Reflection Continuum Model} The X-ray continuum of NGC~6300 can be modeled as pure Compton reflection. For solar abundance, the iron line equivalent width relative to the reflection continuum is predicted to be between 1 and 2 keV depending on inclination (Matt, Perola \& Piro\markcite{18} 1991). When we fit a power law continuum to the spectra, the observed equivalent width is nearly 1~keV. However, when the reflection continuum is fitted, the measured equivalent width is reduced to 470~eV. The reason for the reduction in the measured equivalent width is that the reflection continuum model includes a substantial iron edge. In low resolution data, the iron line and iron edge overlap in the response-convolved spectra. Therefore, when the continuum is modeled by a power law, the iron line models both the line and the edge, so the measured equivalent width is larger than when the continuum is modeled by the reflection continuum which includes the iron edge explicitly. This effect can be seen in Figure 2. A significant improvement is obtained when the iron abundance in the reflection continuum model is allowed to be subsolar. The iron abundance is determined by the depth of the iron edge. Therefore, the subsolar abundance fits because the iron edge is apparently not as deep as the model predicts. The fact that the iron line has a lower equivalent width than predicted in the Compton reflection continuum model may also support subsolar abundance. Alternatively, however, the apparent subsolar abundances may at least partially be due to calibration uncertainties in the {\it RXTE} PCA. (The resolution of the {\it RXTE} PCA is under some debate; see, Weaver, Krolik \& Pier\markcite{26} 1998). The iron line and edge overlap in the response-convolved spectra. If the true energy resolution is worse than the current estimated value, then because the line is an excess and the edge is a deficit, both would be measured to be smaller than they really are. Another source of uncertainty may come from the models themselves, which depend strongly on the geometry of the illuminating and reprocessing material. Models generally assume a point source of X-rays located above a disk and illuminating it with high covering fraction. Such an ideal case may not be attained in nature. \subsection{Dual Absorber Continuum Model} The dual absorber model can also describe the spectra. That such a fit is successful is not surprising, as any flat continuum can be modeled as a sum of absorbed power laws (e.g.\ the X-ray background). The iron line equivalent width for the dual absorber model is similar to that found for the Compton-reflection model, and smaller than that found for the power law model. The reason for the difference is that, like the Compton reflection continuum model, the dual absorber model explicitly includes an iron edge. A plausible origin for the iron line in the dual absorber model is in the absorbing material itself. However, the iron line equivalent width appears to be too large to have been produced in the absorbing material. We investigate this possibility by comparing the observed iron line flux to the predicted value from a spherical shell of gas surrounding an isotropically illuminating point source (Leahy \& Creighton\markcite{11} 1993). A line flux of $2.3\times 10^{-5}\rm \, photons\,cm^{-2}\,s^{-1}$ is predicted for the absorption columns and covering fractions determined by the fit; this is about half of what is observed ($4.7\times 10^{-5}\rm \, photons\,cm^{-2}\,s^{-1}$). The dual absorber model also requires a reflection component with $R=1$ and therefore an additional iron line with equivalent width $\sim 100$~eV is expected from the reflection. Then the predicted flux increases to $3.3\times 10^{-5}\rm \, photons\,cm^{-2}\,s^{-1}$, about 70\% of what is observed. The predicted line flux would be smaller if the absorbing material does not completely cover the source, a circumstance that would exacerbate the difference between predicted and observed flux. Thus, the iron line equivalent width, at least to first approximation, appears to be too large to have been produced in the absorbing material required by the dual absorber model. Therefore, either an iron overabundance is required in the dual absorber model, or the alternative model, the Compton-reflection dominated model, is favored. Discrimination between models will come with observations using detectors with better energy resolution. If NGC~6300 is a Compton-thick Seyfert 2, then we should detect soft X-ray emission lines (e.g.\ Reynolds et al.\markcite{22} 1994). The lack of any observable soft excess in the {\it RXTE} data may imply that there is absorption by the host galaxy (see below), or that there is little contamination from thermal X-rays or scattering from warm gas. If the latter case is true, NGC~6300 will be a particularly clean example of a Compton reflection-dominated Seyfert 2 galaxy, and the observed soft X-ray emission lines should be unambiguously attributable to fluorescence. Such proof should be easily attainable as NGC~6300 is bright compared with known Compton-reflection dominated Seyfert 2 galaxies. \subsection{Information from Other Wavebands} Optical observations provide some support for Compton-thick absorption in NGC 6300. Intrinsically, both hard X-rays and forbidden optical emission lines should be emitted approximately isotropically; therefore, the ratio of these quantities should be the same from object to object. However, if the absorption is Compton-thick, the observed hard X-ray luminosity and therefore $L_X/L_{[OIII]}$ will be significantly reduced; the ratio of the power law to reflection component 2--10 keV fluxes in the {\it pexrav} model is $\approx 15$. Care must be taken when applying this test, as the [\ion{O}{III}] flux must be corrected for reddening in the narrow-line region, and determination of the reddening using narrow-line Balmer decrements can be difficult. The optical spectrum of NGC~6300 is dominated by starlight (e.g.\ Storchi-Bergmann \& Pastoriza\markcite{24} 1989); to remove the Balmer absorption from the stars, an accurate galaxy spectrum subtraction must be done. Another complication could be narrow Balmer lines from star formation. NGC~6300 has a well-studied starburst ring (e.g.\ Buta\markcite{3} 1987), but H$\alpha$ images show that the \ion{H}{II} regions are located $>0.5^\prime$ from the nucleus, with a little diffuse emission inside of that (e.g.\ Evans et al.\markcite{7} 1996). High quality long-slit spectra yield $A_v \approx 2.5-3$ from both the red continuum and the Balmer decrement (Storchi-Bergmann 1999, P.\ comm.). The observed 2--10 keV flux is $6.4\times 10^{-12}\rm \,erg\,cm^{-2}\,s^{-1}$ (corresponding to a luminosity of $2.5\times 10^{41}\,\rm erg\,s^{-1}$). Then $L_X/L_{[OIII]}$ is approximately 1.1--1.9. This value is quite low and comparable to those obtained from Compton-thick Seyfert 2s by Maiolino et al.\markcite{13} (1998). In particular, Circinus has $F_X=1.4\times 10^{-11} \rm ergs\,cm^{-2}\,s^{-1}$ (Matt et al.\markcite{16} 1999) and reddening-corrected $F_{[OIII]}=1.95\times 10^{-11}$ ($A_v=5.2 \pm 0.4$; Oliva et al.\markcite{21} 1994), yielding $L_X/L_{[OIII]}=0.7$. For comparison, a reddening-corrected sample of Seyfert 1s taken from Mulchaey et al.\markcite{19} 1994 as a mean ratio of 14.8 ($1\sigma$ range 6.4--33.9). In contrast, the intrinsic luminosity (i.e.\ corrected for absorption) for the dual absorber model is $6.9\times 10^{41}\,\rm erg\,s^{-1}$, implying $L_X/L_{[OIII]} \approx $3.0--5.2. It is notable also that NGC~6300 shows the reddest continuum toward the nucleus in a sample of objects studied with long slit spectroscopy (Cid Fernandes, Storchi-Bergmann \& Schmitt\markcite{5} 1998). Furthermore, there seems to be a correlation between the presence of a bar in the host galaxy and presence of a Compton thick Seyfert 2 nucleus (Maiolino, Risaliti \& Salvati\markcite{14} 1999); NGC 6300 has a bar (e.g.\ Buta\markcite{3} 1987). \subsection{Consistency with Einstein IPC Observation} The X-ray spectra from Seyfert 2 galaxies frequently includes a soft spectral component which may originate in scattering by warm gas or diffuse thermal X-rays. However, lack of detection in an {\it Einstein} IPC observation, combined with the {\it RXTE} observation presented here, shows that there is no evidence for such a component in NGC~6300. In 1979 NGC~6300 was observed with the {\it Einstein} IPC for 990 seconds. It was not detected and the three sigma upper limit to the count rate was $1.19\times 10^{-2}\rm\, counts\,s^{-1}$ between 0.2 and 4.0 keV (Fabbiano, Kim, \& Trinchieri\markcite{8} 1992). The Compton reflection-dominated model predicts a count rate of $1.3\times 10^{-2}\rm\, counts\,s^{-1}$, just the same order as the upper limit, and probably consistent within the uncertainties of the model and the relative flux calibrations of the two instruments. There may be intrinsic absorption in the system, for example, from the host galaxy. The observed optical reddening of $A_v=2.5-3.0$ corresponds to an absorption column of 4--5$\times 10^{21}\rm cm^{-2}$, assuming a standard dust to gas ratio. Including this column in the Compton-reflection dominated model leads to a predicted IPC flux of $1.1\times 10^{-2}\rm\, counts\,s^{-1}$, consistent with the upper limit. Because the {\it RXTE} band pass is truncated at 3~keV, it is impossible to estimate with accuracy the intrinsic absorption. For the Compton-reflection model, the 90\% upper limit for one parameter of interest is $1.8\times 10^{22}\rm\,cm^{-2}$. Alternatively, there may have been variability within the 18 years between the IPC and the {\it RXTE} observation. A probable site for the reflection is the inner wall of the molecular torus which blocks the line of sight to the nucleus. In unified models for Seyfert galaxies, this material is located outside of the broad line region, and can be 1--100 pc from the nucleus. Short term variability is not expected; long term variability is possible but requires a long term trend in flux. \acknowledgements KML acknowledges useful discussions on the {\it RXTE} background with Keith Jahoda. KML gratefully acknowledges support by NAG-4112 ({\it RXTE}) and NAG5-7971 (LTSA).
\section{Introduction} Strangeness as a possible signature of the phase transition from a hadronic state to a QGP state was put forward about 16 years ago \cite{rafelski1}. It was based on the prediction that the production of strange quark pairs would be enhanced as a result of the approximate chiral symmetry restoration in a QGP state in comparison with a hadronic state. The strangeness enhancement in pA and AA collisions with respect to the superposition of nucleon-nucleon collisions has been investigated and confirmed by many experimental groups \cite{wa85,na351,na36,wa97,naga}. However, alternative explanations exist, they are based on the `conventional' physics in the hadronic regime, like rescattering, string-string interaction, etc. \cite{satai,venus,rqmd}. The first detailed theoretical study of strangeness production can be found in \cite{rafelski2}, where the enhanced relative yield of strange and multi-strange particles in nucleus-nucleus collisions with respect to proton-nucleus interactions has been suggested as a sensitive signature of a QGP. We have done a series of studies in recent years investigating strangeness enhancement with a hadron and string scenario \cite{satai,satai2,satai3,taisa1,taisa2}, from which a Monte-Carlo event generator, LUCIAE, was developed \cite{luciae}. Those studies indicate that including rescattering of the final state hadrons is still not enough to reproduce the NA35 \cite{na351} data of strange particle production. To reproduce the NA35 data needs to rely further on the mechanism of reduction of the strange quark suppression in string fragmentation, which contributes to the enhancement of strange particle yield in nucleus-nucleus collisions with respect to the superposition of the nucleon-nucleon collisions \cite{satai2,satai3,taisa1,taisa2}. Similarly, in order to reproduce the NA35 data, the RQMD generator, equipped with rescattering though, has to resort to the colour rope mechanism \cite{rqmd}. In this picture it is assumed that the neighboring interacting strings might form a string cluster called colour rope in pA and AA collisions. The colour rope then fragments in a collective way and tends to enhance the production of the strange quark pairs from the colour field of strings through the increase of the effective string tension. It has been known for years that the strange quark suppression factor ($\lambda$ hereafter), i.e., the suppression of s quark pair production in the color field with respect to u or d pair production, in hadron-hadron collisions is not a constant, but energy-dependent, increasing from a value of 0.2 at the ISR energies to about 0.3 at the top of the SPS energies \cite{kapa}. In \cite{taisa1} we proposed a mechanism to investigate the energy dependence of $\lambda$ in hh collisions by relating the effective string tension to the production of hard gluon jets (mini-jets). A parameterization form was then obtained, which reproduces the energy dependence of $\lambda$ in hh collisions reasonably well \cite{taisa1}. When the same mechanism is used in the study of pA and AA collisions it is found that $\lambda$ would increase with the increase of energy, mass and centrality of a colliding system as a result of mini-jet(gluon) production stemming from the string-string interaction. Our model reproduced nicely the data of strange particle production in hh \cite{taisa1}, pA, and AA \cite{satai3,taisa2} collisions. In this work we use above ideas \cite{taisa1,taisa2} to study the recently published WA97 data of the enhanced production of singly and multiply strange particles in p-Pb and Pb-pb collisions at 158A GeV/c. The study indicates that the WA97 data, which revealed that the enhancement of strange particle yield increases with the increasing of centrality and of s quark content in multiply strange particles in Pb-Pb collisions with respect to p-Pb collisions, could be explained in a hadron-string model except for $\Omega$ yield in the Pb-Pb data. \section{Brief review of the LUCIAE model} LUCIAE model is developed based on the FRITIOF model \cite{fritiof}. FRITIOF is a string model, which started from the modeling of inelastic hadron-hadron collisions and it has been successful in describing many experimental data from the low energies at the ISR-regime all the way to the SPS energies \cite{B.N,H.P1}. In this model a hadron is assumed to behave like a massless relativistic string. A hadron-hadron collision is pictured as the multi-scattering of the partons inside the two colliding hadrons. In FRITIOF, during the collision two hadrons are excited due to longitudinal momentum transfers and/or a Rutherford Parton Scattering (RPS). The highly excited states will emit bremsstrahlung gluons according to the soft radiation model. They are afterwards treated as excitations i.e. the Lund Strings and allowed to decay into final state hadrons according to the Lund fragmentation scheme. The FRITIOF model has been extended to also describe hadron-nucleus and nucleus-nucleus collisions by assuming that the reactions are superposition of binary hadron-hadron collisions in which the geometry of the nucleus plays an important role because the nuclei should then behave as a ``frozen'' bag of nucleons. However in the relativistic nucleus-nucleus collisions there are generally many excited strings formed close by each other during a collision. Thus in the LUCIAE model a Firecracker model \cite{fire} is proposed to deal with the string-string collective interaction. In the Firecracker model it is assumed that several string from a relativistic heavy ion reaction will form a cluster and then the strings inside such a cluster will interact in a collective way. We assume that the groups of neighbouring strings in a cluster may form interacting quantum states so that both the emission of gluonic bremsstrahlung as well as the fragmentation properties can be affected by the large common energy density. In relativistic nucleus-nucleus collision there are generally a lot of hadrons produced, however, FRITIOF does not include the final state interactions. Thus in LUCIAE a rescattering model \cite{satai} is devised to consider the reinteraction of the produced hadrons with each other and with the surrounding cold spectator matter. The distributions of the final state hadrons will be affected by the rescattering process. We refer to the Refs. \cite{satai,luciae} for the details and we just give here the list of the reactions involving in LUCIAE, which are cataloged into \begin{tabbing} ttttttttttttttt\=ttttttttttttttt\=tttttt\=tttttttttttttttt\= \kill \>$\pi$$N$$\rightleftharpoons$ $\Delta$$\pi$ \> \>$\pi$$N$$\rightleftharpoons$ $\rho$$N$\\ \> $N$$N$$\rightleftharpoons$ $\Delta$$N$ \> \>$\pi\pi \rightleftharpoons k\bar{k}$\\ \>$\pi N \rightleftharpoons kY$ \> \>$\pi\bar{N} \rightleftharpoons \bar{k}\bar{Y}$\\ \>$\pi Y \rightleftharpoons k\Xi$ \> \>$\pi\bar{Y} \rightleftharpoons \bar{k}\bar{\Xi}$\\ \>$\bar{k}N \rightleftharpoons \pi Y$ \> \>$k\bar{N} \rightleftharpoons \pi\bar{Y}$\\ \>$\bar{k}Y \rightleftharpoons \pi\Xi$ \> \>$k\bar{Y} \rightleftharpoons \pi\bar{\Xi}$\\ \>$\bar{k}N \rightleftharpoons k\Xi$ \> \>$k\bar{N} \rightleftharpoons \bar{k}\bar{\Xi}$\\ \>$\pi\Xi \rightleftharpoons k\Omega^- $ \> \>$\pi\bar{\Xi} \rightleftharpoons \bar{k}\overline{\Omega^-}$\\ \>$k\bar{\Xi} \rightleftharpoons \pi\overline{\Omega^-}$ \> \>$\bar{k}\Xi \rightleftharpoons \pi\Omega^-$\\ \>$\bar{N}N$ annihilation\\ \>$\bar{Y}N$ annihilation\\ \end{tabbing} where $Y$ refers to the $\Lambda$ or $\Sigma$ and $\Xi$ refers to the $\Xi^-$ or $\Xi^0$. There are 364 reactions involved altogether. In addition, the reduction mechanism of s quark suppression, i. e., the s quark suppression factor increasing with energy, centrality, and mass of the colliding system, which is linked to string tension, is included in LUCIAE via the parameterized formulas \cite{taisa1,taisa2} \begin{equation} \kappa_{eff}=\kappa_{0} (1-\xi)^{-\alpha}, \label{f2} \end{equation} where $\kappa_{0}$ is the string tension of the pure $q\bar{q}$ string, $\alpha$ is a parameter $\sim$ 3, and $\xi$ ($\leq$ 1) is calculated by \begin{equation} \xi =\frac{\ln(\frac{k_{\perp max}^2}{s_{0}})}{\ln (\frac{s}{s_{0}}) + \sum_{j=2}^{n-1} \ln (\frac{k_{\perp j}^2}{s_{0}})}, \label{f3} \end{equation} which represents the scale that a multigluon string is deviated from a pure $q\bar{q}$ string. The s quark suppression factor, $\lambda$, of two string states can thus be calculated by \begin{equation} \lambda_{2} = \lambda_{1}^ {\frac{\kappa_{eff1}}{\kappa_{eff2}}}, \label{f1} \end{equation} where $\kappa_{eff}$ refers to the effective string tension of a multigluon string. Since $\lambda$ is always less than one, above equation indicates the larger effective string tension the more reduction of s quark suppression. The effective string tension is then relevant to the hard gluon kinks (mini-(gluon) jets) created on the string. It should be mentioned that the LUCIAE (FRITIOF) event generator runs together with JETSET routine. In JETSET routine there are model parameters PARJ(2) (i.e., $\lambda$) and PARJ(3). PARJ(3) is the extra suppression of strange diquark production compared to the normal suppression of strange quark pair. Both PARJ(2) and PARJ(3) are responsible for the s quark (diquark) suppression and related to the effective string tension (the relation of Eq. (3) holds true for PARJ(3) as for $\lambda$). Besides $\lambda$ and PARJ(3) there is PARJ(1), which stands for the suppression of diquark-antidiquark pair production in the color field in comparison with the quark-antiquark pair production and is related to the effective string tension as well. The mechanism mentioned above is performed via these parameters in program. How these three parameters affect the multiplicity distribution of final state particles can be found in \cite{satai2,satai3}. \section{Results and discussions} In Table 1 is given the results of the JETSET parameters PARJ(1), PARJ(2) (i.e., $\lambda$), and PARJ(3) varying with the centrality and the size of collision system in p-Pb and Pb-Pb collisions at 158A GeV/c. That seems quite reasonable. Fig. 1a shows the calculated $\Lambda+\bar{\Lambda}$, $\Xi^-+ \overline{\Xi^-}$, and $\Omega^-+\overline{\Omega^-}$ yields per event ($|y-y_{cm}| \leq$ 0.5 and p$_T$ $\geq$ 0 GeV/c) as a function of the number of participant in minimum bias p-Pb collisions and in central (b=2) Pb-Pb collisions at 158A GeV/c (open labels) comparing with WA97 data (full labels) \cite{wa97}. The corresponding results in Pb-Pb collisions after recaling each yield according to its value in p-Pb are given in Fig. 1b. One knows from Fig. 1a that the agreement between theory and experiment is quite well for $\Lambda+\bar{\Lambda}$ and $\Xi^-+\overline{\Xi^-}$, however, for $\Omega^-+\overline{\Omega^-}$ the theoretical results are lower than experiments. That should be study further both theoretically and experimentally. In Fig. 1b the theoretical results of $\Omega^-+\overline{\Omega^-}$ are also lower than experiments, however, the trend of the strangeness enhancement increasing with increase of the centrality and of the s quark content in strange particles is reproduced quite well. In Fig. 2 and 3 are given, respectively, the calculated m$_T$ spectra ($|y-y_ {cm}| \leq$ 0.5) of $\Lambda$, $\bar{\Lambda}$, $\Xi^-$, $\overline{\Xi^-}$ and $\Omega^-+\overline{\Omega^-}$ in p-Pb and Pb-Pb collisions at 158A GeV/c (open labels). The corresponding full labels in those figures are the corresponding WA97 data \cite{wa97}. One sees from figure 2 that the agreement between theory and experiment is reasonably good, except that the fluctuation in theoretical results of $\Omega^-+\overline{\Omega^-}$ m$_T$ spectrum has to be improved. However, the situations in figure 3 is much better, i.e., the agreement between theory and experiment is reasonably good. In summary, we have used a hadron and string cascade model, LUCIAE, to investigate the WA97 data of the strangeness enhancement increasing with the increase of the centrality and of the s quark content in strange particles. Relying on the mechanism of the reduction of s quark suppression in string fragmentation leads to the enhancement of strange particle yield in nucleus-nucleus collisions the WA97 data could be reproduced nicely except $\Omega$ yield in Pb+Pb collisions, which need to be studied further. \section{ACKNOWLEDGMENTS} We would like to thank T. Sj\"{o}strand for detailed instructions of using PYTHIA. This work was supported by national Natural Science Foundation of China and Nuclear Industry Foundation of China.
\subsection{Two Painlev\'e\ branches and the singular manifold method} These equations (\ref{1.2},\ref{1.3}) have the property that the leading analysis of the expansion (\ref{1.1}) provides $\alpha=1$ and $a_0=\pm 1$, where the $\pm$ sign in $a_0$ indicates that there are two possible Painlev\'e\ branches. Equations with two Painlev\'e\ branches and how to extend the SMM to them have been studied in previous papers (cf.\ \cite{EG93,EG97}). According to these, when an equation, such as (\ref{1.2}) or (\ref{1.3}), has two Painlev\'e\ branches, then the truncation of (\ref{1.1}) should be made for both branches simultaneously. This means that we should work with truncated solutions $\~y$ of the following form \cite{EG97} \begin{equation} \~y=y+{g'\over g}-{h'\over h},\label{1.4}\end{equation} where $y(x)$ is also a solution of the equation, and $g(x) =0$ is the singularity for the $+$ expansion and $h(x)=0$ the singularity for the $-$ expansion. Henceforth we shall call $g$ and $h$ {\it singular manifolds} \cite{Weiss83}. The SMM usually requires that each coefficient in the different powers of the singular manifold that arises from substitution of the truncated expansion in the differential equation be equal to zero. Due to the non-linearity of PII and PIV, the substitution of (\ref{1.4}) in (\ref{1.2}) and (\ref{1.3}) respectively provides terms that mix the powers in $h$ and $g$. To solve this problem we use a {\it decoupling ansatz} \cite{EG97} given by \begin{equation} {g'\over g}{h'\over h}=A{g'\over g}+B{h'\over h},\label{1.5}\end{equation} where $A$ and $B$ are functions of $y$, and $g$ and $h$ to be determined from the truncation itself. Furthermore, taking the derivative of (\ref{1.5}) with respect to $x$ we have \begin{equation} A'=A(r-A-B)\qquad \qquad B'=B(v-A-B),\label{1.6}\end{equation} where $v={g''/g'}$ and $r={h''/h'}$. Within this framework, the objective of this paper is to prove that for PII and PIV the improved version of the SMM including two singular manifolds provides the following results: \begin{itemize} \item[(i)] Modified versions of PII and PIV (namely mPII and mPIV). \item[(ii)] Miura transformations between PII and mPII and between PIV and mPIV. \item[(iii)] B\"acklund\ transformations for mPII and mPIV. \item[(iv)] By regarding the parameter $\ell$ as a discrete variable, discrete equations can be derived associated with mPII and mPIV. \item[(v)] Solutions of these discrete equations allow us to construct solutions of PII and PIV as a ``linear" superposition of solutions of the discrete equations. \end{itemize} \setcounter{equation}{0{Painlev\'e\ II} Substituting (\ref{1.4}) into PII (\ref{1.2}) and the use of (\ref{1.5}) to decouple the crossed terms yields \begin{eqnarray} A&=&y+\tfr{1}{2}v,\label{2.2}\\ B&=&-y+\tfr{1}{2}r,\label{2.3}\\ 0&=&v'+v^2-6y^2-x+6A(2y+B-A),\label{2.4}\\ 0&=&r'+r^2-6y^2-x-6B(2y+B-A).\label{2.5}\end{eqnarray} Using (\ref{1.6}) in (\ref{2.2}--\ref{2.3}) gives $$AB=k, $$ where $k$ is a constant. By substituting (\ref{2.2}--\ref{2.3}) into (\ref{2.4}--\ref{2.5}), the following singular manifold equations are obtained \begin{equation} v'-\tfr12{v^2}+6k-x=0,\qquad\qquad r'-\tfr12{r^2}+6k-x=0.\label{2.7}\end{equation} \subsection{A Miura transformation for modified PII} If we return to equation (\ref{1.6}), substituting $A$ and $B$ as given by (\ref{2.2}) and (\ref{2.3}), respectively, and using (\ref{2.7}), we have \begin{equation} y'-y^2-\tfr12{x}+v'+2k=0,\qquad-y'-y^2-\tfr12{x}+r'+2k=0,\label{2.8}\end{equation} which means that the expression $y'-y^2-\tfr12{x}$ depends only on the singular manifold $g$ whilst $-y'-y^2-\tfr12{x}$ depends only on $h$. Hence it is useful to define the following functions $m'$ and $n'$ as follows \begin{equation} 2m'=y'-y^2-\tfr12{x},\qquad 2n'=-y'-y^2-\tfr12{x}.\label{2.9}\end{equation} With the aid of (\ref{1.2}), these equations (\ref{2.9}) can be integrated to give \begin{equation} 2m=(y')^2-(y^2+\tfr12{x})^2-(2\ell-1)y,\qquad 2n=(y')^2-(y^2+\tfr12{x})^2-(2\ell+1)y,\label{2.10}\end{equation} respectively. In order to identify the equations that $m$ and $n$ satisfy, we take the derivative of (\ref{2.9}), which gives \begin{equation} 2m''=-4ym'+\left(\ell-{\tfr12}\right),\qquad 2n''=4yn'-\left(\ell-{\tfr12}\right).\label{2.11}\end{equation} Now if we take $y'$ from (\ref{2.9}) and $y$ from (\ref{2.11}) and substitute them in (\ref{2.10}), then we obtain the following equation for $m$ and $n$ \begin{equation} {(M_{\ell}'')^2\over M_{\ell}'}+4(M_{\ell}')^2 +2(xM_{\ell}'-M_{\ell})-{(2\ell-1)^2\over16M_{\ell}'}=0,\label{2.12}\end{equation} where $m=M_{\ell}$ and $n=M_{\ell+1}$. It is easy to prove that this equation has the Painlev\'e\ property. In fact, equation (\ref{2.12}) is the potential version of the equation 34 of the Gambier classification \cite{G10} (see also \cite{FA82,Ince,RG92}), which is commonly referred to as P34. Indeed, if we set $M_{\ell}'=Q_{\ell}$, then (\ref{2.12}) becomes \begin{equation} Q_{\ell}''-{(Q_{\ell}')^2\over 2Q_{\ell}}+4Q_{\ell}^2+xQ_{\ell}+{(\ell-{\tfr12})^2\over 8Q_{\ell}}=0,\label{2.15}\end{equation} which is precisely P34. Equations (\ref{2.9}) and (\ref{2.11}) can be written as $$\begin{array}{ll} \displaystyle 2Q_{\ell}=y'-y^2-\tfr12{x},\qquad\qquad &\displaystyle 2Q_{\ell+1}=-y'-y^2-\tfr12{x},\\[10pt] \displaystyle y={-2Q_{\ell}'+\ell-{\tfr12}\over 4Q_{\ell}},\qquad\qquad &\displaystyle y={-2Q_{\ell+1}'+\ell+{\tfr12}\over 4Q_{\ell+1}}.\end{array}$$ These can be interpreted as a Miura transformation between PII (\ref{1.2}) and P34 (\ref{2.15}) \cite{FA82}. According to Ramani and Grammaticos \cite{RG92}, P34 (\ref{2.15}) may be also be thought of as a modified PII (mPII), since the relationship between their solutions is analogous to that between solutions of the Korteweg-de Vries and modified Korteweg-de Vries equations (see also \cite{FA82}). \subsection{Auto-B\"acklund\ transformations for modified PII} By subtracting the two equations of (\ref{2.10}), we have \begin{equation}y(x)=M_{\ell}(x)-M_{\ell+1}(x),\label{2.18}\end{equation} which combined with (\ref{2.11}) gives \begin{equation} M_{\ell+1}= M_{\ell}+{4M_{\ell}''-(2\ell-1)\over 8M_{\ell}'},\qquad M_{\ell}=M_{\ell+1}+{4M_{\ell+1}''+(2\ell+1)\over 8M_{\ell+1}'},\label{2.19} \end{equation} which are auto-B\"acklund\ transformations for potential mPII. \subsection{Linear superposition for PII} Suppose we have two solutions $M_{\ell}$ and $M_{\ell+1}$ of potential mP34 (\ref{2.12}), related by the B\"acklund\ transformations (\ref{2.19}), then we can construct a solution of PII (\ref{1.2}) by using (\ref{2.18}). \subsection{Discrete equations for potential P34} If we let $\ell\rightarrow \ell-1$ in the second of equations (\ref{2.19}) then \begin{equation} M_{\ell+1}= M_{\ell}+{4M_{\ell}''-(2\ell-1)\over 8M_{\ell}'},\qquad M_{\ell-1}= M_{\ell}+{4M_{\ell}''+(2\ell-1)\over 8M_{\ell}'}.\label{2.20}\end{equation} Adding and subtracting these two equations gives \begin{equation}M_{\ell}'=-\,{2\ell-1\over 4(M_{\ell+1}-M_{\ell-1})},\qquad M_{\ell}''=-\,{(2\ell-1)(M_{\ell+1}+M_{\ell-1}-2M_{\ell})\over 4(M_{\ell+1}-M_{\ell-1})}.\label{2.21}\end{equation} By substituting (\ref{2.21}) in (\ref{2.12}), the result is the nonautonomous discrete equation \begin{eqnarray*} &&(M_{\ell-1}-M_{\ell+1})\left\{\left(\ell-{\tfr12}\right)\left[(M_{\ell-1}-M_{\ell})(M_{\ l+1}-M_{\ell})+\tfr12{x}\right]\right.\\ &&\qquad\left. -M_{\ell}(M_{\ell-1}-M_{\ell+1})\right\} +\left(\ell-{\tfr12}\right)^2=0, \end{eqnarray*} where the parameter $\ell$ can be interpreted as the discrete variable. \setcounter{equation}{0{Painlev\'e\ IV} In an analogous way as we studied PII in the previous section, in this section we study PIV (\ref{1.3}) by using the truncated expansion (\ref{1.4}) together with the decoupling ansatz (\ref{1.5}). The result is \begin{eqnarray} A&=&y+x+\tfr{1}{2}v,\label{3.2}\\ B&=&-y-x+\tfr{1}{2}r,\label{3.3}\\ 0&=&(24x+6v)(A-y)-4x^2+8\ell+v^2+2(v'+y')\nonumber\\ &&\qquad-18y^2-16A^2+36yA+8AB-2rA,\label{3.4}\\ 0&=&(-24x+6r)(B+y)-4x^2+8\ell+r^2+2(r'-y')\nonumber\\ &&\qquad-18y^2-16B^2-36yB+8AB-2vB.\label{3.5} \end{eqnarray} Equations (\ref{3.2}--\ref{3.3}) combined with (\ref{1.6}) gives $AB=k$, with $k$ a constant. Then substituting (\ref{3.2}--\ref{3.3}) into (\ref{3.4}--\ref{3.5}), for the singular manifolds we have the equations \begin{equation} v'-\tfr12{v^2}+6k+2x^2+8\ell-2=0,\qquad r'-\tfr12{r^2}+6k+2x^2+8\ell+2=0.\label{3.7}\end{equation} \subsection{A Miura transformation for modified PIV} The substitution of (\ref{2.2}-\ref{2.3}) into (\ref{1.6}) gives \begin{equation} y'-y^2-2xy+v'+2(k+2\ell)=0,\qquad -y'-y^2-2xy+r'+2(k+2\ell)=0.\label{3.8}\end{equation} It is reasonable therefore to define new functions $m$ and $n $ in the following form \begin{equation} 2m'=y'-y^2-2xy,\qquad 2n'=-y'-y^2-2xy.\label{3.9}\end{equation} These equations can be integrated, taking (\ref{1.3}) into account, as \begin{equation} \begin{array}{l} \displaystyle 2m={{(y')^2+2\mu -(y^2+2xy)^2}\over 4y}+(2\ell+1)y,\\[10pt] \displaystyle 2n={{(y')^2+2\mu -(y^2+2xy)^2}\over 4y}+(2\ell-1)y. \end{array} \label{3.10}\end{equation} Furthermore, the derivation of (\ref{3.9}) provides \begin{equation} \begin{array}{l} \displaystyle 2m''={2(m')^2+\mu \over y}-2(m'+2\ell+1)y,\\[10pt] \displaystyle 2n''=-{2(n')^2+\mu \over y}+2(n'+2\ell-1)y. \end{array}\label{3.11}\end{equation} Taking $y'$ as defined by (\ref{3.9}) and substituting in (\ref{3.10}) we have \begin{equation} \begin{array}{l} \displaystyle 4(m-xm')={2(m')^2+\mu \over y}+2(m'+2\ell+1)y,\\[10pt] \displaystyle 4(n-xn')={2(n')^2+\mu \over y}+2(n'+2\ell-1)y. \end{array}\label{3.12}\end{equation} Adding and subtracting (\ref{3.11}) and (\ref{3.12}) the result is \begin{equation} \begin{array}{ll} \displaystyle y={2(m-xm')-m''\over 2(m'+2\ell+1)},\qquad & \displaystyle{1\over y}={2(m-xm')+m''\over 2(m')^2+\mu },\\[10pt] \displaystyle y={2(n-xn')+n''\over 2(n'+2\ell-1)},\qquad & \displaystyle {1\over y}={2(n-xn')-n''\over 2(n')^2+\mu }.\end{array}\label{3.13}\end{equation} We can eliminated $y$ by multiplication of the left equation by the right one. Thus we then have \begin{equation} \left(M_{\ell}''\right)^2-4\left(M_{\ell}-xM_{\ell}'\right)^2 +2\left(M_{\ell}'+2\ell+1\right)\left[2\left(M_{\ell}'\right)^2+\mu \right]=0,\label{3.14}\end{equation} where $m=M_{\ell}$ and $n=M_{\ell-1}$. Equation (\ref{3.14}) is of Painlev\'e\ type, and it can be considered as the modified version of PIV (mPIV). The Miura transformation that relates PII and mPII are (\ref{3.11}), which can now be written as \begin{equation} 2M_{\ell}'=y'-y^2-2xy,\qquad 2M_{\ell-1}'=-y'-y^2-2xy.\label{3.15}\end{equation} \subsection{Auto-B\"acklund\ transformations} Subtracting the two equations of (\ref{3.10}), and with the aid of (\ref{3.15}), we have \begin{equation} y=M_{\ell}-M_{\ell-1},\label{3.17}\end{equation} then combined with the left part of (\ref{3.13}) yields to \begin{equation} \begin{array}{l} \displaystyle M_{\ell-1}=M_{\ell}+{M_{\ell}''-2(M_{\ell}-xM_{\ell}')\over 2(M_{\ell}'+2\ell+1)},\\[10pt] \displaystyle M_{\ell}=M_{\ell-1}+{M_{\ell-1}''+2(M_{\ell-1}-xM_{\ell-1}')\over 2(M_{\ell-1}'+2\ell-1)}, \end{array}\label{3.18}\end{equation} which are auto-B\"acklund\ transformation for mPIV (\ref{3.14}). \subsection{Linear superposition for PIV} If we have two solutions $M_{\ell}$ and $M_{\ell+1}$ of potential mPIV (\ref{3.14}), related by the B\"acklund\ transformations (\ref{3.18}), then we can construct a solution of PIV (\ref{1.3}) by using (\ref{3.17}). \subsection{Discrete equations for mPIV} As we did in the previous section, the B\"acklund\ transformations (\ref{3.18}) can be written as \begin{equation} \begin{array}{l} \displaystyle M_{\ell-1}=M_{\ell}+{M_{\ell}''-2(M_{\ell}-xM_{\ell}')\over 2(M_{\ell}'+2\ell+1)},\\[10pt] \displaystyle M_{\ell+1}=M_{\ell} +{M_{\ell}''+2(M_{\ell}-xM_{\ell}')\over 2(M_{\ell}'+2\ell+1)}. \end{array}\label{3.19}\end{equation} Then by addition and subtraction, it is easy to show that \begin{equation} \begin{array}{l} \displaystyle M_{\ell}'={2M_{\ell}-(2\ell+1)(M_{\ell+1}-M_{\ell-1})\over M_{\ell+1}-M_{\ell-1}+2x},\\[10pt] \displaystyle M_{\ell}''={(M_{\ell}'+2\ell+1)(M_{\ell+1}+M_{\ell-1}-2M_{\ell})\over M_{\ell+1}-M_{\ell-1}+2x},\end{array}\label{3.20}\end{equation} whose substitution in (\ref{3.15}) yields the discrete equation \begin{eqnarray*} &&(M_{\ell+1}-M_{\ell-1}+2x)(M_{\ell}-M_{\ell-1})(M_{\ell}-M_{\ell+1})\left[M_{\ell}+2\left(\ell +{\tfr12}\right)x\right]\\ &&\qquad+\left\{2\left[M_{\ell}-\left(\ell+{\tfr12}\right)\right] (M_{\ell+1}-M_{\ell-1})\right\}^2 +\tfr12{x}\left(M_{\ell+1}-M_{\ell-1}+2x\right)^2=0. \end{eqnarray*} where $\ell$ is the discrete parameter. \setcounter{equation}{0{Conclusions} In this paper we have derived Miura transformations, modified equations and associated discrete equations for the second and fourth Painlev\'e\ equations. Recently there have been several studies of the derivation of discrete equations, and in particular the discrete Painlev\'e\ equations, from B\"acklund\ transformations of the (continuous) Painlev\'e\ equations \cite{refFGR,refGNR,refGra,GR98,refGTii,refGTiii,NSKGR96,refTsegb,refTsegc}. Hierarchies of solutions of PII and PIV are well-known (cf.\ \cite{refAirault,BCH95,PAC90,refGromaki,Gromak,refLuka,Murata,refOka,refUW}). Since there is an explicit relationship between PII and PIV and discrete equations, then these Hierarchies of solutions of PII and PIV also satisfy difference equations in addition to the ordinary differential equations. This is analogous to the situation for the classical special functions, such as Bessel, hypergeometric, Legendre, Weber-Hermite and Whittaker functions, which satisfy both an ordinary differential equation and a recurrence relation, which is a discrete equation. This provides further evidence that the Painlev\'e\ equations may be thought of as nonlinear special functions and that there is a deep relationship between the classical special functions, the Painlev\'e\ equations and the discrete Painlev\'e\ equations (see, for example, \cite{refTRGK}). \def\frenchspacing\it{\frenchspacing\it}
\section*{Introduction} Our introduction is rather brief since this paper mainly is a continuation of \cite{prep:wir}. There we a gave a formula for a Lorentz invariant renormalization in one coordinate. We use the same methods to construct a covariant solution for more variables. We review the EG subtraction and give some useful formulas in section~\ref{sec:sub}. The cohomological analysis that leads to a group covariant solution is reviewed in section~\ref{sec:g}. In sections~\ref{sec:tensor},~\ref{sec:spinor} we give an inductive construction for these solutions in the case of tensorial and spinorial Lorentz covariance. Since the symmmetry of the one variable problem is absent in general some permutation group calculus will enter. Necessary material is in the appendix. In the last section~\ref{sec:bphz} we give a covariant BPHZ subtraction for arbitrary (totally spacelike) momentum by fourier transformation. \section{The EG subtraction} \label{sec:sub} We review the extension procedure in the EG approach \cite{pap:ep-gl}. A more general introduction can be found in \cite{pap:prange1}, a generalization to manifolds in \cite{proc:fred-brun,prep:fred-brun2}. Let $ \ensuremath{\mathcal{D}}^{\omega}(\Rd) $ be the subspace of test functions vanishing to order $ \omega $ at $ 0 $. As usual, any operation on distributions is defined by the corresponding action on testfunctions. Define \begin{gather} \W{\omega}{w}:\ensuremath{\mathcal{D}}(\Rd)\rightarrow\ensuremath{\mathcal{D}}^{\omega}(\Rd), \quad\varphi\rightarrow\W{\omega}{w}\varphi, \notag \\ \left(\W{\omega}{w}\varphi\right)(x)=\varphi(x)-w(x)\sum_{|\alpha|\leq\omega} \frac{x^{\alpha}}{\alpha!}\ensuremath{\partial}_{\alpha}\left(\varphi w^{-1}\right)(0), \label{def:W} \end{gather} with $ w\in\ensuremath{\mathcal{D}}(\Rd), w(0)\not=0 $. If ${^0t}$ is a distribution on $\ensuremath{\mathcal{D}}(\Rdon)$ with singular order $ \omega $ then ${^0t}$ can be defined uniquely on $ \ensuremath{\mathcal{D}}^{\omega}(\Rd) $, and \begin{equation} \scp{\tR{\omega}{w}}{\varphi}\doteq\scp{{^0t}}{\W{\omega}{w}\varphi} \label{def:tR} \end{equation} defines an extension -- called renormalization -- of $^0t $ to the whole testfunction space, $ \tR{\omega}{w}\in\ensuremath{\mathcal{D}}'(\Rd) $. With the Leibnitz rule \begin{equation} \ensuremath{\partial}_{\gamma}(fg) =\gamma!\sum_{\mu+\nu=\gamma}\frac{1}{\mu!\nu!}\ensuremath{\partial}_{\mu}f\,\ensuremath{\partial}_{\nu}g, \label{eq:leibnitz} \end{equation} we find \begin{equation} \ensuremath{\partial}_{\gamma}\biggl(w\sum_{|\alpha|\leq\omega} \frac{x^{\alpha}}{\alpha!}\ensuremath{\partial}_{\alpha}(\varphi w^{-1})(0)\biggr)(0) =\ensuremath{\partial}_{\gamma}\varphi(0), \label{eq:diffrest} \end{equation} for $ |\gamma|\leq\omega $, verifying the projector properties of $W$. The $W$-operation (\ref{def:W}) is simplified if we require $w(0)=1$ and $\partial_{\alpha}w(0)=0$, for $0<|\alpha|\leq\omega$ (this was our assumption in \cite{prep:wir}). It can be achieved by the following $V$-operation ($\ensuremath{\partial}_\mu w^{-1}$ means $\ensuremath{\partial}_{\mu}(w^{-1})$): \begin{equation} V_{\omega}:\ensuremath{\mathcal{D}}(\Rd)\mapsto\ensuremath{\mathcal{D}}(\Rd),\quad (V_{\omega}w)(x)\doteq w(x)\sum_{|\mu|\leq \omega}\frac{x^{\mu}}{\mu!}\ensuremath{\partial}_{\mu}w^{-1}(0), \label{def:V} \end{equation} where $w(0)\not=0$ is still assumed. We can write $W$ as \begin{equation} \label{eq:varW} \left(\W{\omega}{w}\varphi\right)(x) =\varphi(x)-\sum_{|\alpha|\leq\omega} \frac{x^{\alpha}}{\alpha!}V_{\omega-|\alpha|}w\,\ensuremath{\partial}_{\alpha}\varphi(0). \end{equation} The extension \eqref{def:tR} is not unique. We can add any polynomial in derivatives of $\delta$ up to order $\omega$: \begin{align} \scp{\tR{\omega}{w}}{\varphi}&=\scp{\tiltR{\omega}{w}}{\varphi} +\sum_{|\alpha|\leq\omega} \frac{a^{\alpha}}{\alpha!}\ensuremath{\partial}_{\alpha}\varphi(0), \\ \intertext{or rearranging the coefficients} &=\scp{\tiltR{\omega}{w}}{\varphi} +\sum_{|\alpha|\leq\omega} \frac{c^{\alpha}}{\alpha!}\ensuremath{\partial}_{\alpha}\left(\varphi w^{-1}\right)(0) \label{def:tiltR} \end{align} Since $\W{\omega}{w}( w x^\alpha) = \W{\omega}{w} (x^\alpha V_{\omega-|\alpha|}w) = 0$ for $|\alpha| \leq \omega$, $c$ resp.\ $a$ are given by \begin{align} a^\alpha&=\scp{\tiltR{\omega}{w}}{x^\alpha V_{\omega-|\alpha|}w}, & c^\alpha&=\scp{\tiltR{\omega}{w}}{x^\alpha w}. \label{eq:cundaaust} \end{align} They are related through: \begin{align} a^\alpha&=c^\alpha\sum_{|\mu|\leq\omega-|\alpha|} \frac{c^{\mu}}{\mu!}\ensuremath{\partial}_{\mu}w^{-1}(0), & c^\alpha&=a^\alpha\sum_{|\mu|\leq\omega-|\alpha|} \frac{a^{\mu}}{\mu!}\ensuremath{\partial}_{\mu}w(0), & 1\leq|\alpha|\leq\omega, \notag \\ a^0&=\sum_{|\mu|\leq\omega} \frac{c^{\mu}}{\mu!}\ensuremath{\partial}_{\mu}w^{-1}(0), & c^0&=\sum_{|\mu|\leq\omega} \frac{a^{\mu}}{\mu!}\ensuremath{\partial}_{\mu}w(0). \label{eq:causa} \end{align} The equation for $a$ follows from the Leibnitz rule in \eqref{def:tiltR}, while the equation for $c$ is derived from \eqref{eq:cundaaust}. In quantum field theory the coefficients $a$ are called counter terms. They are not arbitrary. They have to be chosen in such a way that Lorentz covariance of $^{0}t$ is preserved in the extension. This follows in the next sections. The remaining freedom is further restricted by discrete symmetries like permutation symmetry or $C,P$ and $T$ symmetries. At the end gauge invariance or renormalization constraints will fix the extension uniquely. But up to now there is no local prescription of the latter. \section{The $G$-covariant extension} \label{sec:g} We will first define the notion of a $G$-covariant distribution. So let $G$ be a linear transformation group on \Rd\ i.e. $x\mapsto gx$, $g\in G$. Then \begin{equation} x^{\alpha}\mapsto{g^{\alpha}}_{\beta}x^{\beta}=(gx)^{\alpha} \label{} \end{equation} denotes the corresponding tensor representation. $G$ acts on functions in the following way: \begin{equation} (g\varphi)(x)\doteq\varphi(g^{-1}x), \label{def:gphi} \end{equation} so that \ensuremath{\mathcal{D}}\ is made a $G$-module. We further have \begin{align} g(\varphi\psi)&=(g\varphi)(g\psi),\label{eq:gprod} \\ x^{\alpha}\ensuremath{\partial}_{\alpha}(g^{-1}\varphi) &=(gx)^{\alpha}g^{-1}(\ensuremath{\partial}_{\alpha}\varphi), \label{eq:1} \\ x^{\alpha}\ensuremath{\partial}_{\alpha}(g^{-1}\varphi)(0) &=(gx)^{\alpha}\ensuremath{\partial}_{\alpha}\varphi(0). \label{eq:2} \end{align} Now assume we have a distribution ${^0t}\in\ensuremath{\mathcal{D}}'(\Rdon)$ that transforms covariantly under the Group $G$ as a density, i.e. \begin{equation} {^0t}(gx)|\det g|=D(g){^0t}(x), \label{eq:tcov} \end{equation} where $D$ is the corresponding representation. That means: \begin{equation} \scp{{^0t}}{g\psi}=\scp{D(g){^0t}}{\psi}\doteq D(g)\scp{{^0t}}{\psi}. \label{def:tcov} \end{equation} We will now investigate what happens to the covariance in the extension process. We compute: \begin{align} D(g)&\scp{\tR{\omega}{w}}{g^{-1}\varphi} -\scp{\tR{\omega}{w}}{\varphi} \label{eq:cov1} \\ &=D(g)\scp{{^0t}}{\W{\omega}{w}g^{-1}\varphi} -\scp{{^0t}}{\W{\omega}{w}\varphi} \label{eq:cov2} \\ &\stackrel{\eqref{eq:varW}}{=} D(g)\scp{{^0t}}{g^{-1}\varphi-\sum_{|\alpha|\leq\omega} \frac{x^{\alpha}}{\alpha!}V_{\omega-|\alpha|}w\,\ensuremath{\partial}_{\alpha}(g^{-1}\varphi)(0)} -\scp{{^0t}}{\W{\omega}{w}\varphi} \label{eq:cov3} \\ &\stackrel{(\ref{eq:gprod},\ref{eq:2})}{=} D(g)\scp{{^0t}}{g^{-1}\biggl(\varphi- \sum_{|\alpha|\leq\omega} \frac{x^{\alpha}}{\alpha!}\left(gV_{\omega-|\alpha|}w\right) \ensuremath{\partial}_{\alpha}\varphi(0)\biggr)} -\scp{{^0t}}{\W{\omega}{w}\varphi} \label{eq:cov4} \displaybreak[0]\\ &\stackrel{(\ref{def:tcov})}{=} \sum_{|\alpha|\leq\omega} \scp{{^0t}}{x^{\alpha}(\eins-g)(V_{\omega-|\alpha|}w)} \frac{\ensuremath{\partial}_{\alpha}\varphi(0)}{\alpha!} \label{eq:cov8} \\ &\doteq\sum_{|\alpha|\leq\omega}b^{\alpha}(g) \frac{\ensuremath{\partial}_{\alpha}\varphi(0)}{\alpha!}. \label{def:balpha} \end{align} Then (\ref{def:balpha}) defines a map from $G$ to a finite dimensional complex vectorspace. Now we follow \cite{prep:stora-pop}, \cite{bk:scharf}[chapter 4.5]: Applying two transformations \begin{align} b^{\alpha}(g_{1}g_{2}) &=\scp{{^0t}}{x^{\alpha}(\eins-g_{1}g_{2})(V_{\omega-|\alpha|}w)} \\ &=\scp{{^0t}}{x^{\alpha}\left((\eins-g_{1}) +g_{1}(\eins-g_{2})\right)(V_{\omega-|\alpha|}w)} \\ &=b^{\alpha}(g_{1})+|\det g_{1}|\scp{{^0t}(g_{1}x)} {(g_{1}x)^{\alpha}(\eins-g_{2})(V_{\omega-|\alpha|}w)}, \end{align} and omitting the indices we see $b(g_{1}g_{2})=b(g_{1})+D(g_{1})g_{1}b(g_{2}), $ which is a 1-cocycle for $ b(g) $. Its trivial solutions are the 1-coboundaries \begin{equation} b(g)=(\eins-D(g)g)a, \label{def:cobound} \end{equation} and these are the only ones if the first cohomology group of $G$ is zero. In that case we can restore $G$-covariance by adding the following counter terms: \begin{equation} \scp{\tR{\omega}{w}^{G-\mathrm{cov}}}{\varphi} \doteq\scp{\tR{\omega}{w}}{\varphi} +\sum_{|\alpha|\leq\omega} \frac{1}{\alpha!}a^{\alpha}(w)\ensuremath{\partial}_{\alpha}\varphi(0). \label{def:tRGcov} \end{equation} The task is to determine $a$ from (\ref{def:cobound}) and (\ref{eq:cov8}): \begin{equation} \scp{{^0t}}{x^{\alpha}(\eins-g)(V_{\omega-|\alpha|}w)} =\bigl[(\eins-D(g)g)a\bigr]^\alpha \label{def:a} \end{equation} \section{Tensorial Lorentz covariance} \label{sec:tensor} The first cohomology group of $\Lcal_{+}^{\uparrow}$ vanishes \cite{bk:scharf}[chapter 4.5 and references there]. We determine $ a $ from the last equation. The most simple solution appears in the case of Lorentz invarinance in one coordinate. This situation was completely analyzed in \cite{prep:wir} for $\ensuremath{\partial}_{\alpha}w(0) = \delta_{\alpha}^{0}$. The following two subsections generalize the results to arbitrary $w,\ w(0)\not=0$. \subsection{Lorentz invariance in \Rv} \label{subsec:Linv4d} If we expand the index $\alpha$ into Lorentz indices $\mu_1, \dots, \mu_n$, \eqref{def:a} is symmetric in $\mu_1, \dots, \mu_n$ and therefore $a$ will be, too. We just state our result from \cite{prep:wir} which is modified through the generalization for the choice of $w$: \begin{multline} a^{(\mu _1 \ldots \mu_n)} = \frac{ (n-1)!!}{(n+2)!!} \sum_{s=0}^{\left[\frac{n-1}{2} \right] } \frac{(n-2s)!!}{(n-2s-1)!!} \eta^{(\mu_1 \mu _2} \ldots \eta^{\mu_{2s-1}\mu _{2s}} \times \\ \times \left\langle {^0t}, (x^2)^s x ^{\mu _{2s+1}} \ldots x ^{\mu_{n-1}} \left(x^2\ensuremath{\partial}^{\mu_{n})}-x^{\mu_{n})}x^{\beta}\ensuremath{\partial}_{\beta}\right) V_{\omega-n}w\right\rangle, \label{eq:amun} \end{multline} if we choose the fully contracted part of $ a $ to be zero in case of $ n $ being even. We used the notation \begin{align*} b^{(\mu_{1}\dots\mu_{n})}&=\frac{1}{n!}\sum_{\pi\in S_{n}} b^{\mu_{\pi(1)}\dots\mu_{\pi(n)}}, & b^{[\mu_{1}\dots\mu_{n}]}&=\frac{1}{n!}\sum_{\pi\in S_{n}} \sgn(\pi)b^{\mu_{\pi(1)}\dots\mu_{\pi(n)}}, \end{align*} for the totally symmetric resp.\ antisymmetric part of a tensor. \subsection{Dependence on $w$} \label{subsubsec:scale} Performing a functional derivation of the Lorentz invariant extension with respect to $w$, only Lorentz invariant counter terms appear. \begin{definition} The functional derivation is given by: \begin{equation*} \scp{\frac{\delta}{\delta g}F(g)}{\psi}\doteq \left.\frac{\dif}{\dif\lambda}F(g+\lambda\psi)\right\vert_{\lambda=0}. \end{equation*} \end{definition} This definition implies the following functional derivatives: \begin{align} \scp{\frac{\delta}{\delta w}\tR{\omega}{w}(\varphi)}{\psi} &=-\sum_{|\alpha|\leq\omega}\frac{1}{\alpha!} \scp{\tR{\omega}{w}}{x^\alpha\psi} \ensuremath{\partial}_\alpha\left(\varphi w^{-1}\right)(0),\label{eq:derivtR} \\ \scp{\frac{\delta}{\delta w}\scp{S}{V_\omega w}}{\psi} &=\sum_{|\alpha|\leq\omega}\frac{1}{\alpha!} \scp{S}{\W{\omega}{w}(x^\alpha\psi)} \ensuremath{\partial}_\alpha w^{-1} (0), \label{eq:derivS} \end{align} for any distribution $S$. \begin{proof} We show how to derive the first relation. Inserting the definition we find: \begin{align*} \left.\frac{\dif}{\dif\lambda} \tR{\omega}{w+\lambda\psi}(\varphi)\right\vert_{\lambda=0} &=\scp{^0t}{-\psi\sum_{|\alpha|\leq\omega} \frac{x^\alpha}{\alpha!}\ensuremath{\partial}_\alpha\left(\varphi w^{-1}\right)(0) +w\sum_{|\alpha|\leq\omega}\frac{x^\alpha}{\alpha!} \ensuremath{\partial}_\alpha\left(\varphi\psi w^{-2}\right)(0)}, \\ \intertext{using Leibnitz rule and rearranging the summation in the second term,} &\begin{aligned} =\sum_{|\alpha|\leq\omega}\frac{1}{\alpha!} \Biggl\langle{^0t},-&\psi x^\alpha\ensuremath{\partial}_\alpha\left(\varphi w^{-1}\right)(0) \\ +&w x^\alpha\ensuremath{\partial}_\alpha\left(\varphi w^{-1}\right)(0) \sum_{|\nu|\leq\omega-|\alpha|}\frac{x^\nu}{\nu!} \ensuremath{\partial}_\nu\left(\psi w^{-1}\right)(0) \Biggr\rangle \end{aligned} \\ &=-\sum_{|\alpha|\leq\omega}\frac{1}{\alpha!} \scp{^0t}{x^\alpha\W{\omega-|\alpha|}{w}\psi} \ensuremath{\partial}_\alpha\left(\varphi w^{-1}\right)(0) \\ &=-\sum_{|\alpha|\leq\omega}\frac{1}{\alpha!} \scp{\tR{\omega}{w}}{x^\alpha\psi}\ensuremath{\partial}_\alpha\left(\varphi w^{-1}\right)(0), \end{align*} where we used the relation $x^\alpha\W{\omega-|\alpha|}{w}\varphi = \W{\omega}{w}(x^\alpha\varphi)$ on the last line. The second equation follows from a similar calculation. \end{proof} We calculate the dependce of $a$ on $w$. With \eqref{eq:derivS} we get: \begin{multline*} \frac{\delta}{\delta w} a^{(\mu_1\dots\mu_n)}(w) = \frac{(n-1)!!}{(n+2)!!} \sum_{s=0}^{\left[\frac{n-1}{2} \right] } \frac{(n-2s)!!}{(n-1-2s)!!} \eta^{(\mu_1\mu_2} \dots \eta^{\mu _{2s-1} \mu _{2s}} \times \\ \times\sum_{|\beta|\leq\omega-n}\frac{1}{\beta!} \scp{^0t}{(x^2)^s x^{\mu_{2s+1}} \dots x ^{\mu_{n-1}} \left(x^2\ensuremath{\partial}^{\mu _{n})} -x^{\mu_{n})}x^{\beta}\ensuremath{\partial}_{\beta}\right)\W{\omega-n}{w}(x^\beta\psi)} \ensuremath{\partial}_\beta w^{-1}(0). \label{eq:damun} \end{multline*} To condense the notation we again use $\beta$ as a multiindex. Since $\W{\omega-n}{w}(x^\beta\psi)$ is sufficient regular, we can put the $x$'s and derivatives on the left and the same calculation like in \cite{prep:wir} applies. The result is \begin{multline*} \scp{\frac{\delta}{\delta w}a^{\mu_{1}\cdots\mu_{n}}(w)}{\psi}= \sum_{|\beta|\leq\omega-n}\frac{\ensuremath{\partial}_\beta w^{-1}(0)}{\beta!} \Biggl[ \scp{\tR{\omega}{w}}{x^{\mu_{1}}\cdots x^{\mu_{n}}x^\beta\psi} + \\ -\begin{cases} 0,&n\text{ odd}, \\ \frac{2(n-1)!!}{(n+2)!!} \scp{\tR{\omega}{w}}{(x^2)^\frac{n}{2}x^\beta\psi} \eta^{(\mu_1\mu_2}\cdots\eta^{\mu_{n-1}\mu_n)}, &n\text{ even.} \end{cases} \Biggr] \end{multline*} Using this result and \eqref{eq:derivtR} we find: \begin{align*} \scp{\frac{\delta}{\delta w}\scp{\tRli{\omega}{w}}{\phi}}{\psi} &=-\sum_{\substack{n=0\\ n\text{ even}}}^{\omega}\frac{d_{n}}{n!} \square^{\frac{n}{2}}\phi(0), \\ d_{n}&\doteq \frac{2(n-1)!!}{(n+2)!!}\sum_{|\beta|\leq\omega-n}\frac{1}{\beta!} \scp{\tR{\omega}{w}}{(x^{2})^{\frac{n}{2}}x^\beta\psi} \ensuremath{\partial}_\beta w^{-1}(0), \end{align*} where we set $ d_{0}=1 $. \subsection{General Lorentz covariance} If the distribution ${^{0}t}$ depends on more than one variable, ${^{0}t}x^{\alpha}$ will not be symmetric in all Lorentz indices in general. Since $x^{\alpha}$ transforms like a tensor, it is natural to generalize the discussion to the case, where ${^{0}t}$ transforms like a tensor, too. Assume rank$( {^0t} )=r$, then $D(g)g$ is the tensor representation of rank $p=r+n,n=|\alpha|$, in (\ref{def:a}). From now on we will omit the indices. So if $t\in\ensuremath{\mathcal{D}}(\Rvmon)$, we denote by \xbar\ -- formerly $x^{\alpha}$ -- a tensor of rank $n$ built of $x_{1},\dots,x_{m}$. To solve \eqref{def:a} we proceed like in \cite{prep:wir}. Since the equation holds for all $g$ we will solve for $a$ by using Lorentz transformations in the infinitesimal neighbourhood of $\eins$. If we take $\theta_{\alpha\beta}=\theta_{[\alpha\beta]} $ as six coordinates these transformations read: \begin{equation} g \approx\eins+\frac{1}{2}\theta_{\alpha\beta}l^{\alpha\beta}, \label{eq:gbyone} \end{equation} with the generators \begin{equation} {(l^{\alpha\beta})^{\mu}}_{\nu} =\eta^{\alpha\mu}\delta^{\beta}_{\nu}-\eta^{\beta\mu}\delta^{\alpha}_{\nu}. \label{def:generator} \end{equation} Then, for an infinitesimal transformation one finds from \eqref{def:a}: \begin{gather} B^{\alpha\beta}\doteq 2\biggl\langle{^0t},\xbar\sum_{j=1}^{m} x^{[\alpha}_j\ensuremath{\partial}^{\beta]}_j(V_{\omega-n}w)\biggr\rangle =(l^{\alpha\beta}\otimes\dots\otimes\eins+\dots +\eins\otimes\dots\otimes l^{\alpha\beta})a, \label{eq:atensor} \end{gather} $\alpha,\beta$ being Lorentz four-indices. In \cite{prep:wir} our ability to solve that equation heavily relied on the given symmetry, which is in general absent here. Nevertheless we can find an inductive construction for $a$, corresponding to equation (29) in \cite{prep:wir}. We build one Casimir operator on the r.h.s. (the other one is always zero, since we are in a $(1/2,1/2)^{\otimes p}$ representation). \subsubsection*{The case $p=1$} Just to remind that $p$ is the rank of $\xbar t$, this occurs if either $t$ is a vector and $\xbar=1, (n=0)$, or $t$ is a scalar and $\xbar=x_{1},\dots,x_{m}$. (\ref{eq:atensor}) gives: \begin{equation} \frac{1}{2}l_{\alpha\beta}B^{\alpha\beta} =\frac{1}{2}l_{\alpha\beta}l^{\alpha\beta}a =-3\eins a, \label{eq:cas1} \end{equation} since the Casimir operator is diagonal in the irreducible $(1/2,1/2)$ representation. \subsubsection*{The case $p=2$} We get \begin{equation} \frac{1}{2}(l_{\alpha\beta}\otimes\eins +\eins\otimes l_{\alpha\beta})B^{\alpha\beta} =(-6\eins+l_{\alpha\beta}\otimes l^{\alpha\beta})a. \label{eq:acas2} \end{equation} Since $a$ is a tensor of rank 2, let us introduce the projector onto the symmetric resp. antisymmetric part and the trace: \begin{align} {P_{S}^{\mu\nu}}_{\rho\sigma} &=\frac{1}{2}(\delta^{\mu}_{\rho}\delta^{\nu}_{\sigma} +\delta^{\nu}_{\rho}\delta^{\mu}_{\sigma}), & {P_{A}^{\mu\nu}}_{\rho\sigma} &=\frac{1}{2}(\delta^{\mu}_{\rho}\delta^{\nu}_{\sigma} -\delta^{\nu}_{\rho}\delta^{\mu}_{\sigma}), \label{def:P} & {P_\eta^{\mu\nu}}_{\rho\sigma} &=\frac{1}{4}\eta^{\mu\nu}\eta_{\rho\sigma}, \\ P^2&=P, & P_S+P_A&=\eins, & P_S-P_A&=\tau, \end{align} where $\tau$ denotes the permutation of the two indices. Using (\ref{def:generator}), we find \begin{equation} \frac{1}{2}l_{\alpha\beta}\otimes l^{\alpha\beta}=4P_{\eta}-\tau. \label{eq:genten} \end{equation} Now we insert (\ref{eq:genten}) into (\ref{eq:acas2}). The trace part will be set to zero again. Acting with $P_A$ and $P_S$ on the resulting equation gives us two equations for the antisymmetric and symmetric part respectively. This yields: \begin{equation} a=-\frac{1}{16}(P_S+2P_A)(l_{\alpha\beta}\otimes\eins +\eins\otimes l_{\alpha\beta})B^{\alpha\beta}. \label{eq:a2} \end{equation} \subsubsection*{Inductive assumption} Now we turn back to equation \eqref{def:a}. We note that any contraction commutes with the (group) action on the rhs. Hence, if we contract \eqref{eq:atensor}, we find on the rhs: \begin{multline*} \eta_{ij}(l^{\alpha\beta}\otimes\dots\otimes\eins+\dots +\eins\otimes\dots\otimes l^{\alpha\beta})a= \\ (l^{\alpha\beta}\otimes\dots\otimes\eins+\dots +\widehat{i}+\dots+\widehat{j}+\dots +\eins\otimes\dots\otimes l^{\alpha\beta})(\eta_{ij}a), \end{multline*} where $i,j$ denote the positions of the corresponding indices, and the $\widehat{\text{hat}}$ means omission. Therefore the rank of \eqref{eq:atensor} is reduced by two and we can proceed inductively. With the cases $p=1, p=2$ solved, we assume that all possible contractions of $a$ are known. \subsubsection*{Induction step} Multiplying (\ref{eq:atensor}) with the generator and contracting the indices yields: \begin{multline} \biggl(3p\eins+2\sum_{\tau\in S_p}\tau\biggr)a =-\frac{1}{2}(l_{\alpha\beta}\otimes\dots\otimes\eins+\dots +\eins\otimes\dots\otimes l_{\alpha\beta})B^{\alpha\beta} +8\sum_{i<j\leq p}{P_\eta}_{ij}a. \label{eq:acasp} \end{multline} The transposition $\tau$ acts on $a$ by permutation of the corresponding indices. For a general $\pi\in S_p$ the action on $a$ is given by: $\pi a^{\mu_1\dots\mu_p}=a^{\mu_{\pi^{-1}(1)}\dots\mu_{\pi^{-1}(p)}}$. In order to solve this equation we consider the representation of the symmetric groups. We give a brief summary of all necessary ingredients in appendix~\ref{app:permrep}. So let $k_\tau\doteq\sum_{\tau\in S_p}\tau$ be the sum of all transpositions of $S_p$. Then $k_\tau$ is in the center of the group algebra $\Acal_{S_p}$. It can be decomposed into the idempotents $e_{(m)}$ that generate the irreducible representations of $S_p$ in $\Acal_{S_p}$. \begin{equation} k_\tau=h_\tau\sum_{(m)}\frac{1}{f_{(m)}}\chi_{(m)}(\tau)e_{(m)}. \label{eq:decompktau} \end{equation} The sum runs over all partitions $(m)=(m_1,\dots,m_r), \sum_{i=1}^r m_i=p, m_1\geq m_2\geq\dots\geq m_r$ and $h_\tau=\frac{1}{2} p(p-1)$ is the number of transpositions in $S_p$. $\chi_{(m)}$ is the character of $\tau$ in the representation generated by $e_{(m)}$ which is of dimension $f_{(m)}$. We use \eqref{eq:decompktau}, the ortogonality relation $e_{(m)}e_{(m')} =\delta_{(m)(m')}$ and the completeness $\sum_{(m)}e_{(m)}=\eins$ in (\ref{eq:acasp}). The expression in brackets on the l.h.s may be orthogonal to some $e_{(m)}$. The corresponding $e_{(m)}a$ contribution will be any combinations of $\eta$'s and $\epsilon$'s --$\epsilon$ being the totally antisymmetric tensor in four dimensions -- transforming correctly and thus can be set to zero. We arrive at \begin{multline} a=\sum_{\substack{(m)\\c(m)\not=0}}\frac{e_{(m)}}{c(m)} \left( -\frac{1}{2}(l_{\alpha\beta}\otimes\dots\otimes\eins+\dots +\eins\otimes\dots\otimes l_{\alpha\beta})B^{\alpha\beta} +8\sum_{i<j\leq p}{P_\eta}_{ij}a \right), \\ c(m)\doteq 3p+p(p-1)\frac{\chi_{(m)}(\tau)}{f_{(m)}} =3p+\sum_{i=1}^r\left( b_i^{(m)}(b_i^{(m)}+1)-a_i^{(m)}(a_i^{(m)}+1) \right), \label{eq:atenrec} \end{multline} with $a=(a_1,\dots,a_r), b=(b_1,\dots,b_r)$ denoting the characteristics of the frame $(m)$, see appendix~\ref{app:permrep}. Let us take $p=4$ as an example: \[ \begin{array}{cccc} \text{idempotent} & \text{Young frame} & \text{dimension} & \text{character} \\ & & & \\ e_{(4)} & \yng(4) & f_{(4)}=1 & \chi_{(4)}(\tau)=1 \\ & & & \\ e_{(3,1)} & \yng(3,1) & f_{(3,1)}=3 & \chi_{(3,1)}(\tau)=1 \\ & & & \\ e_{(2,2)} & \yng(2,2) & f_{(2,2)}=2 & \chi_{(2,2)}(\tau)=0 \\ & & & \\ e_{(2,1,1)} & \yng(2,1,1) & f_{(2,1,1)}=3 & \chi_{(2,1,1)}(\tau)=-1 \\ & & & \\ e_{(1,1,1,1)} & \yng(1,1,1,1) & f_{(1,1,1,1)}=1 & \chi_{(1,1,1,1)}(\tau)=-1 \end{array} \] We find for (\ref{eq:acasp}) \begin{equation} a=\frac{1}{48}(2e_{(4)} + 3e_{(3,1)} +4e_{(2,2)} +6e_{(2,1,1)})\times \text{r.h.s}(\ref{eq:acasp}). \end{equation} We see that no $e_{(1,1,1,1)}$ appears in that equation. It corresponds to the one dimensional $sgn$-representation of $S_4$, so $e_4a\propto\epsilon$. \section{Spinorial Lorentz covariance} \label{sec:spinor} In this section we follow \cite{bk:sexl-urb}. The most general representation of $\Lcal_{+}^{\uparrow}$ can be built of tensor products of \ensuremath{\mathit{SL}(2,\menge{C})}\ and $\overline{\ensuremath{\mathit{SL}(2,\menge{C})}}$ and direct sums of these. A two component spinor $\Psi$ transforms according to \begin{equation} \Psi^{A}={g^{A}}_{B}\Psi^{B}, \label{eq:spintrafo} \end{equation} where $g$ is a $2\times2$-matrix in the \ensuremath{\mathit{SL}(2,\menge{C})}\ representation of $\Lcal_{+}^{\uparrow}$. For the complex conjugated representation we use the dotted indices, i.e. \begin{equation} \overline{\Psi}^{\Xdot}={\overline{g}^{\Xdot}}_{\Ydot}\overline{\Psi}^{\Ydot}, \label{eq:ccspintrafo} \end{equation} with $ {\overline{g}^{\Xdot}}_{\Ydot}=\overline{{g^{X}}_{Y}} $ in the $ \overline{\ensuremath{\mathit{SL}(2,\menge{C})}} $ representation. The indices are lowered and raised with the $ \epsilon $-tensor. \begin{gather} \epsilon_{AB}=\overline{\epsilon}_{\dot{A}} \newcommand{\Bdot}{\dot{B}\Bdot} \doteq\epsilon_{\dot{A}} \newcommand{\Bdot}{\dot{B}\Bdot}, \label{eq:eps} \\ \epsilon^{AB}\epsilon_{AC} =\epsilon^{BA}\epsilon_{CA} =\delta^{B}_{C}. \end{gather} We define the Van-der-Waerden symbols with the help of the Pauli matrices $\sigma_{\mu}$ and ${\widetilde{\sigma}}_{\mu}\doteq\sigma^{\mu}$: \begin{align} {\sigma_{\mu}}^{A\Xdot}&\doteq\frac{1}{\sqrt{2}}(\sigma_{\mu})^{AX}, & {\sigma_{\mu}}_{A\Xdot}&\doteq\frac{1}{\sqrt{2}} ({{\widetilde{\sigma}}_{\mu}}^{T})_{AX}. \label{def:vdw} \end{align} They satisfy the following relations \begin{align} {\sigma_{\mu}}^{A\Xdot}{\sigma_{\nu}}_{A\Xdot} &=\eta_{\mu\nu} & {\sigma_{\mu}}_{A\Xdot}{\sigma^{\mu}}_{B\Ydot} &=\epsilon_{AB}\epsilon_{\Xdot\Ydot} \label{eq:vdw} \end{align} With the help of these we can build the infinitesimal spinor transformations \begin{equation} g\approx\eins+\frac{1}{2}\theta_{\alpha\beta}S^{\alpha\beta}, \label{eq:spininf} \end{equation} with the generators \begin{equation} {(S^{\alpha\beta})^{A}}_{B} ={\sigma^{[\alpha}}^{A\Xdot}{\sigma^{\beta]}}_{B\Xdot}. \label{eq:spingen} \end{equation} Note that the $ \sigma $'s are hermitian: $ \overline{{\sigma_{\mu}}^{A\Xdot}}={\sigma_{\mu}}^{X\dot{A}} \newcommand{\Bdot}{\dot{B}} $. Again we define the projectors for the tensor product. But we have only two irreducible parts: \begin{align} {{P_S}^{AB}}_{CD}&=\frac{1}{2} (\delta^A_C\delta^B_D+\delta^A_D\delta^B_C), & {{P_\epsilon}^{AB}}_{CD}&=\frac{1}{2} \epsilon^{AB}\epsilon_{CD}, \\ P^2=P, & P_S+P_\epsilon&=\eins. \end{align} We get the following identities: \begin{align} S^{\alpha\beta}S_{\alpha\beta} &=\overline{S}^{\alpha\beta}\overline{S}_{\alpha\beta} =-3\eins, \label{eq:sgen1} \\ S^{\alpha\beta}\otimes S_{\alpha\beta} &=\overline{S}^{\alpha\beta}\otimes\overline{S}_{\alpha\beta} =4P_\epsilon-\eins \label{eq:sgen2}, \\ S^{\alpha\beta}\otimes\overline{S}_{\alpha\beta} &=\overline{S}^{\alpha\beta}\otimes S_{\alpha\beta} =0. \label{eq:sgen3} \end{align} In order to have (\ref{def:a}) in a pure spinor representation we have to decompose the tensor $\xbar$ into spinor indices according to \begin{equation} x^{A\Xdot}\doteq x^\mu{\sigma_{\mu}}^{A\Xdot}. \end{equation} Assume $t\widetilde{x}$ transforms under the $u$-fold tensor product of \ensuremath{\mathit{SL}(2,\menge{C})}\ times the $v$-fold tensor product of $\overline{\ensuremath{\mathit{SL}(2,\menge{C})}}$ then, for infinitesimal transformations, (\ref{def:a}) yields: \begin{equation} B^{\alpha\beta} =(S^{\alpha\beta}\otimes\dots\otimes\eins+\dots +\eins\otimes\dots\otimes\overline{S}^{\alpha\beta})a, \end{equation} with $B^{\alpha\beta}$ from equation \eqref{eq:atensor} in the corresponding spinor representation. The sum consists of $u$ summands with one $S^{\alpha\beta}$ and $v$ summands with one $\overline{S}^{\alpha\beta}$ with $u,v>n$. Multiplying again with the generator and contracting the indices gives twice the Casimir on the r.h.s. Inserting (\ref{eq:sgen1}-\ref{eq:sgen3}) yields: \begin{multline} (S_{\alpha\beta}\otimes\dots\otimes\eins+\dots +\eins\otimes\dots\otimes\overline{S}_{\alpha\beta})B^{\alpha\beta} \\ =\biggl( -3(u+v)\eins +2\sum_{1\leq i<j\leq u}(4P_{\epsilon_{ij}}-\eins) + 2\sum_{1\leq i<j\leq v}(4P_{\overline{\epsilon}_{ij}}-\eins) \biggr)a. \end{multline} The sum over $u$ runs over $\frac{u}{2}(u-1)$ possibilities and similar for $v$, so we find the induction: \begin{multline} a=\frac{1}{u(u+2)+v(v+2)}\Biggl[ -(S_{\alpha\beta}\otimes\dots\otimes\eins+\dots +\eins\otimes\dots\otimes\overline{S}_{\alpha\beta})B^{\alpha\beta}+ \\ 8\biggl( \sum_{1\leq i<j\leq u}P_{\epsilon_{ij}} +\sum_{1\leq i<j\leq v}P_{\overline{\epsilon}_{ij}} \biggr)a \Biggr]. \label{eq:spinind} \end{multline} It already contains the induction start for $a^{(AB)}, a^{(XY)}$ and $a^{A\Xdot}$. \section{General covariant BPHZ subtraction} \label{sec:bphz} In this section we shrink the distribution space to $\Scal'$ since we are dealing with Fourier transformation. Let $x,q,p\in\Rvm$. BPHZ subtraction at momentum $q$ corresponds to using $w=e^{iq\cdot}$ in the EG subtraction \cite{pap:prange1}. \begin{align} \widehat{\tR{\omega}{e^{iq\cdot}}}(p) &\doteq\scp{\tR{\omega}{e^{iq\cdot}}} {e^{ip\cdot}} \\ &=\scp{^{0}t}{e^{ip\cdot}-\sum_{|\alpha|\leq\omega} \frac{(p-q)^\alpha}{\alpha!}\ensuremath{\partial}^q_{\alpha}e^{iq\cdot}}. \label{eq:BPHZ} \end{align} It is normalized at the subtraction point $q$, i.e.: $\ensuremath{\partial}_{\alpha}\widehat{\tR{\omega}{e^{iq\cdot}}}(q) = 0,\ |\alpha| \leq \omega$. This is always possible for $q$ totally spacelike, $(\sum_{j\in I}q_{j})^2<0, \forall I\subset\{1,\dots,m\}$ \cite{pap:ep-gl,priv:duetsch}. In massive theories one can put $q=0$ and has the usual subtraction at zero momentum which preserves covariance. But this leads to infrared divergencies in the massless case. There we can use the results from above to construct a covariant BPHZ subtraction for momentum $q$ by adding $\sum_{|\alpha|\leq\omega} \frac{i^{|\alpha|}}{\alpha!} a^\alpha p_\alpha$ to (\ref{eq:BPHZ}), according to equation \eqref{def:tRGcov}. For $|\beta|\geq\omega+1$, ${^0t}x^\beta$ is a well defined distribution on $\Scal$ and so is $\ensuremath{\partial}_\beta\widehat{^0t}$. \subsection{Lorentz invariance on \Rv} We have \begin{align} V_k e^{iq\cdot}&=e^{iq\cdot}\sum_{m=0}^k\frac{1}{m!}(-iqx)^m, & \ensuremath{\partial}_\sigma V_k e^{iq\cdot} &=iq_\sigma e^{iq\cdot}\frac{1}{k!}(-iqx)^k. \end{align} Inserting this into (\ref{eq:amun}) we find: \begin{multline} a^{(\mu _1\dots\mu_n)}=\frac{i^n(-)^{\omega+1}}{(\omega-n)!} \frac{ (n-1)!!}{(n+2)!!} q_{\sigma_1}\dots q_{\sigma_{\omega-n}} \sum_{s=0}^{\left[\frac{n-1}{2} \right] } \frac{(n-2s)!!}{(n-2s-1)!!} \left(q_\rho\ensuremath{\partial}^\rho\ensuremath{\partial}^{(\mu_1}-q^{(\mu_1}\square\right)\times \\ \times \eta^{\mu_2\mu_3}\dots\eta^{\mu_{2s}\mu_{2s+1}} \ensuremath{\partial}^{\mu_{2s+2}}\dots\ensuremath{\partial}^{\mu_n)}\square^s \ensuremath{\partial}^{\sigma_1}\dots\ensuremath{\partial}^{\sigma_{\omega-n}}\widehat{{^0t}}(q). \label{eq:amunq} \end{multline} \subsubsection*{Example} Take the setting sun in massless scalar field theory: $^{0}t=\frac{i^{3}}{6}D_{F}^{3} \Rightarrow \omega=2$. \begin{align*} a^\mu&=-\frac{i}{3}(q_\sigma q_\rho \ensuremath{\partial}^\rho \ensuremath{\partial}^\sigma \ensuremath{\partial}^\mu -q^\mu q_\sigma\ensuremath{\partial}^\sigma\square)\widehat{^{0}t}(q), \\ a^{\mu\nu}&=\frac{1}{4}(q_\rho \ensuremath{\partial}^\rho \ensuremath{\partial}^\mu \ensuremath{\partial}^\nu -q^{(\mu}\ensuremath{\partial}^{\nu)}\square)\widehat{^{0}t}(q), \end{align*} and adding $ip_{\mu}a^{\mu} - \frac{1}{2}p_{\mu}p_{\nu}a^{\mu\nu}$ restores Lorentz invariance of the setting sun graph subtracted at $q$. \subsection{General induction} We only have to evaluate $B^{\alpha\beta}$ with $w=e^{iq\cdot}$ and plug the result into the induction formulas \eqref{eq:spinind} and \eqref{eq:atenrec}. \begin{equation} B^{\alpha\beta} =2i^{n}(-)^{\omega+1}\sum_{j=1}^{m}\sum_{|\gamma|=\omega-n} \frac{q^{\gamma}}{\gamma!}q_{j}^{[\alpha}\ensuremath{\partial}_{j}^{\beta]} \ensuremath{\partial}_{\gamma}\widetilde{\ensuremath{\partial}}\,\widehat{^{0}t}(q). \end{equation} Here, $q_{j}$ are the $m$ components of $q$ hence $\gamma$ is a $4m$ index and $\alpha,\beta$ are four indices. The tensor (spinor) structure of $\widetilde{\ensuremath{\partial}}$ is given by $\widetilde{x}$ in \eqref{eq:atensor}. \section{Summary and Outlook} The subtraction procedure in EG renormalization makes use of an auxiliary (test) function and hence breaks Lorentz covariance, since no Lorentz invariant test function exists. But this symmetry can be restored by an appropriate choice of counterterms. We give an explicit formula for their calculation in lowest order and an inductive one for higher orders. Using the close relationship to BPHZ subtraction this directly translates into a covariant subtraction at totally spacelike momentum. We expect our solution to be useful for all calculations for which the central solution ($w=1$, see \cite{bk:scharf}) does not exist, namely all theories that contain loops of only massless particles. \section{Acknowledgment} I would like to thank Klaus Bresser and Gudrun Pinter for our short but efficient teamwork and K. Fredenhagen for permanent support. \clearpage
\section{Introduction} With the advent of HERA there has been a great increase in the scope of the theoretical effort to understand the physics of hadronic scattering at high energy. This is a challenging subject especially as it might provide a bridge between perturbative partonic physics of short distance processes (e.g. DIS at moderate $x_{Bj}$) and soft physics of hadronic states which, presumably, dominate the high energy asymptotics. The borderline between the two --- the "semihard" physics --- is interesting also in its own right. In essence it is the physics of partonic systems which are, on one hand dense enough for new collective phenomena to play important role but, on the other hand are perturbative since the average momentum transfer between the partons is high enough. In this semihard regime one expects to see perturbativelly controllable nonlinear effects which depart from the standard linear evolution of DGLAP \cite{dglap} or BFKL \cite{bfkl} type and subsequently lead to unitarization of hadronic cross sections. An approach to this high partonic density regime from the partonic side has been spearheaded by Levin and collaborators \cite{levin} based on earlier work by Mueller \cite{mueller} and is an "all twist" generalization of the GLR recombination picture \cite{glr,mq}. It lead to the formulation of a nonlinear evolution equation which exhibits a perturbative mechanism of unitarization. Analysis of this equation suggests that already at present HERA energies the nonlinearities in the gluon sector are considerable and linear evolution for gluons should break down. Better experimental data on gluon distributions would be extremely valuable in order to verify/falsify this assessment. A complementary approach was pioneered some years ago by McLerran and Venugopalan \cite{mv}. It was later somewhat reshaped conceptually and considerably developed technically in a series of papers \cite{jkmw,jklw,moap,soap}. Here the idea is that in the high density regime rather than using partonic language, it is more appropriate to use the language of classical fields. The hadron then is considered as an ensemble of configurations of the gluon field. The statistical weight that governs the contributions of different configurations to the ensemble averaging changes when one probes the hadron on different time scales. Decreasing $x_{Bj}$ corresponds to increasing the time resolution and therefore corresponds to probing the hadron on shorter time scales. The change of the statistical weight with $x_{Bj}$ is governed by the evolution equation. As long as the field intensity is large enough this evolution should be perturbative in $\alpha_s$ but essentially nonlinear in the field intensity itself. This evolution equation to first order in $\alpha_s$ was derived in \cite{jklw,moap}. We will refer to it as the JKLW equation in the following. In \cite{soap} the double logarithmic limit of this evolution was considered. It was shown that in this limit the evolution of the gluon density becomes unitary at large density. Qualitatively, the evolution is very similar to that discussed in \cite{levin}, although the details are different. A detailed numerical study of the doubly logarithmic limit of the JKLW evolution was recently performed in \cite{jw}. Technically the derivation of \cite{moap} is fairly involved. Several consistency checks were performed in \cite{jklw} and \cite{soap} to make sure that the known results are recovered from the general evolution equation in the weak field limit. This includes the BFKL equation, the doubly logarithmic approximation to the DGLAP equation and the GLR equation. It is, however, desirable to have some additional independent checks on the equation which do not involve the weak field limit. It is the aim of the present paper to provide one such check. In a nutshell the issue we address is the following. The evolution equation in \cite{jklw,moap,soap} was derived invoking a two step procedure. Rather than considering directly the evolution of the correlators of the gluon field, one first considers the evolution of the colour charge density. In the second step, one re-expresses the evolution equation for the charge density as the evolution equation for the vector potential (chromoelectric field). The evolution of the field correlators is in fact what one is after, since it is the vector potential and not the charge density that couples directly to fermions and, therefore, are more directly related to physical observables. Our observation in this paper is that one can avoid the introduction of the colour charge density altogether and derive the evolution equations directly for the field correlators. This procedure has the advantage of being somewhat simpler both technically and conceptually. Nevertheless, the final result for the evolution should be the same as in the two step procedure of \cite{jkmw,jklw,moap,soap}. The comparison of our results with the earlier derived formulae provides us with a consistency check on the calculation. We find, in fact, that the results presented in \cite{moap} are not entirely correct. However, after correcting some algebraic mistakes in \cite{moap} we show explicitly that the two approaches yield identical results. We provide the corrected expressions for the "kernels" of the evolution equation, which are somewhat simpler than the expressions found in \cite{moap}. We also clarify the issue of possible Gribov ambiguity and show explicitly that the divergent Jacobians, which appeared in the intermediate steps of the derivations in \cite{soap}, cancel completely in the final expressions for the correlators of the chromoelectric field. Therefore, the Gribov ambiguity, although affecting the relation between the colour charge density and the chromoectric field, does not affect the evolution of the field correlators, at least to order $\alpha_s$. Since the procedure discussed in the present paper avoids the introduction of the charge density entirely, the whole approach is free from the Gribov problem. Perhaps somewhat surprisingly the corrections to the results of \cite{moap} that we find do not affect either the weak field limit discussed in \cite{jklw} or the doubly logarithmic limit of \cite{soap}. They therefore have no bearing on the derivation of the BFKL equation in our approach and also do not help to reconcile the doubly logarithmic limit of the JKLW equation \cite{soap} with the nonlinear equation studied in \cite{levin}. The plan of the paper is the following. In Section 2 we briefly recap the procedure of the derivation of the evolution equation as described in \cite{jklw,moap,soap} and reformulate it directly in terms of the gluon field correlators. In Section 3, using some of the results of \cite{moap}, we calculate the real and virtual parts of the evolution in terms of the field correlators and provide the corrections to the results of \cite{moap}. Finally, in Section 4 we discuss our results. \section{The JKLW evolution} First, let us briefly recall the framework and the results of \cite{jklw, moap,soap}. In this approach the averages of gluonic observables in a hadron are calculated via the following path integral \begin{multline} <O(A)> = \int D\rho DA^\mu O(A) \exp\Big\{ -\int d^2 x_\perp F[\rho ^a(x_\perp)] -i\int d^4 x \frac{1}{4}{\rm tr} F^{\mu\nu}F_{\mu\nu} \\ +{\frac{1}{N_c}} \int d^2 x_\perp dx^- \delta (x^-) \rho_{a}(x_\perp) {\rm tr}T_a W_{-\infty,\infty} [A^-](x^-,x_\perp)\Big\} \label{action} \end{multline} where the gluon field strenght tensor is given by \begin{equation} F^{\mu\nu}_{a} = {\partial}^{\mu} A^{\nu}_{a} - {\partial}^{\nu} A^{\mu}_{a} - g f_{abc}A^{\mu}_{b}A^{\nu}_{c} \end{equation} and $W$ is the Wilson line in the adjoint representation along the $x^+$ axis \begin{equation} W_{-\infty ,+\infty}[A^{-}](x^- , x_{\perp}) = P\exp\Bigg[+i g \int dx^+ A^-_a (x^+,x^- , x_{\perp}) T_a \Bigg] \end{equation} The hadron is represented by an ensemble of colour charges localized in the plane $x^-=0$ with the (integrated across $x^-$) colour charge density $\rho(x_\perp)$. The statistical weight of a configuration $\rho(x_\perp)$ is \begin{equation} Z=\exp \{-F[\rho]\} \label{z} \end{equation} In the tree level approximation (in the light cone gauge $A^+=0$) the chromoelectric field is determined by the colour charge density through the equations \begin{equation} F^{+i}={1\over g}\delta(x^-)\alpha^{i}(x_\perp) \label{chrom} \end{equation} and the two dimensional vector potential $\alpha^i(x_\perp)$ is "pure gauge", related to the colour charge density by \begin{align} {\partial}^{i}{\alpha}_{a}^{j} &- {\partial}^{j}{\alpha}_{a}^{i} - f_{abc}{\alpha}_{b}^{i}{\alpha}_{c}^{j}=0 \nonumber \\ {\partial}^{i}{\alpha}_{a}^{i} & =-{\rho}_{a} \label{sol} \end{align} Integrating out the high longitudinal momentum modes of the vector potential generates the renormalization group equation, which has the form of the evolution equation for the statistical weight $Z$ \cite{jklw} \footnote{All the functions in the rest of this paper depend only on the transverse coordinates. For simplicity of notation we drop the subscript $\perp$ in the following.} \begin{equation} {d\over d\zeta}Z= \alpha_s \left\{{1\over 2}{\delta^2 \over\delta\rho(u)\delta\rho(v)}\left[Z\chi(u, v) \right] - {\delta\over\delta\rho(u)}\left[Z\sigma(u)\right]\right\} \label{final} \end{equation} In the compact notation used in Eq.~(\ref{final}), both $u$ and $v$ stand for pairs of colour index and transverse coordinates, with summation and integration over repeated occurrences implied. The evolution in this equation is with respect to the rapidity $\zeta$, related to the Feynman $x$ by \begin{equation} \zeta=\ln 1/x \end{equation} Technically it arises as a variation of $Z$ with the cutoff imposed on the longitudinal momentum of the fields $A^\mu$. The quantities $\chi[\rho]$ and $\sigma[\rho]$ have the meaning of the mean fluctuation and the average value of the extra charge density induced by the high longitudinal momentum modes of $A^\mu$. They are functionals of the external charge density $\rho$. The explicit expressions have been given in \cite{moap} and it is our aim in this paper to provide a check on these expressions. Eq.~(\ref{final}) can be written directly as an evolution equation for the correlators of the charge density. Multiplying Eq.~(\ref{final}) by $\rho(x_1)...\rho(x_n)$ and integrating over $\rho$ yields \begin{multline} {d\over d\zeta}<\rho(x_1)...\rho(x_n)>= \\ = \alpha_s \Bigg[\sum_{0<m<k<n+1} <\rho(x_1)...\rho(x_{m-1})\rho(x_{m+1})... \rho(x_{k-1})\rho(x_{k+1})...\rho(x_n)\chi(x_m, x_k)> \\ +\sum_{0<l<n+1}<\rho(x_1)...\rho(x_{l-1})\rho(x_{l+1}) ...\rho(x_n)\sigma(x_l)>\Bigg] \label{correl} \end{multline} This set of equations for the correlators of the colour charge density completely specifies the evolution of the hadronic ensemble as one moves to higher energies (or lower values of $x$). The evolution equations for the correlators of the charge density can be rewritten as equations for the correlators of the vector potential \cite{soap}. \begin{multline} \label{correlf} {d\over d\zeta} <\alpha_{a_1}^{i_1}(x_1)...\alpha_{a_n}^{i_n}(x_n)> = \\ = \alpha_s \Bigg[\sum_{0<l<n+1} <\alpha_{a_1}^{i_1}(x_1)... \alpha_{a_{l-1}}^{i_{l-1}}(x_{l-1}) \alpha_{a_{l+1}}^{i_{l+1}}(x_{l+1})... \alpha_{a_n}^{i_n}(x_n)\sigma_{a_l}^{i_l}(x_l)> \\ + \sum_{0<m<k<n+1}<\alpha_{a_1}^{i_1}(x_1)... \alpha_{a_{m-1}}^{i_{m-1}} (x_{m-1})\alpha_{a_{m+1}}^{i_{m+1}}(x_{m+1})... \\ \times \alpha_{a_{k-1}}^{i_{k-1}}(x_{k-1}) \alpha_{a_{k+1}}^{i_{k+1}}(x_{k+1})... \alpha_{a_n}^{i_n}(x_n)\chi_{a_ma_k}^{i_mi_k}(x_m, x_k)> \Bigg] \end{multline} The quantities $\chi_{ab}^{ij}$ and $\sigma_a^i$ have a very simple physical meaning. The high momentum modes of the vector field which have been integrated out in order to arrive at the evolution equation induce extra colour charge density $\delta\rho$. The average value of this induced density and its mean fluctuation appear in the evolution equations eq.(\ref{correl}) as $\sigma_a^{ }$ and $\chi_{ab}^{ }$. The appearance of the induced colour charge density leads to the change in the value of the chromoelectric field through the solution of Eq.~(\ref{sol}) with $\rho+\delta\rho$ on the right hand side. The quantities $\sigma_a^i$ and $\chi_{ab}^{ij}$ are the average value and the mean fluctuation of the induced field respectively. It is perhaps helpful to explain how $\sigma_a^i$ and $\chi_{ab}^{ij}$ were obtained in \cite{soap}. As shown in \cite{moap}, the induced charge density can be decomposed into two pieces \footnote{The reason for the notation $\tilde\rho$ rather than simply $\rho$ will be explained in the next section.} \begin{equation} \delta\rho=\delta\tilde\rho_1+\delta\tilde\rho_2 \end{equation} The first piece $\delta\tilde\rho_1$ is order $g$ while the second piece $\delta\tilde\rho_2$ is order $g^2$. The $\delta\tilde\rho_1$ is time dependent, and has zero average value, while its mean fluctuation is order $g^2$. The $\delta\tilde\rho_2$ being $O(g^2)$ contributes only to the average value of $\delta\tilde\rho$ and not to the mean fluctuation. Assuming that the classical equations eq.(\ref{sol}) hold not only for the background field but also for the relevant part of the fluctuation field one can solve those equations perturbatively. Writing \begin{equation} \delta\alpha^i=\delta\alpha_1^i+\delta\alpha_2^i \end{equation} with $\delta\alpha_1$ being $O(g)$ and $\delta\alpha_2$ being $O(g^2)$ and keeping in the classical equations all terms to order $g^2$ we have \begin{align} &D_{ab}^i\delta\alpha_{1b}^j -D_{ab}^j\delta\alpha_{1b}^i +D_{ab}^i\delta\alpha_{2b}^j -D_{ab}^j\delta\alpha_{2b}^i -f_{abc}\delta\alpha_{1b}^i\delta\alpha_{1c}^j=0\nonumber \\ &\partial^i\delta\alpha_{1a}^i+\partial^i\delta\alpha_{2a}^i =-(\delta\tilde\rho_{1a}+\delta\tilde\rho_{2a}) \label{solp} \end{align} We have defined for convenience \begin{align} \alpha_{ab}^i &= f_{abc}\alpha^i_c \nonumber \\ D_{ab}^i &= \partial^i\delta_{ab}+\alpha^i_{ab} \label{def} \end{align} To order $g$ we find \begin{equation} \delta\alpha^i_1=-D^i{1\over \partial D}\delta \tilde\rho_1 \label{corresp1} \end{equation} Therefore, to order $g^2$ \begin{equation} \chi_{ab}^{ij}(x,y)= r_{ac}^i(x,u)\chi_{cd}^{ }(u,v)r_{db}^{\dagger j}(v,y) \label{indcor} \end{equation} with \begin{equation} r_{ab}^i(x,y)= - <x|[D^i\frac{1}{\partial D}]_{ab}|y> \label{r} \end{equation} Here $<x|O|y>$ denotes a configuration space matrix element in the usual sense. At order $g^2$ we have \begin{equation} \delta\alpha^i_2=-D^i{1\over \partial D} \delta \tilde\rho_2-{1\over 2}\epsilon^{ij}\partial^j {1\over D\partial}\delta\alpha_{1}\times\delta\alpha_{1} \label{corresp2} \end{equation} Here the cross product is defined as \begin{equation*} A\times B = f_{abc}\epsilon^{ij}A_a^iB_b^j \end{equation*} We thus have \begin{equation} \sigma_{a}^{i}(x)= r_{ab}^i(x,u)\sigma_{b}^{ }(u)+p_{abc}^i(x,u,v)\chi_{bc}^{ }(u,v) \label{indfield} \end{equation} with \begin{equation} p^i_{abc}(x,y,z)= - \frac{1}{2}(\epsilon^{ij}\partial^j[\frac{1}{D\partial}]_{ad})(x,u) f_{dfe}\epsilon^{kl} r^k_{fb}(u,y)r^l_{ec}(u,z) \label{p} \end{equation} The procedure of deriving eq.(\ref{correlf}) employed in \cite{moap,soap} consists, therefore, of two steps. One first splits the gluon field into the classical background field $\alpha^\mu$ and the fluctuation field $a^\mu$. The modes of the fluctuation field with longitudinal momenta in some range $\alpha_s\ln {\Lambda^+\over \Lambda^{'+}}$ are assumed to be small. One defines operatorialy the induced charge density $\delta\rho$ in terms of the fluctuation fields $a^\mu$ and the quantities $\sigma$ and $\chi$ are calculated by integrating out the fluctuation fields perturbatively. In the second step, one solves classical equations of motion which include the induced charge density and calculates $\sigma^i$ and $\chi^{ij}$. Clearly, consistency requires that the two step procedure that leads from eq.(\ref{action}) through eqs.(\ref{final},\ref{correl}) to the evolution equations eq.(\ref{correlf}, \ref{indcor}, \ref{indfield}) must be equivalent to the following. Start with the equivalent of eq.(\ref{action}) \footnote{We note that the statistical weight $Z[\alpha_i]$ which appears in eq.(\ref{actionf}) is not equal to $Z$ of eq.(\ref{z}) since going from eq.(\ref{action}) to eq.(\ref{actionf}) involves the change of variables $\rho\rightarrow\alpha_i$. The two statistical weights, therefore, differ by an appropriate Jacobian.} \begin{multline} <O(A)> =\int D\alpha^i DA^\mu O(A) Z[\alpha_i(x_\perp)] \exp\{ -i\int d^4 x {1\over 4}{\rm tr} F^{\mu\nu}F_{\mu\nu} \\ -{{1}\over{N_c}} \int d^2 x_\perp dx^- \delta (x^-) \partial^i\alpha_a^i(x_\perp) {\rm tr}T_a W_{-\infty,\infty} [A^-](x^-,x_\perp)\} \label{actionf} \end{multline} Integrate out the high longitudinal momentum components of $a^\mu$ as before, but instead of calculating the induced charge density $\sigma_a^{ }$ and $\chi_{ab}^{ }$, calculate directly the induced chromoelectric field $\sigma_a^i$ and $\chi_{ab}^{ij}$. Technically this calculation is somewhat simpler, since there is no need to consider the operator $\delta\rho$, which is nonlinear in the fluctuation field $a^\mu$. Instead, one directly calculates the distribution of the static component of $a^\mu$. The resulting evolution equations should coincide with eq.(\ref{correlf}). With this formulation one circumvents completely the need to introduce the colour charge density $\rho$ and to solve classical equations for $\alpha^i$ in terms of $\rho$. While one may want to introduce $\rho$ for reasons of convenience, our present understanding is that it is not necessary from the point of view of physics. The physics that our approach is meant to address is that of the evolution of the hadronic ensemble. The relation between $\alpha^i$ and $\rho$ on the other hand is supposed to hold at every value of $\zeta$, and therefore itself is unrelated to evolution in $\zeta$. The concept of $\rho$ may be sometimes useful to formulate models for the statistical weight $Z$ at some particular value of $\zeta$ as was the original motivation of \cite{mv}. This could then serve as an initial condition for the evolution. This, however, is a separate question and we do not intend to address it here. Before we proceed further, we wish to make one more comment about the relation between the chromoelectric field and the colour charge density eqs.(\ref{corresp1},{\ref{corresp2}). Both these equations contain the dangerous factor $(\partial D)^{-1}$. The operator $\partial D$ has zero, as well as negative eigenvalues and is very reminiscent of the operators usually associated with the Gribov ambiguity in nonabelian gauge theories. In fact, it is quite clear that it has precisely the same origin. The second equation in eq.(\ref{solp}) has the form of the Lorentz like gauge fixing condition on the fluctuation field $\delta\alpha$. Since the calculation is performed in a nonvanishing background field, the Lorentz gauge indeed suffers from Gribov ambiguity precisely due to negative eigenvalues of the operator $\partial D$. Given this, one may worry that our perturbative calculation is plagued with the Gribov ambiguity \footnote{In standard perturbation theory, the Gribov ambiguity does not show up in any finite order. This is due to the fact that one expands the operator $\partial D$ and its inverse in powers of the coupling constant. To leading order then the operator does not have any negative eigenvalues, which ensures that no problems arise in finite order perturbative calculations. Our situation is, however, different. Since our background field is not assumed to be $O(g)$, the operator cannot be expanded. Therefore, there is no guarantee that the problem does not show up even in perturbation theory.}. However this is not necessarily the case. The point is that $\delta\rho$ itself is not arbitrary. It is calculated through the fluctuation field and, at the end of the day, is averaged over with some statistical weight $Z[\delta\rho]$. It could well be that the statistical weight is such that it only allows induced charge density of the form $\delta\rho=\partial D X$ with regular $X$. If that is the case, the dangerous denominator cancels and the induced field is well defined and regular. In fact, in our present formulation where the calculation is performed directly in terms of the field, it is almost clear that this should indeed happen. In this setup one calculates directly $\delta\alpha$, and eqs.(\ref{corresp1},\ref{corresp2}) should be read from right to left, as equations determining an auxilliary quantity $\delta\rho$ through $\delta\alpha$ rather than the other way round. The operator $\partial D$ then appears in the numerator and all expressions are regular. In fact we will show in the next section by explicit calculation that all ``dangerous factors'' indeed cancel in the final expressions for $\chi_{ab}^{ij}$ and $\sigma_a^i$. Note that, if one insists on formulating the problem in terms of the colour charge density, the absence of the Gribov ambiguity implies a nontrivial consistency condition on the statistical weight $Z[\rho]$. Taking an arbitrary weight $Z$ will render the calculation of chromoelectric field correlators ill defined especially at strong fields (strong coupling). This was indeed observed in the numerical calculation \cite{gv} where a simple Gaussian in $\rho$ was used as the weight function \footnote{This problem does not arise in the more recent numerical work \cite{kv} since in effect this work uses a different definition of $\rho$ for which the relations analogous to eqs.(\ref{corresp1},\ref{corresp2}) do not involve singular factors.}. In the next section we will calculate $\sigma_a^i$ and $\chi_{ab}^{ij}$ induced by high longitudinal momentum modes. \section{The Induced Chromoelectric Field} The main ingredients needed for the calculation of the induced chromoelectric field are the eigenfunctions of the quadratic action for the small fluctuations in the static background $\alpha^i$. Solving the classical equations of motion that follow from the action eq.(\ref{actionf}) at fixed $\alpha^i$ we find (in the gauge $\partial^i A^i(x^+\rightarrow -\infty)=0$) the classical solution \begin{equation} A_{cl}^{-}=0, \quad A^i_{cl} =\alpha^i (x_\perp)\theta(x^-) \end{equation} Defining the quantum fluctuation field $a^\mu$ by $A^\mu =A^\mu_{cl}+a^\mu$ and expanding the action to second order in $a^\mu$ we have \begin{equation} S=\frac{1}{2g^2} \Bigg\{ a^-_{x} K_{xy} a^-_{y} + 2a^-(\partial^+ Da - 2 fa ) + 2\partial^+a^i\partial^-a^i + a^i \bigg[ D^2 \delta^{ij} - D^{i}D^{j} \bigg] a^j \Bigg\} \label{eq:action} \end{equation} Here we are using the notation \begin{align} &[fa]_{a}(x^+,x^-,x_\perp ) = \delta (x^-) \alpha^i_{ab}(x_\perp ) a^i_{b}(x^+,x^-,x_\perp ) \nonumber \\ &Da = D^{i}[\alpha]a^{i}=(\partial^i\delta_{ab}+ \theta(x^-)\alpha^i_{ab})a_b^i \label{cov} \end{align} and as previously \begin{equation} \alpha^i_{ab}=f_{abc}\alpha^i_c \end{equation} The operator $K$ is \begin{equation} K_{ab}^{xy} = - \bigg[(\partial^{+})^2 \delta_{ab} +\partial^i\alpha^i_{ab} \delta (x^-) \frac{1}{\partial^-} \bigg] \label{K} \end{equation} Note that there is no ambiguity in the definition of the operator $1/\partial^- $ in this expression. It is defined in the sense of principal value. This follows directly from the fact that the matrix $\alpha^i_{ab}$ is antisymmetric and therefore the term involving $1/\partial ^-$ in eq.(\ref{eq:action}) vanishes for zero frequency fields. This eigenfunctions of the quadratic fluctuation operator have been found in \cite{moap} and we cite here the relevant results. The calculation is performed in the lightcone gauge $A^+=0$ with the residual gauge freedom fixed by the condition \begin{equation} \partial^i A^i(x^-\rightarrow -\infty)=0 \end{equation} It is convenient to define an auxiliary field \begin{equation} \tilde a^- = a^-+ K^{-1} (\partial^+ Da - 2 fa ) \label{tildea} \end{equation} This field can be seen to decouple from $a_i$. Its correlator is \begin{equation} <\tilde a^-_x\tilde a^-_y>=K^{-1}_{x,y} \end{equation} The operator $K$ Eq.(\ref{K}) has zero modes. Defining the projector matrices $\eta$ and $\mu$ by \begin{equation} \mu_{ab} \partial^i\alpha^i_{bc} \frac{1}{\partial^-} = 0,\quad \eta_{ab} \partial^i\alpha^i_{bc} \frac{1}{\partial^-} = \rho_{ac} \frac{1}{\partial^-} \end{equation} and \begin{equation} \mu + \eta = 1 ,\quad \mu^{2} = \mu ,\quad \eta^{2} = \eta \label{eq:project} \end{equation} we can write the normalizable zero modes of $K$ in the form \begin{equation} f_a(x_\perp ,x^-,p^-)=\mu_{ab}f(x_\perp ,p^-) \label{zero} \end{equation} The operator $K$ is therefore, strictly speaking, non invertible. The operator $K^{-1}$ in eq.(\ref{tildea}) has to be understood as the inverse of $K$ on the space of functions which does not include the functions Eq.(\ref{zero}). Further, it is only the nonzero mode part of $a^-$ that enters the definition of $\tilde a^-$ in eq.(\ref{tildea}). For our calculation we will need the properly normalized solutions of the equations of motion that follow from the action eq.(\ref{eq:action}). The complete set of these solutions was found in \cite{moap} \begin{multline} a^{i}_{p^-,r} = ge^{ip^{-}x^{+}} \int d^{2}p_{\perp} \bigg[ \theta (-x^-) \exp\left( i{p^{2}_{\perp} \over 2p^{-}}x^{-} - ip_{\perp}x_{\perp}\right) v^{i}_{-, r}(p_{\perp}) \\ + \theta (x^{-}) U (x_\perp ) \exp\left( i{p^{2}_{\perp} \over 2p^{-}}x^{-} - ip_{\perp}x_{\perp}\right) \left[U^{\dagger}v^{i}_{+,r}\right](p_\perp ) + \theta (x^{-}) \gamma^{i}_{+,r}\bigg] \label{solut} \end{multline} The frequency $p^-$ is a good quantum number since the background field is static. Here $r$ is the degeneracy label, which labels independent solutions with the frequency $p^-$. In the free case it is conventionally chosen as the transverse momentum, $\{r\}=\{p^i\}$. The matrix $U(x_\perp )$ is the $SU(N)$ matrix that parametrizes the two dimensional ``pure gauge'' vector potential $\alpha^i(x_\perp)$ \begin{equation} \alpha^i(x_\perp )= {i\over g}U(x_\perp )\partial^i U^\dagger(x_\perp )\nonumber \end{equation} The auxiliary functions $\gamma^i_+, v^i_\pm$ are all determined in terms of one vector function. Choosing this independent function as $v^i_-$ we have \begin{align} v^{i}_{+,r}&= \bigg[T^{ij} -L^{ij} \bigg] \bigg[t^{jk} -l^{jk}\bigg] v^{k}_{-,r} \\ \gamma^{i}_{+,r}&= 2 D^{i} \bigg[{D^j\over D^2} - {\partial^j\over \partial^{2}}\bigg] \bigg[t^{jk} -l^{jk}\bigg]v^k_{-,r} \end{align} where we have defined the projection operators \begin{alignat}{2} T^{ij} & \equiv \delta^{ij} - {D^i D^j \over D^2}, & \qquad L^{ij} & \equiv {D^i D^j \over D^2} \nonumber\\ t^{ij} & \equiv \delta^{ij} - {\partial^i \partial^j \over \partial^2}, & \quad l^{ij} & \equiv {\partial^i \partial^j \over \partial^2} \label{eq:tproj} \end{alignat} The proper normalization of the eigenfunctions requires $v^i_-$ to be chosen as complete set of eigenfunctions of the two dimensional Hermitian operator $O^{-1}$ \begin{multline} \qquad [(t-l)O^{-1}(t-l)]_{ab}^{ij}(x_\perp,y_\perp)= \\ <x_\perp|\delta_{ab}^{ij} - 2 \bigg[[\partial^i{1\over\partial^2}-D^i{1\over D^2}] S^{-1}[{1\over\partial^2}\partial^j-{1\over D^2}D^j] \bigg]_{ab}|y_\perp> \qquad \end{multline} such that \begin{equation} \int d^2 r_\perp v^{i}_{-,r,a}(x_\perp ) v^{\ast j}_{-,r,b}(y_\perp ) = {1\over 4\pi |p^-|}[O^{-1}]_{ab}^{ij} (x_\perp ,y_\perp ) \label{ort} \end{equation} The rotational scalar operator $S$ is \begin{equation} S = {1\over D^2}+2 [{\partial^i\over\partial^2}- {D^i\over D^2}][{\partial^i\over\partial^2}- {D^i\over D^2}]= {1\over D^2}-2{1\over \partial^2} \partial\alpha{1\over D^2} + 2{1\over D^2}D\alpha{1\over\partial^2} \end{equation} For further use we also need the expression for the $a^-$ component of the fluctuation field. Using the explicit expression for the operator $K$ from \cite{moap} we get from eq.(\ref{tildea}) \begin{multline} a^-(x^-,x_\perp,p^-) = \tilde a^- -\theta(x^-) \int_{x^-}^\infty dy^-D^i(a^i-\gamma^i_+) \\ -\theta(-x^-)\bigg[\int_0^\infty dy^-D^i(a^i-\gamma^i_+) +\int_{x^-}^0 dy^-\partial^ia^i\bigg] + 2ip^-\eta[{D^i\over D^2} - {\partial^i\over \partial^2}] (t-l)^{ij}v^j_-(x_\perp) \label{a-} \end{multline} We note that this expression differs by a $x^-$-independent constant from the one given in \cite{moap}. The reason is that in \cite{moap} a constant have been subtracted from $a^-$ such that $\int_{-\infty}^{+\infty}dx^-a^-(x^-)=0$. This corresponds to the symmetric definition of the integral in eq.(\ref{a-}). This is incorrect, since it violates the residual gauge fixing $\partial^ia^i(x^-\rightarrow -\infty)$ at the one loop level. We will see this explicitly later in this section. At any rate, straightforward albeit somewhat tedious calculation gives the result eq.(\ref{a-}) and this is the expression that will be used in the rest of this paper. So far the formulae presented in this section (except for the corrected expression for $a^-$ eq.(\ref{a-})) are identical to those that appear in \cite{moap} with the only difference that the background charge density $\rho$ has been substituted by the background field via $\rho=-\partial^i\alpha^i$. Now however we will take a different route. Our aim is to calculate the order $O(\alpha_s\ln 1/x)$ correction to the background chromoelectric field eq.(\ref{chrom}) directly, rather than to the background charge density. According to the discussion in the previous section (see also \cite{moap}), we are therefore interested in the following two quantities \begin{align} \alpha_s\ln 1/x\: \chi_{ab}^{ij}(x_\perp,y_\perp) &= <a_{a}^i(x_\perp,x^-\rightarrow\infty, x^+)a_b^j(y_\perp,y^-\rightarrow\infty, x^+)> \\ \alpha_s\ln 1/x\: \sigma_a^i(x_\perp) &= <a_a^i(x_\perp,x^-\rightarrow\infty,x^+)> \end{align} It should be noted that, since the background is static, none of the quantities defined above depend on $x^+$. \subsection{The real part - the mean fluctuation} It is a straightforward matter to calculate $\chi^{ij}_{ab}$. Recall that we need this quantity to order $g^2$. The fluctuation fields $a^\mu$ are formally of order $g$ themselves, and therefore to calculate the mean fluctuation we do not have to include loop corrections. Examining the expression for the general solution eq.(\ref{solut}) we see that it contains oscillating pieces, which do not contribute to the value of the field at infinity as well as the $\gamma_+$ piece, which does not vanish at infinity and, therefore, determines the distribution of the vector potential there. \begin{equation} \chi_{ab}^{ij}(x_\perp,y_\perp)=4\pi\int d p^- <\gamma^{i}_{+,a}(x_\perp,p^-)\gamma^{j}_{+,b}(y_\perp,-p^-)> \end{equation} Using the explicit expressions for $\gamma_+^i$ we find after some trivial algebra \begin{equation} \chi_{ab}^{ij}(x_\perp,y_\perp)= 2<x_\perp|\{{D^i\over D^2}[D^2-S^{-1}]{D^j\over D^2}\}_{ab}|y_\perp> \label{chif} \end{equation} We now want to compare this with the corresponding result of \cite{moap}. The induced charge density $\delta\rho$ in \cite{moap} is \begin{equation} \delta\rho=\delta\rho_1+\delta\rho_2 \end{equation} with \begin{multline} \label{rho11} \delta\rho_{1a}(x_\perp ) = f_{abc} \alpha_{b}^{i}(x_\perp ) \Bigg[a_{c}^{i}(x^-=0)-\int_0^{\infty}dx^- \partial^+ a^i_c(x^-)\Bigg] \\ - {{1}\over{2}} f_{abc} \partial^i \alpha^{i}_{b}(x_\perp ) \int dy^+ \Bigg[\theta (y^+ - x^+) - \theta (x^+ - y^+) \Bigg] a^{-}_{c}(y^+,x_\perp ,x^-=0) \end{multline} and \begin{multline} \label{rho21} \delta \rho_{2a}(x_\perp ) = f_{abc} \int d x^- [\partial^+ a_{b}^{i}(x) ]a_{c}^{i}(x) \\ - {1\over{2}} \partial^i \alpha^{i}_{b}(x_\perp ) \int\! dy^+ a^{-}_{c}(y^+,x_\perp ,x^-=0) \int\! dz^+ a^{-}_{d}(z^+,x_\perp ,x^-=0) \\ \times \Bigg[f_{ace}f_{bde}\theta (z^+ -x^+)\theta (x^+ -y^+) + f_{abe} f_{cde}\theta (x^+ -z^+) \theta (z^+ -y^+) \Bigg] \end{multline} Only $\delta\rho_1$ contributes to $\chi$. Substituting the expressions for $a^i$ and $a^-$ into eq.(\ref{rho11}) we find \begin{equation} \delta\rho_1= -2(\partial D) [{D\over D^2}-{\partial\over\partial^2}] (t-l)v_- \end{equation} Thus, we obviously have \begin{equation} \gamma_+^i=-D^i{1\over \partial D}\delta\rho_1 \label{relat} \end{equation} This reproduces exactly eq.(\ref{corresp1}). Obviously the relation between $\chi_{ab}^{ }$ and $\chi_{ab}^{ij}$, eq.(\ref{indcor}) is also reproduced by this result. We note that our result for $\chi_{ab}^{ }$ is somewhat different than the one presented in \cite{moap}. As discussed before this is due to an incorrect treatment of the $x^-$-independent component of $a^-$ in \cite{moap}. \subsection{The virtual part - the average value of the field} We now proceed to calculate the virtual part of the evolution kernel. For this purpose we have to calculate the zero frequency part of the $\{ij\}$ and $\{i-\}$ components of the fluctuation propagator. The calculation of the $\{ij\}$ at zero frequency is straightforward. The result is \begin{multline} \lim_{p^-\rightarrow 0} G_{ab}^{ij}(x^-,y^-;x_\perp,y_\perp,p^-) \equiv \lim_{p^-\rightarrow 0} <a_a^i(x^-,x_\perp,p^-) a_b^j(y^-,y_\perp,p^-)>= \\ -i\delta^{ij} \delta(x^--y^-) \Big [\theta(x^-) <x_\perp|({1\over D^2})_{ab}|y_\perp>+ \theta(-x^-)<x_\perp|({1\over \partial^2})_{ab}|y_\perp> \Big] \end{multline} The $\{i-\}$ component is then calculated immediately using this result, eq.(\ref{a-}) and the fact noted earlier that the field $\tilde a$ decouples from $a^i$. The result is \begin{multline} \lim_{p^-\rightarrow 0} G_{ab}^{i-}(x^-,y^-;x_\perp,y_\perp,p^-) \equiv \lim_{p^-\rightarrow 0} <a_a^i(x^-,x_\perp,p^-) a^-_b(y^-,y_\perp,p^-)>= \\ i\theta(x^--y^-) \Big[\theta(x^-) <x_\perp|({D^i\over D^2})_{ab}|y_\perp>+ \theta(-x^-)<x_\perp|({\partial^i\over \partial^2})_{ab}|y_\perp> \Big] \end{multline} \begin{fmffile}{moap1pics} \unitlength=1mm \gdef\T#1#2#3#4{ \begin{fmfgraph*}(40,16) \fmfpen{.6thin} \fmfi{wiggly}{(0,.5h) -- (.6w,.5h)} \fmfi{wiggly}{fullcircle scaled .4w shifted (.8w,.5h)} \def\V##1##2##3{ \fmfiv{dec.shape=circle, dec.size=3, lab.angle=##3, label=\noexpand\texttt{\noexpand\small ##1}}{##2}} \V{#1}{(0,.5h)}{95} \V{#2}{(.6w,.5h)}{140} \V{#3}{(.6w,.5h)}{40} \V{#4}{(.6w,.5h)}{-40} \end{fmfgraph*}} \begin{figure}[t] \centering \mbox{ \subfigure[]{\T{$i$}{$j$}{$k$}{$l$}}\qquad \subfigure[]{\T{$i$}{$-$}{$j$}{$j$}}\qquad \subfigure[]{\T{$i$}{$-$}{$-$}{$-$}}} \caption{\small One loop tadpole diagrams contributing to $\sigma^i_a$. The tadpole is calculated at $x^-\rightarrow\infty$.} \end{figure} \end{fmffile} We are now ready to calculate $\sigma_a^i$. It is given by the one loop tadpole diagrams of Fig. 1. The vertex 1c comes from the expansion of the Wilson line term in the action to third order in the fluctuation. The separate contributions of the diagrams can be written in terms of the fluctuation propagator $G^{\mu\nu}\equiv<a^\mu a^\nu>$ in the following form \begin{align} \text{1a} &= \frac{i}{2}\int d y^-d^2 y_\perp G^{ij}_{ab}(x^-, y^-, x_\perp,y_\perp, p^-=0) \epsilon^{jk}D^k_{bc}f_{cde} \epsilon^{mn} G^{mn}_{de}(y^-,y^-;y_\perp,y_\perp,y^+,y^+) \nonumber \\ \text{1b} &= - i \int d y^-d^2 y_\perp G^{i-}_{ab}(x^-, y^-, x_\perp,y_\perp, p^-=0) f_{bcd}{\partial}^{+}_{\tilde{y}^- = y^-} G^{jj}_{cd}(y^-,\tilde{y}^-;y_\perp,y_\perp,y^+,y^+) \nonumber \\ \begin{split} \text{1c} &= \frac{i}{N_c} \int d y^-d^2 y_\perp d y^+ d w^+ d z^+ \\ & \qquad\qquad\times \delta (y^-) ({\partial}^{i}{\alpha}^{i}_{b}(y_\perp)) G^{i-}_{ac}(x^+, y^+,x^-, y^-, x_\perp,y_\perp) G^{--}_{de}(w^+, z^+,y^-, y^-, y_\perp,y_\perp) \\ & \qquad\qquad\quad\times \bigg[ \theta (z^+ - y^+)\theta (y^+ - w^+) f_{bef}f_{cdf} - \theta (y^+ - z^+)\theta (z^+ - w^+) f_{bcf}f_{def} \bigg] \end{split} \end{align} The diagram Fig.~1a corresponds directly to the second term in eq.(\ref{indfield}). For this diagram we immediately find \begin{equation} \delta\sigma^i_{a(1)}(x)= -{1\over 2}\epsilon^{ij} \left[{D^j\over D^2}\right]_{ab}\negthickspace (x,y) f_{bcd}\epsilon^{kl}\chi^{kl}_{cd}(y,y) \label{finals1} \end{equation} The diagramms Fig.~1b and 1c correspond to the first term in eq.(\ref{indfield}) and can be written as \begin{equation} \delta\sigma^i_{a(2)}(x)= -{D^i\over D^2}<\delta\rho_2> \label{stum} \end{equation} with $\delta\rho_2$ (cf eq.(\ref{rho21})): \begin{multline} <\delta\rho_2>_a = f_{abc} \int d x^- < (\partial^+ a_{b}^{i}(x) )a_{c}^{i}(x)> + \, {1\over{2}} \, (f_{ace}f_{bde} - \frac{1}{4} f_{abe}f_{cde}) \, \partial^i \alpha^i_b (x_\perp ) \\ \times \int\! \frac{d\lambda}{\lambda +i\epsilon} dp^- \frac{1}{(p^-)^2} <a^{-}_{c}(p^-, x_\perp ,x^-=0) a^{-}_{d}(-p^-, x_\perp ,x^-=0)> \label{finals} \end{multline} Using the results for the equal time propagators from \cite{moap} we obtain \begin{equation} \begin{split} <\delta\rho_2>_a &= - {1\over 2}\, (f_{ace}f_{bde} - \frac{1}{4} f_{abe}f_{cde})\, \partial^i \alpha^i_b (x_\perp) \\ &\qquad\qquad\times <x_\perp| {1\over \partial^2}+ {1\over 2}\mu{1\over D^2}\mu -2\left [{1\over \partial^2}\alpha D+{\mu\over 2} \right ] {1\over D^2}S^{-1}{1\over D^2} \left [D\alpha{1\over \partial^2} +{\mu\over 2}\right ]|x_\perp>_{cd} \\ &\quad+f_{abc}<x_\perp| \left [t^{ij}-l^{ij} -2\alpha^i\partial^j{1\over \partial^2} \right] \left [\delta^{jk}-2({\partial^j\over\partial^2} -{D^j\over D^2})S^{-1} ({\partial^k\over\partial^2} -{D^k\over D^2})\right ] \\ &\hspace{74mm} \times \left [T^{ki}-L^{ki} -2{1\over \partial^2}\partial^k\alpha^i\right ]|x_\perp>_{bc}\\ &\quad +R^a(x_\perp) \end{split} \label{well} \end{equation} with \begin{equation} \begin{split} R^a(x_\perp) &= f_{abc}\int d^2y_\perp d^2z_\perp {d^2p_\perp d^2k_\perp\over (2\pi)^4} {p_\perp^2\over p_\perp^2-k_\perp^2} e^{ip_\perp (x_\perp-y_\perp)-ik_\perp (x_\perp-z_\perp)}\\ &\qquad\times \Big \{ <y_\perp|\left [t^{ij}-l^{ij}\right] \left [\delta^{jk} -2({\partial^j\over\partial^2}-{D^j\over D^2})S^{-1} ({\partial^k\over\partial^2}-{D^k\over D^2})\right ] \left [t^{ki}-l^{ki}\right ]|z_\perp> \\ &\hspace{20mm} -U(x_\perp)<y_\perp|U^\dagger\left [T^{ij}-L^{ij}\right] \left [\delta^{jk} -2({\partial^j\over\partial^2}-{D^j\over D^2})S^{-1} ({\partial^k\over\partial^2}-{D^k\over D^2})\right ] \\ &\hspace{85mm}\times \left [T^{ki}-L^{ki}\right ]U|z_\perp> U^\dagger(x_\perp) \Big \}_{bc} \end{split} \label{R} \end{equation} Here the singularity in the integrand at $p_\perp^2=k_\perp^2$ has to be understood in the sense of the principal value $${1\over p_\perp^2-k_\perp^2}={p_\perp^2-k_\perp^2 \over(p_\perp^2-k_\perp^2)^2+\epsilon^2}.$$ Our final result for the induced field is given by the sum of eq.(\ref{finals1}) and eq.(\ref{stum}) (supplemented by eqs.(\ref{well},\ref{R})). \section{Conclusions} To summarize, the final results of this paper are eqs.(\ref{chif}) and (\ref{stum},\ref{well}). They supercede the corresponding results of \cite{moap} and \cite{soap}. We now want to comment on this result. The first thing to observe is that the dangerous denominator $\partial D$ does not appear in these expressions. The Gribov problem mentioned earlier therefore does not affect our calculation, at least to order $\alpha_s$. The result for the induced field differs from the corresponding formulae in \cite{moap} and \cite{soap} in two ways. One reason is the improved treatment of $a^-$ relative to \cite{moap}. Now we are in the position to understand why the expression for $a^-$ used in \cite{moap} is inconsistent with the residual gauge fixing. In the previous section we have calculated the induced vector potential far at infinity $x^-\rightarrow\infty$. It is not much more difficult to calculate it everywhere in space. Diagrammatically it is given by the same diagrams as Fig.1 except the coordinate on the free end of the propagator is some finite $x^-$. The difference in the analytic expressions eq.(\ref{stum}) is that the surface charge density $\delta\rho_2$ is substituted by the local charge density integrated up to the longitudinal coordinte $x^-$ \begin{equation} -\theta(x^-){D^i\over D^2}\int_{-\infty}^{x^-}dy^-<\delta j^+_2(y^-)>+ -\theta(-x^-) {\partial^i\over \partial^2}\int_{-\infty}^{x^-}dy^-<\delta j^+_2(y^-)> \label{stump} \end{equation} This expression makes it explicit that the induced field vanishes at $x^-\rightarrow-\infty$. Therefore, our calculation clearly preserves the residual gauge condition $\partial^ia^i(x^-\rightarrow-\infty)=0$. However if we were to subtract the zero momentum piece from the field $a^-$ as done in \cite{moap}, the integration limits in eq.(\ref{a-}) would become symmetric $\int_x^\infty \rightarrow {1\over 2}(\int_x^\infty+\int_x^{-\infty})$. The effect of this would be that $G^{i-}(x^-,y^-)$ would not vanish at $x^-\rightarrow -\infty$. It is then obvious that we would have $\partial^ia^i(x^-\rightarrow -\infty)\ne 0$. The expression obtained in the present paper does not suffer from this problem. It is consistent with the perturbative $i\epsilon$ prescription for regulating the $1/p^+$ gauge pole used in the earlier work \cite{kovchegov}. Another difference between our present result and \cite{moap} is the appearance of $D^2$ rather than $\partial D$ and $D\partial$ in the denominators in eqs.(\ref{finals1},\ref{stum}). This deserves an explanation. This is also related to another point we want to address. Comparing eq.(\ref{stum}) with eq.(\ref{indfield}) one could wonder whether the present method of calculation of $\chi^{ij}$ is consistent with the two step procedure of \cite{moap, soap}. It may look like the relation between the induced field and the induced charge density we obtained here (eq.(\ref{stum})) is different from the equation eq.(\ref{indfield}) which was used in the previous work. This however is not the case. The reason is that the $O(g^2)$ induced charge density $\delta\tilde\rho_2$ which appears in eq.(\ref{indfield}) is not quite the same as $<\delta\rho_2>$ in eq.(\ref{stum}). The $\delta\tilde\rho_2$ was defined as complete $O(g^2)$ contribution to the average of induced density. In other words \begin{equation} \delta\tilde\rho_2=<\delta\rho_1+\delta\rho_2> \label{tilnottil} \end{equation} with $\delta\rho_{1,2}$ defined in eqs.(\ref{rho11},\ref{rho21}). As we discussed above, the fluctuating part of the operator $\delta\rho_1$ is of $O(g)$ and therefore indeed $\delta\tilde\rho_1$ can be identified with $\delta\rho_1$. However, the vacuum average of $\delta\rho_1$ is $O(g^2)$ and does contribute in eq.(\ref{tilnottil}). It can be shown that \begin{equation} <\delta\rho_{1a}>=f_{abc}\alpha_b^i<a^i_c(x^+\rightarrow\infty)> \label{tr} \end{equation} This extra contribution turns $\partial^i\delta\alpha_2^i$ into $D^i\delta\alpha_2^i$ in the second equation in eq.(\ref{solp}) if we use $\delta\rho_2$ rather than $\delta\tilde\rho_2$ in its right hand side. Taking account of this we see that the procedure described in section 2 is consistent with eqs.(\ref{chif},\ref{stum},\ref{well}). In \cite{moap} it was assumed that $<\delta\rho_1>=0$ and thus the extra contribution of eq.(\ref{tr}) was overlooked. This lead to an apparent noncancellation of spurious factors ${1\over \partial D}$ which as we see now, do indeed cancel in the final result. Importantly, the corrections we find vanish in the limit of weak field considered in \cite{jklw} and also in the double logarithmic limit, where the field is considered not necessarily weak but slowly varying in the transverse plane \cite{soap}. This can be seen in the following way. Comparing eq.(\ref{a-}) to the appropriate expression in \cite{moap} we find that the difference between the two is proportional to $\rho$. In the weak field limit one only needs to know $a^-$ to order $1$ and therefore the correction is unimportant. For slowly varying fields all terms proportional to $\rho$ are also negligible. Therefore, the real part - $\chi^{ij}$ - in these two limits is insensitive to the correction we found here. The virtual part - $\sigma^i$ - does not contribute at all in the DLA limit. In the weak field limit the correction is negligible since $<\delta\rho_1>=O((\alpha^{i})^2)$ and one only needs $\delta\rho$ to order $\alpha^i$. \pagebreak {\bf Acknowledgements} We are greatful to J. Jalilian-Marian, L. McLerran and H. Weigert for useful discussions. The work of J.G.M. is supported by PRAXIS XXI/BD/11277/97 grant (Subprograma Ci\^encia e Tecnologia do 2$^{\underline o}$ Quadro Comunit\'ario de Apoio --- Portugal). The work of A.K. is supported by PPARC Advanced fellowship.
\section{Introduction} The question of whether short range Edwards Anderson (EA) spin glasses share the remarkable features of the infinite range Sherrington Kirkpatrick (SK) model \cite{SPINGLASS} is still an open one. In this work we present Monte Carlo simulations of the $4D$ EA Ising Spin Glass \cite{BHAYOU,REBHYO,PARRIT,CIPARI,BCPRPR,PARIRU,BERCAM} with a bimodal distribution of the quenched couplings, performed on large lattice volumes (thanks to the tempering and parallel tempering simulation technique \cite{PARTEM,OPTI}), with a large number of samples, and down to low values of the temperature $T$. In this way we are able to obtain detailed information about the nature of the transition, to determine with good precision critical temperature and exponents, and to give strong evidence supporting the fact that the low-temperature phase is mean-field-like. A great deal of effort has gone in ensuring reliability of the data on delicate issues such as thermalization checks and consistency of data analysis. The paper is organized as follows: first of all we describe the model and the parameters of our MC simulation. We then present data related to the Binder cumulant and to the determination of $T_c$ and $\nu$. By analyzing the overlap susceptibility we determine the value of $\eta$. Finally we discuss in detail about the probability distribution of the overlap $P(q)$. We present, among others, evidence for non-triviality of single sample $P_J(q)$ and for a non-zero value of the position of the maximum of $P(q)$, $q_{max}$, in the thermodynamic limit. \section{The Numerical Simulation} It is very difficult to run reliable numerical simulation of finite dimensional spin glasses. The main reason for such difficulties is the presence of many meta-stable states (responsible for aging effects as well as for many other peculiarities of spin glasses \cite{SPINGLASS}). The Monte Carlo dynamics gets easily trapped, and the system only probes a restricted part of phase space. Many algorithmic solutions have been proposed to improve the speed of thermalization of these systems. All these techniques are related to density scaling methods (see \cite{OPTI} for a review and references): we use here the maybe simplest implementation of these ideas, the parallel tempering~\cite{PARTEM}, where a number of configurations of the system are allowed to exchange their temperature (for multi-canonical methods, that are strongly related and have in principle an even wider range of applicability, see for example \cite{BERG}). Thanks to parallel tempering we have been able to thermalize systems of volume $V=10^4$ down to $T\simeq 1.2 (0.6 T_c)$. We study the $4D$ Edwards-Anderson Ising spin glass with binary couplings, with Hamiltonian \begin{equation} H \equiv - \sum_{\langle i,j \rangle} J_{ij} \sigma_i \sigma_j \ , \end{equation} where the sum runs over nearest neighboring sites, the $\sigma_i$ are $\pm 1$ Ising spins, and the couplings are quenched variables drawn with probability $\frac12$ among the two values $\{ -1, +1 \}$. The {\em overlap} among two different systems is defined as \begin{equation} q^{\alpha, \beta} \equiv \frac{1}{V} \sum_i \sigma_i^{\alpha} \sigma_i^{\beta} \ , \end{equation} where $\alpha$ and $\beta$ denote two configurations of the system in the same realization of the quenched disorder. The overlap probability distribution for a given sample is \begin{equation} P_J(q) \equiv \mt{\delta(q-q^{\alpha, \beta})} \ , \end{equation} where $\mt{...}$ denotes the usual Gibbs average. Its average over samples is \begin{equation} P(q) \equiv \md{P_J(q)} \ , \end{equation} and its moments are defined as \begin{equation} q^{(n)} = \md{\mt{q^n}} = \int dq\ q^n\ P(q) \ . \end{equation} We always denote by $\mt{\cdots}$ the thermal averages and by $\md{\cdots}$ the disorder averages. Our simulation have been performed on a set of workstations, using a multi-spin-coding program that was inspired by the work of \cite{RIEGER}. We have selected the parameters of our Monte Carlo and parallel tempering runs such to guarantee a complete thermalization of the measured observables. We will discuss this issue in some detail. Table (\ref{T-PARAME}) summarizes the relevant parameters used in the simulation: we give among others the number of thermalization steps, of measurements steps, the number of different disorder realizations and the temperature ranges investigated by tempering. Temperature values have been chosen uniformly spaced in the interval between $T_{min}$ and $T_{max}$. \begin{table} \centering \vspace{3mm} \begin{center} \begin{tabular}{|c||c|c|c|c|c|c|c|} \hline $L$ & Thermalization & Equilibrium & Samples & $N_{\beta}$ & $\delta T$ & $T_{min}$& $T_{max}$\\ \hline \hline 3 & 100000 & 100000 & 3200 & 17 & 0.1 & 1.2 & 2.8\\ \hline 4 & 100000 & 100000 & 2944 & 17 & 0.1 & 1.2 & 2.8\\ \hline 5 & 100000 & 100000 & 1920 & 17 & 0.1 & 1.2 & 2.8\\ \hline 6 & 100000 & 100000 & 1120 & 33 & 0.05 & 1.2 & 2.8\\ \hline 8 & 100000 & 100000 & 1376 & 33 & 0.05 & 1.2 & 2.8\\ \hline 10 & 150000 & 150000 & 512 & 56 & 0.04 & 1.2 & 3.4\\ \hline \end{tabular} \end{center} \caption[0]{Parameters of the Tempered Monte Carlo runs. \protect\label{T-PARAME}} \end{table} We have used different methods to verify that we have correctly thermalized the systems. Using ``parallel tempering'', one is actually performing a generalized Markov chain where systems at different temperatures are allowed to ``move'' in temperature-space too. A necessary condition for the Markov chain to be effective in de-correlating different measurements is the fact that each system spans at least a few times all the allowed temperature range during the simulation. In this respect we check {\em a posteriori} that the probability of swapping temperature has been of order $0.5$ (ensuring in this way that a single system did not get stuck at a specific value of $T$) and that the histogram counting the time that each system has spent at each temperature is fairly flat. This requirement is fulfilled in all our simulations, for all $T$ and $L$ values. Another very strong check of thermalization is the fact that the single sample $P_J(q)$ are symmetric in the limits of the statistical significance of the histogram. This is very well verified as can be seen for example in figure \figref{fig:pjq} where we plot $P_J(q)$ for selected samples. \section{The Binder Parameter, $T_c$ and $\nu$} We start by discussing the overlap Binder parameter. We will use it to qualify the phase transition, and to determine the critical temperature and the first of the critical exponents, $\nu$. We will use and describe different methods to compute the quantities we are interested in. Our statistical sample of configurations is a large sample, and our set of data precise (even as far as the dependence over the lattice volume $V$ is concerned): we will show that different analysis styles give compatible (precise) results. We define the usual overlap Binder parameter as \begin{equation} g = \frac{1}{2} \left( 3 - \frac{ \md{\mt{q^4}}}{\md{\mt{q^2}}^2}\right) \ . \end{equation} The Binder parameter is an adimensional quantity, and its value at the critical point is universal. Close to $T_c$ its leading behavior is \begin{equation} g(L,T) \simeq \bar{g}\left(L^{\frac{1}{\nu}} \left(T - T_c\right)\right) \ . \label{eq:binderscaling} \end{equation} In usual ferromagnetic systems the infinite volume limit of the magnetization Binder cumulant is $0$ in the warm phase (where the distribution of the order parameter is Gaussian) and $1$ in the broken phase: for a spin glass with replica symmetry breaking and hence a non-trivial distribution of the overlap order parameter, the transition is signaled by a non-trivial value of $g$ in the broken phase (in the warm phase one expects an infinite volume limit of zero). In both cases the location of $T_c$ is signaled by the crossing of the curves of $g$ versus $T$ for different values of the lattice size $L$ (asymptotically for large $L$): large $L$ curves are lower for $T>T_c$ and higher for $T<T_c$. We show in figure \figref{fig:binder_allsides} $g$ versus $T$ for different $L$ values. The crossing point is close to $T\simeq 2$ for all lattice values, and the value of the Binder cumulant at criticality, $g_c$ is close to $0.45$. Also error analysis has been a sensitive issues. We have always used a jackknife or a bootstrap error analysis \cite{JACKKNIFE} {\em directly} on the fitted parameters to determine errors. Still one has to keep in mind that statistical errors come together with systematic errors, due to the functional form one decides to try to fit (typically the asymptotic scaling form, that on finite size lattices is affected by power corrections). The two types of errors have to be kept under control separately. \begin{figure} \centering \includegraphics[width=0.8\textwidth]{binder_allsides.eps} \caption[qqq]{Binder parameter $g$ versus $T$, for different values of the linear lattice size $L$ (see caption in the plot).} \label{fig:binder_allsides} \end{figure} In figure \figref{fig:binder_allsides} the crossing of the different $g_L$ curves is very clear. It is interesting to stress the difference with the three dimensional case \cite{MAPARU} , where the crossing at $T_c$ looks more like a merging of the different curves. $3d$ is (very) close to the lower critical dimension, while in $4d$ we are in a safe region: potentially this is important to make the physical picture easier to understand. Let us discuss a first naive approach to the data. By looking at the crossing of the curves $g_L(T)$ versus $T$ for different $(L,L+1)$ values one sees that one cannot extract a systematic dependence of the crossing point (and hence of the estimate of the effective critical temperature $T_c(L,L+1)$) over $L$. Any systematic trend is smaller than the statistical error (maybe just showing a systematic average decrease of the estimate $T_c(L,L+1)$ when going from smaller to larger $L$ values). The preferred value of $T_c$ is slightly larger than $2.00$. A first naive estimate of $\nu$ can be done by linearizing $g_L(T)$ around the estimate we have given for $T_c$, and by evaluating the logarithm of the slope ratio (divided by the logarithm of the two lattice sizes ratio, $\log\left(\frac{L}{L+1}\right))$. With this method, one gets a first estimate for a set of effective exponents $\nu(L,L+1)$. Here too, one cannot distinguish any clear strong dependence over the lattice size: the error one gets on $\nu$ is completely correlated to the variation of the estimate of $T_c$. For larger $T_c$ one estimates a lower value of $\nu$, while for lower estimates of $T_c$ one gets larger estimates for $\nu$. The error is dominated by this effect. The estimate for $\nu$ is close to $1$. To get a reliable estimate of $T_c$ and of $\nu$ we have used two methods of analysis of $g$ (see for example the discussion of the analysis of reference \cite{INPARU,MAPARU}). In the first approach we linearize the data close to $T_c$ (for all $L$ values) and we run a global fit to all data: we fit $T_c$ and $\nu$ for the two variable function $g_L(T)$ (as we said, linearized close to $T_c$). We use data in a $T$ range around the interval $1.9-2.1$. We estimate the errors over the fit parameters ($T_c$ and $\nu$) by a jackknife approach~\cite{JACKKNIFE}: we repeat the fit approximately $K$ times over a subsample of the data containing all of our statistical sample but a fraction $\frac{1}{K}$. The error is estimated by looking at fluctuations of the results of the $K$ fits, and by accounting for the fact they are correlated \cite{JACKKNIFE}. We also repeat the fits by discarding the smaller $L$ values, to check if we can observe any systematic drift (again with good accuracy the average value of the result does not seem to depend systematically over the $L$ range selected). Results are very stable, and the value we estimate for $\nu$ systematically comes out to be close to $1.10$. In the second approach, that comes in different flavors, one only uses data in the warm phase. This method leads to a smaller statistical error, that is balanced from a larger systematic incertitude (since we only select data at a given distance from $T_c$, and approaching $T_c$ leads to a systematic drift of the estimate). In this case we start by selecting a threshold value for $g$, $g^* \le g_c$. We start with low values of $g^*$, and we approach $g_c$ from below: we cannot get too close to $g_c$ or the merging of the curves for different sizes makes the error over the measurement too large (we use values of $g^*$ going from $=0.2$ to $0.4$. We use a polynomial fit to interpolate the data for $g(T)$, at different $L$ values. We have decided to use a polynomial of degrees four (we have checked it guarantees stable fits and consistent results), and we fit a $T$ range in the critical region (for $L=3$ we use the data in the $T$ range $1.5-2.8$, for $L=10$ we use the range $1.88-2.16$). We define now $T_c(L,g^*)$ as the crossing point of the fitted polynomial with the horizontal line at $g^*$, and $\nu^*$ as \begin{equation} \lim_{L\to\infty} T_c(L,g^*) = T_c(g^*) + \frac{A}{L^{\frac{1}{\nu^*}}}\ . \end{equation} When $g^*\to g_c$ $\nu^*\to \nu$. If $g^*$ is too small violations of scaling are dominant, while if one approaches too much $g_c$ the merging of the $g$ curves makes the error over the determination of $\nu^*$ overwhelming. The errors have been estimated by using a {\em bootstrap} approach (very similar in spirit to the jackknife technique, see \cite{JACKKNIFE}): one emulates fake sets of data with a Gaussian distribution around the real measurements, fits these multiple sets of fake data and compute the errors over the fit parameter. We note at last that we have also used a variation of this second method, described in \cite{MAPARU}, based on the direct analysis of the derivative of $g$ with respect to $T$. Also this method gives results that are compatible with the other ones. \begin{figure} \centering \includegraphics[width=0.8\textwidth]{binderrescaled.eps} \caption[qqq]{$g$ versus $L^{\frac{1}{\nu}}(T-T_c)$, with $\nu=1.0$, $T_c=2.03$.} \label{fig:binderrescaled} \end{figure} Our final estimates, averaged over the results obtained using these different approaches, are \begin{equation} T_c = 2.03 \pm 0.03\ , \end{equation} and \begin{equation} \nu = 1.00 \pm 0.10\ . \end{equation} In the rest of this paper we will use these two values as our best estimates of $T_c$ and $\nu$. We show in figure \figref{fig:binderrescaled} the data for $g_L(T)$ rescaled by using these two values: the scaling turns out to be very satisfactory. \section{The Overlap Susceptibility and $\eta$} The determination of the overlap susceptibility, $\chi_q$, provides various possible ways to determine the exponent $\eta$ (and hence of the exponent $\gamma$). In a spin glass in the RSB phase the overlap susceptibility \begin{equation} \chi_q \equiv V \langle q^2 \rangle \end{equation} is expected to diverge for all values of $T\le T_c$. We show $\chi_q$ versus $T$ in figure \figref{fig:susc}. \begin{figure} \centering \includegraphics[width=0.8\textwidth]{susc.eps} \caption[qqq]{Overlap susceptibility $\chi_q$ versus $T$, for different $L$ values.} \label{fig:susc} \end{figure} The first method is based on the fact that we expect that at $T=T_c$ \begin{equation} \chi_q(L,T=T_c) \simeq L^{2-\eta}\ . \end{equation} We use a linear interpolation of the data in the region close to $T_c$. As in the case of $g$ the error in the estimate is mainly related to the choice of $T_c$. The fit at $T=2.03$ by using $L>3$ gives an estimate of $0.28$. In the second method we use data where $L\gg\xi$. We go as close to $T_c$ as possible, under the condition that data on our larger lattice ($L=10$) coincide, in our statistical accuracy, with the ones at $L=8$. Here we expect that \begin{equation} \chi_q(T) \simeq (T-T_c)^{-(2-\eta)\nu}\ . \end{equation} We can use data down to $T=2.5$ (i.e. at a $\Delta T\simeq 0.5$ from $T_c$), where finite size effect start to be sizable even at $L=10$. We show our best fit (in a $T$ interval of $=.2$) in figure \figref{fig:susc_warm}. In this region we have a stable fit, with $\eta$ close to $-0.4$. Even if this second measurement is not very precise (we have to stay quite far from the critical region) it is interesting the fact that we get a coherent determination of $\eta$, by using a completely different scaling region than in the former analysis (the new analysis also depends on the value of $\nu$ we have determined by using $g$). \begin{figure} \centering \includegraphics[width=0.8\textwidth]{susc_warm.eps} \caption[qqq]{Best fit to the overlap susceptibility $\chi_q$ versus $T-T_c$, at $T>T_c$.} \label{fig:susc_warm} \end{figure} The last approach we use for determining $\eta$ is based on the analysis of the scaling properties of the distribution probability $P(q)$ of the overlap order parameter $q$ in the region $q\simeq 0$ at $T=T_c$. We analyze the behavior of $P(q)$ in the next section, but we discuss now the scaling of $P(0)$ of $T_c$ in order to define our determination of $\eta$. At $T=T_c$ we expect \begin{equation} P(q\simeq 0) \simeq L^{\frac{d-2+\eta}{2}}\ , \end{equation} i.e. in $d=4$ a scaling with $L^{\frac{2+\eta}{2}}$. We find a very good best fit (we do not include the $L=3$ data), with an $\eta$ value close to $-0.3$. By considering all the methods we have discussed in this section we give our final estimate \begin{equation} \eta = -0.30 \pm 0.05\ , \end{equation} that we will use in the rest of our analysis. In figure \figref{fig:suscrescaled} we plot $\chi_q$ rescaled by using our best fits. The rescaling works fine. \begin{figure} \centering \includegraphics[width=0.8\textwidth]{suscrescaled.eps} \caption[qqq]{ Rescaled overlap susceptibility $\frac{\chi_{SG}}{L^{2-\eta}}$ versus $L^{\frac{1}{\nu}}(T-T_c)$, with $T_c=2.03$, $\nu=1.0$ and $\eta=-0.30$.} \label{fig:suscrescaled} \end{figure} Let us also notice that we have a good agreement with the results reported in \cite{PARIRU} for the $4d$ EA model with Gaussian couplings. There the authors find $\nu\simeq 1.06$, and $\eta \simeq -0.35$. Universality seems to work. \section{$P(q)$} \begin{figure} \centering \includegraphics[width=0.8\textwidth]{pdiqlowtallsides.eps} \caption[qqq]{$P(q)$ at $T= 1.2$ (broken phase), for different lattice sizes.} \label{fig:pqt12} \end{figure} In the two former sections we have shown that the $4D$ EA model undergoes a phase transition, and we have determined its location and the critical exponents. Now we will try to qualify it in better detail, by determining and analyzing the probability distribution of the order parameter, $P(q)$. In figure \figref{fig:pqt12} we show our average $P(q)$ (averaged over the different disorder realizations) at $T=1.2<T_c$. When increasing the lattice size the peak where $P(q)$ is maximum shifts to lower $q$ values: for showing that there is a phase transition to a phase with a non zero expectation value of $q$ we have to show that the peak does not go to $q=0$ when $L\to\infty$. The {\em plateau} of $P(q)$ for $q\simeq 0$ does not lower when increasing $L$, as we will discuss better in the following. We remind the reader that in the RSB Parisi Mean Field scenario the $P(q)$ is (in zero magnetic field) a non trivial function, that in the infinite volume limit is formed by a $\delta$ function at $q=q_{EA}$ and by a regular part that extends down to $q=0$. On the contrary if the broken phase has the same structure of the one of an ordered ferromagnet $P(q)$ has to become asymptotically the sum of two $\delta$ functions at $\pm q_{EA}$. \begin{figure} \centering \includegraphics[width=0.8\textwidth]{p_di_q_allT2200.eps} \caption[qqq]{$P(q)$ at $T= 2.2$ (warm phase), for different lattice sizes.} \label{fig:pqt22} \end{figure} For sake of comparison we show in figure \figref{fig:pqt22} what happens in the warm phase, where the average $P(q)$ shrinks to a Gaussian distribution around $q=0$ when $L\to\infty$. \begin{figure} \centering \includegraphics[width=0.8\textwidth]{p_di_q_alltempL8.eps} \caption[qqq]{$P(q)$ at $L=8$, for all values of $T$. From single peak in $q=0$ to continuous part and double peak at large $q$ for increasing $T$.} \label{fig:pqL8} \end{figure} In figure \figref{fig:pqL8} we compare $P(q)$ at different values of $T$ on the same lattice size. From the single peaked shape at high $T$ one gets a clear double peaked structure, with a clear {\em plateau} at low $q$, in the low $T$ region. \begin{figure} \centering \includegraphics[width=0.8\textwidth]{qmax_vs_L1.3r0_025.eps} \caption[qqq]{$q_{max}$ versus $L^{-1.3}$.} \label{fig:qmL13} \end{figure} As we have said, in order to establish that we are having a real phase transition in the infinite volume limit we have to show that the value of $q=q_{max}$ where $P(q)$ is maximum does not go to zero. We start by plotting in figure \figref{fig:qmL13} $q_{max}$ versus $L^{-1.3}$ (the exponent $1.3$ comes from our best fit, see later). It is easy to see that an asymptotic value $q_{max}=0$ seems unplausible. \begin{figure} \centering \includegraphics[width=0.8\textwidth]{qmax_t1300.eps} \caption[qqq]{$q_{max}$ versus $L$ and our two best fits.} \label{fig:qmax_t1300} \end{figure} Making this last statement more quantitative needs a more careful analysis. In order to do that we fit \begin{equation} q_{max}(L) = q_{max}(\infty) + \frac{A}{L^{\alpha}}\ , \end{equation} both with $q_{max}(\infty)=0$ and by allowing for it a non zero value. In figure \figref{fig:qmax_t1300} we plot the values of $q_{max}$ versus $L$ and the results of the two best fits, one with a fitted value of $q_{max}(\infty)$ and the second with fixed $q_{max}(\infty)=0$. This second fit is clearly unsuitable, and it has a very high value of $\chi^2$. In the first fit we get \begin{equation} q_{max}(\infty) = 0.548 \pm 0.006\ , \end{equation} that is our best estimate for the position of the $\delta$ function at $q_{EA}$ in the infinite volume limit. We estimate $\alpha=1.3\pm 0.1$ (in the fit with a zero asymptotic value one finds the very small value $\alpha\simeq 0.2$). \begin{figure} \centering \includegraphics[width=0.8\textwidth]{pdizeromediata1200.eps} \caption[qqq]{$P(q\simeq 0)$ versus $L$. $T=1.2$.} \label{fig:p0} \end{figure} In figure \figref{fig:p0} we show the value of $P(q)$ close to $q=0$ (averaged over a small $q$ range, where $P(q)$ is remarkably constant, in order to diminish statistical fluctuations) as a function of $L$. One cannot observe any statistically significant decrease of this value for increasing large lattice volume (there is a small decrease only for small volumes). The most plausible implication of this evidence is that the system has many stable states, and that the cold $T$ phase is characterized by Replica Symmetry Breaking (even if it has to be stressed that this evidence is not as strong as the one implied by the figure \figref{fig:qmax_t1300}). \begin{figure} \centering \includegraphics[width=0.8\textwidth]{2pdiq.eps} \caption[qqq]{$P_J(q)$ for selected samples. $T=1.2$, $L=10$.} \label{fig:pjq} \end{figure} \begin{figure} \centering \includegraphics[width=0.8\textwidth]{isto_max.eps} \caption[qqq]{ Percentage of disorder configurations such that $P_J(q)$ has $1$, $2$, $4$ and $6$ peaks versus $L$. The number of configurations with a complex phase space ($P_J(q)$ with many peaks) increases strongly with $L$.} \label{fig:isto_max} \end{figure} In figure \figref{fig:pjq} we plot $P_J(q)$ for selected samples, at $T=1.2$, $L=10$. One can see here that they are very complex distributions: such a pattern is typically related to the presence of many states (it has to be notice however that the small side peaks are not always there because of the presence of a real state). To be more quantitative we have measured the percentage of disorder configurations such that $P_J(q)$ has $1$, $2$, $4$ and $6$ peaks versus $L$, and we plot it in figure \figref{fig:isto_max}. The number of configurations with a complex phase space ($P_J(q)$ with many peaks) increases strongly with $L$. We use this evidence to rule out the picture of a {\em modified droplet model}, that has been discussed, among others, in \cite{NS} and in references therein. The picture of the modified droplet models implies that for each realization of the quenched disorder there are (in the cold phase) only two ground states, but that the value of $q_{EA}$ (i.e. the support of the $\delta$ function that constitutes the $P_J(q)$) depends on the sample. Here, on the contrary, the number of states for a given sample is strongly increasing with $L$ (and with decreasing $T$). \section{Sum Rules} \begin{figure} \centering \includegraphics[width=0.8\textwidth]{lhs_rhs_sumrule.eps} \caption[qqq]{ $\md{\mt{q^2}^2}$, $\md{\mt{q^2}}^2$ and $\md{\mt{q^4}}$ vs. $T$. } \label{fig:lhs_rhs_sumrule} \end{figure} In this section we discuss another important feature of the broken phase of the $4D$ EA model. The starting point for this analysis can be for example the relation: \begin{equation} \md{\mt{q^2}^2} = \frac23 \md{\mt{q^2}}^2 + \frac13 \md{\mt{q^4}} \ . \protect\label{G-A} \end{equation} This is one of a set of relations that are valid in the Mean Field Theory of Spin Glasses \cite{MPSTV}. The work contained in \cite{SUMRULE} has established numerically that these relations are satisfied with good accuracy also in finite dimensional spin glasses. Following these findings a rigorous and theoretical analysis has improved our understanding of such set of sum rules \cite{GUERRA,AIZCON,PARISISR,NSLONG}: they are strongly related to the ultrametric properties of the phase space. First of all we show evidence that the relation (\ref{G-A}) has a non-trivial content in the low-temperature phase (in the high $T$ phase it is satisfied in the form $0=0$). Figure \figref{fig:lhs_rhs_sumrule} shows that the values of the three quantities involved in (\ref{G-A}) are significantly different from zero below $T_c$ (we have already shown in better detail that the infinite volume of such quantities is non-zero). \begin{figure} \centering \includegraphics[width=0.8\textwidth]{lhs_minus_rhs_sumrule.eps} \caption[qqq]{Left hand side minus right hand side of equation \eref{G-A} vs. $T$.} \label{fig:lhs_minus_rhs_sumrule} \end{figure} In figure \figref{fig:lhs_minus_rhs_sumrule} we show the difference between the left hand side and the right hand side of \eref{G-A}. The two contributions cancel out with good accuracy (to $3$ significant figures), and asymptotically for large lattice size the difference extrapolates to zero. \begin{figure} \centering \includegraphics[width=0.8\textwidth]{ratio_sumrule.eps} \caption[qqq]{Ratio of left hand side and right hand side of equation \eref{G-A} versus $T$.} \label{fig:ratio_sumrule} \end{figure} Another possible way to visualize the result is plotting the ratio of the left hand side and the right hand side. As figure \figref{fig:ratio_sumrule} shows, for $T$ below $T_c$ we get identically one, while for $T \to \infty$ we get the value $\frac35$, expected for a Gaussian $P(q)$. \begin{figure} \centering \includegraphics[width=0.8\textwidth]{scaling_sumrule.eps} \caption[qqq]{ Left hand side minus right hand side of equation \eref{G-A} versus $L$, at $T=1.4$, and and best fit to a zero constant value with a simple power correction.} \label{fig:scaling_sumrule} \end{figure} We have also fitted the difference plotted in figure \figref{fig:lhs_minus_rhs_sumrule}, for various values of temperatures $T<T_c$: in all cases a fit to an asymptotic zero value with power corrections works very well, and the exponent of the corrections is close to $3$ for all $T$ values. As an example we plot the data together with the best fit for $T=1.4$ in figure \figref{fig:scaling_sumrule}. \section{Conclusions} In this note we have been able to give strong evidence for mean field behavior of the $4d$ Ising spin glass with binary couplings. Life in the $4d$ model is easier than in $3d$, where even after a large number of intense numerical simulations the evidence for a phase transition is still slightly marginal (even if, at this point, convincing enough). In our case already the crossing of the Binder cumulants is the very clear signature of a typical phase transition (as opposed to the quasi-merging, quasi-Kosterlitz-Thouless behavior of the $3d$ case). It is clear that $3d$ is very close to the lower critical dimension, and that there observing the effects of the physical critical point is dramatically difficult. $4d$ is on the safer side, and numerical simulations show that very clearly. We have been able to determine critical exponents precisely, and to enter in the large volume region with good accuracy. For example we have been able to show that the peak of $P(q)$ is not going to $q=0$ for increasing lattice size, and (with a slightly worst accuracy and level of reliability) that the plateau at $q\simeq 0$ does not decrease with increasing lattice size. Also we remind the reader that non-trivial sum rules are satisfied with very good accuracy. So, thinks look quite clear in the $4d$ case. \section*{Acknowledgments} We thank Giorgio Parisi, Federico Ricci-Tersenghi and Juan Ruiz-Lorenzo for a number of interesting discussions.
\section{Introduction} This is the ninth paper of a series whose purpose is to obtain Cepheid distances to galaxies that have produced supernovae of type Ia (SNe~Ia), thereby calibrating their absolute magnitudes at maximum light. The Hubble diagram for SNe~Ia that are not abnormal in either their intrinsic colors or their spectra at maximum (\cite{bra93} 1993) is exceedingly tight (\cite{sata93} 1993; \cite{tasa95} 1995; \cite{saha97} 1997 ; \cite{par99} 1999) even before second-order corrections for light curve decay rate or color (\cite{ham95} 1995, 1996a, b; \cite{rie96} 1996) are applied. Hence, the absolute magnitude calibrations lead directly to a good estimate of the global value of the Hubble constant because the SNe~Ia Hubble diagram is defined at redshifts that are well beyond any local velocity anomalies in the Hubble flow. The previous papers of this series concern Cepheids in IC~4182 for SN~1937C (\cite{sanetal92} 1992, Paper~I; \cite{saha94} 1994, Paper~II), NGC~5253 for the two SNe~Ia 1895B and 1972E (\cite{sanetal94} 1994, Paper~III; \cite{saha95} 1995, Paper~IV), NGC~4536 for SN~1981B (\cite{saha96a} 1996a, Paper~V); NGC~4496A for SN~1960F (\cite{saha96b} 1996b, Paper~VI); NGC~4639 for SN~1990N (\cite{sanetal96} 1996, Paper~VII; \cite{saha97} 1997, Paper~VIII). The purpose of this paper is to set out the data for the discovery and photometry of Cepheids in NGC~3627, the parent galaxy of the type Ia SN~1989B. This case, together with the new SN~Ia 1998bu in NGC~3368 for which \cite{tan95} (1995) have a Cepheid distance, and the new Cepheid distance of NGC~4414 (\cite{tur98} 1998), parent galaxy to SN~1974G, and the compilation of the extant photometric data on SN~1998bu (\cite{sun98} 1998) and on SN~1974G (\cite{schae98} 1998), now increase the number of SNe~Ia calibrators from seven in \cite{sanetal96} (1996) and \cite{saha97} (1997) to nine here. It will be recalled that we have used the absolute magnitude of SN~1989B in two previous discussions (\cite{sanetal96} 1996; \cite{saha97} 1997) but based on the then unproven premise that the \cite{tan95} (1995) Cepheid distance to NGC~3368 would be the same as the distance to NGC~3627, as both galaxies are in the loose Leo association. SN~1989B was discovered by visual inspection of NGC~3627 by \cite{eva89} (1989) on January 30, 1989 which was 7 days before maximum light in $B$. A detailed light curve in {\it UBVRI} and a determination of the reddening of $E(B-V) = 0\fm37 \pm 0\fm03$ was made by \cite{wel94} (1994), following the earlier analysis by \cite{bar90} (1990). The magnitudes at maximum corrected for the extinction are $B({\rm max}) = 10\fm86 \pm 0\fm13$ and $V({\rm max}) = 10\fm88 \pm 0\fm10$. The parent galaxy NGC~3627 (M~66) is one of the brightest spirals (Sb(s)II.2) in the complicated region of the Leo group, first isolated by \cite{hum56} (1956, Table~XI) where 18 possible members were identified including NGC~3627 and NGC~3368 (M~96). Both galaxies are illustrated in the Hubble Atlas (\cite{san61} 1961, panels 12 and 23), and the Carnegie Atlas (\cite{sanbed94} 1994, Panels 118, 137, and S14). \cite{dev75} (1975) has divided the larger Leo group complex into three groups. G9 in his Table~3 is a spiral--dominated subgroup near NGC~3627 (M~66). His G11 is dominated by NGC~3368 (M~96), parent to SN~1998bu and contains NGC~3351, for which a Cepheid distance is also available (\cite{gra97} 1997). The agreement of the Cepheid distances of NGC~3368 and NGC~3351 to within $0\fm36 \pm 0\fm25$ is satisfactory. The smaller group G49 surrounds NGC~3607. The division into three subgroups is supported by the three-dimensional hierarchical clustering analysis of \cite{mat78} (1978). A catalog of 52 possible members of the subgrouping in the field of NGC~3368, NGC~3379, and NGC~3384 is given by \cite{ferg90} (1990). Because of the complication of three subgroups rather than a well defined single group, we have never been totally convinced that the distance to NGC~3627, required to calibrate SN~1989B, could be taken to be that of the Cepheids in NGC~3368 from \cite{tan95} (1995), although we analyzed our calibration data in Papers~VII and VIII on that premise. This assumption is now no longer necessary with the discovery and analysis of Cepheids in NGC~3627. Figure~\ref{fig1} shows a ground-based image of NGC~3627 made from the Mount Wilson 100-inch Hooker blue plate (E40) that was used in the Hubble Atlas and for one of the frames of the Carnegie Atlas. The position of the four WFPC2 chips of the $HST$ is superposed to show our search area for the Cepheids. The position of SN~1989B is marked, taken from the photographs in \cite{bar90} (1990). Figure~\ref{fig2} is a color composite $HST$ montage of stacked frames. The dust lanes are striking both here and in Fig.~\ref{fig1}. The extinction clearly varies over short scales on the image. This forewarns about the importance of differential extinction to the Cepheids, requiring good color information for them to obtain a reliable true modulus. The extinction in NGC~3627 is more severe than in any galaxy of our six previous calibrations in Papers~I--VIII of the series. The journal of the $HST$ observations and the photometry of the master template frame are in the next section. The identification and classification of the variables are in \S~3. The apparent period-luminosity relations, the analysis of the severe absorption problem and how we have corrected for it, the resulting distance modulus, and the absolute magnitude at maximum for SN~1989B are in \S~4. In \S~5 we combine the data with our six previous calibrators, as well as SN~1974G in NGC~4414 and SN~1998bu in NGC~3368 from external sources. This paper provides a direct Cepheid distance to NGC~3627 and removes the uncertainty, thereby strengthening the weight of SN1989B in the overall calibration. A first value of $H_0$ is given. A discussion of the second-parameter corrections and their effect on $H_0$ is in the penultimate section \S~6. \section{Observations and Photometry} \subsection{The Data} Repeated images of the field in NGC~3627, shown in Fig.~\ref{fig2}, were obtained using the WFPC2 (\cite{hol95a} 1995a) on the $HST$ between November 1997 and January 1998. There are 12 discrete epochs in the $F555W$ passband, and 5 epochs in the $F814W$ passband, spanning a period of 58 days. The duration of this period is constrained by the time window during which this target can be observed with $HST$ without altering the field orientation. This window was further curtailed due to logistics for accommodating a campaign with the NICMOS camera, for which a change of the telescope focus was necessary, thus rendering the telescope useless for other kinds of observations. The epochs were spaced strategically over this period to provide maximum leverage on detecting and finding periods of Cepheid variables over the period range 10 to 60 days. Each epoch in each filter was made of two sub-exposures taken back-to-back on successive orbits of the spacecraft. This allows the removal of cosmic rays by an anti-coincidence technique. The images from various epochs were co-aligned to within 3--4 pixels on the scale of the PC chip, which is 1--2 pixels on the scale of the other three wide-field chips. The journal of observations is given in Table~\ref{tbl1}. \subsection{Photometry} The details of processing the images, combining the sub-exposures for each epoch while removing cosmic rays and performing the photometry with a variant of DoPHOT (\cite{schec93} 1993) optimized for WFPC2 data has been given in \cite{saha96a} (1996a) and need not be repeated here. The data reduction procedure is identical to that described in Paper~V, with the one exception of a change in the definition of the ``partial aperture''. Instead of the $9 \times 9$ pixel aperture, a circular aperture of 5 pixel radius was used, and the local background is defined as the value for which the aperture growth curve is flat from 6 to 8 pixels. The details of these changes are given in \cite{ste98} (1998) in their \S~2.2. This change produces no known systematic differences, but improves the S/N with which aperture corrections are measured. In keeping with the precepts in Paper~V, the measurements in any one passband are expressed in the magnitude system defined by \cite{hol95b} (1995b) that is native to the WFPC2. These are the $F555W$ and $F814W$ ``ground system'' magnitudes calibrated with $HST$ ``short'' exposure frames. The issue of the discrepancy of the photometric zero-points for the ``long'' and ``short'' WFPC2 exposures, originally found by \cite{ste95} (1995) is described in some detail in Paper~V. In any eventual accounting for this zero-point correction, one must {\it add} $0\fm05$ in {\it both} passbands to the \cite{hol95b} (1995b) calibration whenever the exposures are longer than several hundred seconds. The cause of this zero-point difference is not fully understood at the time of this writing, and, as in previous papers of this series, we continue to present the basic photometry (Tables~\ref{tbl3} and \ref{tbl4}) on the ``uncorrected'' \cite{hol95b} (1995b) ``short exposure'' calibration. And, because all of our WFPC2 observations have exposure times that are ``long'', we make the $0\fm05$ adjustment only at the resulting distance modulus of NGC~3627 (cf. \S~4.2.2.). Correspondingly the distance moduli in Table~\ref{tbl5} are corrected to the `long' calibration scale by $0\fm05$, except the moduli of IC~4182 and NGC~5253 which were observed with the older WF/PC. \section{Identification and Classification of the Variable Stars} Armed with measured magnitudes and their reported errors at all available epochs for each star in the object list, the method described by \cite{saha90} (1990) was used to identify variable stars. The details specific to WFPC2 data have been given in various degrees of detail in Papers~V, VI, and VIII. All variable stars definitely identified are marked in Fig.~\ref{fig3}. However, some of the identified variables cannot be seen in Fig.~\ref{fig3} because of their extreme faintness and/or because of the large variation in surface brightness over the field. Hence, to complement these charts, we set out in Table~\ref{tbl2} the X and Y pixel positions for all variable stars as they appear in the images identified in the $HST$ data archive as U3510701R and U3510702R. The photometry on the \cite{hol95b} (1995b) ``short exposure'' calibration system for the final list of 83 variable stars is presented in Table~\ref{tbl3} for each epoch and each filter. The periods were determined with the \cite{laf65} (1965) by using only the $F555W$ passband data. Aliassing is not a serious problem for periods between 10 and 58 days because the observing strategy incorporated an optimum timing scheme as before in this series. The resulting light curves in the $F555W$ passband, together with periods and mean magnitudes (determined by integrating the light curves, converted to intensities, and then converting the average back to magnitudes, and called the ``phase-weighted intensity average'' in \cite{saha90} 1990), are shown in Fig.~\ref{fig4}, plotted in the order of descending period. Four objects, C2-V23, C2-V25, C2-V36, and C2-V37 are definitely variable but are unlikely to be Cepheids. They may be periodic with periods greater than the time spanned by our 12 epochs, or they may be transient variables such as novae. They are plotted in Fig.~\ref{fig4} with periods set artificially at 100 days (much larger than the observing time base of 58 days) for the purpose of visualization. Four obvious objects, C1-V1, C2-V38, C4-V3, and C4-V14 have periods that are just a little larger than the observing time base. Best guesses of the periods have been made from the light curve shapes. As the periods of these variables are not definitive, they should be used with caution in deriving distances. The remaining variables have light curves and periods that are consistent with being Cepheids. The available data for the variables in $F814W$ were folded with the ephemerides derived above using the $F555W$ data. The results are plotted in Fig.~\ref{fig5}. The long-period variables, transient variables, and long-period Cepheids (as discussed above) are shown with the same assigned ephemerides as in Fig.~\ref{fig4}. Not all the variables discovered from the $F555W$ photometry were found in the $F814W$ images. The fainter variables, either because they are intrinsically faint or else appear faint due to high extinction, may not register clearly on the $F814W$ frames which do not reach as faint a limiting magnitude as those in $F555W$. Since photometry of such objects was obviously impossible in $F814W$, these variables are dropped from Fig.~\ref{fig5} and also from further analysis. Only the 68 variables from the $F555W$ frames that were recovered in at least one of the $F814W$ epochs are considered further. The mean magnitudes in $F814W$ (integrated as intensities over the cycle) were obtained from the procedure of \cite{lab97} (1997) whereby each $F814W$ magnitude at a randomly sampled phase is converted to a mean value $\langle{F814W}\rangle$ using amplitude and phase information from the more complete $F555W$ light curves. Note that each available observation of $F814W$ can be used independently to derive a mean magnitude. Hence, the scatter of the individual values about the adopted mean $F814W$ value is an {\it external} measure of the uncertainty in determining $\langle{F814W}\rangle$. This value is retained as the error in $\langle{F814W}\rangle$, and propagated in the later calculations. The prescription given in Paper~V for assigning the light-curve quality index QI (that ranges from 0 to 6) was used. In this scheme, two points are given for the quality of the $F555W$ light curves, two points for the evenness in phase coverage of the five $F814W$ observation epochs, and three points for the amplitude and phase coherence of the $F814W$ observations compared with the $F555W$ light curve. Hence, a quality index of 6 indicates the best possible light curve quality. A quality index of 2 or less indicates near fatal flaws such as apparent phase incoherence in the two passbands. This is generally the indication that object confusion by crowding and/or contamination by background is likely. Table~\ref{tbl4} lists the characteristics of the 68 objects whose light curves in $F555W$ are consistent with those of Cepheids, and for which an $F814W$ measurement exists for at least one epoch. The $F555W$ and $F814W$ instrumental magnitudes of Table~\ref{tbl3} have been converted to the Johnson $V$ and Cousins (Cape) $I$ standard photometric system by the color equations used in Papers V to VIII of this series, as set out in equations (2) and (3) of Paper~V, based on the transformations of \cite{hol95b} (1995b). The magnitude scatter $\sigma_{\langle{V}\rangle}$ in Table~\ref{tbl4} is based on the estimated measuring errors in the photometry of the individual epochs. The determination of the scatter $\sigma_{\langle{I}\rangle}$ is described above. The quality index discussed above is also listed. Other columns of Table~\ref{tbl4} are explained in the next section. \section{The Period-Luminosity Relation and the Distance Modulus} \subsection{The P-L Diagrams in $V$ and $I$} As in the previous papers of this series we adopt the P-L relation in $V$ from \cite{mad91} (1991) as \begin{equation} M_{V} ~~=~~ -2.76 ~\log P - 1.40~, \end{equation} whose companion relation in $I$ is \begin{equation} M_{I} ~~=~~ -3.06 ~\log P - 1.81~. \end{equation} The zeropoint of equations (1) and (2) is based on an adopted LMC modulus of 18.50. The P-L relations in $V$ and $I$ for the 68 Cepheids in Table~\ref{tbl4} are shown in Fig.~\ref{fig6}. The filled circles show objects with periods between 20 and 58 days that have a quality index of 4 or higher. These are the best observed Cepheids. No selection based on color has been made here, hence the total range of differential extinction values is contained in the data as plotted, explaining part of the large scatter. The continuous line in each of the two panels shows equations (1) and (2) as the ridge-line relation using an apparent distance modulus of 30.2. The faint dashed upper and lower envelope lines indicate the expected scatter about the mean due to the intrinsic width of the Cepheid instability strip in the HR diagram (\cite{sata68} 1968). The large observed scatter of the data outside these envelope lines are due to the combination of (1) measuring and systematic errors due to background and contamination, (2) the random error of photon statistics, and (3) the large effects of the variable extinction evident from Fig.~\ref{fig2}. Any modulus inferred directly from Fig.~\ref{fig6} would be unreliable. No matter how the lines defining the P-L strip are shifted vertically in the two panels of Fig.~\ref{fig6}, a large fraction of the points of the total sample will remain outside the boundaries of the instability strip. Analysis of the scatter is the subject of the next two subsections. \subsection{Deriving the Distance Modulus} \subsubsection{A Preliminary Analysis of the P-L Relation} Inspection of Fig.~\ref{fig6} reveals that the deviations from the ridge-lines in $V$ and $I$ of individual Cepheids are correlated. Stars that deviate faintward in $V$ also generally deviate faintward in $I$ and vice versa. This, of course, is a signature of variable extinction, but it can also be caused if the {\it systematic} measuring errors are correlated due, for example, to confusion or to errors in the compensation for background contamination that are not independent in the $V$ and $I$ passbands. To explore these possibilities and to correct for them we use the tools developed in Paper~V and used again in Papers~VII and VIII. For each Cepheid we calculate the apparent distance moduli separately in $V$ and in $I$ from the P-L relations of equations (1) and (2) and the observed $\langle{V}\rangle$ and $\langle{I}\rangle$ magnitudes from Table~\ref{tbl4}. These apparent distance moduli, called $U_{V}$ and $U_{I}$ in columns (7) and (8) of Table~\ref{tbl4}, are calculated by \begin{equation} U_{V} ~~=~~ 2.76 ~\log P + 1.40 + \langle{V}\rangle~, \end{equation} and \begin{equation} U_{I} ~~=~~ 3.06 \log P + 1.81 +\langle{I}\rangle~. \end{equation} They are the same as equations (6) and (7) of Paper~V. If the differences between the $V$ and $I$ moduli are due solely to reddening, and if the dependence of the reddening curve on wavelength is the normal standard dependence as in the Galaxy, then the true modulus $U_T$ is given by \begin{equation} U_{T} ~~=~~ U_{V} - R'_{V} \cdot (U_{V} - U_{I})~, \end{equation} where $R$ is the ratio of total to selective absorption, $A_{V}/E(V-I)$. This is equation (8) of Paper~V. However, equation (5) is valid only if the difference between $U_{V}$ and $U_{I}$ is due to extinction, not to correlated measuring errors, in which case the value of $R$ would {\it not} be given by the normal extinction curve where $A_{V}/A_{I} = 1.7$ and the ratio of absorption to reddening is $R'_{V} = A_{V}/E(V-I) = 2.43$ (\cite{schef82} 1982). Such coupled errors clearly exist (based on Fig.~\ref{fig7} later where the slope of the $U_{V}$ vs. $U_{I}$ correlation is closer to 1 than to the required slope of $A_{V}/A_{I} = 1.7$ if the correlation were to be due entirely to differential extinction rather than to measuring errors). Hence, the interpretation of the values calculated from equations (3), (4), and (5) is considerably more complicated than would be the case in the absence of the correlated systematic measuring errors. Nevertheless, in the initial pass at the data to derive the true modulus, we use equations (3) to (5) as a first approximation. The second approximation, based on knowledge gained by the methods of this section, is in the next section 4.2.2. The values of $U_{T}$ are listed in column 9 of Table~\ref{tbl4}. These would be the true moduli, as corrected for normal extinction, assuming that there are no systematic measuring errors. The total rms uncertainty for each $U_{T}$ value is listed in column 10. This uncertainty includes contributions from the estimated random measuring errors in the mean $V$ and $I$ magnitudes, (in columns 4 and 6), as propagated through the de-reddening procedure, as well as the uncertainty associated with the intrinsic width of the P-L relation, i.e. a given Cepheid may not be on the mean ridge-line of the P-L relation. The de-reddening procedure amplifies the measuring errors. Therefore many Cepheids are needed to beat down these large errors (notice the very large values in column 10) in any final value of the modulus. The values shown in column 10 of Table~4 were calculated using equations 9, 10 and 11 of Paper~V, and correspond to $\sigma_{tot}^2$ as defined in Paper~V. However, note that equation (11) in Paper~V should be : \begin{displaymath} \sigma_{width}^2 = (R'_{V}-1)^2 \cdot \rho_{V}^2 + {R'_{V}}^2 \cdot \rho_{I}^2~. \end{displaymath} Our records show that while this equation was given incorrectly in Paper~V, the calculations were done with the correct relation. Various arithmetics done on the $U_{T}$ values in column 9 of Table~\ref{tbl4} give the first indication of the true modulus. Consider first all Cepheids of all quality indices, but excluding C1-13 and C3-V15 because of their extreme values of $U_{T}$. The 66 Cepheids in this sample give $\langle{U_{T}}\rangle = 29.90 \pm 0.08$ (the error estimate is based on the adopted rms of a single Cepheid of $0\fm646$). Restricting the sample to only those Cepheids with good to excellent data defined by a quality index of 4 or higher gives a subsample of 41 Cepheids for which $\langle{U_{T}}\rangle = 30.09 \pm 0.085$ (error based on an rms of $0\fm535$). Note that the 3-sigma upper and lower limits on the true modulus from this arithmetic are 30.34 and 29.83. These are therefore very strong upper and lower limits on the true distance modulus of NGC~3627. These are the best values we can derive without making cuts in the data according to period, QI, and/or color that would select the bluest and least reddened, and that would be least affected by selection bias at the low-period end. We consider now such subsamples, first using cuts in period and QI, but not yet in color which follows in the next section. A plot of $U_{T}$ vs. period (not shown) reveals a trend that objects with the shortest periods yield smaller $U_{T}$ moduli. This result has been seen by us in the previously analyzed galaxies, and by other investigators analyzing similar data in yet other galaxies. It is due to a combination of selection bias at the short-period end (\cite{san88} 1988) as well as non-symmetrical observational bias in measuring colors near the faint limit, skewing $U_{T}$ via the color effect in equation (5). We note that this trend disappears once periods are restricted to longer than 25 days. We also note that the variation of $U_{T}$ with Quality Index is not as acute for these data as we have seen in previous cases. A likely reason is that $I$ magnitudes were obtained at five epochs, while in all except one of our previous papers fewer epochs were available. At any rate there is no compelling trend in $U_{T}$ once the sample is restricted to objects with QI $\geq 3$. Using the sample of 27 Cepheids that have periods $\geq 25$ days but shorter than the baseline of 58 days, and that have QI $\geq 3$, and weighting the individual $U_{T}$ values by $(1/{\rm rms})^2$, gives the mean de-reddened modulus of $(m - M)_0 = 30.04 \pm 0.12$. If we make more restrictive cuts by accepting only objects with QI $\geq 4$, and then $\geq 5$, and then 6, (with the same period cuts as above), we obtain respectively weighted mean ``true'' moduli of $29.99 \pm 0.13$, $30.10 \pm 0.15$, and $30.07 \pm 0.18$. This shows the general stability of the result. For the unweighted average of the 27 Cepheids we obtain $(m - M)_0 = 30.10 \pm 0.14$. Due to the fact that the P-L relations in $V$ and $I$ have non-negligible width at a constant period, a Cepheid that is {\it intrinsically} redder and fainter will, on the average, carry {\it larger} measuring errors. This can contribute to systematically underestimating the distance when individual $U_{T}$'s are weighted as above. The unweighted solution is therefore preferred, since the uncertainties are similar. We note again that the $U_{T}$ values so derived depend on the assumption that the differences between $U_{V}$ and $U_{I}$ are due to reddening alone, in the absence of appreciable systematic and correlated measuring errors, or when the errors for $U_{V} - U_{I}$ {\it are distributed symmetrically}. If equation (5) is used for Cepheids {\it where correlated and/or asymmetrical errors in $V$ and $I$ dominate over differential reddening}, thereby producing a ratio of the $V$-to-$I$ errors that is different from 2.43, the $U_{T}$ derived via equation (5) {\it will be in error}. In particular, several Cepheids which were discovered in $V$ are too faint in $I$ to be measured, as already mentioned. This introduces a selection effect that biases against Cepheids with bluer colors. The effect is most pronounced at short periods where the {\it intrinsic} colors are bluest. This effect gives an asymmetrical distribution of errors in $U_{V} - U_{I}$ in the sense that it makes the de-reddened modulus {\it too small}. In Paper~V we devised a method to test for the presence of differential extinction or for the fact that the scatter about the P-L relation is due predominantly to measuring errors, or a combination of both. The method, shown in Fig.~11 of Paper~V for NGC~4536 and explained in the Appendix there, was used in Paper~VI for NGC~4496A (Fig.~9 there) and in Paper~VIII for NGC~4639 (Fig.~7 there). The method is to plot the difference in the apparent$V$ and $I$ moduli for any given Cepheid as ordinate against the apparent $V$ modulus as abscissa. If there is a systematic trend of the data along a line of slope $dU_{V}/d(U_{V} - U_{I}) = 2.43$, then equation (5) applies and there is clearly differential reddening. If, on the other hand, there is a general scatter with no trend, that scatter is dominated by measuring errors. While in the latter case true differential extinction can be hidden by measuring errors, trying to correct for putative reddening will result in interpreting any asymmetry in the error distribution as specious extinction. In two of the three previous cases, NGC~4536 (Paper~V), and NGC~4496A (Paper~VI) there is no trend along a differential reddening line. In the third case of NGC~4639 (Paper~VIII), there is a slight trend but also large scatter showing that the spread of points appears to be due to a mixture of measurement errors as well as from differential extinction. The diagnostic diagram just described is shown for the NGC~3627 data from Table~\ref{tbl4} in Fig.~\ref{fig7}. The filled circles show Cepheids with periods between 25 and 58 days. The solid line indicates the reddening vector for the P-L ridge line, if the true (de-reddened) distance modulus is 30.05 which is close to the mean derived earlier in this section. The dashed lines show the bounds due to the intrinsic dispersion of the P-L relation as explained in Paper~V. The slope of the lines is $A_{V}/E(V-I) = 2.43$ as before. There is only marginal evidence from Fig.~\ref{fig7} for a general trend along the solid line. The spread of points clearly spills outside these bounds, indicating that a very significant fraction of the scatter is due to measuring errors and related biasses. If such errors are distributed symmetrically, equation (5) will yield the correct answer, but if there are correlated errors in $V$ and $I$, or if there are selection effects that depend on color, using equation (5) will introduce errors. Note that the scatter of points is skewed along a direction orthogonal to the reddening vector: the lower right side appears to be more sparsely filled, indicating a possible selection effect. A more detailed inspection of the spread in Fig.~\ref{fig7} shows that the largest scatter occurs in the reddest Cepheids, using the $(\langle{V}\rangle - \langle{I}\rangle)$ colors computed from columns (3) and (5) of Table~\ref{tbl4}. This implies that the reddest Cepheids are so more because of skewed measurement errors than due to bona-fide reddening. This is not to deny the presence of differential reddening, but an acknowledgement that equation (5) alone is not adequate for obtaining a bias free result. We proceed by making an additional restriction of the data by color. Note that such a restriction used in conjunction with equation (5) does not introduce a procedural bias in the distance modulus. \subsubsection{The Distance Modulus By Restricting The Data By Color} A plot (not shown) of the color-period relation from the data in Table~\ref{tbl4} shows a distinctive separation into two major color groups, one close to the intrinsic $({\langle{V}\rangle}_0 - {\langle{I}\rangle}_0)$-period relation known for unreddened Cepheids in the Galaxy, LMC, and SMC as summarized in \cite{sanetal99} (1999) from data by \cite{dea78} (1978), \cite{cal85} (1985), and \cite{fern90} (1990). The other group of Cepheids with $(\langle{V}\rangle - \langle{I}\rangle)$ colors larger than 1.15 are far removed from the intrinsic domain in the color-period plot. They are also the Cepheids that show the largest deviation faintward in the P-L relations of Fig.~\ref{fig6}. Excluding these as the Cepheids with the largest reddening leaves a subsample of 29 Cepheids (the blue group) with $QI \geq 3$, and with $1.15 < \log P < 1.76$. We also have excluded C1-V13 because it is obviously an outlier. C1-V1, although blue, is excluded because its proposed period of 75 days is outside the baseline of 58 days. Figure~\ref{fig8} shows again that differential reddening is not the major factor in the scatter of the P-L relation, where $U_{V}$ is plotted vs. $U_{I}$ for the 29 blue Cepheids of the subsample. There is a clear correlation of $U_{V}$ and $U_{I}$, but the slope is not $A_{V}/A_{I} = 1.7$ as required if the Cepheids below the ridge lines of Fig.~\ref{fig6} were fainter because of a larger differential extinction. Rather, the slope near 1 in Fig.~\ref{fig8} can only be due to {\it correlated measuring errors} as we suspected in the last section. Of course, a part of the correlation must also be due to reddening, if for no other reason than the obvious dust pattern in Fig.~\ref{fig1} and \ref{fig2}. Figure~\ref{fig9} shows the P-L relations for the subset of the 29 bluest Cepheids. The scatter is markedly reduced from that in Fig.~\ref{fig6}, showing that the color cut has produced a subset with the smallest extinction and/or measuring error. We use the $U_{V}$ and $U_{I}$ apparent moduli in Table~\ref{tbl4} calculated from equations (3) and (4) of \S~4.2.1, and calculate mean values $\langle{U_{V}}\rangle$ and $\langle{U_{I}}\rangle$ using the 29 Cepheids of this subset. Assuming that the measuring errors for this sample are random and that they cancel in the mean permits the premise that the {\it difference} in the mean apparent moduli in the $V$ and $I$ passbands {\it is} now due to reddening. Multiplying the mean modulus difference by $A_{V}/E(V-I) = 2.43$ then gives $A_V$, which when subtracted from $\langle{U_{V}}\rangle$ gives the true modulus. This, of course, is what equation (5) does automatically, hence we need only analyze the $U_{T}$ values in Table~\ref{tbl4} for the 29 Cepheid subsample. The weighted mean $\langle{U_{T}}\rangle_{W}$ for the subset of 29 Cepheids gives $\langle{U_{T}}\rangle = 30.12 \pm 0.11$. The individual $U_{T}$ values are plotted vs. $\log P$ in Fig.~\ref{fig10}. As noted in the last section and seen in Fig.~\ref{fig10}, there is a tendency for the shortest period Cepheids to have the smallest individual moduli. Making the period cut at $\log P > 1.25$ removes the tendency and gives \begin{equation} {\langle{U_{T}}\rangle}_W = 30.17 \pm 0.12~, \end{equation} as the weighted mean from the 25 Cepheids with $1.25 < \log P < 1.78$, which we adopt. The unweighted mean is $\langle{U_{T}}\rangle = 30.24 \pm 0.09$. Justification for our restriction to the subsample of 25 Cepheids is given in Fig.~\ref{fig11} which is the diagnostic diagram of Fig.~\ref{fig7} but using only this subsample. Plotted are again the individual apparent moduli $U_{V}$ vs. the difference between the individual $V$ and $I$ apparent moduli. Figure~\ref{fig11} is much cleaner than Fig.~\ref{fig7}, and the data points now scatter nearly symmetrically about the differential reddening line. Applying, as in previous papers of this series, the correction for the ``long'' vs. ``short'' exposure effect of $0\fm05$ to equation (6), the de-reddened modulus of NGC~3627 is \begin{equation} (m - M)_0 = 30.22 \pm 0.12~, \end{equation} which we adopt. \section{The Absolute Magnitude at Maximum of SN~1989B Added to Previous Calibrators; the Present Status of the Calibration} The light curves in the {\it UBVRI} passbands of SN~1989B are well determined near maximum light, giving $B_{\rm max} = 12.34 \pm 0.05$, and $V_{\rm max} = 11.99 \pm 0.05$ (\cite{wel94} 1994). These authors have also determined the reddening of SN~1989B itself to be $E(B-V) = 0.37 \pm 0.03$. Hence, the de-reddened magnitudes at maximum light are $B_{\rm max}^{0} = 10.86 \pm 0.13$ and $V_{\rm max}^{0}= 10.88 \pm 0.10$. The absolute magnitudes at maximum for SN~1989B are \begin{equation} M^0_B({\rm max}) = -19.36 \pm 0.18~, \end{equation} and \begin{equation} M^0_V({\rm max}) = -19.34 \pm 0.16~, \end{equation} based on equation (7) and on the Cepheid zero points of equations (1) and (2). These values are combined in Table~\ref{tbl5} with our previous calibrations of the six SNe~Ia from Papers I--VII of this series. In addition, two new values are included from data for SN~1974G in NGC~4414 (\cite{tur98} 1998; \cite{schae98} 1998), and SN~1998bu in NGC~3368 (M~96) whose distance modulus is by \cite{tan95} (1995) with photometry of the SN reported by \cite{sun98} (1998). All Cepheid distances from the WFPC2 are corrected by $0\fm05$ for the short vs. long exposure photometric zeropoint difference (\cite{ste95} 1995; \cite{saha96a} 1996a). Neglecting the listed $M_B$ of SN~1895B, which is the most uncertain of the group, and because it is not absolutely certain whether it was spectroscopically normal at all phases, gives the straight mean values of $\langle{M_B}\rangle = -19.49 \pm 0.03$ and $\langle{M_V}\rangle = -19.49 \pm 0.03$. We adopt the {\it weighted} means, giving the mean calibration without any second parameter corrections for decay rate (see \S~7) as \begin{equation} \langle{M_B(\rm max)}\rangle = -19.49 \pm 0.07~, \end{equation} and \begin{equation} \langle{M_V(\rm max)}\rangle = -19.48 \pm 0.07 \end{equation} on the Cepheid distance scale of equations (1) and (2). Inclusion of SN~1895B would give $\langle{M_B}\rangle = -19.54 \pm 0.06$ for the straight mean and $\langle{M_B}\rangle = -19.52 \pm 0.07$ for the weighted mean. We have been criticized for using calibrators such as SN~1960F and SN~1974G, for which the photometry is more uncertain than others. We should point out that the net worth is not just the uncertainty in the photometry of the supernova, but the combined uncertainty with that of the Cepheid distance determination. Given the range of uncertainties in the Cepheid distance determinations (which depend on distance/faintness, crowding, etc), even photometric uncertainties considerably worse than for SN~1960F can be tolerated if the Cepheid determinations are as good as they are for its host galaxy NGC~4496A. In addition, we weight the contribution of individual calibrators by the inverse variance from the combined uncertainty of distance and supernova photometry, a procedure consistent with Bayesian inference. Thus even bona-fide `poor' cases like SN~1974G (which has also a relatively poor Cepheid distance to its host galaxy NGC~4414), enter only with appropriately lowered weight. Equations (10) and (11), based now on eight calibrators (neglecting SN~1895B), are similar to equations (12) and (13) of Paper~VIII (\cite{saha97} 1997) based there on six calibrators (again neglecting SN~1895B). Equation (10) here is $0\fm03$ fainter than in Paper~VIII. Equation (11) here is identical with that of Paper~VIII. Because of the similarity of the equations (10) and (11) here with equations (12) and (13) in Paper~VIII, the interim value of the Hubble constant, sans decay-rate corrections, for this stage of our $HST$ experiment is nearly identical with the values set out in Table~7 of Paper~VIII. The means of all values in that table, depending on how the SNe~Ia Hubble diagram is divided between {\it spirals} observed before and after 1985, and/or with redshifts larger or smaller than $\log v_{220} = 3.8$, and using equation (10) here rather than equation (12) of Paper~VIII, are then \begin{equation} \langle{H_0(B)}\rangle = 58 \pm 2~~({\rm internal})\ksm~, \end{equation} and \begin{equation} \langle{H_0(V)}\rangle =59 \pm 2~~({\rm internal})\ksm~. \end{equation} \section{The Hubble Diagram of SNe~Ia and the Value of $H_0$ if Second Parameter Corrections are Made} The Hubble diagram for blue SNe~Ia with $(B_{\rm max}-V_{\rm max}) < 0.20$ , based on many sources, historical as well as modern, was given in Paper~VIII (\cite{saha97} 1997, Fig.~10). When calibrated with the SNe absolute magnitudes, this diagram gives the Hubble constant directly. The details of the sample are being published in a paper by \cite{par99} (1999), and we make use of these data in this section to calibrate again correlations of absolute magnitudes of SNe~Ia with the often adopted second parameters of decay rate of the light curve, color of the SNe at maximum, and galaxy type. The precept adopted in the early papers of this series was that by restricting both our calibrator SNe~Ia and the Hubble diagram of SNe~Ia to ``Branch normal'' events (\cite{bra93} 1993), we have selected a homogeneous sample of SNe~Ia for which any systematic variation of absolute magnitude among the sample will be small enough to be neglected to first order at the 10\% level (\cite{cad85} 1985; \cite{lei91b} 1991b; \cite{sata93} 1993; \cite{tasa95} 1995). In the meantime, thanks to the observational program of the Calan/Tololo Chilean consortium (\cite{ham95} 1995; 1996a, b) and the theoretical insights of variations in the pre-explosion conditions of the progenitors (cf. \cite{vHipp97} 1997, \cite{hof98} 1998, \cite{nad98} 1998 for summaries), it has become clear that there is in fact a {\it continuous} variation of SNe~Ia properties, including absolute magnitude, that can be detected from observed second- parameter properties and which can now be accounted for even at this $< 10\%$ level. The most apparent of these second parameters is the change in the shape of the light curve, quantified by the rate of decay from maximum light (\cite{rus74} 1974; \cite{dev76} 1976; \cite{psk77} 1977, 1984; \cite{phi87} 1987; \cite{bar90} 1990; \cite{bra92} 1992; Hamuy et al.\ 1996a, b). \cite{phi93} (1993) derived a very steep relation between decay rate and absolute magnitude using distances determined by a variety of methods and including also very red SNe~Ia. A much flatter dependence was found by \cite{tasa95} (1995) for blue SNe~Ia using more reliable relative distances from recession velocities. The flatter slope was confirmed for only the blue SNe~Ia by the subsequent extensive Calan/Tololo data (Hamuy et al.\ 1996a, b; \cite{saha97} 1997). A detailed discussion of the second order corrections to the SNe~Ia distance scale depending on second and third parameter correlations is the subject of the accompanying paper by \cite{par99} (1999) where it is shown that besides the decay rate, color at maximum is also a principal second parameter, confirming a similar result by \cite{tri98} (1998). The Parodi et al.\ paper lists relative kinematic absolute magnitudes as if the local Hubble redshift-to-distance ratio is the same for redshifts smaller than $10,000\kms$ as for the remote $H_0$ for $v > 10,000\kms$, neglecting the evidence set out in Paper~VIII (\cite{saha97} 1997, \S~8; cf. also \cite{zeh98} 1998) that the local value may be 5\% to 10\% larger than the global value. This possible change of the Hubble ratio outward, decreasing with distance by $\leq 10\%$ for $v$ out to $ 10,000\kms$, is also consistent with other external data on first ranked cluster galaxies (\cite{lau94} 1994; \cite{tam98a} 1998a). The suggestion is also consistent with the derived shallow slope of $d\log v/dm = 0.192$ of the local ($v < 10,000\kms$) Hubble diagram by Parodi et al.\ (1999, their eqs. 7 and 8) rather than 0.200 required if $H_0$ did not vary with distance. In the analysis of this section we consider the consequences of a variable $H_0$ decreasing outward, that is required to give a slope of 0.192 to the Hubble diagram locally, and then carry again the analysis of the second parameter effects of decay rate, color, and galaxy type using the derived kinematic absolute magnitudes. \subsection{Kinematic Absolute Magnitudes Using a Variable Hubble Constant with Local Distance for $v < 10,000\kms$ } We accept the premise that the slope of the Hubble diagram for redshifts smaller than $10,000\kms$ is 0.192 (\cite{par99} 1999). If we adopt an arbitrary global (remote field) value of $H_0 = 55$ (to be adjusted later by our calibrators in Table~\ref{tbl5}), then, by an obvious calculation, the variation of $H_0$ in the distance interval of $1000 < v < 10,000$ is well approximated (within $1\%$) by \begin{equation} H_0(v) = -5.39 \log v + 76.50~. \end{equation} This gives $H_0 = 60.3$ at $v = 1000\kms$ and $H_0 = 54.9$ at $v = 10,000\kms$. Equation (14) has been used to recalculate the absolute magnitudes of all SNe~Ia in the fiducial sample in Table~1 of \cite{par99} (1999) for which $\log v < 4.00$. $H_o = 55$ has been assumed for $\log v > 4.00$. The results are set out in Table~\ref{tbl6} which is divided into three parts according to Hubble type to better understand the type dependence of the Hubble diagram seen in Fig.~10 of Paper~VIII (\cite{saha97} 1997). The first section of Table~\ref{tbl6} lists the SNe in late type spirals ($T$ of 3 and greater, meaning Sb to Im types). The second section for $T = 1$ and 2 are for Sa and Sab parent galaxies. The third section lists parent galaxy $T$ types of 0 and smaller (E and S0). Columns (1) through (6) repeat data from Table~1 of \cite{par99} (1999) with magnitudes at the respective maxima denoted by $B_0$, $V_0$, and $I_0$. The redshifts in column (3) are corrected for peculiar motions. For $v < 3000\kms$ the redshifts, reduced to the frame of the centroid of the Local Group (\cite{yah77} 1977), were then corrected again to the frame of the Virgo cluster using the self-consistent Virgocentric infall model with the local infall vector (actually retarded expansion of the Local Group relative to Virgo) of $220\kms$ (\cite{kra86ab} 1986a, b). For $v > 3000\kms$ an additional correction of $630\kms$ relative to the CMB frame due to the CMB dipole anisotropy (\cite{bou81} 1981; \cite{wil84} 1984) was applied according to the model of \cite{tasa85} (1985, their Fig.~2). For more distant galaxies we have adopted the corrected velocities from \cite{ham96a} (1996a). The sources for the photometry in columns (4) to (6) are listed in \cite{par99} (1999). The magnitudes are corrected for Galactic absorption (\cite{bur84} 1984). Column (7) is the adopted local value of $H_0$ calculated from equation (14) for $\log v < 4.00$, and using $H_0 = 55$ for larger redshifts. The resulting distance moduli are in column (8). The corresponding absolute magnitudes are in columns (9) to (11). The decay rates $\Delta m_{15}(B)$ in column (12) are from the sources listed by \cite{par99} (1999). The data for the eight calibrators in Table~\ref{tbl5}, except for the observed apparent magnitudes, are not shown in Table~\ref{tbl6} because their absolute distance moduli are on the Cepheid system, not based on redshifts. The means of columns (9) to (11) for the absolute magnitudes in $B$, $V$, and $I$ are shown at the foot of each of the three sections of Table~\ref{tbl6}. The systematic progression, becoming fainter for the earlier galaxy types, is evident and is significant at the 2-sigma level. It is this difference that causes the separation of the ridge lines in the Hubble diagram of spirals and E and S0 galaxies seen in Fig.~10 of \cite{saha97} (1997). However, this is not a type dependence per se. Note in column 12 that the mean decay rates, shown at the foot of each section, differ significantly between the three sections of Table~\ref{tbl6}. The decay rates are much longer for the E to S0 types, averaging $\langle{\Delta m_{15}(B)}\rangle = 1.44 \pm 0.04$ for these early Hubble types, compared with $\langle{\Delta m_{15}(B)}\rangle = 1.03 \pm 0.03$ for the spirals in the first section of Table~\ref{tbl6}. Because the decay rate itself is correlated with absolute magnitude (next section), the {\it apparent} dependence of $\langle{M({\rm max})}\rangle$ on Hubble type is in fact due to the decay rate correlation. This, of course, is a clue as to differences in the progenitor mass of the pre-SNe~Ia as a function of Hubble type, and goes to the heart of the physics of the phenomenon (eg. \cite{nad98} 1998). In any case, the apparent correlation of $\langle{M}\rangle$ with Hubble type disappears when the decay rate corrections of the next section are applied. Said differently, the {\it apparent} dependence of mean absolute magnitude with Hubble type is due to the difference in mean decay rate between early type galaxies (E to S0 types) and late type spirals (the first section of Table~\ref{tbl6}), together with the dependence of decay rate on absolute magnitude at maximum (Fig.~\ref{fig12} in the next section). \subsection{The Decay Rate Dependence} The absolute magnitudes in columns (9) to (11) of Table~\ref{tbl6} are plotted in Fig.~\ref{fig12}. It is clear that there is a relation of decay rate with (kinematic) absolute magnitude (\cite{phi93} 1993), but that the relation is strongly non-linear. There is a clear plateau between abscissa values of decay rates that are greater and smaller than 1.0 to 1.5. In this plateau region, the absolute magnitude is nearly independent of the decay rate. Figure~\ref{fig12} shows the correlations of columns (9), (10), and (11) (of Table~6) with the decay rate in column (12). The correlation with absolute magnitude for small and for large $\Delta m_{15}(B)$ is obvious. Nevertheless, the correlation is clearly non-linear. There is a central region ($1.0 < \Delta m_{15}(B) < 1.5$) where the correlation is weak. This is the core of the ``Branch normal'' majority of SNe~Ia, comprising 95\% of the observed SNe~Ia (only 1 in 20 of the local SNe~Ia discoveries are ``Branch abnormal''; see \cite{bra93} 1993). It is only when the few known abnormal SNe~Ia are added to the ``Branch normal'', i.e. blue SNe~Ia, that they show a wide variation in luminosity. Consider now the apparent correlation of $\langle{M_{\rm max}}\rangle$ with Hubble type, shown in Fig.~10 of \cite{saha97} (1997, Paper~VIII). It is of central physical interest how the Hubble type of the parent galaxy can produce slow-decay rate SNe~Ia in E galaxies, and faster-decay rate SNe~Ia in later-type galaxies. One supposes that the mass of the pre-SNe~Ia stars must be a function of chemical evolution in any given parent galaxy, causing the change in the character of the SNe~Ia explosion with Hubble type (\cite{nad98} 1998). Nevertheless, this is an astrophysical problem, not an astronomical problem of how to use the data to determine reliable distances. It suffices here to note that the type dependence on $\langle{M_{\rm max}}\rangle$ disappears when the correction, now described for decay rate (Fig.~\ref{fig12}), is applied. The correlation of decay rate $\Delta m_{15}(B)$ with absolute magnitudes in $B$, $V$, and $I$ in Fig.~\ref{fig12} is striking, confirming Pskovskii's (1977, 1984) initial suggestions. But again, the correlations are much less steep than suggested by \cite{phi93} (1993) and by Hamuy et al.\ (1996b, their Fig.~2), especially in the region of the ``Branch normal'' SNe~Ia with $0.95 < \Delta m_{15}(B) < 1.3$ where there is no strong correlation. \cite{ham96a} (1994a) distinguish between the slope of the decay-rate correlation with $M({\rm max})$ obtained from using TF, PN and SBF distances to nearby SNe~Ia (as originally done by \cite{phi93} 1993), versus that from using relative distances from redshifts at intermediate distances. The latter sample gives a shallower slope than that of Phillips. By restricting to a Branch normal sample, the slope is further reduced. In Fig.~\ref{fig12} we have fitted cubic polynomials to the data contained in Table~\ref{tbl6} for the correlations of absolute magnitudes in $B$, $V$, and $I$ in columns (9) to (11) with $\Delta m(B)_{15}$ in column (12). Denote the correction to $M_i$ for decay rate by $y_i$, such that the ``corrected'' absolute magnitude, reduced to $\Delta m_{15}(B) = 1.1$, is defined by \begin{displaymath} M_i^{15} = M_i(~{\rm{Table~\ref{tbl6}}}) - y_i~. \end{displaymath} The correlations in Fig.~\ref{fig12}, based on the decay-rate data and absolute magnitude data in Table~\ref{tbl6}, give the following cubic corrections for decay rate vs. absolute magnitude in $B$, $V$, and $I$. The lines drawn in Fig.~\ref{fig12} are calculated from these equations. We reduce all data to $\Delta m_{15}(B) = 1.1$, and therefore define $x = \Delta m_{15}(B) - 1.1$. \begin{eqnarray} y_B = 0.693x - 1.440x^2 + 3.045x^3\,, \\ y_V = 0.596x - 2.457x^2 + 4.493x^3\,, \\ y_I = 0.360x - 2.246x^2 + 4.764x^3\,. \end{eqnarray} Note that these polynomials are not well constrained outside the range of $\Delta m(B)_{15}$ spanned by the SNe~Ia in this sample. In particular, the upturn for very slow declining SNe~Ia is uncertain. However, the cubic characterization given here is adequate for the present arguments and calculations: it is not used in a region where it is ill constrained. The corrections for decay rate, $y_i$, from equations (15) to (17), have been applied to columns (9) to (11) of Table~\ref{tbl6} to obtain absolute magnitudes freed from the $\Delta m(B)_{15}$ parameter effect. The arithmetic is not shown but the mean corrected magnitudes $\langle{M_i^{15}}\rangle$, still binned into the three morphological groups of Table~\ref{tbl6} are listed in the first four lines of Table~\ref{tbl7}. The corresponding mean magnitudes $\langle{M_B^{15}}\rangle$ and $\langle{M_V^{15}}\rangle$ of the eight calibrators from Table~\ref{tbl5} are shown in the last line. No mean value$\langle{M_I^{15}}\rangle$row is shown in the last line because too few calibrators have known $I_{\rm max}$. Table~\ref{tbl7} permits two conclusions: (1) The type dependence shown in Fig.~10 of Paper~VIII between E galaxies and spirals has now disappeared to within one sigma differences after the $\Delta m_{15}$ corrections are applied. (2) Comparing the $\langle{M_B^{15}}\rangle$ and $\langle{M_V^{15}}\rangle$ values in the fourth line (the total sample) with the corrected values of the calibrators in the fifth line shows the statistical difference of $0\fm14 \pm 0\fm058$ in $B$ and $0\fm15 \pm 0\fm076$ in $V$ between the fiducial sample and the Table~\ref{tbl5} calibrators. Because the absolute magnitudes of the fiducial sample are based on $H_0 = 55$, these data, corrected for decay rate, give \begin{equation} H_0(B)(~{\rm decay~rate}) = 58.8 \pm 2\ksm\,, \end{equation} and \begin{equation} H_0(V)(~{\rm decay~rate}) = 59.1 \pm 2\ksm\,. \end{equation} It is shown in \cite{par99} (1999), that the effect of the decline rate correction taken alone, is to increase $H_0$ by 7\%. This correction depends on the extent to which the decline rate distribution is different for the calibrating and distant samples of SNe~Ia. \subsection{The Color Dependence} Figure~\ref{fig13} shows that there is still a dependence of the corrected $M_i^{15}$ magnitudes on color that is not removed by the $\Delta m_{15}$ corrections, contrary to the removal of the dependence on Hubble type. The symbols are the same as in Fig.~\ref{fig12}. The ordinates are the kinematic ($H_0 = 55$) absolute magnitudes, $M_i^{15}$, based on variable local values of $H_0$ and the decay rate corrections from the last subsection. The abscissa is the observed $(B_0 - V_0)$ color that can be derived from columns (4) and (5) of Table~\ref{tbl6}. The equations for the least squares solutions for the lines are \begin{equation} M_B^{15} = 1.712\cdot (B_0 - V_0) - 19.648\,, \\ \end{equation} \begin{equation} M_V^{15} = 0.587\cdot (B_0 - V_0) - 19.562\,, \end{equation} and \begin{equation} M_I^{15} = -0.216\cdot (B_0 - V_0) - 19.312\,. \end{equation} Figure~\ref{fig14} is the same as Fig.~\ref{fig13} but with $(V_0- I_0)$ colors derived from columns (10) and (11) of Table~\ref{tbl6}. The least squares solutions for the lines are \begin{equation} M_B^{15} = 0.257\cdot (V_0 - I_0) - 19.540\,, \\ \end{equation} \begin{equation} M_V^{15} = -0.260\cdot (V_0 - I_0) - 19.621\,, \end{equation} and \begin{equation} M_I^{15} = -1.115\cdot (V_0 - I_0) - 19.607\,. \end{equation} The immediate consequence of equations (20) to (25) is that these color dependencies {\it are not due to reddening and absorption} because the slope coefficients in $B$ and $V$, nor their ratios, conform to the canonical normal interstellar values of $A_B/E(B-V) = 4$, $A_V/E(B-V) = 3$, and $A_I/E(B-V) = 1.76$, and $A_B/E(V-I)= 3.2$, $A_V/E(V-I) = 2.4$, and $A_I/E(V-I) = 1.4$ (\cite{schef82} 1982). If, for example, the variation of absolute magnitude with color index from SN to SN were due entirely to absorption and reddening, then the slope coefficients in the correlations of Fig.~\ref{fig13}, (Eqs.\,20 to 22) would have to be $dM_B/E(B-V) = 4$, $dM_V/E(B-V) = 3$, and $dM_I/E(B-V) = 1.76$, based on $A_V/A_I = 1.7$. Not only are the coefficients for $B$ and $V$ in equations (20) and (21) very different from these requirements for reddening, but even the sign of the variation in $I$ with $(B-V)$ is negative. This is impossible if the cause is internal reddening with its corresponding dimming. The situation is even more decisive from equations (23) to (25) using $(V-I)$ colors in Fig.~\ref{fig14}. From E(V-I)/E(B-V) = 1.25 (Eq.\,1 of \cite{dea78} 1978), the predictions are $dM_B/E(V-I) = 3.2$, $dM_V/E(V-I) = 2.4$, and $dM_I/E(V-I) = 1.41$. These are not only far different from the coefficients in equations (23) to (25), but again the sense of the observed correlation in equations (24) and (25) belies a reddening/extinction explanation. {\it Brighter} magnitudes for redder colors are not possible for any known reddening and absorption law. The conclusion is that the correlations in Figs.~\ref{fig13} and \ref{fig14} are due to intrinsic properties internal to the physics of the SNe themselves rather than to reddening and absorption. Hence, any corrections to the absolute magnitudes based solely on assumed absorption-to-reddening ratios for normal extinction are most likely to be incorrect. Nevertheless, the correlations in Figs.~\ref{fig13} and \ref{fig14} are definite, and can be used to further reduce the absolute magnitudes to some fiducial value of observed color. We thus reduce to a fiducial color of $(B_0-V_0) = 0.00$ as a second parameter, in addition to the decay-rate parameter, in agreement with \cite{tri98} (1998). Our approach is different in principle from that of \cite{phi99} (1999), where they assert that the color range at given decline rate is due to reddening, and proceed accordingly. Our conclusion from the above analysis is that for our sample of SNe~Ia, with $(B_{\rm max} - V_{\rm max}) < 0.20$, the color spread at given decline rate is intrinsic in nature, since correlation of the peak brightness with the residual color for this sample is different from what is expected from reddening. Applying the color-term corrections of equations (20) and (21) to the $M_B$ and $M_V$ magnitudes of Table~\ref{tbl6}, and taking the mean values over the complete sample of 34 SNe~Ia in Table~\ref{tbl6} where the decay rates are known, gives \begin{equation} \langle{M_B^{\rm corr}}\rangle = \langle{M_B^{15} + ~{\rm color~term}}\rangle = -19.64 \pm 0.026\,, \end{equation} and \begin{equation} \langle{M_V^{\rm corr}}\rangle = \langle{M_V^{15} + ~{\rm color~term}}\rangle = -19.61 \pm 0.026\,. \end{equation} These magnitudes are calculated with variable $H_0$ following the precepts of \S 6.1 and reduced to $\Delta m_{15} = 1.1$ and $(B_0 - V_0)= 0.00$. For the calibrators in Table~\ref{tbl5} one finds in analogy \begin{equation} \langle{M_B^{\rm corr}}\rangle = -19.45 \pm 0.060\,, \end{equation} and \begin{equation} \langle{M_V^{\rm corr}}\rangle = -19.44 \pm 0.079\,. \end{equation} Comparison of equations (28) and (29) with equations (26) and (27) shows that the $H_0 = 55$ assumption must be changed to accommodate the differences of $0\fm19 \pm 0\fm08$ in $B$ and $0\fm17 \pm 0\fm09$ in $V$. They require the global value of $H_0$, now fully corrected for the three effects of (1) the change of $H_0$ outward, (2) the light curve decay rate, and (3) the color variation with $M({\rm max})$, to be \begin{equation} H_{0}(B) = 60.2 \pm 2\ksm\,, \end{equation} and \begin{equation} H_{0}(V) = 59.6 \pm 2\ksm\,. \end{equation} These values are only 2\% higher than equations (18) and (19) which use only the decay-rate correction, and 1\% to 4\% higher than equations (12) and (13) which are weighted towards SNe~Ia in spirals. Other authors have adopted $\langle{M({\rm max})}\rangle$ values that differ from those in equations (10) and (11) for the calibrator absolute magnitudes. For example, \cite{ken98} (1998), and \cite{fre99} (1999) have discarded several of our calibrators, and added two that are not based on direct Cepheid distances to the host galaxy. They have consequently derived a fainter mean absolute $B$ magnitude than in equation (10). Specifically, they assume that the distances of the early-type galaxies NGC~1316 and NGC~1380 in the Fornax cluster, parent galaxies to SN~1980N and SN~1992A, are identical with that of the spiral NGC~1365 for which there is a Cepheid distance. \cite{sun98} (1998) have also considered the questionable SN~1980 and SN~1992A as possible calibrators. However, there are reasons to suspect that NGC~1365 is in the foreground of the Fornax cluster (\cite{satasa99} 1999), and therefore that the precept of the fainter calibration used by \cite{ken98} (1998), calibrating the two Fornax SNe~Ia via NGC~1365, is not correct. The evidence is that \cite{wel94} (1994) have demonstrated that the multi-color light curves of SN~1989B in NGC~3627 and SN~1980N in NGC~1316 are virtually identical, and in fact establish the reddening and extinction to SN~1989B by comparing the magnitude shifts in different passbands relative to SN~1980N. Asserting then that SN~1980N has the same peak brightness as SN~1989B yields the distance modulus difference of $1\fm62 \pm 0\fm03$ (\cite{wel94} 1994). There is additional uncertainty of $\pm 0.17$ to allow for scatter in the difference in peak brightness of two SNe~Ia with the same decline rate. With our modulus of $(m - M)_0 = 30.22 \pm 0.12 $ (Eq.~7) for SN~1989B, the derived modulus of NGC~1316, host to SN~1980N, is $31.84 \pm 0.21$. This is $0\fm5$ more distant than the Cepheid distance of NGC~1365 (\cite{mad98} 1998), but is close to the value found for the early-type galaxies of the Fornax cluster by independent methods (\cite{tam98b} 1998b). In any case the implication of \cite{ken98} (1998) and \cite{fre99} (1999) that the two SNe~Ia 1989B and 1980N differ by $0\fm5$ in luminosity, which is based on the unproved assertion that NGC~1365 and NGC~1316 are at the same distance, is not credible. The further consequence of the precept of equating the Cepheid distance of NGC~1365 to that of the unknown distance to NGC~1316 and NGC~1380 is that the decay-rate absolute magnitude relation which \cite{fre99} (1999) deduce using their faint absolute magnitudes of the two Fornax SNe~Ia is steeper than that in Fig.~\ref{fig12} by an amount that compromises their conclusions concerning $H_0$, explaining their abnormally high value of $H_0 \approx 73$. \section{Summary and Conclusions} (1) Comparison of 12 epoch $HST$ frames in the $F555W$ band and five epoch frames in the $F814W$ band has isolated 68 definite Cepheid variables with periods between 3 and $\sim75$ days in the dust-rich galaxy NGC~3627 in the Leo Group. (2) The adopted true modulus of $(m - M)_0 = 30.22 \pm 0.12$ that results from analysis of various subsamples of the data according to the degree of internal absorption, agrees well with the distance moduli of $(m - M)_0 = 30.37 \pm 0.16$ for NGC~3368 (parent of SN~1998bu) (\cite{tan95} 1995) and $(m - M)_0 = 30.01 \pm 0.19$ for NGC~3351 (\cite{gra97} 1997), showing that the extended Leo Group in fact exists (\cite{hum56} 1956, Table XI). (3) Combining the NGC~3627 Cepheid modulus of $(m -M)_0 = 30.22$ with the photometry of \cite{wel94} (1994) for the daughter SN~Ia SN~1989B (corrected for extinction) gives the absolute magnitudes in $B$ and $V$ at maximum light of $M_B = -19.36 \pm 0.18$, and $M_V = -19.34 \pm 0.16$ for SN~1989B as listed in Table~\ref{tbl5}. (4) Combining these absolute magnitudes with those for the previous six calibrators determined in previous papers of this series, and with two additional calibrators recently available (SN~1974G in NGC~4414 and SN~1998bu in NGC~3369), gives a mean calibration (without second parameter corrections) of $\langle{M_B}\rangle = - 19.49 \pm 0.07$ and $\langle{M_V}\rangle = - 19.48 \pm 0.07$. (5) The absolute magnitudes $M_{\rm max}$ of blue SNe~Ia with $(B_0 - V_0) <0.20$ correlate with the light curve decay rate $\Delta m_{15}$. The correlation is approximated by a cubic equation which has a flat plateau at intermediate values of $\Delta m_{15}$. In addition, $M_{\rm max}$ depends on the color $(B_0 - V_0)$. For the derivation of the dependence on these second parameters, a slight decrease of $H_0$ out to $10,000 \kms$ has been taken into account. Once corrections for these two second parameters are applied, no residual dependence of $M_{\rm max}$ on Hubble type is seen. (6) The calibrators have somewhat smaller mean values of $\Delta m_{15}$ and bluer mean colors $(B_0 - V_0)$ than the distant SNe~Ia defining the Hubble diagram. The application of second parameter corrections therefore tends to increase $H_0$, but the effect is less than 10\%. (7) The overall value of the Hubble constant determined here from the current eight calibrators as applied to the distant SNe~Ia with $1000 \kms < v < 30,000 \kms$ is $H_0 = 60 \pm 2$ in both $B$ and $V$ (Eqs. 30 and 31). (8) The Hubble constant derived here rests on Cepheid distances whose zeropoint is set to an LMC modulus of 18.50. If the latter is revised upwards by $\approx 0\fm06$ (\cite{fed98} 1998), the consequent value of the Hubble constant is \begin{equation} H_0 = 58 \pm 2\ksm\,. \end{equation} \acknowledgments{Acknowledgment} We thank the many individuals at STScI who worked hard behind the scenes to make these observations possible, and wish to particularly mention Doug van Orsow, George Chapman, Bill Workman, Merle Reinhardt and Wayne Kinzel. A.S. and A.S. acknowledge support from NASA through grant GO-5427.02-93A from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy. L.L. and G.A.T. thank the Swiss National Science Foundation for continued support. \clearpage
\section{Introduction} Two photon reactions are an important part of physics which is being studied in current $e^+e^-$ experiments at LEP1 and LEP2 and which will also be intensively analyzed in future $e^+e^-$ colliders. The available photon-photon energy and photon virtualities continously increase with the increasing energy of the $e^+e^-$ pair. Therefore the data from LEP1 and LEP2 and the expected results from the TESLA and NLC provide us with an excellent oportunity to study virtual photon scattering in the diffractive regime. Moreover, with proper experimental cuts, it is possible to study observables dominated by the perturbative QCD contributions. The theoretical description of such processes is based on expectations concerning high energy limit in perturbative QCD which is at present theoretically fairly well understood \cite{GLR,LIPAT1}. The leading high energy behaviour is controlled by the pomeron singularity which corresponds to the sum of ladder diagrams with reggeized gluons along the chain. This sum is described by the Balitzkij, Fadin, Kuraev, Lipatov (BFKL) equation \cite{BFKL1}. The perturbative QCD pomeron exchange effects can be observed only in specific conditions and even then not in the unambigous form. In order to minimize the contribution of the other mechanisms competing with the QCD pomeron and to guarantee the validity of the calculations based on perturbative QCD one has to chose carefully the processes to analyze. The virtualities of the gluons along the ladder should be large enough to assure the applicability of the perturbative expansion. The neccesary hard scale may be provided either by coupling of the ladder to scattering particles, that contain a hard scale themselves, or by large momentum transfer carried by the gluons. Moreover, to distinguish the genuine BFKL from DGLAP evolution effects it is convenient to focus on procesess in which the scales on both ends of the ladder are of comparable size. Finally, one requires that the non-perturbative effects should factor out in order to minimize the theoretical uncertainties. The two classical processes which can probe the QCD pomeron in $ep$ and in $\gamma^* p$ collisions are the deep inelastic events accompanied by an energetic (forward) jet {\cite{MUELLERJ, DISJET} and the production of large $p_T$ jets separated by the rapidity gap \cite{JETGAP}. The former process probes the QCD pomeron in the forward direction while the latter reflects the elastic scattering of partons via the QCD pomeron exchange with non-zero (and large) momentum transfer. Another possible probe of the QCD pomeron at (large) momentum transfers can be provided by the diffractive vector meson photoproduction accompanied by proton dissociation in order to avoid nucleon form-factor effects \cite{FORSHAW,BARTLQ}. In this talk we shall analyze two measurements in $e^+ e^-$ collisions, complementary to those listed above. Namely we focus on double diffractive $J/\psi$ production in $\gamma\gamma$ collisions and on the total $\gamma^* \gamma^*$ cross section. The former process is unique since in principle it allows to test the QCD pomeron for arbitrary momentum transfers \cite{KMPSI}. The hard scale is given by the relatively large mass of the $c$-quark. The total $\gamma^*\gamma^*$ cross-section has been studied by several authors \cite{GGSTAR,BRODSKY}, however our approach has the novel feature of taking into account dominant non-leading corrections to the BFKL equation. This re-analysis has become necessary when the next-to-leading corrections to the BFKL kernel were obtained \cite{BFKLNL}, which alter substantially the results obtained at the leading order. It turns out that the magnitude of the next-to-leading (NLO), i.e. $O(\alpha_s^2)$, contribution to the QCD pomeron intercept is very large for the values of the QCD coupling within the range which is relevant for most experiments. This means that the NLO approximation alone is not reliable and one has to perform resummation to all orders. Unfortunately the exact result of this resummation is unknown. It may however be possible to pin down certain dominant contributions of well defined physical origin and perform their exact resummation \cite{RESUM,KMSG}. In our approach we shall use the so called consistency constraint which limits the available phase space for the real gluon emission by imposing the requirement that the virtuality of the exchanged gluons along the chain is dominated by their transverse momentum squared. Let us remind that the form of the LO BFKL kernel where the gluon propagators contain only the gluon transverse momentum squared etc. is only valid within the region of phase space restricted by this constraint. Formally however, the consistency constraint generates subleading corrections. It can be shown that at the NLO accuracy it generates about 70 \% of the exact result for the QCD pomeron intercept. The very important merit of this constraint is also the fact that it automatically generates resummation of higher order contributions which stabilizes the solution \cite{KMSG}. \section{The total ${\gamma^* \gamma^*}$ cross-section} The collisions of virtual photons may be studied experimentally only as subprocesses of reactions between charged particles. In principle, one is able to unfold the photonic cross-section from the leptonic data, however this procedure requires additional assumptions which increase the systematic uncertainty of the result. It seems to be more sensible to formulate the predictions for the $e^+e^-$ cross-sections with the properly chosen cuts and compare them directly with the $e^+ e^-$ data. Therefore we use the equivalent photon approximation which allows us to express the leptonic cross-section through a convolution of the photonic cross-section and the standard flux factors. Thus the cross-section for the process $e^+e^- \rightarrow e^+e^- + X$ (averaged over the angle $\phi$ between the lepton scattering planes in the frame in which the virtual photons are aligned along the $z$ axis) is given by the following formula \cite{BRODSKY}: $$ {Q_1^2 Q_2^2 d\sigma \over dy_1 dy_2 dQ_1^2 dQ_2^2} = \left({\alpha\over 2 \pi}\right)^2 [P^{(T)}_{\gamma/e^+}(y_1)P^{(T)}_{\gamma/e^-}(y_2) \sigma^{TT}_{\gamma^* \gamma^*}(Q_1^2, Q_2^2,W^2)+ $$ $$ P^{(T)}_{\gamma/e^+}(y_1)P^{(L)}_{\gamma/e^-}(y_2) \sigma^{TL}_{\gamma^* \gamma^*}(Q_1^2,Q_2^2,W^2)+ P^{(L)}_{\gamma/e^+}(y_1)P^{(T)}_{\gamma/e^-}(y_2) \sigma^{LT}_{\gamma^* \gamma^*}(Q_1^2,Q_2^2,W^2)+ $$ \begin{equation} P^{(L)}_{\gamma/e^+}(y_1)P^{(L)}_{\gamma/e^-}(y_2) \sigma^{LL}_{\gamma^* \gamma^*}(Q_1^2, Q_2^2,W^2)] \label{conv} \end{equation} where \begin{equation} P^{(T)}_{\gamma/e}(y) = {1 + (1-y)^2\over y} \label{pt} \end{equation} \begin{equation} P^{(L)}_{\gamma/e}(y) = 2{1-y\over y} \label{pl} \end{equation} where $y_1$ and $y_2$ are the longitudinal momentum fractions of the parent leptons carried by virtual photons, $Q_i^2 = -q_i^2$ ($i=1,2$) where $q_{1,2}$ denote the four momenta of the virtual photons and $W^2$ is the total CM energy squared of the two (virtual) photon system, i.e. $W^2=(q_1+q_2)^2$. The cross-sections $\sigma^{ij}_{\gamma^* \gamma^*}(Q_1^2,Q_2^2,W^2)$ are the total cross-sections for the process $\gamma^* \gamma^* \rightarrow X$ and the indices $i,j=T,L$ denote the polarization of the virtual photons. The functions $P^{(T)}_{\gamma/e}(y)$ and $P^{(L)}_{\gamma/e}(y)$ are the transverse and longitudinal photon flux factors. \noindent \begin{figure} \leavevmode \begin{center} \parbox{6.5cm}{ {\large a)}\\ \epsfxsize = 5.5cm \epsfysize = 4.2cm \epsfbox[60 545 311 741]{ggfig1.eps}}\qquad \parbox{6.5cm}{ {\large b)}\\ \epsfxsize = 5.5cm \epsfysize = 4.2cm \epsfbox{ppfig1.eps}} \\ \end{center} \caption{\small The QCD pomeron exchange mechanism of the processes a) $\gamma_1 ^* (Q_1 ^2) \gamma_2 ^* (Q_2 ^2) \to X$ and b) $\gamma\gamma \rightarrow J/\psi J/\psi$.} \end{figure} The ladder diagram corresponding to the perturbative contribution to the diffractive subprocess $\gamma_1 ^* (Q_1 ^2) \gamma^* (Q_2 ^2) \to X$ is shown in Fig.~1a. The cross-sections $\sigma^{ij}_{\gamma^* \gamma^*}(Q_1^2,Q_2^2,W^2)$ are given by the following formulae: $$ \sigma^{ij}_{\gamma^* \gamma^*}(Q_1^2, Q_2^2,W^2) = P_S(Q_1^2,Q_2^2,W^2)\delta_{iT}\delta_{jT} + $$ \begin{equation} {1\over 2 \pi}\sum_q\int_{k_0^2}^{k_{max}^2(Q_2^2,x)} {d^2k\over \pi k^4} \int _{\xi_{min}(k^2,Q_2^2)}^{1/x} d\xi G^{0j}_q(k^2,Q_2^2,\xi) \Phi_i(k^2,Q_1^2,x\xi) \label{csx} \end{equation} where \begin{equation} k_{max}^2(Q_2^2,x)=-4m_q^2+Q_2^2 \left( {1\over x} - 1 \right) \label{kmax} \end{equation} \begin{equation} \xi_{min}(k^2,Q^2)=1+{k^2+4m_q^2\over Q^2} \label{ximin} \end{equation} and \begin{equation} x={Q_2^2\over 2q_1q_2} \label{x} \end{equation} In Eq.~(\ref{csx}) we sum over four quark flavours with $m_q \to 0$ for light quarks and $m_c=1.5\rm \; GeV$. The lower limit of integration over $k^2$ appearing in Eq.~(\ref{csx}) is taken to be $k_0 ^2 = 1\rm \; GeV ^2$ in order to subtract the contribution from the nonperturbative region from the perturbative part of the amplitude. The functions $G^{0i}_q(k^2,Q^2,\xi)$ are defined as below: \cite{BRODSKY,KMSTAS} $$ G^{0T}_q(k^2,Q^2,\xi)= $$ $$ 2 \alpha_{em} \alpha_s(k^2+m_q^2)e_q^2\int_0^{\lambda_{max}} d \lambda \int {d^2p^{\prime }\over \pi}\; \delta\left[\xi-\left(1+{p^{\prime 2}+m_q^2\over z(1-z)Q^2} + {k^2\over Q^2}\right)\right] \times $$ \begin{equation} \left\{ \left[(z^2 + (1-z)^2)\left({\tdm p\over D_1} - {\tdm p + \tdm k\over D_2} \right)^2 \right] +m_q^2 \left( {1\over D_1} - {1\over D_2} \right)^2 \right\} \label{g0t} \end{equation} $$ G^{0L}_q(k^2,Q^2,\xi)= $$ $$ 8 \alpha_{em} \alpha_s(k^2+m_q ^2)e_q^2\int_0^{\lambda_{max}} d \lambda \int {d^2p^{\prime }\over \pi}\; \delta\left[\xi-\left(1+{p^{\prime 2}+m_q^2\over z(1-z)Q^2} + {k^2\over Q^2}\right)\right] \times $$ \begin{equation} \left[z^2 (1-z)^2 \, \left({1\over D_1} - {1\over D_2} \right)^2 \right] \label{g0l} \end{equation} where \begin{equation} z={1+\lambda\over 2} \label{zlam} \end{equation} \begin{equation} \tdm p=\tdm p^{\prime} + (z-1) \tdm k \label{pprime} \end{equation} $$ D_1=p^2 + z(1-z)Q^2 + m_q^2 $$ \begin{equation} D_2=(p+k)^2 + z(1-z)Q^2 + m_q^2 \label{d12} \end{equation} In the formulae given above as well as throughout the rest of the text we are using the one loop approximation for the QCD coupling $\alpha_s$ with the number of flavours $N_f=4$ and set $\Lambda_{QCD}=0.23 {\rm \; GeV}$. The function $P_S(Q_1^2,Q_2^2,W^2)$ corresponds to the contribution from the region $k^2 \le k_0^2$ in the corresponding integrals over the gluon transverse momenta. It is assumed to be dominated by the soft pomeron contribution which is estimated from the factorisation of its couplings, i.e. \begin{equation} P_S(Q_1^2,Q_2^2,W^2) = {\sigma^{SP}_{\gamma^*(Q_1^2)p}(Q_1^2,W^2) \sigma^{SP}_{\gamma^*(Q_2^2)p}(Q_2^2,W^2)\over \sigma_{pp}^{SP}} \label{sp} \end{equation} We assume that this term is only contributing to the transverse part. In equation (\ref{sp}) the cross-sections $\sigma^{SP}_{\gamma^*(Q_i^2)p}(Q_i^2,W^2)$ and $\sigma_{pp}^{SP}$ are the soft pomeron contributions to the $\gamma^*p$ and $pp$ total cross sections and their parametrisation is taken from Refs. \cite{DLTOTCX,DLDIS}. Their $W^2$ dependence is, of course, universal i.e. $$ \sigma_{pp}^{SP}=\beta_p^2\left({W^2\over W_0^2}\right)^{\alpha_{SP}(0)-1} $$ \begin{equation} \sigma^{SP}_{\gamma^*(Q_i^2)p}(Q_i^2,W^2) = \beta_{\gamma^*}(Q^2)\beta_p \left({W^2\over W_0^2}\right)^{\alpha_{SP}(0)-1} \label{sppar} \end{equation} with $W_0 = 1\rm \; GeV$ and $\alpha_{SP}(0) \approx 1.08$. The function $ \Phi_T(k^2,Q^2,x_g) $ satisfies the Balitzkij, Fadin, Kuraev, Lipatov (BFKL) equation which, in the leading $\ln (1/x)$ approximation has the following form: $$ \Phi_i(k^2,Q^2,x_g)=\Phi^0_i(k^2,Q^2,x_g)+\Phi^S (k^2,Q^2,x_g)\delta_{iT}+ {3\alpha_s(k^2)\over \pi} k^2\int_{x_g}^1 {dx^{\prime}\over x^{\prime}} \int_{k_0^2}^{\infty} {dk^{\prime 2} \over k^{\prime 2}} $$ \begin{equation} \left [ {\Phi_i(k^{\prime 2},Q^2,x^{\prime}) - \Phi_i(k^{ 2},Q^2,x^{\prime}) \over |k^{\prime 2} - k^{ 2}|} + {\Phi_i(k^{ 2},Q^2,x^{\prime})\over \sqrt{4 k^{\prime 4} + k^{4}}}\right] \label{bfklll} \end{equation} In what follows we shall consider the modified BFKL equation in which we restrict the available phase-space in the real gluon emission by the consistency constraint: \begin{equation} k^{\prime 2} \le k^2{ x^{\prime}\over x_g} \label{cc} \end{equation} This constraint follows from the requirement that the virtuality of the exchanged gluons is dominated by their transverse momentum squared. The consistency constraint (\ref{cc}) introduces the non-leading $\ln(1/x)$ effects and in the next-to-leading approximation exhausts about 70\% of the entire next-to-leading corrections to the QCD pomeron intercept. The modiffied BFKL equation takes the following form: $$ \Phi_i(k^2,Q^2,x_g)=\Phi^0_i(k^2,Q^2,x_g)+\Phi^S(k^2,Q^2,x_g)\delta_{iT}+ {3\alpha_s(k^2)\over \pi} k^2\int_{x_g}^1 {dx^{\prime}\over x^{\prime}} \int_{k_0^2}^{\infty} {dk^{\prime 2} \over k^{\prime 2}} $$ \begin{equation} \left [ {\Phi_i(k^{\prime 2},Q^2,x^{\prime})\Theta \left(k^2{ x^{\prime}\over x_g} -k^{\prime 2}\right) - \Phi_i(k^{ 2},Q^2,x^{\prime}) \over |k^{\prime 2} - k^{ 2}|} + {\Phi_i(k^{ 2},Q^2,x^{\prime})\over \sqrt{4 k^{\prime 4} + k^{4}}}\right] \label{bfklcc} \end{equation} The inhomogeneous terms in equations (\ref{bfklll}, \ref{bfklcc}) are the sum of two contributions $\Phi^0_i(k^2,Q^2,x_g)$ and $\Phi^S(k^2,Q^2,x_g)\delta_{iT}$. The first term $\Phi^0 _i (k^2, Q^2, x_g)$ corresponds to the diagram in which the two gluon system couples to a virtual photon through a quark box and are given by following equations: \begin{equation} \Phi^0_i(k^2,Q^2,x_g)=\sum_q \int_{x_g}^1 dz \; \tilde G^0_{iq}(k^2,Q^2,z) \label{phii0} \end{equation} where $$ \tilde G^0_{Tq}(k^2,Q^2,z)=2\alpha_{em} e_q^2\alpha_s(k^2+m_q^2) \int_0^1 d\lambda \left\{ {[\lambda^2 + (1-\lambda)^2][z^2+(1-z)^2] k^2 \over \lambda(1-\lambda)k^2+ z(1-z)Q^2 + m_q^2} \right. + $$ \begin{equation} 2m_q^2 \left. \left[ {1\over z(1-z)Q^2 + m_q^2} - {1\over \lambda(1-\lambda)k^2+z(1-z)Q^2 + m_q^2}\right] \right\} \label{tgt0} \end{equation} $$ \tilde G^0_{Lq}(k^2,Q^2,z)=16\alpha_{em}Q^2 k^2 e_q^2\alpha_s(k^2+m_q^2) \times $$ \begin{equation} \int_0^1d\lambda \left\{ {[\lambda (1-\lambda)][z^2(1-z)^2] \over [\lambda(1-\lambda)k^2+ z(1-z)Q^2 + m_q^2][z(1-z)Q^2 + m_q^2]} \right\} \label{tgl0} \end{equation} The second term $\Phi^S(k^2,Q^2,x_g)\delta_{iT}$, which is assumed to contribute only to the transverse component, corresponds to the contribution to the BFKL equation from the nonperturbative soft region $k^{\prime 2} < k_0^2$. Adopting the strong ordering approximation $k^{\prime 2} \ll k^2$ it is given by the following formula: \begin{equation} \Phi^S (k^2,Q^2,x_g)= {3 \alpha_s(k^2)\over \pi} \int_{x_g}^1 {dx^{\prime}\over x^{\prime}} \int_{0}^{k_0^2} {dk^{\prime 2} \over k^{\prime 2}} \Phi_T(k^{\prime 2},Q^2,x^{\prime}) \label{stophi} \end{equation} The last integral in equation (\ref{stophi}) can be interpreted as a gluon distribution in a virtual photon of virtuality $Q^2$ evaluated at the scale $k_0^2$. At low values of $x^{\prime}$ it is assumed to be dominated by a soft pomeron contribution and can be estimated using the factorisation of the soft pomeron couplings: \begin{equation} \int_{0}^{k_0^2} {dk^{\prime 2} \over k^{\prime 2}} \Phi_T(k^{\prime 2},Q^2,x^{\prime})=\pi^2 x^{\prime}g_p(x^{\prime},k_0^2) {\beta_{\gamma^*}(Q^2)\over \beta_p} \label{inhos} \end{equation} where $g_p(x^{\prime},k_0^2)$ is the gluon distribution in a proton at the scale $k_0^2$ and the couplings $\beta_{\gamma^*}(Q^2)$ and $\beta_p$ are defined by equation (\ref{sppar}). We adopt the parametrization of the gluon structure function taken from Ref.\cite{KMSTAS} i.e. $xg(x,k_0 ^2) = 1.57 (1-x)^{2.5}$ which is consinstent with the DIS data. \noindent \begin{figure}[hbpt] \begin{center} \epsfxsize = 13cm \epsfysize = 13cm \epsfbox[18 260 555 775]{ggtot.ps} \end{center} \caption{\small Energy dependence of the cross-section $\sigma^{TT} _{\gamma^* \gamma^*}(Q_1 ^2, Q_2 ^2, W^2)$ for the process $\gamma^*(Q_1 ^2) \gamma^*(Q_2^2) \rightarrow X$ for various choices of virtualities $Q^2 = Q_1 ^2 = Q_2 ^2$ corresponding to Eq.~(\ref{csx}). For each choice of the virtuality four curves are shown taking into account hard effects only (``hard part''), hard amplitude with soft pomeron contributions added in the source term of the BFKL equation (``mixed''), the full cross-section including both soft and hard pomeron contributions (``full result''). We also show the ``full result'' with the low scale of $\alpha_s$ in the impact factors: $\mu^2 = (k^2 + m_q^2)/4$. } \end{figure} In Fig.~2 we show our results for $\sigma^{TT}_{\gamma^* \gamma^*} (Q_1 ^2, Q_2 ^2, W^2)$ plotted as the function of the CM energy $W$ for three different values of $Q^2$ where $Q_1^2 = Q_2^2 = Q^2$. We plot in this figure: \begin{enumerate} \item the pure QCD (i.e. ``hard'') contribution obtained from solving the BFKL equation with the consistency constraint included (see Eq.~(\ref{bfklcc})) and with the inhomogeneous term containing only the QCD impact factor defined by equations (\ref{phii0},\ref{tgt0},\ref{tgl0}), \item the ``mixed" contribution generated by the BFKL equation (\ref{bfklcc}) with the soft pomeron contribution defined by equations (\ref{stophi}, \ref{inhos}) included in the inhomogeneous term, \item The ``full" contribution which also contains the soft pomeron term (\ref{sp}). \end{enumerate} We also show results obtained by changing the scale of the strong coupling $\alpha_s$ in the impact factors from $k^2+m_q^2$ to $(k^2+m_q^2)/4$. The scale of $\alpha_s$ in the BFKL equation is the same in the both cases. The components of the cross-section for which at least one of the photons is longitudinally polarized have very similar energy dependence to $\sigma^{TT} _{\gamma^* \gamma^*} (Q_1^2,Q^2_2,W^2)$ and give together about 60\% of the transverse-transverse contribution. We see from this figure that the effects of the soft pomeron contribution are non-negligible at low and moderately large values of $Q^2 < 10 \rm \; GeV^2$ and for moderately large values of $W < 100 \rm \; GeV$. The QCD pomeron however dominates already at $Q^2=40 \rm \; GeV^2$. We also see from this figure that for low energies $W<40 \rm \; GeV$ the phase-space effects are very important. For $W>40\rm \; GeV$ or so one observes that the cross-section exhibits the effective power-law behaviour $\sigma_{\gamma^* \gamma^*}(W) \sim (W^2)^{\lambda_P}$. The (effective) exponent increases weakly with increasing $Q^2$ and varies from $\lambda_P=0.28$ for $Q^2=2.5\rm \; GeV^2$ to $\lambda_P=0.33$ for $Q^2 = 40 \rm \; GeV^2$. This (weak) dependence of the effective exponent $\lambda_P$ with $Q^2$ is the result of the interplay between soft and hard pomeron contributions, where the former becomes less important at large $Q^2$.\\ \begin{table} \caption{ Comparison of the theoretical results to L3 data for $e^+ e^- \to e^+ e^- X$ with $E_{tag} > 30$ GeV, 30 mrad $ < \theta_{tag} <$ 66 mrad. We show in the table $d\sigma / dY$ binned in $Y$ obtained from experiment and the results of our calculation which take into account perturbative pomeron only (hard) and both perturbative and soft pomerons (hard + DL) for two different choices of scale of the $\alpha_s$ in impact factors and for $e^+e^-$ CM energy 91~GeV and 183~GeV.} \begin{center} \begin{tabular}{||c|c||c|c||c|c||} \hline\hline & \multicolumn{5}{|c||}{$\langle d\sigma / dY \rangle$ [fb] }\\ \cline{2-6} & & \multicolumn{4}{c||}{ Theory (BFKL+DL)} \\ \cline{3-6} $\Delta Y$ & Data --- QPM & \multicolumn{2}{c||}{$\alpha_s[(\tdm k^2+m_q^2)/4]$} & \multicolumn{2}{c||}{$\alpha_s (\tdm k^2+m_q^2)$} \\ \cline{3-6} & & Hard & Hard + DL & Hard & Hard + DL \\ \hline\hline \multicolumn{6}{||c||}{91 GeV} \\ \hline 2 -- 3 & $480 \pm 140 \pm 110$ & 76 & 206 & 34 & 163\\ \hline 3 -- 4 & $240 \pm 60 \pm 50 $ & 114& 237 & 53 & 173\\ \hline 4 -- 6 & $110 \pm 30 \pm 10 $ & 60 & 109 & 29 & 74\\ \hline\hline \multicolumn{6}{||c||}{183 GeV} \\ \hline 2 -- 3 & $180 \pm 120 \pm 50$ & 51 & 68 & 25 & 42\\ \hline 3 -- 4 & $160 \pm 50 \pm 30 $ & 70 & 86 & 34 & 49\\ \hline 4 -- 6 & $120 \pm 40 \pm 20 $ & 70 & 85 & 35 & 47\\ \hline\hline \end{tabular} \end{center} \end{table} Using Formula~(\ref{conv}) integrated over the virtualities in the range allowed by the relevant experimental cuts, we have calculated the total cross-section for the process $e^+ e^- \rightarrow e^+ e^- + X$ for LEP1 and LEP2 energies and confronted results of our calculation with the recent experimental data obtained by the L3 collaboration at LEP \cite{L3}. Comparison of our results with experimental data is sumarised in Table~1. We show comparison for $d\sigma/dY$, where $Y=\ln (W^2/Q_1Q_2)$ with subtracted Quark Parton Model (QPM) contribution. We see that the contamination of the cross-section by soft pomeron is substantial. The data do also favour the smaller value of the scale of $\alpha_s$. In general, the results of our calculation lay below the data, however the error bars are still quite large, so that the discrepancy is not very pronounced. Let us also mention that cuts applied to obtain the data shown in Table~1 admit rather low $\gamma\gamma$ energies i.e. below 10~GeV \cite{L3}, which probably is not sufficient to justify the validity of high energy limit in QCD. \section{Exclusive $J/\psi$ production} The experimental aspects of the measurement of double exclusive $J/\psi$ production are different from those for the virtual photons scattering. Namely, since the $c$-quark provides the energy scale, we may perturbatively describe the cross-section for the process of exclusive $J/\psi$ production in which almost real photons take part. It is an important feature beacause the photon flux in electron is dominated by low virtualities. On the other hand one may measure the produced $J/\psi$-s through theirs decay products with no need of tagging of the electrons. Thus, it is prefered to focus on events with anti-tagged leptons. The cross-section for the process $e^+e^- \rightarrow e^+e^- + Y$ for anti-tagged $e^{\pm}$ corresponds to the production of the hadronic state~$Y$ in $\gamma\gamma$ collision and is given by the following convolution integral: \cite{GGREV} \begin{equation} \sigma_{e^+e^- \to e^+e^- + Y} = \int_0^1 dy_1 \int_0^1 dy_2 \Theta ( W^2 - W^2_{Y0}) \sigma_{\gamma \gamma \rightarrow Y}(W^2) f_{\gamma/e}(y_1) f_{\gamma/e}(y_2). \label{conv2} \end{equation} where the $\gamma\gamma$ system invariant mass squared $W^2$ is related to the lepton CM energy squared $s$ by the simple formula: $W^2 = y_1y_2 s$. The flux factor takes the form: \begin{equation} f_{\gamma/e}(y)={\alpha_{em}\over 2 \pi} \left[ {1 + (1-y)^2 \over y}\;{\rm ln}\,{Q_{max}^2\over Q_{min}^2} - 2 m_e ^2 y \left({1\over Q_{min}^2} - {1\over Q_{max} ^2}\right) \right]. \label{flux} \end{equation} and \begin{equation} Q_{min}^2 = {m_e^2 y^2 \over ( 1-y)} \label{pmin} \end{equation} \begin{equation} Q_{max}^2=(1-y)E_{beam}^2 \theta_{max}^2. \label{pmax} \end{equation} The lower limit follows from the kinematics of photon emission from a lepton whereas the upper one arises from the upper limit $\theta_{max}$ for the lepton scattering angle. The minimal invariant mass squared of the hadronic system $W^2_{Y0}$, the angle $\theta_{max}$ and the beam energy $E_{beam}$ depend on the process and experimental conditions. For diffractive $J/\psi$ production we shall choose $\theta_{max}=30$~mrad in accordance with LEP conditions and $W_{Y0}=15\rm \; GeV$.\\ The formalism that we shall employ to evaluate the cross-section of the sub-process $\gamma \gamma \rightarrow J/\psi J/\psi$ is very similar to this used in the previous section. However some modification are neccessary in order to adopt to specific features of the process. First of all we have to go beyond the forward configuration of the pomeron by the use of the BFKL equation with non-zero momentum transver. Besides that, we introduce a parameter $s_0$ in the propagators of exchanged gluons instead of the infra-red cut-off $k_0 ^2$ applied in the previous case. This parameter can be viewed upon as the effective representation of the inverse of the colour confinement radius squared. Sensitivity of the cross-section to its magnitude can serve as an estimate of the sensitivity of the results to the contribution coming from the infrared region. It should be noted that formula (\ref{ima}) gives finite result in the limit $s_0=0$. While analyzing this process we use the asymptotic (high-energy) form of the amplitude, neglecting the phase space effects.\\ The imaginary part ${\rm Im} A(W^2,t=-Q_P ^2)$ of the amplitude for the considered process which corresponds to the diagram in Fig.~1b can be written in the following form: \begin{equation} {\rm Im} A(W^2,t=- Q_P^2) = \int {d^2\tdm k\over\pi}{\Phi_0(k^2, Q_P ^2)\Phi(x,\tdm k,\tdm Q_P)\over [(\tdm k + \tdm Q_P /2)^2 +s_0][(\tdm k - \tdm Q_P /2)^2+s_0]} \label{ima} \end{equation} In this equation $x=m_{J/\psi}^2/W^2$ where $W$ denotes the total CM energy of the $\gamma \gamma$ system, $m_{J/\psi}$ is the mass of the $J/\psi$ meson, $\tdm Q_P / 2 \pm \tdm k$ denote the transverse momenta of the exchanged gluons and $\tdm Q_P$ is the transverse part of the momentum transfer. \noindent \begin{figure} \epsfxsize = 12cm \epsfysize = 8cm \epsfbox{ppfig2.eps} \caption{\small The diagrams describing the coupling of two gluons to the $\gamma \rightarrow J/\psi$ transition vertex. } \end{figure} The impact factor $\Phi_0(k^2, Q_P^2)$ describes the $\gamma J/\psi$ transition induced by two gluons and the diagrams defining this factor are illustrated in Fig.~3. In the nonrelativistic approximation they give the following formula for $\Phi_0(k^2, Q_P^2)$ \cite{FORSHAW,GINZBURG}: \begin{equation} \Phi_0(k^2, Q_P^2)= {C\over 2}\sqrt{\alpha_{em}}\alpha_s(\mu^2) \left[{1\over \bar q^2} - {1\over m_{J/\psi}^2/4+k^2}\right] \label{impf0} \end{equation} where \begin{equation} C=q_c{8\over 3} \pi m_{J/\psi} f_{J/\psi} \label{c} \end{equation} with $q_c=2/3$ denoting the charge of a charm quark and \begin{equation} \bar q ^2= {m_{J/\psi}^2+Q_P^2\over 4} \label{qbar2} \end{equation} \begin{equation} f_{J/\psi}= \sqrt{ {3m_{J/\psi}\Gamma_{J/\psi\rightarrow l^+ l^-} \over 2\pi \alpha_{em} ^2} } \label{fpsi} \end{equation} where $\Gamma_{J/\psi \rightarrow l^+ l^-}$ is the leptonic with of the $J/\psi$ meson. In our calculations we will set $f_{J/\psi}=0.38{\rm \; GeV}$. The function $\Phi(x,\tdm k,\tdm Q_P)$ satisfies the non-forward BFKL equation which in the leading $\ln(1/x)$ approximation has the following form: $$ \Phi(x,\tdm k,\tdm Q_P)=\Phi_0(k^2, Q_P^2)+ {3\alpha_s(\mu^2)\over 2\pi^2}\int_x^1{dx^{\prime}\over x^{\prime}} \int {d^2\tdm k' \over (\tdm k' - \tdm k)^2 + s_0} \times $$ $$ \left\{\left[{{\tdm k_1^2}\over {\tdm k_1^{\prime 2}} + s_0} + {{\tdm k_2^2}\over {\tdm k_2^{\prime 2}} + s_0} - Q_P^2 {(\tdm k' - \tdm k)^2+s_0 \over ({\tdm k_1^{\prime 2}} + s_0) ({\tdm k_2^{\prime 2}} + s_0)} \right] \Phi(x',\tdm k' ,\tdm Q_P) - \right. $$ \begin{equation} \left. \left[{{\tdm k_1^2}\over {\tdm k_1^{\prime 2}} + (\tdm k' - \tdm k)^2 +2s_0} + {{\tdm k_2^2}\over {\tdm k_2^{\prime 2}} + (\tdm k' - \tdm k)^2 +2s_0} \right] \Phi(x',\tdm k,\tdm Q_P) \right\} \label{bfkl} \end{equation} where $$ {\tdm k_{1,2}} = {\tdm Q_P \over 2}\pm \tdm k $$ and \begin{equation} {\tdm k_{1,2}^{\prime}} = {\tdm Q_P \over 2} \pm \tdm k^{\prime} \label{k12} \end{equation} denote the transverse momenta of the gluons. The scale of the QCD coupling $\alpha_s$ which appears in equations (\ref{impf0}) and (\ref{bfkl}) will be set $\mu^2=k^2+Q_P^2/4 +m_c^2$ where $m_c$ denotes the mass of the charmed quark. The differential cross-section is related in the following way to the amplitude~$A$: \begin{equation} {d \sigma \over dt} = {1\over 16 \pi} |A(W^2,t)|^2 \label{dsdt} \end{equation} Generalization of the consistency constraint (\ref{cc}) to the case of non-forward configuration with $Q_P^2 \ge 0$ takes the following form: \begin{equation} k'^2 \le (k^2+ Q_P^2/4) {x'\over x} \label{kc2} \end{equation} Besides the BFKL equation (\ref{bfkl}) in the leading logarithmic approximation we shall also consider the equation which will embody the constraint (\ref{kc2}) in order to estimate the effect of the non-leading contributions. \\ The corresponding equation which contains constraint (\ref{kc2}) in the real emission term reads: $$ \Phi(x,\tdm k,\tdm Q_P)=\Phi_0(k^2, Q_P^2)+ {3\alpha_s(\mu^2)\over 2\pi^2}\int_x^1{dx^{\prime}\over x^{\prime}} \int {d^2\tdm k' \over (\tdm k' - \tdm k)^2 + s_0} \times $$ $$ \left\{\left[{{\tdm k_1^2}\over {\tdm k_1^{\prime 2}} + s_0} + {{\tdm k_2^2}\over {\tdm k_2^{\prime 2}} + s_0} - Q_P^2 {(\tdm k' - \tdm k)^2+s_0 \over ({\tdm k_1^{\prime 2}} + s_0) ({\tdm k_2^{\prime 2}} + s_0)} \right] \Theta \left((k^2+Q_P^2/4)x'/x-k^{\prime 2}) \right) \times \right. $$ \begin{equation} \left. \Phi(x',\tdm k',\tdm Q_P) - \left[{{\tdm k_1^2}\over {\tdm k_1^{\prime 2}} + (\tdm k' - \tdm k)^2 +2s_0} + {{\tdm k_2^2}\over {\tdm k_2^{\prime 2}} + (\tdm k' - \tdm k)^2 +2s_0} \right] \Phi(x',\tdm k,\tdm Q_P) \right\} \label{bfklkc} \end{equation} We solved equations (\ref{bfkl}) and (\ref{bfklkc}) numerically setting $m_c=m_{J/\psi} /2$. Brief summary of the numerical method and of the adopted approximations in solving equations (\ref{bfkl},\ref{bfklkc}) has been given in Ref.\cite{KMPSI}. Let us recall that we used running coupling with the scale $\mu^2=k^2+Q_P^2/4+m_c^2$. The parameter $s_0$ was varied within the range $0.04 {\rm \; GeV}^2 < s_0<0.16 {\rm \; GeV}^2$. It should be noted that the solutions of equations (\ref{bfkl}, \ref{bfklkc}) and the amplitude (\ref{ima}) are finite in the limit $s_0=0$. This follows from the fact that both impact factors $\Phi_0(k^2, Q_P^2)$ and $\Phi(x,\tdm k,\tdm Q_P)$ vanish for $\tdm k=\pm \tdm Q_P/2$ (see equations (\ref{impf0}, \ref{bfkl}, \ref{bfklkc})). The results with finite $s_0$ are however more realistic. \noindent \begin{figure}[hbpt] \leavevmode \begin{center} \epsfxsize = 13cm \epsfysize = 13cm \epsfbox[18 200 565 755]{ppfig3.ps} \end{center} \caption{\small Energy dependence of the cross-section for the process $\gamma\gamma \rightarrow J/\psi J/\psi$. The two lower curves correspond to the calculations based on equation (\ref{bfklkc}) which contains the non-leading effects coming from the constraint (\ref{kc2}). The continuous line corresponds to $s_0=0.04 {\rm \; GeV}^2$ and the dashed line to $s_0=0.16 {\rm \; GeV}^2$. The two upper curves correspond to equation (\ref{bfkl}) i.e. to the BFKL equation in the leading logarithmic approximation. The dashed-dotted line corresponds to $s_0=0.04 {\rm \; GeV}^2$ and short dashed line to $s_0=0.16{\rm \; GeV}^2$. } \end{figure} In Fig.~4 we show the cross-section for the process $\gamma \gamma \rightarrow J/\psi J/\psi$ plotted as the function of the total CM energy $W$. We show results based on the BFKL equation in the leading logarithmic approximation as well as those which include the dominant non-leading effects. The calculations were performed for the two values of the parameter $s_0$ i.e. $s_0=0.04 {\rm \; GeV}^2$ and $s_0=0.16 {\rm \; GeV}^2$. In Fig.~5 we show the $t$-dependence of the cross-section calculated for $s_0 = 0.10 {\rm \; GeV}^2$. We show in this figure results for two values of the CM energy $W$ ($W=50 {\rm \; GeV}$ and $W=125 {\rm \; GeV}$) obtained from the solution of the BFKL equation with the non-leading effects taken into account (see Eq.~(\ref{bfklkc})) and confront them with the Born term which corresponds to the two (elementary) gluon exchange. The latter is of course independent of the energy $W$. The values of the energy $W$ were chosen to be in the region which may be accessible at LEP2. Let us discuss crucial features of the obtained results: \begin{enumerate} \item {\bf Non leading corrections}. We see from Fig.~4 that the effect of the non-leading contributions is very important and that they significantly reduce magnitude of the cross-section and slow down its increase with increasing CM energy $W$. \item {\bf Energy dependence}. The cross-section exhibits approximate $(W^2)^{2\lambda_P}$ dependence. The parameter $\lambda_P$, which slowly varies with the energy~$W$ takes the values $\lambda_P \sim 0.23 - 0.28$ within the energy range $20{\rm \; GeV} < W < 500{\rm \; GeV}$ relevant for LEP2 and for possible TESLA measurements. These results correspond to the solution of the BFKL equation (\ref{bfklkc}) which contains the non-leading effects generated by the constraint (\ref{kc2}). The (predicted) energy dependence of the cross-section ($(W^2)^{2\lambda_P}, \lambda_P \sim 0.23 - 0.28$) is marginally steeper than that observed in $J/\psi$ photo-production \cite{VMPHOTOP}. It should however be remebered that the non-leading effects which we have taken into account although being the dominant ones still do not exhaust all next-to-leading QCD corrections to the BFKL kernel \cite{BFKLNL}. The remaining contributions are expected to reduce the parameter $\lambda_P$ but their effect may be expected to be less important than that generated by the constraint (\ref{kc2}). The cross-section calculated from the BFKL equation in the leading logarithmic approximation gives much stronger energy dependence of the cross-section (see Fig.~4). \item {\bf The value of the cross-section}. Enhancement of the cross-section is still appreciable after including the dominant non-leading contribution which follows from the constraint (\ref{kc2}). Thus while in the Born approximation (i.e. for the elementary two gluon exchange which gives energy independent cross-section) we get $\sigma_{tot} \sim 1.9-2.6$~pb the cross-section calculated from the solution of the BFKL equation with the non-leading effects taken into account can reach the value 4~pb at $W=20 \rm \; GeV$ and 26~pb for $W=100 \rm \; GeV$ i.e. for energies which can be accessible at LEP2. \item {\bf Infrared sensitivity}. The magnitude of the cross-section decreases with increasing magnitude of the parameter $s_0$ which controls the contribution coming from the infrared region. This effect is however much weaker than that generated by the constraint (\ref{kc2}) which gives the dominant non-leading contribution. The energy dependence of the cross-section is practically unaffected by the parameter $s_0$. \item {\bf The $t$-dependence}. Plots shown in Fig.~5 show that the BFKL effects significantly affect the $t$-dependence of the differential cross-section leading to steeper $t$-dependence than that generated by the Born term. Possible energy dependence of the diffractive slope is found to be very weak (see Fig.~5). Similar result was also found in the BFKL equation in the leading logarithmic approximation \cite{BARTLQ}. \end{enumerate} \begin{figure} \leavevmode \begin{center} \epsfxsize = 13cm \epsfysize = 13cm \epsfbox{ppfig4.ps} \end{center} \caption{\small The differential cross-section of the process $\gamma \gamma \rightarrow J/\psi J/\psi$ corresponding to the solution of equation (\ref{bfklkc}) which contains the non-leading effects coming from the consistency (kinematical) constraint (\ref{kc2}) shown for two values of the CM energy $W$, $W=50{\rm \; GeV}$ (continuous line) and $W=125 {\rm \; GeV}$ (dashed line). The short dashed line corresponds to the Born term i.e. to the elementary two gluon exchange mechanism which gives the energy independent cross-section. The parameter $s_0$ was set equal to $0.10{\rm \; GeV}^2$. } \end{figure} In our calculations we have assumed dominance of the imaginary part of the production amplitude. The effect of the real part can be taken into account by multiplying the cross-section by the correction factor $1+tg^2(\pi\lambda_P/2)$ which for $\lambda_P \sim 0.25$ can introduce additional enhancement of about 20~\%.\\ The photonic cross-sections that we obtained in this section are rather low in terms of the expected number of events, at least for the LEP2 luminosity. Therefore we consider the most inclusive observables relevant for double $J/\psi$ production in $e^+e^-$ collisions which is the total cross-section $\sigma_{tot} (e^+ e^- \to e^+ e^- J/\psi J/\psi)$. In fact, it is convenient to impose additionally the anti-tagging condition. Taking $\theta_{max}=30$~mrad we get for the $\sigma_{tot} (e^+ e^- \to e^+ e^- J/\psi J/\psi)$ the values of about 0.14~pb at $\sqrt{s}=175\rm \; GeV$ and 0.74~pb at $\sqrt{s}=500\rm \; GeV$ (i.e. for typical energies at LEP2 and TESLA respectively). Therefore, assuming the LEP2 luminosity to be about 500~pb$^{-1}$ we predict about 70~events, which is far below the previous expectations~\cite{GGREV}. Besides, if one measures both the $J/\psi$-s through the leptonic decay channels the rate should be divided by factor of about 20, which cuts down the statistics to only a few events. \section{Discussion and summary} From the theoretical point of view, there exist excellent oportunities to study the exchange of the QCD pomeron in $e^+ e^-$ colliders. The two golden-plated measurements for this purpose are exclusive $J/\psi$ production and the total $\gamma^* \gamma^*$ cross-section. Both these processes allow to reduce substantially the contribution of unknown, nonperturbative elements. However, the leptonic cross-sections in both cases are well below 1~pb in LEP2 conditions, which makes the measurement rather difficult there. Nevertheless this problem does not appear at the future linear colliders $e^+ e^-$ for which the luminosity is expected to be much larger than at LEP and moreover the cross-section for diffractive processes is enhanced due to the photon flux and the pomeron effects. The large expected statistics enables one to reach the region of large photon virtualities (for double tagged events) where the perturbative calculations are more reliable. \\ The important point that should be stressed once more is the existence of large non-leading corrections to BFKL equation, which influence dramatically the theoretical estimate of the pomeron intercept i.e. the behaviour of the cross-sections as functions of the energy. The recently calculated magnitude of next-to-leading contribution to the intercept (for any relevant value of the strong coupling constant) is comparable or even greater than the leading term. This implies a very poor convergence of the perturbative series. Thus one is forced to rely on a resummation scheme. We adopt the so called consistency constraint, which is based on the requirement that the virtualities of gluons exchanged along the ladder are dominated by transverse momenta squared. This constraint introduces at the next-to-leading order a correction to the pomeron intercept which exhausts about 70\% of the exact QCD result. The main advantage of this approach is that there is a good physical motivation behind it. Moreover it also offers an approximate resummation scheme for the perturbative expansion of the intercept.\\ Employing this scheme we found significant reduction of the predicted value of the intercept in comparison to the leading value. We find that the calculated behaviour of the $\gamma^* \gamma^*$ total cross-section exhibits approximate power law dependence $(W^2)^{\lambda_P}$ with $0.28<\lambda_P<0.35$. It is also found that the cross-section for $\gamma\gamma \to J/\psi J/\psi$ increases with increasing energy $W$ as $(W^2)^{2\lambda_P}$ with $\lambda_P$ varying from~0.23 to~0.28. This has important consequences for the phenomenology, since the enhancement of the cross-section although still quite appreciable is much smaller than that which follows from estimates based on the leading logarithmic approximation \cite{GGREV}. The results of our calculation are in fair agreement with the existing data for $\gamma^*\gamma^*$ cross-section from LEP, although the theoretical calculations have a tendency to underestimate experimental results. They are also much more realistic than the predictions following from the leading order BFKL equation, which are an order of magnitude larger. The encouraging element is that even this very first data with rather low statistics, are enough to show clearly the importance of non-leading corrections. We may therefore expect that when the excellent data from linear colliders will be available we will acquire very good opportunity to test our models and to understand more deeply the physics of the QCD pomeron. \\ \section*{Acknowledgments} We are grateful to the Organizers for the interesting and stimulating Conference. We thank Albert De Roeck for his interest in this work and useful discussions. This research was partially supported by the Polish State Committee for Scientific Research (KBN) grants 2~P03B~184~10, 2~P03B~89~13, 2~P03B~084~14 and by the EU Fourth Framework Programme 'Training and Mobility of Researchers', Network 'Quantum Chromodynamics and the Deep Structure of Elementary Particles', contract FMRX--CT98--0194.
\section{Introduction\protect\\} \label{sec:intro} In the absence of a complete formulation of M/string theory, BPS $p$-brane solutions from various supergravities, which are the low energy and/or weak-coupling limits of this intrinsically non-perturbative unified theory, are almost the only sources from which reliable non-perturbative information about this theory can be extracted. These solutions played vital role in every major development in the past and will continue to do so. BPS properties obtained in the low energy and weak-coupling limits remain to be valid non-perturbatively for the corresponding BPS states. These properties should also be independent of whether they are obtained from fields in bulk spacetime or from fields on the corresponding worldvolume. For this reason, BPS properties obtained from the supergravity $p$-brane solutions are often used to obtain information about fields living on the corresponding worldvolume and vice-versa. For example, the mass per unit $p$-brane volume calculated from a spacetime BPS Dp-brane solution should correspond to the Dp-brane tension of the corresponding Born-Infeld action. This is trivial for simple Dp-brane solutions. But it is quite non-trivial for complicated solutions such as BPS bound state solutions. For example, the tension formula of $(m, n)$-string bound state of Schwarz \cite{schone}, with $m, n$ relatively prime integers, can be used to determine the corresponding D-string worldsheet $U(1)$ field strength, in the linear approximation, as $g m$ with $g$ the string coupling \cite{calm}. This in turn can be used to determine the point charge due to the ending of a fundamental string (for short, F-string) on a general Dp-brane. After all, a $p$-brane configuration represents the field configuration created by the corresponding $p$-brane source just like a static electric field due to a point charge represents the field surrounding the point source. Static interactions should be the same whether we calculate from the fields or from the sources. As mentioned above, BPS properties should be the same no matter whether they are obtained from the field configuration or from the source. However, apart from these BPS properties, one must be careful in using them since in general the fields and the source are not independent and they actually interact with each other through the so-called backreaction except in the case where the so-called decoupling limit is taken. In this limit, the modes propagating on the worldvolume will decouple from the modes propagating in the bulk spacetime even though these spacetime modes may have their origins from the worldvolume. So we expect that in the decoupling limit either the worldvolume fields or the bulk fields can be used to describe the same physics. Recently proposed $AdS/CFT$ correspondence by Maldacena \cite{mal} is one such example. So BPS $p$-brane solutions can provide not only the non-perturbative information but also a basis for new types of strong-weak dualities. These are certainly important for us to seek the eventual formulation of M/string theory, along with the powerful D-brane picture of Polchinski\cite{pol}. In this paper, we will construct new spacetime solutions for BPS bound states of Dp-branes carrying certain units of the quantized worldvolume constant electric field strength. We argued in our previous paper \cite{lurone} that such BPS bound states should exist for $ 1 \le p \le 8$ in both type IIA (when $p$ is even) and type IIB (when $p $ is odd) theories\footnote{ In this paper, we specialize in cases of $2 \le p \le 7$.}. The existence of these bound states was also discussed in \cite{arfs} but in a different approach of mixed boundary conditions. The non-vanishing $U(1)$ field strength on the worldvolume indicates that such a bound state carries information about F-strings ending on the Dp-brane, yet preserving one half of the spacetime supersymmetries. These Dp-brane bound states should be identified with the so-called (F, Dp) bound states which can be obtained from the $(m,n)$-string of Schwarz or (F, D1) bound state in type IIB string theory by T-dualities along the transverse directions. Here integers $m, n$ are relatively prime. This is precisely the method which we will use here to construct these solutions which was also mentioned in \cite{rust,cosp}. The $p = 3, 4, 6$ configurations for (F, Dp) have been given in \cite{rust,grelpt,cosp}, respectively. Other non-threshold bound states for one $p'$-brane within another p-brane with $p' < p$ in M and Type II string theories have been discussed in \cite{rust,grelpt,cosp,col}. Our worldvolume picture of these Dp-brane bound states indicates clearly that the notation `F' in (F, Dp) represents actually an infinite number of parallel NS-strings. Precisely, we have one NS-string per $(2\pi)^{p - 1} \alpha'^{(p - 1)/2}$ area over a $(p - 1)$-dimensional plane which is perpendicular to this string. Each NS-string is $m$ F-strings if the quantized worldvolume constant field strength $F_{01} = g m$, with $g$ the string coupling constant (where the field lines are chosen along $x^1$ axis). These were obtained in \cite{lurone} by noting that the (F, D1) worldvolume action can be obtained from a (F, Dp), for $ p \ge 2$, worldvolume action by T-dualities. By this, we also determined the tensions for (F, Dp) bound states in \cite{lurone} which clearly indicates that (F, Dp) is a non-threshold bound state. We will show in this paper from the corresponding spacetime solutions that indeed these are all true. Since we have constructed in this paper the space-time configurations for all the (F, Dp) bound states in type IIA and IIB theories, it would be natural to look at how they affect the $AdS/CFT$ correspondences conjectured by Maldacena. By examining the (F, D3) configuration carefully, we find that the decoupling limit in this case does not automatically imply the space-time geometry to be $AdS_5 \times S^5$ as happens in the case of Maldacena, i.e., for simple D3-branes. We also find that the string coupling constant in this case is replaced by a finite but a smaller effective string coupling constant. However, we observed that by a suitable rescaling of the coordinates the $AdS_5/CFT_4$ correspondence may still hold true, but now with respect to (F, D3) bound state rather than the simple D3-branes. The effective string coupling constant in the near horizon region is quantized in terms of the integer $n$, the 5-form flux and the integer $m$, the number of F strings per $(2\pi)^2 \alpha'$ area over the two dimensional plane perpendicular to the F strings, where $m$, $n$ are relatively prime integers. This coupling can be independent of the usual string coupling at spatial infinity in the limit $m \gg n$. This paper is organized as follows. In the following section, we discuss in some detail the so-called vertical and diagonal (or double) dimensional reductions/oxidations in type IIA and type IIB supergravities which are the basic methods we use to construct the (F, Dp) configurations. In section 3, we first give a detail construction of (F, D2) as an illustrating example. We then list results for each of (F, Dp) for $ 3 \le p \le 7$. We also present the (W, Dp) (with W representing the waves) solutions for $0 \le p \le 6$ from these newly found (F, Dp) configurations by T-dualities along the direction of F-strings. In section 4, we calculate the charge per unit $(p - 1)$-dimensional area for the F-strings and show that we indeed have $m$ F-strings per $(2\pi)^{p -1} \alpha'^{(p - 1)/2}$ of ($p - 1$)-dimensional area for these bound states. We also calculate the mass per unit $p$-brane volume for these solutions and show that the corresponding tensions agree precisely with what we obtained in \cite{lurone} based on the worldvolume study. In section 5, we study the decoupling limit for the (F, D3) bound state. Finally, we conclude this paper in section 6. \section{Diagonal and Vertical Dimensional Reductions/Oxidations\protect\\} \label{sec:dvdr} It is well-known that type IIA and type IIB superstring theories compactified on a circle are equivalent by T-duality i.e. order by order in perturbation theory. More precisely, IIA and IIB theories are interchanged by the T-duality transformation $R \rightarrow 1/R$, with $R$ the compactified radius, along with an interchange of momentum or KK modes with the winding modes. However, this equivalence does not extend to the respective $S^1$-compactified supergravity theories due to the absence of the string winding modes. But this will not have any effect on the massless modes in $D = 9$. Therefore, the $D = 9, N = 2$ supergravity obtained from type IIA supergravity by the dimensional reduction must be equivalent by T-duality to that obtained from type IIB supergravity. This also follows from supersymmetry because $D = 9, N = 2$ supergravity is unique up to field redefinitions. Hence, the T-duality acting on the $D = 9$ massless fields must map fields in IIA basis to those in IIB basis. This feature, as we will see, is precisely what we need to obtain a BPS solution in type IIA theory from a BPS solution in type IIB theory and vice versa by a T-duality transformation. It is well-known that type II theories contain the following simple BPS states\footnote{ By simple, we mean an object whose charge is associated with only one antisymmetric tensor field in type II theories. We have not included here the D9 branes, i.e., the so-called spacetime-filling branes\cite{ber}.}: \begin{eqnarray} {\rm Type~ IIA}~\qquad&& {\rm W}~~{\rm F}~~{\rm NS5}~~{\rm KK}~~ \qquad~{\rm D0}~~{\rm D2}~~{\rm D4}~~{\rm D6}~~{\rm D8}\nonumber\\ {\rm Type~ IIB}~\qquad&& {\rm W}~~{\rm F}~~{\rm NS5}~~{\rm KK}~ \qquad~{\rm D(-1)}~~{\rm D1}~~{\rm D3}~~{\rm D5}~~{\rm D7}\nonumber\\ \end{eqnarray} where W, F, NS5 and KK denote waves, fundamental strings, NS fivebranes, and KK monopoles, respectively, and they are associated with the NSNS fields. Also, Dp ($ - 1\le p \le 8$) are the so-called D-branes and they are associated with the so-called RR fields. The action of T-duality along parallel or transverse direction on the above objects present in type II theories is given in the following table. \begin{center} \begin{tabular}{|c|c|c|} \hline & Parallel & Transverse\\ \hline Dp& D(p $-$ 1) & D(p + 1)\\ F& W & F \\ W& F & W \\ NS5& NS5 &KK \\ KK & KK & NS5 \\ \hline\end{tabular} \end{center} The T-duality may be performed along either one of the worldvolume directions or one of the transverse directions (for KK monopole, the transverse direction is taken to be the nut direction). It is clear from the above table that if we start from a (F, D1) bound state in the type IIB theory, we will have (F, D2) bound state in type IIA theory by T-dualizing (F, D1) along one of its transverse directions. Then we can have (F, D3) bound state in the type IIB theory by T-dualizing (F, D2) along one of its transverse directions. Repeating this procedure, we can have (F, D4) from (F, D3), (F, D5) from (F, D4), (F, D6) from (F, D5) and finally (F, D7) from (F, D6). We cannot obtain (F, D8) from (F, D7) by this procedure. We will explain possible reasons behind this at the end of this section. If we T-dualize the above bound states along the longitudinal direction of F in (F, Dp) for $ 1 \le p \le 7$, we have (W, Dp) bound states for $ 0 \le p \le 6$ with W representing the waves. The question now is how to implement the above T-duality at the level of solutions of the IIA and IIB supergravities. As discussed above, the $S^1$ compactified IIA and IIB supergravities are not equivalent by T-duality except for the massless sector of $D = 9, N = 2$ supergravity. This implies that T-duality can be implemented in obtaining solutions in one supergravity from the known solutions each with one isometry along the would-be compactified direction in the other supergravity. This is exactly the rationale behind the procedure of the so-called vertical dimensional reduction and diagonal (or double) dimensional oxidation or the diagonal reduction and vertical oxidation, depending on whether the T-duality acts along a transverse direction or a worldvolume direction. In this paper, we choose a convention such that $H_n$ denotes an $n$-form field strength in type IIB supergravity or in $D = 9, N = 2$ supergravity in terms of the type IIB basis and similarly $F_n$ denotes an $n$-form in type IIA supergravity or in $D = 9, N = 2$ supergravity in terms of the type IIA basis. If an $n$-form is reduced to an $(n-1)$-form from $D = 10$ to $D = 9$, we denote the resulting $(n-1)$-from with a superscript `1' indicating that it is obtained by one-step reduction. For example, if the reduction is along one of the indices of $H_n$, we have $H_n \rightarrow H^{(1)}_{n - 1} \wedge dz$. In the case of T-duality along a transverse direction of a BPS $p$-brane solution ($ p \le 6$) as discussed in \cite{dabghw,lups}, we first use the ``no-force" condition for the $p$-brane configuration to obtain a multi-center solution from a single center one by placing many $p$-branes parallel to each other. Without loss of generality and to be specific, let us assume that this $p$-brane is in type IIB supergravity. The multi-center solution can be obtained by the following replacement in the corresponding Harmonic function. For example, if a single-center Harmonic function $H = 1 + Q_p /r^{7 - p}$ with $r^2 = (y^1)^2 + (y^2)^2 + \cdots + (y^{9 - p})^2$, then the multi-center Harmonic function will be given as, \begin{equation} H = 1 + \sum_n \frac{Q_p^{(n)}}{\mid \vec{y} - \vec{y}_n \mid^{7 - p}}. \label{eq:mcs} \end{equation} In order to generate one isometry along the T-dual direction, say, $y^{9 - p}$-direction, as pointed out in \cite{lups}, a continuous stack of $p$-branes with uniform charge density should be placed along the coordinate $z = y^{9 - p}$ such that the summation in Eq.\ (\ref{eq:mcs}) can be replaced by an integration\footnote{This seems to work also for non-BPS solutions as discussed in \cite{lups}.}. For the purpose of this paper, let us take a close look at this continuum limit. We first take $\vec{y}_n = 2 \pi n a \hat{z}$ in Eq.\ (\ref{eq:mcs}), with $n \in Z$, $\hat{z}$ the unit vector along $y^{9 - p}$-direction and $a = \sqrt{\alpha'}$ where $\alpha'$ is the string constant. If we allow the range of the summation in Eq.\ (\ref{eq:mcs}) to be $-\infty < n < \infty$ and take all charges to be equal, i.e., $Q_p^{(n)} = Q_p$, then the multi-center solution is an infinite periodic array of $p$-branes. The solution would be a stringy one if we take $\alpha' \rightarrow 0$. Under this limit, the summation in Eq.\ (\ref{eq:mcs}) can also be replaced by an integration, i.e., \begin{equation} Q_p \sum_{n = - \infty}^\infty \frac{1}{\mid \vec{y} - 2\pi n a \hat{z} \mid^{7 - p}} \rightarrow \frac{Q_p}{2\pi a} \int_{- \infty}^\infty \frac{dz} {(r^2 + z^2)^{(7 - p)/2}} = \frac{\tilde{Q}_p}{ r^{6 - p}}, \end{equation} where $\tilde{Q}_p = c_p Q_p$ with constant $c_p = Q_{p + 1}/Q_p$. The explicit expressions for $Q_{p + 1}$ and $Q_p$ are given in the next section. In the above, we have now $r^2 = (y^1)^2 + \cdots + (y^{8 - p})^2$. In other words, one isometry is generated along the T-dual direction by the infinite periodic array of $p$-branes. Once this solution is known, a BPS $p$-brane solution in $D = 9$ can be obtained from it. The Einstein metric for $D = 9$ $p$-brane can be obtained from the corresponding metric of $D = 10$ $p$-brane as (see, for example, the appendix of \cite{duflp}) \begin{equation} ds^2_{10}~ ({\rm type~ IIB})= e^{-\varphi_B/(2\sqrt{7})} ds_9^2 + e^{\sqrt{7} \varphi_B/2} d z^2, \label{eq:mriib} \end{equation} where $\varphi_B$ is a dilatonic scalar originating from the dimensional reduction. Since we know $d s^2_{10}~({\rm type~ IIB})$ and it also has one isometry along $z$-direction, we can read from the above the new field $\varphi_B$ and therefore the $D = 9$ Einstein metric $d s_9^2$. The $D = 10$ dilaton $\phi_B$ remains the same as the one for the solution of the infinite periodic array of $p$-branes in $D = 10$. The rule for reducing an $n$-form field strength $H_n$ is as follows: if it carries an electric-like charge, it will remain the same in $D = 9$ whereas it will reduce to an $(n-1)$-form, i.e., $H_n \rightarrow H^{(1)}_{n - 1}\wedge dz$, if it carries a magnetic charge. So we now obtain a $D = 9$ BPS $p$-brane solution from a $D = 10$ BPS $p$-brane solution by the so-called vertical dimensional reduction. We expect that the T-dual of the above $p$-brane in type IIB gives a $(p + 1)$-brane in type IIA. Let us first pretend that we know this $(p + 1)$-brane solution in type IIA. We denote the brane coordinate $x^{ p + 1} = z$ ( We choose the brane along $x^1, \cdots, x^{p + 1}$-directions). All the fields for a static BPS $(p + 1)$-brane are usually independent of the brane coordinates. So we have one isometry already along the $z$-direction. Thus a BPS $p$-brane solution in $D = 9$ can be obtained simply by the so-called diagonal or double dimensional reduction \cite{dufhis} on the BPS $(p + 1)$-brane solution of type IIA supergravity. The two Einstein metrics are now simply related to each other by \begin{equation} d s^2_{10}~ ({\rm type~ IIA}) = e^{-\varphi_A/(2\sqrt{7})}ds_9^2 + e^{\sqrt{7} \varphi_A/2} d z^2, \label{eq:mriia} \end{equation} where $\varphi_A$ is another dilatonic scalar due to this dimensional reduction. The IIA dilaton $\phi_A$ remains unchanged. The $n$-form field strength $F_n$ is reduced according to the following: if it carries an electric charge, it will become an $(n-1)$-form in $D = 9$, i.e., $F_n \rightarrow F^{(1)}_{n -1} \wedge dz$, whereas it will remain the same if it carries a magnetic charge, just the opposite to the case of vertical reduction. Here we do not really change anything. It is just the same whether we call the solution a $(p + 1)$-brane in $D = 10$ or a $p$-brane in $D = 9$. In other words, if we know a $p$-brane solution in $D = 9$, we should obtain a $(p + 1)$-brane solution in $D = 10$ right away and vice-versa. The process of obtaining a $(p + 1)$-brane in $D = 10$ from a $p$-brane in $D = 9$ is called the diagonal oxidation. The $p$-brane solution in $D = 9$ obtained from a known type IIB $p$-brane solution is described in terms of fields in type IIB basis while in obtaining a $(p + 1)$-brane in type IIA, we need a $p$-brane solution in $D = 9$ described in terms of the fields in type IIA basis. So we need to map the fields in type IIB basis to fields in type IIA basis. This mapping is nothing but the T-duality transformation for the solution. This is also the T-duality transformation acting on the fields of $D = 9, N = 2$ supergravity discussed earlier. Then T-duality on field strengths in $D = 9, N = 2$ supergravity or the corresponding $p$-brane solutions is the identification $H_n = F_n$ with the understanding that if $H_n$ is defined with a Chern-Simons term so is $F_n$. The mapping for the $D = 10$ dilaton $\phi$ and the dilatonic scalar $\varphi$ between the two versions of $D = 9, N = 2$ supergravity is \begin{equation} \left(\begin{array}{c} \phi_A\\ \varphi_A \end{array}\right) = \left(\begin{array}{cc} \frac{3}{4}& - \frac{\sqrt{7}}{4}\\ - \frac{\sqrt{7}}{4} & - \frac{3}{4} \end{array}\right) \left(\begin{array}{c} \phi_B\\ \varphi_B \end{array}\right). \label{eq:tdvdr} \end{equation} The above relations have been given or can be deduced, for example, from the appendix in \cite{duflp} where the relations between the corresponding gauge potentials are also given\footnote{The precise relations between various fields in two different bases of the nine dimensional theory are given in a recent paper \cite{lupsone}. We would like to thank Chris Pope for letting us know their results prior to publication. These relations are also given in \cite{berho} but in a different conventions than we use here.}. Using the above relations, we can obtain a $p$-brane solution in $D = 9$ in type IIA basis from a $p$-brane solution in type IIB basis. Then we obtain a BPS $(p + 1)$-brane solution in type IIA by oxidizing this $p$-brane solution back to $D = 10$ as described above. If we reverse the above process, we then obtain a BPS $p$-brane solution in one theory from a known BPS $(p + 1)$-brane solution in the other theory now by a T-duality along a world volume direction. The oxidation involved in this process is the so-called vertical one\cite{lups}. Now we conclude this section by explaining why we cannot obtain D8 brane solution in the massive type IIA supergravity from a D7 solution in the type IIB supergravity by a T-duality transformation along one of its transverse directions. First, it is not clear even now whether the massive type IIA supergravity has its origin in any of the known superstring theories or even in M-theory (though we believe so). So it is not clear that there exists a theory from which the massive type IIA supergravity originates and is T-dual to the type IIB superstring theory. Even if we have such a theory, the massless fields in $D = 9$ from the $S^1$-compactified massive type IIA supergravity cannot describe the 8-brane in $D = 9$. Therefore, we do not expect that the D8-brane solution is related to a D7 brane solution in the type IIB supergravity by a T-duality transformation\footnote{We merely mean this at the level of supergravity solutions. As we know, in string theory, D8 brane does seem to be related to D7 brane by a T-duality.}. Technically, as explained in \cite{lups}, the vertical reduction does not seem to apply when $p = 7$. By the same token, we cannot obtain (F, D8) solution from (F, D7) solution by a T-duality transformation. \section{(F, D$_p$) Bound States\protect\\} \label{sec:fdpbs} In this section, we will construct explicitly the (F, Dp) bound state configurations for $ 2 \le p \le 7$ from the known (F, D1) bound state configuration of Schwarz \cite{schone} by a T-duality transformation along one of the transverse directions described in the previous section. We will give a detail construction for the case of (F, D2) as an example and present only the results for the rest. Once we have (F, Dp) configurations for $ 1\le p \le 7$, we will also present the results for (W, Dp) for $0 \le p \le 6$, i.e., a Dp brane carrying waves in it, by a T-duality along the longitudinal direction of F-strings. All these configurations preserve one half of the spacetime supersymmetries. The field configurations of the $(m,n)$-string of Schwarz (or (F, D1) bound state), with $m, n$ relatively prime integers, in type IIB theory \cite{schone} are given in terms of our notations as: the Einstein metric, \begin{equation} ds^2~ ({\rm type ~IIB}) = H^{-3/4} [ - (d x^0)^2 + (d x^1)^2 ] + H^{1/4} dy^i dy^i, \label{eq:fd1m} \end{equation} with $i = 1, \cdots, 8$; the type IIB dilaton, \begin{equation} e^{\phi_B} = e^{\phi_{B0}}\, H' H^{-1/2}, \label{eq:iibd} \end{equation} the axion, \begin{equation} \chi_B = \frac{mn (H - 1) + \chi_{B0} \,\Delta_{(m, n)} \,e^{\phi_{B0}}} { n^2 H + (m - \chi_{B0} n)^2 \,e^{2\phi_{B0}}}, \label{eq:iiba} \end{equation} and the NSNS 3-form field strength $H_3 ~({\rm NSNS})$ and RR 3-form $H_3~({\rm RR})$, \begin{eqnarray} H_3~ ({\rm NSNS}) &=& - \Delta_{(m, n)}^{-1/2} \,e^{\phi_{B0}} \,(m - \chi_{B0} n) d H^{-1} \wedge d x^0 \wedge d x^1,\nonumber\\ H_3~ ({\rm RR})& = &\Delta_{(m, n)}^{-1/2} \left[ \chi_{B0}\, (m - \chi_{B0} n)\, e^{\phi_{B0}} - n \,e^{-\phi_{B0}}\right] d H^{-1}\wedge d x^0 \wedge d x^1. \label{eq:3f} \end{eqnarray} In the above, $\phi_{B0}$ and $\chi_{B0}$ represent the asymptotic values of the type IIB dilaton and axion, respectively. The SL(2, Z) invariant $\Delta$-factor is \begin{equation} \Delta_{(m, n)} = (m - \chi_{B0} n)^2 e^{\phi_{B0}} + n^2 e^{-\phi_{B0}}, \label{eq:df} \end{equation} and the SL(2, Z) invariant Harmonic function $H$ is \begin{equation} H = 1 + \frac{Q_1}{r^6}, \label{eq:hf1} \end{equation} where the radial distance $r^2 = (y^1)^2 + (y^2)^2 + \cdots + (y^8)^2$ and the quantized central charge $Q_1$ is \begin{equation} Q_1 = 2^5 \pi^2 \alpha'^3 \Delta_{(m, n)}^{1/2}, \label{eq:cc1} \end{equation} with $\alpha'$ the string constant. We have also introduced a second Harmonic function \begin{equation} H' = \frac{ (m - \chi_{B0} n)^2 e^{\phi_{B0}} + n^2 H e^{-\phi_{B0}}} {\Delta_{(m, n)}}, \label{eq:nhf2} \end{equation} which approaches unity as $r \rightarrow \infty$. Before we move on to the constructions of (F, Dp) bound states, we fix a few conventions. The RR-charge of the Dp-brane in (F, Dp) is defined \cite{dufkl} as, \begin{equation} e_p = \frac{1}{\sqrt{2} \kappa_0} \int_{S^{8 - p}} e^{ - a(p) \phi} \ast G_{p +2}, \label{eq:ecd} \end{equation} for Noether ``electric" charge, and \begin{equation} g_p = \frac{1}{\sqrt{2} \kappa_0} \int_{S^{p + 2}} G_{p + 2}, \label{eq:mcd} \end{equation} for topological or magnetic-like charge. In the above $\sqrt{2} \kappa_0 = (2\pi)^{7/2} \alpha'^2$. For a Noether charge, the integration is over an asymptotic $(8 - p)$-sphere surrounding the Dp-brane while for a magnetic-like charge, the integration is over an asymptotic $(p + 2)$-sphere surrounding the Dp-brane. Also, $G_{p + 2} = H_{p + 2}$ for type IIB theory while $G_{p + 2} = F_{p + 2}$ for type IIA theory. The constant $a (p) = (p - 3)/2$ for Dp-branes in both type IIA and type IIB theories. The above definitions are also valid for NSNS branes, i.e., NSNS strings and fivebranes, but with $a (p) = - (p - 3)/2$. However, the charge associated with the F-strings in the (F, Dp) bound states cannot be calculated using the above simple formula. We will discuss this in the next section. In this paper, $\ast$ always denotes the Hodge dual. $\epsilon_n$ denotes the volume form on an $n$-sphere where the volume of a unit $n$-sphere is \begin{equation} \Omega_n = \frac{2 \pi^{(n + 1)/2}}{\Gamma ((n + 1)/2)}, \label{eq:usv} \end{equation} and the unit charge for a Dp-brane is given as, \begin{equation} Q_0^p \equiv (2\pi)^{(7 - 2p)/2} \alpha'^{(3 - p)/2}. \label{eq:puc} \end{equation} {\bf (F, D2) Bound State}: Now in order to construct this bound state we first need to construct, as described in the previous section, the solution corresponding to an infinite periodic array of $(m,n)$-strings along $z = y^8$ axis. The Harmonic function in that case would be given as, \begin{equation} H = 1 + Q_1 \sum_{n = -\infty}^\infty \frac{1}{\mid \vec{y} - 2\pi n a \hat{z}\mid^6}, \label{eq:mcs1} \end{equation} where $\hat{z}$ denotes the unit vector and $a = \sqrt{\alpha'}$. In the limit $\alpha' \rightarrow 0$, the summation in the above Harmonic function can be replaced by an integration, i.e., \begin{equation} \sum_{n = -\infty}^\infty \rightarrow \int_{-\infty}^\infty \frac{dz}{2\pi a}. \label{eq:sir} \end{equation} Note that writing $\vec{y} = \tilde{\vec{y}} + z \hat{z}$ and so, $\mid \vec{y} - 2\pi na \hat{z} \mid^6 = [ {\tilde{r}}^2 + (z - 2\pi na)^2]^3$, with ${\tilde r}^2 = (y^1)^2 + \cdots + (y^7)^2$, we have, \begin{equation} \sum_{n = -\infty}^\infty \frac{1}{\mid \vec{y} - 2 \pi n a \hat{z}\mid^6} \rightarrow \int_{-\infty}^\infty \frac{dz}{2\pi a} \frac{1} {({\tilde r}^2 + z^2)^3} = \frac{3}{16 a} \frac{1}{{\tilde r}^5}. \label{eq:ir1} \end{equation} So the new Harmonic function describing the infinite array of $(m,n)$-strings along the $z$-direction is \begin{equation} H = 1 + \frac{Q_2}{r^5}, \label{eq:nh2} \end{equation} where we have dropped the `tilde' above the radial distance and will continue to do so from now on. The central charge $Q_2$ in (3.16) is \begin{equation} Q_2 = \frac{3 Q_1}{16 \alpha'^{1/2}} = 6 \pi^2 \alpha'^{5/2} \Delta_{(m, n)}^{1/2}, \label{eq:cc2} \end{equation} The Harmonic function is now independent of the $z$-coordinate. In other words, the Einstein metric (Eq.\ (\ref{eq:fd1m})) expressed in terms of this Harmonic function possesses one additional isometry along the $z$-direction. If we now compactify the metric along $z$-direction, Eq.\ (\ref{eq:mriib}) will give \begin{equation} e^{\sqrt{7} \varphi_B /2} = H^{1/4}, \label{eq:vd2} \end{equation} and the 9-d Einstein metric will be given as, \begin{equation} d s^2_9 = H^{- 5/7} [ - (d x^0)^2 + (d x^1)^2 ] + H^{2/7} dy^i dy^i, \label{eq:m9} \end{equation} where $i = 1, 2, \cdots, 7$. For the 9-dimensional solution, the type IIB dilaton, axion and NSNS and RR 3-forms remain unchanged with the replacement of the old Harmonic function by the new one. We have therefore obtained $(m,n)$-string bound state in $D = 9$ in the type IIB basis. We now express all the relevant fields in type IIA basis. The $d s_9^2$ remains unchanged. $F_3 = H_3 ~({\rm NSNS})$, $F_3^{(1)} = H_3 ~({\rm RR})$ and $F_1^{(1)} = d \chi$. The type IIA dilaton $\phi_A$ and the dilatonic scalar $\varphi_A$ can be obtained from $\phi_B$ and $\varphi_B$ through Eq.\ (\ref{eq:tdvdr}) as \begin{eqnarray} e^{\phi_A} &=& e^{3\phi_{B0}/4} H'^{3/4} H^{- 1/2},\nonumber\\ e^{\sqrt{7} \varphi_A} &=& e^{-7 \phi_{B0}/4} H'^{- 7/4} H^{1/2}, \label{eq:iiadvd} \end{eqnarray} where the Harmonic function $H'$ continues to be given by the expression \ (\ref{eq:nhf2}) but in terms of the present Harmonic function $H$. Once we express $D = 9$ (F, D1) configuration in the type IIA basis, we can read the (F, D2) configuration in type IIA theory using the diagonal oxidation described in the previous section. For example, the Einstein metric in $D = 10$ for this configuration can be read from Eq.\ (\ref{eq:mriia}) with $d s^2_9$, $\phi_A$ and $\varphi_A$ given in (3.19) and (3.20). We here collect the complete results for the (F, D2) configuration for the metric, \begin{equation} d s^2_{10}~ ({\rm type~ IIA}) = e^{\phi_{B0}/8} H'^{1/8} H^{1/4} \left[ H^{-1} (- (d x^0)^2 + (d x^1)^2 ) + e^{ - \phi_{B0}} H'^{ - 1} (dx^2)^2 + d y^i d y^i\right], \label{eq:iiadtwom} \end{equation} for the dilaton, \begin{equation} e^{\phi_A} = e^{3\phi_{B0}/4} H'^{3/4} H^{-1/2}, \label{eq:iiadtwod} \end{equation} and for the remaining non-vanishing fields\footnote{In type IIA theory, $F_4' = dA_3 + A_1\wedge F_3$}, \begin{eqnarray} F_2 &=& n \Delta_{(m,n)}^{-1} ~e^{-\phi_{B0}} (m -\chi_{B0} n)~ H'^{-2} ~d H\wedge dx^2,\nonumber\\ F_3 &=& - \Delta_{(m,n)}^{-1/2} ~e^{\phi_{B0}} ~(m - \chi_{B0} n) ~d H^{-1}\wedge d x^0\wedge d x^1,\nonumber\\ F_4' &=& n \,e^{- 3 \phi_{B0}/8}\, H'^{-3/8} \, H^{1/4}\, \frac{\sqrt{2} \kappa_0 Q_0^2} {\Omega_5} \,\ast \epsilon_6. \label{eq:tstwo} \end{eqnarray} In the above, $i = 1, 2, \cdots 7$, the Harmonic functions $H'$ and $H$ are given by Eqs.\ (\ref{eq:nhf2}) and (\ref{eq:nh2}), respectively. {\bf (F, D3) Bound State}\footnote{This configuration with zero asymptotic values of $\phi_{B0}$ and $\chi_{B0}$ was also given in \cite{rust} for different purpose. Some non-threshold bound states of M theory, not considered here, were also discussed there.}: Once we obtain (F, D2), we can repeat the above process to obtain (F, D3) bound state configuration. The results are for the Einstein metric, \begin{eqnarray} ds^2_{10}~ ({\rm type ~IIB}) = &&e^{\phi_{B0}/4} \,H'^{1/4}\, H^{1/4} [ H^{-1} ( - (d x^0)^2 + (d x^1)^2) \nonumber\\ && + e^{-\phi_{B0}}\, H'^{-1} ( (d x^2)^2 + (d x^3)^2) + d y^i d y^i], \label{eq:miibthree} \end{eqnarray} with $i = 1, 2, \cdots, 6$; for the type IIB dilaton, \begin{equation} e^{\phi_B} = e^{\phi_{B0}/2}\, H'^{1/2}\, H^{-1/2}, \label{eq:iibdthree} \end{equation} and for the remaining non-vanishing fields, \begin{eqnarray} H_3 ~({\rm NSNS}) &=& - \Delta_{(m,n)}^{-1/2} \,e^{\phi_{B0}} \,(m - \chi_{B0} n) \, d H^{-1}\wedge d x^0 \wedge d x^1,\nonumber\\ H_3 ~({\rm RR}) &=& n \,\Delta_{(m,n)}^{-1}\, e^{-\phi_{B0}} (m - \chi_{B0} n) H'^{-2 } \,d H \wedge d x^2 \wedge d x^3,\nonumber\\ H_5 &= & n \,\frac{\sqrt{2} \kappa_0 Q_0^3 }{\Omega_5} \,(\ast \epsilon_5 + \epsilon_5). \label{eq:iibthreerf} \end{eqnarray} In the above, the Harmonic function $H'$ continues to be given by Eq.\ (\ref{eq:nhf2}) but the Harmonic function $H$ is now \begin{equation} H = 1 + \frac{Q_3}{r^4}, \label{eq:hfthree} \end{equation} where $Q_3 = \Delta_{(m,n)}^{1/2} \sqrt{2} \kappa_0 Q_0^3 /(4 \Omega_5)$. All the other quantities have already been defined. {\bf (F, D4) Bound State}\footnote{The classical configuration of this bound state was also given in \cite{grelpt}, obtained by dimensional reductions from the (M2, M5) bound state in D = 11.}: Repeating the same procedure the various field configuration for this solution are: the metric, \begin{eqnarray} d s^2_{10}~ ({\rm type~ IIA}) =&& e^{3 \phi_{B0}/8}\, H'^{3/8} \, H^{1/4}\, [ H^{-1}\, (- (d x^0)^2 + (d x^1)^2) \nonumber\\ && + e^{-\phi_{B0}}\, H'^{-1} \,( (d x^2)^2 + (d x^3)^2 + (d x^4)^2) + d y^i d y^i], \label{eq:iiamfour} \end{eqnarray} with $i = 1, 2, \cdots, 5$; the type IIA dilaton, \begin{equation} e^{\phi_A} = e^{\phi_{B0}/4}\, H'^{1/4} \,H^{- 1/2}, \label{eq:iiadfour} \end{equation} and the remaining non-vanishing fields, \begin{eqnarray} F_3 &= &- \Delta_{(m,n)}^{-1/2} \, e^{\phi_{B0}}\, (m - \chi_{B0} n)\, d H^{-1} \wedge d x^0 \wedge d x^1,\nonumber\\ F_4' &=& n \frac{\sqrt{2} \kappa_0 Q_0^4}{\Omega_4} \,\epsilon_4 + n \, \Delta_{(m,n)}^{-1} \,e^{-\phi_{B0}}\, (m - \chi_{B0} n) \,H'^{-2}\, d H \wedge d x^2 \wedge d x^3 \wedge d x^4. \label{eq:iiarffour} \end{eqnarray} In the above, the Harmonic function $H'$ is as given by Eq.\ (\ref{eq:nhf2}) but the Harmonic function $H$ is now \begin{equation} H = 1 + \frac{Q_4}{r^3}, \label{eq:iiahffour} \end{equation} where $Q_4 = \Delta_{(m,n)}^{1/2} \sqrt{2} \kappa_0 Q_0^4 /(3 \Omega_4)$. {\bf (F, D5) Bound State}: The field configurations for this bound state are: the Einstein metric, \begin{eqnarray} d s^2_{10} ~({\rm type~ IIB}) =&& e^{\phi_{B0}/2}\, H'^{1/2} \, H^{1/4} \, [ H^{-1} (- (d x^0)^2 + (d x^1)^2) \nonumber\\ &&+ e^{-\phi_{B0}}\, H'^{-1} ( (d x^2)^2 + (d x^3)^2 + (d x^4)^2 + (d x^5)^2) + d y^i d y^i], \label{eq:iibmfive} \end{eqnarray} with $i = 1, 2, 3, 4$; the type IIB dilaton, \begin{equation} e^{\phi_B} = H^{- 1/2}, \label{eq:iibdfive} \end{equation} and the remaining non-vanishing fields, \begin{eqnarray} H_3 ~({\rm NSNS}) &=& - \Delta_{(m,n)}^{-1/2}\, e^{\phi_{B0}}\, (m - \chi_{B0} n)\, d H^{-1}\wedge d x^0 \wedge d x^1, \nonumber\\ H_3 ~({\rm RR}) & = & n \,\frac{\sqrt{2} \kappa_0 Q_0^5}{\Omega_3}\, \epsilon_3,\nonumber\\ H_5 &=& n \,\Delta_{(m,n)}^{-1} \,e^{-\phi_{B0}} \,(m - \chi_{B0} n)\, H'^{-2} \,d H \wedge d x^2 \wedge d x^3 \wedge d x^4 \wedge d x^5. \label{eq:iibrffive} \end{eqnarray} Again the Harmonic function $H'$ remains the same as given in Eq.\ (\ref{eq:nhf2}) but the Harmonic function $H$ is now \begin{equation} H = 1 + \frac{Q_5}{r^2}, \label{eq:iibhffive} \end{equation} where $Q_5 = \Delta_{(m,n)}^{1/2} \sqrt{2} \kappa_0 Q_0^5 /(2 \Omega_3)$. {\bf (F, D6) Bound State}\footnote{The classical configuration of this bound state was also given in \cite{cosp}.}: In this case we have the Einstein metric, \begin{eqnarray} d s^2_{10} ~({\rm type ~IIA}) = &&e^{5 \phi_{B0}/8}\, H'^{5/8} \,H^{1/4} \, [ H^{-1} (- (d x^0)^2 + (d x^1)^2) \nonumber\\ &&+ e^{-\phi_{B0}} H'^{-1} ( (d x^2)^2 + \cdots + (d x^6)^2) + d y^i d y^i], \label{eq:iiamsix} \end{eqnarray} with $i = 1, 2, 3$; the type IIA dilaton, \begin{equation} e^{\phi_A} = e^{ - \phi_{B0}/4} \,H'^{- 1/4}\, H^{- 1/2}, \label{eq:iiadsix} \end{equation} and the remaining non-vanishing fields, \begin{eqnarray} F_2 &=& n \,\frac{\sqrt{2} \kappa_0 Q_0^6}{\Omega_2}\, \epsilon_2,\nonumber\\ F_3 &=&- \Delta_{(m,n)}^{-1/2} \,e^{\phi_{B0}}\, (m - \chi_{B0} n) \,d H^{-1}\wedge d x^0 \wedge d x^1, \nonumber\\ F_4' &=& - n \,\Delta_{(m,n)}^{- 1/2} \,e^{\phi_{B0}}\, (m - \chi_{B0} n)\, H^{-1}\, \frac{\sqrt{2} \kappa_0 Q_0^6}{\Omega_2}\, d x^0 \wedge d x^1 \wedge \epsilon_2, \label{eq:iiarfsix} \end{eqnarray} Once again the Harmonic function $H'$ continues to be given by Eq.\ (\ref{eq:nhf2}) but the Harmonic function $H$ is \begin{equation} H = 1 + \frac{Q_6}{r}, \label{eq:iiahfsix} \end{equation} where $Q_6 = \Delta_{(m,n)}^{1/2} \sqrt{2} \kappa_0 Q_0^6 /\Omega_2$. {\bf (F, D7) Bound State}: The various field configurations in this case are described by the following Einstein metric, \begin{eqnarray} d s^2_{10}~ ({\rm type~ IIB}) =&& e^{3 \phi_{B0}/4}\, H'^{3/4}\, H^{1/4} \, [ H^{-1} (- (d x^0)^2 + (d x^1)^2)\nonumber\\ && + e^{-\phi_{B0}} \, H'^{-1} ( (d x^2)^2 + (d x^3)^2 + \cdots + (d x^7)^2) + d y^i d y^i], \label{eq:iibmseven} \end{eqnarray} with $i = 1, 2$; the type IIB dilaton, \begin{equation} e^{\phi_B} = e^{- \phi_{B0}/2}\, H'^{ -1/2} \,H^{- 1/2}, \label{eq:iibdseven} \end{equation} and the remaining non-vanishing fields, \begin{eqnarray} d \chi &=& n \,\frac{\sqrt{2} \kappa_0 Q_0^7}{\Omega_1}\, \epsilon_1,\nonumber\\ H_3 ~({\rm RR}) &=& - n \,\Delta_{(m,n)}^{- 1/2}\, e^{\phi_{B0}}\, (m - \chi_{B0} n)\, H^{-1}\, \frac{\sqrt{2} \kappa_0 Q_0^7}{\Omega_1} \,d x^0 \wedge d x^1 \wedge \epsilon_1, \nonumber\\ H_3 ~({\rm NSNS}) & = & - \Delta_{(m,n)}^{-1/2}\, e^{\phi_{B0}}\, (m - \chi_{B0} n)\, d H^{-1}\wedge d x^0 \wedge d x^1. \label{eq:iibrfseven} \end{eqnarray} Here the Harmonic function $H'$ is again given by Eq.\ (\ref{eq:nhf2}) but the Harmonic function $H$ is now \begin{equation} H = 1 - Q_7 ~{\rm ln} r, \label{eq:iibhfseven} \end{equation} where $Q_7 = \Delta_{(m,n)}^{1/2}\sqrt{2} \kappa_0 Q_0^7 /\Omega_1$. A detail discussion of the meaning and the other properties of the above solutions will be given in the following section. We, however, continue this section to briefly indicate how to construct (W, Dp) bound states in type II theories from the (F, Dp) bound states already constructed. So far, we have performed T-duality only along the transverse directions of the F-strings to obtain (F, Dp) bound states from the known (F, D1) bound state. But we note that following the prescription given in the previous section, we can also T-dualize the above (F, Dp) solutions for $1\le p \le 7$ along $x^1$-direction of the F-strings. This will give (W, Dp) bound state solutions for $0\le p \le 6$. We here present the Einstein metric, the dilaton and the Kaluza-Klein vector potential for each of these solutions. The remaining non-vanishing fields can be obtained easily from the corresponding ones of (F, Dp). For the (W, Dp) solutions we have the following Einstein metric, \begin{eqnarray} d s^2_{10} =&& e^{(p + 1) \phi_{B0}/8} \,H'^{( p + 1)/8}\, [ - H^{ -1} (d x^0)^2 + e^{-\phi_{B0}}\, H'^{-1} \, H (d x^1 + {\cal A}_0 d x^0)^2 \nonumber\\ &&+ e^{-\phi_{B0}}\, H'^{ -1} ( (d x^2)^2 + \cdots + (d x^p)^2) + d y^i dy^i], \label{eq:gwdpm} \end{eqnarray} with $i = 1, 2, \cdots, (9 - p)$; the dilaton, \begin{equation} e^\phi = e^{(3 - p)\phi_{B0}/4} \,H'^{(3 - p)/4}, \label{eq:gwdpd} \end{equation} and the Kaluza-Klein vector potential \begin{equation} {\cal A}_0 = - \Delta_{(m,n)}^{- 1/2}\, e^{\phi_{B0}} (m - \chi_{B0} n)\, H^{ - 1}. \label{eq:gwdpv} \end{equation} In the above, the Harmonic function $H$ is given by, \begin{equation} H = \left\{\begin{array}{ll} 1 + \frac{Q_{p + 1}}{r^{6 - p}},& 0 \le p \le 5,\\ 1 - Q_7~ \mbox{ln}~r, & p = 6, \end{array} \right. \end{equation} and the Harmonic function $H'$ continues to be given by Eq.\ (\ref{eq:nhf2}). All the other quantities are already given before. Note that for $m = 0$ (also $\chi_{B0} = 0$ and for simplicity we also set $\phi_{B0} = 0$\footnote{We will discuss this in the following section.}), we have $H' = H$ and ${\cal A}_0 = 0$. Then the above solution represents an infinite periodic array of Dp-branes along the $x^1$-axis. While for $n = 0$, $H' = 1$, and the above solution represents a gravitational wave propagating in $x^1$-direction with isometries along $(x^1, x^2, \cdots, x^p)$-directions. We will not discuss the (W, Dp) bound states any further in this paper and focus on (F, Dp) ones for the rest. \section{Properties of (F, D$_p$) Bound States\protect\\} \label{sec:pfdp} The worldvolume picture for a (F, Dp) bound state described in \cite{lurone} consists of constant electric field lines flowing along, say, $x^1$-axis in the Dp-brane worldvolume. This picture applies for each of our spacetime configurations of (F, Dp) given in the previous section. In order to see that, one can examine that for $m = 0$ (with $\chi_{B0} = 0$ and $\phi_{B0} = 0$ for simplicity), the metric describes Dp-branes lying along ($x^1, x^2, \cdots, x^p$)-directions while for $n = 0$ (hence $H' = 1$), it describes F-strings lying along the $x^1$-direction with isometries along $(x^1,x^2, \cdots, x^p$)-directions. So, this configuration indeed describes F-strings within the D-branes in consistency with our worldvolume picture. In order to make such an identification more precise, we need to calculate the charges carried by the Dp-brane and by the F-strings, and also the mass per unit $p$-brane worldvolume, and compare them with what we obtained in \cite{lurone} based on the worldvolume approach. Let us first calculate the RR charge for the Dp-brane in a (F, Dp) bound state configuration given in the previous section. This can be done easily using either Eq.(3.9) or Eq.(3.10) depending on whether the charge is electric-like or magnetic-like with the corresponding explicit $(p + 2)$-form field strength given in the previous section. We have $e_p = n\, Q_0^p$ for an electric-like RR charge or $g_p = n\, Q_0^p$ for a magnetic-like RR charge. The integer $n$ originates from that of the D-string in the (F, D1) bound state. So the RR charge is automatically quantized given the quantization of D-string charge, a well-known fact that the charge quantization for one extended object will imply charge quantizations for the rest of the extended objects in string/M theory. This is consistent with our worldvolume result. Note that $Q_0^p \,Q_0^{6 - p} = 2 \pi$. As hinted earlier, Eq.\ (\ref{eq:ecd}) cannot be applied simply to calculate the electric-like charge associated with the F-strings in (F, Dp). The reason is, as explained in \cite{lurone}, that the notation `F' in (F, Dp) means actually an infinite number of parallel strings in the bound state. As mentioned in \cite{lurone}, there is one NS-string (or equivalently $m$ F-strings) per $(2\pi)^{p -1} \alpha'^{(p - 1)/2}$ of ($p - 1$)-dimensional area. We will show later in this section that this is indeed true. In 1 + 3 dimensional electrostatics, we know that in order to use Gauss law to obtain the charge per unit length for a uniform line distribution of charge, we have to choose a cylinder (with the line charge at the center) as the Gauss surface rather than a 2-sphere as for a point charge. For the present case, we therefore should choose the integration in Eq.\ (\ref{eq:ecd}) over $R^{p -1} \times S^{8 - p}$ rather than over an asymptotic $S^7$. We can therefore have charge per unit $(p -1)$-dimensional area for the F-strings. In order to give meaningful calculations for this quantity and for the mass per unit Dp-brane worldvolume which will be used to determine the tension for a (F, Dp) bound state, we need a good asymptotic behavior for the metric such that the $(p - 1)$-dimensional area and the worldvolume can be defined with respect to certain frame metric. This is also necessary for us to make comparisons with the results obtained in \cite{lurone} based on the worldvolume analysis. Let us take a close look at each Einstein frame metric for those (F, Dp) bound state configurations. They, along with the corresponding dilatons, can actually be expressed in a unified way as \begin{eqnarray} d s^2 = &&e^{(p - 1) \phi_{B0}/8}\, H'^{(p - 1)/8}\, H^{1/4}\, [ H^{ -1} (- (d x^0)^2 + (d x^1)^2) \nonumber\\ &&+ e^{- \phi_{B0}}\, H'^{-1} \,( (d x^2)^2 + \cdots + (d x^p)^2) + d y^i d y^i], \label{eq:gem} \end{eqnarray} for the metric with $i = 1, 2, \cdots, 9 - p$; and \begin{equation} e^\phi = e^{(5 - p)\phi_{B0}/4} \,H'^{(5 - p)/4} \, H^{ - 1/2}. \label{eq:gd} \end{equation} for the dilaton. In the above, the Harmonic function $H$ is \begin{equation} H = 1 + \frac{Q_p}{r^{7 - p}}, \label{eq:ghf} \end{equation} with \begin{equation} Q_p = \frac{\Delta_{(m, n)}^{1/2} \sqrt{2} \kappa_0 Q_0^p} {(7 - p) \Omega_{8 - p}}, \label{eq:gcc} \end{equation} and the Harmonic function $H'$ is given by Eq.\ (\ref{eq:nhf2}). Note first that the constants $\phi_{B0}$ and $\chi_{B0}$ appearing in all the above solutions are no longer the asymptotic values for the dilaton and the axion except for the original (F, D1) bound state configuration. Actually, the axion for $p = 3, 5, 7$ has nothing to do with the constant $\chi_{B0}$. In fact, $\chi_{B0}$ is the asymptotic value of the $(p - 1)$-form RR gauge potential in the corresponding (F, Dp) configuration. The asymptotic value of dilaton for every $p$ except for $p = 5$ is related to the original $\phi_{B0}$ but not equal. This can be understood from the fact that a T-duality transformation shifts the dilaton value\cite{bus}. From the dilaton expression for each of the above solutions, we can see that the asymptotic value of the dilaton is reduced by a quarter of $\phi_{B0}$ for every T-duality transformation. For $p = 5$, the asymptotic value vanishes. Second, even though we start with an Einstein metric for (F, D1) which is asymptotically Minkowski, none of the metrics derived by T-duality transformations remains to be so. Even worse, none of them remains asymptotically as a constant scaling factor times Minskowski metric. However, there exists a constraint on the asymptotic behavior of the Einstein metric for the (F, D1) bound state configuration which preserves not only the asymptotic behavior for each of the derived Einstein metrics but also the asymptotic value for the dilatons\footnote{This remains true also for the (W, Dp) bound states.}. This turns out to be the one which imposes the corresponding string metric rather than the Einstein metric to be asymptotically Minkowski. This constraint has been used in the literature and once again we see its significance. We will discuss possible reasons behind this and other related issues in a separate paper. Once we make such a choice for the asymptotic metric, Eq.\ (\ref{eq:gem}) and (\ref{eq:gd}) become \begin{eqnarray} d s^2 = && e^{- \phi_{B0}/2} \,H'^{(p - 1)/8} \,H^{1/4}\,\left[ H^{ -1} (- (d x^0)^2 + (dx^1)^2)\right. \nonumber\\ && \left. + H'^{-1} ( (d x^2)^2 + \cdots + (d x^p)^2) + d y^i d y^i \right], \label{eq:ngem} \end{eqnarray} and \begin{equation} e^\phi = e^{\phi_{B0}}\, H'^{(5 - p)/4}\, H^{ - 1/2}. \label{eq:ngd} \end{equation} The Harmonic function $H$ is given by Eq.\ (\ref{eq:ghf}) but with $Q_p \rightarrow e^{3 \phi_{B0}/2} Q_p$ with $Q_p$ given in Eq.\ (\ref{eq:gcc}). The expressions (4.5) and (4.6) have the above mentioned properties. For example, the asymptotic value for the dilaton is always $\phi_{B0}$, the one in the original (F, D1) bound state. In what follows, we simply use $\phi_0$ rather than $\phi_{B0}$ as the asymptotic value for the dilaton. We give below the explicit form of NSNS 3-from field strength\footnote{$G_3 = F_3$ in type IIA while $G_3 = H_3 ~({\rm NSNS})$ in type IIB.} $G_3$ for the purpose of calculating the charge per unit $(p-1)$-dimensional area for F-strings which is \begin{equation} G_3 = - \Delta_{(m, n)}^{- 1/2} \,e^{\phi_{0}/2} \,(m - \chi_{B0} n) \, d H^{ -1}\wedge d x^0 \wedge d x^1. \label{eq:gthreef} \end{equation} As discussed above, the charge per unit $(p-1)$-dimensional area for the F-strings should be calculated as, \begin{equation} e_1 = \frac{1}{\sqrt{2} \kappa_0} \int_{R^{p - 1}\times S^{8 - p}} (e ^{ - \phi} \ast G_3 + \cdots), \label{eq:gcfs} \end{equation} where $\cdots$ indicates possible non-vanishing Chern-Simons terms. Actually, the Chern-Simons terms do contribute to this charge. The charge $e_1$ itself must be infinite since we have an infinite number of F-strings. But we expect that $e_1 / A_{p -1}$ should be finite with $A_{p - 1} = \int d x^2\wedge \cdots \wedge d x^p$ the coordinate $(p -1)$-dimensional area. Indeed, we find \begin{equation} |e_1|/A_{p -1} = m Q_0^p, \label{eq:gcpfs} \end{equation} As mentioned in \cite{lurone}, we have one NSNS-string (or $m$ F-strings) per $(2\pi)^{p -1} \alpha'^{(p -1)/2}$ of ($p -1$)-dimensional area from the worldvolume point of view. So, in order to compare with our interpretation for the F-strings along $x^1$-direction we must multiply the left side of the above equation with this $(p -1)$-dimensional area. Thus we obtain, \begin{equation} (2 \pi)^{p -1} \alpha'^{(p -1)/2} \frac{|e_1|}{ A_{p -1}} = \sqrt{2} \kappa_0 \,m T_f, \label{eq:conf} \end{equation} Now in order for the left hand side of Eq.(4.10) to represent the charge of F-strings per $(2\pi)^{p-1} \alpha'^{(p-1)/2}$ of $(p-1)$-dimensional area, we must replace the coordinate area $A_{p-1}$ by the one measured in string metric asymptotically. But, since we have chosen our string metric to be asymptotically flat, they are the same. Note also that according to our definition the charge $e_p$ of the $p$-brane is related to the corresponding tension $T_p$ as $e_p = {\sqrt 2} \kappa_0 T_p$ and therefore, Eq.(4.10) clearly states that there are indeed $m$ F strings in the (F, Dp) bound states per $(2\pi)^{p-1} \alpha'^{(p-1)/2}$ of $(p-1)$-dimensional area, confirming our interpretation. There is another way to obtain the above result which gives a direct link between the spacetime F-strings and the worldvolume electric flux lines. This will be important for our discussion of the $AdS_5/CFT_4$ correspondence given in section 5. The low-energy effective field theory on a Dp brane worldvolume contains a coupling, \begin{equation} \frac{1}{T^p_0 g} \int d^{p + 1} \sigma B_{\mu\nu} (2\pi \alpha' F^{\mu\nu}), \label{eq:coupling} \end{equation} where $B_2$ is the pull-back of the NSNS 2-form potential on the worldvolume, $F$ is the worldvolume gauge field strength\footnote{Strictly speaking, we should use ${\cal F} = F - B_2$ instead.}, $T_0^p = 1/[(2\pi)^p \alpha'^{(p + 1)/2}]$ is the $p$-brane tension units and $g = e^{\phi_0}$ is the string coupling. Because of this term the equation of motion takes the form (in terms of our conventions)\footnote{Here, we do not include an F-string source to the equation of motion since we are considering an infinite number of parallel F-strings in the bulk.} \begin{equation} d (e^{- \phi}\ast G_3 + \cdots ) = \frac{ 2 \,\kappa_0^2} { g \,T_f \,T^p_0} \ast F \wedge \delta^{9 - p}, \label{eq:ceom} \end{equation} where $\ast$ in $\ast G_3$ denotes the Hodge dual in $D = 10$ spacetime while the $\ast$ in $\ast F$ denotes the Hodge dual within the $(p + 1)$-dimensional worldvolume. In the above, $\delta^{9 - p}$ denotes a $(9 - p)$-form delta function on the Dp brane worldvolume. We use this equation for the purpose of relating the F-string charge to the electric flux of gauge field on the worldvolume. (Note that a similar equation has been used in \cite{str} for the charge conservation.) In order to get the charge of F-strings in (F, Dp) bound state, we should integrate the above equation on $R^{p - 1} \times R^{9 - p}$. We have \begin{eqnarray} \int_{R^{p - 1} \times R^{9 - p}} d (e^{- \phi} \ast G_3 + \cdots) &=& \int_{R^{p - 1} \times S^{8 - p}} (e^{-\phi} \ast G_3 + \cdots ),\nonumber\\ &=& \frac{ 2\, \kappa_0^2}{ g \,T_f \,T^p_0} \int_{R^{ p - 1}} \ast F. \label{eq: iceom} \end{eqnarray} This gives, \begin{eqnarray} e_1 &=& \frac{1}{\sqrt{2} \kappa_0} \int_{R^{p - 1}\times S^{8 - p}} (e ^{ - \phi} \ast G_3 + \cdots), \nonumber\\ &=& \frac{ \sqrt{2} \kappa_0}{ g T_f T^p_0} F_{01} A_{p - 1}, \label{eq:cfr} \end{eqnarray} where we have taken $F_{01}$ as the only non-vanishing constant component of the $U(1)$ gauge field strength on the worldvolume of Dp-brane. If we denote $Q_1$ as the charge of F-strings per $(2\pi)^{p -1} \alpha'^{(p -1)/2}$ area, we have \begin{equation} \frac{Q_1}{\sqrt{2}\, \kappa_0} = F_{01} / g, \label{eq:cfrone} \end{equation} where $Q_1/\sqrt{2} \kappa_0$ is the total tension of these F-strings. So we establish the precise relationship between the F-strings and the electric flux lines. If we take $Q_1/\sqrt{2} \kappa_0 = m T_f$ from the above calculated value, we have $F_{01} = g m T_f$ which agrees precisely with what we obtained in \cite{lurone}. Certainly, the other way around is also true. We now use the metric Eq.\ (\ref{eq:ngem}) to calculate the mass per unit $p$-brane volume for a (F, Dp) bound state and compare the result with what we obtained in \cite{lurone}. In doing so, we need to generalize the ADM formula in [18] to accommodate the following metric, \begin{equation} d s^2 = - A(r) dt^2 + B(r) d r^2 + r^2 C (r) d \Omega_{D - p - 2} + D (r) \delta_{ij} d x^i d x^j + E (r) \delta_{kl} d x^k d x^l, \label{eq:gm} \end{equation} where $D$ is the spacetime dimensions, $p$ is the spatial dimensions of a $p$-brane, and indices $(i, j = 1, 2, \cdots, m)$ and $(k, l = m + 1, \cdots, p)$. If $ A(\infty) = B(\infty) = C(\infty) = D (\infty) = E (\infty) = a_p$ (with $a_p$ a constant) [19], we have the ADM mass per unit $p$-brane volume as \begin{eqnarray} M_p = - \frac{\Omega_{\tilde{d} + 1}}{2 \kappa^2_0} [ && (\tilde{d} + 1) \,r^{\tilde{d} + 1}\partial_r C(r) + m \,r^{\tilde{d} + 1} \partial_r D (r) + (p - m) \,r^{\tilde{d} + 1} \partial_r E (r) \nonumber\\ &&- (\tilde{d} + 1)\, r^{\tilde{d}} \left(B (r) - C (r)\right) ]_{r \rightarrow \infty}, \label{eq:gadm} \end{eqnarray} with $\tilde{d} = D - p - 3$. Applying the above formula to our case, we get the mass per unit $p$-brane volume for a (F, Dp) bound state as \begin{equation} M_p = g \sqrt{n^2 + g^2 (m - \chi_{B0} n)^2} ~T^0_p, \label{eq:gmass} \end{equation} where we have used $Q_p = \Delta_{(m,n)}^{1/2} e^{3\phi_{0}/2} \sqrt{2}\kappa_0 Q_0^p/[(7 - p) \Omega_{8 - p}]$, the expressions for $Q_0^p$ from Eq.\ (\ref{eq:puc}) and the $\Delta$-factor from Eq.\ (\ref{eq:df}), respectively. Here $T^p_0 = 1/[(2\pi)^p \alpha'^{(p + 1)/2}]$ is the $p$-brane tension units and $g = e^{\phi_{0}}$ is the string coupling. We can relate the $M_p$ and the tension $T_p (m, n)$ of (F, Dp) in many ways. {}For example, they can be related by looking at the scaling behavior of the energy-momentum tensor due to the contribution of the Dp-brane worldvolume action under a constant rescaling of the spacetime metric. But we here use a very simple approach. We know that $M_p$ must be proportional to $T_p (m, n)$ and the proportionality constant must be some power of the constant $a_p$, i.e., $M_p = a_p^\alpha T_p (m, n)$ with $\alpha$ an as yet undetermined constant. Note that here $a_p = g^{- 1/2}$. When $m = 0, \chi_{B0} = 0$, we know $T_p (0, n) = n T^p_0/ g$ and from Eq.\ (\ref{eq:gadm}) we also have $M_p = g n T^p_0$. So we have $\alpha = - 4$ and we get $M_p = g^2 T_p (m, n) $ in general. Using this relation and Eq.(4.14) we find the tension $T_p (m, n)$ for the (F, Dp) bound state as \begin{equation} T_p (m, n) = \frac{1}{g} \sqrt{n^2 + (m - \chi_{B0} n)^2 g^2}~ T^p_0, \label{eq:gtension} \end{equation} which agrees precisely with what we obtained in \cite{lurone} for $\chi_{B0} = 0$. This also shows that (F, Dp) is a non-threshold bound state. \section{Decoupling Limit for the (F, D3) bound state\protect\\} \label{sec:ads/cft} The properties of the (F, Dp) bound states studied in the previous section are either the BPS ones or the consequences of BPS properties. Therefore, it is not surprising that we find agreements in both the spacetime and the worldvolume approaches even though, strictly speaking, a flat background is always assumed in the worldvolume study for these bound states in \cite{lurone}. These calculations clearly demonstrate that our spacetime bound state configurations (F, Dp) are identical to those bound states obtained from the worldvolume study in\cite{lurone}. The F-strings in (F, Dp) should be identified with the electric flux or field lines in the corresponding gauge theory living on the worldvolume. Now the question is: Can the same $AdS/CFT$ correspondences conjectured by Maldacena in \cite{mal} for $p = 3$ and by Itzhaki et al in \cite{itzmsy} for $p \neq 3$ be proposed in a similar fashion based on the (F, Dp) bound states rather than on the simple Dp branes? If so, can we learn anything new? We will examine these for the $p = 3$ case in the following. The analysis for $p \neq 3$ can be made similarly as for the $p = 3$ case along the line given in \cite{itzmsy}. In general, when the so-called D-branes appear in string theory, we need to consider not only the modes that propagate in the bulk but also the modes that propagate on the D-branes. The modes propagating on the D-branes are associated with open strings ending on the D-branes. In general, these modes interact not only among themselves but also with the modes propagating in the bulk. However, there exists a limit that decouples the modes propagating on the branes from the modes propagating in the bulk and is also typically a low energy limit. The latter says that the modes on the brane are just the massless ones of the open strings. In this limit, the D-brane theory becomes the corresponding SYM theory (for $p \le 3$). In the case of $p = 3$, the D3 brane theory is described by ${\cal N} = 4$ SYM in $1 + 3$ dimensions under this limit. But the D3 brane configuration is also described in terms of the metric and other fields in the bulk. The so-called decoupling or field theory limit is \begin{equation} g_{\rm YM}^2 = 2\pi g = {\rm fixed},~\qquad \alpha' \rightarrow 0, \label{eq:dcl} \end{equation} where $g = e^{\phi_0}$ is the string coupling constant and $g_{\rm YM}$ is the Yang-Mills coupling constant. Note that the ten dimensional Newton constant $\kappa^2 \sim g^2 \alpha'^4$ vanishes in this limit as it should be. Under this limit, the BPS D3-brane configuration in the bulk becomes flat Minkowski for any fixed non-zero $r$. Therefore, to have something nontrivial, we need to consider an additional limit, \begin{equation} U \equiv \frac{r}{\alpha'} = {\rm fixed},~\qquad \alpha' \rightarrow 0. \label{eq:esl} \end{equation} The above says that we keep the mass of the stretched strings between D3 branes fixed. This $U$ actually sets the energy scale in the field theory since it is the expectation value of the Higgs. This also implies that we are considering finite energy configurations in the field theory. As discussed in \cite{mal}, if we consider $n$ parallel D3 branes, then the above decoupling limits say that we are bringing the branes together but the Higgs expectation value corresponding to this separation remains fixed. The resulting theory on the brane is ${\cal N} = 4$ $U(n)$ SYM theory in $1 + 3$ dimensions. In what follows, we simply call Eqs.\ (\ref{eq:dcl}) and \ (\ref{eq:esl}) as the decoupling limit. Under the decoupling limit, the string metric describing the D3 branes becomes $AdS_5 \times S^5$. The isometry of $AdS_5$ is $SO(2, 4)$ and this is also the conformal group in $1 + 3$ dimensions. We also have isometries of $S^5$ as $SO(6)\sim SU(4)$. This symmetry is identical to the R-symmetry of the ${\cal N} = 4$ SYM. After including fermionic generators required by supersymmetry, we have the full isometry supergroup $SU(2, 2\mid 4)$ for the $AdS_5 \times S^5$ background, which is identical to the ${\cal N} = 4$ superconformal group. We would like to emphasize that we have the conformal symmetry $SO(2,4)$ only at $U = 0$. For any fixed non-zero $U$, this symmetry is spontaneously broken since $U$ represents the expectation value of the Higgs. To validate the configuration of D3 branes as a stringy one, the large $n$ limit has to be taken given the fixed but small string coupling\footnote{For large $g$, the D-string frame is chosen.}. It is well-known that there is an SL(2,Z) strong-weak duality symmetry in both type IIB supergravity and ${\cal N} = 4$ SYM. These symmetry identifications, among other things, led Maldacena to conjecture that the large $n$, ${\cal N } = 4$ SYM theory with gauge group $U(n)$ is actually equivalent to the ten dimensional type IIB supergravity on $AdS_5 \times S^5$. However, the supergravity itself is not a consistent quantum theory. On the other hand, the SYM is a unitary quantum theory. Moreover, the above symmetry identifications are independent of the large $n$ limit (even though it is needed to validate the solution). These two facts led Maldacena further to conjecture that type IIB string on $AdS_5 \times S^5$ is equivalent to the ${\cal N} = 4$ SYM theory for a general $n$. Let us now apply the same process to the (F, D3) configuration obtained in section 3, i.e., taking the decoupling limit given by Eqs.\ (\ref{eq:dcl}) and \ (\ref{eq:esl}). In other words, we should examine the near-horizon geometry of (F, D3) configuration with the string coupling $g$ fixed\footnote{For simplicity, we set $\chi_{B0} = 0$ from now on.}. The near horizon string-frame metric in this case is \begin{eqnarray} d s^2 =&& \alpha' \left[ \frac{U^2}{\sqrt{4\pi n g}} \left(\frac{ n^2 e^{-\phi_0}}{\Delta_{(m,n)}}\right)^{3/4} ( - (d x^0)^2 + (d x^1)^2 ) \right.\nonumber\\ && + \frac{U^2}{\sqrt{4\pi n g}} \left( \frac{ n^2 e^{-\phi_0}}{\Delta_{(m,n)}}\right)^{ - 1/4} ((d x^2)^2 + (d x^3)^2) \nonumber\\ && \left. + \sqrt{4 \pi n g} \left(\frac{ n^2 e^{-\phi_0}}{\Delta_{(m,n)}}\right)^{1/4} \left(\frac{d U^2}{U^2} + d \Omega_5^2 \right) \right], \label{eq:nhg} \end{eqnarray} and the dilaton is \begin{equation} e^\phi = \left(n^2 e^{-\phi_0} /\Delta_{(m, n)}\right)^{1/2} e^{\phi_0}. \label{eq:dthree} \end{equation} It is clear from (5.3) and (5.4) that unlike the simple D3 brane, the near-horizon geometry in this case is not automatically $AdS_5 \times S^5$ and the effective string coupling, $e^\phi$, is still a finite constant but is less than the string coupling\footnote{The dilaton for (F, D3) is actually not a constant in general, nevertheless it is bounded as $ \left(n^2 e^{-\phi_0} /\Delta_{(m, n)}\right)^{1/2} e^{\phi_0}< e^\phi < e^{\phi_0}$, where the upper bound is obtained at $r = \infty$ and the lower bound at $r = 0$.} $g$. The above features may be expected given the appearance of an infinite number of F-strings in (F, D3) and the non-threshold bound state nature of this configuration. On the other hand, in the decoupling limit, we expect that D3 brane can be described equivalently either by the SYM theory on the brane or by the type IIB supergravity (or string) in the bulk. Depending on whether we consider excitations with respect to the simple D3-brane vacuum or the (F, D3) bound state (chosen as a new vacuum), we have two descriptions which are different in appearance but probably equivalent in essence. Let us first discuss the possible underlying IIB string/SYM correspondence for each of the above descriptions in the strong sense, i.e., not taking the large $n$ limit. We will discuss the large $n$ limit in the end. If we choose the simple BPS D3 brane configuration in the bulk or equivalently no excitations in the SYM on the brane as the vacuum, then the F-strings in the bulk or equivalently the electric flux lines in the SYM should be formed due to excitations with respect to this vacuum. We know that the energy per unit 3-brane volume associated with the F-strings in the bulk or the energy density associated with the electric flux lines on the brane is finite. This implies that we have an infinite amount of energy associated with either the F-strings or the electric flux lines. In other words, we should have $U = \infty$, which implies that the original conformal symmetry group $SO(2,4)$ must be broken spontaneously to some smaller symmetry group consisting of at most some translational and rotational symmetries. It is not difficult to figure out the origin of $U = \infty$ if we recall the method we used in \cite{lurone} to find the (F, D3) bound state. As discussed in \cite{lurone}, the charge conservation must imply that the infinite parallel F-strings in (F, D3) originate from F-strings along the radial coordinate $r$ ending on the $x^2x^3$-plane placed at $x^1 = - \infty$ \footnote{This is chosen in order to preserve the SO(6) isometry of $S^5$.}. These F-strings ending on $x^2x^3$ plane are the open strings, each semi-infinitely long, giving $U = \infty$. Inside the bulk, we have F-strings along $x^1$-axis, therefore we expect an $SO(1,1)$ symmetry. Because of the ending or the charges on $x^2x^3$-plane placed at $x^1 = - \infty$, we expect that $SO(2,4) \rightarrow SO(1, 1) \times SO(2 )$ for this near-horizon geometry. Thus we find that we actually have $SO(1,1) \times SO(2 ) \times SO(6)$ isometries for the near-horizon geometry of the (F, D3) bound state for the choice of vacuum. On the SYM side, we have the same answer. Dilation and the special conformal transformations are broken spontaneously for the same reason mentioned above. Only the subgroup $SO(1,1) \times SO(2)$ of $SO(2, 4)$ will leave $F_{01} = g m T_f$ invariant. So, following the similar argument as for the $AdS_5/CFT_4$ correspondence, we may suggest that the type IIB string on this geometry is equivalent to the ${\cal N } = 4$ SYM theory in a state with quantized electric flux lines. Let us investigate how the (F, D3) bound state emerges as our vacuum for the second description. Examining carefully the metric and the dilaton, we can see that $ n^2 e^{-\phi_0} /\Delta_{(m, n)}$ factor is the source which causes the near-horizon geometry to be not automatically $AdS_5 \times S^5$ and the effective string coupling to be less than the string coupling $g$. Explicitly, we have this factor \begin{equation} \frac{n^2 e^{-\phi_0}}{ \Delta_{(m,n)}} = \left(1 + \left(\frac{g m}{n}\right)^2 \right)^{-1}. \label{eq:fac} \end{equation} Now if we rescale the coordinate time $x^0$ and the coordinate $x^1$ by a factor $ \left(n^2 e^{-\phi_0} /\Delta_{(m, n)}\right)^{-1/2}$, then in terms of the new coordinates the near-horizon geometry is $AdS_5 \times S^5$. So, if we rewrite the metric in terms of the effective string coupling \begin{equation} g_{\rm eff} \equiv e^\phi = \frac{g}{\left(1 + \left(\frac{g m}{n}\right)^2 \right)^{1/2}}, \label{eq:esc} \end{equation} we have \begin{equation} d s^2 = \alpha' \left[ \frac{U^2}{\sqrt{4\pi n g_{\rm eff}}} \left( - (d x^0)^2 + (d x^1)^2 + (d x^2)^2 + (d x^3)^2 \right) + \sqrt{4 \pi n g_{\rm eff}} \left(\frac{d U^2}{U^2} + d \Omega_5^2 \right) \right]. \label{eq:enhg} \end{equation} The radius for the $AdS_5$ and the $S^5$ is now given as \begin{equation} R^2/\alpha' = \sqrt{4\pi n g_{\rm eff}}, \label{eq:radius} \end{equation} The information about the F-strings in the (F, D3) bound state disappears in the above metric. Since the resulting near-horizon geometry differs from that of a simple D3 brane with 5-form flux $n$ only in the string coupling, we expect that the effects of the F-strings are encoded in the effective string coupling $g_{\rm eff}$. We will see that this is indeed true. With respect to the new coordinates, we have $SO(2,4) \times SO(6)$ isometries in the bulk at least in appearance. How can we reconcile this with what we just discussed above on the SYM side? Moreover, what is the physics behind such a rescaling? In the case of simple D3 branes, the string coupling $g = e^{\phi_0}$ remains the same whether we are at $r = \infty$ or at $r = 0$. This fact enables us to fix the gauge coupling as $g_{\rm YM}^2 = 2 \pi g$ for the SYM theory describing the D3 branes in the decoupling limit. We do not expect that such a relation will be modified for any finite energy excitations\footnote{ But it could be modified if the energy involved is infinite, for example, the case we are presently studying. If this happens, the SYM theory may no longer be valid in describing the underlying physics in general and string theory should be used instead. However, our discussion above for the first description may still be valid since it is based on the BPS (F, D3) configuration.}. However, the story here is different. The gauge coupling should, in the present case, be related to $g_{\rm eff}$ rather than $g$ as $g_{\rm YM}^2 = 2\pi g_{\rm eff}$. With respect to the metric in Eq.\ (\ref{eq:enhg}), we may say that a simple D3 brane with 5-form flux $n$ plays as an ``effective" vacuum configuration in the bulk with the string coupling $g_{\rm eff}$. This simple D3 brane is also described by a ${\cal N} = 4$ $U(n)$ SYM theory in 1 + 3 dimensions with gauge coupling $g_{\rm YM}^2 = 2\pi g_{\rm eff}$. This is just the usual $AdS_5/CFT_4$ correspondence except we have a new coupling constant. Let us dig out the mystery behind the above ``effective" vacuum. We expect that the tension for the simple D3 brane is $n T^3_0 / g_{\rm eff}$. If we use the explicit expression for $g_{\rm eff}$ from Eq.\ (\ref{eq:esc}), this tension is \begin{equation} \frac{n T_0^3} {g_{\rm eff}} = \frac{T_0^3}{g} \sqrt{n^2 + g^2 m^2}, \label{eq:ident} \end{equation} which is just the tension for the (F, D3) bound state. So this ``effective" vacuum is nothing but the (F, D3) bound state. As anticipated, the effects of the F-strings in (F, D3) is encoded into the string coupling constant. On the SYM side, we have already hinted that the gauge coupling is now given by $g_{\rm YM}^2 = 2 \pi g_{\rm eff}$. In the linear approximation, \begin{equation} \frac{1} {g_{\rm eff}} = \frac{1}{g} + \frac{1}{2 g} \left(\frac{ g m}{n}\right)^2, \label{eq:lgc} \end{equation} where the second term for $n = 1$ is nothing but the contribution from the energy density of the electric flux lines. In other words, the effect of the energy of the quantized electric flux lines is absorbed into the gauge coupling. This is in accordance with our spacetime picture just discussed. Therefore, the resulting SYM theory should be the one describing the excitations with respect to the (F, D3) bound state. The superconformal symmetry should then be restored. We therefore have a consistent picture both in the bulk and on the brane in the decoupling limit. Our study lends further support for Maldacena's $AdS_5/CFT_4$ correspondence and shows that it holds true even for the non-trivial D3 brane configurations. The new input for this correspondence is that we should use the effective string coupling constant instead of the usual one. The two descriptions studied may be equivalent but the picture for the first one is surely complicated. The reason for this is our improper choice of the vacuum state. If we take the large $n$ limit, i.e., $n \gg m$ for fixed $g$, the effective string coupling becomes the usual string coupling $g$. The F-strings in the (F, D3) now play a minor role with respect to the D3 branes which can be seen from the tension formula $T_3 (m, n) = n/g \left[ 1 + (m /n)^2 g^2 /2\right] T^3_0$. On the SYM side, on the other hand, we have the gauge coupling $g_{\rm YM}^2 = 2 \pi g (1 - (m/n)^2 g^2 /2)$ which implies that electric flux lines plays the minor role. Thus the $AdS_5/CFT_4$ correspondence with respect to (F, D3) bound state is simply reduced to that of the simple D3 brane configuration. We also expect that the full isometry supergroup $SU(2,2\mid 4)$ in the bulk and the superconformal symmetry in the SYM theory are restored under this limit in our first description above. This description is therefore also reduced to the usual $AdS_5/CFT_4$ correspondence. This large $n$ limit also validates the (F, D3) configuration as a stringy one. Let us now comment on the effective string coupling $g_{\rm eff}$ given by Eq.\ (\ref{eq:esc}) in the near-horizon region. It is clearly quantized in terms of relatively prime integers $n$ and $m$ and is always less than the string coupling $g$. In particular, in the limit $m \gg n$, it becomes $g_{\rm eff} = n /m \ll 1$, independent of the string coupling $g$. In other words, the effective string coupling in the near-horizon region is completely determined by integer $n$, the 3-brane charge, and $m$, the number of F-strings per $(2\pi)^2 \alpha'$ area over the $x^2x^3$-plane, in this limit. We will continue this study in a more general D3-brane configuration, namely, ((F, D1), D3) bound state in a forthcoming paper \cite{lurthree}. \section{Conclusion\protect\\} \label{sec:c} To summarize, we have constructed, in this paper, explicitly the (F, Dp) non-threshold bound state configurations for $2 \le p \le 7$, starting from the known (F, D1) configuration of Schwarz by T-duality transformation along the transverse directions of the strings. We have also presented the solutions for (W, Dp) non-threshold bound state configurations for $0 \le p \le 6$ by T-duality transformations on the newly obtained (F, Dp) ones along the longitudinal direction of F-strings in (F, Dp). We have shown explicitly that there are $m$ F-strings per $(2\pi)^{p - 1} \alpha'^{(p - 1)/2}$ of $(p - 1)$-dimensional area which agrees with our previous result based on the worldvolume study. We have also calculated the tensions for the (F, Dp) bound states which once again agree with our previous results. These two facts confirm our assertion made in our earlier paper that the bound state studied in \cite{lurone} can indeed be identified with the (F, Dp) bound states here. All of these bound states preserve one half of the spacetime supersymmetries. From the D-brane worldvolume point of view, each of these bound states consists of a Dp-brane carrying certain units of quantized electric flux or field line. But from the spacetime point of view, we have type II strings (either type IIA or IIB strings) lying along one direction in the corresponding Dp-brane worldvolume. In this paper, we have shown that for each of the bound states considered, the type II strings can be identified with the electric flux lines in the gauge theory living on the corresponding Dp brane worldvolume. In the $p = 3$ case, we have studied the corresponding decoupling limit and found that the $AdS_5/CFT_4$ correspondence may still hold true but now with respect to the (F, D3) bound state with an effective string coupling constant. The F-strings in the (F, D3) bound state or the electric flux lines in the corresponding SYM theory play the role of reducing the respective coupling constant in general. The string coupling in the near-horizon region is found to be quantized in terms of the relatively prime integers $n$, the 3-brane charge, and $m$, the number of F-strings per $(2 \pi)^2 \alpha'$ area over the $x^2x^3$-plane. It becomes independent of the asymptotic string coupling $g$ in the limit $m \gg n$. \acknowledgments We would like to thank Mike Duff for reading the manuscript and Chris Pope for discussions. JXL acknowledges the support of NSF Grant PHY-9722090.
\section{Introduction} \label{intro} Soft gamma--ray repeaters (SGRs) are a class of astrophysical sources that emit bursts of high energy x--ray and gamma--ray radiation which are among the most energetic events in the Galaxy. The apparent association of their positions with supernova remnants and the detection of pulse periods in their nonbursting emission strongly suggest that the SGRs are young neutron stars (e.g. Mazets et al. 1979, and review by Rothschild 1995). The SGRs may also be related to the anomalous X--ray pulsars (AXPs: \cite{mer97}), which have comparable long ($>$ few second) periods. The observed SGR burst energies, assuming isotropic emission, range from typical values of $\sim 10^{41}$ ergs to as much as $10^{44}$ ergs in rare giant flares, such as that of 5 March 1979 from the SGR 0529--66 in the Large Magellanic Cloud. Suggested energy sources for these bursts have included, i) the rotational energy of the neutron star, $\sim 10^{45} {(P/3.1\,\rm{s})}^{-2}$ ergs, where $P$ is the spin period, which might be tapped by pulsar glitches (e.g. \cite{baym71}), ii) the magnetic field energy $\sim 10^{44}{(B/B_{q})}^{2}$ ergs of {\it magnetars} with surface magnetic fields much greater than the quantum critical field $B_{q}=m_{e}^{2}c^{3}/ e\hbar\approx 4.4\times 10^{13}$ G tapped by magnetic-stress driven crustal quakes and magnetic reconnection (\cite{thompson95}), and iii) the gravitational binding energy of the neutron star, $\sim 10^{53}$ ergs, tapped by quakes (e.g Ramaty et al. 1980), and driven by plate tectonics (\cite{ruderman91}). Recent measurements of the rapid spindown rates of the SGR pulsars have been taken (e.g. \cite{kouv98}, 1999) as evidence for the magnetar hypothesis, in which the magnetic energy of the neutron star exceeds the rotational energy. Pulsations have been observed from three of the SGRs: SGR 0526--66 ($8$ s: \cite{mazets79}), SGR 1806--20 ($7.47$ s: \cite{kouv98}), and SGR 1900+14 ($5.16$ s: \cite{hurley99b}). The period derivatives ($\dot{P}$) of these pulsars have been found by either direct measurement (SGRs 1806--20 and 1900+14) or by $\dot{P}=0.5 P/t_{snr}$, where $P$ is the pulse period and $t_{snr}$ is the estimated age of the associated supernova remnant (SGR 0526--66). If the spindown is driven by magnetic dipole radiation from an orthogonally rotating vacuum magnetic dipole, it can be shown (\cite{pacini68}) that the surface magnetic field is given by $B_{0}\approx 3.2\times 10^{19}\sqrt{P\dot{P}}$ G, which would yield surface magnetic fields of $6\times 10^{14}$, $8\times 10^{14}$, and $5\times 10^{14}$ G for SGRs 0526--66 (\cite{thompson95}), 1806--20 (\cite{kouv98}), and 1900+14 (\cite{kouv99}), respectively. Here we present {\it RXTE} observations, however, which suggest that the spindown rate of SGR 1900+14 is due to torques other than those provided by the magnetic field, and thus does not provide evidence of a supercritical surface dipole field. \section{Observations \& Analysis} \label{obervations} SGR 1900+14 was observed by the Proportional Counter Array (PCA) and High Energy X--ray Timing Experiment (HEXTE) instruments aboard the {\it Rossi X--ray Timing Explorer} on a number of occasions during the period September 4-18, 1996. The total exposure time was $\sim 47$ ks, with a temporal baseline of $15.4$ days. For the first $22$ ks, {\it RXTE} was pointed at a position RA (J2000)$= 286^{\circ}.82$ and Dec (J2000)$=9^{\circ}.32$, which is $\sim 48 \arcsec$ from the precise VLA position of SGR 1900+14 (\cite{frail99}), but well inside the $1^{\circ}$ FWHM field of view of the {\it RXTE} pointed instruments. Midway through the observations, the pointing position was changed to exclude the bright $438$ s binary x--ray pulsar 4U 1907+09 (\cite{zand98}) from the field of view. The second half of the observation ($25$ ks) was then conducted at the pointing position RA$=286^{\circ}.43$ and Dec$=8^{\circ}.98$, which is $\sim 0^{\circ}.35$ from the position of the SGR. As luck would have it, this field also contained a relatively bright confusing source, the $89$ s transient x--ray pulsar XTE J1906+09, which was discovered during the observation ({\cite{marsden98}). Finally, the Galactic Ridge emission is also a significant contributor to the x--ray flux in the {\it RXTE} field of view (\cite{valinia98}), due to the low Galactic latitude of SGR 1900+14 ($b\sim 0^{\circ}.75$). Because of these complications, we do not attempt to determine the x--ray spectrum of the SGR with the {\it RXTE} data, and instead concentrate on the temporal analysis. For information on the x--ray spectrum of the source, the reader is referred to Hurley et al. (1999b), Kouveliotou et al. (1999), and Murakami et al. (1999). The pointed x--ray instruments aboard {\it RXTE} are the High Energy X--ray Timing Experiment (HEXTE) and the Proportional Counter Array (PCA). HEXTE consists of two clusters of collimated NaI/CsI phoswich detectors with a total net area of $\sim 1600$ cm$^{2}$ and and effective energy range of $\sim 15-250$ keV (\cite{rothschild98}). The PCA instrument consists of five collimated Xenon proportional counter detectors with a total net area of $7000$ cm$^{2}$ and an effective energy range of $2-60$ keV (\cite{jahoda96}). The uncertainty in the timing of x--ray photons by the PCA and HEXTE is $<<1$ ms (\cite{rots98}), and is therefore negligible in the temporal analysis presented here. The PCA and HEXTE photon times were corrected to the Solar System barycenter using the JPL DE200 ephemeris and the SGR coordinates RA(J2000)$=19^{\rm{h}} 07^{\rm{m}} 14\fs 33$ and Dec(J2000)$=+09^{\circ}19\arcmin 20\arcsec.1$ (\cite{frail99}). The PCA data were searched for pulsations using the chi-squared folding method, which calculates the value of chi-squared for a pulsar lightcurve (versus a constant rate) folded on a range of trial pulsar periods. Here the pulse phase $\phi$ for a given photon time $t$ is defined by the relation $\phi(t)=f(t-t_{0})+ {1\over 2}\dot{f}{(t-t_{0})}^{2}$, where the pulsar frequency $f$ and frequency derivative $\dot{f}$ are related to the period $P$ and period derivative $\dot{P}$ by the expressions $P=1/f$ and $\dot{P}=-\dot{f}P^{2}$. A maximum value of chi-squared occurs when the data are folded on the true pulsar period and period derivative. The PCA data were initially searched for pulsations using a range of $\sim 500$ periods about $5.153642$ s, the SGR 1900+14 period predicted from the timing ephemeris given in Kouveliotou et al. (1999). A significant chi-squared peak was seen, and a finer search was then conducted on a grid in $P-\dot{P}$ space around the peak, for a broad range of $\dot{P}$ including the value of $\dot{P}\sim 10^{-10}$ s s$^{-1}$ found by Kouveliotou et al. (1999). The results of the grid search are shown in Figure $1$. To estimate the confidence regions of $P$ and $\dot{P}$ indicated by the peak in chi-squared, we folded the $2-10$ keV PCA data with $P$ ($\dot{P}$) values slightly displaced from the peak value, while holding $\dot{P}$ ($P$) fixed at its peak value. The resultant lightcurves were then compared to a template lightcurve using the chi-squared test, and the $90\%$ confidence contours were calculated using the chi-squared probability distribution. A folding time midway through the {\it RXTE} observation was used throughout the analysis to minimize correlations between $P$ and $\dot{P}$. Using this analysis, we obtain a timing solution of $P=5.1558199\pm 0.0000029$ s and $\dot{P}=(6.0\pm 1.0)\times 10^{-11}$ s s$^{-1}$, referenced to $t_{0}=50338.216$ (MJD). The errors are $90\%$ confidence. A search of the $15-100$ keV HEXTE data for the pulsar, using the PCA timing solution, failed to produce evidence of significant pulsations, which is not surprising given the faintness of the source and the presence of the bright confusing sources. The folded SGR 1900+14 pulsar lightcurve for three PCA energy ranges, using the above timing parameters, is shown in Figure $2$. The pulsed fraction of the SGR 1900+14 is not constrained by these data, due to the uncertain x--ray flux from XTE J1906+09, 4U 1907+09, and the Galactic Ridge in the {\it RXTE} bandpass. \section{Discussion} \label{discussion} The $2-10$ keV SGR 1900+14 lightcurve obtained here is virtually identical to the lightcurves obtained just before (\cite{hurley99b}) and just after (\cite{kouv99}) the commencement of the May 1999 active period of the source. This indicates that the x--ray emitting geometry is stable on timescales of years while the source is inactive. The lightcurve appears to have multiple components which vary differently with energy. There are three peaks in the $2-10$ keV lightcurve, with a single relatively broad central peak surrounded by two narrower peaks. The narrow peaks have harder spectra than the broad peak, as the narrow peak emission dominates the emission from the broad peak above $10$ keV. A simple explanation for the lightcurve morphology is that the pulsed emission consists of different emission components arising from different regions of the stellar surface. The narrow components may be beamed emission from a collimated wind off of relatively small hotspots, while the broader component could be more isotropic emission from a larger and cooler area of the crust. The two narrow components are greatly reduced in the pulsar lightcurves obtained just after the giant flare of August 27, 1999 (\cite{kouv99}; \cite{murakami99}), suggesting that the energy of the small hotspots may have been depleted during the active period. The observed temporal history of the SGR 1900+14 pulsar is shown in Figure $3$. The additional timing parameters of the present observations are important because they constrain the pulsar parameters long before the source went into outburst. Although the temporal coverage is incomplete, the {\it secular} spindown rate seems to change abruptly sometime close to the initiation of bursting, at which point the spindown continues steadily at an increased rate. These two different spindown rates are denoted by the dotted lines in Figure $3$, which are linear fits to the data before the outburst [up to and including the first observation of Kouveliotou et al. (1999)] and the data during and after the outburst [beginning with the first observation of Kouveliotou et al. (1999) and ending with the Shitov (1999) observation]. The third data point in Figure $3$, from Kouveliotou et al. (1999), appears to be near the change point in the spindown behavior because the period is consistent with the extrapolation of the pre-outburst timing solution, yet the $\dot{P}$ value measured during this observation is consistent with the outburst values. The fit to the data taken during and after the outburst period yields a value of $\dot{P}=(12.77\pm 0.01)\times 10^{-11}$ s s$^{-1}$ for the mean spindown rate, and the corresponding pre-outburst value is $\dot{P}=(6.126\pm 0.006)\times 10^{-11}$ s s$^{-1}$. Using these mean $\dot{P}$ values, the mean inferred dipole field strengths before and after the initiation of bursting would be $5.7\times 10^{14}$ G and $8.2\times 10^{14}$ G, respectively, if the spindown were driven by dipole radiation losses. These two values, which differ to a high degree of significance, would imply an abrupt increase in the SGR 1900+14 magnetic field energy of more than $100\%$ around the time the source started bursting, which is contrary to the predictions of models in which the bursting is dissipating magnetic field energy. This discrepancy clearly suggests that the SGR 1900+14 spindown is not dominated by magnetic dipole radiation, and that the observed value of $P\dot{P}$ provides no direct measurement of $B$, and no direct evidence for a magnetar. Instead, the measured values of $P$ and $\dot{P}$ suggest that the SGR spindown may be due to {\it winds}, if we take the pulsar age to be that of the associated (\cite{hurley99a}) supernova remnant G42.8+0.6. Assuming that the initial period of the pulsar was much smaller than it is now, and that the braking index is constant in time, the pulsar age $t_{age}=P/[(n-1)\dot{P}]$, where the braking index $n$ is $3$ for pure dipole radiation but much less ($n\sim 1$) for spindown due to wind torques. Taking the estimated age of G42.8+0.6 to be $\sim 10^{4}$ yr (\cite{vasisht94}, \cite{hurley96}), we find that the braking index for SGR 1900+14 must be $\sim 1$, i.e. $n = 1 + 0.16/(t_{age}/10^4\,\rm{yr})$, which indicates that the pulsar spindown is dominated by winds. The remnant age would have to be an order of magnitude smaller in order for the braking index to be consistent with that of dipole radiation, and in addition such an age would require an unreasonably large pulsar velocity of $\sim 2.5\times10^4$ km s$^{-1}$ for it to have traversed from the center of the remnant to its present position, assuming a distance of 5 kpc (\cite{vasisht94}, \cite{hurley96}). Thus the observations provide strong evidence that torques due to wind emission, and not magnetic dipole torques, dominate the spindown dynamics of SGR 1900+14. The spindown behavior of SGR 1900+14 can be explained simply if we assume that the spindown is due almost entirely to wind emission, as was also considered by Kouveliotou et al. (1999). Possible mechanisms for the generation of this wind include thermal radiation from hotspots and Alfv\'{e}n wave emission (\cite{thompson98}). In this interpretation, the SGR emits a robust wind of particles and fields, both during bursting and quiescent intervals, which carries away angular momentum from the star. The emission of a relativistic wind produces an exponential spindown of the pulsar $\Omega(t)=\Omega_{0}\exp(-kt)$, where $k$ is a constant parameterizing the rotational energy loss rate due to the wind (\cite{thompson98}). Using this relation, and the values of $P$ and $\dot{P}$ from our observations, we obtain $k=\dot{P}/P\sim 2700^{-1}$ yr$^{-1}$. Given an age of $(1-2)\times 10^{4}$ yr for G42.8+0.6, we obtain an initial pulsar spin period of $P_{0}\sim 3-120$ ms for SGR 1900+14, which is similar to the spin periods of young isolated pulsars such as the Crab. This $P_{0}$ is most likely an upper limit, given the likelihood of active periods (with higher spindown rates) in the past. As mentioned above, one scenario is that the spindown of SGR 1900+14 is due to Alfv\'{e}n wave emission, in which a stream of particles and fields escape the star along magnetic field lines forced open by the wind pressure (\cite{thompson98}). A supercritical magnetic field is not required for this mechanism to explain the SGR 1900+14 spindown. From Thompson \& Blaes (1998), the spindown constant is given by \begin{equation} k=1.5\times 10^{-11}{\left({B_{\ast}\over 3\times 10^{12}\,\rm{G} }\right )}^{2}{\left({\delta B_{\ast}/B_{\ast}\over0.01}\right )}^{4/3} \,\rm{Hz}, \end{equation} where $B_{\ast}$ is the dipole field strength, $\delta B_{\ast}$ is the wave amplitude, and we have assumed a neutron star moment of inertia and radius of $1.1\times 10^{45}$ g cm$^{2}$ and $10$ km, respectively. This value of $k$ is comparable to the measured value $k=\dot{P}/P\sim 10^{-11}$ Hz for SGR 1900+14, indicating that this mechanism can explain the spindown of the SGR with conventional ($\sim 10^{12}$ G) field strengths, assuming that there is a mechanism to continuously generate Alfv\'{e}n waves. Even though a supercritical magnetic field on a global scale can not account for the SGR pulsar spindown, such fields on much smaller localized scales may nevertheless play an important role in the bursting process. Since the wind torques initially operate to spin down the neutron star crust, one might expect that if the core is not rigidly coupled to the crust, then the core could be spinning slightly faster and the resulting differential rotation could wind up any magnetic field threading between the core and crust, building up large internal magnetic field pressures. By analogy to the Sun, we might expect that the growing pressure of the internal field is episodically released by the surface break out of intense magnetic fields in localized regions, similar to the appearance of sunspots, which have local fields of 10$^2$ to 10$^3$ times the average global surface field of the Sun. Such spots of emerging magnetic flux (EMF) on a neutron star may thus contain supercritical, or larger, localized fields, $B_s$ within radii $r_s$, with total magnetic energies $> 3\times 10^{41}(B_s/B_{q})^2(r_s/1\, \rm{km})^3$ erg, and they may be accompanied by comparable tectonic stresses and heating from field diffusion in the crust. To contain the giant flare of August 27, 1999, for example, a local field with $B\sim B_{q}$ can contain the $3\times 10^{42}$ ergs of energy released (\cite{frail99}) within a bubble of radius $r_{s}\sim 2$ km, which is a small fraction of the surface area of the star. The occurrence of such EMF-spots could thus provide an episodic source of both magnetic and tectonic-gravitational energy release, both thermal and nonthermal, that power both the steady localized winds and the impulsive bursts of SGRs, much as the sunspot fields are dissipated in winds, flares and diffusion on the Sun. The solar analogy was also discussed by Sturrock (1986) for Galactic gamma--ray bursts. The SGR wind hypothesis can also explain other observed features of the burst and quiescent emission from SGRs. If both the quiescent x--ray emission and the spindown torque of SGR 1900+14 are due to wind emission, the persistent x--ray flux and the spindown luminosity should be correlated (this is not true of SGR 1806--20, because of the surrounding plerion --- see below). Between the {\it ASCA} observations of Hurley et al. (1999b) and Murakami et al. (1999), the persistent x--ray flux of SGR 1900+14 increased by $(140\pm 20) \%$. Using the appropriate mean $\dot{P}$ values from Figure $3$, the spindown luminosity increased by $\sim 120\%$ over the same time interval, which is consistent with the steady x--ray flux and spindown arising from the wind. The radio signature of SGR winds have been observed from SGRs 1900+14 (\cite{frail99}) and 1806--20 (\cite{kulkarni94}). In the latter case, the SGR winds power a plerionic nebula with a total energy content ($\sim 10^{45}$ ergs) much greater than the energy given off in a typical burst interval ($\sim 10^{43}$ ergs, \cite{kouv99}), explaining the lack of variability seen from the SGR 1806--20 x--ray and radio counterparts (\cite{sonobe94}; \cite{vasisht95}). In the case of SGR 1900+14, a {\it transient} wind nebula from relativistic particles injected during the giant flare of August 27, 1999 (\cite{hurley99c}) was observed by the VLA (\cite{frail99}). The different radio properties of the SGR 1806--20 and SGR 1900+14 counterparts are probably due to the different external pressures for the two sources, since SGR 1806--20 is still inside its high pressure SNR while SGR 1900+14 is outside its associated supernova remnant, where the confining pressure is relatively low. The weak confining pressure of SGR 1900+14 inhibits the formation of a bright plerion (\cite{frail99}). The observed nonthermal (photon index $\sim 2.2$: \cite{sonobe94}; \cite{hurley99b}) quiescent x--ray spectra of the active SGR sources is characteristic of emission from a magnetized wind (\cite{tavani94}). Finally, the burst spectra of SGRs can be explained by the Compton upscattering of soft photons in a mildly relativistic wind, without involving a supercritical stellar field (\cite{fat96}). \acknowledgments We thank Duane Gruber for suggesting improvements in the timing analysis. This work was funded by NASA grant NAS5-30720.
\section{Introduction and statement of the results} \subsection{Introduction} \subsubsection{} Scattering theory is a collection of methods and results for studying the continuous spectrum of operators. Classical fields of application of scattering theory are the propagation of waves in mechanics, electro-dynamics and quantum mechanics. Often one considers a model that is close to some ideal case in which the wave equation can be solved explicitely. Scattering theory provides the tools for comparison. \subsubsection{} In many cases it is possible to separate the time and space variables. In this case one can apply methods from stationary scattering theory. This branch of scattering theory investigates the generalized eigenfunctions contributing to the continuous spectrum. In good situations these eigenfunctions come in families which extend meromorphically. Often they are studied via a meromorphic continuation of the distribution kernel of the resolvent of the spacial part of the operator. \subsubsection{} The problem of stationary scattering theory that is considered in the present paper has mainly an internal mathematical motivation. The model situation is a Laplace-type operator on a globally symmetric space of negative curvature. Due to the presence of a large symmetry group it is possible to get an essentially complete description of the generalized eigenfunctions. The real problem is to understand the generalized eigenfunctions on associated locally symmetric spaces. The special case of quotients of the globally symmetric space by arithmetic subgroup groups has many applications in number theory and arithmetic. While arithmetic quotients have finite volume the main emphasis in the present work is on spaces of infinite volume. The question is, how far the interplay between global and local symmetries of the problem can be used in order to get a complete picture. \subsubsection{} There is lot of literature on the stationary scattering theory for the function Laplacian on spaces which become close to the global symmetric space asymptotically at infinity. The major part is devoted to the asymptotically hyperbolic case. The main results are meromorphic continuation of the resolvent kernel, the scattering matrix and the parametrization of the relevant generalized eigenfunctions. Most of this work is based on a fine study of the resolvent kernel. We refer to \cite{MR2153454} for one of the latest works and the discussion of the literature therein. \subsubsection{} In the more rigid case of a locally symmetric space one can obtain better results. In this case one can approach stationary scattering theory by pushing the analysis to the boundary. Using this method we obtained in \cite{MR1749869} the spectral decomposition of all locally invariant differential operators on vector bundles in the convex-cocompact case. We refer to this paper for a review of the literature about the convex-cocompact case. \subsubsection{} The goal of the present paper is to generalize this method to geometrically finite spaces. These are locally symmetric spaces of negative curvature which at infinity look like the symmetric space or a cusp. From the point of view of analysis the part at infinity which can be compared with the globally symmetric space is easy with the results of \cite{MR1749869} at hand. The complications are due to the presence of cusps, in particular of those of non-maximal rank. \subsubsection{} Let $G$ be a real simple linear connected Lie group of real rank one. It covers the connected component of the group of isometries of the associated symmetric space $X$. We consider a geometrically finite group $\Gamma\subset G$ which determines the locally symmetric space $\Gamma\backslash X$. For technical reasons we exclude the exceptional symmetric space $X={H\mathbb{O}}^2$ from our considerations. The classical problems of meromorphic continuation of the Eisenstein series, the scattering matrix, and their functional equations was previously adressed by Guillop\'e \cite{guillope92} in the special case that $X$ is the hyperbolic plane ${H\mathbb{R}}^2$ and by Froese/Hislop/Perry \cite{MR1111571} for $X={H\mathbb{R}}^3$ and spherical Eisenstein series. Cusps of non-maximal rank in the three-dimensional hyperbolic case have been investigated first in \cite{MR1128217} via the resolvent. For the meromorphic continuation of the resolvent on $\Gamma\backslash{H\mathbb{R}}^n$ see \cite{MR1710792},\cite{MR2209763}. All these papers are restricted to the case of so-called rational cusps. \subsubsection{} The main result of the present paper is, in the geometrically finite case, the meromorphic continuation of the Eisenstein series, i.e. of the families of generalized eigensections for all locally invariant differential operators on bundles, see Corollary \ref{main} below. Note that this result also includes the case of irrational cusps. At the moment we are not able to obtain results, which are as complete as in the convex cocompact case. In particular, we are still quite far from showing a spectral decomposition. \subsubsection{} The class of geometrically finite discrete subgroups $\Gamma\subset G$ subsumes cocompact, convex-cocompact groups, and subgroups of finite covolume as well as various kinds of combinations of these cases with cusps of non-maximal rank. The paper can be considered as a continuation of \cite{MR1749869}, where the special case of a convex-cocompact group $\Gamma$ is considered. We will frequently refer to this paper for notations and conventions as well as for technical results. \subsubsection{} In the present paper the analysis takes place on the boundary $\partial X$ of the symmetric space. The group $\Gamma$ acts on vector bundles over this boundary. As in \cite{MR1749869} the main players of the geometric scattering theory are the push-down $\pi^\Gamma_*$ (average of sections over $\Gamma$), the extension $ext^\Gamma$ (the adjoint ot the push-down), the restriction $res^\Gamma$, and the scattering matrix $S^\Gamma$. Our task is to construct these as families of maps between suitable function/distribution spaces depending meromorphically on the spectral parameter. Due to the presence of cusps of non-maximal rank the detailed description of these spaces turns out to be quite difficult. \subsection{Notion of geometrical finiteness}\label{geomf} \subsubsection{} In the present paper the symmetric space $X$ belongs to one of the series of hyperbolic spaces ${H\mathbb{R}}^n$, ${H\mathbb{C}}^n$, and ${H\mathbb{H}}^n$ (real, complex, and quaternionic) of real dimension $n$, $2n$, $4n$. We exclude the exceptional symmetric space of rank one, the Cayley hyperbolic plane ${H\mathbb{O}}^2$, for technical reasons since we are going to employ the fact that $X$ belongs to a series in several places. \subsubsection{} By $G$ we denote a real simple linear connected Lie group of real rank one covering the connected component of the group of isometries of $X$. By $\partial X$ we denote the geodesic boundary of $X$. The union $\bar X:= X\cup \partial X$ has the structure of a compact $G$-manifold with boundary. While $X$ parametrizes the set of maximal compact subgroups the boundary $\partial X$ parametrizes the set of parabolic subgroups of $G$. If $P\subset G$ is a parabolic subgroup, then we denote by $\infty_P\in \partial X$ the unique fixed point of $P$. Let $\Omega_P:=\partial X\setminus \infty_P$. \subsubsection{} Let $P\subset G$ be parabolic, and $N\subset P$ be its nil-radical. Then we can write $P$ as an extension \begin{equation} \label{eepp} 0\rightarrow N\rightarrow P\stackrel{l}{\rightarrow} L\rightarrow 0\ .\end{equation} Here $L$ is a reductive Lie group which is canonically isomorphic to a product $MA$ of a compact group $M$ and a group $A\cong {\mathbb{R}}_+^*$. \subsubsection{} Let $\Gamma\subset G$ be a torsion-free discrete subgroup. \begin{ddd}\label{weih111} A parabolic subgroup $P\subset G$ is called $\Gamma$-cuspidal if $U_P:=\Gamma\cap P$ is an infinite subgroup such that $l(U_P)\subset L$ is precompact, i.e. $l(U_P)\subset M$. \end{ddd} Let $p$ denote the $\Gamma$-conjugacy class of the $\Gamma$-cuspidal parabolic subgroup $P$. We call such classes cusps and say that the cusp $p$ has full rank (as opposed to smaller rank) if $U_P\backslash \Omega_P$ is compact for one (and hence for any) $P\in p$. For each cusp $p$ and $P\in p$ we form the manifold with boundary $\bar Y_{U_P}:= U_P\backslash (X\cup \Omega_P)$. \subsubsection{}\label{weih133} The boundary $\partial X$ of $X$ has a $\Gamma$-equivariant decomposition $\partial X=\Omega_\Gamma\cup\Lambda_\Gamma$ into a limit set $\Lambda_\Gamma$ and a domain of discontinuity $\Omega_\Gamma$ (\cite{MR0336648}, Prop.~8.5). Let $\bar Y_\Gamma$ denote the manifold with boundary $\bar Y_\Gamma:=\Gamma\backslash (X\cup \Omega_\Gamma)$. \begin{ddd}\label{t67} The group $\Gamma$ is called geometrically finite if the following conditions hold: \begin{enumerate} \item The set ${\mathcal{P}}_\Gamma$ of $\Gamma$-conjugacy classes of $\Gamma$-cuspidal parabolic subgroups is finite. \item There is a bijection ${\rm end}(\bar Y_\Gamma)\stackrel{\sim}{\rightarrow} {\mathcal{P}}_\Gamma$, where ${\rm end}(\bar Y_\Gamma)$ denotes the set of ends of the manifold with boundary $\bar Y_\Gamma$. \item For all $p\in{\mathcal{P}}_\Gamma$ and $P\in p$ there is a representative $\bar Y_p$ of the end corresponding to $p$ and an isometric embedding $e_P: \bar Y_p \rightarrow \bar Y_{U_P}$ such that its image $e_p(\bar Y_p)$ represents the end of $\bar Y_{U_P}$. \end{enumerate} \end{ddd} By Bowditch \cite{bowditch951}, Corollary 6.3, this definition is equivalent to the slightly weaker definition \cite{bowditch951}, F1 (which can be derived from \ref{t67} by replacing ``set ${\mathcal{P}}_\Gamma$ of $\Gamma$-conjugacy classes of $\Gamma$-cuspidal parabolic subgroups'' by ``set ${\rm end}(\bar Y_\Gamma)$ of ends of $\bar Y_\Gamma$" in 1., ``bijection" by ``map $c$" in 2. and ``all $p\in{\mathcal{P}}_\Gamma$" by ``all $p\in{\mathcal{P}}_\Gamma$ in the range of $c$" in 3.). \subsubsection{}\label{weih132} Let $Y_\Gamma:=\Gamma\backslash X$ denote the locally symmetric space associated to $\Gamma$. By $\bar Y_0$ we denote the compact subset $\bar Y_\Gamma\setminus \bigcup_{p\in{\mathcal{P}}_\Gamma} \bar Y_{p}$ of $\bar Y_\Gamma$. The boundary of $\bar Y_\Gamma$ is the manifold $B_\Gamma:=\Gamma\backslash \Omega_\Gamma$. Let ${\mathcal{P}}_\Gamma^<\subset {\mathcal{P}}_\Gamma$ denote the subset of cusps of smaller rank. The decompositions of $\bar Y_\Gamma$ into a compact piece and its ends induces a decomposition $B_\Gamma=B_0\cup \bigcup_{p\in {\mathcal{P}}^<_\Gamma} B_p$, where $B_p:=B\cap \bar Y_p$, $p\in {\mathcal{P}}_\Gamma^<\cup\{0\}$. \subsubsection{}\label{holle} Set ${\mathcal{P}}_\Gamma^{max}:={\mathcal{P}}_\Gamma\setminus{\mathcal{P}}_\Gamma^<$. It will be convenient to fix a set $\tilde{\mathcal{P}}$ of parabolic subgroups representing the elements of ${\mathcal{P}}_\Gamma$. It comes with a partition $\tilde{\mathcal{P}}=\tilde{\mathcal{P}}^<\cup\tilde{\mathcal{P}}^{max}$. \subsubsection{}\label{weih115} Our analysis will require an additional assumption on the cusps. By Lemma \ref{martin} this assumption is automatically satisfied in the cases ${H\mathbb{R}}^n$, ${H\mathbb{C}}^n$, but by Lemma \ref{contrmartin} it is non-trivial in the case ${H\mathbb{H}}^n$. We are now going to describe this assumption in detail. Let $P\subset G$ be parabolic. A Langlands decomposition $P=MAN$ is the same as a split $s:MA\rightarrow P$ of the extension (\ref{eepp}), where we identify $M$ and $A$ with their images $s(M)$ and $s(A)$. Let $p$ be a cusp of $\Gamma$ and $P\in p$. In Subsection \ref{cuspgeom} we will construct a Langlands decomposition $P=MAN$, a discrete subgroup $V\subset N$, and a homomorphism $m:N_V\rightarrow M$ from the Zariski closure $N_V$ of $V$ in $N$ such that the group $U^0:=\{m(v)v|v\in V\}$ is a subgroup of $U_P$ (see Def.~\ref{weih111}) of finite index. \begin{ddd}\label{t799} The cusp $p$ is called regular if the Langlands decomposition of $P$ above can be adjusted such that $N_V$ is invariant with respect to conjugation by $A$. We call such a Langlands decomposition adapted. \end{ddd} \begin{ass} In the remainder of the present paper we assume that $\Gamma$ is geometrically finite and that all its cusps are regular. \end{ass} \subsection{Twists and bundles} \subsubsection{}\label{weih112} We fix a parabolic subgroup $P$ of $G$ and write $\partial X=G/P$. We further fix a Langlands decomposition $P=MAN$ and let ${\mathfrak{a}}$ denote the Lie algebra of $A$. Let ${\mathfrak{n}}$ denote the Lie algebra of $N$, and let $\alpha\in{\mathfrak{a}}^*$ denote the short root of ${\mathfrak{a}}$ on ${\mathfrak{n}}$. We define $\rho\in{\mathfrak{a}}^*$ by $\rho(H):=\frac12 {\mbox{\rm tr}}({\mbox{\rm ad}}(H)_{|{\mathfrak{n}}})$, $H\in{\mathfrak{a}}$. \subsubsection{}\label{weih135} If $\lambda\in {\aaaa_\C^\ast}$ and $(\sigma,V_\sigma)$ is a representation of $M$, then we define the representation $(\sigma_\lambda,V_{\sigma_\lambda})$ of $P$ by $V_{\sigma_\lambda}:=V_\sigma$ and $\sigma_\lambda(man):=a^{\rho-\lambda}\sigma(m)$. By $1$ we denote the trivial one-dimensional representation of $M$. \subsubsection{} If $(\theta,V_\theta)$ is a representation of $P$, then we define the $G$-homogeneous bundle $V(\theta):=G\times_P V_\theta$ and denote by $\pi^\theta$ the representation of $G$ on spaces of sections of $V(\theta)$. \subsubsection{} If $({\varphi},V_{\varphi})$ is a finite-dimensional representation of $G$ (or $\Gamma$), then we denote by $V(\theta,{\varphi})$ the tensor product of $V(\theta)$ with the trivial bundle $\partial X\times V_{\varphi}$, and by $\pi^{\theta,{\varphi}}$ the representation of $G$ (or $\Gamma$) on spaces of sections of $V(\theta,\phi)$. Note that we can identify $$C^\infty(\partial X,V(\theta,{\varphi}))\cong C^\infty(\partial X,V(\theta))\otimes V_{\varphi},\quad \pi^{\theta,{\varphi}}=\pi^\theta\otimes{\varphi}\ .$$ \subsubsection{} The representation ${\varphi}$ of $\Gamma$ is called twist. Our analysis requires further assumptions on the twist going under the name``admissible''. Let $p$ be a cusp of $\Gamma$, $P\in p$, and $MAN$ be an adapted Langlands decomposition of $P$ (see Definition \ref{t799}). We define $M_U:=\overline{l(U_P)}$. Then $P_U:=M_UN_V\subset P$ is a subgroup containing $U_P$. Let $({\varphi},V_{\varphi})$ be a twist. \begin{ddd}\label{weih134} The twist $({\varphi},V_{\varphi})$ is called admissible at the cups $p$ if its restriction to $U_P$ extends to a continuous representation of $AP_U$ such that $A$ acts algebraically by semisimple endomorphisms. A twist is called admissible if it is admissible at all cusps of $\Gamma$. \end{ddd} Note that the algebraic functions on $A$ are linear combinations of $A\mapsto a^{n\alpha}$, $n\in {\mathbb{Z}}$ (see \ref{weih112} for the definition of $\alpha$). \subsubsection{} If $\Gamma$ has cusps, then the condition of admissibility excludes most unitary representations of $\Gamma$. Examples of admissible twists are restrictions of finite-dimensional representations of $G$ to $\Gamma$. As explained in Subsection \ref{ttw} twists are used as a technical device. \subsubsection{} If $(\gamma,V_\gamma)$ is a representation of $K$, then we can form the homogeneous bundle $V(\gamma):=G\times_KV_\gamma$ over $X$. For a twist ${\varphi}$ let $V(\gamma,{\varphi})$ denote the product $V(\gamma)\otimes V_{\varphi}$. Furthermore let $V_\Gamma(\gamma,{\varphi}):=\Gamma\backslash V(\gamma,{\varphi})$ denote the corresponding bundle over $Y$. \subsection{Exponents} \subsubsection{} The main numerical invariant associated to a discrete subgroup $\Gamma\subset G$ is its exponent $\delta_\Gamma\in {\mathfrak{a}}^*$. Note that each $g\in G$ can be written as $g=ka_gh\in KA_+K$, where $a_g$ is uniquely determined. Here $A_+:=\{a\in A\mid a^\alpha\ge 1\}$. \begin{ddd} The exponent $\delta_\Gamma\in{\mathfrak{a}}^*$ is defined as the infimum of the set $$\{\nu\in{\mathfrak{a}}^*\:|\:\sum_{g\in\Gamma} a_g^{-\nu-\rho}<\infty\}\ .$$ \end{ddd} If follows from the discreteness of $\Gamma$ that $\delta_\Gamma\le\rho$. \subsubsection{} The critical exponent $\delta_\Gamma$ has been extensively studied, in particular by Patterson \cite{patterson762}, Sullivan \cite{sullivan79}, \cite{MR766265}, and Corlette \cite{corlette90}, Corlette and Iozzi \cite{corletteiozzi99}. From these papers we know that $\delta_\Gamma\in [-\rho,\rho]$, if $\Gamma$ is infinite. Moreover, we have $\delta_\Gamma=\rho$ if and only if $\Gamma$ has finite covolume. If $\Lambda_\Gamma$ contains at least $2$ points, then $\delta_\Gamma+\rho=\dim_H(\Lambda)\alpha$, where $\dim_H(\Lambda)$ denotes the Hausdorff dimension of the limit set with respect to the natural class of sub-Riemannian metrics on $\partial X$. If $Y=\Gamma\backslash X$ is an infinite volume quotient of a quaternionic hyperbolic space or the Cayley hyperbolic plane, then $\delta_\Gamma$ can not be arbitrary close to $\rho$. In these cases we have $\delta_\Gamma\le (2n-1)\alpha$ and $\delta_\Gamma\le 5\alpha$, respectively \cite{corlette90},\cite{corletteiozzi99}. \subsubsection{} Let ${\varphi}$ be a twist. \begin{ddd} We define the exponent $\delta_{\varphi}\in{\mathfrak{a}}^*$ of ${\varphi}$ as the infimum of the set $$\{\nu\in{\mathfrak{a}}\:|\:\sup_{g\in\Gamma} a_g^{-\nu} \|{\varphi}(g)\|<\infty\}\ ,$$ where $\|.\|$ denotes any norm on ${\mathrm{End}}(V_{\varphi})$. \end{ddd} \subsection{Description of the main results} \subsubsection{} Let $\Gamma\subset G$ be a discrete, torsion-free, geometrically finite subgroup such that all its cusps are regular. Furthermore let ${\varphi}$ be an admissible twist. Let $B_\Gamma:=\Gamma\backslash \Omega_\Gamma$ and $V_{B_\Gamma}(\sigma_\lambda,{\varphi}):=\Gamma\backslash V(\sigma_\lambda,{\varphi})$. Under the name push-down we subsume several constructions related to the average of elements of $C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$ with respect to $\Gamma$. It is a crucial matter to construct a space $B_\Gamma(\sigma_\lambda,{\varphi})$ which contains the result of the average. If $\lambda\in{\aaaa_\C^\ast}$ varies, then these spaces assemble as a projective limit of locally trivial bundles of Fr\'echet spaces in the sense of Subsection \ref{llim}. Therefore we can speak of meromorphic families of continuous maps from and to these spaces. \subsubsection{}\label{weih123} We define the bundle $B_\Gamma(\sigma_\lambda,{\varphi})$ as a direct sum $$B_\Gamma(\sigma_\lambda,{\varphi}):=B_\Gamma(\sigma_\lambda,{\varphi})_1 \oplus {R}_{\Gamma}(\sigma_\lambda,{\varphi})_{max}\ ,$$ where the first component satisfies $$C_c^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))\subset B_\Gamma(\sigma_\lambda,{\varphi})_1\subset C^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$$ (equality occurs if all cusps have full rank). The second component is associated to the cusps of full rank and is finite-dimensional. \subsubsection{}\label{neuj913} The elements of $B_\Gamma(\sigma_\lambda,{\varphi})_1$ are characterized as smooth sections of $V_{B_\Gamma}(\sigma_\lambda,{\varphi})$ with a certain asymptotic expansion near the cusps. We postpone the detailed description of the asymptotics until Subsection \ref{asz}. The bundle ${R}_{\Gamma}(\sigma_\lambda,{\varphi})_{max}$ has a further decomposition $${R}_{\Gamma}(\sigma_\lambda,{\varphi})_{max}=\bigoplus_{P\in\tilde{\mathcal{P}}^{max}} {R}_{U_P}(\sigma_\lambda,{\varphi})\ .$$ Recall from \ref{holle} that $\tilde {\mathcal{P}}^{max}$ is in bijection with the set of cusps of full rank. In order to define ${R}_{U_P}(\sigma_\lambda,{\varphi})$ we first consider the sheaf ${\mathcal{E}}_{\infty_P}(\tilde\sigma,\tilde{\varphi})$ on ${\aaaa_\C^\ast}$ of holomorphic families $\phi_\nu$, $\nu\in{\aaaa_\C^\ast}$, of $U_P$-invariant distribution sections of $V(\tilde\sigma_{\nu},\tilde{\varphi})$ supported in $\infty_P$. Here $\tilde\sigma,\tilde{\varphi}$ are the dual representations to $\sigma,{\varphi}$. The sheaf ${\mathcal{E}}_{\infty_P}(\tilde\sigma,\tilde{\varphi})$ is the sheaf of holomorphic sections of a trivial finite-dimensional holomorphic vector bundle over ${\aaaa_\C^\ast}$, and we denote by $E_{\infty_P}(\tilde\sigma_\lambda,\tilde{\varphi})$ the fibre of this bundle over $\lambda\in{\aaaa_\C^\ast}$ (see Lemma \ref{vermaolbrich} for an alternative description of the space $E_{\infty_P}(\tilde\sigma_\lambda,\tilde{\varphi})$ and \ref{weih121} for the finite-dimensionality). Then we define $${R}_{U_P}(\sigma_\lambda,{\varphi}):= E_{\infty_P}(\tilde\sigma_{-\lambda},\tilde{\varphi})^*\ .$$ \subsubsection{} We can now state the definition of the push-down. \begin{ddd}\label{pushdowndef} For $f\in C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$ we define the push-down $\pi^\Gamma_*(f)\in B_\Gamma(\sigma_\lambda,{\varphi})$ as the direct sum of $\pi^\Gamma_*(f)_1\in B_\Gamma(\sigma_\lambda,{\varphi})_1$ and $\pi^\Gamma_*(f)_2\in {R}_{\Gamma}(\sigma_\lambda,{\varphi})_{max}$. Here $\pi^\Gamma_*(f)_1$ is given by $\pi^\Gamma_*(f)_1:=\sum_{g\in\Gamma} \pi^{\sigma_\lambda,{\varphi}}(g) f_{|\Omega_\Gamma}$ (if the sum converges). The component $\pi^\Gamma_*(f)_{2,P}\in {R}_{U_P}(\sigma_\lambda,{\varphi})$ for $P\in\tilde{\mathcal{P}}^{max}$ is defined by the condition that $$\langle \phi,\pi^\Gamma_*(f)_{2,P}\rangle = \sum_{[g]\in\Gamma/U_P}\langle \pi^{\tilde\sigma_{-\lambda},\tilde{\varphi}}(g)\phi,f \rangle$$ for all $\phi\in E_{\infty_P}(\tilde\sigma_{-\lambda},\tilde{\varphi})$ (if the sum converges). \end{ddd} \subsubsection{} In \ref{pushdowndef} the push-down is defined if certain sums converge. The first part of the following theorem gives a range of $\lambda\in {\aaaa_\C^\ast}$ for which these convergence conditions are satisfied. The second part asserts its meromorphic continuation to all of ${\aaaa_\C^\ast}$. \begin{theorem}\label{t119} If $f\in C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$, then the push-down $\pi^\Gamma_*(f)\in C^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$ converges for ${\rm Re }(\lambda)<-\delta_\Gamma-\delta_{\varphi}$. The push-down induces a meromorphic family on ${\aaaa_\C^\ast}$ of continuous maps $$\pi^\Gamma_*:C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))\rightarrow B_\Gamma(\sigma_\lambda,{\varphi})$$ with finite-dimensional singularities. \end{theorem} We will first prove the part of the theorem asserting the convergence of the push-down for ${\rm Re }(\lambda)<-\delta_\Gamma-\delta_{\varphi}$. Then we consider the adjoint of the push-down, the extension $ext^\Gamma$, and obtain the meromorphic continuation of the latter (see Theorem \ref{t110}). The remainder of Theorem \ref{t119} then follows from this result by duality. \subsubsection{}\label{weih126} By $D_\Gamma(\sigma_\lambda,{\varphi})$ we denote the topological dual space of $B_\Gamma(\tilde \sigma_{-\lambda},\tilde{\varphi})$. It is a dual Fr\'echet and Montel space. By the latter property it is reflexive so that $$D_\Gamma(\sigma_\lambda,{\varphi})^\prime\cong B_\Gamma(\tilde \sigma_{-\lambda},\tilde{\varphi})\ .$$ Varying $\lambda\in{\aaaa_\C^\ast}$ the spaces $D_\Gamma(\sigma_\lambda,{\varphi})$ form a direct limit of locally trivial holomorphic bundles of dual Fr\'echet spaces in the sense of Subsection \ref{llim}. \begin{ddd} For ${\rm Re }(\lambda)>\delta_\Gamma+\delta_{\varphi}$ the extension $$ext^\Gamma:D_\Gamma(\sigma_\lambda,{\varphi})\rightarrow C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$$ is defined to be the adjoint of $\pi^\Gamma_*:C^\infty(\partial X,V(\tilde \sigma_{-\lambda},\tilde{\varphi}))\rightarrow B_\Gamma(\tilde \sigma_{-\lambda},\tilde{\varphi})$. \end{ddd} \begin{theorem}\label{t110} The extension induces a meromorphic family of continuous maps on ${\aaaa_\C^\ast}$ $$ext^\Gamma:D_\Gamma(\sigma_\lambda,{\varphi})\rightarrow C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$$ with finite-dimensional singularities and values in ${}^\Gamma C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$. \end{theorem} \subsubsection{} The meromorphic continuation of the extension is the main goal of the present paper. We obtain the continuation in close connection with the meromorphic continuation of the scattering matrix which will be defined below. In order to define this scattering matrix we need the restriction map $res^\Gamma$ which is a left-inverse of the extension. Due to the presence of cusps the definition of the restriction map is more complicated than in \cite{MR1749869}. \subsubsection{} We first recall the definition of $res^\Gamma$ in the case of convex-cocompact groups $\Gamma$ (i.e. in the case without cusps, compare Sec.~4 of \cite{MR1749869}). In this case $$res^\Gamma:{}^\Gamma C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))\rightarrow C^{-\infty}(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$$ is given by the composition of the restriction of distributions to $\Omega_\Gamma$ and the identification ${}^\Gamma C^{-\infty}(\Omega_\Gamma,V(\sigma_\lambda,{\varphi}))\cong C^{-\infty}(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$. For the purpose of the present paper we employ a different description. \subsubsection{} Let us still assume that $\Gamma$ is convex-cocompact. We choose a cut-off function $\chi^\Gamma\in C_c^\infty(\Omega_\Gamma)$ such that $\sum_{g\in\Gamma} g^* \chi^\Gamma=1$. Then we define $$\pi^{*}: C^\infty(B_\Gamma,V_{B_\Gamma}(\tilde\sigma_{-\lambda},\tilde{\varphi})) \rightarrow C^\infty(\partial X,V(\tilde\sigma_{-\lambda},\tilde{\varphi}))$$ by $\pi^*(f)=\chi^\Gamma f$ (in order to understand the right-hand side properly one must identify sections of bundles over $B_\Gamma$ with $\Gamma$-equivariant sections on $\Omega_\Gamma$). We define $$\widetilde{res}:C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi})) \rightarrow C^{-\infty}(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$$ to be the adjoint of $\pi^*$. Then $res^\Gamma$ coincides with the restriction of $\widetilde{res}$ to the subspace ${}^\Gamma C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$. While $\widetilde{res}$ depends on the choice of $\chi^\Gamma$, the restriction $res^\Gamma$ is independent of choices. \subsubsection{} If $\Gamma$ has cusps of smaller rank, then we can not assume that $\chi^\Gamma$ has compact support. The definition of $\pi^*$ breaks down. Our way arround that problem is as follows. For each $k\in{\mathbb {N}}_0$ we consider the Banach spaces $C^k(\partial X,V(\tilde\sigma_{-\lambda},\tilde{\varphi}))$ and define a meromorphic family of maps $\pi^*_k:B_\Gamma(\tilde\sigma_{-\lambda},\tilde{\varphi})\rightarrow C^k(\partial X,V(\tilde\sigma_{-\lambda},\tilde{\varphi}))$. The map $\pi^*_k$ is a right-inverse of the push-down $\pi^\Gamma_*$, and its definition is similar in spirit to that of $\pi^*$ above, but the details are more complicated since we have to take into account the asymptotic expansions of the elements of $B_\Gamma(\tilde\sigma_{-\lambda},\tilde{\varphi})$ near cusps. One problem is that we can only deal with a finite number (depending on $k$) of these terms at a time in order to define $\pi_k^*$. The situation is similar to the problem of the construction of smooth functions with given Taylor series at a given point. It is impossible to construct a continuous map from the space of formal power series to smooth functions which is right-inverse to the surjective map which takes the Taylor series. Back to the definition of the restriction, let $C^{-k}(\partial X,V(\sigma_\lambda,{\varphi}))$ denote the distributions of order $k$, i.e. the topological dual of $C^k(\partial X,V(\tilde\sigma_{-\lambda},\tilde{\varphi}))$, and define $res_k^\Gamma:C^{-k}(\partial X,V(\sigma_\lambda,{\varphi}))\rightarrow D_\Gamma(\sigma_\lambda,{\varphi})$ as the adjoint of $\pi^*_k$. The collection of maps $res_k^\Gamma$ play the role of the map $\widetilde{res}$ above. These maps depend on choices and are, in particular, not compatible if we vary $k$. \subsubsection{} Nevertheless, we have the following uniqueness property. If $f_\nu\in {}^\Gamma C^{-\infty}(\partial X,V(\sigma_\nu,{\varphi}))$, $\nu\in{\aaaa_\C^\ast}$, is a meromorphic family, and $W\subset {\aaaa_\C^\ast}$ is compact, then for sufficiently large $k$ the meromorphic family $res^\Gamma_k(f_\nu)$, $\nu\in W$, is well-defined. In fact, due to the compactness of $W$ the order of $f_\nu$ as a distribution is uniformly bounded for $\nu\in W$. In \ref{neuj3001} we will see that $res^\Gamma_k(f_\nu)$ is independent of the choices. \subsubsection{} In general the restriction maps $res^\Gamma_k$ may have poles. Moreover, not every element of ${}^\Gamma C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$ can be written as evaluation of a holomorphic family $f_\nu\in {}^\Gamma C^{-\infty}(\partial X,V(\sigma_\nu,{\varphi}))$ defined near $\lambda$. But for generic $\lambda\in {\aaaa_\C^\ast}$ the restriction map $res^\Gamma_k$ is regular and every element $f\in {}^\Gamma C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$ extends to a family (in the present paper we will prove a weaker statement only, which suffices for our purposes). In this case $res^\Gamma_k(f)$ is well-defined independent of the choice of (the sufficiently large) $k$, and we can omit the subscript $k$ and write $res^\Gamma(f)$ for $res^\Gamma_k(f)$. We have the identity $$res^\Gamma\circ ext^\Gamma={\mathrm {id}}_{D_\Gamma(\sigma_\lambda,{\varphi})}\ ,$$ which holds true for generic $\lambda\in {\aaaa_\C^\ast}$ or as an identity of maps defined on meromorphic families (see Lemma \ref{weih114}). \subsubsection{} If $f\in {}^\Gamma C^{-k}(\partial X,V(\sigma_\lambda,{\varphi}))$, $k\in{\mathbb {N}}_0$, then $f_{|\Omega_\Gamma}$ is a $\Gamma$-invariant distribution on $\Omega_\Gamma$, hence can be considered as an element of $C^{-\infty}(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$. If $$i:D_\Gamma(\sigma_\lambda,{\varphi}) \rightarrow C^{-\infty}(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$$ is the natural map (adjoint to the inclusion $C_c^\infty(B,V_B(\tilde\sigma_{-\lambda},\tilde{\varphi})) \hookrightarrow B_\Gamma(\tilde\sigma_{-\lambda},\tilde{\varphi})$), then $i\circ res^\Gamma(f)$ is defined (even though $res^\Gamma$ may have a pole at $\lambda$) and coincides with $f_{|\Omega_\Gamma}$. In particular, the condition $i\circ res^\Gamma(f)=0$ is equivalent to ${\mathrm{supp}}(f)\subset \Lambda_\Gamma$. If $\Gamma$ has cusps, then $i$ is not injective. In this case the condition ${\mathrm{supp}}(f)\in \Lambda_\Gamma$ does not imply that $res^\Gamma(f)=0$ (provided the latter is defined). \subsubsection{} Vanishing of $res^\Gamma(f)$ is a rather strong condition and should play the role of the condition ${\mathrm{supp}}(f)\subset \Lambda_\Gamma$ in the convex-cocompact case which was successfully exploited in \cite{MR1749869} and \cite{MR1689342}. One application of this condition is the following result which is employed in proving the functional equation of the scattering matrix in the domain of convergence. \begin{theorem}[Corollary \ref{wieder}]\label{micro} Assume that ${\rm Re }(\lambda)>\max\left(\{\delta_\Gamma\}\cup\{\rho_{U_P}\mid P\in\tilde{\mathcal{P}}\}\right)+\delta_{\varphi}$. If $f_\mu\in {}^\Gamma C^{-\infty}(\partial X,V(1_\mu,{\varphi}))$ is a germ of a meromorphic family at $\lambda$, then $ext^\Gamma\circ res^\Gamma(f_\mu) = f_\mu$. \end{theorem} Further applications of the condition $res^\Gamma(f)=0$ are contained in \cite{MR1926489} (non-existence of such $f$ for ${\rm Re }(\lambda)$ large) and \cite{math.DG/0103144} (estimates of the regularity of $f$). In fact, the proof of Theorem \ref{micro} is based on the main result of \cite{MR1926489}. \subsubsection{} We now turn to the scattering matrix. We assume that $\sigma$ is either a Weyl-invariant irreducible representation of $M$ or of the form $\sigma^\prime\oplus(\sigma^\prime)^w$, where $\sigma^\prime$ is irreducible and not Weyl-invariant, and $(\sigma^\prime)^w(m):=\sigma^\prime(w^{-1}mw)$ is the Weyl-conjugate representation of $\sigma^\prime$. Here $w\in N_K(A)$ is a representative of the non-trivial element of the Weyl group $W=N_K(A)/M\cong {\mathbb{Z}}/2{\mathbb{Z}}$. \subsubsection{}\label{gaa101} Our starting point is the scattering matrix $$S^{\{1\}}_\lambda=J_\lambda:C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi})) \rightarrow C^{-\infty}(\partial X,V(\sigma_{-\lambda},{\varphi}))$$ associated to the trivial group $\{1\}$. In representation theory it firms under the name Knapp-Stein intertwining operator. Here we employ a suitably normalized version as in the paper \cite{MR1749869}, Sec. 5, to which we refer for further details. Let $I_{\mathfrak{a}}\subset{\mathfrak{a}}^*$ be the ${\mathbb{Z}}$-module spanned by $\alpha$ if $2\alpha$ is a root of $({\mathfrak{g}},{\mathfrak{a}})$, and by $\frac{1}{2}{\alpha}$ otherwise. The operators $J_\lambda$ form a meromorphic family of continuous maps with singularities in $I_{\mathfrak{a}}$ and satisfy the functional equation $J_\lambda\circ J_{-\lambda}={\mathrm {id}}$. \subsubsection{} If $\Gamma$ is non-trivial, then we obtain the scattering matrix $S_\lambda^\Gamma$ from $J_\lambda$ using restriction and extension. \begin{ddd} For ${\rm Re }(\lambda)>\delta_\Gamma+\delta_{\varphi}$ we define the scattering matrix $$S^\Gamma_\lambda:D_\Gamma(\sigma_\lambda,{\varphi})\rightarrow D_\Gamma(\sigma_{-\lambda},{\varphi})$$ as the meromorphic family of continuous maps $$S^\Gamma_\lambda:=res^\Gamma\circ J_\lambda\circ ext^\Gamma\ .$$ \end{ddd} \begin{theorem}\label{t113} The scattering matrix has a meromorphic continuation to all of ${\aaaa_\C^\ast}$. It satisfies the functional equation $S^\Gamma_\lambda\circ S^\Gamma_{-\lambda}={\mathrm {id}}$ and the relation $J_\lambda\circ ext^\Gamma=ext^\Gamma\circ S^\Gamma_\lambda$. \end{theorem} As noted above this theorem is proved in a multistep procedure which involves the meromorphic continuation of $ext^\Gamma$ at the same time. The basic ideas are adapted from \cite{MR1749869}. The non-compactness of $B_\Gamma$ due to the presence of cusps of non-maximal rank is responsible for the various complications which have to be resolved on the way. \subsubsection{} The following application to the Eisenstein series is an easy consequence of the preceding results and can be derived exactly as in \cite{MR1749869}, Cor.~10.2. Let $\gamma$ be a finite-dimensional representation of $K$ and $T\in {\mbox{\rm Hom}}_M(V_\sigma,V_\gamma)$. Then we have a holomorphic family of $G$-equivariant maps $$P^T_\lambda: C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))\rightarrow C^\infty(X,V(\gamma,{\varphi}))$$ which are called Poisson transformations. We refer to \cite{MR1749869}, Def. 4.8, for a definition of the Poisson transformation and to \cite{MR1749869}, Sec. 6, for a discussion of its properties. \begin{ddd}\label{neuj1001} If $\lambda\in{\aaaa_\C^\ast}$, $f\in D_\Gamma(\sigma_\lambda,{\varphi})$, $ext^\Gamma(f)$ is regular, and $T\in {\mbox{\rm Hom}}_M(V_\sigma,V_\gamma)$, then we define the Eisenstein series $E(\lambda,f,T)\in C^\infty(Y,V_Y(\gamma,{\varphi}))$ by $P^T_\lambda\circ ext^\Gamma(f)$. \end{ddd} In the special case that $\sigma$, $\gamma$, and ${\varphi}$ are trivial, $T$ is the identity, ${\rm Re }(\lambda)>\delta_\Gamma$, and $f=\delta_b$ is the delta-distribution located at $b\in B_\Gamma$, this definition coincides with the classical definition of the Eisenstein series as the $\Gamma$-average of the Poisson kernel, i.e. the integal kernel of $P^{{\mathrm {id}}}_\lambda$. Let $c_\sigma$ and $c_\gamma$ be the $c$-functions introduced in \cite{MR1749869}, Sec. 5. We further employ the notation $(c_\gamma(\lambda) T)^w$ as introduced in the same section of \cite{MR1749869}. \begin{kor}\label{main} The Eisenstein series forms a meromorphic family of continuous maps with finite-dimensional singularities $$E(\lambda,.,T): D_\Gamma(\sigma_\lambda,{\varphi}) \rightarrow C^\infty(Y,V_Y(\gamma,{\varphi}))\ .$$ It satisfies the functional equation $$E(\lambda,S_{-\lambda} f,T)=E(-\lambda,f,(\frac{c_\gamma(\lambda)}{c_\sigma(\lambda)}T)^w)\ .$$ \end{kor} \section{Some machinery} \subsection{Limits of bundles}\label{llim} \subsubsection{} In the present paper a main issue is the construction of families of topological vector spaces $V_\lambda$, $\lambda\in{\mathbb {C}}$, and the study of holomorphic (meromorphic) families of vectors $f_\lambda\in V_\lambda$ or holomorphic (meromorphic) families of homomorphisms $h_\lambda\in {\mbox{\rm Hom}}(V_\lambda,W_\lambda)$ between such families of spaces. In order to make the notion of a holomorphic family precise we must relate the spaces $V_\lambda$ for neighbouring $\lambda$ in some holomorphic manner. One way to do this is to equip the family of spaces with the structure of a locally trivial holomorphic bundle of topological vector spaces. If we fix two local trivializations $U_0\times W_0$, $U_1\times W_1$ of the family $V_\lambda$ with non-trivial overlap, then the transition function is a holomorphic map from $U_0\cap U_1$ to ${\mbox{\rm Hom}}(W_0,W_1)$ with values in the subspace of isomorphisms. \subsubsection{} All topological vector spaces in the present paper will be locally convex. Spaces of homomorphisms between topological vector spaces will always be equipped with the topology of uniform convergence on bounded sets. We refer to \cite{MR1689342}, Sec.2.2 for more details. If we have two holomorphic families $h_\lambda\in{\mbox{\rm Hom}}(W_0,W_1)$ and $g_\lambda\in{\mbox{\rm Hom}}(W_1,W_2)$, then we can consider the composition $g_\lambda\circ h_\lambda\in {\mbox{\rm Hom}}(W_0,W_2)$. This composition is again holomorphic (see \cite{gloeckner}). If the evaluation $W_0\to W^{\prime\prime}_0$ is continuous, then the adjoint $h_\lambda^\prime\in {\mbox{\rm Hom}}(W_1^\prime,W_0^\prime)$ is holomorphic, too (see again \cite{gloeckner}). In order to show that the composition of two holomorphic families of maps is again holomorphic in \cite{MR1689342}, Sec.2.2, we assumed that $W_0$ is a Montel space. The discussion of \cite{gloeckner} shows that this assumption can be omitted. \subsubsection{} In this paper we do not show that the spaces $B_\Gamma(\sigma_\lambda,{\varphi})$ ($D_\Gamma(\sigma_\lambda,{\varphi})$) form locally trivial bundles of Fr\'echet (dual Fr\'echet) spaces. These families of spaces arise instead as projective (symbol $\lim$) and inductive limits (symbol $\mathrm{colim}$) of locally trivial bundles, respectively. E.g. the bundle $B_U(\sigma_\lambda,{\varphi})$ comes as a limit $\lim B_{\Gamma,k}(\sigma_\lambda,{\varphi})$ of trivial bundles $B_{\Gamma,k}(\sigma_\lambda,{\varphi})$, $k\in {\mathbb {N}}_0$. The connecting maps (in the example $B_{\Gamma,k}(\sigma_\lambda,{\varphi})\to B_{\Gamma,k+1}(\sigma_\lambda,{\varphi})$) of the diagrams are holomorphic families of continuous maps which are injective and have dense range (resp. are injective). But they are not compatible with the trivializations. \subsubsection{}\label{weih211} We must extend the notions of a holomorphic or meromorphic family of continuous of maps between such limits of locally trivial holomorphic bundles. Let $$\dots\subset E_{n+1}\subset E_{n} \subset\dots\ ,$$ $$\dots\subset F_{n+1}\subset F_{n} \subset\dots$$ be decreasing families of locally trivial bundles of Fr\'echet spaces defined over some open subset $U\subset {\mathbb {C}}$ such that all inclusion maps are holomorphic. Let $E:=\lim E_n$, $F:=\lim F_n$. \begin{ddd} A family of maps $\phi_z:E_z\rightarrow F_z$, $z\in U$, is called holomorphic (meromorphic) if for each $x\in U$ and $n\in {\mathbb {N}}_0$ there exists a neighbourhood $U_{x,n}\subset U$ of $x$, $m(n)\in{\mathbb {N}}$, and a holomorphic (meromorphic) bundle map $\Phi_n: (E_{m(n)})_{|U_{x,n}}\rightarrow (F_{n})_{|U_{x,n}}$ such that $\phi_y$ is the restriction of $\Phi_n(y)$ to $E_y$ for all $y\in U_{n,x}$. \end{ddd} The composition of two holomorphic families is again a holomorphic family (compare \cite{MR1749869}, Subsection 2.2). \subsubsection{} We now consider the dual situation. Let $$\dots \rightarrow E^\prime_n\rightarrow E^\prime_{n+1}\rightarrow\dots\ ,$$ $$\dots \rightarrow F^\prime_n\rightarrow F^\prime_{n+1}\rightarrow\dots$$ be direct systems of holomorphic locally trivial bundles of dual Fr\'echet spaces defined over some open subset $U\subset {\mathbb {C}}$. Let $E^\prime:=\mathrm{colim} E^\prime_n$, $F^\prime:=\mathrm{colim} F^\prime_n$. \begin{ddd} A family of maps $\phi_z:F_z^\prime\rightarrow E_z^\prime$, $z\in U$, is called holomorphic (meromorphic) if for each $x\in U$ and $n\in{\mathbb {N}}_0$ there exists a neighbourhood $U_{x,n}\subset U$ of $x$, $m(n)\in{\mathbb {N}}$, and a holomorphic (meromorphic) bundle map $\Phi_n^\prime: (F^\prime_{n})_{|U_{x,n}}\rightarrow (E^\prime_{m(n)})_{|U_{x,n}}$ such that for each $y\in U_{x,n}$ the restriction of $\phi^\prime_y$ to $(F^\prime_n)_y$ is the composition of $(\Phi_n^\prime)_y$ with the natural map $(E^\prime_{m(n)})_y\rightarrow E^\prime_y$ . \end{ddd} Again the composition of two such families is a holomorphic (meromorphic) family. The adjoint of a holomorphic family is again a holomorphic (meromorphic) family. \subsection{Embedding}\label{embedd} \subsubsection{} The symmetric spaces $X$ considered in the present paper belong to a series of symmetric spaces. Let $X^n$ denote the $n$'th space of the series. We will use the superscript ${}^n$ as a decoration of symbols for other objects in order to indicate that they are associated to $X^n$. For a number of arguments we need that $\delta_\Gamma^n$ is sufficiently negative. Note that $\delta_\Gamma^n\to-\infty$ as $n\to\infty$. E.g., we will first obtain a meromorphic continuation of $ext^{\Gamma,n+k}$ for sufficiently large $k$, and then use the propositions below in order to conclude that $ext^{\Gamma,n}$ has a meromorphic continuation, too. \subsubsection{} The main point of this subsection is to explain that the concept of embedding is compatible with the function and distribution spaces introduced in \ref{weih123} and \ref{weih126}. Furthermore we need compatibility with the push-down and extension maps. Note that the concept of embedding is only applied to the spherical case $\sigma=1$. \subsubsection{}\label{neuj700} We have embeddings $X^n\hookrightarrow X^{n+1}$ and $i:\partial X^n\hookrightarrow \partial X^{n+1}$. In order to have compatible embeddings $G^n\hookrightarrow G^{n+1}$ of the groups we assume at this point that $G^n$ is one of $$\{Spin(1,n),SO(1,n)_0,SU(1,n),Sp(1,n)\}\ .$$ If we realize the last three groups of the list using ${\mathbb{F}}$-valued matrices, ${\mathbb{F}}\in\{{\mathbb{R}},{\mathbb {C}},{\mathbb{H}}\}$, then the embedding $G^n\hookrightarrow G^{n+1}$ is the usual embedding into the left upper corner. We have compatible Iwasawa decompositions such that $A=A^n=A^{n+1}$, and in particular $P^n\hookrightarrow P^{n+1}$. Let $\zeta$ be $\alpha/2$, $\alpha$, and $2\alpha$ in the cases $X={H\mathbb{R}}^n$, $X={H\mathbb{C}}^n$, and $X={H\mathbb{H}}^n$, respectively. Then $\zeta=\rho^{n+1}-\rho^n$, and we have $(V_{1^{n+1}_\lambda})_{|P^n}=V_{1^n_{\lambda-\zeta}}$, and hence $V(1^{n+1}_\lambda,{\varphi})_{|\partial X^n}=V(1^n_{\lambda-\zeta},{\varphi})$. \subsubsection{}\label{weih301} Let $$i^*:C^\infty(\partial X^{n+1},V(1^{n+1}_\lambda,{\varphi}))\rightarrow C^\infty(\partial X^n, V(1^n_{\lambda-\zeta},{\varphi}))$$ be the restriction of sections. It is a $\Gamma$-equivariant and surjective continuous linear map. \begin{prop}\label{rrttee} There exists a meromorphic family of maps $$i^*_\Gamma:B_\Gamma(1^{n+1}_\lambda,{\varphi})\rightarrow B_\Gamma(1^n_{\lambda-\zeta},{\varphi})$$ such that the following diagram commutes: $$\begin{array}{ccc} C^\infty(\partial X,V(1^{n+1}_\lambda,{\varphi})) &\stackrel{i^*}{\rightarrow}&C^\infty(\partial X^n, V(1^n_{\lambda-\zeta},{\varphi}))\\ \downarrow \pi^{\Gamma,n+1}_*&&\downarrow \pi^{\Gamma,n}_*\\ B_\Gamma(1^{n+1}_\lambda,{\varphi})&\stackrel{i_\Gamma^*}{\rightarrow }& B_\Gamma(1^n_{\lambda-\zeta},{\varphi}) \end{array}\ .$$ If $\Gamma$ does not have cusps of full rank, then $i^*_\Gamma$ is holomorphic. \end{prop} This proposition will be proved in various stages. First we consider a version for the spaces associated to the cusps, see Lemma \ref{klopp}. The global result is stated in Lemma \ref{comppp}. \subsubsection{} In the following we explain the construction of $i_\Gamma^*$. If all cusps of $\Gamma$ as a subgroup of $G^n$ have smaller rank, then $i^*_\Gamma$ is induced by the ususal restriction $i^*_\Gamma: C^\infty(B_\Gamma,V_{B_\Gamma}(1_\lambda^{n+1},{\varphi}))\rightarrow C^\infty(B_\Gamma,V_{B_\Gamma}(1^n_{\lambda-\zeta},{\varphi}))$, and both $\pi^{\Gamma,n}_*$ and $\pi^{\Gamma,n+1}_*$ are given as the average of sections over $\Gamma$. The relation $i^*_\Gamma\circ \pi^{\Gamma,n+1}_*=\pi^{\Gamma,n}_*\circ i^*$ is obvious in the domain of convergence ${\rm Re }(\lambda)>\delta^{n+1}_\Gamma+\delta_{\varphi}$, and follows by meromorphic continuation for all $\lambda\in{\aaaa_\C^\ast}$. We postpone the proof of the fact that $i^*_\Gamma$ really maps $B_\Gamma(1^{n+1}_\lambda,{\varphi})$ to $B_\Gamma(1^{n}_{\lambda-\zeta},{\varphi})$ until these spaces are defined. \subsubsection{} If $\Gamma$ has cusps of full rank (when considered as a subgroup of $G^n$), then $i^*_\Gamma$ is the sum of two maps $(i^*_\Gamma)_1:B_\Gamma(1^{n+1}_\lambda,{\varphi}) \rightarrow B_\Gamma(1^n_{\lambda-\zeta},{\varphi})_1$ and $(i^*_\Gamma)_2: B_\Gamma(1^{n+1}_\lambda,{\varphi})\rightarrow {R}_\Gamma(1^n_{\lambda-\zeta},{\varphi})_{max}$. The map $(i^*_\Gamma)_1$ is the restriction of smooth sections, and we can apply the argument above in order to show commutativity of the part of the commutative diagram involving $(\pi^{\Gamma,n}_*)_1$ and $(i^*_\Gamma)_1$. \subsubsection{} We define $(i^*_\Gamma)_2$ such that the corresponding diagram is commutative. We choose $k\in{\mathbb {N}}_0$ so large that $E_{\infty_{P^n}}(1^n_{-\lambda+\zeta},\tilde{\varphi}) \subset C^{-k}(\partial X^n,V(1^n_{-\lambda+\zeta},\tilde{\varphi}))$ for all $P^n\in\tilde{\mathcal{P}}^{max,n}$. For $f\in B_\Gamma(1^{n+1}_\lambda,{\varphi})$ the component $i^*_\Gamma(f)_{2,P^n} \in {R}^{U_{P^n}}(1^n_{\lambda-\zeta},{\varphi})$ is characterized by $$\langle\phi,i^*_\Gamma(f)_{2,P^n}\rangle=\langle res^{U_{P^{n+1}},n+1}_k\circ i_*(\phi), T_{P^{n+1}}(f) \rangle$$ for all $\phi\in E_{\infty_{P^n}}(1^n_{-\lambda+\zeta},\tilde{\varphi})$, where $P^{n+1}\subset G^{n+1}$ is the unique parabolic subgroup containing $P^n$, and $$T_{P^{n+1}}:C^\infty(B^{n+1}_\Gamma,V_{B_\Gamma}(1^{n+1}_\lambda,{\varphi}))\rightarrow C^\infty(B^{n+1}_{U_{P^{n+1}}},V_{B^{n+1}_{U_{P^{n+1}}}}(1^{n+1}_\lambda,{\varphi}))$$ is the natural map defined using the map $e_{P^{n+1}}$ and a cut-off function $\chi_p\in C^\infty(B_\Gamma)$ which is supported in $B_p$ and one on a neighbourhood of infinity of $B_p$. This definition is independent of the choice of $k$. In order to show that $(i^*_\Gamma)_2\circ \pi^{\Gamma,n+1}_*=(\pi^{\Gamma,n}_*)_2\circ i^*$ we compute in the domain of convergence for generic $\lambda$ (so that the restrictions are defined) \begin{eqnarray*} \langle\phi,(i^*_\Gamma)_2\circ \pi^{\Gamma,n+1}_*(f)\rangle&=& \langle res^{U_{P^{n+1}},n+1}_k\circ i_*(\phi),T^*_{P^{n+1}}\circ \pi^{\Gamma,n+1}_*(f)\rangle\\ &=&\sum_{g\in\Gamma^{P^{n+1}}} \langle res^{U_{P^{n+1}},n+1}_k\circ i_*(\phi),\chi_{P^{n+1}}\circ \pi^{U_{P^{n+1}},n+1}_*(\pi^{1^{n+1}_\lambda,{\varphi}}(g)f)\rangle\\ &\stackrel{(*)}{=}&\sum_{g\in\Gamma^{P^{n+1}}} \langle i_*(\phi), \pi^{1^{n+1}_\lambda,{\varphi}}(g)f\rangle\\ &=&\sum_{g\in\Gamma^{P^{n+1}}} \langle \pi^{1^{n+1}_{-\lambda+\zeta},\tilde{\varphi}}(g^{-1}) \phi, i^*(f)\rangle\\ &=&\langle \phi,(\pi^{\Gamma,n}_*)_2\circ i^*(f)\rangle\ , \end{eqnarray*} where $\Gamma^{P^{n+1}}$ denotes any system of representatives of $U_{P^{n+1}}\backslash \Gamma$, and $\chi_{P^{n+1}}:= e_{P^{n+1}}^* \chi_p$ (see Def.~\ref{t67} for the map $e_p$). In order to get the equality marked by $(*)$ we use that $$(1- \chi_{P^{n+1}})\circ res^{U_{P^{n+1}},n+1}_k \circ i_*(\phi)=0$$ and $$ext^{U_{P^{n+1}},n+1}\circ res_k^{U_{P^{n+1}},n+1}\circ i_*(\phi)=i_*(\phi)\ .$$ \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Let $i_*$ and $i_*^\Gamma$ denote the dual maps to $i^*$ and $i^*_\Gamma$. Dualizing Prop.~\ref{rrttee} we obtain the following corollary. \begin{kor}\label{rrttee1} We have a commutative diagram of meromorphic families of maps $$\begin{array}{ccc} C^{-\infty}(\partial X,V(1^{n}_\lambda,{\varphi})) &\stackrel{i_*}{\rightarrow}&C^{-\infty}(\partial X^{n+1}, V(1^{n+1}_{\lambda-\zeta},{\varphi}))\\ \uparrow ext^{\Gamma,n}_*&&\uparrow ext^{\Gamma,n+1}_*\\ D_\Gamma(1^{n}_\lambda,{\varphi})&\stackrel{i^\Gamma_*}{\rightarrow }&D_\Gamma(1^{n+1}_{\lambda-\zeta},{\varphi}) \end{array}\ .$$ \end{kor} \subsubsection{} \begin{lem}\label{jjjjjjjjjjjtwzew} There exists holomorphic families of continuous maps $$j^*:C^{-\infty}(\partial X^{n+1},V(1^{n+1}_{\lambda-\zeta},{\varphi}))\rightarrow C^{-\infty}(\partial X^n, V(1^n_\lambda,{\varphi}))$$ and $$j_*: C^{\infty}(\partial X^n, V(1^n_{-\lambda},\tilde{\varphi}))\rightarrow C^{\infty}(\partial X^{n+1},V(1^{n+1}_{-\lambda+\zeta},\tilde{\varphi}))$$ such that $j^*\circ i_* ={\mathrm {id}}$, $i^*\circ j_*={\mathrm {id}}$. \end{lem} {\it Proof.$\:\:\:\:$} We choose a tubular neighbourhood $T:{\mathbb{F}}\times\partial X^n\hookrightarrow \partial X^{n+1}$ of $\partial X^n$ such that $T(\{0\}\times \partial X^n)=\partial X^n$. Furthermore we choose a cut-off function $\chi\in C_c^\infty({\mathbb{F}})$ such that $\chi(0)=1$. Then we define the map $t: C^{\infty}(\partial X^n, V(1^n_{\rho^n},\tilde{\varphi}))\rightarrow C^{\infty}(\partial X^{n+1},V(1^{n+1}_{\rho^{n+1}},\tilde{\varphi}))$ setting $(tf)(T(u,x)):=\chi(u)f(x)$ and extending this function by zero outside the tubular neighbourhood. Here we use the canonical identifications $C^{\infty}(\partial X^n, V(1^n_{\rho^n},\tilde{\varphi}))=C^{\infty}(\partial X^n)\otimes V_{\tilde{\varphi}}$ and $C^{\infty}(\partial X^{n+1},V(1^{n+1}_{\rho^{n+1}},{\varphi}))=C^{\infty}(\partial X^{n+1})\otimes V_{\tilde{\varphi}}$. We choose a positive section $s_{n+1}\in C^{\infty}(\partial X^{n+1},V(1^{n+1}_{\rho^{n+1}+\alpha}))$ and put $i^* s_{n+1}=:s_n\in C^{\infty}(\partial X^n, V(1^n_{\rho^n+\alpha}))$. Then we define $\Phi_{\rho^n}:C^{\infty}(\partial X^n, V(1^n_{-\lambda},\tilde{\varphi})) \rightarrow C^{\infty}(\partial X^n, V(1^n_{\rho^n},\tilde{\varphi}))$ and $\Phi_{\rho^{n+1}}: C^{\infty}(\partial X^{n+1},V(1^{n+1}_{-\lambda+\zeta},\tilde{\varphi}))\rightarrow C^{\infty}(\partial X^{n+1},V(1^{n+1}_{\rho^{n+1}},\tilde{\varphi}))$ as multiplication by $s_n^{(\rho^n+\lambda)/\alpha}$ and $s_{n+1}^{(\rho^{n+1}+\lambda-\zeta)/\alpha}$, respectively. We define $j_*:=\Phi_{\rho^{n+1}}^{-1}\circ t\circ \Phi_{\rho^n}$, and $j^*$ as the adjoint of $j_*$. It is now easy to check that $j_*$ and $j^*$ have the required properties. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{}\label{ohu} If $G^n$ does not belong to the list $\{Spin(1,n),SO(1,n)_0,SU(1,n),Sp(1,n)\}$, then there is a finite covering $p:\tilde G^n\rightarrow G^n$ with $\tilde G^n\in \{Spin(1,n),SO(1,n)_0,SU(1,n),Sp(1,n)\}$. In this case can find a normal subgroup $\Gamma^0\subset \Gamma$ of finite index and a discrete subgroup $\tilde \Gamma^0\subset \tilde G^n$ such that $p$ induces an isomorphism from $\tilde \Gamma^0$ to $\Gamma^0$. Indeed, using Selberg's Lemma we can take a torsion-free subgroup $\tilde\Gamma^0$ of $p^{-1}(\Gamma)$ of finite index and set $\Gamma^0:=p(\tilde\Gamma^0)$. We can apply the concept of embedding to the subgroup $\tilde\Gamma^0$. In order to transfer results for $\tilde\Gamma^0$ to $\Gamma$ we use averages over the finite group $\Gamma/\Gamma^0$. \subsection{Twisting}\label{ttw} \subsubsection{} Twisting is an important technical device of the present paper. We explain in \ref{weih125} and \ref{weih1266} how this concept is used. But first we must introduce some notation. \subsubsection{}\label{weih122} If $(\pi,V_\pi)$ is a finite-dimensional representation of $G$, then we can form the bundles $V(\sigma_\lambda\otimes\pi,{\varphi})$ and $V(\sigma_\lambda,\pi\otimes {\varphi})$. There is an isomorphism $T:V(\sigma_\lambda\otimes\pi,{\varphi})\stackrel{\sim}{\rightarrow} V(\sigma_\lambda,\pi\otimes {\varphi})$, which is given on the level of sections by $T:C^\infty(\partial X,V(\sigma_\lambda\otimes\pi,{\varphi})) \stackrel{\sim}{\rightarrow} C^\infty(\partial X,V(\sigma_\lambda,\pi\otimes{\varphi}))$, $T(f)(g):=\pi(g)f(g)$. \subsubsection{}\label{weih144} Given an irreducible representation $\sigma$ of $M$ and $\mu_0\in{\mathfrak{a}}^*$ there exists a finite-dimensional representation $(\pi_{\sigma,\mu},V_{\pi_{\sigma,\mu}})$ of $G$ with highest $A$-weight $\mu\ge\mu_0$ such that $V_\sigma=V_{\pi_{\sigma,\mu}}(\mu)$ as representations of $M$. Here $V_{\pi_{\sigma,\mu}}(\mu)$ denotes the subspace of $V_{\pi_{\sigma,\mu}}$ on which $A$ acts with weight $\mu$. Note that $V_{\pi_{\sigma,\mu}}(-\mu)= V_{\sigma^w}$ as representations of $M$. \subsubsection{}\label{weih241} If $\sigma$ is a representation of $M$ of the form $\sigma^\prime\oplus (\sigma^\prime)^w$, where $\sigma^\prime$ irreducible and not equivalent to its Weyl conjugate, then given $\mu_0\in{\mathfrak{a}}^*$ there exist $(\pi_{\sigma^\prime,\mu},V_{\pi_{\sigma^\prime,\mu}})$ and $(\pi_{(\sigma^\prime)^w,\mu},V_{\pi_{(\sigma^\prime)^w,\mu}})$ for suitable $\mu\ge \mu_0$. In this case we set $\pi_{\sigma,\mu}:=\pi_{\sigma^\prime,\mu} \oplus \pi_{(\sigma^\prime)^w,\mu}$. In connection with twisting without further notice we will always assume that $\sigma$ is Weyl-invariant and either irreducible or of the form $\sigma^\prime\oplus (\sigma^\prime)^w$, where $\sigma^\prime$ is irreducible and not Weyl invariant. Note that $\pi_{\tilde\sigma,\mu}=\tilde\pi_{\sigma,\mu}$. \subsubsection{}\label{weih142} We have an embedding of $P$-modules $V_{\sigma_\lambda}\hookrightarrow V_{1_{\lambda+\mu}}\otimes V_{\pi_{\sigma,\mu}}$ and a corresponding embedding of $\Gamma$-equivariant bundles $V(\sigma_\lambda,{\varphi})\hookrightarrow V(1_{\lambda+\mu}\otimes\pi_{\sigma,\mu},{\varphi})$. Composing this embedding with the isomorphism $T$ (introduced in \ref{weih122}) we obtain the $\Gamma$-equivariant embedding $$i_{\sigma,\mu}:C^{\pm\infty}(\partial X,V(\sigma_\lambda,{\varphi}))\hookrightarrow C^{\pm\infty}(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))\ .$$ Similarly we have a projection of $P$-modules $V_{1_{\lambda-\mu}}\otimes V_{\pi_{\sigma,\mu}} \rightarrow V_{\sigma_{\lambda}}$ and a corresponding projection of $\Gamma$-equivariant bundles $V(1_{\lambda-\mu}\otimes\pi_{\sigma,\mu},{\varphi})\rightarrow V(\sigma_{\lambda},{\varphi})$. Composing this projection with the inverse of $T$ we obtain the $\Gamma$-equivariant projection $$p_{\sigma,\mu}:C^{\pm\infty}(\partial X,V(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes{\varphi})) \rightarrow C^{\pm\infty}(\partial X,V(\sigma_{\lambda},{\varphi}))\ .$$ \subsubsection{}\label{neuj1000} Since $i_{\sigma,\mu}$ and $p_{\sigma,\mu}$ are induced by homomorphisms of $\Gamma$-equivariant bundles restriction to $\Gamma$-equivariant sections over $\Omega_\Gamma$ provides the embedding $$i^\Gamma_{\sigma,\mu}:C^{\infty}(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))\hookrightarrow C^{\infty}(B_\Gamma,V_{B_\Gamma}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))$$ and the projection $$p^\Gamma_{\sigma,\mu}:C^{\infty}(B_\Gamma,V_{B_\Gamma}(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes{\varphi})) \rightarrow C^{\infty}(B_\Gamma,V_{B_\Gamma}(\sigma_{\lambda},{\varphi}))\ .$$ In Lemma \ref{compat3} will show that $i^\Gamma_{\sigma,\mu}$ (resp. $p^\Gamma_{\sigma,\mu}$) maps $B_\Gamma(\sigma_{\lambda},{\varphi})_1$ (resp. $B_\Gamma(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes{\varphi})_1$) to $B_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})_1$ (resp. $B_\Gamma(\sigma_{\lambda},{\varphi})_1$). Hence we can consider the adjoint of $$p^\Gamma_{\tilde\sigma,\mu}:B_\Gamma(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes\tilde{\varphi})_1\rightarrow B_\Gamma(\tilde\sigma_{-\lambda},\tilde{\varphi})_1$$ which will be denoted by by $$(i^\Gamma_{\sigma,\mu})_1:D_\Gamma(\sigma_{\lambda},{\varphi})_1 \rightarrow D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})_1\ .$$ In a similar manner the adjoint of $$i^\Gamma_{\tilde\sigma,\mu}:B_\Gamma(\tilde\sigma_{-\lambda},\tilde{\varphi})_1\rightarrow B_\Gamma(1_{-\lambda+\mu},\pi_{\tilde\sigma,\mu}\otimes\tilde{\varphi})_1$$ will be a map $$(p^\Gamma_{\sigma,\mu})_1:D_\Gamma(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes{\varphi})_1\rightarrow D_\Gamma(\sigma_{\lambda},{\varphi})_1\ .$$ \subsubsection{} In the presence of cusps of full rank we must take into account their contribution to the function spaces (see \ref{weih123}). It is clear that $i_{\sigma,\mu}$ (resp. $p_{\sigma,\mu}$) maps $E_{\infty_P}(\sigma_\lambda,{\varphi})$ (resp. $E_{\infty_P}(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes{\varphi})$) to $E_{\infty_P}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})$ (resp. $E_{\infty_P}(\sigma_{\lambda},{\varphi})$). We get second components \begin{eqnarray*} (i^\Gamma_{\sigma,\mu})_2&:&{R}_{\Gamma}(\sigma_\lambda,{\varphi})_{max}\to {R}_{\Gamma}(1_{\lambda+\mu},\pi_{\sigma,\mu}{\varphi})_{max}\\(p^\Gamma_{\sigma,\mu})_2&:&{R}_\Gamma(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes{\varphi})_{max} \rightarrow {R}_\Gamma(\sigma_{\lambda},{\varphi})_{max}\end{eqnarray*} Combining all these maps we obtain maps \begin{eqnarray*} i^\Gamma_{\sigma,\mu}&:&D_\Gamma(\sigma_{\lambda},{\varphi})\rightarrow D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})\\ p^\Gamma_{\sigma,\mu}&:&D_\Gamma(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes{\varphi})\rightarrow D_\Gamma(\sigma_{\lambda},{\varphi}) \end{eqnarray*} and dually corresponding maps between spaces of functions. \subsubsection{} Using the isomorphism $T$ we transfer the action $\pi^{1_{-\lambda-\mu}\otimes\pi_{\tilde\sigma,\mu},{\mathrm {id}}_{\tilde{\varphi}}}$ of ${\mathcal{Z}}({\mathfrak{g}})$ to $$C^\infty(\partial X,V(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes \tilde{\varphi}))\ .$$ Since this action commutes with $\Gamma$ and is implemented by local operators we obtain an action of ${\mathcal{Z}}({\mathfrak{g}})$ on $C^\infty(B_\Gamma,V_{B_{\Gamma}}(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes \tilde{\varphi}))$ as well. We will show that $B_\Gamma(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes\tilde{\varphi})_1$ is a $Z({\mathfrak{g}})$-invariant subspace of $C^\infty(B_\Gamma,V_{B_{\Gamma}}(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes \tilde{\varphi}))$. By duality we obtain an action of $Z({\mathfrak{g}})$ on $D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})_1$. Since ${\mathcal{Z}}({\mathfrak{g}})$ also acts on $E_{\infty_P}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})$, $P\in \tilde{\mathcal{P}}^{max}$, the space $D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})$ has the structure of a ${\mathcal{Z}}({\mathfrak{g}})$-module. \subsubsection{}\label{weih146} Let $\Omega\in{\mathcal{Z}}({\mathfrak{g}})$ denote the Casimir operator. Let $\{\sigma^i | i\in I_\nu\}$, be the set of Weyl-invariant representations of $M$ (irreducible or sum of two non-Weyl invariant irreducible) occuring in the restriction of $V_{\pi_{\sigma,\mu}}(\nu)$ to $M$, where we set $I_\mu:=\{1\}$. Let $\nu_i(\lambda):=\pi^{\sigma^i_{\lambda+\nu}}(\Omega)\in{\mathbb {C}}$ be the eigenvalue of the Casimir operator on the principal series representation $\pi^{\sigma^i_{\lambda+\nu}}$. We set $I:=\bigcup_{\nu\not=\mu} I_\nu$ and define the meromorphic function $${\aaaa_\C^\ast}\ni\lambda\mapsto \Pi(\lambda):=1-\prod_{i\in I} \frac{\Omega-\nu_i(\lambda)}{\nu_1(\lambda)-\nu_i(\lambda)}\in{\mathcal{Z}}({\mathfrak{g}})\ .$$ This function has a finite number of poles all contained in the subset $I_{\mathfrak{a}}\subset {\mathfrak{a}}^*$. Furthermore, we define $$Z(\lambda):=\pi^{1_{\lambda+\mu}\otimes\pi_{\sigma,\mu},{\mathrm {id}}_{{\varphi}}}(\Pi(\lambda))\ .$$ If $\lambda\not\in I_{\mathfrak{a}}$ (or more precisely, if $\Pi(\lambda)$ is regular), then $Z(\lambda)$ is a projection. \subsubsection{} For $\lambda\in I_{\mathfrak{a}}$ we will show in Lemma \ref{compat6} that $$i_{\sigma,\mu}^\Gamma:D_\Gamma(\sigma_{\lambda},{\varphi})\rightarrow \ker \{Z(\lambda):D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})\rightarrow D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})\}=:\ker_\Gamma(Z(\lambda))$$ is an isomorphism. We then define the continuous map $$j_{\sigma,\mu}^\Gamma:D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}) \stackrel{1-Z(\lambda)}{\longrightarrow} \ker_\Gamma(Z(\lambda))\stackrel{(i_{\sigma,\mu}^\Gamma)^{-1}}{\longrightarrow} D_\Gamma(\sigma_{\lambda},{\varphi})\ .$$ In Lemma \ref{compat6} we will furthermore show that as a function of $\lambda$ the maps $j_{\sigma,\mu}^\Gamma$ form a meromorphic family of continuous maps which is holomorphic on ${\aaaa_\C^\ast}\setminus I_{\mathfrak{a}}$. If $\Gamma$ is trivial, then we omit the superscript and write $j_{\sigma,\mu}$ for $j_{\sigma,\mu}^{\{1\}}$. \subsubsection{}\label{weih125} We now explain by examples how the concept of twisting is applied. Assume that we have obtained a meromorphic continuation of the extension $ext^\Gamma:D_\Gamma(1_\lambda,{\varphi})\rightarrow C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))$ to some half-plane $W=\{{\rm Re }(\lambda)>\lambda_0\}$ for all admissible twists ${\varphi}$. Then we can obtain a meromorphic continuation of $ext^\Gamma:D_\Gamma(\sigma_\lambda,{\varphi})\rightarrow C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$ to some larger half plane $W-\mu_0$, $\mu_0\ge 0$, and for all $\sigma$ as follows. We choose $(\pi_{\sigma,\mu}, V_{\pi_{\sigma,\mu}})$ for some $\mu\ge \mu_0$. Then we can define a meromorphic family $\widetilde{ext}^\Gamma:D_\Gamma(\sigma_\lambda,{\varphi})\rightarrow C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$ for $\lambda\in W-\mu_0$ by the diagram $$\begin{array}{ccc} D_\Gamma(\sigma_\lambda,{\varphi})&\stackrel{i^\Gamma_{\sigma,\mu}}{\rightarrow}&D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))\\ \downarrow \widetilde{ext}^\Gamma&&\downarrow ext^\Gamma\\ C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))&\stackrel{j_{\sigma,\mu}}{\leftarrow}& C^{-\infty}(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})) \end{array}\ .$$ We then show that $ext^\Gamma=\widetilde{ext}^\Gamma$ for all $\lambda$ with sufficiently large real part. Thus $\widetilde{ext}^\Gamma$ provides a meromorphic continuation of $ext^\Gamma$. \subsubsection{}\label{weih1266} Assume that we have defined the scattering matrix $S^\Gamma_\lambda:D_\Gamma(1_\lambda,{\varphi})\rightarrow D_\Gamma(1_{-\lambda},{\varphi})$ in the spherical case for all admissible twists ${\varphi}$. Then we could define the scattering matrix $S^\Gamma_\lambda:D_\Gamma(\sigma_\lambda,{\varphi})\rightarrow D_\Gamma(\sigma_{-\lambda},{\varphi})$ for all $\sigma$ using the diagram \begin{equation}\label{qw11} \begin{array}{ccc} D_\Gamma(\sigma_\lambda,{\varphi})&\stackrel{i^\Gamma_{\sigma,\mu}}{\rightarrow}&D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})\\ \downarrow S^\Gamma_\lambda&&\downarrow S^\Gamma_{\lambda+\mu}\\ D_\Gamma(\sigma_{-\lambda},{\varphi})&\stackrel{p^\Gamma_{\sigma,\mu}}{\leftarrow}&D_\Gamma(1_{-\lambda-\mu},\pi_{\sigma,\mu}\otimes{\varphi}) \end{array}\ . \end{equation} For generic $\lambda\in{\aaaa_\C^\ast}$ we have a well-defined map $res^\Gamma:{}^\Gamma C^{-\infty}(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})) \rightarrow D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})$, and we can define $res^\Gamma:{}^\Gamma C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi})) \rightarrow D_\Gamma(\sigma_\lambda,{\varphi})$ using the diagram $$\begin{array}{ccc} {}^\Gamma C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi})) &\stackrel{i_{\sigma,\mu}}{\rightarrow}&{}^\Gamma C^{-\infty}(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))\\ \downarrow res^\Gamma &&\downarrow res^\Gamma \\ D_\Gamma(\sigma_{\lambda},{\varphi})&\stackrel{j^\Gamma_{\sigma,\mu}}{\leftarrow}&D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}) \end{array}\ .$$ We then check that $S^\Gamma_\lambda$ defined by (\ref{qw11}) coincides with $res^\Gamma\circ J_\lambda\circ ext^\Gamma$. \subsection{Holomorphic vector bundles over ${\mathbb {C}}$} \subsubsection{} It is a general fact that each finite-dimensional holomorphic vector bundle over ${\mathbb {C}}$ or a half-plane is trivial. Let $U\subset {\mathbb {C}}$ be open and $E$ be a finite-dimensional holomorphic vector bundle over $U$. Furthermore let $U\times V\rightarrow U$ be a trivial bundle of Fr\'echet spaces over $U$. By ${\mathcal{E}}$ and ${\cal V}$ we denote the corresponding sheaves of holomorphic sections. Let $\phi:E\rightarrow V$ be a meromorphic family of linear maps. The assertions of the following lemma are well-known in the case where $V$ is finite-dimensional. \begin{lem}\label{bunle} \begin{enumerate} \item There exists a unique finite-dimensional subbundle $F\subset U\times V$ such that $\phi$ factors over a meromorphic family of maps $\psi:E\rightarrow F$ which is surjective for generic $z\in U$. \item If we have meromorphic families of bundle maps $Z_E:E\rightarrow E$ and $Z_V:V\rightarrow V$ such that $\phi\circ Z_E=Z_V\circ \phi$, then $Z_V$ restricts to a meromorphic family of bundle maps of $F$. \item Furthermore, if $U={\mathbb {C}}$ or a half-plane, then there exists a meromorphic right-inverse $\eta:F\rightarrow E$ such that $\phi\circ \eta={\mathrm {id}}_F$. \end{enumerate} \end{lem} {\it Proof.$\:\:\:\:$} The proof of this lemma will occupy the remainder of the present subsection. The main point which we will explain in detail is the reduction to the finite-dimensional case. \subsubsection{} We first show that $F$ exists. Let $\phi$ have singularities in the discrete subset $A\subset U$, and let $n_z$, $z\in A$, denote the order of the corresponding singularity. We consider the divisor $D:=\sum_{z\in A} n_z z$, the associated line bundle $L(D)$, and its sheaf of sections ${\cal L}(D)$. Let ${\cal V}(D)$ be the sheaf of sections of $V\otimes L(D)$. Then $\phi$ induces a holomorphic map $\phi_D:E\rightarrow V\otimes L(D)$. Let ${\mathcal{ W}}$ be the quotient sheaf $${\mathcal{E}} \stackrel{\phi_D}{\rightarrow}{\cal V}(D)\rightarrow {\mathcal{ W}}\rightarrow 0$$ and denote by $Tor({\mathcal{ W}})$ its torsion subsheaf. We define ${\mathcal{F}}(D)$ as the kernel $$0\rightarrow {\mathcal{F}}(D)\rightarrow {\cal V}(D)\rightarrow {\mathcal{ W}}/Tor({\mathcal{ W}})\rightarrow 0\ .$$ There is a natural factorization of $\phi_D$ over $\psi_D:{\mathcal{E}}\rightarrow {\mathcal{F}}(D)$. \subsubsection{}\label{weih129} We claim that ${\mathcal{F}}(D)$ is a coherent sheaf. Let $x\in U$. If $v\in {\cal V}(D)_x$, then let $lp(v)\in V$ denote the leading part of $v$ at $x$. Note that $lp(v)=0$ implies that $v=0$ since a Laurent series must start somewhere. We define the subspace $Z\subset V$ as the set of all leading parts $lp(\phi_D(e))$, $e\in {\mathcal{E}}_x$. \subsubsection{} We now show that $\dim(Z)<\infty$. We choose a connected neighbourhood $U_x\subset U$ of $x$ and a holomorphic trivialization $E_{|U_x}=U_x\times E_x$. Let $\phi=\sum_{n\ge m} \phi_n z^n$ denote the Laurent expansion of $\phi$ in this trivialization, where $\phi_n\in {\mbox{\rm Hom}}(E_x,V)$. Then using that $\dim E_x<\infty$ we have $$Z=\phi_m(E_x)+\phi_{m+1}(\ker\phi_m)+\phi_{m+2}(\ker\phi_m\cap\ker\phi_{m+1})+\dots + \phi_{m+k}(\bigcap_{i=0}^{k-1} \ker\phi_{m+i})\ ,$$ where $k\in{\mathbb {N}}_0$ is sufficiently large such that $\bigcap_{i=0}^{k-1} \ker\phi_{m+i}=\bigcap_{i=0}^{l-1} \ker\phi_{m+i}$ for all $l\ge k$. This proves that $\dim(Z)<\infty$. \subsubsection{} We choose some closed subspace $Y\subset V$ of finite codimension such that $Z\cap Y=\{0\}$. We consider the composition $\overline{\phi}_{D}:E\stackrel{\phi}\rightarrow V \rightarrow V/Y$. We form the quotient of sheaves $${\mathcal{E}} \stackrel{\overline{\phi}_D}{\rightarrow}{\cal V}(D)/{\cal Y}(D)\rightarrow\tilde {\mathcal{ W}}\rightarrow 0\ ,$$ and we define $\tilde {\mathcal{F}}(D)$ as the kernel $$0\rightarrow\tilde{\mathcal{F}}(D)\rightarrow {\cal V}(D)/{\cal Y}(D)\rightarrow\tilde {\mathcal{ W}}/Tor(\tilde {\mathcal{ W}})\rightarrow 0\ .$$ \subsubsection{} Consider the following commutative diagram of sheaves on $U_x$: $$\begin{array}{ccccccccc} &&0&&0&&0&&\\ &&\downarrow&&\downarrow&&\downarrow&&\\ 0&\rightarrow&0&\rightarrow&{\cal Y}(D)&\stackrel{\cong}{\rightarrow}& {\cal Y}(D)&\rightarrow &0\\ &&\downarrow&&\downarrow&&\downarrow&&\\ 0&\rightarrow&{\mathcal{E}}/\ker \phi_D&\rightarrow&{\cal V}(D)&\rightarrow& {\mathcal{ W}}&\rightarrow &0\\ &&\downarrow\cong &&\downarrow&&\downarrow&&\\ 0&\rightarrow&{\mathcal{E}}/\ker \overline{\phi}_D&\rightarrow&{\cal V}(D)/{\cal Y}(D)&\rightarrow& \tilde{\mathcal{ W}}&\rightarrow &0\\ &&\downarrow&&\downarrow&&\downarrow&&\\ &&0&&0&&0&&\end{array} $$ The two lower rows are exact by construction. We now show that the left column is exact. We have a sequence $$\ker(\phi_D)\subset \ker(\overline{\phi_D})\subset {\mathcal{E}}$$ of inclusioins of torsion-free sheaves. In particular, $\ker(\phi_D)$ and $\ker(\overline{\phi_D})$ are sheaves of holomorphic sections of vector bundles on $U_x$. If we show that the inclusion induces an isomorphism \begin{equation}\label{weih128} \ker(\phi_D)_x\cong \ker(\overline{\phi_D})_x\ , \end{equation} then we conclude an isomorphism of sheaves $\ker(\phi_D) \cong \ker(\overline{\phi_D})$ after shrinking $U_x$, if necessary. To see (\ref{weih128}) consider $h\in \ker (\overline{\phi}_D)_x$. Then $lp(\phi_D(h))\in Y$. Since on the other hand $lp(\phi_D(h))\in Z$ we conclude that $lp(\phi_D(h))=0$, hence $h\in\ker (\phi_D)_x$. We now conclude that the last column is exact, too. \subsubsection{} We now consider the diagram $$\begin{array}{ccccccccc} &&0&&0&&0&&\\ &&\downarrow&&\downarrow&&\downarrow&&\\ 0&\rightarrow&0&\rightarrow&Tor {\mathcal{ W}}&\stackrel{\cong}{\rightarrow}& Tor\tilde {\mathcal{ W}} &\rightarrow &0\\ &&\downarrow&&\downarrow&&\downarrow&&\\ 0&\rightarrow&{\cal Y}(D)&\rightarrow&{\mathcal{ W}}&\rightarrow&\tilde {\mathcal{ W}}&\rightarrow &0\\ &&\downarrow\cong &&\downarrow&&\downarrow&&\\ 0&\rightarrow&{\cal Y}(D)&\rightarrow&{\mathcal{ W}} /Tor {\mathcal{ W}}&\rightarrow&\tilde {\mathcal{ W}}/ Tor\tilde{\mathcal{ W}}&\rightarrow &0\\ &&\downarrow&&\downarrow&&\downarrow&&\\ &&0&&0&&0&&\end{array}\ . $$ Since ${\cal Y}(D)$ is torsion-free we have ${\cal Y}(D)\cap Tor{\mathcal{ W}} =0$ This implies the isomorphism in the upper row. It follows that all rows and columns of this diagram are exact. \subsubsection{} We now consider the following diagram of sheaves on $U_x$: $$\begin{array}{ccccccccc} &&0&&0&&0&&\\ &&\downarrow&&\downarrow&&\downarrow&&\\ 0&\rightarrow&0&\rightarrow&{\cal Y}(D)&\stackrel{\cong}{\rightarrow}& {\cal Y}(D)&\rightarrow &0\\ &&\downarrow&&\downarrow&&\downarrow&&\\ 0&\rightarrow&{\mathcal{F}}(D)&\rightarrow&{\cal V}(D)&\rightarrow& {\mathcal{ W}}/Tor {\mathcal{ W}}&\rightarrow &0\\ &&\downarrow\cong &&\downarrow&&\downarrow&&\\ 0&\rightarrow&\tilde{\mathcal{F}}(D)&\rightarrow&{\cal V}(D)/{\cal Y}(D)&\rightarrow& \tilde{\mathcal{ W}}/Tor \tilde{\mathcal{ W}}&\rightarrow &0\\ &&\downarrow&&\downarrow&&\downarrow&&\\ &&0&&0&&0&&\end{array} $$ The rows and the middle column are exact. We just have shown that the right column is exact. \subsubsection{} We conclude that the natural map ${\mathcal{F}}(D)\rightarrow \tilde {\mathcal{F}}(D)$ is an isomorphism of sheaves. This proves the claim \ref{weih129} since $\tilde {\mathcal{F}}(D)$ is obviously coherent. \subsubsection{} Since ${\mathcal{F}}(D)$ is torsion-free it is the sheaf of sections of a holomorphic vector bundle $F(D)$ (here we use the fact that the base space is smooth and one-dimensional). By construction ${\mathcal{F}}(D)/\psi_D({\mathcal{E}})$ is torsion and therefore $\psi_{D,x}:E_x\rightarrow F(D)_x$ is surjective for generic $x\in U$. We define $F:=F(D)\otimes L(-D)$ and let $\psi:E\rightarrow F$ be the corresponding meromorphic family of maps. The uniqueness part and assertion 2. are left to the reader. \subsubsection{} We now construct the meromorphic right-inverse $\eta$. Note that we can assume that $E$ and $F$ are trivial. We fix trivializations and a constant Hermitian metric on $E$. Let $z\in {\mathbb {C}}$ be such that $\phi:E_z\rightarrow F_z$ is regular and surjective. Let $P$ be the orthogonal projection onto the orthogonal complement of $\ker\phi_z$. We extend $P$ constantly over $U$ and let $P(E)$ be the range of $P$. The composition $\phi\circ P:P(E)\rightarrow F$ is invertible at $z$ and therefore has a meromorphic family of inverses $\eta:F\rightarrow P(E)\subset E$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} \begin{ddd}\label{imagebundle} The bundle $F$ constructed in Lemma \ref{bunle} is called the image bundle of $\phi$. \end{ddd} \section{Pure cusps} \subsection{Geometry of cusps}\label{cuspgeom} \subsubsection{} In this subsection we analyze the geometry of cusps. First we recall the following theorem of Auslander. \begin{theorem}[\cite{auslander61}, \cite{MR1490024}] \label{ausl} Let $N$ be a connected, simply-connected nilpotent Lie group and $M$ be a compact group of automorphisms of $N$. Furthermore let $U\subset N\rtimes M$ be a discrete subgroup and $U^*:=\overline{NU}_0\cap U=p_{|U}^{-1}(\overline{p(U)}_0)$, where $p:N\rtimes M\rightarrow M$ is the projection, $()_0$ stands for connected component of the identity, and "$\bar{.}$" means the closure of the set in the argument. Then \begin{enumerate} \item $U^*\subset U$ is a normal subgroup of finite index. \item There exists $b\in N$ and a connected subgroup $N_V\subset N$ such that $(U^*)^b:=bU^*b^{-1}$ acts effectively and cocompactly by translations on $N_V$, i.e. there is lattice $V^*\subset N_V$ and an isomorphism $\theta:(U^*)^b\rightarrow V^*$ such that $u.x = \theta(u)x$, $x\in N_V$, where $(N\rtimes M)\times N\ni (a,x)\mapsto a.x\in N$ denotes the natural action of $N\rtimes M$ on $N$. \item The element $b\in N$ of 2. can be choosen such that $U^b$ leaves the space $N_V$ invariant. \item $M_{U^0}:=\overline{p(U^*)}\subset M$ is a torus. \end{enumerate} \end{theorem} The third assertion is due to Apanasov \cite{MR1490024}. \subsubsection{} We now apply Theorem \ref{ausl} to a discrete torsion-free subgroup $U\subset P$ of a parabolic subgroup $P\subset G$ such that $P$ is $U$-cuspidal. We have the exact sequence \begin{equation}\label{yy77} 0\rightarrow N\rightarrow P\stackrel{l}{\rightarrow} L\rightarrow 0\ ,\end{equation} where $N$ is the nil-radical of $P$. The group $L$ decomposes as $L=M\times A$, and we denote by $l_M:P\rightarrow M$ the composition of $l$ with the projection from $L$ to $M$. Let $E:=l^{-1}(M\times\{1\})$. If we choose a split $s_1:M\rightarrow E$ of the sequence $0\rightarrow N\rightarrow E\stackrel{l_M}{\rightarrow} M\rightarrow 0$, then $m\in M$ acts by the automorphism $s_1(m)$ on $N$, and we have an isomorphism $T_{s_1}:E\rightarrow N\rtimes_{s_1}M$. Since $P$ is $U$-cuspidal we have $U\subset E$. Applying Theorem \ref{ausl} to $T_{s_1}(U)$ we obtain a subgroup $T_{s_1}(U)^*\subset T_{s_1}(U)$ of finite index, $b\in N$, a connected subgroup $N_V$, a lattice $V^*\subset N_V$, and an isomorphism $\theta:bT_{s_1}(U)^*b^{-1}\rightarrow V^*$ such that $bT_{s_1}(U)b^{-1}$ leaves $N_V$ invariant and acts by translations on $N_V$ via $\theta$. \subsubsection{} If we replace the split $s_1$ and $N_V$ by $b^{-1}N_Vb$ and $s:=b^{-1}s_1b$, then $T_s(U)$ itself leaves $N_V$ invariant, and $T_s(U)^*$ acts by translations on $N_V$. From now on we use the split $s$ in order to identify $M$ with the subgroup $s(M)\subset E\subset P$, to write $E$ as the product $NM$, and to indentify $T_s(U)$ with $U$. \subsubsection{}\label{weih131} Any element $u\in U^*$ ($u\in U$) can be written as $n_um_u\in NM$ such that $n_u\in N_V$ and $m_u$ centralizes (normalizes) $N_V$. We have $V^*:=\{n_u|u\in U^*\}\subset N_V$. The map $V^*\ni n_u \mapsto m_u\in M$ defines a homomorphism $m^*:V^*\rightarrow M_{U^0}$. \begin{lem} There exists a subgroup $V\subset V^*$ of finite index such that the restriction $m:=m^*_{|V}$ extends to a homomorphism $m:N_V\rightarrow M_{U^0}$. \end{lem} {\it Proof.$\:\:\:\:$} Let $\pi:V^*\rightarrow N_V/[N_V,N_V]$ be the projection. Then $\pi(V^*)\subset N_V/[N_V,N_V]$ is a lattice (see e.g. \cite{raghunathan72}, proof of Thm. 2.10). Let $g_1,\dots g_n\in V^*$, $n=\dim(N_V/[N_V,N_V])$ be such that $\pi(g_1),\dots,\pi(g_n)$ generate the lattice $\pi(V^*)$. We define $V:=\langle g_1,\dots,g_n\rangle$. Furthermore we put $X_i:=\log(g_i)\in{\mathfrak{n}}_V$ and consider $\pi(X_i)\in {\mathfrak{n}}_V/[{\mathfrak{n}}_V,{\mathfrak{n}}_V]$. Then $\pi(X_i)$ is a basis of ${\mathfrak{n}}_V/[{\mathfrak{n}}_V,{\mathfrak{n}}_V]$, and $\{X_1,\dots,X_n\}$ generate the Lie algebra ${\mathfrak{n}}_V$. We conclude that $N_V$ is the smallest connected group containing $V$, and by \cite{raghunathan72}, Ch.2, $V$ is a lattice in $N_V$. Therefore $V\subset V^*$ has finite index. We claim that $V\cap [N_V,N_V]=[V,V]$. Let $g\in V\cap [N_V,N_V]$. Then there are finite sequences $i_k\in\{1,\dots,n\}$ and $e_k\in\{1,-1\}$, $k=1,\dots r$, such that $g=g_{i_1}^{e_1}\dots g_{i_r}^{e_r}$. Applying $\pi$ we obtain $1=\pi(g_{i_1})^{e_1}\dots \pi(g_{i_r})^{e_r}$. Using the fact that $N_V/[N_V,N_V]$ is free abelian we conclude that there is a permutation $\sigma\in S_r$ such that $1=g_{i_{\sigma(1)}}^{e_{\sigma(1)}}\dots g_{i_{\sigma(r)}}^{e_{\sigma(r)}}$. We conclude that $1=g$ modulo $[V,V]$, i.e. $g\in [V,V]$. This proves the claim. We first define the derivative $dm:{\mathfrak{n}}_V\rightarrow {\mathfrak{m}}_{U^0}$ of $m$ as the composition $${\mathfrak{n}}_V\stackrel{\pi}{\rightarrow} {\mathfrak{n}}_V/[{\mathfrak{n}}_V,{\mathfrak{n}}_V]\stackrel{q}{\rightarrow} {\mathfrak{m}}_{U^0}\ ,$$ where $q$ is given by $q(\pi(X_i)):= Z_i$, $i=1,\dots n$, where $Z_i\in{\mathfrak{m}}_{N_V}$ is any element satisfying $\exp(Z_i)=m(g_i)$. Integrating the derivative we obtain a representation $\tilde m:N_V\rightarrow M_{U^0}$ such that $m(g_i)=\tilde m(g_i)$, $i=1\dots,n$. Since $M_{U^0}$ is a torus we see that $m$ must factor over $[V,V]$ and conclude that $m=\tilde m$. This finishes the proof of the lemma. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{}\label{weih251} We consider the subgroup of finite index $U^1:=\{v m(v)| v\in V\}\subset U$. \begin{ddd}\label{tu0} We define $U^0$ to be the largest normal subgroup of $U$ such that $U^0\subset U^1\subset U$, i.e. $$U^0:=\bigcap_{u\in U} (U^1)^u\ .$$ \end{ddd} $U^0\subset U$ has finite index, too. The torus $M_{U^0}$ coincides with $\overline{m(N_V)}$. \begin{ddd}\label{weih1422} We set $P_{U^0}:=N_VM_{U^0}$. Furthermore we define $M_U:=\overline{l_M(U)}$ and $P_U:=N_VM_U$. \end{ddd} Then $U\subset P_U$, $M_{U^0}\subset M_U$ has finite index, and $M_U$ normalizes $N_V$. \subsubsection{} We have an exact sequence $$0\rightarrow E\rightarrow P\stackrel{l_A}{\rightarrow} A\rightarrow 0 \ ,$$ where the decomposition $E=NM$ is already fixed by the split $s$. The split extends uniquely to a split $s:MA\rightarrow P$ of (\ref{yy77}) and induces a split of the sequence above. In particular we obtain a Langlands decomposition $P=NAM$, where we identify $A$ with its image by $s$. The split furthermore defines an action of $A$ on $N$ by automorphisms commuting with the action of $M$. \subsubsection{} We call the cusp associated to $U\subset P$ regular if we can choose $s$ such that $N_V$ is $A$-invariant (compare Definition \ref{t799}). \begin{lem}\label{martin} If $X$ is a real or complex hyperbolic space, then every cusp is regular. \end{lem} {\it Proof.$\:\:\:\:$} If $X$ is real-hyperbolic, then for any split $s$ the group $A$ acts on ${\mathfrak{n}}$ as multiplication by scalars. Any subspace, in particular ${\mathfrak{n}}_V$, is invariant with respect to $A$. We now consider the case that $X={H\mathbb{C}}^n$. Assume that we have chosen a split $s$. Then we can decompose ${\mathfrak{n}}={\mathfrak{n}}_{\alpha}\oplus{\mathfrak{n}}_{2\alpha}$ with respect to the action of $A$ such that $a\in A$ acts on ${\mathfrak{n}}_{i\alpha}$ as multiplication by $a^{i\alpha}$. Here ${\mathfrak{n}}_{\alpha}$ is a symplectic vector space of dimension $2(n-2)$, where the symplectic form with values in the one-dimensional space ${\mathfrak{n}}_{2\alpha}$ is given by the commutator. Taking $M_{U^0}$-invariants we obtain a decomposition ${\mathfrak{n}}^{M_{U^0}}={\mathfrak{n}}_1\oplus{\mathfrak{n}}_2$, where ${\mathfrak{n}}_1\subset {\mathfrak{n}}_{\alpha}$ is a symplectic subspace (\cite{guilleminsternberg77}, Prop. 4.2.1). If ${\mathfrak{n}}_1=\{0\}$, then ${\mathfrak{n}}_V={\mathfrak{n}}_2$ is invariant with respect to $A$, and we are done. We now assume that ${\mathfrak{n}}_1\not=\{0\}$. We have two cases. If ${\mathfrak{n}}_2\subset {\mathfrak{n}}_V$, then ${\mathfrak{n}}_V={\mathfrak{n}}_V\cap{\mathfrak{n}}_1\oplus {\mathfrak{n}}_2$, and this space is invariant with respect to $A$. Thus we assume that ${\mathfrak{n}}_2\not\subset{\mathfrak{n}}_V$. Since $\dim({\mathfrak{n}}_2)=1$ we then have ${\mathfrak{n}}_2\cap {\mathfrak{n}}_V=\{0\}$. Let $p_i:{\mathfrak{n}}_V\rightarrow {\mathfrak{n}}_i$ denote the projections. Then $p_1:{\mathfrak{n}}_V\rightarrow {\mathfrak{n}}_1$ is injective. Let $\lambda:{\mathfrak{n}}_1\rightarrow {\mathfrak{n}}_2$ be a linear extension of $p_2\circ p_1^{-1}:p_1({\mathfrak{n}}_V)\rightarrow {\mathfrak{n}}_2$. Since the symplectic form is non-degenerate there exists a unique $Y\in {\mathfrak{n}}_1$ such that $\lambda(X)=[Y,X]$ for all $X\in{\mathfrak{n}}_1$. Let $h:=\exp(-Y)\in N^{M_{U^0}}$. We claim that ${\mathfrak{n}}_V^h\subset {\mathfrak{n}}_1$. Indeed, if $X\in {\mathfrak{n}}_V$, then $p_2(X^h)=p_2(X-[Y,X])=\lambda(p_1(X))-[Y,X]=[Y,X]-[Y,X]=0$. If we replace the split $s:M\times A\rightarrow P$ by $hsh^{-1}$, then ${\mathfrak{n}}_V$ gets replaced by ${\mathfrak{n}}_V^h$. Thus by an appropriate choice of the split we can assume that ${\mathfrak{n}}_V\subset{\mathfrak{n}}_1$, and we are done. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} The next lemma shows that regularity of cusps is a proper restriction in the case of ${H\mathbb{H}}^n$. \begin{lem}\label{contrmartin} If $X={H\mathbb{H}}^n$, $n\ge 2$, then there exist non-regular cusps. \end{lem} {\it Proof.$\:\:\:\:$} It suffices to provide an example for $n=2$. We can identify ${\mathfrak{n}}={\mathbb{H}}\oplus {\rm Im}_{\mathbb{H}}({\mathbb{H}})$ (${\rm Im}_{\mathbb{H}}({\mathbb{H}})$ denoting the imaginary quaternions), and the commutator of $X,Y\in{\mathbb{H}}$ is given by $[X,Y]= \bar X Y-\bar Y X \in {\rm Im}_{\mathbb{H}}({\mathbb{H}})$. Implicitly we have fixed some split $s\times t$ such that $a\in A$ acts on ${\mathbb{H}}$ as multiplication by $a^\alpha$ and on ${\rm Im}_{\mathbb{H}}({\mathbb{H}})$ by $a^{2\alpha}$. Let $1,I,J,K$ be a base of the copy of ${\mathbb{H}}$ and $i,j,k$ be the base of ${\rm Im}_{\mathbb{H}}({\mathbb{H}})$. Then we consider ${\mathfrak{n}}_V={\mathrm{span}}_{\mathbb{R}}\{1,I+j,i\}$ and let $U\subset N_V$ be any lattice. It is not possible to conjugate this subspace into an $A$-invariant one. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Assume that $U\subset P$ defines a regular cusp. \begin{ddd}\label{weih215} We define $\rho_U\in{\mathfrak{a}}^*$ by $\rho_U(H):=\frac12 {\mbox{\rm tr}}({\mbox{\rm ad}}(H)_{|{\mathfrak{n}}_V})$, where ${\mathfrak{n}}_V$ is the Lie algebra of $N_V$. Furthermore, we set $\rho^U:=\rho-\rho_U$. \end{ddd} \subsection{Schwartz spaces}\label{schw} \subsubsection{} In this subsection we start the description of our function spaces. Here we introduce the Schwartz spaces ${S}_{U}(\sigma_\lambda,{\varphi})$. We will show that these spaces form trivial holomophic bundles of Fr\'echet spaces for $\lambda\in {\aaaa_\C^\ast}$, and that they are compatible with twisting. The Schwartz spaces are basic building blocks of the function spaces $B_U(\sigma_\lambda,{\varphi})$ which we will define later (see Subsection \ref{asz}) . \subsubsection{} We fix a representative $w\in K$ of the non-trivial element of the Weyl group $W({\mathfrak{g}},{\mathfrak{a}})$ such that $w^2=1$. Let $\infty_P\in\partial X$ be the fixed point of $P$ and $0_P:=w\infty_P$. The subset $\Omega_U=\Omega_P=\partial X\setminus\infty_P$ is the $N$-orbit of $0_P$ (see Subsection \ref{geomf} for notation). Indeed, the map $N\ni x\mapsto xw\infty_P\in \Omega_U$ is a diffeomorphism from $N$ to $\Omega_U$. \subsubsection{} Let $(\sigma,V_\sigma)$ be a finite-dimensional unitary representation of $M$. Given $\lambda\in {\aaaa_\C^\ast}$ we form the representation $(\sigma_\lambda,V_{\sigma_\lambda})$ of $P$ (see \ref{weih135}). Furthermore, let $({\varphi},V_{\varphi})$ be an admissible twist (see Definition \ref{weih134}). \subsubsection{}\label{weih243} The action of $A$ on ${\mathcal{U}}({\mathfrak{n}})$ induces a grading. We choose a basis $\{A_i\}$ of ${\mathcal{U}}({\mathfrak{n}})$ such that the subset $\{A_i\:|\: \deg(A_i)\le d\alpha\}$ spans ${\mathcal{U}}({\mathfrak{n}})^{\le d\alpha}$. Furthermore, we choose a norm $|.|$ on $V_{\varphi}$. For any $d\in{\mathbb {N}}_0$, $k\in {\mathbb{R}}$, and compact subset $W\subset N\setminus N_V$ (see Theorem \ref{ausl} for the definition of $N_V$) we define the seminorm $q_{W,d,k}$ on $C_c^\infty(B_{U},V_{B_{U}}(\sigma_\lambda,{\varphi}))$ by $$q_{W,d,k}(f):=\sup_{\{i|\deg(A_i)\le d\alpha\}} \sup_{x\in W}\sup_{a\in A_+} a^{k\alpha-2(\lambda-\rho^U)} |{\varphi}(a)^{-1} f(x^aA_iw)|\ .$$ Here we consider a section of $V_{B_{U}}(\sigma_\lambda,{\varphi})$ as a $U$-invariant function on $G\setminus P$ with values in $V_{\sigma_\lambda}\otimes V_{\varphi}$ satisfying the corresponding invariance conditions with respect to the right $P$-action, and where $u\in U$ acts by $(\pi^{\sigma_\lambda,{\varphi}}(u)f)(xw)={\varphi}(u) f(u^{-1}xw)$. \subsubsection{}\label{weih259} We fix $W$ and another compact subset $W^\prime\subset N$ such that $W^\prime\cup W^{A_+}$ projects surjectively onto $B_{U}$. We add an arbitrary $C^d$-norm over $W^\prime$ to $q_{W,d,k}$ in order to obtain a norm $\|.\|_{k,d}$. We first define $S_{U,k,d}(\sigma_\lambda,{\varphi})$ to be the Banach space closure of $C_c^\infty(B_{U},V_{B_{U}}(\sigma_\lambda,{\varphi}))$ with respect to $\|.\|_{k,d}$. This Banach space is independent (up to equivalent norms) of the choices of $W$, $W^\prime$, $|.|$, and the base of ${\mathcal{U}}({\mathfrak{n}})$. \begin{ddd}\label{weih136} We define ${S}_{U,k}(\sigma_\lambda,{\varphi})$ to be the intersection of the spaces ${S}_{U,k,d}(\sigma_\lambda,{\varphi})$ for all $d\in{\mathbb {N}}_0$. The space ${S}_{U,k}(\sigma_\lambda,{\varphi})$ is a Fr\'echet space which is topologized by the countable set of norms $\|.\|_{k,d}$, $d\in{\mathbb {N}}_0$. \end{ddd} The space ${S}_{U,k}(\sigma_\lambda,{\varphi})$ is a space of smooth sections with a fixed growth rate at infinity of $B_U$ measured by $k\in {\mathbb {N}}_0$. \subsubsection{} We now show that the family of spaces $\{{S}_{U,k}(\sigma_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$ forms a trivial holomorphic bundle of Fr\'echet spaces. To this end we construct a certain holomorphic family of functions $s^z$, $z\in {\mathbb {C}}$. In the proof of Lemma \ref{ytra} we use multiplication by these functions in order to identify the Schwartz spaces for different $\lambda\in {\aaaa_\C^\ast}$. \subsubsection{} \begin{lem}\label{ingf} There exists a positive $P_{U}$-invariant function $s\in C^\infty(N)$ such that for any compact subset $W\subset N\setminus N_V$ there is $a_0\in A$ with $s(x^a)=a^{2\alpha}s(x)$ for all $a\ge a_0$ and $x\in W$. \end{lem} {\it Proof.$\:\:\:\:$} Let $\bar{N}:=N^w$ and consider the decomposition $$G\setminus wP=\bar NMAN\ , \quad g=\bar n(g) m(x)a(x)n(x)\ .$$ Note that $a(xw)$ satisfies $a(x^aw)=a^2a(xw)$ for all $a\in A$ and $x\in N\setminus\{1\}$ (see (\ref{weih141}) below). We consider the function $N\setminus\{1\}\ni x\mapsto a(xw)^{\alpha}$ as a positive function $s_1\in C^\infty(N\setminus \{1\})$. Let $s_2\in C^\infty(N)$ be any positive function. Using a partition of unity we glue $s_1$ and $s_2$ to obtain a positive function $s_3\in C^\infty(N)$ which coincides with $s_1$ outside of a compact subset $W_1$ of $N$. We now choose $z\in{\mathbb{R}}$ such $z<-\rho_U/\alpha$ and define $s_4$ as the average $$s_4(x):=\int_{P_{U}} s_3^z(y.x) dy$$ (see Definition \ref{weih1422} for $P_U$ and Theorem \ref{ausl}, (2) for the action $(y,x)\mapsto y.x$). The integral converges (compare Lemma \ref{210}) and defines a positive $P_{U}$-invariant smooth function on $N$. If $W\subset N\setminus N_V$ is compact, then there is $a^0\in A$ such that $(N_VW)^{aM_U}\cap W_1=\emptyset$ for all $a\ge a_0$. If $a\ge a_0$ and $x\in W$, then we have \begin{eqnarray*} s_4(x^a)&=&\int_{P_{U}} s_3^z(y.x^a) dy\\ &=&\int_{M_U}\int_{N_V} s_3^z(vx^{au}) dv du\\ &=&a^{2\rho_U}\int_{M_U}\int_{N_V} s^z_3((vx)^{au}) dvdu\\ &=&a^{2(z\alpha+\rho_U)} s_4(x)\ . \end{eqnarray*} We define $s$ as the $\alpha/(z\alpha+\rho_U)$'th power of $s_4$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{}\label{neuj911} We now consider the function $s$ constructed in Lemma \ref{ingf} as a section $s\in C^\infty(\Omega_P,V(1_{\rho+\alpha}))$ by defining $s(xw):=s(x)$. If $\lambda\in{\mathfrak{a}}^*$, then the bundle $V(1_\lambda)$ is a complex line bundle which can be written as the complexification of a trivial $G$-equivariant real line bundle $V(1_\lambda)^{\mathbb{R}}$. It makes sense to speak of a positive section of $V(1_\lambda)^{\mathbb{R}}$. A section of $V(1_\lambda)$ is called positive if it is a positive section of the real subbundle $V(1_\lambda)^{\mathbb{R}}$. In particular, the section $s$ of $V(1_{\rho+\alpha})$ constructed above is positive. \subsubsection{}\label{neuj910} For each $z\in {\mathbb {C}}$ we can form $s^z$, which is a section of $V(1_{\rho+z\alpha})$. There is a natural identification $V(\sigma_\lambda)\otimes V(1_{\rho+z\alpha})\cong V(\sigma_{\lambda+z\alpha})$. Multiplication by $s^z$ identifies $C^\infty(\Omega_P,V(\sigma_\lambda,{\varphi}))$ with $C^\infty(\Omega_P,V(\sigma_{\lambda+z\alpha},{\varphi}))$. \begin{lem}\label{ytra} For each $d\in{\mathbb {N}}_0$ and $z\in{\mathbb {C}}$ multiplication by $s^z$ defines a continuous map from ${S}_{U,k,d}(\sigma_\lambda,{\varphi})$ to ${S}_{U,k,d}(\sigma_{\lambda+z\alpha},{\varphi})$. The family $\{{S}_{U,k}(\sigma_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$ is a trivial holomorphic bundle of Fr\'echet spaces. \end{lem} {\it Proof.$\:\:\:\:$} Since $s$ is $U$-invariant and non-vanishing multiplication by $s^z$ is an isomorphism of $C^\infty(B_{U},V_{B_{U}}(\sigma_\lambda,{\varphi}))$ with $C^\infty(B_{U},V_{B_{U}}(\sigma_{\lambda+z\alpha},{\varphi}))$ and of the subspaces of sections with compact support. It suffices to show that there is a constant $C\in {\mathbb{R}}$ such that for all $f\in C_c^\infty(B_{U},V_{B_{U}}(\sigma_\lambda,{\varphi}))$ we have $\|s^z f\|_{k,d}\le C \|f\|_{k,d}$. This follows from the Leibniz rule and the following estimate. Let $A\in{\mathcal{U}}({\mathfrak{n}})$ be homogeneous of degree $d\alpha$. Then for any compact subset $W\subset N\setminus N_V$ there is a constant $C\in{\mathbb{R}}$ and $a_0\in A$ such that for $x\in W$ and $a\ge a_0$ \begin{eqnarray*} |s^z(x^aAw)|&=&|s^z((xA^{a^{-1}})^aw)|\\ &=&a^{(2z-d)\alpha} |s^z(xAw)|\\ &\le& C a^{(2z-d)\alpha} \ . \end{eqnarray*} For any $\lambda_0$ we now define the trivialization $\Phi_{\lambda_0}:\bigcup_{\lambda\in{\aaaa_\C^\ast}} {S}_{U,k}(\sigma_{\lambda},{\varphi})\rightarrow {S}_{U,k}(\sigma_{\lambda_0},{\varphi}) \times {\aaaa_\C^\ast}$ such that the restriction of $\Phi_{\lambda_0}$ to the fibre ${S}_{U,k}(\sigma_{\lambda},{\varphi})$ is multiplication by $s^{(\lambda_0-\lambda)/\alpha}$. The transition map $\Phi_{\lambda_0}\circ \Phi_{\lambda_1}^{-1}$ is multiplication by $s^{(\lambda_0-\lambda_1)/\alpha}$ and thus independent of $\lambda$. In particular, it is a holomorphic family of continuous maps. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} The Schwartz space is the space of smooth rapidly decaying sections on $B_U$. \begin{ddd}\label{weih245} We define the Schwartz space $S_{U}(\sigma_\lambda,{\varphi})$ as the intersection of the spaces $S_{U,k}(\sigma_\lambda,{\varphi})$ over all $k\in{\mathbb {N}}_0$. \end{ddd} \subsubsection{} Since the trivializations $\Phi_\lambda$ are compatible with the inclusions $S_{U,k^\prime}(\sigma_\lambda,{\varphi})\hookrightarrow S_{U,k}(\sigma_\lambda,{\varphi})$, $k^\prime\ge k$, we obtain the following corollary. \begin{kor} The family $\{S_{U}(\sigma_\lambda,{\varphi})\}_{\lambda\in {\aaaa_\C^\ast}}$ is a trivial holomorphic vector bundle of Fr\'echet spaces. \end{kor} \subsubsection{} Note that the inclusions $S_{U,k+1}(\sigma_\lambda,{\varphi})\hookrightarrow S_{U,k}(\sigma_\lambda,{\varphi})$ are compact. The proof of this fact is a simple application of the Lemma of Arzela-Ascoli. The compactness of these inclusions imply the following fact. \begin{kor} The Schwartz space $S_{U}(\sigma_\lambda,{\varphi})$ is a Montel space. In particular, it is reflexive. \end{kor} \subsubsection{} If the cusp associated to $U\subset P$ has full rank, then we have for all $k\in{\mathbb {N}}_0$ $$S_{U,k}(\sigma_\lambda,{\varphi})={S}_{U}(\sigma_\lambda,{\varphi})=C^\infty(B_{U},V_{B_{U}}(\sigma_\lambda,{\varphi}))\ .$$ \subsubsection{} Next we show that the Schwartz spaces are compatible with twisting. Let $\sigma$ be a Weyl invariant representation of $M$ and $(\pi_{\sigma,\mu},V_{\pi_{\sigma,\mu}})$ be a finite-dimensional representation of $G$ as in \ref{weih144}. See \ref{weih146} for the definition of $Z(\lambda)$ and $\Pi(\lambda)$. \begin{lem}\label{compat1} \mbox{}\\ \begin{enumerate} \item The Schwartz space ${S}_{U}(\sigma_\lambda,{\varphi})$ coincides with the closed subspace of ${}^U C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$ of sections which vanish at $\infty_P$ of infinite order. \item We have holomorphic families of continuous maps \begin{eqnarray*} \{i_{\sigma,\mu}^U\}&:&S_{U}(\sigma_\lambda,{\varphi})\rightarrow S_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})\\ \{p_{\sigma,\mu}^U\}&:&S_U(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})\rightarrow {S}_{U}(\sigma_\lambda,{\varphi})\ . \end{eqnarray*} \item $S_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$ is a $Z({\mathfrak{g}})$-module. \item If $\lambda\not\in I_{\mathfrak{a}}$, then $\{i_{\sigma,\mu}^U\}$ maps ${S}_{U}(\sigma_\lambda,{\varphi})$ isomorphically onto $$\ker\{Z(\lambda):S_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})\rightarrow S_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})\}\ ,$$ and the restriction of $\{p_{\sigma,\mu}^U\}$ to $$\ker\{\tilde Z(\lambda):S_U(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})\rightarrow S_U(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})\}$$ is an isomorphism onto ${S}_{U}(\sigma_\lambda,{\varphi})$ (here $\tilde Z(\lambda)$ is the adjoint of $\pi^{1_{-\lambda+\mu}\otimes \tilde\pi_{\sigma,\mu},{\mathrm {id}}_{{\varphi}}}(\Pi(\lambda))$). \item The composition $$\{j_{\sigma,\mu}^U\}:S_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi}) \stackrel{1-Z(\lambda)}{\rightarrow} \ker(Z(\lambda))\stackrel{(\{i_{\sigma,\mu}^U\})^{-1}}{\rightarrow} {S}_{U}(\sigma_\lambda,{\varphi})$$ which is intitially defined for $\lambda\not\in I_{\mathfrak{a}}$ extends to a meromorphic family of continuous maps. Similarly, the composition $$\{q_{\sigma,\mu}^U\}:{S}_{U}(\sigma_\lambda,{\varphi})\stackrel{(\{p_{\sigma,\mu}^U\})^{-1}}{\rightarrow} \ker(\tilde Z(\lambda)) \rightarrow S_U(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})$$ extends to a meromorphic family of continuous maps. \end{enumerate} \end{lem} {\it Proof.$\:\:\:\:$} We leave the proof of 1. at this place to the interested reader. Alternatively the assertion can be considered as an immediate consequence of the material of Subsection \ref{trivas}. \subsubsection{} 2. follows from 1. and the fact that $\{i_{\sigma,\mu}^U\}$ and $\{p_{\sigma,\mu}^U\}$ are induced by an inclusion $$i_{\sigma,\mu}:V(\sigma_\lambda,{\varphi})\rightarrow V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$$ and a projection $$p_{\sigma,\mu}:V(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})\rightarrow V(\sigma_\lambda,{\varphi})$$ of bundles. Another argument for the assertion about $\{i_{\sigma,\mu}^U\}$ would be to check that $\{i_{\sigma,\mu}^U\}$ maps $S_{U,k,d}(\sigma_\lambda,{\varphi})$ continuously to $S_{U,k^\prime,d}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$, $k^\prime=k+\mu/\alpha$, for all $k,d\in{\mathbb {N}}_0$ by comparing the norms explicitly. A similar argument works for $\{p_{\sigma,\mu}^U\}$ as well. \subsubsection{} In order to see 3. note that the action of $cZ({\mathfrak{g}})$ is implemented by differential operators. Using the fact that the action $\pi^{1_{\lambda+\mu}\otimes \pi_{\sigma,\mu},{\mathrm {id}}_{\varphi}}$ of ${\mathcal{Z}}({\mathfrak{g}})$ commutes with the action $\pi^{1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi}}$ of $P_UA$ which is used in order to characterize the growth of elements of $S_{U,k}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$ one could alternatively check that this space is a ${\mathcal{Z}}({\mathfrak{g}})$-module. \subsubsection{} We now prove the first assertion of 4. and leave the second to the reader since the argument is similar. The representation $V_{1_{\lambda+\mu}}\otimes V_{\pi_{\sigma,\mu}}$ of $P$ fits into an exact sequence $$0\rightarrow V_{\sigma_\lambda} \stackrel{i_{\sigma,\mu}}{\rightarrow} V_{1_{\lambda+\mu}}\otimes V_{\pi_{\sigma,\mu}} \rightarrow W\rightarrow 0\ ,$$ which induces a corresponding exact sequence of bundles \begin{equation} \label{hj112}0\rightarrow V(\sigma_\lambda,{\varphi})\stackrel{i_{\sigma,\mu}}{\rightarrow} V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}) \rightarrow V(w ,{\varphi}) \rightarrow 0 \ ,\end{equation} where $w $ denotes the representation of $P$ on $W $. Using 1. we obtain an exact sequence of sections which vanish of infinite order at $\infty_P$. Going over to $U$-invariant sections we obtain the exact sequence $$0\rightarrow S_U(\sigma_\lambda,{\varphi})\stackrel{\{i_{\sigma,\mu}^U\}}{\rightarrow} S_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}) \stackrel{p}{\rightarrow} S_U(w ,{\varphi}) \ .$$ The action of ${\mathcal{Z}}({\mathfrak{g}})$ commutes with $p$. The operator $\Pi(\lambda)$ is designed such that $\pi^{w ,{\varphi}}(\Pi(\lambda))=1$ and $\pi^{\sigma_\lambda,{\varphi}}(\Pi(\lambda))=0$ (note that $\lambda\not\in I_{\mathfrak{a}}$). Since $\{i_{\sigma,\mu}^U\}$ is a morphism of $Z({\mathfrak{g}})$-modules we see that $\{i_{\sigma,\mu}^U\}$ maps to $\ker(Z(\lambda))$. In order to show that it is onto take $f\in \ker(Z(\lambda))$. Then $$p(f)=p((1-Z(\lambda))f)=(1-\pi^{w,{\varphi}}(\Pi(\lambda)))p(f)=0\ ,$$ and $f$ is in the range of $\{i_{\sigma,\mu}^U\}$. \subsubsection{} Finally we show the first assertion of 5. and leave the second to the reader since the argument is similar, again. We choose a holomorphic family of splits $j :V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}) \rightarrow V(\sigma_\lambda,{\varphi})$ of the sequence of bundles (\ref{hj112}) (not necessarily $U$-invariant). The split $j $ induces a holomorphic family of maps $J :C^\infty(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))\rightarrow C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$. The map $J$ is compatible with the subspaces of sections vanishing of infinite order at $\infty_P$. We define the meromorphic family of continuous maps $\tilde j_{\sigma,\mu}^U$ as the restricition of $J \circ (1-Z(\lambda))$ to $S_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$. Since $$\tilde j_{\sigma,\mu}^U\circ i_{\sigma,\mu}^U = J \circ (1-Z(\lambda))\circ i_{\sigma,\mu}^U={\mathrm {id}}$$ and the restriction of $\tilde j_{\sigma,\mu}^U$ to $\ker(1-Z(\lambda))$ vanishes we see by 4. that $\tilde j_{\sigma,\mu}^U$ maps to $S_U(\sigma_\lambda,{\varphi})$, and that it coincides with $\{j_{\sigma,\mu}^U\}$ for $\lambda\not\in I_{\mathfrak{a}}$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsection{Asymptotics for the trivial group}\label{trivas} \subsubsection{}\label{weih600} We consider the decomposition $G\setminus wP=\bar{N}MAN$, $g=\bar{n}(g)m(g)a(g)n(g)$. There is a unique diffeomorphism $F:N\setminus \{1\} \rightarrow \bar{N}\setminus \{1\}$ such that $nw\in F(n)MAN$. Indeed, $F(n)=\bar{n}(nw)$. One can check that \begin{eqnarray} a(n^{ma}w)&=&a^2 a(nw)\label{weih141}\\ m(n^aw)&=&m(nw)\nonumber\\ F(n^{ma})&=& F(n)^{ma}\nonumber\ . \end{eqnarray} \subsubsection{}\label{weih213} Let $({\varphi},V_{\varphi})$ be an admissible twist for $U$. For simplicity we normalize the restriction of ${\varphi}$ to $A$ such that the lowest $A$-weights of all its irreducible coponents are zero. Let $l_{\varphi}\in{\mathfrak{a}}^*$ be the highest weight of ${\varphi}$. \subsubsection{} In the present subsection we describe the space $C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))=:B_{\{1\}}(\sigma_\lambda,{\varphi})$ as an extension $$0\rightarrow {S}_{\{1\}}(\sigma_\lambda,{\varphi})\rightarrow B_{\{1\}}(\sigma_\lambda,{\varphi}) \rightarrow {R}_{\{1\}}(\sigma_\lambda,{\varphi})\rightarrow 0\ .$$ The Schwartz space ${S}_{\{1\}}(\sigma_\lambda,{\varphi})$ coincides with the subspace of $C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$ of all section that vanish at $\infty_P$ of infinite order. Note that $\bar N$ is diffeomorphic to $\partial X\setminus 0_P$ by $\bar n\mapsto \bar n \infty_P$ (recall that $0_P=w\infty_P$). Using a theorem of E. Borel to the effect that each formal power series can be realized as a Taylor series of a smooth function we obtain an exact sequence \begin{equation}\label{borell}0\rightarrow S_{\{1\}}(\sigma_\lambda,{\varphi})\rightarrow C^\infty(\partial X,V(\sigma_\lambda,{\varphi})) \stackrel{TS}{\rightarrow} {\mbox{\rm Hom}}({\mathcal{U}}(\bar{\mathfrak{n}}),V_{\sigma_\lambda}\otimes V_{\varphi}) \rightarrow 0\ ,\end{equation} where $TS$ is given by $TS(f)(A):=f(Ae)$. \subsubsection{}\label{neuj100} The space ${\mbox{\rm Hom}}({\mathcal{U}}(\bar{\mathfrak{n}}),V_{\sigma_\lambda}\otimes V_{\varphi})$ admits an action of $AM_U$ by $(u.f)(A):=(\sigma_\lambda(u)\otimes {\varphi}(u))f(A^{u^{-1}})$. With respect to the action of $A$ it can be decomposed as $${\mbox{\rm Hom}}({\mathcal{U}}(\bar{\mathfrak{n}}),V_{\sigma_\lambda}\otimes V_{\varphi})=\prod_{n\in{\mathbb {N}}_0} {\mbox{\rm Hom}}({\mathcal{U}}(\bar{\mathfrak{n}}),V_{\sigma_\lambda}\otimes V_{\varphi})^n\ ,$$ where $A$ acts on the $M_U$-module ${\mbox{\rm Hom}}({\mathcal{U}}(\bar{\mathfrak{n}}),V_{\sigma_\lambda}\otimes V_{\varphi})^n$ with weight $\rho-\lambda+\alpha n$. \subsubsection{} By definition a function $p:\bar N\rightarrow V_{\sigma_\lambda}\otimes V_{\varphi}$ is a polynomial iff $p\circ \exp:\bar{\mathfrak{n}}\rightarrow V_{\sigma_\lambda}\otimes V_{\varphi}$ is a polynomial. Such a function $p$ is called homogeneous of degree $n$, iff $$(\sigma_{\lambda}(a)\otimes {\varphi}(a))p(\bar n^{a^{-1}})= a^{\rho-\lambda+n\alpha} p(\bar n)\ .$$ \subsubsection{}\label{neuj101} The space ${\mbox{\rm Hom}}({\mathcal{U}}(\bar{\mathfrak{n}}),V_{\sigma_\lambda}\otimes V_{\varphi})^n$ can be identified $AM_U$-equivariantly with the space of polynomials ${\rm Pol}(\bar N,V_{\sigma_\lambda}\otimes V_{\varphi})^n$ on $\bar N$ with values in $V_{\sigma_\lambda}\otimes V_{\varphi}$ being homogeneous of degree $n$. Here a polynomial $p\in {\rm Pol}(\bar N,V_{\sigma_\lambda}\otimes V_{\varphi})^n$ corresponds to the map ${\mathcal{U}}(\bar{\mathfrak{n}})\ni A\mapsto p(Ae)\in V_{\sigma_\lambda}\otimes V_{\varphi}$. \subsubsection{}\label{weih503} If a homogeneous polynomial $p\in {\rm Pol}(\bar N,V_{\sigma_\lambda}\otimes V_{\varphi})^n$ is considered as a section $f_p$ of $V(\sigma_\lambda,{\varphi})$ defined over $\partial X\setminus \{0_P\}$, then we have for $x\in N$ $$f_p(xw)=f_p(F(x)m(xw)a(xw)n(xw))=\sigma(m(xw))^{-1} a(xw)^{\lambda-\rho} f_p(F(x))\ .$$ In particular, for $x\in N$, $x\not=1$, and $a\in A$ we have \begin{eqnarray*} {\varphi}(a)^{-1} f_p(x^aw)&=&{\varphi}(a)^{-1}a(x^aw)^{\lambda-\rho} \sigma(m(x^aw))^{-1}f_p(F(x^a))\\ &=& a^{2(\lambda-\rho)-n\alpha} a(xw)^{\lambda-\rho}\sigma(m(xw))^{-1} f_p(F(x))\ . \end{eqnarray*} \begin{ddd}\label{weih231} We define $A_{\{1\}}(\sigma_\lambda,{\varphi})^n\subset C^\infty(\partial X\setminus \{\infty_P,0_P\},V(\sigma_\lambda,{\varphi}))$ as the subspace of sections $f$ satisfying \begin{equation}\label{weih217} {\varphi}(a)^{-1}f(x^aw)=a^{2(\lambda-\rho)-n\alpha} f(xw) \end{equation} for all $x\in N\setminus \{1\}$, $a\in A$. We define the subspace ${R}_{\{1\}}(\sigma_\lambda,{\varphi})^n\subset A_{\{1\}}(\sigma_\lambda,{\varphi})^n$ as the subspace spanned by the sections $f_p$ corresponding to $p\in{\rm Pol}(\bar N,V_{\varphi})^n$. We further define ${R}_{\{1\},k}(\sigma_\lambda,{\varphi}):=\bigoplus_{n=0}^k {R}_{\{1\}}(\sigma_\lambda,{\varphi})^n$ and ${R}_{\{1\}}(\sigma_\lambda,{\varphi}):=\prod_{n=0}^\infty{R}_{\{1\}}(\sigma_\lambda,{\varphi})^n$. \end{ddd} \subsubsection{} The spaces $A_{\{1\}}(\sigma_\lambda,{\varphi})^n$ for the various $\lambda$ can be identified using multiplication by suitable powers of the function $xw\mapsto a(xw)^\alpha$ and therefore form a trivial vector bundle of Fr\'echet spaces over ${\aaaa_\C^\ast}$. The spaces ${R}_{\{1\}}(\sigma_\lambda,{\varphi})^n$ form trivial finite-dimensional subbundles trivialized by the identification with the trivial bundles ${\aaaa_\C^\ast}\times {\rm Pol}(\bar N,V_{\sigma} \otimes V_{\varphi})^n\rightarrow {\aaaa_\C^\ast}$. Thus ${R}_{\{1\},k}(\sigma_\lambda,{\varphi})$ has the structure of a trivial holomorphic vector bundle. \subsubsection{}\label{neuj202} We fix a smooth cut-off function $\chi\in C^\infty(A)$ such that $\chi(a)=0$ in a neighbourhood of $A_-$ and $\chi(a)=1$ if $a^\alpha>2$. Multiplication by the function $$\Omega_P\ni x\,0_P\mapsto \chi(a(xw))$$ defines inclusions $L: {R}_{\{1\}}(\sigma_\lambda,{\varphi})^n \rightarrow C^\infty(B_{\{1\}},V_{\{1\}}(\sigma_\lambda,{\varphi}))$ for each $n$, and summing these maps up we obtain inclusions $$L: {R}_{\{1\},k}(\sigma_\lambda,{\varphi}) \rightarrow C^\infty(B_{\{1\}},V_{B_{\{1\}}}(\sigma_\lambda,{\varphi}))\ .$$ In the present paper the symbol $L$ is used for various maps of this kind. It will always be clear from the context which version of $L$ is meant. \subsubsection{} If $f\in {S}_{\{1\},k}(\sigma_\lambda,{\varphi})$, then $$\lim_{a\to\infty} a^{k\alpha-2(\lambda-\rho)}{\varphi}(a)^{-1} f(x^aw) =0$$ uniformly for $x$ in compact subsets of $N\setminus \{1\}$. Thus $L({R}_{\{1\},k}(\sigma_\lambda,{\varphi}))$ intersects ${S}_{\{1\},k}(\sigma_\lambda,{\varphi})$ trivially. We define $$B_{\{1\},k}(\sigma_\lambda,{\varphi}):={S}_{\{1\},k}(\sigma_\lambda,{\varphi})\oplus L({R}_{\{1\},k}(\sigma_\lambda,{\varphi}))\ .$$ This space fits into the exact sequence \begin{equation}\label{weih210} 0\rightarrow {S}_{\{1\},k}(\sigma_\lambda,{\varphi}) \rightarrow B_{\{1\},k}(\sigma_\lambda,{\varphi}) \stackrel{AS}{\rightarrow}{R}_{\{1\},k}(\sigma_\lambda,{\varphi})\rightarrow 0\ , \end{equation} where $AS$ takes the finite asymptotic expansion. To $AS$ applies the same remark as as above for $L$. This symbol appears in various versions which will be denoted by the same symbol, and it will be clear from the context which version is meant. \subsubsection{} The sequence (\ref{weih210}) is split by $L$ and defines $B_{\{1\},k}(\sigma_\lambda,{\varphi})$ as a topological vector space and the family of spaces $\{B_{\{1\},k}(\sigma_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$ as a trivial bundle of Fr\'echet spaces. Moreover, we have continuous and compact inclusions $B_{\{1\},k+1}(\sigma_\lambda,{\varphi})\hookrightarrow B_{\{1\},k}(\sigma_\lambda,{\varphi})$. The intersection of these spaces over all $k\in{\mathbb {N}}_0$ is the Fr\'echet and Montel space $B_{\{1\}}(\sigma_\lambda,{\varphi})$, which concides with $C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$ because of exactness of (\ref{borell}). Note that the sequence $$0\rightarrow {S}_{\{1\}}(\sigma_\lambda,{\varphi})\rightarrow B_{\{1\}}(\sigma_\lambda,{\varphi}) \rightarrow {R}_{\{1\}}(\sigma_\lambda,{\varphi})\rightarrow 0$$ does not admit any continuous split. \subsubsection{} Though this is clearly possible we will not attempt to trivialize the family of spaces $B_{\{1\}}(\sigma_\lambda,{\varphi})$. But we keep in mind that $B_{\{1\}}(\sigma_\lambda,{\varphi})$ is an intersection of trivial holomorphic bundles (see \ref{weih211}). Below we will obtain a similar description of the spaces $B_{U,k}(\sigma_\lambda,{\varphi})$. \subsection{The spaces $B_{U,k}(\sigma_\lambda,{\varphi})$}\label{asz} \subsubsection{} In the present subsection $\sigma$ is any finite-dimensional representation of $M$, and ${\varphi}$ denotes a twist. Below we define the spaces $B_{U}(\sigma_\lambda,{\varphi})$. In particular, we construct the spaces ${R}_{U}(\sigma_\lambda,{\varphi})$ which describe the asymptotic behaviour of elements of $B_{U}(\sigma_\lambda,{\varphi})$. For simplicity we assume that the twist ${\varphi}$ is normalized as in \ref{weih213}. \subsubsection{}\label{weih244} We assume that $U\subset P$ defines a cusp of smaller rank. The case of cusps of full rank will be discussed in Subsection \ref{maxrank}. We want to define a map $\pi^{P_{U}}_*:A_{\{1\}}(\sigma_\lambda,{\varphi})^n\rightarrow C^\infty(\Omega_P\setminus N_V \{0_P\},V(\sigma_\lambda,{\varphi}))$ by $$\pi^{P_{U}}_*(f):=\int_{P_U} \pi^{\sigma_\lambda,{\varphi}}(u) fdu \ .$$ If $x\not\in N_V$, then the integrand $(\pi^{\sigma_\lambda,{\varphi}}(u) f)(xw)$ is well-defined for all $u\in P_{U}$. We will see below that this integral converges for ${\rm Re }(\lambda)<\rho^U-l_{\varphi}/2$. Let $S(N_V):=\{x\in N_V|a(xw)=1\}$ (see \ref{weih213} for a definition of $l_{\varphi}$ and Definition \ref{weih215} for $\rho^U$). \subsubsection{} Employing the assumption that $N_V$ is invariant under conjugation by $A$ (regularity of the cusp, Definition \ref{t799}) we can consider polar coordinates $S(N_V)\times A\ni (\xi,a)\mapsto \xi^a\in N_V$ of $N_V\setminus \{1\}$. \begin{lem}\label{210} There is a measure $d\xi$ on $S(N_V)$ such that the Haar measure $dx$ of $N_V$ is given by $ a^{2\rho_U} dad\xi $. \end{lem} {\it Proof.$\:\:\:\:$} There is a family of measures $d\xi(a)$ on $S(N_V)$ such that for any $f\in C_c(N_V)$ we have $$\int_{N_V} f(x) dx =\int_A \int_{S(N_V)} f(\xi^a) d\xi(a) a^{2\rho_U} da\ .$$ We compute for any $b\in A$ \begin{eqnarray*} \int_{N_V} f(x^{b^{-1}}) dx &=& b^{2\rho_U}\int_{N_V} f(x) dx\\ &=& b^{2\rho_U} \int_A \int_{S(N_V)} f(\xi^a) d\xi(a) a^{2\rho_U} da\ ,\\ \int_{N_V} f(x^{b^{-1}}) dx &=&\int_A \int_{S(N_V)} f(\xi^{ab^{-1}}) d\xi(a) a^{2\rho_U} da\\ &=&b^{2\rho_U} \int_A \int_{S(N_V)} f(\xi^a) d\xi(ab) a^{2\rho_U} da\ . \end{eqnarray*} We conclude that $d\xi(ab)=d\xi(a)$ for all $b\in A$ and hence $d\xi(a)=d\xi(1)=:d\xi$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{}\label{weih221} Recall the the Definition \ref{weih1422} of $P_U=N_V M_U$. We introduce the abbreviation $$f_0:= \int_{M_{U}} \pi^{\sigma_\lambda,{\varphi}}(t) f dt\ .$$ Then we can write for $x\not\in N_V$ and using the homogeneity (\ref{weih217}) \begin{eqnarray} \pi^{P_{U}}_*(f)(xw)&=&\int_A \int_{S(N_V)} {\varphi}(\xi^a)^{-1} f_0( \xi^a xw) a^{2\rho_U} d\xi da\label{hj11}\\&=&\int_A \int_{S(N_V)} {\varphi}(a) {\varphi}(\xi)^{-1} {\varphi}(a)^{-1}f_0( a\xi a^{-1} xw) a^{2\rho_U} d\xi da\nonumber\\ &=&\int_A \int_{S(N_V)} {\varphi}(a) {\varphi}(\xi)^{-1} {\varphi}(a)^{-1}f_0( a\xi x^{a^{-1}}a^{-1} w) a^{2\rho_U} d\xi da\nonumber\\ &=& \int_A\int_{S(N_V)} {\varphi}(a) {\varphi}(\xi)^{-1} f_0(\xi x^{a^{-1}}w) a^{2(\lambda-\rho^U)-n\alpha} d\xi da\ .\label{hj12} \end{eqnarray} We decompose the outer integral into the integrals over $A_+$ and $A_-$ and obtain $\pi^{P_{U} }_+(f)$, $\pi^{P_{U} }_- (f)$ such that $\pi^{P_{U}}_*(f)=\pi^{P_{U} }_+(f) +\pi^{P_{U} }_-(f) $. It is clear from (\ref{hj11}) that $\pi^{P_{U} }_-(f)$ converges for all $\lambda\in{\aaaa_\C^\ast}$. We conclude from (\ref{hj12}) that $\pi^{P_{U}}_+(f)$ converges for ${\rm Re }(\lambda)<\rho^U+(n\alpha-l_{\varphi})/2$. In this domain of convergence we compute for $b\in A$ $${\varphi}(b)^{-1}\pi^{P_{U} }_*(f)(x^bw)= b^{2(\lambda-\rho^U)-n\alpha}\pi^{P_{U} }_*(f)(xw)\ .$$ \subsubsection{} Motivated by this calculation we make the following definition. \begin{ddd} We define $A_{P_U}(\sigma_\lambda,{\varphi})^n$ to be the space of all $P_U$-invariant $f\in C^\infty(\Omega_P\setminus N_V 0_P,V(\sigma_\lambda,{\varphi}))$ satisfying $${\varphi}(a)^{-1}f(x^aw)=a^{2(\lambda-\rho^U)-n\alpha} f(xw)$$ for all $a\in A$, $x\in N\setminus N_V$. \end{ddd} \subsubsection{} \begin{lem} The family of spaces $\{A_{P_U}(\sigma_\lambda,{\varphi})^n\}_{\lambda\in{\aaaa_\C^\ast}}$ forms a trivial holomorphic family of Fr\'echet and Montel spaces. \end{lem} {\it Proof.$\:\:\:\:$} For some $z\in{\mathbb{R}}$ with $z<-\rho_U/\alpha$ we define the function $$s_1(x):=\int_{P_{U}} a(y.xw)^{z\alpha} dy\ .$$ We further define $s$ as the $\alpha/(z\alpha+\rho_U)$'th power of $s_1$. Now we consider the function $s$ as a section $s\in C^\infty(\Omega_P\setminus N_V0_P,V(1_{\rho+\alpha}))$ defining $s(xw):= s(x)$. Multiplication by $s^\mu$ defines an continuous isomorphism of $A_{P_U}(\sigma_\lambda,{\varphi})^n$ with $A_{P_U}(\sigma_{\lambda+\mu},{\varphi})^n$. We employ these isomorphisms in order to define a holomorphic trivialization of the bundle. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} For ${\rm Re }(\lambda)<\rho^U+(n\alpha-l_{\varphi})/2$ we have defined above a holomorphic family of continuous maps $$\pi^{P_{U} }_*:A_{\{1\}}(\sigma_\lambda,{\varphi})^n\rightarrow A_{P_{U}}(\sigma_\lambda,{\varphi})^n\ .$$ \begin{lem}\label{aaww} The family $\pi^{P_{U} }_*$ extends meromorphically to all of ${\aaaa_\C^\ast}$ with at most first order poles in the set $\rho^U+\frac{n\alpha-l_{\varphi}}{2} +\frac{1}{2}{\mathbb {N}}_0$. The residues are finite-dimensional. \end{lem} {\it Proof.$\:\:\:\:$} Recall the decomposition $\pi^{P_{U} }_*=\pi^{P_{U} }_++\pi^{P_{U} }_-$ introduced in \ref{weih221}. We have already seen that $\pi^{P_{U} }_-$ has a holomorphic continuation. It suffices to show that $\pi^{P_{U} }_+$ has a meromorphic continuation. Consider $f\in A_{\{1\}}(\sigma_\lambda,{\varphi})^n$. Let $$F(x):=\int_{S(N_V)} {\varphi}(\xi)^{-1} f_0(\xi xw) d\xi$$ (see \ref{weih221} for the definition of $f_0$). This function is smooth in a small neighbourhood of $1\in N$ and on $N\setminus N_V$. Using the Taylor formula for each $r\in{\mathbb {N}}_0$ we can write $F(x^a)=\sum_{q=0}^r F_q(x) a^{q\alpha}+ a^{(r+1)\alpha}R_r(x,a)$, where $F_q$ is a homogeneous polynomial of degree $q\alpha$ on $N$ and $R_r(x,a)$ is uniformly bounded as $a\to 0$. We write $\pi^{P_{U} }_+(f)(xw):=\int_{A_+} {\varphi}(a) F(x^{a^{-1}}) a^{2(\lambda-\rho^U)-n\alpha}da$ and insert the expansion for $F$ in order to obtain $$\pi^{P_{U} }_+(f)(xw):=J_r^1(f)(xw)+J_r^2(f)(xw)\ ,$$ where \begin{eqnarray*} J_r^1(f)(xw)&:=& \sum_{q=0}^r \int_{A_+} {\varphi}(a) a^{2(\lambda-\rho^U)-(q+n)\alpha}da F_q(x)\\ J_r^2(f)(xw)&:= &\int_{A_+} {\varphi}(a)R_r(x,a^{-1}) a^{2(\lambda-\rho^U)-(r+1+n)\alpha}da\ .\end{eqnarray*} The integral $J_r^2$ converges for ${\rm Re }(\lambda)<\rho^U+ (n+r+1)\alpha/2-l_{\varphi}/2$. In order to evaluate $J_r^1$ we introduce the operator $B:=(d/da)_{|a=1}{\varphi}(a)\in {\mathrm{End}}(V_{\varphi})$. In order to define this derivative we embed $A$ into the multiplicative group of ${\mathbb{R}}$. Since ${\varphi}$ is assumed to be algebraic as a representation of $A$ the eigenvalues of $B$ are integral. We have $$J_r^1(f)(xw) = - \sum_{q=0}^r (B+2(\lambda-\rho^U)-(n+q)\alpha)^{-1} F_q(x)\ .$$ In particular we see that $J_r^1$ has a meromorphic continuation to all of ${\aaaa_\C^\ast}$. Since we can choose $r$ arbitrarily large we obtain a meromorphic continuation of $\pi^{P_{U} }_+$ to all of ${\aaaa_\C^\ast}$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Recall the Definition \ref{weih231} of the subspace of polynomials ${R}_{\{1\}}(\sigma_\lambda,{\varphi})^n\subset A_{1}(\sigma_\lambda,{\varphi})^n$. We consider the following diagram $$\xymatrix{{R}_{\{1\}}(\sigma_\lambda,{\varphi})^n\ar[d]\ar[r]&A_{1}(\sigma_\lambda,{\varphi})^n\ar[d]^{\pi^{P_{U} }_*}\\?\ar[r]&A_{P_{U}}(\sigma_\lambda,{\varphi})^n}\ .$$ We can complete the diagram using the concept of an image bundle Definition \ref{imagebundle}. \begin{ddd}\label{weih254} We define ${R}_{U}(\sigma_\lambda,{\varphi})^n \subset A_{P_{U}}(\sigma_\lambda,{\varphi})^n$ as the image bundle of the restriction of $\pi^{P_{U} }_*$ to ${R}_{\{1\}}(\sigma_\lambda,{\varphi})^n$. We define the meromorphic family of maps $$[\pi^{U}_*]:{R}_{\{1\}}(\sigma_\lambda,{\varphi})^n\rightarrow {R}_{U}(\sigma_\lambda,{\varphi})^n$$ to be induced by $\pi^{P_{U}}_*$. Furthermore, we define ${R}_{U,k}(\sigma_\lambda,{\varphi}):=\bigoplus_{n=0}^k {R}_{U}(\sigma_\lambda,{\varphi})^n$ and let $$[\pi^{U}_*]:{R}_{\{1\},k}(\sigma_\lambda,{\varphi})\rightarrow {R}_{U,k}(\sigma_\lambda,{\varphi})$$ denote the corresponding meromorphic family of maps. \end{ddd} Similar to the usage of $AS$ and $L$ the symbol $[\pi^{\Gamma}_*]$ denotes various versions of the push-down on the level of asymptotic terms. It will be clear from the context wich version is meant. \subsubsection{}\label{weih500} Using the second assertion of Lemma \ref{bunle} we choose once and for all meromorphic families of right-inverses $$[Q]: {R}_{U}(\sigma_\lambda,{\varphi})^n \rightarrow {R}_{\{1\}}(\sigma_\lambda,{\varphi})^n $$ of $[\pi^{U}_*]$. \subsubsection{}\label{weih253} Let $\chi\in C^\infty(B_U)$ be a cut-off function such that $1-\chi$ has compact support, and which vanishes in a neighbourhood of $U\backslash (P_U 0_P)$. We can assume that $\chi$ is $P_U$-invariant (otherwise we replace it by the average $xw\mapsto \int_{U\backslash P_U} \chi(u^{-1}xw) du$). Multiplication by $\chi$ defines an inclusion $L:{R}_{U,k}(\sigma_\lambda,{\varphi})\rightarrow C^\infty(B_{U},V_{B_{U}}(\sigma_\lambda,{\varphi}))$. As in Subsection \ref{trivas} we see that $L({R}_{U,k}(\sigma_\lambda,{\varphi}))\cap {S}_{U,k}(\sigma_\lambda,{\varphi})=\{0\}$. \subsubsection{}\label{neuj915} \begin{ddd}\label{neuj104} We define the Fr\'echet space $$B_{U,k}(\sigma_\lambda,{\varphi}):= {S}_{U,k}(\sigma_\lambda,{\varphi})\oplus L({R}_{U,k}(\sigma_\lambda,{\varphi}))\ .$$ Furthermore we define the Fr\'echet and Montel space $B_{U}(\sigma_\lambda,{\varphi})$ to be the intersection of the spaces $B_{U,k}(\sigma_\lambda,{\varphi})$ for all $k\in{\mathbb {N}}$. \end{ddd} Note that $B_{U,k}(\sigma_\lambda,{\varphi})$ fits into the split exact sequence \begin{equation}\label{sse0}0\rightarrow {S}_{U,k}(\sigma_\lambda,{\varphi})\rightarrow B_{U,k}(\sigma_\lambda,{\varphi}) \stackrel{AS}{\rightarrow} {R}_{U,k}(\sigma_\lambda,{\varphi})\rightarrow 0\ ,\end{equation} where $AS$ takes the finite asymptotic expansion. Since the spaces ${S}_{U,k}(\sigma_\lambda,{\varphi})$ and ${R}_{U,k}(\sigma_\lambda,{\varphi})$ form trivial holomorphic bundles we can employ the split $L$ of the sequence in order to equip the family of spaces $\{B_{U,k}(\sigma_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$ with the structure of a trivial holomorphic bundle over ${\aaaa_\C^\ast}$. The space $B_{U}(\sigma_\lambda,{\varphi})$ fits into the exact sequence \begin{equation}\label{sse1}0\rightarrow S_{U}(\sigma_\lambda,{\varphi})\rightarrow B_{U}(\sigma_\lambda,{\varphi})\rightarrow {R}_{U}(\sigma_\lambda,{\varphi})\rightarrow 0\end{equation} which does not admit any continuous split. The spaces $B_{U}(\sigma_\lambda,{\varphi})$ form a projective limit of locally trivial holomorphic bundles in the sense of \ref{weih211}. \subsubsection{} We now show compatibility of the spaces $B_{U}(\sigma_\lambda,{\varphi})$ with twisting. We assume that $\sigma$ is Weyl invariant as in \ref{weih241}. Let $(\pi_{\sigma,\mu},V_{\pi_{\sigma,\mu}})$ be a finite-dimensional representation of $G$ as in \ref{weih144}. Recall the notation $p_{\sigma,\mu}$ and $i_{\sigma,\mu}$ from \ref{weih142}. \begin{lem}\label{compat2} \begin{enumerate} \item For each $n\in{\mathbb {N}}_0$ we have the following commutative diagrams \begin{eqnarray}&& \begin{array}{ccc} R_{\{1\}}(\sigma_\lambda,{\varphi})^n&\stackrel{i_{\sigma,\mu}^n}{\rightarrow}& R_{\{1\}}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m\\ \downarrow [\pi^{U}_*]&&\downarrow [\pi^{U}_*]\\ R_{\{U\}}(\sigma_\lambda,{\varphi})^n&\stackrel{[i_{\sigma,\mu}^{U}]}{\rightarrow}& R_{\{U\}}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m \end{array}\label{uz76}\\&& \begin{array}{ccc} R_{\{1\}}(\sigma_\lambda,{\varphi})^n&\stackrel{p_{\sigma,\mu}^n}{\leftarrow}& R_{\{1\}}(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m\\ \downarrow [\pi^{U}_*]&&\downarrow [\pi^{U}_*]\\ R_{\{U\}}(\sigma_\lambda,{\varphi})^n&\stackrel{[p_{\sigma,\mu}^{U}]}{\leftarrow}& R_{\{U\}}(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m \end{array} \ ,\nonumber \end{eqnarray} where $m:=n+\mu/\alpha$. \item We have holomorphic families of continuous maps $$i_{\sigma,\mu}^U:B_{U}(\sigma_\lambda,{\varphi})\rightarrow B_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$$ and $$p_{\sigma,\mu}^U:B_U(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})\rightarrow B_{U}(\sigma_\lambda,{\varphi})\ .$$ \item $B_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$ is a $Z({\mathfrak{g}})$-module. \item If $\lambda\not\in I_{\mathfrak{a}}$, then $i_{\sigma,\mu}^U$ maps $B_{U}(\sigma_\lambda,{\varphi})$ isomorphically onto $$\ker\{Z(\lambda):B_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})\rightarrow B_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})\}\ ,$$ and the restriction of $p_{\sigma,\mu}^U$ to $$\ker\{\tilde Z(\lambda):B_U(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})\rightarrow B_U(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})\}$$ is an isomorphism onto $B_{U}(\sigma_\lambda,{\varphi})$ (here $\tilde Z(\lambda)$ is the adjoint of $\pi^{1_{-\lambda+\mu}\otimes\tilde\pi_{\sigma,\mu}, {\mathrm {id}}_{\tilde{\varphi}}}(\Pi(\lambda))$). \item The composition $$j_{\sigma,\mu}^U:B_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi}) \stackrel{1-Z(\lambda)}{\rightarrow} \ker_U(Z(\lambda))\stackrel{(i_{\sigma,\mu}^U)^{-1}}{\rightarrow} B_{U}(\sigma_\lambda,{\varphi})$$ which is initially defined for $\lambda\not\in I_{\mathfrak{a}}$ extends to a meromorphic family of continuous maps. Similarly, the composition $$q_{\sigma,\mu}^U:B_{U}(\sigma_\lambda,{\varphi})\stackrel{(p_{\sigma,\mu}^U)^{-1}}{\rightarrow} \ker(\tilde Z(\lambda)) \rightarrow B_U(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})$$ extends to a meromorphic family of continuous maps. \end{enumerate} \end{lem} {\it Proof.$\:\:\:\:$} \subsubsection{} We consider the first diagram of 1. and leave the second to the reader since the argument is similar. First one checks that the inclusion $i_{\sigma,\mu}$ induces an inclusion $$i_{\sigma,\mu}^{P_U,n}:{\cal A}_{P_U}(\sigma_\lambda,{\varphi})^n\rightarrow {\cal A}_{P_U}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})^{m}$$ for all $n\in{\mathbb {N}}_0$, $m:=n+\mu/\alpha$. Now we obtain diagram (\ref{uz76}), where the map $[i_{\sigma,\mu}^{U}]$ is the restriction of $i_{\sigma,\mu}^{P_U,n}$ which is well-defined by the naturality of the image bundle construction Lemma \ref{bunle}. \subsubsection{} Assertion 2. follows from Lemma \ref{compat1}, 2. and 1. Indeed, $i^{U}_{\sigma,\mu}$ maps $B_{U,k}(\sigma_\lambda,{\varphi})$ to $$\{i^U_{\sigma,\mu}\}(\{S_{U,k}(\sigma_\lambda,{\varphi})\})+L([i_{\sigma,\mu}^{U}](R_{U,k}(\sigma_\lambda,{\varphi}))) \subset B_{U,k^\prime}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi}) \ ,$$ where $k^\prime=k+\mu/\alpha$. The argument for the second assertion of 2. is similar. \subsubsection{} In order to prove 3. we first show that the spaces $R_{U}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m$ are a ${\mathcal{Z}}({\mathfrak{g}})$-modules. Note that the representations $\pi^{1_{\lambda+\mu}\otimes\pi_{\sigma,\mu},{\mathrm {id}}_{\varphi}}$ of ${\mathcal{Z}}({\mathfrak{g}})$ and $\pi^{1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi}}$ of $AP_U$ commute. Since we employ the latter representation in order to define homogeneities we see that the spaces $A_{P_U}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m$ are ${\mathcal{Z}}({\mathfrak{g}})$-modules. The space $R_{\{1\}}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$ is clearly a $Z({\mathfrak{g}})$-module since it is the quotient of $C^\infty(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi}))$ by the submodule $S_{\{1\}}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$ (Lemma \ref{compat1}, 3.). Thus the subspaces $R_{\{1\}}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m$ are $Z({\mathfrak{g}})$-modules. We obtain the $Z({\mathfrak{g}})$-module structure on $R_{U}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m$ from Lemma \ref{bunle} since $\pi^{P_U}_*$ is ${\mathcal{Z}}({\mathfrak{g}})$-equivariant. Assertion 3. now follows from Lemma \ref{compat1}, 3. and the fact that for any $A\in {\mathcal{Z}}({\mathfrak{g}})$ the commutator $[\pi^{1_{\lambda+\mu}\otimes\pi_{\sigma,\mu},{\mathrm {id}}_{\varphi}}(A),\chi]$ is a differential operator with compactly supported coefficients on $B_U$. \subsubsection{} We now prove the first assertion of 4. and leave the second to the reader. The exact sequence of bundles (\ref{hj112}) induces an exact sequence $$0\rightarrow R_U(\sigma_\lambda,{\varphi})^n\stackrel{[i^U_{\sigma,\mu}]}{\rightarrow} R_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m \stackrel{p}{\rightarrow} R_U(w,{\varphi})^k\ ,$$ where $m=n+\mu/\alpha$ and we employ an appropriate definition of homogeneity for the last space. If $\lambda\not\in I_{\mathfrak{a}}$, then we have $\ker(Z(\lambda):R_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m \rightarrow R_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})^m)=\ker(p)$. For these $\lambda$ the map $[i^U_{\sigma,\mu}]$ identifies $R_U(\sigma_\lambda,{\varphi})^n$ with the kernel of $Z(\lambda)$. If $f\in B_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$ satisfies $Z(\lambda)(f)=0$, then $Z(\lambda)AS(f)=0$. From what was shown above and (\ref{sse1}) it follows that there exists $g\in B_U(\sigma_\lambda,{\varphi})$ such that $[i^U_{\sigma,\mu}]\circ AS(g) = AS(f)$. Then $f-i^U_{\sigma,\mu}(g)\in S_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$, and by Lemma \ref{compat1}, 4. there exists $h\in S_U(\sigma_\lambda,{\varphi})$ such that $\{i^U_{\sigma,\mu}\}(h)=f-i^U_{\sigma,\mu}(g)$. Thus $f= i^U_{\sigma,\mu}(h+g)$, and this finishes the proof of 4. \subsubsection{} We now prove the first assertion of 5. and leave the second to the reader. We employ the notation introduced in the proof of Lemma \ref{compat1}, 5. For each $n\in{\mathbb {N}}_0$ there is $m\in{\mathbb {N}}_0$ such that $J:S_{U,m}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi}) \rightarrow S_{U,n}(\sigma_\lambda,{\varphi})$ is a holomorphic family of maps. We define the meromorphic family of continuous maps $\tilde j^U_{\sigma,\mu}:B_{U,m}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi}) \rightarrow B_{U,n}(\sigma_\lambda,{\varphi})$ by $$\tilde j^U_{\sigma,\mu}(f):= L\circ [i^U_{\sigma,\mu}]^{-1}\circ AS\circ (1-Z(\lambda))(f) + \{j^U_{\sigma,\mu}\}(f- L\circ AS \circ (1-Z(\lambda))(f))\ .$$ Then $\tilde j^U_{\sigma,\mu}\circ i^U_{\sigma,\mu}(f)=f$ for any $f\in B_{U}(\sigma_\lambda,{\varphi})$. Since $\tilde j^U_{\sigma,\mu}$ vanishes on $\ker(1-Z(\lambda))$ we conclude that the restriction of $\tilde j^U_{\sigma,\mu}$ to $B_{U}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$ coincides with $j^U_{\sigma,\mu}$ for $\lambda\not\in I_{\mathfrak{a}}$. This proves 5. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsection{The push-down for Schwartz spaces}\label{psuscg} \subsubsection{} In this subsection we show that the push-down induces a map between Schwartz spaces $$\{\pi^U_*\}:{S}_{\{1\},k}(1_\lambda,{\varphi})\rightarrow {S}_{U,k}(1_\lambda,{\varphi}) \ .$$ Note that we only consider the spherical case $\sigma=1$. \begin{lem}\label{schwpush} \begin{enumerate} \item The push-down $\{\pi^U_*\}$ converges for ${\rm Re }(\lambda)<\rho^U+(k\alpha-l_{\varphi})/2$. \item For $k_1<k$ it induces a holomorphic family of maps $\{\pi^U_*\}:{S}_{\{1\},k}(1_\lambda,{\varphi})\rightarrow {S}_{U,k_1}(1_\lambda,{\varphi})$. \end{enumerate} \end{lem} {\it Proof.$\:\:\:\:$} \subsubsection{} Let $f\in C_c^\infty(B_{\{1\}},V_{B_{\{1\}}}(1_\lambda,{\varphi}))$. Then $\{\pi^{U}_*\}(f)=\sum_{u\in U}\pi^{1_\lambda,{\varphi}}(u)f$ converges and defines an element of $C_c^\infty(B_{U},V_{B_{U}}(1_\lambda,{\varphi}))$. \subsubsection{} Let $h\in{\mathbb {N}}_0$, $W\subset N\setminus N_V$ be compact, $|.|$ be a ${\varphi}(M_{U})$-invariant norm on $V_{\varphi}$, and $D\in {\mathcal{U}}({\mathfrak{n}})^{\le d \alpha}$ (see \ref{weih243} for notation). Then we consider the seminorm $$q_{W,D,h}(f):=\sup_{x\in W}\sup_{a\in A_+} a^{h\alpha-2(\lambda-\rho^U)}|{\varphi}(a)^{-1}(f)(x^aDw)|\ .$$ The first assertion of the Lemma will easily follow from estimates of the form $$q_{W,D,k}(\{\pi^{U}_*\}f)<C\|f\|_{k,d}\ ,$$ and the second assertion will follow from the convergence of sums of the form $$\sum_{u\in U} q_{W,D,k_1}(\pi^{1_\lambda,{\varphi}}(u) f)<C\|f\|_{k,d} \ .$$ \subsubsection{} If $u\not=1$, then we can write $u^{-1}=m_u\xi_u^{b_u}$ for suitable $\xi_u\in S(N_V)$, $b_u\in A$, and $m_u\in M_{U}$ (see \ref{weih244} for the notation $S(N_V)$). There is a constant $C\in R$ such that for all $f\in C_c^\infty(B_{\{1\}},V_{B_{\{1\}}}(1_\lambda,{\varphi}))$, $x\in W$, $a\in A$, $1\not=u\in U$ with $b_u\ge a$ we have the estimate \begin{eqnarray*} |{\varphi}(a)^{-1}\pi^{1_\lambda,{\varphi}}(u) f(x^aDw)|&=& |{\varphi}(a)^{-1}{\varphi}(\xi_u^{b_u})^{-1} f(m_u\xi_u^{b_u}x^aDw)|\\ &=& |{\varphi}(a^{-1}b_u){\varphi}(\xi_u)^{-1} {\varphi}(b_u^{-1})f(m_u(\xi_ux^{ab_u^{-1}})^{b_u}Dw)|\\ &\le& C\|f\|_{d,k} |{\varphi}(a^{-1}b_u)| b_u^{2(\lambda-\rho)-k\alpha} \end{eqnarray*} and for $b_u\le a$ \begin{eqnarray*} |{\varphi}(a)^{-1}\pi^{1_\lambda,{\varphi}}(u) f(x^aDw)|&=& |{\varphi}(a)^{-1} {\varphi}(\xi_u^{b_u})^{-1} f(m_u\xi_u^{b_u}x^aDw)|\\ &=& |{\varphi}(\xi_u^{a^{-1}b_u})^{-1} {\varphi}(a)^{-1}f(m_u(\xi^{b_ua^{-1}}x)^{a}Dw)|\\ &\le& C\|f\|_{d,k} a^{2(\lambda-\rho)-k\alpha}\ . \end{eqnarray*} \subsubsection{} If ${\rm Re }(\lambda)<\rho^U+(k\alpha-l_{\varphi})/2$, then summing up these estimates over $U$ and estimating the sum by an integral over $P_{U}$ we obtain $C_1\in{\mathbb{R}}$ such that all $x\in W$, $a\in A^+$, $f\in C_c^\infty(B_{\{1\}},V_{B_{\{1\}}}(1_\lambda,{\varphi}))$ $$|{\varphi}(a)^{-1}\{\pi^{U^0}_*\}(f)(x^aDw)|< C_1 a^{2(\lambda-\rho^U)-k\alpha}\ .$$ This implies $$q_{W,D,k}(\{\pi^{U}_*\}f)<C_1\|f\|_{k,d}\ .$$ \subsubsection{} Multiplying the two inequalities above by $a^{k_1\alpha-2(\lambda-\rho^U)}$ and taking the supremum over $x$ and $a$ we obtain $q_{W,D,k_1}(\pi^{1_\lambda,{\varphi}}(u) f)<b_u^{-2\rho_U-(k-k_1)\alpha}$. Summing this over $U$ and estimating the sum over $U$ by the integral over $P_{U}$ we obtain $$\sum_{u\in U} q_{W,D,k_1}(\pi^{1_\lambda,{\varphi}}(u) f)<C\|f\|_{k,d}\ .$$ \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Recall (Definition \ref{weih245}) that ${S}_{\{1\}}(1_\lambda,{\varphi})$ is the intersection of the spaces ${S}_{\{1\},k}(1_\lambda,{\varphi})$ for $k\in {\mathbb {N}}_0$. Hence Lemma \ref{schwpush} implies the following corollary. \begin{kor}\label{tzu} We have a holomorphic family of maps $$\{\pi^U_*\}:{S}_{\{1\}}(1_\lambda,{\varphi})\rightarrow {S}_U(1_\lambda,{\varphi})$$ defined on all of ${\aaaa_\C^\ast}$. \end{kor} \subsubsection{} We now turn to the problem of constructing a right-inverse of the push-down for Schwartz spaces. Its adjoint plays an important role in the construction of the restriction map. Below we encounter the effect of a loss of regularity ($-2\rho_U$). This is one of the places where the presence of cusps makes the theory much more complicated in comparison with the convex cocompact case. \subsubsection{} \begin{lem} \label{schwspl} For all $k_1\in{\mathbb {N}}_0$ satisfying $k_1 \alpha<k\alpha-2\rho_U$ there exists a holomorphic family of maps $$\{Q\}:{S}_{U,k}(1_\lambda,{\varphi})\rightarrow {S}_{\{1\},k_1}(1_\lambda,{\varphi})$$ such that $\{\pi^U_*\}\circ \{Q\}$ is the natural inclusion ${S}_{\{1\},k}(1_\lambda,{\varphi})\hookrightarrow {S}_{\{1\},k_1}(1_\lambda,{\varphi})$. \end{lem} {\it Proof.$\:\:\:\:$} We define a holomorphic family $\{Q\}$ of right-inverses of $\{\pi^U_*\}$. We employ a cut-off function $\chi^U\in C^\infty(\Omega_P)$ with $\sum_{u\in U} u^*\chi^U=1$, and such that $\chi^U(xD)$ is bounded for each $D\in{\mathcal{U}}({\mathfrak{n}})$. Here we have identified $\Omega_P$ with $N$. Then $\chi^U$ considered as a function on $N$ can be derived with respect to the left invariant differential operator $D$. To see that such a function exists we equip $\Omega_P\cong N$ with a $MN$-invariant Riemannian metric such that $U$ acts isometrically. Note that the Riemannian manifold $U\backslash N$ admits a lower bound of the injectivity radius. For $f\in {S}_{U,k}(1_\lambda,{\varphi})$ we define $\{Q\}(f)$ to be the lift of $f$ to $\Omega_P$ multiplied by $\chi^U$. It is then easy to see that $\{Q\}: {S}_{U,k}(1_\lambda,{\varphi}) \rightarrow {S}_{\{1\},k_1}(1_\lambda,{\varphi})$ for all $k_1\in{\mathbb {N}}_0$ satisfying $k_1 \alpha<k\alpha-2\rho_U$. Therefore we obtain a holomorphic family of maps $\{Q\}:{S}_{U}(1_\lambda,{\varphi})\rightarrow {S}_{\{1\}}(1_\lambda,{\varphi})$. By definition $\{\pi^U_*\}\circ \{Q\}={\mathrm {id}}$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsection{Push down for cusps of smaller rank}\label{weih261} \subsubsection{} We assume that the cusp associated to $U\subset P$ does not have full rank. In order to construct the push-down it remains to extend $\{\pi^U_*\}$ (constructed in Lemma \ref{schwpush}) for each $k\in{\mathbb {N}}$ from ${S}_{\{1\},k}(1_\lambda,{\varphi})$ to $B_{\{1\},k}(1_\lambda,{\varphi})$ such that the following diagram is commutative: $$ \begin{array}{ccccccccc} 0&\rightarrow&{S}_{\{1\},k}(1_\lambda,{\varphi})&\rightarrow& B_{\{1\},k}(1_\lambda,{\varphi})&\stackrel{AS}{\rightarrow}& {R}_{\{1\},k}(1_\lambda,{\varphi})&\rightarrow &0\\ & &\downarrow \{\pi^{U}_*\} & & \downarrow \pi^{U}_* & & \downarrow [\pi^{U}_*] &&\\ 0&\rightarrow&{S}_{U,k}(1_\lambda,{\varphi})&\rightarrow& B_{U,k}(1_\lambda,{\varphi})& \stackrel{AS}{\rightarrow}&{R}_{U,k}(1_\lambda,{\varphi})&\rightarrow &0\ .\end{array} $$ The main technical result in this direction is the following Proposition \ref{mainpure}. \subsubsection{} Note that ${R}_{\{1\}}(1_\lambda,{\varphi})^n$ is a finite-dimensional representation of the torus $M_{U^0}$ (see \ref{weih251} for the explanation of $M_{U^0}$). Let ${\mathcal{X}}(M_{U^0})$ denote the set of characters of $M_{U^0}$. For any $\theta\in {\mathcal{X}}(M_{U^0})$ let ${R}_{\{1\}}(1_\lambda,{\varphi})^n_\theta$ be the subspace of all $f\in {R}_{\{1\}}(1_\lambda,{\varphi})^n$ such that $$\pi^{1_\lambda,{\varphi}}(t)f=\theta(t)f\ .$$ Then we have a finite decomposition $${R}_{\{1\}}(1_\lambda,{\varphi})^n= \bigoplus_{\theta\in{\mathcal{X}}(M_{U^0}) } {R}_{\{1\}}(1_\lambda,{\varphi})^n_\theta\ .$$ \subsubsection{} Recall the construction of $L$ from \ref{weih253}. Furthermore see Definition \ref{weih254} for $[\pi^{U^0}_*]$, Definition \ref{weih215} for $\rho^U$ and \ref{weih213} for $l_{\varphi}$. \begin{prop}\label{mainpure} The composition $$\pi^{U^0}_*\circ L:{R}_{\{1\}}(1_\lambda,{\varphi})^n_\theta\rightarrow B_{U^0}(1_\lambda,{\varphi})$$ converges for ${\rm Re }(\lambda)<\rho^U+(n\alpha-l_{\varphi})/2$ and has a meromorphic continuation to all of ${\aaaa_\C^\ast}$ such that $AS\circ \pi^{U^0}_*\circ L = [\pi^{U^0}_*]$. \end{prop} {\it Proof.$\:\:\:\:$} \subsubsection{} We decompose the push-down $\pi^{U^0}_*$ into two intermediate maps. The first map is the push-down $\pi^{[U^0,U^0]}_*\circ L : {R}_{\{1\}}(1_\lambda,{\varphi})^n_\theta\rightarrow B_{[U^0,U^0]}(1_\lambda,{\varphi})$ with respect to the commutator group $[U^0,U^0]$, and the second is a relative push-down $\pi^{U^0/[U^0,U^0]}$. Note that $M_{U^0}$ centralizes $[U^0,U^0]$ (see \ref{weih131}) and therefore acts on the sequence $$0\rightarrow {S}_{[U^0,U^0]}(1_\lambda,{\varphi})\rightarrow B_{[U^0,U^0]}(1_\lambda,{\varphi})\rightarrow {R}_{[U^0,U^0]}(1_\lambda,{\varphi})\rightarrow 0\ .$$ The representation of $M_{U^0}$ on ${R}_{[U^0,U^0]}(1_\lambda,{\varphi})$ is compatible with the decomposition \linebreak[4] $\bigoplus_{n=0}^\infty {R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n$. \subsubsection{}\label{weih257} Let ${R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n_\theta$ be the subspace of all $f\in {R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n$ satisfying $$\pi^{1_\lambda,{\varphi}}(t)f=\theta(t)f\ .$$ Then we have a further finite decomposition $${R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n= \bigoplus_{\theta\in{\mathcal{X}}(M_{U^0}) } {R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n_\theta\ .$$ \subsubsection{} We will first show \begin{lem} \label{lpro1} The push-down over $[U^0,U^0]$ defines a meromorphic family of maps $$\pi^{[U^0,U^0]}:{R}_{\{1\}}(1_\lambda,{\varphi})^n_\theta\rightarrow {S}_{[U^0,U^0]}(1_\lambda,{\varphi}) \oplus L({R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n_\theta)$$ such that $$AS\circ \pi^{[U^0,U^0]}\circ L=[\pi^{[U^0,U^0]}]\ .$$ \end{lem} It is clear that the average $\{\pi_*^{U^0/[U^0,U^0]}\}: {S}_{[U^0,U^0]}(1_\lambda,{\varphi})\rightarrow {S}_{U^0}(1_\lambda,{\varphi})$ over $U^0/[U^0,U^0]$ converges and depends holomorphically on $\lambda$. A simple way to see this formally is to write $$\{\pi_*^{U^0/[U^0,U^0]}\}=\{\pi^{U^0}_*\}\circ\{Q\}\circ \pi_*^{[U^0,U^0]}$$ using the split $\{Q\}$ constructed in Lemma \ref{schwspl} and Corollary \ref{tzu}. Proposition \ref{mainpure} now immediately follows from \begin{lem}\label{lpro2} The composition $$\pi^{U^0/[U^0,U^0]}_*\circ L: {R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n_\theta \rightarrow B_{U^0}(1_\lambda,{\varphi})$$ defines a meromorphic family of maps such that $$AS\circ \pi^{U^0/[U^0,U^0]}_* \circ L \circ [\pi_*^{[U^0,U^0]}]= [\pi^{U^0}_*]\ .$$ \end{lem} \subsubsection{} We now start with the proof of Lemma \ref{lpro1}. Since $M_{U^0}$ is abelian the homomorphism $m:N_V\rightarrow M_{U^0}$ vanishes on $[N_V,N_V]$. Therefore $[U^0,U^0]$ is a discrete subgroup of the center of $N$. If $f\in {R}_{\{1\}}(1_\lambda,{\varphi})^n$, then for all $d\in{\mathbb {N}}_0$ we have $\|L(f)\|_{n,d}<\infty$. Therefore the same arguments as in the proof of Lemma \ref{schwpush} show that the push-down $\pi^{[U^0,U^0]}_*\circ L$ converges for ${\rm Re }(\lambda)<\rho^{{[U^0,U^0]}}+(n\alpha-l_{\varphi})/2$. \subsubsection{}\label{weih256} We consider the abelian group $Z:=P_{{[U^0,U^0]}}=[N_V,N_V]$. If we are given an unitary character $\vartheta\in {\mathcal{X}}(Z)$, then we consider the push-down $\pi^{Z,\vartheta}_*$, which associates to $f\in C^\infty(\partial X,V(1_\lambda,{\varphi}))$ the average $$\pi^{Z,\vartheta}_*(f):=\int_{Z} \vartheta(z)^{-1} \pi^{1_\lambda,{\varphi}}(z)(f_{|\Omega}) dz$$ provided the integral converges. \subsubsection{} Observe that $[U^0,U^0]$ is a cocompact discrete subgroup of $Z$. We normalize the Haar measure of $Z$ such that ${\rm vol}({[U^0,U^0]}\backslash Z)=1$. Let ${\mathcal{X}}(Z,{[U^0,U^0]})$ denote the set of all unitary characters of $Z$ which are trivial on ${[U^0,U^0]}$. We identify ${\mathcal{X}}(Z)$ with $\imath \mathfrak{z}^*$ and ${\mathcal{X}}(Z,{[U^0,U^0]})$ with a lattice in $\imath \mathfrak{z}^*$, where $\mathfrak{z}$ denotes the Lie algebra of $Z$. Our approach is based on the Poisson summation formula which states that $$\pi^{[U^0,U^0]}_*(f)=\sum_{\vartheta\in{\mathcal{X}}(Z,{[U^0,U^0]})} \pi^{Z,\vartheta}_*(f)$$ in the domain of convergence. \subsubsection{} We choose an euclidean structure on $\mathfrak{z}$. This structure induces norms on $\mathfrak{z}$ and $\mathfrak{z}^*$. Furthermore we use the euclidean structure in order to define a Laplace operator $\Delta\in {\mathcal{U}}(\mathfrak{z})$ which is normalized such that if $\vartheta\in{\mathcal{X}}(Z)$, then $\vartheta(\Delta^l x)=|\vartheta|^{2l} \vartheta(x)$. Let $COP:{\mathcal{U}}(\mathfrak{z})\rightarrow {\mathcal{U}}(\mathfrak{z})\otimes {\mathcal{U}}(\mathfrak{z})$ denote the coproduct on ${\mathcal{U}}(\mathfrak{z})$. There are $A_{l,j},B_{l,j},C_{l,j}\in{\mathcal{U}}(\mathfrak{z})$ such that $$(COP\otimes 1)\circ COP(\Delta^l)=\sum_{j} A_{l,j}\otimes B_{l,j}\otimes C_{l,j}\ .$$ Let $D\mapsto \tilde D$ be the canonical antiautomorphism of ${\mathcal{U}}(\mathfrak{z})$. \subsubsection{} Fix $l\in{\mathbb {N}}$ and let $f\in {R}_{\{1\}}(1_\lambda,{\varphi})^n_\theta$. Recall that $L(f)(xw)=\chi(a(xw)) f(xw)$, where $\chi\in C^\infty(A)$ was some cut-off function which vanishes on a neighbourhood of $A_-$ and is equal to one near $\infty$. By $S(Z)$ we denote the unit-sphere in $Z$. For $\xi\in S(Z)$ and $a\in A$ we compute \begin{eqnarray} &&\pi^{1_\lambda,{\varphi}}(\Delta^l\xi^a) L(f)(.w)\nonumber\\ &=&\pi^{1_\lambda,{\varphi}}(a)\pi^{1_\lambda,{\varphi}}((\Delta^l)^{a^{-1}}\xi) \pi^{1_\lambda,{\varphi}}(a^{-1})\chi (a(.w))f(.w)\nonumber\\ &=&a^{\rho-\lambda-4l\alpha}\pi^{1_\lambda,{\varphi}}(a)\pi^{1_\lambda,{\varphi}}(\Delta^l\xi) {\varphi}(a^{-1}) \chi(a(a.a^{-1}w)) f(a. a^{-1}w) \nonumber \\ &=&a^{\lambda-\rho-(n+4l)\alpha}\pi^{1_\lambda,{\varphi}}(a)\pi^{1_\lambda,{\varphi}}(\Delta^l) {\varphi}(\xi) \chi(a^2a(\xi^{-1}.w)) f(\xi^{-1}.w)\nonumber\\ &=&a^{\lambda-\rho-(n+4l)\alpha}\sum_{j} \pi^{1_\lambda,{\varphi}}(a) {\varphi}(A_{l,j}\xi) \chi(a^2 a(\xi^{-1}\tilde B_{l,j}.w)) f(\xi^{-1}\tilde C_{l,j}.w)\nonumber\\ &=& a^{2(\lambda-\rho)-(n+4l)\alpha}\sum_{j} {\varphi}(a) {\varphi}(A_{l,j}\xi) \chi(a^2 a(\xi^{-1}\tilde B_{l,j}a^{-1}.aw)) f(\xi^{-1}\tilde C_{l,j}a^{-1}.aw)\label{zzz}\ . \end{eqnarray} \subsubsection{} If $W\subset N$ is any compact subset and $D\in {\mathcal{U}}({\mathfrak{n}})$, then there is a constant $C\in{\mathbb{R}}$ such that $$|\pi^{1_\lambda,{\varphi}}(\Delta^l\xi^a) L(f)(yDw)|\le C a^{2(\lambda-\rho)-(n+4l)\alpha+l_{\varphi}}$$ for all $\xi\in S(Z)$ and $a\in A_+$. If ${\rm Re }(\lambda)$ is sufficiently small, then by partial integration for $\vartheta\in {\mathcal{X}}(Z,{[U^0,U^0]})$, $\vartheta\not=0$ \begin{eqnarray*} \pi^{Z,\vartheta}_*(L(f))&=&\int_{Z} \vartheta(x)^{-1}\pi^{1_\lambda,{\varphi}}(x) L(f) dx\\ &=&\frac{1}{|\vartheta|^{2l}}\int_{Z} \vartheta(\Delta^lx)^{-1} \pi^{1_\lambda,{\varphi}}(x) L(f) dx\\ &=&\frac{1}{|\vartheta|^{2l}}\int_{Z} \vartheta(x)^{-1}\pi^{1_\lambda,{\varphi}}(\Delta^l x) L(f) dx\\ &=&\frac{1}{|\vartheta|^{2l}}\int_{S(Z)} \int_A \vartheta(\xi^a)^{-1}\pi^{1_\lambda,{\varphi}}(\Delta^l \xi^a) L(f)a^{2\rho_{{[U^0,U^0]}}} da d\xi\ . \end{eqnarray*} By the estimate above the integral converges locally uniformly on $\{{\rm Re }(\lambda)<\rho^{{[U^0,U^0]}}+(n+4l)\alpha/2 -l_{\varphi}/2\}$, and for any compact subset of this region, $D\in{\mathcal{U}}({\mathfrak{n}})$, and compact subset $W\subset N$ there is a constant $C_1$ such that $$|\pi^{Z,\vartheta}_*(L(f))(yDw)|\le \frac{C_1}{ |\vartheta|^{2l}}$$ for all $y\in W$. \subsubsection{} Choosing $2l>\dim(Z)$ we see that the sum $$\sum_{0\not=\vartheta\in{\mathcal{X}}(Z,{[U^0,U^0]})} \pi^{Z,\vartheta}_*(L(f))$$ converges in $C^\infty(\Omega_P,V(1_\lambda,{\varphi}))$. \subsubsection{} Refining the estimates above we now show that this sum in fact converges in the space of rapidly decreasing functions. Let $W_1\subset N\setminus Z$ be a compact subset and $D\in{\mathcal{U}}({\mathfrak{n}})$. It immediately follows from (\ref{zzz}), that there is a constant $C_2$ such that for all $\xi\in S(Z)$, $a,b\in A_+$, with $a \ge b$ and $y\in W_1$ $$|{\varphi}(b)^{-1}(\pi^{1_\lambda,{\varphi}}(\Delta^l\xi^a) L(f))(y^bDw)|\le C_2 a^{2(\lambda-\rho)-(n+4l)\alpha}| {\varphi}(ab^{-1})| \ .$$ Let $COP(D)=\sum_{h} D_h^l\otimes D_h^r$ denote the coproduct of $D$, where here $COP:{\mathcal{U}}({\mathfrak{n}})\rightarrow {\mathcal{U}}({\mathfrak{n}})\otimes{\mathcal{U}}({\mathfrak{n}})$. We find a constant $C_3\in {\mathbb{R}}$ such that for all $\xi\in S(Z)$, $a,b\in A_+$ with $a \le b$ and $y\in W_1$ \begin{eqnarray*} &&|{\varphi}(b)^{-1}(\pi^{1_\lambda,{\varphi}}(\Delta^l\xi^a) L(f))(y^bDw)|\\ &=& a^{2(\lambda-\rho)-(n+4l)\alpha}|\sum_{j,h} {\varphi}(ab^{-1}) {\varphi}(A_{l,j}\xi) \chi(a^2 a(\xi^{-1}\tilde B_{l,j}a^{-1}y^bD^l_haw)) f(\xi^{-1}\tilde C_{l,j}a^{-1}y^bD^r_haw)|\\ &=& a^{-4l\alpha} b^{2(\lambda-\rho)-n\alpha} |\sum_{j,h} {\varphi}((A_{l,j}\xi)^{ab^{-1}}) \chi(b^2 a((\xi^{-1}\tilde B_{l,j})^{b^{-1}a}y(D_h^l)^{b^{-1}}w)) f((\xi^{-1}\tilde C_{l,j})^{b^{-1}a}y(D_h^r)^{b^{-1}}w)|\\ &\le& C_3 b^{2(\lambda-\rho)-(n+4l)\alpha}\ . \end{eqnarray*} Using this estimate we see that $${\varphi}(b)^{-1}\sum_{0\not=\vartheta\in{\mathcal{X}}(Z,{[U^0,U^0]})} \pi^{Z,\vartheta}_*(L(f))(y^bDw)$$ can be estimated by $C b^{2(\lambda-\rho^{{[U^0,U^0]}})-(n+4l)\alpha}$, where $C$ can be choosen uniformly for $y\in W_1$ and $\lambda$ in compact subsets of $\{{\rm Re }(\lambda)<\rho^{{[U^0,U^0]}}+(n+4l)\alpha/2 -l_{\varphi}/2\}$. Since we can choose $l$ arbitrary large we obtain a holomorphic continuation of the sum above to all of ${\aaaa_\C^\ast}$. Moreover we see that this sum is rapidly decreasing with respect to $b$. \subsubsection{} We now consider $\pi^{Z,0}_*(L(f))$. If ${\rm Re }(\lambda)$ is sufficiently small then we can write $$\pi^{Z,0}_*(L(f))=\int_A \int_{S(Z)} \pi^{1_\lambda,{\varphi}}(\xi^a) L(f) d\xi a^{2\rho_{[U^0,U^0]}} da\ .$$ We again employ the relation \begin{eqnarray*} &&\pi^{1_\lambda,{\varphi}}(\xi^a) L(f)(yw)\\ &=& a^{2(\lambda-\rho)-n\alpha} {\varphi}(a) {\varphi}(\xi) \chi(a^2 a(\xi^{-1}a^{-1}yaw)) f(\xi^{-1}a^{-1}yaw)\ . \end{eqnarray*} Note that $$F(a,y):= \int_{S(Z)}{\varphi}(\xi) \chi(a^2 a(\xi^{-1}yw)) f(\xi^{-1}yw) d\xi$$ is a smooth function of $y$ near $y=0$ which is independent of $a$ for large $a$. Let $$F(a,y^{a^{-1}})=\sum_{q=0}^r F_q(y)a^{-q\alpha} + a^{-(r+1)\alpha} R_r(a,y)$$ be the asymptotic expansion for large $a\in A$ obtained from the Taylor series of $F(a,y)$ at $y=0$. The remainder $R_r(a,y)$ remains bounded as $a\to\infty$. We write $$\pi^{Z,1}_*(L(f))(yw)=I_+(y)+I_-(y)\ ,$$ where \begin{eqnarray*} I_+(y)&:=&\int_{A_+} a^{2(\lambda-\rho^{[U^0,U^0]})-n\alpha} {\varphi}(a) F(a,y^{a^{-1}}) da\\ I_-(y)&:=&\int_{A_-}\int_{S(Z)} \pi^{1_\lambda,{\varphi}}(\xi^a) L(f)(yw) d\xi a^{2\rho_{[U^0,U^0]}} da \ .\end{eqnarray*} The integral $I_-$ converges for all $\lambda\in{\aaaa_\C^\ast}$ and defines a holomorphic family of smooth functions. We write $I_+(y):=J_r^1(y)+J_r^2(y)$, where \begin{eqnarray*} J_r^1(y)&:=& \int_{A_+} a^{2(\lambda-\rho^{[U^0,U^0]})-n\alpha} {\varphi}(a) \sum_{q=0}^r F_q(y)a^{-q\alpha} da\\ J_r^2(y)&:=& \int_{A_+} a^{2(\lambda-\rho^{[U^0,U^0]})-n\alpha} {\varphi}(a) a^{-(r+1)\alpha} R_r(a,y) da\ . \end{eqnarray*} The integral $J_r^2$ converges for ${\rm Re }(\lambda)<\rho^{[U^0,U^0]}+(n+r+1)\alpha/2-l_{\varphi}/2$ and defines a smooth function in $y$. The integral $J_r^1$ can be evaluated: $$J_r^1(y)=-\sum_{q=0}^r (B+2(\lambda-\rho^{[U^0,U^0]})-(n+q)\alpha)^{-1} F_q(y)\ ,$$ where $B:=\frac{d}{da}_{|a=1}{\varphi}(a)\in {\mathrm{End}}(V_{\varphi})$ (compare with the proof of Lemma \ref{aaww}). It obviously defines a meromorphic family of smooth functions. Since we can choose $r$ arbitrary large we obtain a meromorphic continuation of $\pi^{Z,0}_*(L(f))$ to all of ${\aaaa_\C^\ast}$. Note that if $y\in N\setminus Z$ and $b\in A$ is sufficiently large, then we have $\pi^{Z,0}_*(L(f))(y^bw)=\pi_*^{P_{[U^0,U^0]}}(f)(y^bw)$, where $\pi_*^{P_{[U^0,U^0]}}$ was discussed in Subsection \ref{asz}. We conclude that $\pi^{Z,0}_*(L(f))\in S_{{[U^0,U^0]}}(1_\lambda,{\varphi})\oplus L({R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n)$. We have shown that $$\pi^{[U^0,U^0]}_*\circ L:{R}_{\{1\}}(1_\lambda,{\varphi})^n\rightarrow S_{{[U^0,U^0]}}(1_\lambda,{\varphi})\oplus L({R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n_\theta)\subset B_{{[U^0,U^0]}}(1_\lambda,{\varphi})$$ has a meromorphic continuation to all of ${\aaaa_\C^\ast}$ such that $AS\circ \pi^{[U^0,U^0]}_*\circ L = [\pi^{[U^0,U^0]}_*]$. This finishes the proof of the lemma. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} \label{weih258} We now prove Lemma \ref{lpro2} in a similar manner. Since $M_{U^0}$ centralizes $N_V$ we can form the direct product $P_1:=P_{U^0}/Z=M_{U^0}T$ where $T:=N_V/Z$ (the group $Z$ was defined in \ref{weih256}). By construction of $U^0$ there is a lattice $V_1 \subset T$ and a homomorphism $m:T\rightarrow M_{U^0}$ such that ${U^0/{[U^0,U^0]}}=\{m(v)v|v\in {V_1}\}\subset P_1$. For $f\in {R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n_\theta$ (this space is defined in \ref{weih257}) we have $$\pi^{1_\lambda,{\varphi}}(m(v)v)L(f)(yw)= \theta^m(v) {\varphi}(\tilde v) f(\tilde v^{-1}yw)\ ,$$ where $\tilde v\in N$ is any lift of $v\in T$, and $\theta^m=\theta\circ m\in {\mathcal{X}}(T)$. \subsubsection{} We normalize the Haar measure on $T$ such that ${\rm vol}({V_1}\backslash T)=1$. Given a unitary character $\vartheta\in {\mathcal{X}}(T)$, we consider the push-down $\pi^{T,\vartheta}_*$, which associates to $f\in C^\infty(B_Z,V_{B_Z}(1_\lambda,{\varphi}))$ the average $$\pi^{T,\vartheta}_*(f):=\int_{T} \vartheta(z)^{-1} \pi^{1_\lambda,{\varphi}}(z) f dz$$ provided the integral converges. Note that we define $L:{R}_{[U^0,U^0]}(1_\lambda,{\varphi})_\theta^n\rightarrow B_{[U^0,U^0]}(1_\lambda,{\varphi})$ using a $Z$-invariant cut-off function (recall that $Z=P_{[U^0,U^0]}$). It follows from the proof of Lemma \ref{lpro1} that all $f\in {R}_{[U^0,U^0]}(1_\lambda,{\varphi})_\theta^n$ are $Z$-invariant. Therefore $L(f)\in C^\infty(B_Z,V_{B_Z}(\theta,{\varphi}))$ and $\pi^{T,\vartheta}_*(L(f))$ is well-defined (up to convergence). \subsubsection{} We fix an isomorphism of abelian groups $T\cong {\mathbb{R}}^{\dim(T)}$ preserving the Haar measure such that we obtain an euclidean structure on $T$ and its Lie algebra ${\mathfrak{t}}$. As before let ${\mathcal{X}}(T,{V_1})$ denote the set of all unitary characters of $T$ which are trivial on ${V_1}$. We identify ${\mathcal{X}}(T)$ with $\imath {\mathfrak{t}}^*$ and ${\mathcal{X}}(T,{V_1})$ with a lattice in $\imath {\mathfrak{t}}^*$. If $f\in {R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n_\theta$, then by the Poisson summation formula $$\pi^{U^0/{[U^0,U^0]}}_*(L(f))=\sum_{\vartheta\in{\mathcal{X}}(T,{V_1})} \pi^{T,\vartheta-\theta^m}_*(L(f))$$ in the domain of convergence. \subsubsection{} Note that if $\theta^m\in {\mathcal{X}}(T,{V_1})$, then $\theta=0$ by the construction of $M_{U^0}$. If $\theta\not=0$, then will show that $\sum_{\vartheta\in{\mathcal{X}}(T,{V_1})} \pi^{T,\vartheta-\theta^m}_*(L(f))$ gives rise to a rapidly decreasing section for all $\lambda\in {\aaaa_\C^\ast}$. If $\theta=0$, then again $\sum_{0\not=\vartheta\in{\mathcal{X}}(T,{V_1})} \pi^{T,\vartheta-\theta^m}_*(L(f))$ is rapidly decreasing. The remaining term $\pi^{T,0}_*(L(f))$ contributes to ${S}_{U^0}(1_\lambda,{\varphi})\oplus L({R}_{U^0}(1_\lambda,{\varphi})^n_\theta)$ and depends meromorphically on $\lambda$. \subsubsection{} Observe that $A$ normalizes $Z$. This is a consequence of the assumption that the cusp is regular, and that the Langlands decomposition of $P$ is adapted (Definition \ref{t799}). It is in fact the reason for making this assumption. We see that $A$ acts on $T$. In particular ${\mathfrak{t}}^*$ decomposes into two eigenspaces ${\mathfrak{t}}={\mathfrak{t}}_1^*\oplus{\mathfrak{t}}_2^*$ (we write $\vartheta=\vartheta_1\oplus \vartheta_2$ for the corresponding decomposition of $\vartheta\in{\mathfrak{t}}^*$) with respect to $A$ such that $A$ acts on ${\mathfrak{t}}_i^*$ by $a^{i\alpha}$, $i=1,2$. We define the $A$-homogeneous "norm" on ${\mathfrak{t}}^*$ by $|\vartheta|:=(|\vartheta_1|^4+|\vartheta_2|^2)^{1/2}$. Furthermore we use the euclidean structure on ${\mathfrak{t}}_i$ in order to define Laplace operators $\Delta_i\in {\mathcal{U}}({\mathfrak{t}}_i)$ and the $A$-homogeneous operator $\Delta:=\Delta_1^2+\Delta_2$. We fix the normalizations such that for $\vartheta\in{\mathcal{X}}(T)$ we have $\vartheta(\Delta^l x)=|\vartheta|^{2l} \vartheta(x)$. Let $\rho_{U^0/{[U^0,U^0]}}\in{\mathfrak{a}}^*$ be such that $A$ acts on $\Lambda^{max}{\mathfrak{t}}$ by the character $a^{2\rho_{U^0/{[U^0,U^0]}}}$. Note that $\rho_{U^0/{[U^0,U^0]}}+\rho_{[U^0,U^0]}=\rho_{U}$. \subsubsection{} Let $COP:{\mathcal{U}}({\mathfrak{t}})\rightarrow {\mathcal{U}}({\mathfrak{t}})\otimes {\mathcal{U}}({\mathfrak{t}})$ denote the coproduct on ${\mathcal{U}}({\mathfrak{t}})$. There are $A_{l,j},B_{l,j},C_{l,j}\in{\mathcal{U}}({\mathfrak{t}})$ such that $$(COP\otimes 1)\circ COP(\Delta^l)=\sum_{j} A_{l,j}\otimes B_{l,j}\otimes C_{l,j}\ .$$ Let $D\mapsto \tilde D$ be the canonical antiautomorphism of ${\mathcal{U}}({\mathfrak{t}})$. Recall that $L(f)(xw)=\chi(xw) f(xw)$, where $\chi$ is $Z$-invariant. By $S(T)$ we denote the unit-sphere in $T$. For $\xi\in S(T)$ and $a\in A$ we compute \begin{eqnarray} &&\pi^{1_\lambda,{\varphi}}(\Delta^l\xi^a) L(f)(.w)\nonumber\\ &=&\pi^{1_\lambda,{\varphi}}(a)\pi^{1_\lambda,{\varphi}}((\Delta^l)^{a^{-1}}\xi) \pi^{1_\lambda,{\varphi}}(a^{-1})\chi f(.w)\nonumber\\ &=&a^{\rho-\lambda-4l\alpha}\pi^{1_\lambda,{\varphi}}(a)\pi^{1_\lambda,{\varphi}}(\Delta^l\xi) {\varphi}(a^{-1}) \chi(a.a^{-1}w) f(a. a^{-1}w) \nonumber \\ &=&a^{\lambda+\rho-2\rho^{[U^0,U^0]}-(n+4l)\alpha}\pi^{1_\lambda,{\varphi}}(a)\pi^{1_\lambda,{\varphi}}(\Delta^l) {\varphi}(\xi) \chi(a\xi^{-1}.a^{-1}w) f(\xi^{-1}.w)\nonumber\\ &=&a^{\lambda+\rho-2\rho^{[U^0,U^0]}-(n+4l)\alpha}\sum_{j} \pi^{1_\lambda,{\varphi}}(a) {\varphi}(A_{l,j}\xi) \chi(a\xi^{-1}\tilde B_{l,j}.a^{-1}w) f(\xi^{-1}\tilde C_{l,j}.w)\nonumber\\ &=& a^{2(\lambda-\rho^{[U^0,U^0]})-(n+4l)\alpha}\sum_{j} {\varphi}(a) {\varphi}(A_{l,j}\xi) \chi(a\xi^{-1}\tilde B_{l,j}a^{-1}.w) f(\xi^{-1}\tilde C_{l,j}a^{-1}.aw)\label{zzz1}\ . \end{eqnarray} If $W\subset N$ is any compact subset and $D\in {\mathcal{U}}({\mathfrak{n}})$, then there is a constant $C\in{\mathbb{R}}$ such that $$|\pi^{1_\lambda,{\varphi}}(\Delta^l\xi^a) L(f)(yDw)|\le C a^{2(\lambda-\rho^{[U^0,U^0]})-(n+4l)\alpha+l_{\varphi}}$$ for all $\xi\in S(T)$ and $a\in A_+$. \subsubsection{} If ${\rm Re }(\lambda)$ is sufficiently small, then by partial integration for $\vartheta\in {\mathcal{X}}(T,{V_1})$, $\vartheta-\theta^m\not=0$ \begin{eqnarray*} \pi^{Z,\vartheta-\theta^m}_*(L(f))&=&\int_{Z} (\vartheta-\theta^m)^{-1}(x)\pi^{1_\lambda,{\varphi}}(x) L(f) dx\\ &=&\frac{1}{|\vartheta-\theta^m|^{2l}}\int_{T} (\vartheta-\theta^m)^{-1}(\Delta^lx) \pi^{1_\lambda,{\varphi}}(x) L(f) dx\\ &=&\frac{1}{|\vartheta-\theta^m|^{2l}}\int_{T} (\vartheta-\theta^m)^{-1}(x)\pi^{1_\lambda,{\varphi}}(\Delta^l x) L(f) dx\\ &=&\frac{1}{|\vartheta-\theta^m|^{2l}}\int_{S(T)} \int_A (\vartheta-\theta^m)^{-1}(\xi^a)\pi^{1_\lambda,{\varphi}}(\Delta^l \xi^a) L(f)a^{2\rho_{{U^0/{[U^0,U^0]}}}} da d\xi\ . \end{eqnarray*} By the estimate above the integral converges locally uniformly on $\{{\rm Re }(\lambda)<\rho^{U}+(n+4l)\alpha/2 -l_{\varphi}/2\}$, and for any compact subset of this region, $D\in{\mathcal{U}}({\mathfrak{n}})$, and compact subset $W\subset N$ there is a constant $C_1$ such that $$|\pi^{T,\vartheta-\theta^m}_*(L(f))(yDw)|\le \frac{C_1}{ |\vartheta-\theta^m|^{2l}}$$ for all $y\in W$. \subsubsection{} Choosing $2l>\dim(T)$ we see that the sum $$\sum_{\vartheta\in{\mathcal{X}}(T,{V_1}), \vartheta-\theta^m\not=0} \pi^{T,\vartheta-\theta^m}_*(L(f))$$ converges in $C^\infty(\Omega,V(1_\lambda,{\varphi}))$. \subsubsection{} Refining the estimates above we now show that this sum in fact converges in the space of rapidly decreasing functions. Let $W_1\subset N\setminus N_V$ be a compact subset and $D\in{\mathcal{U}}({\mathfrak{n}})$. It immediately follows from (\ref{zzz1}), that there is a constant $C_2$ such that for all $\xi\in S(T)$, $a,b\in A_+$, with $a \ge b$ and $y\in W_1$ $$|{\varphi}(b)^{-1}(\pi^{1_\lambda,{\varphi}}(\Delta^l\xi^a) L(f))(y^bDw)|\le C_2 a^{2(\lambda-\rho^{[U^0,U^0]})-(n+4l)\alpha}| {\varphi}(ab^{-1})| \ .$$ Let $COP(D)=\sum_{h} D_h^l\otimes D_h^r$ denote the coproduct of $D$, where here $COP:{\mathcal{U}}({\mathfrak{n}})\rightarrow {\mathcal{U}}({\mathfrak{n}})\otimes{\mathcal{U}}({\mathfrak{n}})$. We find a constant $C_3$ such that for all $\xi\in S(T)$, $a,b\in A_+$, with $a \le b$ and $y\in W_1$ \begin{eqnarray*} &&|{\varphi}(b)^{-1}(\pi^{1_\lambda,{\varphi}}(\Delta^l\xi^a) L(f))(y^bDw)|\\ &=& a^{2(\lambda-\rho^{[U^0,U^0]})-(n+4l)\alpha}|\sum_{j,h} {\varphi}(ab^{-1}) {\varphi}(A_{l,j}\xi) \chi(a\xi^{-1}\tilde B_{l,j}a^{-1}y^bD^l_hw) f(\xi^{-1}\tilde C_{l,j}a^{-1}y^bD^r_haw)\\ &=& a^{-4l\alpha} b^{2(\lambda-\rho^{[U^0,U^0]})-n\alpha} |\sum_{j,h} {\varphi}((A_{l,j}\xi)^{ab^{-1}}) \chi(a\xi^{-1}\tilde B_{l,j}a^{-1}y^bD^l_hw) f((\xi^{-1}\tilde C_{l,j})^{b^{-1}a}y(D^r_h)^{b^{-1}}w)\\ &\le& C_3 b^{2(\lambda-\rho^{[U^0,U^0]})-(n+4l)\alpha}\ . \end{eqnarray*} Using this estimate we see that $${\varphi}(b)^{-1}\sum_{\vartheta\in{\mathcal{X}}(T,{V_1}),\vartheta-\theta^m\not=0} \pi^{T,\vartheta-\theta^m}_*(L(f))(y^bDw)$$ can be estimated by $C b^{2(\lambda-\rho^{U})-(n+4l)\alpha}$, where $C$ can be choosen uniformly for $y\in W_1$ and $\lambda$ in compact subsets of $\{{\rm Re }(\lambda)<\rho^{U}+(n+4l)\alpha/2 -l_{\varphi}/2\}$. Since we can choose $l$ arbitrary large we obtain a holomorphic continuation of the sum above to all of ${\aaaa_\C^\ast}$ and that it is rapidly decreasing with respect to $b$. \subsubsection{} Let now $\theta=0$ and consider $\pi^{T,0}_*(L(f))$. We write for ${\rm Re }(\lambda)$ sufficiently small $$\pi^{T,0}_*(L(f))=\int_A \int_{S(T)} \pi^{1_\lambda,{\varphi}}(\xi^a) L(f) d\xi a^{2\rho_{U^0/{[U^0,U^0]}}} da\ .$$ We again employ \begin{eqnarray*} &&\pi^{1_\lambda,{\varphi}}(\xi^a) L(f)(yw)\\ &=& a^{2(\lambda-\rho^{[U^0,U^0]})-n\alpha} {\varphi}(a) {\varphi}(\xi) \chi(a\xi^{-1}y^{a^{-1}}a^{-1}w) f(\xi^{-1}a^{-1}yaw)\ . \end{eqnarray*} Note that $$F(a,y):= \int_{S(T)}{\varphi}(\xi) \chi(a\xi^{-1}ya^{-1}w) f(\xi^{-1}yw) d\xi$$ is a smooth function of $y$ near $y=0$ which is independent of $a$ for large $a$. Let $$F(a,y^{a^{-1}})=\sum_{q=0}^r F_q(y)a^{-q\alpha} + a^{-(r+1)\alpha} R_r(a,y)$$ be the asymptotic expansion for large $a\in A$ obtained from the Taylor series of $F(a,y)$ at $y=0$. The remainder $R_r(a,y)$ remains bounded as $a\to\infty$. We write $$\pi^{T,0}_*(L(f))(yw)=I_+(y)+I_-(y)\ ,$$ where \begin{eqnarray*} I_+(y)&=&\int_{A_+} a^{2(\lambda-\rho^{U})-n\alpha} {\varphi}(a) F(a,y^{a^{-1}}) da\\ I_-(y)&=&\int_{A_-}\int_{S(T)} \pi^{1_\lambda,{\varphi}}(\xi^a) L(f)(yw) d\xi a^{2\rho_{U^0/{[U^0,U^0]}}} da\ . \end{eqnarray*} The integral $I_-$ converges for all $\lambda\in{\aaaa_\C^\ast}$ and defines a holomorphic family of smooth functions. We write $I_+(y):=J_r^1(y)+J_r^2(y)$, where \begin{eqnarray*} J_r^1(y)&:=& \int_{A_+} a^{2(\lambda-\rho^U)-n\alpha} {\varphi}(a) \sum_{q=0}^r F_q(y)a^{-q\alpha} da\\ J_r^2(y)&:=& \int_{A_+} a^{2(\lambda-\rho^U)-n\alpha} {\varphi}(a) a^{-(r+1)\alpha} R_r(a,y) da\ . \end{eqnarray*} The integral $J_r^2$ converges for ${\rm Re }(\lambda)<\rho^U+(n+r+1)\alpha/2-l_{\varphi}/2$ and defines a smooth function in $y$. The integral $J_r^1$ can be evaluated: $$J_r^1(y)=\sum_{q=0}^r (B+2(\lambda-\rho^U)-(n+q)\alpha)^{-1} F_q(y)\ .$$ It obviously defines a meromorphic family of smooth functions. Since we can choose $r$ arbitrary large we obtain a meromorphic continuation of $\pi^{U,1}_*(L(f))$ to all of ${\aaaa_\C^\ast}$. \subsubsection{} Let now $f\in {R}_{\{1\}}(1_\lambda,{\varphi})_0^n$. If $y\in N\setminus T$ and $b\in A$ is sufficiently large, then we have \begin{eqnarray*} \pi^{T,0}_*\circ L\circ [\pi^{U^0/{[U^0,U^0]}}_*](f)(y^b)&=& \pi^{T,0}_*\circ \pi^{Z,0}_*(f)(y^b)\\ &=&\pi^{P_{U^0}}_*(f)(y^b)\\ &=&[\pi^{U^0}_*](f)(y^b)\ . \end{eqnarray*} We now have shown that if $f\in {R}_{[U^0,U^0]}(1_\lambda,{\varphi})^n_\theta$, then $\pi^{U^0/[U^0,U^0]}_*(f)\in {S}_{U^0}(1_\lambda,{\varphi})\oplus L({R}_{U^0}(1_\lambda,{\varphi})^n)$, and that $\pi^{U^0/[U^0,U^0]}_*$ depends meromorphically on $\lambda$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} The finite group $U/U^0$ acts on ${S}_{U^0,k}(1_\lambda,{\varphi})$ and ${R}_{U^0}(1_\lambda,{\varphi})^n$. We can define $\{\pi^{U/U^0}\}:{S}_{U^0,k}(1_\lambda,{\varphi})\rightarrow {S}_{U,k}(1_\lambda,{\varphi})$, $\pi^{U/U^0}:B_{U^0}(1_\lambda,{\varphi})\rightarrow B_U(1_\lambda,{\varphi})$, and $[\pi^{U/U^0}_*]:{R}_{U^0}(1_\lambda,{\varphi})^n\rightarrow {R}_U(1_\lambda,{\varphi})^n$. Using $\pi^{U}_*:=\pi^{U/U^0}_*\circ \pi^{U^0}_*$ and Proposition \ref{mainpure} (and its proof) we obtain \begin{prop}\label{muncor} The composition $$\pi^{U}_*\circ L:{R}_{\{1\}}(1_\lambda,{\varphi})^n_\theta\rightarrow B_{U}(1_\lambda,{\varphi})$$ converges for ${\rm Re }(\lambda)<\rho^U+(n\alpha-l_{\varphi})/2$ and has a meromorphic continuation to all of ${\aaaa_\C^\ast}$ such that $AS\circ \pi^{U}_*\circ L = [\pi^{U}_*]$. If $\theta\not=0$, then it is in fact a holomorphic family of maps $\pi^{U}_*\circ L:{R}_{\{1\}}(1_\lambda,{\varphi})^n_\theta\rightarrow {S}_{U}(1_\lambda,{\varphi})$. If $\theta=0$, then $\pi^{U}_*\circ L$ has at most first order poles in the set $\rho^U+\frac{n\alpha-l_{\varphi}}{2} +\frac{1}{2}{\mathbb {N}}_0$. \end{prop} \subsubsection{} Combining Proposition \ref{muncor} with Lemma \ref{schwpush} we obtain \begin{kor}\label{u76} Assume that the cusp associated to $U\subset P$ does not have full rank. Then for any $k_1< k$ the push-down $\pi_*^U: B_{\{1\},k}(1_\lambda,{\varphi}) \rightarrow B_{U,k_1}(1_\lambda,{\varphi})$ forms a meromorphic family of continuous maps with finite-dimensional singularities defined on $\{{\rm Re }(\lambda)<\rho^U+(k\alpha-l_{\varphi})/2\}$. It fits into the following commutative diagram $$ \begin{array}{ccccccccc} 0&\rightarrow&{S}_{\{1\},k}(1_\lambda,{\varphi})&\rightarrow& B_{\{1\},k}(1_\lambda,{\varphi})&\stackrel{AS}{\rightarrow}& {R}_{\{1\},k}(1_\lambda,{\varphi})&\rightarrow &0\\ & &\downarrow \{\pi^{U}_*\} & & \downarrow \pi^{U}_* & & \downarrow [\pi^{U}_*] &&\\ 0&\rightarrow&{S}_{U,k}(1_\lambda,{\varphi})&\rightarrow& B_{U,k}(1_\lambda,{\varphi})& \stackrel{AS}{\rightarrow}&{R}_{U,k}(1_\lambda,{\varphi})&\rightarrow &0\ .\end{array} $$ Moreover we have a meromorphic family of maps $\pi_*^U:B_{\{1\}}(1_\lambda,{\varphi}) \rightarrow B_{U}(1_\lambda,{\varphi})$ with finite-dimensional singularities and defined on all of ${\aaaa_\C^\ast}$ such that $$ \begin{array}{ccccccccc} 0&\rightarrow&{S}_{\{1\}}(1_\lambda,{\varphi})&\rightarrow& B_{\{1\}}(1_\lambda,{\varphi})&\stackrel{AS}{\rightarrow}& {R}_{\{1\}}(1_\lambda,{\varphi})&\rightarrow &0\\ & &\downarrow \{\pi^{U}_*\} & & \downarrow \pi^{U}_* & & \downarrow [\pi^{U}_*] &&\\ 0&\rightarrow&{S}_{U}(1_\lambda,{\varphi})&\rightarrow& B_{U}(1_\lambda,{\varphi})& \stackrel{AS}{\rightarrow}&{R}_{U}(1_\lambda,{\varphi})&\rightarrow &0\ .\end{array} $$ is commutative. \end{kor} \subsection{Push-down for cusps of full rank and for general $\sigma$} \label{maxrank} \subsubsection{} In this subsection $(\sigma,V_\sigma)$ denotes a Weyl-invariant representation of $M$ as in \ref{weih241}. Let $C^{-\infty}(\infty_P,V(\sigma_\lambda,{\varphi}))\subset C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$ be the space of distributions which are supported on $\infty_P$. This space can be identified $P_U$-equivariantly with the tensor product of a generalized Verma module by $V_{\varphi}$ $$({\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}})\otimes V_{\varphi}$$ such that $(X\otimes s\otimes v)\in ({\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}})\otimes V_{\varphi}$ maps $f\in C^\infty(\partial X,V(\tilde \sigma_{-\lambda},\tilde{\varphi}))$ to $(s\otimes v)(f(Xe))$. We can further identify $C^{-\infty}(\infty_P,V(\sigma_\lambda,{\varphi}))$ with $R_{\{1\}}(\tilde\sigma_{-\lambda},\tilde{\varphi})^*$. \subsubsection{}\label{weih121} We now assume that the cusp associated to $U\subset P$ has full rank. In this case $N$ is the Zariski closure of $U^0$. The space $({\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}})\otimes V_{\varphi}$ carries an algebraic representation of $N$. Therefore \begin{eqnarray*} {}^U[({\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}})\otimes V_{\varphi}]&\subset& {}^{U^0}[({\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}})\otimes V_{\varphi}]\\ &=&{}^N[({\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}})\otimes V_{\varphi}] \end{eqnarray*} is finite-dimensional since the space of highest weight vectors of the ${\mathfrak{g}}$-module ${\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}}$ is finite-dimensional \cite{MR552943}. \subsubsection{}\label{weih300} Let ${\mathcal{E}}_{\infty_P}(\sigma,{\varphi})$ denote the sheaf of holomorphic families $f_\nu\in{}^UC^{-\infty}(\infty_P,V(\sigma_\nu,{\varphi}))$. Since ${\mathcal{E}}_{\infty_P}(\sigma,{\varphi})$ is torsion-free it is the space of sections of a unique holomorphic vector bundle $E_{\infty_P}(\sigma,{\varphi})$ over ${\aaaa_\C^\ast}$. By $E_{\infty_P}(\sigma_\lambda,{\varphi})$ we denote the fibre of $E_{\infty_P}(\sigma,{\varphi})$ at $\lambda\in{\aaaa_\C^\ast}$. We will discuss this bundle in detail in Lemma \ref{vermaolbrich}. \subsubsection{} We now define the function space $B_U(\sigma_\lambda,{\varphi})$ in the case of a cusp of full rank. \begin{ddd} We define $R_U(\sigma_\lambda,{\varphi}):= E_{\infty_P}(\tilde\sigma_{-\lambda},\tilde{\varphi})^*$. Furthermore we set $$B_U(\sigma_\lambda,{\varphi}):={S}_U(\sigma_\lambda,{\varphi})\oplus {R}_U(\sigma_\lambda,{\varphi})\ .$$ Let $AS:B_U(\sigma_\lambda,{\varphi}) \rightarrow {R}_U(\sigma_\lambda,{\varphi})$ be the projection and $L:{R}_U(\sigma_\lambda,{\varphi})\rightarrow B_U(\sigma_\lambda,{\varphi})$ be the inclusion. \end{ddd} These families of spaces form trivial holomorphic bundles of Fr\'echet and Montel spaces over ${\aaaa_\C^\ast}$. \subsubsection{} \begin{ddd}\label{rolf} We define $$[ext^U]:E_{\infty_P}(\tilde\sigma_{-\lambda},\tilde{\varphi})\hookrightarrow {R}_{\{1\}}(\tilde\sigma_{-\lambda},\tilde{\varphi})^*$$ as the natural inclusion. We define the push-down $$[\pi^U_*]:{R}_{\{1\}}(\sigma_{\lambda},{\varphi})\rightarrow R_U(\sigma_\lambda,{\varphi})$$ to be the adjoint of $[ext^U]$. Furthermore we define $$(\pi^U_*)_1:C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))\rightarrow {S}_U(\sigma_\lambda,{\varphi})$$ by $(\pi^U_*)_1(f):=\sum_{u\in U}\pi^{\sigma_\lambda,{\varphi}}(u)(f_{|\Omega})$ (convergence provided). Finally we set $$\pi^U_*:=(\pi^U_*)_1\oplus [\pi^U_*]:C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))\rightarrow B_U(\sigma_\lambda,{\varphi}) \ .$$ \end{ddd} Note that $[\pi^U_*]$ is a holomorphic family of surjective maps. Once and for all we fix a right-inverse $[Q]$ of $[\pi^U_*]$. We leave it to the interested reader to show that the statements of Lemma \ref{compat2} hold true for cusps of full rank as well. \subsubsection{} \begin{lem} The push-down $(\pi^U_*)_1:C^\infty(\partial X,V(1_\lambda,{\varphi}))\rightarrow {S}_U(1_\lambda,{\varphi})$ converges for ${\rm Re }(\lambda)<-l_{\varphi}/2$ and has a meromorphic continuation to all of ${\aaaa_\C^\ast}$ with finite-dimensional singularities. \end{lem} {\it Proof.$\:\:\:\:$} One checks that the corresponding parts of the proofs of Lemma \ref{schwpush} and Proposition \ref{mainpure} apply to the case of cusps of full rank as well. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Note that in Subsection \ref{weih261} we have considered the push-down in the spherical case $\sigma=1$. We now deal with the general case using the concept of twisting. We assume that $U\subset P$ defines a cusp of smaller rank and consider a Weyl invariant $\sigma$ (see \ref{weih241}). We show the existence of a meromorphic family of push-down maps $$\pi^{U}_*:C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))\rightarrow B_U(\sigma_\lambda,{\varphi})$$ using twisting. We employ a finite-dimensional representation $(\pi_{\sigma,\mu},V_{\pi_{\sigma,\mu}})$ of $G$ as in \ref{weih144}. Note that $$\pi^U_*:C^\infty(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})) \rightarrow B_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})$$ is ${\mathcal{Z}}({\mathfrak{g}})$-equivariant. Therefore we can make the following definition: \begin{ddd}\label{cv1} If $\lambda\not\in I_{\mathfrak{a}}$, then using Lemma \ref{compat2}, 4., we define the push-down $$\pi^U_*:C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))\rightarrow B_U(\sigma_\lambda,{\varphi})$$ by the following coummutative diagram: $$\begin{array}{ccccccc} 0&\rightarrow&C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))&\stackrel{i_{\sigma,\mu}}{\rightarrow}&C^\infty(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))&\stackrel{Z(\lambda)}{\rightarrow}& C^\infty(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))\\ &&\downarrow \pi^U_*&&\downarrow \pi^U_*&&\downarrow \pi^U_*\\ 0&\rightarrow&B_U(\sigma_\lambda,{\varphi})&\stackrel{i^U_{\sigma,\mu}}{\rightarrow}& {\mathcal{B}}_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})&\stackrel{Z(\lambda)}{\rightarrow}& {\mathcal{B}}_U(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})\end{array}\ .$$ \end{ddd} \subsubsection{} It is clear that in the domain of convergence ${\rm Re }(\lambda)\ll 0$ this definition coincides with Definition \ref{pushdowndef}. In order to see that $\pi^U_*$ extends to a meromorphic family note that $i^U_{\sigma,\mu}$ admits a meromorphic family of left inverses $j^U_{\sigma,\mu}$ (Lemma \ref{compat2}, 5.), and that we can express the push-down for $\sigma$ through the spherical push-down by $j^U_{\sigma,\mu}\circ \pi^U_* \circ i^U_{\sigma,\mu}$. \subsubsection{} We claim that $\pi^U_*$ has finite-dimensional singularities. Note that given $\nu\in {\mathfrak{a}}^*$ there exists $k\in{\mathbb {N}}_0$ such that the restriction of $\pi^U_*$ to $S_{\{1\},k}(\sigma_\lambda,{\varphi})\cap B_{\{1\}}(\sigma_\lambda,{\varphi})$ converges for all ${\rm Re }(\lambda)<\nu$. The rank of the singularities of $\pi^U_*$ for ${\rm Re }(\lambda)<\nu$ is bounded by the codimension of $S_{\{1\},k}(\sigma_\lambda,{\varphi})\cap B_{\{1\}}(\sigma_\lambda,{\varphi})$ in $B_{\{1\}}(\sigma_\lambda,{\varphi})$ which is finite. This proves the claim. \subsubsection{} The following corollary is a consequence of the discussion above and Corollary \ref{u76}. \begin{kor}\label{neuj106} Let $U\subset P$ define a regular cusp and ${\varphi}$ be an admissible twist. Then the push-down is a meromorphic family of continuous maps $\pi_*^U:B_{\{1\}}(\sigma_\lambda,{\varphi}) \rightarrow B_{U}(\sigma_\lambda,{\varphi})$ with finite-dimensional singularities such that $$ \begin{array}{ccccccccc} 0&\rightarrow&{S}_{\{1\}}(\sigma_\lambda,{\varphi})&\rightarrow& B_{\{1\}}(\sigma_\lambda,{\varphi})&\stackrel{AS}{\rightarrow}& {R}_{\{1\}}(\sigma_\lambda,{\varphi})&\rightarrow &0\\ & &\downarrow \{\pi^{U}_*\} & & \downarrow \pi^{U}_* & & \downarrow [\pi^{U}_*] &&\\ 0&\rightarrow&{S}_{U}(\sigma_\lambda,{\varphi})&\rightarrow& B_{U}(\sigma_\lambda,{\varphi})& \stackrel{AS}{\rightarrow}&{R}_{U}(\sigma_\lambda,{\varphi})&\rightarrow &0\ .\end{array} $$ is commutative. \end{kor} \subsection{Compatibility with embedding}\label{weih116} \subsubsection{} In this subsection we assume that $G^n$ belongs to the list $$\{Spin(1,n), SO(1,n)_0, SU(1,n), Sp(1,n)\}\ .$$ First assume that $U\subset P$ defines a cusp of smaller rank. Then we have a commutative diagram $$\begin{array}{ccc} A_{\{1\}}(1^{n+1}_{\lambda},{\varphi})^m&\stackrel{i^*}{\rightarrow} &A_{\{1\}}(1^{n}_{\lambda-\zeta},{\varphi})^m\\ \downarrow \pi^{P^{n+1}_U}_*&&\downarrow \pi^{P^n_U}_*\\ A_{P_U}(1^{n+1}_{\lambda},{\varphi})^m&\stackrel{i_U^*}{\rightarrow} &A_{P_U}(1^{n}_{\lambda-\zeta},{\varphi})^m \end{array}\ .$$ In fact, commutativity is obvious in the domain of convergence. \subsubsection{} By the definition of the spaces $R_{\{1\}}(1^{n+1}_{\lambda},{\varphi})$, $R_{\{1\}}(1^{n}_{\lambda-\zeta},{\varphi})$ as the spaces of asymptotics of smooth sections and the fact that that $i^*$ maps smooth sections to smooth sections we obtain $U$-equivariant maps $R_{\{1\}}(1^{n+1}_{\lambda},{\varphi})^m\stackrel{i^*}{\rightarrow} R_{\{1\}}(1^{n}_{\lambda-\zeta},{\varphi})^m$, $m\in{\mathbb {N}}_0$. It follows from the naturality of the image bundle Lemma \ref{bunle} that we have a commutative diagram $$\begin{array}{ccc} R_{\{1\}}(1^{n+1}_{\lambda},{\varphi})^m&\stackrel{i^*}{\rightarrow} &R_{\{1\}}(1^{n}_{\lambda-\zeta},{\varphi})^m\\ \downarrow [\pi^{U,n+1}_*]&&\downarrow[\pi^{U,n}_*]\\ R_U(1^{n+1}_{\lambda},{\varphi})^m&\stackrel{i_U^*}{\rightarrow} &R_{U}(1^{n}_{\lambda-\zeta},{\varphi})^m \end{array}\ .$$ \subsubsection{} On the level of Schwartz spaces we have for all $k\in{\mathbb {N}}_0$ $$\begin{array}{ccc} S_{\{1\},k}(1^{n+1}_{\lambda},{\varphi})&\stackrel{i^*}{\rightarrow} &S_{\{1\},k}(1^{n}_{\lambda-\zeta},{\varphi})\\ \downarrow \{\pi^{U,n+1}_*\}&&\downarrow\{\pi^{U,n}_*\}\\ S_{U,k}(1^{n+1}_{\lambda},{\varphi})&\stackrel{i_U^*}{\rightarrow} &S_{U,k}(1^{n}_{\lambda-\zeta},{\varphi}) \end{array}\ .$$ In order to see that $i^*$ and $i^*_U$ (initially defined on spaces of smooth sections) induce maps between Schwartz spaces one checks that these maps are bounded with respect to the norms $\|.\|_{k,d}$ introduced in \ref{weih259}. Commutativity of the diagram is clear. \subsubsection{} We now easily obtain the diagram $$\begin{array}{ccc} B_{\{1\},k}(1^{n+1}_{\lambda},{\varphi})&\stackrel{i^*}{\rightarrow} &B_{\{1\},k}(1^{n}_{\lambda-\zeta},{\varphi})\\ \downarrow \pi^{U,n+1}_*&&\downarrow\pi^{U,n}_* \\ B_{U,k}(1^{n+1}_{\lambda},{\varphi})&\stackrel{i_U^*}{\rightarrow} &B_{U,k}(1^{n}_{\lambda-\zeta},{\varphi}) \end{array}\ .$$ It gives Proposition \ref{rrttee} in the case of a pure cusp of lower rank. \subsubsection{} We now assume that $U\subset P^n$ defines a cusp of full rank. We are going to construct a map $i_U^*:B_U(1^{n+1}_{\lambda},{\varphi})\rightarrow B_U(1^n_{\lambda-\zeta},{\varphi})$ as the direct sum of $$(i_U^*)_1:B_U(1^{n+1}_{\lambda },{\varphi})\rightarrow S_U(1^{n}_{\lambda },{\varphi})$$ and $$(i_U^*)_2:B_U(1^{n+1}_{\lambda },{\varphi})\rightarrow R_U(1^{n}_{\lambda-\zeta},{\varphi})\ .$$ While $(i_U^*)_1$ is just the restriction of sections the definition of $(i_U^*)_2$ is more complicated. In order to verify its properties we need some results of Subsection \ref{samel}. \subsubsection{} In the following we discuss $(i_U^*)_2$. This map comes as a meromorphic family and will essentially be fixed by the condition \begin{equation}\label{tobsucht}\pi^{U,n}_*\circ i^*= i_U^*\circ \pi^{U,n+1}_*\ .\end{equation} The details are as follows. \subsubsection{} Since $E_{\infty_P}(1^{n}_{-\lambda+\zeta},\tilde{\varphi})$ (see \ref{weih300} for notation) is finite-dimensional it consists of distributions of uniformly bounded order. Further, since these distributions are supported in $\infty_P$ we can choose $k\in {\mathbb {N}}$ be such that $E_{\infty_P}(1^{n}_{-\mu+\zeta},\tilde{\varphi})$ pairs trivially with the space ${S}_{\{1\},k-2\rho_U}(1^n_{\mu-\zeta}, {\varphi})$ for all $\mu$ in some compact neighbourhood of $\lambda$. Thus we have an inclusion $E_{\infty_P}(1^{n}_{-\mu+\zeta},\tilde{\varphi})\subset{R}_{\{1\},k}(1^n_{\mu-\zeta}, {\varphi})^*$ (see Definition \ref{weih231} for notation and use the dual of (\ref{weih210})). Note that the target of $(i_U^*)_2$ is the dual of $E_{\infty_P}(1^{n}_{-\lambda+\zeta},\tilde{\varphi})$. We define $$(i^*_U)_2: B_U(1^{n+1}_{\lambda }, {\varphi}) \rightarrow {R}_U(1^n_{ \lambda-\zeta},{\varphi})$$ by the condition that \begin{equation}\label{weih404} \langle\phi,(i^*_U)_2(f)\rangle=\langle i_*(\phi), L\circ [Q]\circ AS(f)\rangle \end{equation} for all $\phi\in E_{\infty_P}(1^{n}_{-\lambda+\zeta},\tilde{\varphi})$, where $i_*:C^{-\infty}(\partial X^n,V(1^n_{-\lambda+\zeta},\tilde{\varphi}))\rightarrow C^{-\infty}(\partial X^{n+1},V(1^{n+1}_{-\lambda },\tilde{\varphi}))$ is the natural inclusion adjoint to $i^*$ (see \ref{weih301}), $L$ is the split of (\ref{weih210}), $[Q]$ is defined in \ref{weih500}, and $AS(f)\in {R}_{\{1\},k}(1^n_{\lambda}, {\varphi})$. This formula defines $(i^*_U)_2(f)$ for generic $\lambda\in {\aaaa_\C^\ast}$ where $[Q]$ is regular. The meromorphic family is given by $$(i^*_U)_2:=(i_*)_{|E_{\infty_P}(1^{n}_{-\lambda+\zeta},\tilde{\varphi})}^*\circ L\circ [Q]\circ AS\ .$$ \subsubsection{} \begin{lem}\label{sucht} The definition of $(i^*_U)_2$ is independent of the choice of $k$, the split $L$ and $[Q]$. \end{lem} {\it Proof.$\:\:\:\:$} Let $L^\prime\circ [Q^\prime]\circ AS^\prime$ be defined with different choices (assume that $k^\prime\ge k$). Let $f_\mu\in B_U(1^{n+1}_\mu,{\varphi})$ be the germ of a holomorphic family near $\lambda$, and consider a family $\phi_\mu\in E_{\infty_P}(1^{n}_{-\mu+\zeta},\tilde{\varphi})$. Then we have \begin{eqnarray*} \lefteqn{AS\circ \pi_*^{U,n+1}(L^\prime\circ [Q^\prime]\circ AS^\prime(f)- L\circ [Q]\circ AS(f))}&&\\&\stackrel{Cor. \ref{u76}}{=}& [\pi_*^{U,n+1}]\circ AS (L^\prime\circ [Q^\prime]\circ AS^\prime(f)-L\circ [Q]\circ AS(f))\\ &=&[\pi_*^{U,n+1}] ([Q^\prime]\circ AS^\prime(f)-[Q]\circ L\circ AS(f))\\ &=&AS^\prime(f)-AS(f)\ . \end{eqnarray*} We define (see Lemma \ref{schwspl} for $\{Q\}$) \begin{eqnarray*} \Delta&:=&L^\prime\circ [Q^\prime]\circ AS^\prime(f)+\{Q\}\circ L(AS^\prime(f)-AS(f))\\&-& L\circ [Q]\circ AS(f)\ . \end{eqnarray*} Then by construction we have \begin{equation}\label{weih402} \pi_*^{U,n+1}(\Delta)=0\ . \end{equation} Using results which we will prove (independently of the present stuff) in Subsection \ref{samel} we show that $\langle i_*(\phi),\Delta\rangle=0$. Proposition \ref{maertins} states an equality of two spaces defined in \ref{weih400} and \ref{weih401}: $${\mathrm{ Ext}}_U(1^{n+1}_{-\mu},\tilde{\varphi})=E_U(1^{n+1}_{-\mu},\tilde{\varphi})\ .$$ On the one hand, $$E_U(1^{n+1}_{-\mu},\tilde{\varphi})\subset {}^UC^{-\infty}(\partial X^{n+1},V(1^{n+1}_{-\mu},\tilde {\varphi}))$$ is the space of evaluations at $\mu$ of holomorphic families of $U$-invariant distributions. In particular we have $i_*(\phi_\mu)\in E_U(1^{n+1}_{-\mu},\tilde{\varphi})$. On the other hand, for generic $\mu$ the space ${\mathrm{ Ext}}_U(1^{n+1}_{-\mu},\tilde{\varphi})$ is contained in the annihilator of the kernel of $\pi^{U,n+1}_*$. Using (\ref{weih402}) we obtain $\langle i_*(\phi_\mu),\Delta_\mu\rangle=0$. By our choice of $k$ we have $\langle i_*(\phi_\mu),g\rangle=0$ for every $g\in {S}_{\{1\},k-2\rho_U}(1^{n+1}_{\mu}, {\varphi})$ and $\mu$ near $\lambda$ since $g$ vanishes at $\infty_P$ with order larger than $k$. Note that $L(AS^\prime(f_\mu)-AS(f_\mu))\in {S}_{U,k}(1^{n+1}_{\mu}, {\varphi})$. Since $\{Q\}\circ L(AS^\prime(f_\mu)-AS(f_\mu))\in {S}_{\{1\},k-2\rho_U}(1^{n+1}_{\mu}, {\varphi})$ it follows that $$\langle i_*(\phi_\mu), \{Q\}\circ L(AS^\prime(f_\mu)-AS(f_\mu))\rangle=0$$ for generic $\mu$ near $\lambda$. Finally we conclude for these $¸\mu$ that \begin{eqnarray*} \lefteqn{ \langle i_*(\phi_\mu), L^\prime\circ [Q]^\prime\circ AS^\prime(f_\mu)\rangle}&&\\&=& \langle i_*(\phi_\mu), L^\prime\circ [Q^\prime]\circ AS^\prime(f_\mu)+\{Q\}\circ L(AS^\prime(f_\mu)-AS(f_\mu))\rangle\\ &=&\langle i_*(\phi_\mu), L\circ [Q]\circ AS(f_\mu)\rangle\ .\end{eqnarray*} \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} \begin{lem} The equality (\ref{tobsucht}) holds true. \end{lem} {\it Proof.$\:\:\:\:$} Let $f\in B_{\{1\}}(1^{n+1}_{\lambda }, {\varphi})$. Then we have \begin{eqnarray}\lefteqn{ \pi_*^{U,n+1}(f-L\circ [Q]\circ AS\circ \pi^{U,n+1}_*(f)}&&\nonumber\\&&-\{Q\}(\pi^{U,n+1}_*(f)- \pi^{U,n+1}_*\circ L\circ [Q]\circ AS\circ \pi^{U,n+1}_*(f))\, =\,0\ .\label{tob} \end{eqnarray} We have seen in the proof of Lemma \ref{sucht} that $i_*(\phi)$ annihilates the kernel of $\pi^{U,n+1}_*$ (for generic $\lambda$, where all the maps are regular). We combine this fact with (\ref{tob}) in order to derive Equality (\ref{entscheident}) below. For all $\phi\in E_{\infty_P}(1^{n}_{-\lambda+\zeta},\tilde{\varphi})$ we have \begin{eqnarray} \langle\phi, AS\circ i^*_U\circ \pi^{U,n+1}_*(f)\rangle&\stackrel{def}{=}& \langle\phi, (i^*_U)_2 \circ \pi_*^{U,n+1}(f)\rangle\nonumber\\ &\stackrel{(\ref{weih404})}{=}&\langle i_*(\phi),L\circ [Q]\circ AS\circ \pi^{U,n+1}_*(f)\rangle \nonumber\\ &=&\langle i_*(\phi),L\circ [Q]\circ AS\circ \pi^{U,n+1}_*(f)\rangle \nonumber\\ &&+\{Q\}(\pi^{U,n+1}_*(f)-\pi^{U,n+1}_*\circ L\circ [Q]\circ AS\circ \pi^{U,n+1}_*(f))\rangle\nonumber\\ &=&\langle i_*(\phi),f \rangle \label{entscheident}\\ &=&\langle \phi,i^*(f) \rangle \nonumber\\ &=&\langle \phi,[\pi^{U,n}_*]\circ AS\circ i^*(f) \rangle\nonumber\ . \end{eqnarray} Since clearly $\{\pi^{U,n}_*\}\circ i^*=(i_U^*)_1\circ \{\pi^{U,n+1}_*\}$ we conclude the required indentity (\ref{tobsucht}). \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Let us combine the results of the present subsection into one statement. Let now $U\subset P^n$ define a cusp of arbitrary rank. Note that $(i_U^*)_2$ and therefore $i^*_U$ may have poles in the case of a cusp of full rank. The following Proposition settles Proposition \ref{rrttee} in the case of pure cusps. \begin{prop}\label{klopp} We have the following commutative diagram (to be understood as an identity of meromorphic families if $i_U^*$ has poles) \begin{equation}\label{weih502} \begin{array}{ccc} C^\infty(\partial X,V(1^{n+1}_{\lambda },{\varphi})) &\stackrel{i^*}{\rightarrow}&C^\infty(\partial X^n, V(1^n_{\lambda-\zeta},{\varphi}))\\ \downarrow \pi^{U,n+1}_*&&\downarrow \pi^{U,n}_*\\ B_U(1^{n+1}_{\lambda },{\varphi})&\stackrel{i_U^*}{\rightarrow }& B_U(1^n_{\lambda-\zeta},{\varphi}) \end{array}\ . \end{equation} \end{prop} \subsection{Extension and restriction}\label{samel} \subsubsection{} In the present subsection we assume for simplicity that the twist ${\varphi}$ is normalized such that all its highest $A$-weights are zero. This differs from the convention adopted in \ref{weih213}. Our present convention has the effect that the dual $\tilde {\varphi}$ has the normalization adopted in \ref{weih213}. Let $l_{\varphi}$ be the highest weight of $\tilde{{\varphi}}$, i.e. $l_{\varphi}:=l_{\tilde{{\varphi}}}$. \subsubsection{} The space $C^{-\infty}(\infty_P,V(\sigma_\lambda,{\varphi}))$ carries an action $\pi^{\sigma_\lambda,{\varphi}}$ of $P_UA$. Using the isomorphism of $P_UA$-modules $$C^{-\infty}(\infty_P,V(\sigma_\lambda,{\varphi}))\cong ({\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}})\otimes V_{\varphi}$$ and of $M_UA$-modules (compare \ref{neuj100}) $$({\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}})\otimes V_{\varphi} \cong {\mathcal{U}}(\bar{\mathfrak{n}})\otimes V_{\sigma_{\lambda+2\rho}}\otimes V_{\varphi}$$ given by the PBW-theorem we see that $A$ acts semisimply. Let $C^{-\infty}(\infty_P,V(\sigma,{\varphi}))^n$ denote the subspace on which $A$ acts with weight $-\lambda-\rho-n\alpha$. Then we have $$C^{-\infty}(\infty_P,V(\sigma_\lambda,{\varphi}))^n= (R_{\{1\}}(\tilde\sigma_{-\lambda},\tilde{\varphi})^n)^*\ .$$ \subsubsection{} Since the action of $P_U$ is algebraic and $U\subset P_U$ is Zariski dense we have $${}^U C^{-\infty}(\infty_P,V(\sigma_\lambda,{\varphi}))= {}^{P_U} C^{-\infty}(\infty_P,V(\sigma_\lambda,{\varphi}))\ .$$ Since $A$ normalizes $P_U$ we conclude that ${}^U C^{-\infty}(\infty_P,V(\sigma_\lambda,{\varphi}))$ is an $A$-invariant subspace. In particular, we obtain a decomposition $${}^U C^{-\infty}(\infty_P,V(\sigma_\lambda,{\varphi}))=\bigoplus_{n=0}^\infty {}^U C^{-\infty}(\infty_P,V(\sigma_\lambda,{\varphi}))^n= \bigoplus_{n=0}^\infty{}^U (R_{\{1\}}(\tilde\sigma_{-\lambda},\tilde{\varphi})^n)^*\ .$$ \subsubsection{}\label{neuj703} Let ${\mathcal{E}}_{\infty_P}(\sigma,{\varphi})^n$ be the sheaf of holomorphic families $f_\nu\in {}^U (R_{\{1\}}(\tilde\sigma_{-\nu},\tilde{\varphi})^n)^*$. This sheaf is torsion-free, and it is therefore the sheaf of holomorphic sections of a holomorphic vector bundle $E_{\infty_P}(\sigma,{\varphi})^n$. By $E_{\infty_P}(\sigma_\lambda,{\varphi})^n$ we denote its fibre at $\lambda$. For each $n\in{\mathbb {N}}_0$ we define the space $\bar Q_(\sigma_\lambda,{\varphi})^n$ by the exact sequence $$0\rightarrow E_{\infty_P}(\sigma_\lambda,{\varphi})^n\rightarrow {}^U(R_{\{1\}}(\tilde\sigma_{-\lambda},\tilde{\varphi})^n)^*\rightarrow \bar Q_{\infty_P}(\sigma_\lambda,{\varphi})^n\rightarrow 0\ .$$ Furthermore let $E_{\infty_P}(\sigma_\lambda,{\varphi}):=\bigoplus_{n=0}^\infty E_{\infty_P}(\sigma_\lambda,{\varphi})^n$ and $\bar Q_{\infty_P}(\sigma_\lambda,{\varphi}):=\bigoplus_{n=0}^\infty \bar Q_{\infty_P}(\sigma_\lambda,{\varphi})^n$. \begin{ddd}\label{neuj107} We call the elements of $E_{\infty_P}(\sigma_\lambda,{\varphi})^n$ deformable. An element of\linebreak[4] ${}^U(R_{\{1\}}(\tilde\sigma_{-\lambda},\tilde{\varphi})^n)^*$ is called undeformable, if it represents a nontrivial class in $\bar Q_{\infty_P}(\sigma_\lambda,{\varphi})^n$. \end{ddd} \subsubsection{} Let ${\mathfrak{t}}\subset {\mathfrak{m}}$ be a Cartan subalgebra of ${\mathfrak{m}}$. Then ${\mathfrak{h}}:={\mathfrak{t}} \oplus {\mathfrak{a}}$ is a Cartan subalgebra of ${\mathfrak{g}}$. By $\Delta^+({\mathfrak{g}},{\mathfrak{h}})$ we denote a positive root system which is compatible with the orientation of ${\mathfrak{a}}$. By $\Delta^+({\mathfrak{m}},{\mathfrak{h}})\subset \Delta^+({\mathfrak{g}},{\mathfrak{h}})$ we denote the subsystem of roots of ${\mathfrak{m}}$. For each $\sigma\in \hat{M}$ we define \begin{equation}\label{neuj103} {\mathfrak{a}}^*\ni d(\sigma):=-\rho+\max\{ \frac{\langle \mu_\sigma,\varepsilon\rangle}{\langle \alpha,\varepsilon\rangle} \:|\:\varepsilon\in \Delta^+({\mathfrak{g}},{\mathfrak{h}})\setminus \Delta^+({\mathfrak{m}},{\mathfrak{h}})\} \alpha \ , \end{equation} where $\mu_\sigma$ is the highest weight of $\sigma$, and $\langle.,.\rangle$ is any Weyl-invariant scalar product on ${\mathfrak{h}}$. There is a natural action of $P_UA$ on the space ${\rm Pol}(N,V_{\sigma^w_{-\lambda}}\otimes V_{\varphi})$ of polynomials on $N$ with values in $V_{\sigma^w_{-\lambda}}\otimes V_{\varphi}$ (compare \ref{neuj101}) given by $$(man.f)(x) = (\sigma^w_{-\lambda}\otimes {\varphi})(man) f((n^{-1}x)^{m^{-1}a^{-1}}),\quad m\in M_U,a\in A, n\in N_V\ .$$ \subsubsection{} \begin{lem}\label{vermaolbrich} \begin{enumerate} \item There is a holomorphic family of $A$-equivariant maps $$j_\lambda : E_{\infty_P}(\sigma_\lambda,{\varphi}){\rightarrow}\ {}^{P_U}{\rm Pol}(N,V_{\sigma^w_{-\lambda}}\otimes V_{\varphi}) \ .$$ \item If ${\rm Re }(\lambda)> d(\sigma)$ or $\lambda\not\in I_{\mathfrak{a}}$, then $j_\lambda$ is an isomorphism and $\bar Q_{\infty_P}(\sigma_\lambda,{\varphi})=0$. \end{enumerate} \end{lem} {\it Proof.$\:\:\:\:$} \subsubsection{}\label{gaa100} Let $$\hat{J}^w_\lambda:C^{-\infty}(\partial X,V(\sigma_\lambda)) \rightarrow C^{-\infty}(\partial X,V(\sigma^w_{-\lambda}))$$ be the unnormalized Knapp-Stein intertwining operator (compare \cite{MR1749869}, Sec. 5, (15)). In order to fix the conventions we recall its definition. The restriction of $\hat{J}^w_\lambda$ to smooth sections is given for ${\rm Re }(\lambda)<0$ by $$(\hat J^w_{\lambda})f(g)=\int_{\bar N} f(gw\bar n) d\bar n\ .$$ For the rest of parameters it is defined by meromorphic continuation, and it extends by continuity to distributions. \subsubsection{} By $$j_\lambda:C^{-\infty}(\infty_P,V(\sigma_\lambda)) \rightarrow {\rm Pol}(N,V_{\sigma^w_{-\lambda}})$$ we denote the off-diagonal part of the Knapp-Stein intertwining operator. Here we identify ${\rm Pol}(N,V_{\sigma^w_{-\lambda}})$ with a subspace of $C^\infty(\Omega_P,V(\sigma^w_{-\lambda}))$ such that $p\in {\rm Pol}(N,V_{\sigma^w_{-\lambda}})$ corresponds to $f_p\in C^\infty(\Omega_P,V(\sigma^w_{-\lambda}))$ with $f_p(xw)=p(x)$. The off-diagonal part of $\hat{J}^w_\lambda$ maps to polynomials since it is $P$-equivariant and the elements of $C^{-\infty}(\infty_P,V(\sigma_{\lambda}))$ are $P$-finite. Alternatively, using the identification $$C^{-\infty}(\infty_P,V(\sigma_{\lambda}))\cong\linebreak[4] {\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}}\ ,$$ we can write $$j_\lambda(X\otimes v)(n)=\pi^{\sigma^w_{-\lambda}}(X) f_{1_v}(nw)\ ,$$ where $1_v\in {\rm Pol}(N,V_{\sigma^w_{-\lambda}})$ is the constant polynomial with value $v\in V_{\sigma^w_{-\lambda}}$. \subsubsection{} The map $j_\lambda$ is in fact ${\mathfrak{g}}$-equivariant, where the action $\pi^{\sigma^w_{-\lambda}}$ of ${\mathfrak{g}}$ on ${\rm Pol}(N,V_{\sigma^w_{-\lambda}})$ is induced by the embedding ${\rm Pol}(N,V_{\sigma^w_{-\lambda}})\subset C^\infty(\Omega_P,V(\sigma^w_{-\lambda}))$. Therefore $\ker(j_\lambda)$ is a ${\mathfrak{g}}$-submodule of the Verma module ${\mathcal{U}}({\mathfrak{g}})\otimes_{{\mathcal{U}}({\mathfrak{p}})} V_{\sigma_{\lambda+2\rho}}$. By \cite{MR552943}, Satz 1.17, this Verma module is irreducible for ${\rm Re }(\lambda)> d(\sigma)$. If $\lambda\not\in I_{\mathfrak{a}}$, then $j_\lambda$ is injective by \cite{MR1749869}, Lemma 6.7. \subsubsection{} Since $$\dim {\rm Pol}(N,V_{\sigma^w_{-\lambda}})^n= \dim C^{-\infty}(\infty_P,V(\sigma_\lambda))^n = \dim ({\mathcal{U}}(\bar{\mathfrak{n}})\otimes V_{\sigma_\lambda})^n\ ,$$ we conclude that $j_\lambda$ is in fact an isomorphism. After tensoring with $V_{\varphi}$ and taking $U$-invariants we obtain an isomorphism $$j_\lambda:{}^U(R_{\{1\}}(\tilde\sigma_{-\lambda},\tilde{\varphi})^n)^* \cong {}^{P_U}{\rm Pol}(N,V_{\sigma^w_{-\lambda}}\otimes V_{\varphi})^n\ .$$ Since $j_\lambda$ and its inverse depend holomorphically on $\lambda$ we have $${}^{P_U}{\rm Pol}(N,V_{\sigma^w_{-\lambda}}\otimes V_{\varphi})^n\cong E_{\infty_P}(\sigma_\lambda,{\varphi})^n$$ for $\lambda\in{\aaaa_\C^\ast}$ with ${\rm Re }(\lambda) > d(\sigma)$ or $\lambda\not\in I_{\mathfrak{a}}$. This proves the lemma. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Recall the definition of the function spaces $B_{U,k}(\tilde\sigma_{-\lambda},\tilde{\varphi})$ given in \ref{neuj104}. \begin{ddd}\label{neuj1002} We define $D_{U,k}(\sigma_\lambda,{\varphi})$ to be the dual space to $B_{U,k}(\tilde\sigma_{-\lambda},\tilde{\varphi})$. Furthermore let $D_{U}(\sigma_\lambda,{\varphi}):=\bigcup_{k\in{\mathbb {N}}_0} D_{U,k}(\sigma_\lambda,{\varphi})$. \end{ddd} As a consequence of the corresponding properties of the family of function spaces the family of spaces $D_{U,k}(\sigma_\lambda,{\varphi})$, $\lambda\in{\aaaa_\C^\ast}$, forms local trivial holomorphic bundles of dual Fr\'echet spaces. Furthermore, the spaces $D_{U}(\sigma_\lambda,{\varphi})$ are Montel and form a direct limit of locally trivial holomorphic bundles. \subsubsection{} Recall the definition of the push-down Definition \ref{pushdowndef}. Its meromorphic continuation was finally established in Corollary \ref{neuj106}. \begin{ddd} We define the extension map $$ext^U:D_{U}(\sigma_\lambda,{\varphi})\rightarrow D_{\{1\}}(\sigma_\lambda,{\varphi})$$ as the adjoint of the push-down $$\pi^U_*:B_{\{1\}}(\tilde\sigma_{-\lambda},\tilde {\varphi}) \rightarrow B_{U}(\tilde\sigma_{-\lambda},\tilde {\varphi})\ .$$ \end{ddd} It follows from the corresponding properties of the push-down that the extension maps form a meromorphic family of continuous maps with finite-dimensional singularities. \subsubsection{} In the remainder of the present subsection we discuss the case $\sigma=1$. Assume that the cusp associated to $U\subset P$ has lower rank. For $k\in {\mathbb {N}}_0$ and ${\rm Re }(\lambda)>-\rho^U+(l_{\varphi}-k\alpha)/2$ such that $$\pi^U_*:B_{\{1\},k}(1_{-\lambda},\tilde {\varphi}) \rightarrow B_{U,k}(1_{-\lambda},\tilde {\varphi})$$ is regular (compare Corollary \ref{u76}) we also have a map $$ext^U:D_{U,k}(1_\lambda,{\varphi})\rightarrow D_{\{1\},k}(1_\lambda,{\varphi})$$ defined as the adjoint $\pi^U_*$. If $k_1 > k$, then $ext^U:D_{U,k}(1_\lambda,{\varphi})\rightarrow D_{\{1\},k_1}(1_\lambda,{\varphi})$ is a meromorphic family of continuous maps with finite-dimensional singularities defined for ${\rm Re }(\lambda)>-\rho^U+(l_{\varphi}-k\alpha)/2$. \subsubsection{}\label{weih400} \begin{ddd}\label{neuj108} We define $$Ext_{U}(1_\lambda,{\varphi})\subset D_{\{1\}}(1_\lambda,{\varphi})$$ to be the subspace of all $f\in D_{\{1\}}(1_\lambda,{\varphi})$ of the form $ext^U(h)_\lambda$, where $h_\mu\in D_U(1_\mu,{\varphi})$ is a meromorphic family defined near $\lambda$ such that $\mu\mapsto ext^U(h)_\mu$ is regular at $\mu=\lambda$. In a similar manner we define $Ext_{U,k}(1_\lambda,{\varphi})$ to by requiring in addition that $h_\mu\in D_{U,k}(1_\mu,{\varphi})$. \end{ddd} The subspaces $Ext_{U,k}(1_\lambda,{\varphi})\subset Ext_{U}(1_\lambda,{\varphi})$ are defined for ${\rm Re }(\lambda)>-\rho^U+(l_{\varphi}-k\alpha)/2$. The space $Ext_{U}(1_\lambda,{\varphi})$ plays the role of the range of the extension. \subsubsection{}\label{neuj471} It is clear that $$Ext_{U}(1_\lambda,{\varphi})\subset {}^U D_{\{1\}}(1_\lambda,{\varphi})\ .$$ In order to describe to which extent the space ${}^U D_{\{1\}}(1_\lambda,{\varphi})$ is exhausted by $Ext_{U}(1_\lambda,{\varphi})$ we define $Q_{U}(1_\lambda,{\varphi})$ to be the following quotient: $$0\rightarrow Ext_{U}(1_\lambda,{\varphi}) \rightarrow {}^U D_{\{1\}}(1_\lambda,{\varphi}) \rightarrow Q_{U}(1_\lambda,{\varphi})\rightarrow 0\ .$$ The elements in the space $Q_{U}(1_\lambda,{\varphi})$ turn out to be somewhat uncontrollable. We therefore take much effort to show that this space is trivial under certain conditions. \subsubsection{}\label{weih401} We now define the space of deformable $U$-invariant distributions (compare Def.~\ref{neuj107} for a similar definition with an additional support condition). \begin{ddd}\label{neuj1088} We define the subspace $$E_U(1_\lambda,{\varphi})\subset {}^UD_{\{1\}}(1_\lambda,{\varphi})$$ as the space of evaluations of germs at $\lambda$ of holomorphic families $f_\nu\in {}^UD_{\{1\}}(1_\nu,{\varphi})$. In a similar manner we define $E_{U,k}(1_\lambda,{\varphi})\subset {}^UD_{\{1\},k}(1_\lambda,{\varphi})$ as the subspace of evaluations of families with the additional property that $f_\nu\in{}^UD_{\{1\},k}(1_\nu,{\varphi})$. \end{ddd} \subsubsection{} We define the space $\bar Q_{U}(1_\lambda,{\varphi})$ as the quotient $$0\rightarrow E_{U}(1_\lambda,{\varphi}) \rightarrow {}^U D_{\{1\}}(1_\lambda,{\varphi}) \rightarrow \bar Q_{U}(1_\lambda,{\varphi})\rightarrow 0\ .$$ \begin{ddd} An element of ${}^U D_{\{1\}}(1_\lambda,{\varphi})$ which represents a non-trivial class in $\bar Q_{U}(1_\lambda,{\varphi})$ is called undeformable. \end{ddd} (compare Definition \ref{neuj107}). \subsubsection{} It follows immediately from the definitions that $$Ext_{U}(1_\lambda,{\varphi}) \subset E_U(1_\lambda,{\varphi})\ .$$ Hence we have a surjection $$Q_U(1_\lambda,{\varphi})\to \bar Q_U(1_\lambda,{\varphi})\ .$$ One of the goals of the present subsection is to show that $$Ext_U(1_\lambda,{\varphi})=E_U(1_\lambda,{\varphi})\ .$$ Furthermore, we want to show that for many (generic) $\lambda\in {\aaaa_\C^\ast}$ every element of ${}^U D_{\{1\}}(1_\lambda,{\varphi})$ is deformable, i.e. $Q_{U}(1_\lambda,{\varphi})\cong 0$. The results are stated in Proposition \ref{maertins}. \subsubsection{} Now we come to the definition of a left-inverse of $ext^U$: the restriction map $res^U$. We fix $k\in {\mathbb {N}}_0$. Recall the construction of the meromorphic family of right-inverses of $[\pi_*^U]$ $$[Q]:R_{U,k}(1_\lambda,\tilde{{\varphi}})\to R_{\{1\},k}(1_\lambda,\tilde {\varphi})$$ from \ref{weih500}. Let $k_1\in{\mathbb {N}}_0$ be such that $k_1 \alpha<k\alpha-2\rho_U$, and let $$\{Q\}:{S}_{U,k}(1_\lambda,{\varphi})\rightarrow {S}_{\{1\},k_1}(1_\lambda,{\varphi})$$ be as in Lemma \ref{schwspl}. We define a meromorphic family of maps $$Q:B_{U,k}(1_\lambda,\tilde{{\varphi}})\rightarrow B_{\{1\},k_1}(1_\lambda,\tilde{{\varphi}})$$ by $$Q(h):= \{Q\}\left(h - \pi^U_*\circ L\circ [Q]\circ AS(h)\right) + L\circ [Q]\circ AS(h)$$ Then one easily checks that $\pi^U_*\circ Q$ is just the inclusion $B_{U,k}(1_\lambda,\tilde{{\varphi}})\rightarrow B_{U,k_1}(1_\lambda,\tilde{{\varphi}})$. \begin{ddd}\label{neuj707} We define the meromorphic family of restriction maps $$res^U: D_{\{1\},k_1}(1_\lambda,{\varphi})\rightarrow D_{U,k}(1_\lambda,{\varphi})$$ as the adjoint of $Q$. \end{ddd} Note that $res^U$ depends on choices (we do not indicate these choices in the notation for the restriction map). In particular, these maps are not compatible if we change $k$ and $k_1$. \subsubsection{}\label{neuj2001} Note that the composition $res^U\circ ext^U$ coincides with the inclusion $$D_{U,k_1}(1_\lambda,{\varphi}) \hookrightarrow D_{U,k}(1_\lambda,{\varphi})\ .$$ Recall the Definition \ref{neuj108} of $Ext_{U,k_1}(1_\lambda,{\varphi})$. \begin{lem}\label{weedef} Let $k,k_1\in{\mathbb {N}}_0$ and $\lambda\in{\aaaa_\C^\ast}$ be such that $res^U:D_{\{1\},k_1}(1_\lambda,{\varphi})\rightarrow D_{U,k}(1_\lambda,{\varphi})$ is defined. Furthermore we assume that $\pi^U_*:B_{\{1\}}(1_{-\lambda},\tilde{\varphi})\rightarrow B_U(1_{-\lambda},\tilde{\varphi})$ is regular. Then the restriction of $res^U$ to $Ext_{U,k_1}(1_\lambda,{\varphi})$ is independent of choices. \end{lem} {\it Proof.$\:\:\:\:$} This follows from $res^U\circ ext^U(f)=f$ and the fact that $ext^U$ is regular at $\lambda$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{}\label{neuj300} Assume that $\lambda\in{\aaaa_\C^\ast}$ is such that $$\pi^U_*:B(1_{-\lambda},\tilde{\varphi})\rightarrow B_U(1_{-\lambda},\tilde{\varphi})$$ and $$[Q]:R_{U}(1_{-\lambda},\tilde{{\varphi}})^n\to R_{\{1\}}(1_{-\lambda},\tilde {\varphi})^n$$ are regular for all $n\in {\mathbb {N}}_0$. We call $\lambda\in{\aaaa_\C^\ast}$ satisfying these conditions admissible. \subsubsection{}\label{neuj704} Note that ${R}_{\{1\}}(1_{-\lambda},\tilde{\varphi})^*$ is the space of those distribution sections of $V(1_\lambda,{\varphi})$ that are supported at the point $\infty_P$. We define $$Ext_{\infty_P}(1_\lambda,{\varphi}):={R}_{\{1\}}(1_{-\lambda},\tilde{\varphi})^*\cap Ext_U(1_\lambda,{\varphi})\ .$$ Recall the definition \ref{neuj471} of the space $Q_{U}(1_\lambda,{\varphi})$. The point of the following lemma is that every element of $Q_{U}(1_\lambda,{\varphi})$ can be represented by an invariant distribution supported in $\infty_P$. Note that $$[ext^{U}]:{R}_{U}(1_{-\lambda},\tilde{\varphi})^*\rightarrow {}^U {R}_{\{1\}}(1_{-\lambda},\tilde{\varphi})^*$$ generates a subspace of $Ext_{\infty_P}(1_\lambda,{\varphi})$, where $[ext^{U}]$ is the restriction of $ext^{U}$ to ${R}_{U}(1_{-\lambda},\tilde{\varphi})^*$, or equivalently, the adjoint of $[\pi^{U}_*]$ (see Def.~\ref{weih254}). \begin{lem}\label{homjk} \begin{enumerate} \item The family of maps $[ext^{U}]$ generates all of $Ext_{\infty_P}(1_\lambda,{\varphi})$. \item If $\lambda\in{\aaaa_\C^\ast}$ is admissible in the sense of \ref{neuj300}, then there is an exact sequence $$0\rightarrow Ext_{\infty_P}(1_\lambda,{\varphi})\rightarrow {}^{U}{R}_{\{1\}}(1_{-\lambda},\tilde{\varphi})^* \rightarrow Q_{U}(1_\lambda,{\varphi})\rightarrow 0$$ of semisimple $A$-modules.\end{enumerate} \end{lem} {\it Proof.$\:\:\:\:$} Recall that $${}^{U}{R}_{\{1\}}(1_{-\lambda},\tilde{\varphi})^*= \bigoplus_{n\in{\mathbb {N}}_0} {}^U({R}_{\{1\}}(1_{-\lambda},\tilde{\varphi})^n)^*$$ is a weight-decomposition of the $A$-module. Moreover, we have $$[ext^{U}]:({R}_{U}(1_{-\lambda},\tilde{\varphi})^n)^*\rightarrow ({R}_{\{1\}}(1_{-\lambda},\tilde{\varphi})^n)^*\ .$$ If $f\in Ext_{\infty_P}(1_\lambda,{\varphi})$ is represented by $ext^{U}(h)$ for some meromorphic family $h_\mu\in D_{U}(1_\mu,{\varphi})$, then $\{res^{U}\} \{ext^{U}(h)\} =\{h\}$ vanishes at $\lambda$, where $\{h\}$ denotes the restriction of $h$ to ${S}_{\{1\}}(1_{-\lambda},\tilde{\varphi})$, and $\{res^U\}:=\{Q\}^*$. Therefore all non-positive Laurent-coefficients of the expansion of $h$ at $\lambda$ belong to ${R}_{U}(1_{-\lambda},\tilde{\varphi})^*$. Thus we can choose the family $h$ such that $h_\mu\in{R}_{U}(1_{-\mu},\tilde{\varphi})^*$, and such that $[ext^{U}](h)_\lambda=f$. This shows that $[ext^{U}]$ generates $Ext_{\infty_P}(1_\lambda,{\varphi})$, and that $A$ acts semisimply on $Ext_{\infty_P}(1_\lambda,{\varphi})$. \subsubsection{} It remains to show that any element of $Q_U(1_\lambda,{\varphi})$ can be represented by some element of ${}^{U}{R}_{\{1\}}(1_{-\lambda},\tilde{\varphi})^*$. Let $f\in {}^{U}D_{\{1\}}(1_\lambda,{\varphi})$. Then there is $k\in{\mathbb {N}}_0$ such that $f\in {}^{U}D_{\{1\},k}(1_\lambda,{\varphi})$. We choose $k_1$ such that $$\{res^{U}\}:{S}_{\{1\},k}(1_{-\lambda},\tilde{\varphi})^* \rightarrow {S}_{U,k_1}(1_{-\lambda},\tilde{\varphi})^*$$ is defined. We then put $h:=\{res^{U}\}\{f\}$. Let $$T:{S}_{U,k_1}(1_{-\lambda},\tilde{\varphi})^* \rightarrow D_{U}(1_\lambda,{\varphi})$$ be the split induced by the dual split $L$. We form $f-ext^{U}\circ T(h)$. This difference represents the same element in $Q_{U}(1_\lambda,{\varphi})$ as $f$, but its restriction to ${S}_{\{1\}}(1_{-\lambda},\tilde{\varphi})$ vanishes. Indeed, for $g$ in ${S}_{\{1\}}(1_{-\lambda},\tilde{\varphi})$ we have $$ \langle ext^{U}\circ T(h),g\rangle= \langle f, \{Q\}\circ\{\pi_*^U\}(g)\rangle=\langle f, g\rangle \ .$$ \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \newcommand{{\mathrm{pr}}}{{\mathrm{pr}}} \newcommand{{\mathbb{F}}}{{\mathbb{F}}} \subsubsection{} \begin{prop}\label{argumentprop} Let $n\ge 0$. If $\lambda\in{\mathfrak{a}}^*$ is sufficiently large, then the inclusion $$Ext_{\infty_P}(1_\lambda,{\varphi})^n\hookrightarrow {}^U(R_{\{1\}}(1_{-\lambda},\tilde{\varphi})^n)^*$$ is an isomorphism. \end{prop} {\it Proof.$\:\:\:\:$} If $U$ defines a cusp of full rank, then the proposition is a direct consequence of Lemma \ref{vermaolbrich} and the definition of $[ext^U]$ (Definition \ref{rolf}). Thus we may and will assume in the following that $U$ defines a cusp of smaller rank. Note that both spaces appearing in the proposition are finite-dimensional. The proposition is an immediate consequence of the following lemma. \begin{lem}\label{argumentlem1} If $\lambda\in{\mathfrak{a}}^*$ is sufficiently large, then we have the inequality $$\dim Ext_{\infty_P}(1_\lambda,{\varphi})^n\ge \dim {}^U(R_{\{1\}}(1_{-\lambda},\tilde{\varphi})^n)^*\ .$$ \end{lem} {\it Proof.$\:\:\:\:$} By Lemma \ref{vermaolbrich},2, if $\lambda\in{\mathfrak{a}}^*$ is sufficiently large we have $\dim {}^U(R_{\{1\}}(1_{-\lambda},\tilde{\varphi})^n)^*=\dim {}^{P_U} {\rm Pol}(N,V_{1_{-\lambda}}\otimes V_{\varphi})^n$, where the action of $AP_U$ on a polynomial is given by $$(p \: f)(n):={\varphi}(p)f(n^{p^{-1}})\ .$$ The degree-$n$ subspace of the polynomial maps is characterized by \begin{equation}\label{argumentdegree1}a\:f=a^{-n\alpha} f\ ,\quad a\in A\ .\end{equation} We now consider $$[\pi^U_*]:R_1(1_{-\lambda},\tilde {\varphi})^n\to R_U(1_{-\lambda},\tilde {\varphi})^n$$ which is regular for ${\rm Re }(\lambda)$ sufficiently large. Its adjoint is $$[ext^U]:(R_U(1_{-\lambda},\tilde {\varphi})^*)^n\to Ext_{\infty_P}(1_\lambda,{\varphi})^n\ ,$$ and we have $\dim {\mbox{\rm im}} [\pi^U_*] =\dim {\mbox{\rm im}} [ext^U]$. \subsubsection{} Therefore, Lemma \ref{argumentlem1} follows directly from the following lemma. \begin{lem}\label{argumentlem2} If $\lambda\in{\mathfrak{a}}^*$ is sufficiently large, then we have $$\dim {\mbox{\rm im}} [\pi^U_*] \ge \dim {}^{P_U} {\rm Pol}(N,V_{1_{-\lambda}}\otimes V_{\varphi})^n\ .$$ \end{lem} {\it Proof.$\:\:\:\:$} Let $\langle.,.\rangle$ be some non-degenerate invariant bilinear form on ${\mathfrak{g}}$. We define the linear subspace \begin{equation}\label{argumenteq33}\bar {\mathfrak{n}}^U:=\{Y\in \bar {\mathfrak{n}}\:|\: \langle [Y,H],X\rangle=0\:\:\forall X\in {\mathfrak{n}}_V\}\end{equation} and the submanifold $\bar N^U:=\exp(\bar {\mathfrak{n}}^U)\subset \bar N$. We furthermore choose a Cartan involution $\theta$ of ${\mathfrak{g}}$ compatible with ${\mathfrak{a}}$. \begin{lem}\label{argumentlem6} \begin{enumerate} \item The submanifold $\bar N^U$ is $AM_U$-invariant. \item The multiplication map $\bar N_V\times \bar N^U\to \bar N$ is a diffeomorphism, where $\bar N_V:=N_V^\theta$. \item The composition $$\bar N\stackrel{\sim}{\to}\bar N_V\times \bar N^U\stackrel{{\mathrm{pr}}}{\to} \bar N^U$$ is an $AM_U$-equivariant polynomial map. \end{enumerate} \end{lem} {\it Proof.$\:\:\:\:$} The subspace $\bar {\mathfrak{n}}^U$ is $AM_U$-invariant since ${\mathfrak{n}}_V$ is $AM_U$-invariant. The first assertion now follows from the $AM_U$-equivariance of the exponential map By the $A$-invariance of $\bar N^U$, $\bar N_V$ and the equivariance of the multiplication, it suffices to show that the multiplication map is a diffeomorphism near $(1,1)$. Infinitesimally it is given by the map $\bar{\mathfrak{n}}_V\times \bar {\mathfrak{n}}^U\to \bar {\mathfrak{n}}$, $(Y_V,Y^U)\mapsto \bar Y_V+Y^U$. This map is an isomorphism. In fact, the dimensions of the domain and the target coincide, and $\bar {\mathfrak{n}}_V\cap \bar {\mathfrak{n}}^U=\{0\}$. The last assertion follows from the diagram $$\xymatrix{\bar {\mathfrak{n}}\ar[d]^{\exp}&\bar {\mathfrak{n}}_V\times\bar {\mathfrak{n}}^U\ar[l]_{q\quad}\ar[d]^{(\exp,\exp)}\ar[r]^{\quad pr}&\bar {\mathfrak{n}}^U\ar[d]^{\exp}\\\bar N&\bar N_V\times \bar N^U\ar[l]_{mult\quad}\ar[r]^{\quad{\mathrm{pr}}}&\bar N^U}$$ and the fact that $q$ is an $AM_U$-equivariant polynomial map with a polynomial inverse. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Note that $\bar N_VAM_U$ acts on ${\rm Pol}(\bar N,V_{1_{-\lambda}}\otimes V_{\tilde{\varphi}})$ by $$\bar n_Vam\: f(\bar n)=\tilde{\varphi}(am) f((\bar n_V^{-1}\bar n)^{(am)^{-1}})\ .$$ We now consider the space $$I(\lambda)^n:={}^{\bar N_VM_U}{\rm Pol}(\bar N,V_{1_{-\lambda}}\otimes V_{ \tilde {\varphi}})^n\subset{\rm Pol}(\bar N,V_{1_{-\lambda}}\otimes V_{\tilde {\varphi}})^n\ ,$$ where the degree $n$-subspace is distinguished by the condition \begin{equation}\label{argumentdegree2}a\: f=a^{n\alpha} f\ .\end{equation} We have a degree-preserving inclusion $$I(\lambda)^n\hookrightarrow R_1(1_{-\lambda},\tilde {\varphi})^n\ .$$ Furthermore, it follows from Lemma \ref{argumentlem6} that the restriction to $\bar N^U$ induces an isomorphism $$I(\lambda)^n\stackrel{\sim}{\to} {}^{M_U}{\rm Pol}(\bar N^U,1_{-\lambda}\otimes V_{\tilde {\varphi}})^n\ .$$ Lemma \ref{argumentlem2} now follows from the following two assertions. \begin{lem}\label{argumentlem3} $\dim I(\lambda)^n= \dim {}^{P_U} {\rm Pol}(N,V_{1_{-\lambda}}\otimes V_{\varphi})^n$ \end{lem} \begin{lem}\label{argumentlem4} If $\lambda\in{\mathfrak{a}}^*$ is sufficiently large, then the restriction of $[\pi^U_*]$ to $I(\lambda)^n$ is injective. \end{lem} \subsubsection{} We first show Lemma \ref{argumentlem3}. We define $N^U:=(\bar N^U)^\theta\subset N$. Then we have an $A$-equivariant diffeomorphism $N_V\times N^U\stackrel{mult}{\to} N$, and the projection $N\stackrel{\sim}{\to} N_V\times N^U\stackrel{{\mathrm{pr}}}{\to} N^U$ is an $AM_U$-equivariant polynomial map. These facts follow from Lemma \ref{argumentlem6} by an application of the Cartan involution $\theta$. We conclude that the restriction to $N^U$ induces an isomorphism $${}^{P_U} {\rm Pol}(N,V_{1_{-\lambda}}\otimes V_{\varphi})^n\stackrel{\sim}{\to} {}^{M_U}{\rm Pol}(N^U, V_{1_{-\lambda}}\otimes V_{\varphi})^n\ .$$ Note that there is a canonical $AM_U$-equivariant isomorphism between $S(({\mathfrak{n}}^U)^*)\otimes V_{\varphi}$ and \linebreak[4] ${\rm Pol}(N^U, V_{1_{-\lambda}}\otimes V_{\varphi})$. Here $S(.)$ stands for the symmetric algebra. Similarly, we have $$S((\bar{\mathfrak{n}}^U)^*)\otimes V_{\tilde{\varphi}}\cong {\rm Pol}(\bar N^U,V_{1_{-\lambda}}\otimes V_{\tilde {\varphi}})\ .$$ The natural pairing between $\bar{\mathfrak{n}}^U$ and ${\mathfrak{n}}^U:= (\bar{\mathfrak{n}}^U)^\theta$ via the $G$-invariant form $\langle.,.\rangle$ induces a nondegenerate pairing between the two symmetric algebras above. We conclude that the spaces ${}^{M_U}{\rm Pol}(N^U, V_{1_{-\lambda}}\otimes V_{\varphi})^n$ and $ {}^{M_U}{\rm Pol}(\bar N^U,V_{1_{-\lambda}}\otimes V_{\tilde {\varphi}})^n\cong I(\lambda)^n$ are each others duals (the pairing is degree-preserving in view of the characterizations (\ref{argumentdegree1}) and (\ref{argumentdegree2})). Lemma \ref{argumentlem3} now follows. \subsubsection{} We now start with the proof of Lemma \ref{argumentlem4}. First of all note that the push-down (see Def.~\ref{weih254}) $$[\pi^U_*]:I(\lambda)^n\to R_U(1_{-\lambda},\tilde {\varphi})^n\subset A_U(1_{-\lambda},\tilde {\varphi})^n $$ is given by a convergent integral $$[\pi^U_*](f)(\bar n)=\int_{N_V}{\varphi}(n_V)^{-1} f(\bar n(n_V\bar n))a(n_V\bar n)^{-\lambda-\rho} dn_V\ ,\bar n\in \bar N\setminus\{1\}\ .$$ Here we have employed again the Bruhat decomposition $g=\bar n(g) m(g) a(g) n(g)$. The vector space $V_{\tilde {\varphi}}$ has a filtration induced by the action of $A$ that is preserved by $M_UN_V$. The induced action of $N_V$ on the associated graded vector space ${\mathrm{Gr}}(V_{\tilde {\varphi}})$ is trivial. The filtration of $V_{\tilde {\varphi}}$ induces filtrations on $I(\lambda)^n$ and $A_U(1_{-\lambda},\tilde {\varphi})^n$. From the integral representation of $[\pi^U_*]$ we see that that this map preserves the filtrations and induces a map ${\mathrm{Gr}}[\pi_U^*]:{\mathrm{Gr}} I(\lambda)^n\to {\mathrm{Gr}} A_U(1_{-\lambda},\tilde {\varphi})^n$. Since injectivity of the associated graded map implies injectivity of a filtration preserving map, Lemma \ref{argumentlem4} is a consequence of \begin{lem}\label{argumentlem44} If $\lambda\in{\mathfrak{a}}^*$ is sufficiently large, then $${\mathrm{Gr}} [\pi^U_*]:{\mathrm{Gr}} I(\lambda)^n\to {\mathrm{Gr}} A_U(1_{-\lambda},\tilde {\varphi})^n$$ is injective. \end{lem} \subsubsection{} Note that $${\mathrm{Gr}} [\pi^U_*](f)(\bar n)=\int_{N_V} f(\bar n(n_V\bar n))a(n_V\bar n)^{-\lambda-\rho}dn_V\ ,\bar n\in \bar N\setminus\{1\}\ .$$ We define $$c_\lambda(\bar n):={\mathrm{Gr}} [\pi^U_*](1)(\bar n)= \int_{N_V} a(n_V\bar n)^{-\lambda-\rho}dn_V\ ,\bar n\in \bar N\setminus\{1\}\ .$$ We will show the following lemma. \begin{lem}\label{argumentlem8} For $\bar n^U\in \bar N^U\setminus\exp(\bar{\mathfrak{n}}_{-2\alpha})$ we have $$\frac{1}{c_\lambda(\bar n^U)} {\mathrm{Gr}} [\pi^U_*](f)(\bar n^U)= f(\bar n^U)+O(|\lambda|^{-1})\ .$$ \end{lem} \subsubsection{} Let us first show that Lemma \ref{argumentlem8} implies \ref{argumentlem44}. The natural identification $V_{1_{-\lambda}}\cong {\mathbb {C}}$ induces identifications ${\rm Pol}(\bar N^U,1_{-\lambda}\otimes V_{\tilde {\varphi}})^n\cong {\rm Pol}(\bar N^U,V_{\tilde {\varphi}})^n$ for all $\lambda$. Note that $\bar {\mathfrak{n}}^U\cap\bar{\mathfrak{n}}_{-\alpha}$ is non-trivial. Otherwise we would have ${\mathfrak{n}}_V\cap{\mathfrak{n}}_{\alpha}={\mathfrak{n}}_{\alpha}$ and therefore ${\mathfrak{n}}_V={\mathfrak{n}}$, i.e., $U$ would define a cusp of full rank. Hence we can choose a finite sequence of base points $\bar n^U_i\in \bar N^U\setminus \exp(\bar{\mathfrak{n}}_{-2\alpha})$ and vectors $v_i\in V_{\varphi}$, $i=1,\dots,r:=\dim {}^{M_U}{\rm Pol}(\bar N^U,V_{\tilde {\varphi}})^n$ such that the following map is an isomorphism: $$\Phi:{}^{M_U}{\rm Pol}(\bar N^U,V_{\tilde {\varphi}})^n\to{\mathbb {C}}^r\ , f\mapsto (\langle v_1,f(n^U_1)\rangle,\dots,\langle v_r,f(n^U_r)\rangle)\ .$$ We now consider the composition $$A(\lambda):=\Phi\circ \frac{1}{c_\lambda} {\mathrm{Gr}} [\pi^U_*]\circ \Phi^{-1}:{\mathbb {C}}^n\to {\mathbb {C}}^n\ .$$ Lemma \ref{argumentlem8} implies that $$A(\lambda)=1+O(|\lambda|^{-1})\ .$$ In particular, $A$ is injective, if $\lambda\in{\mathfrak{a}}$ is sufficiently large. This implies the assertion of Lemma \ref{argumentlem44}. \subsubsection{} We now show Lemma \ref{argumentlem8}. Fix $\bar n^U\in \bar N^U\setminus \exp(\bar{\mathfrak{n}}_{-2\alpha})$. We set $$\Psi(n_V):=\log a(n_V\bar n^U) \ ,\quad g(n_V):=f(\bar n(n_V\bar n^U))$$ Then we can write $$\int_{N_V} f(\bar n(n_V\bar n^U))a(n_V\bar n^U)^{-\lambda-\rho}dn_V=\int_{N_V} {\rm e}^{(-\lambda-\rho)\Psi( n_V)}g(n_V)dn_V\ .$$ \begin{lem}\label{argumentlem9} The function $\Psi(n_V)$ has a unique non-degenerate absolute minimum $\Psi(1)=0$ at $n_V=1$. Furthermore, there exists a compact neighbourhood $K\subset N_V$ of $1$ such that $\alpha(\Psi(n_V))\ge 1$ for $n_V\not\in K$. \end{lem} We first show that Lemma \ref{argumentlem9} implies \ref{argumentlem8}. We split the integral as $\int_{N_V}=\int_{K}+\int_{N_V\setminus K}$. Note that $\bar n(N_V\bar n^U)\subset \bar N$ is a pre-compact subset. It follows that $g$ is smooth and uniformly bounded. We approximate the first summand by a Gaussian integral at the minimum of $\Psi$ and get $$\int_K {\rm e}^{(-\lambda-\rho)(\Psi(n_V))}g(n_V)dn_V= \int_K {\rm e}^{(-\lambda-\rho)(\Psi(n_V))}dn_V \left(g(0)+O(\lambda^{-1})\right)\ .$$ Furthermore, $\int_K {\rm e}^{(-\lambda-\rho)(\Psi(n_V))}dn_V$ decreases at most as $|\lambda|^{-\dim N_V/2}$. We claim that the contributions $\int_{N_V\setminus K}$ decrease exponentially so that these parts of the integrals can only contribute exponentially small error terms. This claim implies Lemma \ref{argumentlem8}. In order to see the claim we write $$\int_{N_V\setminus K}{\rm e}^{(-\lambda-\rho)(\Psi(n_V))}dn_V =\min_{n_V\in N_V\setminus K}{\rm e}^{(-\lambda+\alpha)(\Psi(n_V))}\int_{N_V\setminus K}{\rm e}^{(-\alpha-\rho)(\Psi(n_V))}dn_V\ .$$ The integral on the right-hand side converges, and $$\min_{n_V\in N_V\setminus K}{\rm e}^{(-\lambda+\alpha)(\Psi(n_V))}\le {\rm e}^{-c|\lambda|}$$ for a suitable constant $c>0$. This finishes the proof of Lemma \ref{argumentlem8} under the assumption of Lemma \ref{argumentlem9}. \subsubsection{} We now show Lemma \ref{argumentlem9}. In order to compute $a(n_V\bar n^U)$ we may assume that $G$ is the subgroup of $GL(n+1,{\mathbb{F}})$ that preserves the ${\mathbb{F}}$-valued Hermitian scalar product \begin{equation}\label{argumenteq45}\bar v_0 w_n+ \bar v_n w_0+\bar v_1 w_1+ \dots + \bar v_{n-1} w_{n-1}\ \end{equation} on the right ${\mathbb{F}}$-vector space ${\mathbb{F}}^{n+1}$. We choose $$A:=\left\{\left(\begin{array}{ccc}a&0&0\\0&1_{n-1\times n-1}&0\\0&0&a^{-1}\end{array}\right)\:|\: a\in {\mathbb{R}}^+\right\}$$ and get $$N:=\left\{\left(\begin{array}{ccc}1&w&p-\frac{\|w\|^2}{2}\\0&1&-\bar w^t\\0&0&1\end{array}\right)\:|\: w\in {\mathbb{F}}^{n-1}\ ,p\in {\mbox{\rm im}}\, {\mathbb{F}}\right\}\ .$$ The parametrization of $N$ given here is via the exponential map, if we identify ${\mathfrak{n}}={\mathfrak{n}}_\alpha\oplus{\mathfrak{n}}_{2\alpha}\cong {\mathbb{F}}^{n-1}\oplus {\rm Im}\,{\mathbb{F}}$. Furthermore, $$\bar N:=\left\{\left(\begin{array}{ccc}1&0&0\\-\bar v^t&1&0\\q-\frac{\|v\|^2}{2}&v&1\end{array}\right)\:|\: v\in {\mathbb{F}}^{n-1}\ ,q\in {\rm Im}\, {\mathbb{F}}\right\}\ .$$ Since they are $A$-invariant, the subspaces ${\mathfrak{n}}_V\subset {\mathfrak{n}}$ and $\bar {\mathfrak{n}}^U\subset \bar {\mathfrak{n}}$ are given in this identification as $W\oplus P\subset {\mathfrak{n}}$ and $V\oplus Q\subset \bar {\mathfrak{n}}$ for real subspaces $W,V\subset {\mathbb{F}}^{n-1}$ and $P,Q\subset {\rm Im}\, {\mathbb{F}}$. We consider the invariant form $\langle A,B\rangle:={\rm Re }\,{\rm Tr} AB$ on ${\mathfrak{g}}$. Given $W\oplus P$, the space $V\oplus Q$ is characterized by $\langle{\mbox{\rm ad}}(V\oplus W)(H),W\oplus H\rangle=0$ (see (\ref{argumenteq33})). Explicitly, $$H=\left(\begin{array}{ccc}1&0&0\\0&0&0\\0&0&-1\end{array}\right)\ ,\langle{\mbox{\rm ad}}(v+q)(H),w+p\rangle=-2{\rm Re } (\bar w v^t+qp)\ .$$ Therefore we have \begin{equation}\label{argumenteq88}V=W^\perp\ , Q=P^\perp\end{equation} with respect to the natural euclidean pairings on ${\mathbb{F}}^{n-1}$ and ${\rm Im}\, {\mathbb{F}}$. The vector $e_0:=(1,\dots,0)\in {\mathbb{F}}^{n+1}$ is the highest weight vector of the standard representation of $G$ with weight $\alpha$. Similarly, $e_n:=(0,\dots,0,1)$ is the lowest weight vector with weight $-\alpha$. Let $g=\bar n man\in \bar NMAN$. If $\langle.,.\rangle$ denotes the ${\mathbb{F}}$-valued scalar product (\ref{argumenteq45}), then $$\langle ge_0,e_n\rangle=\langle\bar ma n e_0,e_n\rangle=\langle ma n e_0,\bar n^{-1}e_n\rangle=a^\alpha\langle m e_0,e_n\rangle\ .$$ The subspace $e_0{\mathbb{F}}\cong {\mathbb{F}}$ is invariant under the group $M$. In particular, we have a homomorphism $\vartheta:M\to {\mathbb{F}}^*$ such that $m e_0=e_0 \vartheta(m)$. It now follows that $\langle m e_0,e_n\rangle=\bar \vartheta(m)$. Since $M$ is compact, we have $\|\vartheta(m)\|=1$ and therefore $a^\alpha=\|\langle ge_0,e_n\rangle\|$. If we parametrize $(w,p)=n_V\in N_V$ and $(v,q)=\bar n^U\in \bar N^U$ as above, then we get \begin{equation}\label{argumenteq100} a(n_Vn^U)^{2\alpha}=\|1-w\bar v^t+(p-\frac{\|w\|^2}{2})(q-\frac{\|v\|^2}{2})\|^2\ .\end{equation} We fix $(v,q)\in V\oplus Q$ with $v\not=0$. We must show that the right-hand side has a unique absolute minimum $1$ at $(w,p)=(0,0)$, and that this minimum is non-degenerate. Using (\ref{argumenteq88}) we get \begin{eqnarray*} {\rm Re }(1-w\bar v^t+(p-\frac{\|w\|^2}{2})(q-\frac{\|v\|^2}{2}))&=&1+\frac{\|v\|^2\|w\|^2}{4}\\ {\rm Im}(1-w\bar v^t+(p-\frac{\|w\|^2}{2})(q-\frac{\|v\|^2}{2}))&=&pq-w\bar v^t-\frac{\|w\|^2}{2}q-\frac{\|v\|^2}{2}p\ . \end{eqnarray*} First of all, $a(n_Vn^U)^{2\alpha}\ge 1$ and $a(1)^{2\alpha}=1$. Moreover, if $a(n_V\bar n^U)^{2\alpha}=1$, then $w=0$. Since ${\rm Re }(\overline{pq}p)={\rm Re } \bar q\|p\|^2)=0$ we have $pq\perp \frac{\|v\|^2}{2}p$. Hence, the equality $a(n_V\bar n^U)=1$ implies in addition to $w=0$ that also $p=0$. We thus have shown that $a(n_V\bar n^U)^{2\alpha}$ takes its unique absolute minimum at $1$. Next we show that it is non-degenerate. The Hessian $h(p,w)$ is the part of the polynomial (\ref{argumenteq100}) which is quadratic in $(w,p)$. It can be written as $$h(w,p):=\|w\|^2\frac{\|v\|^2}{2}+\|p(q-\frac{\|v\|^2}{2})-w\bar v^t\|^2\ .$$ If $h(w,p)=0$, then from the first summand and $v\not=0$ we get $w=0$ and $p(q-\frac{\|v\|^2}{2})=0$. Since $(q-\frac{\|v\|^2}{2})\not=0$ we conclude that $p=0$. The last assertion of Lemma \ref{argumentlem9} follows from $\lim_{n_V\to \infty}a(n_V\bar n^U)^{2\alpha}=\infty$, which is easy to check using the explicit formula (\ref{argumenteq100}). This finishes the proof of Lemma \ref{argumentlem9}. We now have also finished the proof of Proposition \ref{argumentprop}.\hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} \begin{prop}\label{maertins} \begin{enumerate} \item We have $Ext_{U}(1_\lambda,{\varphi})=E_{U}(1_\lambda,{\varphi})$ for all $\lambda\in{\aaaa_\C^\ast}$. \item If $\lambda\not\in I_{\mathfrak{a}}$ or ${\rm Re }(\lambda)>-\rho$, then $Q_U(1_\lambda,{\varphi})=0$.\end{enumerate} \end{prop} {\it Proof.$\:\:\:\:$} \subsubsection{} As in\ref{neuj703} let ${\mathcal{E}}_{\infty_P}(1,{\varphi})^n$ be the torsion-free coherent sheaf on ${\aaaa_\C^\ast}$ of holomorphic families of $U$-invariant distributions supported on $\infty_P$, on which $A$ acts by multiplication by the function $\lambda\mapsto a^{\lambda-\rho-n\alpha}$. By ${\mathcal{E}} xt_{\infty_P}(1,{\varphi})^n$ we denote the subsheaf generated by the restriction of $[ext^{U}]$ to the homogeneous part $({R}_{U}(1_{-.},\tilde{\varphi})^n)^*$. The space $Ext_{\infty_P}(1_\lambda,{\varphi})^n$ is the geometric fibre of ${\mathcal{E}} xt_{\infty_P}(1,{\varphi})^n$ for the generic set of $\lambda\in{\aaaa_\C^\ast}$, where $[ext^U]$ is regular. \subsubsection{}\label{neuj705} The sheaf ${\mathcal{E}}_{\infty_P}(1,{\varphi})^n$ is the sheaf of sections of a finite-dimensional trivial holomorphic vector bundle $E_{\infty_P}(1,{\varphi})^n\to {\aaaa_\C^\ast}$. The torsion-free subsheaf ${\mathcal{E}} xt_{\infty_P}(1,{\varphi})^n$ corresponds to a bundle ${\mathrm{ Ext}}_{\infty_P}(1,{\varphi})^n$. This discussion shows the following: if the inclusion $Ext_{\infty_P}(1_\lambda,{\varphi})^n\hookrightarrow E_{\infty_P}(1_\lambda,{\varphi})^n$ is surjective at one point $\lambda\in {\aaaa_\C^\ast}$, then it is surjective for generic $\lambda$, i.e. outside a discrete set. However, we know from Proposition \ref{argumentprop} that this inclusion is surjective for many $\lambda$, hence it is so generically. \subsubsection{}\label{miau} In view of Lemma \ref{homjk}, 2. we have \begin{equation}\label{neuj708} E_{\infty_P}(1_\lambda,{\varphi})^n/E xt_{\infty_P}(1_\lambda,{\varphi})^n\cong \ker\left( Q_U(1_\lambda,{\varphi})^n\to \bar Q_{\infty_P}(1_\lambda,{\varphi})^n\right)\ . \end{equation} By \ref{neuj705} the quotient on the left hand side is trivial generically. The same is true for $\bar Q_{\infty_P}(1_\lambda,{\varphi})^n$ by Lemma \ref{vermaolbrich}. We conclude that $Q_U(1_\lambda,{\varphi})^n$ is trivial outside a discrete set. \subsubsection{} We now prove the first assertion of Prop.~\ref{maertins}. We know that $Ext_{U}(1_\lambda,{\varphi})\subset E_{U}(1_\lambda,{\varphi})$. Let now $f\in E_{U}(1_\lambda,{\varphi})$ be given as the value at $\lambda$ of a meromorphic family $f_\mu\in {}^{U} D_{\{1\},k}(1_\mu,{\varphi})$ for some sufficiently large $k$. Let $res^U$ be the meromorphic family of continuous maps $res^U:{}^{U} D_{\{1\},k}(1_\mu,{\varphi}):\rightarrow D_{U,k_1}(1_\mu,{\varphi})$ for suitable $k_1\in{\mathbb {N}}$ (see Definition \ref{neuj707}). By \ref{miau} for generic $\mu$ we can write $f_\mu=ext^{U} g_\mu$ for some $g_\mu\in D_{U,k}(1_\mu,{\varphi})$. We conclude $ext^{U}\circ res^{U}(f_\mu)= ext^{U}\circ res^{U} \circ ext^U(g_\mu)= ext^U(g_\mu)=f_\mu$ by Lemma \ref{weedef}. Thus \begin{equation}\label{wau} ext^{U}\circ res^{U}(f_\mu)=f_\mu \quad\mbox{ for all }\mu\ . \end{equation} We conclude that $f\in Ext_{U}(1_\lambda,{\varphi})$. \subsubsection{} We now turn to the second assertion. The first assertion implies that the left hand side of (\ref{neuj708}) is trivial for all $\lambda$. Therefore the map $Q_U(1_\lambda,{\varphi})^n\to \bar Q_{\infty_P}(1_\lambda,{\varphi})^n$ is always injective. We now apply Lemma \ref{vermaolbrich}. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} The argument that proofs Equation (\ref{wau}) also shows the following. \begin{lem}\label{wauwau} The composition $ext^U\circ res^U$ is the identity on $Ext_{U}(1_{\lambda},{\varphi})$. \end{lem} \subsection{The scattering matrix} \subsubsection{} Recall that the family of intertwining operators (see \ref{gaa100} for the unnormalized version and \cite{MR1749869} for normalizations) forms a meromorphic family of operators. It therefore maps (holomorphic) families of invariant sections to (meromorphic) families. It follows that $J_\lambda$ maps $E_U(1_\lambda,{\varphi})$ to $E_U(1_{-\lambda},{\varphi})$ if $\lambda\in {\aaaa_\C^\ast}$ is such that $J_\lambda$ is regular (e.g. $\lambda\not\in I_{\mathfrak{a}}$, see \ref{gaa101}). By Proposition \ref{maertins} we get a mapping \begin{equation}\label{uhu} J_\lambda:Ext_{U}(1_\lambda,{\varphi})\rightarrow Ext_{U}(1_{-\lambda},{\varphi})\ . \end{equation} Moreover, for given $k\in{\mathbb {N}}_0$ we have $$J_\lambda:Ext_{U,k}(1_\lambda,{\varphi})\rightarrow Ext_{U,k_1}(1_{-\lambda},{\varphi})$$ if $k_1\in{\mathbb {N}}_0$ is sufficiently large. \subsubsection{}\label{neuj2004} We can now define the scattering matrix $$S^U_\lambda:D_U(1_\lambda,{\varphi})\rightarrow D_U(1_{-\lambda},{\varphi})\ .$$ Fix $k\in{\mathbb {N}}_0$ and a compact subset $W\subset {\mathfrak{a}}^*$. Then we choose $k_0>k$ and $k_1,k_2$ such that $$J_\lambda: Ext_{U,k_0}(1_\lambda,{\varphi})\rightarrow Ext_{U,k_1}(1_{-\lambda},{\varphi})$$ for generic $\lambda$ (e.g. non-integral) with ${\rm Re }(\lambda)\in W$ and $$res^U:D_{\{1\},k_1}(1_\lambda,{\varphi})\rightarrow D_{U,k_2}(1_\lambda,{\varphi})$$ is defined (Definition \ref{neuj707}) as a meromorphic family on $W$. Then we consider the composition $$S^U_\lambda:=res^U\circ J_\lambda\circ ext^U :D_{U,k}(1_\lambda,{\varphi})\rightarrow D_{U,k_2}(1_{-\lambda},{\varphi})\ .$$ By definition $S^U_\lambda$ is a meromorphic family of continuous maps. Since $$J_\lambda\circ ext^U:D_U(1_\lambda,{\varphi})\to Ext_U(1_{-\lambda},{\varphi})$$ Lemma \ref{weedef} now implies that $S^U_\lambda$ is well-defined independently of the choices made for $res^U$. \subsubsection{} Letting $k$ tend to infinity and $W$ run over a sequence of compact subsets exhausting ${\mathfrak{a}}^*$ we are arrive at the following definition. \begin{ddd}\label{neuj2003} We define the scattering matrix as the meromorphic family of continuous maps $$S^U_\lambda:D_U(1_\lambda,{\varphi})\rightarrow D_U(1_{-\lambda},{\varphi})\ ,$$ which is given by the composition $$S^U_\lambda(f):=res^U\circ J_\lambda\circ ext^U(f)\ $$ whenever the constituents are regular. \end{ddd} \subsubsection{} If $-\lambda$ is admissible in the sense of \ref{neuj300} (with ${\varphi}$ replaced by $\tilde {\varphi}$), then the push-down $$\pi^U_*:B_{\{1\}}(1_\lambda,\tilde{{\varphi}})\to B_U(1_\lambda,\tilde{{\varphi}})$$ is regular and admits a right-inverse. Hence it induces an isomorphism $$B_U(1_\lambda,\tilde{{\varphi}})\cong B_{\{1\}}(1_\lambda,\tilde{{\varphi}})/\ker\pi^U_*\ .$$ If $\lambda\not\in I_{\mathfrak{a}}$, then $J_\lambda$ is regular. We claim that it maps $\ker\pi^U_*$ to the kernel of $\pi^*_U$ at $-\lambda$. Let $f\in\ker\pi^U_*$ and $\phi\in D_U(1_{\lambda},{\varphi})$. By (\ref{uhu}) there is an $\psi\in\ D_U(1_{-\lambda},{\varphi})$ such that $J_\lambda\circ ext^U(\phi)=ext^U(\psi)$. We find $$ \langle \pi^U_*\circ J_\lambda (f),\phi\rangle= \langle f,J_\lambda\circ ext^U(\phi)= \langle f,ext^U(\psi)\rangle= \langle \pi^U_*(f),\psi\rangle=0\ . $$ The claim follows. Therefore, if $\pm\lambda$ is admissible and non-integral, then the operator $J_\lambda$ descends to a map \begin{equation}\label{neuj801} \tilde S^U_\lambda: B_U(1_\lambda,\tilde{{\varphi}})\rightarrow B_U(1_{-\lambda},\tilde{{\varphi}})\ . \end{equation} On the other hand, we have a meromorphic family of maps given by the adjoint of the scattering matrix $${}^tS^U_\lambda:B_U(1_\lambda,\tilde{{\varphi}})\rightarrow B_U(1_{-\lambda},\tilde{{\varphi}})\ .$$ \begin{lem} \label{gen} If $\pm\lambda$ is non-integral and admissible, then we have $\tilde S^U_\lambda={}^tS^U_\lambda$. \end{lem} {\it Proof.$\:\:\:\:$} Recall from Lemma \ref{wauwau} that $ext^U\circ res^U$ is the identity on $Ext_{U}(1_{-\lambda},{\varphi})$. For non-integral $\lambda$ we have $Ext_{U}(1_{-\lambda},{\varphi})={}^U C^{-\infty}(\partial X,V(1_{-\lambda},{\varphi}))$ by Proposition \ref{maertins},~2. The assumptions on $\lambda$ imply that $\tilde S^U_\lambda$ is defined, that $\pi^U_*:B_{\{1\}}(1_\lambda,\tilde{{\varphi}})\rightarrow B_U(1_\lambda,\tilde{{\varphi}})$, $ext^U: D_U(1_{\lambda},{\varphi})\rightarrow D_{\{1\}}(1_{\lambda},{\varphi})$, and $J_\lambda$ are regular. Let $f\in B_U(1_\lambda,\tilde{{\varphi}})$ be given by $\pi^U_*(F)$, $F\in B_{\{1\}}(1_{\lambda},{\varphi})$. Furthermore let $\phi\in D_U(1_{\lambda},{\varphi})$. Then we compute \begin{eqnarray*} \langle \tilde S^U_\lambda(f),\phi\rangle&=& \langle \pi^U_*\circ J_\lambda(F),\phi\rangle\\ &=&\langle F,J_\lambda\circ ext^U(\phi)\rangle\\ &=&\langle F,ext^U\circ res^U\circ J_\lambda\circ ext^U(\phi)\rangle\\ &=&\langle f,S^U_\lambda(\phi)\rangle\\ &=&\langle {}^tS^U_\lambda(f),\phi\rangle\ . \end{eqnarray*} \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Note that the normalization of ${\varphi}$ in the present subsection differs from that in \ref{weih213}, since we want $\tilde {\varphi}$ to be normalized as required there. As a consquence the space $R_{U}(1_\lambda,{\varphi})$ contains summands with negative index. If we put $k_{\varphi}:=l_{\varphi}/\alpha$, then we have $$R_{U}(1_\lambda,{\varphi})=\prod^\infty_{n=-k_{\varphi}} R_{U}(1_\lambda,{\varphi})^n\ ,\quad B_U(1_\lambda,{\varphi})=\bigcap_{k\ge -k_{\varphi}}B_{U,k}(1_\lambda,{\varphi})\ .$$ \subsubsection{} \begin{lem}\label{fundd} Assume that $U$ defines a cusp of smaller rank. If $l_{\varphi}<2\rho^U$, then there is a natural non-degenerate pairing between $B_{U,-k_{\varphi}}(1_\lambda,{\varphi})$ and $B_{U,0}(1_{-\lambda},\tilde{\varphi})$ given by integration over $B_U$. We obtain an inclusion $$B_{U}(1_\lambda,{\varphi})\subset B_{U,-k_{\varphi}}(1_\lambda,{\varphi})\hookrightarrow D_U(1_\lambda,{\varphi})\ .$$ \end{lem} {\it Proof.$\:\:\:\:$} \subsubsection{} Let $f\in B_{U,-k_{\varphi}}(1_\lambda,{\varphi})$ and $\phi\in B_{U,0}(1_{-\lambda},\tilde{\varphi})$. Let $W\subset N\setminus N_V$ be any compact subset. Then there exists a constant $C\in {\mathbb{R}}$ such that for all $\xi\in W$ and $a\in A_+$ we have \begin{eqnarray} |\langle f(\xi^aw),\phi(\xi^aw) \rangle | &=& |\langle{\varphi}(a){\varphi}(a)^{-1} f(\xi^aw),\phi(\xi^aw) \rangle|\nonumber\\ &=&|\langle {\varphi}(a)^{-1} f(\xi^aw),\tilde{\varphi}(a)^{-1}\phi(\xi^aw)\rangle|\nonumber\\ &\le& C \|f\|_{-k_{\varphi},0} \|\phi\|_{0,0} a^{l_{\varphi}-4\rho^U}\label{stegh} \end{eqnarray} (see \ref{weih136} for the definition of the norms). \subsubsection{} Let $F\subset N_V$ be any compact fundamental domain for the lattice $V\subset N_V$. We can now write \begin{eqnarray*} &&\int_{U\backslash N}|\langle f(xw),\phi(xw) \rangle | dx \\&=&\frac{1}{[U:U^0]}\int_{N_V\backslash N} \int_{F} |\langle f(yvw),\phi(yvw) \rangle |dv dy\\ &=&\frac{1}{[U:U^0]}\int_A \int_{S(N_V\backslash N)} \int_{F} |\langle f(\xi^a vw),\phi(\xi^avw) \rangle |dv d\xi a^{2\rho^U} da\\ &=&\frac{1}{[U:U^0]}\int_A \int_{S(N_V\backslash N)} \int_{F} |\langle f((\xi v^{a^{-1}})^aw),\phi((\xi v^{a^{-1}})^aw) \rangle |dv d\xi a^{2\rho^U} da\\ &=& I_+ + I_-\ , \end{eqnarray*} where $I_\pm$ are the integrals over $A_\pm$, respectively. Inserting (\ref{stegh}) we obtain $$ \int_{S(N_V\backslash N)} \int_{F} |\langle f((\xi v^{a^{-1}})^aw),\phi((\xi v^{a^{-1}})^aw) \rangle |dv d\xi \le C_1 \|f\|_{-k_{\varphi},0} \|\phi\|_{0,0}a^{l_{\varphi}-4\rho^U}$$ and therefore $$I_+ \le C_2 \|f\|_{-k_{\varphi},0} \|\phi\|_{0,0} \int_{ A_+}a^{l_{\varphi}-2\rho^U}da \ .$$ The integral on the right-hand side converges since $l_{\varphi}-2\rho^U<0$ by assumption. Since $I_-$ can clearly be estimated by $C_3 \|f\|_{-k_{\varphi},0} \|\phi\|_{0,0}$, we have shown the lemma. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{}\label{neuj2005} Using the inclusion (Lemma \ref{fundd}) $$B_{U}(1_\lambda,{\varphi})\hookrightarrow D_U(1_\lambda,{\varphi})$$ we can consider the restriction of the scattering matrix $$(S^U_\lambda)_{|B_{U}(1_\lambda,{\varphi})}:B_{U}(1_\lambda,{\varphi})\to D_{U}(1_{-\lambda},{\varphi})\ .$$ If $f\in B_{U}(1_\lambda,{\varphi})$, then the restriction of the distribution $ext^U(f)$ to $\Omega_P$ is smooth. Since $J_\lambda$ is pseudo-local it it follows that $J_\lambda\circ ext(f)$ is smooth on $\Omega_P$. Hence the restriction of the distribution $S^U_\lambda(f)$ to the Schwartz space ${S}_U(1_\lambda,\tilde{\varphi})$ is given by integration against a smooth section. The following lemma asserts that $S^U_\lambda(f)={}^tS_\lambda^U(f)\in B_{U}(1_{\lambda},{\varphi})$ considered as an element of $D_{U}(1_{-\lambda},{\varphi})$ via Lemma \ref{fundd}. \subsubsection{} \begin{lem}\label{inside} Assume that $U$ defines a cusp of smaller rank and that $l_{\varphi} <2\rho^U$. Then we have an equality of meromorphic families \begin{equation}\label{neuj900} (S^U_\lambda)_{|B_{U}(1_\lambda,{\varphi})}={}^tS^U_\lambda\ . \end{equation} \end{lem} {\it Proof.$\:\:\:\:$} By meromorphic continuation it suffices to show the equality (\ref{neuj900}) for non-integral $\lambda$ in the open subset $\{|{\rm Re }(\lambda)|<\rho^U-l_{\varphi}/2\}\subset {\aaaa_\C^\ast}$ such that $-\lambda$ is admissible in the sense of \ref{neuj300}. Then the push-down converges at $\pm\lambda$. Let $f\in B_{\{1\}}(1_\lambda,\tilde{\varphi})$. We consider an element $\phi\in {S}_{\{1\}}(1_\lambda, {\varphi})$. We view $\tilde S^U_\lambda\circ \pi^U_*(f)$ (see (\ref{neuj801})) as an element of $D_U(1_{-\lambda},\tilde{\varphi})$ and compute \begin{eqnarray*} \langle \tilde S^U_\lambda\circ \pi^U_*(f),\pi^U_*(\phi)\rangle_{B_U} &=& \langle \pi^U_*\circ J_\lambda (f),\pi^U_*(\phi)\rangle_{B_U}\\ &=&\langle \pi^U_*\circ J_\lambda (f),\phi\rangle_{\partial X}\\ &=&\sum_{u\in U} \langle \pi^{1_{-\lambda},\tilde{\varphi}}(u) J_\lambda(f),\phi\rangle_{\partial X}\\ &=&\sum_{u\in U} \langle f,\pi^{1_{-\lambda},{\varphi}}(u) J_\lambda(\phi)\rangle_{\partial X}\\ &=&\langle f,\pi^U_* \circ J_\lambda (\phi)\rangle_{\partial X}\\ &=&\langle \pi_*^U(f),\pi^U_* \circ J_\lambda (\phi)\rangle_{B_U}\\ &=&\langle ext^U\circ \pi_*^U(f),J_\lambda (\phi)\rangle_{\partial X}\\ &=&\langle J_\lambda\circ ext^U\circ \pi^U_*(f),\phi\rangle_{\partial X}\\ &=&\langle ext^U\circ res^U\circ J_\lambda\circ ext^U\circ \pi^U_*(f),\phi\rangle_{\partial X}\\ &=&\langle S^U_\lambda\circ \pi^U_*(f),\pi^U_*(\phi)\rangle_{B_U}\ . \end{eqnarray*} Since the $\pi^U_*(\phi)$, $\phi\in {S}_{\{1\}}(1_\lambda,{\varphi})$ (resp. $\pi_*^U(f)\in B_{U}(1_\lambda,\tilde{\varphi})$, $f\in B_{\{1\}}(1_\lambda,\tilde{\varphi})$), exhaust ${S}_U(1_\lambda,{\varphi})$ (resp. $B_{U}(1_\lambda,\tilde{\varphi})$) we conclude that $$\tilde S^U_\lambda(h) =S^U_\lambda(h) $$ as smooth sections for all $h\in B_{U}(1_\lambda,\tilde{\varphi})$ and hence $${}^tS^U_\lambda = (S^U_\lambda)_{|B_{U}(1_\lambda,{\varphi})} $$ by Lemma \ref{gen}. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Next we show that the scattering matrix is off-diagonally smoothing. Let $\chi,\tilde\chi$ be smooth cut-off functions on $B_U$ with compact support such that $\tilde\chi \chi = \chi$. \begin{lem}\label{offside} Assume that $U$ defines a cusp of smaller rank and that $l_{\varphi} <2\rho^U$. Then we have a meromorphic family of continuous maps $$(1-\tilde\chi)\circ S^U_\lambda \circ \chi:D_U(1_\lambda,{\varphi})\rightarrow B_U(1_{-\lambda},{\varphi})\ .$$ \end{lem} {\it Proof.$\:\:\:\:$} We are going employ the fact that $J_\lambda$ is off-diagonally smoothing. Let $\chi^U\in C^\infty(\Omega_P)$ be a cut-off function such that $\sum_{u\in U} u^* \chi^U\equiv 1$, and such that the restriction of the projection $\Omega_P\rightarrow B_U$ to ${\mathrm{supp}}(\chi^U)$ is proper. Then there exists a cut-off function $\bar\chi$ on $\partial X$ such that $\bar\chi \chi\chi^U=0$ and $\bar\chi(1-\tilde\chi)=1-\tilde\chi$. Finally we choose a compactly supported cut-off function $\hat\chi$ on $B_U$ such that $\chi^U \bar \chi \hat \chi=0$ and $\hat \chi \chi=\chi$. Let $\phi\in {S}_U(1_{\lambda},\tilde{\varphi})$ be any test function and $f\in D_U(1_\lambda,{\varphi})$. We approximate $\chi f$ by a sequence of smooth functions $f_\alpha$ with compact support. We can in addition assume that ${\mathrm{supp}} f_\alpha\subset \{\hat \chi=1\}$ for all $\alpha$. Then we compute \begin{eqnarray*} \langle (1-\tilde\chi)S^U_\lambda(\chi f),\phi\rangle_{B_U}&=& \lim_{\alpha} \langle (1-\tilde\chi)S^U_\lambda f_\alpha,\phi\rangle_{B_U}\\ &\stackrel{\mathrm{Lemma}\,\ref{inside}}{=}&\lim_{\alpha} \langle (1-\tilde\chi){}^tS^U_\lambda f_\alpha,\phi\rangle_{B_U}\\ &=&\lim_{\alpha} \langle f_\alpha, S^U_\lambda( (1-\tilde\chi) \phi)\rangle_{B_U}\\ &=&\lim_{\alpha} \langle f_\alpha, \{res^U\} \circ J_\lambda \circ ext^U((1-\tilde\chi) \phi)\rangle_{B_U}\\ &=&\lim_{\alpha} \langle \chi^U f_\alpha , J_\lambda \circ ext^U((1-\tilde\chi) \phi)\rangle_{\partial X}\\&=&\lim_{\alpha} \langle \chi^U\hat \chi f_\alpha , J_\lambda \circ \bar \chi ext^U((1-\tilde\chi) \phi)\rangle_{\partial X}\\ &=&\lim_{\alpha}\langle \pi^U_* (\bar\chi J_\lambda (\chi^U \hat \chi f_\alpha)),(1-\tilde\chi) \phi \rangle_{B_U}\\ &\stackrel{(*)}{=}&\langle \pi^U_* \left(\bar\chi J_\lambda ( \chi^U \chi f)-\tilde\chi\pi^U_* (\bar\chi J_\lambda (\chi \chi^U f))\right) , \phi \rangle_{B_U}\ . \end{eqnarray*} In order to see the equality marked by $(*)$ note that that $\bar\chi J_\lambda \hat\chi\chi^U$ is a continuous map from $D_U(1_\lambda,{\varphi})$ to $B_{\{1\}}(1_{-\lambda},{\varphi})$ (since $J_\lambda$ is off-diagonal smoothing) and that $\lim_{\alpha}\chi^U \hat \chi f_\alpha=\chi^U\chi f$. The assertion of the Lemma follows from the identity proved above: $$(1-\tilde\chi)\circ S^U_\lambda\circ \chi (.) = \pi^U_* \left(\bar\chi J_\lambda (\chi \chi^U .)-\tilde\chi \pi^U_* (\bar\chi J_\lambda (\chi \chi^U .))\right)\ .$$ \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \section{The general case} \subsection{The space $B_\Gamma(\sigma_\lambda,{\varphi})$}\label{samel1} \subsubsection{} Let $\Gamma\subset G$ be a torsion-free geometrically finite discrete subgroup (see Definition \ref{t67}) such that all its cusps are regular (Definition \ref{t799}). Furthermore let $({\varphi},V_{\varphi})$ be an admissible twist (Definition \ref{weih134}). We are going to employ the notation introduced in Subsection \ref{geomf}. \subsubsection{} On $\bar Y$ we choose a partition of unity $\{\chi_p\}_{p\in {\mathcal{P}}_\Gamma\cup \{0\}}$ such that $\chi_p\in C^\infty(\bar Y)$ and ${\mathrm{supp}}(\chi_p)\subset \bar{Y}_p$ for all $p$. Restriction of these functions to $B_\Gamma$ gives a partition of unity $\{\chi_p\}_{p\in{\mathcal{P}}_\Gamma^<\cup\{0\}}$ on $B_\Gamma$ such that ${\mathrm{supp}}(\chi_p)\subset B_p$. Here we denote the restriction of $\chi_p$ to the boundary by the same symbol $\chi_p$. In a similar manner for $P\in p\in {\mathcal{P}}^<_\Gamma$ we let $e_P:B_p\rightarrow B_{U_P}$ denote the map defined as restriction of the map $e_P:\bar Y_p\rightarrow \bar Y_{U_P}$. Let $\chi_P\in C^\infty(\bar{Y}_{U_P})$ be the cut-off function which is supported on the range of $e_P$ and satisfies $\chi_p=e_P^*\chi_P$. \subsubsection{}\label{neuj3005} Using cut-off with $\chi_p$ and the map $e_P:B_p\rightarrow B_{U_P}$ for $P\in p\in{\mathcal{P}}^<_\Gamma$ we define maps \begin{eqnarray*} T_P&:&C^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))\rightarrow C^\infty(B_{U_P},V_{B_{U_P}}(\sigma_\lambda,{\varphi}))\\ T^P&:&C^\infty(B_{U_P},V_{B_{U_P}}(\sigma_\lambda,{\varphi}))\rightarrow C^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))\ . \end{eqnarray*} \subsubsection{} We define the Schwartz space for $\Gamma$ as the space of smooth sections which belong to the Schwartz space \ref{weih245} near all cusps. \begin{ddd} We define ${S}_{\Gamma,k}(\sigma_\lambda,{\varphi})$ to be the subspace of all $f\in C^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$ such that $T_P(f)\in {S}_{U_P,k}(\sigma_\lambda,{\varphi})$ for all $P\in p\in {\mathcal{P}}_\Gamma^<$. The inclusion $${S}_{\Gamma,k}(\sigma_\lambda,{\varphi})\hookrightarrow C^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$$ and the maps $$T_P:{S}_{\Gamma,k}(\sigma_\lambda,{\varphi})\rightarrow {S}_{U_P,k}(\sigma_\lambda,{\varphi})$$ induce on ${S}_{\Gamma,k}(\sigma_\lambda,{\varphi})$ the structure of a Fr\'echet space. Furthermore we define the Fr\'echet and Montel space $${S}_{\Gamma}(\sigma_\lambda,{\varphi}):=\bigcap_{k\in{\mathbb {N}}_0}{S}_{\Gamma,k}(\sigma_\lambda,{\varphi})\ .$$ \end{ddd} \subsubsection{} \begin{lem}\label{loctrivg} The families $\{{S}_{\Gamma,k}(\sigma_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$ and $\{{S}_{\Gamma}(\sigma_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$ form locally trivial holomorphic bundles. \end{lem} {\it Proof.$\:\:\:\:$} Let $\Phi_{P,\lambda_0}$ denote the trivialization constructed in Lemma \ref{ytra} which is given by multiplication by $s_P^{(\lambda_0-\lambda)/\alpha}$ (here we add the index $P$ to the notation for the section $s$ constructed in Lemma \ref{ingf} in order to indicate its dependence on the parabolic subgroup). Let $s_p:=T^P(s_P)$, where $P\in\tilde {\mathcal{P}}^<$ represents $p$. Let $s_0\in C^\infty(B_\Gamma,V_{B_\Gamma}(1_{\rho+\alpha},{\varphi}))$ be any positive section (compare \ref{neuj911}). Then we form $s_\Gamma:=\sum_{p\in {\mathcal{P}}_\Gamma} s_p + \chi_0 s_0$. Multiplication by $s_\Gamma^{(\lambda_0-\lambda)/\alpha}$ defines isomorphisms $$\Phi_{\Gamma,\lambda_0}: {S}_{\Gamma,k}(\sigma_\lambda,{\varphi})\rightarrow {S}_{\Gamma,k}(\sigma_{\lambda_0},{\varphi})$$ and $$\Phi_{\Gamma,\lambda_0}: {S}_{\Gamma}(\sigma_\lambda,{\varphi})\rightarrow {S}_{\Gamma}(\sigma_{\lambda_0},{\varphi})\ .$$ We employ the maps $\Phi_{\Gamma,\lambda_0}$ in order to obtain the required local trivializations and to implement the holomorphic structures. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Similarly to the case of the Schwartz space we define the space $B_{\Gamma}(\sigma_\lambda,{\varphi})$ as the space of smooth sections with the same asymptotic expansions near the cusps as the spaces as the space defined in \ref{neuj104}. Recall the notation $\tilde {\mathcal{P}}^{max}$ from \ref{holle}. \begin{ddd} We define $$B_{\Gamma,k}(\sigma_\lambda,{\varphi}):=B_{\Gamma,k}(\sigma_\lambda,{\varphi})_1 \oplus \bigoplus_{P\in \tilde{\mathcal{P}}^{max}} {R}_{U_P}(\sigma_\lambda,{\varphi})\ ,$$ where $$B_{\Gamma,k}(\sigma_\lambda,{\varphi})_1\subset C^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$$ is the subspace of all $f$ such that $T_P(f)\in B_{U_P,k}(\sigma_\lambda,{\varphi})$ for all $P\in p\in {\mathcal{P}}_\Gamma^<$. The map $$B_{\Gamma,k}(\sigma_\lambda,{\varphi})\hookrightarrow C^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))\ ,$$ the maps $$T_P:B_{\Gamma,k}(\sigma_\lambda,{\varphi})\rightarrow B_{U_P,k}(\sigma_\lambda,{\varphi})\ ,$$ $P\in \tilde {\mathcal{P}}^<$, and the natural projections $$AS_P:B_{\Gamma,k}(\sigma_\lambda,{\varphi})\rightarrow {R}_{U_P}(\sigma_\lambda,{\varphi})\ ,$$ $P\in \tilde{\mathcal{P}}^{max}$, equip $B_{\Gamma,k}(\sigma_\lambda,{\varphi})$ with the structure of a Fr\'echet space. We further define the Fr\'echet and Montel space $$B_{\Gamma}(\sigma_\lambda,{\varphi}):= \bigcap_{k\in{\mathbb {N}}_0} B_{\Gamma,k}(\sigma_\lambda,{\varphi})\ .$$ \end{ddd} \subsubsection{} We define $${R}_{\Gamma,k}(\sigma_\lambda,{\varphi}):=\bigoplus_{P\in \tilde{\mathcal{P}}} {R}_{U_P,k}(\sigma_\lambda,{\varphi}) \ ,\quad {R}_{\Gamma}(\sigma_\lambda,{\varphi}) :=\bigoplus_{P\in \tilde{\mathcal{P}}} {R}_{U_P}(\sigma_\lambda,{\varphi})\ .$$ We have asymptotic term maps $$AS_P: B_{\Gamma,k}(\sigma_\lambda,{\varphi})\rightarrow {R}_{U_P,k}(\sigma_\lambda,{\varphi})\ ,$$ $AS_P:=AS\circ T_P$, where $AS$ was defined in \ref{neuj915} for pure cusps. The asymptotic term maps admit right-inverses $$L_P:{R}_{U_P,k}(\sigma_\lambda,{\varphi})\rightarrow B_{\Gamma,k}(\sigma_\lambda,{\varphi})$$ given by the natural inclusion for $P\in\tilde{\mathcal{P}}^{max}$, and by $L_P:=T^P\circ L$ if $P\in \tilde{\mathcal{P}}^<$, where $L$ was defined in \ref{weih253} for pure cusps. Let $$AS:B_{\Gamma,k}(\sigma_\lambda,{\varphi})\rightarrow {R}_{\Gamma,k}(\sigma_\lambda,{\varphi})$$ be induced by the maps $AS_P$ and $$L: {R}_{\Gamma,k}(\sigma_\lambda,{\varphi}) \rightarrow B_{\Gamma,k}(\sigma_\lambda,{\varphi})$$ be given by the sum of the maps $L_P$ for the various $P\in\tilde{\mathcal{P}}$. \subsubsection{} \begin{lem}\label{spst} The family $\{B_{\Gamma,k}(\sigma_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$ forms a trivial holomorphic bundle of Fr\'echet spaces. We have a split (by $L$) exact sequence \begin{equation}\label{trtra}0\rightarrow {S}_{\Gamma,k}(\sigma_\lambda,{\varphi})\rightarrow B_{\Gamma,k}(\sigma_\lambda,{\varphi}) \stackrel{AS}{\rightarrow} {R}_{\Gamma,k}(\sigma_\lambda,{\varphi})\rightarrow 0\ .\end{equation} Furthermore, the spaces $\{B_{\Gamma}(\sigma_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$ form a limit of locally trivial holomorpic bundles in the sense of Subsection \ref{llim} and fit into the exact sequence (which does not admit any continuous split) $$0\rightarrow {S}_{\Gamma}(\sigma_\lambda,{\varphi}) \rightarrow B_{\Gamma}(\sigma_\lambda,{\varphi}) \rightarrow {R}_{\Gamma}(\sigma_\lambda,{\varphi})\rightarrow 0\ .$$ \end{lem} {\it Proof.$\:\:\:\:$} The assertions follow from Lemma \ref{loctrivg} and the fact that the spaces ${R}_{\Gamma,k}(\sigma_\lambda,{\varphi})$ form locally trivial holomorphic vector bundles. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} We now show that the spaces $B_\Gamma(\sigma_\lambda,{\varphi})$ are compatible with twisting. Let $(\pi_{\sigma,\mu},V_{\pi_{\sigma,\mu}})$ be a finite-dimensional representation of $G$ as in \ref{weih144}. \begin{lem}\label{compat3} \begin{enumerate} \item We have holomorphic families of continuous maps (see \ref{neuj1000} for notation) $$i_{\sigma,\mu}^\Gamma:B_{\Gamma}(\sigma_\lambda,{\varphi})\rightarrow B_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$$ and $$p_{\sigma,\mu}^\Gamma:B_\Gamma(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})\rightarrow B_{\Gamma}(\sigma_\lambda,{\varphi})\ .$$ \item $B_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})$ is a ${\mathcal{Z}}({\mathfrak{g}})$-module. \item If $\lambda\not\in I_{\mathfrak{a}}$, then $i_{\sigma,\mu}^\Gamma$ maps $B_{\Gamma}(\sigma_\lambda,{\varphi})$ isomorphically onto $$\ker_\Gamma(Z(\lambda)):=\ker\{Z(\lambda): B_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})\rightarrow B_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi})\}$$ (see \ref{weih146} for notation), and the restriction of $p_{\sigma,\mu}^\Gamma$ to $$\ker\{\tilde Z(\lambda):B_\Gamma(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})\rightarrow B_\Gamma(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})\}$$ is an isomorphism onto $B_{\Gamma}(\sigma_\lambda,{\varphi})$ (here $\tilde Z(\lambda)$ is the adjoint of $\pi^{1_{-\lambda+\mu}\otimes \tilde\pi_{\sigma,\mu}, {\mathrm {id}}_{\tilde{\varphi}}}(\Pi(\lambda))$). \item The composition $$j_{\sigma,\mu}^\Gamma:B_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes {\varphi}) \stackrel{1-Z(\lambda)}{\rightarrow} \ker_\Gamma(Z(\lambda))\stackrel{(i_{\sigma,\mu}^\Gamma)^{-1}}{\rightarrow} B_{\Gamma}(\sigma_\lambda,{\varphi})$$ which is initially defined for $\lambda\not\in I_{\mathfrak{a}}$ extends to a meromorphic family of continuous maps. Similarly, the composition $$q_{\sigma,\mu}^\Gamma:B_{\Gamma}(\sigma_\lambda,{\varphi}) \stackrel{(p_{\sigma,\mu}^\Gamma)^{-1}}{\rightarrow} \ker_\Gamma(\tilde Z(\lambda)) \rightarrow B_\Gamma(1_{\lambda-\mu},\pi_{\sigma,\mu}\otimes {\varphi})$$ extends to a tame meromorphic family of continuous maps. \end{enumerate} \end{lem} {\it Proof.$\:\:\:\:$} \subsubsection{} In order to see 1. we employ Lemma \ref{compat2} 2., and the identities $$i_{\sigma,\mu}^{U_P}\circ T_P=T_P\circ i_{\sigma,\mu}^\Gamma\ ,\quad p_{\sigma,\mu}^{U_P}\circ T_P=T_P\circ p_{\sigma,\mu}^\Gamma\ .$$ \subsubsection{} In order to see 2. we use Lemma \ref{compat2}, 3., and the fact that $[\pi^{1_{\lambda+\mu}\otimes \pi_{\sigma,\mu},{\mathrm {id}}_{\varphi}}(A),\chi_p]$ is a differential operator with compactly supported coefficients on $B_\Gamma$ for any $A\in{\mathcal{Z}}({\mathfrak{g}})$. \subsubsection{} We show the first assertion of 3. and leave the second to the reader, since the argument is similar. Since $i^\Gamma_{\sigma,\mu}$ is ${\mathcal{Z}}({\mathfrak{g}})$-equivariant it maps to $\ker_\Gamma(Z(\lambda))$ by construction of $Z(\lambda)$. Let now $f\in\ker_\Gamma(Z(\lambda))$. We claim that for any $h\in C^\infty(B_\Gamma)$ we have $Z(\lambda)(hf)=0$. Let $k$ be the order of the differential operator $Z(\lambda)$ which is a projection. Then we have \begin{eqnarray*} Z(\lambda)(hf)&=& Z(\lambda)^{k+1}(hf)\\ &=&Z(\lambda)^{k}[Z(\lambda),h]f\\ &=&Z(\lambda)^{k-1}[Z(\lambda),[Z(\lambda),h]]f\\ &\dots&\\ &=&\underbrace{[Z(\lambda),\dots,[Z(\lambda),h]\dots]}_{k+1}f\\ &=&0\ . \end{eqnarray*} This shows that claim. We conclude that $T_P(f)\in \ker_U(Z(\lambda))$ for all $P\in\tilde P$. By Lemma \ref{compat2} we find $g_P\in B_U(\sigma_\lambda,{\varphi})$, $P\in\tilde{\mathcal{P}}$, such that $T_P(f)=i_{\sigma,\mu}^{U_P}(g_P)$. Let $$f_0:=f-\sum_{P\in\tilde {\mathcal{P}}} i_{\sigma,\mu}^\Gamma\circ T^P (g_P)\in C_c^\infty(B_\Gamma,V_{B_\Gamma}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))\ .$$ As in the proof of Lemma \ref{compat1},4. we can find $g_0\in C_c^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$ such that $f_0=i^\Gamma_{\sigma,\mu}(g_0)$. Thus $f= i^\Gamma_{\sigma,\mu}\left(\sum_{P\in\tilde {\mathcal{P}}\cup\{0\}} g_P\right)$. \subsubsection{}\label{neuj3006} We show the first assertion of 4. and leave the second to the reader. As in the proof of Lemma \ref{compat1}, 5. we construct a holomorphic family $$J:C_c^\infty(B_\Gamma,V_{B_\Gamma}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})) \rightarrow C_c^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$$ as the composition $J:=j\circ (1-Z(\lambda))$, where $$j:C_c^\infty(B_\Gamma,V_{B_\Gamma}(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})) \rightarrow C_c^\infty(B_\Gamma,V_{B_\Gamma}(\sigma_\lambda,{\varphi}))$$ comes from a bundle homomorphism. Then we define $$\tilde j^\Gamma_{\sigma,\mu}:= J \chi_0 (1-Z(\lambda))+ \sum_{P\in\tilde {\mathcal{P}}} \tilde T^P\circ j^{U_P}_{\sigma,\mu} \circ T_P\circ (1-Z(\lambda))\ .$$ Here $\tilde T^P$ is defined as $T^P$ but using a cut-off function $\tilde\chi_p\in C^\infty(B_\Gamma)$, $P\in p$, satisfying ${\mathrm{supp}}(\tilde\chi_p)\in B_p$ and $\tilde\chi_p\chi_p=\chi_p$ if $\tilde{\mathcal{P}}^<$. If $P\in\tilde{\mathcal{P}}^{max}$, then we set $\tilde T^P:=T^P$. One checks that $\tilde j^\Gamma_{\sigma,\mu}\circ i^\Gamma_{\sigma,\mu}={\mathrm {id}}$. Since $\tilde j^\Gamma_{\sigma,\mu}$ vanishes on $\ker(1-Z(\lambda))$ we conclude that it coincides with $j^\Gamma_{\sigma,\mu}$ for $\lambda\not\in I_{\mathfrak{a}}$. We thus have shown that $j^\Gamma_{\sigma,\mu}$ extends to a meromorphic family of continuous maps. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsection{Push-down} \subsubsection{} For the first part of the present subsection we choose an Iwasawa decomposition $G=KAN$ and a parabolic subgroup $P=MAN$, $M\subset K$. Then we write $X=G/K$ and $\partial X=G/P=K/M$. \subsubsection{} For $f\in C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$ we consider the push-down $$\pi^\Gamma_*(f)\in C^\infty(B_\Gamma,B_{B_\Gamma}(\sigma_\lambda,{\varphi}))\oplus \bigoplus_{P\in\tilde{\mathcal{P}}^{max}} {R}_{U_P}(\sigma_\lambda,{\varphi})\ .$$ It is given by the sum $$(\pi^\Gamma_*(f))_1:=\sum_{g\in\Gamma}\pi^{\sigma_\lambda,{\varphi}}(g) (f_{|\Omega_\Gamma})\in{}^\Gamma C^\infty(\Omega_\Gamma,V(\sigma_\lambda,{\varphi}))$$ in the first component. Its second component $$(\pi^\Gamma_*(f))_2\in \bigoplus_{P\in\tilde{\mathcal{P}}^{max}} {R}_{U_P}(\sigma_\lambda,{\varphi})$$ will be constructed below. The goal of the present section is to show that the push-down converges for ${\rm Re }(\lambda)<-\delta_\Gamma-\delta_{\varphi}$ and defines a holomorphic family of maps $$\pi^\Gamma_*:C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))\rightarrow B_{\Gamma}(\sigma_\lambda,{\varphi})\ .$$ \subsubsection{} In the following lemma we consider points of $\partial X$ as subsets of $K$ using the identification $\partial X\cong K/M$. \begin{lem}\label{ggll} If $W\subset\Omega_\Gamma$ is compact, then $\Gamma\cap WMA_+K$ is finite. \end{lem} {\it Proof.$\:\:\:\:$} The set $ WMA_+K$ is a precompact subset of $X\cup \Omega_\Gamma$. Since $\Gamma$ acts properly discontinuously on $X\cup \Omega_\Gamma$ the intersection of the orbit $\Gamma K$ with $WMA_+K$ is finite. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} \begin{lem}\label{barpi} For ${\rm Re }(\lambda)<-\delta_\Gamma-\delta_{\varphi}$ the sum $$\sum_{g\in\Gamma}\pi^{\sigma_\lambda,{\varphi}}(g)f_{|\Omega_\Gamma}$$ converges in $C^\infty(\Omega_\Gamma,V(\sigma_\lambda,{\varphi}))$ and defines a holomorphic family of continuous maps $$C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))\to {}^\Gamma C^\infty(\Omega_\Gamma,V(\sigma_\lambda,{\varphi}))\ .$$ \end{lem} {\it Proof.$\:\:\:\:$} Using Lemma \ref{ggll} we employ the same argument as in the proof of \cite{MR1749869}, Lemma 4.2. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Next we study the behaviour of the sum $\sum_{g\in\Gamma}\pi^{\sigma_\lambda,{\varphi}}(g)f_{|\Omega_\Gamma}$ near the cusps. Let $P\in\tilde{\mathcal{P}}$. We choose a system $\Gamma^P$ of representatives of $U_P\backslash \Gamma$ by taking in each class $[g]\in U_P\backslash \Gamma$ an element $h\in [g]$ which minimizes ${\rm dist}_X({\mathcal{O}},h{\mathcal{O}})$. \begin{lem}\label{wwe} There is a neighbourhood $W\subset\partial X$ of $\infty_P$ such that $WMA_+K\cap \Gamma^P$ is finite. \end{lem} {\it Proof.$\:\:\:\:$} Let $D({\mathcal{O}},U_P):=\{x\in X|{\rm dist}_X(x,{\mathcal{O}})\le {\rm dist}_X(gx,{\mathcal{O}})\:\forall g\in U_P\}$ be the Dirichlet domain of $U_P$. If $h\in \Gamma^P$, then $h{\mathcal{O}}\in D({\mathcal{O}},U_P)$. Let $\pi:D({\mathcal{O}},U_P)\rightarrow Y_{U_P}$ denote the projection and consider $E:=\pi^{-1}(e_P(Y_p))$. Since $e_P:Y_p\rightarrow Y_{U_P}$ is an isometry we have $\sharp(E\cap \Gamma^P{\mathcal{O}})\le 1$. Now the closure in $\bar X$ of $D({\mathcal{O}},U_P)\setminus E$ does not contain $\infty_P$ and is therefore disjoint from a neighbourhood $W\subset \partial X$ of $\infty_P$. Since $WMA_+K\cap (D({\mathcal{O}},U_P)\setminus E)$ has a compact closure inside $X$ we conclude that $WMA_+K\cap (D({\mathcal{O}},U_P)\setminus E)K\cap \Gamma^P$ is finite. Since $\Gamma^P\subset D({\mathcal{O}},U_P)K$ we have shown the lemma.\hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Let $P\in \tilde {\mathcal{P}}$ and $W_P$ be a neighbourhood of $\infty_P$ as constructed in Lemma \ref{wwe}. Using Lemma \ref{wwe} and the arguments of the proof of \cite{MR1749869}, Lemma 4.2, we show \begin{lem}\label{bbarpi} For ${\rm Re }(\lambda) < -\delta_\Gamma-\delta_{\varphi}$ and $f\in C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$ the sum $$\bar\pi^P_*(f):=\sum_{g\in\Gamma^P}(\pi^{\sigma_\lambda,{\varphi}}(g)f)_{|W_P\cup\Omega_\Gamma}$$ converges in $C^\infty(W_P\cup\Omega_\Gamma,V(\sigma_\lambda,{\varphi}))$. For varying $\lambda$ the maps $\bar\pi^P_*$ form a holomorphic family of continuous maps. \end{lem} \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} We choose a cut-off function $\kappa_P\in C^\infty_c(W_P)$ such that $\kappa_P$ is equal to one near $\infty_P$. Then $\kappa_P \bar\pi^P_*(f))\in C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$, and we can write $$(\pi^\Gamma_*(f))_1=\left((\pi^{U_P}_*(\kappa_P \bar\pi^P_*(f)))_1 + \sum_{u\in U_P} \pi^{\sigma_\lambda,{\varphi}}(u)(1-\kappa_P) \bar\pi^P_*(f)\right)_{|\Omega_\Gamma}\ .$$ Here $$(\pi^{U_P}_*(\kappa_P \bar\pi^P_*(f)))_1=\pi^{U_P}_*(\kappa_P \bar\pi^P_*(f))$$ if the cusp associated to $U_P\subset P$ has smaller rank. The second sum is locally finite and defines a smooth section on $\Omega_\Gamma$. Moreover we have ($\bar \pi^P_*$ was defined in Lemma \ref{bbarpi}) $$\chi_P \sum_{u\in U_P} \pi^{\sigma_\lambda,{\varphi}}(u)(1-\kappa_P) \bar\pi^P_*(f)_{|\Omega_\Gamma}\in C^\infty_c(B_{U_P},V_{B_{U_P}}(\sigma_\lambda,{\varphi}))\ .$$ \subsubsection{} We define $$(\pi^\Gamma_*(f))_2:=\bigoplus_{P\in\tilde{\mathcal{P}}^{max}} [\pi^{U_P}_*]\circ AS (\kappa_P \bar\pi^P_*(f))\ .$$ This definition is independent of the choice of $\kappa_P$. Finally we define $\pi^\Gamma_*(f)$ to be the sum $(\pi^\Gamma_*(f))_1\oplus (\pi^\Gamma_*(f))_2$. \subsubsection{} \begin{lem} For ${\rm Re }(\lambda) < -\delta_\Gamma-\delta_{\varphi}$ the push-down $\pi^\Gamma_*$ induces a holomorphic family of continuous maps $$\pi^\Gamma_*:C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))\rightarrow B_\Gamma(\sigma_\lambda,{\varphi})\ .$$ \end{lem} {\it Proof.$\:\:\:\:$} This follows from Lemmas \ref{barpi}, \ref{bbarpi}, the observation that \begin{equation}\label{xxxt}T_P(\pi^\Gamma_*(f))_1\in \chi_P \pi^{U_P}_*(\kappa_P \bar\pi^P_*(f)) + C_c^\infty(B_{U_P},V_{B_{U_P}}(\sigma_\lambda,{\varphi}))\subset B_{U_P}(\sigma_\lambda,{\varphi})\end{equation} for $P\in \tilde{\mathcal{P}}^<$, and the definition of $(\pi^\Gamma_*(f))_2$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{}\label{neuj1003} Let $\chi^\Gamma\in C^\infty(\Omega_\Gamma)$ be a smooth cut-off function such that $\sum_{g\in\Gamma} g^*\chi^\Gamma \equiv 1$. We choose $\chi^\Gamma$ such that it coincides with $\chi^{U_P}$ in a neighbourhood of $\infty_P$, and such that the restriction of the projection $\Omega_\Gamma\rightarrow B_\Gamma$ to ${\mathrm{supp}}(\chi^\Gamma)$ is proper. Then multiplication by $\chi^\Gamma$ defines a holomorphic family of right-inverses of the push-down (see Lemma \ref{schwspl}) $$\{Q\}:{S}_\Gamma(\sigma_\lambda,{\varphi})\rightarrow C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))$$ such that $\pi^{\Gamma}_*\circ \{Q\}={\mathrm {id}}$. \subsubsection{} Occasionally we will need the partial push-down $$\pi^{\Gamma/U_P}_*:B_{U_P,k}(\sigma_\lambda,{\varphi})\rightarrow B_{\Gamma,k}(\sigma_\lambda,{\varphi})$$ which is defined as the average over $\Gamma/U_P$ if the cusp associated to $U_P\subset P$ has smaller rank. If the cusp associated to $U_P\subset P$ has full rank, then we define $$\pi^{\Gamma/U_P}_*(f\oplus v):=\pi^\Gamma_*(\{Q\}(f-(\pi^{U_P}_*\circ L\circ [Q](v))_1)+L \circ [Q](v))\ ,$$ where $$f\oplus v\in C^\infty(B_{U_P},V_{B_{U_P}}(\sigma_\lambda,{\varphi}))\oplus {R}_{U_P}(\sigma_\lambda,{\varphi})=B_{U_P}(\sigma_\lambda,{\varphi})\ .$$ If $P\in \tilde{\mathcal{P}}^<$, then we have $$\pi^{\Gamma/U_P}_*(f):=\pi^\Gamma_*(\{Q\}(f-\pi^{U_P}_*\circ L\circ [Q]\circ AS(f))+L \circ [Q] \circ AS (f))\ .$$ This formula together with (\ref{xxxt}) can be used to verify that $\pi^{\Gamma/U_P}_*$ has the required mapping properties. In particular we conclude that if $P\in\tilde{\mathcal{P}}^<$, then the partial push-down converges for ${\rm Re }(\lambda)<-\delta_\Gamma-\delta_{\varphi}$ and depends holomorphically on $\lambda$. If $P\in\tilde{\mathcal{P}}^{max}$, then the partial push-down is defined as a meromorphic family of continuous maps for ${\rm Re }(\lambda)<-\delta_\Gamma-\delta_{\varphi}$. \subsubsection{} We collect the following useful identities \begin{equation}\label{useid}\pi^{\Gamma/U_P}_*\circ \pi^{U_P}_*=\pi^\Gamma_*,\quad \pi^{\Gamma/U_P}_* \circ T_P = \chi_p,\quad \pi^{\Gamma/U_P}_* \chi_P = T^P\ ,\end{equation} where $P\in \tilde{\mathcal{P}}^<$ for the last two equations and multiplication by $\chi_p$ implicitly involves a projection onto the first component. If $P\in \tilde{\mathcal{P}}^{max}$, then we have \begin{equation}\label{useid2} \pi^{\Gamma/U_P}_* \circ AS_P =AS_P \ .\end{equation} \subsubsection{} We define $q_{\tilde\sigma,\mu}:=q^{\{1\}}_{\tilde\sigma,\mu}$ as in Lemma \ref{compat3}, 4. (applied to the trivial group). \begin{lem}\label{compat5} We have the following commutative diagrams of meromorphic families of maps: \begin{enumerate} \item $$\begin{array}{ccccc} C^\infty(\partial X,V(\sigma_\lambda,{\varphi}))&\stackrel{i_{\sigma,\mu}}{\rightarrow}&C^\infty(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))&\stackrel{Z(\lambda)}{\rightarrow}& C^\infty(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))\\ \downarrow \pi^\Gamma_*&&\downarrow \pi^\Gamma_*&&\downarrow \pi^\Gamma_*\\ B_\Gamma(\sigma_\lambda,{\varphi})&\stackrel{i^\Gamma_{\sigma,\mu}}{\rightarrow}& {\mathcal{B}}_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})&\stackrel{Z(\lambda)}{\rightarrow}& {\mathcal{B}}_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})\end{array}\ .$$ \item $$\begin{array}{ccc} C^\infty(\partial X,V(\tilde\sigma_{-\lambda},\tilde{\varphi})) &\stackrel{q_{\tilde\sigma,\mu}}{\rightarrow}& C^\infty(\partial X,V(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes\tilde{\varphi}))\\ \downarrow \pi^\Gamma_*&&\downarrow \pi^\Gamma_*\\ B_\Gamma(\tilde\sigma_{-\lambda},\tilde{\varphi}) &\stackrel{p^\Gamma_{\tilde \sigma,\mu}}{\leftarrow}& B_\Gamma(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes\tilde{\varphi})\\ \end{array}\ .$$ \end{enumerate} \end{lem} {\it Proof.$\:\:\:\:$} The push-down is ${\mathcal{Z}}({\mathfrak{g}})$-equivariant. This implies commutativity of the right square of the diagram 1. The commutativity of the left square is obvious for the part involving $(\pi^\Gamma_*)_1$, and for $(\pi^\Gamma_*)_2$ we invoke Definition \ref{cv1}. For the second diagram consider the enlarged diagram $$\xymatrix{C^\infty(\partial X,V(1_{-\lambda+\mu^\prime},\pi_{\tilde\sigma,\mu^\prime}\otimes\tilde{\varphi})) \ar[d]^{\pi^\Gamma_*}&C^\infty(\partial X,V(\tilde\sigma_{-\lambda},\tilde{\varphi}))\ar@/_-0.5cm/[r]^{q_{\tilde\sigma,\mu}} \ar[l]^{\qquad\quad i_{\tilde \sigma,\mu^\prime}} \ar[d]^{\pi^\Gamma_*}&C^\infty(\partial X,V(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes\tilde{\varphi})) \ar[d]^{\pi^\Gamma_*}\ar[l]^{p_{\tilde \sigma,\mu}\quad}\\ B_\Gamma(1_{-\lambda+\mu^\prime},\pi_{\tilde\sigma,\mu^\prime}\otimes\tilde{\varphi})& B_\Gamma(\tilde\sigma_{-\lambda},\tilde{\varphi}) \ar[l]^{\qquad i^\Gamma_{\tilde \sigma,\mu^\prime}}& B_\Gamma(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes\tilde{\varphi}) \ar[l] ^{p^\Gamma_{\tilde \sigma,\mu}}}\ .$$ The left square commutes by the Definition \ref{cv1} of the push-down in the middle. The outer rectangle commutes by the naturality of the definition of the push-down. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} Let $G^n$ be one of $\{Spin(1,n),SO(1,n)_0,SU(1,n),Sp(1,n)\}$. Recall the definition of $i^*_\Gamma$ given in Subsection \ref{embedd}. \begin{lem}\label{comppp} We have the following commutative diagram: $$\begin{array}{ccc} C^\infty(\partial X^{n+m},V(1^{n+1}_{\lambda },{\varphi})) &\stackrel{i^*}{\rightarrow}&C^\infty(\partial X^n, V(1^n_{\lambda-\zeta},{\varphi})) \\ \downarrow \pi^{\Gamma,n+1}_*&&\downarrow \pi^{\Gamma,n}_* \\ B_\Gamma(1^{n+1}_{\lambda },{\varphi})&\stackrel{i_\Gamma^*}{\rightarrow }& B_\Gamma(1^n_{\lambda-\zeta},{\varphi}) \end{array}\ .$$ \end{lem} {\it Proof.$\:\:\:\:$} We use Proposition \ref{klopp}. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsection{Extension, restriction, and the scattering matrix} \subsubsection{} We now define the distribution spaces $D_{\Gamma}(\sigma_\lambda,{\varphi})$ associated to $\Gamma$. These spaces are the domain of the extension $ext^\Gamma$ and the Eisenstein series (see Definition \ref{neuj1001}). \begin{ddd} We define $D_{\Gamma,k}(\sigma_\lambda,{\varphi})$ to be the dual space to $B_{\Gamma,k}(\tilde\sigma_{-\lambda},\tilde{\varphi})$. Furthermore let $D_{\Gamma}(\sigma_\lambda,{\varphi}):=\bigcup_{k\in{\mathbb {N}}_0} D_{\Gamma,k}(\sigma_\lambda,{\varphi})$. \end{ddd} Lemma \ref{spst} has the following consequence: \begin{kor} The spaces $\{D_{\Gamma,k}(\sigma_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$, form a trivial holomorphic bundle of dual Fr\'echet spaces and fit into the split exact sequence $$0\rightarrow {R}_{\Gamma,k}(\tilde\sigma_{-\lambda},\tilde{\varphi})^*\stackrel{AS^*}{\rightarrow} D_{\Gamma,k}(\sigma_\lambda,{\varphi})\rightarrow {S}_{\Gamma,k}(\tilde\sigma_{-\lambda},\tilde{\varphi})^*\rightarrow 0\ .$$ The spaces $\{D_{\Gamma}(\sigma_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$, are dual Fr\'echet and Montel spaces, and they form a direct limit of trivial bundles. Furthermore we have the exact sequence $$0\rightarrow {R}_{\Gamma}(\tilde\sigma_{-\lambda},\tilde{\varphi})^*\stackrel{AS^*}{\rightarrow} D_{\Gamma}(\sigma_\lambda,{\varphi})\rightarrow {S}_{\Gamma}(\tilde\sigma_{-\lambda},\tilde{\varphi})^*\rightarrow 0\ .$$ \end{kor} Let $(\pi_{\sigma,\mu},V_{\pi_{\sigma,\mu}})$ be a finite-dimensional representation of $G$ as in Subsection \ref{ttw}. \subsubsection{} The following Lemma states that the distribution spaces are compatible with twisting. \begin{lem}\label{compat6} \begin{enumerate} \item $D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})$ is a ${\mathcal{Z}}({\mathfrak{g}})$-module. \item There is a natural inclusion $i^\Gamma_{\sigma,\mu}:D_\Gamma(\sigma_{\lambda},{\varphi}) \rightarrow D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})$ which identifies $D_\Gamma(\sigma_{\lambda},{\varphi})$ with $$\ker_\Gamma(Z(\lambda)):=\ker(Z(\lambda): D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}) \rightarrow D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))$$ for $\lambda\not\in I_{\mathfrak{a}}$. \item Moreover, the map $$j_{\sigma,\mu}^\Gamma:D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}) \stackrel{1-Z(\lambda)}{\longrightarrow} \ker_\Gamma(Z(\lambda))\stackrel{(i_{\sigma,\mu}^\Gamma)^{-1}}{\longrightarrow} D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})$$ extends to a tame meromorphic family of continuous maps. \end{enumerate} \end{lem} {\it Proof.$\:\:\:\:$} We employ Lemma \ref{compat3}. The map $i^\Gamma_{\sigma,\mu}$ is defined as the adjoint of $$p^\Gamma_{\tilde\sigma,\mu}:B_\Gamma(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes\tilde{\varphi})\rightarrow B_\Gamma(\tilde\sigma_{-\lambda},\tilde{\varphi})\ .$$ $j_{\sigma,\mu}^\Gamma$ is just the adjoint of $$q^\Gamma_{\tilde\sigma,\mu}:B_\Gamma(\tilde\sigma_{-\lambda},\tilde{\varphi})\rightarrow B_\Gamma(1_{-\lambda-\mu},\pi_{\tilde\sigma,\mu}\otimes\tilde{\varphi})\ .$$ \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} \begin{ddd}For ${\rm Re }(\lambda)>\delta_\Gamma+\delta_{\varphi}$ we define the extension map $$ext^\Gamma:D_{\Gamma}(\sigma_\lambda,{\varphi})\rightarrow C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$$ to be the adjoint of the push-down $$\pi^\Gamma_*:C^\infty(\partial X,V(\tilde\sigma_{-\lambda},\tilde{\varphi}))\rightarrow B_{\Gamma}(\tilde\sigma_{-\lambda},\tilde{\varphi})\ .$$ \end{ddd} The extension $ext^\Gamma$ is a holomorphic family of continuous maps and has values in $\Gamma$-invariant distributions. We will also need the partial extensions $ext^{\Gamma/U_P}:D_{\Gamma,k}(\sigma_\lambda,{\varphi})\rightarrow D_{U_P,k}(\sigma_\lambda,{\varphi})$, which are defined as the adjoints of $\pi^{\Gamma/U_P}_*$. Taking the adjoint of the relations (\ref{useid}) and (\ref{useid2}) we obtain $$ext^{U_P}\circ ext^{\Gamma/U_P}=ext^\Gamma,\quad T_P^*\circ ext^{\Gamma/U_P}=\chi_p,\quad \chi_P ext^{\Gamma/U_P} = (T^P)^*$$ (where $P\in\tilde{\mathcal{P}}^<$ for the last two, and multiplication by $\chi_p$ implicitly involves projection onto the first component), and $$AS_P^*\circ ext^{\Gamma/U_P}=AS_P^*\ ,$$ if $P\in\tilde{\mathcal{P}}^{max}$. \subsubsection{} The space $D_{\{1\},k}(\sigma_\lambda,{\varphi})$ defined in Def.~\ref{neuj1002} depends on the choice of a parabolic subgroup $P$. In the present subsection we write $D_{P,k}(\sigma_\lambda,{\varphi})$ for that space and let $$D_{\{1\},k}(\sigma_\lambda,{\varphi}):=\bigcap_{p\in {\mathcal{P}}_\Gamma,P\in p} D_{P,k}(\sigma_\lambda,{\varphi})\ .$$ Using the relation $ext^{U_P}\circ ext^{\Gamma/U_P}=ext^\Gamma$ we see that $$ext^\Gamma: D_{\Gamma,k}(\sigma_\lambda,{\varphi}) \rightarrow D_{\{1\},k}(\sigma_\lambda,{\varphi})\ ,$$ and that it is holomorphic as a map $ext^\Gamma: D_{\Gamma,k}(\sigma_\lambda,{\varphi}) \rightarrow D_{\{1\},k_1}(\sigma_\lambda,{\varphi})$ for any $k_1>k$ (and, of course, for ${\rm Re }(\lambda)>\delta_\Gamma+\delta_{\varphi}$). \subsubsection{}\label{neuj3002} From now on we consider the spherical case $\sigma=1$. We introduce the restriction maps $$res^\Gamma: D_{\{1\},k}(1_\lambda,{\varphi})\rightarrow D_{\Gamma,k_1}(1_\lambda,{\varphi})$$ defined for all $k$ and suitable $k_1\ge k$ depending on $k$. The map $res^\Gamma$ is a refinement of the naive restriction $$\{res^\Gamma\}: C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))\rightarrow {S}_\Gamma(1_{-\lambda},\tilde{\varphi})^*$$ which is given as the adjoint of $\{Q\}$ (see \ref{neuj1003}). We define the meromorphic family of maps $res^\Gamma$ by \begin{equation}\label{rreessdef} res^\Gamma(f):=\pi^\Gamma_*(\chi^\Gamma\chi_0 f) + \sum_{P\in\tilde{\mathcal{P}}} T_P^* \circ res^{U_P}(f)\ ,\end{equation} where we interprete $T_P^*$ as $AS_P^*$, if $P\in\tilde{\mathcal{P}}^{max}$. Here $\pi^\Gamma_*$ simply stands for the average over $\Gamma$. It is well-defined and holomorphic for all $\lambda\in{\aaaa_\C^\ast}$ since $\chi^\Gamma\chi_0 f\in C_c^{-\infty}(\Omega_\Gamma,V(1_\lambda,{\varphi}))$ and thus $\pi^\Gamma_*(\chi^\Gamma\chi_0 f)\in C_c^{-\infty}(B_\Gamma,V_{B_\Gamma}(1_\lambda,{\varphi}))$. Alternatively, one could write $\chi_0\{res^\Gamma\}(f)$ for $\pi^\Gamma_*(\chi^\Gamma\chi_0 f)$. Like the maps $res^{U_P}$ (see Definition \ref{neuj707}) the restriction map $res^\Gamma$ depends on choices. \subsubsection{} A priori the composition $res^\Gamma\circ ext^\Gamma$ is meromorphic and depends on many choices. The following lemma shows that the situation is much better. It generalizes \ref{neuj2001}. \begin{lem}\label{weih114} The composition $res^\Gamma\circ ext^\Gamma$ is regular and coincides with the inclusion of $$D_{\Gamma,k}(1_\lambda,{\varphi})\to D_{\Gamma,k_1}(1_\lambda,{\varphi})\ .$$ \end{lem} {\it Proof.$\:\:\:\:$} Assume that $\lambda\in {\aaaa_\C^\ast}$ is such that $res^\Gamma$ is regular. Note that we still assume ${\rm Re }(\lambda)>\delta_\Gamma+\delta_{\varphi}$ in order ensure convergence of $ext^\Gamma$. Let $f\in D_{\Gamma,k}(1_\lambda,{\varphi})$. It is easy to see that $\pi^\Gamma_*(\chi^\Gamma\chi_0 ext^\Gamma(f))=\chi_0 f$. Furthermore, if $P\in\tilde{\mathcal{P}}^<$, then \begin{eqnarray*} T_P^* \circ res^{U_P} \circ ext^\Gamma(f) &=&T_P^* \circ res^{U_P} \circ ext^{U_P}\circ ext^{\Gamma/U_P}(f)\\ &=&T_P^* \circ ext^{\Gamma/U_P}(f)\\ &=&\chi_p f\ . \end{eqnarray*} If $P\in\tilde{\mathcal{P}}^{max}$, then we have \begin{eqnarray*} AS_P^* \circ res^{U_P} \circ ext^\Gamma(f) &=&AS_P^* \circ res^{U_P} \circ ext^{U_P}\circ ext^{\Gamma/U_P}(f)\\ &=&AS_P^* \circ ext^{\Gamma/U_P}(f)\\ &=&AS_P (f)\ . \end{eqnarray*} Summing these equations over $\tilde{\mathcal{P}}\cup\{0\}$, then we obtain the desired identity $$res^\Gamma\circ ext^\Gamma(f)=f\ .$$ Since this equation holds true for generic $\lambda$, the assertion of the lemma follows. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} We also have the following useful identity $$T_P^*\circ res^{U_P}= \chi_p res^\Gamma$$ for all $P\in p \in {\mathcal{P}}^<_\Gamma$. \subsubsection{}\label{weih113} Consider $P\in \tilde{\mathcal{P}}$ and recall the Definition \ref{neuj108} of $Ext_{U_P}(1_\lambda,{\varphi})$. If $\lambda$ is admissible (for $P$) in the sense of \ref{neuj300}, then on this space the restriction $res^{U_P}$ is well-defined independent of choices. We now say that $\lambda\in {\aaaa_\C^\ast}$ is admissible, if it is admissible for all $P\in \tilde{\mathcal{P}}$. Furthermore we define $${}^\Gamma C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))^0:= {}^\Gamma C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))\cap\bigcap_{P\in \tilde{\mathcal{P}}} Ext_{U_P}(1_\lambda,{\varphi})\ .$$ It immediately follows from (\ref{rreessdef}) that $res^\Gamma$ is well-defined on this space independent of choices. \subsubsection{}\label{neuj3001} If we assume that $\lambda\not\in I_{\mathfrak{a}}$ or $\lambda>-\rho$, then $${}^\Gamma C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))^0={}^\Gamma C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))$$ by Prop.~\ref{maertins}, 2. Thus $res^\Gamma$ is pointwise well-defined at admissible $\lambda$ satisfying these additional conditions. \subsubsection{} \begin{ddd} For ${\rm Re }(\lambda)>\delta_\Gamma+\delta_{\varphi}$ we define the scattering matrix $$S_\lambda^\Gamma:D_\Gamma(1_\lambda,{\varphi})\rightarrow D_\Gamma(1_{-\lambda},{\varphi})$$ as a meromorphic family of operators, which is given by $$S_\lambda^\Gamma:=res^\Gamma \circ J_\lambda\circ ext^\Gamma$$ provided $-\lambda$ is non-integral and admissible in the sense of \ref{weih113}. \end{ddd} Using the scattering matrices for the cusps (Definition \ref{neuj2003}) we can rewrite the formula for $S^\Gamma_\lambda$ as follows: \begin{equation}\label{spltt} S_\lambda^\Gamma =\pi^\Gamma_* \circ\chi^\Gamma\chi_0 J_\lambda \circ ext^\Gamma + \sum_{P\in \tilde {\mathcal{P}}} T^*_P\circ S^{U_P}_\lambda \circ ext^{\Gamma/U_P}\ .\end{equation} Using this formula and the results of \ref{neuj2004} we see that $S_\lambda^\Gamma$ is a well-defined meromorphic family of continuous maps. \subsubsection{}\label{gaga} As in the case of pure cusps (Lemma \ref{fundd}) we obtain a natural inclusion $$B_\Gamma(1_\lambda,{\varphi})\hookrightarrow D_\Gamma(1_\lambda,{\varphi})$$ provided that $2\rho^{U_P}>l_{{\varphi}_{|U_P}}$ for all $P\in \tilde {\mathcal{P}}^<$. Note that $l_{{\varphi}_{|U_P}}=2\delta_{{\varphi}_{|U_P}}$. \begin{lem}\label{restrf} We assume that all cusps of $\Gamma$ have smaller rank. Assume that $\delta_{{\varphi}_{|U_P}}<\rho^{U_P}$ for all $P\in\tilde{\mathcal{P}}$. Then the restriction of the scattering matrix to $B_\Gamma(1_\lambda,{\varphi})$ induces a meromorphic family of continuous maps $$( S_\lambda^\Gamma)_{|B_\Gamma(1_\lambda,{\varphi})}:B_\Gamma(1_\lambda,{\varphi})\rightarrow B_\Gamma(1_{-\lambda},{\varphi})\ .$$ \end{lem} {\it Proof.$\:\:\:\:$} First note that if $f\in B_\Gamma(1_\lambda,{\varphi})$ (considered as distribution), then $ext^\Gamma(f)$ is smooth on $\Omega_\Gamma$. It follows that $J_\lambda\circ ext^\Gamma(f)$ is smooth on $\Omega_\Gamma$, and hence $S^\Gamma_\lambda(f)=res^\Gamma\circ J_\lambda\circ ext^\Gamma(f)$ is represented by a smooth function. We choose a cut-off function $\tilde\chi_P$ on $B_{U_P}$ which is supported in $e_{P}(B_p)$ such that $\tilde\chi_P\chi_P=\chi_P$. In order to see that $S^\Gamma_\lambda(f)\in B_\Gamma(1_{-\lambda},{\varphi})$ we employ Equation (\ref{spltt}). Let $P\in\tilde{\mathcal{P}}$. Then we have \begin{eqnarray} T_P\circ S^\Gamma_\lambda(f)&=& T_P(\chi^\Gamma \chi_0 J_\lambda\circ ext^\Gamma(f))\\ &&+T_P\circ T_P^*\circ S^{U_P}_\lambda\circ ext^{\Gamma/U_P}(f)\nonumber\\ &=&T_P(\chi^\Gamma \chi_0 J_\lambda\circ ext^\Gamma(f))\label{ppo0}\\ &&+\chi_P^2 S_\lambda^{U_P}\circ \tilde\chi_P ext^{\Gamma/U_P}(f)\label{ppo1}\\ &&+\chi_P^2 S_\lambda^{U_P}\circ(1- \tilde\chi_P) ext^{\Gamma/U_P}(f)\label{ppo2}\ . \end{eqnarray} The term (\ref{ppo0}) belongs to $C_c^\infty(B_\Gamma,V_{B_\Gamma}(1_\lambda,{\varphi}))$. We have $$\tilde\chi_P ext^{\Gamma/U_P}(f)=(\tilde T^P)^*(f)=\tilde T_P(f)\in B_{U_P}(1_{-\lambda},{\varphi})\ ,$$ where $\tilde T_P, \tilde T^P$ are defined as $T_P,T^P$, but using $\tilde\chi_P$ instead of $\chi_P$. We can now apply Lemma \ref{inside} to (\ref{ppo1}) and Lemma \ref{offside} to (\ref{ppo2}) in order to see that these terms belong to $B_\Gamma(1_{-\lambda},{\varphi})$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsection{Vanishing results} \subsubsection{} If $\lambda\in{\aaaa_\C^\ast}$ is admissible in the sense of \ref{weih113} and ${\rm Re }(\lambda)>-\rho$, then $res^\Gamma$ is well-defined on ${}^\Gamma C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))$ (see \ref{neuj3001}). \subsubsection{} Let $f\in {}^\Gamma C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))$. The condition $\{res^\Gamma\}(f)=0$ (see \ref{neuj3002}) is equivalent to the condition that the support of $f$ as a distribution is contained in the limit set $\Lambda_\Gamma$. In the case that $\Gamma$ is convex cocompact we know from \cite{MR1749869}, Thm.~4.7 that the space $$\{f\in {}^\Gamma C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))|\{res^\Gamma\}(f)=0\}$$ is trivial if ${\rm Re }(\lambda)>\delta_\Gamma+\delta_{\varphi}$. Already the example of pure cusps shows that this is not true in the presence of cusps. Note that in the convex-cocompact case $\{res^\Gamma\}=res^\Gamma$. It is at the heart of the matter that for geometrically finite groups $\Gamma$ and ${\rm Re }(\lambda)$ large the stronger condition $res^\Gamma(f)=0$ implies $f=0$. In \cite{MR1926489} we introduced a related condition ``$f$ is strongly supported on the limit set'' in order to show such vanishing results. \subsubsection{} \begin{prop}\label{vani} We assume that $\lambda$ is admissible in the sense of \ref{weih113} and satisfies \begin{equation}\label{neuj3004} {\rm Re }(\lambda)>\max\left(\{\delta_\Gamma\}\cup\{\rho_{U_P}\mid P\in\tilde{\mathcal{P}}\}\right)+\delta_{\varphi} \ . \end{equation} Let $f\in {}^\Gamma C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))$. If $res^\Gamma(f)=0$, then $f=0$. \end{prop} {\it Proof.$\:\:\:\:$} \subsubsection{} Let $\lambda$ be admissible and $f\in {}^\Gamma C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))$. From the defining formula (\ref{rreessdef}) for $res^\Gamma$ we conclude that $res^\Gamma(f)=0$ if and only if $f$ is supported on $\Lambda_\Gamma$ and $res^{U_p}(f)\in C_c^{-\infty}(B_{U_P},V_{B_{U_P}}(1_\lambda,{\varphi}))\subset D_{U_P}(1_\lambda,{\varphi})$ for all $P\in\tilde{\mathcal{P}}$. The latter condition implies that \begin{equation}\label{wuff} res^{U_P}(f)=\{res^{U_P}\}(f) \end{equation} for all $P\in\tilde{\mathcal{P}}$. \subsubsection{} If ${\rm Re }(\lambda)>-\rho^{U_P}$ and $h\in C^{\infty}(\partial X,V(1_{-\lambda},\tilde{\varphi}))$, then Proposition \ref{maertins}, 2, Lemma \ref{wauwau}, and Equation (\ref{wuff}) imply $$ \langle f,h\rangle =\langle ext^{U_P}\circ res^{U_P}(f),h\rangle =\langle \{res\}^{U_P}(f),\pi_*^{U_P}(h)\rangle $$ and therefore \begin{equation}\label{woff} \langle f,h\rangle =\sum_{u\in U_P} \langle\chi^{U_P}f,\tilde{\varphi}(u)h(u^{-1}.)\rangle\ . \end{equation} Thus, if $res^\Gamma(f)=0$, then $f$ is supported on the limit set and satisfies (\ref{woff}) for all $P\in\tilde{\mathcal{P}}$, $h\in C^{\infty}(\partial X,V(1_{-\lambda},\tilde{\varphi}))$. These are precisely the defining conditions for being ``strongly supported on the limit set'' in the sense of \cite{MR1926489}. The main result of \cite{MR1926489} now states that an element $f\in {}^\Gamma C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))$ that is ``strongly supported on the limit set'' vanishes provided that (\ref{neuj3004}) holds. The proposition follows. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} \begin{kor}\label{wieder} Assume $\lambda\in{\aaaa_\C^\ast}$ satisfies (\ref{neuj3004}). If $f_\mu\in {}^\Gamma C^{-\infty}(\partial X,V(1_\mu,{\varphi}))$ is a germ of a meromorphic family at $\lambda$, then $ext^\Gamma\circ res^\Gamma(f_\mu) = f_\mu$. \end{kor} {\it Proof.$\:\:\:\:$} We apply $res^\Gamma$, and we use $res^\Gamma\circ ext^\Gamma={\mathrm {id}}$ and the injectivity of $res^\Gamma$ for generic $\mu$ near $\lambda$ proved in Proposition \ref{vani}. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent. \subsection{Meromorphic continuations and general $\sigma$} \subsubsection{} Recall that we have defined $ext^\Gamma$ for ${\rm Re }(\lambda)>\delta_\Gamma+\delta_{\varphi}$. The scattering matrix $S^\Gamma_\lambda$ was defined on the same range of $\lambda$ for the trivial $M$-type $1$. In the present subsection we extend the definition of the scattering matrix to general $M$-types. Further we show that the extension and the scattering matrix have meromorphic continuations to all of ${\aaaa_\C^\ast}$. This finishes the proof of theorems \ref{t119}, \ref{t110}, and \ref{t113}. \subsubsection{} We start with the meromorphic continuation of $S^\Gamma_\lambda$ to a certain half-plane $W\subset {\aaaa_\C^\ast}$ in the spherical case. Under these conditions we can construct a meromorphic family of parametrices for $S_\lambda^\Gamma$ with finite-dimensional singularities. In a second step we employ twisting in order to show meromorphy of the extension and the scattering matrix on all of ${\aaaa_\C^\ast}$ and for general $\sigma$. \subsubsection{} \begin{prop}\label{meroext} The extension $$ext^\Gamma:D_\Gamma(1_\lambda,{\varphi})\rightarrow C^{-\infty}(\partial X,V(1_\lambda,{\varphi}))$$ and the scattering matrix $$S_\lambda^\Gamma:D_\Gamma(1_\lambda,{\varphi})\rightarrow D_\Gamma(1_{-\lambda},{\varphi})$$ have meromorphic continuations to $$W:=\{\lambda\in{\aaaa_\C^\ast}\:|\: {\rm Re }(\lambda)>-\rho+\beta\}\ ,$$ where $\beta=0$ for $X={H\mathbb{R}}^n, {H\mathbb{C}}^n$, and $\beta=2\alpha$ for $X={H\mathbb{H}}^n$. The family $ext^\Gamma$ has finite-dimensional singularities. \end{prop} {\it Proof.$\:\:\:\:$} \subsubsection{} We first show the proposition under an additional assumption on $\Gamma$ and ${\varphi}$. \begin{lem}\label{wahn1} We assume that \begin{equation}\label{kuckuck} \delta_\Gamma+\delta_{\varphi}<-\max\left(\{0\}\cup\{\rho_{U_P}+\delta_{\varphi}\mid P\in\tilde{\mathcal{P}}\}\right) \ . \end{equation} Then the extension $ext^\Gamma$ and the scattering matrix $S_\lambda^\Gamma$ have a meromorphic continuation to the half-plane $W$. The family $ext^\Gamma$ has finite-dimensional singularities. \end{lem} {\it Proof.$\:\:\:\:$} \subsubsection{} The additional assumption in particular implies that $\rho^{U_P}>\delta_{\varphi}\ge 0$ for all $P\in\tilde{\mathcal{P}}$. Therefore all cusps of $\Gamma$ have smaller rank, and there is an embedding $$B_\Gamma(1_\lambda,{\varphi})\hookrightarrow D_\Gamma(1_\lambda,{\varphi})$$ (see \ref{gaga}). We consider the non-empty open subset $$U:=\{\lambda\in{\aaaa_\C^\ast}\:|\:\max\{\delta_\Gamma+\delta_{\varphi},-\rho+\beta\}<{\rm Re }(\lambda)<\rho-\beta\}\subset {\aaaa_\C^\ast}\ .$$ Based on Lemma \ref{restrf} we first consider the restriction of the scattering matrix to $B_\Gamma(1_\lambda,{\varphi})$. For $\lambda\in U$ we construct a parametrix (see \ref{neuj3005} for $T_P$ and \ref{neuj3006} for $\tilde T^P$) $$Q_{-\lambda}:B_\Gamma(1_{-\lambda},{\varphi})\rightarrow B_\Gamma(1_\lambda,{\varphi})$$ of $S^\Gamma_\lambda$ by $$Q_{-\lambda} (f) := \pi^\Gamma_*(\chi^\Gamma \chi_0 J_{-\lambda} (\chi^\Gamma \tilde\chi_0 f)) + \sum_{P\in\tilde{\mathcal{P}}} T_P^* \circ S_{-\lambda}^{U_P}\circ (\tilde T^P)^* (f)\ .$$ Here $\tilde \chi_0$ is some cut-off function on $B_\Gamma$ of compact support such that $\chi_0\tilde \chi_0=\chi_0$, and $\tilde T^P$ is defined in the same way as $T^P$ but using a cut-off function $\tilde\chi_p$ with support on $B_p$ such that $\chi_p\tilde\chi_p=\chi_p$. Note that $Q_{-\lambda}$ has a continuous extension to a map $$Q_{-\lambda}:D_\Gamma(1_{-\lambda},{\varphi})\rightarrow D_\Gamma(1_\lambda,{\varphi})\ .$$ This is clear for the second term $\sum_{P\in\tilde{\mathcal{P}}} T_P^* \circ S_{-\lambda}^{U_P}\circ (\tilde T^P)^*$ since the scattering matrices $S_{-\lambda}^{U_P}$ extend to distributions. The map $f\mapsto\chi^\Gamma\chi_0J_{-\lambda}(\chi^\Gamma\tilde\chi_0f)$ extends to a continuous map from $D_\Gamma(1_{-\lambda},{\varphi})$ to $C^{-\infty}_c(\Omega_\Gamma,V(1_\lambda,{\varphi}))$. The push-down extends to a continuous map from $C^{-\infty}_c(\Omega_\Gamma,V(1_\lambda,{\varphi}))$ to $C^{-\infty}_c(B_\Gamma,V_{B_\Gamma}(1_\lambda,{\varphi}))$. This implies that the first term $\pi^\Gamma_*(\chi^\Gamma \chi_0 J_{-\lambda} (\chi^\Gamma \tilde\chi_0 f))$ extends to distributions, too. \subsubsection{} We claim that for $|{\rm Re }(\lambda)|<\rho-\beta$ the scattering matrix $S_\lambda^{U_P}$ has at most finite-dimensional singularities. To see the claim we write $S_\lambda^{U_P}=res^{U_P}\circ J_\lambda\circ ext^{U_P}$. Since $J_\lambda$ is regular and bijective for these $\lambda$, and $ext^{U_P}$ has finite-dimensional singularities, the only term which may contribute infinite-dimensional singularities is $res^{U_P}$. Since $\{res^{U_P}\}$ is always regular the singular part of $res^{U_P}$ has values in the space ${R}_{U_P}(1_{\lambda},\tilde{\varphi})^*$. Now $S_\lambda^{U_P}$ is the continuous extension of its restriction to the dense subspace $B_{U_P}(1_\lambda,{\varphi})$. Since $S^{U_P}_\lambda$ maps $B_{U_P}(1_\lambda,{\varphi})$ to $B_{U_P}(1_{-\lambda},{\varphi})$ the range of the singular part of $S_\lambda^{U_P}$ does not contain non-trivial elements of ${R}_{U_P}(1_{\lambda},\tilde{\varphi})^*$. Let $D_{U_P}(1_\lambda,{\varphi})^0\subset D_{U_P}(1_\lambda,{\varphi})$ be a closed subspace of finite codimension on which $ext^{U_P}$ is regular and put $B_{U_P}(1_\lambda,{\varphi})^0:= D_{U_P}(1_\lambda,{\varphi})^0\cap B_{U_P}(1_\lambda,{\varphi})$. Then $B_{U_P}(1_\lambda,{\varphi})^0$ has finite codimension in $B_{U_P}(1_\lambda,{\varphi})$, and the closure of $B_{U_P}(1_\lambda,{\varphi})^0$ in $D_{U_P}(1_\lambda,{\varphi})$ is $D_{U_P}(1_\lambda,{\varphi})^0$. Looking at Laurent expansions we conclude that the restriction of $S_\lambda^{U_P}$ to $B_{U_P}(1_\lambda,{\varphi})^0$ and hence to $D_{U_P}(1_\lambda,{\varphi})^0$ is regular. This proves the claim. We conclude that for $\pm\lambda\in U$ the family $Q_\lambda$ has at most finite-dimensional singularities. \subsubsection{} We define the remainder $$R_\lambda:=Q_{-\lambda}\circ S^\Gamma_\lambda-{\mathrm {id}} \ .$$ We are going to show that $R_\lambda$ is a meromorphic family of smoothing operators with finite-dimensional singularities. We start with \begin{eqnarray*} R_\lambda(f)&=&Q_{-\lambda}\circ S^\Gamma_\lambda(f)-f\\ &=&\pi^\Gamma_* (\chi^\Gamma \chi_0 J_{-\lambda}( \chi^\Gamma\tilde\chi_0 S^\Gamma_\lambda(f)))\\ && + \sum_{P\in\tilde{\mathcal{P}}} T_P^* S_{-\lambda}^{U_P}(\tilde T^P)^* S^\Gamma_\lambda(f)-f \end{eqnarray*} Next we insert the definition $S^\Gamma_\lambda=res^\Gamma\circ J_\lambda\circ ext^\Gamma$ and the definition of $res^\Gamma$ (\ref{rreessdef}). We obtain \begin{eqnarray*} R_\lambda(f)&=& \pi^\Gamma_*(\chi^\Gamma \chi_0 J_{-\lambda} (\chi^\Gamma\tilde\chi_0 \pi^\Gamma_*(\chi^\Gamma\chi_0 J_\lambda\circ ext^{\Gamma}(f)))) \\ &&+\sum_{P\in\tilde{\mathcal{P}}}\pi^\Gamma_* (\chi^\Gamma \chi_0 J_{-\lambda} (\chi^\Gamma\tilde\chi_0 T_P^* res^{U_P}\circ J_\lambda \circ ext^{\Gamma}(f)))\\ &&+\sum_{P\in\tilde{\mathcal{P}}} T_P^* S_{-\lambda}^{U_P} (\tilde T^P)^*\pi^\Gamma_*( \chi^\Gamma\chi_0 J_\lambda \circ ext^\Gamma(f))\\ &&+\sum_{P,Q\in\tilde{\mathcal{P}}} T_P^* S_{-\lambda}^{U_P} (\tilde T^P)^* T_Q^* (res^{U_Q}\circ J_\lambda \circ ext^{\Gamma} (f)) - f\ . \end{eqnarray*} Now we employ \begin{eqnarray*} \chi^\Gamma\tilde\chi_0 \pi^\Gamma_* \chi^\Gamma \chi_0&=& \chi^\Gamma\chi_0\\ \chi^\Gamma\tilde\chi_0\sum_{P\in\tilde{\mathcal{P}}} T_P^*\circ res^{U_P} &=& \tilde\chi_0 (1-\chi_0)\chi^\Gamma\\ \sum_{Q\in\tilde{\mathcal{P}}} (\tilde T^P)^* T_Q^* res^{U_Q} + (\tilde T^P)^*\pi^\Gamma_* \chi^\Gamma\chi_0 & =& \tilde \chi_P res^{U_P}\\ res^{U_P}\circ J_\lambda\circ ext^\Gamma &=& S^{U_P}_\lambda\circ ext^{\Gamma/U_P} \end{eqnarray*} in order to obtain \begin{eqnarray*} R_\lambda(f) &=&\pi^\Gamma_*(\chi^\Gamma \chi_0 J_{-\lambda} (\chi^\Gamma\tilde\chi_0 J_\lambda \circ ext^{\Gamma}(f))) \\ &&+\sum_{P\in\tilde{\mathcal{P}}} T_P^* S_{-\lambda}^{U_P} (\tilde\chi_P S_{\lambda}^{U_P} \circ ext^{\Gamma/U_P}(f))- f\ . \end{eqnarray*} Using the functional equations of the intertwining operators and scattering matrices we can further write \begin{eqnarray*} R_\lambda(f) &=&\pi^\Gamma_*(\chi^\Gamma \chi_0 ext^{\Gamma}(f)) \\ &&-\pi^\Gamma_*(\chi^\Gamma \chi_0 J_{-\lambda}( (1-\chi^\Gamma\tilde\chi_0) J_\lambda \circ ext^{\Gamma}(f))) \\ &&+\sum_{P\in\tilde{\mathcal{P}}} T_P^* ext^{\Gamma/U_P}(f)\\ &&-\sum_{P\in\tilde{\mathcal{P}}} T_P^* S_{-\lambda}^{U_P} ((1-\tilde\chi_P) S_{\lambda}^{U_P} \circ ext^{\Gamma/U_P}(f))- f\\ &=&-\pi^\Gamma_*(\chi^\Gamma \chi_0 J_{-\lambda}( (1-\chi^\Gamma\tilde\chi_0) J_\lambda\circ ext^{\Gamma}(f))) \\ &&-\sum_{P\in\tilde{\mathcal{P}}} T_P^* S_{-\lambda}^{U_P} ((1-\tilde\chi_P) S_{\lambda}^{U_P}\circ ext^{\Gamma/U_P}(f))\ . \end{eqnarray*} Note that $\chi^\Gamma \chi_0 (1-\chi^\Gamma\tilde\chi_0)=0$, and that $J_{-\lambda}$ is off-diagonal smoothing. Furthermore note that $\chi_P (1-\tilde\chi_P)=0$, and that $S_{-\lambda}^{U_P}$ is off-diagonal smoothing by Lemma \ref{offside}. We conclude that $R_\lambda$ extends to a meromorphic family of continuous maps $$R_\lambda:D_\Gamma(1_\lambda,{\varphi})\rightarrow B_\Gamma(1_\lambda,{\varphi})$$ with finite-dimensional singularities. Since $J_0={\mathrm {id}}$ and $S_0^{U_P}={\mathrm {id}}$ we have $R_0=0$. \subsubsection{} We define the Banach space ${S}_{\Gamma,0,0}(1_\lambda,{\varphi})$ to be the subspace of all $f\in C(B_\Gamma,V_{B_\Gamma}(1_\lambda,{\varphi}))$ such that $T_P(f)\in {S}_{U_P,0,0}(1_\lambda,{\varphi})$ for all $P\in \tilde{\mathcal{P}}$ (see \ref{weih259}). Then we have compact inclusions $$ B_\Gamma(1_\lambda,{\varphi})\hookrightarrow {S}_{\Gamma,0,0}(1_\lambda,{\varphi}) \hookrightarrow D_\Gamma(1_\lambda,{\varphi})$$ (in order to see the second inclusion use Lemma \ref{fundd}). The family of Banach spaces $\{{S}_{\Gamma,0,0}(1_\lambda,{\varphi})\}_{\lambda\in{\aaaa_\C^\ast}}$ forms a trivial holomorphic bundle. \subsubsection{} We now apply meromorphic Fredholm theory \cite{reedsimon78}, Thm. VI.13, (or better its version for families of operators on Banach spaces) to the family $$1+R_\lambda:{S}_{\Gamma,0,0}(1_\lambda,{\varphi})\rightarrow {S}_{\Gamma,0,0}(1_\lambda,{\varphi})\ .$$ As in \cite{MR1749869}, Lemma 3.6, we conclude that $(1+R_\lambda)^{-1}$ exists as a meromorphic family of maps on ${S}_{\Gamma,0,0}(1_\lambda,{\varphi})$ of the form $(1+T_\lambda)$, where $T_\lambda$ is a meromorphic family of continuous operators from $D_\Gamma(1_\lambda,{\varphi})$ to $B_\Gamma(1_\lambda,{\varphi})$ with finite-dimensional singularities. \subsubsection{} We see that the inverse of the scattering matrix is given on $U$ by $$(S^\Gamma_\lambda)^{-1} := (1+R_\lambda)^{-1} Q_{-\lambda}\ .$$ as a meromorphic family and with finite-dimensional singularities. \subsubsection{} We have $-U\cup \{\lambda\in{\aaaa_\C^\ast}\:|\: {\rm Re }(\lambda)>\delta_\Gamma+\delta_{\varphi}\}=W$. By our assumption the set $$W_0=\left\{\lambda\in{\aaaa_\C^\ast}\:|\: \delta_\Gamma+\delta_{\varphi} < {\rm Re }(\lambda)< -\delta_{\varphi}-\max\left(\{\delta_\Gamma\}\cup\{\rho_{U_P}\mid P\in\tilde{\mathcal{P}}\}\right)\right\}$$ is non-empty. By Corollary \ref{wieder} we find for $\lambda\in W_0$ the meromorphic identity \begin{eqnarray*} S_{-\lambda}^\Gamma\circ S^\Gamma_\lambda&=& res^\Gamma\circ J_{-\lambda}\circ ext^\Gamma\circ res^\Gamma\circ J_\lambda\circ ext^\Gamma\\ &=&res^\Gamma\circ J_{-\lambda}\circ J_\lambda\circ ext^\Gamma\\ &=&res^\Gamma\circ ext^\Gamma\\ &=&{\mathrm {id}} \ . \end{eqnarray*} Thus defining $S^\Gamma_\lambda:=(S_{-\lambda}^\Gamma)^{-1}$ for $-\lambda\in U$ we obtain the continuation of $S^\Gamma_\lambda$ to $W$. \subsubsection{} We obtain the meromorphic continuation of $ext^\Gamma$ to $W$ with finite-dimensional singularities using the identity $$ext^\Gamma=J_{-\lambda}\circ ext^\Gamma\circ S_\lambda^\Gamma\ .$$ See the proof of \cite{MR1749869}, Lemma 5.11, for the corresponding argument. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} The following lemma finishes the proof of Proposition \ref{meroext}. \begin{lem}\label{hjdhdjwqdwqdwqdwq} Lemma \ref{wahn1} holds true without the assumption (\ref{kuckuck}). \end{lem} {\it Proof.$\:\:\:\:$} We are going to employ the embedding trick. First we assume that $G^n$ belongs to the list $\{Spin(1,n),SO(1,n)_0,SU(1,n),Sp(1,n)\}$. If we consider $\Gamma\subset G^n$ as a subgroup of $G^{n+m}$, then for sufficiently large $m$ the inequality (\ref{kuckuck}) is satisfied. Indeed, all ingredients of (\ref{kuckuck}) but $\delta_\Gamma$ are independent of $m$, while $\delta_\Gamma^{n+m}=\delta^n_\Gamma-m\zeta$, $\zeta=\rho^{n+1}-\rho^n$. It thus suffices to show that a meromorphic continuation of $ext^{\Gamma,n+m}$ implies a meromorphic continuation of $ext^{\Gamma,n}$. Assume that we have a meromorphic continuation of $ext^{\Gamma,n+m}$ to $W^{n+m}:=\{\lambda\in{\aaaa_\C^\ast}\:|\: {\rm Re }(\lambda)>-\rho^{n+m}+\beta\}$ with finite-dimensional singularities. Then we employ the diagram $$\begin{array}{ccc} C^{-\infty}(\partial X,V(1^{n}_\lambda,{\varphi})) & \stackrel{ j^*}{ \leftarrow} &C^\infty(\partial X^{n+m}, V(1^{n+m}_{\lambda-m\zeta},{\varphi})) \\ \uparrow ext^{\Gamma,n}_*&&\uparrow ext^{\Gamma,n+m}_* \\ D_\Gamma(1^{n}_\lambda,{\varphi})&\stackrel{i^\Gamma_*}{\rightarrow }&D_\Gamma(1^{n+m}_{\lambda-m\zeta},{\varphi}) \end{array}$$ which apriori holds in the domain of convergence ${\rm Re }(\lambda)>\delta_\Gamma^n+\delta_{\varphi}^n$ by Lemma \ref{comppp}. Note that the left-inverse $j^*$ of $i^*$ was constructed in Lemma \ref{jjjjjjjjjjjtwzew} (or better its iterate for the $m$-fold embedding). We can define the meromorphic continuation to $W^n=W^{n+m}+m\zeta$ of $ext^{\Gamma,n}$ using the identity $ext^{\Gamma,n}=j^*\circ ext^{\Gamma,n+m}\circ i^\Gamma_*$. Let now $G$ be general. Then we can find a subgroup $\Gamma^0\subset \Gamma$ of finite index to which we can apply the concept of embedding (see \ref{ohu}). In particular we obtain a meromorphic continuation of $ext^{\Gamma^0}$ to $W$. If we identify $D_\Gamma(1_\lambda,{\varphi})$ with the subspace ${}^{\Gamma/\Gamma^0} D_{\Gamma^0}(1_\lambda,{\varphi})$, then we obtain the meromorphic continuation of $ext^\Gamma$ using the identity $ext^{\Gamma}=ext^{\Gamma^0}_{|D_\Gamma(1_\lambda,{\varphi})}$. Using the meromorphic continuation of $ext^{\Gamma,n}$ we then obtain the meromorphic continuation of the scattering matrix $S^\Gamma_\lambda$ by $S_\lambda^\Gamma=res^\Gamma\circ J_\lambda\circ ext^\Gamma$. \hspace*{\fill}$\Box$ \\[0.5cm]\noindent \subsubsection{} In Proposition \ref{meroext} we have obtained (in the spherical case) a meromorphic extension of $ext^\Gamma$ and the scattering matrix $S_\lambda^\Gamma$ to a half space $W\subset {\aaaa_\C^\ast}$. In the following argument, using twisting, we find the meromorphic continuation of these objects to all of ${\aaaa_\C^\ast}$. \subsubsection{} Let $\sigma$ be a Weyl invariant representation of $M$, and $(\pi_{\sigma,\mu},V_{\pi_{\sigma,\mu}})$ be a finite-dimensional representation of $G$ as in \ref{weih144}. The adjoint of the diagram 2. of Lemma \ref{compat5} gives the following commutative diagram of meromorphic families of maps \begin{equation}\label{dieletzte}\begin{array}{ccc} D_\Gamma(\sigma_\lambda,{\varphi}) &\stackrel{i^\Gamma_{\sigma,\mu}}{\rightarrow}&D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi})\\ \downarrow ext^\Gamma&&\downarrow ext^\Gamma\\ C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi})) &\stackrel{j_{\sigma,\mu}}{\leftarrow}& C^{-\infty}(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu} \otimes{\varphi})) \end{array} \end{equation} for ${\rm Re }(\lambda)>\delta_\Gamma+\delta_{\varphi}$. Using Proposition \ref{meroext} we conclude that $$ext^\Gamma:D_\Gamma(\sigma_\lambda,{\varphi})\rightarrow C^{-\infty}(\partial X,V(\sigma_\lambda,{\varphi}))$$ is meromorphic on the set $W-\mu$. Since we can choose $\mu$ arbitrary large we obtain the meromorphic continutaion of $ext^\Gamma$ to all of ${\aaaa_\C^\ast}$. From (\ref{dieletzte}), injectivity of $i_{\sigma,\mu}$, and Proposition \ref{meroext} we conclude that $ext^\Gamma$ has finite-dimensional singularities. This finishes the proof of Theorem \ref{t110}. The adjoint of the extension is the push-down. Therefore Theorem \ref{t119} follows from \ref{t110}. \subsubsection{} Let for a moment be $\sigma=1$. Using the meromorphic continuation of the extension we obtain the meromorphic continuation of the scattering matrix to all of ${\aaaa_\C^\ast}$ by the formula $S^\Gamma_\lambda=res^\Gamma\circ J_\lambda\circ ext^\Gamma$. In order to proceed similarly for general $\sigma$ we first have to define the restriction map in this situation. Let now $\sigma$ be a Weyl-invariant representation of $M$ and $(\pi_{\sigma,\mu},V_{\pi_{\sigma,\mu}})$ be a finite-dimensional representation of $G$ as in \ref{weih144}. For $k\in {\mathbb {N}}$ we define restriction maps $$res^\Gamma:{}^\Gamma C^{-k}(\partial X,V(\sigma_\lambda,{\varphi})) \rightarrow D_\Gamma(\sigma_\lambda,{\varphi})$$ using the diagrams $$\begin{array}{ccc} C^{-k}(\partial X,V(\sigma_\lambda,{\varphi})) &\stackrel{i_{\sigma,\mu}}{\rightarrow}&C^{-k}(\partial X,V(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}))\\ \downarrow res^\Gamma&&\downarrow res^\Gamma\\ D_\Gamma(\sigma_\lambda,{\varphi})&\stackrel{j^\Gamma_{\sigma,\mu}}{\leftarrow}& D_\Gamma(1_{\lambda+\mu},\pi_{\sigma,\mu}\otimes{\varphi}) \end{array}\ .$$ The right column depends on choices (including $\mu$), and so does the left column. But it is easy to check that \begin{equation}\label{juhu} res^\Gamma\circ ext^\Gamma ={\mathrm {id}} \end{equation} \subsubsection{}\label{hjshqwjsdwhqdwqdwqdwqdqw} Note that (\ref{dieletzte}) implies the meromorphic identity $ext^\Gamma\circ j_{\sigma,\mu}^\Gamma=j_{\sigma,\mu}\circ ext^\Gamma$. Let $f_\nu\in {}^\Gamma C^{-\infty}(\partial X,V(\sigma_\nu,{\varphi}))$ be a meromorphic family defined on {\em all} of ${\aaaa_\C^\ast}$. We compute for $\nu\gg 0$ using Corollary \ref{wieder} and any set of choices for $res^\Gamma$ \begin{eqnarray}\label{jetztaber} ext^\Gamma\circ res^\Gamma(f_\nu)&=& ext^\Gamma\circ j_{\sigma,\mu}^\Gamma \circ res^\Gamma\circ i_{\sigma,\mu}(f_\nu)\nonumber\\ &=&j_{\sigma,\mu}\circ ext^\Gamma\circ res^\Gamma\circ i_{\sigma,\mu}(f_\nu)\nonumber\\ &=&j_{\sigma,\mu}\circ i_{\sigma,\mu}(f_\nu)\nonumber\\ &=&f_\nu\ . \end{eqnarray} Formula (\ref{juhu}) implies that $ext^\Gamma$ is injective on families. Thus we can conclude from (\ref{jetztaber}) that $res^\Gamma(f_\nu)$ is independent of choices. \subsubsection{} Now we can define the scattering matrix for general $\sigma$ by the same formula as in the spherical case $$S_\lambda^\Gamma:=res^\Gamma\circ J_\lambda\circ ext^\Gamma\ .$$ It is meromorphic on ${\aaaa_\C^\ast}$. In order to finish the proof of Theorem \ref{t113} it remains to show the functional equation. Using (\ref{jetztaber}) we find \begin{eqnarray*} S^\Gamma_{-\lambda}\circ S_\lambda^\Gamma&=& res^\Gamma\circ J_{-\lambda} \circ ext^\Gamma\circ res^\Gamma\circ J_\lambda\circ ext^\Gamma\\ &=&res^\Gamma\circ J_{-\lambda}\circ J_\lambda\circ ext^\Gamma\\ &=&res^\Gamma\circ ext^\Gamma\\ &=&{\mathrm {id}}\ . \end{eqnarray*} \bibliographystyle{plain}
\section{Introduction} It has long been believed that trans-Alfv\'{e}nic shock waves (TASWs), at which the flow velocity passes over the Alfv\'{e}n velocity, cannot exist in the real world. Since a stationary trans-Alfv\'{e}nic shock transition was obtained in a numerical simulation \cite{wu87}, this conventional view point was replaced by an opposite view point. The overall claim was that there is no principal difference between TASWs and fast and slow shocks, at which the flow is super- and sub-Alfv\'{e}nic, respectively. At the same time, the contradiction inherent in a stationary TASW, which follows from an analytical theory, was not lifted. To reconcile this contradiction, it was suggested that a TASW exists in an unsteady state in which it is repeatedly destroyed and recovered \cite{m98}. In the present paper, we show by way of magnetohydrodynamic (MHD) simulation that the evolution of a TASW may have the form of oscillatory disintegration, i.e., reversible transformation into another TASW. The disintegration of an arbitrary hydrodynamic discontinuity was considered for the first time by Kotchine \cite{kotchine}. After that, Bethe \cite{bethe} studied the disintegration of shock waves. In the absence of a magnetic field, the shock may disintegrate only in a medium with anomalous thermodynamic properties. The magnetic field enlarges the number of possible discontinuous structures thus giving additional degrees of freedom for the disintegration. The disintegration configurations of arbitrary MHD discontinuities were obtained in Refs. \cite{arbit}. Furthermore, it has been shown that trans-Alfv\'{e}nic shock transitions can be realized also through a set of several discontinuities \cite{shock}, in contrast with fast and slow transitions. However, this fact on its own does not assure that the shock disintegrates. The important feature that predetermines the disintegration of TASWs is their nonevolutionarity. The problem of evolutionarity was initially formulated for the fronts of combustion \cite{landau} and hydrodynamic discontinuities \cite{cf}. Evolutionarity is a property of a discontinuous flow to evolve in such a way that the flow variation remains small under the action of a small perturbation. This is not the case for a nonevolutionary discontinuity. At such a discontinuity, the system of boundary conditions, which follow from the conservation laws, does not have the unique solution for the amplitudes of outgoing waves generated by given incident waves. From a mathematical view point this means that the number of unknown parameters (the amplitudes of the outgoing waves and the discontinuity displacement) is incompatible with the number of independent equations. Since a physical problem must have the unique solution, the assumption that the perturbation of a nonevolutionary discontinuity is infinitesimal leads to a contradiction. In fact, the infinitesimal perturbation results in disintegration, i.e., finite variation of the initial flow, or transformation into some other unsteady configuration. The evolutionarity requirement gives additional restrictions on the flow parameters at a shock, compared to the condition of the entropy increase. The restrictions appear because the direction of wave propagation (toward the discontinuity surface or away from it), and thus the number of the outgoing waves, depends on the flow velocity. If the velocity is large enough then the given wave may be carried down by the flow. Therefore, at an evolutionary discontinuity, the flow velocity must be such that it provides the compatibility of the boundary equations. This form of evolutionarity condition was applied to MHD shock waves in Refs. \cite{mhd,s58}. As a result, the fast and slow shocks are evolutionary, while the TASWs are nonevolutionary. This classical picture was challenged when Wu \cite{wu87} obtained a stationary TASW in a numerical simulation. The existence of a stationary numerical solution does not mean of course that the shock is stable with respect to disintegration or transition into another unsteady flow. Wu \cite{wu88} demonstrated that a TASW, which is subfast upstream and subslow downstream, disintegrates under the action of a small Alfv\'{e}n perturbation with a large enough characteristic time. Nevertheless, this numerical result was interpreted as being in a contradiction with the principle of evolutionarity and stimulated the efforts to modify or even disprove this principle. It was suggested that the free parameters that describe a nonunique structure of a TASW \cite{wu90} or the amplitudes of strongly damping dissipative waves \cite{hada} should be included in the number of unknown parameters when solving the problem of evolutionarity. This would make the TASW evolutionary. In both cases, however, the perturbation is confined within the shock transition layer. Consequently, it does not enter into the boundary conditions, which relate the quantities far enough from the transition layer, and thus it does not contribute to the evolutionarity \cite{evol}. Wu \cite{wu90} also argued that the TASW whose nonevolutionarity is based on separation of Alfv\'{e}n perturbations from the remaining perturbations \cite{s58} becomes evolutionary in the case of a nonplanar shock structure because in this case the separation formally does not take place. However, as shown by Markovskii \cite{evol}, the coupling of the small-amplitude Alfv\'{e}n modes with a low enough frequency to the remaining modes is weak (unless the shock is of the type close to one of the degenerate types, Alfv\'{e}n discontinuity or switch shocks). Therefore the coupling becomes essential only when the small perturbation generates large variation of the flow, which is the same result as predicted by the principle of evolutionarity. There is one more finding that favors the nonexistence of stationary TASWs. As discussed by Kantrowitz and Petschek \cite{kp}, the TASWs are isolated solutions of Rankine-Hugoniot problem, which do not have neighboring solutions corresponding to small deviations of boundary conditions. Wu and Kennel \cite{wk} introduced a new class of trans-Alfv\'{e}nic shock-like structures with noncoplanar boundary states. The thickness of such a structure increases in the course of time, and it eventually evolves to a large-amplitude Alfv\'{e}n wave. It was thus shown that neighboring to a TASW are time-dependent configurations, which are not solutions of the Rankine-Hugoniot problem. In addition, Falle and Komissarov \cite{fk} recently considered stationary TASWs of all possible types and showed that the shocks disintegrate if the boundary values deviate from their initial values. Strictly speaking, a TASW, in contrast with fast and slow shocks, becomes a time-dependent shock-like structure once it is perturbed by a small-amplitude Alfv\'{e}n wave because the Alfv\'{e}n wave violates the coplanarity condition. This fact, already on its own, means that the TASW becomes unsteady under the action of the small perturbation. However, the scenario for its evolution depends on the initial configuration and on the nature of the perturbation. After the disintegration, the magnetic field reversal given at the initial nonevolutionary shock may be taken either by a secondary TASW or by an Alfv\'{e}n discontinuity. Both structures are nonevolutionary \cite{m98,evol}. Therefore single disintegration does not lift the contradiction inherent in a TASW. The main question that we solve in this paper is what happens to the post-disintegration nonevolutionary configuration. We show that the secondary TASW is again unstable with respect to disintegration and that the evolution of a TASW may have the form of oscillatory disintegration. In Sec. II, we describe the simulation method. In Sec. III, we discuss the results of the calculations. Our conclusions are presented in Sec. IV. \section{Numerical method} We take the MHD equations in the following form \begin{mathletters} \label{1} \begin{equation} {\partial \rho \over \partial t} + {\partial \rho v_{x} \over \partial x} = 0, \label{1a} \end{equation} \begin{equation} {\partial \rho v_{x} \over \partial t} + {\partial \over \partial x} \biggl( p + \rho v_{x}^{2} + {\textstyle {1\over 2}}{\bf B}_{\perp}^{2} - {\textstyle { 4\over 3}} \eta {\partial v_{x} \over \partial x}\biggr) = 0, \label{1b} \end{equation} \begin{equation} {\partial \rho {\bf v}_{\perp} \over \partial t} + {\partial \over \partial x} \biggl( \rho v_{x} {\bf v}_{\perp} - B_{x} {\bf B}_{\perp} - \eta {\partial {\bf v}_{\perp} \over \partial x} \biggr) = 0, \label{1c} \end{equation} \begin{equation} {\partial {\bf B}_{\perp} \over \partial t} + {\partial \over \partial x} \biggl( v_{x} {\bf B}_{\perp} - B_{x} {\bf v}_{\perp} - \nu_{m} {\partial {\bf B}_{\perp} \over \partial x} \biggr) = 0, \label{1d} \end{equation} \begin{eqnarray} && {\partial \over \partial t} \biggl( {\textstyle {1\over 2}} \rho v^{2} + { p \over \gamma - 1 } + {\textstyle {1\over 2}} {\bf B}_{\perp}^{2} \biggr) + {\partial \over \partial x} \biggl[ \rho v_{x} \biggl({\textstyle {1\over 2}} v^{2} \nonumber \\ && \quad + { \gamma \over \gamma - 1} { p\over \rho} \biggr) + \biggl( {\bf B}_{\perp} \cdot \biggl( v_{x} {\bf B}_{\perp} - B_{x} {\bf v}_{\perp} - \nu_{m} {\partial {\bf B}_{\perp} \over \partial x} \biggr) \biggr) \nonumber \\ && \quad - \eta \biggl( {\textstyle{4\over3} } v_{x} {\partial v_{x}\over \partial x} + \biggl( {\bf v}_{\perp} \cdot {\partial {\bf v}_{\perp}\over \partial x} \biggr)\biggr)\biggr] = 0. \label{1e} \end{eqnarray} \end{mathletters} Here the subscript "$\perp$" denotes the vector component perpendicular to the $x$ axis, $B_{x}={\rm const},$ magnetic diffusivity $\nu_{m}$ and viscosity $\eta$ are put constant and equal to 0.1 in all calculations, and we use the units such that the factor $4\pi$ does not appear. The initial distribution of the MHD quantities is given by the following formulas \begin{mathletters} \label{2} \begin{equation} \rho = { \textstyle {1\over 2}} (\rho_{\uparrow} + \rho_{\downarrow}) - { \textstyle {1\over 2}} (\rho_{\uparrow} - \rho_{\downarrow}) {\rm tanh} ( { x/ L}) , \label{2a} \end{equation} \begin{equation} v_{x} = { \textstyle {1\over 2}} (v_{x\uparrow} + v_{x\downarrow}) - {\textstyle {1\over 2}} (v_{x\uparrow} - v_{x\downarrow}) {\rm tanh} ( { x/ L} ) , \label{2b} \end{equation} \begin{equation} p = {\textstyle {1\over 2}} (p_{\uparrow} + p_{\downarrow}) - { \textstyle {1\over 2}} (p_{\uparrow} - p_{\downarrow}) {\rm tanh} ( { x/ L}) , \label{2c} \end{equation} \begin{equation} B_{y} = B_{\tau} {\rm cos} (\theta), \qquad B_{z} = B_{\tau} {\rm sin} (\theta), \label{2d} \end{equation} \begin{equation} v_{y} = v_{\tau} {\rm cos} (\theta), \qquad v_{z} = v_{\tau} {\rm sin} (\theta), \label{2e} \end{equation} \begin{eqnarray} B_{\tau} = && {\textstyle {1\over 2}} (\mid B_{\perp\uparrow}\mid + \mid B_{\perp\downarrow}\mid ) \nonumber \\ && - {\textstyle {1\over 2}} (\mid B_{\perp\uparrow} \mid - \mid B_{\perp\downarrow}\mid ) {\rm tanh} ( { x/ L} ), \label{2f} \end{eqnarray} \begin{eqnarray} v_{\tau} = && {\textstyle {1\over 2}} (\mid v_{\perp\uparrow}\mid + \mid v_{\perp\downarrow}\mid ) \nonumber \\ && - {\textstyle {1\over 2} } (\mid v_{\perp\uparrow} \mid - \mid v_{\perp\downarrow}\mid ) {\rm tanh} ( { x/ L} ), \label{2g} \end{eqnarray} \begin{equation} \theta = {\textstyle {\pi\over 2}} (1 + {\rm tanh} ( { x/ L})), \label{2h} \end{equation} \end{mathletters} where the subscripts "$\uparrow$" and "$\downarrow$" denote the quantities in the asymptotic upstream and downstream regions, respectively. After the configuration relaxes to a steady state, it is perturbed by an Alfv\'{e}n wave specified by the expression \begin{mathletters} \label{3} \begin{equation} B_{z} = {\textstyle {1 \over 2}} \delta B_{z} \biggl( 1 + {\rm tanh} \biggl( {x - x_{0} \over l} \biggr) \biggr) , \label{3a} \end{equation} \begin{equation} v_{z} = - B_{z} / \sqrt{\rho}. \label{3b} \end{equation} \end{mathletters} This wave moves downstream. The configuration is set by putting $B_{x}=0.89,$ $L=1.45,$ and \begin{mathletters} \label{4} \begin{equation} B_{y\uparrow} = 0.93, \qquad B_{y\downarrow} = -0.8, \label{4a} \end{equation} \begin{equation} B_{z\uparrow} = v_{z\uparrow} = 0., \qquad B_{z\downarrow} = v_{z\downarrow} = 0., \label{4b} \end{equation} \begin{equation} v_{x\uparrow} = 1., \qquad v_{x\downarrow} = 0.55042, \label{4c} \end{equation} \begin{equation} v_{y\uparrow} = 1.04494, \qquad v_{y\downarrow} = -0.49476, \label{4d} \end{equation} \begin{equation} \rho_{\uparrow} = 1., \qquad \rho_{\downarrow} = 1.81681, \label{4e} \end{equation} \begin{equation} p_{\uparrow} = 0.00116, \qquad p_{\downarrow} = 0.56319. \label{4f} \end{equation} \end{mathletters} This corresponds to a ${\rm II \rightarrow III}$ shock, for which $V_{+\uparrow} > v_{x\uparrow} > V_{Ax\uparrow}$ and $V_{Ax\downarrow} > v_{x\downarrow} > V_{-\downarrow},$ where $V_{+}$ and $V_{-}$ are the fast and slow magnetosonic velocities. We solve Eq. (\ref{1}) using a uniform grid and an explicit conservative Lax-Wendroff finite-difference scheme with physical dissipation \cite{pt}. The time step is limited by the Courant-Friedrichs-Lewy (CFL) condition and by the dissipation timescale. The boundary values are obtained by hyperbolic interpolation. The numerical interval $-50 < x < +300$ is covered by 2600 grid points. The interval is chosen in such a way that no large-amplitude wave reaches the boundaries during the computation time. Small-amplitude waves pass through the boundaries without any detectable reflection which could affect the flow inside the simulation region. We have tested our code for a smaller mesh and a corresponding time step determined by the CFL condition as well as for the same mesh and a time step smaller than that determined by the CFL condition. The test showed that there is no considerable dependence of our results on the mesh size and time step. \section{Results of simulations} Equation (\ref{2}) does not exactly describe the shock structure. Therefore the flow undergoes time variations until it adjusts to a stationary shock transition. The resulting boundary values differ slightly from those given by Eq. (\ref{4}) but the difference is less than 1\%. The conservation laws for these new values are fulfilled with the precision less than 0.1\%. The stationary configuration is then perturbed by a small-amplitude Alfv\'{e}n wave with $l=-L,$ $x_{0}= -40,$ and $\delta B_{z}= 0.025$ or $\delta B_{z}= -0.025$ (Fig.~\ref{f1}). Note that $\delta B_{z}$ is about 50 times smaller than $\mid {\bf B}_{\uparrow} \mid.$ Although in the case of an upstream incident wave the perturbation of $B_{z}$ and $v_{z}$ (not shown) is carried to the downstream region, the boundary conditions for the Alfv\'{e}n waves are incompatible. Therefore the given Alfv\'{e}n perturbation pumps $B_{z}$ and $v_{z}$ into the shock or out of the shock, depending on the sign of $B_{z}$ inside the transition layer. Since $B_{z}$ inside the transition layer is nonzero, the shock behaves in different ways under the action of the perturbations with positive and negative $\delta B_{z}.$ If the shock and the perturbation carry $B_{z}$ of the same sign, the shock disintegrates into a ${\rm II \rightarrow III}$ shock of a smaller amplitude, a large-amplitude slow shock, and some other structures of a much smaller amplitude (Fig.~\ref{f2}a,b). If the shock and the perturbation carry $B_{z}$ of opposite signs, the situation is somewhat peculiar. The main secondary structures are a TASW and a slow rarefaction (Fig.~\ref{f3}a,b). However, these structures do not become separated. The reason is that the secondary TASW is of a so-called ${\rm II \rightarrow IV=III}$ type \cite{kbw}. This means that the downstream velocity at the shock is exactly equal to the slow magnetosonic velocity. Therefore there is no disintegration in the usual sense but the configuration becomes unsteady because the right boundary of the slow rarefaction moves away from the TASW, while the left boundary remains attached to the shock. Note that the rarefaction wave is attached to the TASW not at the density peak but somewhere to the right of the peak. This is related to the fact that the density profile of a ${\rm II \rightarrow IV}$ shock has a maximum (see, e.g., Ref. \cite{wu88}), in contrast with the monotonic profile of a ${\rm II \rightarrow III}$ shock. From the moment when the Alfv\'{e}n wave with $\delta B_{z}>0$ arrives to the shock, the disintegration starts almost immediately, in contrast with the result of Wu \cite{wu88}. The reason is that the disintegration time depends on the shock type and on its initial state. This can be understood as follows. The important characteristic of a TASW, introduced by Kennel {\it et al.} \cite{kbw}, is the integral of $B_{z}$ over the transition layer, \begin{equation} I_{z} = \int \limits_{x\downarrow}^{x\uparrow} B_{z} dx. \label{5} \end{equation} This integral fixes the nonunique structure of a TASW. For a ${\rm II \rightarrow III}$ shock, the quantity $I_{z}$ takes two distinct values, $I_{z0}$ and $-I_{z0},$ and, for a ${\rm I \rightarrow III}$ or ${\rm II \rightarrow IV}$ shock, it falls into the interval $-I_{z0} < I_{z} < I_{z0}.$ The quantity $I_{z0}$ depends on the boundary values, and it tends to infinity when the shock approaches an Alfv\'{e}n discontinuity or a switch shock, which is intermediate between evolutionary and nonevolutionary shocks. This result was obtained for almost parallel small-amplitude shocks, but one may expect that it remains qualitatively valid in the general case. When an Alfv\'{e}n wave is incident on a TASW, it changes $I_{z}.$ If we start from a planar ${\rm I \rightarrow III}$ or ${\rm II \rightarrow IV}$ shock ($I_{z}=0$), as in the case studied by Wu \cite{wu88}, the quantity $\mid I_{z} \mid $ first has to reach the value $I_{z0}.$ Only after that it falls into the forbidden region, and the disintegration starts. In the case of a ${\rm II \rightarrow III}$ shock, there is a different situation. Since $I_{z}$ takes only distinct values $I_{z0}$ and $-I_{z0},$ the disintegration starts immediately, and the disintegration time is close to its minimum value $L/V,$ approximately equal to 30 in our case, where $V$ is a relative velocity of the secondary discontinuities. Let us now follow the further evolution of the post-disintegration configuration under the action of a small perturbation. Our main conclusion is that the secondary TASW is again unstable with respect to disintegration. At the same time, the way of evolution depends on a form of the perturbation. We first discuss the case where the perturbation of the secondary TASW is such that $I_{z}$ continues to increase or decrease, in particular where the perturbation is equal to its initial positive (Fig.~\ref{f2}b,c) or negative (Fig.~\ref{f3}b,c) value. If the perturbation of $B_{z}$ is positive then the shock spreads in space, with all the jumps, except for $\Delta B_{y}$ and $\Delta v_{y},$ decreasing in time. It thus approaches a large-amplitude Alfv\'{e}n wave. If the perturbation is negative, the shock first passes through the state in which $I_{z}=0.$ This is not in a contradiction with the analytical theory, because a ${\rm II \rightarrow IV}$ shock may have a planar structure, in contrast with a ${\rm II \rightarrow III}$ shock \cite{kbw}. When $\mid I_{z} \mid$ reaches a critical value, the shock disintegrates (Fig.~\ref{f3}b), and after that it spreads in space approaching a large-amplitude Alfv\'{e}n wave (Fig.~\ref{f3}c). The precursor of the disintegration is the peak in $B_{y}$ curve in Fig. \ref{3}b. We now turn to a cyclic perturbation. We impose the perturbation described by Eq. (\ref{3}) in such a way that $B_{z}$ changes sign at $x=-40$ and the Alfv\'{e}n wave now carries the perturbation of the same amplitude but opposite sign. After the first disintegration starts (at $t=20$ for $\delta B_{z} >0$ and at $t=670$ for $\delta B_{z}<0$), the opposite sign perturbation arrives to the TASW each 150 units of time. The resulting configuration is such that the increase of $\mid I_{z} \mid$ is repeatedly replaced by its decrease, and the shock undergoes oscillatory disintegration. The disintegration configurations after several cycles are shown in Figs.~\ref{f4} and \ref{f5}. As can be seen from the figures, the configuration emits a sequence of contact discontinuities. The contact discontinuities move with the flow velocity, which is approximately equal to that given by Eq. (\ref{4c}). The corresponding time interval between the discontinuities is equal to 150. Downstream of the TASW, there is a wave train, which consists of slow shock and rarefaction waves. These structures are not standing in the flow. They consecutively emerge at the left edge of the train and merge at the right edge. The merging is seen in Fig. \ref{4}b at $x\approx 150.$ We note that, in the case of a negative initial perturbation, the transition through the state with $I_{z}=0$ is not necessary for the oscillatory disintegration to occur. If the perturbation changes sign for the first time before $I_{z}$ becomes negative, the disintegration configuration is similar to that shown in Fig. \ref{5}, except for the sign of $I_{z}$ inside the shock. Finally, the shock comes to a steady state only in a degenerate case where the perturbation of the secondary TASW exactly compensates the nonzero value of $B_{z}$ and $v_{z}$ outside of the transition layer. We emphasize that in all but the degenerate cases the small Alfv\'{e}n perturbation makes the TASW unsteady, in contrast with fast and slow shocks. However, there remains a question. Formally, the initial TASW becomes a time-dependent structure, much like the secondary TASW, since the Alfv\'{e}n perturbation arrives to the initial shock. The question is why the initial TASW disintegrates when $I_{z}$ increases monotonically, while the secondary TASW does not. To answer this question, we first mention that the secondary TASW is more close to a finite-amplitude Alfv\'{e}n wave than the initial shock. Alfv\'{e}n waves, as well as switch shocks, are singular structures. As shown by Kennel {\it et al.} \cite{kbw}, the quantity $dI_{z0}/dq$ tends to infinity as the shock approaches these singular structures. Here $q=B_{y\uparrow}/B_{y\downarrow}$ characterizes the jumps of the boundary values at the shock with a given $I_{z},$ and $I_{z0}(q)$ is an allowed curve in which a ${\rm II \rightarrow III}$ shock has a stationary structure. Assume now that the initial shock is in the state $I_{z}=I_{z0}(q_{0}).$ A small Alfv\'{e}n perturbation changes $I_{z}.$ For the shock to remain in the curve $I_{z0}(q),$ a change of $q$ is required. In the general case, the variation of $I_{z}$ is comparable with the variation of $q,$ and thus with the jumps of the boundary values at the TASW. In this case, the evolution has the form of disintegration. By contrast, if the shock is close to the singular structure, the given variation of $I_{z}$ requires a small variation of $q,$ and the jumps of the boundary values are adjusted to $I_{z0}(q)$ in a diffusion-like manner. It should be mentioned that the curves $I_{z0}(q)$ were obtained by Kennel {\it et al.} \cite{kbw} for small-amplitude shocks propagating almost parallel to the magnetic field. Nevertheless, we speculate that, in our simulation, the initial TASW has a small enough $dI_{z0}/dq$ to disintegrate, while for the secondary TASW the quantity $dI_{z0}/dq$ is large enough to dim the disintegration. Such an explanation does not imply that a TASW cannot disintegrate more than one time in principle. Furthermore, in our simulation run with a positive constant $\delta B_{z},$ there is an evidence for a possible second disintegration at $t=380.$ However, the second disintegration is too faint to contend that it indeed takes place. \section{Conclusions} We have performed a numerical simulation of a trans-Alfv\'{e}nic shock wave. The shock that we have considered is of a ${\rm II \rightarrow III}$ type, i.e., it is subfast upstream and superslow downstream. We have shown that the shock disintegrates under the action of a small Alfv\'{e}n perturbation. The resulting configuration includes a secondary TASW, a large-amplitude slow shock or rarefaction wave, and other small-amplitude structures. We have also demonstrated that the secondary TASW is again unstable with respect to disintegration. When the perturbation has a cyclic nature, the shock undergoes an oscillatory disintegration. This result is in a qualitative agreement with our previous finding \cite{m98}. This process shows up as a train of slow shock and rarefaction waves, which consecutively emerge at one edge of the train and merge at the other edge. At the same time, the disintegration configuration of a small-amplitude almost parallel TASW discussed by Markovskii \cite{m98} includes alternating TASWs and Alfv\'{e}n discontinuities rather than alternating TASWs. This discrepancy is explained by the fact that, in the approximation used in Ref. \cite{m98}, the difference between the secondary TASW and the Alfv\'{e}n discontinuity manifests itself in higher orders. In contrast with the results of Wu \cite{wu88}, the disintegration starts almost immediately after the Alfv\'{e}n perturbation arrives to the initial shock. The characteristic time of this process is equal to that required for the secondary structures to become separated. The reason for this can be seen as follows. TASWs have a nonunique structure. A ${\rm II \rightarrow IV}$ shock transition studied by Wu \cite{wu88}, as well as a ${\rm I \rightarrow III}$ transition, allows a continuous family of integral curves, while the ${\rm II \rightarrow III}$ shock has two distinct integral curves. For given boundary values, each integral curve is fixed by the definite parameter. The incident Alfv\'{e}n wave changes the parameter and thus the shock structure. In the case of a ${\rm I \rightarrow III}$ or ${\rm II \rightarrow IV}$ shock, some time passes until the parameter falls into a forbidden region, and only after that the shock disintegrates. In the case of a ${\rm II \rightarrow III}$ shock, its structure immediately becomes inconsistent with the boundary values under the action of the Alfv\'{e}n wave, which initiates the disintegration. Thus, our simulations confirm that a TASW becomes unsteady when it is perturbed by a small-amplitude incident wave. Furthermore, an almost vanishing perturbation results in considerable dynamics at relatively small timescales. The scenario for the shock evolution depends on its initial state and on the nature of the perturbation. In particular, the evolution may have the form of oscillatory disintegration in which the shock repeatedly transforms into another TASW. \acknowledgments{This work is supported in part by Russian Foundation for Basic Research (grants 99-02-16344 and 98-01-00501).}
\section{INTRODUCTION} Not long ago a variety of complex magnetic structures formed by many strongly curved and entangled vortices was discovered in bulk superconductors [1]. The origin of these structures can not be explained if treat the motion of vortices like that of stick-like objects. It is necessary to consider the evolution of three-dimensional magnetic flux lines with potentially arbitrary shape. For the first, it is useful to investigate the motion and shaping of a single vortex but interacting with a surface supercurrent which represents either transport current or Meissner current induced by external field. As far as we know, even this simple task previously was not under careful consideration. Of course, a vortex never lives alone, without interactions with other vortices, and no stable many-vortex structure could exist without mutual repulsion of vortices. However, one can suppose that the scenario of magnetic flux penetration into a current-carrying bulk superconductor should be dominated only by vortex interaction with the surface current distribution, i.e. eventually by geometry of superconducting sample, not by the vortex-vortex interaction. The latter can not seriously affect this scenario, merely because it itself is unable to ensure a deep penetration at all. To prove this statement, let us imagine the steady flow of vortices which arise at the flat current-carrying boundary of half-infinite superconductor and then move deep into the sample due to their repulsion. Clearly, because of viscous character of vortex motion, such the flow needs in nonzero gradient of concentration of vortices. As a consequence, both the concentration and the local drift velocity of vortices must be decreasing functions of the depth. Hence, their product is not constant, that is the magnetic flux conservation can not be satisfied. This discrepancy means that no steady flow could be supported by the inter-vortex forces only. In particular, it is impossible to realize the stationary lasting penetration of vortices from infinite flat boundary parallel to external magnetic field. Therefore, the only force what can push a vortex through the sample interior is nothing but self-action of vortex caused by its distorsion. But in order to involve this force into the evolution, the vortex must feel the shape of the sample boundary. Hence, the true picture looks as follows. After nucleation in a surface layer with thickness of order of London penetration depth $\lambda $ , a vortex firstly expands over the sample boundary remaining in this layer. At this stage only the end fragments of vortex are factually moving. The ends slide along the boundary, and the resulting shape of vortex core reflects that of boundary. This process lasts until the curvature of main middle part of the core becomes sufficiently strong in order to cause the deepening of vortex as a whole. In view of these reasonings, the geometry of steady transport of magnetic flux into a bulk supeconductor looks rather insensible to inter-vortex interactions and thus can be testified in terms of a single vortex, at least if not consider details of vortex nucleation and processes like annihilation and reconnections of vortices [1] which take place deep inside the sample. For example, many aspects of resistivity in supercurrent-carrying wires can be described as evolution of ring-like vortices as if thats instantly arise near the boundary, then contract independently one on another and finally self-annihilate [2]. However, more correct consideration should include the first stage when vortex transforms from small nucleus into a ring. We shall see that in fact this stage may result also in a non-ring penetration geometry, and more detailed theory can predict what the scenario realizes under given transport current value and sample dimensions. Though a lot of works were published previously touching upon a role of vortex distorsions, for instance, under a pinning by randomly distributed centers [3], always some preliminary restrictions of the vortex geometry were attracted. In the present work the general equations of evolution of arbitrarily curved vortex lines in isotropic superconductors are formulated and analysed. We shall especially discuss the true formulation of boundary conditions for these equations. It will be shown that in a sample whose dimensions noticably exceed $\lambda $ the vortex can penetrate either as flexible stick or as elastic (similarly to a rubber thread attracted by its ends through water). of surface supercurrent. The latter case occurs only if surface supercurrent exceeds $H_{c1}c/4\pi $ (in CGS units) and is characterized by giant stretching of the vortex core along the sample boundary in the direction parallel to drift of the ends. The stretching is accompanied by decrease of both the vortex energy and viscous dissipation per unit drift velocity, and results in strong increase of the vortex drift velocity under given transport current. In the framework of this scenario, the vortex core firstly tranforms into a ring-like curve winding round the wire cross-section (or into a spiral, if there is an external magnetic field parallel to current), and only later the vortex begins to cut the wire and enter deep into its interior. This general picture is in agreement with the known simplific model of magnetic flux penetration into round wires. Additionally, our approach allows to scope very different stages of vortex evolution in unified manner and obtain quantitative estimates for each stage. \section{LONDON APPROXIMATION} We shall confine ourselves by the London approximation. Of course, it would have no sense if one could not apply it to actually moving vortices. But in any case the requirement must be satisfied that characteristic velocity $u_0$ of viscous vortex motion influenced by magnetic fields comparable with low critical magnetic field $H_{c1}$ , must be significantly smaller than the speed of electromagnetic waves. The velocity scale $u_0$ can be naturally estimated as \[ u_0\equiv \mu \varepsilon /\lambda \] where \[ \varepsilon =\Phi _0H_{c1}/4\pi \] is the self-energy per unit length of long straight-line vortex, $\varepsilon /\lambda $ is the characteristic scale of Lorentz force also related to unit length, and $\mu $ is mobility of the vortex core. Below it will be seen that so defined $u_0$ really serves as the velocity unit. If combine this definition and the known relations [4] \[ \frac{c^2}{\Phi _0\mu }\sim \sigma _nH_{c2}\,\,\,,\,\,\,\sigma _n\sim \frac \hbar \Delta (\frac c\lambda )^2 \] with standard notations, $\sigma _n$ being the normal conductivity and $\Delta \sim 2k_BT_c$ being the order parameter, one obtains \[ u_0\sim \frac{\lambda k_BT_c}{2\pi \hbar }\frac{H_{c1}}{H_{c2}} \] As a typical example, at $T_c\sim 100\,K$ , $\lambda \sim 3\cdot 10^{-5}\,cm$ and $H_{{c2}}/H_{c1}\sim 500$ , one gets the estimate $u_0\sim 10^5\,cm/s$ . This value looks small enough to allow for applicability of quazi-static London approximation. In fact, such an approach was used in large number of works on motion of separate vortices as well as vortex lattices. The obvious exception is very dense lattice, with small inter-vortex distancies of order of coherence length. But our present subjects of interest are far from such complications. \section{EVOLUTION EQUATIONS} In the framework of London approximation, the free energy $E$ of vortex, placed into a given surroundings, is completely determined by the shape of its core, $R(p)=\{X(p),Y(p),Z(p)\}$ , with $X,Y,Z$ being coordinates of the core points and $p$ being a scalar parameter. In accordance with the principles of mechanics and nonequilibrium thermodynamics, the simplest equation of a massless viscous evolution of the core line looks as \begin{equation} \mu^{-1} \partial R/\partial t= f(R) \end{equation} with Lorentz force on the right-hand side and friction force on the left, both being related to unit core length. By its sense, the parameter $\mu ^{-1}$ is the effective drag coefficient which is determined by all the dissipative energy losses conjugated with the core motion. Generally, there are at least two sorts of dissipative processes accompanying the motion (see, for example, the review [4]), namely, relaxation of the order parameter and normal currents induced by time-dependent own magnetic field of the vortex. A concrete expression for $\mu $ can be derived from more detailed theory, for instance, from the Ginzburg-Landau functional approach, under its reduction to London approximation [4]. The reduction is possible because normal self-current of moving vortex and corresponding dissipation are located mainly in a close vicinity of the core line, at distance comparable with coherence length. After the transition to London's description, the effect of normal currents becomes hidden in $\mu $ , but these currents give no contribution to the Lorentz force [4]. Therefore, the reduction results in the identity whose meaning is balance of friction force and Lorentz force, as it is stated by the Eq.1, with $f(R)$ being determined only by supercurrents. To write $f(R)$ , one has not to evaluate the supercurrent distribution. Instead, as in general in mechanics and statistical thermodynamics, $f(R)$ can be expressed by means of $E$'s variation under a small displacement of a local core fragment, that is as the functional derivative $\delta E/\delta R(p)$ . However, the latter itself is not invariant with respect to arbitrary (non-degenerated) transformations of the parametrization $R(p)$ and to physical dimensionality of $p$ . In case of isotropic media, the only true invariant expression for the Lorentz force is \begin{equation} f=-\frac{dp}{dL}\frac{\delta E}{\delta R(p)}=-\left| \frac{\partial R}{\partial p}\right| ^{-1} \frac{\delta E}{\delta R(p)}=\frac{\Phi _0}c\left[ J\times N\right] \end{equation} Here the vector $\partial R/\partial p\equiv R^{\prime }$ is locally parallel to the core, $dL=|R^{\prime }|\,dp$ is the differential of the core length, $N\equiv R^{\prime }\left| R^{\prime }\right| ^{-1}=\partial R/\partial L$ , and $J$ is the density of full effective supercurrent which streams around core and pushes a given core fragment. The energy $E=E\{R(p)\}$ includes self-interaction of vortex and its interaction with surroundings, in particular, with other vortices. Correspondingly, in general $J$ consists of external currents and self-current of vortex determined by its distorsion. The Eqs.1 and 2 could be directly extended to a number of interacting vortices. Besides, in principle, one may add into $E$ also interactions with pinning potentials. However, below we are interested only in motion of separate vortex in absence of pinning. The parameter $p$ enumerates strictly the core points. But in practice it is preferable to use another kind of parametrization, concretely, to introduce the parameter $q$ which enumerates some suitable continuum of surfaces $Q(r)=q$ , $r=\{x,y,z\}$ , each possessing only one intersection with core line. The connection between $p$ and new parameter $q$ is implied by the obvious relation $Q(R(p(q,t),t))=q$ , and simple algebraic manipulations lead to the modified form of the evolution equations, \begin{equation} \frac{\partial R}{\partial t}=\mu \left[ 1-\frac{\partial R}{\partial q} \otimes \frac{\partial Q(R)}{\partial R}\right]f(R)\,\,,\,\,\, \,f(R)=-\left| \frac{\partial R}{\partial q} \right| ^{-1}\frac{\delta E}{\delta R(q)} \end{equation} where the symbol $\otimes $ denotes the tensor product of two vectors. These equations describe how the intersection points marked by $q$ move along the corresponding surfaces $Q(r)=q$ . Clearly, this is factually two-dimensional motion. This feature becomes quite obvious if it is possible to identify $q$ as one of cartezian coordinates, that is to use parallel planes as the marking surfaces. For instance, if thats are XY-planes, $q=Z$ and $Q(r)=z$ , then the Eqs.3 reduces to the equation \begin{equation} \frac \partial {dt}\left( \begin{array}{c} X \\ Y \end{array} \right) =-\frac \mu {\sqrt{1+X^{\prime 2}+Y^{\prime 2}}}\left( \begin{array}{cc} 1+X^{\prime 2} & X^{\prime }Y^{\prime } \\ Y^{\prime }X^{\prime } & 1+Y^{\prime 2} \end{array} \right) \left( \begin{array}{c} \delta E/\delta X(Z) \\ \delta E/\delta Y(Z) \end{array} \right) \end{equation} with shortened notations $X^{\prime }\equiv \partial X/\partial Z$ , $Y^{\prime }\equiv \partial Y/\partial Z$ . The Eq.4 describes the time evolution of $X$ and $Y$ coordinates of the core points marked with their $Z$-coordinate. \section{BOUNDARY CONDITIONS AND LOCAL APPROXIMATION} In absence of pinning and more vortices, the vortex energy $E=E_s+E_i$ consists of two parts: the energy $E_i$ of the vortex interaction with transport or Meissner supercurent and the self-energy $E_s$. Therefore, the current in the Eq.2 also can be devided into two parts, $J=J_s+J_i$ . Formally, $E_s$ is a complicated spatially non-local functional [5] depending on both the core configuration $R(p)$ and the shape of sample. Among other factors, $E_s$ includes the vortex interaction with the sample boundary what can be interpreted as attraction of the end fragments of the core to their mirror images placed outside superconductor. But, if the curvature radius of the core everywhere is not too small as compared with $\lambda $ , and besides, if the core nowhere is too close to itself, then the so-called local approximation is possible, \begin{equation} E_s\approx \varepsilon L=\varepsilon \int \left| dR(p)\right| \end{equation} where $L$ is the core length. This well known approximation was argued and used as long ago as in 1968 by Galaiko [6], and later by many other authors (in particular, in [2-5]). Our own computer simulations showed that the relative error of evaluation of self-action force by means of local approximation does not exceed a few percents even if the curvature radius is as small as $0.1\lambda $ . In the local approximation the Eqs.1 and 2 take the form \begin{equation} \frac{\partial R}{\partial t}=\mu \frac{\Phi _0}c\left[ (J_s+J_i)\times N\right] \,\,\,\,\,,\,\,\,\,\,J_s=\frac{cH_{c1}}{4\pi }\left[ N\times \frac{\partial ^2R}{\partial p^2}\right] \left| \frac{\partial R}{ \partial p}\right| ^{-2} \end{equation} Here $J_s$ is the self-current what flows through the very core. Its absolute value is inversely proportional to the local curvature radius of core. However, the local approximation needs to be accompanied by correct boundary conditions. The true conditions should take into account the vortex interaction with superconductor boundary. There are two ways to show that this interaction results in the orthogonality of the end fragments of core to the boundary. Thought these conditions are known at least since [6], sometimes thats are neglected, so it is desirable to present more argumentation. First, let us note that the force vector $f(R)$ is always perpendicular to the local core direction. Indeed, any variarion $\delta R$ parallel to this core direction, $\delta R\parallel \partial R/\partial p$ , merely is identical to a change of parametrization, without factual change of the shape, so it has no physical meaning and should result in $\delta E=0$ (therefore the last expression in (2) always is consistent with previous ones). The same is seen from (6). As a consequense, any core point displaces perpendicularly to the core, in particular, the end points do which are placed just on the boundary. Hence, we must conclude that the end fragments always are oriented to be orthogonal with respect to the boundary. Secondly, the non-orthogonality would mean that the contour formed by core and its mirror image is broken at the end point, i.e. has infinitely small curvature radius here. From the point of view of exact $E_s$ [5], if such a sharp "knee" occured it would cause infinitely strong Lorentz self-action force and consequently would be immediately straightened thus restoring the orthogonality. But, we must to underline that the principal conclusions to be deduced do not refer to the local approximation and can be derived from general non-local Eqs.1 and 2 only. \section{STICK-ELASTIC VORTEX TRANSFORMATION\\ IN CURRENT-CARRYING PLATE} To avoid a complicated mathematics, we confine ourselves by simplific superconductor geometry. Consider the vortex evolution in an infinitely wide plate, $-D<Z<D$ , without pinning but in presence of transport surface supercurrent uniformly distributed over the boundary planes and obeying the London equation. If this current flows along Y-axis then \[ J_x=J_z=0\,\,\,\,\,,\,\,\,\,J_y=\frac c{4\pi \lambda }H_{c1}j(Z)\,\,\,\,,\,\,\,j(Z)\equiv h\frac{\cosh (Z/\lambda )}{\cosh (D/\lambda )} \] with $h$ being the dimensionless measure of current density. Let initially the vortex pierces the plate in Z-direction being described with $R=\{X(Z,0)=0,0,Z\}$ . It has similar orientation soon after nucleation near the edge of a real finite plate. Then, due to obvious spatial symmetry, the vortex will remain inside the XZ-plane $Y=0$ and keep only one intersection with any of XY-planes. In this situation the Eq.4 can be applied and, besides, reduced to only equation for X-coordinate, $X=X(Z,t)$ , as a function of time and Z -coordinate: \begin{equation} \frac{\partial X}{\partial t}=-\mu \sqrt{1+X^{\prime 2}}\frac{\delta E}{\delta X(Z)} \end{equation} The energy can be expressed as \begin{equation} E=E_s+E_i=E_s-\frac \varepsilon \lambda \int X(Z,t)j(Z)dZ \end{equation} where the integral represents the energy $E_i$ of vortex interaction with transport current (this expression differs only by some constant from the general $E_i$ representation [5]). In the local approximation (5), the Eq.7 looks as strongly nonlinear diffusion-type equation \begin{equation} \frac{\partial X}{\partial t}=u_0\left[ \lambda \frac{X^{\prime \prime }}{1+X^{\prime 2}}+\sqrt{1+X^{\prime 2}}j(Z)\right] \end{equation} with notation $X^{\prime \prime }\equiv \partial ^2X/\partial Z^2$ and characteristic velocity $u_0$ introduced in Sec.2. Here the left side is responsible for the friction, and two terms on the right-hand side represent the self-action force and transport current-induced Lorenz force, respectively. Clearly, because of the latter force both the vortex ends will forwardly move in one and the same direction (to opposite edge of the plate), while the middle of vortex will be more or less backward, and the larger is transport current the longer should be the distance $\Delta X = X(\pm D,t)-X(0,t) $ (below termed vortex stretching). Some predictions of further vortex behaviour can be deduced merely from the energy expression (8). Just after start the middle is still in rest. As the Eq.8 shows, in thick plate ($D>>\lambda $) the unit displacement of every end leads to the $E_i$ 's decrease by $h\epsilon $ . At the same time, the corresponding lengthening of each of two symmetrical core branches results in the $E_s$ 's increase by $\epsilon $ per unit length. Consequently, if $h>1$ then the total energy decreases, and the vortex stretchening along the drift direction becomes profitable. The lengthening process should last until the curvature of the most backward central part of the core becomes so large that the self-action force makes this part moving as quickly as the ends do. Thus, at $h>1$ the vortex gets over the friction like elastic in water. Oppositely, at $h<1$ the stretchening is energetically unprofitable, and the vortex should move as deformed flexible stick. The transition from this stick-like behaviour to elastic-like one, when transport current increases from $h<1$ to $h>1$ , is the example of so-called "nonequilibrium phase transitions". Let us consider the steady drift of vortex as a whole, without change of shape. The corresponding solution on the Eqs. 7 or 9 can be written as $X(Z,t)=ut+X(Z)$ . The stationary shape $X(Z)$ and the drift velocity $u=u(h,D)$ should be obtained from (7) or (9) with the help of above discussed orthogonality boundary conditions \[ \frac{dX}{dZ}(\pm D)=0 \] Besides, due to the mirror symmetry, the condition $X^{\prime }(\pm 0)=0$ should be satisfied. In this steady nonequilibrium state the self-energy $E_s$ is constant, therefore, the work $M_j$ produced by transport current per unit time, \[ M_j=-\frac{dE_i}{dt}=\frac{u\Phi _0}c\int_{-D}^DJ_y(Z)dZ=2uh\varepsilon \tanh (D/\lambda ) \] coinsides with the energy dissipation per unit time $M_d$ . In accordance with (1) and (2), \[ W_d=\frac 1\mu \int \left| \frac{\partial R}{\partial t}\right| ^2dL=\frac 1\mu \int \left( \frac uQ\right) ^2QdZ=2\delta u^2/\mu \] where the notations \[ Q\equiv \sqrt{1+X^{\prime 2}}=\frac{dL}{dZ}\,\,\,,\,\,\delta \equiv \int\limits_0^D\frac{dZ}Q\, \] are introduced. We took into account that actual displacement of the core always is locally perpendicular to its orientation. Only such displacements are physically meaningful and really cause the friction. Therefore, the drift velocity and the local core velocity are connected by the relation \[ \left| \frac{\partial R}{\partial t}\right| =\frac uQ \] Evidently, the factor $Q$ determines at one and the same time local orientation of the core and degree of its stretching. Hence, the equality $M_d=M_j$ yields \begin{equation} U\equiv \frac u{u_0}=\frac{h\lambda }\delta \tanh (D/\lambda )\approx \frac{ h\lambda }\delta \end{equation} In view of above reasonings, at $h<1$ the vortex stretchening is weak, therefore, $X^{\prime 2}$ is comparable with unit, $Q\sim 1$, $\Delta X\sim D$ and $\delta \sim D$. Then the Eq.10 shows that in this stick-like regime $U\sim h\lambda /\delta \approx h\lambda /D<<h$ , i.e. the drift velocity is inversely proportional to the plate thickness. This is quite natural, because the surface current-induced Lorentz force acts only on the ends, while the friction almost equally acts on any core fragment. In general, the parameter $\delta $ serves as the effective plate half-thickness. Obviously, always $\delta <D$ . In the stretched elastic-like regime in thick plate anywhere at $D-|Z|>> \lambda $ the inequalities $\left| X^{\prime }\right| >>1$ and $Q>>1$ take place. Hence, $\delta <<D$ and what is essential it becomes almost insensitive to thickness. As a consequence, both the drift velocity and mobility $u/h$ strongly increase as compared with stick-like regime and both become independent on thickness (below we shall see that $\delta \sim \lambda $ and $U\sim h$ , i.e. $U$ becomes approximately $D/\lambda $ times larger). According to the $M_d$'s expression, the matter is that thought the energy dissipation $Q$ times increases due to the core lengthening this effect is overpowered by its $Q^2$ decrease because of $Q$ times decrease of the factual core velocity $\left| \frac{\partial R}{\partial t}\right| $ . As the result, the vortex stretching leads to smaller friction and smaller entropy production, under fixed vortex velocity, and to larger velocity ander fixed transport current. The picture looks as if most part of core slides along itself, but this process does not mean a real motion of core and so does not cause a friction and dissipation. \section{DRIFT OF THE VORTEX ENDS} To be convinced in what was said, let us consider vortex shape in the stretched elastic-like regime. Because at $\Delta X>>D$ most part of the core inevitably has a small curvature, it can be considered with neglecting the self-action. Then any of the Eqs.7 and 9 reduces to \begin{equation} U\approx \sqrt{1+X^{\prime 2}}j(Z) \end{equation} Here from the characteristic exponential asymptotics does follow, \begin{equation} X(Z)\approx -\frac{\lambda U}h\left[ \exp \left( \frac{D-Z}\lambda \right) -1\right] \end{equation} Here $Z>0$ , $X(-Z)=X(Z)$ , and for definitness the position $X=0$ is prescribed to the end fragments. It is easy to verify that corresponding self-action contribution in the Eq.9 indeed is negligibly small as compared with the current-induced force. We can get a rough estimate of the stretching if put on $Z=0$ in (12) and take into account that $U>h\lambda /D$. Then the Eq.12 yields \[ \Delta X/\lambda \approx \frac Uh\exp \left( \frac D\lambda \right) >\frac \lambda D\exp \left( \frac D\lambda \right) \] Hence, $\Delta X/\lambda $ possesses exponentially strong dependence on $D/\lambda $ , and it can be giantly large if $D$ exceeds $\lambda $ by an order of value or more. In view of this circumstance, the ratio $\Delta X /W $ with $W$ being the width of a real finite plate, becomes of principal importance. Clearly, if $\Delta X >> W $ then the steady drift of the vortex as a whole is impossible: the ends of vortex will achieve the opposite edge before the displacement of its backward central part will be comparable with $W$ (all the more, before the velocity of this part becomes equal to that of the ends). Consider the drift of the ends in such a non-stationary situation. Because the vortex lengthening is profitable, this drift can do independently on the motion of deepened backward part, as if thickness was infinitely large ($D/\lambda \rightarrow \infty $) . To estimate the drift velocity, let us multiply the Eq.7 or 9 (with $\frac{\partial X}{\partial t}\Rightarrow u$ ) by $Q^{-1}$ and integrate over variable $z=D-Z$ from zero to infinity, with the condition $X^{\prime }(z\rightarrow \infty )=\infty $ which evidently corresponds to infinitely far backward center. Then both the Eqs.7 and 9 result in \[ U=\frac \lambda \delta (h-1)\,\,\,,\,\,\delta =\int_0^\infty \frac{dz}Q \] To evaluate this integral, note that in accordance with the orthogonality boundary conditions the shape of the end fragments of the core is parabolic, for instance, at upper end $X(Z)=X(D)-(Z-D)^2/2\rho $ , with $\rho $ being the curvature radius at the end point. It follows from the Eq.9 that $\lambda /\rho =h-U$ . In this parabolic region the integration divergers but becomes cut after transition to exponential asymptotics (12). The estimate of the integral leads to approximate equation \begin{equation} U\approx h(h-1)/\{h-1+\ln [2z_0(h-U)/\lambda ]\} \end{equation} where $z_0$ is the depth of the crossover point, $z_0\sim 4\lambda $ . The Eq.12 helps to estimate the end drift velocity in thick plate. Obviously, it turns into zero at $h\rightarrow 0$, in agreement with $D\rightarrow \infty $ limit of the estimate for stick-like regime. It can be shown that velocity of the steady drift of the vortex as a whole is only slightly smaller differing by a multiplier of order of unit. \section{GIANT VORTEX STRETCHING} The exponentially large vortex stretching is the most significant possibility of vortex evolution in thick plate, as well as in bulk samples in general. Therefore it would be useful to more correctly justify the above simplific estimate of $\Delta X $ . Note that $\Delta X>L/2-D$ . Divide both sides of (9) by $j(Z)$ and integrate from zero to $D$ . This results in \[ L/2=B-A\,\,,\,\,B\equiv U\int \frac{dZ}{j(Z)}\,\,,\,\,\,\,A\equiv \int \arctan (X^{\prime })\left| \frac d{dZ}\frac \lambda {j(Z)}\right| dZ \] It is easy to notice that \[ A<A_0\equiv \frac{\pi \lambda }2\left[ \frac 1{j(0)}-\frac 1{j(D)}\right] \] so $\Delta X>B-A_0-D$. The calculation of integral $B$ gives \[ B=\frac{2U\lambda }h\cosh (D/\lambda )\{\arctan [\exp (D/\lambda )]-\frac \pi 4\} \] Then, after simplifications possible due to $D>>\lambda $, one finally obtains \begin{equation} \Delta X/\lambda >\frac{\pi (U-k)}{4h}\exp (D/\lambda )\, \end{equation} with $k<1$ . Because $u$ is monotonously growing function of $h$ , the coefficient in front of exponent is positive if $h$ exceeds some level larger than unit, for example, if $h>2$ . Hence, ate least at $h>2$ the vortex stretching is exponentially strong. This estimate is obtained in the framework of local approximation. More correct estimate should give a lesser value, because of self-attraction of the core in middle part of the plate where two symmetrical exponential tails described by (12) meet one another and form an arc. Such a non-local effect is most essential just under the specific plate geometry. However, the non-local correction can not change the shape of the front vortex part where the non-local interaction is weak as compared with other forces. It is not hard to show that the latter requirement is satisfied if $|Z|>Z_0$, where $Z_0$ is the solution on equation \[ h\exp [-(D-\left| Z\right| )/\lambda ]\approx \sqrt{\lambda /2\pi \left| Z\right| }\exp (-2\left| Z\right| /\lambda ) \] If take into account that the more is $h$ the less is $Z_0$, then this equation yields $Z_0<D/3$ . Hence, at $|Z|>D/3 $ the mutual attraction of two core branches can be neglected, and the asymptotics (12) remains valid. This means that the maximally possible effect of non-locality is the replacing $D$ in the exponent by $\alpha D$ with $\alpha >2/3$ . Consequently, the lower bound for the stretching with confidence can be estimated as \[ \Delta X/\lambda >\frac \lambda D\exp (2D/3\lambda ) \] Thus, even in the worst case the non-local effects do not abolish the exponential character of stretching. For example, if $\lambda \approx 3\cdot 10^{-5}\,cm $ and $h$ equals to a few units, then even at $D\sim 20\lambda <10^{-3}\,cm$ one gets $\Delta X >m\lambda \exp(2D/3\lambda ) $ , with $m \sim 1 $ , i.e. $\Delta X > 1\,cm $ what exceeds a width of any realistic sample. Thus at first the vortex should form a ring whose shape approximately copies that of the sample cross-section. During this process the velocity of backward deepened core part is primarity determined by its distorsion which is created in the beginning of stretchening and thus has curvature radius of order of $D$ . Hence, this velocity is of order of $u_0\lambda /D$ , and at the moment when the ends will meet one another the displacement of most backward point will be yet as small as $\sim \lambda W/Dh<<W$ . \section{DISCUSSION AND RESUME} It seems clear that both the above conclusions can be extended to bulk current-carrying superconductors with another geometry, for instance, to round wires, if treat $2D$ and $W$ as minimal and maximal diameters of cross-section of the wire, respectively. Due to possibility of giant deformation and stretching of vortices, the thermodynamically nonequilibrium process of vortex penetration can promote formation of complicated many-vortex dynamical configurations which seem rather strange and unprofitable from the point of view of equilibrium thermodynamics. The presence of an external magnetic field parallel to transport current should lead to formation of spiral-like configuration instead of ring-like one and thus especially ensure the entangling of vortices. The Eq.4 enables us to describe this scenario in details, if choose Z-axis to be directed along the wire. Besides, the presence of weak pinning should amplifier the stretching of vortex and additionly complicate its shaping, because the motion of deepened part of vortex is characterized by relatively small forces of order of $\epsilon \lambda /D $ (much smaller than forces $\sim \epsilon $ what act on the end fragments) and so may be easily held back by pinning centers. We would like to underline the role of orthogonality boundary conditions. In the work [3] the equation was under use similar to our Eq.9, but boundary conditions was formulated in terms of the tension of core line. One can see from [3] that such conditions make it impossible to consider the case of high surface transport current $>H_{c1} $ corresponding to the elastic-like regime. To resume, we formulated the invariant equations of viscous motion of arbitrarily shaped 3D vortex lines, and applied them to careful analysis of the scenario of vortex penetration into a thick superconducting sample. As it was argued, the vortex-vortex interaction does not significantly affect the penetration process. But, of course, a full description of resistive state leads to more complicated tasks about vortex-vortex interactions deep inside the sample. \,\,\,\, ACKNOWLEDGEMENTS I would like to thank Dr. M.Indenbom, Dr. Yu.Genenko and Dr. A.Radievskiy for usefull discussions. \,\,\,\, REFERENCES 1. M.V.Indenbom, C.J.van der Beek, V.Berseth, W.Benoit, G.D'Anna, A.Erb, E.Walker and R.Flukiger, Nature, 1997, Feb.20 . 2. Yu.A.Genenko, Phys.Rev., B 49, 1994, 6950. 3. Chao Tang, Shechao Feng and L.Golubovich, Phys.Rev.Lett., 472, 1994, 1264. 4. L.P.Gorkov and N.B.Kopnin, Uspekhi fizicheskikh nauk, 116, 1975, 411 (transl. in English in Sov.Phys.-Usp., 1975). 5. Yu.E.Kuzovlev, Physica, C 292, 1997, 117. 6. V.P.Galaiko, Zh.Teor.Eksp.Fiz., 50 , 1966, 1322. \end{document}
\section{Introduction} Hybrid baryons are bound states of three quarks with an explicit excitation in the gluon field of QCD. The construction of (hybrid) baryons in a model motivated from the strong coupling expansion of the hamiltonian formulation of lattice QCD, the non--relativistic flux--tube model of Isgur and Paton \cite{paton85}, was detailed in ref. \cite{hadron}. This model predicts the adiabatic potentials of (hybrid) mesons at large interquark separations, as well as the mass of the $J^{PC}=1^{-+}$ hybrid meson, consistent with recent estimates from lattice QCD \cite{paton85,morning}. In ref. \cite{hadron} we studied the detailed flux dynamics and built the flux hamiltonian. We restrict our discussion to cases where the flux settles down in a Mercedez Benz configuration (as motivated by lattice QCD \cite{bali}). A minimal amount of quark motion is allowed in response to flux motion, in order to work in the centre of mass frame. Otherwise, we make the so--called ``adiabatic'' approximation, where the flux motion adjusts itself instantaneously to the motion of the quarks. The main result is that the lowest flux excitation can to a high degree of accuracy (about 5\%) be simulated by neglecting all flux--tube motions except the vibration of a junction. This result was obtained within the small oscillation approximation. The junction acquires an effective mass $M_{\mbox{\small eff}}$ from the motion of the remainder of the flux--tube and the quarks. The model is then simple: a junction is connected via a linear potential to the three quarks. The ground state of the junction motion corresponds to a conventional baryon and the various excited states to hybrid baryons. The junction can move in three directions, and correspondingly be excited in three ways, giving the hybrid baryons $H_1, H_2$ and $ H_3$. The junction motion is depicted in Fig. \ref{rough}. \begin{figure}[t] \vspace{0cm} \begin{centering} \epsfig{file=captalkfig1.ps,width=15cm,angle=0} \vspace{-0.2cm} \caption[x]{The junction connects strings coming from the three quarks. The vectors $\br$ and $\bl_i$ respectively point from the equilibrium position of the junction to its current position and the quark positions.} \plabel{rough} \end{centering} \end{figure} \vspace{0.3cm} The hamiltonian for the junction motion in the Mercedez Benz configuration is simply the kinetic energy of the junction added to the sum of the lengths from the junction to the quarks multiplied by the string tension $b$, \begin{eqnarray} \plabel{ham3} {H}_{\mbox{\small flux}}=\frac{1}{2}M_{\mbox{\small eff}}\; {\bf \dot{r}^2}+ b\sum_{i=1}^{3}|\bl_i-\br| \end{eqnarray} We shall be taking ansatz wave functions of the form \begin{equation} \plabel{psih} {\mbox{\boldmath $\hat{\eta}$}}_{-}\cdot\br \;\;\Psi_B(\br)\hspace{1cm} \end{equation} for $H_1$ hybrid baryons, where $\Psi_B(\br)$ is an exponential function. It is not difficult to show that ${\mbox{\boldmath $\hat{\eta}$}}_{-}$ lies in the plane spanned by the three quarks (the ``QQQ plane''). \section{Quantum numbers of low--lying hybrid baryons} {\bf Angular Momentum:} The hamiltonian in Eq. \ref{ham3} is not invariant under rotations in the junction position $\br$, with fixed quark positions. When the junction wave function, which is hence not an eigenfunction of angular momentum, is combined with the quark motion wave functions, which are eigenfunctions of angular momentum, it must be done in such a way that the total angular momentum of the junction and quark motion is well--defined. Obtaining a well--defined total angular momentum is a technically challenging problem that is an artifact of the adiabatic approximation, which separates junction and quark motion. We here merely give an intuitive argument why the total angular momentum $L$ of the $H_1$ baryons is expected to be 1. The hybrid baryon wave function is proportional to ${\mbox{\boldmath $\hat{\eta}$}}_-\cdot\br$, and since ${\mbox{\boldmath $\hat{\eta}$}}_-$ lies in the QQQ plane, it can be regarded as the x--axis, so that ${\mbox{\boldmath $\hat{\eta}$}}_-\cdot\br = \sqrt{\frac{2\pi}{3}} r (-Y_{11}({\bf \hat{r}}) + Y_{1-1}({\bf \hat{r}}))$ in terms of spherical harmonics. If the mathematics of conservation of angular momentum is followed through, it is found that if the angular momentum of the quark motion is $L_q=0$ (corresponding to the lowest energy quark motion states), then the total angular momentum projection just equals the angular momentum projection of the junction wave functions, which in this case is $\pm 1$. Hence the total angular momentum projection is $\pm 1$ so that $L$ cannot be zero, and should most likely be 1. {\bf Exchange symmetry:} Exchange symmetry transformations $S_{ij}$ exchange the positions of the quarks $\bl_i \leftrightarrow \bl_j$. Since the physics does not depend on the quark position labelling convention, the junction hamiltonian should be exchange symmetric, as can be seen explicitly in Eq. \ref{ham3}, noting that the junction position $\br$ is not determined by the positions of the quarks. We now argue that the junction wave functions of (hybrid) baryons should transform either totally symmetrically or totally anti--symmetrically under exchange symmetry. Since the hamiltonian is invariant under exchange symmetry we have the commutation relation $[H_{\mbox{\small flux}},S_{ij}]=0$. Combining this with the Schr\"{o}dinger equation \begin{equation}\plabel{ex} H_{\mbox{\small flux}}\Psi = V(l_1,l_2,l_3)\Psi \hspace{1cm} \mbox{gives} \hspace{1cm} H_{\mbox{\small flux}} (S_{ij}\Psi) = V(l_1,l_2,l_3) (S_{ij}\Psi)\end{equation} so that $S_{ij}\Psi$ is degenerate in energy with $\Psi$. Now since the baryon and each of the hybrid baryons $H_i$ have different energies (except when $l_1=l_2=l_3$) it follows that $S_{ij}\Psi$ must be a multiple of $\Psi$, i.e. that $S_{ij}\Psi = \varsigma \Psi$, where $\varsigma$ is complex number. Now note that the product of two exchange symmetry transformations is the identity, i.e. that \begin{equation} S_{ij}S_{ij} = 1 \hspace{1cm} \mbox{which implies that} \hspace{1cm} \varsigma^2 = 1 \end{equation} or $\varsigma=\pm 1$. Hence $S_{ij}\Psi = \pm \Psi$. Assume that $S_{12}\Psi = \varsigma\Psi$. We now show\footnote{This result also follows by noting that $[H_{\mbox{\small flux}},S_{ij}]=0$ implies that $\Psi$ must be an irreducible representation of the exchange symmetry group, i.e. totally symmetric, anti--symmetric or mixed symmetry. But since we already showed that $S_{ij}\Psi = \pm \Psi$, it follows that $\Psi$ is in either the totally symmetric or anti--symmetric irreducible representation.} that $S_{23}\Psi =S_{13}\Psi= \varsigma\Psi$, i.e. that $\Psi$ is either totally symmetric or totally anti--symmetric under label exchange. This follows by the two identities \begin{eqnarray} \lefteqn{S_{12}S_{23}S_{13}S_{23}= 1 \hspace{1cm} \mbox{which implies that} \hspace{1cm} S_{12}\Psi = S_{13}\Psi \nonumber } \\ & & \hspace{-0.7cm} S_{23}S_{12}S_{13}S_{12}= 1 \hspace{1cm} \mbox{which implies that} \hspace{1cm} S_{23}\Psi = S_{13}\Psi \end{eqnarray} For each of the hybrid baryons $H_i$, there are hence two varieties: the junction wave function is totally symmetric (S) or totally anti--symmetric (A) under quark label exchange, denoted by $H_i^S$ and $H_i^A$. {\bf Parity:} The inversion of all coordinates $\bl_i\rightarrow -\bl_i$ and $\br\rightarrow -\br$, called ``parity'', is a symmetry of the junction hamiltonian in Eq. \ref{ham3}. ${\mbox{\boldmath $\hat{\eta}$}}_{-}$ is a vector in the QQQ plane and is a linear combination of the $\hat{\bf l}_i$, which span the plane, with coefficients which are functions of $l_i$. The lengths $l_i$ remain invariant under parity. However, $\hat{\bf l}_i \rightarrow - \hat{\bf l}_i$ under parity. It follows that ${\mbox{\boldmath $\hat{\eta}$}}_{-}$ is {\it odd under parity}. The junction wave function in Eq. \ref{psih} is thus even under parity, since ${\mbox{\boldmath $\hat{\eta}$}}_{-}\rightarrow -{\mbox{\boldmath $\hat{\eta}$}}_{-}$ and $\br\rightarrow -\br$. For a low--lying hybrid the quark motion wave function is even under parity, so that the full hybrid baryon wave function has even parity. Since quarks are fermions, the wave function should be totally antisymmetric under quark label exchange, called the Pauli principle. Since our philosophy is that (hybrid) baryon dynamics is dominated by (non--perturbative) long distance physics, we consider the colour structure of the (hybrid) baryon to be motivated from the long distance limit, i.e. from the strong coupling limit of the hamiltonian formulation of lattice QCD \cite{paton85}. Here, the quarks are sources of triplet colour, which flows along the string connected to the quarks into the junction, where an $\epsilon_{ijk}$ neutralizes the colour. The colour wave function $\epsilon_{ijk}$ is hence totally antisymmetric under exchange of quarks for {\it both} the conventional and hybrid baryon. This imposes constraints on the combination of flavour and non--relativistic spin $S$ of the three quarks that is allowed. For a totally symmetric hybrid baryon junction wave function, the flavour--spin wave functions must be totally symmetric. This is because we are interested in the low--lying hybrid baryons which have the quark motion wave function in ground state, i.e. totally symmetric. If the flavour is $\Delta$, which is totally symmetric, this implies that the spin must be totally symmetric, i.e. $S=\frac{3}{2}$. Similarly for flavour $N$ the spin must be $\frac{1}{2}$. For a totally antisymmetric junction wave function, the flavour--spin wave function must be totally antisymmetric. For $\Delta$ flavour this implies that the spin must be totally antisymmetric, which is not realizable. Hence there are no $\Delta$ hybrid baryons with totally antisymmetric junction wave functions. The $N$ flavour is found to have spin $\frac{1}{2}$. The quantum numbers of the lowest--lying states that can be constructed on the $H_1$ adiabatic surface are indicated in Table \ref{tabqu}. The total angular momentum ${\bf J} = {\bf L} + {\bf S}$. Since $L=1$ for ground state $H_1$ hybrid baryon, $J=\frac{1}{2},\frac{3}{2}$ for $S=\frac{1}{2}$, and $J=\frac{1}{2},\frac{3}{2},\frac{5}{2}$ for $S=\frac{3}{2}$. These assignments are indicated in Table \ref{tabqu}. One notes from Table \ref{tabqu} that amongst the $H_1^S$ hybrid baryons, there are $N \frac{1}{2}^+$ and $\Delta \frac{3}{2}^+$ states which have identical quantum numbers to the conventional $N$ and $\Delta$ baryons. \begin{table}[t] \begin{center} \caption{\small Quantum numbers of low--lying hybrid baryons for the adiabatic surface $H_1$. In the absense of spin dependent forces all these states are degenerate. $N,\Delta$ are the flavour structure of the wave function (i.e. those of the conventional baryons $N,\Delta$ respectively) and $P$ the parity. } \label{tabqu} \begin{tabular}{|c||l|c|l|} \hline Hybrid Baryon & $L$ & $S$ & $(N,\Delta)^{2S+1}J^P$ \\ \hline $H_1^S $ & 1 & $\frac{1}{2},\frac{3}{2}$ & $N^2 {\frac{1}{2}}^+, \; N^2 {\frac{3}{2}}^+, \; \Delta^4 {\frac{1}{2}}^+, \; \Delta^4 {\frac{3}{2}}^+, \; \Delta^4 {\frac{5}{2}}^+$\\ $H_1^A $ & 1 & $\frac{1}{2}$ & $N^2 {\frac{1}{2}}^+, \; N^2 {\frac{3}{2}}^+$\\ \hline \end{tabular} \end{center} \end{table} It is interesting to compare our hybrid baryons to the predictions of the bag model. Out of all the states listed under $H_1^S$ and $H_1^A$ in Table \ref{tabqu}, only one pair of $N^2 {\frac{1}{2}}^+, \; N^2 {\frac{3}{2}}^+$ states have the same flavour, spin $S$, total angular momentum and parity as the low--lying hybrid baryons in the bag model \cite{bag}. In fact, for the $H_1^S$ hybrid baryons, the bag model swaps the $N$ and $\Delta$ flavours from our assignments, keeping other quantum numbers the same. Both our model and the bag model has seven low--lying hybrid baryons \cite{bag}. \section{Numerical estimate of the hybrid baryon mass} The difference between the hybrid and conventional baryon adiabatic potentials (or junction energies) as a function of quark positions, $V_{H_1}(l_1,l_2,l_3)-V_B(l_1,l_2,l_3)$, was determined numerically from the first part of Eq. \ref{ex} by using the hamiltonian in Eq. \ref{ham3}, and were displayed in ref. \cite{conf}. Now define the hybrid baryon potential as \begin{equation}\plabel{pot2} \mbox{\={V}}_{H_1} (l_1,l_2,l_3) \equiv \mbox{\={V}}_B (l_1,l_2,l_3) + V_{H_1} (l_1,l_2,l_3) - V_B (l_1,l_2,l_3) \end{equation} where $\mbox{\={V}}_B (l_1,l_2,l_3)$ is the phenomenologically successful relativized baryon hamiltonian with Coulomb and linear potential terms of ref. \cite{capstick86} (with spin--spin, spin--orbit and tensor interactions neglected); and the parameters are also those of ref. \cite{capstick86}. Note that the Coulomb interaction of the conventional and hybrid baryon is assumed to be identical. We solve the Schr\"{o}dinger equation for the hamiltonian in Eq. \ref{pot2} with 95 spin--space basis states incorporating $L_q =0,1,2$ harmonic oscillator wave functions for the $J=\frac{1}{2}$ case, i.e. construct 95 $\times$ 95 dimensional matrices. These matrices are subsequently diagonalized. The differences between the energies for the hybrid and the conventional baryon is then added to the experimental mass of the lowest baryon, taken as the spin--averaged mass of the $N$ and $\Delta$, i.e. 1085 MeV \protect\cite{pdg98}. The first three quark orbital excitations $L_q=0,1,2$ of hybrid baryons composed of up and down quarks are found to have masses 1976, 2341 and 2619 MeV respectively. Hence, for the lowest hybrid baryon level, with the quantum numbers in Table \ref{tabqu}, we obtain that $M_{H_1} - M_B = 891$ MeV, giving a mass estimate of $M_{H_1} = 1976$ MeV. This mass estimate is substantially higher than other mass estimates in the literature: $\sim 1.5$ GeV in the bag model \cite{bag} and $1.5\pm 10\%$ GeV in QCD sum rules \cite{zpli}. There are two crucial assumptions that were made in the early work on (hybrid) meson masses in the flux--tube model: the adiabatic motion of quarks and the small oscillation approximation for flux motion \cite{paton85}. It was later shown that when the adiabatic approximation is lifted, the masses goes up, and when the small oscillation approximation is lifted, the masses goes down \cite{paton85}. In our study of (hybrid) baryons we have partially lifted the adiabatic approximation by working in the centre of mass frame. We have fully lifted the small oscillation approximation. The effects on the masses of (hybrid) baryons when the various approximations are lifted are the same as those found for (hybrid) mesons. In our simulation, we obtain the average values $\sqrt{\langle\rho^2\rangle}=\sqrt{\langle\lambda^2\rangle} = 2.12,\; 2.52$ GeV$^{-1}$ for the low--lying baryon and $H_1$ hybrid baryon respectively. ${\langle\rho^2\rangle}={\langle\lambda^2\rangle}$ is expected since the spatial parts of the wave functions of the low--lying states are totally symmetric under exchange symmetry. The hybrid baryon is 20\% larger than the conventional baryon. \section{Phenomenology} The sign of the the Coulomb interaction is expected to be the same for both conventional and hybrid baryons \cite{bag}. This means that the hyperfine interaction has the same sign in both situations, so that the $\Delta$ hybrid baryons are always heavier than the $N$ hybrids. This implies that only four of the original seven low--lying baryons, the $N$ hybrids, are truely low--lying. We expect {\it a priori} the most phenomenologically interesting decay of the low--lying hybrid baryons to be the P--wave decay to $N\rho$ and $N\omega$, simply because the phase space is favourable and $\rho$ and $\omega$ are easily isolated experimentally. The $N\rho$ decay would be especially relevant to the electro-- and photoproduction of hybrid baryons at TJNAF via the vector meson dominated coupling of the photon to the $\rho$. Indeed, a search for excited $N^*$ resonances with mass $< 2.2$ GeV is currently underway in Hall B \cite{kees}. Given the mass estimate for the low--lying hybrid baryons, the detection of hybrid baryons in $N\rho$ or $N\omega$ is feasible at TJNAF. There are also planned experiments in $\pi N$ scattering by Crystal Ball E913 at the new D--line at Brookhaven with the capability of searching for states in $N\{\eta,\rho,\omega\}$, which would isolate states in the mass region $\sim 2$ GeV \cite{bris}. The decay $\psi\rightarrow p\bar{p}\omega$ has been observed with a branching ratio of $1.30\pm 0.25\; 10^{-3}$ and $\psi\rightarrow p\bar{p}\eta^{'}$ with branching ratio $9\pm 4\; 10^{-4}$ \cite{pdg98}. Since gluonic hadron production is expected to be enhanced above conventional hadron production in the glue--rich decay of the $\psi$, it is possible that a partial wave analysis of the $p\omega$ or $p\eta^{'}$ invariant masses would yield evidence for hybrid baryons. Future work at BEPC and an upgraded $\tau$--charm factory would be critical here. \section{Conclusions} The spin and flavour structure of the low--lying hybrid baryons have been specified, and differ from their structure in the bag model. Exchange symmetry constrains the spin and flavour of the (hybrid) baryon wave function. The orbital angular momentum of the low--lying hybrid baryon is argued to be unity, with the parity even, contrary to conventional baryons where $L=1$ would imply the parity to be odd. The low--lying hybrid baryon adiabatic potential and mass has been estimated numerically. The mass estimate is considerably higher than bag model and QCD sum rule estimates.
\section{Introduction} \label{sec:introduction} \indent\indent In this paper, we consider extensions of the bar-attendance problem introduced by Arthur \cite{arthur94} and simplified into a minority game by Challet and Zhang \cite{challet9798}. In its simplest form, the minority game mimics the internal dynamic of the exchange of one commodity. Agents are allowed to buy or sell this commodity at each time step. No attempt is made to model any external factors that influence the market. Here, we introduce symmetric and asymmetric three sided games as extensions of the minority game. In the symmetric three sided model, the agents have to choose between three identical sides at each time step. These three sides are trading with each other, agents on one side buying from the second side to sell to the third. This model mimics the cyclic trading of goods. If we group any two sides together and consider the trading between this imaginary group and the third side, the model reduces to a kind of minority game with an uneven distribution of the agents. Hence, the connection between this model and the minority game is very strong. In the asymmetric three sided model, the agents can buy or sell a commodity at each time step, but they can also be inactive, that is, they are allowed to miss a turn. In contrast to the symmetric model, the three choices are not equivalent, as being inactive appears as a compromise between buying and selling. This model can be thought of as an open minority game in the sense that the agents buying and selling are playing a minority game with a variable number of agents at each turn. In Sec. \ref{sec:themodels}, the minority game is briefly recalled and the two new three sided models are described in detail. In Sec. \ref{sec:symmetric}, the symmetric three sided model is numerically investigated, while in Sec. \ref{sec:asymmetric}, the asymmetric three sided model is investigated. Sec. \ref{sec:conclusions} presents a comparison between the minority game and the two three sided models, as well as our conclusions. \section{The models} \label{sec:themodels} \indent\indent In the minority game, an odd number $N$ of agents have to choose between two sides, $1$ or $2$, at each time step. An agent wins if he chooses the minority side. The record of which side was the winning side for the last $m$ time steps constitutes the history of the system. The agents analyze the history of the system in order to make their next decision. In the symmetric three sided model, a number of agents $N$ have to choose between three sides, $1$, $2$ or $3$, at each time step. $N$ is not a multiple of 3. The agents on side 1 buy from side 2 to sell to side 3, the agents on side 2 buy from side 3 to sell to side 1 and the agents on side 3 buy from side 1 to sell to side 2. This cyclic trading pattern is shown in Fig. 1. It is assumed that the profit or loss at a side is reflected in the difference between the number of agents they are selling to and the number of agents they are buying from. For instance, $N_3 -N_2$ is a measure of the profit of side 1. Agents choosing the side with the highest profit win and are rewarded with a point. Agents choosing the side with the lowest profit lose and consequently lose a point. Agents choosing the side with the intermediate profit neither lose nor gain a point. Agents strive to maximize their total number of points. \end{multicols} \begin{figure}[h] \centering\begin{picture}(5,5)(-2,1) \put(-0.2,4.9){$N_1$} \put(0,5){\circle{2}} \put(-1.932,1.9){$N_2$} \put(-1.732,2){\circle{2}} \put(1.532,1.9){$N_3$} \put(1.732,2){\circle{2}} \put(0.732,2){\vector(-1,0){1.364}} \put(-1.232,2.866){\vector(1,2){0.682}} \put(0.5,4.134){\vector(1,-2){0.682}} \put(-3.5,5){buying from 2} \put(-4,3){selling to 1} \put(-3,0.8){buying from 3} \put(1,0.8){selling to 2} \put(2,3){buying from 1} \put(1,5){selling to 3} \end{picture} \caption{Schematic representation of the symmetric three sided model. The arrows indicate the direction of the exchange. $N_i$ is the number of agents choosing side $i$. \label{fig1}} \end{figure} \begin{multicols}{2} In the asymmetric three sided model, a number of agents $N$ also have to choose between three sides, $1$, $2$ or $3$, at each time step. $1$ corresponds to selling, $2$ to doing nothing and $3$ to buying. The agents buying or selling are said to be active, while the agents doing nothing are said to be inactive. The agents choosing the smaller group among buyers and sellers win and are rewarded with a point. The agents choosing the larger group among buyers and sellers lose and they lose a point. The points of the inactive agents don't change. If there is the same number of buyers and sellers, the points of all the agents remain unchanged. However, the inactive agents are recorded as winners in the history of the system, on the grounds that they achieved the same result as the buyers and sellers, but without taking any risk. Again, agents strive to maximize their total number of points. \end{multicols} \begin{figure} \begin{minipage}[b]{.46\linewidth} \begin{center} \begin{tabular}{|c|c|c|} \hline History & $\sigma$ & $\sigma'$\\ \hline (1,1) & 3 & 2 \\ (1,2) & 1 & 3 \\ (1,3) & 2 & 2 \\ (2,1) & 2 & 1 \\ (2,2) & 3 & 3 \\ (2,3) & 1 & 1 \\ (3,1) & 3 & 1 \\ (3,2) & 2 & 2 \\ (3,3) & 3 & 1 \\ \hline \end{tabular} \end{center} \vfill \end{minipage} \hfill \begin{minipage}[b]{.46\linewidth} \noindent Table 1: The first column lists all the possible histories of the system for the last 2 time steps $(m=2)$. A strategy is a set of decisions for all the different possible histories. Two example strategies $\sigma$ and $\sigma'$ are shown in the second and third columns. \end{minipage} \end{figure} \begin{multicols}{2} In each model, the record of which side was the winning side for the last $m$ time steps constitutes the history of the system. For a given $m$, there are $3^m$ different histories. The 9 different histories for $m=2$ are listed in the first column of table 1. Every agent makes a decision for the next time step according to the history of the system. To be able to play, an agent must have a strategy that allows him to make a decision for any of the $3^m$ different histories. The second and third columns of table 1 list two possible sets of decisions, $\sigma$ and $\sigma'$, that we will call strategies. Each agent has at his disposal a fixed set of $s$ strategies chosen at random, multiple choices of the same strategy being allowed. At any one moment in time, the agent only uses one of these strategies to make a decision. To allow an agent to decide which strategy to use, every strategy is awarded points, which are called virtual points. The virtual points of a strategy are the points the agent thinks he could have earned had he played with this strategy. Hence, the virtual points are rewarded using the same scheme as the points given to the agents, the prediction of a strategy being compared to the actual decisions. A strategy predicting the winning side is awarded a virtual point, a strategy predicting the losing side loses a virtual point and a strategy predicting the third side does not gain or lose any points. In the asymmetric model, in the case of an equal number of buyers and sellers, the virtual points of all strategies remain unchanged. An agent always plays with the strategy with the highest number of virtual points. When more than one strategy has the highest number of virtual points, one of them is chosen at random. If we compare two strategies $\sigma$ and $\sigma'$ component by component, we see that for some histories they can make the same prediction and for others they can make different predictions. In the example in table 1, the decisions differ when the history is (1,1), (1,2), (2,1), (3,1) and (3,3). To consider this feature, we have to distinguish between the symmetric model and the asymmetric one. For the former, the three sides are equivalent and only the number of differences between the strategies can give a measure of the difference between two strategies in the strategy space. For the latter, there is a qualitative difference between the three sides. This qualitative difference should appear in the definition of the difference between strategies. Consider first the symmetric three sided model. As the three sides are equivalent, a geometrical representation should put them at the same distance from one another. A convenient measure of the differences between two strategies $\sigma$ and $\sigma'$ is \begin{equation} d_s = {1\over 3^m} \sum_{i=1}^{3^m} \delta (\sigma_i - \sigma'_i ) \label{eq:symmetric distance definition} \end{equation} where $\delta (0) = 1$, and $\delta (x) = 0$ otherwise. $d_s$ is defined as the distance between strategies in the symmetric model. This definition takes into account the geometrical structure of the strategy space, including the equivalence between the three sides. In the example of table 1, $d_s =5/9$. By definition, the symmetric distance is a number ranging from 0 to 1. As Eq. (\ref{eq:symmetric distance definition}) shows, the symmetric distance $d_s$ is defined as a sum of $3^m$ terms, which we label $d_s^{(i)}$'s. Each of these terms is equal to 0 with probability 1/3 or equal to 1 with probability 2/3. The average distance between two strategies is $\overline{d}_s = 2/3$, while the variance of the symmetric distance distribution is $\sigma^2_s = 2/3^{m+2}$. The symmetric distance between two strategies corresponds to the probability that these two strategies will give different predictions, assuming that all the histories are equally likely to occur. The symmetric distance corresponds to the distance defined in the minority game \cite{dhulst99}. Two strategies at $d_s = 0$ are correlated, two strategies at $d_s =2/3$ are uncorrelated and two strategies at $d_s = 1$ are anticorrelated. In the asymmetric three sided minority game, selling is just the opposite decision to buying while doing nothing is a compromise. Consequently, the normalized asymmetric distance between strategies $\sigma$ and $\sigma'$, \begin{equation} d_a = {1\over 3^m} \sum_{i=1}^{3^m} {| \sigma_i - \sigma'_i |\over 2} \label{eq:asymmetric distance definition} \end{equation} is a measure of the difference between the two strategies. $d_a$ is defined as the distance between strategies in the asymmetric model. This definition takes into account the fact that buying is more different from selling than it is from being inactive in an arbitrary way. In the example of table 1, $d_a =4/9$. By definition, the asymmetric distance is a number ranging from 0 to 1. As shown by Eq. (\ref{eq:asymmetric distance definition}), the asymmetric distance $d_a$ is defined as a sum of $3^m$ terms we label $d_a^{(i)}$'s. When the component of a strategy is equal to 2, this component can never give a $d_a^{(i)}$ equal to 1. In other words, the inactive side has no side at distance 1 from itself. Considering all the possibilities, the probability to find a $d_a^{(i)}$ of 0 is 1/3, of 0.5 is 4/9 and of 1 is 2/9. The average asymmetric distance between strategies is $\overline{d}_a = 4/9$, while the variance of the asymmetric distance distribution is equal to $\sigma^2_a = 11/3^{m+4}$. The interpretation of this asymmetric distance is ambiguous. In fact, the opposite to selling is buying, but the opposite to being inactive is being inactive. Hence, $d_a$ is not a measure of the probability that two strategies would give opposite decisions. Two correlated strategies are at $d_s = 0$ from each other, two uncorrelated strategies are at $d_a = 1/2$ from each other, but two anticorrelated strategies can be at $d_a = 0$ or $d_a = 1$ from each other. \section{Numerical results for the symmetric model} \label{sec:symmetric} \indent\indent In this section, we report on numerical investigations of the properties of the symmetric three sided model, interpreting the results using the symmetric distance. Fig. 2 presents a typical result for the time evolution of the attendance at one side. The simulation is for $N=101$ agents with $s=2$ strategies each and a memory of $m=3$. The result for the attendance at one side is very similar to the results of the minority game, the mean attendance being shifted to $N/3$ instead of $N/2$. Given an agent choosing one side, the average distance between the strategy used by this agent and the strategies used by the other agents is $\overline{d} = 2/3$. That is, around $2/3$ of the agents should choose one of the two other sides. Hence, the average attendance at one side is $N/3$. \end{multicols} \begin{figure}[h] \centerline{\epsfig{file=fig2.eps, width=15cm,height=8cm}} \caption{Numerical simulation of the attendance at one side for the symmetric three sided model. The choice for the parameters is $N=101$, $s=2$ and $m=3$.} \label{fig2} \end{figure} \begin{multicols}{2} The variance of the attendance at one side as a function of the size of the memory $m$ is presented at Fig. 3 for $N=101$ agents with $s=2$ strategies. The result is again very similar to the minority game, with a very high variance for $m<3$, a minimum at $m=3$ and the variance going to $2N/9$, the random value, as $m$ goes to infinity. Curves of the same shape are obtained for the variance of the number of winners or the variance of the number of losers. Also, the maximum profit or the number of agents on the more crowded side exhibit the same behaviour. Each of these curves has a minimum for $m=3$. For $m<3$, the number of strategies used at each time step is a representative sample of the space of the strategies. Consequently, the variance of the attendance is directly related to the variance of the distance distribution, $\sigma^2_s = 2/3^{m+2}$. In fact, the variance of the attendance scales like $1/ \sigma^2_s$. On the contrary, for $m>3$, the space of the strategies is very large, so that most of the strategies used are uncorrelated. As a result, the kinetics of the system are the same as the kinetics of a random walk. Between these two behaviours, for $m$ around 3, the agents organize themselves better, a crowd-anticrowd effect being obtained \cite{johnson98-1}. \end{multicols} \begin{figure}[h] \centerline{\epsfig{file=fig3.eps, width=15cm,height=8cm}} \caption{Variance of the number of agents at one side as a function of the size of the memory $m$ ($\times$) for $N=101$ and $s=2$ (symmetric model). The analytical results using the calculation of Johnson {\it et al.} [4] is also shown ($\Box$).} \label{fig3} \end{figure} \begin{multicols}{2} Even for small values of $m$, the space of strategies is very large, of size $3^{3^m}$. But as in the minority game, not all the strategies are uncorrelated. If we suppose that $1/ \sigma^2_s$ gives an estimate of the number of uncorrelated strategies, the method of Johnson {\it et al.} \cite{johnson98-1} can be used to find an analytical expression for the variance of the attendance at one side. We followed the original calculation in \cite{johnson98-1}, with a size $a = 3^{m+2}/2$ for the space of strategies and a variance of 2/9 for an independent agent. The analytical result obtained by this method is compared in Fig. 3 to the result of the numerical simulations. The curves agree qualitatively. \end{multicols} \begin{figure} \begin{tabular}{cc} \begin{picture}(6,2)(0,0) \epsfig{figure=fig4a.eps,width=0.5\linewidth} \end{picture} & \begin{picture}(8,2)(0,0) \epsfig{figure=fig4b.eps,width=0.5\linewidth} \end{picture} \end{tabular} \caption{(a) Average points earned by the agents and their strategies as a function of the time for $N=101$, $s=2$ and $m=3$. The ordinate on the left refers to the points of the agents (bold line) while the ordinate on the right refers to the points of their strategies (simple line). (b) Profit rate of the strategies as a function of $m$ for $N=101$ and $s=2$.} \label{fig4} \end{figure} \begin{multicols}{2} Fig. 4 (a) presents a typical result for the average number of points given to the agents and their strategies. The parameters of the simulation are $N=101$, $s=2$ and $m=3$. Note that there are two different ordinate scales. As Fig. 4 (a) shows, the virtual points are steadily decreasing with time. In contrast, the points given to the agents display a more complex behaviour. The points given to the agents increase very slowly for $m<3$ and then oscillate around 0 for $m>3$. There seem to be no special behaviour for $m=3$. The time evolution of the virtual points can be approximated by a linear relation with a profit rate $\tau$. We define $\tau$ as the average number of points earned by a strategy at each time step. Fig. 4 (b) presents $\tau$ as a function of the memory $m$ for $N=101$ and $s=2$. For $m<3$, the strategies are slowly losing points, the worst results being obtained for $m=3$. For $m>3$, the virtual points oscillate around 0. Hence, the agents seem to be able to choose their strategy efficiently, in the sense that the strategies they choose win more often than the average strategy. This behaviour is to be contrasted with the minority game where the agents are not able to choose a strategy efficiently. As a summary, the symmetric model is a direct extension of the minority game to three sides. The results found are very similar, with a glassy phase transition \cite{savit97} when the size of the memory of the agents is increased. We numerically identified a critical value $m_c$ for the size of the memory. For $m<m_c$, the space of strategies is crowded and its geometrical structure is apparent in the results. As this structure is encoded in the distance definition, the system is driven by its distance distribution . For $m>m_c$, the number of strategies used is not relevant as most of the strategies used are uncorrelated. The kinetics of the system reduce to agents choosing one of the three sides at random. Hence, there is a transition from a system driven by its distance distribution to a random system. \section{Numerical results for the asymmetric model} \label{sec:asymmetric} \indent\indent We investigated numerically the different properties of the asymmetric three sided game. In the figures, 1 denotes buying, 2, doing nothing and 3, selling. The attendance of the three different sides as a function of the size of the memory $m$ is plotted at Fig. 5 for $N=101$ agents, playing with $s=2$ strategies each. The number of agents in the winning side is also presented. For small $m$ values, most of the agents are buying or selling (the two superimposed upper curves). Just a few of them are doing nothing (the lower curve for small $m$ values). As the size of the memory is increased, the system corresponds more and more to the agents guessing at random between the three possibilities. Also, for small values of $m$, the number of winners is significantly more than 1/3, the random guess value. Fig. 5 is interesting because the difference between the three sides is clearly apparent. \end{multicols} \begin{figure}[h] \centerline{\epsfig{file=fig5.eps, width=15cm,height=8cm}} \caption{Numerical result for the average attendances at the three sides as functions of $m$ for $N=101$ and $s=2$. The two upper curves ($\triangle$ and $\bigcirc$) are for the selling and buying options. The lower curve ($\times$) is for the inactive agents. The average number of winners is also presented ($\Box$) as a function of $m$ for the same choice of parameters.} \label{fig5} \end{figure} \begin{multicols}{2} In Fig. 6, the variance of the attendances at the three sides and the variance of the number of winners are presented as functions of $m$ for $N=101$ and $s=2$. For $m<6$, the variance of the number of inactive agents is significantly higher than $2N/9$, the value for agents guessing at random. The variances of the number of buyers and sellers has a minimum at $m=2$. The variances of the three sides increase to $2N/9$ as $m$ increases. Hence, there seems to be an organization of the agents for $m$ around 2. The variance of the number of winners has a shape very similar to the one found in the minority game. For small value of $m$, the variance diverges like a power law of $m$; at $m\simeq 7$, it seems to reach a minimum and for higher values of $m$, it goes asymptotically to a value near $N/9$. However, the existence of a minimum at $m=7$ could not be confirmed unequivocally by the numerical simulations. \end{multicols} \begin{figure}[h] \centerline{\epsfig{file=fig6.eps, width=15cm,height=8cm}} \caption{Variances of attendances at the three sides as functions of $m$ for $N=101$ and $s=2$. $\triangle$ and $\bigcirc$, selling and buying options, $\times$ inactive agents. $\Box$ is the variance of the number of winners as a function of $m$ for the same choice of parameters.} \label{fig6} \end{figure} \begin{multicols}{2} Fig. 5 and 6 show that for small $m$ values, the behaviour of the system is directly related to the properties of the distance distribution. The proportion of people buying or selling is of the same order as the average distance, 4/9, while the variance of the number of winners scales as $1 /\sigma^2_a$, the inverse of the variance of the asymmetric distance distribution. These properties were also present in the minority game. In this asymmetric model, the variance of the attendance at one side does not represent the wasted number of points. The wasted number of points is defined to be the difference between the maximum points that can be earned by the system at each time step and the average points actually earned by the system at each time step. This is why we also have to consider the properties of the number of winners in addition to the properties of the attendances. For higher values of $m$, the strategy space is so large that most of the used strategies are uncorrelated. The system is similar to a system with agents choosing at random from the three sides. In the minority game, the relative attendance predicted by the distance distribution is the same as the one predicted by random guesses, that is 1/2. In the present three sided minority game, these two ratios are 4/9 and 1/3 respectively. Hence, the transition from a system driven by the distance distribution to a system of agent guessing at random is seen directly in the attendance of the different sides. Fig. 7 presents the average success rate of one side, that is, the probability that at any one moment in time, one side will win. As expected, the sides corresponding to buying and selling are symmetric and more likely to win than the inactive side. In fact, there are $(N+1)(N+2)/2$ different configurations for the attendances of the 3 sides. Among these, only $(N+2)/2$ make the inactive agents winners if $N$ is even, $(N+1)/2$ if $N$ is odd. Hence, if all the situations were equally likely to occur, the inactive agents would win at most about every $N+1$ time steps. This is the order of the asymptotical value for the success rate of this side. For low values of $m$, the success rate of the inactive side is higher than the asymptotical value, implying that the agents playing are organizing themselves rather well. The transition between organized and non-organized agents is for $m=2$ in Fig. 7. \end{multicols} \begin{figure}[h] \centerline{\epsfig{file=fig7.eps, width=15cm,height=8cm}} \caption{Success rates of the three different sides as functions of $m$ for $N=101$ and $s=2$. The two upper curves ($\triangle$ and $\bigcirc$) are for selling and buying while the lower curve ($\times$) is for the inactive agents.} \label{fig7} \end{figure} \begin{multicols}{2} Fig. 8 confirms the organization of the agents. The profit rates of the agents and their strategies are shown as functions of $m$. We define a profit rate as the average number of points earned at each time step. For values of $m$ less than $m=5$, the agents are able to choose strategies which are more successful than the average ones. On the contrary, for $m>5$, they are doing worse than guessing at random. The curve of the profit rate of the strategies suggests that the transition takes place for $m=2$. \end{multicols} \begin{figure}[h] \centerline{\epsfig{file=fig8.eps, width=15cm,height=8cm}} \caption{Average profit rates of the agents and their strategies as functions of $m$ for $N=101$ and $s=2$. $\Box$ is for the profit rate of the agents while $\bigcirc$ is for the profit rate of their strategies.} \label{fig8} \end{figure} \begin{multicols}{2} As a summary, in the asymmetric three sided minority game, agents playing with a small memory win more points on average than agents playing with a bigger memory in a pure population, that is, a population of all the agents with the same memory size $m$. As in the minority game and the symmetric three sided model, a glassy phase transition \cite{savit97} is found at a particular value of the memory $m_c$. For $m<m_c$, the geometrical properties of the space of strategies is apparent, especially the asymmetry between the three sides. Most of the agents are playing and the system is driven by the distance distribution. In contrast to the minority game, this property is seen directly in the number of agents on each side. For $m>m_c$, the strategies used are uncorrelated and the system is similar to a system of agents guessing at random. Considering the adaptation of the agents, they are unable to realize that the wiser choice is to decide to be inactive. In fact, more than half of the active agents will lose. The agents are fooled because they base their confidence in virtual points, not on their profit. Hence, the agents are always tempted to play even if they are unlikely to win. \section{Conclusions} \label{sec:conclusions} \indent\indent We introduced two three sided models as extensions of the minority game. In the symmetric three sided model, agents are given three equivalent choices while in the asymmetric three sided model, agents have the opportunity to miss a turn and not play. We have investigated these two new models numerically and compared the results with the original minority game. In both models, we defined a distance between the strategies of the agents. These distances incorporate in their definitions the geometrical structure of the space of strategies. In the symmetric model, the geometrical structure of the space of strategies is very similar to the one in the minority game. The distance gives a measure of the correlation between two strategies. Conversely, the distance in the asymmetric model has no obvious interpretation. A transition between a system driven by the distance distribution and a system of agents guessing at random was identified numerically in both models. However, in contrast to the minority game, the agents make their highest profit for $m$ small and not at the transition value of $m$. In the distance driven phase, the agents organize themselves, as in the minority game. In contrast to the minority game, however, the average profit rate of the agents is higher than the average profit rate of the strategies, indicating that the agents are choosing their strategies efficiently. In the symmetric model, the transition is apparent in the variance of the number of agents choosing one side while in the asymmetric model the transition is seen in the number of agents itself. This latter property of the asymmetric model is a direct consequence of the geometrical structure of the space of strategies. In the future, we intend to investigate both models analytically. The symmetric model, in particular, should be amenable to analytical treatment, perhaps following the methods introduced in \cite{challet99} for the two sided minority game.
\section{Introduction} $B$ meson decays with non-charmed final states will play an important role in the detailed investigations at future $B$--factories. Among these processes the ones involving the quark transitions $b \to u \ell \bar{\nu}_\ell$ and $b \to s \gamma$ are of interest with respect to the determination of $V_{ub}$ and $V_{ts}$ as well as to discover or constrain effects of physics beyond the Standard Model (SM). From the theoretical side a lot of progress has been made by employing an expansion in inverse powers of the heavy quark mass $m_b$. Using operator product expansion (OPE) and the symmetries of Heavy Quark Effective Theory (HQET) \cite{HQET} the nonperturbative uncertainties can be reduced to a large extent in inclusive decays \cite{inclusive}. Concerning the transitions mentioned above the inclusive semileptonic or radiative processes such as $B \to X_u \ell \bar{\nu}_\ell$ and $B \to X_s \gamma$ are the ones which we shall address in the present paper. Looking at decay spectra of the leptons and photons in these processes, it has been pointed out that in certain regions of phase space (such as the endpoint region of the lepton energy spectrum in $B \to X_u \ell \bar{\nu}_\ell$ or the photon energy spectrum in $B \to X_s \gamma$ close to the endpoint of maximal energy) the correct description requires more than the naive $1/m_b$ expansion \cite{Neubert,Bigietal,MannelNeubert}. In these endpoint regions it is required to sum the leading twist contributions into a non-perturbative ``shape'' function, which describes the distribution of the light-cone component of the $b$ quark residual momentum inside the $B$ meson. Unfortunately, there is no way to avoid these endpoint regions due to experimental cuts. In $B \to X_u \ell \bar{\nu}_\ell$ cuts on the lepton energy and/or on the hadronic invariant mass of the final state are required to suppress the much larger charm contribution \cite{BaBarBook,Uraltsev,Falk}. Likewise, in $B \to X_s \gamma$ a cut on the photon energy is mandatory to reduce the background from ordinary bremsstrahlung in inclusive $B$ decays \cite{bsgammaexp}. These cuts more or less reduce the accessible part of phase space to the endpoint regions. Hence it is unavoidable to get a theoretical handle on this light-cone distribution function. The distribution function is universal, since it depends only on the properties of the initial-state $B$ hadron. This fact may be used to establish a model independent relation between the two inclusive decays $B \to X_u \ell \bar{\nu}_\ell$ and $B \to X_s \gamma$ \cite{Neubert1}. Parametrizations of this function have been proposed, which include the known features such as the few lowest moments \cite{BaBarBook,MannelNeubert}. It has also been shown that the popular ACCMM model \cite{ACCMM} for a certain range of its parameters is indeed consistent with this QCD based approach \cite{BigiUraltsev}. There have been various suggestions how to overcome the non-perturbative uncertainties induced by the distribution function. One way is to consider moments of appropriate distributions \cite{Neubert,Bigietal,MannelNeubert,Bauer,WiseLigeti}. The first few moments are then sensitive only to the first few moments of the distribution function which are known. However, due to experimental cuts only parts of the distributions can be measured such that only moments involving cuts can be obtained. Depending on the cut, these quantities are sensitive to large moments of the distribution function and in this way the non-perturbative uncertainties reappear. In the present paper we propose a different approach. We suggest to directly compare the light-cone spectra of the final state hadrons in $B \to X_u \ell \bar{\nu}_\ell$ and $B \to X_s \gamma$. The individual rates depend on the shape function while the ratio of the two spectra is not very sensitive to non-perturbative effects even if cuts are included. We discuss the kinematic variables which allow a direct comparison of the two processes and consider the radiative corrections entering their relation. In the next section we give a derivation of the non-perturbative light cone distribution. Based on this we calculate in section 3 the perturbative corrections and combine these with the non-perturbative light cone distribution function. Finally we study the comparison between $B \to X_u \ell \bar{\nu}_\ell$ and $B \to X_s \gamma$ quantitatively and conclude. \section{Non-perturbative Contributions: Light cone distribution function of the heavy quark} Although the derivation of how the light-cone distribution function emerges in the heavy-to-light transitions at hand is well known \cite{Neubert,MannelNeubert,Bigietal}, we consider it useful to rederive it here, since our discussion of radiative corrections will be based on this derivation. We are going to consider heavy to light transitions such as $b \to u$ and $b \to s$ decays. The relevant effective Hamiltonian is obtained from the Standard Model by integrating out the top quark and the weak bosons. The dominant QCD corrections are taken into account as usual by running this effective interaction down to the scale of the bottom quark. The corresponding expressions are known to next-to-leading order accuracy \cite{Buras,AliGreub,ChetMM} and one obtains for the relevant pieces \begin{eqnarray} && H_{eff}^{\rm sl} = \frac{G_F}{\sqrt{2}} (\bar{b} \gamma_\mu (1-\gamma_5) u)(\bar{\nu} \gamma_\mu (1-\gamma_5) \ell) \\ && H_{eff}^{rare} = \frac{G_F}{\sqrt{2}} \frac{e}{8 \pi^2} C_7 (\mu) m_b (\mu) (\bar{b} \sigma_{\mu \nu} (1-\gamma_5) s)|_\mu F^{\mu \nu} \end{eqnarray} mediating semileptonic $b \to u$ transitions and radiative $b \to s$ decays respectively. $C_7 (\mu)$ is a Wilson coefficient, depending on the renormalization scale $\mu$, and $F^{\mu \nu}$ is the electromagnetic field strength. In the following we shall consider only these two contributions; at next-to-leading order (which is the accuracy we need here) there are also contributions of other operators in the effective Hamiltonian. However, these contributions are small and can be neglected here \cite{WiseLigeti}, although they should be included when analyzing experimental data. Considering the inclusive $B$ decays mediated by these two effective interactions, we shall look at the generic quantity \begin{equation} R = \sum_X (2\pi)^4 \delta^4 (p_B - p_X - q) \langle B(p_B)| \bar{b} \Gamma q | X (p_X) \rangle \langle X (p_X)| \bar{q} \Gamma^\dagger b | B(p_B) \rangle \end{equation} where $q$ is the momentum transferred to the non-hadronic system and $\Gamma$ is either $\gamma_\mu (1-\gamma_5)$ or $\sigma_{\mu \nu}$. We first redefine the phase of the heavy quark field \begin{equation} b(x) = \exp(-im_b v \cdot x) b_v (x) \end{equation} where $v$ is the velocity of the decaying $B$ meson \begin{equation} \label{velo} v = \frac{p_B}{M_B} \, ; \end{equation} this corresponds to a splitting of the $b$ quark momentum according to $p_b = m_b v + k$, where $k$ is a small residual momentum. Going through the usual steps, we can rewrite $R$ as \begin{equation} R = \int d^4 x \exp(-ix[m_b v - q]) \langle B(p_B)| \bar{b}_v (0) \Gamma q(0) \bar{q} (x) \Gamma^\dagger b_v(x) | B(p_B) \rangle \end{equation} The momentum $P=m_b v -q$ is the momentum of the final state partons, which is considered to be large compared to to any of the scales appearing in the matrix element. This allows us to set up an operator product expansion (OPE). If we assume \begin{equation} (m_b v - q)^2 = {\cal O}(m^2) \mbox{ and } (m_b - v q) = {\cal O}(m) \end{equation} the OPE is a short distance expansion yielding the usual $1/m_b$ expansion of the rates. Close to the endpoint, where $P=m_b v - q$ is a practically light-like vector, which has still large components, i.e. \begin{equation} \label{kinema} (m_b v - q)^2 = {\cal O}(\Lambda_{QCD} m_b) \mbox{ and } (m_b - v q) = {\cal O}(m_b) \end{equation} one has to switch to a light cone expansion very similar to what is known in deep inelastic scattering. In order to study the latter case in some more detail, it is useful to define light-cone vectors as \begin{equation} n_\pm = \frac{1}{\sqrt{(vP)^2 - P^2}} \left[ v \left(\sqrt{(vP)^2 - P^2} \mp (vP)\right) \pm P \right] \quad n_\pm^2 = 0 \end{equation} which satisfy the relations $n_+ n_- = 2$, $vn_\pm = 1$ and $2v = n_+ + n_-$. These vectors allow us to write \begin{equation} P = \frac{1}{2} ( P_+ n_- + P_- n_+ ) \qquad P_\pm = P n_\pm \end{equation} where \begin{eqnarray} && P_+ = (vP) - \sqrt{(vP)^2 - P^2} \to - \frac{P^2}{2 (vP)} \mbox { for } P^2 \ll (vP)^2 \\ && P_- = (vP) + \sqrt{(vP)^2 - P^2} \to 2 (vP) \mbox { for } P^2 \ll (vP)^2 \end{eqnarray} The kinematic region in which the light cone distribution function becomes relevant is the one where $P_+$ is much smaller than $P_-$: Here the main contributions to the integral come from the light cone $x^2 \approx 0$, since in order to have a contribution to the integral we need to have \begin{equation} x\cdot P = \frac{1}{2} (x_- P_+ + x_+ P_-) = x_+ (vP) + \mbox{ const} < \infty \end{equation} in the limit in which $ P_- \to \infty$. Hence $x_+$ has to be small, restricting the integration to the light cone. We consider first the tree level contribution, which corresponds to a contraction of the light quark line; we get \cite{MannelNeubert} \begin{eqnarray} R &=& \int d^4 x \int \frac{d^4 Q}{(2\pi)^4} \Theta(Q_0) (2\pi) \delta(Q^2) \exp(-ix[m_b v - q - Q]) \\ \nonumber && \langle B(v)| \bar{b}_v (0) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger b_v(x) | B(v) \rangle \end{eqnarray} where now $|B(v)\rangle$ is the static $B$ meson state. Performing a (gauge covariant) Taylor expansion of the remaining $x$ dependence of the matrix element, we get \begin{eqnarray} \label{taylor} R &=& \int d^4 x \int \frac{d^4 Q}{(2\pi)^4} \Theta(Q_0) (2\pi) \delta(Q^2) \exp(-ix[m_b v - q - Q]) \\ \nonumber && \sum_n \frac{1}{n!} \langle B(v)| \bar{b}_v (0) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger (-ix \cdot iD)^nb_v(0) | B(v) \rangle \\ \nonumber &=& \int d^4 x \int \frac{d^4 Q}{(2\pi)^4} \Theta(Q_0) (2\pi) \delta(Q^2) \\ \nonumber && \langle B(v)| \bar{b}_v (0) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger {\cal P} \exp(-ix[m_b v - q - Q + iD]) b_v(0) | B(v) \rangle \end{eqnarray} where the symbol ${\cal P}$ means the usual path--ordering of the exponential. Using spin symmetry and the usual representation matrices of the $0^-$ B meson states, the matrix element which appears in (\ref{taylor}) becomes \begin{eqnarray} && \langle B(v)| \bar{b}_v (0) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger (iD_{\mu_1}) (iD_{\mu_2}) \cdots (iD_{\mu_n}) b_v(0) | B(v) \rangle \\ \nonumber && = \frac{M_B}{2} \mbox{Tr}\{\gamma_5 (\@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{v}+1) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger (\@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{v}+1)\gamma_5 \} [a_1^{(n)} v_{\mu_1} v_{\mu_2} \cdots v_{\mu_n} + a_2^{(n)} g_{\mu_1 \mu_2} v_{\mu_3} \cdots v_{\mu_n} + \cdots ] \end{eqnarray} where the elipses denote terms with one or more $g_{\mu \nu}$'s and also antisymmetric terms and $a_i^{(n)}$ are non-pertrubative parameters. Contracting with $x^{\mu_1} x^{\mu_2} \cdots x^{\mu_n}$ all antisymmetric contributions vanish; furthermore, since the relevant kinematics restricts the $x_\mu$ to be on the light cone, also all the $g_{\mu \nu}$ terms are suppressed relative to the first term, which has only $v_\mu$'s. Hence \begin{equation} \label{term} \langle B(v)| \bar{b}_v (0) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger (-ix \cdot iD)^n b_v(0) | B(v) \rangle = M_B \mbox{Tr}\{(\@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{v}+1) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger \} a_1^{(n)} (v\cdot x)^n \end{equation} In this way we have made explicit that the kinematics we are studying here forces us to resum the series in $1/m_b$. Defining the twist $t$ of an operator ${\cal O}$ in the usual way $t = dim[{\cal O}] - \ell$, where $\ell$ is the spin of the operator, we find that the resummation corresponds to the contributions of leading twist, $t=3$. Since $x_\mu$ is light like, it projects out only the light cone component $D_+$ of the covariant derivative in (\ref{term}). Hence we may write $a_1^{(n)}$ as \begin{equation} 2 M_B a_1^{(n)} = \langle B(v)| \bar{b}_v (iD_+)^n b_v | B(v) \rangle \end{equation} and one obtains as a final result \begin{eqnarray} R &=& \int d^4 x \int \frac{d^4 Q}{(2\pi)^4} \Theta(Q_0) (2\pi) \delta(Q^2) \frac{1}{2} \mbox{Tr}\{(\@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{v}+1) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger \} \\ \nonumber && \langle B(v)| \bar{b}_v \exp(-ix[(m_b+iD_+) v - q - Q]) b_v | B(v) \rangle \end{eqnarray} Introducing the shape function (or light cone distribution function) as \cite{Neubert,MannelNeubert,Bigietal} \begin{equation} 2 M_B f(k_+) = \langle B(v)| \bar{b}_v \delta(k_+ - iD_+) b_v | B(v) \rangle \end{equation} we can write the result as \begin{eqnarray} \label{final} R &=& \int dk_+ f(k_+) \int \frac{d^4 Q}{(2\pi)^4} \Theta(Q_0) (2\pi) \delta(Q^2) \\ \nonumber && M_B \mbox{Tr}\{(\@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{v}+1) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger \} (2 \pi)^4 \delta^4 ([m_b + k_+]v -Q - q) \end{eqnarray} This result shows that the leading twist contribution is obtained by convoluting the partonic result with the shape function, where in the partonic result the $b$ quark mass $m_b$ is replaced by the mass \begin{equation} \label{mstar} m_b^* = m_b + k_+ \end{equation} Let us take a closer look at the kinematics. The final-state quark is massless, $$ 0 = ([m_b+k_+] v - q)^2 $$ in which case we have for the light cone component of the heavy quark residual momentum\footnote{% There are actually two solutions, but only one vanishes at the endpoint} \begin{equation} \label{kplus} k_+ = -m_b + (vq) + \sqrt{(vq)^2 - q^2} \end{equation} Note that this variable -- up to a minus sign -- is just $P_+$ and thus close to the endpoint \begin{equation} k_+ \approx - \frac{P^2}{2(vP)} \end{equation} If we now look at the hadronic kinematics \begin{equation} M_B v - q = p_X \, , \end{equation} where $p_X$ is the four momentum of the final state hadrons, and use the relation between the heavy quark mass and the meson mass \begin{equation} \label{mass} M_B = m_b + \bar{\Lambda} + {\cal O}(1/m_b) \end{equation} we find \begin{equation} p_X = P + \bar{\Lambda} v \quad \mbox{ or } \quad p_{X\pm} = P_\pm + \bar{\Lambda} \end{equation} Thus the light cone component of the hadronic momentum of the final state \begin{equation} L_+ = - p_{X+} = - M_B + (vq) + \sqrt{(vq)^2 - q^2} \end{equation} ranging between $-M_B \le L_+ \le 0$ is directly related to the light-cone component of the residual momentum of the heavy quark \begin{equation} L_+ = k_+ - \bar{\Lambda} \end{equation} where the range of $k_+$ is given by $-m_b \le k_+ \le \bar{\Lambda}$. However, large negative values of $k_+$ close to $-m_b$ are beyond the validity of the heavy mass limit. The observable $L_+$ directly measures the light cone component of the residual momentum of the heavy quark, at least for small values of $L_+$. In the case of $b \to s \gamma$ we have $q^2 = 0$ and $L_+$ is directly related to the energy $E_\gamma$ of the photon in the rest frame of the decaying $B$ $$ L_+ = - M_B + 2 E_\gamma . $$ For $b \to u \ell \bar{\nu}_\ell$ a reconstruction of $L_+$ requires both a measurement of the hadronic energy and the hadronic invariant mass. Reexpressing the convolution (\ref{final}) in terms of $L_+$ and using that the light-cone distribution function is non-vanishing only for $-m_b \to - \infty < k_+ < \bar{\Lambda}$ we may rewrite (\ref{final}) as \begin{eqnarray} \label{final1} R &=& \int\limits_{-M_B}^0 dL_+ f(L_+ +\bar{\Lambda}) \int \frac{d^4 Q}{(2\pi)^4} \Theta(Q_0) (2\pi) \delta(Q^2) \\ \nonumber && M_B \mbox{Tr}\{(\@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{v}+1) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger \} (2 \pi)^4 \delta^4 ([M_B + L_+]v -Q - q) \end{eqnarray} where we have made use of (\ref{mass}). Equivalently, we may write the spectrum in the variable $L_+$ as \begin{eqnarray} \label{final2} \frac{dR}{dL_+} &=& f(L_++\bar{\Lambda}) \int \frac{d^4 Q}{(2\pi)^4} \Theta(Q_0) (2\pi) \delta(Q^2) \\ \nonumber && M_B \mbox{Tr}\{(\@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{v}+1) \Gamma \@ifnextchar[{\fmsl@sh}{\fmsl@sh[0mu]}{Q} \Gamma^\dagger \} (2 \pi)^4 \delta^4 ([M_B + L_+]v -Q - q) \end{eqnarray} The two relations (\ref{final1}) and (\ref{final2}) exhibit an interesting feature of the summation of the leading twist terms, namely that the dependence on the heavy quark mass has completely disappeared, only the shape function $f$ still depends on $\bar\Lambda$. The full result should be independent of this unphysical quantity and hence the explicit dependence on $\bar{\Lambda}$ of the shape function has to cancel against the one appearing in the argument of the shape function. In other words, if we write the shape function as $f = f(k_+,\bar{\Lambda})$, this function of two variables has to satisfy \begin{equation} \label{Labindependence} \left(\frac{\partial}{\partial k_+} + \frac{\partial}{\partial \bar{\Lambda}} \right) f(k_+,\bar{\Lambda}) = 0 \quad \mbox{ or } \quad f(k_+,\bar{\Lambda}) = g(k_+ - \bar{\Lambda}) \end{equation} In this way the total result becomes independent of any reference to the quark mass $m_b$ or equivalently on $\bar{\Lambda}$. In particular, (\ref{Labindependence}) ensures the cancellation of ambiguities related to the renormalon in the heavy quark mass \cite{BigiUraltsev}. Putting everything together, the light-cone spectrum (\ref{final2}) for the decay $B \to X_u \ell \bar{\nu}_\ell$ becomes \begin{equation} \label{final3} \frac{d \Gamma}{dL_+} (B \to X_u \ell \bar{\nu}_\ell) = \frac{G_F^2 (M_B + L_+)^5 |V_{ub}|^2}{192 \pi^3} g(L_+) \end{equation} which means that the spectrum is proportional to the total rate in which the $m_b$ has been replaced by $M_B + K_+$. Similarly, for $B \to X_s \gamma$ one finds \begin{equation} \label{final4} \frac{d \Gamma}{dL_+} (B \to X_s \gamma)= \frac{G_F^2 (M_B + L_+)^5 |V_{tb} V_{ts}^*|^2 \alpha_{em} }{32 \pi^4} [C_7 (M_B)]^2 g(L_+) \end{equation} Note that (\ref{final3}) and (\ref{final4}) do not depend on any unphysical parameter, since the $\bar\Lambda$ dependence has disappeared. In particular, all ambiguities induced by renormalons should cancel in these equations. Based on this it has been suggested to give a definition for a heavy quark mass $\hat{m}_b$ through (\ref{final3}) and (\ref{final4}) using e.g. the mean energy $\langle E_\gamma \rangle$ of the photon in $B \to X_s \gamma$ \cite{Bauer}. To this end we consider the moments of the function $g$, which can be obtained from the moments of the shape function $f$. The usual HQET relations are \begin{eqnarray} \label{moments} && \int_{-\infty}^{\bar\Lambda} dk_+ f(k_+) = 1 \label{zero}\\ && \int_{-\infty}^{\bar\Lambda} dk_+ k_+ f(k_+) = \delta m_b \label{one}\\ && \int_{-\infty}^{\bar\Lambda} dk_+ k_+^2 f(k_+) \label{two} = -\frac{1}{3} \lambda_1 \end{eqnarray} which define the normalization, the residual mass term \cite{FalkNeubert} and the kinetic energy parameter. Reexpressing relation (\ref{one}) through the function $g$ we have \begin{equation} \int_{-\infty}^0 dL_+ L_+ g(L_+) = \delta m_b - \bar\Lambda \equiv \hat\Lambda \end{equation} which gives a definition of the heavy quark mass $\hat{m_b} = M_B - \hat\Lambda$ free of renomalon ambiguities which cancel between the $\delta m_b$ and $\bar\Lambda$ \cite{Bauer}. For the second moment one obtains from (\ref{two}) \begin{equation} \label{two1} \int_{-\infty}^0 dL_+ L_+^2 g(L_+) = -\frac{1}{3} \lambda_1 - \bar\Lambda^2 - 2 \bar\Lambda \hat\Lambda \equiv -\frac{1}{3} \hat\lambda_1 \end{equation} where $\hat\lambda_1$ again has to be free of renormalon ambiguities. We shall later need a model shape function to discuss our results. In order to keep things simple, we use the one-parameter model of \cite{MannelNeubert} \begin{eqnarray} \label{sf} g(L_+) &=& {32 \over \pi^2 \sigma^3} L_+^2 \exp\left[- {4 \over \pi \sigma^2 } L_+^2 \right] \Theta(-L_+) \end{eqnarray} which is easy to deal with. Here $\sigma$ is the width parameter of the shape function and will be varied in our numerical studies between 400 MeV and 800 MeV. \section{Including Perturbative Contributions} The next step is to include radiative corrections to the relations (\ref{final3}) and (\ref{final4}). The starting point of the considerations is (\ref{final}) or eqivalently (\ref{final1}). However, now the partonic result $d\Gamma_{part}$ including the radiative corrections is convoluted with the shape function \begin{equation} d\Gamma_{had} = \int\limits_{-m_b}^{\bar{\Lambda}} d\Gamma_{part} (m_b^* = m_b + k_+) f(k_+) dk_+ \end{equation} We shall compute the spectrum in the light cone variable of the hadronic momentum $L_+$, and the partonic counterpart of this variable is \begin{equation} l_+ = - m_b + (vq) + \sqrt{(vq)^2 - q^2} \, . \end{equation} in which we compute a differential spectrum as some function $G$ of $l_+$ and $m_b$ \begin{equation} \frac{d\Gamma_{part}}{d l_+} = G(m_b,l_+) \, . \end{equation} To apply the convolution formulae (\ref{final}) and (\ref{final1}) we replace in the first step the heavy quark mass by $m_b^*$ and convolute with the shape function. The variable $l_+$ is also replaced by \begin{equation} l_+^* = - m_b^* + (vq) + \sqrt{(vq)^2 - q^2} = l_+ - k_+ \end{equation} Note that $l_+^*$ is again a light cone variable for a process in which the heavy quark mass is replaced by $m_b^*$, and hence it ranges between $-m_b^* \le l_+^* \le 0$, which means that the $k_+$ integration becomes restricted to \begin{equation} -l_+ \le k_+ \le \bar{\Lambda} \end{equation} The second step towards hadronic variables is to replace $l_+$ by $L_+$, the hadronic light cone momentum. At the same time we perform a shift in the integration variable $k_+ \to K_+ - \bar{\Lambda}$ and obtain \begin{equation} \label{pert} \frac{d\Gamma_{had}}{dL_+} = \int\limits_{L_+}^0 dK_+ \, G(M_B+K_+,L_+ -K_+) g(K_+) \end{equation} Relation (\ref{pert}) holds for any $b \to q$ transition where $q$ is a light quark. We shall exploit (\ref{pert}) for a comparison between the inclusive processes $B \to X_s \gamma$ and $B \to X_u \ell \bar{\nu}_\ell$. The partonic $l_+$ spectra of the two processes can be calculated; both take the generic form \begin{eqnarray} \label{partonic} G(m_b,l_+) &=& \gamma_0 m_b^5 \left[ (1+ b_0 \frac{\alpha_s}{3\pi}) \delta(l_+) + \vphantom{\int\limits_t^t} \right. \\ \nonumber && \left. \frac{\alpha_s}{3\pi} \left( b_1 \left(\frac{\ln(-l_+ / m_b)}{-l_+} \right)_+ + b_2 \left(\frac{1}{-l_+} \right)_+ + \frac{1}{m_b} \Phi \left(\frac{-l_+}{m_b} \right) \right) \right] \end{eqnarray} where we have defined the ``+''-distributions in the usual way \begin{equation} \int_0^1 dx D_+ (x) = 0 \, , \end{equation} and $b_0$, $b_1$, $b_2$ and $\Phi$ are known quantities. For $b\to u \ell \bar{\nu}_\ell$ we find \begin{eqnarray} \label{b2uln} \gamma_0 &=& {G_F^2 \left|V_{ub}\right|^2 \over 192 \pi^3} \\ b_0 &=& -{13\over 36}-2\pi^2 \nonumber \\ b_1 &=& -4 \nonumber\\ b_2 &=& -26/3 \nonumber \\ \Phi(x) &=& \frac{158}{9} + \frac{407\,x}{18} - \frac{367\,{x^2}}{6} + \frac{118\,{x^3}}{3} - \frac{100\,{x^4}}{9} + \frac{11\,{x^5}}{6} - \frac{7\,{x^6}}{18} \nonumber \\ && - \frac{4\,\ln (x)}{3} + \frac{46\,x\,\ln (x)}{3} + 6\,{x^2}\,\ln (x) - \frac{16\,{x^3}\,\ln (x)}{3} \nonumber\\ && - 12\,{x^2}\,{{\ln (x)}^2} + 8\,{x^3}\,{{\ln (x)}^2} \nonumber \end{eqnarray} While for $b\to s\gamma$ \begin{eqnarray} \label{b2sg} \gamma_0 &=& {G_F^2 \left|V_{ts}V_{tb}^*\right|^2 \left| C_7\right|^2 \alpha_{\rm em} \over 32 \pi^4} \\ b_0 &=& -5-\frac{4}{3} \pi^2 \nonumber \\ b_1 &=& -4 \nonumber \\ b_2 &=& -7 \nonumber \\ \Phi(x) &=& 6 + 3\,x - 2\,{x^2} - 4\,\ln (x) + 2\,x\,\ln (x) \nonumber \end{eqnarray} where $C_7 = -0.306$ is obtained from the next-to-leading order calculation of \cite{ChetMM}. Our result for $B \to X_s \gamma$ coincides with the ones obtained in the literature \cite{AliGreub,ChetMM,KaganNeubert}. To determine $b_0$ for $B \to X_u \ell \bar{\nu}_\ell$ and $B \to X_s \gamma$ the respective total rates at ${\cal O}(\alpha_s)$ \cite{Timo,Pott} have been used. The partonic result up to order $\alpha_s$ contains terms of all orders in the $1/m_b$ expansion and we shall first consider the leading twist contribution. Taking the limit $m_b \to \infty$ of the partonic calculation leaves us only with the ``+''-distributions and the $\delta$ function. The convolution becomes \begin{eqnarray} \frac{d\Gamma_{had}}{dL_+} &=& \int\limits_{L_+}^0 dK_+ \, g(K_+) \gamma_0 (M_B+K_+)^5 \left[ (1+ b_0 \frac{\alpha_s}{3\pi}) \delta(L_+-K_+) \vphantom{\int\limits_t^t} \right. \\ \nonumber && + \frac{\alpha_s}{3\pi} \left. \left( b_1 \left( \frac{\ln\left(\frac{-L_+ + K_+}{M_B + K_+}\right)}{-L_+ + K_+} \right)_+ + b_2 \left(\frac{1}{-L_+ + K_+} \right)_+ \right) \right] \end{eqnarray} The shape function $g(K_+)$ is restricted to values $K_+$ small compared to $M_B$, and one may expand the dependence on $M_B + K_+$ for small $K_+$. This induces terms of higher orders in $1/m_b$, which again may be dropped. Hence the leading twist terms become \begin{eqnarray} \frac{d\Gamma_{had}}{dL_+} &=& \gamma_0 M_B^5 \left[ (1+ b_0 \frac{\alpha_s}{3\pi}) g(L_+) + \vphantom{\int\limits_t^t} \right. \\ \nonumber && \left. \frac{\alpha_s}{3\pi} \int\limits_{L_+}^0 dK_+ g(K_+) \left( b_1 \left( \frac{\ln\left(\frac{-L_+ + K_+}{M_B}\right)}{-L_+ + K_+} \right)_+ + b_2 \left(\frac{1}{-L_+ + K_+} \right)_+ \right) \right] \end{eqnarray} It is well known that the coefficient $b_1$ is universal as well as the shape function. This suggests to define a scale dependent shape function, which to order $\alpha_s$ becomes \begin{equation} \label{scale} {\cal F}(K_+,\mu) = g(K_+) + \frac{\alpha_s b_1}{3 \pi} \int\limits_{L_+}^0 dK_+ g(K_+) \left( \frac{\ln\left(\frac{-L_+ + K_+}{\mu}\right)}{-L_+ + K_+} \right)_+ \end{equation} The scale dependence of the distribution function has been discussed in \cite{BMK}, where the evolution equation for the distribution function has been set up. The terms proportional to $(1/l_+)_+$ are not universal, which is natural, since the relavant scale in two different processes may be different. A change of scale changes the contribution of the $(\alpha_s / \pi) (1/l_+)_+$ pieces according to \begin{equation} {\cal F}(K_+,\mu) = {\cal F}(K_+,\mu^\prime) + \frac{\alpha_s b_1}{3 \pi} \int\limits_{L_+}^0 dK_+ g(K_+) \ln\left(\frac{\mu^\prime}{\mu}\right) \left(\frac{1}{-L_+ + K_+} \right)_+ \end{equation} and hence we find for the leading twist contribution \begin{equation} \label{part1} \frac{d\Gamma_{had}}{dL_+} = \gamma_0 M_B^5 (1+ b_0 \frac{\alpha_s}{3\pi}) {\cal F}(L_+,\mu) \end{equation} where the scale $\mu$ is given by \begin{equation} \label{scalemu} \ln\left(\frac{M_B}{\mu}\right) = \frac{b_2}{b_1} \end{equation} Inserting the results for the two processes under consideration we find \begin{equation} \label{scales} \mu_{b \to s \gamma} = 0.1738 M_b \qquad \mu_{b \to u \ell \bar{\nu}_\ell} = 0.1146 M_b \end{equation} which are in fact scales small compared to the mass of the $b$ quark. Furthermore, the ratio of the two scale is of order one, \begin{equation} \frac{\mu_{b \to s \gamma}}{\mu_{b \to u \ell \bar{\nu}_\ell}} \approx 1.52 \end{equation} The physical meaning of the scale $\mu$ can be understood in a picture as proposed in \cite{KorchemskySterman}. Here the amplitude is decomposed into a hard part, a ``jetlike'' part incorporating the collinear singularities and a soft part. The typical scales corresponding to these pieces are $m_b^2$ for the hard, $\Lambda_{QCD} m_b$ for the ``jetlike'' and $\Lambda_{QCD}^2$ for the soft part. Comparing the present approach to \cite{KorchemskySterman} the light-cone distribution function corresponds to the convolution of the soft and the ``jetlike'' piece, both of which are universal but scale dependent. Thus the scale appearing in the shape function should be of the order $\mu^2 \sim m_b \Lambda_{QCD}$ since it incorporates all lower scales. In particular, the shape function combines both the collinear and the soft contributions which according to \cite{KorchemskySterman} could be factorized. Since we are working to next--to--leading order, we can fix the scale according to (\ref{scalemu}), and we expect to obtain a scale $\mu^2$ of order $\Lambda_{QCD} m_b \sim 0.9$ GeV${}^2$, which is confirmed by (\ref{scales}). The rest of the partonic result, i.e. the function $\Phi$, is a contribution of subleading terms, suppressed by at least one power of the heavy mass. We shall use these terms to estimate, how far we can trust the leading twist terms, in particular in the comparison between $B \to X_u \ell \bar{\nu}_\ell$ and $B \to X_s \gamma$. To this end we add the two contributions and use the expression \begin{equation} \label{partfull} \frac{d\Gamma_{had}}{dL_+} = \gamma_0 M_B^5 \left[ (1+ b_0 \frac{\alpha_s}{3\pi}) {\cal F}(L_+,\mu) + \frac{\alpha_s}{3\pi M_B} \Phi \left(-\frac{L_+}{M_B} \right) \right] \end{equation} We note that the functions $\Phi$ for both processes contain terms diverging logarithmically as $L_+ \to 0$, but which are suppressed by $1/m_b$. These divergencies will be cured once the analogon of the light-cone distribution function at subleading order is taken into account. However, for the quantitative analyis of the process these terms are not important as long as we do not get too close to the endpoint. \section{Comparison of $B \to X_u \ell \bar{\nu}_\ell$ to $B \to X_s \gamma$} \begin{figure} \hspace*{-15mm} \leavevmode \epsfxsize=8cm \epsffile{plot1a.eps} \epsfxsize=8cm \epsffile{plot1b.eps} \caption{Light-cone distribution spectra for $B \to X_u \ell \bar{\nu}_\ell$ (solid) and $B \to X_s \gamma $ (dashed) in units of $\gamma_0 M_B^5$. The dotted line is the leading contribution, which is the same for both processes. The three sets of curves correspond to values of $\sigma$ of 400, 600 and 800 MeV.} \label{cfig1} \end{figure} We shall first study the $L_+$ spectra of the two processes using (\ref{part1}) and (\ref{partfull}). These results depend on the model for the shape function one is using and we shall employ the simple one-parameter model (\ref{sf}). In fig.\ref{cfig1} we show the $L_+$ spectra of $B \to X_u \ell \bar{\nu}_\ell$ and $B \to X_s \gamma$. We divide the rates by the prefactors $\gamma_0 M_B^2$, such that at leading order only the shape function remains. The left plot shows the shape function (\ref{sf}) together with the leading twist contribution (\ref{part1}) for the two processes for three different values of the width parameter $\sigma$. The leading twist contribution can only be trusted close to the endpoint $L_+ =0$; it even becomes negative for values $L_+ \le - \sigma$, indicating the breakdown of the leading twist approximation. Adding the subleading terms as in (\ref{partfull}) cures this problem and the results are shown in the right hand figure of fig.\ref{cfig1}. The sharp rise of the full results is due to the logarithms $\ln L_+$ of order $1/m_b$, which become relevant only very close to $L_+ = 0$. These contributions are integrable and will not be relevant once the rates are binned with reasonably large bins. Next we shall compare $B \to X_u \ell \bar{\nu}_\ell$ to $B \to X_s \gamma$ by considering the ratio of the two $L_+$ distributions. From the experimental point of view many systematic uncertainties cancel, while from the theoretical side one expects to reduce nonperturbative uncertainties, which indeed at tree level cancel completely. The perturbative result for the spectrum is a distribution and the ratio of the perturbative expressions is meaningless, even if we avoid the region around $L_+ = 0$. However, once we have combined both perturbative and non-perturbative contributions we can take the ratio without encountering the problem of distributions. The price we have to pay is that the resulting ratio is not independent of the shape function, an effect which has been observed already in \cite{Neubert1}. Still the ratio of the two processes is a useful quantity, since one may expect that much of the non-perturbative uncertainties still cancel. \begin{figure} \hspace*{-15mm} \leavevmode \epsfxsize=8cm \epsffile{plot2a.eps} \epsfxsize=8cm \epsffile{plot2b.eps} \caption{Ratio of the light cone distribution spectra for the two processes. The leading twist contribution (\protect{\ref{comp}}) is given in the left plot while the right plot shows the full result. The solid, dashed and dotted lines correspond to values of $\sigma$ of 600, 400 and 800 MeV, respectively.} \label{cfig2} \end{figure} The two spectra are due to the ``smearing'' with the nonperturbative shape function smooth curves and we can perform a comparison of the two processes by taking the ratio. At leading twist this ratio becomes \begin{equation} \label{comp} \frac{(d \Gamma_{B \to X_u \ell \bar{\nu}_\ell} / dL_+)} {(d \Gamma_{B \to X_s \gamma} / dL_+)} = \frac{|V_{ub}|^2}{|V_{ts} V_{tb}^*|^2}\frac{\pi}{6 \alpha_{em} |C_7|^2} \frac{{\cal F}(L_+,\mu_{b \to s \ell \bar{\nu}_\ell})} {{\cal F}(L_+,\mu_{b \to s \gamma})} \left(1 + \frac{\alpha_s (m_b)}{3 \pi} \left[\frac{167}{36} - \frac{2}{3} \pi^2 \right] \right) \end{equation} where the scales of the distribution functions have been given in (\ref{scales}). It is interesting to note that although the radiative corrections at leading twist (the constants $b_0$) are big for the individual processes, they turn out to be small in the ratio (\ref{comp}). For values of $L_+$ within the nonperturbative region $L_+ \ge -\sigma$ the leading twist contribution is dominant and the comparison may be performed using the leading twist terms only. This is the region with the largest fraction of the total rate and hence the subleading terms will not play a role, also due to experimental cuts. The ratio of the leading twist terms is shown in the left hand plot of fig.\ref{cfig2} for three values of the parameter $\bar\Lambda$. The region of validity of the leading twist contribution may be studied by looking at the ratio of the full results (\ref{partfull}) shown in fig(\ref{cfig2}). The subleading terms lead to a minimum at the point where they are becomming the dominant contribution to the rates. This happens at values of $L_+$ which correspond to the typical width $\sigma$ of the non-perturbative shape function. For values of $L_+$ between $-\sigma$ and $0$ one may then use our results to determine CKM matrix elements, since the ratio of the two rates depends on $$ \frac{|V_{ub}|^2}{|V_{ts} V_{tb}^*|^2}\frac{\pi}{6 \alpha_{em} |C_7|^2} \approx 5 $$ i.e. the results shown in fig.\ref{cfig2} have to be multiplied by this factor. Hence one may determine $ |V_{ub}|/|V_{ts}|$ in this way; however, the radiative corrections induce a small hadronic uncertainty, which is hard to quantify without knowledge of the non-perturbative distribution function. Still one may expect only a small uncertainty, since at tree level there is no hadronic uncertainty at all. \section{Conclusions} Due to experimental constraints severe cuts on the phase space in both $b \to s \gamma$ and $b \to u \ell \bar{\nu}_\ell$ inclusive transitions have to be imposed making these decays sensitive to non-perturbative effects. Within the framework of the heavy mass expansion of QCD these effects are encoded in a light-cone distribution function which is universal for all heavy to light processes. The universality of this function may be exploited to eliminate the non-perturbative uncertainties in $b \to s \gamma$ using the input of $b \to u \ell \bar{\nu}_\ell$ or vice versa. To this end, one may use the comparison of the two processes to determine the ratio $V_{ub} / V_{ts}$ or --- for fixed CKM parameters --- to test the Wilson coefficient $C_7$. The comparison between the two processes is best performed in terms of the spectra of the light-cone component of the final state hadrons, which for the case of $B \to X_s \gamma$ is equivalent to the photon energy spectrum, while for $B \to X_u \ell \bar{\nu}_\ell$ this requires a measurement of both the hadronic invariant mass and the hadronic energy. For small values, the light cone component of the final state hadrons is the same as the light cone component of the heavy quark residual momentum. The corresponding spectra are at tree level directly proportional to the light cone distribution function and hence one may access this function by a measurement of the the light cone distribution of the momentum of the final state hadrons. However, radiative corrections change this picture. The main effect is that the shape function does not cancel any more in the ratio of the two spectra. The partonic radiative corrections to the light-cone distribution of the final state partons exhibit the well known singularities of the type $\ln l_+ / l_+$ and $1/l_+$ which can be absorbed into a radiatively corrected, scale dependent light cone distribution function. It turns out that the relevant scales are small in both cases and different. The scale dependent distribution functions involve a convolution such that the nonperturbative effects cannot be cancelled anymore in the ratio. We have also computed the subleading terms partonically and use this to estimate the reliability of the leading twist comparison of $B \to X_s \gamma$ and $B \to X_u \ell \bar{\nu}_\ell$. This comparison depends on the width of the non-perturbative function once radiative corrections are taken into account. For values of $L_+$ close to the endpoint (i.e. within the width of the distribution function) one may use the leading twist contribution to extract $|V_{ub}| / |V_{ts}|$ with reduced hadronic uncertainties. This method should become feasible at the future $B$ physics experiments. \section*{Acknowledgements} We thank Kostja Chetyrkin, Timo van Ritbergen and Marek Jezabek for discussions and comments. SR and TM acknowledge the support of the DFG Graduiertenkolleg ``Elementarteilchenphysik an Beschleunigern''; TM acknowledges the support of the DFG Forschergruppe ``Quantenfeldtheorie. Computeralgebra und MonteCarlo Simulationen''.
\section{Introduction} The ordering of holes and spins into stripes in the CuO$_2$-planes of Nd-doped La$_{2-x}$Sr$_{x}$CuO$_4$ (LSCO) has attracted much attention, because charge separation and antiferromagnetic spin fluctuations are believed to be important for understanding the mechanism of high temperature superconductivity. The idea of pinned stripes in the low temperature tetragonal (LTT) phase of Nd-doped LSCO or La$_{2-x}$Ba$_{x}$CuO$_4$ \cite{john_nature,axe89} provides an explanation for the anomaly observed at $x \sim \frac{1}{8}$, where superconductivity is destroyed, or at least strongly suppressed \cite{mood88,craw91,buch94,mood97}. The tilting of the oxygen octahedra along the $[1\,0\,0]$ and $[0\,1\,0]$ directions (i.e., parallel to Cu--O bonds) in the LTT phase introduces a pinning potential for horizontal and vertical stripes, whereas in the low temperature orthorhombic (LTO) phase it is absent due to the rotation of the tilting axis into the $[1\,1\,0]$ direction. In a single crystal of La$_{1.48}$Nd$_{0.4}$Sr$_{0.12}$CuO$_4$, neutron diffraction allowed the observation of both magnetic and charge-ordering superstructure reflections \cite{john_nature}. Consistent with the idea of pinning by the LTT lattice modulation, static stripe ordering within the CuO$_2$ planes appears at the transition temperature from the LTO to the LTT phase, which is $\rm \sim 68 K$. From the positions of the superlattice peaks and the nominal hole concentration it follows that stripes of holes are approximately half-filled, and act as antiphase domain walls with respect to the antiferromagnetically ordered Cu spins. Thus, a Sr doping level of $\sim\frac{1}{8}$ yields a spacing between stripes of $4a$, where $a$ is the lattice constant. High energy x-ray studies have been succesful in confirming the results for the charge stripe ordering in an $x=0.12$ sample \cite{martin98}. It is still an open question as to whether charge stripes are limited to hole concentrations near $\frac{1}{8}$ in Nd-doped LSCO or whether they influence the physics of cuprate superconductivity in general, but there is accumulating evidence for the latter. Inelastic neutron scattering experiments indicate the possibility of moving, fluctuating stripes in LSCO \cite{cheo91,yamada98} and even in YBCO \cite{dai98,mook98}. Local charge ordering in La$_{2-x}$Sr$_{x}$CuO$_4$ with $x\leq\frac18$ is suggested by a recent nuclear quadrupole resonance (NQR) study \cite{hunt99}. It is interesting to imagine that dynamic stripe correlations might be necessary for superconductivity \cite{emer97}, whereas pinning the stripes leads to a strongly reduced critical temperature $\rm T_c$, but the experimental evidence for such a scenario is still incomplete. So far the Nd-doped LSCO system offers a unique opportunity to study pinned stripe patterns in diffraction experiments, giving information about the nature of charge and spin ordering. The superstructure reflections due to the ordering of holes are shifted by $2\epsilon$ in $h$- or $k$-direction relative to fundamental reflections, whereas the magnetic peaks are located around the antiferromagnetic peak position $(\frac12,\frac12,0)$, shifted by $\epsilon$ also in $h$- or $k$-direction. In neutron scattering experiments, the splitting $\epsilon$ has been observed to increase slightly as the Sr content increases from 0.12 to 0.15 and 0.20, implying a decrease in the average stripe spacing; however, it has only been practical to study the superstructure reflections due to the antiferromagnetic ordering \cite{john97}. It is now imperative to directly characterize the charge order at Sr concentrations away from $\frac18$. In this work we present x-ray studies with 120~keV photons on $\rm La_{1.45}Nd_{0.4}Sr_{0.15}CuO_4$. Our results are in complete agreement with the stripe model and complementary to the experimental evidence of antiferromagnetic stripe ordering in this sample \cite{john97}. As observed in the $x=0.12$ case \cite{john_nature}, the superstructure reflections due to the ordering of the holes sets in at a higher temperature compared to the magnetic signal, indicating that the transition into the stripe phase is driven by the charge separation \cite{zach98}. \section{Experiment} The experiments have been performed on the triple-axis diffractometer designed for the use of $\sim 100$~keV photons at the high-field wiggler beamline BW5 at HASYLAB, Hamburg \cite{bouchard98}. X-ray diffraction in this energy range has proven to be very successful in studying charge ordering in cuprates, nickelates and manganates \cite{martin98,titi97,thomas99}. As in neutron scattering experiments, the large penetration depth ($\sim 1$ mm) allows one to probe the bulk of the sample, enabling a direct comparison of x-ray and neutron diffraction data. In contrast to former experiments on $\rm La_{1.48}Nd_{0.4}Sr_{0.12}CuO_4$ \cite{martin98}, the Si/TaSi$_2$ monochromator and analyzer crystals have been replaced by the new Si$_{1-x}$Ge$_x$ gradient crystal material. These crystals show very high reflectivity values of 96\%\ (not corrected for absorption) and variable widths of the rocking curves depending on the Ge content \cite{steffen98,steffen99}. By use of this new material, the scattered intensity at the stripe peak positions in Nd doped LSCO is about four times higher compared to the results obtained previously with the utilization of $\rm Si/TaSi_2$ crystals as monochromator and analyzer \cite{steffen98user}. Figure~1 shows a comparison of a superstructure reflection measured with the $\rm Si/TaSi_2$ and Si$_{1-x}$Ge$_x$ gradient crystals. The count rate collected with the gradient crystals is nearly two times higher, and a slightly better signal to background ratio could be reached by reducing the beam spot on the sample, i.e. by probing only the center of the crystal. The lower curve has been measured with the $\rm Si/TaSi_2$ crystals, illuminating a two times larger sample volume. In this experiment 120~keV photons have been employed, with a monochromatic beam intensity of $1.2\times 10^{11}$ photons/$\rm mm^2$. The resolution (FWHM) at a (2,0,0) reflection of $\rm La_{1.45}Nd_{0.4}Sr_{0.15}CuO_4$ was 0.020~\AA$^{-1}$ in the longitudinal and 0.0014~\AA$^{-1}$ in the transverse direction, the latter being limited by the sample mosaicity. At smaller diffraction angles the longitudinal resolution improves because the diffraction geometry becomes less dispersive, e.g. the FWHM of the (1,1,0) reflection is 0.011~\AA$^{-1}$. A closed-cycle cryostat has been utilised and temperatures between 9~K and 300~K could be reached at the sample position. The studied crystal ($\sim2\times2\times4$ mm$^3$) is a piece of a cylindrical rod that was grown by the travelling-solvent floating-zone method. \section{Results} Previous hard x-ray diffraction experiments on a $\rm La_{1.48}\-Nd_{0.4}Sr_{0.12}CuO_4$ crystal have shown that the stripe peaks are displaced not only within the $(h,k,0)$ plane, but also in the $\ell$ direction \cite{martin98}. Due to the better resolution in reciprocal space compared to neutron scattering experiments it has been possible to find a modulated intensity of the stripe peaks along the $\ell$-direction with maxima at $\ell = \pm 0.5$, indicating a correlation of the charge stripes in next-nearest-neighbor layers along the $c$-axis. The highest scattering intensities for the stripe peaks are expected at the positions $(2-2\epsilon,0,\pm0.5)$ and $(2+2\epsilon,0,\pm0.5)$. Figure 2 clarifies the positions of the stripe peaks in reciprocal space relative to the CuO$_2$ planes. We have used the tetragonal unit cell with $a=b=3.775$~\AA, $c=13.10$~\AA. Longitudinal scans along $(h,0,0.5)$ are shown in Fig.~3. As anticipated, small superstructure reflections are observed at $h=2\pm2\epsilon$ with $2\epsilon = 0.256(1)$. The peaks that are present at $T=9$~K have disappeared after raising the temperature to 70~K. The signal to background ratio is $\sim 0.2$ for the $2-2\epsilon$ reflection and $\sim 0.1$ at the $2+2\epsilon$ position. In both scans the background rises due to the vicinity of the (2,0,0) reflection. At this fundamental Bragg reflection the count rate in the peak maximum is about $10^8$ times larger than in the stripe peaks. The curves through the data points in Fig.~3 are least-squares fits. In both (a) and (b) the charge-order peaks are modelled with a Gaussian. The background in (a) is approximated by the tail of a second Gaussian centered near $h=2$ plus a linear contribution, while only a linear background is used in (b). Usually, the tail of a fundamental reflection is Lorentzian shaped, but nevertheless it is possible that locally other functions are a better approximation to a non-linear background, since the shape of the background is related to the sample quality. The studied $\rm La_2CuO_4$ crystal with Sr and Nd as dopants on the La-site certainly incorporates defects and strain which are responsible for the rather high background, which is $\sim 150$ photons per second at the superlattice positions. In the inset of Fig.~3(a), two more low-temperature longitudinal scans, slightly shifted in the $k$-direction, are displayed. The absence of any peaks in these scans indicates that the peak found at $(2-2\epsilon,0,0.5)$ is narrow in $k$. This conclusion is confirmed by the 9~K transverse scan shown in Fig.~4. The background in the transverse scans is linear, and a fit of the stripe peak with a Lorentzian is slightly more successful than utilizing a Gaussian. From the present data, it is difficult to determine the true peak shape, given the small signal-to-background ratio. Figure~5(a) shows the temperature variation of the amplitudes of the $h$ and $k$ scans at the $(2-2\epsilon,0,0.5)$ position. With rising temperature the amplitudes decrease rather linearly; the peak vanishes at 62(5)~K, which is $\sim8$~K below the structural transition from the LTT into the LTO phase. The intensity of the (3,0,0) reflection, also shown in Fig.~5(a), is a measure of this transition because it only occurs in the LTT phase. From the longitudinal and transverse scans a FWHM of 0.028(5)~r.l.u. can be inferred, resulting in a correlation length of 43(8)~{\AA}. The change of the peak widths with temperature can be seen in Fig.~5(b). Up to $\sim55$~K it is rather constant, but above this temperature the width of the peak in $h$- and $k$-directions increases considerably. Note that the increasing width as well as the vanishing peak intensity are not directly connected to the structural transition from the LTT into the LTO phase, but occur at a remarkably lower temperature. \section{Discussion} Our x-ray results provide direct evidence for charge-stripe order in $\rm La_{1.45}Nd_{0.4}Sr_{0.15}CuO_4$. Such charge order had previously been inferred from the magnetic order observed by neutron scattering \cite{john97}. This is an important result, because high-field magnetization \cite{oste97} and muon-spin-rotation ($\mu$SR) \cite{muon98} studies have indicated that bulk superconductivity also exists in this sample. Another conclusion of the $\mu$SR study is that static magnetic order is present throughout essentially the entire sample volume. Thus, it appears that stripe order and superconductivity coexist intimately at $x=0.15$. Compared to the $x=0.12$ composition, both the charge ordering temperature and the superstructure peak intensities are reduced. The superlattice intensities [normalized to the (200) peak] are roughly 30\%\ weaker in $x=0.15$ relative to $x=0.12$. A similar reduction in magnetic ordering temperature and intensity have been observed previously. At both Sr concentrations, charge ordering appears at a higher temperature than magnetic order. Within the error bars the correlation length of the stripe order in the $x=0.15$ sample is the same as in the $x=0.12$ sample, and in both samples the correlation length of the stripes is constant over a large temperature range. In contrast to this finding, it is important to note that neutron scattering and muon-spin rotation experiments revealed that the spins are only quasi-static, with a decreasing correlation length, in the temperature range above 30~K in the $x=0.12$ sample \cite{muon98,john98}. This continuous loss of correlation is only observed in the magnetic signal, whereas the correlation length for the charge order shows a decrease only very close to the disordering temperature. The temperature range in which a residual signal with an increased width is observed can be interpreted with fluctuating stripes, since the detected signal in our x-ray diffraction experiment is a sum of elastic and inelastic contributions. One might conclude that fluctuations of stripes set in if the pinning potential in the LTT phase is destroyed at the structural transition from LTT to LTO, as observed in the x~=~0.12 sample, or if rising temperature competes with a pinning potential via thermal activation. In the sample with $x=0.15$, the second possibility holds, since the suppression of the stripe ordering is not connected to the structural transition. The pinning potential is smaller in the $x=0.15$ sample than in the $x=0.12$ sample because the tilting angle of the oxygen octahedra is reduced with increasing Sr content in Nd doped LSCO \cite{buch94}. On the other hand, the idea that the LTT structure is required for charge-stripe order is challenged by the interpretation of the Cu NQR results in LSCO by Hunt {\it et al.} \cite{hunt99}. That work suggests that static stripes can occur even within the LTO phase, and that the ordering temperature continues to increase as $x$ decreases below 0.12. As an initial test of this picture, we intend to study charge order in a Nd-doped crystal with $x=0.10$ in the near future. To summarize, we have been successful in validating the existence of stripe order in $\rm La_{1.45}Nd_{0.4}Sr_{0.15}CuO_4$, a sample which is superconducting below $\sim 10$~K. Charge-order peaks due to stripe ordering have been observed at $(2-2\epsilon,0,\pm0.5)$, $(2+2\epsilon,0,0.5)$, and at $(0,2-2\epsilon,\pm0.5)$. The $2\epsilon$ value of 0.256(1) is in very good agreement with neutron scattering results for the magnetic peaks. This is a further step in establishing the picture of static stripes in Nd doped LSCO. The intensity of the stripe signal decreases almost linearly with rising temperature and vanishes at 62(5)~K, $\sim8$~K below the structural transition from the LTT to the LTO phase. The formation of the stripe pattern in $\rm La_{1.45}Nd_{0.4}Sr_{0.15}CuO_4$ is less pronounced than in $\rm La_{1.48}Nd_{0.4}Sr_{0.12}CuO_4$. \\ Acknowledgements:\\ Work at Brookhaven is supported by Contract No.\ DE-AC02-98CH10886, Division of Materials Sciences, U.S. Department of Energy.
\section{Introduction} Numerical studies of QCD on the lattice provide us a quantitative understanding of the dynamics of strong interactions. One of the main goals is to derive the hadron mass spectrum from the first principle. In order to give a precise prediction with lattice QCD simulations we have to control the following three physical sources of systematic errors besides the quenching error: (i) scaling violation, (ii) finite size effects and (iii) chiral extrapolations. In the quenched approximation the GF11 collaboration carried out an extensive calculation to reduce these three systematic errors and found that the quenched spectrum is consistent with experiment within $5-10\%$ errors\cite{GF11}. The aim of our work is to tell definitely how the quenched spectrum deviates from experiment, by considerably diminishing the statistical and systematic uncertainties. On the other hand, in spite of much effort devoted in the full QCD spectrum study we have not yet been able to answer a fundamental question: to what extent the dynamical quarks affect the light hadron spectrum. Before embarking upon a realistic full QCD calculation in the future we attempt to settle this question. In the first part of this report we present our results for the quenched light hadron spectrum, examining the validity of QChPT for the chiral extrapolation. The two-flavor full QCD results obtained so far are presented in the second part, where we investigate possible signs of dynamical quark effects on the light hadron spectrum. More details are found in Ref\cite{latt98}. \section{Quenched light hadron spectrum with the Wilson quark action} \subsection{Details of numerical simulation} \begin{table}[t] \caption{\label{tab:para_quench} Simulation parameters for quenched QCD.} \begin{tabular}{llllllllllll} $\beta$ & $L^3\times T$ & $a^{-1}$[GeV] & $La$[fm] & \#conf. & $\begin{array}{c}{\rm sweep}\\{\rm /conf.}\end{array}$ & \multicolumn{5}{c}{$m_\pi/m_\rho$} & $\delta$ \\ \tableline $5.90$ & $32^3\times 56$ & $1.934(16)$ & $3.26(3)$ & $800$ & $200$ & 0.752(1) & 0.692(1) & 0.593(1) & 0.491(2) & 0.415(2) & 0.106(5) \\ $6.10$ & $40^3\times 70$ & $2.540(22)$ & $3.10(3)$ & $600$ & $400$ & 0.751(1) & 0.684(1) & 0.581(2) & 0.474(2) & 0.394(3) & 0.103(6) \\ $6.25$ & $48^3\times 84$ & $3.071(34)$ & $3.08(3)$ & $420$ & $1000$ & 0.760(1) & 0.707(2) & 0.609(2) & 0.502(2) & 0.411(3) & 0.117(7) \\ $6.47$ & $64^3\times 112$ & $3.961(79)$ & $3.18(6)$ & $150$ & $2000$ & 0.759(2) & 0.708(3) & 0.584(3) & 0.493(4) & 0.391(4) & 0.113(13) \\ \end{tabular} \end{table} Our simulations are carried out with the plaquette gauge action and the Wilson quark action. To control the systematic errors we carefully choose our run parameters, which are summarized in Table~\ref{tab:para_quench}. We employ four $\beta$ values so that the lattice spacing covers the range $a\approx 0.1-0.05$fm to remove the scaling violation effects by extrapolation of the data to the continuum limit. To avoid finite size effects we keep the physical spatial lattice size approximately constant at $La\approx 3$fm. A previous study showed that the finite size effects are $2\%$ or less already at $La\approx 2$fm\cite{finiteV}. The most subtle issue in controlling the systematic errors is associated with chiral extrapolations of the hadron masses. Although the mass formulae predicted by QChPT are considered to be plausible candidates for the fitting functions of chiral extrapolations, their validities should be checked employing a wide range of quark masses. At each value of $\beta$ we choose five values of the hopping parameter $K$ corresponding to $m_\pi/m_\rho\approx 0.75$, $0.7$, $0.6$, $0.5$ and $0.4$. The heaviest two values $m_\pi/m_\rho\approx 0.75$ and $0.7$ are taken to be around the physical strange quark mass. In terms of these quark masses we calculate hadron masses both for degenerate and non-degenerate cases. Gauge configurations are generated with the 5-hit pseudo heat-bath algorithm incorporating the over-relaxation procedure quadruply and employing periodic boundary condition. Quark propagators are solved in the Coulomb gauge both for the point and the smeared sources with periodic boundary conditions imposed in all four directions. For the hadron mass measurement we use the smeared source quark propagators. The physical point for the degenerate up and down quark mass is determined by $m_\pi(135.0)$ and $m_\rho(768.4)$. For the strange quark mass we employ $m_K(497.7)$ or $m_\phi(1019.4)$. The lattice scale $a^{-1}$ is set with $m_\rho$. \subsection{Chiral extrapolations} We first examine the validity of QChPT investigating the presence of quenched chiral logarithm in the pseudoscalar(PS) meson sector. For the quark mass dependence of the PS meson mass $m_{PS}$ the QChPT predicts\cite{QChPT} \begin{equation} \frac{m_{PS}^2}{m_s+m}=A\left\{1-\delta\left[{\rm ln} \left(\frac{2mA}{\Lambda_\chi^2}\right) +\frac{m_s}{m_s-m}{\rm ln}\left(\frac{m_s}{m}\right)\right]\right\} +B(m_s+m)+O((m,m_s)^2), \label{eq:m_ps} \end{equation} where $m$ and $m_s$ are masses of two valence quarks in the PS meson. This ratio diverges logarithmically toward the chiral limit due to the $\delta$ term. To detect the contribution of the $\delta$ term in a direct manner we introduce the two variables: $x=2-(m_s+m){\rm ln}(m_s/m)/(m_s-m)$ and $y=4 m m_s/(m_s+m)^2\times m_K^4/(m_\pi^2 m_\eta^2)$, for which eq.(\ref{eq:m_ps}) leads to the relation $y=1+\delta\cdot x$. Here $\pi$ and $\eta$ are the degenerate PS mesons with quark mass $m$ or $m_s$ and $K$ is the non-degenerate one with $m$ and $m_s$. It should be noted that the quark masses are defined by an extended axial vector current Ward identity(AWI) $\nabla_\mu A_\mu^{\rm ext}=2m_q^{\rm AWI} P$\cite{m_q^AWI}, where we are free from ambiguities originating from the determination of the critical hopping parameter $K_c$. Figure~\ref{fig:delta} shows the distribution of $y$ as a function of $x$. Our data fall in a wedge shaped with $y=1+0.08 x$ and $y=1+0.12 x$. A PS meson decay constant ratio $y=f_K^2/(f_\pi f_\eta)$, for which QChPT predicts $y=1-\delta/2\cdot x$, is another quantitative test for $\delta$. Our data indicate that the value of $\delta$ is within $\delta=0.08-0.16$. Since we observe clear evidence for the existence of the quenched chiral logarithm, we now try to fit the PS meson masses with the functional form of eq.(\ref{eq:m_ps}). In Fig.~\ref{fig:ps_chl_quench} we show a typical result for chiral extrapolations of the degenerate PS meson mass and the AWI quark mass. There are two features to be remarked. One is a good description of the AWI quark mass with a linear function of $1/K$, which means the AWI quark mass is proportional to the vector Ward identity(VWI) quark mass $m_q^{\rm VWI}=(1/K-1/K_c)/2$. The other is a good agreement between the critical hopping parameter $K_c$ determined from the QChPT fit of $m_{PS}^2$ and that from the linear fit of $m_q^{\rm AWI}$. They are consistent within $2.5\sigma$, which should be compared with $17\sigma$ ($12\sigma$) discrepancy for the case of a quadratic (cubic) polynomial fit of $m_{PS}^2$. Our examinations for the chiral properties of the PS meson masses strongly suggest the validity of QChPT with $\delta\approx 0.10(2)$. It would be legitimate to apply the QChPT mass formulae for the chiral extrapolation of the vector meson masses and the baryon ones. \begin{figure}[b] \begin{minipage}[b]{80mm} \centerline{\epsfxsize 80mm \epsfbox{kuramashi603fig1.eps}} \caption[]{\label{fig:delta} Test of the quenched chiral logarithm using the PS meson masses.} \vspace*{3.5mm} \end{minipage} \hspace{\fill} \begin{minipage}[b]{80mm} \centerline{\epsfxsize 73mm \epsfbox{kuramashi603fig2.eps}} \caption[]{\label{fig:ps_chl_quench} Chiral extrapolations of the degenerate PS meson mass and the AWI quark mass as a function of $1/K$ at $\beta=5.9$.} \end{minipage} \end{figure} For the vector meson masses we make simultaneous but uncorrelated fits of the degenerate and non-degenerate data employing the functional form predicted by QChPT: \begin{eqnarray} m_{V}&=&m_V^0+\frac{C_{1/2}}{6} \left\{\frac{3}{2}(m_\pi+m_\eta) +2\frac{m_\eta^3-m_\pi^3}{m_\eta^2-m_\pi^2}\right\} +\frac{C_1}{2}(m_\pi^2+m_\eta^2) +C_D(m_\pi^3+m_\eta^3)+C_N m_K^3, \label{eq:m_vfit} \end{eqnarray} where $C_{1/2}=-4\pi g_2^2\delta$ with $g_2$ a phenomenological coupling constant of the vector meson quenched chiral Lagrangian. It should be noticed that the $O(m_{PS})$ term is a characteristic of QChPT. This model function up to the $O(m_{PS}^2)$ terms describes our vector meson mass data well as illustrated in Fig.~\ref{fig:v_chl_quench}. A small bending toward the chiral limit found in the fitting result of $m_\rho$ reflects the contribution of the $O(m_{PS})$ term. However, the fitted values for the coefficient $C_{1/2}$ of $O(m_{PS})$ term are much smaller than expected. The average value of $C_{1/2}$ over four $\beta$ points gives $C_{1/2}=-0.071(8)$, whose magnitude is ten times smaller than the phenomenological estimate $C_{1/2}=-4\pi g_2^2\delta\approx -0.71$ with $\delta=0.1$ and $g_2=0.75$. The magnitude of the $O(m_{PS})$ term in eq.(\ref{eq:m_vfit}) is $0.07\times m_\pi\approx 10$MeV, which means about a $1\%$ contribution to $m_\rho$. If we employ the fitting function up to the $O(m_{PS}^3)$ terms, we find a few times larger value for $C_{1/2}$ at each beta. In this case, however, the fitting results are very unstable against the $\chi^2$ value. \begin{figure}[b] \begin{minipage}[b]{80mm} \centerline{\epsfxsize 73mm \epsfbox{kuramashi603fig3.eps}} \caption[]{\label{fig:v_chl_quench} Chiral extrapolations of vector meson masses as a function of $m_\pi^2$ at $\beta=5.9$. Open symbols in the inset represent extrapolated values at the physical degenerate up and down quark mass.} \end{minipage} \hspace{\fill} \begin{minipage}[b]{80mm} \centerline{\epsfxsize 73mm \epsfbox{kuramashi603fig4a.eps}} \centerline{\epsfxsize 73mm \epsfbox{kuramashi603fig4b.eps}} \caption[]{\label{fig:b_chl_quench} Same as Fig.~\protect{\ref{fig:v_chl_quench}} for octet and decuplet baryon masses.} \end{minipage} \end{figure} \begin{figure}[b] \begin{minipage}[b]{80mm} \centerline{\epsfxsize 73mm \epsfbox{kuramashi603fig5a.eps}} \centerline{\epsfxsize 73mm \epsfbox{kuramashi603fig5b.eps}} \caption[]{\label{fig:vb_con_quench} Continuum extrapolations of quenched hadron masses with $m_K$ as input for the strange quark mass.} \end{minipage} \hspace{\fill} \begin{minipage}[b]{80mm} \centerline{\epsfxsize 73mm \epsfbox{kuramashi603fig6.eps}} \vskip +.3 cm \caption[]{\label{fig:spectrum_quench} Quenched light hadron spectrum in the continuum limit. GF11 results\protect{\cite{GF11}} are also plotted for comparison. Horizontal bars denote experimental values.} \end{minipage} \end{figure} For fitting of the octet and decuplet baryon masses we take the same strategy as for the vector meson mass case. Assuming the QChPT mass formulae we fit the degenerate and non-degenerate data simultaneously without including correlations. While we find that $O(m_{PS})$ and $O(m_{PS}^2)$ terms are sufficient to reproduce our decuplet baryon mass data as shown in Fig.~\ref{fig:b_chl_quench}, for the octet baryon masses a $O(m_{PS}^3)$ term is inevitably required to explain the convex curvature of the nucleon mass data. The fitted value for the coefficient $C_{1/2}$ of the $O(m_{PS})$ term in the nucleon is $C_{1/2}=-0.118(4)$ as an averaged value over four $\beta$ points. This is less than half of the phenomenological estimate $C_{1/2}=-(3\pi/2)(D-3F)^2\delta\approx -0.27$ with $\delta=0.1$, $F=0.5$ and $D=0.75$. The $O(m_{PS})$ term contribution to the nucleon mass is about $16$MeV or $2\%$. We find a similar value $C_{1/2}=-0.14(1)$ for $\Delta$. If the $O(m_{PS}^3)$ terms are included in the fitting function of the decuplet baryon masses, we obtain much smaller value for $C_{1/2}$ in $\Delta$. Whereas we observe that the mass formulae based on the QChPT reproduce our data adequately, it is instructive to try another chiral ansatz employing polynomial fitting functions in $1/K$ (cubic for $N$, quadratic for others). We employ independent fits for the different degeneracies in the vector meson and the octet and decuplet baryons. The fitting results are shown by dashed lines in Figs.~\ref{fig:ps_chl_quench}, \ref{fig:v_chl_quench} and \ref{fig:b_chl_quench}. Although the QChPT fits seem to give better descriptions of our hadron mass data than the polynomial ones, the differences between them are fairly small. \subsection{Quenched light hadron spectrum in the continuum limit} The lattice spacing dependences of the hadron masses are shown in Fig.~\ref{fig:vb_con_quench}. We expect that the leading scaling violation effects are of $O(a)$ in the Wilson quark case. A linear extrapolation to the continuum limit assuming the fitting function $m=m_0(1+\alpha a)$ yields $\alpha \sim 0.2$GeV, for which we find that higher order terms are safely negligible, {\it e.g.}, $(\alpha a)^2\sim 0.01$ at $a=0.5$GeV$^{-1}$. While the results of the QChPT fits and the polynomial ones differ by about $3\%$($5\sigma$) in the largest case after the chiral extrapolation at each $\beta$, the differences in the continuum limit are within $1.5\%$($1.5\sigma$) of the results of the QChPT fits. At a few percent level of statistical errors it is hard to appreciate the differences between the QChPT chiral extrapolations and the polynomial ones. Our final result of the quenched light hadron spectrum is presented in Fig.~\ref{fig:spectrum_quench}. The spectrum deviates from experiment systematically and unambiguously. The following discrepancies should be noticed: The $K$-$K^*$ meson hyperfine splitting is definitely underestimated by $9.5\%$($4.3\sigma$) for the $m_K$ input case and by $16\%$($6.1\sigma$) for the $m_\phi$ input case; The nucleon mass is appreciably smaller than experiment by $7\%$($2.5\sigma$); For the strange octet baryon masses our data with the $m_K$ input are systematically smaller than experiment by $6-9\%$($4-7\sigma$), which is much reduced in the $m_\phi$ input case; While the $m_K$ input leads $30\%$ smaller estimates for the decuplet mass splittings in average, we observe rather good agreement for the $m_\phi$ input results. \section{Light hadron spectrum in two-flavor full QCD} \subsection{Details of numerical simulation} \nopagebreak The simulation of full QCD requires a huge amount of computing time compared to the quenched approximation. In order to secure an adequate physical spatial lattice size for avoiding finite size effects, the lattice spacing is compelled to be coarse $a^{-1}\rlap{\lower 3.5 pt\hbox{$\mathchar \sim$}}\raise 1pt \hbox {$<$} 2$GeV, which urges us to employ improved actions. Our choice is a RG-improved gauge action\cite{RGaction} and the SW quark action\cite{SWaction} with a mean-field improved value of $c_{SW}$. This decision is based on a former comparative study of various combinations of improved gauge and quark actions at $a^{-1}\approx 1$GeV\cite{prestudy}. Our full QCD study is performed with two flavors of sea quarks which are supposed to be the degenerate dynamical up and down quarks. Our simulation parameters are summarized in Table~\ref{tab:para_full}. Four $\beta$ values covering the lattice spacing in the range $a^{-1}\approx 1-2$GeV are employed to examine the scaling behavior of the light hadron spectrum. We keep the physical spatial lattice size approximately constant at $La\approx 2.4$fm except for the finest lattice where $La\approx 2.0$fm. At each value of $\beta$ we choose four values of sea quark mass corresponding to $m_\pi/m_\rho\approx 0.8$, $0.75$, $0.7$ and $0.6$. Gauge configurations including the dynamical sea quark effects are generated with the HMC algorithm. We calculate hadron masses using the five valence quarks whose masses correspond to $m_\pi/m_\rho\approx 0.8$, $0.75$, $0.7$, $0.6$ and $0.5$. Hadron propagators are constructed with the smeared source quark propagators at every fifth trajectory. Hadron masses are extracted by an exponential fit ignoring correlations between time slices. Errors are estimated by the jackknife method with a bin size of 10 configurations (50 HMC trajectories). The light hadron spectrum is obtained by setting the sea quark mass on the physical up and down quark mass and the valence quark mass on the physical up and down quark mass or the strange one. The physical point of the degenerate up and down quark mass is fixed with $m_\pi$ and $m_\rho$ as input. For the strange quark mass the physical point is determined from $m_K$ or $m_\phi$. The lattice scale $a^{-1}$ is set with $m_\rho$. \subsection{Chiral extrapolations in the sea and valence quark masses} In Fig.~\ref{fig:msn_chl_full} we plot a result for the sea and valence quark mass dependences of hadron masses at $\beta=1.95$. The data are parameterized by the averaged hopping parameter $1/K_{av}=(1/K_{val(1)}+1/K_{val(2)})/2$ of the two kind of valence quarks constituting the hadrons. $S$ represents a valence quark with $K_{val}=K_{sea}$ and $V$ a valence quark with $K_{val}\ne K_{sea}$. \begin{table}[t] \caption{\label{tab:para_full} Simulation parameters for full QCD.} \begin{tabular}{lccccllll} $\beta$ & $L^3\times T$ & $c_{SW}$ & $a$[fm] & $La$[fm] & \multicolumn{4}{c}{$m_{\pi}/m_{\rho}$ for sea quarks} \\ & & & & & \multicolumn{4}{c}{\#traj.} \\ \tableline 1.80 & $12^3\times 24$ & $1.60$ & 0.215(2) & 2.58(3) & 0.8060(7) & 0.753(1) & 0.696(2) & 0.548(4) \\ &&&&& 6250 & 5000 & 7000 & 5250 \\ 1.95 & $16^3\times 32$ & $1.53$ & 0.153(2) & 2.45(3) & 0.8048(9) & 0.751(1) & 0.688(1) & 0.586(3) \\ &&&&& 7000 & 7000 & 7000 & 5000 \\ 2.10 & $24^3\times 48$ & $1.47$ & 0.108(2) & 2.59(5) & 0.806(2) & 0.757(2) & 0.690(3) & 0.575(6) \\ &&&&& 2000 & 2000 & 2000 & 2000 \\ 2.20 & $24^3\times 48$ & $1.44$ & 0.086(3) & 2.06(6) & 0.800(2) & 0.754(2) & 0.704(3) & 0.629(5) \\ &&&&& 2000 & 2000 & 2000 & 2000 \\ \end{tabular} \end{table} We observe that the data of the meson masses are distributed on different four lines in accordance with the four sea quarks. Partially quenched data on each sea quark show almost linear behavior both for $m_{PS}^2$ and $m_{Vec}$. Their slopes, however, slightly depend on the sea quark masses: As the sea quark mass decreases the slope decreases for $m_{PS}^2$, while the slope increases for $m_{Vec}$. In the $S$-$S$ channel $m_{PS}^2$ behaves almost linearly in $1/K_{sea}$, whereas $m_{Vec}$ exhibit a convex curvature. Chiral extrapolations of the meson masses are made by global fits of all the data assuming quadratic functions in terms of $1/K_{sea}$ and $1/K_{val}$. For the PS meson mass we employ \begin{eqnarray} m_{PS}^2&=& B_s m_{sea}+B_v{\overline m}_{val}+C_s m_{sea}^2 +C_v{\overline m}_{val}^2+C_{sv}m_{sea}{\overline m}_{val} +C_{12}m_{val(1)}m_{val(2)}, \end{eqnarray} where bare quark masses are defined by $m_{sea/val(i)}=(1/K_{sea/val(i)}-1/K_c)/2$ and ${\overline m}_{val}$ is the averaged value of two valence quark masses. Similar fitting function without the valence-valence cross term $m_{val(1)}m_{val(2)}$ is used for the vector meson masses and the decuplet baryon masses. The fitting results for the meson masses are drawn in Fig.~\ref{fig:msn_chl_full}. As for the octet baryon masses we find a rather complicated situation in Fig.~\ref{fig:msn_chl_full}: Partially quenched baryon masses are not functions of ${\overline m}_{val}$. For the fitting function we take a combination of linear terms based on the ChPT prediction and general quadratic terms of individual $m_{val(i)}$. This function has 12 free parameters in all, which are determined by a combined fit of $\Sigma$-like and $\Lambda$-like baryon masses. \begin{figure}[b] \begin{minipage}[b]{80mm} \centerline{\epsfxsize 68.5mm \epsfbox{kuramashi603fig7a.eps}} \centerline{\epsfxsize 68.5mm \epsfbox{kuramashi603fig7b.eps}} \centerline{\epsfxsize 68.5mm \epsfbox{kuramashi603fig7c.eps}} \caption[]{\label{fig:msn_chl_full} Sea and valence quark mass dependence of hadron masses at $\beta=1.95$. Solid and broken lines represent results of quadratic fits for chiral extrapolations. See text for the labels $S$ and $V$.} \end{minipage} \hspace{\fill} \begin{minipage}[b]{80mm} \centerline{\epsfxsize 73mm \epsfbox{kuramashi603fig8a.eps}} \centerline{\epsfxsize 73mm \epsfbox{kuramashi603fig8b.eps}} \vspace*{1mm} \centerline{\epsfxsize 73mm \epsfbox{kuramashi603fig8c.eps}} \caption[]{\label{fig:spectrum_full} Light hadron spectrum in two-flavor full QCD with $m_K$ as input for the strange quark mass as a function of the lattice spacing. For comparison quenched QCD results in the last section are also plotted.} \end{minipage} \end{figure} \subsection{Sea quark effects on the light hadron spectrum} The results for the light hadron spectrum are presented in Fig.~\ref{fig:spectrum_full} as a function of the lattice spacing. For comparative purpose we also plot the quenched results(open symbols) from the last section, which would help us distinguish the dynamical sea quark effects. We observe an intriguing feature in the meson sector: The $K^*$ and $\phi$ meson masses in full QCD are heavier than those in quenched QCD at finite lattice spacing and become closer to experimental values as the lattice spacing decreases. However, our full QCD results do not necessarily agree with experiment in the continuum limit. We have two possible reasons. One is the relatively heavy sea quark masses corresponding to $m_\pi/m_\rho\approx 0.6-0.8$, for which the chiral extrapolations are rather ambiguous. The other is the quenched treatment of the strange quark. For the baryon sector it is hard to read any meaningful implication in comparison between the full QCD results and the quenched ones. At the finite lattice spacing the full QCD spectrum is consistent with the quenched ones within rather large errors, while the lattice spacing dependence of the decuplet baryon masses in full QCD is opposite to that in quenched case. \section{Conclusions} We have presented our results for the quenched light hadron spectrum. After examining the validity of QChPT in the pseudoscalar meson masses we applied its hadron mass formulae to the chiral extrapolations. Although the QChPT predictions reproduce our hadron mass data well, for the vector meson and baryon masses it is hard to confirm the contributions of the term specific to QChPT. The spectrum in the continuum limit deviates systematically and unambiguously from experiment on a level of about $10\%$ far beyond the statistical error of $1-2\%$ for mesons and $2-3\%$ for baryons. Our full QCD study is an exploring step toward a realistic QCD simulation. For the light hadron spectrum we found an encouraging result that the deficiency of the meson hyperfine splitting in the quenched approximation is largely compensated with the dynamical sea quark effects. At this stage, however, the most important conclusion is as follows: To perform a close investigation of the sea quark effects we need a direct comparison between the full QCD results and the quenched ones employing the same gauge and quark actions with the same $a^{-1}$, $m_\pi/m_\rho$ and $La$. Work in this direction is now in progress. I am grateful to all the members of the CP-PACS Collaboration for their help in preparing this manuscript. I would particularly like to thank R.~Burkhalter and T.~Yoshi{\'e}. This work is supported in part by the Grants-in-Aid of Ministry of Education(Nos. 08NP0101 and 09304029).
\section{Introduction} In spite of the extraordinary successes of the Standard Model (SM) in explaining all the high energy experiments with large luminosity up to now \cite{Altarelli98}, there have been various attempts to search for physics beyond the SM. Even though some astrophysical and collider measurements have intimated that the SM is not the whole story, the signals have been regarded as hints, not as facts. A major portion of the motivations to extend the SM comes from conceptual discomfort in the theoretical viewpoints, such as the hierarchy problem: the SM cannot provide a satisfactory answer to why the nature allows such an enormous ratio between two fundamental mass scales, the electroweak scale at $\sim 100$ GeV and the Planck mass scale at $\sim 10^{19}$ GeV. This problem has prompted the extensive studies of the supersymmetric models \cite{Susy}, the technicolor models \cite{Technicolor}, etc. Recently Arkani-Hamed, Dimopoulos, and Dvali (ADD) have approached the hierarchy problem in a salient way by removing one prerequisite of the problem itself: the Planck mass is not fundamental, that is, there exists only one fundamental mass scale, $M_S$, in the nature \cite{ADD1}. The observed extremely small Newton's constant $G_N$ or alternatively huge Planck mass scale can be obtained by introducing the existence of extra compact $N$ dimensions. Then the Planck mass is related to the $M_S$ and the size of extra dimensions. For example, if the extra dimensional space is the $N$-dimensional torus with the same compactification radii $R$, the relation is \begin{equation} \label{Planck} \kappa^2 R^N = 16 \pi (4 \pi)^{N \over 2} \Gamma\left( \frac{N}{2} \right) M_S^{-(N+2)} \,, \end{equation} where $\kappa^2\equiv 16 \pi G_N$ \cite{Han}. The excellence of Newtonian mechanics in describing the solar system excludes the $N=1$ case where the $R$ is order of $10^{13}$ cm. The $N=2$ case which implies mm scale extra dimensions is not excluded by the current macroscopic measurement of gravitational force \cite{macro}. The cases of $N>2$ are also acceptable but difficult to probe through macroscopic observations. In order to be phenomenologically consistent with the SM at the electroweak scale, the model assumes that the SM fields are confined in our 4-dimensional brane while gravitons are freely propagating in the whole $(4+N)$-dimensional bulk. Even though this discrimination can be achieved by considering our world as a topological defect of a higher dimensional theory and by localizing the SM fields at the vortex \cite{ADD1}, its natural realization had been already discussed in string theories \cite{ADD2,Sundrum}. Strong string coupling obtained by T-dualising transforms the Kaluza-Klein modes of opens strings into the winding modes of open strings of which the two ends are attached on a brane, which are good candidates of the SM particles. And closed strings are still able to propagate orthogonal to the brane, identified as gravitons. The existence of extra dimensions reveals through the interactions between gravitons and the SM particles on our brane. Although the required invariance under general coordinate transformations in the brane and bulk can, in principle, specify the interactions, the presence of non-trivial metric in the extra dimension complicates them. Unless gravitons take momentum compatible with or larger than $M_S$ at high energy collisions, the spacetime region where collisions occur can be regarded as flat \cite{Wells}. Then the linear approximation is valid so that the metric in $(4+N)$-dimensions can be expanded around the Minkowski metric as \begin{equation} \label{2} \hgmn = \hetamn +\hkp \hhmn \,, \end{equation} where the $\hhmn$ is the canonical graviton fields and the hatted indices denote the $(4+N)$-dimensional spacetime. After the compactification on a sub-manifold of extra dimensions, our brane effectively possesses Kaluza-Klein towers of massive spin 2 gravitons $\thmnn$, massive vector particles $\tilde{A}_{\mu i}^\vn $, and massive spin zero particles $\tilde{\phi}_{ij}^\vn $. The matter-graviton couplings are specified by the minimal coupling of gravity, yielding the Feynman rules of the interactions between the Kaluza-Klein gravitons and the SM particles \cite{Sundrum,Han,Wells}. Following the results of Ref.\cite{Han}, we use the effective action to the leading order in $\kappa=1/M_{\rm pl}$ \begin{equation} \label{3} I=-\frac{\kappa}{2} \sum_\vn \int d^4 x \left[ \tilde{h}^{\mu\nu,\vn} T_{\mu\nu} + w \tilde{\phi}^\vn T^\mu_\mu \right] \,, \end{equation} where $w=\sqrt{2/3(n+2)}$ and $T^{\mu\nu}$ is the energy-momentum tensor. It is to be noted that new kinds of interactions due to the low scale quantum gravity are neutral current ones. Recently the idea of the existence of large extra dimensions has drawn quite explosive attentions of particle physicists. Above all, attractive is that the quantum gravity scale can be as low as TeV so to be testable. It would be worthwhile to search for collider tests which can confirm or exclude the validity of the model. Various phenomenological studies have been performed to constrain the scale of $M_S$ or the number of extra dimensions $N$ from the existing data of collider experiment \cite{Peskin,Hewett,Rizzo}, and in the future experiments \cite{Hewett,Rizzo,top,Agashe,Balazs,Cheung}. Two kinds of processes at colliders are computable by using the effective action in Eq.(\ref{3}), single graviton emission and virtual exchange of gravitons. For the first case, the presence of extra dimension induces the Kaluza-Klein multiplicity of phase factor, proportional to $(\Delta E \cdot R)^N$, where $\Delta E$ is the energy of emitting graviton. Thus any process involving single graviton emission shows the large dependence on $N$ since the branching ratio is proportional to $(\Delta E/M_S)^{N+2}$ \cite{ADD1}. Studies of various processes show that even in the future NLC or LHC, the $N\geq 5$ cases are practically impossible to distinguish from the SM background \cite{Hewett,Rizzo,top,Agashe}. Virtual exchange of gravitons does not have such dependence; the scattering amplitudes, suppressed by $\kappa^2$ and mediated by Kaluza-Klein towers, are proportional to $1/M_S^4$. Their dependences on the $N$, according to compactification models, may not be as critical as in the single graviton emission case \cite{Hewett,Rizzo}. Moreover in this case, the polarizations of incoming and outgoing particles are affected by the fact that gravitons are of spin two while photons or $Z$ bosons, mediating the SM neutral currents, are of spin one. Therefore, one of the most useful and significant measurements to probe the existence of the extra dimensions would be the measurements of the polarizations at virtual graviton exchange processes. In this paper we concentrate on the process $e^+ e^- \to W^+ W^-$ in the future Linear Colliders (LC) \cite{LC}. The spin-one nature of the $W$ bosons provides more channels to signal new physics, and the measurements of the $W$ polarizations are attainable from the decay angular distributions \cite{Wpol}. Therefore it is desirable to derive in detail the effects of large extra dimensions on the $W$ and beam polarizations. \vskip 1.5cm \begin{center} \begin{picture}(230,100)(0,0) \Text(55, 5)[]{(a)} \Text(15,100)[]{$e^-$} \Text(15,20)[]{$e^+$} \Text(40,60)[]{$\nu_e$} \ArrowLine(10,90)(50,90) \ArrowLine(50,90)(50,30) \ArrowLine(50,30)(10,30) \Photon(50,90)(90,90){3}{7} \Photon(50,30)(90,30){3}{7} \Text(95,100)[]{$W^-$} \Text(95,20)[]{$W^+$} \Text(170, 5)[]{(b)} \ArrowLine(120,90)(150,60) \ArrowLine(150,60)(120,30) \Text(125,100)[]{$e^-$} \Text(125,20)[]{$e^+$} \Text(170,80)[]{$\gamma,Z,\tilde{h}_{\mu\nu}^{\vec{n}}$} \Photon(150,60)(190,60){3}{5} \Photon(190,60)(220,90){3}{5} \Photon(190,60)(220,30){3}{5} \Text(215,100)[]{$W^-$} \Text(215,20)[]{$W^+$} \end{picture} \end{center} \smallskip { } \noindent Figure~1: {\it Feynman Diagrams contributing to the process $e^+e^-\rightarrow W^+ W^-$ including large extra dimension effects. } \smallskip \smallskip For the process \begin{equation} e^-(p_1,\kp)+e^+(p_2,\overline{\kp})\to W^-(q_1,\lm)+W^+(q_2,\overline{\lm}) \,, \end{equation} there are four Feynman diagrams, one $t$-channel diagram mediated by the neutrino and three $s$-channel ones mediated by the photon, $Z$ boson, and spin-2 gravitons as depicted in Fig.1. At high energies where the electron mass is negligible, there are 18 different helicity amplitudes ${\mathcal M}^\pm (\lm,\overline{\lm})$, where superscripts denote the electron helicities. Since the CP-invariance relates some amplitudes such as \begin{equation} {\mathcal M}^\pm (\lm,\overline{\lm}) = {\mathcal M}^\pm (-\overline{\lm},-\lm) \,, \end{equation} thus only 12 amplitudes are independent \cite{Denner}. By using the helicity formalism in Ref.~\cite{HAmp} and the polarization convention in Ref.~\cite{Denner}, we calculate the helicity amplitudes for the left-handed electron beam as \begin{eqnarray} \label{LH} {\mathcal M}^-_{++} &=& s_\theta [ f_L\bw + \frac{g^2 s}{4 t}(c_\theta-\bw) -2f_D c_\theta(1-\bw^2) ] \,, \\ \no {\mathcal M}^-_{+0} &=& -\frac{\sqrt{2}(1+c_\theta)}{\mw}[ f_L\bw + \frac{g^2 s}{8 t} (2c_\theta - 1-2\bw+\bw^2) -f_D(2c_\theta-1)(1-\bw^2)] \,, \\ \no {\mathcal M}^-_{+-} &=& 2 s_\theta (1+c_\theta)[ \frac{g^2 s}{8 t} -f_D ] \,, \\ \no {\mathcal M}^-_{0+} &=& \frac{\sqrt{2}(1-c_\theta)}{\mw}[ f_L\bw + \frac{g^2 s}{8 t} (2c_\theta+1-2\bw-\bw^2) -f_D(2c_\theta+1)(1-\bw^2)] \,, \\ \no {\mathcal M}^-_{00} &=& -\frac{s_\theta}{\mw^2} [ f_L\bw (3-\bw^2) + \frac{g^2 s}{4 t} (2c_\theta-3\bw+\bw^3) -2f_D c_\theta(2-3\bw^2+\bw^4) ] \,, \\ \no {\mathcal M}^-_{-+} &=& -2s_\theta (1-c_\theta) [ \frac{g^2 s}{8 t} -f_D ]\,, \end{eqnarray} and for the right-handed electron beam we have \begin{eqnarray} \label{RH} {\mathcal M}^+_{++} &=& s_\theta[ f_R\bw - 2f_D c_\theta(1-\bw^2) ] \,, \\ \no {\mathcal M}^+_{+0} &=& \frac{\sqrt{2}(1-c_\theta)}{\mw}[ f_R\bw -f_D(2c_\theta+1)(1-\bw^2) ] \,, \\ \no {\mathcal M}^+_{+-} &=& 2 f_D s_\theta(1-c_\theta) \,, \\ \no {\mathcal M}^+_{0+} &=& -\frac{\sqrt{2}(1+c_\theta)}{\mw}[ f_R\bw -f_D(2c_\theta-1)(1-\bw^2) ] \,, \\ \no {\mathcal M}^+_{00} &=& -\frac{s_\theta}{\mw^2}[ f_R\bw(3-\bw^2)-2f_D c_\theta(2-3\bw^2+\bw^4) ] \,, \\ \no {\mathcal M}^+_{-+} &=& -2 f_D s_\theta(1+c_\theta)\,. \end{eqnarray} The CP-invariance implies \begin{equation} {\mathcal M}^\pm_{--}={\mathcal M}^\pm_{++}, \quad {\mathcal M}^\pm_{-0}={\mathcal M}^\pm_{0+}, \quad {\mathcal M}^\pm_{0-}={\mathcal M}^\pm_{+0}. \end{equation} In Eqs.(\ref{LH}) and (\ref{RH}) we denote $s_\theta=\sin\theta$, $c_\theta=\cos\theta$, and \begin{eqnarray} \es_L&=& -\hf +\sin^2\theta_W,\quad \es_R =\sin^2\theta_W\,, \\ \no f_{L,R} &=& \frac{g^2 \es_{L,R}}{1-M_Z^2/s} -e^2\,,\quad \\ \no \mw &=& \frac{M_W}{\sqrt{s}/2} =\sqrt{1-\beta^2_W} \,. \end{eqnarray} The terms proportional to $f_{L,R}$ denote the contributions from the $\gamma$- and $Z$-mediated diagrams, and those proportional to the $g^2 s/t$ from the neutrino-mediated one. It can be easily seen that the preparation of right-handed electron beam switches off the $t$-channel $\nu$-mediated contributions. These SM results are consistent with those in Ref.\cite{Denner,Agashe}. The low scale quantum gravity effects are included in $f_D$, defined by \begin{eqnarray} \label{f_D} f_D \equiv -\frac{\pi s^2}{2 M_{\rm pl}^2 } \sum_{\vn} \frac{1}{s-m_\vn^2} &\simeq & \frac{\pi s^2}{2M_S^4} \ln ( \frac{M_S^2}{s} ) ~\quad {\rm for} ~~ N=2 \\ \no & \simeq& \frac{\pi s^2}{(N-2) M_S^4} \quad\quad {\rm for} ~~ N>2 \,. \end{eqnarray} It is to be noted that the two helicity amplitudes ${\mathcal M}^+_{+-} $ and ${\mathcal M}^+_{-+} $ vanish at the tree level in the SM, but retain extra dimension effects as sizable as the other amplitudes. In Fig.~2 and 3, we plot the differential cross sections for the $W$ pair production at $e^+ e^-$ collisions with respect to the $W^-$ scattering angle against the electron beam at $\sqrt{s}=1$ TeV in the case of $N=2$ and $M_S=2.5$ TeV, broken down to the transverse and longitudinal helicity components of the $W$ bosons. The cases with $N>2$ unless $N$ is too large shows similar behaviors. The effects of large extra dimensions with respect to the SM background can be enhanced by using the right-handed electron beam and selecting the $W^+ W^-$ polarizations to be both transverse. This is because the employment of the right-handed electron beam eliminates the dominant SM contributions of the $t$-channel $\nu$-mediated diagram in Fig.1. And the SM background from the $s$-channel with the right-handed electron beam is proportional to \begin{equation} f_R =e^2 \left[ \frac{1}{1-M_Z^2/s}-1 \right] \propto \frac{M_Z^2}{s}\qquad{\rm for}\quad s \gg m_Z, \end{equation} which $decreases$ as the beam energy becomes larger, while the extra dimension correction proportional to $f_D$ in Eq.(\ref{f_D}) $increases$. Furthermore, at the tree-level, ${\mathcal M}^{+{\rm (SM)}}_{+-}={\mathcal M}^{+{\rm (SM)}}_{-+}=0$ in the SM. The $\sigma_{_{\rm TT}}$ including large extra dimension effects is about $\sim10^4$ times the SM background. In practice, the generation of 100\% polarized electron beam is infeasible. We consider the expected polarization of the electron beam as 90\%. Figure 4 shows the differential cross sections against the $W$ scattering angle according to the $W$ polarizations at $\sqrt{s}=1$ TeV when $N=2$ and $M_S=2.5$ TeV. The dominant $t$-channel SM background contaminates the unique behavior of $d \sigma_{TT}$, however, substantial corrections to the SM background still remain. We do not consider the polarization of the positron beam since it is more difficult to generate, expected presumably in the range of 60 $\%$ to $65\%$ \cite{LC}. No radiative corrections are included here. The SM radiative corrections have the effects of the same orders of magnitude on the results with or without considering this model since the effects of the extra dimensions mainly come from the interference with the SM amplitudes. \vskip 1.0cm \noindent \begin{center} \begin{tabular}{|c|cc|cc|}\hline &\multicolumn{2}{|c|}{~~$\sqrt{s}=0.5$ TeV ( $\int{\mathcal L} = $50 fb$^{-1}$)~~} &\multicolumn{2}{|c|}{~~$\sqrt{s}=1$ TeV ( $\int{\mathcal L} = $200 fb$^{-1}$)~~} \\ \hline {} &~~ $N=2$~~ &~~ $N=6$~~ & ~~$N=2$~~ &~~ $N=6$~~ \\ \hline ~~$\sigma_{\rm tot}^{\rm unpol}$ ~~& 4.6 & 2.6 & 9.3 & 5.3 \\ $\sigma_{\rm TT}^{\rm unpol}$ & 4.6 & 2.6 & 9.3 & 5.4 \\ $~~\sigma_{\rm tot}^{\rm 90\%\,RH}~~$ & 3.4 & 2.0 & 6.8 & 4.1 \\ $\sigma_{\rm TT}^{\rm 90\%\,RH}$ & 3.4 & 2.0 & 7.0 & 4.1 \\ $\sigma_{\rm tot}^{\rm 90\%\,LH}$ & 5.1 & 2.9 & 10.1 & 5.8 \\ $\sigma_{\rm TT}^{\rm 90\%\,LH}$ & 5.1 & 2.9 & 10.2 & 5.8 \\ \hline \end{tabular} \end{center} {Table~1}. {\it The LC bounds of $M_S$ in TeV at 95\% confidence level according to the beam or $W$ pair polarizations.} \vskip 0.5cm The LC bounds on the $M_S$ are derived by the statistical errors with the angular cut $|\cos\theta|<0.95$ at 95\% confidence level at $\sqrt{s}=0.5$, 1.0 TeV and $N=2,6$ from six different observables according to the beam and $W$ pair polarizations. As for the beam polarization effects on the $M_S$ bounds, the preparation of the left-handed electron beam is expected to yield higher bounds. With a given beam polarization, the $\sigma_{_{\rm TT}}$'s are likely to give higher $M_S$ bounds. On account of the smaller numbers of transversely polarized $W$ pair events, these results imply that large extra dimension corrections in the $\sigma_{_{\rm TT}}$'s are much larger than those in the $\sigma_{\rm tot}$. It is concluded that valuable information about the models with large extra dimensions can be obtained by observing the $W$ pair and beam polarizations at the $e^+ e^-\to W^+ W^-$ process. In particular, the measurements of the cross section for transversely polarized $W$ pair with the right-handed electron beam highly enhances the possibilities to see the low scale quantum gravity effects. The current inability to generate purely polarized beam at $e^+ e^-$ colliders contaminates this feature. Almost purely polarized beams are possible in the future $\mu^+ \mu^-$ colliders \cite{muon-collider}, since the muons are prepared through the pion decays accompanied by purely chiral neutrinos. We expect definite signal of the large extra dimension effects through the observations of transverse polarizations of $W$ pair with the right-handed muon beam at the muon colliders. It has been shown that for the LC bounds of the string scale $M_S$ the use of left-handed electron beam is preferred, and for the probe of large extra dimension effects the measurements of the cross section for transversely polarized $W$ pair are. \acknowledgments We would like to appreciate valuable discussions with S.Y. Choi and T. Lee. The work was supported in part by the Korea Science and Engineering Foundation (KOSEF) through the Center for Theoretical Physics, Seoul National University. HSS would like to acknowledge the financial support of the Korea Research Foundation through the 97 Sughak Program and 98--015-D00054. \bigskip \bigskip
\section*{ ACKNOWLEDGMENTS } \end{center} The authors would like to thank the partial financial support by National Natural Science Foundation of China under Grant Nos. 19675044 and 19875026.
\section{Introduction} The Weyl representation places quantum mechanics in phase space. The density operator is mapped into a real phase function that projects onto the position or momentum probability densities. The unitary operators corresponding to linear canonical transformations transform the Weyl symbol of any operator as a classical variable. Therefore, it is not surprising that the study of the semiclassical limit of nonintegrable systems has relied heavily on the Weyl representation as reviewed by reference \cite {ozrep}. The study of systems that are chaotic in the classical limit has developed several models supported by a compact phase space. The simplest choice is that of a $\left( 2L\right) $-dimensional torus, corresponding to a system with $L$ degrees of freedom. Indeed, if the system propagates in discrete time ( a mapping of the torus onto itself) even the special case $L=1$ may be chaotic as is the case of the cat map \cite{arnold}, \cite{hanay} or the baker's map \cite{balavoros}, \cite{saraceno}. It is well known that the Hilbert space corresponding to a classical torus is finite. We may picture the $N$ allowed positions and momenta as forming a lattice, the {\it quantum phase space} (QPS), even though no rigorous definition of position and momentum operators is available on the torus \cite{pqtor}. The semiclassical limit is then obtained as $N\sim \left( 2\pi \hbar \right) ^{-1}\rightarrow \infty .$ Though it is a great advantage to investigate numerically the propagation of finite vectors or matrices defined in the QPS, it is somewhat disconcerting that the difference between classical and quantum motion is more extreme on the torus than on the plane. This problem also manifests itself in the adaptation of the Weyl representation to the torus. The existing literature \cite{wooters}-\cite{kaperpeev} relies mainly on formal procedures, so that the achievement of the classical limit as $N\rightarrow \infty $ may not be preceded by the emergence of classical structures even for finite $\hbar $, such as have been found for a plane phase space. Clearly, a way to avoid this difficulty is to consider the classical torus as a periodic plane phase space, to quantize the latter and then to project this on QPS. We thus generalize the procedure of Hannay and Berry \cite {hanay}, allowing for arbitrary ''Bloch '' or ''Floquet ''angles for each circuit of the torus. It is then possible to project the appropriate plane translation and reflection operators onto the torus. Hence, we define the Weyl (or center) representation and its conjugate chord representation while maintaining the main geometrical features characteristic of the plane. In section 2 we present the translation operators of momentum and position for a torus considered as the fundamental unit cell of the periodic plane. By also allowing tori made up of more than one cell, we show that the unit quantum torus may be obtained as the projection of the Hilbert space corresponding to such a larger torus. Finally, by connecting the Hilbert space for the plane to that of an infinitely large torus, we obtain the projection of plane operators onto the torus. Section 3 summarizes important features of the Weyl representation on the plane. Defining the translation operators and their Fourier transform, the reflection operators, we derive the Weyl symbols and their product rules in terms of integrals over phase space polygons. In section 4 we project appropriate translation operators onto QPS. However, it is found that the reflection operators are supported on a lattice with half the spacing of QPS . In consequence, the trace of these operators is not homogeneous, which leads to complications in the formulae for products of operators. Only in the case that $N$ is odd, can we simplify the product rule for the Weyl symbols into a form that is analogous to the theory in the plane. In any case, we need only half the number of sums derived in the previous work of Galleti and Toledo Pisa \cite{galeti1} and these sums depend on the symplectic areas of the same polygons arising in the plane theory. Finally, we discuss the restricted form of symplectic invariance that holds for the Weyl and chord representation on the torus: The Weyl symbols transform classically under the action of quantum cat maps. Though we have motivated our paper through discrete time models, periodic Hamiltonians have obvious applications in solid state physics. Thus, section 5 is dedicated to the derivation of a path integral for the Weyl symbol of the propagator. This relies on the symplectic area of polygons as in the plane theory. In the semiclassical limit, the propagator is expressed in terms of the center generating function presented in reference \cite{ozrep}. Throughout this work we differentiate operators on the plane by italic $% \widehat{A}$, as opposed to bold operators $\widehat{{\bf A}}$ on the torus. \section{Hilbert space for tori} \setcounter{equation}{0} Classical phase space of ($2L)$ dimensions may be considered to be periodic, so that we confer to it the topology of a $(2L)$ dimensional torus. Evidently, we may use invariance, with respect to symplectic transformations, to equate the periods $\Delta q=\Delta p=\nu $ of the position and momentum coordinates. The usual choice is $\nu =1$, but we will leave this as a free integer parameter so as to study the nesting of tori, that is, the case where quantization is imposed on a larger (periodic)\ region than the unit cell. Thus, the number of unit cells will be $\nu ^{2L}. $ It is important to treat the specification of the Hilbert space of quantum states for the torus, or {\it prequantization}, independently from the dynamics of the system. That is, we treat the quantum kinematics, corresponding to the geometrical description of phase space at the classical level. A complete description for prequantization must include boundary conditions; which are here that the wave functions satisfy Bloch conditions: \begin{eqnarray} {\bf \Psi }(q+\nu ) &=&e^{2\pi i\chi _p}{\bf \Psi }(q), \label{eq1} \\ {\bf \tilde{\Psi}}(p+\nu ) &=&e^{-2\pi i\chi _q}{\bf \tilde{\Psi}}(p) \label{eq:eq2} \end{eqnarray} where \begin{equation} {\bf \tilde{\Psi}}(p)=(2\pi \hbar )^{-1/2}\int e^{-ipq/\hbar }{\bf \Psi }% (q)dq \label{eq:91} \end{equation} and $2\pi i\chi _p$ and $2\pi i\chi _q$ are fixed arbitrary Floquet angles; that is, the prequantization depends on the vector $\chi =(\chi _p,\chi _q)$ whose coordinates are in the range $0\le \chi _q,\chi _q<1$. It is a well known kinematical restriction \cite{debievre} for torus quantization that there are \begin{equation} \nu ^2N=\nu ^2(2\pi \hbar )^{-1} \label{eq:nh} \end{equation} basis states for each degree of freedom, so that $N=(2\pi \hbar )^{-1}$ is the number of states corresponding to the unit cell .This is a crucial point: the compactness of the phase space implies in the finiteness of the dimension of the Hilbert space. Recalling the translation operators $\widehat{T}_p=\exp \left( \frac{i\alpha }\hbar \widehat{q}\right) \,$ and $\widehat{T}_q=\exp \left( -\frac{i\beta }% \hbar \widehat{p}\right) $ that respectively translates momentum by $\alpha $ and position by $\beta $ in the plane, we define minimal translators on the torus $\widehat{{\bf T}}_{p,\nu ^2N}$ and $\widehat{{\bf T}}_{q,\nu ^2N}$ with their discrete eigenstates $|{\bf q}_n,\nu ^2N>$ and $|{\bf p}_m,\nu ^2N>$ such that \begin{equation} \begin{array}{cc} \widehat{{\bf T}}_{p,\nu ^2N}|{\bf q}_n,\nu ^2N>=e^{\left[ \frac{2\pi i}{\nu ^2N}(n+\chi _q)\right] }|{\bf q}_n,\nu ^2N> & \widehat{{\bf T}}_{p,\nu ^2N}|% {\bf p}_m,\nu ^2N>=|{\bf p}_{m+1},\nu ^2N> \\ \widehat{{\bf T}}_{q,\nu ^2N}|{\bf q}_n,\nu ^2N>=|{\bf q}_{n+1},\nu ^2N> & \widehat{{\bf T}}_{q,\nu ^2N}|{\bf p}_m,\nu ^2N>=e^{\left[ -\frac{2\pi i}{% \nu ^2N}(m+\chi _p)\right] }|{\bf p}_m,\nu ^2N> \end{array} \end{equation} The product of $\nu ^2N$ translations on the basis states $|{\bf q}_n,\nu ^2N>$ or $|{\bf p}_m,\nu ^2N>\,$must return these to the same state, i.e. these Schwinger operators \cite{schwinger} satisfy \begin{equation} \left( \widehat{{\bf T}}_{p,\nu ^2N}\right) ^{\nu ^2N}=e^{2\pi i\chi _q}% \widehat{{\bf 1}}_{\nu ^2N}^\chi \end{equation} and \begin{equation} \left( \widehat{{\bf T}}_{q,\nu ^2N}\right) ^{\nu ^2N}=e^{-2\pi i\chi _p}% \widehat{{\bf 1}}_{\nu ^2N}^\chi \ . \end{equation} To define the Hilbert space ${\cal H}_{\nu ^2N}^\chi $, we add the Hermitian structures \begin{equation} <{\bf q}_n,\nu ^2N|{\bf q}_{n^{\prime }},\nu ^2N>=\delta _{n,n^{\prime }}^{(\nu ^2N)}e^{\frac{2\pi i}{\nu ^2N}(n-n^{\prime })\chi _p}, \label{eq:QQsim} \end{equation} and \begin{equation} <{\bf p}_m,\nu ^2N|{\bf p}_{m^{\prime }},\nu ^2N>=\delta _{m,m^{\prime }}^{(\nu ^2N)}e^{-\frac{2\pi i}{\nu ^2N}(m-m^{\prime })\chi _q}. \label{qbastgp} \end{equation} Here we define the {\it $N$-periodic Kronecker delta } \begin{equation} \delta _{m,n}^{(N)}\equiv \sum_{j=-\infty }^\infty \delta _{m,n+jN}. \label{Nkrone} \end{equation} The bases are exchanged with the transformation kernel \begin{equation} <{\bf p}_m,\nu ^2N|{\bf q}_n,\nu ^2N>=\frac 1{\nu N^{1/2}}e^{2\pi i\frac 1{\nu ^2N}(m+\chi _p)(n+\chi _q)}\equiv F_{m,n}, \label{eq:PQ} \end{equation} forming a unitary matrix ( finite Fourier transformation). Clearly, this last expression allows us to interpret the position ${\bf q}_n$ as corresponding to $q_n=\frac 1{\nu N}(n+\chi _q)$, whereas ${\bf p}_m$ corresponds to $p_m=\frac 1{\nu N}(m+\chi _p)$, leading to \begin{equation} <{\bf p}_m,\nu ^2N|{\bf q}_n,\nu ^2N>=(2\pi \hbar )^{-\frac 12}\exp \left( \frac i\hbar p_mq_n\right) . \end{equation} Likewise the $q$-Translator $\widehat{T}_q$, corresponds to a translation in the plane by $\Delta q=\frac 1{\nu N}$ and the phase change $\exp \left[ 2\pi i\chi _q\right] $ results from the translation $\Delta q=\nu $ $\,$% around the torus. Although the indices $n$ and $m$ can run over all integers, only $\nu ^2N$ successive values will form a basis for the torus Hilbert space ${\cal H}_{\nu ^2N}^\chi .$ We may keep to the fundamental range $\left[ 0,\nu ^2N-1\right] $, corresponding to the square with side $% \nu $, or extend to the periodic plane, by taking into account the phases $% \chi $. These considerations apply to each of the $L$ degrees of freedom, so that in general the fundamental domain is a $\left( 2L\right) $-hypercube. We see that position and momentum form a discrete web on the torus, as shown in Fig.~\ref{fig.3.1}, that we call the quantum phase space (QPS), following reference \cite{galeti1}: \begin{equation} x=\left( \begin{array}{c} p \\ q \end{array} \right) =\frac 1N\left( \begin{array}{c} m+\chi _p \\ n+\chi _q \end{array} \right) . \label{xtorop} \end{equation} \begin{figure}[htb] \centerline {\epsfxsize=5in \epsffile{fig3-1.eps} } \caption{ The quantum Phase space for $N=4$. The intersection of the bold lines in the unit square determine QPS for $\nu =1$, whereas the full figure corresponds to the choice $\nu =3$. } \label{fig.3.1} \end{figure} We now consider the relation between the Hilbert spaces of two nested tori $% {\cal H}_N^\chi $ and ${\cal H}_{\nu ^2N}^{\chi ^{\prime }}$ with $\nu >1.$ The consideration for the QPS corresponding to ${\cal H}_N^\chi $ to be a sublattice of the QPS of ${\cal H}_{\nu ^2N}^{\chi ^{\prime }}$ is first that \begin{equation} \chi ^{\prime }=\nu \chi -k, \label{nunup} \end{equation} where $k=(k_p,k_q)$ is an integer vector that denotes the number of loops around the torus made by $\chi $ when multiplied by $\nu .$ Thus $\chi ^{\prime }$ is uniquely determined by $\chi .$ The indices for the larger Hilbert space ${\cal H}_{\nu ^2N}^{\chi ^{\prime }}$ for points in the fundamental domain of the smaller QPS are given by \begin{eqnarray} m^{\prime } &=&\nu m+k_p \\ n^{\prime } &=&\nu n+k_q. \end{eqnarray} We can now define the Hilbert space ${\cal H}_N^\chi $ as a projection of the larger space ${\cal H}_{\nu ^2N}^{\chi ^{\prime }}.$ Indeed, it is easy to verify that if $|{\bf q}_{n^{\prime }},\nu ^2N>$ are orthogonal position eigenstates for ${\cal H}_{\nu ^2N}^{\chi ^{\prime }}$, then \begin{equation} |{\bf q}_n,N>_\chi =\frac 1{\sqrt{\nu }}\sum_{r=0}^{\nu -1}e^{i2\pi \chi _pr}|{\bf q}_{\nu n+k}+r,\nu ^2N>_{\chi ^{\prime }} \label{eq:q1q2} \end{equation} form an appropriate orthonormal basis for ${\cal H}_N^\chi $. Thus, defining the projection operator \begin{equation} \widehat{{\bf 1}}_N^\chi =\sum_{n=0}^{N-1}|{\bf q}_n,N><{\bf q}_n,N| \label{uno} \end{equation} we verify that this is a Hermitian operator in ${\cal H}_{\nu ^2N}^{\chi ^{\prime }}$, and that \begin{equation} \widehat{{\bf 1}}_N^\chi \widehat{{\bf 1}}_N^\chi =\widehat{{\bf 1}}_N^\chi . \end{equation} Furthermore the states are obtained as \begin{equation} |{\bf \Psi ,}N>=\widehat{{\bf 1}}_N^\chi {\bf |\Psi ,}\nu ^2N>. \end{equation} For all operators $\widehat{{\bf A}}_{\nu ^2N}$ acting in ${\cal H}_{\nu ^2N}^{\chi ^{\prime }},$ there is a projected operator which acts on ${\cal H% }_N^\chi $: \begin{equation} \widehat{{\bf A}}_N=\widehat{{\bf 1}}_N^\chi \widehat{{\bf A}}_{\nu ^2N}% \widehat{{\bf 1}}_N^\chi . \end{equation} If $\widehat{{\bf A}}_{\nu ^2N}$ leaves ${\cal H}_N^\chi $ invariant, i.e. if \begin{equation} \lbrack \widehat{{\bf 1}}_N^\chi ,\widehat{{\bf A}}_{\nu ^2N}]=0, \label{acomu} \end{equation} then \begin{equation} \widehat{{\bf A}}_N=\widehat{{\bf A}}_{\nu ^2N}\widehat{{\bf 1}}_N^\chi . \end{equation} We also verify that, for any pair of operators satisfying (\ref{acomu}) \begin{equation} \widehat{{\bf A}}_N\widehat{{\bf B}}_N=\widehat{{\bf A}}_{\nu ^2N}\widehat{% {\bf B}}_{\nu ^2N}\widehat{{\bf 1}}_N^\chi . \label{eq:abpt} \end{equation} In particular, we obtain the Schwinger translation operators for ${\cal H}% _N^\chi $ as the projection of those in ${\cal H}_{\nu ^2N}^{\chi ^{\prime }} $: \begin{equation} \widehat{{\bf T}}_{q,N}=\widehat{{\bf T}}_{q,\nu ^2N}\widehat{{\bf 1}}% _N^\chi \qquad ;\qquad \widehat{{\bf T}}_{p,N}=\widehat{{\bf T}}_{p,\nu ^2N}% \widehat{{\bf 1}}_N^\chi . \end{equation} Let us now take the limit $\nu \rightarrow \infty $. Clearly, the variable $% q_n$ becomes continuous in this limit as the volume of the torus $\nu ^{2L}\rightarrow \infty $.Throughout the limit, the relation (\ref{nunup}) defines an appropriate $\chi ^{\prime }\left( \chi \right) $. The main step to recover the Banach space ${\cal H}_{{\Bbb R}}$ for the plane $\left( 2L\right) $-dimensional phase space is to redefine the normalization so that \begin{equation} <q|q^{\prime }>=\delta (q-q^{\prime }), \label{qbastg2} \end{equation} introducing the Dirac delta function on the right, and the continuous Fourier integral for the change of basis: \begin{equation} <p|\psi >=\left( 2\pi \hbar \right) ^{-L/2}\int dq\exp \left( \frac{ipq}% \hbar \right) <q|\psi > \end{equation} In all other respects, the kinematics in the plane will coincide with that of an infinitely large torus. However, the normalization condition (\ref {qbastg2}) implies a change in the way we express the sates in ${\cal H}% _N^\chi $ in terms of those of ${\cal H}_{{\Bbb R}}$. For that purpose, we recall that for an unit torus \begin{equation} \sum_{r=0}^{N-1}e^{i\frac{2\pi }N(m-n)r}=N\delta _{m,n}^{(N)}, \label{eq:114} \end{equation} so we extend the definition of the $N$-periodic Kroeneker delta function to real numbers $x$ and $y$: \begin{equation} \delta _{x,y}^{(N)}\equiv \left\langle e^{i\frac{2\pi }N\left( x-y\right) k}\right\rangle _k, \label{eq:kroneinf} \end{equation} where $\left\langle ...\right\rangle _k$ denotes the average over $k,$ \begin{equation} \left\langle ...\right\rangle _k=\lim_{r\rightarrow \infty }\frac 1r\sum_{k=-\frac r2}^{\frac r2}. \label{kprom} \end{equation} From (\ref{eq:kroneinf}) we see that $\delta _{x,y}^{(N)}$ only depends on the difference $x-y$, so let us take $y=0$ for simplicity. For $x=r$ an integer number, the argument in (\ref{eq:kroneinf})\ has period $N$. Then, the average is just (\ref{eq:114}) divided by $N,$ so the definition (\ref {eq:kroneinf})\ is consistent with (\ref{Nkrone}) for integer arguments. Let us now suppose $x=\frac rd,$ a rational number. The argument in (\ref {eq:kroneinf})\ thus has period $Nd$, so that \begin{equation} \delta _{x,0}^{(N)}=\frac 1{Nd}\sum_{k=0}^{Nd-1}e^{i\frac{2\pi }N\frac rdk}=\frac 1{Nd}\frac{1-e^{i2\pi r}}{1-e^{i\frac{2\pi }{Nd}r}}=\left\{ \begin{array}{cc} 1 & \mbox{ if }r=0\qquad \mbox{mod}(Nd) \\ 0 & \mbox{ otherwise} \end{array} \right. . \label{krorac} \end{equation} Hence, once more the $N$-periodic Kroeneker delta function is different from zero only for $x$ being zero modulo $N$. By allowing $d\rightarrow \infty $ in (\ref{krorac}) we can extend the definition of $\delta _{x,y}^{(N)}$ to irrational numbers, so that \begin{equation} \delta _{x,y}^{(N)}=\left\{ \begin{array}{cc} 1 & \mbox{ only if }\left( x-y\right) =0\qquad \mbox{mod}(N) \\ 0 & \mbox{ otherwise} \end{array} \right. . \end{equation} The definition (\ref{eq:kroneinf}) is an interpolation of (\ref{Nkrone}), which will allows us to perform, not only sums with the $N$ periodic Kroeneker delta function, but also integrals. Indeed, we will relie on the formal equivalence: \begin{equation} \delta _{x,y}^{(N)}=\left\langle \delta (\frac{x-y}N-k)\right\rangle _k. \label{eq:dirkro} \end{equation} This is a consequence of the definition(\ref{eq:kroneinf}) and the Poisson sum formula, \begin{equation} \sum_{t\in {\Bbb Z}}\delta (x-t)=\sum_{k\in {\Bbb Z}}e^{i2\pi kx}. \label{eq:poisson} \end{equation} Indeed, from the definition (\ref{eq:kroneinf}) we have \begin{equation} \frac 1\nu \sum_{k=-\frac \nu 2}^{\frac \nu 2}e^{i\frac{2\pi }N\left( x-y\right) k}=\frac 1\nu \sum_{k=-\frac \nu 2}^{\frac \nu 2}\delta (\frac{x-y% }N-k)+\frac 1\nu R_\nu (x-y) \end{equation} with \begin{equation} R_\nu (x-y)=\sum_{k=-\frac \nu 2}^{\frac \nu 2}e^{i\frac{2\pi }N\left( x-y\right) k}-\sum_{k=-\frac \nu 2}^{\frac \nu 2}\delta (\frac{x-y}N-k). \end{equation} From the Poisson sum formula (\ref{eq:poisson}), we have that, $\lim_{\nu \rightarrow \infty }R_\nu (x-y)=0,$ upon an appropriate ordering of the limits $\nu \rightarrow \infty $ and the width of the delta function$% \rightarrow 0$.A\ consequence of (\ref{eq:dirkro}) is that, for any function $f(t),$ \begin{equation} \left\langle \delta (\frac{x-y}N-t)f(t)\right\rangle _t=\delta _{x,y}^{(N)}\;f(\frac{x-y}N). \label{eq:dirkrox} \end{equation} Changing the origin, so as to keep the unit torus at the center of the larger torus, (\ref{eq:q1q2}) must be replaced by \begin{equation} |{\bf q}_n,N>=\lim_{\nu \rightarrow \infty }\frac 1\nu \sum_{r=-\frac \nu 2}^{\frac \nu 2}|\frac{n+\chi _q}N+r>e^{2\pi ir\chi _p}. \label{qpltor} \end{equation} A straightforward calculation using (\ref{eq:dirkrox}) shows that the orthonormality conditions (\ref{eq:QQsim}) are obtained for the states defined in (\ref{qpltor}) with the normalization (\ref{qbastg2}). So we consider that the states and the operators in the unit torus are obtained from the plane by projections. Recalling the definition (\ref{kprom}) of the average, (\ref{qpltor}) can be written as \begin{equation} |{\bf q}_n,N>=\left\langle |\frac{n+\chi _q}N+r>e^{2\pi ir\chi _p}\right\rangle _r \label{eq:qQ} \end{equation} and the projection operator $\widehat{{\bf 1}}_N^\chi \,$ is then given by (% \ref{uno}). In the case of $\chi =0\,$ we thus retrieve the definition of Hannay and Berry \cite{hanay} for ${\cal H}_N^0$ as the average over a periodic array of Dirac delta distributions. We will now derive the Weyl representation on the torus by projecting the properties that have been well established in the plane. \section{The Weyl representation on the plane} \setcounter{equation}{0} We here summarize the results obtained for the plane in reference \cite {ozrep} that will be projected onto the torus in the following sections. We define the operator corresponding to a general translation in phase space by the $\left( 2L\right) $-dimensional vector $\xi =(\xi _p,\xi _q)$ as \begin{equation} \hat{T}_\xi \equiv \exp \left( \frac i\hbar \xi \wedge \hat{x}\right) \equiv \exp \left[ \frac i\hbar (\xi _p.\hat{q}-\xi _q.\hat{p})\right] , \label{eq:tcor} \end{equation} where naturally $\hat{x}=(\hat{p},\hat{q})$ and the symplectic product $\xi \wedge \eta $ is defined as \begin{equation} \xi \wedge \eta =({\frak J}\xi ).\eta \end{equation} with \begin{equation} {\frak J}=\left[ \begin{array}{c|c} 0 & -1 \\ \hline 1 & 0 \end{array} \right] . \end{equation} $\hat{T}_\xi $ is also known as a {\it Heisenberg operator}. In the case where either $\xi _p$ or $\xi _q=0$, we obtain respectively the operators $% \hat{T}_q$ or $\hat{T}_p$ mentioned in the preceding section. Acting on the Banach space ${\cal {H}_{{\Bbb R}}}$ we have \begin{equation} \widehat{T}_\xi |q_a>=e^{\frac i\hbar \xi _p(q_a+\frac{\xi _q}2)}|q_a+\xi _q> \label{eq:tq} \end{equation} and \begin{equation} \widehat{T}_\xi |p_a>=e^{-\frac i\hbar \xi _q(p_a+\frac{\xi _p}2)}|p_a+\xi _p>. \end{equation} The classical group property is maintained within a phase factor: \begin{equation} \hat{T}_{\xi _2}\hat{T}_{\xi _1}=\hat{T}_{\xi _1+\xi _2}\ \exp [\frac{-i}{% 2\hbar }\xi _1\wedge \xi _2]=\hat{T}_{\xi _1+\xi _2}\ \exp [\frac{-i}\hbar D_3(\xi _1,\xi _2)], \label{eq:tt} \end{equation} where $D_3$ is the symplectic area of the triangle determined by two of its sides. Evidently, the inverse of the unitary operator $\hat{T}_\xi ^{-1}=% \hat{T}_\xi ^{\dag }=\hat{T}_{-\xi }$ and we can generalize (\ref{eq:tt}): \begin{equation} \widehat{T}_{\xi _1}...\widehat{T}_{\xi j}=\widehat{T}_{\xi _1+....+\xi _j}e^{\frac i\hbar D_{j+1}(\xi _1,....\xi _j)}, \label{eq:ttt} \end{equation} where $D_{j+1}(\xi _1,....\xi _j)$ denotes the symplectic area of the $% \left( j+1\right) $ sided polygon formed by the chords, $(\xi _1,....\xi _j)$% . The operator corresponding to a phase space reflection about a point $% x=(p,q) $ is \cite{ozrep} \begin{equation} \widehat{R}_x\equiv (4\pi \hbar )^{-L}\int d\xi \quad e^{\frac i\hbar x\wedge \xi }\widehat{T}_\xi . \label{eq:rint} \end{equation} This operator has the following properties \cite{ozrep} : \begin{equation} \widehat{T}_\xi =(\pi \hbar )^{-L}\int dx\quad e^{-\frac i\hbar x\wedge \xi }% \widehat{R}_x\ , \label{eq:tintr} \end{equation} \begin{equation} \widehat{R}_x\widehat{T}_\xi =\widehat{R}_{x-\xi /2}e^{-\frac i\hbar x\wedge \xi }\ , \label{eq:rt} \end{equation} \begin{equation} \widehat{T}_\xi \widehat{R}_x=\widehat{R}_{x+\xi /2}e^{-\frac i\hbar x\wedge \xi }\ \label{eq:tr} \end{equation} and \begin{equation} \widehat{R}_{x_1}\widehat{R}_{x_2}=\widehat{T}_{2(x_2-x_1)}e^{\frac i\hbar 2x_1\wedge x_2}, \label{eq:rr} \end{equation} so that \begin{equation} \widehat{R}_x\widehat{R}_x=\widehat{1}\ . \end{equation} The trace of the translation is \begin{eqnarray} Tr\hat{T}_\xi &=&\int <q|T_\xi |q>dq=(2\pi \hbar )^L\delta (\xi _p)\delta (\xi _q) \nonumber \\ &=&(2\pi \hbar )^L\delta (\xi )\ \end{eqnarray} and then taking the Fourier transform, \begin{equation} Tr\hat{I\!\!\!\!R}_x=Tr2^L\hat{R}_x=(2\pi \hbar )^{-L}\int d\xi \exp \left[ \frac i\hbar \ x\wedge \xi \right] Tr\hat{T}_\xi =1\ , \end{equation} where it is now also convenient to define the exact Fourier transform $\hat{% I\!\!\!\!R}_x$ of $\hat{T}_\xi $. We recall that the classical transformation $R_x$ has a single fixed point ($x$ itself), whereas $T_\xi $ has fixed points only if $\xi =0$, when all points are fixed. These results are in general agreement with our intuition as to the classical correspondence of the traces of unitary operators. i.e. that the trace is related to the classical fixed points. The above properties allow any operator $\hat{A}$ to be expressed as a linear superposition of elementary translation operators: \begin{equation} \hat{A}=\int \frac{d\xi }{(2\pi \hbar )^L}\ A(\xi )\hat{T}_\xi \ . \label{eq:cora} \end{equation} The confirmation results from \begin{equation} Tr(\hat{T}_{-\xi }\hat{A})=A(\xi )\ , \label{eq:acor} \end{equation} The analogy with the classical chord generating function for canonical transformations is discussed in \cite{ozrep}. We can equally represent any operator $\hat{A}$ as a superposition of reflections: \begin{equation} \hat{A}=\int \frac{dx}{(2\pi \hbar )^L}\ A(x)\hat{I\!\!\!\!R}_x=\int \frac{dx% }{(\pi \hbar )^L}\ A(x)\hat{R}_x\ . \label{eq:cena} \end{equation} Again we obtain the expansion coefficient by calculating \begin{equation} Tr(\hat{I\!\!\!\!R}_x\hat{A})=A(x)\ . \label{eq:acen} \end{equation} Notice that comparison of (\ref{eq:cora}) and (\ref{eq:cena}) with (\ref {eq:rint}) and (\ref{eq:tintr}) yields \begin{equation} R_x(\xi )=2^{-L}\exp \left[ \frac i\hbar \ x\wedge \xi \right] \ \ \ % \mbox{and}\ \ \ T_\xi (x)=\exp \left[ -\frac i\hbar \ x\wedge \xi \right] \ . \end{equation} In analogy with our previous result, we may refer to $A(x)$ as the {\it % center representation} of the operator $\hat{A}$, but the historic term is the {\it Weyl representation}. The product laws of center and chord representations are of fundamental interest. Starting with the chord representation, we have, for the product $% \hat{A}_n\hat{A}_{n-1}\cdots \hat{A}_1$, \begin{eqnarray} &&A_n.A_{n-1}\cdots A_1(\xi )=\left( \frac 1{2\pi \hbar }\right) ^{L(n-1)}\int d\xi _n\cdots d\xi _1A_n(\xi _n)\cdots \nonumber \\ &&A_1(\xi _1)\delta (\xi _1+\cdots \xi _n-\xi )\exp \left[ -\frac i\hbar D_{n+1}(\xi _1,\cdots ,\xi _n)\right] , \label{eq:ancor} \end{eqnarray} where we note that the Dirac $\delta $-function has reduced the $(n+2)$% -sided polygon with symplectic area $D_{n+2}$ to an $(n+1)$-sided polygon, with $n$ free sides. Evidently, we can now use the $\delta $-function to remove one of the variables in the integral, but (\ref{eq:ancor}) is in its most symmetric form. We shall also need integral formulae for the product of operators in the center representation. The result depends crucially on the parity of the number of operators\cite{ozrep}, so we will start with the simplest case where $n=2$. Proceeding from the definition ~(\ref{eq:cena}), we obtain \begin{equation} A_2.A_1(x)=\left( \frac 1{\pi \hbar }\right) ^{2L}\int dx_2dx_1A_2(x_2)A_1(x_1)\exp \left[ \frac i\hbar \Delta _3(x,x_1,x_2)\right] . \label{eq:aacen} \end{equation} where $\Delta _3(x,x_1,x_2)=2(x_1\wedge x_2+x_2\wedge x+x\wedge x_1)$ is the symplectic area of the triangle whose midpoints are, $x,x_1,x_2$. The extension to $(2n)$ operators is \cite{ozrep} \begin{eqnarray} &&A_{2n}\cdots A_1(x)= \nonumber \\ &&\left( \frac 1{\pi \hbar }\right) ^{2nL}\int dx_{2n}\cdots dx_1A_{2n}(x_{2n})\cdots A_1(x_1)\exp \left\{ \frac i\hbar \Delta _{2n+1}(x,x_1,\cdots ,x_{2n})\right\} . \label{eq:ancen} \end{eqnarray} Here the symplectic area $\Delta _{2n+1}$ corresponds to the $(2n+1)$-sided polygon circumscribed around the centers $(x,x_1,\cdots ,x_{2n})$. The main advantage of the chord and center representation is their symplectic invariance. It is well known that linear classical canonical transformations $x^{\prime }=Mx$ correspond to unitary transformations in $% {\cal {H}_{{\Bbb R}}}$% \begin{equation} \widehat{A}\rightarrow \widehat{U}_M\widehat{A}\widehat{U}_M^{-1}. \end{equation} The effect of such a unitary transformation on the chord and center representation is merely \begin{equation} A(x)\rightarrow A(Mx)\quad \mbox{and}\quad A(\xi )\rightarrow A(M\xi ). \end{equation} \section{Weyl representation in the torus} \setcounter{equation}{0} \subsection{ Translation and Reflection Operators on the torus} In this section we will project the translations $\widehat{T}_\xi $ and reflections $\widehat{R}_x$ operators defined on ${\cal H}_{{\Bbb R}}$ onto the Hilbert space ${\cal H}_N^\chi $. Again we will treat the case of one degree of freedom explicitly, since the generalization is obvious. For this purpose we first investigate the action of the translation operators $% \widehat{T}_\xi $ on the $|{\bf q}_n,N>$ basis vectors defined by (\ref {eq:qQ}). From the effect of a translation (\ref{eq:tq}) on a single position in the plane, we have \begin{equation} \widehat{T}_\xi |{\bf q}_n,N>=\left\langle |\frac{n+\chi _q}N+k+\xi _q>e^{2\pi ik\chi _p}e^{\frac i\hbar \xi _p(\frac{n+\chi _q}N+k+\frac 12\xi _q)}\right\rangle _k, \end{equation} using the relation (\ref{eq:nh}) between $N$ and $\hbar $. This vector will belong to ${\cal H}_N^\chi $ only if it has the form (\ref{eq:qQ}), i.e., if we can write $\xi _q=\frac sN$ with $s$ an integer. A similar treatment in the $|{\bf p}_m,N>$ representation implies that \begin{equation} \widehat{T}_\xi |{\bf p}_m,N>=\left\langle |\frac{m+\chi _p}N+k+\xi _p>e^{-2\pi ik\chi _q}e^{-\frac i\hbar \xi _q(\frac{m+\chi _q}N+k+\frac{\xi _p}2)}\right\rangle _k, \end{equation} which does not belong to ${\cal H}_N^\chi $ unless we can write $\xi _p=\frac rN$ with $r$ an integer. So, as was already pointed in \cite {debievre}, the only translations that leave ${\cal H}_N^\chi $ invariant are those whose chords are $\xi =(\frac rN,\frac sN)$. For these cases we have \begin{eqnarray} \widehat{T}_\xi |{\bf q}_n,N &>&=\left\langle |\frac{n+\chi _q}N+k+\frac sN>e^{2\pi ik\chi _p}e^{i2\pi N\frac rN(\frac{n+\chi _q}N+k+\frac s{2N})}\right\rangle _k \nonumber \\ &=&e^{\frac{i2\pi }Nr(n+\frac s2+\chi _q)}\left\langle |\frac{n+s+\chi _q}% N+k>e^{2\pi ik(\chi _p+r)}\right\rangle _k \nonumber \\ &=&e^{\frac{i2\pi }Nr(n+\frac s2+\chi _q)}|{\bf q}_{n+s},N>. \label{eq:tQ} \end{eqnarray} In short, we obtain the torus operator $\widehat{{\bf T}}_\xi ^\chi $ in terms of the plane operator $\widehat{T}_\xi $ as \begin{eqnarray} \widehat{{\bf 1}}_N^\chi \widehat{T}_\xi \widehat{{\bf 1}}_N^\chi &=&\left\{ \begin{tabular}{lll} $\widehat{{\bf T}}_\xi ^\chi $ & $\mbox{if }$ $\xi =\left( \frac rN,\frac sN\right) ,$where $r$ and $s$ are integers & \\ $0$ & otherwise & \end{tabular} \right. \\ &=&\widehat{{\bf T}}_\xi ^\chi \left\langle \delta \left( \xi -\frac 1N\left( r,s\right) \right) \right\rangle _{\left( r,s\right) }=\widehat{T}% _\xi \widehat{{\bf 1}}_N^\chi \ , \label{projt} \end{eqnarray} where the torus translation operators $\widehat{{\bf T}}_\xi ^\chi \equiv \widehat{{\bf T}}_{r,s}^\chi $ are defined through \begin{equation} \widehat{{\bf T}}_{r,s}^\chi |{\bf q}_n,N>=e^{i\frac{2\pi }Nr(n+\chi _q+s/2)}|{\bf q}_{n+s},N> \label{eq:TQ} \end{equation} and \begin{equation} \widehat{{\bf T}}_{r,s}^\chi |{\bf p}_m,N>=e^{-i\frac{2\pi }Ns(m+\chi _p+r/2)}|{\bf p}_{m+r},N>. \label{eq:TP} \end{equation} The last equality in (\ref{projt}) holds because $\widehat{T}_\xi $ and $% \widehat{{\bf 1}}_N^\chi $ commute for $\xi =\left( \frac rN,\frac sN\right) $. We then see that the only translation operators that do not vanish on projection to the torus are those that leave ${\cal H}_N^\chi $ invariant. They correspond precisely to those classical transformations that preserve the QPS web. To simplify the notation, we will usually assume implicitly the $\chi $ dependence. Let us now study some properties of the torus translation operators. For the case where the chords are the minimal translations in any one of the $q$ or $% p$ directions, we recover the Schwinger operators \cite{schwinger} so that, \begin{equation} \widehat{{\bf T}}_{0,1}|{\bf q}_n,N>\equiv \widehat{{\bf T}}_q|{\bf q}_n,N>=|% {\bf q}_{n+1},N> \end{equation} and, \begin{equation} \widehat{{\bf T}}_{1,0}|{\bf p}_m,N>\equiv \widehat{{\bf T}}_p|{\bf p}_m,N>=|% {\bf p}_{m+1},N>. \end{equation} The kernel (\ref{eq:PQ}) implies that, \begin{equation} \widehat{{\bf T}}_p\widehat{{\bf T}}_q=e^{-\frac{2\pi i}N}\widehat{{\bf T}}_q% \widehat{{\bf T}}_p. \label{eq:grupo} \end{equation} so any translation operator in ${\cal H}_N^\chi $ is defined as \begin{equation} \widehat{{\bf T}}_\xi \equiv \widehat{{\bf T}}_{r,s}=e^{-\frac{i\pi rs}N}% \widehat{{\bf T}}_p^r\widehat{{\bf T}}_q^s\ , \label{eq:Tcor} \end{equation} with chords $\xi =(\frac rN,\frac sN).$ We can express the matrix elements of the translation operators in the $|{\bf q}_n,N>$ basis, \begin{equation} <{\bf q}_m,N|\widehat{{\bf T}}_{r,s}|{\bf q}_n,N>=e^{-i\frac{2\pi }Nr(\frac{% m+n}2+\chi _q)}\delta _{m,n+s}^{(N)}e^{i\frac{2\pi }N(\frac r2+\chi _p)(m-n-s)}, \label{eq:TQQ} \end{equation} using the orthonormality relations of the states (\ref{eq:QQsim}).The fact that the Hilbert space has finite dimension implies that linear operators acting on it will be represented by $N\times N$ matrices. Then $N^2$ linearly independent matrices will form a basis for the operators in ${\cal H% }_N^\chi $. This is clear from the symmetries of the translation operators; through their action on ${\cal H}_N^\chi $ (\ref{eq:TQ}) we see that \begin{equation} \widehat{{\bf T}}_{\xi +{\bf k}}=(-1)^{sk_p-rk_q+k_pk_qN}e^{i2\pi (k_p\chi _q-k_q\chi _p)}\widehat{{\bf T}}_\xi =e^{-i2\pi N\left[ \left( \frac \xi 2+\frac \chi N\right) \wedge {\bf k}+\frac 14{\bf k}\widetilde{{\frak J}}% {\bf k}\right] }\widehat{{\bf T}}_\xi , \label{eq:TTsim} \end{equation} where ${\bf k}=(k_p,k_q)$ is a vector with integer components denoting chords that perform respectively $k_p$ and $k_q$ loops around the irreducible circuits of the torus. We have also defined the symmetric matrix \begin{equation} \widetilde{{\frak J}}=\left[ \begin{array}{c|c} 0 & 1 \\ \hline 1 & 0 \end{array} \right] . \end{equation} If we perform ${\bf k}$ loops around the torus, (\ref{eq:TTsim}) implies that we recover the identity operator only up to a phase: \begin{equation} \widehat{{\bf T}}_{{\bf k}}=(-1)^{k_pk_qN}e^{i2\pi (k_p\chi _q-k_q\chi _p)}% \widehat{{\bf 1}}_N^\chi . \label{tNQ} \end{equation} Thus, to have a basis of operators we only need $r$ and $s$ in the range $% [0,N-1]$, that is, we only need translations that perform less than one loop around the torus. The second phase factor in expression (\ref{eq:TTsim}) comes from the Bloch boundary conditions, but the $(-1)^{sk_p-rk_q+k_pk_qN}$ factor shows that we need two loops around the torus to recover the same operator, doubling the expected periodicity. This will have crucial importance in the construction of the reflection operators. An important property of the translation operators, which can be deduced from (\ref{eq:grupo}) is \begin{equation} \widehat{{\bf T}}_{\xi _2}\widehat{{\bf T}}_{\xi _1}=\widehat{{\bf T}}_{\xi _1+\xi _2}e^{i\pi N\xi _1\wedge \xi _2}, \label{eq:TT} \end{equation} which generalizes to \begin{equation} \widehat{{\bf T}}_{\xi _1}...\widehat{{\bf T}}_{\xi j}=\widehat{{\bf T}}% _{\xi _1+....+\xi _j}e^{i\pi ND_{j+1}(\xi _1,....\xi _j)}, \end{equation} where $D_{j+1}(\xi _1,....\xi _j)$ denotes the area of $j+1$ sided polygon formed by the chords, $(\xi _1,....\xi _j)$, exactly as (\ref{eq:ttt}) in the plane case. From (\ref{eq:TT}) and the unitarity of $\widehat{{\bf T}}% _\xi ,$ we can see that \begin{equation} \widehat{{\bf T}}_\xi ^{\dag }=\widehat{{\bf T}}_\xi ^{-1}=\widehat{{\bf T}}% _{-\xi }. \label{eq:T-1} \end{equation} Notice that (\ref{eq:TT}) reduces to (\ref{eq:grupo}) for the particular case that $\xi _1$ and $\xi _2$ are vectors along the coordinate axis. For the reflection operators, we use their definition (\ref{eq:rint}) in terms of plane translations and then (\ref{projt}) projects the translations onto the torus, so that \begin{eqnarray} \widehat{{\bf R}}_x^\chi &=&\widehat{{\bf 1}}_N^\chi \widehat{R}_x\widehat{% {\bf 1}}_N^\chi =(4\pi \hbar )^{-L}\int d\xi \quad e^{\frac i\hbar x\wedge \xi }\widehat{{\bf T}}_\xi \left\langle \delta \left( \xi -\frac{{\bf j}}% N\right) \right\rangle _{{\bf j}} \nonumber \\ &=&(4\pi \hbar )^{-L}\left\langle e^{i2\pi x\wedge {\bf j}}\widehat{{\bf T}}% _{\frac{{\bf j}}N}\right\rangle _{{\bf j}}. \end{eqnarray} To perform the average we use the periodicity of the torus translation (\ref {eq:TTsim}), but we perform two loops around the torus, so that \begin{eqnarray} \widehat{{\bf R}}_x^\chi &=&(4\pi \hbar )^{-L}\left( \frac 1{2N}\right) ^2\sum_{r=0}^{2N-1}\sum_{s=0}^{2N-1}\left\langle e^{-i2\pi N\left[ \left( \frac{{\bf j}}{2N}+\frac \chi N\right) \wedge 2{\bf k}+{\bf k}\widetilde{% {\frak J}}{\bf k}\right] }e^{i2\pi x\wedge ({\bf j+}2{\bf k}N{\bf )}}% \widehat{{\bf T}}_{\frac{{\bf j}}N}\right\rangle _{{\bf k}} \nonumber \\ &=&\frac 1{2N}\sum_{r=0}^{2N-1}\sum_{s=0}^{2N-1}e^{-i2\pi x\wedge {\bf j}}% \widehat{{\bf T}}_{\frac{{\bf j}}N}\left\langle e^{-i2\pi \left[ \left( \chi -xN\right) \wedge 2{\bf k}\right] }\right\rangle _{{\bf k}}. \end{eqnarray} The ${\bf k}$ average is different from zero only if the point $x$ is such that $x=x_{a,b}\equiv \frac 1N\left( \begin{array}{c} a+\chi _p \\ b+\chi _q \end{array} \right) ,\,$ with $a\,$ and $b$ half integers. In short, \begin{eqnarray} \widehat{{\bf 1}}_N^\chi \widehat{R}_x\widehat{{\bf 1}}_N^\chi &=&\left\{ \begin{tabular}{lll} $\widehat{{\bf R}}_{x_{a,b}}^\chi $ & $\mbox{if }$ $x=\left( \frac{a+\chi _p}% N,\frac{b+\chi _q}N\right) ,$ where $a$ and $b$ are half-integers & \\ $0$ & otherwise & \end{tabular} \right. \\ &=&\widehat{{\bf R}}_{x_{a,b}}^\chi \left\langle \delta \left( x-\frac{{\bf k% }}{2N}+\frac \chi N\right) \right\rangle _{{\bf k}}=\widehat{R}_x\widehat{% {\bf 1}}_N^\chi , \label{rtoro} \end{eqnarray} where \begin{equation} \widehat{{\bf R}}_{x_{a,b}}^\chi =\frac 1{2N}\sum_{r=0}^{2N-1}\sum_{s=0}^{2N-1}e^{i2\pi Nx\wedge \xi }\widehat{{\bf T% }}_\xi ^\chi \label{eq:RRT} \end{equation} is the torus reflection on the center point $x_{a,b}$. The last equality in (% \ref{rtoro}) holds because $\widehat{{\bf R}}_{x_{a,b}}^\chi $ commutes with $\widehat{{\bf 1}}_N^\chi .$ Again, we will usually omit the explicit $\chi $ dependence. The construction of the reflection operators on the torus replaces the Fourier transform (\ref{eq:rint}) by a Fourier sum on the torus translation operators. The sums are to be taken over operators on one complete period; the symmetry properties (\ref{eq:TTsim}) show that this period is obtained with chords that perform two loops around the torus, i.e. , the period is double that expected. So, although the basis of operators is formed with chords that perform up to one loop around the torus in the Fourier sum, we have to sum over chords that perform up to two loops. Thus, the basis operators are summed twice, but with different Fourier phases. In what follows, the subscripts $(a,b)$ for the discrete centers and $(r,s)$ for the lattice of chords are implicit and they will be explicitly written only to avoid possible confusion. With the use of (\ref{eq:114}) we can derive the following extensively used relations, \begin{equation} \sum_{a=0}^{N-1/2}\sum_{b=0}^{N-1/2}e^{i2\pi Nx\wedge \xi }=(2N)^2\delta _{r,0}^{(2N)}\delta _{s,0}^{(2N)}, \label{eq5} \end{equation} where all the sums over $a$ and $b$ are taken with step $\frac 12$, and \begin{equation} \sum_{r=0}^{2N-1}\sum_{s=0}^{2N-1}e^{i2\pi Nx\wedge \xi }=(2N)^2\delta _{b,0}^{(N)}\delta _{a,0}^{(N)}. \label{eq6} \end{equation} Here $\delta _{b,0}^{(2N)}$ is a period-$2N$ Kronecker delta function. Inserting (\ref{eq:TQ}) in (\ref{eq:RRT})we find the action of the reflection operators on the Hilbert space: \begin{equation} \widehat{{\bf R}}_x|{\bf q}_n,N>=e^{i\frac{2\pi }N2(b-n)(a+\chi _p)}|{\bf q}% _{2b-n},N> \label{eq:RQ} \end{equation} and \begin{equation} \widehat{{\bf R}}_x|{\bf p}_m,N>=e^{i\frac{2\pi }N2(a-m)(b+\chi _q)}|{\bf p}% _{2a-m},N>. \label{eq:RP} \end{equation} The unitarity of $\widehat{{\bf R}}_x$ is ensured by (\ref{eq:RQ}) and (\ref {eq:RP}). We then see that $\widehat{{\bf R}}_x$ reflects the (QPS) web about the point $x=(\frac{a+\chi _p}N,\frac{b+\chi _q}N)$. We need to include half-integer values of $a$ and $b$ so that with a given $|{\bf q}_n,N>$ we can span all ${\cal H}_N^\chi $ by applying different $\widehat{{\bf R}}_x$. This is in complete agreement with the fact that the reflections that leave invariant the web formed by the QPS must include half-integer values of $a$ and $b$, conferring on these half-integers a clear geometrical meaning. So the centers of the reflections form a web whose spacing is half that of the QPS, as shown in Fig.~\ref{fig.3.2}. Once more, the only operators that do not vanish on projection to the torus are those that leave ${\cal H}_N^\chi $ invariant. These correspond classically to those transformations that leave the QPS web invariant. \begin{figure}[htb] \centerline {\epsfxsize=3in \epsffile{fig3-2.eps} } \caption{ The quantum Phase space QPS for $N=4$ (solid line). The Weyl phase space WPS (doted line) is the grid of points $x$, centers of the reflection operators in QPS. The area in the bold square is the quarter-torus in which lie the centers $x,$ which label the basis of operators ${\bf R}_x$. } \label{fig.3.2} \end{figure} The matrix elements of the reflection operators in the $|{\bf q}_n,N>$ basis are \begin{equation} <{\bf q}_m,N|\widehat{{\bf R}}_x|{\bf q}_n,N>=e^{i\frac{2\pi }N(m-n)(a+\chi _q)}\delta _{m,2b-n}^{(N)}e^{i\frac{2\pi }Na(2b-n-m)}. \label{eq:RQQ} \end{equation} From (\ref{eq:RQ}) we can see the symmetry properties of these operators, \begin{equation} \widehat{{\bf R}}_{x+\frac{{\bf k}}2}=(-1)^{bk_p+ak_q+k_pk_qN}\widehat{{\bf R% }}_x \label{eq:symR} \end{equation} where ${\bf k}=(k_p,k_q)$ is a vector with integer components. It is important to see that the domain of the variables $a$ and $b$ being integer and half-integer values, we have $(2N)^2$ different reflection operators in the unit square. But the symmetry properties (\ref{eq:symR}) show that only $N^2$ of them are independent, so we take the values of $a$ and $b$ that belong to $[0,\frac{N-1}2]$ ; this forms a complete set of independent operators. That is, only one quarter of the torus is needed to define a complete set of reflection operators. The values of $x=(\frac{a+\chi _p}N,% \frac{b+\chi _q}N)$ generated by these values of $a$ and $b$ do not all belong to the QPS; indeed we define here another space, the {\it Weyl phase space}, WPS, formed by the support of $x$ this is shown by the bold face area in Fig.~\ref{fig.3.2}. In the case where $N$ is odd, we will see later that WPS can be defined such that it coincides with the QPS. By the use of (\ref{eq5}) we can see that, \begin{equation} \widehat{{\bf T}}_\xi =\frac 1{2N}\sum_{a=0}^{N-1/2}\sum_{b=0}^{N-1/2}e^{-i2\pi Nx\wedge \xi }\widehat{% {\bf R}}_x, \label{eq:tsumr} \end{equation} where we are again taking the sum with the indices running in an interval twice as large as that needed to define a basis of operators. This is so because (\ref{eq:symR}) implies that classically equivalent reflections, through points diametrically opposed on any of the circuits of the torus, are only equal up to a phase. We now investigate the group or cocycle properties of the translations and reflections defined in this section. It is important to note that the transformations treated here are such that they leave the web formed by the QPS invariant at the classical level, as well as the Hilbert space ${\cal H}% _N^\chi $. With the help of (\ref{eq:tsumr}) and (\ref{eq:TT}), we obtain the following properties for these operators, \begin{equation} \widehat{{\bf R}}_x\widehat{{\bf T}}_\xi =\widehat{{\bf R}}_{x-\xi /2}e^{-i2\pi Nx\wedge \xi }, \label{eq:RT} \end{equation} \begin{equation} \widehat{{\bf T}}_\xi \widehat{{\bf R}}_x=\widehat{{\bf R}}_{x+\xi /2}e^{-i2\pi Nx\wedge \xi }, \label{eq:TR} \end{equation} \begin{equation} \widehat{{\bf R}}_{x_1}\widehat{{\bf R}}_{x_2}=\widehat{{\bf T}}% _{2(x_2-x_1)}e^{i2\pi N(2x_1\wedge x_2)}. \label{eq:RR} \end{equation} We then have the same cocycle properties as in the plane: (\ref{eq:tt})-(\ref {eq:rr}). This is a consequence of the commutation of operator products with projection (\ref{eq:abpt}) and will be of crucial importance when we derive the properties of the center and chord representations on the torus. Note that the characterization of the chords $\xi $ by integers and the centers $% x $ by half-integers is respected by the group of operations above. Another property which results from the last cocycle relation (\ref{eq:RR}) is \begin{equation} \widehat{{\bf R}}_x\widehat{{\bf R}}_x=\widehat{{\bf 1}}_N^\chi , \end{equation} in accordance with classical reflections. This means that \begin{equation} \widehat{{\bf R}}_x^{-1}=\widehat{{\bf R}}_x^{\dag }=\widehat{{\bf R}}_x, \end{equation} that is, reflection operators on the torus are unitary and Hermitian. It is important at this stage to examine the trace of these operators. Using (\ref{eq:TQQ}) and (\ref{eq:114}), we have: \begin{equation} {\bf Tr}(\widehat{{\bf T}}_\xi )=Ne^{i\frac{2\pi }N(\frac{rs}2+r\chi _p-s\chi _p)}\delta _{r,0}^{(N)}\delta _{s,0}^{(N)}\equiv N\;e^{i\frac{2\pi }% N(\frac{rs}2+r\chi _q-s\chi _p)}\;\delta _\xi ^{(N)}. \label{eq:trT} \end{equation} For the trace of the reflection operators, we recall (\ref{eq:RQQ}), so \begin{equation} <{\bf q}_n,N|\widehat{{\bf R}}_x|{\bf q}_n,N>=\delta _{n,2b-n}^{(N)}e^{i% \frac{2\pi }Na(2b-2n)}\ , \label{eq:Rnn} \end{equation} which is different from zero only if \begin{equation} n=b\qquad \mbox{mod}\left( \frac N2\right) . \label{eq:nb} \end{equation} However, if $b$ is half-integer and $N$ is even, for example, there would be no $n$ such that (\ref{eq:nb}) is satisfied. In general we can have up to 2 solutions of (\ref{eq:nb}) for $n\in \left[ 0,N-1\right] $, but they can have different phase contributions in (\ref{eq:Rnn}). A careful inspection leads to: \begin{eqnarray} {\bf Tr}(\widehat{{\bf R}}_x) &=&f_N(x)=\frac 12(1+(-1)^{2a}+(-1)^{2b}+(-1)^{2a+2b+N})= \nonumber \label{eq:trRT1} \\ &=&\left\{ \begin{tabular}{ll} $0$ & $\mbox{ if N is even and }a\mbox{ or }b\mbox{ semi-integers}$ \\ $2$ & $\mbox{ if N is even and }a\mbox{ and }b\mbox{ integers}$ \\ $1$ & $\mbox{if N is odd and }a\mbox{ or }b\mbox{ integers}$ \\ $-1$ & $\mbox{ if N is odd and }a\mbox{ and }b\mbox{ semi-integers}$% \end{tabular} \right. \label{eq:trRT} \end{eqnarray} The importance of this result for the following theory calls for some intuitive explanation in terms of the reflections of the discrete periodic lattice. As in the plane case, we can relate ${\bf Tr}(\widehat{{\bf R}}_x)$ to the number of fixed points of the corresponding classical map. Indeed, for $N$ odd there is always a single fixed point, agreeing with the modulus of (\ref{eq:trRT}). If $N$ is even, there will only be fixed points if $x$ is characterized by integer numbers ($a,b$), in which case there are two. \subsection{Operators and their Symbols} Once we have defined the reflection and translation operators, we can decompose any operator as their linear combination. To construct the chord, or translation representation of an operator, we express any operator as a linear combination of translations. To have a complete basis, we need just $% N^2$ operators, so that $r$ and $s$ run from $0$ to $N-1$. The chords $\xi =(\frac rN,\frac sN)$ having this property are said to belong to the fundamental domain. The other translation operators are obtained from these through the symmetry properties; that is, the fundamental translations are those which have chords smaller than one loop around any of the irreducible circuits of the torus in a given direction. The chord representation of an operator is defined as \begin{equation} {\bf A}(\xi )\equiv {\bf Tr}\left( \widehat{{\bf A}}\widehat{{\bf T}}_{-\xi }\right) . \label{eq:Acor} \end{equation} From the symbol, we recover the operator: \begin{equation} \widehat{{\bf A}}=\frac 1N\sum_{r,s=0}^{N-1}{\bf A}(\xi )\widehat{{\bf T}}% _\xi \equiv \frac 1N\sum_\xi {\bf A}(\xi )\widehat{{\bf T}}_\xi . \label{eq:corA} \end{equation} Although, to recover the operator we only need the symbol defined in the fundamental domain (i.e. $r,s$ in $[0,N-1]$), (\ref{eq:Acor}) can be used to extend the definition of the symbol for $r$ and $s$ running among all integer numbers. Of course, these will not be independent of the symbols in the fundamental domain and, from the symmetry properties of $\widehat{{\bf T}% }(\xi )$ (\ref{eq:TTsim}), we see that ${\bf A}(\xi )$ satisfies \begin{equation} {\bf A}(\xi +{\bf k})=(-1)^{sk_p+rk_q+k_pk_qN}e^{i2\pi (k_p\chi _q-k_q\chi _p)}{\bf A}(\xi )=e^{-i2\pi N\left[ \left( \frac \xi 2+\frac \chi N\right) \wedge {\bf k}+\frac 14{\bf k}\widetilde{{\frak J}}{\bf k}\right] }{\bf A}% (\xi ), \label{eq:Acorsim} \end{equation} where ${\bf k}=(k_p,k_q)$ is a vector with integer components denoting chords that perform respectively $k_p$ and $k_q$ loops around the irreducible circuits of the torus. This is an important consequence of the fact that the symmetry properties of the symbol of operators are the same as those of the basis operators used to generate this symbol. We now expand the operators in term of reflections; this is the center or Weyl representation. It is important to recall that we must take values of $% a $ and $b$ that belong to $[0,\frac{N-1}2]$, that is, only one quarter of the torus is needed to define a complete set of reflection operators . The values of $x=(\frac{a+\chi _p}N,\frac{b+\chi _q}N)$ generated by these values of $a$ and $b$ define the Weyl phase space ,WPS, shown by the bold face area in Fig.~\ref{fig.3.2}. We define the center symbol of an operator $\widehat{{\bf A}}${\bf \ }such that, \begin{equation} {\bf A}(x)\equiv {\bf Tr}\left( \widehat{{\bf A}}\widehat{{\bf R}}_x\right) . \label{eq:Acen} \end{equation} From the symbol, we recover the operator through \begin{equation} \widehat{{\bf A}}=\frac 1N\sum_{a,b=0}^{\frac{N-1}2}\widehat{{\bf R}}_x{\bf A% }(x)\equiv \frac 1N\sum_x\widehat{{\bf R}}_x{\bf A}(x). \label{eq:cenA} \end{equation} The symmetry properties of $\widehat{{\bf R}}_x$ (\ref{eq:symR}) imply \begin{equation} {\bf A}(x+\frac{{\bf k}}2)=(-1)^{bk_p+ak_q+k_pk_qN}{\bf A}(x), \label{eq:Acensim} \end{equation} for any vector ${\bf k}=(k_p,k_q)$ with integer components. This result had already been obtained by Hannay and Berry \cite{hanay} for the Wigner function and we see here that it is general for any Weyl symbol on the torus. As in the plane case, we derive some important properties of the translations and Weyl symbols. Notice first that: \begin{equation} {\bf R}_x(\xi )={\bf Tr}(\widehat{{\bf R}}_x\widehat{{\bf T}}_{-\xi })={\bf % Tr}(\widehat{{\bf R}}_{x+\xi /2})e^{i2\pi Nx\wedge \xi }=f_N(x+\xi /2)e^{i2\pi Nx\wedge \xi } \end{equation} and \begin{equation} {\bf T}_\xi (x)={\bf Tr}(\widehat{{\bf T}}_\xi \widehat{{\bf R}}_x)={\bf Tr}(% \widehat{{\bf R}}_{x+\xi /2})e^{-i2\pi Nx\wedge \xi }=f_N(x+\xi /2)e^{-i2\pi Nx\wedge \xi }. \end{equation} The trace is now obtained as \begin{eqnarray} {\bf Tr}\left( \widehat{{\bf A}}\right) &=&{\bf A}(\xi =0) \label{eq:trAcor} \\ &=&\sum_x{\bf A}(x)f_N(x)=\frac 12\sum_{a,b=0}^{N-\frac 12}{\bf A}(x). \end{eqnarray} In the last equality we use the fact that the Weyl symbols for the entire torus are obtained from those of a quarter of it through the symmetry relations (\ref{eq:Acensim}) and the definition of $f_N(x)$ (\ref{eq:trRT1}). The representation of the identity on the torus Hilbert space ${\cal H}% _N^\chi $ has now the form: \begin{equation} {\bf 1}_N^\chi (x)=f_N(x)\ \ \ \mbox{ and }\ \ \ {\bf 1}_N^\chi (\xi )=N\;\delta _\xi ^{(N)}. \end{equation} Hermitian operators are associated to the observables of the system and, in particular, the Hamiltonian generates the dynamics. Defined as $\widehat{% {\bf H}}=\widehat{{\bf H}}^{\dag }$, we obtain \begin{equation} {\bf H}^{\dag }(\xi )=[{\bf H}(-\xi )]^{*}\ \ \ \mbox{and}\ \ \ {\bf H}% ^{\dag }(x)=[{\bf H}(x)]^{*}\ , \end{equation} just as for the plane case \cite{ozrep}. The role played in the plane case by the Fourier transform will be taken by the finite Fourier transform, since it allows us to exchange chords and centers as well as to change from center or chord to the position representation. But there are some small differences due to the $f_N(x)$ factors peculiar to the torus. Thus, in the exchange of centers and chords we have, \begin{eqnarray} {\bf A}(\xi ) &=&{\bf Tr}\left( \frac 1N\sum_x\widehat{{\bf R}}_x{\bf A}(x)% \widehat{{\bf T}}_{-\xi }\right) =\frac 1N\sum_x{\bf A}(x){\bf Tr}(\widehat{% {\bf R}}_{x+\xi /2})e^{i2\pi Nx\wedge \xi } \nonumber \\ &=&\frac 1N\sum_x{\bf A}(x)f_N(x+\xi /2)e^{i2\pi Nx\wedge \xi }, \end{eqnarray} whereas \begin{equation} {\bf A}(x)=\frac 1N\sum_\xi {\bf A}(\xi )f_N(x+\xi /2)e^{-i2\pi Nx\wedge \xi }. \end{equation} Using (\ref{eq:TQQ}) and (\ref{eq:Acorsim}) we obtain the position representation of an operator $\widehat{{\bf A}}$% \begin{eqnarray} <{\bf q}_m,N|\widehat{{\bf A}}|{\bf q}_n,N> &=&\frac 1N\sum_\xi {\bf A}(\xi )<{\bf q}_m,N|\widehat{{\bf T}}_\xi |{\bf q}_n,N> \nonumber \\ &=&\frac 1N\sum_{r,s=0}^{N-1}{\bf A}(\xi _{r,s})e^{-i\frac{2\pi }Nr(\frac{m+n% }2+\chi _q)}\delta _{m,n+s}^{(N)}e^{i\frac{2\pi }N(\frac r2+\chi _p)(m-n-s)} \nonumber \\ &=&\frac 1N\sum_{r=0}^{N-1}{\bf A}(\xi _{r,m-n})e^{-i\frac{2\pi }Nr(\frac{m+n% }2+\chi _q)}. \label{eq:AQsig} \end{eqnarray} Note that in this last equation we are employing chords that may not belong to the fundamental domain; that is $m-n$ may not belong to $[0,N-1]$. However, the symbol for this chord is well defined through (\ref{eq:Acor}). If we restrict ourselves to chords that belong to the fundamental domain, we then have a supplementary $e^{i\frac{2\pi }N(\frac r2+\chi _p)(m-n)}$ phase factor in the last sum. This kind of difficulty may appear in the following formulae, but, by allowing the indices to run over all integer numbers, the formulae become indeed much simpler, as is the case for (\ref{eq:AQsig}). Using the position representation \begin{equation} \widehat{{\bf A}}=\sum_{m,n=0}^{N-1}|{\bf q}_m,N><{\bf q}_m,N|\widehat{{\bf A% }}|{\bf q}_n,N><{\bf q}_n,N|, \end{equation} we retrieve the chord representation as \begin{eqnarray} {\bf A}(\xi ) &=&{\bf Tr}\left( \sum_{m,n=0}^{N-1}<{\bf q}_m,N|\widehat{{\bf % A}}|{\bf q}_n,N>|{\bf q}_m,N><{\bf q}_n,N|\widehat{{\bf T}}_{-\xi }\right) \nonumber \\ &=&\sum_{m,n=0}^{N-1}<{\bf q}_m,N|\widehat{{\bf A}}|{\bf q}_n,N>e^{-i\frac{% 2\pi }Nr(\frac{m+n}2+\chi _q)}\delta _{m,n+s}^{(N)}e^{i\frac{2\pi }N(\frac r2+\chi _q)(m-n-s)} \nonumber \\ &=&\sum_{n=0}^{N-1}<{\bf q}_{n+s},N|\widehat{{\bf A}}|{\bf q}_n,N>e^{-i\frac{% 2\pi }Nr(n+\frac s2+\chi _q)}. \label{AcorQ} \end{eqnarray} Using (\ref{eq:RQQ}) and (\ref{eq:Acensim}) we exchange the coordinate and the center representation: \begin{eqnarray} <{\bf q}_m,N|\widehat{{\bf A}}|{\bf q}_n,N> &=&\frac 1N\sum_x{\bf A}(x)<{\bf % q}_m,N|\widehat{{\bf R}}_x|{\bf q}_n,N> \nonumber \\ &=&\frac 1N\sum_{a,b=0}^{\frac{N-1}2}{\bf A}(x_{a,b})e^{i\frac{2\pi }% N2(b-n)(a+\chi _p)}\delta _{m,2b-n}^{(N)}e^{i\frac{2\pi }N2\chi _p(m-2b+n)} \nonumber \\ &=&\frac 1N\sum_{a=0}^{\frac{N-1}2}{\bf A}(x_{a,\frac{m+n}2})e^{i\frac{2\pi }% N(m-n)(a+\chi _p)}. \label{eq:AQx} \end{eqnarray} Note that in this last formula we are taking the center point $x$ that does not belong to the fundamental domain (i.e. $\frac{m+n}2$ may not belong to $% [0,\frac{N-1}2]$ ). We recover the center representation through \begin{eqnarray} {\bf A}(x) &=&{\bf Tr}\left( \sum_{m,n=0}^{N-1}<{\bf q}_m,N|\widehat{{\bf A}}% |{\bf q}_n,N>|{\bf q}_m,N><{\bf q}_n,N|\widehat{{\bf R}}_x\right) \nonumber \\ &=&\sum_{m,n=0}^{N-1}<{\bf q}_m,N|\widehat{{\bf A}}|{\bf q}_n,N>e^{i\frac{% 2\pi }N(m-n)(a+\chi _p)}\delta _{m,2b-n}^{(N)}e^{i\frac{2\pi }Na(2b-n-m)} \nonumber \\ &=&\sum_{n=0}^{N-1}<{\bf q}_{2b-n},N|\widehat{{\bf A}}|{\bf q}_n,N>e^{i\frac{% 2\pi }N2(b-n)(a+\chi _p)}. \label{AxQ} \end{eqnarray} It would be possible to define the chord and the Weyl representations by equations (\ref{AcorQ}) and (\ref{AxQ}) respectively. However, the geometrical structure, the role of the translations and reflection operators and the relation to the plane theory would then be relegated to curiosities. \subsection{Symbols of the product of operators} We now derive the product law of the symbols of the operators in these representations. Let us start with the chord representation (\ref{eq:corA}). For this purpose we use ,(\ref{eq:TT}) and (\ref{eq:trT}) to obtain \begin{eqnarray} {\bf AB}(\xi ) &=&{\bf Tr}\left( \widehat{{\bf A}}\widehat{{\bf B}}\widehat{% {\bf T}}_{-\xi }\right) =(\frac 1N)^2\sum_{\xi _1,\xi _2}{\bf A}(\xi _1){\bf % B}(\xi _2){\bf Tr}\left( \widehat{{\bf T}}_{\xi _1}\widehat{{\bf T}}_{\xi _2}% \widehat{{\bf T}}_{-\xi }\right) \nonumber \\ &=&(\frac 1N)^2\sum_{\xi _1,\xi _2}{\bf A}(\xi _1){\bf B}(\xi _2)Ne^{i2\pi ND_4(\xi _1,\xi _2,-\xi )}{\bf Tr}\left( \widehat{{\bf T}}_{\xi _1+\xi _2-\xi }\right) \nonumber \\ &=&(\frac 1N)\sum_{\xi _1}{\bf A}(\xi _1){\bf B}(\xi -\xi _1)e^{i2\pi ND_3(\xi _1,-\xi )}, \label{eq:ABcor} \end{eqnarray} where we allow chords $\xi _2=\xi -\xi _1$ not to be in the fundamental domain. Let us now take the trace of the product; inserting (\ref{eq:trAcor}% ) in (\ref{eq:ABcor}) leads to \begin{equation} {\bf Tr}\left( {\bf AB}\right) =(\frac 1N)\sum_{\xi _1}{\bf A}(\xi _1){\bf B}% (-\xi _1). \end{equation} The generalization of (\ref{eq:ABcor}) for the product of an arbitrary number of operators is \begin{eqnarray} {\bf A}_n...{\bf A}_1(\xi ) &=&(\frac 1N)^{n-1}\sum_{\xi _1...\xi _{n-1}}% {\bf A}_1(\xi _1)... \nonumber \\ &&\!{\bf A}_{n-1}(\xi _{n-1}){\bf A}_n(\xi -\sum_{i=1}^{n-1}\xi _i)\exp \left[ -i2\pi ND_{n+1}(\xi _1,...,\xi _{n-1},-\xi )\right] . \label{prodtr} \end{eqnarray} Thus, the product rule for the chords is obtained from that in the plane by simply substituting the integral in (\ref{eq:ancor}) by the corresponding sum. For the center symbol (\ref{eq:Acen}) the trace of the product is obtained using (\ref{eq:RR}) and (\ref{eq:trT}): \begin{eqnarray} {\bf Tr}\left( {\bf AB}\right) &=&(\frac 1N)^2\sum_{x_1,x_2}{\bf A}(x_2){\bf % B}(x_1){\bf Tr}\left( \widehat{{\bf R}}_{x_2}\widehat{{\bf R}}_{x_1}\right) \nonumber \\ &=&(\frac 1N)^2\sum_{x_1,x_2}{\bf A}(x_2){\bf B}(x_1)e^{i2\pi N(2x_1\wedge x_2)}{\bf Tr}\left( \widehat{{\bf T}}_{2(x_2-x_1)}\right) \nonumber \\ &=&(\frac 1N)\sum_{x_1}{\bf A}(x_1){\bf B}(x_1). \end{eqnarray} We will now derive the full product properties in the center representation (% \ref{eq:cenA}); with the help of the group properties (\ref{eq:RR}), (\ref {eq:RT}) and (\ref{eq:trRT}) we have \begin{eqnarray} {\bf AB}(x) &=&{\bf Tr}\left( \widehat{{\bf A}}\widehat{{\bf B}}\widehat{% {\bf R}}_x\right) =(\frac 1N)^2\sum_{x_1,x_2}{\bf A}(x_2){\bf B}(x_1){\bf Tr}% \left( \widehat{{\bf R}}_{x_2}\widehat{{\bf R}}_{x_1}\widehat{{\bf R}}% _x\right) \nonumber \\ &=&(\frac 1N)^2\sum_{x_1,x_2}{\bf A}(x_2){\bf B}(x_1)e^{i2\pi N2(x_1\wedge x_2+x_2\wedge x+x\wedge x_1)}{\bf Tr}\left( \widehat{{\bf R}}% _{x+x_2-x_1}\right) \nonumber \\ &=&(\frac 1N)^2\sum_{x_1,x_2}{\bf A}(x_2){\bf B}(x_1)e^{i2\pi N\Delta _3(x,x_1,x_2)}f_N(x+x_2-x_1), \label{eq:ABcen} \end{eqnarray} where the symplectic area of the triangle $\Delta _3(x,x_1,x_2)$ was defined in section 3. Note that the sides of these triangles must be integer vectors in this case, because the symmetry of each side about its center implies that all the corners will be of the same type regardless of whether either $% a $ or $b$ are integer or half-integer. The argument of the function $f_N$ defined in (\ref{eq:trRT}) can thus be any corner of the triangle as shown in Fig.~\ref{fig.3.3}(a). We thus find that the reflection properties of the QPS lead to a more complex product rule than for the plane (\ref{eq:aacen}). \begin{figure}[htb] \centerline {\epsfxsize=5in \epsffile{fig3-3.eps} } \caption{ Two examples of polygons displaying the uniform nature of the vertices of the polygon. (a) all vertices lie on half integers: $f_N=0$ or $-1$ for $N$ respectively even or odd (b) the vertices lie on integers: $% f_N=2 $ or $1$ for $N$ respectively even or odd. } \label{fig.3.3} \end{figure} The generalization for the product of $2n$ operators is \begin{eqnarray} &&{\bf A}_{2n}...{\bf A}_1(x)=(\frac 1N)^{2n}\sum_{x_1...x_{2n}}{\bf A}% _{2n}(x_{2n}) \nonumber \\ &&{\bf ...A}_1(x_1)e^{i2\pi N\Delta _{2n+1}(x,x_{2n},...,x_1)}f_N(x+\sum_{j=1}^{2n}(-1)^jx_j), \label{3.77} \end{eqnarray} where again the argument of $f_N$ is any corner of the polygon whose centers are $x,x_{2n},...,x_1$ (see an example in Fig.~\ref{fig.3.3}(b)). For an odd number of operators we just choose $\widehat{{\bf A}}_1=\widehat{{\bf 1}}$, that is $% {\bf A}_1(x_1)=f_N(x_1)$ in (\ref{3.77}). The product laws are the main result of this section. In contrast with the Weyl-like representation obtained by Galleti and Toledo Pisa \cite{galeti2}, we only need half the number of sums (including the implicit sums in the trace of their formula (21)). Kaperskovitz and Peev \cite{kaperpeev} also have a Weyl-like representation, but only for the case of $N$ even. They perform products of 2 operators and the product law that they obtain is very similar to ours, although their result is not compatible with our geometrical interpretation, because we need half-integer vectors to completely describe the reflections of QPS. Most important is the fact that our formalism prescribes the product of an arbitrary even number of operators, just as for the plane, whereas previous results could only cope explicitly with the product of two operators at a time. \subsection{Weyl representation in QPS} If $N$ is odd, we can redefine the WPS so that it coincides with the QPS. For this purpose we define $X=(\frac{\alpha +\chi _p}N,\frac{\beta +\chi _q}% N)$ so that , \begin{eqnarray} \alpha &=&a+j\frac N2\qquad \mbox{ where }j=\left\{ \mbox{\begin{tabular}{ll} $0$ & $\quad \mbox{ if }a\mbox{ is integer}$ \\ $1$ & $\quad \mbox{ otherwise}$\end{tabular} }\right. \\ \beta &=&b+k\frac N2\qquad \mbox{ where }k=\left\{ \mbox{\begin{tabular}{ll} $0$ & $\quad \mbox{ if }b\mbox{ is integer}$ \\ $1$ & $\quad \mbox{ otherwise}$\end{tabular} }\right. \end{eqnarray} We then have that $\alpha $ and $\beta $ are integers for the case were $N$ is odd. In other words for any $x$ there is a point $X$ such that \begin{equation} X=\frac 1N\left( \begin{array}{c} \alpha \\ \beta \end{array} \right) +\frac \chi N=x+\frac 12{\bf n} \label{xentero} \end{equation} with ${\bf n}$ an integer vectors. If $N$ is even, the $\alpha $ and $\beta $ will have the same character ( integer or half-integer) as $a$ and $b$, so we cannot recover the QPS. In the rest of this section we will then restrict ourselves to the case where $N$ is odd. The symmetry relation (\ref{eq:symR}% ) shows then that \begin{equation} \widehat{{\bf R}}_X=(-1)^{(2jb+2ja+jk)}\widehat{{\bf R}}_x, \end{equation} and with the use of (\ref{eq:trRT}) we have \begin{equation} {\bf Tr}(\widehat{{\bf R}}_X)=1. \end{equation} We now see that letting $a$ and $b$ run over the half-integers in $[0,\frac{% N-1}2]$, we then have $\alpha $ and $\beta $ integers in $[0,N-1]$ and we recover the QPS. For this space we will now have a new Weyl representation \begin{equation} {\bf A}(X)={\bf Tr}\left( \widehat{{\bf A}}\widehat{{\bf R}}_X\right) , \end{equation} from which we recover the operator: \begin{equation} \widehat{{\bf A}}=\frac 1N\sum_{\alpha ,\beta =0}^{N-1}\widehat{{\bf R}}_X% {\bf A}(X_{\alpha ,\beta })\equiv \frac 1N\sum_X\widehat{{\bf R}}_X{\bf A}% (X). \end{equation} We will now examine the properties of the product in this representation: \begin{eqnarray} {\bf AB}(X) &=&{\bf Tr}\left( \widehat{{\bf A}}\widehat{{\bf B}}\widehat{% {\bf R}}_{X_{\alpha ,\beta }}\right) =(\frac 1N)^2\sum_{X_1,X_2}{\bf A}(X_2)% {\bf B}(X_1){\bf Tr}\left( \widehat{{\bf R}}_{X_2}\widehat{{\bf R}}_{X_1}% \widehat{{\bf R}}_X\right) \nonumber \\ &&(\frac 1N)^2\sum_{X_1,X_2}{\bf A}(X_2){\bf B}(X_1)e^{i2\pi N2(X_1\wedge X_2-X_2\wedge X+X\wedge X_1)}{\bf Tr}\left( \widehat{{\bf R}}% _{X+X_2-X_1}\right) \nonumber \\ &&(\frac 1N)^2\sum_{X_1,X_2}{\bf A}(X_2){\bf B}(X_1)e^{i2\pi N\Delta _3(X,X_1,X_2)}. \label{eq:ABinter} \end{eqnarray} This last expression is very similar to the general case described in the previous section, but slightly simplified by the absence of the $f_N$ term, in close analogy to the plane formalism. For the product of $2n$ operators this generalizes to \begin{equation} {\bf A}_{2n}...{\bf A}_1(X)=(\frac 1N)^{2n}\sum_{X_1...X_{2n}}{\bf A}% _{2n}(X_{2n}){\bf ...A}_1(X_1)e^{i2\pi N\Delta _{2n+1}(X,X_{2n,...},X_1)}. \end{equation} The absence of the $f_N$ factor in these simplified formulae may be understood from the fact that a polygon whose centers all lie on an integer lattice always has corners on the same lattice. Hence, $f_N$ is always unity for all corners if $N$ is odd. In the same way, for $N$ odd, we can perform a transformation, similar to (% \ref{xentero}), to a set of chords $\tilde{\xi}$ that are even multiples of $% \frac 1N$. This set of chords will be complete if we now allow them to perform up to two loops around the circuits of the torus. This scheme can be generalized to perform quantization on centers or chords that are multiples of $\frac \phi N$ only if $2\phi $ and $N$ are coprime numbers. Then, the chord (or center) will be supported by a lattice of spacing $\frac \phi N$ and length $\phi $. This transformation will have importance for cat maps and will be studied in more details in reference \cite{loxocat}. \subsection{Relation between symbols} There are many ways to represent a given operator $\widehat{A}$ that acts on the Banach space of the plane ${\cal H}_{{\Bbb R}}$. Among the different representations, the center and chord symbols are of special interest in this work. Projecting the operator $\widehat{A}$ onto the torus Hilbert space ${\cal H}_N^\chi $ through $\widehat{{\bf A}}=\widehat{{\bf 1}}_N^\chi \widehat{A}\widehat{{\bf 1}}_N^\chi $, it can be represented in terms of torus translations or reflections. We shall now show how the symbols on the torus can be obtained from their counterparts on the plane. Starting with the chord representation, we calculate the torus symbol at points $\xi =\left( \frac rN,\frac sN\right) $. From the fact that $\widehat{% {\bf T}}_{-\xi }$ and $\widehat{{\bf 1}}_N^\chi $ commute, we have \begin{equation} {\bf A}(\xi )={\bf Tr}\left( \widehat{{\bf 1}}_N^\chi \widehat{A}\widehat{% {\bf 1}}_N^\chi \widehat{{\bf T}}_{-\xi }\right) ={\bf Tr}\left( \widehat{% {\bf 1}}_N^\chi \widehat{A}\widehat{T}_{-\xi }\widehat{{\bf 1}}_N^\chi \right) ={\bf Tr}\left( \widehat{A}\widehat{T}_{-\xi }\widehat{{\bf 1}}% _N^\chi \right) . \end{equation} Then, we express the operator $\widehat{A}$ in terms of translations (\ref {eq:acor}) and use the group properties of the translation operators (\ref {eq:tt}) to obtain \begin{equation} {\bf A}(\xi )=\int \frac{d\xi _1}{(2\pi \hbar )}\ A(\xi _1)e^{\frac i{2\hbar }\xi _1\wedge \xi }{\bf Tr}\left( \widehat{T}_{\xi _1-\xi }\widehat{{\bf 1}}% _N^\chi \right) . \end{equation} We now use the projection properties of the translation operators on the torus (\ref{projt}), so that \begin{equation} {\bf A}(\xi )=\int \frac{d\xi _1}{(2\pi \hbar )}\ A(\xi _1)e^{\frac i{2\hbar }\xi _1\wedge \xi }{\bf Tr}\left( \widehat{{\bf T}}_{\xi _1-\xi }\right) \left\langle \delta \left( \xi _1-\xi -\frac{{\bf k}}N\right) \right\rangle _{{\bf k}}. \end{equation} Performing the integral, with the help of the trace properties (\ref{eq:trT}% ), we obtain \begin{eqnarray} {\bf A}(\xi ) &=&\left\langle (-1)^{sk_p+rk_q+k_pk_qN}e^{i2\pi (k_p\chi _q-k_q\chi _p)}A\left( \frac rN+k_p,\frac sN+k_q\right) \right\rangle _{{\bf % k}}= \nonumber \\ &&\left\langle e^{i2\pi N\left[ \left( \frac \xi 2-\frac \chi N\right) \wedge {\bf k+}\frac 14{\bf k}\widetilde{{\frak J}}{\bf k}\right] }A\left( \xi +{\bf k}\right) \right\rangle _{{\bf k}}, \label{eq:Acorprom} \end{eqnarray} where the $(2L)$-dimensional vectors ${\bf k}$ have integer components. Note that we have to perform a phase weighted average on equivalent points to obtain the symbol on the torus. This is similar to the way Hannay and Berry quantize the cat map\cite{hanay} in the coordinate representation. We now proceed in a similar manner to derive the symbols in the center representation at the points $x=(\frac{a+\chi _p}N,\frac{b+\chi _q}N)$. Using the commutation of $\widehat{{\bf R}}_x$ with $\widehat{{\bf 1}}% _N^\chi $ and (\ref{eq:acor}), we have \begin{equation} {\bf A}(x)={\bf Tr}\left( \widehat{{\bf 1}}_N^\chi \widehat{A}\widehat{{\bf R% }}_x\right) ={\bf Tr}\left( \widehat{{\bf 1}}_N^\chi \int \frac{dy}{(\pi \hbar )}\ A(y)\hat{R}_y\widehat{R}_x\widehat{{\bf 1}}_N^\chi \right) , \end{equation} which combined with the cocycle properties (\ref{eq:rr}), becomes \begin{equation} {\bf A}(x)={\bf Tr}\left( \widehat{{\bf 1}}_N^\chi \int \frac{dy}{(\pi \hbar )}\ A(y)e^{\frac i\hbar 2y\wedge x}\widehat{T}_{2(x-y)}\widehat{{\bf 1}}% _N^\chi \right) . \end{equation} Projecting the translations on the torus (\ref{projt}), we have \begin{equation} {\bf A}(x)=\int \frac{dy}{(\pi \hbar )}\ A(y)e^{\frac i\hbar 2y\wedge x}{\bf % Tr}\left( \widehat{{\bf T}}_{2(x-y)}\right) \left\langle \delta \left( 2(x-y)-\frac{{\bf k}}N\right) \right\rangle _{{\bf k}}, \end{equation} so that performing the integral we obtain, \begin{eqnarray} {\bf A}(x) &=&\left\langle e^{i2\pi (ak_q-bk_p+\frac N2k_qk_p)}A\left( \frac{% a+\chi _p}N+\frac{k_p}2,\frac{b+\chi _q}N+\frac{k_q}2\right) \right\rangle _{% {\bf k}} = \nonumber \\ & &\left\langle e^{i2\pi N(\left[ \left( x-\frac \chi N\right) \wedge {\bf k+% }\frac 14{\bf k}\widetilde{{\frak J}}{\bf k}\right] )}A\left( x+\frac{{\bf k}% }2\right) \right\rangle _{{\bf k}}, \label{eq:Axplator} \end{eqnarray} with the help of the trace properties (\ref{eq:trT}). We again have a phase weighted average on equivalent points, this is a general feature of any representation of projected operators ; only the phase will depend on the specific representation we are taking. An important feature of the center representation is that the phases do not have any dependence on the $\chi $ parameters of the quantization; this is best seen in (\ref{eq:Axplator}). Note also that comparing (\ref{eq:Acorprom}) and (\ref {eq:Axplator}) with (\ref{eq:Acorsim}) and (\ref{eq:Acensim}) respectively, the phases are a consequence of the periodicity conditions of the symbols. It is important to note that if $\widehat{A}$ and $\widehat{{\bf 1}}% _N^\chi $ commute the restriction of $\widehat{A}$ on the Hilbert space $% {\cal H}_N^\chi $ denotes an automorphism. Indeed, the commutation of $% \widehat{A}$ and $\widehat{{\bf 1}}_N^\chi $ implies that the symbols $A(x)$ and $A(\xi )$ are periodic functions. Otherwise the average defined in (\ref {eq:Acorprom}) and (\ref{eq:Axplator}) may not exist, it may happen that the projected operator $\widehat{{\bf A}}=\widehat{{\bf 1}}_N^\chi \widehat{A}% \widehat{{\bf 1}}_N^\chi =0.$ \subsection{Symplectic invariance} At the end of section 3, we remarked that the center and the chord representations in the plane are invariant with respect to the quantum equivalents of linear canonical transformations, or symplectic transformations $x^{\prime }=Mx.$ The transformations for which the symplectic matrix $M$ is made up of integers are known colloquially as {\it % cat maps}. These have the property that they leave invariant the unit torus. Because of the commutation of operator products with projection from the plane to the torus, the effect of a similarity transformation $\widehat{{\bf % A}}\rightarrow \widehat{{\bf U}}_M\widehat{{\bf A}}\widehat{{\bf U}}_M^{-1}$ performed by a quantized cat map on any operator defined on the torus will be purely classical in the center or the chord representations: \begin{equation} {\bf A}(x)\rightarrow {\bf A}(Mx)\quad \mbox{and\quad }{\bf A}(\xi )\rightarrow {\bf A}(M\xi ) . \label{cat1} \end{equation} Evidently, the matrix $M=1$ is a cat map; the product of cat maps is also a cat map, as is the inverse of a cat map. It follows that the set of all cat maps forms a subgroup of the symplectic transformations, which we will refer to as the {\it feline group}. Likewise, relations (\ref{cat1}) indicate the feline invariance of the chord and center representations. In a companion paper \cite{loxocat} we use the chord and center representations to study the properties of quantum cat maps of more than one degree of freedom. This extends previous work by Hannay and Berry \cite {hanay} and Keating \cite{keat1}, \cite{keat2} on two-dimensional cat maps. For completeness, we will just note that the symbol corresponding to $% \widehat{{\bf U}}_M$ is \begin{equation} {\bf U}_M(X)=\frac 1{\sqrt{N^L}}\exp \left( \frac i\hbar XBX\right) , \label{catx} \end{equation} whereas the chord symbol is \begin{equation} {\bf U}_M(\tilde{\xi})=\frac 1{\sqrt{N^L}}\exp \left( -\frac i\hbar \tilde{% \xi}\beta \tilde{\xi}\right) . \label{catcor} \end{equation} For quantization performed on $\chi =0$, with $N$ an odd integer, $B$ and $% \beta $ are integer symmetric matrices and the chords $\tilde{\xi}=\frac 2N(r,s)$ (we perform quantization on a set of chords that are multiple of $% \frac 2N$). The symmetric matrices $B$ and $\beta $ in the above quadratic forms define the Cayley parametrization of the symplectic matrix $M:$% \begin{equation} M=\frac{1-{\frak J}B}{1+{\frak J}B}=\frac{1+{\frak J}\beta }{1-{\frak J}% \beta }. \end{equation} In general, The Cayley matrices for a cat map may not be integer, leading to less transparent relation between the symplectic classical matrix $M$ and the symmetric matrices $B$ and $\beta $ in (\ref{catx}) and (\ref{catcor}). They depend on Gaussian sums, as presented in reference \cite{loxocat}. \section{Hamiltonians on the Torus and Path Integrals} \setcounter{equation}{0} We will now treat Hamiltonian systems on the torus. Let us first recall that the Poisson bracket relation that defines the symplectic product is the same for the torus as in the case of the plane. Therefore the classical generating function for canonical transformations are governed by the same composition laws as defined in \cite{ozrep} for the plane case. The only difference is that there would be different chords for a given center due to the periodic boundary conditions that identify centers with half the period as that of the whole torus. We will then study dynamical systems with $L$ degrees of freedom for which there is defined a Hamiltonian function that generates the dynamics through Hamilton's equations and that is periodic in all its $2L$ variables. For $% L=1 $ this kind of system has applications in solid state physics; it has been used to model electron eigenstates in a one-dimensional solid with an incommensurate modulation of the structure \cite{aubry} and in models of Bloch electrons in a magnetic field \cite{harp}. It has also been shown \cite {willk} that this model presents a critical behavior giving rise to hierarchical structures in the solutions throughout the spectrum of the kind known as a {\it Hofstadter butterfly} \cite{hofs} and to localization transitions from extended to localized states. The Fourier theorem ensures that a classical Hamiltonian that is periodic in the plane can be written as \begin{equation} H(p,q)=\sum_{r,s=-\infty }^{+\infty }H_{r,s}e^{i2\pi (rp-sq)}. \end{equation} To quantize this Hamiltonian, different ways may be taken involving different orderings. We choose the Weyl ordering, which is such that \begin{equation} \widehat{H}(\hat{p},\hat{q})=\sum_{r,s=-\infty }^{+\infty }H_{r,s}e^{i2\pi (r% \hat{p}-s\hat{q})}. \end{equation} With the definition of the translation operators on the plane (\ref{eq:tcor}% ), we immediately see that this is equivalent to \begin{equation} \widehat{H}(\hat{p},\hat{q})=\sum_{r,s=-\infty }^{+\infty }H_{r,s}\widehat{T}% _{\xi _{r,s}} \label{eq:hqt} \end{equation} where $\xi _{r,s}=(2\pi \hbar r,2\pi \hbar s)=(\frac rN,\frac sN)$. The Hamiltonian is then a linear combination of translation operators that leaves ${\cal H}_N^\chi $ invariant. If another ordering is chosen, there will be corrections to (\ref{eq:hqt}) of the order of $\frac 1N$. The quantal evolution of the system is determined by the propagator: \begin{equation} \widehat{U}_t=e^{\frac i\hbar t\widehat{H}}=\sum_{n=0}^\infty \frac 1{n!}(\frac i\hbar t\widehat{H})^n. \end{equation} This last relation implies that the propagator is a combination of products of torus translations in the expansion (\ref{eq:hqt}). These form a cocycle, as we already saw, so we can write \begin{equation} \widehat{U}_t=\sum_{r,s=-\infty }^{+\infty }U_{r,s}\widehat{T}_{\xi _{r,s}}. \end{equation} Thus, the evolution operator also leaves ${\cal H}_N^\chi $ invariant. Written in this way we can see that the evolution operator and the Hamiltonian in the periodic plane have their chord representation in terms of torus translations only, in the form \begin{equation} U_t(\xi )=\sum_{r,s=-\infty }^{+\infty }U_{r,s}\delta (\xi_{r,s} -\xi). \end{equation} Let us now project this operator on ${\cal H}_N^\chi $ and follow the evolution. We may first note that (\ref{eq:abpt}) and (\ref{eq:hqt}) allows us to write: \begin{eqnarray} \widehat{{\bf U}}_t &=&\widehat{U}_t\widehat{{\bf 1}}_N^\chi =\sum_{r,s=-\infty }^{+\infty }U_{r,s}\widehat{{\bf T}}_\xi =\sum_{n=0}^\infty \frac 1{n!}(\frac i\hbar t\widehat{H})^n\widehat{{\bf 1}}% _N^\chi \nonumber \\ &=&\sum_{n=0}^\infty \frac 1{n!}(\frac i\hbar t\widehat{H}\widehat{{\bf 1}}% _N^\chi )^n=e^{\frac i\hbar t{\bf \hat{H}}} , \label{eq:UHT} \end{eqnarray} where ${\bf \hat{H}=}\widehat{{\bf 1}}_N^\chi \hat{H}\widehat{{\bf 1}}% _N^\chi =\hat{H}\widehat{{\bf 1}}_N^\chi $ is the Hamiltonian acting in the torus Hilbert space ${\cal H}_N^\chi .$ The unitarity evolution operators form a group, such that \begin{equation} \widehat{U}_t=(\widehat{U}_{\frac tM})^M. \end{equation} Projecting onto the torus and using (\ref{eq:abpt}) we obtain \begin{eqnarray} \widehat{{\bf U}}_t &=&\widehat{{\bf 1}}_N^\chi \widehat{U}_t\widehat{{\bf 1}% }_N^\chi =\widehat{{\bf 1}}_N^\chi (\widehat{U}_{\frac tM})^M\widehat{{\bf 1}% }_N^\chi \label{eq:prodevol1} \\ &=&(\widehat{{\bf U}}_{\frac tM})^M. \label{eq:prodevol2} \end{eqnarray} This last result is very important; the evolution and the projection commute. We have two alternatives to obtain the center representation for (\ref {eq:prodevol2}) , i.e. to work from the plane relations or to work directly with the torus. First we note that (\ref{eq:UHT}) implies \begin{equation} \lim_{t\rightarrow 0}{\bf U}_t(x)=e^{\frac i\hbar t{\bf H}(x)}+0(t^2). \end{equation} The use of (\ref{eq:prodevol2}) and (\ref{3.77}) results in \begin{eqnarray} {\bf U}_t(x) &=&\lim_{M\rightarrow \infty }\left( \frac 1N\right) ^{2LM}\sum_{x_i=0}^{\frac{N-1}2}f_N(x+\sum_{j=1}^{2M}(-1)^jx_j) \nonumber \\ &&\exp \left\{ \frac i\hbar \left[ \,\Delta _{2M+1}(x,x_1,...,x_{2M})-\frac t{2M}\sum_{i=1}^{2M}{\bf H(}x_i)\right] \right\} . \label{eq:utxT} \end{eqnarray} For the odd $N$ case we obtain the representation on the points $X$ of QPS, \begin{equation} {\bf U}_t(X)=\lim_{M\rightarrow \infty }\left( \frac 1N\right) ^{2LM}\sum_{X_i=0}^{N-1}\exp \left\{ \frac i\hbar \left[ \,\Delta _{2M+1}(X,X_1,...,X_{2M})-\frac t{2M}\sum_{i=1}^{2M}{\bf H(}X_i)\right] \right\} . \label{eq:utxTni} \end{equation} Notice that this expression for the projector relies on our original product rule for an arbitrary number of operators. To take the projection, using (\ref{eq:prodevol1}), we can use the already known result about the propagator in the center representation on the plane \cite{ozrep}, obtained as a {\it path integral} \begin{eqnarray} U_t(x) &=&\lim_{M\rightarrow \infty }\int \frac{dx_1\cdots dx_{2M}}{(\pi \hbar )^{2ML}}\ \exp \left\{ \frac i\hbar [\Delta _{2M+1}(x,x_1,\cdots ,x_{2M}))-\frac tM\sum_{n=1}^{2M}H(x_n)]\right\} \label{eq:utx2} \\ &=&\int_\gamma {\cal D}_\gamma \,e^{i/\hbar S_\gamma (x)}. \end{eqnarray} Here we can see that for the odd $N$ case the propagator (\ref{eq:utxTni}) is similar to (\ref{eq:utx2}) replacing the integral by the appropriate sums. The phase of the integral in (\ref{eq:utx2}) coincides with the center action $S_\gamma (x)$ for the polygonal path $\gamma $ with endpoints centered on $x$ and whose $k^{\prime }$th side is centered on $x_k$. The center variational principle ensures that this center action is stationary for the classical trajectories centered on $x$. In Fig.~\ref{fig.1.8.}. we show two possible paths, whose actions are compared by the center variational principle. \begin{figure}[h] \vspace{7cm} \caption{ Two possible paths, whose actions are compared by the center variational principle. } \label{fig.1.8.} \end{figure} Now we project the symbol on the torus through (\ref{eq:Axplator}). We then obtain \begin{eqnarray} {\bf U}_t(x) &=&\left( \frac 12\right) ^L\left\langle e^{i2\pi N(\left[ \left( x-\frac \chi N\right) \wedge {\bf k+}\frac 14{\bf k}\widetilde{{\frak % J}}{\bf k}\right] )}\lim_{M\rightarrow \infty }\int \frac{dx_1\cdots dx_{2M}% }{(\pi \hbar )^{2ML}}\right. \nonumber \\ &&\left. \exp \left\{ \frac i\hbar [\Delta _{2M+1}(x+\frac k2,x_1,\cdots ,x_{2M}))-\frac t{2M}\sum_{n=1}^{2M}H(x_n)]\right\} \right\rangle _{{\bf k}} \nonumber \\ &=&\left( \frac 12\right) ^L\left\langle e^{i2\pi N(\left[ \left( x-\frac \chi N\right) \wedge {\bf k+}\frac 14{\bf k}\widetilde{{\frak J}}{\bf k}% \right] )}\int_\gamma d\gamma \,\;e^{\frac i\hbar S_\gamma (x+\frac{{\bf k}}% 2)}\right\rangle _{{\bf k}}. \label{eq:utxT2} \end{eqnarray} Although it is not immediately evident, (\ref{eq:utxT}) and (\ref{eq:utxT2}) are the same object; in (\ref{eq:utxT}), we first project on the torus and then perform the evolution, while in (\ref{eq:utxT2}) we first evolve on the plane and the projection on the torus is performed later. But, since the projection (\ref{eq:prodevol2}) and evolution are commuting operations, (\ref {eq:utxT}) and (\ref{eq:utxT2}) coincide. If we had defined the Weyl transformation intrinsically in the torus, without projecting from the plane, we could still derive a formula equivalent to (\ref{eq:utxT2}) with the help of a Poisson transformation applied to (\ref{eq:utxT}). To take the semiclassical limit, (\ref{eq:utxT2}) is the adequate expression. Indeed, to apply this semiclassical limit we must evaluate the integrals in (\ref{eq:utx2}) by the stationary phase approximation as in \cite{ozrep}, so \begin{equation} U_t(x)_{SC}\sim \sum_j2^L|\det (1+{\cal M}_j)|^{-\frac 12}\ \exp \left\{ i\hbar ^{-1}S_{tj}(x)+i\mu _j\right\} , \label{eq:utxsc} \end{equation} where ${\cal M}$ is the symplectic matrix for the linearized transformation between the neighborhood of the tips of the chord $\xi (x)$ generated by $% S_t(x)$ as a center function.The index runs over all the contributing classical orbits. In the case of a single orbit, the corresponding {\it % Morse index} $\mu _j=0$. Hence on the torus we obtain \begin{eqnarray} {\bf U}_t(x)_{SC} &\sim &\left\langle e^{i2\pi N(\left[ \left( x-\frac \chi N\right) \wedge {\bf k+}\frac 14{\bf k}\widetilde{{\frak J}}{\bf k}\right] )}\right. \nonumber \\ &&\left. \sum_j2^L|\det (1+{\cal M}_j)|^{-\frac 12}\ \exp \left\{ i\hbar ^{-1}S_{tj}(x+\frac k2)+i\gamma _j\right\} \right\rangle _{{\bf k}}. \label{eq:UTxsc} \end{eqnarray} The sum over $k$ is a sum over center points that are equivalent on the torus because of the boundary conditions, but are different points on the plane. To obtain the correct periodicity, the contribution of each term must be summed with different phases. But for each point there are several classical orbits whose chord is centered on it. The contribution of those orbits are obtained in the $j$ sum. Then, for the semiclassical propagator on the torus, we have a multiplicity of chords for any center point, due to the boundary conditions. \section{Conclusions} \setcounter{equation}{0} Our construction of the Weyl representation on the torus naturally generates the conjugate chord representation. This appears to be more useful on the torus than on the plane where it also arises. The advantage of our derivation of the Weyl representation resides in the clear geometrical interpretation of the operator basis in terms of translations and reflections in QPS, so that the law for the symbol of the product of operators acquires a simple form and generalizes to multiple products. It is important to note that the parity of the number of states $N$ plays an important role and the product law for $N$ odd is related to that in the plane case, by merely replacing the integrals by the appropriate sums. Although the geometric interpretation is valid for toral geometries, the construction can be applied to any system whose Hilbert space has finite dimension irrespective of the geometric structure of the underlying phase space, except for its compactness. Indeed this operator basis and symbols can be applied, for example, to spin systems or many-body fermionic systems \cite{lipgal}. However, such a generalization destroys the intuitive interpretation of the semiclassical limit. By defining the operators on the torus as the projections of their analogues on the plane, some important properties of the plane can then be used on the torus. We exploit this fact for periodic Hamiltonian systems where we map the continuous problem on a finite dimensional one. The path integral formulation of Hamiltonian systems on the plane allows us to obtain that on the torus, thus illuminating the semiclassical limit. The symplectic invariance of the Weyl representation on the plane translates to the torus as the Feline invariance; this fact will be used to study cat maps of general dimension \cite{loxocat}. {\it Acknowledgments: }We thanks A. Voros, M. Saraceno and R.O. Vallejos for helpful discussions. We acknowledge financial support from Pronex-MCT and A.M.F.R. also thanks support from CLAF-CNPq. \clearpage
\section{Introduction} A foam is a disordered collection of densely-packed polydisperse gas bubbles in a relatively small volume of liquid \cite{prudhommebook,djddaw,weaire84}. Foams have a rich rheological behavior; they act like elastic solids for small deformations but they flow like viscous liquids at large applied shear stress \cite{kranikrev}. The stress is relaxed by discrete rearrangement events that occur intermittently as the foam is sheared. Three-dimensional foams are opaque, which makes it difficult to observe these bubble movements directly. However, measurements \cite{gopal,gopal2} by diffusing-wave spectroscopy of three-dimensional foams subjected to a constant shear rate suggest that the number of bubbles involved in the rearrangements is small, of the order of four bubbles. Bubble rearrangements can be observed directly by fluorescence microscopy in two-dimensional foams found in insoluble monolayers at the air-water interface. A study of shear in such foams \cite{dennin} also revealed no large-scale rearrangements. While analytical theories for the response to applied steady shear may be constructed for periodic foams, only simulation approaches are possible for disordered foams. Kawasaki's \cite{okuzono} vertex model was the first to incorporate dissipative dynamics. It applies to a two-dimensional foam in the limit in which the area fraction of gas is unity (a dry foam). Bubble edges are approximated by straight line segments that meet at a vertex that represents a Plateau border. The equations of motion for the vertices are solved by balancing viscous dissipation due to shear flow within the borders by surface tension forces. At low shear rates, the elastic energy of the foam, which is associated with the total length of the bubble segments, shows intermittent energy drops with a distribution of event rate vs. energy release that follows a broad power law, consistent with self-organized criticality. The rearrangements associated with the largest events consist of cooperative motions of bubbles that extend over much of the system. Weaire and coworkers \cite{bolton0,bolton,hutzler} were the first to develop a model appropriate to a disordered wet foam. The model does not include dissipation, so it is quasi-static by construction. Thus the system is allowed to relax to an equilibrium configuration after each of a series of infinitesimal shear steps. The size of rearrangements is measured by the number of changes in nearest-neighbor contacts. For dry foams, the average event size is small, inconsistent with a picture of self-organized criticality. However, as the liquid content increases, the event-size distribution broadens, with the largest events involving many bubbles. Although the statistics are limited, this is consistent with a picture of criticality at the point where the foam loses its rigidity. The first model capable of treating wet, disordered foams at nonzero shear rate was proposed by Durian \cite{durian1}. His model pictures the foam as consisting of spherical bubbles that can overlap. Two pairwise-additive interactions between neighboring bubbles are considered, a harmonic repulsive force that mimics the effect of bubble deformation and a force proportional to the velocity difference between neighboring bubbles that accounts for the viscous drag. He found \cite{durian2} that the probability density of energy drops followed a power law, with a cutoff at very high energy events. The largest event observed consisted of only a few bubbles changing neighbors. This is inconsistent with a picture of self-organized criticality, although the effect of the liquid content on the topology statistics was not examined. Most recently, Jiang et al.\cite{jiang} have employed a large-Q Potts model to examine sheared foams. In this lattice model bubbles are represented by domains of like spin, and the film boundaries are the links between regions of different spins. Each spin merely acts as a label for a particular bubble, and the surface energy arises only at the boundaries where the spins differ. The evolution of the foam is studied by Monte Carlo dynamics with a Hamiltonian consisting of three terms: the coupling energy between neighboring spins at the boundaries of the bubbles; an energy penalty for changes in the areas of the bubbles, which inhibits coarsening of the foam; and a shear term that biases the probability of a spin reassignment in the strain direction. The spatial distribution of T1 events was examined and no system-wide rearrangements were observed. Nevertheless, Jiang, et al. found a power-law distribution of energy changes. They also found that the number of events per unit strain displayed a strong shear rate dependence, suggesting that a quasi-static limit does not exist. These four simulation approaches thus offer conflicting pictures as to (1) the existence of a quasistatic limit, (2) whether or not rearrangement dynamics at low shear rates are a form of self-organized criticality, and (3) whether or not the melting of foams with increasing liquid content is a more usual form of criticality. One possible reason for this disagreement is differences in the treatment of dissipation, and hence in the treatment of the {\it dynamics} of the rearrangements. In principle, the only accurate way in which to include dissipation in a sheared foam is to solve for the Stokes flow in the liquid films and Plateau borders. This approach has been adopted by Li, Zhou and Pozrikidis\cite{pozrikidis}, but so far it has only been applied to periodic foams. The statistics of rearrangement events are fundamentally different in periodic and disordered foams; in sheared periodic foams, all the bubbles rearrange simultaneously at periodic intervals, while in a disordered foam, the rearrangements can be localized and intermittent. Nonetheless, the Stokes-flow approach is the only one that can be used as a benchmark for more simplified models. In order to gain a better understanding of the origin of the discrepancies between the various models, as well as between the models and experiments, we report here a systematic study of the properties of a sheared foam using Durian's model. We begin by reviewing his model and discussing our numerical implementation using two different forms of dissipation. After confirming that there are no significant system-size effects for dry samples, we examine shear-rate dependence and establish the existence of a true quasistatic limit for the distribution and rate of energy drops and topology changes. This limit is shown to be independent of the dissipation mechanism for foams of different gas fractions. Finally, we examine dramatic changes in the behavior of these quantities as the liquid content is tuned toward the melting point. \section{Bubble model} Durian's model\cite{durian1,durian2} is based on the wet-foam limit, where the bubbles are spherical. The foam is described entirely in terms of the bubble radii $\{ R_i \}$ and the time-dependent positions of the bubble centers $\{ \vec{r}_i \}$. The details of the microscopic interactions at the level of soap films and vertices are subsumed into two pairwise additive interactions between bubbles, which arise when the distance between bubble centers is less than the sum of their radii. The first, a repulsion that originates in the energy cost to distort bubbles, is modeled by the compression of two springs in series with individual spring constants that scale with the Laplace pressures $\sigma/R_i$, where $\sigma$ is the liquid-gas surface tension and $R_i $ is the bubble radius. Bubbles that do not overlap are assumed not to interact. The repulsive force on bubble $i$ due to bubble $j$ is then \begin{equation} \vec{F}_{ij}^r = k_{ij} \left [(R_i + R_j) - |\vec{r}_i - \vec{r}_j| \right ] \hat{r}_{ij} \end{equation} where $\hat{r}_{ij}$ is the unit vector pointing from the center of bubble $j$ to the center of bubble $i$, and $k_{ij} = F_0/(R_i +R_j)$ is the effective spring constant, with $F_0 \approx \sigma \langle R \rangle$. The second interaction is the viscous dissipation due to the flow of liquid in the films. It, too, is assumed to be pairwise additive and is modeled by the simplest form of drag, where the force is proportional to the velocity difference between overlapping bubbles. The viscous force on bubble $i$ due to its neighbor $j$ is \begin{equation} \vec{F}_{ij}^v = -b(\vec{v}_i - \vec{v}_j), \label{drag} \end{equation} where the constant $b$ is proportional to the viscosity of the liquid, and is assumed to be the same for all bubble neighbors. The net force on each bubble sums to zero, since inertial effects are negligible in this system. Summing over those bubbles $j$ that touch bubble $i$, the equation of motion for bubble $i$ is \begin{equation} \sum_j (\vec{v}_i - \vec{v}_j) = \frac{F_0}{b} \sum_j \left [ \frac{1}{|\vec{r}_i - \vec{r}_j|} - \frac{1}{R_i + R_j} \right ] (\vec{r}_i - \vec{r}_j) + \frac{\vec{F}_i^a}{b}, \label{eq:vel} \end{equation} where $\vec{F}_i^a$ is an externally applied force, arising, for instance, from interactions with moving walls. Durian \cite{durian1,durian2} employed a further simplification of this model, in which the viscous dissipation is taken into account in a mean-field manner by taking the velocity of each bubble relative to an average linear shear profile. In this case, the total drag force on bubble $i$ due to all of its $N_{i}$ overlapping neighbors is \begin{equation} \vec{F}_{i}^v = -b N_{i} \left ( \vec{v}_i - \dot{\gamma}y_i \hat{x}\right ). \label{mf} \end{equation} In the numerical simulations reported here we use both the mean-field model of dissipation as well as the approximation represented by Eq.~\ref{drag}, which we call the local dissipation model. In the latter, at each integration time step the velocity of a bubble is measured with respect to the average of the velocities of its $N_i$ overlapping neighbors, so that the total drag force on bubble $i$ is \begin{equation} \vec{F}_{i}^v = -b \left (N_{i} \vec{v}_i - \sum_{j={\text{ nn}}} \vec{v}_j \right ) \label{local} \end{equation} For very large $N_{i}$, this reduces to Eq.~\ref{mf}; otherwise, it allows for fluctuations. One aim of our study is to establish the sensitivity of the results to the specific form of dissipation used, Eq.~\ref{mf} or Eq.~\ref{local}. In two dimensions, the area fraction of gas bubbles, $\phi$, can be defined by the total bubble area $\sum\pi R_{i}^{2}$ per system area. Because the bubbles are constrained to remain circular and their interactions are approximated as pairwise-additive\cite{lacasse}, the model necessarily breaks down for very dry foams. In fact, bubble radii can even be chosen so that $\phi$ exceeds one. In a real foam, of course, this is prevented by the divergence of the osmotic pressure. \section{Numerical Method} All the results reported here are based on simulations of a two-dimensional version of Durian's model. We use Eq.~\ref{eq:vel} to study a two-dimensional foam periodic in the $x$--direction and trapped between parallel plates in the $y$--direction. Bubbles that touch the top and bottom plates are fixed to them, and the top plate is moved at a constant velocity in the $x$--direction. (The system can also be sheared with a constant force instead of a constant velocity, but that case will not be discussed here.) Thus, bubbles are divided into two categories --- ``boundary'' bubbles, which have velocities that are determined by the motion of the plates, and ``interior'' bubbles, whose velocities must be determined from the equations of motion. The equation of motion Eq.~\ref{eq:vel} can be written in the form \begin{equation} \label{eq:Meom} {\bf M}(\{{\bf r}\})\cdot\{{\bf v}\} = \{{\bf F}^r\}/b + \{{\bf F}^a\}/b \end{equation} where $\{\bf v\}$ is a vector containing all the velocity components of all of the bubbles, $\{v_0^x, v_0^y, v_1^x, v_1^y, \ldots\}$, $\{{\bf F}^r\}$ is a vector of all of the repulsive bubble--bubble forces, and $\{{\bf F}^a\}$ contains all the forces exerted by the walls. The matrix $\bf M$ depends on the instantaneous positions of the bubbles. The $2\times2$ block submatrix $M_{ij}$ is a unit matrix $\bf 1$ if the distinct bubbles $i$ and $j$ overlap, and $\bf 0$ if they do not overlap. On the diagonal, $M_{ii} = -{\bf 1} N_i$, where $N_i$ is the number of overlapping neighbors of bubble $i$. Eq. \ref{eq:Meom} is of the form ${\bf A}({\bf r}, t)\cdot (d{\bf r}/dt) = f({\bf r}, t)$, which we solve for the bubble positions $\bf r$ with the routine DDRIV3\cite{libs}. DDRIV3 has the ability to solve differential equations in which the left hand side is multiplied by an arbitrary time-dependent matrix. Furthermore, it allows all matrix algebra to be performed by external routines, allowing us to take advantage of the sparse nature of $\bf M$. We use the SPARSKIT2\cite{libs} library for sparse matrix solutions. The only relevant dynamical scale in this problem is set by the characteristic relaxation time arising from the competing mechanisms for elastic storage and viscous dissipation, $\tau_d= b\langle R \rangle/F_0$. This is the characteristic time scale for the duration of bubble rearrangements driven by a drop in total elastic energy. Without loss of generality we set this to unity in the simulation. In these units, the dimensionless shear rate $\dot \gamma$ is the capillary number. To introduce polydispersity, the bubble radii are drawn at random from a flat distribution of variable width; in all the results reported here, the bubble radii vary from 0.2 to 1.8 times the average bubble radius. We note that the size distribution in experimental systems is closer to a truncated Gaussian with the maximum size equal to twice the average radius. The truncated Gaussian distribution arises naturally from the coarsening process \cite{glazierweaire,stavansrev}. We tested the sensitivity of our results to the bubble distribution by doing one run with bubbles drawn from a triangular distribution, and found that the shape of the distribution had no significant effect. Similarly, variation of the width of a triangular distribution has been shown to have no influence on the linear viscoelasticity \cite{durian2}. Note that it is important to include polydispersity because a monodisperse system will crystallize under shear, especially in two dimensions. In all of our runs, the system is first equilibrated with all bubbles treated as interior bubbles, and with a repulsive interaction between the bubbles and the top and bottom plates so that bubbles cannot penetrate the plates. The bubbles that touch the top and bottom plates are then converted to boundary bubbles. The top plate is moved at a constant velocity and data collection begins after any initial transients die away. In addition to recording quantitative measures of the system, we also run movies of the sheared foam in order to observe visually how the flow changes as a function of shear rate, area fraction and other parameters \cite{movies}. \section{Quantities Measured} Before showing results, we discuss the various quantities extracted during a run. Under a small applied shear strain, bubbles in a real foam distort; as the shear strain increases, the structure can become unstable and they may thus rearrange their relative positions. In the bubble model, the distortion of bubbles is measured globally by the total elastic energy stored in all the springs connecting overlapping bubbles: \begin{equation} E = \sum {1 \over 2}k_{ij} \left [(R_i + R_j) - |\vec{r}_i - \vec{r}_j| \right ] ^{2}. \label{Etot} \end{equation} Under steady shear, the elastic energy rises as bubbles distort (overlap) and then drops as bubbles rearrange. Thus, the total elastic energy fluctuates around some average value. The scale of the energy is set by the elastic interaction and is of order $F_0 \langle R \rangle$ per bubble, where $ \langle R\rangle $ is the average bubble radius. Fig.~\ref{fig1}a shows a plot of the total elastic energy as a function of strain for a system of 144 bubbles at area fraction $\phi=1.0$ driven at a constant shear rate of $\dot \gamma =10^{-3}$. Similar plots for stress vs strain are shown in Refs. \cite{durian1,durian2}. Note the precipitous energy drops, $\Delta E$, due to bubble rearrangements. In the literature, these energy drops are often referred to as avalanches. Since the term ``avalanche'' tends to imply the existence of self-organized criticality, we employ the more neutral but less elegant term ``energy drop.'' The time interval between energy drops is much larger than the duration of a single event. This is also illustrated in Fig.~\ref{fig1}b, which shows the magnitude of energy drops that occur as the system is strained. ($\Delta E$ is normalized by the average energy per bubble $E_b$, which has been computed by averaging the elastic energy over the entire duration of a run and dividing by the total number of bubbles in the system, $N_{bub}$.) These recurring precipitous rearrangements represent the only way for the foam to relax stress: there is no mechanism involving a gradual energy release, as illustrated in Fig.~\ref{fig1}a. Note that we compute only the total elastic energy of the system; because events can be localized and intermittent, the elastic energy may be dropping in one region of the sample and rising in other regions. This would limit the size of the energy drop measured. While useful for building intuition, the distribution of energy drops does not yield direct information about bubble rearrangements. Therefore, we also measure the number $N$ of bubbles that experience a change in overlapping neighbors during an energy drop. We exclude events in which two bubbles simply move apart or together; thus the smallest event is $N=3$. A typical sequence of configurations before, during, and after an event is shown in the first three frames of Fig.~\ref{edropav}. In this energy drop the magnitude of the drop and the number of bubbles that change neighbors are close to the average. In the second and third frame of the sequence, we have marked the bubbles that changed neighbors since the beginning of the energy drop (shown in the first frame). As the system is strained, more bubbles change neighbors. For the particular energy drop chosen, roughly one-sixth of the bubbles eventually change overlapping neighbors. The fourth frame shows the final configuration of bubbles (colored gray) superimposed on the initial configuration at the start of the energy drop (colored black). Most of the bubble motions that lead to this average-sized energy drop are rather subtle shifts; there are no topological rearrangements. A large energy drop, from the tail of the distribution, is shown in Fig.~\ref{edropbig}. Again, the first three frames show the configurations at the beginning, middle and end of the drop, with the bubbles that change overlapping neighbors marked in gray. The fourth frame shows the extensive rearrangements that occur from the beginning to the end of the drop. The configuration shown is the final one, and the short segments are the tracks made by the centers of the bubbles during the energy drop. Typically, larger drops involve larger numbers of bubbles. Fig.~\ref{fig1}c depicts $N$ during each energy drop in the same run as in Fig.~\ref{fig1}a and b. (Here, $N$ is normalized by the total number of bubbles in the system, $N_{bub}$.) The correlation between energy drops and the number of bubbles involved is shown by a scatter plot of these quantities in Fig.~\ref{scat} for a 900-bubble system strained from 0 to 10. We see that indeed there is a strong correlation between these two measures of the size of an event. Larger drops in energy involve larger numbers of bubbles and are therefore spatially more extended. The correlation is particularly good at the large-event end. There is more variability for midsize and small events -- a large range of energy drops corresponds to the same small number of rearranging bubbles, suggesting that typical rearrangements involve only a few bubbles. Besides counting statistics for energy drops and changes in number of bubble overlaps, another direct measure of bubble rearrangements is the rate of T1 events, i.e. of topology changes of the first kind \cite{weaire84}. For a perfectly dry two-dimensional foam consisting of thin films, these are said to occur when a bubble edge shrinks to zero, such that a common vertex is shared by four bubbles, two moving apart and two moving together. These events were the only property used by Dennin and Knobler~\cite{dennin} to characterize the response of their monolayer foam to shear because they were unable to measure changes in the energy. While the time at which a T1 event occurs is well defined in a dry foam, it is somewhat ambiguous for a wet foam because there can be an exchange of nearest neighbors without a common point of contact. Moreover, while the number of bubbles involved in a T1 event is four by definition, large clusters of bubbles can rearrange, with some of the interior bubbles being involved in two or three T1 events simultaneously. It is then much harder to assign an exact time to a T1 event. To make contact with the monolayer experiments, we may define T1 events within the bubble model as follows. First we broaden the definition of ``nearest neighbors'' to also include bubbles that do not necessarily overlap, but that are nonetheless so close such that $|\vec{r}_i - \vec{r}_j| < a(R_i + R_j)$, where $a>1$ is a suitably chosen factor that may depend on $\phi$. We then say that a T1 event begins when two nearest neighbors move apart, and we say that it ends when a new nearest neighbor pair intrudes between them; the time at which the event occurs is taken as the midpoint in this sequence. This definition is illustrated in the time sequence of a T1 event shown in Fig.~\ref{T1}. While the duration of an actual T1 event in a dry foam is instantaneous, the duration within the bubble model may vary greatly. Furthermore, the midpoint in the sequence does not necessarily coincide with the exact moment the switching occurs. In many instances it takes a long time after two bubbles separate for the remaining pair to come into contact. To compare with our other measures of rearrangement, we depict in Fig.~\ref{fig1}d the number of T1 events as a function of strain for the same run as in Figs.~\ref{fig1}a, b and c. There appears to be good correlation between the largest energy drops and instances in which many T1 events occur simultaneously. However, there are many more T1 events than energy drops. This is because many T1 events can occur when a large cluster of bubbles rearranges, and because our definition also includes topology changes that cause an {\it increase} in the total elastic energy. We can examine the consequences of our definition of a T1 event by studying the distribution of the number of rearrangement events as a function of their total duration in units of the strain. This is done for both energy drops and T1 events, as shown in Fig.~\ref{durdist}. The duration of an energy drop is taken as the difference in strain between a decrease in the elastic energy and the next increase. It is evident from the duration distribution for energy drops, Fig \ref{durdist}a, that most energy drops occur over a relatively short strain scale. In units of time, the longest events are comparable to a hundred times the characteristic time scale in the problem ($\tau_d=1$ in our simulations). We find a good correlation between the number of bubbles that change overlapping neighbors and the duration of the event; the more bubbles involved in the event, the longer it lasts. The distribution for T1 events, shown in Fig \ref{durdist}b, has a qualitatively similar shape, exhibiting a slightly more rapid decrease for both fast and slow events. However, the scale on which T1 events occur is an order of magnitude larger than the characteristic duration of the energy drops. By examining the bubble motions we see that the largest energy drops are associated with many T1 events, but the difference in strain scales makes it difficult to demonstrate an exact correlation between the number of overlap changes and the number of T1's. In counting the T1 events, we include only events that have a total strain duration of less than 2. Fig.~\ref{durdist}b shows that we have included all the T1 events for this run. \section{Simulation Results} For a given system size, strain rate, dissipation mechanism and gas fraction, we now collect statistics on the following measures of bubble dynamics: (1) The probability distribution $P(\Delta E)$ for energy drops of size $\Delta E$; (2) The probability distribution $P(N)$ for the number of bubbles $N$ that change overlapping neighbors during a energy drop event; and (3) The event rates for both energy drops and T1 events, ${S}(T1)$ and ${S}(\Delta E)$, both defined as the number of events per bubble per unit strain. \subsection{System Size} We first address the important issue of the finite size of the simulation sample. This is done for dry foams, $\phi=1.0$, driven at a slow strain rate, $\dot \gamma=10^{-3}$. The results for four system sizes, $N_{bub}=36$, 144, 324 and 900, are shown in Fig.\ref{size}. In these runs, the systems were strained up to 80, 80, 31 and 10, respectively. The top plot shows the energy drop distribution scaled by $E_b$, the average energy per bubble. It shows that energy drops vary greatly in size over the course of a single run. The general features of this distribution have been reported earlier \cite{durian2}. There is a power-law region with an exponent of -0.7 that extends over several decades in $\Delta E/E_b$, followed by a sharp cutoff that occurs above a characteristic event size. Such a distribution has a well-defined average energy drop, which is near the cutoff between 2$E_b$ and 3$E_b$ for the systems shown here. The slight deviation from power-law behavior for small $\Delta E$ was absent in the earlier simulations \cite{durian2}, which did not exclude two-bubble events, and which had a different roundoff error. Also, as seen earlier \cite{durian2}, the two largest systems, with 324 and 900 bubbles, respectively, have nearly identical distributions. This has two important implications; namely, that the sharp cutoff of the power-law distribution is not a finite-size effect, and that the system does not exhibit self-organized criticality. The presence of a characteristic energy-drop size can be corroborated by examining the number of bubbles that participate in rearrangements for the same set of runs, which is given in the middle plot, Fig. \ref{size}b. This quantity has not been studied previously within the bubble model. We plot the probability distribution $P(N)$ of the number of bubbles $N$ that change overlapping neighbors during a rearrangement. The distribution decreases monotonically with a sharp cutoff at the large-event end. This indicates that most of the rearrangements are local and involve only a few bubbles. Fig.~\ref{size}b shows that as the system size increases, the largest events represent a smaller fraction of the total number of bubbles. Indeed, the tail of the distribution extends to smaller and smaller values of $N/N_{bub}$ with no signs of saturation as the system size $N_{bub}$ increases, indicating diminishing finite size effects. We next look at the system-size dependence of event rates, $S(T1)$ and ${S}(\Delta E)$, for the number of T1 events and energy drops per bubble per unit strain. This is shown in the bottom plot, Fig. \ref{size}c, for the same runs as in Figs. \ref{size}a-b. We find that ${S}(\Delta E)$ decreases very slightly with increasing system size, but saturates for the largest systems. The results for ${S}(T1)$ show a stronger system-size dependence, increasing slightly with $N_{bub}$. This could be due to the fact that bubbles on the top and bottom boundaries of the system are fixed, which lowers the number of possible T1 events per bubble. As the system size grows, the boundary bubbles represent a smaller fraction of the system so the event rate increases towards its bulk value. In short, all of our measurements at $\phi=1.0$ and $\dot \gamma=10^{-3}$ indicate that the rearrangement events are localized and that there is no self-organized criticality. This agrees with observations of rearrangements in both monolayer and bulk foams. \subsection{Shear Rate Dependence} Now that size effects have been ruled out for dry foams, we may examine the influence of shearing the sample at different rates. Experiments by Gopal and Durian\cite{gopal2} on three-dimensional foams show a marked change in the character of the flow with increasing shear rate. At low shear rates, the flow is characterized by intermittent, jerky rearrangement events occurring at a rate proportional to the strain rate. As the shear rate increases, so that the inverse shear rate becomes comparable to the duration of a rearrangement event, the flow becomes smoother and laminar, with all the bubbles gradually rearranging all the time. This was attributed to a dominance of viscous forces over surface tension forces when the strain rate exceeds the yield strain divided by the duration of a rearrangement event. In movies of our simulation runs, we also observe a crossover from intermittent, jerky rearrangements to smooth laminar flow. Similar smoothing has also been seen in stress vs. strain at increasing shear rates for the mean-field version of bubble dynamics \cite{durian2}. This raises the question of how the statistics of rearrangement events change with shear rate. Specifically, how is the ``smoothing out'' of the flow reflected in the statistics at high rates, and is there a quasistatic limit at low shear strain rates, in which rearrangement behavior is independent of strain rate? Earlier numerical studies by Bolton and Weaire\cite{bolton} were restricted, by construction, to the quasistatic limit. Okuzono and Kawasaki\cite{okuzono} examined nonzero shear rates, but focused only on establishing the low shear-rate limit. Recently, Jiang and coworkers found a strong dependence of the T1 event rate on shear rate \cite{jiang}. They found that the number of T1 events per bubble per strain, ${S}(T1)$, decreases sharply with strain rate with no evidence of a quasi-static limit. Our results for rearrangement behavior vs strain rate are collected in Fig. \ref{shear} for a 144-bubble system at $\phi=1.0$. The top plot for the probability distribution of energy drops indicates that there is no gross change in $P(\Delta E)$ with shear rate, even though our movies show a smoothing with less frequent energy drops. However, there is some suppression of small energy drops with an accompanying increase at large energy drops, as reflected in a somewhat smaller power-law exponent and larger cutoff at high values of $\Delta E/E_{b}$. It is not apparent from $P(\Delta E)$ vs $\Delta E/E_{b}$, but we find that the average energy drop $\langle \Delta E \rangle$ and the average energy per bubble $E_{b}$ both increase with shear rate, and that $\langle \Delta E \rangle$ increases more rapidly. The reason why $E_{b}$ increases with shear rate is, of course, that viscous forces become more important than elastic forces and lead to increasing deformation (or in our model, overlaps) of bubbles. The net result is that there are fewer, relatively larger, rearrangements at high strain rates. The tendency that small events are suppressed with increasing shear rates is also borne out by the distribution of the number of bubbles that change neighbors during an energy drop, as shown in Fig.~\ref{shear}b. Note that unlike the previous curves, $P(N)$ is plotted here on a linear scale. Two systematic trends emerge with increasing $\dot{\gamma}$: there are relatively fewer small events, i.e. $P(N)$ decreases significantly at small $N/N_{bub}$, and the tail extends to slightly higher $N/N_{bub}$. For $\dot{\gamma} \geq 10^{-1}$ the distribution is fairly flat, suggesting that no one event size is dominant and there are numerous large events of the order of the system size. This suggests that at this shear rate the system no longer relaxes stress by intermittent rearrangements, but by continuous flow, as confirmed by our movies of the runs \cite{movies}. The trend in $P(N)$ is seen in larger systems as well. For the 900-bubble system we also find that as the shear rate increases from $10^{-5}$ to $10^{-3}$, the distribution flattens and extends to higher values of $N$. The average number of rearranging bonds increases with shear rate, consistent with the picture of many bubbles in motion as the system becomes more liquid-like. We cannot, however, probe the system at very high shear rates. Data above a shear rate of about 1 cannot be trusted because of the nature of the model used. At high rates of strain the viscous term dominates and the elastic forces are not strong enough to prevent clumping of bubbles. This is actually an artifact of the assumption that only overlapping bubbles interact viscously; such clumping does not occur until much higher strain rates in the mean-field version of dynamics. Another reason why we do not study shear rates higher than unity is because we do not allow bubble breakup under flow (recall that $\dot\gamma$ is the capillary number). The gradual smoothing with increasing shear rate is most apparent in Fig \ref{shear}c, where we see that the event rates of T1 events and energy drops both decrease with increasing strain rate. For the T1 events, the decrease is slight, and is primarily due to the fact that the event duration becomes even longer. The decrease is more dramatic for the energy drop events. With increasing strain rate, the average energy drop increases and the rate of energy drops decreases. Let us now re-examine the behavior of all quantities in Fig.\ref{shear}, focusing on behavior at low shear strain rates. Note that all quantities appear to approach a reasonably well-defined ``quasistatic'' limit insensitive to the value of $\dot\gamma$. We thus have the following picture. For small $\dot{\gamma}$, the time between rearrangements is typically much longer than the duration of a rearrangement, implying there is adequate time for the system to relax stress. As the shear rate increases, bubbles are constantly in motion and cannot fully rearrange into local-minimum-energy configurations. Therefore, the viscous interactions dominate, and the system flows like an ordinary liquid. \subsection{Mean-Field vs Local Dissipation} In the bubble model at higher strain rates, the behavior was seen to depend on the form of dissipation: clumping for local dissipation, Eq.~\ref{local}, as opposed to no clumping for mean-field dissipation, Eq.~\ref{mf}. In this section we will investigate whether dissipation affects the low-strain-rate behavior as well. If there truly exists a quasi-static limit as $\dot{\gamma}\rightarrow 0$, as suggested by the plots in the previous section, then the form of dissipation should have no influence. This need not occur, since once a rearrangement starts it proceeds with finite speed according to dynamics set by a competition between surface tension and dissipation forces. For example, it is conceivable that the mean-field dynamics might discourage the mushrooming of a tiny shift in bubble position into a large avalanche, whereas local dynamics might not. Another important issue is that differences in mean-field vs local dissipation could be relevant to true physical differences between bulk foams and Langmuir monolayers at an air/water interface. For three-dimensional foams, the shear is transmitted through the sample via bubble-bubble interactions, so the dissipation might be better captured by the local dissipation model. In contrast, for two-dimensional Langmuir monolayer foams the subphase imposes shear on the monolayers, and the dissipation might therefore be closer to that calculated with the mean-field model. To investigate the influence of mean-field vs local dynamics, we can simply compare avalanche statistics. This is done in Fig. \ref{MFL} for 144-bubble systems at four different area fractions, all sheared at $\dot\gamma=10^{-3}$. The top plot shows results for the energy-drop distribution, $P(\Delta E)$, with solid/dashed curves for local/mean-field dissipation respectively. There is no significant difference seen between the two choices of dissipative dynamics. This is also true of the spatial extent of the rearrangements, as seen in the middle plot for the probability distribution $P(N)$ of rearranging bubbles. The bottom plot for the rate of energy-drop and T1 events also shows little significant difference between mean-field and local dynamics. The only distinction is a slightly greater rate of T1 events in the mean-field case. This reflects the difference in duration of T1 events within the two models; we find that T1 events tend to last longer within the local dissipation model. Since we do not count T1 events that last longer than a strain of 2, we count fewer events within the local model than the mean-field version. Thus, the differences in $ S(T1)$ may simply be due to our method of counting T1 events. Taken together, the three plots in Fig.~\ref{MFL} encourage us to believe that the rearrangement dynamics predicted by the model are robust against details of the dissipation. They also provide further evidence for the existence of a true quasi-static limit, where the effect of strain rate is $\it{only}$ to set the rate of rearrangements. \subsection{Gas Area fraction} Finally, we turn to the issue of how the elastic character of a foam disappears with increasing liquid content, and possibility of critical behavior at the melting transition. The principal signature of the melting, or rigidity-loss, transition is that the shear modulus $G=\lim_{t\rightarrow\infty} \sigma(t)/\gamma$ vanishes and the foam can no longer support a nonzero shear stress without flowing. In two-dimensional systems, this happens at a critical gas fraction corresponding to that of randomly packed disks, $\phi_c \approx 0.84$. This has been seen in several different simulations, where the gas fraction was tuned to within 0.05 of the transition\cite{bolton0,bolton,hutzler} and where it was tuned through, and even below, the transition\cite{durian1,durian2}. Other signatures of melting are that the osmotic pressure vanishes as a power-law\cite{durian1,durian2,lacasse} the coordination number decreases towards about 4 as a power-law\cite{bolton0,bolton,hutzler,durian1,durian2,weaire}, and that the time scale for stress relaxation following an applied step-strain appears to diverge \cite{durian1,durian2}. Here we look for signs of melting in the statistics of avalanches during slow, quasi-static flow. Within our model, an increase in liquid content causes a decrease in the average overlap between neighboring bubbles. This in turn produces a decrease in the average elastic energy of the system, $E_{b}$ and sets the scale for the average energy drop $\langle\Delta E\rangle$ per rearrangement. It therefore should also decrease at lower gas fractions $\phi$. The energy drop and size statistics of rearrangement events for increasingly wet foams were shown already in Fig.~\ref{MFL}, but were discussed only in the context of mean-field vs local dissipative dynamics. A clear trend emerges when we examine the $\phi$ dependence specifically. In the top plot Fig.~\ref{MFL}a for $P(\Delta E)$, we see that the power-law behavior for small events does not change, but that the exponential cut-off moves towards larger values of $\Delta E/E_b$ as $\phi\rightarrow\phi_{c}$. Though both $\langle\Delta E\rangle$ and $E_{b}$ decrease towards zero, the latter evidently vanishes more rapidly. This results in a broader distribution of event sizes near the melting transition; as the system becomes more liquid, large events are more prevalent. The probability distribution $P(N)$ for the numbers of bubbles involved in rearrangement events is shown in Fig. \ref{MFL}b. It displays similar trends as a function of $\phi$, but not as pronounced as in $P(\Delta E)$. Namely, the power law for small $N$ is unaffected by $\phi$, but the exponential cut-off moves towards slightly larger events as $\phi\rightarrow\phi_{c}$. Thus, although the scale of energy drops increases dramatically, the number of broken bonds only increases marginally. Note, however, that the largest events include almost all the bubbles in the system; thus, the relatively weak dependence of $P(N)$ on $\phi$ could be a finite-size effect in these $N_{bub}=144$ systems, as we will show below. The behavior of ${S}$, the number of energy drops and T1 events per bubble per strain, is shown in Fig.\ref{MFL}c. As the system becomes wetter, there is no noticeable change in the event rate ${S}(\Delta E)$ for energy drops. In contrast, if our definition of nearest neighbors only includes overlapping bubbles, we find that ${S}(T1)$ decreases as $\phi$ decreases. This runs counter to expectations--bubbles in a wet foam should have more freedom to move and rearrange because the energy barrier between rearrangements is lower and the yield strain is smaller. The apparent drop arises because the bubble coordination number is much higher in a dry foam (roughly 6) than in a wet foam (roughly 4). As a result there are more overlapping neighbors for each bubble in a dry foam, and more possibilities for the occurrence of T1 events. In the wet foam, however, there are many T1 events that do not satisfy the stringent starting or ending configurations because neighboring bubbles do not overlap. It is therefore appropriate in wet foams to modify the criterion for neighbors to $|{\bf r}_i - {\bf r}_j| < a(R_i + R_j)$, where the proximity coefficient $a$ is taken as $1/\phi$. When T1's are computed with this definition, we find no significant dependence on area fraction. The fact that the power-law region of the energy drop distribution is more extended at lower area fractions suggests the possibility of a critical point as the close-packing density, $\phi_{c}$, is approached from above. This would imply a pure power-law distribution $P(\Delta E)$ for the energy drops at $\phi_{c}$, which would presumably be accompanied by a growing correlation length, as well as the growing relaxation time observed previously in Refs.~\cite{durian1,durian2}. Note, however, that the distribution of the number of bubbles involved in a rearrangement, $P(N)$, does not depend very strongly on $\phi$ for the 144-bubble systems of Fig. \ref{MFL}; furthermore, the cut-off to power-law behavior is always present, no matter how closely $\phi_{c}$ is approached. This raises the question of whether finite system size effects are more important at values of $\phi$ near $\phi_{c}$ (recall from Fig.~\ref{size} that there were no significant system size effects near $\phi=1$). To examine this, we have plotted the dependence of $P(\Delta E)$, $P(N)$ and $ S$ on system size in Fig.~\ref{size:85}. We indeed find a strong system size dependence in $P(\Delta E)$ at $\phi=0.85$ just above the melting transition, with no saturation at the largest size studied (900 bubbles). This is consistent with the existence of a long correlation length. The distribution of the number of bubbles per energy drop, $P(N)$ also shows signs of criticality. Recall from Fig.~\ref{size}b that at $\phi=1$, the tail of $P(N)$ was cut off at smaller and smaller values of $N/N_{bub}$ with increasing system size at $\phi=1$. This was consistent with a short correlation length, characteristic of localized rearrangement events. At $\phi=0.85$, the behavior with increasing $N_{bub}$ is quite different, as shown in Fig.~\ref{size:85}b. The distribution falls off slightly more rapidly with $N/N_{bub}$ at larger system sizes (probably because $\phi=0.85$ still lies above $\phi_{c}$), but the largest events in the system still involve the same fraction $N/N_{bub} \approx 0.75$ of bubbles, indicating a correlation length that is comparable to the largest system size studied (30 bubble diameters across). The event rates for energy drops and T1 events for the different system sizes at $\phi=0.85$ are shown in Fig. \ref{size:85}c. The behavior is not markedly different from that found for the drier foam. Recall, however, that we have adjusted our definition of a T1 event by changing the proximity coefficient $a$ with area fraction, so little can be expected to be learned from this measure. \section{Discussion} We have reported the results of several different measures of rearrangement event dynamics in a sheared foam. A comparison of the probability distribution of energy drops $P(\Delta E)$ with the probabilty distribution of bubbles changing neighbors $P(N)$ shows that the size of an energy drop correlates well with the number of bubbles involved in a rearrangement (see Fig.~\ref{scat}). This is valuable because the energy drop-distribution has been widely studied theoretically, but is very difficult to measure experimentally. The number of bubbles involved in rearrangements, however, can be probed with multiple light scattering techniques on three-dimensional foams\cite{gopal} and by direct visualization of two-dimensional foams\cite{dennin}. A study of the rate of occurrence of topological changes (T1 events) provides a further link to experiments. In general, our results agree with experiments on three-dimensional and two-dimensional foams. Despite its simplicity, the bubble model appears to capture the main qualitative features of a sheared foam remarkably well. For example, we find that the size of rearrangement events is typically small at low shear rates and at area fractions not too close to $\phi_{c}$. This is in accord with experiments of Gopal and Durian\cite{gopal}, and Dennin and Knobler\cite{dennin}, as well as simulation results of Bolton and Weaire\cite{bolton} and Jiang and coworkers\cite{jiang}. Our results do not agree with those of Okuzono and Kawasaki\cite{okuzono}, however, who found power-law distributions of rearrangement events at $\phi=1$ in two dimensions. The largest discrepancies between our results and those of others lie in the statistics of T1 events. We find that the number of T1 events per bubble per unit strain is of order unity and is generally insensitive to shear rate and gas area fraction. Kawasaki et al. \cite {kawasaki} found similar results: ${S}(T1) = 0.5$ and no dependence on shear rate. In the Potts-model simulations \cite {jiang}, however, ${S}(T1)$ is unity at $\dot{\gamma}=10^{- 3}$ but falls to about 0.01 at $\dot{\gamma}=10^{- 1}$. The monolayer experiments \cite{dennin} yielded values of $ S(T1) \approx 0.15$, nearly an order of magnitude lower than predicted by our simulations. Durian \cite{durian2} reported a number of rearrangement events per bubble per unit strain for simulations of a 900-bubble system at $\dot{\gamma}=10^{- 5}$ that was comparable to the monolayer result, but he measured the number of energy drops per bubble per unit strain, $S(\Delta E)$, not the T1 event rate, $S(T1)$. Note that our energy-drop event rate, $S(\Delta E)$, agrees well with Durian's earlier result. One might guess that the discrepancy between our measurement of $ S(T1)$ and that of the monolayer experiment might lie in the method of analysis used to count T1 events. Unlike the simulations, in which the number of T1 events can be computed from an analysis of bubble positions as a function of time, the number of T1's in the monolayer studies was determined by repeated viewing of videotapes of the experiments and counting of the events as the foam cells reach their midpoint configuration. It seemed possible, then, that the difference between the simulation and the experiment was the result of a systematic undercounting of the number of the events. To check this possibility, the number of T1's in a simulation run was determined by observations of the animated bubble motions. The number of events missed in this unautomated counting was only 2\% of the total. We believe that the origin of the discrepancy between the T1 event rates in the simulation and the monolayer experiment lies in the yield strain. While the yield strain in the model system is less than 0.2, which is consistent with that measured in three-dimensional foams, that in the monolayer foams is closer to unity. Bubbles in monolayer foams can therefore sustain very large deformations without inducing rearrangements. The T1 event rate should be inversely proportional to the yield strain. Thus, the ratio of $ S(T1)$ in the simulation to $ S(T1)$ in the experiment should equal the ratio of the yield strain in the experiment to the yield strain in the simulation. This is exactly what we find. One of our main results is that a quasistatic limit exists within the bubble model. We find that the statistics of rearrangement events are independent of shear rate at low shear rates. This agrees with the monolayer experiments\cite{dennin}, which measured T1 event rates at two different shear rates, $\dot \gamma=0.003 s^{-1}$ and 0.11$s^{-1}$. Dennin and Knobler found no noticeable difference in the T1 event rate, despite the fact that the shear rates studied differ by a factor of thirty. In addition, Gopal and Durian found that the number of rearrangement events per bubble per second in a three-dimensional foam is given by the event rate in the absence of shear plus a term proportional to the shear rate. In their case, the event rate was nonzero in the absence of shear because of coarsening; we have neglected this effect in our simulations. However, we do find that the rearrangement event rate per unit time is simply proportional to the shear rate at low shear rates. Thus, experimental results in both two and three dimensions contradict the simulation results of Jiang, et al.\cite{jiang}, which find no quasistatic limit, but agree with our findings. The form of dissipation used in the bubble model is a simple dynamic friction, which does not capture the hydrodynamics of fluid flow in the plateau borders and films in a realistic way. However, our results suggest that we may still be capturing the correct behavior at low shear rates. We find that the rearrangement event statistics are the same whether we use mean-field or local dissipation at low shear rates. This suggests that the statistics are determined by elastic effects rather than viscous ones at low shear rates, and that the behavior in that limit should be independent of the form of viscous dissipation used. Finally, our results as a function of gas area fraction imply that there may be a critical point at the melting transition, as the area fraction approaches the random close-packing fraction from above. Previous studies showed that both the shear modulus and yield stress vanish as power laws at the melting transition\cite{bolton,durian1}, and that the stress relaxation time appears to diverge\cite{durian1}. Here, we have shown by finite-size studies that there is also a correlation length, characterizing the size of rearrangements, which grows as one approaches the melting transition. We also find that the distribution of energy drops appears to approach a pure power law in that limit. The existence of a critical point at the melting transition remains to be tested experimentally. The vanishing of the shear modulus and osmotic pressure at the transition has been measured by Mason and Weitz\cite{mason} for monodisperse, disordered emulsions, and by Saint-Jalmes and Durian for polydisperse gas-liquid foams\cite{arnaud}. However, these small-amplitude-strain rheological measurements could not test whether there is a diverging length scale for rearrangements in a steadily sheared system at the melting transition. On the other hand, Gopal and Durian\cite{gopal} have measured the size of rearrangement events in a gas-liquid foam, but only at packing fractions well above the melting transition. At lower packing fractions close to the melting transition, the liquid drains too quickly from the foam due to gravity to permit such measurements. Experiments under microgravity conditions should be able to resolve whether the melting transition is indeed a critical point. \acknowledgements We thank Narayanan Menon and Ian K. Ono for many helpful discussions, and we thank Michael Dennin for performing the visual analysis of the number of T1 events. This work was supported by the National Science Foundation through grants CHE-9624090 (AJL), CHE-9708472 (CMK), and DMR-9623567 (DJD), as well as by NASA through grant NAG3-1419 (DJD).
\section{Introduction} \indent The current-voltage (I-V) relation of normal metal -superconductor (NS) interfaces~\cite{btk}-\cite{lesovik} is strongly modified both by wave interference phenomena (producing quasi-bound Andreev levels) and ballistic transport at the interface. Several groups have observed such novel superconducting phenomena at the NS interface between a superconductor and semiconductor~\cite{kleins1}-\cite{poirier}. Based on the success of Marsh et al.~\cite{marsh1}-\cite{marsh4} in fabricating In and Sn `alloyed' contacts to a two-dimensional electron gas (2DEG) formed at the AlGaAs/GaAs interface, we have studied `alloyed' In contacts to the 2DEG. We find the mechanism for producing highly transmissive NS contacts is In growth into the AlGaAs `guided' along a preferred crystallographic direction. For an AlGaAs/GaAs heterojunction with a [100] oriented surface, we find that In growth into the AlGaAs occurs preferentially along the \{111\} crystallographic planes. This unusual type of In growth into the AlGaAs produces an `inverted pyramid' or `field emission' point contact tip as shown in Fig.~\ref{cartoon1}(a). Similar microstructure for metallic contacts to GaAs has been observed for both AuGeNi~\cite{triangles} and Au~\cite{gold-GaAs} metallizations. This guiding of In into the AlGaAs also allows the In to maintain its superconducting properties. An AlInGaAs alloy, formed by diffusing In into AlGaAs, would simply be a normal metal. For such a crystallographically defined point contact metallization, we find the closer one can grow the tip of the point contact to the 2DEG without contacting it, the higher the transmission coefficient of an electron incident from the NS contact into the 2DEG. For such nearly ballistic transport through the NS junction, a corresponding excess current results~\cite{btk}. Growing the In down into direct contact with the 2DEG, on the other hand, results in a low-transmission normal metal - insulator - superconductor (NIS) contact and its corresponding Giaever tunneling I-V characteristic~\cite{btk}. We postulate that In in direct contact with the 2DEG depletes the electrons around it, forcing the superconducting electrons to tunnel through a large depletion layer near the contact, shown schematically in Fig.~\ref{cartoon1}(b). Electron depletion around a metallic contact to GaAs is commonly known as a Schottky barrier. A similar mechanism for forming highly transmissive AuGeNi contacts to GaAs was originally postulated by Braslau~\cite{braslau}-\cite{woodall}. This mechanism for forming highly transmissive tunneling type AuGeNi contacts to GaAs is the reason some AlGaAs/GaAs heterojunction transistors can operate at low temperatures with low contact resistance. \begin{figure} \centps{figs1a.eps}{45} \centps{figs1b.eps}{45} \caption{Schematic of the In profile obtained after annealing for (a) a low temperature/short time anneal (Sample 1) and (b) a higher temperature/longer time anneal (Sample 2). Penetration of In into the AlGaAs layer is guided along a preferred crystallographic direction in (a), forming point contacts to the 2DEG. The point contacts grow together and penetrate the 2DEG in (b).} \label{cartoon1} \end{figure} In this paper we correlate transmission electron microscope (TEM) photographs of the superconducting In contacts to the resulting I-V characteristics of the NS junctions. All the different $dI/dV$ characteristics shown in this paper are from nominally identical samples, grown and prepared from the same GaAs wafer at the same time. The only differences between the samples is in post process contact annealing, and hence in the contact geometry. Changes in contact geometry produce widely different $dI/dV$ characteristics. In point contacts grown near the 2DEG result in ballistic transport of electrons through the NS contact and an excess current. In in direct contact with the 2DEG produces lower transmission contacts and Giaever tunneling. Since the GaAs semiconductor forming the normal metal is also weakly localized, we are able to observe weak localization corrections to Giaever tunneling~\cite{vanwees1}. In In/GaAs junctions where some region of the contact is transmissive, that portion of the contact will produce a conductance drop around zero bias~\cite{marmorkos1}. The incoherent addition of the conductance from different regions of the same `contact' could therefore generate the `finite bias anomaly' seen in Ref.~\cite{poirier}. \section{Formation of the NS Contact} \indent A cross section of the unnannealed In/GaAs heterostructure is shown in Fig.~\ref{heterostructure}(a). An undoped Al$_{0.3}$Ga$_{0.7}$As spacer layer, followed by a Si doped Al$_{0.3}$Ga$_{0.7}$As layer and a 50 $A^{o}$ protective Si doped GaAs layer, was grown on an undoped (semi-insulating) (100) GaAs substrate. The resulting mobility of the 2DEG at liquid nitrogen temperature was about 125,000 $cm^{2}$/$V-s$. The In contacts were deposited by thermal evaporation and liftoff. A top view of the device is shown in Fig.~\ref{heterostructure}(b). Two In pads, each of dimension 3.2mm$X$2.5mm and seperated by a nominal gap of 4$\mu$, were deposited on top of the heterostructure using thermal evaporation lift off. After annealing, we diffused In through the AlGaAs barrier layer to contact the 2DEG. The annealing temperature was varied between 500$^{o}$C and 600$^{o}$C. For temperatures of 450$^{o}$C or less the In failed to contact the 2DEG, while annealing temperatures greater than 700$^{o}$C caused thermal deterioration of the interface. Large In grains, 1-2 microns in diameter, grew on the contacts after annealing. \begin{figure} \centps{figs2a.eps}{45} \centps{figs2b.eps}{45} \caption{(a) Cross-section of the AlGaAs/GaAs heterostructure before annealing the superconducting In contacts. (b) Top view of the contact geometry.} \label{heterostructure} \end{figure} A TEM micrograph of Sample 1, annealed for a relatively short duration e.g. at 550 $^{o}C$ for 2 minutes, is shown in Fig.~\ref{tem1}(a). Fig.~\ref{tem1}(a) shows that Indium starts growing into (100) AlGaAs/GaAs preferentially along the $<$111$>$ directions. The growth seems to be guided by the \{111\} crystallographic planes of GaAs inclined at angles of about 55$^{o}$ to the surface. Therefore for short annealing times we get `spikes' of Indium descending towards the interface, forming an array of point contacts pictured in Fig.~\ref{tem1}(a). The spikes are rather non-uniform in size and irregular in their penetration depths, probably nucleating at defects in the suface oxide of AlGaAs. In grain growth on the wafer surface therefore has little effect on the final microstructure and geometry of the NS contact. The net total conductance of the junction is determined by the sum of the conductance of all the point contacts in parallel. This growth mechanism is similar to that for alloyed Au-Ge-Ni contacts with GaAs~\cite{woodall} in which most of the conduction is through isolated Ge rich islands formed on the GaAs. \begin{figure} \begin{center} \setlength{\epsfxsize}{7.0truecm} \setlength{\epsfysize}{7.0truecm} \epsfbox{image3-2.eps} \setlength{\epsfxsize}{7.0truecm} \setlength{\epsfysize}{7.0truecm} \epsfbox{image2-2.eps} \end{center} \caption{TEM photographs of an In-AlGaAs/GaAs contact annealed (a) for a short time (Sample 1) and (b) a longer time (Sample 2). Penetration of In into the sample is guided by the \{111\} planes, forming the In point contacts are clearly observable in (a). The point contacts agglomerate and grow through the AlGaAs/GaAs interface in (b).} \label{tem1} \end{figure} A TEM micrograph of Sample 2, annealed at 550$^{o}$C for 6 minutes and then at 650$^{o}$C for 3 minutes, is shown in Fig.~\ref{tem1}(b). For the longer annealing times and higher annealing temperatures used in sample 2, more such In spikes grow from the deposited In contact. Fig.~\ref{tem1}(b) shows these In spikes coalesce and penetrate completely through the AlGaAs/GaAs interface to physically touch the 2DEG, as depicted schematically in Fig.~\ref{cartoon1}(b). The conductance characteristics of Samples 1 and 2 differ dramatically as described in the next section. A high magnification TEM photograph of one of the inverted pyramid type In spikes in Sample 1 is shown in Fig.~\ref{tem2}. The (100) GaAs surface is towards the top of Fig.~\ref{tem2}. The penetration of In into the AlGaAs is clearly guided by \{111\} crystallographic planes. One can see in the high resolution TEM picture that the In indeed follows the \{111\} AlGaAs planes for several atoms. The In-AlGaAs boundary then moves abruptly along the [010] direction for 1-2 atoms before continuing along the \{111\} planes. The detailed TEM picture in Fig.~\ref{tem2} shows that while the \{111\} planes strongly guide the growth of In into (100), the In does not exactly follow those planes. However, the overall structure of the In contacts still resembles a point contact. \begin{figure} \begin{center} \setlength{\epsfxsize}{7.0truecm} \setlength{\epsfysize}{7.0truecm} \epsfbox{image1small.eps} \end{center} \caption{High magnification TEM photograph of a portion of a single In point contact grown into a (100) AlGaAs surface. In growth into the AlGaAs is strongly guided along the \{111\} planes.} \label{tem2} \end{figure} \section{Current-Voltage Characteristics} \indent Hall measurements were made on the 2DEG at the interface at a temperature of 400 mK. From the slope of transverse resistance $R_{xy}$ we estimate the carrier density to be $1.8X10^{11}$ $cm^{-2}$. The mobility was 215,000 ${{cm}^2}/{V-s}$ which yielded a mean free path $l$ of about 0.4 $\mu$. From the low-field magnetoresistance, we estimate the phase breaking length $l_{\phi}$ to be about 1.6 $\mu$m~\cite{chaud-thesis}, which is less than the 4 $\mu$m separation between the two In pads in Fig.~\ref{heterostructure}(b). We conclude that the conductance of the devices is essentially equivalent to two NS junctions in series separated by a series resistor (the 2DEG). \subsection{NS Junction with an excess current} \indent In Fig.~\ref{base12}(a) we show the conductance characteristics of Sample 1 at a base temperature of 100 mK. Fig.~\ref{base12}(a) shows an increase in conductance of about 10-12\% around a range of $\pm$ 6mV. The 10-12\% excess conductance around zero voltage shows that the junction behaves like a moderately transmissive junction~\cite{btk}, with the majority of the diffused Indium spikes forming transmissive interfaces with the 2DEG of the GaAs. The large oscillations of the conductance in Fig.~\ref{base12}(a) are reproducible with thermal recycling of the device. The oscillations are therefore likely a consequence of electron wave interference due to scattering from the fixed In point contacts to the 2DEG. The parasitic resistance of the 2DEG in series with the two NS junctions stretches the dI/dV versus V characteristics along the voltage axis, and also suppresses any changes observed along the dI/dV axis. Series resistance of the 2DEG explains why we observe only a 10-12\% excess conductance instead of an excess conductance approaching 100\% in Fig.~\ref{base12}(a). \begin{figure} \centps{figs5a.eps}{60} \centps{figs5b.eps}{60} \caption{Conductance vs the DC bias voltage across the two Indium pads for (a) Sample 1 and (b) Sample 2. Sample 1 shows nearly 100\% ballistic transport of electrons (after correcting for series resistance), while Sample 2 shows Giaever tunneling (through a tunnel barrier having transmission of the order $T \simeq 0.1$).} \label{base12} \end{figure} We can crudely estimate the interface transmission in Fig.~\ref{base12}(a) by comparing it to an ideal, ballistic NS junction~\cite{btk}. A perfectly transmissive interface shows a 100\% rise of conductance when the voltage bias satisfies $|eV| \leq \Delta$, where $2\Delta$ superconducting energy gap~\cite{btk}. Setting $2 \Delta =$ 6mV (for two NS junctions in series), and using the BCS formula $2 \Delta = 3.5 k_B T_c$, gives a critical temperature $T_c = 20$K. Since the actual critical temperature of the contacts is about $3.4$K, we infer a 1mV drop across the two NS interfaces and 5mV across the semiconductor. The total conductance at the gap voltage can be read directly from Fig.~\ref{base12}(a) as about 9mS, from which we infer a series resistance of about 92.6$\Omega$ and a resistance of the two NS interfaces in series of about 18.5$\Omega$ (when the voltage across the NS interface is less than the energy gap). When the voltage across the two NS interfaces is large, we can read directly off Fig.~\ref{base12}(a) a conductance of about 7.75mS, or a total resistance of about 129$\Omega$. Subtracting the series resistance, we see the device resistance changes from 18.5$\Omega$ when $V_{\rm interface} \leq \Delta$ to about 36.4$\Omega$ when $V_{\rm interface} \gg \Delta$, roughly a 97\% increase in background conductance. Since strong wave interference is present in Fig.~\ref{base12}(a), this comparison should be regarded as giving an order of magnitude estimate. However, after correcting for series resistance, this estimate indicates the junction in Fig.~\ref{base12}(a) is a nearly ballistic NS interface. \subsection{Giaever-type Tunnel Junctions} \indent In Fig.~\ref{base12}(b) we show the differential conductance for Sample 2 at a temperature of 400mK. The differential conductance is suppressed around zero voltages, indicating Giaever tunneling and a lower transmission contact. We can conclude from Figs.~\ref{cartoon1} and \ref{base12} that In in intimate contact with the 2DEG forms a relatively low transmission contact, whereas In nearby but not directly in contact with the 2DEG forms a high transmission contact. The temperature dependence of the differential conductance for Samples 1 and 2 is also consistent with a ballistic contact and a low transmission contact, respectively. The differential conductance of Sample 1 is relatively constant with temperature, while the differential conductance of Sample 2 greatly decreases as the temperature is lowered (indicating thermionic emission). We have taken several NS contacts displaying Giaever tunneling at low temperature and annealed them for longer times, attempting to obtain a ballistic NS interface. In all cases further annealing simply makes the Giaever tunneling characteristic more pronounced, indicating that further annealing lowers the interface transmission. The room temperature conductance of the sample improves with further annealing, however, simply due to an increase of the effective contact area. Since the room temperature conduction mechanism through the contact of Sample 2 is thermionic emission, it simply scales with the contact area. Further annealing improves the room temperature conductance, but worsens the low temperature conductance. There is an optimal annealing time where In grows down to almost reach the 2DEG, but is not in physical contact with the 2DEG. Any further annealing after this point degrades interface transmission. \subsection{Effect of Weak Localization on Giaever Tunneling} \indent For an NS junction of length L, obeying the condition $l\ll L \leq l_{\phi}$, electrons which initially failed to Andreev refelct from the NS interface can backscatter again to the NS interface. Therefore, weak localization inside the normal conductor gives the electrons additional opportunities for Andreev reflection. The net effect of Giaever tunneling at the NS junction combined with weak localization inside the normal conductor is an enhancement of the total Andreev reflection probability at the Fermi level, leading to an additional conductance peak around zero bias voltage~\cite{kleins1,vanwees1}. Fig.~\ref{weakloc} shows the differential conductance for the NS junction annealed at 550$^{o}$C for 3 minutes, which we call Sample 3. Sample 3 displays the conductance peak around zero bias voltage first observed in Ref.~~\cite{kleins1} and explained in Ref.~\cite{vanwees1}. Disorder assisted backscattering can cause a zero bias conductance peak of magnitude up to 10\% of the background conductance value. In our case the zero bias peak is about a 1.5\% increase over the background conductance. The Giaever tunneling feature is also spread over a large voltage range larger than 1mV. Parasitic resistance from the 2DEG again explains the smaller zero bias conductance peak and the spreading out of the dI/dV versus V along the voltage axis. Since the height of the zero bias conductance peak saturates by 300 mK Fig.~\ref{weakloc}(a), the peak is not the precursor of a supercurrent between the two contacts. The supercurrent should be negligibly small in any case, since the sample satisfies $l\ll L \geq l_{\phi}$. We adequately filtered RF noise away from the devices to observe supercurrents in other superconductor - semiconductor samples with closer pad separation. We would also have observed such a supercurrent if it were present in this sample. Fig.~\ref{weakloc}(b) shows the magnetic field dependence of the conductance for Sample 3. For a junction of length L and width W, where $L,W > L_{\phi}$, the magnetic field required to destroy this zero bias conductance peak is of the order $B_{c}={\phi_{0}}/{{L_{\phi}}^2}$. In Sample 3 we observe $B_{c} \simeq$ 80 Gauss. The calculated field is $B_{c} \simeq$ 20 Gauss, nearly four times smaller than the observed value. It is possible that the $L_{\phi}$ is overestimated, i.e. it may be around $L_{\phi}\simeq$ 0.8 $\mu$m. Since these numbers are not precise data fits based on any quantitative theory, it is comforting that we obtain roughly the coherence length obtained by previous weak localization measurements on the 2DEG. The shift in the background conductance with the magnetic field in Fig.~\ref{weakloc}(b) is also due to the parasitic magnetoresistance of the 2DEG. \begin{figure} \centps{figs6a.eps}{60} \centps{figs6b.eps}{60} \caption{ Conductance vs the DC bias voltage for Sample 3 for (a) different temperatures and (b) different magnetic fields. A conductance peak develops near zero bias voltage, a correction to Giaever tunneling arising from weak localization inside the normal metal.} \label{weakloc} \end{figure} \subsection{Anamalous Weak Localization Corrections to Giaever Tunneling} \indent On most samples where we observed weak localization corrections to Giaever tunneling, we obtained conductance characteristics similar to those in Fig.~\ref{weakloc}. However, Fig.~\ref{anomloc} shows the conductance characteristics for an NS junction annealed at 500$^{o}$ C for 2 minutes which we call Sample 4. Overall Sample 4 displays a background of Giaever tunneling. A zero bias conductance peak (similar to the one in Fig.~\ref{weakloc}) continues to develop for temperatures down to about 800 mK in Fig.~\ref{anomloc}(a). At a temperature of 650 mK in Fig.~\ref{anomloc}(a), however, a conductance dip begins developing around zero bias. This dip in conductance around zero bias, which is superposed on the broader conductance peak, is nearly fully developed by 300 mK as shown in Fig.~\ref{anomloc}(a). There is little change in the differential conductance between 300 mK and 180 mK in Fig.~\ref{anomloc}(a). This anomalous dip feature superimposed on the weak localization correction to Giaever tunneling is reproducible on thermally cycling the NS junction back to room temperature and again down to mK temperatures. \begin{figure} \centps{figs7a.eps}{60} \centps{figs7b-2.eps}{60} \caption{ Conductance vs the DC bias voltage for Sample 4 for (a) different temperatures and (b) different magnetic fields. The additional conductance dip which develops around zero bias voltage can be explained by an inhomogeneous NS contact consisting of both high and low transmission regions. } \label{anomloc} \end{figure} Marmorkos, Beenakker, and Jalabert~\cite{marmorkos1} have numerically simulated the conductance of an NS junction in contact with a dirty normal metal. For low transmission interfaces they numerically observe, in Fig.~2 of Ref.~\cite{marmorkos1}, the zero bias conductance peak associated with the weak localization corrections to Giaever tunneling. However, for high transmission between the NS interface and normal conductor, the numerical simulations of Ref.~\cite{marmorkos1} reveal that the conductance peak changes into a conductance dip around zero bias. Ref.~\cite{marmorkos1} therefore shows that the same weak localization phenomena which causes a zero bias conductance peak in low transmission contacts causes a zero bias conductance dip for highly transmissive NS interfaces. The numerical simulation in Ref.~\cite{marmorkos1} offers one possible way to explain the conductance dip around zero bias we observe in Sample 4. The overall conductance of Sample 4 displays Giaever tunneling. Therefore, the majority of the NS interface area in Sample 4 has an additional tunneling barrier between the superconductor and 2DEG, namely the depletion region shown in Fig.~\ref{cartoon1}(b). However, the type of NS junctions we form by diffusion In into AlGaAs/GaAs are inhomogeneous enough that a significant fraction of the sample can form a transmissive NS interface of the type shown in Fig.~\ref{cartoon1}(a). The conductance we observe in Fig.~\ref{anomloc} will be a parallel combination of these two different types of NS junctions, as shown schematically in Fig.~\ref{cartoon2}. For voltages away from $V=0$, the slow variation of the background Giaever tunneling conductance dominates the dI/dV curve. For voltages very close to zero bias, the weak localization phenomena at the transmissive regions dominate and leads to the observed conductance peak and dip. \begin{figure} \centps{dI-dVguess.eps}{30} \setlength{\epsfysize}{7.0truecm} \caption{Possible explanation for conductance dip around zero voltage observed inside the zero bias conductance peak. Low transmission regions of the interface give both the overall Giaever tunneling shape of the dI/dV versus V and the zero bias conductance peak. A few high transmission regions of the contact could produce the zero bias conductance dip.} \label{cartoon2} \end{figure} The peak at finite bias in Figs.~\ref{anomloc} was first observed by Poirier et al.~\cite{poirier}, who called it the `finite bias anomaly'. One problem with using the simulations of Mormorkos et al.~\cite{} to explain a conductance dip around zero bias is that it requires a high interface transmission, whereas the data of Poirier et al.~\cite{poirier} (and our own data) show a Giaever tunneling background (low average interface transmission). The inherent inhomogeneity of supposedly planar superconducting In contacts to the 2DEG in AlGaAs/GaAs we have demonstrated in this paper overcomes this difficulty. A few high transmission point emitters can produce the conductance dip around zero bias, whereas the majority of the contact can maintain low overall interface transmissivity. The weak localization dip around zero bias can therefore peacefully coexist with a Giaever tunneling background conductance. The weak localization correction to the conductance of a ballistic NS junction could have been more clearly observed in Sample 1, were it actually present in that sample. Similarly Sample 2 (and several other samples we measured) did not exhibit the weak localization correction to the Giaever tunneling conductance. The exact impurity configuration near a particular NS interface will determine whether or not the weak localization correction to the conductance appears in any given sample. Perhaps it is therefore not surprising that the weak localization correction to the conductance can be observed only in a fraction of the samples. A different mechanism which splits the zero bias conductance peak in NI$_1$NI$_2$S junctions was developed in Ref.~\cite{lesovik}. Weak localization inside the middle N region produces the zero bias conductance peak. If the two insulators I$_1$ and I$_2$ have two different transmission coefficients, the zero bias conductance peak is split as shown in Fig.~3 of Ref.~\cite{lesovik}. These two barriers, having different transmissivity, is the same mechanism proposed by Poirier et al.~\cite{poirier} to account for the `finite bias anomaly'. Ref.~\cite{poirier} proposed a model which used the Schottky barrier at the NS interface to produce I$_2$, and an impurity inside the semiconductor as I$_1$. The spacing between I$_1$ and I$_2$ is L, a random number set by the impurity configuration. The McMillan-Rowell resonance nearest the Fermi level survives in the conductance of an NI$_1$NI$_2$S junction upon averaging over different L, producing a finite bias anomaly whose voltage is set by the average L. The composite `point emitter' model for the contact developed in this paper may also provide some support for this NI$_1$NI$_2$S model for the `finite bias anomaly'. An electron moving through the 2DEG past a point emitter would see that emitter as a scattering center, equivalent to an insulating barrier. The distance between the emitters in Sample 1 is of the order 100nm, less than the electron phase coherence length. In Sample 1, therefore, could be regarded as a type of NI$_1$NI$_2$NI$_3$~...~S junction. Each normal metal region N would also be weakly localized. This model may also produce a finite bias anomaly, but would require further numerical support. A two-dimensional numerical simulation, where the electrons could actually move around the point emitter scattering centers, would be required to confirm this picture. \section{Conclusions} \indent We have measured the differential conductance of superconductor-normal metal junctions formed by diffusing Indium into AlGaAs/GaAs heterostructures. In grows into a AlGaAs/GaAs heterostructure having a [100] oriented surface preferentially along the \{111\} crystallographic planes. Instead of a planar diffusion profile, we therefore find that In forms `inverted pyramids' or point contacts to the 2DEG. Supposedly `planar' superconducting In contacts to the electron gas in an AlGaAs/GaAs heterojunction are therefore actually composed of many point emitters. Correlating the contact microstructure observed on different samples with the differential conductance spectroscopy of the NS contact allowed us both to explain many observed features in the conductance and to determine the mechanism of superconducting (ohmic) contacts to the 2DEG in this materials system. For NS junctions annealed at a moderate temperature for a short times, so that the In point contacts do not physically touch the 2DEG, we obtain highly transmissive NS junctions. Due to the contact inhomogeneity, the point emitters nucleate and grow at different rates into the semiconductor. We observed wave interference between these different superconducting emitters in transmissive NS junctions. For identically prepared NS junctions annealed at higher temperatures and for longer times, so that the In point contacts grow together and have direct physical contact with the 2DEG, we obtain a lower transmission NS interface and Giaever tunneling. This is due to a depletion layer which forms around the In which directly touches the 2DEG. Further annealing simply increases the effective strength of the interface barrier between N and S, as regions of the In which previously were not in direct physical contact with the 2DEG come in contact with the 2DEG. Since the semiconductor forming N is disordered and has a reasonable phase coherence length, weak localization corrections to the differential conductance around zero bias voltage are also observed in this materials system. This zero bias conductance peak is a correction to Giaever tunneling which has been previously observed by several other groups~\cite{kleins1,vanwees1}. We observed an additional dip inside this zero bias conductance peak which develops in some samples at low temperature~\cite{poirier}. One possible explanation for the additional dip is due to contact inhomogeneity, where a small percentage of the contact is a nearly ballistic NS interface while most of the NS contact area remains in the tunneling limit. This conductance dip around zero bias voltage is therefore possible evidence for the predicted weak localization correction to the conductance of ballistic normal metal - superconductor junctions in Ref.~\cite{marmorkos1}. In any case, explanations for this `finite bias anomaly' should account for the actual non-planar physical structure of the superconducting contact. \section{Acknowledgments} \indent We gratefully acknowledge support from the David and Lucile Packard Foundation and from the MRSEC of the National Science Foundation under grant No. No. DMR-9400415. We thank Tamer Rizk, Richard Riedel, Manoj Samanta and Supriyo Datta for many useful discussions. \vspace{0.1in} $^1$ Present address: Intel Corporation, RN2-40, 2200 Mission College Blvd. Santa Clara, CA 95052. $^2$ Present address: Xilinx, 2100 Logic Dr., San Jose, CA 95124. $^3$ Present Address: Yale University, Department of Electrical Engineering, New Haven, CT 06520. $^4$ Present address: Dept. of Physics, University of North Florida, Jacksonville, FL 32224.
\section{Introduction} The long awaited recent report \cite{KTev} on a clear observation of direct CP violation in $K\to\pi\pi$ decays, ${\rm Re}(\epsilon'/\epsilon) = (28.0 \pm 3.0 \pm 2.6 \pm 1.0)\times 10^{-4}$, is the first evidence for the important role played by penguin amplitudes in the phenomena of CP violation \cite{Paschos}. $B$ decays are expected to provide a variety of CP asymmetry measurements, as well as measurerments of certain combinations of rates, some of which carry the promise of determining the angles of the unitarity triangle \cite{review}, $\alpha, \beta$ and $\gamma$. This can test the commonly accepted hypothesis that CP violation arises solely from phases in the Cabibbo-Kobayashi-Maskawa matrix \cite{KM}. Let us review \cite{talk} a few of the ideas involved in this study, paying particular attention to the role of penguin amplitudes. \begin{itemize} \item {\bf$\beta$}: In the experimentally feasible \cite{CDF} and theoretically pure example of $B^0(t)\to J/\psi K_S$ the decay amplitude is real to a very high precision. Theoretically \cite{SanBi}, the time-dependent mixing-induced CP asymmetry measures the phase $\beta\equiv -{\rm Arg}V_{td}$ controlling $B^0$-$\bar B^0$ mixing to an accuracy of 1$\%$ \cite{pen}. \item {\bf$\alpha$}:~ $B^0(t)\to \pi^+\pi^-$ involves direct CP violation from the interference between a dominant current-current amplitude carrying a weak phase $\gamma$ and a smaller penguin contribution, which ``pollutes" the measured $\sin\Delta mt$ term in the time-dependent asymmetry \cite{pen}. A ratio of penguin to tree amplitudes $|P/T|=0.3\pm 0.1$ in $B^0\to \pi^+\pi^-$ is inferred \cite{DGR} from the measured rates \cite{CLEO} of $B\to K\pi$ dominated by a penguin amplitude. Such a penguin contribution introduces a sizable uncertainty \cite{MG} in the determination of $\alpha=\pi-\beta-\gamma$ in $B^0\to\pi^+\pi^-$. Isospin symmetry may be used \cite{GL} to remove this unknown correction to $\alpha$ by measuring also the time-integrated rates of $B^{\pm}\to\pi^{\pm}\pi^0$ and $B^0(\bar B^0)\to \pi^0\pi^0$. In the likely case that the decay rate into $\pi^0\pi^0$ cannot be measured with sufficient precision, one can at least use this measurement to set upper limits on the error in $\alpha$ \cite{GQ}. Further out in the future, one may combine the time-dependence of $B^0(t)\to \pi^+\pi^-$ with the U-spin related $B_s(t)\to K^+K^-$ to determine separately $\beta$ and $\gamma$ \cite{Duni}. This involves uncertaities due to SU(3) breaking. \item {\bf$\gamma$}: The angle $\gamma$ is apparently the most difficult to measure. It was suggested some time ago \cite{GLR} to obtain information about this angle from charged $B$ decays to $K\pi$ final states by measuring the relative phase between a dominant real penguin amplitude and a smaller current-current amplitude carrying the phase $\gamma$. This is achieved by relating the latter amplitude through flavor SU(3) \cite{GHLR} to the amplitude of $B^+\to\pi^+\pi^0$, introducing SU(3) breaking in terms of $f_K/f_{\pi}$. \end{itemize} In the above two examples of determining $\alpha$ and $\gamma$, QCD penguin amplitudes were taken into account in terms of their very general properties, whereas electroweak penguin (EWP) contributions were first neglected and later on analyzed in a model-dependent manner \cite{ewp}. Such an approach relies on factorization and on form factor assumptions \cite{model}, and involves theoretical uncertainties in hadronic matrix elements similar to those plaguing $\epsilon'/\epsilon$ \cite{Paschos}. In the present report we will focus on recent developments in the study of EWP contributions, which partially avoid these uncertainties, thereby improving the potential accuracy of measuring $\alpha$ and $\gamma$. \section{Model-independent treatment of electroweak penguins} The weak Hamiltonian governing $B$ decays is given by \cite{Buras} \begin{equation}\label{H} {\cal H} = \frac{G_F}{\sqrt2} \sum_{q=d,s}\left(\sum_{q'=u,c} \lambda_{q'}^{(q)} [c_1 Q_1 + c_2 Q_2] - \lambda_t^{(q)}\sum_{i=3}^{10}c_i Q^{(q)}_i\right)~, \end{equation} where $Q_1=(\bar bq')_{V-A}(\bar q'q)_{V-A}, Q_2=(\bar bq)_{V-A} (\bar q'q')_{V-A},~\lambda_{q'}^{(q)}=V_{q'b}^*V_{q'q},~q=d,s,~q'=u,c,t, ~\lambda_u^{(q)}+\lambda_c^{(q)}+\lambda_t^{(q)}=0$. The dominant EWP operators $Q_9,~Q_{10}$ ($|c_{7,8}|\ll |c_{9,10}|$) have a (V-A)(V-A) chiral structure, similar to the current-current operators $Q_1, Q_2$. Thus, isospin alone relates the matrix elements of these operators in $B^+\to \pi^+\pi^0$ \cite{GPY} \begin{equation}\label{EWpi} \sqrt2 P^{EW}(B^+\to \pi^+\pi^0)=\frac32 \kappa(T+C)~,~~~~~\kappa=\frac{c_9 + c_{10}}{c_1 + c_2} = -0.0088~, \end{equation} where $T+C$ represents graphically \cite{GHLR} the current-current amplitudes dominating $B^+\to \pi^+\pi^0$. Similarly, flavor SU(3) implies \cite{GPY} \begin{equation}\label{EW1} P^{EW}(B^+\to K^0\pi^+) + \sqrt2 P^{EW}(B^+\to K^+\pi^0) = \frac32\kappa (T+C)~, \end{equation} \begin{equation}\label{EW2} P^{EW}(B^0\to K^+\pi^-) + P^{EW}(B^+\to K^0\pi^+) = \frac32 \kappa (C-E)~. \end{equation} In the next three sections we describe briefly applications of these three relations to the determination of $\alpha$ and $\gamma$ from $B\to \pi\pi$ and $B\to K\pi$, respectively. \section{Controlling EWP contributions in $B\to\pi\pi$} The time-dependent rate of $B^0\to \pi^+\pi^-$ includes a term $\sim\sin(2\alpha+ \theta)\sin(\Delta mt)$, where the correction $\theta$ is due to penguin amplitudes \cite{GL}. Using isospin (\ref{EWpi}), the EWP contribution to $\theta$, denoted by $\xi$, is found to be very small \cite{GPY,BF} \begin{equation} \tan\xi = \frac{x\sin\alpha}{1+x\cos\alpha},~~ x\equiv \frac32\kappa |\frac{\lambda_t^{(d)}}{\lambda_u^{(d)}}| = -0.013 |\frac{\lambda_t^{(d)}}{\lambda_u^{(d)}}|~, \end{equation} and is nicely incorporated into the analysis of Ref.~12 which determines $\alpha$. \section{$\gamma$ from $B^+\to K\pi$} Using (\ref{EW1}), EWP terms are included in the triangle construction of Ref.~15 \cite{NR2} \begin{equation} \sqrt2 A(B^+\to K^+\pi^0) + A(B^+\to K^0\pi^+) = \tilde r_u A(B^+\to\pi^+\pi^0) \left(1 - \delta_{EW} e^{-i\gamma}\right)~, \end{equation} where $\tilde r_u = (f_K/f_{\pi})\tan\theta_c\simeq 0.28,~ \delta_{EW}=-(3/2)|\lambda^{(s)}_t/\lambda^{(s)}_u|\kappa \simeq 0.66\pm 0.15$. This relation and its charge-conjugate permit a determination of $\gamma$ \cite{GLR,NR2} under the {\it assumption} that a rescattering amplitude with phase $\gamma$ can be neglected in $B^+\to K^0\pi^+$. This amplitude is bounded by the U-spin related rate of $B^{\pm}\to K^{\pm}\bar K^0$ \cite{Falk,GR,FL}. Present limits are at the level of $20-30\%$ of the dominant penguin amplitude \cite{GPY,Neubert}, and are expected to be improved to the level of 10$\%$. In this case the rescattering effect, which depends strongly on the final state phase difference $\phi$ between $I=3/2$ current-current and penguin amplitudes, introduces an uncertainty at a level of $15^{\circ}$ in the determination of $\gamma$ if $\phi$ is near $90^{\circ}$ \cite{GP}. A considerably smaller theoretical error \cite{Neubert} would be implied if this measurable phase is found to be far from $90^{\circ}$. Other sources of errors in $\gamma$, such as SU(3) breaking, are discussed elsewhere at this meeting \cite{Neubert,Flei}. We note that in this determination of $\gamma$ SU(3) breaking does not occur in the leading penguin amplitudes as it does in some other methods \cite{Duni}. The phase $\gamma$ can also be constrained by measuring only charge-averaged $B^{\pm}\to K\pi$ rates. Defining \begin{equation}\label{R*def} R^{-1}_*=\frac{2[B(B^+\to K^+\pi^0) + B( B^-\to K^-\pi^0)]} {B(B^+\to K^0\pi^+) + B(B^-\to \bar K^0\pi^-)}~, \end{equation} one finds using (\ref{EW1}) \cite{GPY,NR1} \begin{equation} R^{-1}_* = 1 - 2\epsilon \cos\phi (\cos\gamma - \delta_{EW}) + {\cal O}(\epsilon^2, \epsilon^2_A, \epsilon\epsilon_A)~, \end{equation} where \cite{GLR,NR1} $\epsilon = \tilde r_u \sqrt 2 |A(B^{\pm}\to\pi^{\pm}\pi^0)/A(B^{\pm}\to K^0\pi^{\pm})|\sim 0.24$, while $\epsilon_A$ is the suitably normalized rescattering amplitude. The resulting bound \begin{equation}\label{const} |\cos\gamma - \delta_{EW}| \ge \frac{|1-R^{-1}_*|}{2\epsilon}~, \end{equation} which neglects {\it second order} corrections, can be used to exclude an interesting region around $\cos\gamma = \delta_{EW}$ if $R^{-1}_*\ne 1$ is measured. Again, this would be very difficult if $\phi\simeq 90^{\circ}$. The present value of the ratio of rates is \cite{CLEO} $R^{-1}_*=2.1\pm 1.1$. \section{$\gamma$ from the ratio of $B^0\to K^{\pm}\pi^{\mp}$ to $B^{\pm}\to K^0\pi^{\pm}$ rates} Denoting this ratio of charged-averaged rates by $R$ ~\cite{FM}, one finds using (\ref{EW2}) a constraint very similar to (\ref{const}) \cite{GPY,BF,FL} \begin{equation}\label{const'} |\cos\gamma - \delta'_{EW}| \ge \frac{|1-R|}{2\epsilon'}~ \end{equation} where $\delta'_{EW}\sim 0.2\delta_{EW}\sim 0.13$ represents color-suppressed EWP contributions, and \cite{GR}~ $\epsilon'\sim 0.2$ is the ratio of tree to penguin amplitudes in $B^0\to K^+\pi^-$. In contrast to (\ref{const}), this bound neglects {\it first order} rescattering effects, and the values of $\delta'_{EW}$ and $\epsilon'$ are less solid than those of $\delta_{EW}$ and $\epsilon$ in (\ref{const}). Eq.~(\ref{const'}) can exclude a region around $\gamma=90^{\circ}$ if $R\ne 1$ is found. Presently \cite{CLEO} $R=1.07\pm 0.45$. \section{Conclusion} \begin{itemize} \item In $B\to\pi\pi$ strong and electroweak penguins are controlled by isospin. \item In $B\to K\pi$ strong penguins dominate and EWP are controlled by SU(3). \item Interesting bounds on $\gamma$, in one case susceptible to rescattering effects, are implied if the $B\to K\pi$ charge-averaged ratios of rates differ from 1. \item A precise determination of $\gamma$ from $B\to K\pi$ is challenging and requires a combined effort involving further theoretical and experimental studies. \end{itemize} \noindent {\bf Acknowledgment}: This work is supported by the United States $-$ Israel Binational Science Foundation under Research Grant Agreement 94-00253/3.
\section{Introduction} Over the last several years we have learned a great deal about supersymmetric gauge theories following the discovery of dualities between string/M-theory and supersymmetric gauge theories \cite{bfs97,sus97,mald97}. Recently this has been extended to conformal field theories without supersymmetry \cite{klt99}. Evidently, it would be desirable to have a deeper understanding of supersymmetry breaking in order to bridge the gap between the formulation of physics in a supersymmetric world, and its more realistic counterpart, where no such symmetry is manifest. One straightforward approach is to start with a supersymmetric formulation, and then proceed to break supersymmetry `softly' by adding appropriate mass terms. The context within which we will consider this is a theory that has been well studied before: two dimensional SU($N$) gauge theory coupled to an adjoint Majorana fermion \cite{bdk93}. Interestingly, this theory is known to exhibit supersymmetry at a particular value of the fermion mass, $m=m_{SUSY}$ \cite{kut93}. This is believed to be a theory with two parameters $g$ and $m$, both of which have the dimensions of mass. Since the only $g$ dependence is an overall $g^2$ factor in the Hamiltonian the theory depends on one dimensionless parameter $X={m^2 \pi \over g^2 N}$ and therefore all the bound state masses, in units of ${g^2 N \over \pi}$, must be determined in terms of the one parameter $X$. In this work, we provide evidence that, while this viewpoint is still correct, there is still scope for an additional operator (and associated coupling constant) that may be introduced to improve convergence of the DLCQ bound state masses towards their actual continuum values. Of course, these continuum masses will be unaffected by the presence of such an operator, but a judicious choice of coupling will serve to improve the rate of convergence of our numerical results. Naively, when one adds a `soft' breaking term to the two DLCQ formulations of the theory, we appear to arrive at different spectra. The two formulations we are alluding to are the `Principle Value' (PV) and `Supersymmetric Discrete light Cone Quantization' (SDLCQ), and are discussed in detail below. The question is whether these two spectra are truly different or simply rescalings of the same spectrum which become identical in the continuum limit. It would be very good news if they were in fact the same because the PV prescription is generally accepted as correct (\cite{pab85,bpp98}), while the SDLCQ approach is known to converge more rapidly in general. Actually, to understand the relation between these two schemes, it is helpful to present a formulation that {\em interpolates} between the PV and SDLCQ prescriptions by introducing an additional operator \cite{alp99} and associated coupling constant that we will call $Y$. In particular, $Y=0$ will correspond to the PV prescription, while $Y=1$ will imply the SDLCQ prescription. Intermediate values for $Y$ will correspond to a `mixture' of the two schemes. By diagonalizing the DLCQ Hamiltonian matrix, and extrapolating to the continuum limit, we are able to solve for bound state masses and wave functions at different values of the fermion mass parameter $X$ and coupling constant $Y$. We shall show that the continuum bound state masses are independent of the coupling $Y$, as expected from a scheme independent prescription, although the {\em rate of convergence} towards the actual continuum mass will be significantly affected by our choice for $Y$. In fact, it will turn out that the value for $Y$ that arises naturally in the regularization of supersymmetric theories (i.e. $Y=1$) provides the best convergence towards actual continuum masses. Thus, the supersymmetric formulation of DLCQ (SDLCQ), corresponding to $Y=1$ -- first highlighted in the work \cite{mss95} -- yields a method for improving numerical convergence of DLCQ bound state masses even for theories {\em without supersymmetry}. To show this, we study the DLCQ bound state integral equations at high resolution, which is made possible by truncating the Fock space to two particles. In particular, we show that the SDLCQ approach converges more uniformly and rapidly for all values of $X$ that are sufficiently far from the critical value $X=0$. We remark that at the supersymmetric point $X=1$, the SDLCQ prescription preserves supersymmetry even in the discretized theory. The advantages of such an approach have been exploited in a study of a wide class of supersymmetric gauge theories in two \cite{alp98a,alp98b,alpp98,alpp99,alp99} and three dimensions \cite{alp99b}. \section{\bf Formulations of The Theories} In this section we will consider the formulations of $1+1$ dimensional QCD coupled to adjoint Majorana fermions having arbitrary mass (see for example \cite{bdk93}) in the light cone gauge $A^+=0$. After eliminating non-physical degrees of freedom by solving constraint equations, the light--cone components of total momentum are found to be: \begin{eqnarray} \label{momenta} P^+&=&\int dx^- Tr(i\psi\partial_-\psi),\\ P^-&=&\int dx^-Tr\left(-\frac{im^2}{2}\psi\frac{1}{\partial_-}\psi- \frac{g^2}{2}J^+\frac{1}{\partial_-^2}J^+\right) \nonumber \\ &=&\frac{m^2}{2}\int_0^\infty\frac{dk}{k}b^\dagger_{ij}(k)b_{ij}(k)+ \frac{g^2N}{\pi}\int_0^\infty\frac{dk}{k}\int_0^k dp \frac{k}{(p-k)^2} b^\dagger_{ij}(k)b_{ij}(k) +\nonumber\\ &&\frac{g^2}{2\pi}\int_0^\infty dk_1dk_2dk_3dk_4\left(\delta(k_1+k_2-k_3-k_4) A(k)b^\dagger_{kj}(k_3)b^\dagger_{ji}(k_4)b_{kl}(k_1)b_{li}(k_2)+\right.\\ &&\left.\delta(k_1+k_2+k_3-k_4)B(k)(b^\dagger_{kj}(k_4)b_{kl}(k_1)b_{li}(k_2)b_{ ij}(k_4)- b^\dagger_{kj}(k_1)b^\dagger_{jl}(k_2)b^\dagger_{li}(k_3)b_{ki}(k_4)) \right)\nonumber \end{eqnarray} with \begin{eqnarray} A(k)=\frac{1}{(k_4-k_2)^2}-\frac{1}{(k_1+k_2)^2},\nonumber\\ B(k)=\frac{1}{(k_3+k_2)^2}-\frac{1}{(k_1+k_2)^2}. \end{eqnarray} Here $x^\pm =(x^+ \pm x^-)/\sqrt2 $ and $J^+_{ij}=2\psi_{ik}\psi_{kj}$ is the longitudinal component of the fermion current. To avoid introducing an additional mass scale in the theory we will write this in term of mass operators: $M^2=2P^+P^-$. It is well known that at the special value of fermionic mass (namely $m_{SUSY}^2=g^2N/\pi$) this system is supersymmetric \cite{kut93}. We will use a dimensionless mass parameter $X=\frac{\pi m^2}{g^2N}$, and the supersymmetric point is $X=1$ and the masses of all bound states will be quoted in units of $g^2 N/\pi$.The supercharge is given by \begin{eqnarray} \label{sucharge} Q^-&=&2^{1/4}\int dx^-tr(2\psi\psi\frac{1}{\partial_-}\psi). \nonumber \\ &=&\frac{i2^{-1/4}g}{\sqrt{\pi}}\int_0^\infty dk_1dk_2dk_3 \delta(k_1+k_2-k_3) \left(\frac{1}{k_1}+\frac{1}{k_2}-\frac{1}{k_3}\right)\times\nonumber\\ &\times&\left(b^\dagger_{ik}(k_1)b^\dagger_{kj}(k_2)b_{ij}(k_3)+ b^\dagger_{ij}(k_3)b_{ik}(k_1)b_{kj}(k_2)\right), \end{eqnarray} \noindent Using the anticommutator at equal $x^+$: \begin{equation} \{\psi_{ij}(x^-),\psi_{kl}(y^-)\}=\frac{1}{2}\delta(x^- -y^-) \end{equation} it can be checked that at $m=m_{SUSY}$ the SUSY algebra $\{Q^-,Q^-\}=2\sqrt{2}P^-$ is satisfied. In the DLCQ approximation the system lives in a $x^-$ box of length $L$ and one has to sums over discrete variables $k^+ \ne 0$ instead of integrations in the above formulas. For periodic boundary conditions (BC), $k^+= n\pi/L $ where $n=1,2,\dots, K$ and $K$ is called the resolution. One formulation of DLCQ which we will denote as the principal value (PV) prescription \cite{tho74}, treats the singularities of the Hamiltonian using a PV prescription and can be formulated using either anti- periodic or periodic BC. The anti-periodic boundary condition must break the supersymmetry at finite resolution because the fermions and bosons are in different Fock sectors. The PV prescription with periodic BC could in principle give supersymmetric results at finite resolution, although this is not the case. In the PV prescription the supersymmetry at $X=1$ is restored only in the decompactification limit ($K\rightarrow\infty$). This restoration was shown in \cite{bdk93} \footnote{They find that convergence is slower for period BC}. The Hamiltonian for this formulation will be referred to as $P^-_{PV}$. The prescription that preserves supersymmetry at finite resolution will be called SDLCQ. In SDLCQ one simply uses DLCQ to calculate the supercharges and then uses the super charges to calculate the Hamiltonian and longitudinal momentum operator \cite{mss95}. Here we must use periodic BC because the supercharge $Q^-$ is cubic in the fields, while the supercharge $Q^+$ is quadratic. The SUSY algebra is reproduced at a special value of fermion mass and at every finite resolution the supercharge matrices give a representation of the super algebra. Both SDLCQ and $P^-_{PV}$ at $X=1$ give the same results as the resolution goes to infinity \cite{alp98b} \footnote{However SDLCQ converges much faster}. We now want to add identical `soft' SUSY breaking terms (mass terms) to these theories and study the resulting non-supersymmetric theory. Since we already have a mass term in $P^-_{PV}$ this only requires varying $X$, but for SDLCQ this means explicitly adding a mass term. It is very instructive to actually do the numerical calculation differently and introduce a third formulation, $P_{SUSY}$ which includes both SDLCQ and $P^-_{PV}$ . We have found the operator which is the difference between the SDLCQ and the PV formulation \cite{alp99} \footnote{To date we have only found this operator for this particularly simple theory but it should be possible to find it for other theories as well. The calculation of this operator involves a careful study of the intermediate zero modes that contribute to the square of the supercharge}. Thus if we add this operator to the PV Hamiltonian it is now supersymmetric at every resolution and produces exactly the same mass and wave functions as SDLCQ. In the large $N$ approximation the operator take the form. \begin{equation} \label{operator} \frac{g^2NK}{\pi}\sum_n \frac{1}{n^2} B^\dagger_{ij}(n)B_{ij}(n). \end{equation} Numerically, this operator does not alter the actual continuum values observed in the PV approach when $X=1$. In our numerical formulation of $P^-_{SUSY}$ we included this operator with an adjustable coupling $Y$. We can now think of $P^-_{SUSY}$ as a single theory in the coupling constant space $(X,Y)$. The formulation we called PV corresponds to setting $Y=0$ and allowing $X$ to be arbitrary, while the prescription we call SDLCQ corresponds to setting $Y=1$. In the following, we will present results for the lightest bosonic bound states as a function of $X$ and $Y$. For a few values of $X$ and $Y$ we will truncate the Fock space to allow only two particles Fock states, which will permit us to investigate the t'Hooft equation for higher resolutions than would otherwise be possible. \section{\bf `Soft' SUSY Breaking} Our investigation of this theory indicates that at $X=1$ ( the supersymmetric value of the fermion mass) the lightest fermionic and bosonic bound states are degenerate with continuum masses approximately $M^2= 26$ \cite{bdk93,alp98b}. Using $P^-_{SUSY}$ we arrive at the same conclusion for any value of $Y$. Boorstein and Kutasov \cite{kub94} have investigated `soft' supersymmetry breaking for small values of this difference, $X-1$ and they found that the degeneracy between the fermion and boson bound state masses is broken according to \begin{equation} M^2_F(X) - M^2_B(X)= (1-X) M_B(1)+O((X-1)^3). \label{linear} \end{equation} They calculated these masses using the PV prescription ($Y=0$) with anti-periodic BC and found very good agreement with the theoretical prediction. We have compared this theoretical prediction at $Y=1$ and we find that eq (\ref{linear}) is very well satisfied. At resolution $K=5$, for example, the slope is 4.76 and the predicted slope $M_B(1)$ is 4.76. The indication is that this result is true for any value of $Y$. \begin{figure}[h] \begin{center} \epsfig{file=twoplots.eps, width=14cm} \end{center} \caption{(a) The contour plots of $Y=Y(X)$ for the mass squared of the lowest bound state in units of $g^2 N/ \pi$ as a function of $X=m \pi /g^2 N$ and Y (b)The contour plots of $Y=Y(X)$ for the mass squared of the second lowest bound state in units of $g^2 N/ \pi$ as a function of $X=m \pi /g^2 N$ and Y (b) } \end{figure} In Fig. 1 we show the contour plots of the mass squared $M^2$ of the two lightest bosonic bound states as a function of $X$ and $Y$ at resolution $K=10$. These contours are lines of constant mass squared. Selecting a particular value of the mass of the first bound state then fixes a particular contour in Fig. 1a as a contour of fixed mass, which we can write as $Y=Y_p(X)$. Interestingly, constructing the same contour plot for the next to lightest bosonic bound state -- see Fig. 1b -- yields contours that have approximately the same functional dependence implied by Fig. 1a. In fact, one obtains approximately the same contour plots for the next twenty bound states (which is as far as we checked). The simple conclusion is that the coupling $Y$ which represents the strength of the additional operator affects all bound state masses more or less equally. This in turn suggests that at finite resolution, we can smoothly interpolate between different values of fermion mass $X$, and different prescriptions specified by the coupling $Y$, without affecting too much the actual numerical spectrum. Of course, in the decompactification limit $K \rightarrow \infty$, such a dependence on $Y$ disappears, due to scheme independence. Since the lightest bosonic bound state is primarily a two particle state it is reasonable to truncate the Fock basis to two particle states. This will permit very high resolutions, which will be needed to carefully scrutinize any possible discrepancies between the two versions of 'soft' symmetry breaking presented here. In fact, we are able to study the theory for $K$ up to 800. The mass of the lowest state as a function of the resolution for various values of $X$ and $Y$ are shown in Fig. 2. Each converging pair of lines -- which extrapolate the actual data points -- in Fig. 2 corresponds to different values of fermion mass $X$. The top upper curve in each pair runs through data points that were calculated via SDLCQ (i.e. $Y=1$), while the lower corresponds to the PV (i.e. $Y=0$) prescription commonly adopted in the literature. We find that each pair of curves converge to the same point at infinite resolution, although this may not be completely obvious for the lowest pair in the figure (corresponding to the critical mass $X=0$). Away from $X=0$, the SDLCQ formulation is fitted with a linear function of $1/K$, while the PV formulation is fit with a polynomial of $1/K^{2\beta}$, where $\beta$ is the solution of $1-X/2=\pi \beta Cot(\pi \beta)$ \cite{van96}. It now appears that SDLCQ not only provides more rapid convergence for supersymmetric models, but also for the massive t'Hooft model, which is not supersymmetric. For the massless case, the situation is reversed; the SDLCQ formulation converges slower. It is fit by a polynomial in $1/Log(K)$ and gives the same mass at infinite resolution as the PV formulation. This behavior may be understood from the observation that the wave function of this state does not vanish at $x=0$. We have looked closely at `small' masses, such as $X=.1$, and one finds that both PV and SDLCQ vary as a polynomial in $1/K^{2\beta}$ at large resolution. Thus careful extrapolation schemes must be adopted at small masses. We therefore conclude that the continuum of regularization schemes that interpolate smoothly between the SDLCQ and PV prescriptions -- which we characterized by the parameter $Y$ -- yield the same continuum bound state masses, although the rate of convergence of the DLCQ spectrum may be altered significantly. This implies that the contour plots observed in Fig. 1 eventually approach lines parallel to the $Y$ axis, and the sole dependence on the parameter $X$ is recovered. \begin{figure}[h] \begin{center} \epsfig{file=convergenceplot.eps} \end{center} \caption{Mass of the of the lowest bound state in units of $g^2 N/ \pi$ calculated in the t'Hooft model. The top pair is at $X=1$, the second is at $X=.5$, and the bottom pair is at $X=0$ } \end{figure} \section{\bf Discussion} The two dimensional gauge theory of adjoint Majorana fermions has been studied extensively \cite{bdk93,kut93,alp98b,kub94,anp98} and is known to be a theory with an overall mass scale $g^2$, and one real coupling -- the mass of the fermion -- which we write as $X$ in our notation. When one adds a `soft' supersymmetry breaking term, the supersymmetric (SDLCQ) and principle value (PV) prescriptions for regulating the Coulomb singularity appear to give different bound state masses at finite resolution. We observed at finite resolution that these different bound state masses may be smoothly connected -- in an approximate sense -- by introducing a new operator, and an associated coupling $Y$, and then varying the couplings $X$ and $Y$ along an appropriately chosen contour. By truncating the Fock space to two particles, we were able to study the DLCQ bound state equations up to $K=800$, which we summarized in Fig. 2. We concluded that after carefully extrapolating the data, the different prescriptions yielded identical continuum bound state masses. Moreover, we observed that the SDLCQ prescription improved convergence for sufficiently large values of fermion mass. Interestingly, since the two-body equation studied here for the adjoint fermion model is simply the t'Hooft equation with a rescaling of coupling constant, we have arrived at an alternative prescription for regulating the Coulomb singularity in the massive t'Hooft model that improves the rate of convergence towards the actual continuum mass. Thus, a prescription that arises naturally in the study of supersymmetric theories is also applicable in the study of a theory without supersymmetry. We believe that this idea deserves to be exploited further in a wider context of theories. In particular, it is an open question whether this procedure could provide a sensible approach to regularizing softly broken gauge theories with bosonic degrees of freedom, and in higher dimensions. In any case, it appears that the special cancellations afforded by supersymmetry -- especially in the context of DLCQ bound state calculations -- might have scope beyond the domain of supersymmetric field theory. This would be a crucial first step towards a serious non-perturbative study of theories with broken supersymmetry. {\bf Acknowledgments} \noindent The work was supported in part by a grant from the US Department of Energy. The authors are grateful for useful discussion with Brett Van de Sande.
\section{Introduction} Foster and Michael \cite{FM} have studied numerically in lattice gauge models an interesting system consisting of a spinless static color octet source at the origin which modifies the gluonic field of pure color gauge theory in its vicinity. They call this system a ``gluelump'' and we follow their nomenclature. In \cite{FM} and several earlier publications referred to in \cite{FM} they have discussed the spectrum of this system. It is of interest to use this system as a testing ground for models of low energy QCD. Two models which have been much discussed in the literature are the bag model and the flux tube model and both may be applied to this system. We shall restrict ourselves here to the bag model. This would appear to be a particularly simple system to analyse in the bag model since the bag has a definite centre unlike the case of light quarks or gluons confined in a bag. \\There are earlier discussions of the system in the bag model \cite{CS} \cite{Juge} with emphasis on the ground state configuration. We extend the earlier work to higher excited states and make a comparison with the lattice results. Physically the system would be of interest if heavy gluinos were to be produced experimentally and last long enough for them to have ther color neutralized. This was the motivation of the study by Chanowitz and Sharpe \cite{CS} who were interested in the ground state configuration of the system (which they called the ``glueballino''). Another physical connection is to the states of hybrid mesons in the limit of short distances between the quark and antiquark. Juge, Kuti and Morningstar \cite{Juge} have studied static hybrid potentials both on the lattice and in bag models and have shown that the bag gives a good representation of the lattice results at small quark-antiquark separation. The present discussion elaborates their work, as well as earlier work by Hasenfratz et al \cite{Has} and Ono \cite{Ono} at the special point $r=0$ and for very small values of $r$. We confirm in the bag model the results of Foster and Michael \cite{FM} for the quantum numbers of the low lying states, with plausible ordering $ J^{PC}=1^{+-}$,$1^{--}$, $2^{--}$,$3^{+-}$, $2^{+-}$. \\The system we study has similarities to a glueball, which consists entirely of gluons. In the case of the {\it gluelump} one of the gluons is very heavy, and is located at the origin. The analogous heavy meson state, $Q{\bar{q}}$ has also been studied in the bag model \cite{Shuryak} and it was noted that the problem of centre of mass motion (and spurious excited states) disappears. \section{Bag Model Calculation} In the bag model, the system consists (in SU(3) of color) of an octet color charge fixed at the origin and one or several cavity ``gluons'' neutralizing the color charge in a spherical cavity of radius $R$. The one gluon states have lowest energy and we shall study them first, but will later estimate roughly the energies of the lowest two-gluon states. With a single gluon the energy of the system will have three contributions: a volume term $4\pi R^3 \Lambda/3$ where $\Lambda$ is the bag constant, the energy of the gluon mode, $k$, and $V^C$, the Coulomb interaction energy between the gluon and the octet charge at the origin. Initially we neglect $V_C$ so that the gluon in the bag is described by a free field. \\The dynamics of a massless vector field confined to a cavity is well known \cite{Close} and the boundary conditions for the color electric ${\bf E}$ and magnetic ${\bf B}$ fields are ${\bf E \cdot r} = 0$ and ${\bf B \times r = 0}$ at $r = R$. The fields ${\bf E}$ and ${\bf B}$ are solutions of the (vector) wave equation which may be represented as ${\bf L}Y_{lm}j_l$ with $Y_{lm}$ spherical harmonics, $j_l$ spherical Bessel functions and ${\bf L}= {\bf r \times \nabla}$. There are two sets of solutions of the vector wave equation, one in which ${\bf E} \propto {\bf L}Y_{lm}j_l$ and ${\bf B \propto \nabla \times} {\bf L}Y_{lm}j_l$, called TE modes (because ${\bf E \cdot r} = 0$ automatically) and the other set, the TM modes, in which ${\bf E}$ and ${\bf B}$ are interchanged. From these forms one may derive the equations which determine $kR$ through the boundary conditions. For the TE modes $kR$ is a solution of the equation ${\frac{d}{dr}(rj_l(kr)) =0}$ at $r=R$, while for the TM modes $kR$ is a solution of $j_l(kr) = 0$ at $r = R$. The lowest roots of these equations are as follows:\\ TE1: $\:kR= 2.744,\, 6.117,... $ for $J^P=1^+$\\ TE2: $\:kR= 3.870,\, 7.443,... $ for $J^P=2^-$\\ TE3: $\:kR= 4.973,\, 8.772, ... $ for $J^P=3^+$\\ TM1: $\:kR= 4.493,\, 7.72\,,... $ for $J^P=1-$\\ TM2: $\:kR = 5.763,\, 9.09\,,... $ for $J^P=2^+$\\ Since they are all single gluon modes they have $C=-1$.\\ Therefore the five lowest modes, in increasing order of $kR$, are \[ J^P = 1^+ (kR=2.744),\: 2^- (kR =3.870),\: 1^- (kR=4.493),\: 3^+ (kR= 4.973),\: 2^+ (kR=5.763) \] We note that the first three states ($1^+, 2^-, 1^-$) are the same as found on the lattice \cite{FM} but the order of the $2^-$ and $1^-$ states is reversed. However we have still to consider the color Coulomb interaction with the octet charge at the origin, and we shall see that this interaction may easily reverse the order of the $2^-$ and $1^-$ states, to bring agreement with the lattice results. We treat the color Coulomb interaction by perturbation theory. This is justified, for our parameter choice, by the small ratio of the energy shift to the spacing between levels of the same quantum numbers. (\cite{Shuryak})\\ The color Coulomb interaction between the central charge and the gluon field is a consequence of the non-Abelian nature of gluons. It is easy to compute it incorrectly! For example, if we were to take the volume integral of ${\bf E_1 \cdot E_2}$ where ${\bf E_1}$ is the field of the point color charge and ${\bf E_2}$ is that of the gluon, we would get zero since since ${\bf E_1}$ is longitudinal while ${\bf E_2}$ is transverse, so that ${\bf E_1 \cdot E_2}$ vanishes. We first obtain the color charge density of the gluon field which is proportional to ${\bf E_2 \cdot E_2}$, average over the polar angles $\theta, \phi$ and integrate with $(Q/r)$ where $Q$ is the central charge. An earlier calculation of this interaction is reported by Ono \cite{Ono} who uses for the gluon the approximation of uniform color charge density inside the spherical cavity. In Ono's approximation all states shift by the same amount, so that we must go beyond this approximation to find the relative shift of levels.\\ To compute the (color) Coulomb interaction between the central octet charge and the gluon we use a ``confined'' Coulomb potential which obeys the boundary condition that the potential vanishes at the boundary of the bag, $V(R)=0$; so the potential of a central Abelian charge $q$ is $\frac{q}{4\pi}(r^{-1}-R^{-1})$. There is an additional factor of $-3$ associated with the color $SU(3)$. For consistency we also include the change in self-energy, due to the boundary conditions, of the central charge which also depends on $R$, and the self-energy of the gluon which is estimated in the approximation of averaging over angles. The constant term in the potential does not contribute to the final answer. We shall illustrate the computation for the case of the lowest TE1 mode in which ${\bf E_{10}} = c{\bf L}Y_{10} j_1(kr)$ where $c$ is a normalization constant and we have chosen $(l,m) = (1,0)$ to simplify the calculation. We have immediately that $[{\bf E_{10}}]_z =0$ and, from angular momentum algebra, \[ [{\bf E_{10}}]_x =-ic\frac{y}{r}\sqrt{\frac{3}{4\pi}}j_1(kr),\:\: [{\bf E_{10}}]_y =+ic\frac{x}{r}\sqrt{\frac{3}{4\pi}}j_1(kr). \] Therefore ${\bf E_{10}^2} = c^2(1-\cos^2 \theta)\frac{3}{4\pi} j_1^2(kr)$ and integrating over all angles one obtains $<{\bf E_1}^2>$ proportional simply to $j_1(kr)^2$, leading to the following expression for the color Coulomb interaction energy: \[ V^C = -3 \alpha_s \frac{kR}{R} \frac{\int_0^{kR}x^2 (1/x - 1/(kR)) dx j_1(x)^2}{\int_0^{kR}x^2 dx j_1(x)^2} \] where the factor $-3 = <{\bf F_1 \cdot F_2>_1}$ is the coupling of the color octets into a singlet and $\alpha_s$ is the color fine structure constant. The factor $kR$ in this case is 2.744, but the formula also applies to TE2, TE3 modes where we replace $j_1$ by $j_2$ or $j_3$ and use the apropriate value of $kR$. Incorporating also the self-energies detailed above, this leads to the following estimate for the first three TE modes: \[ V^C(TE1) = -2.36 (-1.133) \alpha_s/R, \] \[ V^C(TE2) = -2.13 (-0.837) \alpha_s/R,\] \[V^C(TE3) = -2.01 (-0.680) \alpha_s/R \] where the bracketed numbers correspond to neglecting the self energies. These coefficients were obtained by numerical integration of the spherical Bessel functions in the expression above. They diminish with the angular momentum $l$ as the color charge density shifts further from the origin with increasing $l$.\\ For the TM modes, where the color electric field is proportional to ${\bf \nabla \times} {\bf L }Y_{lm}j_l$, the computation is a little more tedious, and we find for the TM1 mode \[ <{\bf E_1}^2> = c^2\left(\frac{3j_1^2}{r^2} + {{j_1^\prime}}^2 + \frac{2j_1\prime{j_1}}{r}\right) \] and a similar expression with different coefficents for the TM2 mode, from which we compute, including self-energies (excluding self-energies in brackets): \[ V_C(TM1) = -4.74 (-4.04)\frac{\alpha_s}{R},\:\: V_c(TM2) = -3.32(-2.40)\frac{\alpha_s}{R}. \] The Coulomb interactions are larger than for the TE modes because the charge density in the case of the TM modes is closer to the origin. \\We now have the ingredients to compute the energy of the bag by finding the extremal value of R. In this way one obtains \[ E = \frac{4}{3} (4\pi \Lambda)^{\frac{1}{4}} (\alpha_{nj} - \kappa \alpha_s)^{\frac{3}{4}} = .79 (\alpha_{nj} - \kappa \alpha_{s})^{\frac{3}{4}} {\rm GeV} \] where $\Lambda$ is the bag constant, $\alpha_{nj} =(kR)_{nj}$, $\kappa$ is the coefficent of $\alpha_s$ in the color Coulomb interaction. (For example, for the TE1 mode, $\alpha_{nj}=2.744$ and $\kappa =+2.36$). We have followed Juge et al \cite{Juge} and taken ${\Lambda}^\frac{1}{4} =.315 {\rm GeV}$, $\alpha_s = .23$, and so obtain: \[ E(1+) = 1.43 {\rm GeV}, E(2-) = 1.97 {\rm GeV}, E(3+) = 2.44 {\rm GeV}\] \[ E(1-) = 1.98 {\rm GeV}, E(2+) = 2.64 {\rm GeV},\] We note that the states of $J^P = 2^-, 1^-$ are now essentially degenerate for this value of $\alpha_s$. For any larger values of $\alpha_s$ their order is reversed and the order of the first three states is now as found on the lattice \cite{FM}.\\ Before working out further one gluon states one should consider the position of two gluon states, which can also neutralize the central octet charge. The energy of a two gluon state has contributions from the energies of each of the gluons, their color electric interaction with the central charge, their mutual color electric interaction, and a color magnetic interaction between them. There are also the source and gluon self-energies as in the case of the one gluon states. The lowest such state is composed of two TE1 gluons, since TE1 is the lowest single gluon state. The only change which occurs relative to the case of the one gluon state comes from the fact that the interaction between two octet states (either gluon with central charge or between the two gluons, has a different color factor $<{\bf F_1 \cdot F_2}>_{\bf 8} = -3/2$, half the size of the factor $<{\bf F_1 \cdot F_2}>_{\bf 1} = -3$, since the two octets are combined in an octet rather than a singlet. The interaction of each gluon with the central charge is the same as in the single gluon case apart from this factor of $\frac{1}{2}$ but there are two gluons. The interaction energy between the two gluons is a little cumbersome to estimate, so we shall here make the approximation of assuming a uniform color charge distribution for each of the gluons for this evaluation. In this approximation the color Coulomb interaction of the two gluons with the central charge is just the same as that of the single gluon previously ($-1.13 \alpha_s/R$) and the interaction between the two gluons is \[ \frac{1}{5}\frac{\alpha_s}{R}\left(-\frac{3}{2}\right)= -.3 \,\alpha_s/R \] Combining these results and the relevant self-energies, we estimate the two gluon states to have an energy of $.79 \,{\rm GeV}\: (5.5-2.4\alpha_s)^\frac{3}{4}$ $\simeq2.6 \,{\rm GeV}$. This estimate neglects the color magnetic interaction which will split the the various $J^P$ states from each other. With two $J^P=1+$ states one expects $0^+$, $1^+$ and $2^+$ states. Presumably the $0+$ state lies lowest with the $1+$ and $2+$ higher as with glueballs, though the glueball spectrum in the bag model does not correlate too well with that determined on the lattice. Therefore, one finds that the lowest states of the gluelump spectrum are $1^{+-}, 1^{--}, 2^{--}, 3^{+-}, 0^{+ +},2^{+-}$ with the relative positions of the $1^{--}$ and $2^{--}$ states and of the $0^{++}$ and $2^{+-}$ state not well determined by our estimates because of the uncertainty in $\alpha_s$.\\ \section{Hybrid Potentials At Small Quark-Antiquark Separation} We next consider the effect of replacing the central octet charge by a pair of static color triplet and anti-triplet sources (in overall octet state) with the distance between the two sources being very small compared to the bag radius $R$. This system is equivalent to the addition of a small color octet dipole moment on top of the color octet charge. It is easy to see that the $J=1$ states will split into a ``molecular" $\Sigma$ state (coming form the $m=0$ component) and a molecular $\Pi$ doublet (coming from the $m=\pm1$ components) where $m$ is defined relative to a $z$-axis along the direction of the dipole. We argue below that the energy of these ``molecular" states varies quadratically as a function of the distance $\cal R$ between the triplet and anti-triplet with different coefficents for $\Sigma$ and $\Pi$ making the $\Pi$ doublet lower. It should be noted that we are neglecting at this point the color electrostatic interaction between the the triplet anti-triplet pair (which goes like $\alpha_s/{6\cal R}$) but this is the same in both $\Pi$ and $\Sigma$. \\The action of the the dipole operator $\mu_z^{el}$ on the gluon wave function is equivalent to multiplication by the $z$-coordinate, where the $z$-axis is the axis of the dipole. This changes the multipolarity of the the gluon state, and in particular the parity of the gluon state flips. It is clear that all diagonal matrix elements vanish because of parity. Therefore the first non-zero contributions to the energy are in second order in the dipole operator, and therefore proportional to ${\cal R}^2$. We can evaluate approximately the coefficent of this quadratic term by using a single average energy denominator $\Delta E$, which we remove from the sum and use closure. In this way we obtain: \[ E({\cal R}) = E(0) + \frac{9}{4} \alpha_s^2 <\frac{\cos^2 \theta}{r^4}>\frac{1}{\Delta E} {\cal R}^2 \] where the expectation value is over the lowest state TE1,$m$ in the case of the ground state. One obtains two different values for $m=0$ and $m=\pm 1$. For the angular integrals we obtain $(2/5)$ for the $\Pi$ states ($m=\pm1$) and $1/5$ for the $\Sigma$ state ($m=0$). Recalling that for the ground state the energy denominator $\Delta E$ is negative, we find that the lower molecular state is the $\Pi$ state and the next excitation is the $\Sigma$, in agreement with the lattice computations of \cite{Juge}. Evaluating numerically the integrals over spherical Bessel functions we estimate the splitting between the $\Sigma$ and $\Pi$ states to be about $0.2\left(\frac{\cal R}{R_{\rm bag}}\right)^2$ in GeV, which is similar to the lattice estimate at short distances. \cite{Morning} \section{Concluding Remarks} Our bag model results for the order of levels in the gluelump spectrum are gratifyingly close to those of the lattice. The actual magnitude of the splitting between the lowest two levels is given as $.35$ GeV in \cite{Michael2}. With the bag parameters we have chosen it is somewhat larger, $.54$ GeV. It is also amusing that our crude bag model estimate of the $\Sigma-\Pi$ splitting of the lowest hybrid potentials also agree qualitatively with the lattice results. \cite{Michael3}\\ Some lattice gauge results for the gluelump spectrum also exist in the case of $SU(2)$ of color \cite{Michael3}. The corresponding bag model calculation differs from the above only by the change in group theory factors. Normalizing to the string tension which in the bag model is given by \[ SU(3): \sigma = \sqrt{\frac{32 \pi \alpha_s \Lambda}{3}} \], \[ SU(2): \sigma = \sqrt{6 \pi \alpha_s \Lambda} \] and noting that the ``color Coulomb'' potential has about the same strength in SU(2) as in SU(3), we have \[ \Lambda(SU(2))=\Lambda(SU(3)),\:\:\frac{3}{4}\alpha(SU(2))\simeq\frac{4}{3}\alpha(SU(3)) \] so that there is no appreciable change in the bag model spectrum in going from $SU(3)$ to $SU(2)$ (apart from the fact that the two gluon state must now have $C=-$ only). There is a suggestion from the lattice \cite{Michael2} that the separation in $SU(2)$ may be less than in $SU(3)$.\\ \acknowledgements{We thank PPARC for financial support and GK also thanks NSERC, Canada for support.}
\section{Introduction} One of the main challenges of thermal QCD is to get reliable numbers. Though the gauge coupling may be small, Linde's argument\cite{linde} tells us that perturbation theory will fail. The powerlike infrared divergencies one meets in perturbation theory will off-set the powers of the coupling constant. At what order in perturbation theory this will happen depends on the observable in question. For the free energy this happens when the static sector starts to dominate, and a simple dimensional argument shows this will happen at $O(g^6)$. For the Debye mass Linde's phenomenon starts already at next to leading order. So the problem is certainly not academic! One should bear in mind that Linde's argument does not deny the existence of a perturbation series. It says that from a certain order on the coefficients are no longer obtained by evaluating a finite number of diagrams of a given loop order. So we are faced with evaluating non-perturbative effects from the three dimensional sector defined by the static configurations. It was realized some time ago~\cite{dimred} that one could take the static part of the 4D action combined with induced effects by the non-static configurations. This theory gives at large distances the same physics as the 4D theory, and has the advantage of relatively straightforward lattice simulations\cite{debye}. In section 2 we discuss the relation between the 4D and the 3D theory. In particular we show how the phase diagram of the 3D theory has a remarkable property: the curve of 4D physics, and the critical curve as determined by perturbation theory do coincide to one and two loop order. However, perturbation theory has no reason to be trustworthy in determining the critical curve, and this is probably the reason why the fit to the numerical determination is problematic. In section 3 we discuss the physics of the domain wall in some detail. \section{Effective 3D action and symmetry in 4D}\label{sec:eff3dact} Construction of the effective action proceeds along familiar lines. In the case of QCD with $n_f$ quarks its form is given by integrating out the heavy modes of $O(T)$: \begin{equation} S_{3D}=S_{YM,n=0}+S_{ind} \label{eq:s3d} \end{equation} The first term is the static sector of the pure Yang-Mills theory in 4D with coupling constant $g_3=g\sqrt T$. The second term in eq.~\ref{eq:s3d} must contain the symmetries of the original QCD action, as long as they are respected by the reduction process. So we expect the induced action to be of the form: \begin{equation} S_{ind}=V(A_0)+ \mbox{ terms involving derivatives} \end{equation} $V(A_0)$ should be invariant under static gauge transformations, C, CP ($A_0\to -A_0^T$) and this reduces it to a sum of traces of even powers of $A_0$: \begin{equation} V(A_0)=m^2TrA_0^2+\lambda_1(TrA_0^2)^2+\lambda_2 TrA_0^4+.... \label{eq:va0} \end{equation} Only one independent quartic coupling survives for SU(2) and SU(3). We take it to be ${TrA_0^2}^2$. Note that we lost a symmetry present in the 4D action for gluons alone, and less and less conserved when quarks get lighter and lighter: Z(N) symmetry. Remember from the lattice formulation of pure Yang-Mills that one can multiply at a given time slice in the original 4D action all links in the time direction with a factor $\exp{\pm i\ {2\pi\over 3}}$. This will not change the form of the action, but will change by the same factor the value of the Wilson line $P$ wrapping around the periodic time direction: \begin{equation} P(A_0)={\cal {P}}{\exp{i\int A_0 d\tau}} \end{equation} Clearly in eq.~\ref{eq:va0} this symmetry has gone. Apparently the reduction process does not respect $Z(3)$ symmetry! The reason for this is twofold: i)the reduction process does not include the static modes. ii)the values of $A_0/T$ in the effective action are order $g$, whereas the Z(N) symmetry equates the free energy in $A_0 /T$ and $A_0/T+O(2\pi /3)$. To understand this better -- and to prepare the way for the discussion of the domain wall observable in the last section \ref{sec:domain} -- we recall some familiar facts in 4D for SU(3). \subsection{Z(3) symmetry and domain walls in 4D gauge theory}{\label{sec:domain1}} The free energy $U$ as a function of the Wilson line invariants $TrP,TrP^2$ is naturally defined through: \begin{equation} \exp{-{VU(t_1,t_2)\over T}}=\int DA_0 D\vec A \ \delta(t_1-\overline{ TrP}) \delta(t_2-\overline{TrP^2}) \ \exp{-{S(A)\over{g^2}}} \label{eq:wilsonline} \end{equation} where $\overline{TrP}$ is the normalized space average of the trace over the volume $V$. A natural parametrization of the parameters $t_1$ and $t_2$ suggests itself: define the phase matrix $\exp{iC}$ with $C$ being a traceless diagonal 3x3 matrix with entries $C_i \ ,(i=1,2,3)$ and $\sum_i C_i=0$, because we have SU(N), not U(N). Consider pure Yang-Mills. A gauge transformation that is periodic modulo a phase in Z(3) will {\it only} change the arguments in the delta functions in eq.~\ref{eq:wilsonline}. Hence the potential $U$ has degenerate minima in all points of the C-plane, where $\exp{iC}=1$, or $\exp{\pm i2\pi/3}$. This is called Z(3) symmetry (and the degeneracy is lifted by the presence of quarks). This statement is independent of perturbation theory. In fact the potential in eq.~\ref{eq:wilsonline} has been computed in perturbation theory including two loop order. And this potential includes the static modes. Propagators acquire a mass proportional to the phases $C$, because it acts like a VEV of the adjoint Higgs $A_0$. Hence, for small $C$, eventually Linde's argument will apply and the perturbative evaluation becomes impossible. For SU(3) the direction in which the Wilson line phase causes minimal breaking is in the hypercharge direction $C={1\over 3}diag(q,q,-2q)$. Minimal breaking means the maximal number of unbroken massless excitations, that do not contribute to the potential. Hence this is at the same time the valley through which the system tunnels from one minimum to the next. In this "q- valley" the combined 1 and 2 loop result is exceedingly simple: \begin{equation} U^{(1)}+U^{(2)}={4\pi^2\over 3}T^4(N-1)\left(1-5 {g^2N\over{(4\pi)^2}}\right)q ^2(1-q)^2 \label{eq:pot4d2q} \end{equation} For use in the reduced theory we isolate the static part of the one and two loop contribution in the q-valley from eq.~\ref{eq:pot4d2q}: \begin{equation} \left( U^{(1)}+U^{(2)}\right)_{(n=0)} =-T^4(N-1){4\pi^2\over 3}\left(2q^3 +3{g^2N\over{(4\pi)^2}}q^2\right) \label{eq:un0} \end{equation} Note that the two loop contribution is quadratic in q in contrast to the one loop which is cubic. The two-loop cubic part in eq.~\ref{eq:pot4d2q} comes from a combination of static and non-static modes. If we prepare the 4D system conveniently this symmetry will give rise to domain walls. Profile and energy of these wall have been computed semi-classically a long time ago\cite{bhatta}. The method of twisted boundary conditions triggers walls and is most economic computerwise. We will discuss them in the context of the lattice formulation in section~\ref{sec:domain}. Be it enough to mention that these boundary conditions force the Wilson lines to change by a Z(N) phase in going from one side to another side of the volume in some a priori fixed space direction. This will trigger a wall profile for the loop in this direction. It is the long range behaviour of this profile that contains the information on the Debye mass. To one loop order this behaviour comes entirely from the slope of the potential, see above. But to two loop order we have to take the one-loop renormalization of the gradient part of the Wilson line phase into account, and this suffers the Linde effect: there is an infinity of many-loop diagrams contributing to the gradient part. So to next to leading order there are already non perturbative effects in the long range tail of the wall, and hence in the Debye mass, as we mentioned earlier. On the other hand we know that the effective 3D action correctly reproduces the large distance behaviour of the 4D theory. So a 3D projection of the twist should produce a wall with the same tail as the 4D one. The inside of the wall in both formulations may be quite different but the inside is anyway computable by perturbation theory. \subsection{3D action and 4D physics}{\label{sec:4dphysics}} The parameters of the 3D theory ($m^2$ and $\lambda\equiv\lambda_1+\lambda_2$ for SU(3)) in eq.~\ref{eq:va0} can be calculated in perturbation theory by integrating out all modes in a path integral except the mode $A_\mu (\vec x, n=0)$. To one loop order we have the well known result for the Debye mass and for the four point coupling $\lambda$. All higher order terms have a coefficient zero~\cite{these}. To two loop order one has to take care not only of the two loop graphs, but also of the 1-loop renormalization of the three dimensional gauge coupling $g_3$ and the renormalization of the $A_0$ field in the gradient terms. The latter renormalization is taking care of gauge dependence in the two loop graphs. The result~\cite{loop} in the $\overline{MS}$ scheme is that both parameters are expressed in the renormalized 4D coupling $g(\mu )$ where $\mu $ is the subtraction point. Eliminating the 4D coupling gives for the dimensionless quantities $x\equiv {\lambda\over{g_3^2}}$ and $y={{m^2} \over{g_3^4}}$ the result for N=3: \begin{equation} xy_{4D}={3\over{8\pi^2}}(1+{3\over 2}x) \label{eq:physline} \end{equation} whereas for N=2: \begin{equation} xy_{4D}={2\over{9\pi^2}}(1+{9\over 8}x) \label{eq:physlinebis} \end{equation} Note the absence of explicit $\mu$ dependence in this relation. The variable $x$ has a $\mu\over T$ dependence such that as T becomes large $x$ becomes small. In conclusion, it is along this line that we have to simulate the 3D system, in order to get information about the 4D theory. Before we do this, we still have to settle an important question: where are -- in the $xy$ versus $x$ diagram -- possible phase transitions? \subsection{Phase diagram of the 3D theory}{\label{secphase}} To get the phase diagram we must first decide what order parameters to take. In the case of SU(3) there are two: $TrA_0^2$ and $TrA_0^3$. Strictly speaking, only the latter is an order parameter, since it flips sign under C. We will study the analogue of eq.\ref{eq:wilsonline}: \begin{equation} \label{effaction} \exp{-VS_{eff}(D,E)}=\int DA\delta\left(g_3^2D-\overline{TrA_0^2}\right) \delta\left(g_3^3E-\overline{TrA_0^3}\right)\exp{-S} \label{eq:effactiondef} \end{equation} Again as for the Wilson line we parametrize D and E in terms of $D=Tr\left[ C^2 \right]$ and $E=Tr\left[ C^3 \right]$ respectively. Let us first state the result one gets for $S_{eff}$ to one and two loop order: \begin{equation} S_{eff}={U(n=0)\over T}\hskip1cm \hbox{one and two loop only} \label{eq:static} \end{equation} The one and two loop result equals the static part of the 4D Z(3) potential,eq.~\ref{eq:wilsonline}! This static part was explicitely written in the q-valley, eq.\ref{eq:un0}. It has to be added to the tree result and one gets in terms of the dimensionless variables x and y for N= 2 or 3 colours, absorbing a factor $2\pi$ in q: \begin{equation} {S_{eff}\over{g_3^6}}= y\left({N-1\over N}\right)q^2 +x\left({N-1\over N}\right)^2q^4 -(N-1)\left({1\over{3\pi}}q^3 +{N\over{(4\pi)^2}}q^2\right) \label{eq:effaction} \end{equation} The question is now: for what values of x and y we have degenerate minima for q? Keeping only the 1 loop result cubic in q we see that it must be of the order of magnitude of the quartic term of the tree result to get a second degenerate minimum. So q must be of $O({1\over x})$ in that minimum. Thus the quadratic two loop result contributes $O(x)$ less. From eq.\ref{eq:effaction} we find the potential develops two degenerate minima for N=3 when: \begin{equation} xy_{c}={3\over{8\pi^2}}(1+{3\over 2}x) \label{eq:crit} \end{equation} For N=2: \begin{equation} xy_{c}={2\over{9\pi^2}}(1+{9\over 8}x) \end{equation} This is important: slope and intercept of the physics line~\ref{eq:physline} are identical with those of the critical line~\ref{eq:crit}, at least if we can take the low order loop results for the critical line seriously. This was numerically found in ref.\cite{polonyi,loop} The intercept equality is just due to the Z(N) potential in 4D and the effective potential $S_{eff}$ in 3D being {\it identical} to one loop. But to two loop order this simple explanation is no longer true. The cubic term in eq.\ref{eq:pot4d2q} is appearing also in the two loop result, but not in the two loop result for the 3D effective action. It is however true that also in 2 loops the leading contribution is the static part of the Z(N) potential, eq.\ref{eq:un0}. \subsection{Saddle point of the effective potential in 3D}\label{sec:saddle} In this subsection we will investigate in more detail the computation of the 3D effective potential. The saddle point is found by admitting $A_0$ fluctuates around a diagonal and constant background B: \begin{equation} A_0=B+Q_0 \end{equation} whereas the spatial gauge fields fluctuate around zero: \begin{equation} A_i=Q_i. \end{equation} One then goes through the usual procedure of expanding the effective action \ref{eq:effactiondef}. The equations of motion fix the background B to be equal to the matrix C, and the part quadratic in the fluctuations will not contain any reference to the Higgs potential $V(A_0)$. This is clear because the quadratic constraint tells the mass term not to fluctuate. Only the Higgs component parallel to C , $TrCQ_0$, has a mass term due to the Higgs potential, $4\lambda TrC^2$. So apart from this the quadratic part comes entirely from the static part of the 4D action. We can make a convenient gauge choice, namely the static form of the covariant gauge fixing: \begin{equation} S_{gf}=Tr\left([ig_3 B,Q_0]+\partial_kQ_k\right)^2 \label{eq:fixing} \end{equation} This gives propagators which are precisely the static version of the propagators appearing in the Wilson line potential~\ref{eq:wilsonline}. Only the component $Q_0$ parallel to C is the exception: its propagator has a mass from the Higgs potential and can be written as the the sum of the static propagator and a remaining part (``massive'') containing the mass term: \begin{equation} {1\over {\vec p}^2}+ {1\over {\vec p}^2+4\lambda TrC^2}-{1\over {\vec p}^2} \label{eq:prop} \end{equation} The static propagator dominates in diagrams over the rest. The massive propagator will give rise to half integer powers of x in the perturbative expansion of the potential; gauge couplings contribute $O(1)$ in dimensionless units, whereas Higgs couplings contribute $O(x)$. As long as we are interested in intercept and slope of the critical curve, it follows that only the static part of the Feynman rules contributes. Hence the result \ref{eq:static}. Let's from now on work in the q-valley where we evaluate the effective action \ref{eq:effaction}. Then two remarks are crucial: i)The broken minimum occurs for $q=O(1/x)$. Power counting then reveals that from $O(x^{3/2})$ on an infinite number of diagrams contributes to each order. ii)From five loop order on, the potential starts to develop poles in $q=0$. We are bringing this up, because insisting on the low order result \ref{eq:crit} and fitting numerically the coefficients of $x^{3/2}$ and higher order gives an unexpected result: the numerical coefficients are orders of magnitude larger\cite{loop} than the first two in \ref{eq:crit}. In fig. 1, taken from ref.~\cite{loop}, the situation is shown. Only for very small x the critical and the 4D physics line are allowed to become tangent. It seems that this constraint affects the quality of the fit. Dropping it altogether necessitates numerical determination of transition points at $x\le 0.04$. \begin{figure}[h] \begin{center} \hskip 0.truecm \epsfbox{fig.eps} \caption{Phase diagram in the SU(2) case, from ref. 9. The straight line is the 4d $\rightarrow$ 3d curve of eq.\ref{eq:physlinebis}. The thick line is a 4th order fit to the data. The dashed line marks the region where the transition turns into a cross-over.} \end{center} \end{figure} \section{Debye mass from a 3D domain wall}{\label{sec:domain}} After this long discussion of where the physics line lies with respect to the critical curve we have to come to grips with the domain wall method. The idea here is extremely simple and has been explained elsewhere\cite{altesbron}. Twisted boundary conditions in 4D~\cite{groe} have a very simple and intuitive form in the reduced theory. Remember that a twisted plaquette in the time-space direction is of the form $Tr(1-\Omega U(P))$, with $\Omega=\exp{i2\pi/N}$. Thus intuitively one would say that all one has to do in the reduced action is to modify the kinetic part of the Higgs field by the twist, because that's what the plaquette in the time space direction is reducing to. In the next subsection we work out this idea in more detail. \subsection{Construction of the wall} In this section we want to make more precise the action that defines the wall. We follow the notation of ref.\cite{boucaud}, specifically that of hep-lat/9811004 and write the kinetic part of the action as: \begin{eqnarray} {\cal L}_{kin} & = & {36 \over \beta} \sum_{\vec x} Tr \left[ A^2({\vec x}) \right] \ - \ {12 \over \beta} \sum_{{\vec x},j} Tr \left[ A({\vec x}) U_j({\vec x}) A({\vec x}+ a{\vec e_j} ) U_j^+({\vec x}) \right]\nonumber \\ & = & {12 \over \beta} \sum_{{\vec x},j} Tr \left[ {1\over 2} \left( A^2({\vec x})+ A^2({\vec x}+ a{\vec e_j})\right) - A({\vec x}) U_j({\vec x}) A({\vec x}+ a{\vec e_j} ) U_j^+({\vec x}) \right] \nonumber\\ \end{eqnarray} where $\vec x$ is a vector with three components $(x,y,z)$. \noindent Consider the following expression: \begin{eqnarray} {\cal X } \ = \ 1 \ - \ {1\over {\rm N} } {\cal R}e \ Tr \left[ e^{{\rm i} \alpha A(x,y,0)} U e^{-{\rm i} \alpha A(x,y,1)} U^+ \right]\nonumber \label{eq:X} \end{eqnarray} If $\alpha A$ is small we get: \begin{equation} {\cal X }= {1\over {\rm N} } \alpha^2 Tr \left[ {1\over 2} A^2(0) + {1\over 2} A^2(1 ) - A(0) U A(1) U^+\right]\nonumber \end{equation} This is precisely the kind of expression that appears in eq. (19). >From this follows the expression for the modified kinetic energy in the plane $(x,y,0)$: \begin{eqnarray} {\cal L}_{kin}^{mod} = {12 \over \beta} \kern -.4 truecm \sum_{{\vec x},j \atop (z,j) \ne (0,3) } \kern -.3 truecm Tr \left[ {1\over 2} \left( A^2({\vec x})+ A^2({\vec x}+ a{\vec e_j})\right) - A({\vec x}) U_j({\vec x}) A({\vec x}+ a{\vec e_j} ) U_j^+({\vec x}) \right]&& \nonumber\\ + \ {12 \over \beta}\ { {\rm N} \over \alpha^2}\ \sum_{x,y} \left\{ 1 - {1\over {\rm N} } {\cal R}e \ Tr \left[ e^{{\rm i} \alpha A(x,y,0)} U_3(x,y,0) e^{-{\rm i} \alpha A(x,y,1)} U_3^+(x,y,0) \right] \right\}&&\nonumber\\ \end{eqnarray} \noindent So all we need is to put a twist $\Omega \ \in Z({\rm N})\ $ in order to get a wall: \begin{eqnarray} {\cal L}_{kin}^{wall} = {12 \over \beta} \kern -.4 truecm \sum_{{\vec x},j \atop (z,j) \ne (0,3) } \kern -.31 truecm Tr \left[ {1\over 2} \left( A^2({\vec x})+ A^2({\vec x}+ a{\vec e_j})\right) - A({\vec x}) U_j({\vec x}) A({\vec x}+ a{\vec e_j} ) U_j^+({\vec x}) \right]&& \nonumber\\ + {12 N \over \beta \alpha^2} \sum_{x,y} \left\{ 1 - {1\over {\rm N} } {\cal R}e \left( \Omega \ Tr \left[ e^{{\rm i} \alpha A(x,y,0)} U_3(x,y,0) e^{-{\rm i} \alpha A(x,y,1)} U_3^+(x,y,0) \right] \right) \right\}&&\nonumber\\ \end{eqnarray} \noindent What is now the actual value of $\alpha$ to use? We recover the kinetic term in the continuum if we relate the field $A$ on the lattice to the field $A_{cont}$ in the continuum by the relation: \begin{eqnarray} A \ = \ { A_{cont} \over g_3}\nonumber \end{eqnarray} \noindent This is not the usual normalization for the lattice fields. Usually we have: $A_{latt} \ = \ a g_3 A_{cont}$ and $A_{latt} \ \rightarrow \ 0 $ in the continuum limit. Here this is not anymore the case. Remember that to expand the modified action we had to suppose that $\alpha A$ was small. To enforce this condition it seems natural to put~: $ \alpha \ = \ a g_3^2$; in this manner terms of the kind $e^{{\rm i} \alpha A} $ become $e^{{\rm i} a g_3^2 A} $. That is to say, they become of the usual sort~: $e^{{\rm i} a g_3 A_{cont}} $. With this choice the term in the exponential indeed goes to zero as the lattice spacing goes to zero, so: \begin{eqnarray} \alpha \ \equiv \ a g_3^2 \ = \ { 6 \over \beta} \nonumber \end{eqnarray} In the end we obtain as final expression for the kinetic part of the action supporting the wall: \begin{eqnarray} {\cal L}_{cin}^{wall} = \displaystyle {12 \over \beta} \kern -.4 truecm \sum_{{\vec x},j \atop (z,j) \ne (0,3) } \kern -.31 truecm Tr \left[ {1\over 2} \left( A^2({\vec x})+ A^2({\vec x}+ a{\vec e_j})\right) - A({\vec x}) U_j({\vec x}) A({\vec x}+ a{\vec e_j} ) U_j^+({\vec x}) \right] &&\nonumber \\ \nonumber & &\\ \nonumber \displaystyle + \beta \sum_{x,y} \left\{ 1 - {1\over {\rm N} } {\cal R}e \left( \Omega \ Tr \left[ e^{{\rm i} {6 \over \beta} A(x,y,0)} U_3(x,y,0) e^{-{\rm i} {6 \over \beta}A(x,y,1)} U_3^+(x,y,0) \right] \right) \right\}&&\\ \nonumber & & \\ \label{eq:twistkin} \end{eqnarray} \subsection{Excitations of the wall}{\label{subsec:excwall}} Now the system with the wall is defined by adding the 3D gauge field action and the Higgs potential V(A) to eq.~\ref{eq:twistkin}. Let us call the resulting twisted action $S_t$. Both twisted and untwisted action have periodic boundary conditions. When we compute the average of an observable $O$ in the twisted box we average the observable over the $(x,y)$ plane at the point $z$, written as $\overline{O(z)}$, and compute in the twisted box (action $S_t$). It is quite trivial to relate this average to the correlation of the wall and $O$ in the untwisted box (action $S$): \begin{equation} \langle \overline{O(z)}\rangle_{S_t}=\langle\exp{-(S_t-S)} \overline{O(z)}\rangle_{S} \label{eq:twistav} \end{equation} There is no difference between the two actions except at $z=0$, at the location of the wall. The twist is C and P odd, but T even. This means we can expect a signal for the Debye mass by taking any observable $O$ C odd (a necessary condition\cite{arnold}). Whatever operator gives the lowest mass in the correlation \ref{eq:twistav} is the preferred one. Thus one and the same updating with the twisted box can be used for various operators. \section{Conclusions}{\label{sec:concl}} Once we know the 4D physics line we can do a simulation of the twisted box with some convenient observable, and measure the mass through eq.~\ref{eq:twistav}. Care should be taken, as emphasized by Kajantie et al.~\cite{debye}, that we start in the symmetric phase and then move to the 4D physics line. In so doing we will stay on the physical branch of the hysteresis curve for the mass, that we will meet when crossing the transition curve. Nethertheless our discussion of the {\it location} of the critical curve underlines the importance to know wether the 4D physics line lies for small x in the symmetric phase or in the broken phase. \section*{Acknowledgments} One of us (C.P.K.A.) thanks the organizers of this conference for their hospitality and for the occasion to present this material. \section*{References}
\section{Introduction} The quantitative theory of stellar structure is more than 100 years old {\Cse{Emd07}} and our understanding of the stellar interior has improved dramatically during this time, especially since it became possible to construct detailed stellar models with the help of computers in the 1950's. However, even today, our understanding of many observable properties of massive stars (${M_{\mathrm{ZAMS}}} \gtrsim 8 {\mathrm{M}_{\odot}}$, $\log L/{\mathrm{L}_{\odot}} \gtrsim 4$) remains rudimentary. Aside from comparatively minor uncertainties remaining in the opacities and nuclear physics, the major frontiers in the study of stars, and indeed stellar evolution in general, are proper treatments of convection, mass loss, and rotation. This paper is the first in a series concerning the effects of rotation and angular momentum transport on the evolution of stars massive enough that a single one can become a supernova ($M \gtrsim 8\,{\mathrm{M}_{\odot}}$). The first to recognize the importance of rotation for celestial bodies was Sir Isaac Newton. Early studies of rotating, self-gravitating, incompressible fluids were carried out by McLaurin, Jakobi, Poincar\'e, and Schwarzschild. Additional important contributions to the numerical treatment of rotating stars were provided by {\cite{KT70}}. {\cite{KMT70}} performed calculations taking these effects into account using a simple model for angular momentum transport. Studies with artificial rotation laws were carried out by {\cite{ES76}}. In their pioneering work, {\cite{ES78}} considered several rotationally induced instabilities, made order-of-magnitude estimates for their efficiencies, and performed time-dependent stellar evolution calculations of rotating massive stars up to the ignition of carbon burning. Later, {\cite{pin89}} introduced a parameterization of the poorly known efficiencies of the rotationally induced transport processes of {\cite{ES78}} and gauged them to solar models. The formalism we shall employ here is based largely upon these two works. We differ, however, in using more recent data to calibrate the uncertain efficiencies for angular momentum and composition transport in this formalism and especially in following the stars past carbon burning, all the way to the presupernova state. Our formalism (discussed in detail, in $\S$2) is relatively simple compared to others used in recent studies of rotation during hydrogen and helium burning, e.g., {\cite{Lan92}}, {\cite{Den94}}, {\cite{ery94}}, {\cite{CZ92}}, {\cite{Zah92}}, {\cite{USS96}}, {\cite{tal97}}, {\cite{Meyn97}}, and {\cite{MZ98}}, but easier to understand and implement, and more easily extrapolable to the late stages of stellar evolution. Indeed our poor understanding, especially during the late stages of stellar evolution, of both convection and possible modifications to angular momentum transport by magnetic fields (not considered in the present work nor in the papers cited above) suggests that it is worth trying something simple first. Most of the rotation physics described in {\Sect{rotphys}} can already be found, with slight modifications, in previous papers. However, since this is the first in a series of papers, it will facilitate our presentation to have all the relevant equations collected in one place. We also correct several errors. In previous publications, e.g. in the equation for the secular shear instability, and cast the results in a consistent notation. Following a summary of how we model various rotationally induced instabilities (\Sect{rotphys}) and a discussion of the uncertain parameters of the model (\Sect{calib}), we discuss the implementation of this physics in the stellar codes in {\Sect{NumSolve}} and give an overview of the initial models in {\Sect{InitialCond}}. The evolution during hydrogen burning and helium burning is discussed in {\Sects{cHburn}} and {\Sectff{cHeBurn}}, respectively. In {\Sect{CmpOther}} we compare our results to the works of other authors. The late evolution is discussed in {\Sect{LateEv}} and the final angular momentum distribution at the presupernova stage is given in {\Sect{j:preSN}}. Its implications are discussed in {\Sect{youngPSR}} and a summary and our conclusions are given in {\Sect{SumConcl}}. Discussion of the details of observable parameters (evolution in the HR diagram, surface abundances, lifetimes) and presupernova nucleosynthesis are deferred to future papers. \section{Rotation and Mixing in Massive Stars} \lSect{rotphys} \subsection{Modification to the stellar structure equations} \lSect{ModRot} In rotating stars, centrifugal forces act on the matter and lead to deviations from spherical symmetry. For slow to moderate rotation these deformations remain rotationally symmetric {\Cite{Tas78}}. Only if the rotational energy exceeds a notable fraction of the binding energy of the star does genuine triaxial deformation result. In this work we consider only the case of ``slow'' rotation, i.e., where no triaxial deformations are expected. Some stars may reach ``critical'' rotation velocity ({\Sect{Omega}}) at their surfaces during brief stages of their evolution {\Cite{HL98}}. However, except for possibly modifying the mass loss rate ({\Sect{Omega}}), this affects only the very outermost layers and is not expected to have a big influence on the results of this paper. Even for slow rotation, the shapes of surfaces of constant pressure, constant density, and constant temperature are affected by the centrifugal potential and thus deviate from spherical symmetry. The momentum equation and the energy transport equation for spherically symmetric stars must be modified to take this effect into account. In this work, the centrifugal force is included following {\cite{KT70}} in the approximation of {\cite{ES76}} and applied to the hydrodynamic stellar structure equations {\Cite{Fli93}}. In this approach, mass shells correspond to isobars instead of spherical shells. Corrections are applied to the acceleration and the radiative temperature gradient. According to {\cite{Zah75}}, {\cite{CZ92}}, and {\cite{Zah92}}, anisotropic turbulence acts much stronger on isobars than in the perpendicular direction. This enforces ``shellular'' rotation rather than cylindrical rotation {\Cite{MM97}}, and it sweeps out compositional differences on isobars. Therefore it can be assumed that matter on isobars is approximately chemically homogeneous. Together with the shellular rotation this allows us to retain a one-dimensional approximation. The specific angular momentum, $j$, of a mass shell is treated as a local variable and the angular velocity, $\omega$, is computed from the specific moment of inertia, $i$. The time-dependent angular momentum redistribution is discussed in {\Sect{AngTrans}}, and its influence on transport processes in {\Sect{RotMix}}. We begin here by describing the modification to the stellar structure equations of non-rotating stars {\Csa{ES76,MM97}}. Let ${V_{\!\mathrm{P}}}$ be the volume enclosed by a surface of constant pressure, $P$, and $\SP:=\partial {V_{\!\mathrm{P}}}$ its surface area. Then its ``radius'', ${r_{\!\mathrm{P}}}$, is defined as the radius of a sphere of the same volume, ${V_{\!\mathrm{P}}}=4\pi {r_{\!\mathrm{P}}}^3/3$, and the equation of continuity becomes \begin{equation} \dxdycz{\mP}{{r_{\!\mathrm{P}}}}{t}=4\pi{r_{\!\mathrm{P}}}^2\rho \;, \lEq{mDefRot} \end{equation} where $\rho$ is the density and $\mP$ the mass enclosed by $\SP$. For quantities varying on isobars, a mean value is defined by \begin{equation} \av{\;\cdot\;} := \frac{1}{\SP} \oint_{\SP}\cdot\;\;{\mathrm d}\sigma\;, \end{equation} where ${\,\D\!\!\;}\sigma$ is an element of isobaric surface area. The effective gravitational acceleration $\vec{g}$ is normal to $\SP$. For the equation of momentum balance, one finds \Cite{ES76} \begin{equation} \dxdycz{P}{\mP}{t}= -\frac{G \mP}{4\pi {r_{\!\mathrm{P}}}^4} {f_{\mathrm{P}}} -\frac{1}{4\pi {r_{\!\mathrm{P}}}^2}\dxdycz{^2 {r_{\!\mathrm{P}}}}{t^2}{\mP} \;, \lEq{MomBalRot} \end{equation} where $G$ is the gravitational constant, $P$, the pressure, $t$, the time, and the inertia term (last term) is added here. The influence of rotation is described by the quantity ${f_{\mathrm{P}}}$ \begin{equation} {f_{\mathrm{P}}} := \frac{4\pi {r_{\!\mathrm{P}}}^4}{G\mP\SP} \av{g^{-1}}^{-1} \;, \end{equation} where $g:=\abs{\vec{g}}$. The radiative temperature gradient then takes the form \begin{equation} \dxdycz{\ln T}{\ln P}{t} = \frac{3\kappa}{16\pi a c G} \frac{P}{T^4} \frac{\LP}{\mP} \frac{{f_{\mathrm{T}}}}{{f_{\mathrm{P}}}} \SBrak{1+\frac{{r_{\!\mathrm{P}}}^2}{G \mP {f_{\mathrm{P}}}} \dxdycz{^2 {r_{\!\mathrm{P}}}}{t^2}{\mP}}^{\!-1} \;, \lEq{NabRadRot} \end{equation} where $\kappa$ is the opacity, $T$ the temperature, $a$ the radiation constant, and $\LP$, the energy flux through $\SP$. The last factor on the right hand side is included to account for inertia as it follows from the momentum equation {\Cite{Fli93}}, and \begin{equation} {f_{\mathrm{T}}} := \Brak{\frac{4\pi {r_{\!\mathrm{P}}}^2}{\SP}}^{\!2} \Brak{\av{g}\av{g^{-1}}}^{-1} \;. \end{equation} For the derivation of these formulae and for a numerical evaluation of ${f_{\mathrm{T}}}$ and ${f_{\mathrm{P}}}$, see {\cite{ES76}}. The equations for ${f_{\mathrm{T}}}$ and ${f_{\mathrm{P}}}$ are solved iteratively with the stellar structure equations in order to obtain consistent models {\Cite{ES76,Fli93}}. In the rest of this work, the subscript $P$ is omitted (except for ${f_{\mathrm{P}}}$). There is, in principle, an inconsistency between the assumption of shellular rotation and the method described by {\cite{KT70}}, i.e., the assumption of shellular rotation does not generally lead to a conservative potential as it does for a constant rotation rate on cylinders, which is used by {\cite{KT70}}. However, {\cite{MM97}} show that replacing the average $\av{\;\cdot\;}$ by ``appropriate mean values'', i.e., reinterpreting the quantities describing the stellar structure as the mean values over the isobars, allows one to keep the formalism of {\cite{KT70}} as a good approximation. \subsection{Ordinary mixing in the absence of rotation} \lSect{MixInst} Compositional mixing is generally treated as a diffusive process and implemented by solving the diffusion equation \begin{equation} \dxdycz{X_n}{t}{m}=\dxdycz{}{m}{t} \SBrak{(4\pi r^2 \rho)^2 D \dxdycz{X_n}{m}{t}}+\DxDyInd{X_n}{t}{\mathrm{nuc}} \lEq{Diff} \;, \end{equation} where $D$ is the diffusion coefficient constructed from the sum of individual mixing processes and $X_n$, the mass fraction of species $n$. The second term on the right hand side accounts for nuclear reactions. At the inner and outer boundary reflecting conditions are used: \begin{equation} \At{\dxdycz{X_n}{m}{t}}{m=0}=0=\At{\dxdycz{X_n}{m}{t}}{m=M(t)} \;. \end{equation} Mixing, burning, and mass loss are treated as separate, sequential operations. The different contributions to the diffusion coefficient, $D$, are discussed in the following sections. \subsubsection{Convection and overshooting} \lSect{Conv} \pFig{shear} Convection occurs when the temperature gradient exceeds the adiabatic condition, as modified by any gradient in mean molecular weight, $\mu$ (\Fig{shear}). That is, a stratification is stable against convection if \begin{equation} {\nabla_{\!\mathrm{ad}}}-{\nabla}+\frac{\varphi}{\delta}{\nabla_{\!\mathrm{\mu}}} \geq 0 \lEq{Ledoux} \end{equation} {\Ce{KW91}}. This is the so-called Ledoux criterion for convection. Here the common definitions are used: \begin{equation} {\nabla_{\!\mathrm{ad}}}:=\dxdycz{\,\ln T}{\,\ln P}{\mathrm{ad}} \;,\quad {\nabla_{\!\mathrm{\mu}}}:=\DxDy{\,\ln \mu}{\,\ln P} \;,\quad {\nabla}:=\DxDy{\,\ln T}{\,\ln P} \;, \end{equation} \begin{equation} \delta:=-\dxdycz{\ln\rho}{\ln T}{\mu,P} \;,\qquad \varphi:=\dxdycz{\ln\rho}{\ln\mu}{P,T} \;. \end{equation} The index ``ad'' stands here for ``at constant entropy {\emph{and}} composition''. The diffusion coefficient for composition mixing is treated according to the mixing-length theory {\Cite{Vit53,Bom58}}: \begin{equation} {D_{\mathrm{conv}}}:={\alpha_{\mathrm{MLT}}}\HP{v_{\mathrm{conv}}}/3 \;, \end{equation} where ${v_{\mathrm{conv}}}$ is the convective velocity. The pressure scale-height is defined for the hydrostatic case by \begin{equation} \HP:=-\DxDy{r}{\ln P} = \frac{P}{\rho g} \;. \end{equation} The local gravitational acceleration is given by $g=Gm/r^2$. In this work a mixing-length parameter of ${\alpha_{\mathrm{MLT}}}=1.5$ {\Cite{Lan91B}} is used. The mixing performed by convection is fast in comparison to most of the other time-scales relevant for the stellar evolution. It operates on the local dynamical time-scale and usually manages to smooth out any compositional inhomogeneities in the regions where it is active. Only when the time-scale of thermonuclear burning becomes comparable to that of convection, as, e.g., during central silicon and shell oxygen burning, can notable gradients persist. In the present work ``overshooting'' of the convection into the convectively stable regime defined by {\Eq{Ledoux}} is neglected. It will be shown that rotation leads to mixing above the convective core of massive stars during central hydrogen burning and thereby to the formation of more massive helium cores later in the evolution. In order to obtain such mixing, large overshooting is often introduced in literature {\Ce{CS91,sch92}}, but moderate rotation can lead to similar effects. \subsubsection{Semiconvection} \lSect{Semiconv} Semiconvection is a secular instability which can occur in non-rotating stars. According to a local, linear stability analysis by {\cite{Kat66}}, it is an oscillatory instability which appears in regions where an unstable temperature gradient is stabilized against convection by a sufficiently large gradient in the mean molecular weight ($\mu$-gradient), i.e., it lives in the regime \begin{equation} \frac{\varphi}{\delta}{\nabla_{\!\mathrm{\mu}}} \geq {\nabla}-{\nabla_{\!\mathrm{ad}}} \geq 0 \lEq{semiconv} \end{equation} {\CiteA{; and \Fig{shear}}{KW91}}. Heat transfer between a displaced mass element and its surrounding causes the growth of the instability on the local thermal time-scale. In the code STERN (see {\Sect{STERN}}), semiconvection is treated following {\cite{LSF83}}. The diffusion coefficient for this process is computed from \begin{equation} {D_{\mathrm{sem}}}=\frac{{\alpha_{\mathrm{sem}}} K}{6{c_{\mathrm{P}}}\rho} \frac{{\nabla}-{\nabla_{\!\mathrm{ad}}}}{{\nabla_{\!\mathrm{ad}}}-{\nabla}+\frac{\varphi}{\delta}{\nabla_{\!\mathrm{\mu}}}} \;,\quad K=\frac{4acT^3}{3\kappa\rho} \;, \lEq{Dsem+K} \end{equation} where $K$ is the thermal conductivity and ${c_{\mathrm{P}}}$ the specific heat at constant pressure. As proposed by {\cite{Lan91B}}, an efficiency parameter of ${\alpha_{\mathrm{sem}}}=0.04$ is adopted here. In KEPLER (see {\Sect{KEPLER}}) semiconvection is computed from {\Cite{WZW78,WW93}} \begin{equation} {{D'}_{\mathrm{\!\!sem}}}=\frac{1}{6}{\alpha_{\mathrm{MLT}}}^2{v_{\mathrm{sem}}}\HP \;, \end{equation} where the velocity ${v_{\mathrm{sem}}}$ is determined through \begin{equation} {v_{\mathrm{sem}}}=\sqrt{\Brak{{\nabla}-{\nabla_{\!\mathrm{ad}}}}\frac{P\delta}{g\rho^2}\DxDy{P}{r}} \;. \end{equation} The diffusion coefficient is limited to a fraction ${\alpha_{\mathrm{sem}}}$ of the radiative diffusion coefficient \begin{equation} {D_{\mathrm{rad}}}=\frac{K}{\rho\cV} \end{equation} by means of \begin{equation} {D_{\mathrm{sem}}}=\frac{{\alpha_{\mathrm{sem}}}{D_{\mathrm{rad}}}{{D'}_{\mathrm{\!\!sem}}}}{{{D'}_{\mathrm{\!\!sem}}}+{\alpha_{\mathrm{sem}}}{D_{\mathrm{rad}}}} \;. \end{equation} As usual, $\cV$ denotes the specific heat at constant volume. In this work a value of ${\alpha_{\mathrm{sem}}}=10^{-4}$ is used in KEPLER, which results in a comparable efficiency for semiconvection as the value used for STERN {\Cite{Woo97:pc}}. \subsection{Rotationally induced mixing} \lSect{RotMix} In this work, the mixing processes discussed in {\cite{ES78}} are included in a parametric way, following the work of {\cite{pin89}}. Five different processes are considered. To account for the uncertain mixing efficiency of each, they are weighed by efficiency factors {\CiteA{; \Sect{calib}}{pin89}} and then added to the diffusion coefficient, $D$, in the diffusion equation {\Eqff{Diff}}. \subsubsection{Dynamical shear instability} \lSect{DSI} \pFig{DSI} Dynamical shear instability occurs when the energy that can be gained from the shear flow becomes comparable to the work that has to be done against the gravitational potential for the adiabatic turn-over of a mass element (``eddy''). This means that it is stabilized by density gradients. Since there is no work required to mix on isobars, this instability can work very efficiently on those {\CITE{horizontal turbulence; }{Zah92}} and thus enforce rigid rotation horizontally {\Cite{ES78,pin89}}. Thus chemical inhomogeneities are smoothed on isobars. This, together with the so called baroclinic instability, which also acts barotropic for shear on isobars on a dynamical time scale {\Cite{Zah83}}, justifies the assumption of shellular rotation and that the composition is only a function of the isobars (\Sect{ModRot}). The linear condition for stability is given by \begin{equation} {R_{\mathrm{i}}}:=\frac{\rho\delta}{P}\Brak{{\nabla_{\!\mathrm{ad}}}-{\nabla}+\frac{\varphi}{\delta}{\nabla_{\!\mathrm{\mu}}}} \Brak{g \DxDy{\ln r}{\omega}}^{\!2}>{R_{\mathrm{i,c}}} \approx\frac{1}{4} \lEq{def:ri} \end{equation} for the case of a rotating fluid {\Cite{Zah74}}. Here, $\omega$ is the angular velocity, ${R_{\mathrm{i}}}$, Richardson number, and ${R_{\mathrm{i,c}}}$, its critical value, about $1/4$. Note that the term for ${\nabla_{\!\mathrm{\mu}}}$ in {\Eq{def:ri}} was omitted in the original work by {\cite{ES78}} and {\cite{pin89}}. The corresponding diffusion coefficient is computed from the spatial extent of the unstable region ${d_{\mathrm{inst}}}$, limited to a pressure scale-height, and the local dynamical time-scale {\Csa{ES78}}: \begin{equation} {D_{\mathrm{DSI}}}=\SBrak{\Min{{d_{\mathrm{inst}}},\HP} \Brak{1-\Max{\frac{{R_{\mathrm{i}}}}{{R_{\mathrm{i,c}}}},0}}}^2 /{\tau_{\mathrm{dyn}}} \;, \end{equation} where the dynamical time-scale is defined by \begin{equation} {\tau_{\mathrm{dyn}}}:=\sqrt{r^3/(G\,m)} \;. \end{equation} Furthermore, it is assumed that the instability is weaker when the deviation from the Richardson criterion is smaller. This is accounted for by a factor $(1-{R_{\mathrm{i,c}}}/{R_{\mathrm{i}}})^2$, which is limited to the range $[0,1]$. For ${R_{\mathrm{i}}} > {R_{\mathrm{i,c}}}$, the flow is assumed to be stable against the dynamical shear instability and thus ${D_{\mathrm{DSI}}}$ is set to $0$. \subsubsection{Solberg-H{\o}iland instability} \pFig{SHI} The Solberg-H{\o}iland instability arises if an adiabatically displaced mass element experiences a net force (the sum of gravity, buoyancy and centrifugal force) that has components in the direction of the displacement only. {\cite{Was46}} gives a condition for the stability against axisymmetric adiabatic perturbations of this kind. It separates into two scalar conditions. At the equator the condition for stability in the vertical direction is \begin{equation} {R_{\mathrm{SH}}}:=\frac{g}{\rho}\SBrak{\Brak{\DxDy{\rho}{r}}_{\!\mathrm{ad}} -\DxDy{\rho}{r}} +\frac{1}{r^3}\DxDy{}{r}\Brak{r^2\omega}^2 \geq 0 \lEq{SHcA} \end{equation} {\Cite{Tas78,ES78}}. If the specific angular momentum $j\sim r^2\omega$ is constant with $r$, the last term on the left-hand side vanishes and the Ledoux criterion results --- not the Schwarzschild criterion as stated by \cite{ES78}. This can be seen by rewriting the condition for stability \begin{equation} {R_{\mathrm{SH}}}:=\frac{g\delta}{\HP}\SBrak{{\nabla_{\!\mathrm{ad}}}-{\nabla}+\frac{\varphi}{\delta}{\nabla_{\!\mathrm{\mu}}}} +\frac{1}{r^3}\DxDy{}{r}\Brak{r^2\omega}^2 \geq 0 \lEq{SHcB} \end{equation} and comparing it with \Eq{Ledoux}. If, on the other hand, the medium is marginally stable to convection, the first term on the right hand side vanishes and the Rayleigh criterion results {\Cite{Tas78,KW91}}. Note that this instability only occurs in regions of {\emph{decreasing}} specific angular momentum (\Fig{SHI}) and is strongly suppressed in stable stratifications (${\nabla}<{\nabla_{\!\mathrm{ad}}}+\frac{\varphi}{\delta}{\nabla_{\!\mathrm{\mu}}}$). The diffusion coefficient resulting from the Solberg-H{\o}iland instability is estimated in a way similar to that for the dynamical shear instability. The extent of the unstable region, ${d_{\mathrm{inst}}}$, limited to the pressure scale-height, is used as the characteristic length-scale, and the dynamical time-scale is used as characteristic time-scale: \begin{equation} {D_{\mathrm{SHI}}}=\SBrak{\Min{{d_{\mathrm{inst}}},\HP} \Brak{\frac{r\,{R_{\mathrm{SH}}}}{g}}}^2/{\tau_{\mathrm{dyn}}} \;. \end{equation} Again, as for the dynamical shear instability, a factor of order unity ($r\,{R_{\mathrm{SH}}}/g$) was introduced to smoothly turn on the instability as the criterion for stability gets increasingly violated, and ${D_{\mathrm{SHI}}}$ is set to $0$ wherever the stability criterion is fulfilled. \subsubsection{Secular shear instability} \lSect{SSI} The strict criterion for dynamical shear instability can be relaxed considerably by allowing for thermal adjustment of radial perturbations. However, this process then operates only on a thermal time-scale, and is therefore a secular process. Gradients in the mean molecular weight, which may inhibit the occurrence of the instability, also have to be taken into account. According to {\cite{ES78}}, the following two conditions have to be violated simultaneously for this instability to set in (\Fig{shear}): \begin{equation} {R_{\mathrm{is,1}}}:=\frac{{{\mathcal{P}_{\!\mathrm{r}}}}\,{R_{\mathrm{e,c}}}}{8}\frac{\rho\delta}{P}\Brak{{\nabla_{\!\mathrm{ad}}}-\nabla} \Brak{g\DxDy{\ln r}{\omega}}^{\!2} >{R_{\mathrm{i,c}}} \end{equation} {\Cite{Tow58,Zah75}} because of the relaxed condition for the temperature gradient, and \begin{equation} {R_{\mathrm{is,2}}}:=\frac{\rho\varphi{\nabla_{\!\mathrm{\mu}}}}{P} \Brak{g\DxDy{\ln r}{\omega}}^{\!2} >{R_{\mathrm{i,c}}} \lEq{SSIb} \end{equation} since the condition for the $\mu$-gradient is not relaxed. The latter formula follows from the physical arguments of {\cite{ES78}}, but corrects an error in their Eq.~(10). For the critical Reynolds number, ${R_{\mathrm{e,c}}}$, a value of $2500$ is assumed in this work {\CITE{but see also}{RZ99}}. The Prandtl number, ${{\mathcal{P}_{\!\mathrm{r}}}}$, is defined as the ratio of the thermal diffusion time-scale to the angular momentum diffusion time-scale, and is estimated according to {\cite{Tas78}}: \begin{equation} {{\mathcal{P}_{\!\mathrm{r}}}}= \frac{\cV\Brak{{\mu_{\mathrm{p}}}+{\mu_{\mathrm{r}}}}}{\chi} \;, \end{equation} where the coefficients of shear viscosity of the plasma and by radiation are computed according to \begin{equation} {\mu_{\mathrm{p}}}\approx0.406\,\frac{\sqrt{\mi (\kB T)^5}} {\Brak{{Z_{\mathrm{i}}} e}^{4}\ln\Lambda} \;,\qquad {\mu_{\mathrm{r}}}=\frac{4 a T^4}{15 c \kappa \rho} \end{equation} {\Cite{Spi62,Tas78}}, respectively. The quantity $\Lambda$ is the ratio of the cut-off length for ion collisions, which is taken as the ratio of the Debye length, to the impact parameter for a $\pi/2$ deflection for Rutherford scattering of the ions, i.e., \begin{equation} \Lambda=\frac{2}{3 e^3}\sqrt{ \frac{\mi \Brak{\kB T}^{3}}{\pi \rho {Z_{\mathrm{i}}}^5}} \end{equation} \CITE{for details, see}{Spi62}. Here, $e$ is the charge of the electron in e.s.u., $c$ the velocity of light, $\kB$ Boltzmann's constant, ${Z_{\mathrm{i}}}$ the charge number of the ion, and $\mi$ its mass. It should be noted that for burning phases beyond hydrogen burning, as well as for helium, carbon, or oxygen stars, it is important to take the ${Z_{\mathrm{i}}}$-dependence of the plasma viscosity into account. The quantity $\Lambda$ enters only logarithmically and $\ln \Lambda$ is $\sim25$. At temperatures below the Fermi temperature, depending somewhat on the chemical composition, the ion viscosity dominates over the electron contribution. For the evaluation of the formulae above, complete ionization is assumed. If magnetic fields and neutrinos are neglected, the thermal conductivity is given by $\chi\approx K$ {\CITE{\Eq{Dsem+K}; }{Tas78}}. The opacity, $\kappa$, used in this work takes into account the energy transport by radiation as well as heat conduction by degenerate electrons. Following {\cite{ES78}}, the circulation velocity associated with this process is computed from the time-scale and the length-scale of the turbulent elements, \begin{equation} {v_{\mathrm{SSI}}}=\sqrt{\frac{\nu}{{R_{\mathrm{e,c}}}}\DxDy{\omega}{\ln r}} \;, \end{equation} limited to the adiabatic sound velocity, ${c_{\mathrm{s}}}$. The kinematic viscosity, $\nu$, is given by {\Cite{Tas78}} \begin{equation} \nu=\frac{{\mu_{\mathrm{p}}}+{\mu_{\mathrm{r}}}}{\rho} \;. \end{equation} For the typical length-scale the velocity scale height of the flow is assumed, \begin{equation} {H_{\mathrm{v,SSI}}}:=\abs{\DxDy{r}{\ln {v_{\mathrm{SSI}}}}} \;, \end{equation} limited to the pressure scale height. The resulting diffusion coefficient is given by \begin{equation} {D_{\mathrm{SSI}}}=\Min{{v_{\mathrm{SSI}}},{c_{\mathrm{s}}}} \Min{{H_{\mathrm{v,SSI}}},\HP} \Brak{1-\frac{\Max{{R_{\mathrm{is,1}}},{R_{\mathrm{is,2}}}}}{{R_{\mathrm{i,c}}}}}^{\!2} \;. \end{equation} Again, the instability is smoothly turned on with increasing violation of the stability criteria (term in the last bracket). In recent work, {\cite{MM96}},{\cite{Mae97}}, and {\cite{MZ98}} reconsidered the interaction of thermal diffusivity, horizontal turbulence (due to the baroclinic instability), and vertical shear. An important conclusion that can be drawn from their work is that $\mu$-gradients might not completely suppress the occurrence of the shear instability, since the medium is already turbulent due to the baroclinic instability. Consequently, some mixing can occur {\Cite{Mae97}}. In the present work, we parameterize the efficiency of the secular shear instability for chemical mixing and of the $\mu$-gradients in suppressing its occurrence (\Sect{uncertMix}). \subsubsection{Eddington-Sweet circulation} \lSect{ES} As first shown by {\cite{Zei24A,Zei24B}} for rigid rotation, and later by {\cite{BK59}} for a general rotation law, a rotating star cannot be in hydrostatic and radiative thermal equilibrium at the same time. This is so because surfaces of constant temperature and constant pressure do not coincide. Consequently, large-scale circulations develop. Since inhomogeneities on isobars are quickly smoothed out by the horizontal turbulence only the perpendicular ($\approx$ radial) component of the circulation velocity is considered here, and the process is approximated by diffusion along the radial coordinate. \cite{Kip74} estimated the circulation velocity as \begin{equation} {v_{\mathrm{e}}}:=\frac{{\nabla_{\!\mathrm{ad}}}} {\delta\,\Brak{{\nabla_{\!\mathrm{ad}}}-\nabla}} \frac{\omega^2 r^3 l}{\Brak{G m}^2} \Brak{\frac{2 \Brak{{\varepsilon_{\mathrm{n}}}+{\varepsilon_{\mathrm{\nu}}}} r^2 }{l} -\frac{2 r^2}{m} -\frac{3}{4 \pi \rho r}} \;. \lEq{vESO} \end{equation} In the presence of $\mu$-gradients, meridional circulation has to work against the potential and thus might be inhibited or suppressed {\Cite{Mes52,Mes53}}. Formally, this can be written as a ``stabilizing'' circulation velocity, \begin{equation} {v_{\mathrm{\mu}}}:=\frac{\HP}{{\tau^*_{\mathrm{KH}}}} \frac{\varphi{\nabla_{\!\mathrm{\mu}}}}{\delta\,\Brak{\nabla-{\nabla_{\!\mathrm{ad}}}}} \lEq{vmu} \end{equation} {\Cite{Kip74,pin89}}, where \begin{equation} {\tau^*_{\mathrm{KH}}}:=\frac{G m^2}{r \Brak{l - m {\varepsilon_{\mathrm{\nu}}}}} \lEq{tauKHx} \end{equation} is the local Kelvin-Helmholtz time-scale, used here as an estimate for the local thermal adjustment time-scale of the currents {\Cite{pin89}}. The spatial extent of the currents is typically of the order of the radius coordinate $r$. Here, neutrino losses are taken into account, because they reduce the thermal time-scale in the late stages of the stellar evolution significantly. Note that ${\varepsilon_{\mathrm{\nu}}}$ is defined as the energy {\emph{generation}} rate due to neutrino losses and therefore is negative. This increases the numerator in the definition of the local Kelvin-Helmholtz time-scale and thus decreases ${\tau^*_{\mathrm{KH}}}$. For the evaluation of the diffusion coefficient, the sign of the circulation velocity does not matter, but the stabilizing ``currents'' due to $\mu$-gradients always point in the direction opposite to the meridional flow, thus resulting in a reduction of the effective circulation velocity. The velocity is then computed from \begin{equation} {v_{\mathrm{ES}}}:=\Max{\abs{{v_{\mathrm{e}}}}-\abs{{v_{\mathrm{\mu}}}},0} \lEq{vES} \end{equation} {\CiteA{; and {\Fig{GSF}}}{ES78}}. The diffusion coefficient is calculated as the the product of the circulation velocity and a typical length-scale for the circulation. This is assumed to be the minimum of the extent ${d_{\mathrm{inst}}}$ of the instability and the velocity scale-height \begin{equation} {H_{\mathrm{v,ES}}}:=\abs{\DxDy{r}{\ln {v_{\mathrm{ES}}}}} \end{equation} {\Cite{ES78}}, i.e., \begin{equation} {D_{\mathrm{ES}}}:=\Min{{d_{\mathrm{inst}}},{H_{\mathrm{v,ES}}}}{v_{\mathrm{ES}}} \;. \end{equation} In recent work, {\cite{CZ92,Zah92,USS96,tal97,MZ98}} have discussed several improvements to the theory of meridional circulation and its interaction with the baroclinic instability. In contrast to the present work, their method requires the solution of a fourth order differential equation in $\omega$, which is numerically very involved. So far this method has only been used to investigate main sequence stars. An interesting result of these work for the Eddington-Sweet circulation is that the stabilizing effect of $\mu$-gradients, entering through ${v_{\mathrm{e}}}$ in ${v_{\mathrm{ES}}}$ ({\Eqs{vESO}} and {\Eqff{vES}}), may be reduced (\Sect{uncertMix}). The second important change to the above estimate is that the interaction of the baroclinic instability and the large-scale meridional reduces the mixing efficiency of the Eddington-Sweet circulation in agreement with the numerical studies by {\cite{pin89}} for the sun. We consider these effects when we perform an empirical calibration of the mixing efficiencies in {\Sect{calib}}. \subsubsection{Goldreich-Schubert-Fricke instability} \lSect{GSF} \pFig{GSF} {\cite{GS67}} and {\cite{Fri68}} performed an analysis of stability against axisymmetric perturbations (GSF instability). For the inviscid limit {\mbox{(${{\mathcal{P}_{\!\mathrm{r}}}} \ll 1$)}}, which can be well assumed in the interior of stars, they derive two conditions for stability in chemically homogeneous stars {\Cite{Kip69}}: \begin{equation} \dxdy{j}{r} \geq 0 \qquad\mbox{and}\qquad \dxdy{\omega}{z} = 0 \;. \lEq{GSFc} \end{equation} The first condition is the secular analogue to the Solberg-H{\o}iland stability criterion {\Eq{SHcB}}, where the stabilization by the temperature gradient is removed due to thermal conduction. This is similar to the relation between the secular and the dynamical shear instability. The second condition in {\Eq{GSFc}} is the analogue to the Taylor-Proudman theorem for slowly rotating incompressible fluids {\Cite{Kip74,Tas78}}. If the rotational velocity depends on the distance from the equatorial plane, i.e., the rotation profile is not conservative, meridional flows will be driven. Also in this case, the buoyancy force, which acts to suppress the instability, can be removed by heat conduction. However, this occurs only on a thermal time-scale. Interestingly, the typical velocities for both the above processes are quite similar {\Cite{Kip74}}. Since the second condition of {\Eq{GSFc}} is in general in contradiction with the shellular rotation law enforced by the baroclinic instability, except for the case of solid body rotation, the GSF instability will tend to enforce uniform rotation in chemically homogeneous regions {\Cite{ES78}}. The dependence of the GSF instability on differential rotation is stronger than that of Eddington-Sweet circulation, and the large-scale circulation velocity in the equatorial plane can be estimated by \begin{equation} {v_{\mathrm{g}}}=\frac{2 {H_{\mathrm{T}}} r}{{H_{\mathrm{j}}}^2} \Brak{1+2\DxDy{\ln r}{\ln \omega}}^{-1} {v_{\mathrm{e}}} =\frac{2{H_{\mathrm{T}}}}{{H_{\mathrm{j}}}}\,\DxDy{\ln \omega}{\ln r} {v_{\mathrm{e}}} \end{equation} {\Cite{ES78,JK70,JK71}}. Here ${H_{\mathrm{T}}}:=-\Frac{{\mathrm d} r}{{\mathrm d} \ln T}$ is the temperature scale-height and ${H_{\mathrm{j}}}:=\Frac{{\mathrm d} r}{{\mathrm d} \ln j}$, the scale-height of the angular momentum distribution. The GSF instability has the same $\mu$-dependence as Eddington-Sweet circulation {\Cite{ES78}} and therefore the resulting circulation velocity is computed in the same way, taking the stabilizing effect of the $\mu$-gradient into account: \begin{equation} {v_{\mathrm{GSF}}}:=\Max{\abs{{v_{\mathrm{g}}}}-\abs{{v_{\mathrm{\mu}}}},0} \;. \end{equation} Again, the diffusion coefficient is determined from the circulation velocity, ${v_{\mathrm{GSF}}}$, and the minimum of the circulation velocity scale height, ${H_{\mathrm{v,GSF}}}$, and the extent, ${d_{\mathrm{inst}}}$, of the instability: \begin{equation} {D_{\mathrm{GSF}}}:=\Min{{d_{\mathrm{inst}}},{H_{\mathrm{v,GSF}}}}{v_{\mathrm{GSF}}} \;, \end{equation} where we define in the same way as above \begin{equation} {H_{\mathrm{v,GSF}}}:=\abs{\DxDy{r}{\ln {v_{\mathrm{GSF}}}}} \;. \end{equation} {\Fig{GSF}} compares the parameter space in which the GSF and the Eddington-Sweet instability operate. For small angular velocity gradients the Eddington-Sweet circulation dominates, while the GSF instability becomes more important as the differential rotation increases. Note that for strong differential rotation the shear instability also occurs (cf. {\Fig{shear}}) \subsection{Other instabilities} \lSect{OtherInst} The five instabilities discussed in the previous section are not a complete list of all rotationally induced instabilities for massive stellar evolution. However, they appear to be the most relevant ones, or at least the best understood. For the ABCD-instability {\Cite{SKR84}} and the triply diffusive instability {\Cite{KS83}}, no reliable estimates of efficiency exist. Furthermore, non-axisymmetric instabilities may also exist, but are poorly investigated so far. Another important issue is the interaction of the different instabilities, and the interaction of rotation and rotationally induced instabilities with the instabilities listed in {\Sect{MixInst}}. The interaction of the shear instabilities and the Eddington-Sweet circulation has been investigated by, e.g., {\cite{CZ92,Zah92,USS96,MM97,Mae97,TZ97,tal97,MZ98}}, and semiconvection has recently also been included by {\cite{Mae97}} and {\cite{MZ98}}. However, the effects of the interactions are not large and therefore not taken into account in the present work. Perhapes most importantly, we have neglected magnetic fields. Magnetic fields might transport angular momentum by torques {\CITE{$\sim r^3 B_{r}B_{\phi}$;}{Spr97}}, or cause instabilities by magnetic buoyancy resulting from the winding up of magnetic field lines by differential rotation. This could be effective even if the initial field strength is small {\Cite{Spr97,SP98}}. Unfortunately, little is known about either the strength of the initial field or the efficiency of instabilities in amplifying the magnetic field. The Velikhov-Chandrasekhar instability depends only on the presence of magnetic fields, not on their strength, but it is efficiently suppressed by $\mu$-gradients {\Cite{Ach78,Spr97}}. Detailed studies of the action of magnetic fields inside stars must be left to future investigations. \subsection{Angular momentum transport} \lSect{AngTrans} Following {\cite{ES78}} and {\cite{pin89}}, we formulate the transport of angular momentum as a diffusive process, \begin{equation} \dxdycz{\omega}{t}{m}=\frac{1}{i}\dxdycz{}{m}{t} \SBrak{(4\pi r^2 \rho)^2 i\nu \dxdycz{\omega}{m}{t}} - \frac{2\omega}{r}\dxdycz{r}{t}{m} \Brak{\frac{1}{2}\DxDy{\,\ln i}{\,\ln r}} \lEq{AngDiff} \end{equation} {\Cite{ES78}}, where $\nu$ is the turbulent viscosity and $i$, the specific angular momentum of a shell at mass coordinate $m$. For a spherical shell of constant density, inner radius $\ri$ and outer radius ${r_{\!\mathrm{o}}}$, the specific moment of inertia, $i$, is given by $i=0.4\,\Frac{{r_{\!\mathrm{o}}}^5-\ri^5}{{r_{\!\mathrm{o}}}^3+\ri^3}$; for a thin shell of radius $r$ this simplifies to $i=2/3\,r^2$. The last term in {\Eq{AngDiff}}, an advection term, accounts for contraction or expansion of the layers at constant mass coordinate. The factor in the last bracket on the right hand side vanishes if the gyration constant $k:=i/r^2$ does not depend on $r$. {\Eq{AngDiff}} is essentially a diffusion equation for $\omega$ along the ``moment of inertia coordinate'', \begin{equation} I(m):=\int_0^m i(m'){\,\D\!\!\;} m' \;, \end{equation} defined analogously to the mass coordinate {\Ce{KW91}}. This equation conserves angular momentum and leads to rigid rotation in a region of extent $\ell$ whenever the diffusion time-scale, ${\tau_{\mathrm{D}}} := \ell^2/\nu$, is short in comparison to structural changes of the star. Since the Eddington-Sweet circulation may redistribute angular momentum by advection rather than by viscous stress {\Cite{Zah92}}, the equilibrium solution might deviate from rigid rotation assumed here in regions where it is the dominant process. However, for consistency to {\cite{ES78,pin89}} and for simplification of the numerical treatment we stick with the prescription outlined above. Compared to {\cite{TZ97}} we get very similar results at the end of central hydrogen burning (see {\Sect{CmpOther}}). At the inner and outer boundary, reflecting conditions similar to those given in {\Eq{Diff}} for the compositional mixing are used. At the surface of the star, the angular momentum contained in the layers which are lost due to stellar winds is removed from the star (\Sect{AngLoss}). The turbulent viscosity, $\nu$, is determined as the sum of the convective and semiconvective diffusion coefficients, and those from rotationally induced instabilities {\CiteA{; {\Sect{calib}}}{ES78}}. In contrast to {\cite{ES78}} and {\cite{pin89}}, in the present work the transport equation for angular momentum is solved for the entire star as a whole. Since the evolutionary time-scale of the star is in most cases much longer than the convective time-scale, {\Eq{AngDiff}} results in rigid rotation in those regions. Unlike composition, which can show significant gradients even inside convective regions due to burning (e.g., during central silicon burning), angular momentum is locally conserved, and therefore convective regions can more easily reach rigid rotation than chemical homogeneity during hydrostatic burning phases. This, however, does not hold if the respective layers are contracting or expanding rapidly. The approximation that convection leads to rigid rotation rather than constant specific angular momentum seems to be justified, at least if the rotational period is long in comparison to the convective time scale, and it may also hold for more rapid rotation if convective blobs can be assumed to scatter elastically {\Cite{KNL95}}. The latitudinally averaged rotation rate of the solar convection zone deviates from solid body rotation by less than $5\,\%$ {\Ce{ABC97}}. \subsection{Enhanced mass loss due to rotation} \lSect{MassLoss} \lSect{Omega} Mass loss from the stellar surface (``stellar winds'') significantly affects the evolution of massive stars {\Cite{CM86}}. In the present work, the empirical mass loss rate of {\cite{NJ90}} is used. For Wolf-Rayet stars, the prescription of {\cite{Lan89B}} is applied. The uncertainties in these mass loss rates are considerable due to the uncertainties in the observational data and their interpretation. These mass loss rates are further modified to account for the effect of stellar rotation according to {\cite{FA86}} \begin{equation} {\dot{M}}(\omega) := {\dot{M}}(\omega=0) \times \left(\frac{1}{1-\Omega}\right)^{\xi} \;,\qquad \xi\approx0.43 \lEq{MassLossRot} \;, \end{equation} where \begin{equation} \Omega:=\frac{v}{{v_{\mathrm{crit}}}} \;, \lEq{Omega} \end{equation} is the ratio of the equatorial surface rotation rate to the critical rotation rate defined by \begin{equation} {v_{\mathrm{crit}}}^2:=\frac{Gm}{r}\Brak{1-\Gamma} \;. \lEq{vcrit} \end{equation} The Eddington factor, \begin{equation} \Gamma:=\frac{\kappa L}{4 \pi c G m} \;, \lEq{Gamma} \end{equation} is evaluated only in the radiative part of the optical depth range $\tau \in [2/3 , 100]$ {\Cite{Lam93,Lan97:LBV}}, where $\tau(r)=\int_r^{\infty}\kappa\rho{\,\D\!\!\;} r$ has the usual definition. The quantitative result for the $\Omega$-dependence of the mass loss rate obtained by {\cite{FA86}} was questioned by {\cite{OCG96}}, who performed hydrodynamic simulations of the winds of rotating hot stars including the effect of non-radial radiation forces and gravity-darkening in the approximation of {\cite{Zei24A,Zei24B}}. In any case, the latitude dependence of the surface properties (temperature, radiation flux, etc.) of rapidly rotating luminous stars is largely unknown as {\cite{Kip77}} showed in a generalization of the von Zeipel theorem that they depend strongly on the details of the internal rotation law {\Csa{Mae99}}. However, the only crucial ingredient for our model calculations, which is confirmed by {\cite{OG97}}, is the fact that the latitudinally integrated mass loss rate increases strongly as the star approaches the $\Omega$-limit, so that the star cannot exceed critical rotation, but rather loses more mass and angular momentum {\Cite{Lan98}}. \subsection{Angular momentum loss} \lSect{AngLoss} The loss of angular momentum from the surface due to stellar winds is approximated by removing of the angular momentum along with the surface layer, i.e., \begin{equation} {\dot{J}}={\dot{M}}{j_{\mathrm{surf}}} \;, \end{equation} where ${j_{\mathrm{surf}}}$ is the latitudinally averaged specific angular momentum at the surface of the star when the mass loss is assumed independent of latitude. \section{Calibration of the mixing efficiencies} \lSect{uncertMix} \lSect{calib} The diffusion coefficients used in this work are subject to considerable uncertainties, as they result from order-of-magnitude estimates of some of the relevant time- and length-scales. Therefore, efficiency factors of order unity are introduced, in order to calibrate the diffusion coefficients with observational data. This is similar to the treatment of {\cite{pin89}}. The first adjustable parameter is the ratio of the turbulent viscosity to the diffusion coefficient, $\fc:=D/\nu$. The contribution of the rotationally induced instabilities to the diffusion coefficient is assumed to be reduced by the factor $\fc$, while their full value enters the turbulent viscosity, \begin{equation} D={D_{\mathrm{conv}}}+{D_{\mathrm{sem}}}+\fc\Brak{{D_{\mathrm{DSI}}}+{D_{\mathrm{SHI}}}+{D_{\mathrm{SSI}}}+{D_{\mathrm{ES}}}+{D_{\mathrm{GSF}}}} \;, \end{equation} \begin{equation} \nu={D_{\mathrm{conv}}}+{D_{\mathrm{sem}}}+{D_{\mathrm{DSI}}}+{D_{\mathrm{SHI}}}+{D_{\mathrm{SSI}}}+{D_{\mathrm{ES}}}+{D_{\mathrm{GSF}}} \;. \end{equation} The second parameter, ${f_{\!\mathrm{\mu}}}\in[0,1]$, describes the sensitivity of the rotationally induced mixing to $\mu$-gradients, i.e., ${\nabla_{\!\mathrm{\mu}}}$ is replaced by ${f_{\!\mathrm{\mu}}}{\nabla_{\!\mathrm{\mu}}}$. In order to reproduce the surface {\I{7}{Li}} abundance in the sun, {\cite{pin89}} introduced the factor $\fc\in[0,1]$. They found a value of $\fc=0.046$ for their best fit. From theoretical work {\cite{CZ92}} found a similar value, $\fc=1/30$, for the combined action of shear and meridional circulation. This is the value chosen for most of the models presented in this work (cf. {\Tab{InitModels}}). \pFig{M-CNOHe-gauge} \pFig{fc-gauge} The best observational probe of rotationally induced mixing in stars is the evolution of the surface composition during central hydrogen burning. While lithium and boron are depleted early during this phase {\Cite{VLL96,FLV96}}, since they are destroyed at relatively low temperatures, {\I{14}{N}}, is only produced at higher temperature, i.e., much deeper inside the star. Therefore, an increase of nitrogen at the surface should be accompanied by an decrease of carbon ({\I{12}{C}}) or, in the case of even deeper mixing, oxygen ({\I{16}{O}}), which is destroyed at even higher temperatures. An enrichment of nitrogen of order $2\ldots3$ is observed for evolved stars of about $10$ to $20\,{\mathrm{M}_{\odot}}$ {\Cite{GL92,Her94,vra98}}. Since observations can only give the projected rotation rate and are also restricted to low projected rotational velocities (\Cite{GL92,vra98}), only a qualitative comparison with our models is possible. The processing of carbon to nitrogen which occurs at core hydrogen ignition does not introduce large $\mu$-gradients. Therefore, the occurrence of a surface nitrogen enrichment and carbon depletion is rather insensitive to ${f_{\!\mathrm{\mu}}}$. In contrast, any enrichment of helium in O stars {\Cite{her92,HVM98}} strongly restricts ${f_{\!\mathrm{\mu}}}$. Unfortunately, helium abundances are hard to measure and correspondingly uncertain {\Cite{Her94}}. For purposes of calibration, we computed evolutionary sequences for solar metallicity stars in the mass range $4\,{\mathrm{M}_{\odot}}$ to $60\,{\mathrm{M}_{\odot}}$ through core hydrogen burning, adopting a typical zero-age main sequence rotational velocity of $\sim 200\,{\km\,\Sec^{-1}}$ {\Cite{Sle70,Fuk82,Lang91,Hal96,Pen96}}. {\Fig{M-CNOHe-gauge}} shows the surface values of helium, carbon, nitrogen, and oxygen at core hydrogen exhaustion as function of the initial stellar mass for various combinations of ${f_{\!\mathrm{\mu}}}$ and $\fc$. A value of ${f_{\!\mathrm{\mu}}}=0.05$ reproduces an enhancement of nitrogen by a factor of $2$ to $3$ in the mass range $10\,{\mathrm{M}_{\odot}}$ to $20\,{\mathrm{M}_{\odot}}$, and results in a surface helium mass fraction of $\sim40\,\%$ for the $60\,{\mathrm{M}_{\odot}}$ star, while the enrichment remains quite small for stars below $20\,{\mathrm{M}_{\odot}}$. For ${f_{\!\mathrm{\mu}}}=0.01$, nitrogen and helium are clearly enriched too much for stars below $30\,{\mathrm{M}_{\odot}}$. On the other hand, the nitrogen enrichment might be too low for values of ${f_{\!\mathrm{\mu}}}\ge0.1$. Certainly, for ${f_{\!\mathrm{\mu}}}=0.25$ and ${f_{\!\mathrm{\mu}}}=1.0$ the nitrogen abundance for the most massive stars ($30\,{\mathrm{M}_{\odot}}\ldots60\,{\mathrm{M}_{\odot}}$) is inconsistent with the observations. The same is true for the helium abundances. \pTab{InitModels} In summary, ${f_{\!\mathrm{\mu}}}=0.05$ seems to be the best value (provided $\fc=1/30$; see above). This set of parameters is used in the present work for the models whose name ends with ``{\Mod{B}}'' (\Tab{InitModels}). The consequences of a variation of $\fc$ (for fixed ${f_{\!\mathrm{\mu}}}=0.05$) is shown in {\Fig{fc-gauge}} for a $12\,{\mathrm{M}_{\odot}}$ star. For small values of $\fc$ the nitrogen abundance is too low, while for large values, helium becomes quite high. As discussed above, too much surface enrichment occurs with $\fc=1/30$ for small values of ${f_{\!\mathrm{\mu}}}$ ($\simle0.01$). Nevertheless, it is interesting to investigate the case where $\mu$-gradients are completely neglected, since the calibration of $\fc$ and ${f_{\!\mathrm{\mu}}}$ is not unambiguous, and different combinations might result in similar surface enrichments. The surface abundance, however, are the only clear observational constraint, while the degree of internal mixing is not directly observable. Therefore, a second parameter set of $\fc=0.01$ and ${f_{\!\mathrm{\mu}}}=0$ is also used. The resulting surface abundances (displayed as thick grey line in {\Fig{M-CNOHe-gauge}}) show quite similar enrichments. Models with this choice of $\fc$ and ${f_{\!\mathrm{\mu}}}$ do {\emph{not}} carry a ``{\Mod{B}}'' at the end of their name (\Tab{InitModels}). A value of $\fc=0.01$ for ${f_{\!\mathrm{\mu}}}=0$ is also supported by calibrations of the lithium, beryllium and boron surface abundance for the sun by {\cite{Fli93}}. \section{Numerical solution} \lSect{NumSolve} Two different numerical codes were used here to follow the stellar evolution. We now briefly describe each. \subsection{STERN} \lSect{STERN} The STERN code is a pseudo-Lagrangian, implicit hydrodynamic code {\Cite{lan88}}, based on the ``G\"ottinger stellar evolution code''. For numerical solution, relative mass coordinate $q:=m/M$ is used instead of the the mass coordinate $m$, which allows to reserve the distribution of computational grid in the presence of mass loss. The equation of state includes radiation, ionization, relativistic electron degeneracy, and electron-positron pairs. Ions are treated as a Boltzmann gas {\Cite{EL86}}. The chemical evolution due to thermonuclear burning is traced by $35$ isotopes: {\I{}n}, {\I{1,2}H}, {\I{3,4}{He}}, {\I{6,7}{Li}}, {\I{7,9}{Be}}, {\I{8,10,11}B},{\I{11,12,13}C},{\I{12,14,15}N}, {\I{16,17,18}O}, {\I{19}F}, {\I{20,21,22}{Ne}}, {\I{23}{Na}}, {\I{24,25,26}{Mg}}, {\I{26,27}{Al}},{\I{28,29,30}{Si}}, and {\I{56}{Fe}}. Except for {\I{19}F}, {\I{26}{Al}}, and {\I{56}{Fe}}, reactions between them are solved in a $32$ isotope network. These reaction rates are also used to determine the nuclear energy generation rate. The {\El{Ne}}/{\El{Na}} and {\El{Mg}}/{\El{Al}} hydrogen-burning cycles are solved separately using a $13$ isotope network including {\I1H}, {\I{18}O}, {\I{19}F}, {\I{20,21,22}{Ne}}, {\I{23}{Na}}, {\I{24,25,26}{Mg}}, {\I{26,27}{Al}}, {\I{28}{Si}}, and {\I{16}{O }} {\Cite{Braun97}}. The neutrino losses are determined according to {\cite{MKI85}}. The reaction networks are solved separately for each zone between the individual stellar structure integration time-steps. This allows for subcycling of the reaction network with fine time-steps wherever needed. \subsection{KEPLER} \lSect{KEPLER} In the KEPLER code {\Cite{WZW78,WWF84,WW88}} the equation of state includes a crude treatment of Coulomb corrections, beyond what is used in STERN {\Cite{WZW78}}. A $19$-isotope network is employed through oxygen burning, including the elements {\I1H}, {\I3{He}}, {\I4{He}}, {\I{12}C}, {\I{14}N}, {\I{16}O}, {\I{20}{Ne}}, {\I{24}{Mg}}, {\I{28}{Si}}, {\I{32}S}, {\I{36}{Ar}}, {\I{40}{Ca}}, {\I{44}{Ti}}, {\I{48}{Cr}}, {\I{52}{Fe}}, {\I{54}{Fe}}, {\I{56}{Ni}} and neutrons and protons from photodisintegration. Silicon burning is followed using a quasi-equilibrium network of $137$ isotopes, in which subgroups of elements are treated in nuclear statistical equilibrium while reactions between these subgroups are considered explicitly. Beyond silicon burning full nuclear statistical equilibrium is assumed. A more detailed description of the reaction networks in KEPLER can be found in {\cite{WZW78}}. However, an improvement of the treatment of hydrogen burning has been implemented (\App{KepImpr}). For the present work, angular momentum has been added to KEPLER as a new local variable, and rotationally induced mixing processes incorporated according to {\Sect{RotMix}}. However, because changes to the structural model calculations on KEPLER would be difficult, the modifications to the momentum balance and the energy transport ({\Sect{ModRot}}) applied in STERN are not included into KEPLER. The same opacities {\Cite{IR96}} used in STERN are also included in KEPLER (an update to previous versions of the code), which allows for more consistency between the two calculations. For temperatures above $10^9\;{\mathrm{K}}$ the opacities used in KEPLER are still chiefly due to electron scattering with corrections due to relativity and degeneracy {\Cite{WZW78}}. As outer boundary conditions a finite (or zero) boundary pressure is often utilized in KEPLER. The radius of the photosphere is determined as the location where an optical depth of $2/3$ is reached. This treatment of the outer boundary condition, but also the mass loss, is less accurate than that implemented in STERN {\Cite{Heg98}}. For this reason, the stellar evolution from the pre-main sequence until a central temperature of $10^9\,{\mathrm{K}}$, i.e., before central neon ignition, is followed by STERN, and the rest of the evolution until core collapse by KEPLER. At this stage of evolution, the total mass lost in its remaining lifetime ($\simle100\,\yr$) prior to core collapse is negligible. The stellar envelope, and therefore the outer appearance of the star, hardly changes. However, stellar models followed form the pre-main sequence using KEPLER give results similar to those obtained by STERN. \section{Initial models} \lSect{InitialCond} The initial model for the calculations presented in this work is that of a fully convective, rigidly rotating (following our assumption that convection does lead to rigid rotation) pre-main sequence star. In the Hertzsprung-Russell (HR) diagram such stars are located on their Hayashi line. These models are constructed from the Lane-Emden equation {\Ce{KW91}} with a polytropic index of $n=3/2$. Typically, initial stellar radii around $1\,000\,{\mathrm{R}_{\odot}}$ are used. This kind of initial condition is for computational convenience only and is not intended to reproduce the true pre-main sequence evolution {\Csa{BM94,BM96}}. The influence of rotation on the stellar structure is negligible in the initial models, but it becomes more important when the stars contract towards central hydrogen ignition. On the zero-age main sequence (ZAMS) close-to rigid rotation establishes throughout the star, mainly through the action of Eddington-Sweet circulation (\Sect{ES}) and the Gold\-reich\--Schu\-bert\--Fricke instability (\Sect{GSF}). These processes are sufficiently efficient in the early stellar evolution that rigid rotation is established virtually independent of the initial angular momentum distribution assumed. Almost no angular momentum is lost before the star reaches the main sequence. \pTab{InitAbuSTERN} \pTab{InitAbuKEPLER} All models in this work use an approximately solar initial chemical composition with a mass fraction of all elements heavier than helium (``metals'') of $Z=0.02$. The mass fractions of hydrogen and helium are set to $X=0.7$ and $Y=1-X-Z=0.28$, respectively. In STERN (\Sect{STERN}), the abundance ratios of the isotopes within each of these groups are chosen to have the solar system meteoritic abundance ratios according to {\cite{GN93}} (see {\Tab{InitAbuSTERN}}). Calculations performed with the KEPLER code (\Sect{KEPLER}) start on the pre-main sequence with a relative distribution of the metals according to {\cite{AG89}} as given in {\Tab{InitAbuKEPLER}}. For the main set of models in this work the initial angular momentum is determined such that the stars reach a rotational velocity of $\sim200\,{\km\,\Sec^{-1}}$ on the ZAMS. This is a typical observed value for stars in the mass range $8\,{\mathrm{M}_{\odot}}\ldots25\,{\mathrm{M}_{\odot}}$ {\Cite{Sle70,Fuk82,Hal96,Pen96,how97}}. It corresponds to $\sim 35\,\%$ of their ``critical'' rotation speed (\Sect{Omega}). Also models with different initial rotation rates are computed, in order to investigate the influence of this parameter on the evolution of massive stars (see {\Tab{InitModels}}). \section{Central hydrogen burning} \lSect{cHburn} \subsection[Chemical mixing: the example of $20\,{\mathrm{M}_{\odot}}$ stars] {Chemical mixing: the example of $\mathbf{20\,{\mathbf{M_{\odot}}}}$ stars} \lSect{IntHydroMix} \pFig{m-X-20AB} In {\Fig{m-X-20AB}} the internal profiles of the most abundant isotopes in a non-rotating star and two rotating $20\,{\mathrm{M}_{\odot}}$ models are compared at core hydrogen exhaustion. Convection causes flat profiles in the innermost few solar masses. Small convective and/or semiconvective regions (similar to {\ModA{D15}} in {\Fig{D15cnv}}) cause steps in the profile above the convective core. In the non-rotating case no mixing occurs in the envelope. In contrast, the rotating models mix thermonuclear processed matter into the envelope. If no inhibition of rotationally induced instabilities by $\mu$-gradients is assumed an extended gradient in helium (along with other species) reaches from the upper edge of the convective core up to the surface ({\ModA{E20}} in {\FIG{A}{m-X-20AB}}). Due to the increase of the mean molecular weight in the whole envelope, as a consequence of helium enrichment, the mass of the hydrogen-depleted core of {\ModA{E20}} is about $1.5\,{\mathrm{M}_{\odot}}$ larger than in the non-rotating case. The dominant rotationally induced mixing process during central hydrogen burning is Eddington-Sweet circulation. It is fast enough to keep the whole star close to rigid rotation (\Sect{MSjtrans}), and thus renders shear instabilities unimportant. The GSF instability remains one to two orders of magnitude less efficient than the Eddington-Sweet circulation. The $\mu$-gradients above the convective core in {\ModA{E20}} (see also {\Fig{E15cnv}}) are strong enough to suppress the occurrence of extended semiconvective structures. The secular shear instability occurs only in a small layer close to the surface, and never contributes significantly to the mixing. \lSect{MixHmu} If $\mu$-gradients {\emph{are}} taken into account for the rotationally induced instabilities ({\ModA{E20B}}; {\FigBB{B}{m-X-20AB}{}{E15Bcnv}}), the $\mu$-gradient which forms at the upper edge of the convective core is not smoothed out fast enough, but instead almost completely chokes off any mixing between core and envelope quite early during core hydrogen burning. Therefore, below $m=10\,{\mathrm{M}_{\odot}}$ the composition of {\ModA{E20B}} remains quite similar to that of {\ModA{D20}}. The higher concentration of carbon in {\ModA{E20B}}, however, shows the occurrence of {\emph{some}} mixing early on. Above the ``barrier'' due to the $\mu$-gradient ($\mu$-barrier) mixing is efficient (see the small slope of the composition profiles in the envelope of {\ModA{E20B}}; {\FIG{B}{m-X-20AB}}), and stronger than for {\ModA{E20}}, since the efficiency for compositional mixing is assumed to be $\fc=1/30$ in {\ModA{E20B}} instead of $\fc=1/100$ for {\ModA{E20}}. The relative contributions of the different rotationally induced mixing processes above the $\mu$-barrier are similar in {\ModB{E20}{E20B}}, except that close to the $\mu$-barrier the GSF instability becomes important in {\ModA{E20B}}. Within the $\mu$-barrier, almost all rotationally induced mixing is suppressed and the mixing is dominated by semiconvection. The secular shear instability is inhibited by the $\mu$-gradient. Strong angular velocity gradients at the boundaries of convective layers cause, in principle, layers where the shear flow can overcome the stabilizing effect of the $\mu$-gradients. However, they are too thin to be resolved in the present calculations. \subsection{Transport of angular momentum} \lSect{MSjtrans} \pFig{m-wjJ53-E15AB} Similar to chemical mixing, the transport of angular momentum depends strongly on the inhibition of rotationally induced mixing by $\mu$-gradients. {\Fig{m-wjJ53-E15AB}} compares the internal angular velocity profile of two $15\,{\mathrm{M}_{\odot}}$ stars ({\ModB{E15}{E15B}}) which were computed with different values of ${f_{\!\mathrm{\mu}}}$ (\Sect{calib}). In {\ModA{E15}}, the difference between surface and core angular velocity remains less than $30\,\%$ during core H-burning. The over-all decrease of the rotation rate by a factor of $\sim3$ is caused by two effects: mass loss from the surface, which carries away $\sim40\,\%$ of the initial angular momentum, and the expansion of the stellar envelope, which increases the total moment of inertia by a factor of $\sim2$. At the same time, the stellar core contracts. The persistence of almost rigid rotation during core hydrogen burning implies transport of angular momentum from the core to the envelope. This is confirmed by {\FIG{C}{m-wjJ53-E15AB}} which shows a decrease of the core specific angular momentum with time (see also {\FIG{E}{m-wjJ53-E15AB}}). Because of its small radial extent, the core contains only a small fraction of the total angular momentum of the star (\FigBb{E}{m-wjJ53-E15AB}{F}). For higher mass loss rates, i.e., for more massive stars, the spin-down (decrease of $\omega$) is dominated by the mass loss, while at lower mass it is dominated by the expansion of the envelope. {\FIG{B}{m-wjJ53-E15AB}} shows that the inhibition of rotational mixing leads to differential rotation during core hydrogen burning. The ratio of the core to envelope angular velocity in {\ModA{E15B}} becomes $\sim4$ at core hydrogen exhaustion. The envelope rotates slightly faster than in {\ModA{E15}} since the star loses only $20\,\%$ of the initial total angular momentum, i.e., about half as much as {\ModA{E15}}. This is due to the lower luminosity of {\ModA{E15B}} during core hydrogen burning --- due to less efficient chemical mixing (\Sect{IntHydroMix}) --- and consequently about $60\,\%$ less mass loss than in {\ModA{E15}}. {\FigBB{D}{m-wjJ53-E15AB}{F}{m-wjJ53-E15AB}} show that in {\ModA{E15B}} the core angular momentum is constant throughout core hydrogen burning. \lSect{J53def} {\FigBb{E}{m-wjJ53-E15AB}{F}} compare the angular momentum distribution of {\ModA{E15}} and {\ModA{E15B}} at various evolutionary stages using the variable ${J\Brak{m}}/m^{5/3}$, with $J\Brak{m}:=\int_0^m j\Brak{m'}{\,\D\!\!\;} m'$. Since for a rigidly rotating body of constant density, $\rho_0$, the angular momentum ${J\Brak{m}}$ enclosed by the mass coordinate $m$ is \begin{equation} J\Brak{m} =\frac{3\omega k}{5}\Brak{\frac{3}{4\pi\rho_0}}^{\!2/3} m^{5/3} \propto m^{5/3} \; \lEq{J53def} \end{equation} the curves in {\FigBb{E}{m-wjJ53-E15AB}{F}} are more or less flat. The evolution of ${J\Brak{m}}$ illustrates the transport of angular momentum throughout stellar evolution. ${J\Brak{m}}$ drops when angular momentum is transported through the mass shell $m$. If no transport angular momentum through the mass shell $m$ occurs, ${J\Brak{m}}$, and also ${J\Brak{m}}/m^{5/3}$, remain constant. Furthermore, following a line of constant $J$ from one evolutionary stage to a subsequent one shows to what mass coordinate angular momentum has been transported in the star during the time between the two evolutionary stages. We will refer more to {\FigBb{E}{m-wjJ53-E15AB}{F}} in the discussion of the angular momentum transport during the later evolutionary phases. \subsection{The influence of the initial rotation rate} \lSect{inflji} \pFig{m-wjJ53-GF15B} {\Fig{m-wjJ53-GF15B}} shows the evolution of angular velocity and specific angular momentum in {\ModB{G15B}{F15B}}, which both contain the inhibition of rotational mixing due to $\mu$-gradients. The latter model initially has three times more angular momentum than the first. While this difference of a factor of three in the rotation rate is conserved in the envelope throughout core hydrogen burning, it becomes much smaller in the cores. The faster rotation of {\ModA{F15B}} sustains the transport of angular momentum out of the core for a longer time than in {\ModA{G15B}}, where the core angular momentum is almost completely conserved (\Fig{m-wjJ53-GF15B}). That is, the angular momentum is less efficiently trapped in the fast rotating {\ModA{F15B}} than in the {\ModB{G15B}{E15B}} (\FigB{m-wjJ53-E15AB}{m-wjJ53-GF15B}). This feedback process leads to a convergence of the core rotation rates. We note already here that this convergence persists during the later burning stages and leads to very similar iron core angular momenta for a wide range of initial rotation rates (cf.~{\Sect{j:preSN}} below). The stronger core angular momentum depletion in faster rotating models occurs simultaneously with rotationally induced mixing across the $\mu$-barrier: The masses of the convective cores at the end of central hydrogen burning are $2.4\,{\mathrm{M}_{\odot}}$, $2.5\,{\mathrm{M}_{\odot}}$, $2.6\,{\mathrm{M}_{\odot}}$, and $2.8\,{\mathrm{M}_{\odot}}$ for {\ModD{D15}{G15B}{E15B}{F15B}}, respectively. However, {\ModA{E15}}, where the $\mu$-barrier was assumed to be inefficient, has a core of about $3.5\,{\mathrm{M}_{\odot}}$. Thus, even for very rapid rotation the assumption of $\mu$-barriers inhibiting rotational mixing strongly restricts the core growth due to rotation. (see also {\FigC{D15cnv}{E15cnv}{E15Bcnv}}). \section{Central helium burning} \lSect{cHeBurn} After core hydrogen exhaustion, the models become red supergiants (except for {\ModA{H12B}} which first burns helium as a as a blue supergiant for some time) and their extended hydrogen-rich envelopes become convective. The pulsational properties of these envelopes have been discussed by {\cite{heg97}} and the evolution of the surface rotation rates, especially during blue loops, by {\cite{HL98}}. In the following, we investigate the evolution of the cores, using the $15\,{\mathrm{M}_{\odot}}$ models as example. \pTab{tauES} The importance of rotation in the post main sequence evolution can be estimated from {\Tab{tauES}}, which compares the Eddington-Sweet time-scale \begin{equation} {\tau_{\mathrm{ES}}}\sim{\tau_{\mathrm{KH}}}\Brak{\frac{\wk}{\omega}}^{\!2} \;,\qquad \wk:=\sqrt{Gm/r^3} \;. \end{equation} {\Cite{Zah92}} in the cores of our {\ModB{E15}{E15B}} during the various burning stages with the respective nuclear time scales. For the amount of differential rotation in our models, the characteristic time-scale for mixing due to the GSF instability (\Sect{GSF}) is comparable to the Eddington-Sweet time-scale. The core hydrogen burning phase is the only one where mixing and nuclear time scale are comparable. During core helium burning, the mixing time scale is one or two orders of magnitude larger than the nuclear time-scale, which may still allow for some effects of rotational mixing. The later phases are too short to allow for any rotationally induced mixing in the cores; note however, that at the core boundaries some effects of rotational mixing may still be possible in case of strong gradients in the angular velocity (cf.~\Sect{hydMix} below). An energetic limit to the amount of mixing due soley to shear instabilities can be obtained by comparing the rotational energy of the core with the potential energy required to lift processed matter from the upper edge of the convective core to the hydrogen-burning shell source {\Cite{Heg98}}. For a typical value of $\omega/\wk=0.05$ and a difference in the mean molecular weight of fully ionized carbon relative to helium of $\sim0.3$ (oxygen would be even heavier), an enrichment of carbon by at most $\simle0.5\,\%$ is possible. This assumes the carbon to be homogeneously distributed throughout the radiative layer and that all the rotational energy of the core is used to supply the buoyancy energy. Note that this limit does not apply to instabilities which tap the energy flux in the star like the Eddington-Sweet circulation. \subsection{Chemical mixing} \lSect{HeMix} \pFig{m-X-DE15AB} In a non-rotating star using the Ledoux criterion for convection (\Sect{Semiconv}) prevents the growth of the convective helium core that would occur if the Schwarzschild criterion were assumed. Instead, several convective regions, separated by semiconvective layers, form above the convective core ({\FigBB{C}{m-X-DE15AB}{}{D15cnv}}). In the rotating models with ${f_{\!\mathrm{\mu}}}=0.05$ (e.g., {\ModA{E15B}} in {\FigBB{B}{m-X-DE15AB}{}{E15Bcnv}}) the shear across the semiconvective layers is not strong enough to overcome the stabilizing $\mu$-gradient, even for the fast rotating {\ModA{F15B}}. If rotationally induced mixing is assumed to be insensitive to $\mu$-gradients (i.e., ${f_{\!\mathrm{\mu}}}=0$; {\ModA{E15}} in {\FigBB{A}{m-X-DE15AB}{}{E15cnv}}) the dynamical shear instability operates in the semiconvective regions and dissolves them, similar to the case of Schwarzschild convection. In this case, the rotational mixing leads to considerably more massive helium cores. The resulting higher burning temperatures in the cores lead to lower central carbon-to-oxygen ratios at core helium exhaustion. An interesting issue is the mixing (and angular momentum transport) in the radiative helium layer between the convective core and the hydrogen-burning shell. If the products of helium burning could be mixed upward into the hydrogen-burning shell, {\emph{primary}} production of {\I{14}{N}} could occur. If hydrogen were transported down into the helium-burning center, a much stronger than normal s-process could result and build up more heavy or neutron-rich elements. On the other hand, strong instabilities in this region could also lead to a significant slowing-down of the core. The dominant mixing process present in this layer is Eddington-Sweet circulation, with some contribution from the GSF instability. During the early stages of core helium burning of models with ${f_{\!\mathrm{\mu}}}=0$ (e.g., {\ModA{E15}}), the secular shear instability dominates slightly over the Eddington-Sweet circulation at the upper edge of the helium core. Towards central helium exhaustion, the mixing is dominated by the GSF instability. In the case of ${f_{\!\mathrm{\mu}}}=0.05$, the secular shear instability is suppressed by $\mu$-gradients. {\FIG{B}{m-X-DE15AB}} illustrates that some mixing occurs during core helium burning: A gradient in {\I{12}C} and {\I{16}O} extends from the convective core up to the edge of the helium core. In this model, the increase in {\I{12}C} or {\I{16}O} is not sufficient to result in any significant primary nitrogen production in the hydrogen burning shell. Even though this effect is not notably more pronounced in the initially faster rotating {\ModA{F15B}} --- due to the convergence of the core rotation rates; cf.~\Sect{inflji} --- or for the different initial masses investigated here, such a primary nitrogen production appears possible in more favorable conditions, e.g., for higher mixing efficiencies or at lower metallicity. In {\ModA{E15}} (\FIG{A}{m-X-DE15AB}) the rotation of the helium core is slower, and the {\I{12}C} and {\I{16}O} gradients are much steeper, leveling off to the CNO equilibrium values a few tenths of a solar mass above the convective core. In the non-rotating {\ModA{D15}} (\FIG{C}{m-X-DE15AB}), no enrichment of {\I{12}C} and {\I{16}O} appears at all above the outermost semiconvective layer of the convective core. Even though the strong entropy gradient at the location of the hydrogen-burning shell suppresses rotational mixing between the helium core and the hydrogen burning shell, some mixing occurs due to the large angular velocity gradient. This can be seen in {\Fig{E15Bcnv}}: The tail of the energy generation rate at the lower bound of the hydrogen-burning shell source in {\ModA{E15B}} penetrates into the helium core, i.e., some hydrogen is mixed downward. Since the protons burn quite fast as they are mixed deeper inside the helium core, they cannot reach the central convective region. However, some protons may survive and get mixed into the convective helium shell later on (\Sect{hydMix}). In {\ModA{E15}} (\Fig{E15cnv}), where the core is rotating slower, and also in the non-rotating {\ModA{D15}} (\Fig{D15cnv}), this feature is not found. \pFig{E15Bcnv} \subsection{Transport of angular momentum} \lSect{HeAngTrans} After core hydrogen exhaustion, the stars undergo a phase of major restructuring as the core contracts and the envelope expands. This leads to a spin-up of the core ({\FigDdd{A}{m-wjJ53-E15AB}{B}{A}{m-wjJ53-GF15B}{B}}) and a spin-down of the envelope. At the same time, the convective envelope grows in mass and its bottom approaches the helium core. A steep rise in the specific angular momentum occurs at the bottom of the rigidly rotating envelope that persists throughout core helium burning and beyond ({\FigDdd{C}{m-wjJ53-E15AB}{D}{C}{m-wjJ53-GF15B}{D}}). The entire helium core stays close to rigid rotation during central helium burning ({\FigDdd{A}{m-wjJ53-E15AB}{B}{A}{m-wjJ53-GF15B}{B}}). Up to core helium exhaustion, the specific angular momentum of the helium core drops appreciably with time (\FigB{m-wjJ53-E15AB}{m-wjJ53-GF15B}). Three processes contribute to this effect. First, angular momentum is removed from the core during the star's restructuring phase between core hydrogen depletion and helium ignition. Second, the core grows in mass due to hydrogen shell burning and engulfs regions of lower specific angular momentum (\FigBb{C}{m-wjJ53-E15AB}{D}). The reasons for the low specific angular momentum above the core are secular shear instabilities, the {\emph{first dredge-up}}, and short-lived convective regions which temporarily extend down to mass coordinates smaller than the final helium core mass. The regions of outwards decreasing specific angular momentum are not Solberg-H{\o}iland unstable due to strong stabilizing entropy and composition gradients. Third, some angular momentum is transported from the helium core into the envelope through the hydrogen-burning shell. The models of the ``\Mod{B}'' series lose less angular momentum during the restructuring phase because of the inhibiting effect of the $\mu$-gradients, but more during central helium burning, due to their considerably faster rotation. The relative loss of angular momentum in {\ModC{G15B}{E15B}{F15B}} during helium burning increases with the initial amount of angular momentum left at the end of central hydrogen burning. Consequently all three models end up with very similar core angular momenta and rotation rates (\Fig{m-wjJ53-GF15B}), about three times that of {\ModA{E15}} (\Fig{m-wjJ53-E15AB}). \section{Comparison with previous work} \lSect{CmpOther} In contrast to {\cite{KMT70}}, who investigated rapidly rotating $9\,{\mathrm{M}_{\odot}}$ stars ($v\simgr400\,{\km\,\Sec^{-1}}$), our models do not become secularly unstable at the end of central helium burning, since, according to our assumptions, $\mu$-barriers are less efficient in suppressing angular momentum transport compared to {\cite{KMT70}}. {\cite{ES78}} followed the evolution of $7\,{\mathrm{M}_{\odot}}$ and $10\,{\mathrm{M}_{\odot}}$ stars with a ZAMS rotational velocity of $\sim200\,{\km\,\Sec^{-1}}$ using essentially the same method as in the present work, except for some improvements in the input physics of the individual processes applied here {\CITE{{\Sect{MixInst}} and}{ES78}}. They used the Schwarzschild criterion for convection, however, and did not include mass loss. In their work, the $\mu$-barrier above the convective hydrogen-burning core suppressed mixing and transport of angular momentum almost completely. Therefore, their stellar cores lose very little angular momentum during central hydrogen burning. Although we use ${f_{\!\mathrm{\mu}}}<1$ and the inhibiting effect of the $\mu$-gradients is smaller, a similar $\mu$-barrier forms during central hydrogen burning. However, we obtain some mixing between the core and the envelope early during core hydrogen burning, some angular momentum loss from the core to the envelope, and in most cases some enrichment of the surface with H-burning products. In an earlier work, {\cite{ES76}} disregarded rotationally induced angular momentum transport, but imposed various rotation laws. In this case an even more extreme result was obtained: all models reached critical rotation before carbon ignition. {\cite{ery94}} considered turbulent diffusion according to {\cite{Zah83}} in their computation of a rotating $20\,{\mathrm{M}_{\odot}}$ stars with a metallicity of $Z=0.008$. They found a surprisingly large surface {\I{14}N} enrichment at the end of core hydrogen burning of more than $2\,\%$ by mass. Since the CNO cycle conserves the total mass of the CNO isotopes, this result appears implausible and cannot be reproduced in the present work. {\cite{MM97}} used a prescription for the Eddington-Sweet circulation according to {\cite{Zah92}}, and a modified Richardson criterion to account for thermal effects {\Cite{Mae95,MM96}}. They computed the hydrogen-burning evolution of stars from $9\,{\mathrm{M}_{\odot}}$ to $60\,{\mathrm{M}_{\odot}}$ and found a strong inhibiting effect of the $\mu$-gradients on the rotational mixing, which resulted in stronger differential rotation at core hydrogen exhaustion than found in our models of the ``\Mod{B}'' series. Their models did not show any surface enrichment of helium. These models were superseded by those of {\cite{Meyn97}}, who computed the main sequence evolution of $20\,{\mathrm{M}_{\odot}}$ and $40\,{\mathrm{M}_{\odot}}$ stars, using improved physics of rotationally induced mixing as discussed by {\cite{MM97}}, and {\cite{Mae97A}}. The inhibiting effect of $\mu$-gradients on shear mixing and Eddington-Sweet circulation was strongly reduced in the new formulation. {\citeauthor{Meyn97}}'s rotating $20\,{\mathrm{M}_{\odot}}$ model showed a larger envelope helium enrichment than a comparable models of {\ModA{E20}}, and a similar mass of the hydrogen-depleted core. The physics used in {\cite{Meyn97}} has been revised again by {\cite{MZ98}} for a more consistent treatment of $\mu$-gradients. Models with this prescription are not yet available. {\cite{tal97}} followed the main sequence evolution of $9\,{\mathrm{M}_{\odot}}$ stars with ZAMS rotation rates of $100\,{\km\,\Sec^{-1}}$ and $300\,{\km\,\Sec^{-1}}$ until end of central hydrogen burning, using the prescription for the Eddington-Sweet circulation by {\cite{Zah92}}. The helium enrichment in the envelope showed a smooth profile, similar to our {\ModB{E08}{E10}}. At core hydrogen exhaustion, their models showed steep composition gradients close to the stellar surface. This may have resulted from the low mass loss assumed in their calculation in combination with inefficient mixing close to the surface. With a slightly larger mass loss rate, their rapidly rotating model would have a much stronger surface enrichment. Due to the downward advection of angular momentum by meridional circulation in the theory of {\cite{Zah92}}, {\cite{tal97}} found a somewhat stronger envelope differential rotation compared to our {\ModB{E08}{E10}}, but a comparable one to, e.g., {\ModA{E12B}}. We conclude that this downward advection is not a strong effect, which may justify its neglect in the present work. Summarizing, the prescription for rotationally induced mixing used in the model of {\cite{KMT70}} corresponds roughly to ${f_{\!\mathrm{\mu}}}=\fc=\infty$ in terms of the present formulation. {\cite{ES78}} used about ${f_{\!\mathrm{\mu}}}=\fc=1$. Neither work obtained any surface enrichment during core hydrogen burning due to the strong inhibiting effect of $\mu$-gradients {\Cite{MM97}}. The recent picture of interacting Eddington-Sweet circulation, anisotropic turbulence and shear instabilities {\Cite{CZ92,MZ98}} has been continuously improved in the last years {\Ce{USS96,Mae97A,TZ97,tal97}}. The most recent work in this series, {\cite{MZ98}}, includes an improved treatment of compositional gradients, but this type of description for rotationally induced mixing is complex, computationally expensive, and has not yet been successfully tested for post-hydrogen burning stars. However, the results obtained in earlier work {\Cite{tal97,Meyn97}} for massive main sequence stars are not significantly different from those of the present work. \section{Late evolution until core collapse} \lSect{LateEv} After core helium exhaustion, the carbon-oxygen core ({CO} core) contracts and subsequently phases of carbon, neon, oxygen, and silicon central convective and shell burning follow inside this core. {\Tab{FinalModels}} gives some key parameters of the final models: the final mass of the star, the masses of the helium, {CO}, and iron cores, and the angular momenta contained in theses cores. For the iron core additionally the average specific angular momentum is given. {\ModF{D10}{D12}{E08}{G12B}{E12B}{F12B}} develop degenerate neon-oxygen cores and central neon burning starts off-center. Due to the computational difficulties (and expense) these models were not followed until core collapse, but stopped during neon or oxygen shell burning or even before neon ignition. In {\ModA{D10}} even carbon burning ignites off-center. {\ModB{D10}{E08}} experience a dredge-up of almost the entire helium shell by the convective envelope. Therefore the final helium core is small and has little angular momentum. The remaining helium shell above the {CO} core at the point where the calculation is ended is only a few hundreths of a solar mass. The masses of the helium cores before the dredge-up are $2.25$ and $2.1\,{\mathrm{M}_{\odot}}$, respectively. {\ModA{E25}} loses its hydrogen-rich envelope during central helium burning and becomes a Wolf-Rayet star. Strong Wolf-Rayet mass loss sets in and further decreases the mass of the star. It ends up with only $5.45\,{\mathrm{M}_{\odot}}$ at the time of core collapse and very little angular momentum. Note that in {\Tab{FinalModels}} the masses of some of the helium cores in the non-rotating models are apparently larger than those of the slowly rotating models of the ``\Mod{B}'' series. This is an artifact due to the criterion used to measure the mass of the helium core. We define the helium core by the mass coordinate at which the hydrogen mass fraction drops below $\Ep{-3}$. In the rotating models of the ``\Mod{B}'' series, the hydrogen gradient at the top of the helium core is significantly shallower. If instead a hydrogen mass fraction of $\Ep{-2}$ is chosen, the helium core masses are similar or even larger for the rotating models. Note that the size of the {CO} core is defined in a similar way: by the mass coordinate at which the helium mass fraction drops below $\Ep{-3}$. \subsection{Chemical mixing in the helium shell} \lSect{hydMix} After core helium exhaustion, the {CO} core contracts and the burning of helium continues in a shell. At the same time, the outer layers of the helium-rich shell cool down and the hydrogen shell source goes out. Since this implies a reduced entropy barrier, rotationally induced mixing through the hydrogen-helium interface can now operate more efficiently. The protons which are mixed downward into the helium shell do not burn immediately. When later the helium-burning driven convective shell extends upwards, it dredges these protons down into the hot, helium-burning layers (see {\Fig{E15Bcnv}}). This mechanism can open new channels of nucleosynthesis. This will be investigated in more detail in forthcoming papers {\CS{ for a first report}{lan99:CosV}}. \subsection{Chemical mixing inside the {CO} core} During the final remaining stellar burning phases, rotational mixing inside the {CO} core is unimportant. The strongest instabilities are again the Eddington-Sweet circulation and the GSF instabilities, at about same order of magnitude. From {\Tab{tauES}} it can be seen that their time scale is too long in order to be significant. Also, the mixing of {\emph{traces}} of material into regions of neighboring burning phase is not expected to introduce qualitatively new nucleosynthesis channels, since all abundant nuclear species in one burning phase are anyway present in the neighboring one (e.g., mixing traces of oxygen into neon burning is not exciting). In the fast rotating cores of the models of the ``{\Mod{B}}'' series, secular shear instabilities arise above several of the central and shell convection zones for a limited time, but they do not become efficient enough to cause any noticeable mixing. \subsection{Transport of angular momentum} \lSect{lateJtrans} As for the chemical mixing, rotational mixing cannot effectively remove angular momentum from the core during the late burning stages. In particular, transport is too inefficient to keep the {CO} core in rigid rotation. Strong differential rotation occurs ({\FigDdd{A}{m-wjJ53-E15AB}{B}{A}{m-wjJ53-GF15B}{B}}). At this time, the only instability capable of enforcing rigid rotation is convection. Since the radii of the lower boundaries of the major convection zones of carbon, oxygen, and silicon burning are much smaller than that of their upper edges, large differences in the specific angular momentum exist between the bottom and the top of the convection zone. Thus, angular momentum is mainly carried outwards. The typical signature of such a convection zone is a steep drop of the specific angular momentum at its bottom, accompanied by a large increase at its top (\FigC{m-wjJ53-E15AB}{m-wjJ53-GF15B}{jj15B}). Convection zones that subsequently overlap can transport angular momentum efficiently over scales larger than their individual extent. This is most efficient when the lower boundary of a convective shell overlaps with the upper boundary of a preceding convection zone. For the models investigated in this work, such an overlap occurs rather infrequently ({\App{convDiag}}). Subsequent shells, which are driven by nuclear burning, tend to form their lower boundary at the upper edge of a preceding shell, where the fuel for their burning is not yet depleted. The most prominent example of this is the sequence of carbon-burning shells (e.g., {\Fig{E15cnv}}). Exceptions occur only for some of the late carbon burning shells, and for the oxygen burning shells in {\ModB{G15B}{E15B}}. Convective angular momentum transport does not operate across the boundary of the {CO} core. These cores retain their angular momentum after core helium exhaustion. Some redistribution, mainly due to convection, occurs inside the cores. For example, {\FIG{E}{m-wjJ53-E15AB}} shows that after core helium exhaustion in {\ModA{E15}}) no angular momentum is transported through the shells at $m\approx 3.4\,{\mathrm{M}_{\odot}}$ and $m\approx5.1\,{\mathrm{M}_{\odot}}$, i.e., the boundaries of the helium and the {CO} core, respectively. In models with more rapidly rotating cores, ({\ModC{G15B}{E15B}{F15B}}; {\FigCCc{F}{m-wjJ53-E15AB}{E}{m-wjJ53-GF15B}{F}}), the helium core does lose some angular momentum, even though its upper boundary (at $m\approx3.7\,{\mathrm{M}_{\odot}}$) remains a significant barrier for angular momentum transport as indicated by the spike in {\FIG{F}{m-wjJ53-E15AB}}. The loss of angular momentum from the helium core is correlated with the mixing of hydrogen into the helium shell described in {\Sect{hydMix}}. No significant angular momentum was transported across the boundary of the {CO} core in any of the models ({\FigDdd{E}{m-wjJ53-E15AB}{F}{E}{m-wjJ53-GF15B}{F}}). \subsection{Stability to triaxial deformations} \pFig{m-phi} As described in {\Sect{ModRot}}, the approximations employed in this work are limited to slow rotation in the sense that no triaxial deformation appear. In the KEPLER code, the influence of the centrifugal forces on the structure is completely neglected. However, when models from calculations with STERN, where centrifugal forces {\emph{are}} included, are continued by KEPLER at a central temperature of $\Ep9\,{\mathrm{K}}$, the evolution usually proceeds smoothly, i.e., these forces {\emph{are not}} important at this late stage of evolution. On the ZAMS, the rotational energy ${E_{\mathrm{rot}}}$ of the star is negligible in comparison to its gravitational binding energy ${E_{\mathrm{pot}}}$ for all models, even for those which are close to critical rotation at their surface (${E_{\mathrm{rot}}}/\abs{{E_{\mathrm{pot}}}}\ll1\,\%$). However, in the course of their evolution the stellar models contract and --- as outlined in {\Sect{lateJtrans}} above --- the transport of angular momentum out of the core is inhibited or slow, with the consequence of rapidly rotating cores ({\FigDdd{A}{m-wjJ53-E15AB}{B}{A}{m-wjJ53-GF15B}{B}}, and {\Tab{jev}}). For local angular momentum conservation in a shell with given specific angular momentum $j$, the ratio of angular velocity to Keplerian angular velocity scales as \begin{equation} \frac{\omega}{\wk}=\frac{j}{k\sqrt{Gmr}}\sim r^{-1/2} \;,\qquad k\approx2/3 \lEq{awawk} \;. \end{equation} The ratio of the specific rotational energy to the gravitational potential is then given by \begin{equation} \frac{{e_{\mathrm{rot}}}}{{\phi_{\mathrm{grav}}}}=\frac{1}{2}\Brak{\frac{\omega}{\wk}}^{\!2} \;. \end{equation} This ratio is displayed for several $15\,{\mathrm{M}_{\odot}}$ pre-collapse models in {\Fig{m-phi}}. A uniformly rotating, self-gravitating, incompressible, and inviscid fluid (McLaurin spheroid) becomes secularly unstable to triaxial deformations when the ratio of rotational to gravitational potential energy \begin{equation} \frac{{E_{\mathrm{rot}}}}{{E_{\mathrm{pot}}}}=\frac{1}{2}\int_0^m\omega^2(m'){\,\D\!\!\;} m' \left/ \int_0^m\wk^2(m'){\,\D\!\!\;} m'\right. \end{equation} exceeds $\sim0.1375$ {\Ce{OB73,Tas78}}. If this ratio exceeds $\sim0.26$, the object becomes dynamically unstable to non-axisymmetric instabilities and fission may occur {\Cite{OT69,OB73}}. The stars simulated in the present work are well below these limits even at the pre-collapse stages (${E_{\mathrm{rot}}}/{E_{\mathrm{pot}}}\simle30\,\%$ of the critical value) and therefore no triaxial instabilities arise. \section{Angular momentum prior to core collapse} \lSect{j:preSN} \pFig{jj15B} \pFig{m-J53-presn-GEF} Our model sequences are terminated at the onset of core collapse, defined by the infall velocity inside the iron core exceeding $9\E{2}\,{\km\,\Sec^{-1}}$. At this stage of evolution the investigated stars typically have central densities of $\simle1\E{10}\,\gccm$. From the previous discussions it is clear that the distribution of angular momentum in the star at onset of core collapse strongly reflects its recent convective structure. {\Fig{jj15B}} shows the distribution of the specific angular momentum at the pre-collapse stage of $15\,{\mathrm{M}_{\odot}}$ stars with different initial rotation rates {\ModCx{G15B}{E15B}{F15B}}. These three models show a very similar final angular momentum distribution (cf. also {\FIG{A}{m-J53-presn-GEF}}), due to a similarity entire chemical structure. The reason for this is the convergence of the core rotation rates, i.e. their independence from the initial rotation rates, already during hydrogen and helium burning, as outlined in {\Sect{inflji}}. In contrast, {\ModA{E15}} has much less angular momentum left in the core (see also {\Tab{FinalModels}}). It grows a larger helium and {CO} core due to the lack of sensitivity to $\mu$-gradients. \pTab{FinalModels} The total angular momentum in the final models is dominated by that of the envelope (\Tab{FinalModels}). {\ModC{G15B}{E15B}{F15B}} show that for initially faster rotation, a slightly larger helium core results (\Tab{FinalModels}) and therefore the stars become more luminous. This in turn causes more mass and angular momentum loss, which can, for the rapid rotators or for more massive stars, decreases the total angular momentum by a larger factor ({\FigBB{B}{m-J53-presn-GEF}{B}{m-J53-presn}}). \pFig{m-J53-presn} \pFig{m-J53-presn4} At hydrogen ignition, the total and even the mean specific angular momentum of models with a given surface rotational velocity are larger for larger initial masses (\Tab{FinalModels}). On the contrary, the final total angular momentum decreases for larger initial mass ({\Tab{FinalModels}} and {\Fig{m-J53-presn}}). This trend is only interrupted between $12$ and $15\,{\mathrm{M}_{\odot}}$ since our models with initial masses of $12\,{\mathrm{M}_{\odot}}$ or less undergo a blue loop during core helium burning which leads to an additional strong angular momentum loss {\Cite{HL98}}. As in the limit of vanishing mass loss, the angular momentum of our models is conserved, the decrease of the total angular momentum for higher initial masses is solely due to the increase of the mass loss rate for larger initial masses. The total angular momentum of the helium and {CO} cores increases with the the initial stellar mass of our models ({\Tab{FinalModels}}). However, this trend is much weaker for the specific angular momentum of the helium cores, the specific angular momenta of the {CO} cores even decreases a little with increasing initial mass. This illustrates that angular momentum transport from the core into the envelope is stronger for larger cores. Finally, we find that the specific angular momenta of the iron cores are rather insensitive to the initial mass and rotation rate (\Tab{FinalModels} and {\Fig{m-J53-presn4}}), due to the convergence of the core rotation rates discussed in {\Sect{inflji}}. In the models with ${f_{\!\mathrm{\mu}}} = 0$, angular momentum transport was efficient and final values of ${j_{\!\isofont{Fe}}}\simeq 6\E{15}\,{\cm^2\,\Sec^{-1}}$ are found. The value for {\ModA{E25}} is significantly lower since its {CO} core was spun down in a Wolf-Rayet phase. The models with ${f_{\!\mathrm{\mu}}} = 0.05$ all end up with ${j_{\!\isofont{Fe}}}\simeq 1.2\E{15}\,{\cm^2\,\Sec^{-1}}$. Note that, unless the iron core that forms after central silicon burning is already large enough to collapse, one or more subsequent phases of silicon shell burning occur until the critical iron core mass is exceeded. The sizes of these shells depend on the details of the preceding evolution. As a result, the iron core mass does not necessarily increase monotonically with initial mass or rotation. For example, the iron core of {\ModA{D20}} is larger than those of the $20\,{\mathrm{M}_{\odot}}$ models of the ``\Mod{B}'' series. \section{Implications for young pulsars and supernovae} \lSect{youngPSR} \pTab{jev} {\Tab{jev}} shows, for times during the evolution, the specific angular momentum contained in the innermost $1.7\,{\mathrm{M}_{\odot}}$ (the mass of the iron core at core collapse) for {\ModA{E20}}. Due to the continuous contraction of the central region of the star, it spins up and gets closer to critical rotation {\Eqx{awawk}}. If the pre-collapse value of the specific angular momentum is applied to a neutron star with an assumed radius of $12\,\km$, it would rotate with $90\,\%$ of Keplerian rotation (\Tab{jev}). {\ModA{E20}} has the largest iron core mass of all our models (\Tab{FinalModels}), and a lower core specific angular momentum than the models computed with ${f_{\!\mathrm{\mu}}} = 0.05$. Those models have even much more angular momentum in the collapsing iron core than a neutron star can possibly carry ($\omega/\wk\propto jm^{-1/2}$). This much angular momentum would certainly be important in the dynamics of core collapse, and it is expected that significant deviations from spherical symmetry will arise {\Cite{Ims95,aks97,ZM97,RMR98}}. \subsection{Comparison with observed young pulsars} \lSect{pulsCmp} \pTab{jNS} At $90\,\%$ of Keplerian angular velocity, the neutron star which might form in the collapse of the iron core of {\ModA{E20}} would have a rotation period of $1\,{\mathrm{m}\Sec}$ (\Tab{jev}). In {\Tab{jNS}} the periods of the four known young neutron stars associated with supernova remnants {\Cite{mas98}} are given along with their specific angular momentum (with the same assumption regarding moment of inertia as above). Comparing this to the specific angular momentum in the iron cores found in the pre-collapse models in {\Tab{FinalModels}}, we see that the iron cores of our models have roughly $\sim20$ to $100$ times more specific angular momentum than found in these neutron stars. Triaxial deformations and gravitational radiation would result --- even during the explosion. Still it might be expected that the resulting neutron stars would spin much faster than observed. However, the observed ``young'' neutron stars have ages of several hundred years. They might have spun much faster immediately after their formation. In fact, it has been proposed recently that rapidly rotating hot neutron stars are spun down on a time-scale of one year by r-mode oscillations and accompanying emission of gravitational radiation. These oscillations are supposed to cease at spin periods compatible with those observed in the young neutron stars {\Cite{LOM98,owe98}}. Alternatively, an important angular momentum transport mechanism might be missing in our models (see also {\Sect{OtherInst}}). {\cite{SP98}} have assumed, without computing detailed models, that the winding up of weak magnetic fields by differential rotation can cause enough Maxwell stress to keep the entire star in uniform rotation until the end of central carbon burning. This scenario approaches the problem of the young neutron star periods from the other side: It implies initial spin periods of $\sim100\,{\mathrm{s}}$. Since this is much larger than observed, they employed off-center ``kicks'' during the supernova explosion to spin them up to the observed rotation rates. This scenario is speculative at present, since neither the evolution of magnetic instabilities in the stellar interior nor the neutron star kicks have been adequately investigated. \subsection{Formation of Kerr black holes?} \pFig{m-jLSO-E15AB} If the large angular momenta obtained for the iron cores in this work pose a problem for pulsars, they are very favorable for the collapsar model for $\gamma$-ray bursts {\Cite{Woo93}}. If the cores of the stars would collapse to a black hole, the angular momentum calculated here would be enough to support matter in a stable disk outside {\Cite{ST83, Nov97}}. This is indicated in {\Fig{m-jLSO-E15AB}}, where the distribution of the specific angular momentum at the pre-collapse stage of the two $20\,{\mathrm{M}_{\odot}}$ {\ModB{E20}{E20B}} is shown. Thin dashes and dash-dotted lines indicate the specific angular momenta of the last stable orbit around a non-rotating and a maximum rotating black hole with a mass equal to the mass coordinate. If the matter in the star has more angular momentum than necessary to get into the last stable orbit, an accretion disc must form, and efficiently transform gravitational binding energy into heat, up to $42.3\,\%$ of the rest mass for a maximum rotating black hole {\Cite{ST83, Nov97}}. Note that the specific angular momentum displayed in {\Fig{m-jLSO-E15AB}} is the latitudinal average over a shell. Its actual value at the equator is higher than that by $50\,\%$, while it is zero at the pole. Therefore matter might fall in almost freely along the rotation axis, while it hits the centrifugal barrier at the equator. In case a prompt supernova explosion fails and a black hole forms instead of a neutron star, this might be a mechanism for an efficient energy source for supernovae or even a $\gamma$-ray burst {\Cite{Woo93,PWF98,MW99}}. \section{Summary and conclusions} \lSect{SumConcl} We have presented the first complete numerical simulation of the evolution of rotating stars from the ignition of nuclear burning until the supernova stage. Emphasis has been placed on the modification of the evolution induced by rotation. This includes an examination of the transport processes responsible for redistributing each angular momentum and composition and the resultant changes that occur in the stellar structure and nucleosynthesis. The distribution of angular momentum in the presupernova stage is of particular interest. Two different one-dimensional hydrodynamic stellar evolution codes were modified to include angular momentum as a new local variable. The effects of centrifugal forces on the stellar structure were treated in latitudinally averaged way. Rotationally induced instabilities were included (\Sect{MixInst}): secular and dynamic shear instabilities, the Solberg-H{\o}iland instability, the Eddington-Sweet circulation, and the Goldreich-Schubert-Fricke instability. The uncertain parameters of rotationally induced mixing were calibrated using observational constraints on the surface abundances (\Sect{calib}). Observed surface enrichments with CNO-processed matter were reproduced for stars in the mass range from $5\,{\mathrm{M}_{\odot}}$ to $60\,{\mathrm{M}_{\odot}}$, for typical initial stellar rotation rates. Stellar mass loss and its dependence on the surface rotation rate were also taken into account (\Sect{MassLoss}). The evolution of stars of approximately solar composition in the mass range from $10\,{\mathrm{M}_{\odot}}$ to $25\,{\mathrm{M}_{\odot}}$ was modeled up to iron core collapse, the immediate presupernova stage. Models that used different assumptions regarding the stabilizing effect of gradients in the mean molecular weight on rotationally induced instabilities were computed and compared. Observations indicate that gradients in the mean molecular weight inhibit rotationally induced mixing much less than in the pioneering models of {\cite{ES78}}. This conclusion is also supported by recent investigations of the physics of meridional circulations, shear instabilities, and semiconvective mixing {\Cite{MZ98}}. \subsection{Internal stellar structure} During central hydrogen burning, the products of the burning are mixed into the stellar envelope and new fuel is supplied to the convectively burning stellar core by rotationally induced mixing. Since this mixing proceeds on a time-scale comparable to the thermonuclear time-scale of hydrogen burning, a gradient of processed matter builds up inside the radiative envelope. The processed matter has a higher mean molecular weight, $\mu$, than the pristine matter of the star, and therefore a gradient of the mean molecular weight results. If rotationally induced mixing occurs by processes that depend sensitively upon these gradients, they act as a barrier ($\mu$-barrier), and mixing between the core and the envelope is inhibited. Exactly when this inhibition becomes important depends on the initial angular momentum of the star. The amount of mixing that occurs between the core and the envelope is affected accordingly. Mixing inside the envelope also increases for larger initial angular momentum, since the dominant mixing process, Eddington-Sweet circulation, has an efficiency that increases as the square of the stellar rotation rate. As the evolution of the star proceeds to later stages, the time-scale for rotationally induced mixing becomes too long in comparison to the evolutionary time-scales to constitute an important source of large-scale mixing. Also, the mixing is not able to dissolve the molecular weight barrier which forms in the core during central helium burning. In general, rotationally induced mixing does not strongly affect the stellar structure after central helium ignition. The evolution of the star from this point until core collapse is similar to that of a non-rotating star of same structure at this time, except for the differences in the nucleosynthesis discussed below. For models where rotationally induced mixing is assumed to be insensitive to gradients in the mean molecular weight, no $\mu$-barrier inhibits the mixing. This affects the mixing between the core and the envelope during central hydrogen burning. The $\mu$-barrier in the superadiabatic part of the core during central helium burning is eroded by shear instabilities. Consequently, the convective core can grow unhindered. As a result, the helium cores are more massive, corresponding to non-rotating stars with about $25\,\%$ higher initial mass. Inside this helium core, the {CO} core is also larger than that of a non-rotating star with same helium core mass. Towards the end of central helium burning, fresh helium is mixed into the convective core both by the continuing growth of this core and by rotationally induced mixing. The fresh helium preferentially converts carbon into oxygen instead of producing new carbon by the triple-alpha process. This reduces the carbon abundance in the core. Except for this, the effect of rotationally induced mixing is small after helium ignition for the reason outlined above. In particular, also in this case, the hydrogen burning shell constitutes an efficient barrier for mixing processes --- indeed even more efficient, because the core rotates slower as in the case where $\mu$-gradients were considered (see below). A consequence of the enlarged cores is that the limit on the initial stellar mass for core collapse supernovae is somewhat smaller for higher initial rotation rates. \subsection{Angular momentum} At central hydrogen ignition, the stars establish almost uniform rotation. If a molecular weight barrier forms as hydrogen burning progresses, angular momentum is trapped inside the core and differential rotation results, with up to a factor of $\sim3$ variation in the rotation rate between the core and the envelope. If $\mu$-barriers are unimportant for the rotationally induced mixing, the stars stay close to rigid rotation until the end of central hydrogen burning. Since this barrier forms later in the faster rotators, stars having different initial rotation rates may end up with similar specific angular momenta in the core at the end of central hydrogen burning (\FigCCc{D}{m-wjJ53-E15AB}{C}{m-wjJ53-GF15B}{D}). Due to angular momentum transport during core helium burning, they may become even more similar in the pre-collapse stage ({\Fig{jj15B}}). Some angular momentum gets lost from the core during the restructuring that occurs after core hydrogen exhaustion, but during central helium burning the hydrogen-burning shell constitutes an efficient barrier that inhibits the transport of angular momentum out of the core. Even so, the average angular momentum of the core may decrease somewhat, since it grows into regions with lower specific angular momentum on top of it. The helium core itself stays close to uniform rotation (\FigBb{A}{m-wjJ53-E15AB}{B}). During central helium burning rotationally induced mixing processes already become slow compared to the evolution, and after core helium exhaustion they do not cause any relevant transport of angular momentum. Only convective processes are rapid enough to notably redistribute angular momentum during the late stages of stellar evolution. Within the assumptions made, rigid rotation results in convective regions, transporting angular momentum from their bottom to their top. Subsequent phases of convective central and shell burning stages give some outward transport of angular momentum inside the carbon-oxygen core. Since none of the convective shells penetrates through the outer boundary of the {CO} core, the angular momentum remains trapped inside (\FigCCc{F}{m-wjJ53-E15AB}{E}{m-wjJ53-GF15B}{F}). The outer boundary of the helium core constitutes a similar barrier. This has interesting consequence for the final angular momentum in the core. First, the different convective burning shells leave their fingerprint not only in the chemical composition, but also in the angular momentum distribution: a spiky profile results at the onset of core collapse (\Fig{jj15B}). The high peaks correspond to the upper edges of the most recently active convection zones and the deep valleys to their bottoms. Shells of similar composition tend to rotate almost rigidly. Second, even the slowest rotating core of the Type~II supernova progenitor stars considered here would result in a neutron star rotating close to break-up if angular momentum were conserved during the collapse. This is not necessarily in contradiction with observations of young neutron stars in supernova remnants, even though the fastest of these rotates much slower. These pulsars are already hundreds of years old, and recent theoretical investigations of hot, newly born neutron stars indicate they may spin down to the observed rotation rates within about a year by emitting gravitational waves {\Cite{LOM98}}. The electro-magnetic radiation emanating from pulsars is trapped inside the supernova ejecta during that time, but the gravitational radiation of these very young neutron stars might become detectable in the future {\Cite{owe98}}. \acknowledgments We are grateful to T.A.~Weaver for helpful discussions and aid with KEPLER. This work was supported by the Deutsche Forschungsgemeinschaft through grants La~587/15 and La~587/16, by the the National Science Foundation (AST 97-31569), and by the Alexander von Humboldt Foundation. AH was, in part, supported by a ``Doktorandenstipendium aus Mitteln des 2.~Hochschulprogramms''.
\section{Introduction} \bigskip The possibility of analyzing narrow $s$-channel resonances is considered to be one of the most important strengths of muon colliders, at present under consideration by different laboratories \cite{mumucolliders}. Most of the content of this note is indeed addressed to physics of particular relevance to future muon colliders (for general reviews see \cite{physics}). The importance of analyzing accurately an $s$-channel resonance at a lepton collider and the physical interest to be assigned to the extracted information are, of course, not new subjects. Much literature has in fact been devoted to the problem of the $Z$ resonance shape at electron-positron colliders and to the implications of its accurate measurement. Many of the special issues that arise when studying a narrow $s$-channel resonance, such as a light SM Higgs boson, have also been considered \cite{bbgh}. However, these latter studies assumed that the beam energy profile is perfectly known. If a resonance is so narrow that its width is smaller than or comparable to the beam energy spread, uncertainty in the beam's energy profile can introduce substantial errors in the experimental determinations of the resonance parameters. One of the main goals of the present paper is to assess these errors. To this end, we shall present a detailed discussion of methods for studying a narrow $s$-channel resonance or systems of nearly degenerate resonances at a lepton collider, and consider applications to particular cases. We carefully analyze the errors in the measured resonance parameters (such as the total width and partial widths) that arise from the uncertainty in the energy spread of the beam. Procedures for reducing this source of errors are studied and experimental observables with minimal sensitivity to beam energy spread are emphasized. Our study will assume that the beam energy spread is independently measured, as will normally be possible. Expectations and procedures for this determination at a muon collider will be noted. Although the parameters of a narrow $s$-channel resonance can be determined with minimal sensitivity to beam energy spread by measuring the total Breit-Wigner area, the peak total cross section and the cross sections in different final state channels, these measurements are not all easily performed with good statistical accuracy. Generally, a more effective method for determining resonance parameters is to scan the resonance using specific on- and off-resonance energy settings. In this paper, we also discuss the possibly superior new method introduced in Ref.~\cite{ratio} in which one operates the collider always with center of mass energy equal to the resonance mass but uses two different beam energy spreads: one smaller than the resonance width and one larger. All of these possibilities will be compared. The most immediate application is to the study of a light SM-Higgs, where, as already known \cite{bbgh,gunmumu,higgsmumu}, accurate measurements at a muon collider might make it possible to distinguish a standard model Higgs from the lightest Higgs of a supersymmetric model. We shall also analyze in detail resonant production of the lightest pseudogoldstone boson of dynamical electroweak symmetry breaking models. Following that, we consider two almost coincident vector resonances of the Degenerate BESS model \cite{DBESS} when their mass splitting is larger than the beam energy spread. We shall then examine the case of two nearby resonances with mass splitting much smaller than their average mass, assuming that the energy spread of the beam is of the same order as the mass splitting. Different possible procedures for reducing the errors in the determination of the physical parameters will be examined. As an application we shall discuss such a situation in Degenerate BESS. In Section 2 we present the analysis for a narrow resonance and in Section 3 the applications we have mentioned for this case. Section 4 gives the analysis for nearly degenerate resonances, and Section 5 a corresponding application. \bigskip \section{Analysis for a narrow resonance} \label{general} \bigskip In this Section we will discuss several ways of determining the parameters of a narrow resonance at a lepton collider. For a given decay channel, an $s$-channel resonance can be described by three parameters: the mass, the total width and the peak cross section. These quantities can only be accurately measured if the beam parameters are very well known. Measurement of the mass requires precise knowledge of the absolute energy value. Measurement of the width and cross section requires excellent knowledge of the energy spread of the beam and the ability to determine the difference between two beam energy settings with great precision. All of these beam parameters must be known with extraordinary accuracy in order to study a very narrow resonance. For example, consider a light SM Higgs boson. In \cite{bbgh} it was found that the beam energy $E_{\rm beam}$ must be known to better than 1 part in $10^6$ and the beam energy spread $\Delta E_{\rm beam}$ must be smaller than 1 part in $10^4$ in order to scan the resonance and determine its width and other parameters. Below, we shall find that errors in these parameters due to uncertainty in $\Delta E_{\rm beam}$ will only be small compared to statistical errors associated with the measurements of a typical cross section (equivalent to a rate in a particular final state channel) if the error in the measurement of $\Delta E_{\rm beam}$ is smaller than $\sim 1\%$. The independent measurement of $\Delta E_{\rm beam}$ can have both statistical and systematic error. Let us first consider the impact of statistical errors. We will argue that statistical uncertainties in $\Delta E_{\rm beam}$ (and also $E_{\rm beam}$ itself) should not be significant. Consider measurement of some resonance parameter $p$ through a series of observations over a large number ($S$) of spills. In a year of operation at a muon collider there will be something like $S=10^8$ spills. Each spill ({\it i.e.}\ each muon bunch) will contribute to the measurement of $p$, but the value of $p$ interpreted from the cross section $\sigma$ observed for a given spill will be uncertain by an amount $\delta p$. This uncertainty arises (a) because of the limited number of events $N_i$ accumulated during the spill and (b) due to uncertainties in $E_{\rm beam}$ and $\Delta E_{\rm beam}$ for that spill. We have, using the short hand notation of $E_b$ for $E_{\rm beam}$, \begin{equation} \left[\delta p\over p\right]_i=\left[ { c^2\over N_i}+ \left(a {\delta E_b\over E_b}\right)_i^2 + \left(b {\delta \Delta E_b\over \Delta E_b}\right)_i^2 \right]^{1/2}\,, \label{erroreq} \end{equation} where $c={d\ln p\over d \ln \sigma}$, $a={d\ln p\over d\ln E_b}$ and $b={d\ln p\over d\ln \Delta E_b}$. $E_b$ and $\Delta E_b$ are measured by looking at oscillations in the spectra of the muon decay products arising due to spin precession of the muons in the bunch during the course of a very large number of turns around the ring. These measurements are completely independent of the measured rate(s) in question. The fractional statistical errors ${\delta E_b\over E_b}$ and ${\delta \Delta E_b\over \Delta E_b}$ for each spill are expected to be small as we shall summarize below. As a result, the contribution from the ${c^2\over N_i}$ term in Eq.~(\ref{erroreq}) will normally be much larger than that from the $\delta E_b$ and $\delta\Delta E_b$ terms unless the coefficients $a$ and $b$ are very large. We will later learn that $a$ and $b$ will always be small under circumstances in which systematic errors in $E_b$ and $\Delta E_b$ do not badly distort the parameter measurement. To the extent that this is true, the statistical error $\delta p/p$ after a large number of spills, computed as \begin{equation} {\delta p \over p}=\left[ \sum_i {1\over \left({\delta p\over p}\right)_i^2}\right]^{-1/2}\,, \label{finalerror} \end{equation} will be dominated by the usual $c/\sqrt{\sum_i N_i}$ term. Note that if this term were altogether absent, then ${\delta p\over p}$ would be proportional to $1/\sqrt S$ times the per spill error from $\delta E_b$ and $\delta \Delta E_b$. Thus, even if this latter were quite substantial, it would be strongly suppressed after accumulating data over $10^8$ spills. Since $E_{\rm beam}$ and $\Delta E_{\rm beam}$ are expected to vary somewhat from spill to spill, what is important is the statistical error with which they can be measured in a given spill. This has been discussed in \cite{spread,raja,reportusa,reporteurope}. Very roughly, the frequency of oscillation in the signal of secondary positrons from the muon decays (which signal is sensitive to the precession of the naturally-present polarization of the muons) determines $E_{\rm beam}$ and the decay with time of the oscillation signal amplitude determines $\Delta E_{\rm beam}/E_{\rm beam}$. At a resonance mass of $100~{\rm GeV}$,~\footnote{The precise figures given here are from \cite{reporteurope}.} a beam energy spread of $\Delta E_{\rm beam}/E_{\rm beam}\sim 3\times 10^{-5}$ can be achieved by an appropriate machine design while maintaining adequate yearly luminosity ($\geq 0.1~{\rm fb}^{-1}$). Further, in this case, {\it for each muon spill} ({{\it i.e.}\ each muon bunch}), one can determine the beam energy itself to 1 part in $10^7$ (5 keV) and measure the actual magnitude of $\Delta E_{\rm beam}/E_{\rm beam}$ with an accuracy of roughly $1.67\%$. For $\Delta E_{\rm beam}/E_{\rm beam}\sim 10^{-3}$, as might be useful for a much broader resonance than the SM Higgs boson and as would allow for substantially larger yearly integrated luminosity ($\geq 1~{\rm fb}^{-1}$), $E_{\rm beam}$ can be measured to 2 parts in $10^{6}$ (100 keV) and $\Delta E_{\rm beam}/E_{\rm beam}$ can be measured with an accuracy of roughly $0.2\%$. As described above, this level of statistical error for $E_{\rm beam}$ and $\Delta E_{\rm beam}/E_{\rm beam}$ will result in only a tiny $\delta p/p$ error for a given resonance parameter unless the coefficients $a$ and $b$ of Eq.~(\ref{erroreq}) are very large. Thus, in what follows, we will analyze the parameter errors introduced by an uncertainty in $\Delta E_{\rm beam}/E_{\rm beam}$ that is assumed to be systematic in nature. The level of such uncertainty is not well understood at the moment. The energy spread will be affected by possible time dependence of other beam parameters, such as emmittance, by time-dependent backgrounds to the precession measurements, and by other sources of depolarization. Absent a detailed design for the machine and the polarimeter, \cite{reporteurope} states that it is ``quite certain'' that the energy spread can be known with relative systematic error better than $\sim 1\%$, even if $\Delta E_{\rm beam}/E_{\rm beam}=3\cdot 10^{-5}$. Given the precision with which the beam energy can be determined, it is clear that the mass of a resonance will be precisely known. We will not consider here the statistical errors in the measurements of the relevant cross sections, but these will later become important when choosing among the different procedures that we will discuss. Our main focus will be on the errors in the determination of branching ratios and total widths induced by a systematic uncertainty in the energy spread of the beam. Therefore, we will discuss the possibility of reducing such errors, and also discuss measurements which are independent of the energy spread. As already noted, these errors are expected to be important only when the width of the resonance is smaller than or comparable to the energy spread of the beam. We assume that the resonance is well described by a Breit-Wigner shape. For a resonance $R$ of spin $j$ produced in the $s$-channel and decaying into a given final state $F$, one has \begin{equation}\label{BW} \sigma^{F}(E)=4\pi(2 j+1)\frac{\Gamma(R\to\ell^+\ell^-) \Gamma(R\to F)}{(E^2-M^2)^2+M^2\Gamma^2} \end{equation} where $M$ and $\Gamma$ are the mass and the width of the resonance, and $E=\sqrt{s}$. We will work in the narrow width approximation and therefore neglect the running of the width. We will consider also the total production cross section \begin{equation}\label{BWT} \sigma(E)=4\pi(2 j+1)\frac{\Gamma(R\to\ell^+\ell^-) \Gamma}{(E^2-M^2)^2+M^2\Gamma^2} \end{equation} We assume that in the absence of bremsstrahlung the lepton beams have a Gaussian beam energy spread specified by $\Delta E_{\rm beam}/E_{\rm beam}=0.01\,R(\%)$, leading to a spread in the center of mass energy given by \begin{equation}\label{spread} \sigma_E={0.01\,R(\%)\over\sqrt 2}\,E\sim 0.007\,R(\%)\,E\,. \end{equation} In the absence of bremsstrahlung, the energy probability distribution $f(E)$ can be written in the scaling form \begin{equation}\label{scaling} f(E)=\frac 1{\sigma_E} g\left(\frac {E-E_0}{\sigma_E}\right)\,, \end{equation} where $E_0$ is the central energy setting and \begin{equation}\label{normalization} \int f(E)\,dE=1\,. \end{equation} Accurate measurements of a narrow resonance require an accurate knowledge of both $\sigma_E$ and of the shape function $g$. We will explore the systematic uncertainties associated with errors in determining $\sigma_E$, assuming that the shape function $g$ is known to be of a Gaussian form. After including bremsstrahlung (at an $e^+e^-$ collider, beamstrahlung must also be included) it is no longer possible to write the full energy distribution in a scaling form; bremsstrahlung has intrinsic knowledge of the mass of the lepton, which in leading order enters in the form $\log(E/m_\ell)$. Since the bremsstrahlung distribution is completely known it may be convoluted with the scaling form of Eq.~(\ref{scaling}) to obtain a final energy distribution which depends upon both $\sigma_E$ and the (very accurately known) machine energy setting. Some useful figures illustrating the effects of bremsstrahlung at a muon collider are Figs. 30, 31 and 32 in Appendix A of \cite{bbgh}. One typically finds that the peak luminosity at the central beam energy, $E_0$, is reduced to about 60-80\% of the value in the absence of bremsstrahlung, and that $d{\cal L}/dE$ falls below $10^{-3}\left.d{\cal L}/dE\right|_{E=E_0}$ at $E\sim 0.9 E_0$. We will discuss results at a muon collider both before and after including bremsstrahlung, and show that the errors introduced by uncertainty in the beam energy spread can be very adequately assessed without including bremsstrahlung. The measured cross sections are obtained from the convolution of the theoretical cross sections [see (\ref{BW})] with the energy distribution: \begin{equation}\label{convolution} \sigma_c^{F}(E)=4\pi(2j+1) \Gamma(R\to\ell^+\ell^-) \Gamma(R\to F) h(\Gamma,\sigma_E,E) \end{equation} where \begin{equation}\label{convolution2} h(\Gamma,\sigma_E,E)=\int \frac{f(E- E')}{({E'}^2-M^2)^2+M^2\Gamma^2}d E' \end{equation} From these equations we can immediately draw two useful consequences: \begin{itemize} \item the ratio of $\sigma_c^{F}$ to the production cross section is given by \begin{equation}\label{ratio} \frac{\sigma_c^{F}}{\sigma_c}=B_{F},~~~B_{F}=B(R\to F) \end{equation} and does not depend on $\sigma_E$; \item due to the normalization condition (\ref{normalization}), the integral of $\sigma_c^{F}$ is independent of $\sigma_E$ \begin{equation}\label{integral} \int\sigma_c^{F}(E)dE=\int\sigma^{F}(E)dE \,. \end{equation} Of course, one must be very certain that the integral over $E$ covers all of the resonance and all of the beam spread, including the low-energy bremsstrahlung tail. \end{itemize} Let us now consider the narrow resonance approximation. In this case, we can write \begin{equation}\label{NRA} M^2\sigma^{F}(E)=\pi(2 j+1)B_{\ell^+\ell^-} B_{F} \frac{\gamma^2}{x^2+\gamma^2/4} \end{equation} where we have scaled the energy and the total width in terms of the energy spread evaluated at the peak (here we take $\sigma_E\approx \sigma_M$) \begin{equation}\label{newvariables} x=\frac{E-M}{\sigma_M},~~~~\gamma=\frac{\Gamma}{\sigma_M} \end{equation} The convolution can also be written in terms of these variables, with the result (see Eq. (\ref{scaling})) \begin{eqnarray}\label{scaled-conv} M^2\sigma_c^{F}(E)&=&\pi(2j+1)B_{\ell^+\ell^-}B_{F} \int g(x-x')\frac{\gamma^2}{{x'}^2+\gamma^2/4}dx'\nonumber\\&\equiv &B_{\ell^+\ell^-}B_{F}\,\Phi(x,\gamma) \end{eqnarray} In particular, for the production cross section, we get \begin{equation}\label{scaled-conv3} M^2\sigma_c(E)=B_{\ell^+\ell^-}\,\Phi(x,\gamma) \end{equation} Some simple results that are valid in the narrow resonance approximation are the following: \begin{itemize} \item If $\Gamma\gg\sigma_M$ (and $\Gamma$ is also large compared to bremsstrahlung energy spread), then \begin{equation} \sigma_c(E)={4\pi B_{\ell^+\ell^-}\over M^2}{\Gamma^2M^2\over (E^2-M^2)^2+\Gamma^2M^2}\,. \label{largegam} \end{equation} A measurement of the peak cross section (summed over all final state modes) gives a direct measurement of $B_{\ell^+\ell^-}$. In addition, the resonance shape can be scanned by using a series of different energy settings for the machine and $\Gamma$ can then be determined. \item If $\Gamma\ll\sigma_M$, then \begin{equation} \sigma_c(E)={4\pi B_{\ell^+\ell^-} \over M^2}{\Gamma \sqrt\pi\over 2\sqrt 2\sigma_M}\exp\left[{-(E-M)^2\over 2\sigma_M^2}\right]\,, \label{smallgam} \end{equation} where we have assumed a Gaussian form for the beam energy spread. (We also neglected bremsstrahlung, but this could be easily included.) In this case, the peak cross section determines $B_{\ell^+\ell^-}\cdot\Gamma$ {\it provided} $\sigma_M$ is accurately known. $\sigma_M$ could in turn be measured by changing $E$ relative to $M$. However, statistical accuracy may be poor because cross sections are suppressed by $\Gamma/\sigma_M$ in comparison to those obtained if $\Gamma\geq\sigma_M$. \end{itemize} When scanning a resonance, the above limits make it clear that the cross section is largest and that systematic errors associated with $\sigma_M$ are smallest when $\sigma_M$ is as small compared to $\Gamma$ as possible. However, for a sufficiently narrow resonance, the luminosity reduction associated with achieving $\sigma_M$ values significantly smaller than $\Gamma$ is often so great that statistical uncertainties become large. Thus, we are often faced with a situation in which $\Gamma$ is comparable to or only a few times larger than $\sigma_M$. (For example, this will be the case when studying a light SM Higgs boson at a muon collider.) It is in this situation that systematic errors in the resonance parameters deriving from uncertainty in $\sigma_M$ can be enhanced. We now discuss four ways of extracting $B_{\ell^+\ell^-}$, $B_{\ell^+\ell^-}B_{F}$ in a specific final state channel and $\Gamma$ from the data and assess the extent to which their determination will be influenced by uncertainty associated with an independent measurement of $\sigma_M$. \begin{table}[t] \caption{ Errors on $\Gamma$ and $B$ induced by $\Delta\sigma_M/\sigma_M= +0.05$ (upper lines) and $\Delta\sigma_M/\sigma_M= -0.05$ (lower lines) evaluated by choosing one observation at the peak and the other at $E-M=k\,\Gamma$, for $k=1,2,3$. Bremsstrahlung is neglected.} \begin{center} \begin{tabular}{|c|l|l|l|l|l|l|l|l|c|} \hline & \multicolumn{2}{|c|}{$1\,\Gamma$}&\multicolumn{2}{|c|}{$2\,\Gamma$} &\multicolumn{2}{|c|}{$3\,\Gamma$} \\ $\Gamma/\sigma_M$& $\Delta B/B$& $\Delta\Gamma/\Gamma$&$\Delta B/B$& $\Delta\Gamma/\Gamma$&$\Delta B/B$&$\Delta\Gamma/\Gamma$ \\\hline 1 & ${+0.220}$ & ${-0.205}$ & ${+0.171}$ & ${-0.160}$ & ${+0.114}$ & ${-0.101}$\\ & ${-0.140}$ & ${+0.198}$ & ${-0.119}$ & ${+0.153}$ & ${-0.089}$ & ${+0.094}$\\ \hline 2 & ${+0.065}$ & ${-0.075}$ & ${+0.043}$ & ${-0.037}$ & ${+0.035}$ & ${-0.023}$\\ & ${-0.053}$ & ${+0.072}$ & ${-0.039}$ & ${+0.035}$ & ${-0.033}$ & ${+0.023}$\\ \hline 3 & ${+0.031}$ & ${-0.036}$ & ${+0.023}$ & ${-0.017}$ & ${+0.022}$ & ${-0.013}$\\ & ${-0.027}$ & ${+0.034}$ & ${-0.022}$ & ${+0.017}$ & ${-0.021}$ & ${+0.013}$\\ \hline 4 & ${+0.018}$ & ${-0.020}$ & ${+0.016}$ & ${-0.011}$ & ${+0.015}$ & ${-0.009}$\\ & ${-0.017}$ & ${+0.019}$ & ${-0.015}$ & ${+0.010}$ & ${-0.014}$ & ${+0.009}$\\ \hline \end{tabular} \end{center} \label{nobremi} \end{table} \bigskip \bigskip \noindent {\bf{Scan of the resonance}} - If $\Gamma\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}}\sigma_M$, one of the most direct ways to determine the parameters of an $s$-channel resonance is through a three-point scan \cite{bbgh}. In this method one measures the production cross section $\sigma_c$ at three different energies. However, the position of the peak is independent of the energy spread, and we do not expect that uncertainty in the latter will induce an error in the mass of the resonance (we have checked explicitly that this is indeed the case). Therefore we will assume in the following that the mass of the resonance is known with high accuracy, and, as a consequence, we will take into account only two scan points; one will be chosen at the peak and the other one off the peak. We will shortly discuss how the results depend on the position of this last point. The parameters $(B,\Gamma)$ are then extracted by a two parameter fit. Here, $B=B_{\ell^+\ell^-}$ if one is able to sum over all final state modes, and $B=B_{\ell^+\ell^-}B_{F}$ if one focuses on a particular final state. In principle this problem has a unique solution, by deconvoluting the observed cross section. However, the error in the determination of $\sigma_M$ (the energy spread at the peak of the resonance) induces an error on the determination of the parameters of the resonance. Assuming that the measured value of $\sigma_c$ at a given energy $E$ is given by $\sigma_c(E,B,\Gamma,\sigma_M)$ (here again we use the narrow resonance approximation to put $\sigma_E\approx\sigma_M$), the changes in the values of $B$ and $\Gamma$ that result if $\sigma_M$ is shifted by an amount $\Delta\sigma_M$ can be evaluated through the equation \begin{equation}\label{condition1} \sigma_c(E,B,\Gamma,\sigma_M)= \sigma_c(E,B+\Delta B,\Gamma+\Delta\Gamma,\sigma_M+\Delta\sigma_M) \end{equation} or, from Eq. (\ref{scaled-conv3}), \begin{equation}\label{condition2} \Phi(x,\gamma)= \Phi(x+\Delta x,\gamma+\Delta\gamma) \left(1+\frac{\Delta B}{B}\right) \end{equation} where \begin{eqnarray}\label{def} \Delta x&=&x(\sigma_M+\Delta\sigma_M)-x(\sigma_M)\approx -x\frac{\Delta\sigma_M} {\sigma_M}\nonumber\\\Delta\gamma&=&\gamma(\sigma_M+\Delta\sigma_M)-\gamma(\sigma_M) \approx \frac{\Gamma}{\sigma_M}\left( \frac{\Delta\Gamma}{\Gamma}-\frac{\Delta\sigma_M}{\sigma_M}\right) \end{eqnarray} If we measure the cross section at two different energies, and assume that $\Delta\sigma_M$ is the same at these two nearby energies (as would be typical of a systematic error), we can easily employ Eq. (\ref{condition2}) to determine the induced fractional shifts $\Delta\Gamma/\Gamma$ and $\Delta B/B$. We fix one of the two points at the peak ($x=0$), and we let the other one vary between $E-M=\Gamma$ and $E-M=3\Gamma$. The induced systematic errors are illustrated in a series of Tables. In Tables \ref{nobremi} and \ref{nobremii}, we assume a Gaussian energy distribution without including bremsstrahlung and tabulate errors induced by $\Delta\sigma_M/\sigma_M=0.05$ and 0.01, respectively. As clear from Eq. (\ref{condition2}), the fractional errors on $\Gamma$ and $B$ depend only on the $E-M$ choice and on the reduced width $\gamma$, but not on $B$. \begin{table}[h] \caption{Same as Table \protect\ref{nobremi} for $\Delta\sigma_M/\sigma_M= +0.01$ (upper lines) and $\Delta\sigma_M/\sigma_M= -0.01$ (lower lines).} \begin{center} \begin{tabular}{|c|l|l|l|l|l|l|l|l|c|} \hline & \multicolumn{2}{|c|}{$1\,\Gamma$}&\multicolumn{2}{|c|}{$2\,\Gamma$} &\multicolumn{2}{|c|}{$3\,\Gamma$} \\ $\Gamma/\sigma_M$& $\Delta B/B$& $\Delta\Gamma/\Gamma$&$\Delta B/B$& $\Delta\Gamma/\Gamma$&$\Delta B/B$&$\Delta\Gamma/\Gamma$ \\\hline 1 & ${+0.036}$ & ${-0.040}$ & ${+0.029}$ & ${-0.031}$ & ${+0.020}$ & ${-0.020}$\\ & ${-0.033}$ & ${+0.040}$ & ${-0.027}$ & ${+0.031}$ & ${-0.019}$ & ${+0.019}$\\ \hline 2 & ${+0.012}$ & ${-0.015}$ & ${+0.008}$ & ${-0.007}$ & ${+0.007}$ & ${-0.004}$\\ & ${-0.011}$ & ${+0.015}$ & ${-0.008}$ & ${+0.007}$ & ${-0.007}$ & ${+0.004}$\\ \hline 3 & ${+0.006}$ & ${-0.007}$ & ${+0.005}$ & ${-0.003}$ & ${+0.004}$ & ${-0.003}$\\ & ${-0.006}$ & ${+0.007}$ & ${-0.004}$ & ${+0.003}$ & ${-0.004}$ & ${+0.003}$\\ \hline 4 & ${+0.004}$ & ${-0.004}$ & ${+0.003}$ & ${-0.002}$ & ${+0.003}$ & ${-0.002}$\\ & ${-0.003}$ & ${+0.004}$ & ${-0.003}$ & ${+0.002}$ & ${-0.003}$ & ${+0.002}$\\ \hline \end{tabular} \end{center} \label{nobremii} \end{table} As expected from our earlier discussion, the fractional errors become smaller for larger $\Gamma/\sigma_M$. The fractional errors are also decreased by moving the second point of the scan away from the peak (see Tables \ref{nobremi} and \ref{nobremii}). On the other hand the statistical errors for $\sigma_c$ are increasing as \begin{equation}\label{ratio2} \sigma_c(E)/\sigma_c(M) =\frac{\Phi(x,\gamma)}{\Phi(0,\gamma)} \end{equation} becomes smaller. (Due to presence of background, the decrease is not simply proportional to $\sqrt{\sigma_c(M)/\sigma_c(E)}$.) By fixing the amount of luminosity that one can safely lose without incurring substantial statistical error in $\sigma_c(E)$, and for a given ratio $\Gamma/\sigma_M$, one can, by looking at Fig. \ref{fig1}, fix the second scan point. For instance, for $\Gamma=2\sigma_M$, allowing for a loss in luminosity of 60\%, we see that the point to be chosen is $x\approx 2$ corresponding to $E\approx M+\Gamma$. By choosing the second point of the scan at $E=M+2\Gamma$, the resulting behaviour of the systematic errors vs. $\gamma$ is illustrated in Fig. \ref{fig2}. For a given uncertainty on the energy spread the errors decrease for increasing $\gamma$. Therefore, as noted earlier, one should employ the smallest value of $\sigma_M$ consistent with having luminosity large enough to give small statistical $\sigma_c$ errors. For instance, at $\gamma=1$ the errors on the resonance parameters are of order $2.5\div 3.5$ times $\Delta\sigma_M/\sigma_M$, whereas for $\gamma=3$ they are down to $0.3\div 0.5$ times $\Delta\sigma_M/\sigma_M$. Of course, a full study of the optimization of this procedure in a concrete case requires also taking into account the third scan point. In fact, one has the further freedom of moving the position of this point, with corresponding changes in the errors. One can show that by taking the third scan point in a symmetric position with respect to the other off-peak point, one gets the same errors shown in Tables \ref{nobremi} and \ref{nobremii} (in the absence of bremsstrahlung). By taking an asymmetric configuration the results are in between the ones obtained with the corresponding symmetric configurations. For instance, by taking one point at the peak, one at $E=M+2\Gamma$ and the third one at $E=M-\Gamma$, we find, for $\gamma=1$ and $\Delta\sigma_M/\sigma_M=0.05$, $\Delta\Gamma/\Gamma=-0.174$ and $\Delta B/B=+0.186$. \begin{table}[t] \caption{Errors for $\Gamma$ and $B$ induced by $\Delta\sigma_M/\sigma_M= +0.05$ for no bremsstrahlung (first lines, from Table \ref{nobremi}), and after including bremsstrahlung for $R=0.003\%$ (second lines), $R=0.03\%$ (third lines) and $R=0.1\%$ (fourth lines). The energy profile before bremsstrahlung is taken to be a Gaussian specified by the $\sigma_M$ corresponding to a given $R$ taking $M=100~{\rm GeV}$.} \begin{center} \begin{tabular}{|c|l|l|l|l|l|l|l|l|c|} \hline & \multicolumn{2}{|c|}{$1\,\Gamma$}&\multicolumn{2}{|c|}{$2\,\Gamma$} &\multicolumn{2}{|c|}{$3\,\Gamma$} \\ $\Gamma/\sigma_M$& $\Delta B/B$& $\Delta\Gamma/\Gamma$&$\Delta B/B$& $\Delta\Gamma/\Gamma$&$\Delta B/B$&$\Delta\Gamma/\Gamma$ \\\hline 1 & ${+0.22}$ & ${-0.21}$ & ${+0.17}$ & ${-0.16}$ & ${+0.11}$ & ${-0.10}$\\ & ${+0.22}$ & ${-0.21}$ & ${+0.18}$ & ${-0.16}$ & ${+0.12}$ & ${-0.11}$\\ & ${+0.22}$ & ${-0.21}$ & ${+0.18}$ & ${-0.16}$ & ${+0.12}$ & ${-0.11}$\\ & ${+0.22}$ & ${-0.21}$ & ${+0.18}$ & ${-0.16}$ & ${+0.12}$ & ${-0.11}$\\ \hline 2 & ${+0.065}$ & ${-0.075}$ & ${+0.043}$ & ${-0.037}$ & ${+0.035}$ & ${-0.023}$\\ & ${+0.067}$ & ${-0.078}$ & ${+0.046}$ & ${-0.044}$ & ${+0.040}$ & ${-0.032}$\\ & ${+0.065}$ & ${-0.075}$ & ${+0.044}$ & ${-0.041}$ & ${+0.037}$ & ${-0.029}$\\ & ${+0.067}$ & ${-0.077}$ & ${+0.045}$ & ${-0.042}$ & ${+0.039}$ & ${-0.031}$\\ \hline 3 & ${+0.031}$ & ${-0.036}$ & ${+0.023}$ & ${-0.017}$ & ${+0.022}$ & ${-0.013}$\\ & ${+0.031}$ & ${-0.035}$ & ${+0.024}$ & ${-0.020}$ & ${+0.023}$ & ${-0.017}$\\ & ${+0.031}$ & ${-0.037}$ & ${+0.024}$ & ${-0.020}$ & ${+0.023}$ & ${-0.017}$\\ & ${+0.031}$ & ${-0.038}$ & ${+0.024}$ & ${-0.021}$ & ${+0.022}$ & ${-0.016}$\\ \hline 4 & ${+0.018}$ & ${-0.020}$ & ${+0.016}$ & ${-0.011}$ & ${+0.015}$ & ${-0.009}$\\ & ${+0.019}$ & ${-0.022}$ & ${+0.017}$ & ${-0.013}$ & ${+0.015}$ & ${-0.011}$\\ & ${+0.018}$ & ${-0.021}$ & ${+0.016}$ & ${-0.012}$ & ${+0.015}$ & ${-0.011}$\\ & ${+0.018}$ & ${-0.020}$ & ${+0.015}$ & ${-0.012}$ & ${+0.015}$ & ${-0.011}$\\ \hline \end{tabular} \end{center} \label{brema} \end{table} \begin{table}[h] \caption{Errors for $\Gamma$ and $B$ expressed as $\Delta \Gamma/\Gamma=c_\Gamma \Delta\sigma_M/\sigma_M$ and $\Delta B/B=c_B \Delta\sigma_M/\sigma_M$ in the limit of small $\Delta\sigma_M/\sigma_M$. Same entry ordering and bremsstrahlung assumptions as in Table~\ref{brema}.} \begin{center} \begin{tabular}{|c|l|l|l|l|l|l|l|l|c|} \hline & \multicolumn{2}{|c|}{$1\,\Gamma$}&\multicolumn{2}{|c|}{$2\,\Gamma$} &\multicolumn{2}{|c|}{$3\,\Gamma$} \\ $\Gamma/\sigma_M$ & $~~~c_B$ & $~~~c_\Gamma$ & $~~~c_B$ & $~~~c_\Gamma$ & $~~~c_B$ & $~~~c_\Gamma$ \\\hline 1 & 3.4 & $-$4.0 & 2.8 & $-$3.1 & 2.0 & $-$2.0 \\ & 3.6 & $-$4.3 & 2.8 & $-$3.2 & 2.1 & $-$2.1 \\ & 3.2 & $-$3.7 & 2.8 & $-$3.2 & 2.1 & $-$2.1 \\ & 3.4 & $-$3.9 & 2.8 & $-$3.2 & 2.1 & $-$2.2 \\ \hline 2 & 1.2 & $-$1.5 & 0.81 & $-$0.72 & 0.69 & $-$0.45 \\ & 1.2 & $-$1.5 & 0.85 & $-$0.81 & 0.72 & $-$0.59 \\ & 1.2 & $-$1.5 & 0.86 & $-$0.82 & 0.74 & $-$0.61 \\ & 1.2 & $-$1.5 & 0.85 & $-$0.82 & 0.74 & $-$0.61 \\ \hline 3 & 0.58 & $-$0.70 & 0.46 & $-$0.34 & 0.43 & $-$0.26 \\ & 0.62 & $-$0.76 & 0.48 & $-$0.40 & 0.43 & $-$0.32 \\ & 0.60 & $-$0.76 & 0.48 & $-$0.42 & 0.46 & $-$0.36 \\ & 0.57 & $-$0.71 & 0.48 & $-$0.42 & 0.44 & $-$0.36 \\ \hline 4 & 0.35 & $-$0.40 & 0.31 & $-$0.21 & 0.30 & $-$0.17 \\ & 0.39 & $-$0.45 & 0.30 & $-$0.24 & 0.30 & $-$0.26 \\ & 0.39 & $-$0.43 & 0.32 & $-$0.23 & 0.29 & $-$0.23 \\ & 0.35 & $-$0.40 & 0.29 & $-$0.24 & 0.28 & $-$0.20 \\ \hline \end{tabular} \end{center} \label{derapprox} \end{table} Inclusion of bremsstrahlung might complicate the picture developed above, since the induced errors could, in principle, depend upon the resonance mass and the value of $R$, even when holding $\Gamma/\sigma_M$ and $(E-M)/\sigma_M$ fixed. To determine if the effects of bremsstrahlung are important for determining systematic errors coming from uncertainty in $\sigma_M$, we begin by considering again $\Delta \sigma_M/\sigma_M=+0.05$. We adopt $M=100~{\rm GeV}$ and consider $R=0.003\%$, $0.03\%$, and $0.1\%$ at a muon collider. The resulting errors are given in Table~\ref{brema} in comparison to the no-bremsstrahlung results of the upper lines of Table~\ref{nobremi}. The induced errors computed including bremsstrahlung are quite independent of $R$ and (within the numerical errors of our programs) are essentially the same as those computed in the absence of bremsstrahlung. This can be further illustrated in the limit of very small $\Delta\sigma_M/\sigma_M$. In this limit, one can simply compute the errors induced in $B$ and $\Gamma$ by using a linear expansion of Eq.~(\ref{condition1}) and solving the resulting matrix equation. The results for $\Delta B/B$ and $\Delta \Gamma/\Gamma$ expressed as a coefficient times $\Delta\sigma_M/\sigma_M$ are given in Table \ref{derapprox} for four cases. The first case is that of no bremsstrahlung. (Note that for $\Delta\sigma_M/\sigma_M=\pm 0.01$ the errors predicted are quite close to those of Table \ref{nobremii}.) The other three cases are for $M=100~{\rm GeV}$ and $R=0.003\%$, $0.03\%$, and $0.1\%$, including bremsstrahlung. Once again, we see that the error coefficients computed including bremsstrahlung do not depend much on $R$ and are essentially the same (within the numerical errors of our programs) as the error coefficients computed without including bremsstrahlung. Thus, in what follows we will discuss errors from $\Delta\sigma_M/\sigma_M$ without including the effects of bremsstrahlung. Once a particular resonance is discovered, the effects of bremsstrahlung upon the results presented here can be easily incorporated to whatever precision is required. As regards $e^+e^-$ colliders, at which bremsstrahlung is a more substantial effect, a study (similar to that above for the $\mu^+\mu^-$ collider) shows that the Gaussian approximation for studying systematic errors from $\Delta\sigma_M$ is again reasonable. Although beamstrahlung is also important at an $e^+e^-$ collider, we did not study its effects. However, current $e^+e^-$ collider designs are such that the energy spreading from beamstrahlung is typically comparable to or smaller than that from bremsstrahlung. An interesting general question regarding the scan procedure is how to optimize the choice of $\sigma_M$ relative to $\Gamma$ so as to minimize the net statistical plus systematic error. Let us consider using $k=1$ for the scan, assuming a resonance mass of $M=100~{\rm GeV}$ and that $\Delta\sigma_M/\sigma_M$ is small ($\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.02$), so that the linear error expansion using coefficients $c_{\Gamma}$ and $c_B$ is valid. A very rough parameterization for the induced error in $\Gamma$ when $0.2\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} \Gamma/\sigma_M\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 3$ and $\Delta\sigma_M/\sigma_M\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.02$ is \begin{equation} {\Delta\Gamma\over\Gamma}\sim3.3\left({\sigma_M\over\Gamma}\right)^{1.288} {\Delta\sigma_M\over\sigma_M}\,. \label{dgamform} \end{equation} The optimal choice of $R$ for determining the parameters of a given resonance depends upon many factors: how the machine luminosity varies with $R$; the variation of $\Delta\sigma_M/\sigma_M$ with $R$; the variation of the $\Delta\sigma_M$-induced errors as a function of $R$; the magnitude of the resonance width (in particular, as compared to $\sigma_M$); and the size of backgrounds in the important final states to which the resonance decays. At a muon collider, $R$ in the range $0.1\%\div0.15\%$ is the natural result and allows maximal luminosity. Increasing $R$ above this range does not result in significant luminosity increase; in contrast, the luminosity declines rapidly as $R$ is decreased below this range. A convenient parameterization for the luminosity of an $E=100~{\rm GeV}$ muon collider, valid for $0.003\%\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} R\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.12\%$, is:~\footnote{The $L_{\rm year}$ parameterization interpolates the three results of Table 5 in \cite{reporteurope} taken from \cite{reportusa}.} \begin{equation} L_{\rm year}=1.2~{\rm fb}^{-1}\,\left({R\over 0.12\%}\right)^{0.67362}\,. \label{lform} \end{equation} The specific coefficient in Eq.~(\ref{lform}) represents the most pessimistic estimate for the instantaneous luminosities. On occasion we shall also discuss results assuming a factor of 10 larger coefficient, which we shall refer to as the optimistic luminosity estimate. We do not know what to expect for the variation of the systematic error in $\sigma_M$ as a function of $R$. In order to understand how the optimization might work, we have adopted on a purely adhoc basis a form that mimics the per-spill {\it statistical} error:~\footnote{This form for the statistical $\Delta\sigma_M/\sigma_M$ error interpolates the results for the two $R=0.003\%$ and $R=0.1\%$ cases given in Sec. 4.3.2 of \cite{reporteurope} using a power law form.} \begin{equation} {\Delta\sigma_M\over\sigma_M}=2\cdot 10^{-3}\,\left({R\over 0.12\%}\right)^{-0.57477}\,. \label{dsigmaform} \end{equation} This form corresponds to a decreasing fractional systematic error in $\sigma_M$ as $R$ increases, as seems a likely possibility. Finally, it is useful to recall the value of $\sigma_M$ as a function of $R$: $$\sigma_M=86~{\rm MeV}\,{M\over 100~{\rm GeV}}\,\left({R\over 0.12\%}\right)\,.$$ For $R=0.003\%$, the smallest $R$ that is likely to be achievable, these quantities have the values $L_{\rm year}=0.1~{\rm fb}^{-1}$, ${\Delta\sigma_M/ \sigma_M}=0.0167$, and $\sigma_M\sim 2~{\rm MeV}$ (for $M=100~{\rm GeV}$). The rough implications of these dependencies are as follows. \begin{itemize} \item For a broad resonance, defined as one with $\Gamma\gg0.001M$, one should operate the muon collider at its natural $R$ value of order $0.12\%$. The $\Delta\sigma_M$-induced errors will be very tiny, both because $\Gamma/\sigma_M$ is very large and because $\Delta\sigma_M/\sigma_M$ should be small for such $R$. A precision scan of the resonance is readily possible in this case. Further, the resonance cross section is maximal for large $\Gamma/\sigma_M$ and both background and signal rates vary slowly as a function of $R$. \item For a resonance with $\Gamma<0.001M$, one will wish to reduce $R$ until $\Gamma\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} \sigma_M$. The primary reasons to avoid $\Gamma<\sigma_M$ are to avoid the large $\Delta\sigma_M$-induced errors summarized above and to maintain adequate sensitivity to the resonance width. Keeping $\Gamma/\sigma_M>1$, one can combine the above formulae with the predicted variation of statistical error as a function of $R$ at fixed $L$ to determine the optimal $R$ choice. To illustrate, we will focus on the statistical error for $\Gamma$ obtained using the three-point scan with measurements at $E=M,M\pm 2\sigma_M$. If $\Gamma/\sigma_M$ is large (typically $>7\div 10$) the signal cross sections at the three scan points are roughly independent of $R$ (but only for $R$ values smaller than or not too much larger than that corresponding to $\sigma_M$). Since the background cross section is also independent of $R$, statistical errors would not vary much with $R$ if the total integrated luminosity is held fixed. More generally, one will find that, at fixed $L$, $\Delta\Gamma/\Gamma\propto R^p$ with $p>0$. (This is a very crude representation valid only for a limited range of $R$.) The exact rate of increase depends upon both $\Gamma/\sigma_M$ and the background level. Such cases will be discussed shortly. However, even if $p=0$, $\Delta\Gamma/\Gamma$ also varies as $1/\sqrt L$, where the available $L$ is given in Eq.~(\ref{lform}). Defining $f\equiv R/0.003\%$, the net result is that we can write $(\Delta\Gamma/\Gamma)_{\rm stat.}=c_{\rm stat.} f^{p-0.337}$. Meanwhile, if we insert the expressions for $\sigma_M$ and $\Delta\sigma_M/\sigma_M$ in the expression for the $\Delta\sigma_M$-induced $\Delta\Gamma/\Gamma$, we find $(\Delta\Gamma/\Gamma)_{\rm syst.}=c_{\rm syst.} f^{0.713}$, where (for the $k=1$ three-point scan) $c_{\rm syst.}\sim 0.06\left({2~{\rm MeV}\over\Gamma}\right)^{1.288}$ if $M=100~{\rm GeV}$; we have normalized $\Gamma$ to the approximate width of a $100~{\rm GeV}$ SM Higgs boson. If $p-0.337<0$, as would apply for $p\sim 0$, then, if the statistical and systematic errors are added in quadrature, the opposite signs of their $f$ exponents means that there will be a minimum in the combined error as a function of $f$. If we use as a reference the value $f_0$ such that $c_{\rm stat.}=c_{\rm syst.}$, then one finds that the optimal $f$ is such that $f/f_0=R/R_0\sim 0.7$ when $p\sim 0$ (requiring $\Gamma/\sigma_M(R_0)\gg 1$). In other words, one should choose a value of $R$ somewhat smaller than that value which would yield equal statistical and systematic errors. However, large $\Gamma/\sigma_M$ cannot be achieved for the very narrow Higgs and pseudogoldstone bosons discussed later in the paper. Even for the larger $M_H$ and $M_{\pzero}$ values of possible interest for this optimization discussion, we have, at best, $\Gamma/\sigma_M(R=0.003\%)\sim 3\div 5$. Starting from $R=0.003\%$, one finds that, at fixed total $L$, $\Delta\Gamma/\Gamma$ increases quite rapidly with increasing $R$. As two examples, $\Gamma/\sigma_M(R=0.003\%)\sim 3$ ($\sim 5$) for a SM-like Higgs boson with $M_H\sim 140~{\rm GeV}$ ($\sim 150~{\rm GeV}$). In these two cases, $(\Delta\Gamma/\Gamma)_{\rm stat.}$ increases by a factor $\sim \sqrt 10$ ($\sim \sqrt 2$) as $R$ increases from $0.003\%$ to $0.01\%$. Representing this increase using a power law (it is actually somewhat faster than a power law), we can roughly write $(\Delta\Gamma/\Gamma)_{\rm stat.}\sim c R^p$ at fixed $L$, with $p\sim 1$ ($p\sim 0.3$) for $M_H=140~{\rm GeV}$ ($M_H\sim 150~{\rm GeV}$). (Note that for $R$ above the $0.003\%$ to $0.01\%$ range, the $p$ values are much bigger.) From the analysis given in the preceding paragraph, we immediately see that if $p>0.337$, the best statistical and systematic errors will both be achieved by taking $R$ as small as possible. In the SM Higgs case, $\Gamma/\sigma_M>5$, corresponding to $M_H>150~{\rm GeV}$ would be required before $p<0.337$ and there could be some possible gain from increasing $R$. Similarly, in the case of the narrow pseudogoldstone boson $P^0$ it is only at the very largest $M_{\pzero}=200~{\rm GeV}$ mass considered that $p$ falls just slightly below $0.337$. In practice, our a priori knowledge of $\Gamma$ (from the initial scan needed to pin down the precise location of the $H$ or $P^0$) will be too imprecise to allow for such optimization; one should plan on operating the machine at $R=0.003\%$ if initial information indicates that we are dealing with a Higgs or pseudogoldstone boson. \end{itemize} \bigskip \noindent {\bf{Measuring the on-peak cross sections for different \boldmath$\sigma_M$}} - In this technique \cite{ratio} one presumes that $M$ is already quite well known and that one has in hand a rough idea of the size of $\Gamma$. (This is the likely result after the first rough scan used to locate the resonance.) One then operates the collider at $E=M$ for two different values of $\sigma_M$ (spending perhaps a year or two at each value). The results of Eqs.~(\ref{largegam}) and (\ref{smallgam}) show that if $\sigma_M^{\rm min}\ll\Gamma$ and $\sigma_M^{\rm max}\gg\Gamma$, then $\sigma_c(\sigma_M^{\rm min})/\sigma_c(\sigma_M^{\rm max})={2\sqrt 2\sigma_M^{\rm max}\over \Gamma\sqrt\pi}$. The systematic error in $\Gamma$ is then given by $(\Delta\Gamma/\Gamma)_{\rm syst.} =\Delta\sigma_M^{\rm max}/ \sigma_M^{\rm max}$. In practice, $\sigma_M^{\rm max}/\sigma_M^{\rm min}$ will be limited in size. If we define $\sigma_M^{\rm central}=\sqrt{\sigma_M^{\rm max}\sigma_M^{\rm min}}$ (the geometric mean value) and compute $r_c\equiv\sigma_c(\sigma_M^{\rm min})/\sigma_c(\sigma_M^{\rm max})= \Phi(0,\gamma_{\rm min})/\Phi(0,\gamma_{\rm max})$ (where $\gamma_{\rm max,min}=\Gamma/\sigma_M^{\rm max,min}$) as a function of $\Gamma$, our ability to measure $\Gamma$ in this way for any given value of $\sigma_M^{\rm max}/\sigma_M^{\rm min}$ is determined by the slope $|s|$ of $\ln[\Gamma/\sigma_M^{\rm central}]$ plotted as a function of $\ln[r_c]$. The relevant plot from Ref.~\cite{ratio} appears in Fig.~\ref{dlgdlr}. For a known $\sigma_M^{\rm central}$, the $|s|$ at any $\Gamma/\sigma_M^{\rm central}$ gives the relation $(\Delta\Gamma/\Gamma)_{\rm stat.}=|s|(\Delta r_c/r_c)_{\rm stat.}$, where $\Delta r_c/r_c$ is computed by combining the fractional statistical errors for $\sigma_c(\sigma_M^{\rm min})$ and $\sigma_c(\sigma_M^{\rm max})$ in quadrature. The point at which the magnitude of the slope, $|s|$, is smallest indicates the point at which a given fractional statistical error in the cross section ratio will give the most accurate determination (as measured by fractional error) of $\Gamma/\sigma_M^{\rm central}$. We observe that $\Gamma/\sigma_M^{\rm central}\sim 2\div 3$ gives the smallest $|s|$ (and hence smallest statistical error), although $|s|$ at $\Gamma/\sigma_M^{\rm central}\sim 1$ is not that much larger. As expected, the larger $\sigma_M^{\rm max}/\sigma_M^{\rm min}$, the smaller $|s|$ at any given $\Gamma/\sigma_M^{\rm central}$. The systematic error in $\Gamma$ due to uncertainties in $\sigma_M^{\rm max}$ and $\sigma_M^{\rm min}$ is obtained from \begin{equation} {\Phi\left(0,{\Gamma\over \sigma_M^{\rm min}}\right)\over \Phi\left(0,{\Gamma\over\sigma_M^{\rm max}}\right)}= {\Phi\left(0,{\Gamma+\Delta\Gamma\over \sigma_M^{\rm min}+\Delta\sigma_M^{\rm min}}\right)\over \Phi\left(0,{\Gamma+\Delta\Gamma \over\sigma_M^{\rm max} +\Delta\sigma_M^{\rm max}}\right)}\,. \label{dgamratio} \end{equation} In most instances, the $\sigma_M^{\rm min}$ measurement is not very sensitive to the precise value of $\sigma_M^{\rm min}$, in which case the result is $(\Delta\Gamma/\Gamma)_{\rm syst.} =\Delta\sigma_M^{\rm max}/\sigma_M^{\rm max}$. More generally $(\Delta\Gamma/\Gamma)_{\rm syst.}=\Delta\sigma_M/\sigma_M$ to the extent that $\Delta\sigma_M/\sigma_M$ is the same at the different $\sigma_M$ settings (as might apply for systematic errors of a certain type). From this we see one clear advantage of this technique: the systematic error in $\Gamma$ due to systematic error in $\sigma_M$ does not increase with decreasing $\Gamma/\sigma_M^{\rm max,min}$. If statistical errors for measuring the two $\sigma_c$'s can be kept small, this is a clear advantage of the technique as compared to the scan technique in which small $\Gamma/\sigma_M$ leads to large $\Delta\sigma_M$-induced systematic error even if statistics are excellent for all measurements. We will shortly discuss issues related to statistical errors. To continue our analysis of systematic errors, let us note that, in the limit of very high statistics for the cross section measurements ({\it i.e.}\ zero statistical error for $r_c$), precise values of both $\gamma_{\rm min}=\Gamma/\sigma_M^{\rm min}$ and $\gamma_{\rm max}=\Gamma/\sigma_M^{\rm max}$ are obtained by the above procedure. Further, $\sigma_c(\sigma_M^{\rm min})$ and $\sigma_c(\sigma_M^{\rm max})$ are given by $B\Phi(0,\gamma_{\rm min})$ and $B\Phi(0,\gamma_{\rm max})$, respectively, where $B$ stands for $B_{\ell^+\ell^-}$ or $B_{\ell^+\ell^-}B_F$, depending upon whether we are looking at the total rate or the rate in some particular final channel $F$. As a result, there is no systematic error in $B$ from uncertainty in $\sigma_M$, only statistical error associated with the number of events observed. This is another advantage of this technique. Of course small systematic errors are not important if statistical errors for the technique are not also small. We summarize the considerations \cite{ratio}. For the SM Higgs and the $P^0$ we found that it is best to use $\sigma_M^{\rm min}$ corresponding to $R=0.003\%$ and $\sigma_M^{\rm max}$ corresponding to $R=0.03\%$, so that $\sigma_M^{\rm central}$ corresponds to $R=0.003\%\times\sqrt{10}$. The parameterization for the variation of $L_{\rm year}$ given in Eq.~(\ref{lform}) implies that $L_{\rm year}=0.1~{\rm fb}^{-1}$ ($0.47~{\rm fb}^{-1}$) for $R=0.003\%$ ($0.03\%$). If, for example, $\Gamma/\sigma_M^{\rm central}=1$, one finds $\sigma_c(\sigma_M^{\rm min})/\sigma_c(\sigma_M^{\rm max})=4.5$, implying that the signal rate $S(\sigma_M)=L_{\rm year}(\sigma_M)\sigma_c(\sigma_M)$ is nearly the same for $\sigma_M^{\rm max}$ as for $\sigma_M^{\rm min}$. However, the background rate $B$ is proportional to $L$ and thus $B/S$ is a factor of 4.7 times larger at $\sigma_M^{\rm max}$ than at $\sigma_M^{\rm min}$. Consequently, the statistical error in the measurement of $\sigma_c(\sigma_M^{\rm max})$ will be worse than for $\sigma_c(\sigma_M^{\rm min})$ for the same $S$.~\footnote{In the scan procedure, there is a similar difficulty. There, $B/S$ is large for the off-peak measurements.} For a given running time at a given $\sigma_M$, one must compute the channel-by-channel $S$ and $B$ rates, compute the fractional error in $\sigma_c(\sigma_M)$ for each channel, and then combine all channels to get the net $\sigma_c(\sigma_M)$ error. This must be done for $\sigma_M=\sigma_M^{\rm min}$ and $\sigma_M=\sigma_M^{\rm max}$. One then computes the net $r_c$ and net $\sigma_c$ errors as: \begin{eqnarray} {\Delta r_c\over r_c}&=& \left\{\left[{\Delta \sigma_c(\sigma_M^{\rm min})\over\sigma_c(\sigma_M^{\rm min})}\right]^2 +\left[{\Delta \sigma_c(\sigma_M^{\rm max})\over\sigma_c(\sigma_M^{\rm max})}\right]^2 \right\}^{1/2}\,; \label{rcerror}\cr {\Delta\sigma_c\over\sigma_c}&=& \left\{\left[{\Delta \sigma_c(\sigma_M^{\rm min})\over \sigma_c(\sigma_M^{\rm min})}\right]^{-2} +\left[{\Delta \sigma_c(\sigma_M^{\rm max})\over\sigma_c(\sigma_M^{\rm max})}\right]^{-2} \right\}^{-1/2}\,. \label{sigcerror} \end{eqnarray} The ratio of running times at $\sigma_M^{\rm min}$ vs. $\sigma_M^{\rm max}$ cannot be chosen so as to simultaneously minimize the net $\Delta\sigma_c/\sigma_c$ and $\Delta r_c/r_c$. The former is minimized by running only at $\sigma_M^{\rm min}$, while the latter is typically minimized for $t(\sigma_M^{\rm min})/t(\sigma_M^{\rm max})\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 1$. For the SM Higgs, a good compromise is to take $t(\sigma_M^{\rm min})/t(\sigma_M^{\rm max})=1$. As demonstrated in the next section, it turns out that for both the $P^0$ and the SM Higgs boson, the ratio $L_{\rm year}(R=0.03\%)/L_{\rm year}(R=0.003\%)\sim 4.7$ and the predicted cross sections and backgrounds are such that this technique is very competitive with the scan technique as regards statistical errors for $\Gamma$. Of course, for resonances (such as those of the Degenerate BESS model considered later) that have fairly large widths, the normal scan procedure can achieve superior results to the $r_c$-ratio technique. This is because there will be little sacrifice in luminosity associated with choosing an $R$ such that $\Gamma/\sigma_M$ is substantially larger than 1. Measurements on the wings of the resonance will have nearly the same statistical accuracy as measurements at the resonance peak. \bigskip \noindent{\bf{Measurement of $\Gamma$ using the ratio of the $\ell^+\ell^-$ final state and total cross sections}} - First of all let us discuss the measurement of absolute and relative branching ratios. This is possible by simply measuring the cross sections in different final states, holding the energy fixed: $\sigma_c^{F_1}/\sigma_c^{F_2}=B_{F_1}/B_{F_2}$ and $\sigma_c^{F}/\sigma_c=B_{F}$. As apparent from Eq.~(\ref{ratio}), such cross section ratios do not depend on $\sigma_M$. Therefore, we can measure branching ratios and ratios of branching ratios with no systematic error induced by the energy spread. Of course, to measure $\sigma_c$, we must be able to sum over all final states for which $B_{F}$ is substantial, taking into account backgrounds and systematics associated with correctly determining the relative normalizations of different final states. This might not be easy, and could even be impossible if the resonance has some effectively invisible decays unless the branching ratio for invisible decays can be measured in a different experimental setting (in particular, one in which invisible decays can be effectively `tagged' by producing the resonance in association with some other particle). Given $B_{\ell^+\ell^-}$, the value of $\sigma_c$ and Eq.~(\ref{scaled-conv3}) can then be used to determine $\Gamma$ if $\sigma_M$ is known. The error on $\Gamma$ induced by uncertainty in $\sigma_M$, given the absence of systematic error in $B_{\ell^+\ell^-}$, is obtained from Eq. (\ref{condition2}) with $\Delta B=0$, \begin{equation}\label{condition3} \Phi(x,\gamma)=\Phi(x+\Delta x, \gamma+\Delta\gamma)\,. \end{equation} If we measure $\sigma_c$ and $\sigma_c^{\ell^+\ell^-}$ at the resonance peak we have $\Delta x=0$ and Eq.~(\ref{condition3}) then requires $\Delta\gamma=0$; in turn, \begin{equation} \Delta\gamma=0\longrightarrow \frac{\Delta\Gamma}{\Gamma}=\frac {\Delta\sigma_M}{\sigma_M}\,. \end{equation} Therefore, the systematic error induced in $\Gamma$ does not depend on $\gamma$. On the other hand, if we move away from the peak, the situation changes drastically, as illustrated in Fig. \ref{fig3} which gives $\Delta\Gamma/\Gamma$ vs. $\gamma$ for $\Delta\sigma_M/\sigma_M$ equal to 0.05. This figure can be trivially scaled for different values of $\Delta\sigma_M/\sigma_M$. We see that above $\gamma\approx 2$ the error decreases as one moves further away from the resonance peak. For values below $\gamma\approx 2$ the situation is more complicated, but for each choice of the energy there is a zero in the error.~\footnote{A really precise computation of the locations of these zeroes would need to include the effects of bremsstrahlung.} Therefore, by opportunely choosing the energy of the measurement one could try to work in a region where the systematic error is very much reduced. For instance, if $\Gamma\approx 1.4\sigma_M$, by taking $E=M+\Gamma$, one has $\Delta\Gamma=0$. Notice that the locations of the zeroes do not depend on $\Delta\sigma_M$. There are two potentially severe difficulties associated with this technique. First, in many instances the $\ell^+\ell^-$ decay mode has a small branching ratio. If the resulting event rate in the $\ell^+\ell^-$ final state is not large, the statistical error for $\sigma_c^{\ell^+\ell^-}$ will be large. Statistics for $E$ significantly different from $M$ (as required to take advantage of the zeroes discussed above) would be the first to become problematical. Second, we have already noted that measurement of $\sigma_c$ may be quite tricky. One must correctly normalize different final state channels relative to one another and assume that all final state channels with substantial branching ratio are visible. Regarding the latter, there is the alternative of measuring the branching ratio for invisible final states using other experimental situations/techniques and then making the appropriate correction to compute the full $\sigma_c$ given the contributions to $\sum_{F}\sigma_c^{F}$ that can be measured in the $s$-channel setting. Of course, in some instances $B_{\ell^+\ell^-}$ will also be measured at other machines or via other processes. Also in this case, one can immediately determine $\Gamma$ given a measurement of $\sigma_c$. Alternatively, if $B_{F}$ is also known for some final state $F$ (as well as $B_{\ell^+\ell^-}$), $\sigma_c^F/(B_{\ell^+\ell^-}B_F)$ will give a determination of $\Gamma$ with systematic uncertainty from $\Delta \sigma_M$ as described above. \bigskip \noindent {\bf{Measurement of the Breit-Wigner area}} - As apparent from Eq. (\ref{integral}), the energy integral of the cross section does not depend on $\sigma_M$, and, in the narrow width approximation, is given by \begin{equation}\label{integral2} \int\sigma_c(E)\,dE= 4\pi^2(2j+1)\,\Gamma\cdot\,B_{\ell^+\ell^-} \end{equation} Therefore, even if the energy spread is very poorly known, so long as it does not change as one scans over the resonance one could still measure the integral of the cross section, which is proportional to the product $\Gamma\cdot B_{\ell^+\ell^-}$, and obtain $B_{\ell^+\ell^-}$ from the ratio of the peak cross sections $\sigma_c^{\ell^+\ell^-}/\sigma_c$ (or, possibly, in another experimental setting). Of course, as discussed with regard to the previous procedure, the statistical error for $\sigma_c^{\ell^+\ell^-}$ might be large even when measured at the resonance peak. In addition, the event rate substantially off-resonance (as compared to both $\Gamma$ and $\sigma_M$) would be small (while typical backgrounds would be essentially constant). Thus, one can often only measure the cross section with good statistical accuracy over a limited portion of the full range. If $\Gamma\gg\sigma_M$, and $\sigma_M$ is known ahead of time (via spin-precession measurements or the like), this fact will quickly become apparent after measuring the cross section at a few energy settings with $E-M\gg\sigma_M$. A reliable value for $\Gamma\cdot\,B_{\ell^+\ell^-}$ can normally be obtained so long as statistics are good at the peak. The procedure is that based on Eq.~(\ref{largegam}). One will measure $\sigma_c(E)$ for a set of points perhaps out to $E-M\sim \pm 2\Gamma$, and the remainder of the integral will be determined using the Breit-Wigner shape that would have been revealed by the measurements made. Sensitivity to $\sigma_M$ will be very minimal. If $\sigma_M\gg\Gamma$, then the energy dependence of $\sigma_c(E)$ is determined by $\sigma_M$, see Eq.~(\ref{smallgam}), or more generally by $g$ of Eq.~(\ref{scaling}) (bremsstrahlung should be included). Given a known value for $\sigma_M$ and a set of measured points, the remainder of the integral can be computed. However, statistical errors are likely to be quite large because the cross section is suppressed by the ratio $\Gamma/\sigma_M$. Clearly, the most difficult case is that in which the resonance is very narrow and the smallest achievable $\sigma_M$ value is such that $\Gamma\sim\sigma_M$. Unless statistics remain good far off the peak, which is not likely, deconvolution of the effects of $\sigma_M$ and $\Gamma$ is required. The simplest deconvolution procedures for known $\sigma_M$ are the scan and ratio procedures outlined previously. Thus, if $\Gamma\sim \sigma_M$, the three-point scan technique and the technique of varying $\sigma_M$ while keeping $E=M$ are the most efficient for determining the parameters of an $s$-channel resonance so long as systematic uncertainty in the energy spread of the beam is smaller than $1\%\div 2\%$. (Of course, the appropriate strategy for exploring a resonance would be quite different if $\sigma_M$ is not independently measured with high statistical accuracy using the precession measurements.) \bigskip \section{Applications} \bigskip \noindent {\bf{SM-Higgs boson}} - In this Section we will apply the first two methods of Section \ref{general} to the study of a light SM-Higgs boson. Consider first the three-point scan. By using the results of Fig. \ref{fig2} one can easily determine the errors induced by the uncertainty in the energy spread as a function of the Higgs mass. For the evaluation of the Higgs width as a function of the mass, we make use of the expressions given in Ref. \cite{djouadi}. The decay channels we have considered are $b\bar b$, $\tau^+\tau^-$, $c\bar c$, $gg$ and $WW^*$, $ZZ^*$ (one of the two final bosons being virtual). The resulting total width is given in Fig. \ref{fig4}. We have considered the interval $50\le M_H({\rm GeV})\le 150$, but recall that, actually, there is an experimental lower bound at $90\%$ C.L. of about 90 GeV \cite{vancouver}. Below 110 GeV, the width of the Higgs increases approximately linearly with the mass (aside from logarithmic effects due to the running of the quark masses) which means that the ratio $\Gamma_H/\sigma_M$ is approximately constant. By choosing $R=0.003\%$ (see \cite{gunmumu}) we get $\Gamma_H/\sigma_M\approx 1$. From Fig. \ref{fig2}, we see that the $\Delta\sigma_M$-induced errors in $B$ and in $\Gamma_H$ are about 15\% and 2.5\% for $\Delta\sigma_M/\sigma_M=0.05$ and 0.01, respectively, for a $k=2$ scan (as appropriate, given that $\Gamma_H\sim\sigma_M$, for comparing to the $E=M,~E=M\pm 2\sigma_M$ scan summarized below that was used to estimate statistical errors). Above 110 GeV, the systematic errors decrease rapidly due to the fast increase of the width. Fig. \ref{fig5} summarizes the $\Delta B/B$ and $\Delta \Gamma_H/\Gamma_H$ fractional systematic errors as a function of $M_H$ for $R=0.003\%$ and $k=2$. Notice that, at least in the region up to 110 GeV, it will be mandatory to have values of $R$ of the order 0.003\%. For instance, if we take $R=0.01\%$, corresponding to $\Gamma_H/\sigma_M\approx 0.3$ (for $50\le M_H({\rm GeV})\le 110$), the fractional $\Delta\sigma_M$-induced errors in $B$ and $\Gamma_H$ increase to about $14\%\div 16\%$ for $\Delta\sigma_M/\sigma_M=0.01$, as can be seen from the dashed line of Fig. \ref{fig2}. \begin{table}[h] \caption[fake]{\baselineskip 0pt Percentage errors ($1\sigma$) for $\sigma_cB(H\to b\overline b,WW^\star,ZZ^\star)$ (extracted from channel rates) and $\Gamma_H$ for $s$-channel Higgs production at the muon collider assuming beam energy resolution of $R=0.003\%$. Results are presented for two integrated four-year luminosities: $L=4~{\rm fb}^{-1}$ ($L=0.4~{\rm fb}^{-1}$). An optimized three-point scan is employed using measurements at $E=M_H$, $E=M_H+2\sigma_M$ and $E=M_H-2\sigma_M$, with luminosities of $L/5$, $2L/5$ and $2L/5$, respectively. [For the cross section measurements, this is equivalent to $L\sim 2~{\rm fb}^{-1}$ ($L=0.2~{\rm fb}^{-1}$) at the $E=M_H$ peak]. This table is taken from Ref.~\cite{gunmumu}. Efficiencies and cuts are those employed in \cite{bbgh}. The effects of bremsstrahlung are included.} \small \begin{center} \begin{tabular}{|c|c|c|c|c|} \hline Quantity & \multicolumn{4}{c|}{Errors} \\ \hline \hline {\bf Mass (GeV)} & {\bf 80} & {\bf $M_Z$} & {\bf 100} & {\bf 110} \\ \hline $\sigma_cB(b\overline b)$ & $0.8\%(2.4\%)$ & $7\%(21\%)$ & $ 1.3\%(4\%)$ & $ 1\%(3\%)$ \\ \hline $\sigma_cB(WW^\star)$ & $-$ & $-$ & $ 10\%(32\%)$ & $ 5\%(15\%)$ \\ \hline $\sigma_cB(ZZ^\star)$ & $-$ & $-$ & $-$ & $ 62\%(190\%)$ \\ \hline $\Gamma_H$ & $ 3\%(10\%)$ & $ 25\%(78\%)$ & $ 10\%(30\%)$ & $ 5\%(16\%)$ \\ \hline \hline {\bf Mass (GeV)} & {\bf 120} & {\bf 130} & {\bf 140} & {\bf 150} \\ \hline $\sigma_cB(b\overline b)$ & $1\%(3\%)$ & $1.5\%(5\%)$ & $ 3\%(9\%)$ & $ 9\%(28\%)$ \\ \hline $\sigma_cB(WW^\star)$ & $3\%(10\%)$ & $2.5\%(8\%)$ & $ 2.3\%(7\%)$ & $ 3\%(9\%)$ \\ \hline $\sigma_cB(ZZ^\star)$ & $16\%(50\%)$ & $10\%(30\%)$ & $ 8\%(26\%)$ & $ 11\%(34\%)$ \\ \hline $\Gamma_H$ & $ 5\%(16\%)$ & $ 6\%(18\%)$ & $ 9\%(29\%)$ & $ 34\%(105\%)$ \\ \hline \end{tabular} \end{center} \label{fmcerrors} \end{table} It is interesting to compare these systematic errors to the statistical errors. The analysis at a muon collider done in Ref. \cite{gunmumu} gives statistical errors for a three-point scan using scan points at $E=M,~E=M\pm 2\sigma_M$ and $R=0.003\%$, assuming $L=4~{\rm fb}^{-1}$ or $L=0.4~{\rm fb}^{-1}$ total accumulated luminosity (corresponding to 4 years of operation for optimistic or pessimistic, respectively, instantaneous luminosity). The results of that analysis are summarized in Table~\ref{fmcerrors}. Except for $M_H\sim M_Z$, the $L=4~{\rm fb}^{-1}$ statistical error for measuring the total width would be of order $3\%\div 10\%$ when $80\le M_H\le 140~{\rm GeV}$. Therefore, to avoid contaminating this high precision measurement error with systematic uncertainty from $\Delta\sigma_M$ we will certainly want to have $\Delta\sigma_M/\sigma_M\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 1\%$. If one adopts the $L=0.4~{\rm fb}^{-1}$ luminosity assumption (the benchmark value of Refs.~\cite{reportusa,reporteurope}), and if $m_H<130~{\rm GeV}$ and not near $M_Z$, the statistical measurement error for $\Gamma_H$ is in the 10\% to 20\% range. This means that $\Delta\sigma_M/\sigma_M$ as large as $5\%$ would be very undesirable. Finally, we re-emphasize the fact that performing the scan using larger $R$ than $R=0.003\%$ leads to larger statistical errors until $M_H$ approaches the $WW$ decay threshold and $\Gamma_H/\sigma_M(R=0.003\%)>5$. For lower $M_H$, the $R=0.003\%$ results are the best that can be achieved despite the smaller luminosity at $R=0.003\%$ as compared to higher $R$ values. For example, the error in $\Gamma_H$ for a given luminosity using $R=0.01\%$ can be read off from Fig.~13 of \cite{bbgh}. One finds that $L(R=0.01\%)/L(R=0.003\%)=20,10,2$ is required in order that the $\Gamma_H$ statistical errors for $R=0.01\%$ be equal to those for $R=0.003\%$ at $M_H=130,140,150~{\rm GeV}$, respectively. Existing machine designs are such that $L_{\rm year}(R=0.01\%)/L_{\rm year}(R=0.003\%)=0.22~{\rm fb}^{-1}/0.1~{\rm fb}^{-1}=2.2$. Thus, increasing $R$ would not improve the scan-procedure statistical errors until $M_H>150~{\rm GeV}$. In addition, $\Delta\sigma_M$-induced systematic errors always rise rapidly with increasing $R$. For $M_H\leq 150~{\rm GeV}$, one should employ the smallest value of $R$ possible. \begin{table}[h] \caption[fake]{\baselineskip 0pt Percentage errors ($1\sigma$) for $\sigma_cB(H\to b\overline b,WW^\star,ZZ^\star)$ (extracted from channel rates) and $\Gamma_H$ for $s$-channel Higgs production at the muon collider. Operation at $E=M_H$ is assumed and the $r_c$-ratio technique is used for determining $\Gamma_H$. Results are presented assuming accumulated luminosities of $L=2~{\rm fb}^{-1}$ ($L=0.2~{\rm fb}^{-1}$) at $R=0.003\%$ and $L=9.4~{\rm fb}^{-1}$ ($L=0.94~{\rm fb}^{-1}$) at $R=0.03\%$, corresponding to roughly two years of running at each $R$ for optimistic (pessimistic) instantaneous luminosity assumptions. Efficiencies and cuts employed appear in \cite{bbgh}. The effects of bremsstrahlung are included.} \small \begin{center} \begin{tabular}{|c|c|c|c|c|} \hline Quantity & \multicolumn{4}{c|}{Errors} \\ \hline \hline {\bf Mass (GeV)} & {\bf 80} & {\bf $M_Z$} & {\bf 100} & {\bf 110} \\ \hline $\sigma_cB(b\overline b)$ & $0.7\%(2.2\%)$ & $6.3\%(20\%)$ & $ 1.2\%(3.8\%)$ & $ 0.9\%(2.8\%)$ \\ \hline $\sigma_cB(WW^\star)$ & $-$ & $-$ & $ 8.2\%(26\%)$ & $ 3.8\%(12\%)$ \\ \hline $\sigma_cB(ZZ^\star)$ & $-$ & $-$ & $-$ & $ 60\%(190\%)$ \\ \hline $\Gamma_H$ & $ 6\%(19\%)$ & $ 63\%(200\%)$ & $ 14\%(45\%)$ & $ 8\%(25\%)$ \\ \hline \hline {\bf Mass (GeV)} & {\bf 120} & {\bf 130} & {\bf 140} & {\bf 150} \\ \hline $\sigma_cB(b\overline b)$ & $0.9\%(2.8\%)$ & $1.4\%(4.4\%)$ & $ 2.4\%(7.6\%)$ & $ 6.6\%(21\%)$ \\ \hline $\sigma_cB(WW^\star)$ & $2.4\%(7.7\%)$ & $1.8\%(5.7\%)$ & $ 1.6\%(5.0\%)$ & $ 1.8\%(5.6\%)$ \\ \hline $\sigma_cB(ZZ^\star)$ & $15\%(46\%)$ & $7.9\%(25\%)$ & $ 6.3\%(20\%)$ & $ 7.0\%(22\%)$ \\ \hline $\Gamma_H$ & $ 6.3\%(20\%)$ & $ 6\%(19\%)$ & $ 5.4\%(17\%)$ & $ 4.7\%(18\%)$ \\ \hline \end{tabular} \end{center} \label{fmcerrorsrc} \end{table} Let us now compare to the $r_c$-ratio technique. We employ the same total of 4 years of operation as considered for the three-point scan, but always with $E=M_H$. As noted earlier, a good sharing of time is to devote two years to running at $R=0.003\%$, accumulating $L=2~{\rm fb}^{-1}$ ($L=0.2~{\rm fb}^{-1}$) for optimistic (pessimistic) instantaneous luminosity, and a second two years to running at $R=0.03\%$, corresponding to [using the luminosity scaling law of Eq.~(\ref{lform})] $L=9.4~{\rm fb}^{-1}$ ($L=0.94~{\rm fb}^{-1}$) of accumulated luminosity. For $R=0.03\%$, $\Gamma_H/\sigma_M^{\rm max}\sim 0.1$ and, as noted earlier, each $\sigma_c B_F$ cross section is decreased by almost the same factor by which the luminosity has increased, leaving the number of signal events unchanged. However, the background is increased by a factor of 4.7. A complete calculation is required. This was performed in \cite{ratio}. The various $\sigma_c B_F$ statistical errors are summarized in Table~\ref{fmcerrorsrc} along with the $\Delta\Gamma_H/\Gamma_H$ statistical error computed by combining all the listed final state channels and following the procedure of Eq.~(\ref{sigcerror}). We observe that the ratio technique becomes superior to the scan technique for the larger $M_H$ values ($M_H>130~{\rm GeV}$). This is correlated with the fact that $\Gamma_H/\sigma_M^{\rm min}$ (where $\sigma_M^{\rm min}$ is that for $R=0.003\%$) becomes substantially larger than 1 for such $M_H$. In particular, for larger $M_H$, $\Gamma_H/\sigma_M^{\rm central}$ is in a range such that $|s|$ and, consequently, the error in $\Gamma_H$ will be minimal. Thus, the two techniques are actually quite complementary --- by employing the best of the two procedures, a very reasonable determination of $\Gamma_H$ and very precise determinations of the larger channel rates will be possible for all $M_H$ below $2M_W$. Finally, we again note that the $r_c$-ratio technique has the advantage that the $\Delta\sigma_M$-induced systematic error in $\Gamma_H$ is equal to $\Delta\sigma_M/\sigma_M$ and will, therefore, be smaller by a factor of about 2.5 (if $M_H\leq 120~{\rm GeV}$) for the $r_c$-ratio technique than for the scan technique and that there is no $\Delta\sigma_M$-induced systematic error in the $B_{\ell^+\ell^-}B_F$ determinations. Thus, although the statistical errors for the $r_c$-ratio procedure are larger than for the scan procedure when $M_H<130~{\rm GeV}$, if the fractional error in $\sigma_M$ is substantially larger than $0.01$, the $r_c$-ratio could have net overall (statistical plus systematic) error that is smaller than the scan technique down to $M_H$ values significantly below $130~{\rm GeV}$. \bigskip \noindent {\bf{The lightest PNGB}} - The $s$-channel production of the lightest neutral pseudo-Nambu-Goldstone boson (PNGB) ($P^0$), present in models of dynamical breaking of the electroweak symmetry which have a chiral symmetry larger than $SU(2)\times SU(2)$, has recently been explored \cite{mumu,mumualso}. In the broad class of models considered in \cite{mumu}, the $P^0$ is of particular interest because it contains only down-type techniquarks (and charged technileptons) and thus has a mass scale that is most naturally set by the mass of the $b$-quark. Other color-singlet PNGB's will have masses most naturally set by $m_t$, while color non-singlet PNGB's will generally be even heavier. The $M_{\pzero}$ mass range, that is typically suggested by technicolor models \cite{technicolor}, is $10~{\rm GeV}<M_{\pzero}<200~{\rm GeV}$. Discovery of the $P^0$ in the $gg\toP^0\to\gamma\gam$ mode at the Tevatron Run II and at the LHC will almost certainly be possible unless its mass is either very small ($\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 30~{\rm GeV}$?) or very large ($\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 200~{\rm GeV}$?), where the question marks are related to uncertainties in backgrounds in the inclusive $\gamma\gam$ channel. Run I data at Tevatron can already be used to exclude a $P^0$ in the $50-200~{\rm GeV}$ mass range for a number of technicolors $N_{TC}> 12-16$. In contrast, an $e^+e^-$ collider, while able to discover the $P^0$ via $e^+e^-\to\gamP^0$, so long as $M_{\pzero}$ is not close to $M_Z$, is unlikely (unless the TESLA $500~{\rm fb}^{-1}$ per year option is built or $N_{TC}$ is very large) to be able to determine the rates for individual $\gamma F$ final states ($F=b\overline b,\tau^+\tau^-,gg$ being the dominant $P^0$ decay modes) with sufficient accuracy such as to yield more than very rough indications about the parameters of the technicolor model. The $\gamma\gam$ collider mode of operation at an $e^+e^-$ collider would allow one to discover and study the $P^0$ with greater precision. A $\mu^+\mu^-$ collider would play a very special role with regard to determining key properties of the $P^0$. In particular, the $P^0$, being comprised of $D\overline D$ and $E\overline E$ techniquarks, will naturally have couplings to the down-type quarks and charged leptons of the SM. Thus, $s$-channel production ($\mu^+\mu^-\toP^0$) is predicted to have a substantial rate for $\sqrt s\simM_{\pzero}$. Because the $P^0$ has a very narrow width (see Fig. \ref{fig6}), not much larger than that of a SM-Higgs boson of the same mass, in order to maximize this rate it is important that one operates the $\mu^+\mu^-$ collider so as to have extremely small beam energy spread, $R=0.003\%$. For such an $R$, the resolution in $E=\sqrt s$ of the muon collider, $\sigma_E$, is of order $\sigma_E\sim 1~{\rm MeV} (E/50~{\rm GeV})$, whereas the $P^0$ width, $\Gamma_{\pzero}$, varies from $2~{\rm MeV}$ to $20~{\rm MeV}$ as $M_{\pzero}$ ranges from $50~{\rm GeV}$ up to $200~{\rm GeV}$ (small differences with respect to \cite{mumu} come from running fermion masses). Thus, $\sigma_E<\Gamma_{\pzero}$ is possible and leads to very high $P^0$ production rates for typical $\mu^+\mu^-\toP^0$ coupling strength. Assuming that the $P^0$ is discovered at the Tevatron, the LHC or (as might be the only possibility if $M_{\pzero}$ is very small) at an $e^+e^-$ collider (possibly operating in the $\gamma\gam$ collider mode), the $\mu^+\mu^-$ collider could quickly (in less than a year) scan the mass range indicated by the previous discovery (for the expected uncertainty in the mass determination) and center on $\sqrt s\simeqM_{\pzero}$ to within $<\sigma_M$. A first very rough estimate of $\Gamma_{\pzero}$ would also emerge from this initial scan. One would then proceed with a dedicated study of the $P^0$. One technique would be to use the optimal three-point scan \cite{mumu} of the $P^0$ resonance (with measurements at $E=M_{\pzero}$ and $E=M_{\pzero}\pm 2\sigma_M$ using $R=0.003\%$). The three-point scan would determine with high statistical precision all the $\mu^+\mu^-\toP^0\to F$ channel rates and give a reasonably accurate measurement of the total width $\Gamma_{\pzero}$. For the particular technicolor model parameters analysed in \cite{mumu}, 4 years of the pessimistic yearly luminosity ($L_{\rm year}=0.1~{\rm fb}^{-1}$) devoted to the scan yields the results presented in Fig.~19 of \cite{mumu}.~\footnote{This figure gives the errors before taking into account the possible variation of the luminosity with $M_{\pzero}$. If the muon collider is built so that the $\sqrt s=100~{\rm GeV}$ luminosity is maximized, then $L_{\rm year}$ will be smaller (larger) than the $\sqrt s=100~{\rm GeV}$ value for smaller (larger) $\sqrt s$. The relevant luminosity scaling laws are those given in Eq.~(7.2) of Ref.~\cite{mumu}. The effects upon the statistical errors quoted below of such luminosity scaling are given in Fig.~20 of \cite{mumu}, but will not be included in our discussion here.} Sample statistical errors for $\sigma_cB(P^0\to {\rm all})$ and $\Gamma_{\pzero}$ taken from this figure at $M_{\pzero}=60~{\rm GeV}$, $80~{\rm GeV}$, $M_Z$, $110~{\rm GeV}$, $150~{\rm GeV}$ and $200~{\rm GeV}$ are given in Table~\ref{fmcerrorspgb}. \begin{table}[h] \caption[fake]{\baselineskip 0pt Fractional statistical errors ($1\sigma$) for $\sigma_cB(P^0\to {\rm all})$ (combining $b\bar b$, $\tau^+\tau^-$, $c\bar c$ and $gg$ tagged-channel rates --- see \cite{mumu}) and $\Gamma_{\pzero}$ for $s$-channel $P^0$ production at the muon collider. We compare results for an $R=0.003\%$ three-point scan with total integrated luminosity of $L=0.4~{\rm fb}^{-1}$ (corresponding to four years of running at pessimistic luminosity, with distribution $L/5$ at $E=M_{\pzero}$, $2L/5$ at $E=M_{\pzero}+2\sigma_M$ and $2L/5$ at $E=M_{\pzero}-2\sigma_M$) to results obtained using the $r_c$-ratio technique assuming accumulated luminosities at $E=M_{\pzero}$ of $L=0.2f~{\rm fb}^{-1}$ at $R=0.003\%$ and $L=0.94(2-f)~{\rm fb}^{-1}$ at $R=0.03\%$ (corresponding to roughly $2f$ years of running at $R=0.003\%$ and $(4-2f)$ years of running at $R=0.03\%$ for pessimistic instantaneous luminosity assumptions). $f$ (tabulated below) is chosen to minimize the error in $\Gamma_{\pzero}$. We employ the efficiencies, cuts and tagging procedures described in \cite{mumu}. The effects of bremsstrahlung are included.} \small \begin{center} \begin{tabular}{|c|c|c|c|c|c|c|} \hline Quantity & \multicolumn{6}{c|}{Errors for the scan procedure} \\ \hline \hline {\bf Mass (GeV)} & {\bf 60} & {\bf 80} & {\bf $M_Z$} & {\bf 110} & {\bf 150} & {\bf 200} \\ \hline $\sigma_cB$ & 0.0029 & 0.0054 & 0.043 & 0.0093 & 0.012 & 0.018\\ \hline $\Gamma_{\pzero}$ & 0.014 & 0.029 & 0.25 & 0.042 & 0.052 & 0.10 \\ \hline \hline Quantity & \multicolumn{6}{c|}{Errors for the $r_c$-ratio procedure} \\ \hline \hline {\bf Mass (GeV)} & {\bf 60} & {\bf 80} & {\bf $M_Z$} & {\bf 110} & {\bf 150} & {\bf 200} \\ \hline $f$ & 0.8 & 0.7 & 0.6 & 0.8 & 0.9 & 1.0 \\ \hline $\sigma_cB$ & 0.0029 & 0.0062 & 0.055 & 0.010 & 0.011 & 0.016 \\ \hline $\Gamma_{\pzero}$ & 0.014 & 0.028 & 0.24 & 0.041 & 0.039 & 0.053 \\ \hline \end{tabular} \end{center} \label{fmcerrorspgb} \end{table} In the analysis performed in \cite{mumu}, we did not consider the errors induced by a systematic error in the energy spread. In Fig. \ref{fig7} we show the $\Delta\sigma_M$-induced fractional errors for the branching ratio $B$ (where $B=B_{\ell^+\ell^-}B_F$ if we focus on a given final state or $B=B_{\ell^+\ell^-}$ if we sum over all final states) and for $\Gamma_{\pzero}$ predicted for the same three-point scan measurement ($E=M,~E=M\pm 2\sigma_M$, $R=0.003\%$) as employed for the statistical error analysis summarized above. Results are shown for the two cases of $\Delta \sigma_M/\sigma_M=0.05$ (solid line) and $\Delta \sigma_M/\sigma_M=0.01$ (dashed line). For $R=0.003\%$, we find $\Delta\Gamma_{\pzero}/\Gamma_{\pzero}\sim c_{\Gamma_{\pzero}} \Delta\sigma_M/\sigma_M$ with $c_{\Gamma_{\pzero}}\sim 1.5$ for $M_{\pzero}\sim 60~{\rm GeV}$ falling to $c_{\Gamma_{\pzero}}\sim 0.45$ for $M_{\pzero}\sim 200~{\rm GeV}$. For $M_{\pzero}\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 80~{\rm GeV}$, the result is that the induced $\Delta\Gamma_{\pzero}/\Gamma_{\pzero}$ systematic errors are comparable to the expected statistical errors even for $\Delta \sigma_M/\sigma_M=0.01$. For example, at $M_{\pzero}=60~{\rm GeV}$ both the systematic error and the statistical error are of order 1.5\%. In the neighborhood of the $Z$ peak the errors from the optimal three-point scan \cite{mumu} are largely dominated by the $Z$ background and the $\Delta \sigma_M/\sigma_M$ effect can be neglected if $\Delta\sigma_M/\sigma_M$ is not large. For $M_{\pzero}\sim 150~{\rm GeV}$ ($\sim 200~{\rm GeV}$), $c_{\Gamma_{\pzero}}\sim 0.6$ ($0.45$), {\it i.e.}\ $\Delta\Gamma_{\pzero}/\Gamma_{\pzero}\sim 0.6\Delta\sigma_M/\sigma_M$ ($0.45\Delta\sigma_M/\sigma_M$), while the statistical $\Delta\Gamma_{\pzero}/\Gamma_{\pzero}$ error from the three-point scan \cite{mumu} would be of the order $5\%$ ($10\%$). Thus, for $\Delta \sigma_M/\sigma_M\leq 0.01$ the systematic error would be much smaller than the statistical error. As a result, for $M_{\pzero}\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 100~{\rm GeV}$, as far as the $\Delta\sigma_M$-induced errors for $\Gamma_{\pzero}$ are concerned, one could consider employing a value of $R$ larger than $0.003\%$. As an example, $R=0.01\%$ could be chosen. Since $\Gamma_{\pzero}/\sigma_M$ is still $\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 1$ for this $R$ and $M_{\pzero}\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 100~{\rm GeV}$, we can expect that systematic errors will still be under control. The actual systematic errors resulting from an $E=M$, $E=M\pm 2\sigma_M$ three-point scan appear in Fig.~\ref{fig7}. For $M_{\pzero}\sim 150~{\rm GeV}$ ($\sim 200~{\rm GeV}$) and $\Delta\sigma_M/\sigma_M=0.01$, one finds systematic error of $\Delta\Gamma_{\pzero}/\Gamma_{\pzero}\sim 0.028$ ($0.022$), which is indeed an acceptable level. However, as described in the previous Section, despite the factor of 2.2 increase [see Eq.~(\ref{lform})] in yearly luminosity achieved by increasing $R$ from $0.003\%$ to $0.01\%$, the decrease in the signal to background ratio is very substantial and would lead to worse statistical errors unless $M_{\pzero}>200~{\rm GeV}$ [for which $\Gamma_{\pzero}/\sigma_M(R=0.003\%)>5$]. Thus, for $M_{\pzero}<200~{\rm GeV}$ and typical model parameters as embodied in the choices of Ref.~\cite{mumu}, one should employ $R=0.003\%$ for the scan. Let us now consider the $r_c$-ratio technique for the $P^0$. We will compare to the scan technique using the choices $R=0.003\%$ for $\sigma_M^{\rm min}$ and $R=0.03\%$ for $\sigma_M^{\rm max}$. This means $\sigma_M^{\rm central}\sim 6.3~{\rm MeV}\,(M_{P^0}/100~{\rm GeV})$. From Fig.~\ref{fig6}, we then find $\Gamma_{\pzero}/\sigma_M^{\rm central}\sim 2/3$ at $M_{\pzero}=50~{\rm GeV}$ rising slowly to 1.6 at $M_{\pzero}=200~{\rm GeV}$. This region is that for which the slope $|s|$ (see Fig.~\ref{dlgdlr}) is smallest. Consequently, the error in $\Gamma_{\pzero}$ will be small if that for $r_c$ is. The $\Delta\sigma_c/\sigma_c$ errors for 3 years of operation at pessimistic instantaneous luminosity ($L=0.3~{\rm fb}^{-1}$) at $R=0.003\%$ were given in Fig.~19 of Ref.~\cite{mumu}. We rescale these errors to $L=0.2f~{\rm fb}^{-1}$ (corresponding to $2f$ years of operation at $R=0.003\%$), where $f$ will be chosen to minimize the error in $r_c$. We also compute $\Delta\sigma_c/\sigma_c$ for $L=0.94(2-f)~{\rm fb}^{-1}$ devoted to $R=0.03\%$ running (corresponding to $4-2f$ years of operation at this latter $R$). The net $\Delta\sigma_c/\sigma_c$ is computed using Eq.~(\ref{sigcerror}) after combining all final state channels. We also compute $\Delta r_c/r_c$ according to the Eq.~(\ref{sigcerror}) procedure. The corresponding statistical error for $\Gamma_{\pzero}$ is then computed using the appropriate $|s|$ slope value. We then search for the value of $f$ (see above) such that $\Delta\Gamma_{\pzero}/\Gamma_{\pzero}$ is smallest. The value of $f$ and the corresponding errors for the combined-channel $\sigma_c$ and for $\Gamma_{\pzero}$ error are given in Table~\ref{fmcerrorspgb} for the same $M_{\pzero}$ values as considered for the three-point scan procedure. For $\sigma_c$, the $r_c$-ratio procedure statistical errors are very similar to the 4-year three-point scan statistical errors for all $M_{\pzero}$ values considered. For $\Gamma_{\pzero}$, the $r_c$-ratio procedure statistical errors are as good as the tabulated 4-year three-point scan statistical errors for $M_{\pzero}<110~{\rm GeV}$, and become superior for larger $M_{\pzero}$ values where $\Gamma_{\pzero}/\sigma_M^{\rm central}$ is significantly larger than unity. As we have discussed, for the $r_c$-ratio procedure, the fractional systematic error in $\Gamma_{\pzero}$ is equal to that in $\sigma_M$ and there is no systematic error in $B\equiv B_{\mu^+\mu^-}B_F$. In contrast, we have seen that for $M_{\pzero}<80~{\rm GeV}$ the systematic errors in $\Gamma_{\pzero}$ and $B$ from the scan technique will be somewhat larger than $\Delta\sigma_M/\sigma_M$. Taking $\Gamma_{\pzero}$ as an example, the scan systematic errors are of order $1.5\Delta\sigma_M/\sigma_M$. As summarized earlier, this means that the scan systematic error for $\Delta\sigma_M/\sigma_M=0.01$ is essentially the same as the scan statistical error computed assuming pessimistic luminosity. For optimistic luminosity the scan systematic error would be dominant. Thus, the $r_c$-ratio procedure will actually give better overall (systematic plus statistical) error for $\Gamma_{\pzero}$ and, especially, $B$ than the scan procedure for low as well as high $M_{\pzero}$ values. This will become especially important if the luminosity available is better than the pessimistic value or if $\Delta\sigma_M/\sigma_M>0.01$. If a narrow resonance is observed in the $\gamma\gam$ final state at the Tevatron or LHC, and if it has the weak coupling to $ZZ$ that is typical of a pseudogoldstone boson, then the $r_c$-ratio technique for precision measurements of its properties at the muon collider is strongly recommended. \bigskip \noindent {\bf{Degenerate BESS}} - The Degenerate BESS model \cite{DBESS} describes two isotriplets of nearly degenerate vector resonances $(\vec L,\vec R)$ characterized by two parameters $(M,g'')$, the common mass (when the EW interactions are turned off) and their gauge coupling. The main feature of the model is the decoupling property, which implies very loose bounds from existing precision experiments \cite{vancouver} as shown in Fig. \ref{fig8}. Another consequence is that the decay of the resonances into pairs of ordinary gauge vector bosons is quite depressed. The total widths for the two neutral states $(L_3,R_3)$ are given by \begin{equation} \Gamma_{L_3,R_3}=M\,h_{L_3,R_3}(g/g'')\,, \end{equation} where $g$ is the weak coupling constant. The behaviour of the total widths as functions of $g/g''$ is shown in Fig. \ref{fig9}, whereas for $g/g''\ll 1$ one has \begin{equation} h_{L_3}\approx 0.068\left(\frac g{g''}\right)^2,~~~ h_{R_3}\approx 0.01\left(\frac g{g''}\right)^2\,. \end{equation} The ratio of the widths, $\Gamma_{L_3}/\Gamma_{R_3}$, in the interval $0\le g/g''\le .5$ approximately varies between $0.15$ and $0.07$. The weak interactions break the mass degeneracy giving rise to the mass splitting \begin{equation} M^s=M_{L_3}-M_{R_3}=M\,h_{M^s}(g/g'')\,. \end{equation} The behaviour of $h_{M^s}$ for $g/g''\ll 1$ is \begin{equation} h_{M^s}\approx (1-\tan^2\theta_W)\left(\frac g{g''}\right)^2\approx 0.7 \left(\frac g{g''}\right)^2\,. \end{equation} The branching ratios into charged lepton pairs are almost parameter independent and rather sizeable \begin{equation} B(L_3\to\ell^+\ell^-)\approx 4.5\%,~~~~B(R_3\to\ell^+\ell^-)\approx 13.6\%\,. \end{equation} In this Section, we will assume that $\sigma_M$ is much smaller than $M^s$, and therefore we can apply the previous analysis to the two resonances separately. The case $\sigma_M\approx M^s\gg \Gamma_{L_3,R_3}$ will be studied in Section 4. By combining Fig. \ref{fig2} with Fig. \ref{fig9}, we easily obtain the $\Delta\sigma_M$-induced fractional errors in the branching ratios and in the widths of the two resonances $L_3$ and $R_3$ as functions of $g/g''$ for different choices of $\sigma_M$ [equivalently, $R$; see Eq. (\ref{spread})]. The results are given in Figs. \ref{fig10} and \ref{fig11} for $L_3$ and $R_3$, respectively, for the choices $\Delta\sigma_M/\sigma_M=1\%$ and 5\%, and for $R=1\%$ and 0.1\%. (Because the resonance widths are large, we do not need the very small values of $R$ required to study the SM Higgs or the $P^0$.) Combining these results with the bounds on the portion of parameter space still allowed by precision experiments, one can put lower limits on the masses of the resonances such that they have not been excluded and yet one is able to measure the widths with no more than a given systematic uncertainty from $\Delta\sigma_M$. For instance, in the case of a machine with $R=1\%$ (typical of an $e^+e^-$ collider) and for $\Delta\sigma_M/\sigma_M=1\%$, one finds, from Fig. \ref{fig9}, that $g/g''\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.3$ is needed to avoid $\Delta\sigma_M$-induced errors in $\Gamma_{L_3}$ above $2\%$. From Fig. \ref{fig8}, the portion of parameter space that is still allowed by precision experiment can be roughly expressed by \begin{equation} \frac{M}{1000~{\rm GeV}}\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 2.17\left(\frac g{g''}+0.01\right)\,. \end{equation} Therefore, the $g/g''\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.3$ requirement converts to the requirement that $M\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 670~{\rm GeV}$. This means that, at lepton colliders with $R=1\%$ and $\Delta\sigma_M/\sigma_M=1\%$, one will be able to measure $\Gamma_{L_3}$ (for an allowed $L_3$ resonance) with a $\Delta\sigma_M$-induced error of less than 2\% only if the mass of the resonance is greater than 670 GeV. Similar considerations apply to $R_3$, where keeping the systematic error in $\Gamma_{R_3}$ below $2\%$ requires $g/g''\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.7$, leading to $M\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 1540$ GeV. In the case of a muon collider, $R=0.1\%$ can be achieved while maintaining large luminosity. In this case, the $\Delta\sigma_M$-induced fractional error for $L_3$ is below $2\%$, for $\Delta\sigma_M/\sigma_M=1\%$, if $g/g''\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.12$, which converts to $M\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 280~{\rm GeV}$ for $L_3$ resonances not already excluded by the precision data. The corresponding limits for $R_3$ are $g/g''\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.3$ and $M\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 670~{\rm GeV}$. In particular, we see from Fig. \ref{fig8} that, at a machine working at the top threshold (about 350 GeV), only resonances with $g/g''\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.14$ have not already been excluded by the precision data. Therefore, only a machine with $R\le 0.1\%$ would be able to measure the width and branching ratios of an allowed resonance without encountering significant systematic uncertainty coming from $\Delta\sigma_M/\sigma_M\sim 0.01$. \bigskip \section{Analysis for nearly degenerate resonances} \bigskip In this Section we will discuss nearly degenerate resonances, {\it i.e.}\ the case of a mass splitting much smaller than the average mass of the resonances, $M^s=M_2-M_1\ll M=(M_1+M_2)/2$. We will be interested in the case where the energy spread of the beam is of the same order of magnitude as the mass splitting between the two peaks, $\sigma_M\approx M^s$. We will also assume that the widths of the two resonances are much smaller than the mass splitting, {\it i.e.}\ $\Gamma_1,\,\Gamma_2\ll M^s$. It follows also that $\Gamma_1,\,\Gamma_2\ll\sigma_M$. In this approximation, we can safely describe the cross section as the sum of two Breit-Wigner functions, and furthermore we can use the narrow width approximation. Therefore,~\footnote{We focus on the total cross section, but it should be kept in mind that we could also consider the cross section in a given final state. In this case, $\Gamma(R_i\to\ell^+\ell^-)$ ($i=1,2$) should be replaced by $\Gamma(R_i\to\ell^+\ell^-)B_F$ in all that follows. The $\Delta\sigma_M$-induced systematic errors on this product would be the same as for $\Gamma(R_i\to\ell^+\ell^-)$.} \begin{equation}\label{bw2} M^2\sigma_c= B_{\ell^+\ell^-}^1\Phi(x_1,\gamma_1)+B_{\ell^+\ell^-}^2\Phi(x_2,\gamma_2)\,, \end{equation} where $B_{\ell^+\ell^-}^1$ and $B_{\ell^+\ell^-}^2$ are the branching ratios of the resonances into $\ell^+\ell^-$, and \begin{equation}\label{def2} x_1=\frac{E-M_1}{\sigma_M},~~~x_2=\frac{E-M_1-M^s}{\sigma_M}=x_1-m^s \end{equation} and \begin{equation}\label{def3} M_2=M_1+M^s,~~~m^s=\frac{M^s}{\sigma_M}\,. \end{equation} We have also assumed $M\approx M_1\approx M_2$. For a Gaussian beam we recall that the function $\Phi(x,\gamma)$ defined in Eq. (\ref{scaled-conv}) is given by \begin{equation}\label{Phi} \Phi(x,\gamma)=\sqrt{\frac \pi 2} (2 j+1)\int_{-\infty}^{+\infty}e^{-(x-y)^2/2} \frac{\gamma^2}{y^2+\gamma^2/4}dy\,. \end{equation} We may evaluate this expression by performing the Fourier transforms of the Gaussian distribution and of the Breit-Wigner and then taking the inverse Fourier transform. We get \begin{equation}\label{integration} \Phi(x,\gamma)= (2j+1)\pi^{3/2}\gamma\int_{-\infty}^{+\infty}dp e^{-p^2/2+ipx} \left(\theta(p)e^{-p\gamma/2}+ \theta(-p)e^{+p\gamma/2}\right)\,. \end{equation} Since we are assuming that the resonance widths are much smaller than the energy spread, we may evaluate Eq.~(\ref{integration}) in the $\gamma\to 0$ approximation, yielding \begin{equation}\label{approx} \Phi(x,\gamma)=(2j+1)\gamma\sqrt{2}\pi^{3/2} e^{-x^2/2}\,, \end{equation} as earlier given in Eq.~(\ref{smallgam}) for $j=0$. As previously noted, this expression shows that when the energy spread is much bigger than the width the convolution gives rise to a Gaussian function with spread $\sigma_M$, and we loose any information about the total width. In fact, the total cross section $\sigma_c$ depends only on the product $B_{\ell^+\ell^-}\gamma=\Gamma(R\to\ell^+\ell^-)/\sigma_M$. In the case of two resonances with $\gamma_1,\,\gamma_2\to 0$, we thus find \begin{equation}\label{final cross} M^2\sigma_c=(2j+1)\sqrt{2}\pi^{3/2}\left( g_1e^{-x_1^2/2}+g_2 e^{-(x_1-m^s)^2/2}\right)\,, \end{equation} where \begin{equation} g_i=\gamma_iB^i_{\ell^+\ell^-}=\frac{\Gamma(R_i\to\ell^+\ell^-)}{\sigma_M}\,. \end{equation} The function (\ref{final cross}) is invariant under the substitution \begin{equation} g_1\leftrightarrow g_2,~~~x_1\leftrightarrow m^s-x_1 \label{invariance} \end{equation} The behaviour of the function is characterized by the ratio \begin{equation} \frac{g_2}{g_1}=\frac{\Gamma(R_2\to\ell^+\ell^-)}{\Gamma(R_1\to\ell^+\ell^-)} \equiv a \end{equation} and by $m^s$. For small $m^s$, the convolution of the Gaussian with the two Breit-Wigners has a single maximum in between 0 and $m^s$ depending on the value of $a$. For instance, for $a=1$ the maximum is at $m^s/2$. In this situation, the second derivative of the function (\ref{final cross}) has two zeros corresponding to the changes of curvature before and after the peak. By increasing $m^s$ the second derivative acquires a third zero (in fact, a double zero). This is due to the effect of the smaller Breit-Wigner which gives rise to a further change of the curvature. Just to get an idea, we list in Table 5, for several choices of $a$, the critical value $m^s_1$ of $m^s$ at which this third zero occurs. As $m^s$ is increased further, there comes a point at which the two Breit-Wigner maxima start to show up. The minimum value of $m^s$ required to see two maxima, $m^s_2$, is given as a function of the ratio $a$ in Table \ref{doublezero}. Notice that the invariance (\ref{invariance}) implies $m^s_i(a)=m^s_i(1/a)$. \begin{table}[bht] \caption{Values of $m^s$ at which the double zero of the second derivative of the function (\ref{final cross}) occurs ($m^s_1(a)$) and at which the two maxima of the function (\ref{final cross}) start to show up ($m^s_2(a)$).} \begin{center} \begin{tabular}{|c|c|c|} \hline $a$ & $m^s_1(a)$ & $m^s_2(a)$ \\ \hline\hline 1 & 2 &2\\ \hline 2 & 1.85 & 2.63 \\ \hline 3 & 2.02 & 2.85 \\ \hline 4 & 2.20 & 2.98\\ \hline 5 & 2.31 & 3.08\\ \hline \end{tabular} \end{center} \label{doublezero} \end{table} We can now discuss the type of measurements necessary to determine the parameters $M_1$, $M_2$, $g_1$ and $g_2$. In this discussion, we will use the notation $x\equiv x_1$. We first determine the overall location in energy of the resonance structure by locating the absolute maximum of the cross section. For small $m^s$, such that the individual resonance peaks are unresolved, the cross section has a single maximum near the location of the resonance with the larger $g$; we will assume that it is $g_2$ which is largest. For $m^s$ large enough that the peaks are resolved, we may locate the larger maximum. We denote the location of the maximum in the cross section by $E_{\rm max}$, and choose this as our first energy setting. We write \begin{equation} E_{\rm max}=\sigma_{M} x_{\rm max} +M_1\,, \label{energyscale} \end{equation} where $x_{\rm max}$ is a function of $a$ and $m^s$ which can be evaluated numerically from Eq. (\ref{final cross}). For instance, for $a\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 2$ it is a good approximation to assume $x_{\rm max}\approx m^s$ (specially for $m^s$ bigger than the critical value $m^s_2$), implying $E_{\rm max}\simeq M_2$. The value of the cross section at $x_{\rm max}$ provides a second input into determining $M_{1,2}$ and $g_{1,2}$. To complete the process of determining these four parameters, we need two more measurements. If $m^s$ is larger than $m_2^s$, so that a second (lower) maximum is present, we can use the location of the second maximum and the cross section value at this second maximum as our two additional measurements. We effectively have four equations in four unknowns, two equations involving the derivatives of Eq.~(\ref{final cross}) and two involving the absolute magnitude of Eq.~(\ref{final cross}). If no second maximum is present, then we must effectively determine the slope of $\sigma_c(x)$ of Eq.~(\ref{final cross}) at some energy location away from $x_{\rm max}$ and determine the cross section at this same location. Measurements of $\sigma_c(x)$ at two nearby values of $x$ away from $x_{\rm max}$ are needed. That these approaches are really equivalent becomes apparent when we realize that the $m^s>m_2^s$ procedure of locating the second maximum actually requires measuring the slope of $\sigma_c(x)$ and finding its second zero. (Recall that this discussion assumes infinite statistics so that we will end up effectively computing the minimum error that will be induced by uncertainty in $\sigma_M$.) In the absence of a second maximum, the choice of the other two energies can be difficult if $m^s$ is smaller than the critical value $m^s_1$. In fact, in this case the convolution of the two Breit-Wigner looks very much like the convolution with a single Breit-Wigner. Not surprisingly, to obtain reasonable errors it is necessary that $\sigma_M$ be such that $m^s$ is at least bigger than $m^s_1$. Let us assume that for $m^s>m^s_1$ we can approximately locate the energy corresponding to $x\sim 0$ and that we measure the cross section at two points in its vicinity (as well as at $x=x_{\rm max}$). If $m^s$ is greater than $m^s_2$, we may continue to employ the above procedure or we may use the alternative procedure outlined earlier based on the fact that the cross section has two maxima, one near $x=0$ and one near $x=m^s$; in the alternative procedure, we measure the energy and the cross section at the two peaks. We consider first the former procedure that is the only choice if $m^s<m^s_2$. We start our analysis by taking $a=2$ and then later discuss the modifications for different values of $a$. From the measurement which fixes the energy scale (see Eq. (\ref{energyscale})) we find the following relation between the parameter errors and the uncertainty in the energy spread: \begin{equation} \left(1+\frac{\Delta\sigma_M}{\sigma_M}\right)x_{\rm max}(m^s+\Delta m^s) +\Delta m_1-x_{\max}(m^s)=0\,, \label{energy1} \end{equation} where $\Delta M_1=\sigma_M\Delta m_1$ is the error in $M_1$ and $\Delta m^s$ is the error in $m^s$. From this equation we can eliminate $\Delta m_1$ in terms of the other errors. This has been done by using the approximation $x_{\rm max}\approx m^s$. The errors on $M^s$ and on the partial widths can then be determined using the three cross sections --- $\sigma_c(x_{\rm max})$ and $\sigma_c(x)$ at two other $x$ values near 0 --- following a procedure analogous to that discussed for a single resonance using Eq. (\ref{condition1}). In particular, we assume measurements of the cross section at $x_{\max}$ and at $x=0.1$ and $x=0.2$. The resulting fractional errors for $\Gamma(R_i\to\ell^+\ell^-)$ ($i=1,2$) and $M^s$ are given in Fig. \ref{fig12} as a function of $M^s/\sigma_M$ for $\Delta\sigma_M/\sigma_M=5\%$. As expected, the errors grow rapidly once $m^s$ falls below $m^s_1$. As $m^s$ is increased above $m^s_1$, there is a change of curvature in the fractional error curves around the critical value $m^s_2$, after which the fractional errors approach asymptotic limits. Notice that the asymptotic value of $\Delta M^s/M^s$ is zero, whereas $\Delta\Gamma(R_i\to\ell^+\ell^-)/\Gamma(R_i\to\ell^+\ell^-)\to \Delta\sigma_M/\sigma_M$ for both $i=1,2$. This is because the cross sections depend only on the ratio $\Gamma(R_i\to\ell^+\ell^-)/\sigma_M$. In Fig. \ref{fig13} we represent the same quantities but for $a=3$, and we see results very similar to those for $a=2$ except for changes due to the different values of the critical points $m^s_1$ and $m^s_2$. We have tried different choices for the two measured points off the maximum, varying them up to $x=0.3$, without any significant change in the results. We now consider the alternative procedure outlined earlier that is possible when two cross section maxima become visible, that is when the mass splitting is bigger than $m^s_2$. In this case, the four measurements for determining $M_{1,2}$ and $g_{1,2}$ are the energy locations of the two maxima and the cross sections at these two maxima. The measurement of the energy of the second maximum gives rise to a condition similar to the one of Eq. (\ref{energy1}). The resulting fractional errors in $M^s$ and $\Gamma(R_i\to\ell^+\ell^-)$ are given in Fig. \ref{fig14} (for $a=2$), and are essentially the same as obtained in the previous procedure when $m^s>m^s_2$. The basic conclusion from these analyses is that for $m^s$ of the order of the critical value $m^s_2$, the fractional errors in the parameters of the resonances are of the order of the fractional error in $\sigma_M$. For smaller $m^s$ the errors become very large. For $m^s$ significantly bigger than $m^s_2$, the fractional error in the mass splitting rapidly approaches zero while the fractional errors for the $\Gamma(R_i\to\ell^+\ell^-)$ partial widths become equal to the fractional error in the energy spread. In short, we can use the critical value $m^s_2$ in order to discriminate between a good and a bad determination of the mass splitting. \bigskip \section{Application to Degenerate BESS} \bigskip In Degenerate BESS one can show that the condition $M^s \gg\Gamma_L,\,\Gamma_R$ is rather well satisfied (by one and two orders of magnitude respectively). Therefore we can apply the analysis of the previous Section. From Fig. \ref{fig15}, we see that the value of $a$ is almost constant and approximately 2.2 for $g/g''$ up to 0.2, and then $a$ increases up to $\approx 4$ for $g/g''=0.5$. As discussed in the last Section, we can use the values of $m^s_2$ given in Table \ref{doublezero} in order to determine the minimum value of $M^s/\sigma_M$ needed in order to make a good determination of the mass splitting. For $a=2.2$ one finds that the minimum value is $m^s_2=2.68$. As in our earlier single resonance discussions, for any fixed value of the energy resolution $R$, we convert this bound into lower bounds for $g/g''$ and for the mass $M$ of nearly degenerate resonances that have not already been excluded by precision experimental data. From $M^s/\sigma_M\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 2.68$ we get \begin{eqnarray} R=1\% &\to& \frac g{g''}\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.16,~~~ M\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 370~{\rm GeV}\,,\nonumber\\ R=0.1\% &\to& \frac g{g''}\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.05,~~~ M\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}} 130~{\rm GeV}\,.\nonumber \end{eqnarray} We see that for $R=1\%$ a machine with energy near the top threshold would just be at the border of being able to accurately measure the mass difference between the $L_3$ and $R_3$ resonances not already excluded by precision data. \bigskip \section{Conclusions} We have considered the production of a narrow resonance via $s$-channel collisions of leptons ($\ell=e$ or $\mu$). Here, a `narrow' resonance is defined as one that has width $\Gamma$ substantially smaller than the beam energy spread $\Delta E_{\rm beam}$ that is natural for the collider (and therefore is associated with the largest instantaneous luminosity). It will be convenient to use the parameterization $\Delta E_{\rm beam}=0.01R\,E_{\rm beam}$, where $R$ is in per cent. For example, at a muon collider with center of mass energy $E\sim 100~{\rm GeV}$, $R\sim 0.12\%$ allows for maximal $L$ and $L$ declines rapidly as $R$ is forced to smaller values by compression techniques. A resonance with width $\Gamma\ll 0.001M$ would then be narrow. Our focus has been on the systematic error that might be introduced into measurements of the parameters of a narrow resonance due to systematic uncertainty in the value of $\Delta E_{\rm beam}/E_{\rm beam}$. The important parameters that can be measured are the branching ratio of the resonance into the charged leptons ({\it i.e.}\ those that are being collided), the product of this leptonic branching ratio times that for the resonance to decay to a particular final state, and the total width of the resonance. We examined four methods for determining the resonance parameters: (1) a scan of the resonance; (2) sitting on the resonance and changing the beam energy resolution; (3) measurement of the cross section in the $\ell^+\ell^-$ final state; and (4) measurement of the Breit-Wigner area. Methods (3) and (4) avoid the introduction of systematic errors due to uncertainty in $\Delta E_{\rm beam}/E_{\rm beam}$, but for the integrated luminosities that are anticipated to be available the statistical errors associated with these techniques would be quite large for a narrow resonance. Methods (1) and (2) can provide resonance parameter determinations with small statistical error. However, even in the limit of infinite statistical accuracy, determinations of the resonance parameters are sensitive to systematic uncertainties in $\Delta E_{\rm beam}$ if $\Gamma$ is not much larger than $\Delta E_{\rm beam}$. At a muon collider, the smallest $R$ that can be achieved is expected to be $R=0.003\%$, for which $\Delta E_{\rm beam}\sim \Gamma_H$ for a light SM Higgs boson and $\Delta E_{\rm beam}\sim \Gamma_{\pzero}/2$ for the lightest pseudo goldstone boson of a technicolor model. Consequently, in these and other similar cases, a detailed assessment of the systematic errors in resonance parameter determinations introduced by uncertainty in $\Delta E_{\rm beam}$ is very important. We have performed a general analysis to determine the (systematic) errors in the measured resonance parameters induced by a systematic uncertainty in $\Delta E_{\rm beam}$. We find that the induced fractional errors in the leptonic branching ratio (and also the product of the leptonic branching ratio times the branching ratio into any given final state) and in the total width can be expressed as universal functions of the ratio $\Gamma/\sigma_M$, where $\sigma_M$ is the spread in total center of mass energy resulting from the beam energy spreads: $\sigma_M/M=0.01R/\sqrt 2$. In the case of the minimal three-point scan, with sampling at $E=M,E=M\pm k\Gamma$, the error functions also depend on $k$. For a minimal three-point scan with $k=1$, the induced $\Delta\Gamma/\Gamma$ fractional systematic errors were parameterized as a function of $\Gamma/\sigma_M$ in Eq.~(\ref{dgamform}). Very roughly, for $\Gamma/\sigma_M\sim 2.5$ we find that the induced fractional errors in $\Gamma$ and $B=B_{\ell^+\ell^-}$ or $B_{\ell^+\ell^-}B_F$ ($F$=final state) are of the order of the fractional uncertainty $\Delta\sigma_M/\sigma_M$. As $\Gamma/\sigma_M$ increases above 2, the fractional errors smoothly decrease. For values of $\Gamma/\sigma_M$ below 1, the fractional errors in the resonance parameters increase very rapidly. For example, for $\Gamma/\sigma_M\sim 1$ ($\sim 0.2$) the $\Delta\sigma_M$-induced fractional systematic errors in the resonance parameters increase to $\sim 3.5\div 4\Delta\sigma_M/\sigma_M$ ($\sim 20\div 25\Delta\sigma_M/\sigma_M$), for the $k=1$ scan. Thus, to avoid large systematic errors from $\Delta\sigma_M$, it is imperative to operate the collider with $\Gamma/\sigma_M$ no smaller than 1. If $R$ can be adjusted to achieve values significantly larger than 1, one can consider how to optimize the choice of $R$ so as to minimize the total statistical plus systematic error. A discussion was presented leading to the following two basic conclusions. (a) For a broad resonance, defined as one with $\Gamma\gg0.001M$, one should operate the muon collider at its natural $R$ value of order $0.12\%$. The $\Delta\sigma_M$-induced errors will be very tiny, both because $\Gamma/\sigma_M$ is very large and because $\Delta\sigma_M/\sigma_M$ should be small for such $R$. (b) For a resonance with $\Gamma<0.001M$, one will typically wish to operate at an $R$ that is significantly smaller than that value which would yield equal statistical and systematic errors. In a typical case, this would mean a value of $R$ such that $\Gamma/\sigma_M$ is larger than $5\div 10$. Of course, if the resonance is extremely narrow, it may happen that $\Gamma/\sigma_M$ is of order, or not much larger than, unity even for $R=0.003\%$. In this case, it will normally be essential to run with $R=0.003\%$ even though this $R$ yields the smallest machine luminosity. Larger values of $R$ lead to a drastic decline in the signal to background ratio in a typical final state that, in turn, leads to very poor statistical errors (given the rather slow compensating increase with $R$ of the instantaneous luminosity). For a narrow resonance with $\Gamma\ll 0.001M$, the technique in which one sits on the resonance peak and measures the cross section for two different values of $\sigma_M$ ($\sigma_M^{\rm max}>\Gamma$ and $\sigma_M^{\rm min}<\Gamma$) is a strong competitor to the scan technique. $\Gamma/\sigma_M^{\rm central}$ (where $\sigma_M^{\rm central}=\sqrt{\sigma_M^{\rm max}\sigma_M^{\rm min}}$) is determined by the ratio $r_c=\sigma_c(\sigma_M^{\rm min})/\sigma_c(\sigma_M^{\rm max})$, where $\sigma_c$ is the measured cross section. For $\sigma_M^{\rm max}/\sigma_M^{\rm min}$ of order 5 to 20, the statistical error in $\Gamma/\sigma_M^{\rm central}$ is smallest if $\Gamma/\sigma_M^{\rm central}\sim 2\div 3$. The larger $\sigma_M^{\rm max}/\sigma_M^{\rm min}$, the smaller the statistical error for $\Gamma/\sigma_M^{\rm central}$. For a typical choice of $\sigma_M^{\rm max}/\sigma_M^{\rm min}=10$, one finds a statistical error of $\Delta\Gamma/\Gamma\sim 1.8\Delta r_c/r_c$ for $\Gamma/\sigma_M^{\rm central}\sim 2\div 3$. This technique has the advantage that the $\Delta\sigma_M$-induced systematic error in $\Gamma$ is simply given by $\Delta\Gamma/\Gamma=\Delta\sigma_M/\sigma_M$, while there is no systematic error in the determination of any of the $B\equiv B_{\mu^+\mu^-}B_F$ branching ratio products ($F$ = a particular final state). Of course, if the resonance is very narrow ({\it e.g.}\ as narrow as a light SM Higgs boson or a light pseudogoldstone boson), $\Gamma/\sigma_M^{\rm central}\sim 2\div 3$ will not be achievable. In this case, the best that one can do is to employ $\sigma_M^{\rm min}$ ($\sigma_M^{\rm max}$) as given by $R=0.003\%$ ($R=0.03\%$). The statistical error in $\Gamma$ for such a situation is typically still very good. Let us now summarize how these results apply in the specific cases we explored. In the case of a three-point scan of the SM Higgs boson, we have shown that in the region $M_H\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 110~{\rm GeV}$ it is mandatory to have $R\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.003\%$. In fact, even for this very small $R$ value, $\Gamma_H/\sigma_M$ is still $\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 1$, the very minimum needed for accurate measurements of resonance parameters. For $\Gamma_H/\sigma_M\sim 1$ the fractional systematic errors induced in $\Gamma_H$ from uncertainty in the beam energy spread are of order $3\Delta\sigma_M/\sigma_M$ for a $k=2$ scan. This should be compared to the typically-expected statistical errors tabulated in Table~\ref{fmcerrors}. For example, for $M_H=110~{\rm GeV}$ the statistical error in $\Delta\Gamma_H/\Gamma_H$ is $\sim 5\%$ for optimistic 4-year integrated luminosity of $L=4~{\rm fb}^{-1}$ at $R=0.003\%$; $\Delta\sigma_M/\sigma_M\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.01$ would be needed for the systematic error to be smaller than the statistical error. For the pessimistic 4-year integrated luminosity of $L=0.4~{\rm fb}^{-1}$, the statistical error would be much larger ({\it e.g.}\ $\Delta\Gamma_H/\Gamma_H\sim 16\%$ at $M_H=110~{\rm GeV}$) and the $\Delta\sigma_M$-induced error would be much smaller than the statistical error if $\Delta\sigma_M/\sigma_M\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.01$. Note, however, that increasing $R$ is not appropriate as this would push one into the $\Gamma_H/\sigma_M<1$ region, implying large statistical errors and still larger systematic errors. For the $\sigma_M^{\rm max,min}$ on-peak ratio technique, one must choose $\sigma_M^{\rm min}$ corresponding to $R=0.003\%$ ($\Gamma_H/\sigma_M^{\rm min}\sim 1$). Results for statistical errors were presented in Table~\ref{fmcerrorsrc}. As a point of comparison, for optimistic (pessimistic) instantaneous luminosity and 4 years of operation, the net production rate error after summing over important channels is of order 0.8\% (2.7\%) for $M_H=110~{\rm GeV}$, and the $\Delta\Gamma_H/\Gamma_H$ statistical error is of order 8\% (25\%). Although the statistical $\Delta\sigma_c/\sigma_c$ fractional error is somewhat smaller for the ratio technique than for the scan technique, the $\Delta\Gamma_H/\Gamma_H$ statistical error is larger. However, the ratio technique might still be better if $\Delta\sigma_M/\sigma_M$ were as large as $5\%$, especially if the optimistic luminosity level is available. This is because the systematic error in $\Gamma_H$ is equal to $\Delta\sigma_M/\sigma_M$ for the ratio technique as opposed to $3\Delta\sigma_M/\sigma_M$ for the scan technique. The $r_c$-ratio technique becomes increasingly superior as the assumed luminosity increases. For $M_H\geq 130~{\rm GeV}$, the ratio technique gives smaller statistical errors for both $\sigma_cB$ and $\Gamma_H$ than does the scan technique (for which statistical errors rapidly become very large). Indeed, the two procedures are nicely complementary in that at least one of them will allow a measurement of $\Gamma_H$ with statistical accuracy below 6\% (20\%) for optimistic (pessimistic) luminosity. For the lightest PNGB ($P^0$) of an extended technicolor model, the $\Delta\sigma_M$-induced errors for the three-point scan method can be kept smaller than in the case of the SM Higgs boson. This is because, for typical model parameters, the $P^0$ has a width that is larger than that of a SM Higgs boson; $\Gamma_{P^0}/\sigma_M>2$ is quite likely for $R\approx 0.003\%$. For example, the parameter choices of \cite{mumu} give $\Gamma_{P^0}\sim 5~{\rm MeV}$ vs. $\sigma_M\sim 2~{\rm MeV}$ at $M_{P^0}=100~{\rm GeV}$. As we have seen, the resulting $\Gamma_{P^0}/\sigma_M\sim 2.5$ yields $\Delta\sigma_M$-induced resonance parameter fractional errors of order $\Delta\sigma_M/\sigma_M$. This can be compared to the statistical errors given in Table~\ref{fmcerrorspgb}, computed assuming the pessimistic 4-year integrated luminosity of $L=0.4~{\rm fb}^{-1}$. For example, at $M_{\pzero}=110~{\rm GeV}$ the fractional statistical error for $\Gamma_{\pzero}$ would be $\sim 0.04$, which is much larger than the systematic error if $\Delta\sigma_M/\sigma_M\sim 0.01$. For $M_{\pzero} <80~{\rm GeV}$, the statistical error in $\Gamma_{\pzero}$ declines to the $1\%\div 3\%$ level while the systematic error for $\Delta\sigma_M/\sigma_M=0.01$ rises to about this same level. As $M_{\pzero}$ increases from $150~{\rm GeV}$ to $200~{\rm GeV}$, the statistical error for $\Gamma_{\pzero}$ rises from $\sim 5\%$ to $\sim 10\%$ while the systematic error is below $1\%$ if $\Delta\sigma_M/\sigma_M=0.01$. For optimistic integrated luminosity of $L=4~{\rm fb}^{-1}$, the statistical errors would be smaller than quoted above. For $\Delta\sigma_M/\sigma_M=0.01$, the induced systematic errors would generally dominate for $M_{\pzero}\leq 80~{\rm GeV}$. The $P^0$ resonance is sufficiently narrow that one should also consider using the $\sigma_M^{\rm max,min}$ on-peak ratio technique to determine $\Gamma_{\pzero}$. For 4-year pessimistic luminosity operation we find the statistical errors for $\Gamma_{\pzero}$ given in Table~\ref{fmcerrorspgb}. For the on-peak ratio technique, the systematic error in $\Gamma_{\pzero}$ is equal to $\Delta\sigma_M/\sigma_M$ and for lower $M_{\pzero}$ values would only be smaller than the statistical $\Gamma_{\pzero}$ error if $\Delta\sigma_M/\sigma_M\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}} 0.01$. For optimistic luminosity, the $1\%$ systematic error induced in $\Gamma_{\pzero}$ for $\Delta\sigma_M/\sigma_M=0.01$ would dominate over the statistical error for all but $M_{\pzero}\sim M_Z$ and $M_{\pzero}>150~{\rm GeV}$. Most importantly, the statistical and systematic errors of the ratio technique are at least as good as, and often better than, obtained using the scan technique. For $M_{\pzero}\leq 110~{\rm GeV}$, the statistical $\Gamma_{\pzero}$ errors of the ratio technique are almost the same as obtained via the three-point scan (performed with $R=0.003\%$), while the systemic $\Gamma_{\pzero}$ errors from the ratio technique are smaller ($\Delta\sigma_M/\sigma_M$ vs. $\sim 1.5\Delta\sigma_M/\sigma_M$). For $M_{\pzero}\geq 120~{\rm GeV}$, the $\Delta\sigma_M$-induced systematic errors are comparable for the two techniques, but the statistical errors for the ratio technique are substantially smaller than for the three-point scan. Precision measurements of the properties of a $P^0$ resonance would, thus, always be best performed using the ratio procedure. In the case of the two resonances of the Degenerate BESS model, we have determined (for several typical $R$ values) the region of the model parameter space for which the fractional errors in the resonance properties ($\Gamma$, $\ldots$) induced by $\Delta\sigma_M$ are less than a given fixed value. Induced errors are small for large resonance masses. But, for any given choice of $R$, as the resonance mass is decreased, while maintaining model parameter choices such that the model is still consistent with precision experimental data, there comes a point at which, even for $\Delta\sigma_M/\sigma_M\sim 0.01$, the induced error becomes large. For allowed model parameter choices yielding a resonance mass below this, the resonance's properties cannot be measured accurately. The lowest masses of the resonances that correspond to precision-data-allowed BESS model parameters, and for which $\Delta\sigma_M$-induced errors in the measured resonance properties are small, decrease rapidly with decreasing $R$. As a result, the ability to achieve $R<0.1\%$ at a muon collider would be crucial for exploring the low-resonance-mass portions of Degenerate BESS parameter space not currently excluded by precision data. Finally, we have performed the analysis of two nearly degenerate resonances, a situation encountered in a number of theoretical examples, including the Degenerate BESS model and the minimal supersymmetric model. We focused on the case in which the total widths of the resonances are much smaller than the mass splitting. We have shown that, in general, the $\Delta\sigma_M$-induced fractional error in the measured mass splitting, $M^s$, and in the leptonic partial widths of the resonances (or leptonic partial widths times final state branching fraction) depend only on $M^s/\sigma_M$ and on the ratio of the two partial widths, $a$. The main result is that the errors are generally big for $M^s/\sigma_M$ less than a certain critical value (typically in the $2\div 3$ range) that is a function of $a$. As $M^s/\sigma_M$ increases beyond the critical value, the $\Delta\sigma_M$-induced fractional error in the mass splitting approaches zero rapidly, whereas the fractional errors in the partial widths approach $\Delta\sigma_M/\sigma_M$. As a concrete case, we have discussed the application to the two spin-1 resonances of the Degenerate BESS model for beam energy spread of the same order as the mass splitting between them. We determined the regions of the model parameter space in which $M^s$ could be measured with $\Delta\sigma_M$-induced fractional error below a given fixed value assuming a given value of $R$. The smaller the value of $R$ that can be used while maintaining sufficient luminosity for small statistical errors, the larger the fraction of allowed parameter space for which $M^s$ can be measured with small $\Delta\sigma_M$-induced error. In particular, $R<0.1\%$ is required if we are to be able to separate the degenerate resonance peaks for model parameter choices not already excluded by precision experimental data in which the resonance masses are as low as $100~{\rm GeV}$. \bigskip \noindent{\bf Acknowledgements} We wish to thank A. Blondel for the interesting discussions during the CERN Workshop on muon colliders, where this paper was conceived. JFG is supported in part by the U.S. Department of Energy and by the Davis Institute for High Energy Physics.
\section{How many {\it eB} neutrinos are needed?} SK has recently presented a measurement of the energy spectrum of recoil electrons from solar neutrino scattering, corresponding to 504 days of data taking \cite{sk98}. By assuming the SSM estimate of the {\it hep} neutrino flux \cite{bp98}, $\Phi_{hep}^{SSM}\simeq2\times10^{3}~\rm{cm}^{-2}~\rm{s}^{-1}$ and an undeformed $^{8}B$ neutrino spectrum, with an arbitrary normalization, they obtained a $\chi^2/D.O.F.=25.3/15$, corresponding to a 4.6 \% confidence level \cite{sk98}. The poor fit is due mainly to the behaviour of the energy-bins above $13~\rm{MeV}$. Escribano et al. \cite{esc} suggested that a {\it hep} flux significantly larger than the SSM estimate could reproduce the observed spectrum. Bahcall et al. \cite{hep} have shown that a flux $\Phi_{hep} \ge 20 \times \Phi_{hep}^{SSM}$ could actually mimic the SK spectrum. Alternatively, one can keep the SSM prediction for {\it hep} neutrinos and look for other high energy neutrino sources. Since the average energy of {\it eB} neutrinos is roughly twice than that of {\it hep} neutrinos and since the $\nu-e$ scattering cross section increase linearly with energy, one expects that a flux $\Phi_{eB}\simeq 10\times\Phi_{hep}^{SSM}\simeq 2\times 10^{4}~\rm{cm}^{-2} ~\rm{s}^{-1}$ could be sufficient to account for the high energy behaviour of SK data. In order to provide a quantitative estimate, let us analyse the data by using as free parameters $\alpha=\Phi_{B}/\Phi_{B}^{SSM}$ and $\delta=\Phi_{eB}/\Phi_{B}^{SSM} $, where $\Phi_{B}^{SSM}=5.15 \times10^{6}~\rm{cm}^{-2}~\rm{s}^{-1}$ is the SSM prediction for the $^{8}B$ neutrino flux \cite{bp98}. We define, in analogy with \cite{sk98}, the following $\chi^{2}$: \begin{equation} \chi^{2}=\sum_{i=1}^{16} \left\{\frac{\frac{R_{i}}{SSM_{i}} -\frac{\alpha+\delta\times B_{i}/SSM_{i}} {(1+\delta_{i,exp}\times\beta)(1+\delta_{i,cal}\times\gamma)}} {\sigma_{i}} \right\} ^{2} +\gamma^{2}+\beta^{2}\ . \label{chi2} \end{equation} In the previous relation $R_{i}$ is the number of solar neutrino events observed in the i-th energy-bin; $SSM_{i}$ \footnote{The quantities $SSM_{i}$ and $B_{i}$ have been calculated taking into account the energy resolution of SK \cite{skres}, as described e.g. in \cite{noi}} is the number of events in the same energy bin due to $^{8}B$ neutrinos, for a total flux $\Phi_{B}^{SSM}$; $B_{i}$ is the same number due to {\it eB} neutrinos, again for a total flux $\Phi_{B}^{SSM}$; the quantities $\delta_{i,exp}$, $\delta_{i,cal}$, $\sigma_{i}$, defined as in \cite{sk98}, take into account correlated and uncorrelated theoretical and experimental errors; the free parameters $\beta$ and $\gamma$ are used for constraining the variation of correlated systematic errors. For each value of $\delta$ we determined the parameters $\alpha$, $\beta$ and $\gamma$ so as to determine the minimum of eq. (\ref{chi2}), $\chi^{2}_{m}$, see fig. 2. As expected, a large {\it eB} neutrino flux produces a steep increase in the high energy tail of the Superkamiokande normalized spectrum, see fig.3. The best-fit is obtained when $\Phi_{eB}=1.1\times10^{4} \rm{cm}^{-2}\rm{s}^{-1}$, corresponding to $\chi^{2}_{min}/D.O.F.=15.7/14$. Acceptable fits are anyhow obtained for $\Phi_{eB}$ in the range $(0.3-2)\times 10^{4} \rm{cm}^{-2}\rm{s}^{-1}$, see fig. 2. \section{Theoretical evaluation of the {\it eB} neutrino flux} Boron ($^8$B) is produced in the sun, according to the following reaction $$ ^7 Be + p \rightarrow ^8B + \gamma$$ and it undergoes $\beta^+$ decay \begin{equation} ^8B \rightarrow ^8 Be^* + e^+ +\nu_e \rightarrow 2\alpha + e^+ +\nu_e \label{decay} \end{equation} or electron capture reaction \begin{equation} ^8B + e^-\rightarrow ^8 Be^* + \nu_e \rightarrow 2\alpha + \nu_e ~. \label{EC} \end{equation} The process under consideration is an allowed transition: in fact (see ref.\cite{Led78}) $ J^P (^8B) = J^P (^8Be^*) = 2^+$. In this case, the ratio $R$ between electron capture probability ($\Gamma_{eB}$) and $\beta^+ $ decay probability ($\Gamma_{\beta^+}$) does not depend on the matrix elements of the transition operator between the nuclear states. A simple phase--space calculation, assuming that the electron number density at nuclear site $n_{e}(0)$ can be approximated by the average electron number density $n_{e}$, gives immediately \begin{equation} R= \frac{1}{8 \pi}\left(\frac{hc}{m_e c^{2}}\right)^{3} \times\left(\frac{E_{eB}}{m_{e}c^{2}}\right)^{2}\times f^{-1}\times{n}_{e} ~, \label{ratio} \end{equation} where, for later convenience, we show explicity the dimensionless phase--space factor associated to $\beta^{+}$ decay, $f\simeq[(E_{eB}-m_{e}c^{2})/m_{e}c^{2}]^{5}/30\simeq7.1\times10^{5}$. For $n_{e}\simeq5.4\times10^{25}\rm{cm}^{-3}$ as suggested by SSM, one has $R\simeq4\times10^{-8}$ and consequently \begin{equation} \Phi_{eB}=R\times\Phi_{B}^{SSM}=2\times10^{-1}~\rm{cm}^{-2}~\rm{s}^{-1} ~, \end{equation} i.e. five orders of magnitude lower than that required to account for SK data. It is anyhow useful to estimate $\Phi_{eB}$ with a better accuracy. With respect to the naive estimate given previously, one should consider the effects of interactions with the solar plasma. The distortion of the positron wave function in the $\beta^{+}$ decay rate can be described as a modification of the dimensionless phase--space factor $f$, which is now given by $f=5.70\times 10^{5}$ \cite{ec0,ec1}. Moreover the electron density at nucleus $n_{e}(0)$ is larger than $n_{e}$ and, consequently, the ratio $R$ has to be enhanced, with respect to eq. (\ref{ratio}), by a factor \begin{equation} \omega=\frac{n_{e}(0)}{n_{e}} \label{omega} \end{equation} For a precise estimate of $\omega$ one has to take into account: {\it i)} distortion of electron wave functions in the Coulomb field of nucleus\cite{bah62}, {\it ii)} electron capture from bound states \cite{iks}, {\it iii)} screening effects \cite{iks,bm}. Let us discuss the problem in some detail, following the lines of Gruzinov and Bahcall who recently produced a clear and comprehensive analysis of the $^{7}Be$ electron capture in the sun \cite{gb}: \begin{enumerate} \item[\it i)] Because of the Couloumb field of the nucleus, the wave functions of continuum electron states differ from plane waves. The rate of electron capture from continuum has then to be corrected by an enhancement factor $\omega_{c}$ \cite{bah62}: \begin{equation} \omega_{c} = <\left|\frac{\psi_{coul}(0)}{\psi_{free}(0)}\right|^{2}>= \left(\frac{m_{e}c^2}{kT}\right)^{\frac{1}{2}}\times \left(Z\alpha \right)\times 2 \times \left(2\pi\right)^{\frac{1}{2}} \times I(\beta) ~, \end{equation} where the average is taken over electron thermal distribution. In the previous relation $T$ is the Sun temperature, while $I(\beta)$ is a correction factor of order unity, defined e.g. in \cite{ec2}. For $R/R_{\odot}\simeq0.05$, which corresponds to the solar region where the production of $^{8}B$ neutrinos is maximal, the density enhancement at nucleus due to electron in continuum states is $\omega_{c}=3.82$. \item[\it ii)] As pointed out by Iben, Kalata e Schwartz \cite{iks}, under solar conditions bound electrons give a substantial contribution to the electron density at the nucleus. The bound state enhancement factor is given by \cite{gb}: \begin{equation} \omega_{b} = \pi^{\frac{1}{2}}\times\hbar^{3}\times \left(\frac{m_{e}kT}{2}\right)^{-\frac{3}{2}} \sum_{n} \left(\frac{Z}{a_{0}n}\right)^{3} \exp\left(Z^{2}e^{2}/2n^{2}a_{0}kT\right) ~, \end{equation} where $a_{0}$ is the Bohr radius. For $R/R_{\odot}\simeq0.05$, the bound state enhancement factor is $\omega_{b}=2.94$. The total density enhancement factor is then $\omega_{c}+\omega_{b}=6.76$ \item[\it iii)] Screening effects reduce the electron density at nucleus for both bound \cite{iks} and continuum electron states \cite{bm}. If the temperature is sufficiently high and if the screened potential can by described by \begin{equation} V(r)=-\frac{Ze^{2}}{r}\exp(-r/R_{D}) ~, \label{salpeter} \end{equation} where $R_{D}$ is the Debye radius, by using a thermodynamical argument one finds \cite{gb}: \begin{equation} \omega=\exp(-Ze^{2}/kTR_{D})\times(\omega_{b}+\omega_{c}) ~. \label{screen} \end{equation} For $R/R_{\odot}\simeq0.05$ the total density enhancement factor , due to screening effects, is reduced to $\omega=5.34$. The small difference between this value of $\omega$ and that given by \cite{gb} is due to the fact that they have been calculated for slightly different solar regions. Relation (\ref{screen}) is not so straightforward, especially because of the possible inadequacies of the Debye screening theory \cite{joh} and because of the relatively large thermal fluctuations which could results from the small number of ions in a Debye sphere \cite{sal}. For the similar case of $^{7}Be$ electron capture, Gruzinov \& Bahcall have performed a detailed analisys of the problem, concluding that relation (\ref{omega}) is accurate at the 2\% level. \end{enumerate} By using the previous relations we can determine the ratio between electron capture and $\beta^{+}$ decay rates. We obtain: \begin{equation} R=2.6\times 10^{-7}\times(1\pm0.02) \end{equation} This value is about 30\% larger than previous estimates \cite{ec1} which took into account only continuum electron states contribution. By using the SSM estimate of the $^{8}B$ neutrino flux, which is uncertain by about 17\% \cite{bp98}, one concludes \begin{equation} \Phi_{eB}=R\times\Phi_{B}^{SSM}=1.3\times(1\pm0.17)~\rm{cm}^{-2}\rm{s}^{-1} ~. \end{equation} The predicted neutrino flux is lower by a factor $10^{4}$ than required to account for SK data and the calculation method is robust. We conclude that {\it eB} neutrinos cannot explain the spectral distributions of solar neutrino events reported by SK. The author thanks M.R. Quaglia, G. Fiorentini, P. Pizzochero and P. Bortignon for useful discussions and for earlier collaboration on the subject of this paper.
\section{Introduction} Exact Boson or Fermion solutions of the quantum N-body problem in which every particle interacts with every other in three dimensions are very rare. They are almost as rare in classical mechanics although Newton solved one in Principia (1687) (see Cajori 1934 \& Chandrasekhar 1995) and there are also some very special solutions such as Laplace's in which the three unequal masses describe ellipses about their centre of mass while at each time they make an {\em equilateral} triangle. However Newton's solution was for all initial conditions when the force on body $I$ due to body $J$ was of the form $F_{IJ} = k m_I m_J ({\bf x}_J - {\bf x}_I)$. Newton reduced this problem to that of $N$ harmonic oscillators relative to the centre of mass. The quantum solution is similar to the $N$ oscillator solution for solid state physics. The potential energy of Newton's system is $$V = {\textstyle {1 \over 2}} \! \sum_{\ \ I \ <} \! \sum_{\! \! \! J} k m_I m_J ({\bf x} _I - {\bf x} _J)^2 = {\textstyle {1 \over 2}} kM \Sigma m_I ({\bf x}_I - {\overline {\bf x}})^2 \equiv {\textstyle {1 \over 2}} kM^2r^2\ . $$ Here we show that this solution may be generalised to systems in which the total potential energy $V$ is any function of $r$. We have already explored these systems and their generalisations in Classical Mechanics (Lynden-Bell \& Lynden-Bell 1999). Except for Newton's harmonic case all these systems give many-body forces in which the force between any two bodies is approximately linear for separations much less than the mean but with a coefficient that depends on that current mean. Despite the strange global nature of these force laws they may be the only non-trivial quantum many-body problems that have been solved exactly in 3 dimensions. Only in Newton's harmonic case do the forces reduce to simple pair-wise interactions. It could be argued that such global forces are unnatural, however, in some respects the resulting behaviour mimics that found in Nature. Ruth Lynden-Bell (1995, 1996) showed that these systems can be used to give a simple model of a phase transition that can be calculated even when $N$ is small. Also as we now show, such forces can be used to mimic some aspects of gravitation. For a homogeneous sphere that may pulsate in radius, $a$, the gravitational potential energy is $V = -{3 \over 5} GM^2/a$ and the gravitational force on unit mass not outside $a$ is $-{4 \over 3} \pi G \rho {\bf x}$. The mean square radius of the sphere is $r^2 = 3a^2/5$ so $V = - \left( {3 /5} \right)^{3/2} GM^2/r$. Now forget real gravity but adopt this form of $V (r)$ for the potential of one of our extraordinary N-body problems. The force per unit mass on any particle at ${\bf x}$ is given by $$-M^{-1} \partial V/\partial {\bf x} = - \left({3\over 5}\right)^{3/2} GM {\bf x}/r^3 = - GM {\bf x}/a^3 = - {4 \over 3} \pi G \rho {\bf x}$$ where ${\bf x}$ stands for any one of the ${\bf x}_I - {\overline {\bf x}}$. Thus for homogeneous spheres this choice of $V (r)$ in our extraordinary N-body problem exactly mimics the effect of true gravity both for global radial pulsations of the system and for the forces on masses within it. However if the system departs from homogeneity this mimicry is no longer exact. For inhomogeneous spherical systems the true gravitational potential energy $V (r)$ can always be written $-k GM^2/r$ with a $k$ that depends on the radial profile. By taking that to be the $V (r)$ in our extraordinary N-body problem its Virial theorem will perfectly mimic that of the gravitational problem but apart from the homogeneous case the forces on the individual particles of which the system is composed will not be the same in the mimic. Outside gravitational theory the concept of an effective potential is widely used in physics and chemistry, e.g., in the shell model of the nucleus, in quark-quark interactions at low energy, and in modelling inter-molecular forces. Now we have shown that motion in these special potentials can be exactly calculated, they will no doubt be used as approximations in those applications, as well as many others. Since the N-body wave-function is known exactly, so is the correlation energy, but that may not be a useful general guide to correlation because in our systems the net force on each particle is directed radially toward the centre-of-mass whatever the configuration of the other particles may be. Furthermore in many real systems the interaction between any two particles is strongest when the particles are closest together while it is weakest for the systems discussed here. In spite of this it is possible to make systems that are strongly repulsive when all particles try to come close together and ones that behave like gravitating systems in the sense that the overall radius obeys the Virial Theorem for a self-gravitating system. Even without any repulsion the exclusion principle provides support for systems of Fermions so with a $V \propto -GM^2/r$ appropriate for gravity, we find configurations of White Dwarf type. The N-body problems discussed here arose by direct generalisation of Newton's work and so skipped the developments of the intervening centuries. We may nevertheless see how they fit into those developments. Liouville (1855) showed that if a system of $D$ degrees of freedom had $D$ integrals of the motion whose mutual Poisson Brackets vanished, then the remaining $D$ integrals of the motion could be found as quadratures. He also discovered a large class of such separable systems while St\"ackel (1890) proved his necessary and sufficient conditions for separability. Whittaker (1904) gives a good description of those works while he, Eddington (1915) and Eisenhart (1934) helped to determine and classify such systems; De Zeeuw (1985) gives a good historical introduction in his thesis paper. Lynden-Bell (1962) and Hall (1985) developed different ideas for finding classes of systems with integrals or configuration invariants. Carter (1968) extended such results to the motion of charged particles in magnetic fields in General Relativity. Marshall and Wojciechowski (1988) determined those potentials in $D$ dimensions for which the motion of a classical particle separates in suitable coordinates and the hyper-spherical potential of the systems discussed here can be viewed as a highly degenerate member of their general $D$-dimensional ellipsoidally separable potentials. Evans (1990, 1991) has explored systems that are superintegrable, having more than $D$ integrals for $D$ degrees of freedom. They separate in several different coordinate systems and the integrals are the separation constants. This is the case for our hyperspherical systems (see Appendix). In all these works separability was achieved by changing the coordinates only. The idea that separability might be achievable only via canonical transformations involving the momenta as well as the coordinates was not exploited. Thus Kovalevski's top (1888) provided an unexpected new system in which the separation was not of the standard type. Linear soluble systems were known which were not of the simple separable type and Routh's (1877) Adam's prize essay on the Stability of Motion gives a very thorough discussion. Simple examples of both linear (Freeman 1966) and non-linear problems (Vandervoort 1979, Contopoulos \& Vandervoort 1992) that need momentum dependent transformations arose in stellar dynamics but it was only recently, e.g., in the work of Sklyanin (1995) that more general ways of looking for such systems were found. Meanwhile a whole body of work based around Lax (1968) operator pairs and the Inverse Scattering method showed that there were many previously unsuspected exact solutions in both classical and quantum mechanics. This field of endeavour is too large to be reviewed here so the reader is referred to the review volumes Solitons (Bullough \& Caudrey eds., 1980), Dynamical Systems VI Integrable Systems (Arnold ed., 1995) and Soliton Theory: a survey of results (Fordy ed., 1990). The connections between soluble models of N-body problems and field theory are discussed in the book edited by Bazhanov \& Burden (1995). In that volume quantum and classical integrable lattice models in one dimension are considered by Bullough \& Timonen (1995) while two dimensional models in statistical mechanics are discussed by Baxter (1995). Prominent among many exactly soluble N-body models in one dimension are the Toda lattice (Toda 1967, 1980) also discussed by Henon (1974) and the Calogero model and its generalisations. [Calogero (1971), Sutherland (1971), Bullough \& Caudrey (1980), Olshanetsky \& Perelomov (1995).] In two dimensions Baxter's book (1982) and article (1995) contain much of interest and certain solvable models in 3 dimensions have been proposed by Baxter (1986) and Bazhanov \& Baxter (1992, 1993). Probably the most prominent soluble field theory in 2 dimensions is that of Davey-Stewartson (1974), see Anker \& Freeman (1978), the quantum version of which is considered in Pang et al. (1990). Sklyanin (1995) holds out the hope that all these soluble systems may eventually be seen as special cases of the method of Separation of Variables and produces some supporting evidence. Although the above models of interacting systems of many Fermions or Bosons can be solved exactly in one or two dimensions, the calculation that follows may be the only exactly solved non-trivial three dimensional N-body system yet known. Furthermore suitable choices of the function $V(r)$ will allow a study of the way the form of interaction (albeit one of our strange global type) affects Bose-Einstein condensation. It should also prove possible by these methods to study the effect of rotation on the condensation. However this paper is solely concerned with the solutions of Schr\"odinger's equation with the correct symmetry in the wave function, so the statistical mechanics and Bose-Einstein condensation displayed by these models is not further discussed here. With the somewhat more realistic $\delta$ function interaction it has been studied previously for one dimensional chains and their continuum limits, see, e.g., Bogoliubov et al. (1994) and Bullough \& Timonen (1998). It is of course the case that all soluble models are exceptional and a good example of the intricacies of non-soluble models was furnished by Henon (1969). \section{N-body Solutions of Schr\"odinger's Equation} Let $m_I$ be the mass of the $I^{\rm th}$ particle and ${\bf x}_I$ its position vector. Writing $M = \Sigma m_I$ for the total mass, the centre of mass is given by $${\overline {\bf x}} = \Sigma \mu_I {\bf x}_I \ , \eqno (2.1)$$ where $\mu_I = m_I/M$, which implies $$\Sigma \mu_I = 1 \ . \eqno (2.2)$$ We define `mass weighted' coordinates relative to the centre of mass \break ${\bf r}_I = \mu^{1/2}_I \left( {\bf x}_I - {\overline {\bf x}} \right)$ and an associated 3$N$ dimensional vector, ${\bf r}$, in the space of all the ${\bf r}_I$ by $${\bf r} = \left( {\bf r}_1 , {\bf r}_2, {\bf r}_3 \ldots {\bf r}_N \right) \ . $$ The length of ${\bf r}$ is the mass weighted r.m.s. radius of the system since $$r^2 = \Sigma \mu_I \left( {\bf x}_I - {\overline {\bf x}} \right)^2 \ . \eqno (2.3)$$ \vfill \eject \noindent This expression may be rewritten in terms of the mutual separations of the particles since $$\displaylines{r^2 = \sum_I \mu_I \left( {\bf x}_I - {\overline {\bf x}} \right) \cdot {\bf x}_I = \sum_I \sum_J \mu _I \mu _J \left( {\bf x}_I - {\bf x}_J \right) \cdot {\bf x}_I = \cr \hfill{} = \sum_J \sum_I \mu_J \mu_I \left ( {\bf x}_J - {\bf x}_I\right) \cdot {\bf x}_J \ ,\hfill{}\cr} $$ and by adding the last two expressions and halving the result $$r^2 = {\textstyle {1 \over 2}} \sum_I \sum_J \mu_I \mu_J ({\bf x}_I - {\bf x}_J)^2 = \sum_{\ \ I \ <} \! \sum_{\! \! \! J} \mu_I \mu_J ({\bf x}_I - {\bf x}_J)^2 \ . \eqno (2.4)$$ In practice the ${\bf r}$ vector is constrained by the fact that the centre of mass is at the origin so $$\Sigma \mu^{1/2}_I {\bf r}_I = \Sigma \mu_I ({\bf x}_I - {\overline {\bf x}}) = 0 \ . \eqno (2.5)$$ We define three mutually orthogonal unit vectors ${\hat {\bf X}}$, ${\hat {\bf Y}}$ and ${\hat {\bf Z}}$ in our 3$N$ space by $$ {\hat{\bf X}} = \left(\mu^{1/2}_1,\, 0,\, 0,\, \mu^{1/2}_2 ,\, 0 ,\, 0 \ldots \mu^{1/2}_N, \, 0 , \, 0\right)$$ $$ {\hat{\bf Y}} = \left(0,\, \mu^{1/2}_1,\, 0,\, 0,\, \mu^{1/2}_2,\, 0 \ldots 0,\, \mu^{1/2}_N,\, 0\right) \eqno (2.6)$$ $$ \ \ \ {\hat {\bf Z}} = \left(0,\, 0,\, \mu^{1/2}_1,\, 0,\, 0,\, \mu^{1/2}_2 \ldots 0,\, 0,\, \mu^{1/2}_N \right)\ . $$ Then the constraints (2.5) can be rewritten $${\hat {\bf X}} \cdot {\bf r} = {\hat {\bf Y}} \cdot {\bf r} = {\hat {\bf Z}} \cdot {\bf r} = 0 \eqno (2.7)$$ which show that ${\bf r}$ is confined to three hyperplanes through the origin. Defining ${\hat {\bf r}} = {\bf r}/r$ then ${\hat {\bf r}}$ lies on the unit 3$N$ sphere $|{\hat {\bf r}}|^2 = 1$ but the ${\hat {\bf X}}$ constraint confines it to the intersection of that sphere with the hyperplane ${\hat {\bf X}}\cdot {\bf r} = 0$, which is a sphere in 3$N$-1 space; similarly the ${\hat {\bf Y}}$ constraint leaves it on the intersection of that 3$N$-1 sphere with the hyperplane ${\hat{\bf Y}} \cdot {\bf r} = 0$, which is a 3$N$-2 sphere and the third constraint leaves it on the 3$N$-3 sphere orthogonal to ${\hat {\bf X}}$, ${\hat{\bf Y}}$ and ${\hat {\bf Z}}$. We are concerned with the N-body problems whose potential energies, $V$, are functions of the magnitude $r$ only, so Schr\"odinger's equation takes the form $$-{\textstyle{1 \over 2}} \hbar^2 \sum_I m^{-1}_I {\partial^2 \psi \over \partial {\bf x}_I \cdot \partial {\bf x}_I} + V \psi = E_T \psi \ . \eqno (2.8)$$ The key to solving this problem lies in the right choice of coordinates. In what follows upper case indices run over particle labels while lower case indices run over coordinate-vector components. \subsection{Separation of ${\overline x}$} Let $R_{ij}$ be an orthogonal unit $3N \times 3N$ rotation matrix which rotates the basis vectors of our $3N$ space so that ${\hat{\bf X}}$, ${\hat{\bf Y}}$ and ${\hat{\bf Z}}$ are the last three of the new orthogonal basis vectors. Thus with two alternative notations and assuming the summation convention over lower case indices only, $$q_i = R_{ij} r_j = \sum_J {\bf R} _{iJ} \cdot {\bf r}_J \ , \eqno (2.9)$$ with $$q_{3N-2} = X \ , \ q_{3N-1} = Y \ , \ q_{3N} = Z \ . \eqno (2.10)$$ Note $$R_{ij}R_{kj} = \delta _{ik} \ {\rm and} \ \sum_J {\bf R}_{iJ} \cdot {\bf R}_{kJ} = \delta_{ik} \ . \eqno (2.11)$$ Let $a$ run from 1 to 3$N$-3 (rather than from 1 to $3N$). Then the $q_a$ together with the coordinates ${\overline {\bf x}}$ form a complete set of independent orthogonal coordinates for our system. We shall need the partial derivatives from (2.1) and below (2.2) $$\partial {\overline{\bf x}}/\partial {\bf x}_I = \mu_I {\underline{\mbox{\boldmath$ \delta$}}} \ , \eqno (2.12)$$ $$\partial {\bf r}_J/\partial {\bf x}_I = \sqrt{\mu_J} \left ( \delta_{JI} - \mu_I \right) {\underline{\mbox{\boldmath$ \delta$}}} \ , \eqno (2.13)$$ where $ {\underline{\mbox{\boldmath$ \delta$}}}$ is the unit $3 \times 3$ matrix. The centre of mass motion will separate so our wave functions may be taken in the form $\psi = {\overline \psi} ({\overline {\bf x}}) \widetilde \psi (q _a \ldots )$ so $${\partial \psi \over \partial {\bf x}_I} = {\partial {\overline \psi} \over \partial {\overline {\bf x}}}\, \mu_I \widetilde \psi + {\overline \psi} \, {\partial \widetilde \psi \over \partial {\bf x}_I} \ . \eqno (2.14)$$ For Schr\"odinger's equation we shall need $$\sum_I \mu^{-1}_I\, {\partial ^2 \psi \over \partial {\bf x}_I \cdot \partial {\bf x}_I} = {\partial^2 \overline \psi \over \partial {\overline{\bf x}} \cdot \partial {\overline {\bf x}}} \, \widetilde \psi + \overline \psi \Sigma \mu^{-1}_I {\partial^2 \widetilde \psi \over \partial {\bf x}_I \cdot {\bf x}_I} \ , \eqno (2.15)$$ where the cross derivative term has vanished because $\widetilde \psi$ only involves differences of the coordinates $x_I$ so $\Sigma \partial \widetilde \psi/\partial {\bf x}_I = 0$, i.e., $\widetilde \psi$ is independent of where the system-as-a-whole is. To evaluate the second term we need from (2.9) and (2.13) $${\partial \widetilde \psi\over \partial {\bf x}_I} = \sum_K {\partial \widetilde \psi \over \partial q_j} \, {\partial q_j \over \partial {\bf r}_K} \cdot {\partial {\bf r}_K \over \partial {\bf x}_I} = {\partial \widetilde \psi \over \partial q_j} \left( {\bf R}_{jI} \sqrt{\mu_I} - \sum_K \sqrt{\mu_K} \mu_I {\bf R}_{jK} \right) \ . \eqno (2.16)$$ We check that indeed $\Sigma \partial \widetilde \psi / \partial {\bf x}_I = 0$ by summing this over $I$ and noting that the two sums cancel because $\Sigma \mu_I = 1$. We now proceed to the last term in Schr\"odinger's equation (2.15) $$\sum_I {1 \over \mu_I}\, {\partial^2 \widetilde \psi \over \partial {\bf x}_I \cdot \partial {\bf x}_I} = {\partial^2 \widetilde \psi \over \partial q_\ell \partial q_j }\, \sum_I {\bf R}_{jI} \cdot {\bf R}_{\ell I} - \sum_K \sum_I {\partial^2 \widetilde \psi \over \partial q_j \cdot \partial {\bf x}_I} \sqrt{\mu_K} {\bf R}_{jK} \ .$$ But the last term involves $${\partial \over \partial q_j} \left ( \sum_I {\partial \widetilde \psi \over \partial {\bf x}_I} \right)$$ which is zero and $$\sum_I {\bf R}_{jI} \cdot {\bf R}_{\ell I} = \delta_{j\ell}$$ because $\bf R$ is an orthogonal matrix. Hence we have the desired expression $$\sum_I \mu^{-1}_I {\partial^2 \psi \over \partial {\bf x}_I \cdot \partial {\bf x}_I} = {\partial^2 {\overline \psi} \over \partial {\overline {\bf x}} \cdot \partial {\overline {\bf x}}} {\widetilde \psi} + {\overline \psi} {\partial^2 {\widetilde \psi} \over \partial q_a \partial q_a} \ . \eqno (2.17)$$ $a$ has replaced $j$ in the final term because $\widetilde \psi$ is only dependent on the first 3$N$-3 of the $q_j$, and we remember that $q_aq_a = r^2$ since $X$, $Y$ and $Z$ are all zero. On division by $\psi$ Schr\"odinger's equation now takes the form $$-{\textstyle {1 \over 2}} \hbar^2 M^{-1} \left( {1 \over {\overline \psi}} \, {\partial^2 {\overline \psi} \over \partial {\overline {\bf x}} \cdot \partial {\overline {\bf x}}} + {1 \over {\widetilde \psi}} \, {\partial^2 {\widetilde \psi} \over \partial {\bf q} \cdot \partial {\bf q}} \right) + V(r) = E_T \ , \eqno (2.18)$$ where ${\bf q}$ stands for the $3(N-1)$ vector $q_a$. The equation clearly separates with the final three terms dependent on the $q_a$ only and the first dependent on ${\overline {\bf x}}$ only, so it must be constant. Without loss of generality we can take the total momentum to be $\hbar {\bf K}$. Then ${\overline \psi} = \exp (i{\bf K} \cdot {\overline {\bf x}})$ and writing $E = E_T - {1 \over 2} \hbar^2 K^2/M$ we find $$- {\hbar^2 \over 2M} \, {\partial^2 {\widetilde \psi} \over \partial {\bf q} \cdot \partial {\bf q}} + V(r) {\widetilde \psi} = E {\widetilde \psi} \eqno (2.19)$$ where ${\bf q} \cdot {\bf q} = r^2$. \subsection{Separation of Angular Coordinates} Equation (2.19) clearly separates again in hyperspherical polar coordinates but it is simplest to write them symbolically by putting ${\bf q} = r {\hat {\bf r}}$ and regarding $r$ as independent of the angular coordinates ${\hat {\bf r}}$. We need the partial differentials $$\partial r/\partial {\bf q} = {\hat {\bf r}} \eqno (2.20)$$ $$\partial {\hat {\bf r}}/ \partial {\bf q} = \partial /\partial {\bf q} \left( {\bf q}/r\right) = r^{-1} ({\underline {\mbox {\boldmath $\delta$}}} - {\hat {\bf r}} {\hat {\bf r}} ) \eqno (2.21)$$ and by writing ${\hat {\bf r}} = {\bf q}/r$ and using (2.20) $$ \partial / \partial {\bf q} \cdot {\hat {\bf r}} = r^{-1} (3N-4) \ . \eqno (2.22)$$ We write ${\widetilde \psi} = \psi_r (r) {\hat {\psi}} ({\hat {\bf r}})$ and notice that ${\hat \psi}$ is constant on radial lines so that ${\hat {\bf r}} \cdot \partial {\hat \psi}/\partial {\hat {\bf r}} = 0$. Then $${\partial {\widetilde \psi} \over \partial {\bf q}} = {\partial \psi_r \over \partial r} \, {\hat {\bf r}} {\hat \psi} + r^{-1} \psi_r {\partial {\hat \psi} \over \partial {\hat {\bf r}}} \eqno (2.23)$$ and $${\partial^2 {\widetilde \psi} \over \partial {\bf q} \cdot \partial {\bf q}} = {\partial ^2 \psi_r \over \partial r^2} \, {\hat \psi} + (3N-4) r^{-1} {\partial \psi_r \over \partial r} \, {\hat \psi} + r^{-2} \psi_r {\partial^2 {\hat \psi} \over \partial {\hat {\bf r}} \cdot \partial {\hat {\bf r}}} \eqno (2.24)$$ dividing by $\psi$ Schr\"odinger's equation now takes the form $$\psi^{-1}_r \left[ r^2 \partial ^2 \psi_r / \partial r^2 + (3N-4)r \partial \psi_r/\partial r \right] - \alpha^2 r^2 + U(r) r^2 = {\hat {\psi}}^{-1} A ({\hat {\psi}}) \ . \eqno (2.25)$$ $\alpha$ is given by $\alpha^2 = -2ME \hbar^{-2}$ and $U(r) = -2MV(r) \hbar^{-2}$. The angular operator $A$ (the hyper-angular-momentum operator) is given by $$A ({\hat \psi}) = - \partial^2 {\hat \psi}/\partial {\hat {\bf r}} \cdot \partial {\hat {\bf r}} \ . \eqno (2.26)$$ The angular operator $A$ also appears in the generalised $\nabla ^2$ in 3$N$-3 dimensions viz ${\partial^2 / \partial {\bf q} \cdot \partial {\bf q}}$, so we shall first study the hyper-spherically symmetric solutions of $\partial^2 \chi / \partial {\bf q} \cdot \partial {\bf q} = 0$. Evidently $${d^2 \chi \over dr^2} + {(3N-4) \over r} \, {d \chi \over d r} = 0$$ so $d \chi/dr = Cr^{-(3N-4)}$ therefore $\chi = B r^{-(3N-5)}$ where we have omitted a constant of integration which is irrelevant to our purpose. We see this is correct by considering the first non-trivial case in which terms other than ${\overline {\bf x}}$ are involved which is $N=2$. Then the space of the $q_a$ is 3 dimensional and the elementary solution has $\chi = Br^{-1}$. Now our generalised $\nabla^2$ knows no particular origin so if $\chi$ is a solution for $r \neq0$ then so is ${\widetilde \chi} = B |{\bf r} - {\bf r}_0 | ^{-(3N-5)}$ for ${\bf r} \neq {\bf r}_0$. We expand such solutions both for $|{\bf r}| < |{\bf r}_0|$ and for $|{\bf r}| > |{\bf r}_0|$ in powers and the coefficients of these powers are $Y_L ({\hat {\bf r}})$ hyperspherical harmonics, just as they are in 3 dimensions. In particular the $L^{\rm th}$ harmonics have a power in $r$ of either $r^L$ or $r^{-L-(3N-5)}$. Looking for solutions of the form $r^Lf({\hat {\bf r}})$ to our generalised $\nabla^2 =0$ we see that in $D=3N-3$ dimensions $$r^2 {d^2 \over dr^2} \left(r^Lf\right) + (D-1) r {d \over dr} \left( r^Lf \right) = r^L A(f) \ ,$$ i.e., $$L(L + D-2)\, f = A(f) \eqno (2.27)$$ and hence the eigenvalues of $A(f)$ are $L(L+3N-5)$. Notice that for a two body problem this reduces to the $L (L+1)$ in 3 dimensions that we know so well. We shall return later to look for the degeneracies of these different eigenstates but for solving Schr\"odinger's equation the eigenvalues are sufficient. For the detailed separation of the $3N-4$ angular coordinates each in turn see the Appendix. \subsection{The Radial Equation} Schr\"odinger's equation (2.25) now reads $$d^2 \psi_r/dr^2 + (3N-4)r^{-1} d\psi_r/dr - [\alpha^2 - U(r) +L (L+3N-5)r^{-2}] \psi_r = 0 \ . \eqno (2.28)$$ Now in the corresponding classical N-body problem we showed that the solution for the radial pulsations of the whole N-body system could be found in terms of the radial pulsation of the corresponding two body problem with the same $U(r)$. With $N=2$ we have the usual Schr\"odinger equation for a spherical potential with angular momentum $\ell$ $$d^2 \psi_2/dr^2 + 2r^{-1}d\psi_2/dr - (\alpha^2 - U(r) + \ell (\ell +1) r^{-2}) \psi_2 = 0 \ . $$ We shall suppose that this problem has been solved and the corresponding eigenvalues and eigenfunctions are known. Now put $\psi_2 = r^{\beta} \chi_2$; then $\chi_2$ obeys $$\chi_2 '' + (2 \beta +2) r^{-1} \chi'_2 - \left[ \alpha^2 - U(r) + \ell (\ell + 1) - \beta (\beta +1) \right] \chi_2 = 0 \ . \eqno (2.29)$$ Writing $\ell (\ell +1) - \beta (\beta +1) = (\ell - \beta) (\ell + \beta +1)$ one notices that putting $$\beta = {\textstyle {3 \over 2}} (N-2) \eqno (2.30)$$ and $\ell = L+ \beta$, transforms equation (2.29) into precisely (2.28). We deduce the surprising theorem below: {\em Theorem}: If the energy levels of the usual Schr\"odinger equation in 3 dimensions are $E(P, \ell)$, where $P= 1,2,3 \ldots$ is the radial quantum number and $\ell$ is the angular momentum quantum number, then the energy levels of the N-body problem with the `same' `potential' $U(r)$ are $E \left(P, L+ {\textstyle {3 \over 2}} (N-2)\right)$. Furthermore the radial parts of their wave functions are related by $\psi_r (P, L, r) = r^{-{3 \over 2} (N-2)} \psi_2 \left(P, L + {3 \over 2} (N-2), r \right)$. Notice that the ground states are not the same because $L+{3 \over 2}(N-2)$ can not be zero for $N>2$. Thus the ground state of the N-body system corresponds to an $\ell$ of ${3 \over 2} (N-2)$ which can be a high angular momentum state of the two body problem. We see at once that the N-body problem will have proper energy levels even if strongly {\em attractive} $r^{-2}$ potentials are added to $U$ just because of this effective increase in the $\ell$ of the ground state. The theorem above is very powerful in that it enables us to use everything that is known about the solutions to the normal Schr\"odinger equation in spherical potentials and transform it into knowledge of our N-body problems. As is well known, not only the generalised Kepler potential $U(r)\! =\! \zeta r^{-1}\! -\! \zeta_2 r^{-2}$, but also similarly generalised square well potentials, and the $U \propto \delta (r-r_0)$, $\delta$-function potentials as well as oscillator potentials $U(r) = -{1 \over 2} \kappa^2 r^2 - \zeta_2 r^{-2}$ all have pretty solutions for the eigenvalues and eigenfunctions. All this takes over directly. However, when $\zeta_2$ is so negative that the two body problem has no ground state there is perhaps room for doubt as to whether we can use that potential's higher angular momentum states for our N-body problem. To allay any such doubts and provide a pretty illustration of the truth of our general theorem, we now solve the N-body equation for the generalised Kepler potential. \subsection{Generalised Kepler Problem} We have already shown that the wave function takes the form $$\psi = \exp (i {\bf k} \cdot {\overline {\bf x}}) Y_L ({\hat {\bf r}}) \psi_r (r)$$ where $Y_L$ may be given a further 3$N$-5 indices corresponding to various components of $\bf L$. We must now determine $\psi_r$. In solving equation (2.28) we follow the standard method of solution beautifully laid out in the book by Pauling \& Wilson (1935). Setting $\alpha r = \tilde r$ and $\zeta/\alpha = \tilde \zeta$ and keeping only the terms that dominate at large $\tilde r$ we find $$d^2 \psi_r / d \tilde r^2 - \psi_r \sim 0\ .$$ So the asymptotic solutions behave as $\exp \pm \, \tilde r$. Only the minus sign is acceptable so we write $\psi_r = \eta (\tilde r) \exp (- \tilde r)$ and derive the equation for $\eta$ valid for all $\tilde r$, $$\eta '' + (3N\! - 4) \tilde r^{-1} \eta ' - 2 \eta' + \biggl \{ \left [ {\tilde \zeta} - (3N\! \! - \! 4) \right] {\tilde r}^{-1}\! - \! \left[ \zeta _2 +L (L+3N\! \! -\! 5) \right] {\tilde r}^{-2} \biggr\} \eta = 0 \; . \eqno (2.31)$$ We look for power series solutions of the form $$\eta = \sum^\infty_{p=0} a_p \tilde r ^{s+p}$$with $a_0 \neq 0$ and find the recurrence relation $$\displaylines{ \biggl\{ (p+s) (p + s + 3N -5) - \bigl[ \zeta_2 +L (L+3N-5) \bigr] \biggr\} a_p = \cr \hfill = \left[ 2(p+s) + 3(N-2) - \tilde \zeta \right] a_{p-1} \ . \hfill {(2.32)}}$$ The indicial equation has $a_{-1} = p = 0 $ and yields a quadratic equation for $s$ $$s^2 + (3N -5) s - \left[ \zeta_2 + L (L+3N-5) \right] = 0 \ . \eqno (2.33)$$ In the pure Kepler case with $\zeta_2=0$ this yields $s = L$ or $-(L+3N-5)$ of which only the positive $s=L$ solution obeys the boundary condition at the origin. For general $\zeta_2$ the solutions are (cf (2.30)) $$s = -{\textstyle {1 \over 2}} (3N-5) \pm \sqrt{ \left[ L+{\textstyle {1 \over 2}} (3N-5) \right]^2 + \zeta_2} = - {\textstyle {1 \over 2}} - \beta \pm \sqrt{ \left( {\textstyle{1 \over 2}} + L + \beta \right)^2 + \zeta_2} \ , \eqno (2.34)$$ of which only that with the $+$ sign obeys the boundary condition at $r = 0$. When $\zeta_2 < 0$ a more detailed discussion is given later. If the series for $\eta$ does not terminate the asymptotic form of the recurrence relation gives $a_p \simeq 2 a_{p-1}/p$ so $\eta \propto e^{2 \tilde r}$ and $\psi_r$ is divergent at $\infty$; so the series must terminate at $a_{P-1}$ say and in (2.32) $$\tilde \zeta = 2 (P+s) + 3(N-2)=2(P+s+\beta ) \ , \eqno (2.35)$$ with $s$ given by taking the upper sign in (2.34) (i.e., $s = L$ when $\zeta_2 = 0$). Remembering that $\tilde \zeta = \zeta/\alpha$ and that $\alpha^2 = -2 ME \hbar ^{-2}$ expression (2.35) can be recast as the energy spectrum $$E = - {\hbar ^2 \over 8M} \, {\zeta^2 \over (P + s + \beta)^2} = - {\hbar^2 \over 2M} \, {\zeta^2 \over \left[2P-1 + \sqrt{ \left(2L +3N-5 \right)^2 + 4 \zeta_2} \right]^2} \ . \eqno (2.36)$$ In accordance with our theorem the energy levels with general $N$ are given by putting $N=2$ and then replacing $L$ by $L+\beta = L + {3 \over 2} (N-2)$. Of course if $\zeta_2=0$ we have $s=L$ and the theorem is then obvious from the first form. Notice that the theorem really applies to $\alpha^2 = -2ME/\hbar^2$ thus we can only apply it to $E$ itself if we consider a two body problem with the same mass $M$ as the N-body problem; $\zeta^2$ is also taken as unchanged since we require both problems to have the same $U(r)$. However this in no way restricts us to N-body problems with $M$ and $\zeta$ independent of $N$; it merely means that we change correspondingly the masses $M$ and coefficients $\zeta$ in the two body problems with which we compare N-body problems of different $N$. Some may wish to see the precise Schr\"odinger hydrogen atom spectrum with the correct reduced mass emerging when $N=2$; to get this we must evaluate $\zeta$ in terms of $Ze^2$. Our potential energy is $V = -{\textstyle {1 \over 2}} \hbar^2 M^{-1} \zeta/r$ but this $r$ is not the separation of the nucleus from the electron, $R$, but the mass weighted r.m.s. separation of them from the centre of mass. Hence $r = (mm_n/M^2)^{1/2} R$ where $m$ and $m_n$ are the masses of the electron and the nucleus respectively. Setting $V = - {Ze^2 \over R}$ we deduce that $\zeta = 2 (mm_n)^{1/2} \hbar ^{-2} Ze^2$. Inserting this $\zeta$ into (2.36) along with $N=2, \zeta_2 = 0, n = P+L, M = m+m_n$ and putting the reduced mass $mm_n/M = m_r$ the energy levels of hydrogenic atoms are given by $$E = - {m_r \over 2 \hbar^2} \, {(Ze^2)^2 \over n^2} $$ just as they should be. We now return to the question of how negative $\zeta_2$ can be. Since $L$ can be zero the energy of the ground state ceases to be real if $\zeta_2 < - \left( {3N-5 \over 2} \right)^2$ which gives $\zeta_2 < - {1 \over 4}$ for the familiar case $N=2$. Such strongly attractive forces cause the particles to propagate into the nucleus and the ground state ceases to exist. It may be seen that the limiting case has a wave function $\psi \propto r^{- {1 \over 2}}$ near the origin which is easily still square integrable $\int \psi \psi ^\star r^{2} dr < \infty$. This is also true for the limiting case $\zeta_2 = - \left( {3N-5 \over 2} \right) ^2$ for then $\psi \propto r^{-{1 \over 2}(3N-5)}$ and $\int \psi \psi ^\star r^{3N-4} dr < \infty$. The limits are surpassed for the attractions of magnetic monopoles on the magnetic moments of protons and for charged monopoles attracting spinning electrons (Lynden-Bell \& Nouri-Zonoz 1998). \section{Level Degeneracies} In equations (2.26) and (2.27) and the attendant discussion we showed that the solutions of our Schr\"odinger equation consisted of a hyperspherical harmonic in $3N-3$ dimensions times a radial function. Furthermore the hyperspherical harmonics of degree $L$ in $D$ dimensions are the coefficients of $r^L$ in the polynomial solutions of Laplace's equation in $D$ dimensions. Thus the degeneracy of the states of given $L$ and given radial quantum number $P$ will be equal to the number of independent polynomial solutions of Laplace's equation of degree $L$ in $D=3N-3$ dimensions (i.e., harmonic polynomials). To determine this number we first ask how many independent polynomials of degree $L$ exist in $D$ dimensions without the harmonic requirement. Each can be considered as a term of the form $\Pi _i x_i^{l_i}$ where $i$ runs from 1 to $D$ and $\Sigma l_i =L$. That number of polynomials is equal to the number of ways of dividing $L$ objects into $D$ groups where a group is allowed to contain no objects. If we take $L$ units and $D-1$ dividing bars then the number of ways of ordering them is $(L+D -1)!$ and if we disregard the ordering of the $D-1$ bars among themselves and the $L$ units among themselves the number of ways of sorting them into groups is $$G (L,D) = {(L+D-1)! \over (D-1)!L!}\ , \eqno (3.1)$$ so this is the number of independent polynomials of degree $L$. Let $f_L ({\bf x}_a)$ be such a homogeneous polynomial of degree $L$ in the ${\bf x}_a$. In general $$\nabla^2 f_L = \sum_a {\partial \over \partial {\bf x}_a} \cdot {\partial \over \partial {\bf x}_a} \, f_L = f_{L-2}$$ where $f_{L-2}$ is such a polynomial of degree $L-2$ which will have $G(L-2, D)$ independent coefficients. The condition that $f_L$ be harmonic $(\nabla^2f_L = 0)$ thus imposes $G(L-2, D)$ constraints on the $G(L, D)$ free coefficients in $f_L$. Thus the number of independent harmonic polynomials of degree $L$ in $D$ dimensions is $$\displaylines{\ g (L,D) = G(L,D) - G(L-2,D) = \hfill \cr \hfill{}\cr \hfill \quad \ \, = {(L+D-3)! \left [ (L+D-1)(L+D-2)-L(L-1)\right] \over (D-1)!L!} = \hfill (3.2) \cr \hfill{}\cr = {(L+D-3)! \over (D-2)!L!} (2L+D-2) \ . \qquad \qquad \qquad \qquad \qquad \quad \ \ \, }$$ Notice that for the familiar case $D=3$ this gives the correct answer, $2L+1$, for the degeneracy of the states of given $L$. When $\zeta_2=0$ we have the extra degeneracy between the $s, p, d, f$ levels of hydrogen. Then a state of principal quantum number $n$ can be obtained by combining a wave function of given $L$ with a radial wave function with radial quantum number $P = n - L \geq 1$. Thus, to find the degeneracy of states with a given $n$, we need to know the number of harmonic polynomials of degree less than or equal to $n-1$, because the $n-L-1$ extra quanta are taken up by different radial wave functions. To find this number we merely add a dummy group `$o$' to our sorting of $L$ objects into groups and ignore the number of units, $n_o$, that falls into that group. Thus the required answer is $$g_H (n,D) = g (n-1, D+1) = {(n+D-3)! \over (D-1)! (n-1)!} (2n + D-3) \ . \eqno (3.3)$$ For $D=3$ this reduces to $n^2$, which is the familiar degeneracy of the $n^{\rm th}$ hydrogen level before spin is considered. Thus for our problems the degeneracy of levels of hyper-angular-momentum $L$ for a system of $N$ particles is $g(L, 3N-3)$ with $g\ {\rm given}$ by (3.2), while for $\zeta_2=0$ the degeneracy of the $n^{\rm th}$ level is \break $g_H (n, 3N-3) = g (n-1, 3N-2)$. The above degeneracies are worked out assuming that none of the particles are identical. In practice we are more interested in problems with $N$ identical Bosons or $N$ identical Fermions and they require wave functions with even or odd permutational symmetries so we now study that question. \section{Symmetry under permutation of particles} \subsection{Bosons} For Bosons we need wave-functions that are symmetrical for the interchange of any two particle labels ${\bf x}_I, {\bf x}_J$. Both ${\overline {\bf x}}$ and $r$ have the required symmetry when the particles are of equal mass. Just as the magnitude of the angular momentum treats $x, y$ and $z$ symmetrically in 3 dimensions, so the magnitude of the hyperangular momentum $L$ is symmetrical for particle interchange. However the components of the term $L_{ij}$ and the vector ${\hat {\bf r}}$ are not symmetrical under the interchange of particle labels. In \sect2 we found the solutions for our $N$-particle wave functions in the form $$\psi = \exp (i {\bf k} \cdot {\overline {\bf x}}) Y_L ({\hat {\bf r}}) \psi _r (r) \ , \eqno (4.1)$$ \vfill \eject \noindent where $\psi_r$ depends only on the scalar magnitude $L$. The only term that is not automatically symmetrical for particle interchange is $Y_L ({\hat {\bf r}})$ but even that will be automatically symmetrical when $L=0$, because $Y_0$ is constant. Thus the ground state and all $s$-states are automatically symmetrical and are possible states for a system of identical Bosons. The states we have been discussing are {\em not} the one-particle states commonly considered as components of $N$-particle product states (or, for Fermions, Slater determinants); rather our states are themselves $N$-particle states. To get a symmetrical $N$-particle state from one lacking that symmetry we merely add all the wave functions obtained by permuting the labels on the particles. But whereas each of our wave-functions thereby generates one boson $N$-particle state, such a state in general comes from a number of different unsymmetrical wave-functions so we can no longer count the degeneracies by the arguments of \sect3. However the arguments of \sect3 connect the number of $Y_L$ functions with the number of polynomials that are homogeneous and both of degree $L$ and harmonic. If we can count interchange symmetric polynomials independent of ${\overline {\bf x}}$ which are homogeneous of degree $L$ and solutions of Laplace's equation, we have the degeneracy of the quantum states of hyper-angular-momentum $L$. Let $F_L ({\bf x}) = F_L ({\bf x}_1, {\bf x}_2, \ldots {\bf x}_N)$ be a homogeneous polynomial of degree $L$ in ${\bf x}$ which is symmetric under the interchange of any ${\bf x}_I$ with ${\bf x}_J$. Then \break $F_L (\lambda {\bf x}_1, \lambda {\bf x}_2, \ldots \lambda {\bf x}_N) = \lambda^L F_L ({\bf x}_1, {\bf x}_2, \ldots {\bf x}_N) = \lambda^L F_L ({\bf x}_1, {\bf x}_N, \ldots {\bf x}_2), \ {\rm etc.} $ Consider $F_L ({\bf x} - {\overline {\bf x}} ) = F_L ({\bf x}_1 - {\overline {\bf x}}, \ldots {\bf x}_N - {\overline {\bf x}})$. It is also a homogeneous polynomial of degree $L$ in ${\bf x}$ and is also symmetric, but it has the property that it is invariant under the transformation ${\bf x}_I \rightarrow {\bf x}_I + {\mbox {\boldmath $\Delta$}}$ for all $I$ (because then ${\overline {\bf x}} \rightarrow {\overline {\bf x}} + {\mbox {\boldmath $\Delta$}}$). Thus such functions do not depend on the position of the centre of mass. However it can happen that $F_L ({\bf x} - {\overline {\bf x}})$ is identically zero even when $F_L ({\bf x})$ is not. For this to happen $F_L ({\bf x})$ must be of the form $F_L ({\bf x}) = {\overline {\bf x}} \cdot {\bf F}_{L-1} ({\bf x})$ where ${\bf F}_{L-1} ({\bf x})$ is a vector each of whose components is a polynomial of degree $L-1$ in ${\bf x}$ which is symmetric under interchange of ${\bf x}_I$ and ${\bf x}_J$. Now let ${\overline G}(L)$ be the number of independent symmetric polynomials which are homogeneous of degree $L$ in 3$N$ dimensions. Then the number of such polynomials giving rise to non-zero $F_L ({\bf x} - {\overline {\bf x}})$ will be ${\overline G} (L)$ less the number of free coefficients in the $- {\overline {\bf x}} \cdot {\bf F}_{L-1}$ term which we might expect to be $3 {\overline G} (L-1)$. However that is not quite right because a polynomial with a factor ${\overline {x}}\, \overline {y}$ will occur as a possibility in both the $x$ and $y$ components of ${\bf F}_{L-1}$ and in subtracting $3 {\overline G} (L-1)$ we will have subtracted its number of free coefficients not once but twice. The same double counting will have occurred for polynomials with factors ${\overline y} \, {\overline z}$ or ${\overline z} \, {\overline x}$ so we should be subtracting not $3 {\overline G} (L-1)$ but rather $3 {\overline G} (L-1) - 3 {\overline G} (L-2)$. However even that is not quite correct because there may be polynomials with a factor ${\overline x}\, {\overline y}\, {\overline z}$. They will have been subtracted off three times in $3 {\overline G} (L-1)$ but added back in three times in $3 {\overline G} (L-2)$ so they will still be there and they should not be since they clearly vanish when ${\bf x} - {\overline {\bf x}}$ is written for ${\bf x}$. Thus finally the number of independent symmetric polynomials which are homogeneous of degree $L$ and independent of ${\overline {\bf x}}$ is $${\overline G}_1 (L) \equiv {\overline G} (L) - 3 {\overline G} (L-1) + 3 {\overline G} (L-2) - {\overline G} (L-3) \ . \eqno (4.2)$$ However, we still have to impose Laplace's equation $$\sum_I \nabla^2_I F_L \equiv \nabla^2 F_L=0 \ . $$ Now in general $\nabla^2 F_L ({\bf x} - {\overline {\bf x}})$ will be a polynomial of degree $L-2$ in ${\bf x}$. However, since $F_L ({\bf x} - {\overline {\bf x}})$ is invariant to the transformation ${\bf x} \rightarrow {\bf x} + \mbox {\boldmath $\nabla$}$, $\nabla^2 F_L ({\bf x} - {\overline {\bf x}})$ will also have that property. Thus in general we may write $$\nabla^2 F_L ({\bf x} - {\overline {\bf x}}) = F_{L-2} ({\bf x} - {\overline {\bf x}})$$ where $F_{L-2}$ is also symmetrical for particle label interchange since $\nabla^2$ does not destroy that property. Thus the condition $F_{L-2} ({\bf x} - {\overline {\bf x}}) \equiv 0$ will put ${\overline G}_1 (L-2)$ constraints on the ${\overline G}_1 (L)$ free coefficients of the homogeneous $L^{{\rm th}}$ degree polynomial $F_L ({\bf x} - {\overline {\bf x}})$. There will be just ${\overline G}_1 (L) - {\overline G}_1 (L-2)$ free coefficients left in $F_L ({\bf x} - {\overline {\bf x}})$ after imposing the harmonic condition so this is the degeneracy of the $Y_L$ that corresponds to the $(2L+1)$ with $L$ even for the 2 boson problem. Since ${\overline G}_1$ is known in terms of ${\overline G}$, we have reduced our problem to that of determining the number of exchange-symmetric homogeneous polynomials of degree $L$ in 3$N$ dimensions. This is the crux of our problem and it took us considerable thought to solve it. Exchange symmetry involves exchanging vectors ${\bf x}_I$ with ${\bf x}_J$, so we do not need symmetry in all 3$N$ dimensions but only between them taken in triples. We shall begin our considerations with the simpler case of $N$ bosons on a line with each having but one coordinate $x_I$. We then wish to know how many independent exchange-symmetric polynomials there are which are homogeneous of degree $L$ in $N$ dimensions. Let $\Phi = S \Pi_I x^{\ell _I}_I$ be a symmetrical polynomial of degree $L$ with $N$ factors $x^{\ell _I}_I$. $S$ is the symmetrising operator which is the sum over all permutations of the particle labels $I$. Different symmetric polynomials are characterised by different sets of integers $(\ell_1, \ell_2, \ldots \ell_N)$ or partitions of the integer $L$ into $N$ parts, some of which may be zero. We construct the generating function $$B_1 (u,x) = \sum^\infty_{n=1} \sum^\infty_{\ell = 0} p(n, \ell) u^n x^\ell$$ where $p(N,L)$ is the number of partitions of $L$ into $N$ integers that may be zero, and for convenience we have defined $p (0, \ell) = 0$ and $p (n, 0 ) = 1$. We now show how the theory of partitions allows us to construct the function $B_1 (u, x)$. \subsection{Partitions of an integer $L$} We learn from Abramowitz \& Stegun (1964) that `The number of decompositions of an integer $L$ into integer summands without regard to order is called $p (L)$'. For example, five may be written $$5 = 4+1 = 3+2 = 3+1+1 = 2+2+1 = 2+1+1+1 = 1+1+1+1+1 \ , \eqno (4.3)$$ so we deduce that $p(5) =7$. It is easiest to work with the generating function for the $p(L)$ which we call $A(x)$. For this there is a standard result see, e.g., Hardy \& Ramanujan (1918), $$A(x) = 1 + \sum^\infty_1 p(L) x^L = \prod\limits^\infty_{\ell =1} (1 - x^\ell)^{-1} \ . \eqno (4.4)$$ \vfill \eject For what follows it is essential to understand how this standard result comes about. To do so we rewrite the product by expansions in powers of $x$ $$\begin{array}{lccc} A(x) = & \! \! \left( 1 + x^1 + x^2 + \ldots \right) & \! \! \left( 1 + (x^2)^1 + (x^2)^2 + \ldots \right) & \! \! \left( 1 + (x^3)^1 + (x^3)^2 \ldots \right) \cr &({\rm generates \ units}) & ({\rm generates \ twos}) & ({\rm generates \ threes }) \end{array}$$ \vskip -0.3cm \rightline {(4.5)} \vskip -0.6cm $$\hfill \begin{array}{ccc} \left( 1+ (x^4)^1 + \ldots \right) & \left( 1 + (x^5)^1 + \ldots \right) &(\ldots) \cr ({\rm generates \ fours}) & ({\rm generates \ fives})& \qquad \qquad \qquad . \end{array}$$ To see how the coefficient of $x^5$ in this expression is $p(5)=7$ we first notice that we must take the 1 from all brackets after the fifth, since otherwise we would get too high a power of $x$. In the 5$^{\rm th}$ bracket we can take the $x^5$ term but then we must take the 1 from all earlier brackets. Alternatively we take the 1 in the 5$^{\rm th}$ bracket. In the latter case we turn to the fourth bracket. Here we may take the $x^4$ term but that can only be coupled to the $x$ term in the first bracket in which case we get the split of 5 into $4+1$. Turning now to the third bracket and taking the $x^3$ term we can take it with either the $(x^2)^1$ bracket of the second term to yield $3+2$ or with the $(x^1)^2$ term of the first bracket to yield $3+1+1$. Similarly from the second bracket we could take the $(x^2)^2$ term with the $x$ from the first bracket to give $2+2+1$ or the $(x^2)^1$ term with the $x^3$ from the first to give $2+1+1+1$. Finally we could take the $x^5$ from the first bracket to give $1+1+1+1+1$. Thus the first bracket yields the number of ones in the sum, the second the number of twos, the third the number of threes, etc., and in this way the coefficient of $x^L$ yields $p(L)$. However, we need the restricted partition of $L$ into $N$ or fewer non-zero integers $p(N,L)$. These are sums of partitions $p_1(N,L)$ into exactly $N$ non-zero integers. Looking at (4.3) we see, for example, that $p_1(2,5) = 2 = p_1 (3,5)$. If we place a factor $u$ along with each factor $x^{\ell}$ in (4.5), then the power of $u$ in each term will tell us how many parts there are in the partition generated by a particular term. Thus in place of $A(x)$ we consider $$\begin{array}{lcc} A(u, x) = & \left(1 + ux + (ux)^2 + (ux)^3 + \ldots \right) & \left( 1 + (ux^2)^1 + (ux^2)^2 + \ldots \right) \cr & ({\rm generates \ units}) & ({\rm generates \ twos}) \cr &&\hfill (4.6)\cr &\left( 1 + (ux^3)^1 + (ux^3)^2 + \ldots \right) \ \ \ \ldots \cr &({\rm generates \ threes}) \ \ \ \ \end{array}$$ Then the terms in $u^Nx^L$ will have exactly $N$ integers in the corresponding partition of $L$ so (cf. 4.4) $$A(u,x) = 1 + \sum^\infty_{n=1} \sum^\infty_{\ell-1} p_1(n,\ell) u^nx^\ell = \prod\limits^\infty_{\ell =1} (1 - ux^\ell)^{-1} \ . $$ However, we want the number of partitions with $N$ or fewer integers, i.e., \break $p(N,L) = \sum^N_{n=1} p_1 (n,L)$. These sums will be automatically generated if we multiply $A$ by $(1 + u + u^2 + u^3 \ldots)$ before taking the coefficient of $u^N x^L$ so $$B_1 (u,x) = \sum^\infty_{n=1} \sum^\infty_{\ell = 0} p(n,\ell) u^n x^\ell = (1 + u +u^2 + \ldots) A(u,x) = \prod\limits ^\infty _{\ell = 0} (1 - ux^\ell)^{-1} \ . \eqno (4.7)$$ \vfill \eject Readers will recognise the analogy of this expression with the grand partition function for a gas of non-interacting Bosons. So $p(N,L)$ can be found from the product as the coefficient of $u^Nx^L$. Had we been interested in Bosons on a line then $p(N,L)$ would have given us the desired function ${\overline G} (L)$ but our problem is three dimensional. Instead of partitioning $L$ into integers $\ell_I$ we need it partitioned into integer triples $\left( \ell_{Ix}, \ell_{Iy}, \ell_{Iz}\right) $ and the general term in our polynomial will be $x^{\ell_{1x}}_1 y_1^{\ell_{1y}} z_1^{\ell_{1z}} x_2^{\ell_{2x}} \ldots x_N^{\ell_{Nx}} y_N^{\ell_{Ny}} z_N^{\ell_{Nz}}$. When we permute we do so by exchanging $(x_1, y_1, z_1)$ as a triple with say $(x_2, y_2, z_2)$. The degree of our polynomial is $$L= \sum^N_{I=1} \left( \ell_{Ix} + \ell _{Iy} + \ell_{Iz} \right) \ . $$ A triple $(2,1,0)$ corresponding to $\ell_{1x} = 2, \ell_{1y} = 1, \ell_{1z} = 0$ will not be permuted into $(1,0,2)$ by exchanging particle labels so such triples must be regarded as distinct. The number of partitions of 3 into triples, such that the order within a triple matters but the order of the different triples does not, $p_3(L)$ with $L=3$, is already quite a handful. Writing $=$ for an equal total we have $$\left( \begin{array}{c} 3 \\ 0 \\ 0 \end{array} \right) = \left( \begin{array}{c} 0 \\ 3 \\ 0 \end{array} \right) = \left( \begin{array}{c} 0 \\ 0 \\ 3 \end{array} \right) = \left( \begin{array}{c} 2 \\ 1 \\ 0 \end{array} \right) = \left( \begin{array}{c} 0 \\ 2 \\ 1 \end{array} \right) = \left( \begin{array}{c} 1 \\ 0 \\ 2 \end{array} \right) = \left( \begin{array}{c} 1 \\ 2 \\ 0 \end{array} \right) = \left( \begin{array}{c} 0 \\ 1 \\ 2 \end{array} \right) =$$ $$ = \left( \begin{array}{c} 2 \\ 0 \\ 1 \end{array} \right) = \left( \begin{array}{c} 1 \\ 1 \\ 1 \end{array} \right) = \left( \begin{array}{c} 2 \\ 0 \\ 0 \end{array} \right) + \left( \begin{array}{c} 1 \\ 0 \\ 0 \end{array} \right) \ = \begin{array}{ll} & {\rm 8 \ more \ like \ that} \cr &{\rm with} \ x,y,z \ {\rm components} \cr & {\rm permuted.} \end{array} \ = $$ $$ \hskip 3.4cm \ = \left( \begin{array}{c} 1 \\ 1 \\ 0 \end{array} \right) + \left( \begin{array}{c} 1 \\ 0 \\ 0 \end{array} \right) \! = \begin{array}{ll} & {\rm 8 \ more \ like \ that} \cr &{\rm with} \ x,y,z \ {\rm components} \cr & {\rm permuted \ in \ each \ triple.} \end{array} \ = $$ $$ \hskip 1cm = \left( \begin{array}{c} 1 \\ 0 \\ 0 \end{array} \right) + \left( \begin{array}{c} 1 \\ 0 \\ 0 \end{array} \right) + \left( \begin{array}{c} 1 \\ 0 \\ 0 \end{array} \right) = \begin{array}{ll} & {\rm 9 \ more \ like \ that} \cr &{\rm (not \ 26 \ because \ the}\cr & {\rm ordering \ of \ the \ triples} \cr &{\rm does \ not \ count).} \end{array} $$ Hence there are 38 partitions of 3 into triples! Now let $p_3 (L)$ be the number of partitions of $L$ into triples, every triple being counted as different but the different orderings of the same triples being regarded as the same. By analogy with (4.5) we consider the expression \eject $$ A(x,y,z)\equiv $$ $${\setlength{\arraycolsep}{.1em} \begin{array}{lllll} \equiv (1 + x + x^2 ...)& \times & (1 + (x^2)^1 + (x^2)^2 + ...)& \times & (1 + (x^3)^1 + (x^3)^2 + ... )\times ... \cr \times \, (1 + y + y^2 ...) & \times & (1 + (y^2)^1 + (y^2)^2 + ...)& \times & (1 + (y^3)^1 + (y^3)^2 + ... ) \cr \times \, (1 + z+ z^2 ...) & \times & (1 + (z^2)^1 + (z^2)^2 + ...) & \times & (1 + (z^3)^1 + (z^3)^2 + ... ) \cr & \times & (1 + (yz)^1 + (yz)^2 + ...) & \times & (1 + (x^2y)^1 + (x^2y)^2 ...)\cr {\rm This \ column} & \times & (1 + (zx)^1 + (zx)^2 + ...) & \times & (1+ (y^2z)^1 + (y^2z)^2 + ...) \cr {\rm generates} & \times & (1+(xy)^1 +(xy)^2 + ...) & \times & (1+ (z^2x) + ...)\cr {\rm triples} & & {\rm column \ generates} & \times & (1 + (x^2z) + ...)\cr \left( \! \begin{array}{c} 1 \\ 0 \\ 0 \end{array} \! \right) \! {\rm or} \! \left(\! \begin{array}{c} 0 \\ 1 \\ 0 \end{array} \! \right) \! {\rm or} \! \left(\! \begin{array}{c} 0 \\ 0 \\ 1 \end{array}\! \right)&& \hskip -.1cm \left(\! \begin{array}{c} 2 \\ 0 \\ 0 \end{array} \! \right) \! \! \left( \! \begin{array}{c} 0 \\ 2 \\ 0 \end{array} \! \right) \! \! \left( \! \begin{array}{c} 0 \\ 0 \\ 2 \end{array}\! \right) \! \! \left( \! \begin{array}{c} 1 \\ 1 \\ 0 \end{array} \! \right) \! \! \left( \! \begin{array}{c} 0 \\ 1 \\ 1 \end{array}\! \right) \! \! \left( \! \begin{array}{c} 1 \\ 0 \\ 1 \end{array} \! \right)&& \hskip -0.45cm \begin{array}{rl} \times &(1 + (y^2x) + ...)\cr \times & (1 + (z^2y) + ...)\cr \times & (1 + (xyz) + (xyz)^2 + ...) \end{array}\cr {\rm and \ multiple}&& {\rm and \ multiple} && {\rm generates \ triples} \cr {\rm combinations} && {\rm combinations} && {\rm with}\ \ell_x + \ell_y + \ell_z =3 \cr {\rm thereof} && {\rm thereof} && \end{array}} $$ By analogy with the arguments beneath (4.5) one sees how the terms of the third degree generate all the partitions of 3 into triples that we have just enumerated. So the coefficient of $t^L$ in $A(t,t,t)$ will be $p_3(L)$, the number of partitions of $L$ into triples. Furthermore $A(x,y,z)$ may be compactly written in the form $${\setlength{\arraycolsep}{.1em} \begin{array}{lcl} A(x,y,z) = & \prod\limits^\infty_{p=0}\prod\limits^\infty_{q=0} \prod\limits^\infty_{r=0} & (1-x^py^qz^r)^{-1} \ . \cr & ^{p + q + r \neq 0} &\cr \end{array}} $$ \vskip -1.7cm $$\eqno{(4.8)}$$ \vskip .8cm For our $N$ boson problem we are interested not in such partitions into any triples but in partitions constrained to have $N$ or fewer triples as summands. To get exactly $N$ summands we merely insert a $u$ in each term as was done in (4.6). While to get the sum of all terms with $N$ or less summands we have to multiply by $(1+u+u^2 \ldots)$ as in (4.7). Thus putting $x=y=z=t$ the Boson generating function in 3 dimensions is, writing $\ell = p+q+r$, $$B(u,t) = \prod\limits^\infty_{p=0} \prod\limits^\infty _{q=0} \prod\limits^\infty _{r=0} \left(1 - ut^{p+q+r} \right)^{-1} = \prod\limits^\infty_{\ell = 0} \left(1 - ut^\ell \right)^{-{1 \over 2} (\ell +1)(\ell +2)} \, \eqno (4.9)$$ where ${1 \over 2} (\ell +1)(\ell +2)$ is the number of terms in the triple product with $p+q+r = \ell$. The required function ${\overline G} (L)$ is the coefficient of $u^N t^L$ in $B(u,t)$. Again if one only wants ${\overline G} (L)$ for $L<L_{\max}$ then the infinite product can be replaced by a finite product up to $L_{\max}$ without altering the required coefficients. To get ${\overline G}_1(L)$ one merely takes the coefficient of $u^Nt^L$ in $(1 - t)^3B(u,t)$ \break while ${\overline G}_1 (L) - {\overline G}_1 (L-2)$ is the coefficient of $u^Nt^L$ in $(1 -t^2) (1 -t)^3 B(u,t) = (1+t) (1-t)^4 B(u,t)$. In summary the degeneracy $g_B(L)$ of the $N$ Boson state with hyperangular momentum $L$ is the coefficient of $u^N t^L$ in the expression $$(1+t) (1-t)^4 \prod\limits^\infty_{n=0} \left( 1 - ut^{\ell}\right)^{-{1 \over 2} \left(\ell + 1 \right) \left( \ell +2 \right)} \ \ . \eqno (4.10)$$ For two particles $N=2$ we find that the coefficient of $u^2$ is $\left( 1-3t^2 \right) \left( 1 - t^2 \right)^{-2}$. The coefficient of $t^L$ in this expression is $2L+1$ when $L$ is even and zero when $L$ is odd just as it should be. This is just as in the $C^{12} - C^{12}$ homonuclear diatomic molecule of two Bosons with every odd rotational state missing as seen in Carbon star spectra. \subsection{Fermions} The argument of \sect3 relates the number of hyperspherical harmonics of degree $L$ to the number of harmonic polynomials of that degree. Our experience in \sect4$\, a$ leads us to study first the number of antisymmetric polynomials of degree $L$ in $N$ dimensions. If $x^{\ell_1}_1 x^{\ell_2}_2 \ldots x^{\ell_N}_N$ is a term in such a polynomial then $\Sigma \ell_I = L$. Furthermore the antisymmetric polynomial involving that term is the Slater determinant $$\left| \begin{array}{ccccl} x^{\ell_1}_1 & x_2^{\ell_1} & \ldots & x^{\ell_1}_N \cr x^{\ell_2}_1 & x^{\ell_2}_2 & \ldots & x^{\ell_2}_N \cr \vdots &&&\vdots \cr x^{\ell_N}_1 & x^{\ell_N}_2 & \ldots & x^{\ell_N}_N \end{array} \right| $$ Clearly if $\ell_I = \ell_J$ for $I \neq J$ then this determinant vanishes. Furthermore the determinant only changes sign (at most) if the $\ell_I$ are permuted. So for any term that survives we may, without loss of generality, take $\ell_1 < \ell_2 < \ell_3 \ldots < \ell_N$. Thus among the $\ell_I$ only $\ell_1$ can be zero, $\ell_2$ must be at least 1, $\ell_3$ at least 2 and so on with $\ell_N$ at least $N-1$. For a non-zero result $L= \Sigma \ell_I \geq {\textstyle {1 \over 2}} (N-1)N$. By analogy with our study of partitions $p (L)$ for the Boson case we now study partitions of $L$ into distinct parts. Let $q(L)$ be the number of decompositions of $L$ into distinct integer summands without regard to order. Thus $5 = 4+1 = 3+2$ so that $q(5) = 3$. The generating functions for the $q(L)$ is, setting $q(0)=1$, $$\sum^\infty _0 q(L) x^L = \prod\limits^\infty_{\ell = 1} (1 + x^\ell) \ . \eqno (4.11)$$ However not all decompositions of $L$ into distinct parts lead to antisymmetric polynomials in $N$ dimensions. We need a decomposition into either $0 + (N-1)$ unequal integers or into $N$ unequal integers, i.e., we need the coefficient of $u^N x^L$ in $$\prod\limits^\infty_{\ell =0} \left( 1 + ux^\ell \right) = \sum^\infty_{L=0} \sum^\infty _{N=0} q (L,N) u^N x^L \ , \eqno (4.12)$$ which is the expression analogous to (4.7) of the Boson case. Again it is the analogue of the grand partition function for Fermions. The generalisation corresponding to three dimensions follows the argument for (4.8) and yields the generating function $${\setlength{\arraycolsep}{.1em} \begin{array}{lcl} E(x,y,z) = & \prod\limits^\infty_{p=0}\prod\limits^\infty_{q=0} \prod\limits^\infty_{r=0} & (1+x^py^qz^r) \ , \cr & ^{p + q + r \neq 0} &\cr \end{array}} $$ \vskip -1.8cm $$\eqno{(4.13)}$$ \vskip .7cm \noindent which leads analogously to (4.9) to the Fermion generating function $$F(u,t) = \prod\limits^\infty_{p=0} \prod\limits^\infty_{q=0} \prod\limits^\infty_{r=0} \, \left( 1 + u t^{p+q+r} \right) = \prod\limits^\infty _{\ell = 0} \left( 1 + ut^{\ell}\right) ^{{1 \over 2} (\ell +1)(\ell +2)} \ . \eqno (4.14)$$ The coefficient of $u^Nt^L$ in this expression gives the number of homogeneous polynomials of degree $L$ antisymmetric for interchanges of triples in 3$N$ dimensions. For Fermions of spin ${\textstyle {1 \over 2}}$ we do not need complete antisymmetry but only antisymmetry between particles of the same spin state. To allow for the $\alpha$ and $\beta$ spin states being alternatives we generalise $E (x,y,z)$ to $$\prod\limits^\infty_{p=0} \prod\limits^\infty_{q=0} \prod\limits^\infty_{r=0} \, \left( 1 + x^py^qz^r \alpha \right) \left( 1 + x^py^q z^r \beta \right) \eqno (4.15)$$ and in place of $F(u,t)$ we then find $F^2$ $$F^2 = \prod\limits^\infty_{\ell = 0} \left( 1 + u t^\ell \right)^{(\ell + 1)(\ell+2)} \ . \eqno (4.16)$$ The coefficient of $u^Nt^L$ in $F^2$ is the number of Fermionic polynomials for spin ${1 \over 2}$ particles with the correct antisymmetry. The arguments relating the degeneracy of the $N$-Fermion spin ${1 \over 2}$ wave function for the state of hyper-angular momentum $L$ to this expression is the same as for the Boson case. Thus $g_F(L)$ is coefficient of $u^Nt^L$ in the expression $$(1+t) (1-t)^4 \prod\limits^\infty_{\ell = 0} \left( 1 + ut^\ell \right)^{(\ell +1) (\ell +2)} \ . \eqno (4.17)$$ To check this formula we take the coefficient of $u^2$, and find for the 2 Fermion problem the generating function $$\left( 1 + 9t + 3t^2 + 3t^3 \right) \left( 1 - t^2 \right) ^{-2} = \sum_{\ell {\rm even}} (2 \ell + 1 ) t^\ell + 3 \sum_{\ell {\rm odd}} (2 \ell + 1 ) t^{\ell} \ , \eqno (4.18)$$ in which the coefficients of $t^\ell$ will be recognised as the degeneracies of the rotational states of the hydrogen molecule with 2 protons of spin $1 \over 2$. The even $\ell$ values correspond to para-hydrogen and the odd $\ell$ values to ortho-hydrogen. We now turn to the degeneracies in hydrogen-like potentials in which different $L$ states can be degenerate. Here we need the sum of the $g_F(L)$ for all \break $L \leq n-1$ as in equation (3.3). If we multiply our generating function for $g_F(L)$ by $1+t+t^2+t^3+ \ldots = 1/(1-t)$ and then pick the coefficient of $u^Nt^{n-1}$ we will get the required sum so the hydrogenic Fermi degeneracy is that coefficient in $$(1+t) (1-t)^3 \prod\limits^\infty_{\ell = 0} (1 + ut^\ell)^{(\ell + 1)(\ell + 2)}\ . \eqno (4.19)$$ To find the ground state we need the first energy level $n$ for which the coefficient of $u^Nt^{n-1}$ is non-zero. To get $u^N$ and no more with the lowest power of $t$, we need to use the $ut^\ell$ terms rather than the 1 in all the low $\ell$ brackets since others come with higher powers of $t$. Thus if the highest $\ell$ needed is $\ell_m$ we require $$\sum ^{\ell_m}_{\ell = 0} (\ell +1) (\ell +2) = N\ . \eqno (4.20)$$ The sum is $\, {1 \over 3} (\ell _m +1)\, (\ell_m +2)\, (\ell _m + 3)\, $ but $\, N\, $ may not be of precisely this form for integer $\ell_m$, in which case we take the lowest $\ell_m$ that gives \break ${1 \over 3} (\ell _m +1) (\ell_m +2) (\ell _m + 3) \geq N$ so that there will be at least one term in $u^N$. We are interested in the least power of $t$ associated with this term. This will be $$n-1 = \sum^{\ell_m}_{\ell =1} \ell (\ell +1) (\ell +2) = {{\textstyle {1 \over 4}}} \ell _m (\ell _m+1) (\ell_m +2) (\ell _m + 3) \eqno (4.21)$$ whenever $N$ is of the special closed shell form given by the equality. Then \break $\ell_m \simeq (3N)^{1/3} - 2 - (3N)^{-1/3}$ and $n = {3N \over 4} \ell_m + 1 \rightarrow (3N)^{4/3}\big / 4$ as $N \rightarrow \infty$. For large $N$ we then find that the ground state energy behaves as $$E = - {\hbar^2 \over 8M} \, {\zeta^2 \over \left[ {(3N) \over 4} ^{4/3} \right]^2} \ \, . \eqno (4.22) $$ To compare this energy with that of a white dwarf star we must first choose an appropriate value of $\zeta$ so that the potential corresponds to gravity. We showed in an earlier paper (Lynden-Bell \& Lynden-Bell 1999) that at high temperatures our systems have a Gaussian density profile at equilibrium. For such a profile the potential energy may be expressed in terms of the total mass $Nm_H$ and the rms radius at equilibrium $r$ and is $$V = \left( {3 \over 4 \pi}\right)^{1/2} G (Nm_H)^2/r\ . \eqno (4.23) $$ We therefore choose $${\hbar ^2 \over 2 M} \zeta = \left( {3 \over 4 \pi} \right) ^{1/2} G (Nm_H)^2 \ . \eqno (4.24)$$ With this choice our ground state energy level is $$E = - {2 \ G^2m^4_H m_e \over 3^{5/3}\ \pi \ \hbar ^2} N^{7/3} \eqno (4.25)$$ where we have written $M = Nm_e$ for the mass of the degenerate electrons whose wave function we have been evaluating. This expression is of precisely the form found by the standard equation of state of a cold degenerate non-relativistic gas under its own gravity. We have, therefore, established that White Dwarfs may be regarded as `superatoms' -- systems with N-body wave functions which are solutions of Schr\"odinger's equation in a central potential. \section{Conclusions} By treating the N-body problem as a separable system in $3N$ dimensions we have shown that it can be solved in appropriate potentials. Marshall \& Wojciechowski (1988) have given the general form of potentials that allow separability in many dimensions and we have concentrated on the hyper-spherical one. Of the many others possible, a sub-class are symmetrical for exchange of the particles. Whereas we have shown how to classify the wave functions by hyper-angular momentum, those interested in rotating systems will need to develop methods of classification involving the 3 dimensional angular momentum; here the methods of Dragt (1965) and the work by Louck \& Galbraith (1972) may prove useful. It is hoped that study of that problem will throw light on Bose-Einstein condensation of small clusters in rotating systems. The fact that we are only able to treat non-relativistically degenerate white dwarfs suggests that an appropriate generalisation for relativistically moving particles should be sought for systems that do not radiate gravitational waves. In our earlier paper on the classical mechanics of these systems we showed that the fundamental breathing oscillation in $r$ separates for the far-more-general potentials $V= V_0 (r) + r^{-2} V_2 ({\hat {\bf r}})$ where $V_2$ is an arbitrary function of all the coordinates which is independent of scale, $\lambda$, when all the $x_I \rightarrow \lambda x_I$. While this is still the case in quantum mechanics the energy eigenvalues depend on the other motions so this separation does not by itself yield eigenvalues. However, the possibility of solid-like and liquid-like states where $V_0 = {1 \over 2} kr^2$ and \break $r^{-2} V_2 = {\sum \atop I} {\sum \atop {\! \! \! \! \! <J}} k' \left| {\bf r}_I - {\bf r}_J \right|^{-2}$ suggests that such systems are worthy of further study. The statistical mechanics of the systems with $V = -{GM^2 \over r} \, , r<r_e$, gives negative specific heats just as those studied earlier as examples of phase transitions (Lynden-Bell \& Lynden-Bell 1977). However, within one system there is no gravothermal catastrophe (Antonov 1962, Lynden-Bell \& Wood 1968). \begin{acknowledgments} We thank Dr Dragt for sending us his papers relevant to classifying our states by symmetry and Drs Twambey and Sridhar for referring us to the Calogero model. D. Lynden-Bell is currently supported on a PPARC Senior Research Fellowship. We thank the referees for detailed comments that broadened the background of the paper and improved its clarity. \end{acknowledgments}
\section{Introduction} The transient \xray\ source \srcnm\ is an accreting \xray\ pulsar in an eccentric 24 day orbit (\cite{bil97}) with the O9e star, V635 Cassiopeiae (\cite{ung98}). \xray\ outbursts have been observed from \srcnm\ with \emph{Uhuru} (\cite{for76}), {\em HEAO-1}\ (\cite{whe79,ros79}), {\it Ginga}\ (e.g. \cite{tam92}), and {\it CGRO}/BATSE (\cite{bil97}). A cyclotron resonance scattering feature (CRSF) in \srcnm\ was first noted near 20\,keV by Wheaton, \etal\ (1979) with the UCSD/MIT hard \xray\ (10-100\,keV) experiment aboard {\em HEAO-1}. White, Swank \& Holt (1983) analyzed concurrent data from the lower energy (2-50\,keV) {\em HEAO-1}/A2 experiment and found an additional feature at \aprx12\,keV. Two outbursts of \srcnm\ were observed with {\it Ginga}, in 1990 February and 1991 April (\cite{nag91,tam92,mih95}). The pattern of absorption features differed dramatically in the two outbursts. A pair of features similar to the {\em HEAO-1}\ results in the 1990 outburst gave way to a single feature near 17\,keV in 1991 (\cite{mih98}). We discuss here spectral and timing analyses of observations of the 1999 March outburst (\cite{wil99,hei99}) obtained with the \emph{Rossi X-Ray Timing Explorer} ({\em RXTE}). \section{Observations and Analysis} Observations were made with the Proportional Counter Array (PCA) (\cite{jah96}) and High Energy X-ray Timing Experiment (HEXTE) (\cite{rot98}) on board {\em RXTE}. The PCA is a set of 5 Xenon proportional counters sensitive in the energy range 2--60\,keV with a total effective area of \aprx 7000\,$\rm cm^2$. HEXTE\ consists of two arrays of 4 NaI(Tl)/CsI(Na) phoswich scintillation counters (15-250\,keV) totaling \aprx 1600 {$\rm cm^2$}. The HEXTE\ alternates between source and background fields in order to measure the background. The PCA and HEXTE\ fields of view are collimated to the same 1\hbox{$^\circ$}\ full width half maximum (FWHM) region. Beginning on 1999 March 3, daily, short (\aprx 1\,ks) monitoring observations were carried out. In addition, we performed four long pointings (duration \aprx15-35\,ks, labeled A -- D in Fig.~\ref{f_asm}) to search for CRSFs. Observation B, on 1999 March 11.87-12.32, spanned periastron passage at March 11.95 (\cite{bil97}). Figure~\ref{f_asm} shows the RXTE/All Sky Monitor (ASM, 1.5--12\,keV) light curve of \srcnm\ together with the times of the pointed observations. In this work, we concentrate on observation B. \begin{figure} \centerline{\includegraphics[angle=0,width=3.5in]{0115asm_rev.eps}} \figcaption{\label{f_asm} The {\em RXTE}/ASM light curve of \srcnm\ averaged in 6\,h segments. The vertical dotted lines indicate times of periastron passage. The bars at the top indicate the times of the short, public {\em RXTE}\ pointings and of the long observations (heavy bars, A--D).} \end{figure} \subsection{\label{s_anal}Spectral Analysis} The spectrum of \srcnm\ varies significantly with neutron star rotation phase (\cite{nag91}), making fits to the average spectrum difficult to interpret. In order to study the evolution of the spectrum through the pulse, we corrected photon arrival times to both the solar system and the binary system barycenters using the ephemeris of Bildsten, \etal\ (1997). We then applied a Z$^2$ period search (\cite{buc83}) to the HEXTE\ data to determine the pulse period. Figure~\ref{f_flc} shows folded light curves derived from spectra in 50 pulse phase bins for observation B, where the period was 3.614512(33)s. The folded light curve has a sharp main peak, followed by a broader, softer second peak, similar to earlier reports (\cite{wsh83,bil97}). In searching for CRSFs, we followed the spectral analysis methods described in Kreykenbohm, \etal\ (1998). \begin{figure} \centerline{\includegraphics[angle=0,width=4.25in]{flc_rev.ps}} \figcaption{\label{f_flc} Top: The {\em RXTE}\ folded light curve of \srcnm\ in 5 energy bands. Bottom: F-Test probability for the addition of a second and subsequently a third CRSF to a simple continuum fit to the HEXTE data in 20 phase bins. In both panels, points which fall on the dashed line at $\rm 10^0$ indicate no improvement of the fit for addition of that line. Vertical dotted lines delineate pulse phase 0.70--0.76 corresponding to the spectrum in Fig.~\ref{f_spec}.} \end{figure} Because CRSFs at \aprx12 and \aprx22\,keV are known in \srcnm\ (\cite{wsh83,nag91}), we first concentrated on the HEXTE\ data where higher harmonics might appear. We fit a ``cut-off power law'' (a power law times an exponential) to the HEXTE\ spectra in 20 phase bins. The reduced \chisq\ of these fits ranged from 1.3 to 12.3 (62 degrees of freedom, ``dof''). Concentrated at the main pulse and through the rise and peak of the second, there were significant residuals resembling absorption features near 20--25\,keV. In the fall of the second peak, residuals appeared between 30--40\,keV. We then fit for an absorption feature near 20--25\,keV, resulting in reduced \chisq s between 0.7 and 2.0 (59 dof). We used a simple Gaussian model for the optical depth profile. Fig.~\ref{f_flc}, shows the result of an F-Test for adding this line. In the cases where no line was allowed by the fit, the points are plotted with a value of $\rm 10^0$. Next, we allowed a CRSF between 28--45\,keV. This significantly improved the fits in about half of the phase bins, including some near the main peak where large \aprx20\,keV residuals in the initial fits masked the presence of this line. The corresponding F-test results are also plotted in Fig.~\ref{f_flc}. Although there is strong evidence for multiple lines at other phases, the phase range 0.70--0.76 shared both a significant \aprx20\,keV line as well as the most clearly line-like residuals in 30--40\,keV in the no-lines fit. We therefore chose to concentrate on this phase in this \emph{Letter}. We plan to perform detailed analysis of all phases and all four long (A--D) observations in a future paper. Next, we jointly fit the HEXTE and PCA data for phase 0.70--0.76. To account for uncertainties in the response matrix, 1\% systematic errors were applied to the PCA data. None of the continuum models (high energy cut-off power law, Fermi-Dirac cut-off (FDCO) times a power law, and Negative and Positive power law Exponential (NPEX); see Kreykenbohm, \etal\ 1998) typically used for accreting pulsar spectra provided an acceptable fit without the inclusion of absorption features. A black body with kT\aprx0.4\,keV and photoelectric absorption were required to describe the data. No Fe-K line was required. Ultimately, it was necessary to include CRSFs at \aprx12, \aprx21, and \aprx34\,keV in the joint spectrum. The fitted line parameters were insensitive to the details of the continuum model used. The results given here used a Fermi-Dirac cut-off (FDCO) times a power law, given by $F(E) \propto (1 + e^{(E - E_c)/E_f})^{-1} \times E^{-\Gamma}$. F is the photon flux, $\rm E_c$ the cutoff energy, $\rm E_f$ the folding energy, and $\Gamma$ the photon index. $\rm E_c$ was fixed at zero. An F-Test for the significance of adding the \aprx34\,keV line to a model with only two absorption features gave a chance probability of $\rm 10^{-17}$. \subsection{Temporal Variability} Along with standard Fourier techniques, we analyzed the data in the time domain using the linear state space model (LSSM) formalism described by K{\"o}nig \& Timmer (1997) and Pottschmidt, \etal\ (1998). Parameters of the LSSM are related to dynamical timescales of the system such as oscillation periods, decay times of damped oscillators and stochastic noise. As shown in Figure~\ref{f_lc}, the data are well described by a LSSM of order 2. This model, based on an auto-regressive process, is dominated by a stochastically driven sinusoid of period 552\,s which exponentially decays with an e-folding time of $P_{\rm fold} = 282$\,s. This short $P_{\rm fold}$ accounts for the broad QPO peak seen in the PSD (Fig.~\ref{f_psd}). A Kolmogoroff-Smirnoff test shows that the difference between the data and the LSSM is purely attributable to white noise. The light curve is thus consistent with a single exponentially decaying sinusoid, driven by a white noise process. \section{Results and Discussion} \subsection{Spectrum and CRSFs} Fig.~\ref{f_spec} shows the best fit joint count spectrum from pulse phase 0.70--0.76. Also plotted is the inferred incident photon spectrum. Best fit parameters are given in Table~\ref{t_fit}, and the reduced \chisq\ of the fit is 1.66 (74 dof). This is the first time that a fundamental and two harmonic CRSFs have been detected in a single accreting \xray\ pulsar spectrum. Previously, at most a fundamental and second harmonic have been seen: 4U~1907+09 (\cite{cus98}); or suggested: Vela~X-1 (\cite{kre98}), A0535+25 (\cite{ken94}), 1E~2259+586 (\cite{iwa92}). Those earlier observations lacked the broad-band sensitivity of {\em RXTE}. Furthermore, simple fits to the phase-resolved HEXTE\ spectra show that the \xray\ spectrum varies rapidly with neutron star rotation. We observe significant variations between consecutive phase bins which cover only 2\% of the pulse phase. This suggests complex spatial variations of conditions near the neutron star polar cap. Since the 35\,keV CRSF is only present in about half of the pulse (predominantly during the fall of the second, weaker pulse), and both the 22 and 35\,keV lines are only present together in 3 of 20 coarse phase bins, averaging over large phase angles would likely have washed out the line in the variable continuum. \begin{figure} \centerline{\includegraphics[angle=0,width=3.5in]{best_fit_rev.ps}} \figcaption{\label{f_spec} Top: The PCA and HEXTE\ count spectra (crosses) for pulse phase 0.70-0.76. Also shown are the best fit model (histograms) with 3 CRSFs and the inferred incident spectrum (smooth curve). Middle: The ratio of the data to the best fit model. Bottom: The ratio of the data to a model fit with only two CRSFs. The residuals between 30--40\,keV and the underprediction of the continuum above 40\,keV emphasize the presence of the third line. } \end{figure} The fundamental energy of 12.4\,keV implies a neutron star surface field of $\rm 1.1\times 10^{12}(1 + z)$\,G. Contrary to simple Landau theory, the observed line spacing is not quite harmonic. The ratio of the $\rm 2^{nd}$ and $\rm 3^{rd}$ harmonics to the fundamental are $\rm 1.73 \pm 0.08$ and $\rm 2.71 \pm 0.13$ respectively. We tried fits with the $\rm 2^{nd}$ and $\rm 3^{rd}$ lines constrained to be exact harmonics of the first, whose energy was allowed to vary. The resulting fit was unacceptable (reduced \chisq\ of 6.7 with 76 dof). We did, however, find a reasonable fit with the $\rm 3^{rd}$ harmonic tied to the first but the second free to vary. Nevertheless, an F-test comparing this fit to our best model gives a chance improvement probability of $\rm 2 \times 10^{-4}$ for allowing the non-integer energy ratio of the first and third lines. \begin{table} \caption{\label{t_fit} Parameters of the best fit model spectrum (see \S \ref{s_anal}).} \begin{minipage}{\linewidth} \renewcommand{\thefootnote}{\thempfootnote} \begin{tabular}{lccc} \hline \hline Harmonic & Energy & Width\footnote{$\sigma$ of the Gaussian optical depth profile} & Optical Depth \\ & (keV) & (keV) & \\ \hline 1 (Fundamental) & $\rm 12.40^{+0.65}_{-0.35}$ & $\rm 3.3^{+0.1.9}_{-0.4} $ & $\rm 0.72^{+0.10}_{-0.17} $\\ 2 & $\rm 21.45^{+0.25}_{-0.38} $ &$\rm 4.5^{+0.7}_{-0.9} $ & $\rm 1.24^{+0.04}_{-0.06} $\\ 3 & $\rm 33.56^{+0.70}_{-0.90} $ & $\rm 3.8^{+1.5}_{-0.9}$ & $\rm 1.01^{+0.13}_{-0.14} $\\ \hline \end{tabular} \end{minipage} \end{table} In addition to relativistic shifts, line energies may deviate from harmonic for a number of reasons. The main scattering for the harmonics may take place in regions of different magnetic field, either resulting from optical depth effects in the mound of accreting matter, or for lines primarily produced at opposite magnetic poles. It is interesting to note (see Fig.~\ref{f_flc}) that the second and third harmonics are most significant in the main and secondary pulses, respectively, possibly indicating origins at opposite poles. In our initial fits to the HEXTE\ data alone, we observed that the 20\,keV CRSF varied in both strength and energy (by 20\%) through the pulse phase. This was first observed with {\it Ginga}\ by Nagase, \etal\ (1991) in the 1990 Feb. outburst. The line is strongest and the energy highest (\aprx24\,keV), on the falling edge of the main pulse, which is similar to the behavior of the CRSFs in Cen~X-3 and 4U~1626-67 (\cite{hei99b}). The {\it Ginga}\ data showed significant \aprx11 and \aprx22\,keV lines at all 8 pulse phases analyzed (\cite{mih95}). The HEXTE data find that the \aprx20\,keV line is not required just before the rise of the main pulse (Fig.~\ref{f_flc}). However, with the addition of the PCA data, which constrain the continuum and fundamental line, it is possible that this line will be required at all phases. In any case, strong long term variability of the lines is known (\cite{mih95}), so differences between these results and earlier observations are not surprising. \subsection{Temporal Variability: a 2\,mHz QPO} Figures~\ref{f_lc} and~\ref{f_psd} show the PCA light curve of observation~B and its power spectral density (PSD), respectively. Strong variability with an \aprx500\,s period is obvious. At frequencies above 5\,mHz, the PSD can be described by an overall power law $\propto f^{-1}$ plus peaks at the neutron star rotational frequency and its multiples. Some accreting pulsars have shown a QPO at the beat frequency between the neutron star rotation and the Keplerian orbit at the inner edge of the accretion disk (\cite{fin96}). Using the relations in Finger, Wilson \& Harmon (1996) with a surface B field strength of $\rm 1.3 \times 10^{12}$\,G, a distance of 6\,kpc (\cite{neg99}), and a total flux of $\rm 2.1 \times 10^{-8}$\,\eflux, the expected beat QPO frequency is 0.8\,Hz. Overtones of the rotational frequency confuse the search for QPOs in this region (see Fig.~\ref{f_psd}), and no obvious peaks apart from the pulsation were seen in the 1\,ks short pointings or in observation B. \begin{figure} \centerline{\includegraphics[angle=0,width=3.5in]{0115long.eps}} \figcaption{\label{f_lc} PCA (1.5-60\,keV) light curve in 16\,s bins of observation B together with a second order linear state space model of the variability. The bottom panel shows the residuals of the LSSM. } \end{figure} Below 5\,mHz, the PSD is dominated by a broad QPO feature from the 500\,s oscillation. The shape of this feature is complex and asymmetric, it can neither be described by a Lorentzian line nor by the superposition of two Lorentzian lines. The feature peaks at $\sim$1.5\,mHz and has a FWHM of 1\,mHz ($Q=f_0/\Delta f \approx 1.5$). The excess power of the QPO with respect to the underlying red noise component in the range from 1\,mHz to 4\,mHz is \aprx5\% RMS. The \aprx500\,s period of this oscillation is longer than any \xray\ QPO previously reported in an accreting \xray\ pulsar. Soong \& Swank (1989) reported a broad 0.062\,Hz QPO in {\em HEAO-1}\ observations of a flaring state in \srcnm\ that also did not fit into the beat frequency model. The QPO was probably present in several of the short (\aprx1\,ks) observations as well, as there was apparent, slow variability on a several hundred second timescale. In observation A, the QPO was at most present weakly, as no clear peak is evident in the PSD. Two possible explanations for the QPO are: 1) modulation of the accretion flow, and 2) occultation of the beam by intervening matter in an accretion disk. It seems unlikely that the variability is due to modulation of the accretion flow itself. The timescales at the neutron star pole (milliseconds) and the inner edge of the disk (seconds) are too fast. Furthermore, the rotation period (days) of V635Cas is too long. We compared PCA spectra at minima of the 500\,s cycle to the spectra of the following maxima. The spectral shape is unchanged from 2.5 -- 5\,keV, and only \aprx20\% deviations appear at higher energies. Because the spectrum below 5\,keV is steady through the QPO and the flux varies by a factor of two from peak to minimum, the QPO mechanism cannot be absorption in \emph{cold} material. If it were, the low energy spectrum would be highly modified by photoelectric absorption. It is possible that Thomson scattering in ionized matter causes the variability. We suggest two possible mechanisms, which both require that the accretion disk be viewed nearly edge on. Both are consistent with a second order LSSM process. First, an azimuthal warp propagating around the disk could cause the ionized disk surface to intervene in the line of sight. In this case, the 500\,s timescale is the time for the wave to circle the disk (assuming a single-peaked warp). In the second picture, the absorption takes place in a lump in the disk which orbits at a Kepler period of 500\,s. \section{Summary} We have made two discoveries in the {\em RXTE}\ observations of the 1999 March outburst of \srcnm. The HEXTE\ data have revealed for the first time in any pulsar a third harmonic CRSF. The line spacing between the fundamental and second harmonic and between the second and third harmonics are not equal, and are not multiples of the fundamental line energy. We have also discovered the slowest (2\,mHz) QPO yet observed from an accreting pulsar. It was most pronounced during an observation spanning periastron passage of the neutron star around its massive companion. Based on the timescale, amplitude, and energy spectrum of the oscillation, it is most likely due to obscuration of the neutron star by hot accretion disk matter. The longest QPO previously observed had a timescale of 100\,s in SMC~X-1 (\cite{ang91}). \begin{figure}[h] \centerline{\includegraphics[angle=0,width=3.5in]{longpsd01_rev.eps}} \figcaption{\label{f_psd}The PSD (normalized according to Miyamoto et al. 1991) of the observation B PCA light curve. The PSD has been rebinned such that $\rm df/f =$ 0.35, 0.1 and 0.05 in the frequency bands $10^{-4}$--$10^{-3}$\,Hz, $10^{-3}$--$10^{-1}$\,Hz, and $10^{-1}$--$10^{1}$\,Hz respectively. The \aprx2\,mHz QPO and several harmonics of the 3.6\,s pulsation are clear above a 1/f continuum. } \end{figure} \acknowledgements We thank E. Smith, J. Swank, and C. Williams-Heikkila for rapidly scheduling the observations and supplying the realtime data. ASM data are provided by the {\em RXTE}/ASM teams at MIT and at the {\em RXTE}\ SOF and GOF at NASA's GSFC. This work was supported by NASA grants NAS5-30720 and NAG5-7339, DFG grant Sta 173/22, and a travel grant from the DAAD.
\section{Introduction} Semiclassical trace formulas for both chaotic \cite{Gut67,Gut90} and regular \cite{Ber76} systems relate quantum spectra and classical periodic orbits. These formulas have proven to be useful in the analysis of level statistics \cite{Ber85} and long-range correlations \cite{Win87} in the quantum spectra, and it has even become possible to compute individual eigenenergies from these expressions \cite{Cvi89,Sie91,Ber92,Mai97a,Mai98}. Gutzwiller's trace formula \cite{Gut67,Gut90} and the Berry-Tabor formula \cite{Ber76} are semiclassical approximations to the density of states but do not provide information about experimentally measurable observables, i.e., matrix elements of Hermitian operators. The trace formulas have been extended to the calculation of diagonal matrix elements of smooth operators in Refs.\ \cite{Wil88,Eck92}. The extended trace formulas relate the diagonal matrix elements of operators to the periodic orbit means of the corresponding classical observables. However, these formulas cannot be applied directly for the semiclassical calculation of {\em products} of diagonal matrix element where the weighted density of states cannot, in general, be written as a trace formula. Products of diagonal matrix elements are important in several interesting applications of semiclassical theories, e.g., for the semiclassical theory of matrix element fluctuations \cite{Eck95}, with the variance of an operator $\hat A$ in an eigenstate $|n\rangle$ given by ${\rm Var}_n A\equiv\langle n|\hat A^2|n\rangle-\langle n|\hat A|n\rangle^2$. A semiclassical periodic orbit formula for products of diagonal matrix elements is also of crucial importance for the semiclassical quantization technique developed in Ref.\ \cite{Mai99}, where the classical information of a set of observables is used to significantly improve the convergence properties of periodic orbit quantization. In this paper we investigate {\em non-trace} type formulas for the density of states weighted with the diagonal matrix elements of two operators $\hat A$ and $\hat B$, i.e., $\rho^{(A,B)}(E) =\sum_n\langle n|\hat A|n\rangle\langle n|\hat B|n\rangle\delta(E-E_n)$. From the high resolution analysis of quantum spectra in the semiclassical regime we find strong evidence that weighting the $\delta$-functions in the quantum mechanical density of states with the product of diagonal matrix elements, $\langle n|\hat A|n\rangle\langle n|\hat B|n\rangle$, is equivalent to weighting the periodic orbit contributions in the semiclassical periodic orbit sum with the product of the periodic orbit means, $\langle A\rangle_p\langle B\rangle_p$, of the classical observables $A$ and $B$. The outline of the paper is as follows. In Sec.\ \ref{trace:sec} we first briefly review Gutzwiller's trace formula for chaotic systems and the Berry-Tabor formula for integrable systems, and discuss the extension of both equations to the calculation of diagonal matrix elements. We then apply the theories to systems with scaling properties, and introduce the high resolution analysis (harmonic inversion) of quantum spectra as a powerful tool to numerically verify the validity of the semiclassical expressions. In Sec.\ \ref{non-trace:sec} we present our results on the semiclassical non-trace type formulas. Strong numerical evidence for the validity of the non-trace type equations is provided by the harmonic inversion of spectra of two different systems, viz.\ the hydrogen atom in a magnetic field and the circle billiard. Sec.\ \ref{conclusion} concludes with remarks on useful and important applications, and an outlook on possible generalizations of the non-trace type formulas. \section{Semiclassical trace formulas} \label{trace:sec} \subsection{Matrix element extension of periodic orbit theory} The quantum mechanical density of states can be written as the trace of the Green's function, $\rho(E)=\sum_n\delta(E-E_n)=-(1/\pi)\,{\rm Im}\,{\rm tr}\,\hat G_E^+$. Replacing the quantum mechanical Green's function, $\hat G_E^+=(E-\hat H+i\epsilon)^{-1}$ with its semiclassical analogue and calculating integrals and traces in stationary phase approximation Gutzwiller derived the fundamental equation of {\em periodic orbit theory} \cite{Gut67,Gut90}, i.e., the density of states expressed in terms of quantities of the periodic orbits of the classical system. To obtain the density of states weighted with the diagonal matrix elements of an operator $\hat A$ we start from the generalized trace formula \begin{eqnarray} \rho^{(A)}(E) &=& -{1\over\pi}\, {\rm Im}\, {\rm tr} \left(\hat G_E^+\hat A\right) \nonumber \\ &=& -{1\over\pi}\, \lim_{\epsilon\to 0} {\rm Im}\, \sum_n {\langle n|\hat A|n\rangle\over E-E_n+i\epsilon} \nonumber \\ &=& \sum_n \langle n|\hat A|n\rangle \delta(E-E_n) \; . \label{rho_A_qm:eq} \end{eqnarray} The r.h.s.\ of Eq.\ \ref{rho_A_qm:eq} is the density of states weighted with the diagonal matrix elements $\langle n|\hat A|n\rangle$ of the operator $\hat A$. The semiclassical approximation to Eq.\ \ref{rho_A_qm:eq} for a system with $N$ degrees of freedom reads \cite{Wil88,Eck92} \begin{eqnarray} \label{rho_A_sc:eq} & & \rho^{(A)}(E) = \rho^{(A)}_0(E) \\ &+& {1\over\pi\hbar}\, {\rm Re}\, \sum_p A_p \sum_{r=1}^\infty {T_p\over\sqrt{|\det(M_p^r-I)|}}\, e^{i[S_p(E)/\hbar-{\pi\over 2}\mu_p]r} \; , \nonumber \end{eqnarray} where the Weyl term $\rho^{(A)}_0(E)=h^{-N} \int d{\bf q}d{\bf p} A({\bf q},{\bf p}) \delta(E-H({\bf q},{\bf p}))$ is a smooth function of the energy and the fluctuating part is given by the periodic orbit sum, with $T_p$ the time period, $S_p$ the classical action, $M_p$ the monodromy matrix, and $\mu_p$ the Maslov index of the primitive periodic orbit $p$. The integer $r$ is the repetition number of the orbit. The weights $A_p$ in the periodic orbit sum (\ref{rho_A_sc:eq}) are the means of the observable $A$ along the periodic orbit $p$, i.e. \begin{equation} A_p = {1\over T_p} \int_0^{T_p} A({\bf q}_p(t),{\bf p}_p(t)) dt \; . \label{Ap_t:eq} \end{equation} The derivation of Eq.\ \ref{rho_A_sc:eq} requires smoothness of the observable $A$ over regions in phase space of size $h^N$ \cite{Eck92}. A rigorous mathematical proof of the semiclassical trace formula (\ref{rho_A_sc:eq}) using a coherent states decomposition can be found in \cite{Com97}. In Refs.\ \cite{Wil88,Eck92,Com97} formulas for the semiclassical calculation of diagonal matrix elements are obtained for chaotic systems with isolated periodic orbits. For {\em regular} systems the semiclassical trace formula for the density of states has been derived by Berry and Tabor \cite{Ber76}. For simplicity we restrict ourselves to systems with two degrees of freedom. Assuming now that the Hamiltonian is classically integrable, one can express it in action-angle variables $({\bf I},\varphi)$ with $\varphi_1,\varphi_2\in [0,2\pi]$ as $H({\bf I})$. For a given torus, $\omega_i=\partial H/\partial I_i$ $(i=1,2)$ are the corresponding angular frequencies. Periodic orbits are associated with tori such that the rotation number $\alpha\equiv \omega_1/\omega_2$ is rational, i.e., $\alpha=M_1/M_2$ with $M_1$ and $M_2$ integers. The fluctuating part of the Berry-Tabor formula reads \begin{eqnarray} & & \rho_{\rm fl}(E) \nonumber \\ &=& {1\over\pi\hbar^{3/2}}\, {\rm Re}\, \sum_{\bf M} {T_{\bf M}\over M_2^{3/2}|g_E''|^{1/2}} \, e^{i\left(S_{\bf M}(E)/\hbar-{\pi\over 2}\eta_{\bf M}-{\pi\over 4}\right)}\; , \label{BT:eq} \end{eqnarray} with ${\bf M}=(M_1,M_2)$ specifying the periodic orbit, and $T_{\bf M}$, $S_{\bf M}$, and $\eta_{\bf M}$ the time, action and Maslov index of the orbit, respectively. The function $g_E$ in (\ref{BT:eq}) is obtained by inverting the Hamiltonian, expressed in terms of the actions $(I_1,I_2)$ of the corresponding torus, with respect to $I_2$, viz.\ $H(I_1,I_2=g_E(I_1))=E$ \cite{Boh93}. By analogy with Eq.\ \ref{rho_A_sc:eq} for chaotic systems the Berry-Tabor formula (\ref{BT:eq}) can now be generalized straightforwardly to the semiclassical calculation of diagonal matrix elements \cite{Meh99}, yielding \begin{eqnarray} \label{BT_A:eq} & & \rho^{(A)}(E)= \rho^{(A)}_0(E) \\ &+& {1\over\pi\hbar^{3/2}}\, {\rm Re}\, \sum_{\bf M} A_{\bf M} {T_{\bf M}\over M_2^{3/2}|g_E''|^{1/2}} \, e^{i[S_{\bf M}(E)/\hbar-{\pi\over 2}\eta_{\bf M}-{\pi\over 4}]}\, , \nonumber \end{eqnarray} with \begin{equation} A_{\bf M} = {1\over(2\pi)^2}\int_0^{2\pi}d\varphi_1\int_0^{2\pi}d\varphi_2 A(I_1,I_2,\varphi_1,\varphi_2) \label{Ap_reg:eq} \end{equation} the classical average of the observable $A$ on the torus. \subsection{Scaling systems} In the following we will apply Eqs.\ \ref{rho_A_sc:eq} and \ref{BT_A:eq} to systems with scaling properties. In scaling systems the classical phase space structure does not change for all values of an appropriate scaling parameter, $w$. The scaling parameter is usually some power of an external field strength or, for Hamiltonians with homogeneous potentials, the energy. Examples are billiard systems \cite{Meh99} or atoms in magnetic fields \cite{Fri89,Has89,Wat93}. In scaling systems the shape of periodic orbits does not depend on the scaling parameter, $w$, and the classical action $S_p$ scales as \begin{equation} S_p = w s_p \; . \label{S_po} \end{equation} The scaling parameter plays the role of an inverse effective Planck constant, i.e., $w\equiv\hbar_{\rm eff}^{-1}$. For scaling systems the weighted densities of states, Eqs.\ \ref{rho_A_sc:eq} and \ref{BT_A:eq} can be rewritten as a function of the scaling parameter $w$, i.e. \begin{eqnarray} && \rho^{(A)}(w) = \rho^{(A)}_0(w) \nonumber \\ &+& {1\over\pi\hbar}\, {\rm Re}\, \sum_p A_p \sum_{r=1}^\infty {s_p\over\sqrt{|\det(M_p^r-I)|}}\, e^{i[s_p w-{\pi\over 2}\mu_p]r} \label{rho_A_scaled:eq} \end{eqnarray} for chaotic systems, and \begin{eqnarray} &&\rho^{(A)}(w) = \rho^{(A)}_0(w) \nonumber \\ &+& {1\over\pi\hbar^{3/2}}\, {\rm Re}\, \sum_{\bf M} A_{\bf M} {s_{\bf M}\over M_2^{3/2}|g_E''|^{1/2}} \, e^{i[s_{\bf M}w-{\pi\over 2}\eta_{\bf M}-{\pi\over 4}]} \label{BT_A_scaled:eq} \end{eqnarray} for two-dimensional systems with regular dynamics. Note that the time periods $T_p$ and $T_{\bf M}$ in Eqs.\ \ref{rho_A_sc:eq} and \ref{BT_A:eq} must be replaced with the scaled actions $s_p$ and $s_{\bf M}$. Furthermore the time average of the classical observable $A$ (Eq.\ \ref{Ap_t:eq}) must be replaced with the average with respect to the scaled action, \begin{equation} A_p = {1\over s_p} \int_0^{s_p} A({\bf q}_p(s),{\bf p}_p(s)) ds \; . \label{Ap_s:eq} \end{equation} If an observable $A$ is chosen which is invariant under the scaling of the system [or scales $\sim w^\beta$ with a constant exponent $\beta$] the periodic orbit amplitudes and scaled actions in Eqs.\ \ref{rho_A_scaled:eq} and \ref{BT_A_scaled:eq} do not depend on $w$ [despite a possible power law scaling of the amplitudes with $w^\beta$ which can be transfered to the l.h.s.\ of Eqs. \ref{rho_A_scaled:eq} and \ref{BT_A_scaled:eq}]. The attractive feature of scaling systems is that the semiclassical weighted density of states [or more generally the density of states multiplied by $w^\beta$] is a superposition of sinusoidal functions of the scaling parameter $w$. The Fourier transforms of $w^\beta\rho^{(A)}(w)$ should therefore exhibit sharp peaks at the positions of the scaled actions of the periodic orbits. When analyzing quantum spectra, we will make use of the scaling advantages in the following. \subsection{Precision check of the semiclassical trace formulas} We now wish to apply the semiclassical trace formulas, Eqs.\ \ref{rho_A_scaled:eq} and \ref{BT_A_scaled:eq}, to a physical system with chaotic and regular dynamics, respectively, and to check numerically the validity of the semiclassical equations. The numerical check is not motivated by doubts on the validity of these expressions, which have been mathematically proven, rather we want to introduce a powerful numerical technique for the high precision check of equations of this kind. We will demonstrate the accuracy of the method on the well established semiclassical trace formulas here and then apply the same technique to numerically verify our conjecture on semiclassical non-trace type formulas in Sec.\ \ref{non-trace:sec}. The semiclassical trace formulas can be tested, in principle, by the Fourier transform analysis of quantum spectra. The Fourier transformed spectra should exhibit peaks at the periods (scaled actions) of periodic orbits with amplitudes given by the semiclassical expressions. However, the transformation of spectra with finite length yields limited resolution only, due to the uncertainty principle of the Fourier transform, which implies a fundamental restriction to high precision checks of the semiclassical trace formulas. We therefore adopt the method of Ref.\ \cite{Mai97b} where we introduced {\em harmonic inversion} as a high resolution method for the analysis of quantum spectra. We briefly review the basic ideas of the harmonic inversion technique and refer the reader to Ref.\ \cite{Mai97b} for more details. According to Eqs.\ \ref{rho_A_scaled:eq} and \ref{BT_A_scaled:eq} the semiclassical weighted density of states can be written as the sum of a smooth background $\rho^{(A)}_0(w)$ and oscillatory modulations induced by the periodic orbits, \begin{equation} \label{semicl} \rho^{(A)}(w) = \rho^{(A)}_0(w) + {\rm Re} \sum_p d^{(A)}_p e^{i s_p w} \; . \end{equation} The amplitudes $d^{(A)}_p$ and scaled actions $s_p$ of the periodic orbits are obtained from classical calculations and are in general complex quantities. The amplitudes $d^{(A)}_p$ contain the phase information determined by the Maslov indices of orbits and the classical means of the observable $A$ given by Eqs.\ \ref{Ap_reg:eq} and \ref{Ap_s:eq} for regular and chaotic systems, respectively. Instead of using the standard Fourier analysis to extract the amplitudes and actions, we adjust a finite range of the quantum spectrum by the semiclassical expression (\ref{semicl}) with unknown and in general complex parameters $d^{(A)}_p$ and $s_p$. The problem of fitting a ``signal'' $\rho^{(A)}(w)$ to the functional form (\ref{semicl}) is known as harmonic inversion. As a numerical technique for the harmonic inversion of a signal, i.e.\ a quantum spectrum, we apply the method of filter-diagonalization \cite{Wal95,Man97} which allows extracting the spectral quantities in any given interval of interest. Operationally, one proceeds by setting up a small generalized eigenvalue problem. The actions $s_p$ in the chosen spectral domain and amplitudes $d^{(A)}_p$ are obtained from the resulting eigenvalues and eigenvectors. Thus, the recurrence spectrum is effectively discretized, the number of terms being the number of eigenvalues in the spectral domain. This method is a variational one (as opposed to the Fourier transform) and therefore practically has an infinite resolution once the amount of information contained in the signal $\rho^{(A)}(w)$ is greater than the total number of unknowns $d^{(A)}_p$ and $s_p$. As a physical system for the high precision analysis of quantum spectra and the comparison with the semiclassical trace formulas we choose the hydrogen atom in a magnetic field \cite{Fri89,Has89,Wat93}. This is a scaling system, with $w=\gamma^{-1/3}=\hbar_{\rm eff}^{-1}$ the scaling parameter and $\gamma=B/(2.35\times 10^5\, {\rm T})$ the magnetic field strength in atomic units. Introducing scaled coordinates $\gamma^{2/3}{\bf r}$ and momenta $\gamma^{-1/3}{\bf p}$ and choosing the projection of the angular momentum on the magnetic field axis $L_z=0$ one arrives at the scaled Hamiltonian \begin{equation} \tilde H = {1\over 2}{\bf p}^2 - {1\over r} + {1\over 8}(x^2+y^2) = \tilde E \; , \label{H_scal:eq} \end{equation} with $\tilde E=E\gamma^{-2/3}$ the scaled energy. The classical dynamics is near-integrable at low energies, $\tilde E<-0.5$, and undergoes a transition from regularity to chaos in the energy region $-0.5<\tilde E<-0.13$. At energies above $\tilde E=-0.13$ a Poincar\'e surface of section analysis of the classical dynamics does not exhibit any regular structures larger than of microscopic size \cite{Has89}. We compare spectra at constant scaled energy $\tilde E=-0.1$ with the results of the semiclassical trace formula (\ref{rho_A_scaled:eq}) for chaotic systems, and spectra in the near-integrable regime at $\tilde E=-0.5$ with the extended Berry-Tabor formula (\ref{BT_A_scaled:eq}). We choose two different operators. The first, \begin{equation} \hat A = {1\over 2r{\bf p}^2} \label{Adef:eq} \end{equation} has already served to study the distribution of transition matrix elements in classically chaotic and mixed quantum systems \cite{Boo95,Boo96}. The second operator is \begin{equation} \hat B = r{\bf p}^2 \; . \label{Bdef:eq} \end{equation} Eigenvalues of the scaling parameter $w$ are obtained by solving Schr\"odinger's equation (in semiparabolic coordinates $\mu=\sqrt{r+z}$ and $\nu=\sqrt{r-z}$) \begin{eqnarray} & & \left[2\tilde E ({ \mu}^2+{ \nu}^2) - {1\over 4}\mu^2\nu^2(\mu^2+\nu^2) + 4 \right] \, \Psi(\mu,\nu) \nonumber\\ &=& w^{-2} \left(\hat p_\mu^2 + \hat p_\nu^2\right) \, \Psi(\mu,\nu) \; , \label{H_eps} \end{eqnarray} with the radial operators ${{\hat { p}}_\mu}^2$ and ${{\hat { p}}_\nu}^2$ defined as \[ \hat p_\mu^2 = -{1\over\mu}{\partial \over {\partial\mu}} \left( \mu{\partial \over {\partial\mu}}\right) \; , \quad \hat p_\nu^2 = -{1\over\nu}{\partial \over {\partial\nu}} \left( \nu{\partial \over {\partial\nu}}\right) \; . \] Eq.\ \ref{H_eps} can be written in matrix form by using an appropriate basis set. The resulting generalized eigenvalue problem is solved numerically. It has to be noted that the eigenvectors obtained, $|\psi_n\rangle$, are orthonormal with respect to the scaled momentum operator, i.e. \begin{equation} \langle\psi_m|\hat p_\mu^2 + \hat p_\nu^2|\psi_n\rangle = \langle m|n\rangle = \delta_{mn} \; , \end{equation} with modified eigenvectors $|n\rangle$ defined by $|n\rangle\equiv (\hat p_\mu^2+\hat p_\nu^2)^{1/2}|\psi_n\rangle$. The diagonal matrix elements of an operator $\hat A$ are therefore obtained as \begin{equation} A_{nn} = \langle n|\hat A|n \rangle = \langle\psi_n|\hat A (\hat p_\mu^2 + \hat p_\nu^2)|\psi_n \rangle \; . \end{equation} We are now prepared to compare the quantum spectra of the hydrogen atom in a magnetic field with the semiclassical approximations in the chaotic and regular regime of the classical phase space. \subsubsection{Chaotic regime} We have calculated 3181 eigenvalues $w_n<80$ of the scaling parameter and the diagonal matrix elements of the two operators $\hat A=1/(2r{\bf p}^2)$ and $\hat B=r{\bf p}^2$ for the hydrogen atom in a magnetic field at constant scaled energy $\tilde E=-0.1$. The distributions of the matrix elements are presented in Fig.\ \ref{fig1}. The matrix elements are distributed randomly around the mean values without showing any regular pattern, as is typical of systems with chaotic dynamics. The quantum mechanical weighted density of states \begin{equation} \rho^{(A)}(w) = \sum_n \langle n|\hat A|n\rangle \delta(w-w_n) \label{rho_A_qm_scaled:eq} \end{equation} can now be analyzed with the harmonic inversion technique to obtain the scaled actions $s_p$ and the amplitudes $d^{(A)}_p$ (see Eq.\ \ref{semicl}) of the classical periodic orbits. As can be seen from Eq.\ \ref{rho_A_scaled:eq} the periodic orbit amplitudes \newpage \phantom{} \begin{figure} \vspace{9.5cm} \special{psfile=fig1.ps voffset=-25 hoffset=-25 vscale=46 hscale=46} \caption{\label{fig1} Values of the diagonal matrix elements $\langle n|\hat A|n\rangle$ for the hydrogen atom in a magnetic field at scaled energy $\tilde E=-0.1$ in the chaotic region of phase space as functions of the scaling parameter $w=\gamma^{-1/3}$ ($\gamma\sim$ magnetic field strength): (a) operator $\hat A=1/(2r{\bf p}^2)$; (b) $\hat B = r{\bf p}^2$. } \end{figure} \begin{equation} d^{(A)}_p = A_p d_p \label{d_Ap:eq} \end{equation} are given as the product of the amplitudes, $d_p$ of Gutzwiller's original trace formula, and the classical periodic orbit means $A_p$ in Eq.\ \ref{Ap_s:eq}. For the graphical presentation of the results it is therefore convenient to divide the quantum amplitudes $d^{(A)}_p$ obtained by the harmonic inversion of the spectra by the amplitudes, $d_p$ of Gutzwiller's trace formula. The periodic orbit quantities $A_p$ obtained in this way from the quantum spectra at scaled energy $\tilde E=-0.1$ are presented in Fig.\ \ref{fig2} for three different operators, viz.\ (a) the identity $\hat I$, and the operators (b) $\hat A=1/(2r{\bf p}^2)$ and (c) $\hat B=r{\bf p}^2$. The solid lines and crosses in Fig.\ \ref{fig2} mark the periodic orbit means obtained by the harmonic inversion of the quantum spectra. For comparison the dashed lines and squares present the periodic orbit means of the observable obtained classically via Eq.\ \ref{Ap_s:eq}. For the identity the classical periodic orbit averages (squares in Fig.\ \ref{fig2}a) are exactly equal to one. This is in excellent agreement with the harmonic inversion analysis of the quantum mechanical density of states (crosses in Fig.\ \ref{fig2}a), despite the two weakly separated periodic orbit contributions around $s/2\pi\approx 1.1$. For the two operators $\hat A=1/(2r{\bf p}^2)$ in Fig.\ \ref{fig2}b and $\hat B=r{\bf p}^2$ in Fig.\ \ref{fig2}c the agreement between the periodic orbit means obtained by harmonic inversion of the quantum spectra and classically by Eq.\ \ref{Ap_s:eq} is of similar high accuracy as \newpage \phantom{} \begin{figure} \vspace{9.7cm} \special{psfile=fig2.ps voffset=-25 hoffset=-25 vscale=46 hscale=46} \caption{\label{fig2} Periodic orbit means of observables (a) the identity, (b) $A=1/(2r{\bf p}^2)$, and (c) $B = r{\bf p}^2$ for the hydrogen atom in a magnetic field at scaled energy $\tilde E=-0.1$ as functions of the scaled action. Solid lines and crosses: Results of the harmonic inversion of quantum spectra. Dashed lines and squares: Periodic orbit means obtained by classical calculations. The agreement between the quantum and the classical calculations seems to be excellent, except for the nearly degenerate recurrences at $s/2\pi\approx 1.1$. } \end{figure} \noindent for the identity in Fig.\ \ref{fig2}a. The results presented in Fig.\ \ref{fig2} demonstrate that harmonic inversion of quantum spectra \cite{Mai97b} is indeed a powerful tool for the high precision check of semiclassical theories. Fig.\ \ref{fig2} provides an excellent numerical verification, by way of example of the hydrogen atom in a magnetic field and the chosen set of operators, of the validity of the semiclassical trace formula (\ref{rho_A_scaled:eq}) for chaotic systems. \subsubsection{Regular regime} In the same way as described above we have checked the validity of the extended Berry-Tabor formula (\ref{BT_A_scaled:eq}) for integrable systems. As a physical system we again choose the hydrogen atom in a magnetic field, but at low scaled energy $\tilde E=-0.5$, where the classical phase space is regular. We have calculated 5640 eigenvalues $w_n<160$ of the scaling parameter and the diagonal matrix elements of the two operators $\hat A=1/(2r{\bf p}^2)$ and $\hat B=r{\bf p}^2$. The weighted density of states (\ref{rho_A_qm_scaled:eq}) for the identity, and the operators $\hat A=1/(2r{\bf p}^2)$ and $\hat B=r{\bf p}^2$ have been analyzed in the same way as explained above. The results obtained for the regular system at scaled energy $\tilde E=-0.5$ resemble those of Fig.\ \ref{fig2} for the chaotic system. The difference is that the averages of the observables for the resonant tori have been extracted from the quantum spectra by application of the generalized Berry-Tabor formula (\ref{BT_A_scaled:eq}). The quantum results perfectly agree with the classical averages which illustrates the validity of the generalized Berry-Tabor formula. \section{Non-trace type formulas} \label{non-trace:sec} The generalized semiclassical trace formulas (\ref{rho_A_scaled:eq}) and (\ref{BT_A_scaled:eq}) discussed in Sec.\ \ref{trace:sec} allow the semiclassical calculation of the diagonal matrix elements of smooth operators. However, it would be desirable to know even more generalized expressions for the calculation of {\em products} of matrix elements. As mentioned in the introduction, such formulas are important, e.g., in the semiclassical theory of matrix element fluctuations \cite{Eck95} or for the construction of cross-correlated periodic orbit sums \cite{Mai99}. To study matrix element fluctuations of an operator $\hat A$ the density of states can be weighted with the variances ${\rm Var}_n A\equiv\langle n|\hat A^2|n\rangle-\langle n|\hat A|n\rangle^2$, i.e. \begin{eqnarray} \rho^{({\rm Var}~A)}(E) &\equiv& \sum_n \langle n|\hat A^2|n\rangle \delta(E-E_n) \nonumber \\ &-& \sum_n \langle n|\hat A|n\rangle^2 \delta(E-E_n) \; . \label{rho_Var_A_qm:eq} \end{eqnarray} The first term in (\ref{rho_Var_A_qm:eq}) can be written as a semiclassical trace formula (Eqs.\ \ref{rho_A_sc:eq} and \ref{BT_A:eq} for chaotic and regular systems, respectively) with the observable $A$ replaced with its square, $A^2$. However, because of the squares of the matrix elements, the second term in (\ref{rho_Var_A_qm:eq}) in general cannot be expressed in a straightforward fashion with the help of the Green's operator $\hat G_E^+$ as a trace formula. The trivial exception is when the operator $\hat A$ commutes with the Hamiltonian, which means that $A$ is a constant of motion and thus its variance vanishes. Thus the derivation of a semiclassical approximation to the second term in (\ref{rho_Var_A_qm:eq}) constitutes a nontrivial problem. One solution can be obtained by application of periodic orbit sum rules \cite{Ber85}. Using smooth approximations of the $\delta$-functions, e.g.\ Gaussians of width $\epsilon$, \begin{equation} \delta_\epsilon(E) = {1\over\sqrt{2\pi}\epsilon} e^{-E^2/2\epsilon^2} \; , \end{equation} and the relation \begin{equation} \delta_\epsilon^2(E) = {1\over 2\sqrt{\pi}\epsilon} \delta_{\epsilon/\sqrt{2}}(E) \end{equation} the second term in (\ref{rho_Var_A_qm:eq}) can formally be written as the square of the density of states weighted with the diagonal matrix elements \cite{Ber85,Eck95}, viz. \begin{eqnarray} & & \sum_n \langle n|\hat A|n\rangle^2 \delta_{\epsilon/\sqrt{2}} (E-E_n) \nonumber \\ &=& 2\sqrt{\pi}\epsilon\sum_n \langle n|\hat A|n\rangle^2 \delta_\epsilon^2 (E-E_n) \nonumber \\ &=& 2\sqrt{\pi}\epsilon\left[\sum_n \langle n|\hat A|n\rangle \delta_\epsilon (E-E_n)\right]^2 \; . \label{Berry_trick:eq} \end{eqnarray} The width $\epsilon$ in (\ref{Berry_trick:eq}) must be chosen sufficiently small so that the smoothed $\delta$-functions do not overlap. However, it should be noted that this condition cannot be fulfilled for systems with degenerate states. On the r.h.s.\ of Eq.\ \ref{Berry_trick:eq} the weighted density of states can now be written as a trace formula and replaced with its semiclassical approximations (\ref{rho_A_sc:eq}) and (\ref{BT_A:eq}) for chaotic and regular systems, respectively. Evaluation of the square of the periodic orbit sum on the r.h.s.\ of (\ref{Berry_trick:eq}) then finally yields a double sum over all periodic orbits of the classical system. Although this result is formally correct, it is very inconvenient for practical applications for the following reasons. Firstly, the number of periodic orbits proliferates exponentially in chaotic systems and the handling of the single periodic orbit sum is already nontrivial. The practical evaluation of the double sum would be even more cumbersome. Secondly, the width $\epsilon$ in (\ref{Berry_trick:eq}) is a free parameter. Although the results should not depend on the width if $\epsilon$ is chosen sufficiently small, the appropriate choice may render numerical calculations extremely expensive. Thirdly, and most importantly, the r.h.s.\ of Eq.\ \ref{Berry_trick:eq} does not coincide with the ``simple'' trace formulas in those special cases, where the operator $\hat A$ commutes with the Hamiltonian. Even for the simplest operator, the identity $\hat A=\hat I$, we end up with the nontrivial periodic orbit sum rule of Ref.\ \cite{Ber85} instead of Gutzwiller's trace formula for the density of states. Especially the third point indicates that the procedure described above might not be the simplest way to construct a semiclassical approximation to non-trace type formulas such as Eq.\ \ref{rho_Var_A_qm:eq}. It is the main objective of this section to present a semiclassical approximation to non-trace type weighted densities of states. Our semiclassical expressions agree with the well established ``simple'' semiclassical trace formulas when the weighted density of states can be written, for at least one of the operators commuting with the Hamiltonian, as a quantum mechanical trace formula. Starting from a more general equation than (\ref{rho_Var_A_qm:eq}) we study the density of states \begin{equation} \rho^{(A,B)}(E) \equiv \sum_n \langle n|\hat A|n\rangle \langle n|\hat B|n\rangle \delta(E-E_n) \; , \label{rho_AB_qm:eq} \end{equation} weighted with the product of the diagonal matrix elements of two smooth operators $\hat A$ and $\hat B$. Eq.\ \ref{rho_AB_qm:eq} is the starting point to construct a quantum mechanical cross-correlation function from a set of operators in Ref.\ \cite{Mai99}. The variance of matrix elements (Eq.\ \ref{rho_Var_A_qm:eq}) is obtained by setting $\hat B=\hat A$. The weighted density of states (\ref{rho_AB_qm:eq}) can only be written as a trace formula, $\rho^{(A,B)}(E)=(-1/\pi)\,{\rm Im~tr}\,\{\hat A\hat G_E^+\hat B\}$ if either $\hat A$ or $\hat B$ commutes with the Hamiltonian. As discussed in Sec.\ \ref{trace:sec} (see Eq.\ \ref{d_Ap:eq}) the semiclassical expressions for the weighted densities of states differ from Gutzwiller's trace formula and the Berry-Tabor formula in the following way. The periodic orbit amplitudes are multiplied with the classical periodic orbit (or torus) averages of the observable $A$. We now assume that this ansatz is still valid for the non-trace type weighted density of states (\ref{rho_AB_qm:eq}), i.e., its semiclassical analogue has the same functional form as Gutzwiller's periodic orbit sum but with periodic orbit amplitudes $d_p$ multiplied with the classical averages $A_p$ and $B_p$ of both observables $A$ and $B$, \begin{equation} d^{(A,B)}_p = A_p B_p d_p \; , \label{d_ABp:eq} \end{equation} with $A_p$ and $B_p$ given by Eqs.\ \ref{Ap_t:eq} and \ref{Ap_reg:eq} for chaotic and regular systems, respectively. As can easily be seen, this ansatz has the property that the trace formulas (\ref{rho_A_sc:eq}) and (\ref{BT_A:eq}) are recovered if one of the operators is chosen to be the identity or one of the operators commutes with the Hamiltonian. However, the general validity of this ansatz is not at all obvious, and will be checked numerically by the high resolution analysis of quantum spectra in the following. With the ansatz (\ref{d_ABp:eq}) for the periodic orbit amplitudes the semiclassical analogue to the non-trace type formula (\ref{rho_AB_qm:eq}) reads \begin{eqnarray} \label{rho_AB_sc:eq} & & \rho^{(A,B)}(E) = \rho^{(A,B)}_0(E) \\ &+& {1\over\pi\hbar}\, {\rm Re}\, \sum_p A_p B_p \sum_{r=1}^\infty {T_p\over\sqrt{|\det(M_p^r-I)|}}\,e^{i[S_p(E)/\hbar-{\pi\over 2}\mu_p]r} \; , \nonumber \end{eqnarray} for systems with underlying chaotic classical dynamics, and \begin{eqnarray} \label{BT_AB:eq} & & \rho^{(A,B)}(E) = \rho^{(A,B)}_0(E) \\ &+& {1\over\pi\hbar^{3/2}}\, {\rm Re}\, \sum_{\bf M} A_{\bf M} B_{\bf M} {T_{\bf M}\over M_2^{3/2}|g_E''|^{1/2}} \, e^{i[S_{\bf M}(E)/\hbar-{\pi\over 2}\eta_{\bf M}-{\pi\over 4}]}\, , \nonumber \end{eqnarray} for integrable systems. Eqs.\ \ref{rho_AB_sc:eq} and \ref{BT_AB:eq} are the central propositions of this paper, and generalize the semiclassical trace formulas (\ref{rho_A_sc:eq}) and (\ref{BT_A:eq}) to the non-trace type weighted density of states (\ref{rho_AB_qm:eq}). The nontrivial statement is that weighting the quantum mechanical density of states with the product of diagonal matrix elements of smooth operators is equivalent, on the semiclassical level, to weighting the periodic orbit contributions in the periodic orbit sum with the product of the averages of the corresponding classical observables. In analogy to the discussion of scaling properties in Sec.\ \ref{trace:sec}, Eqs.\ \ref{rho_AB_sc:eq} and \ref{BT_AB:eq} can be reformulated for scaling systems, viz. \begin{eqnarray} \label{rho_AB_scaled:eq} & & \rho^{(A,B)}(w) = \rho^{(A,B)}_0(w) \\ &+& {1\over\pi\hbar}\, {\rm Re}\, \sum_p A_p B_p \sum_{r=1}^\infty {s_p\over\sqrt{|\det(M_p^r-I)|}}\, e^{i[s_p w-{\pi\over 2}\mu_p]r} \nonumber \end{eqnarray} for chaotic systems, and \begin{eqnarray} \label{BT_AB_scaled:eq} & & \rho^{(A,B)}(w) = \rho^{(A,B)}_0(w) \\ &+& {1\over\pi\hbar^{3/2}}\, {\rm Re}\, \sum_{\bf M} A_{\bf M} B_{\bf M} {s_{\bf M}\over M_2^{3/2}|g_E''|^{1/2}} \, e^{i[s_{\bf M}w-{\pi\over 2}\eta_{\bf M}-{\pi\over 4}]} \nonumber \end{eqnarray} for two-dimensional systems with regular dynamics. For scaling systems the classical periodic orbit averages $A_p$ and $B_p$ in (\ref{rho_AB_scaled:eq}) must be calculated with respect to the classical action instead of time as defined in Eq.\ \ref{Ap_s:eq}. In the following we will provide convincing numerical evidence for the validity of the semiclassical non-trace type formulas by the high precision analysis (harmonic inversion) of quantum spectra of two different systems, viz.\ the hydrogen atom in a magnetic field and the circle billiard. A rigorous mathematical proof of the expressions given above is still lacking and constitutes a challenge for the further development of semiclassical theories. \subsection{Hydrogen atom in a magnetic field} To demonstrate the validity of the semiclassical non-trace type formulas, Eqs.\ \ref{rho_AB_scaled:eq} and \ref{BT_AB_scaled:eq}, we use the same system and set of operators as in Sec.\ \ref{trace:sec}, viz.\ the hydrogen atom in a magnetic field at scaled energies $\tilde E=-0.1$ and $\tilde E=-0.5$ in the chaotic and near-integrable regime, respectively, and the operators $\hat A=1/(2r{\bf p}^2)$ and $\hat B=r{\bf p}^2$. With the quantum mechanical eigenvalues and diagonal matrix elements at hand, we construct the weighted densities of states (see Eq.\ \ref{rho_AB_qm:eq}) (a) $\rho^{(A,A)}(w)$, (b) $\rho^{(B,B)}(w)$, and (c) $\rho^{(A,B)}(w)$. These spectra are analyzed with the harmonic inversion technique as described in Sec.\ \ref{trace:sec}. The analysis provides the scaled action $s_p$ of the periodic orbits and the periodic orbit amplitudes $d^{(A,A)}_p$ ($d^{(B,B)}_p$ and $d^{(A,B)}_p$). The amplitudes of the weighted densities of states are divided by the amplitudes of the unweighted densities of states to obtain, according to Eq.\ \ref{d_ABp:eq}, the products of the periodic orbit means $A_p^2$ ($B_p^2$ and $A_pB_p$). These values are presented as solid lines and crosses in Fig.\ \ref{fig5} for the spectra in the chaotic regime at scaled energy $\tilde E=-0.1$ and in Fig.\ \ref{fig6} for the spectra at scaled energy $\tilde E=-0.5$ in the near-integrable regime. For comparison, the squares mark the products of the periodic orbit means obtained from the classical calculations. As in Fig.\ \ref{fig2} for the high precision check of the semiclassical trace formula (\ref{rho_A_scaled:eq}), the agreement between the quantum and classical calculations is found to be very good, which strongly supports the validity of the semiclassical non-trace type expressions. Note that the somewhat larger deviations between the crosses and squares for the nearly degenerate recurrencies at $s_p/2\pi\approx 1.1$ in Fig.\ \ref{fig5} have also been observed in Fig.\ \ref{fig2} for the semiclassical trace formulas, i.e., the deviation does not indicate any failure of the non-trace type formula (\ref{rho_AB_scaled:eq}). \newpage \phantom{} \begin{figure} \vspace{9.5cm} \special{psfile=fig5.ps voffset=-25 hoffset=-25 vscale=46 hscale=46} \caption{\label{fig5} Products of periodic orbit means of the two observables $A=1/(2r{\bf p}^2)$ and $B = r{\bf p}^2$ for the hydrogen atom in a magnetic field at scaled energy $\tilde E=-0.1$ as functions of the scaled action. Solid lines and crosses: Results of the harmonic inversion of the non-trace type weighted densities of states. Dashed lines and squares: Results obtained by classical calculations. As in Fig.\ \ref{fig2}, the agreement between the quantum and the classical calculations seems to be excellent, except for the nearly degenerate recurrences at $s/2\pi\approx 1.1$. } \end{figure} \subsection{Circle billiard} We now investigate the validity of the semiclassical non-trace type formula (\ref{BT_AB_scaled:eq}) on a second system, viz.\ the integrable circle billiard. This system also serves as a model example in Ref.\ \cite{Mai99} to construct a semiclassical cross-correlated periodic orbit sum for a given set of smooth observables, and to calculate semiclassical spectra and diagonal matrix elements by harmonic inversion of the cross-correlation function. As is well known, Schr\"odinger's equation for the circle billiard with radius $R$ can be separated in polar coordinates $(r,\varphi)$, and the wave functions can be expressed in terms of Bessel functions, \begin{equation} \psi_{nm}(r,\varphi) = {\cal N}_{nm} J_{|m|}(k_{nm}r)e^{im\varphi} \; , \label{Bessel:eq} \end{equation} with the ${\cal N}_{nm}$ being normalization constants, $m$ the angular momentum quantum number, and $k_{nm}=\sqrt{2ME_{nm}}/\hbar$ the quantized wave numbers obtained as the $n$th zero of Bessel functions, $J_{|m|}(k_{nm}R)=0$. In the following we choose radius $R=1$. We calculated \newpage \phantom{} \begin{figure} \vspace{9.2cm} \special{psfile=fig6.ps voffset=-35 hoffset=-25 vscale=46 hscale=46} \caption{\label{fig6} Same as Fig.\ 5 but at scaled energy $\tilde E=-0.5$. } \end{figure} \noindent 31208 eigenvalues $k_{nm}<500$, and the diagonal matrix elements of the operators $r$ and $r^2$. The quantum spectra of (a) the unweighted density of states $\rho(k)$, (the wave number $k$ is the scaling parameter, $w=k$ for billiard systems \cite{Mai99}) and the density of states weighted with the matrix element expressions (b) $\langle\psi_{nm}|r^2|\psi_{nm}\rangle$, (c) $\langle\psi_{nm}|r|\psi_{nm}\rangle^2$, and (d) the variance $\langle{\rm Var}~r\rangle_{nm}\equiv \langle\psi_{nm}|r^2|\psi_{nm}\rangle-\langle\psi_{nm}|r|\psi_{nm}\rangle^2$ have been analyzed with the harmonic inversion method. The amplitudes obtained, divided by the amplitudes of the Berry-Tabor formula, are presented as solid lines and crosses in Fig.\ \ref{fig7}, and the corresponding classical averages are drawn as squares for comparison. As can be seen, the agreement is perfect, not only for the identity and the periodic orbit means of the observable $r^2$ in Fig.\ \ref{fig7}a and \ref{fig7}b, verifying the Berry-Tabor formula and its extension (\ref{BT_A_scaled:eq}), but also for the squares of the periodic orbit means of $r$ and the variance of this observable in Fig.\ \ref{fig7}c and \ref{fig7}d, where the agreement demonstrates the validity of the non-trace type equation (\ref{BT_AB_scaled:eq}) for the circle billiard with $\hat A=\hat B=r$. The squares in Fig.\ \ref{fig7}d mark the classical variances of the observable $r$ on the various resonant tori. Our conjecture therefore provides a basic formula for semiclassical matrix element fluctuations, since it directly relates the quantum variances ${\rm Var}_n A\equiv\langle n|\hat A^2|n\rangle-\langle n|\hat A|n\rangle^2$ of a smooth operator $\hat A$ to the classical variances ${\rm Var}_p A\equiv \langle A^2\rangle_p-\langle A\rangle_p^2$ of the observable $A$ taken along the periodic orbits or resonant tori. The perfect agreement between the quantum and classical results for the circle billiard in Fig.\ \ref{fig7} compared to \newpage \phantom{} \begin{figure} \vspace{10.5cm} \special{psfile=fig7.ps voffset=-25 hoffset=-25 vscale=46 hscale=46} \caption{\label{fig7} Classical averages on rational tori of (a) the identity, (b) $\langle r^2\rangle_p$, (c) $\langle r\rangle_p^2$, and (d) the variances ${\rm Var}_p r\equiv\langle r^2\rangle_p-\langle r\rangle_p^2$ for the circle billiard with radius $R=1$ as functions of the scaled action. Solid lines and crosses: Results of the harmonic inversion of quantum spectra. Dashed lines and squares: Periodic orbit means obtained by classical calculations. } \end{figure} \noindent the very good but not absolutely perfect results for the hydrogen atom in a magnetic field in Figs.\ \ref{fig5} and \ref{fig6} may be explained by the different number of quantum states used in the harmonic inversion analysis. For the circle billiard we have calculated more than 30000 states, which is by about a factor of 10 (5.5) times more quantum states than for the hydrogen atom in a magnetic field at scaled energy $\tilde E=-0.1$ ($\tilde E=-0.5$). \section{Conclusion and outlook} \label{conclusion} We have extended semiclassical trace formulas for the density of states of regular and chaotic systems, or the density of states weighted with the diagonal matrix elements of smooth operators, to the more general class of {\em non-trace} type equations, where the density of states is weighted with the diagonal matrix elements of two operators $\hat A$ and $\hat B$, i.e., $\rho^{(A,B)}(E) =\sum_n\langle n|\hat A|n\rangle\langle n|\hat B|n\rangle\delta(E-E_n)$. By the high resolution analysis (harmonic inversion) of the quantum spectra of two different systems, viz.\ the hydrogen atom in a magnetic field and the circle billiard, we have given numerical evidence that weighting the quantum mechanical density of states with the product of the diagonal matrix elements $\langle n|\hat A|n\rangle\langle n|\hat B|n\rangle$ is equivalent, on the semiclassical level, to weighting the periodic orbit contributions in the periodic orbit sum with the product of the averages of the corresponding classical observables, $\langle A\rangle_p\langle B\rangle_p$, where the means are taken along the periodic orbits or resonant tori for chaotic and regular systems, respectively. However, a rigorous mathematical derivation of semiclassical non-trace type formulas appears nontrivial, and would be a challenging task for the further development of semiclassical theories. There are several useful and important applications of semiclassical non-trace type formulas. For example, it enables the semiclassical approach to matrix element fluctuations. The variances ${\rm Var}_n A\equiv\langle n|\hat A^2|n\rangle-\langle n|\hat A|n\rangle^2$ of the diagonal matrix elements of a smooth operator $\hat A$ are expressed in terms of the variances ${\rm Var}_p A\equiv\langle A^2\rangle_p-\langle A\rangle_p^2$ of the classical observable $A$ taken along the periodic orbits or resonant tori. Non-trace type formulas also provide the semiclassical approximation to cross-correlated weighted density of states, $\rho_{\alpha\alpha'}(E)=\sum_n\langle n|\hat A_\alpha|n\rangle \langle n|\hat A_{\alpha'}|n\rangle\delta(E-E_n)$ with a set of smooth operators $\hat A_\alpha$, $\alpha=1,\dots,D$. The additional classical information obtained from the set of classical observables can be used to significantly improve the convergence properties of semiclassical quantization methods \cite{Mai99}. In this paper we have investigated non-trace type expressions for products of two diagonal matrix elements. These products have been chosen because of the important applications to semiclassical matrix element fluctuations, i.e., the calculation of variances of matrix elements and to the semiclassical quantization method in Ref.\ \cite{Mai99}. However, our conjecture is not restricted to products of two matrix elements. For example, it appears straightforward to generalize Eqs.\ \ref{rho_AB_sc:eq} and \ref{BT_AB:eq} to products of more than two matrix elements and classical periodic orbit means. The most general case of non-trace type equations would be the analysis of functions $f(A_{nn},B_{nn},C_{nn},\dots)$ of one or more diagonal matrix elements, i.e., $\rho^{(f)}(E)=\sum_nf(A_{nn},B_{nn},C_{nn},\dots)\delta(E-E_n)$, which should be obtained semiclassically by multiplying the periodic orbit amplitudes of Gutzwiller's trace formula or the Berry-Tabor formula with the function $f(\langle A\rangle_p,\langle B\rangle_p,\langle C\rangle_p,\dots)$ of the periodic orbit means of the corresponding classical observables. Certainly the operators and the function $f$ must be smooth. Clearly, further investigations will be necessary to verify that conjecture and to specify the smoothness conditions on operators and functions. In conclusion, the analysis of non-trace type equations will provide a valuable instrument for extending the relation between quantum mechanical matrix elements on the one side and the periodic orbit means of classical observables on the other. \acknowledgements We acknowledge stimulating discussions with J.\ Keating. This work was supported in part by the Son\-der\-for\-schungs\-be\-reich No.\ 237 of the Deutsche For\-schungs\-ge\-mein\-schaft. J.M.\ is grateful to Deutsche For\-schungs\-ge\-mein\-schaft for a Habilitandenstipendium (Grant No.\ Ma 1639/3).\\[-3ex]
\section{Introduction} The determination of the value of the Hubble constant, $H_0$, is one of the classical tasks of {\it observational} cosmology. In the framework of the expanding space paradigm it provides a measure of the distance scale in FRW universes and its reciprocal gives the time scale. This problem has been approached in various ways. A review on the recent determinations of the value of $H_0$ shows that most methods provide values at $H_0\sim55\dots75$ (for brevity we omit the units; all $H_0$ values are in $\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$): Virgo cluster yields $55\pm7$ and clusters from Hubble diagram with relative distances to Virgo $57\pm7$ (Federspiel et al. \cite{Federspiel98}), type Ia supernovae give $60\pm10$ (Branch \cite{Branch98}) or $65\pm7$ (Riess et al. \cite{Riess98}), Tully-Fisher relation in I-band yields $69\pm5$ (Giovanelli et al. \cite{Giovanelli97}) and $55\pm7$ in B-band (Theureau et al. \cite{Theureau97b}, value and errors combined from the diameter and magnitude relations), red giant branch tip gives $60\pm11$ (Salaris \& Cassisi \cite{Salaris98}), gravitational lens time delays $64\pm13$ (Kundi\'c et al. \cite{Kundic97}) and the `sosies' galaxy method $60\pm10$ (Paturel et al. \cite{Paturel98}). Sunyaev-Zeldovich effect has given lower values, $49\pm29$ by Cooray (\cite{Cooray98}), $47^{+23}_{-15}$ by Hughes \& Birkinshaw (\cite{Hughes98}), but the uncertainties in these results are large due to various systematical effects (Cen, \cite{Cen98}). Surface brightness fluctuation studies provide a higher value of $87\pm11$ (Jensen et al., \cite{Jensen99}), but most methods seem to fit in the range 55 - 75 stated above. An important comparison to these local values may be found after the cosmic microwave background anisotropy probes (MAP and Planck) and galaxy redshift surveys (2dF and SDSS) offer us a multitude of high resolution data (Eisenstein et al., \cite{Eisenstein98}). Note that most of the errors cited here as well as given in the present paper are $1\sigma$ errors. The present line of research has its roots in the work of Bottinelli et al. (\cite{Bottinelli86}), where $H_0$ was determined using spiral galaxies in the {\it field}. They used the direct Tully-Fisher relation (Tully \& Fisher \cite{Tully77}): \begin{equation} \label{E1} \\ M\propto\log V_\mathrm{max}, \end{equation} where $M$ is the absolute magnitude in a given band and $\log V_\mathrm{max}$ is the maximum rotational velocity measured from the hydrogen 21 cm line width of each galaxy. Gouguenheim (\cite{Gouguenheim69}) was the first to suggest that such a relation might exist as a distance indicator. Bottinelli et al. (\cite{Bottinelli86}) paid particular attention to the elimination of the so-called {\it Malmquist bias}. In general terms, the determination of $H_0$ is subject to the Malmquist bias of the $2^\mathrm{nd}$ kind: the inferred value of $H_0$ depends on the distribution of the derived distances $r$ for each true distance $r'$ (Teerikorpi \cite{Teerikorpi97}). Consider the expectation value of the derived distance $r$ at a given true distance $r'$: \begin{equation} \label{E2} \\ E(r\vert r')=\int\limits_0^{\infty}\!\mathrm{d}r\,r\,P(r\vert r'). \end{equation} The integral is done over {\it derived} distances $r$. For example, consider a strict magnitude limit: for each true distance the derived distances are exposed to an upper cut-off. Hence the expectation value for the derived distance $r$ at $r'$ is too small and thus $H_0$ will be overestimated. Observationally, the direct Tully-Fisher relation takes the form: \begin{equation} \label{E3} \\ X = \mathrm{slope}\times p + \mathrm{cst}, \end{equation} where we have adopted a shorthand $p$ for $\log V_\mathrm{max}$ and $X$ denotes either the absolute magnitude $M$ or $\log D$, where $D$ labels the absolute linear size of a galaxy in kpc. In the {\it direct} approach the slope is determined from the linear regression of $X$ against $p$. The resulting direct Tully-Fisher relation can be expressed as \begin{equation} \label{E4} \\ E(X\vert p) = ap+b. \end{equation} Consider now the {\it observed} average of $X$ at each $p$, $\langle X\rangle_p$, as a function of the true distance. The limit in $x$ (the observational counterpart of $X$) cuts off progressively more and more of the distribution function of $X$ for a constant $p$. Assuming $X=\log D$ one finds: \begin{equation} \label{E5} \\ \langle X\rangle_p \ge E(X\vert p), \end{equation} The inequality gives a practical measure of the Malmquist bias depending primarily on $p$, $r'$, $\sigma_X$ and $x_\mathrm{lim}$. The equality holds only when the $x$-limit cuts the luminosity function $\Phi(X)$ insignificantly. That the direct relation is {\it inevitably} biased by its nature forces one either to look for an unbiased subsample or to find an appropriate correction for the bias. The former was the strategy chosen by Bottinelli et al. (\cite{Bottinelli86}) where the method of normalized distances was introduced. This is the method chosen also by the KLUN project. KLUN ({\sl Kinematics of the Loal Universe}) is based on a large sample, which consists of 5171 galaxies of Hubble types T=1-8 distributed on the whole celestial sphere (cf. e.g. Paturel \cite{Paturel94}, Theureau et al. \cite{Theureau97b}). Sandage (\cite{Sandage94a}, \cite{Sandage94b}) has also studied the latter approach. By recognizing that the Malmquist bias depends not only on the imposed $x$-limit but also on the rotational velocities and distances, he introduced the triple-entry correction method, which has consistently predicted values of $H_0$ supporting the long cosmological distance scale. As a practical example of this approach to the Malmquist bias cf. e.g. Federspiel et al. (\cite{Federspiel94}). Bottinelli et al. (\cite{Bottinelli86}) found $H_0=72\mathrm{\, km\, s^{-1}\, Mpc^{-1}}$ using the method of normalized distances, i.e. using a sample cleaned of galaxies suffering from the Malmquist bias. This value was based on the de Vaucouleurs calibrator distances. If, instead, the Sandage-Tammann calibrator distances were used Bottinelli et al. (\cite{Bottinelli86}) found $H_0=63\mathrm{\, km\, s^{-1}\, Mpc^{-1}}$ (or $H_0=56\mathrm{\, km\, s^{-1}\, Mpc^{-1}}$ if using the old ST calibration). One appreciates the debilitating effect of the Malmquist bias by noting that when it is ignored the de Vaucouleurs calibration yields much larger values: $H_0\sim100\mathrm{\, km\, s^{-1}\, Mpc^{-1}}$. Theureau et al. (\cite{Theureau97b}) by following the guidelines set out by Bottinelli et al. (\cite{Bottinelli86}) determined the value of $H_0$ using the KLUN sample. $H_0$ was determined not only using magnitudes but also diameters because the KLUN sample is constructed to be complete in angular diameters rather than magnitudes (completeness limit is estimated to be $D_{25}=1\farcm6$). Left with 400 unbiased galaxies (about ten times more than Bottinelli et al. (\cite{Bottinelli86}) were able to use) reaching up to $2000{-}3000\mathrm{\, km\, s^{-1}}$ they found using the most recent calibration based on HST observations of extragalactic cepheids \begin{itemize} \item $H_0=53.4\pm 5.0\mathrm{\, km\, s^{-1}\, Mpc^{-1}}$ from the magnitude relation, and \item $H_0=56.7\pm 4.9\mathrm{\, km\, s^{-1}\, Mpc^{-1}}$ from the diameter relation. \end{itemize} They also discussed in their Sect. 4.2 how these results change if the older calibrations were used. For example, the de Vaucouleurs calibration would increase these values by 11 \%. We expect that a similar effect would be observed also in the present case. In the present paper we ask whether the results of Theureau et al. (\cite{Theureau97b}) could be confirmed by implementing the {\it inverse} Tully-Fisher relation: \begin{equation} \label{E6} \\ p=a'X+b', \end{equation} This problem has special importance because of the ``unbiased'' nature that has often been ascribed to the inverse Tully-Fisher relation as a distance indicator and because of the large number of galaxies available contrary to the direct approach where one is constrained to the so called unbiased plateau (cf. Bottinelli et al. \cite{Bottinelli86}; Theureau et al. \cite{Theureau97b}). The fact that the inverse relation has it own particular biases has received increasing attention during the years (Fouqu\'e et al. \cite{Fouque90}, Teerikorpi \cite{Teerikorpi90}, Willick \cite{Willick91}, Teerikorpi \cite{Teerikorpi93}, Ekholm \& Teerikorpi \cite{Ekholm94}, Freudling et al. \cite{Freudling95}, Ekholm \& Teerikorpi \cite{Ekholm97}, Teerikorpi et al. \cite{Teerikorpi99} and, of course, the present paper). \section{Outlining the approach} As noted in the introduction the KLUN project approaches the problem of the determination of the value of $H_0$ using field galaxies with photometric distances. Such an approach reduces to three steps \begin{enumerate} \item construction of a relative kinematical distance scale, \item construction of a relative redshift-independent distance scale, and \item establishment of an absolute calibration. \end{enumerate} Below we comment on the first two steps. In particular we further develop the concept of a relevant inverse slope which may differ from the theoretical slope, but is still the slope to be used. The third step is addressed in Sect.~6. It is hoped that this review clarifies the methodological basis of the KLUN project and also makes the notation used more familiar. \subsection{The kinematical distance scale} The first step takes its simplest form by assuming the strictly linear Hubble law: \begin{equation} \label{E7} \\ R_\mathrm{kin} = V_\mathrm{o}/H_0' \end{equation} where $V_\mathrm{o}$ is the radial velocity inferred from the observed redshifts and $H_0'$ is some input value for the Hubble constant. Because $V_\mathrm{o}$ reflects the true kinematical distance $R_\mathrm{kin}^*$ via the true Hubble constant $H_0^*$ \begin{equation} \label{E8} \\ R_\mathrm{kin}^* = V_\mathrm{o}/H_0^*, \end{equation} one recognizes that Eq.~\ref{E7} sets up a {\it relative} distance scale: \begin{equation} \label{E9} \\ d_\mathrm{kin} = \frac {R_\mathrm{kin}}{R_\mathrm{kin}^*} = \frac {H_0^*}{H_0'}. \end{equation} In other words, $\log d_\mathrm{kin}$ is known next to a constant. In a more realistic case one ought to consider also the {\it peculiar} velocity field. In KLUN one assumes that peculiar velocities are governed mainly by the Virgo supercluster. In KLUN the kinematical distances are inferred from $V_\mathrm{o}$'s by implementing the spherically symmetric model of Peebles (\cite{Peebles76}) valid in the linear regime. In the adopted form of this model (for the equations to be solved cf. e.g. Bottinelli et al. \cite{Bottinelli86}, Ekholm \cite{Ekholm96}) the centre of the peculiar velocity field is marked by the pair of giant ellipticals M86/87 positioned at some unknown true distance $R^*$ which is used to normalize the kinematical distance scale: the centre is at a distance $d_\mathrm{kin}=1$. The required cosmological velocities $V_\mathrm{cor}$ (observed velocities corrected for peculiar motions) are calculated as \begin{equation} \label{E10} \\ V_\mathrm{cor} = C_1\times d_\mathrm{kin}, \end{equation} where the constant $C_1$ defines the linear recession velocity of the centre of the system assumed to be at rest with respect to the quiescent Hubble flow: \begin{equation} \label{E11} \\ C_1 = V_\mathrm{o}(\mathrm{Vir})+ V_\mathrm{inf}^\mathrm{LG}. \end{equation} $V_\mathrm{o}(\mathrm{Vir})$ is the presumed velocity of the centre and $V_\mathrm{inf}^\mathrm{LG}$ is the presumed infall velocity of the Local Group into the centre of the system. \subsection{The redshift-independent distances} The direct Tully-Fisher relation is quite sensitivite to the sampling of the luminosity function. On the other hand, when implementing the inverse Tully-Fisher relation (Eq.~\ref{E6}) under ideal conditions it does not matter how we sample $X$ (Schechter \cite{Schechter80}) in order to obtain an unbiased estimate for the inverse parameters and, furthermore, the expectation value $E(r\vert r')$ is also unbiased (Teerikorpi \cite{Teerikorpi84}). {\it However, we should sample all $\log V_\mathrm{max}$ for each constant {\it true} $X$ in the sample}. This theoretical prerequisition is often tacitly assumed in practice. For more formal treatments on the inverse relation cf. Teerikorpi(\cite{Teerikorpi84}, \cite{Teerikorpi90}, \cite{Teerikorpi97}) and e.g. Hendry \& Simmons (\cite{Hendry94}) or Rauzy \& Triay (\cite{Rauzy96}). In the inverse approach the distance indicator is \begin{equation} \label{E12} \\ X = A'\langle p\rangle_X+\mathrm{cst.}, \end{equation} where $A'=1/a'$ following the notation adopted by Ekholm \& Teerikorpi (\cite{Ekholm97}; hereafter ET97). The inverse regression slope $a'$ is expected to fulfill \begin{equation} \label{E13} \\ \langle p\rangle_X \equiv E(p\vert X)=a'X+\mathrm{cst}. \end{equation} $\langle p\rangle_X$ is the observed average $p$ for a given $X$. Eq.~\ref{E13} tells that in order to find the correct $a'$ one must sample the distribution function $\phi_X(p)$ in such a way that $\langle p\rangle_X=(p_0)_X$, where $(p_0)_X$ is the central value of the underlying distribution function. $\phi_X(p)$ is presumed to be {\it symmetric} about $(p_0)_X$ for all $X$. ET97 demonstrated how under these ideal conditions the derived $\log H_0$ as a function of the kinematical distance should run horizontally as the adopted slope approaches the ideal, theoretical slope. In practice the parameters involved are subject to uncertainties, in which case one should use instead of the unknown theoretical slope a slope which we call the {\it relevant} inverse slope. We would like to clarify in accurate terms the meaning of this slope which differs from the theoretical slope and which has been more heuristically discussed by Teerikorpi et al. (\cite{Teerikorpi99}). The difference between the theoretical and the relevant slope can be expressed in the following formal way. Define the observed parameters as \begin{equation} \label{E14} \\ X_\mathrm{o}=X+\epsilon_x+\epsilon_\mathrm{kin}, \end{equation} \begin{equation} \label{E15} \\ p_\mathrm{o}=p+\epsilon_p, \end{equation} where $X$ is inferred from $x$ with a measurement error $\epsilon_x$ and the kinematical distance $d_\mathrm{kin}$ has an error $\epsilon_\mathrm{kin}$ due to uncertainties in the kinematical distance scale. $\epsilon_p$ is the observational error on $p$. The theoretical slope $a'_\mathrm{t}$ is\footnote {We make use of the formal definition of the slope of the linear regression of $y$ against $x$ with \begin{equation} \label{E16} \\ \mathrm{Cov}(x,y) = \frac{\sum(x-\langle x\rangle)(y-\langle y\rangle)}{(N-1)}. \end{equation} } \begin{equation} \label{E17} \\ a'_\mathrm{t}= \frac {\mathrm{Cov}(X,p)}{\mathrm{Cov}(X,X)}, \end{equation} while the observed slope is \begin{equation} \label{E18} \\ a'_\mathrm{o} = \frac {\mathrm{Cov}(X_\mathrm{o},p_\mathrm{o})}{\mathrm{Cov} (X_\mathrm{o},X_\mathrm{o})} \sim \frac {\mathrm{Cov}(X,p)+\mathrm{Cov}(\epsilon_x+\epsilon_\mathrm{kin}, \epsilon_p)} {\mathrm{Cov}(X,X)+\sigma^2_x+\sigma^2_\mathrm{kin}} \end{equation} We call the slope $a'_\mathrm{o}$ relevant if it verifies for {\it all} $X_\mathrm{o}$ (Eq.~\ref{E13}) \begin{equation} \label{E19} \\ \langle p_\mathrm{o}\rangle_{X_\mathrm{o}}=E(p\vert X_\mathrm{o})= a'_\mathrm{o}X_\mathrm{o}+\mathrm{cst}. \end{equation} This definition means that the average observed value of $p_\mathrm{o}$ at each fixed value of $X_\mathrm{o}$ (derived from observations and the kinematical distance scale) is correctly predicted by Eq.~\ref{E19}. Note also that in the case of diameter relation, $\epsilon_x$, $\epsilon_\mathrm{kin}$ and $\epsilon_p$ are only weakly correlated. Thus the difference between the relevant slope and the theoretical slope is dominated by $\sigma_x^2+\sigma_\mathrm{kin}^2$. In the special case where the galaxies are in one cluster (i.e. at the same true distance), the dispersion $\sigma_\mathrm{kin}$ vanishes. In order to make the relevant slope more tangible we demonstrate in Appendix A how it indeed is the one to be used for the determination of $H_0$. Finally, also selection in $p$ and type effect may affect the derived slope making it even shallower. Theureau et al. (\cite{Theureau97a}) showed that a type effect exists seen as degenerate values of $p$ for each constant linear diameter $X$. Early Hubble types rotate faster than late types. In addition, based on an observational program of 2700 galaxies with the Nan\c{c}ay radiotelescope, Theureau (\cite{Theureau98}) warned that the detection rate in HI varies continuosly from early to late types and that on average $\sim 10\%$ of the objects remain unsuccessfully observed. Influence of such a selection, which concerns principally the extreme values of the distribution function $\phi(p)$, was discussed analytically by Teerikorpi et al. (\cite{Teerikorpi99}). \section{A straightforward derivation of $\log H_0$} \subsection{The sample} KLUN sample is -- according to Theureau et al. (\cite{Theureau97b}) -- complete up to $B_\mathrm{T}^\mathrm{c}=13\fm25$, where $B_\mathrm{T}^\mathrm{c}$ is the corrected total B-band magnitude and down to $\log D_{25}^\mathrm{c}=1.2$, where $D_{25}^\mathrm{c}$ is the corrected angular B-band diameter. The KLUN sample was subjected to exclusion of low-latitude ($\vert b\vert\ge15\degr$) and face-on ($\log R_{25}\ge0.07$) galaxies. The centre of the spherically symmetric peculiar velocity field was positioned at $l=284\degr$ and $b=74\degr$. The constant $C_1$ needed in Eq.~\ref{E10} for cosmological velocities was chosen to be $1200\mathrm{\, km\, s^{-1}}$ with $V_\mathrm{o}(\mathrm{Vir})=980\mathrm{\, km\, s^{-1}}$ and $V_\mathrm{inf}^\mathrm{LG}=220\mathrm{\, km\, s^{-1}}$ (cf. Eq.~\ref{E11}). After the exclusion of triple-valued solutions to the Peebles' model and when the photometric completeness limits cited were imposed on the remaining sample one was left with 1713 galaxies for the magnitude sample and with 2822 galaxies for the diameter sample. \subsection{The inverse slopes and calibration of zero-points} Theureau et al. (\cite{Theureau97a}) derived a common inverse diameter slope $a'\approx0.50$ and inverse magnitude slope $a'\approx-0.10$ for all Hubble types considered i.e. T=1-8. These slopes were also shown to obey a a simple mass-luminosity model (cf. Theureau et al. \cite{Theureau97a}). \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f1}} \caption{ The slope $a'=0.50$ forced to the calibrator sample with Cepheid distances yielding $b'_{\mathrm{cal}}=1.450$, when no type corrections were made. } \label{F1} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f2}} \caption{ The slope $a'=-0.10$ forced to the calibrator sample with Cepheid distances yielding $b'_\mathrm{cal}=0.117$, when no type corrections were made. } \label{F2} \end{figure} With these estimates for the inverse slope the relation can be calibrated. At this point of derivation we ignore the effects of type-dependence and possible selection in $\log V_\mathrm{max}$. The calibration was done by forcing the slope to the calibrator sample of 15 field galaxies with cepheid distances, mostly from the HST programs (Theureau et al. \cite{Theureau97b}, cf. their Table~1.). The absolute zero-point is given by \begin{equation} \label{E20} \\ b'_\mathrm{cal}=\frac {\sum {(\log V_\mathrm{max}-a'X)}}{N_\mathrm{cal}}, \end{equation} where the adopted inverse slope $a'=0.50$ yields $b'_\mathrm{cal}=1.450$ and $a'=-0.10$ $b'_\mathrm{cal}=0.117$. In Fig.~\ref{F1} we show the calibration for the diameter relation and in Fig.~\ref{F2} for the magnitude relation. \subsection{$H_0$ without type corrections} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,123][460,507]{8156.f3}} \caption{Panel (a): The $\log H_0$ vs. $V_\mathrm{cor}$ diagram for the calibrated inverse Tully-Fisher relation $\log V_\mathrm{max}= 0.50\log D+1.450$. The horizontal solid line corresponds to the average value $\langle\log H_0\rangle=1.92$. Panel (b): the average values $\langle\log H_0\rangle$ (circles) are shown as well as the average of the whole sample. The averages were calculated for velocity bins of size $1000\mathrm{\ km\,s^{-1}}$. Total number of points used was $N=2822$.} \label{F3} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,123][460,507]{8156.f4}} \caption{The sample imposed to the strict magnitude limit $B_\mathrm{T}^\mathrm{c}=13\fm25$ (N=1713). The forced solution yields $\langle\log H_0\rangle=1.857$ or $H_0=71.9\mathrm{\, km\, s^{-1}\, Mpc^{-1}}$. } \label{F4} \end{figure} ET97 discussed in some detail problems which hamper the determination of the Hubble constant $H_0$ when one applies the inverse Tully-Fisher relation. They concluded that once the relevant inverse slope is found, the average $\langle\log H_0\rangle$ shows no tendencies as a function of the distance. Or, in terms of the method of normalized distances of Bottinelli et. al. (\cite{Bottinelli86}), the unbiased plateau extends to all distances. ET97 also noted how one might simultaneously fine-tune the inverse slope and get an unbiased estimate for $\log H_0$. The resulting $\log H_0$ vs. kinematical distance diagrams for the inverse diameter relation is given in Fig.~\ref{F3} and for the magnitude relation in Fig.~\ref{F4}. Application of the parameters given in the previous section yield $\langle\log H_0\rangle=1.92$ correponding to $H_0=83.2\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ for the diameter sample and $\langle\log H_0\rangle=1.857$ or $H_0=71.9\mathrm{\, km\, s^{-1}\, Mpc^{-1}}$ for the magnitude sample. These averages are shown as horizontal, solid straight lines. In panels (a) individual points are plotted and in panels (b) the averages for bins of $1000\mathrm{\, km\, s^{-1}}$ are given as circles. Consider first the diameter relation. One clearly sees how the average follows a horizontal line up to $9000\mathrm{\ km\,s^{-1}}$. At larger distances, the observed behaviour of $\langle H_0\rangle$ probably reflects some selection in $\log V_\mathrm{max}$ in the sense that there is an upper cut-off value for $\log V_\mathrm{max}$. Note also the mild downward tendency between $1000\mathrm{\, km\, s^{-1}}$ and $5000\mathrm{\, km\, s^{-1}}$. Comparison of Fig.~\ref{F4} with Fig.~\ref{F3} shows how $\langle\log H_0\rangle$ from magnitudes and diameters follow each other quite well as expected (ignoring, of course, the vertical shift in the averages). Note how the growing tendency of $\langle\log H_0\rangle$ beyond $9000\mathrm{\, km\, s^{-1}}$ is absent in the magnitude sample because of the limiting magnitude: the sample is less deep. This suggests that the possible selection bias in $\log V_\mathrm{max}$ does not affect the magnitude sample. One might, by the face-value, be content with the slopes adopted as well as with the derived value of $H_0$. The observed behaviour is what ET97 argued to be the prerequisite for an unbiased estimate for the Hubble constant: non-horizontal trends disappear. It is -- however -- rather disturbing to note that the values of $H_0$ obtained via this straightforward application of the inverse relation are significantly {\it larger} than those reported by Theureau et al. (\cite{Theureau97b}). The inverse diameter relation predicts some 50 percent larger value and the magnitude relation some 30 percent larger value than the corresponding direct relations. In what follows, we try to understand this discrepancy. \section{Is there selection in $\log V_\mathrm{max}$?} The first explanation coming to mind is that the apparently wellbehaving slope $a'= 0.5$ ($a'=-0.1$) is incorrect because of some selection effect and is thus {\it not} relevant in the sense discussed in Sect.~2.2 and in Appendix A. The relevant slope brings about an unbiased estimate for the Hubble parameter (or the Hubble constant if one possesses an ideal calibrator sample) {\it if} the distribution function of $\log V_\mathrm{max}$, $\phi(p)_X$, is completely and correctly sampled for each $X$. Fig.~\ref{F3} showed some preliminary indications that this may not be the case as regards the diameter sample. Teerikorpi (\cite{Teerikorpi99}) discussed the effect and significance of a strict upper and/or lower cut-off on $\phi(p)_X$. For example, an upper cut-off in $\phi(p)_X$ should yield a too large value of $H_0$ and, furthermore, a too shallow slope. Their analytical calculations given the gaussianity of $\phi(p)_X$ show that this kind of selection effect has only a minuscule affect unless the cut-offs are considerable. Because the selection does not seem to be significant, we do not expect much improvement in $H_0$. \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f5}} \caption{The inverse Tully-Fisher diagram for the sample used in the analysis. The solid line refers to a linear regression of $a'=0.576$ and $b'=1.256$. The dashed lines give the forced solutions with $a'=0.50$ for Hubble types 1 with $b'=1.448$ and 8 with $b'=1.209$ The dotted lines at $\log V_\mathrm{max}=2.55$ and $\log V_\mathrm{max}=1.675$ are intended to guide the eye. At least the upper cut-off is quite conspicuous.} \label{F5} \end{figure} There is, however, another effect which may alter the slope. As mentioned in Sect.~2.2 the type-dependence of the zero-point should be taken into account. Because the selection function may depend on the morphological type it also affects the type corrections. This is clearly seen when one considers how the type corrections are actually calculated. As in Theureau et al. (\cite{Theureau97b}) galaxies are shifted to a common Hubble type 6 by applying a correction term $\Delta b' = b'(T)-b'(6)$ to individual $\log V_\mathrm{max}$ values, where \begin{equation} \label{E21} \\ b'(T) = \langle \log V_\mathrm{max}\rangle_T-a'\langle X\rangle_T. \end{equation} Different morphological types do not have identical spatial occupation, which is shown in Fig.~\ref{F5} for Hubble types 1 and 8 as dashed lines corresponding to forced solutions using the common slope $a'=0.5$. The strict upper and lower cut-offs would influence the extreme types more. Hence we must first more carefully see if the samples suffer from selection in $\log V_\mathrm{max}$ The inverse Tully-Fisher diagram for the diameter sample is given in Fig.~\ref{F5}. The least squares fit ($a'=0.576$, $b'=1.259$) is shown as a solid line. One finds evidence for both an upper and lower cut-off in the $\log V_\mathrm{max}$-distribution, the former being quite conspicuous. The dotted lines are positioned at $\log V_\mathrm{max}=2.55$ and $\log V_\mathrm{max}=1.675$ to guide the eye. Fig.~\ref{F5} hints that the slope $a'=0.5$ adopted in Sect.~3 may not be impeccable and thus questions the validity of the ``na\"{\i}ve" derivation of $H_0$ at least in the case of the diameter sample. \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f6}} \caption{The differential behaviour of $\langle\log H_0\rangle$ as a function of the normalized distances.The inverse parameters were $a'=0.5$ and $b'=1.450$.} \label{F6} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f7}} \caption{A straightforward linear regression applied to the calibrator sample yielding $a'= 0.749$ and $b'= 1.101$.} \label{F7} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f8}} \caption{ As Fig.~\ref{F6}, but now the parameters $a'= 0.749$ and $b'= 1.101$ were used. One can see how the downward tendency between $\log d_\mathrm{norm}\sim1.45$ and $\log d_\mathrm{norm}\sim2$ has disappeared. Also cf. Fig. 2 in Teerikorpi et al. (\cite{Teerikorpi99}).} \label{F8} \end{figure} In the case of diameter samples, Teerikorpi et al. (\cite{Teerikorpi99}) discussed how the cut-offs should demonstrate themselves in a $\log H_0$ vs. $\log d_\mathrm{norm}$ diagram, where $\log d_\mathrm{norm}=\log D_{25}+\log d_\mathrm{kin}$, which in fact is the log of $D_\mathrm{linear}$ next to a constant. We call $d_\mathrm{norm}$ ``normalized" in analogy to the method of normalized distances, where the kinematical distances were normalized in order to reveal the underlying bias. That is exactly what is done also here. Consider the differential behaviour of $\langle\log H_0\rangle$ as a function of the normalized distance. Differential average $\langle\log H_0\rangle$ was calculated as follows. The abscissa was divided into intervals of 0.01 starting at minimum $\log d_\mathrm{norm}$ in the sample. If a bin contained at least 5 galaxies the average was calculated. In Fig.~\ref{F6}. the inverse parameters $a'=0.5$ and $b'=1.450$ were used. It is seen that around $\log d_\mathrm{n}\sim 2$ the values of $\log H_0$ have a turning point as well as at $\log d_\mathrm{n}\sim 1.45$. The most striking feature is -- however -- the general decreasing tendency of $\log H_0$ between these two points. Now, according to ET97, a downward tendency of $\log H_0$ as a function of distance corresponds to $A/A'>1$, i.e. the adopted slope A is too shallow ($A'$ is the relevant slope). Closer inspection of Fig.~\ref{F1} shows that a steeper slope might provide a better fit to the calibrator sample. One is thus tempted to ask what happens if one adopts for the field sample the slope giving the best fit to the calibrator sample. As such solution we adopt the straightforward linear regression yielding $a'= 0.749$ and $b'= 1.101$ shown in Fig.~\ref{F7}. It is interesting to note that when these parameters are used the downward tendency between $\log d_\mathrm{norm}\sim1.45$ and $\log d_\mathrm{norm}\sim2$ disappears as can be seen in Fig.~\ref{F8}. From hereon we refer to this interval as the ``unbiased inverse plateau". The value of $\log H_0$ in this plateau is still rather high. \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f9}} \caption{ The differential $\langle \log H_0\rangle$ vs. $\mu_\mathrm{norm}$ diagram. One finds no indication of a selection in $\log V_\mathrm{max}$. The adopted slope ($a'=-0.10$) appears to be incorrect. } \label{F9} \end{figure} In the case of the magnitude sample we study the behaviour of the differential average $\langle \log H_0\rangle$ as a function of a ``normalized" distance modulus: \begin{equation} \label{E22} \\ \mu_\mathrm{norm} = B_\mathrm{T}^\mathrm{c}-5\log d_\mathrm{kin}. \end{equation} The $\mu_\mathrm{norm}$ axis was divided into intervals of 0.05 and again, if in a bin is more than five points the average is calculated. As suspected in the view of Fig.~\ref{E4}., Fig.~\ref{E9} reveals no significant indications of a selection in $\log V_\mathrm{max}$. The points follow quite well the straight line also shown. The line however is tilted telling us that the input slope $a'=-0.10$ may not be the relevant one. As already noted the type corrections may have some influence on the slopes. In the next section we derive the appropriate type corrections for the zero-points using galaxies residing in the unbiased plateau ($\log d_\mathrm{norm} \in [1.45,2.0]$) for the diameter sample and for the whole magnitude sample and rederive the slopes. \section{Type corrections and the value of $H_0$} The zero-points needed for the type corrections are calculated using Eq.~\ref{E21}. It was pointed out in Sect.~2.2 that $\log H_0$ should run horizontally in order to find an unbiased estimate for $H_0$. In this section we look for such an slope. Because the type-corrections depend on the adopted slope, this fine-tuning of the slope must be carried out in an iterative manner. This process consists of finding the type corrections $\Delta b'(\mathrm{T})$ for each test slope $a'$. Corrections are made for both the field and calibrator samples. The process is repeated until a horizontal $\langle\log H_0\rangle$ run is found. Consider first the diameter sample. When the criteria for the unbiased inverse plateau were imposed on the sample, 2142 galaxies were left. For this subsample the iteration yielded $a'=0.54$ (the straight line in Fig.~\ref{F10} is the least squares fit with a slope 0.003) and when the corresponding type corrections given in Table~1 were applied to the calibrator sample and the slope forced to it one found $b'_\mathrm{cal}(6)=1.325$. The result is shown in Fig.~\ref{F10}. The given inverse parameters predict an average $\langle\log H_0\rangle=1.897$ (or $H_0=78.9\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$). We treated the magnitude sample of 1713 galaxies in a similar fashion. The resulting best fit is shown in Fig.~\ref{F11}. The relevant slope is $a'=-0.115$ (the least squares fit yields a slope 0.0004). The corresponding type corrections are given in Table~1. The forced calibration gives $b'_\mathrm{cal}(6)=-0.235$. From this sample we find an average $\langle\log H_0\rangle=1.869$ (or $H_0=72.4\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$). In both cases the inverse estimates for the Hubble constant ($H_0\approx80$ for the diameter relation and $H_0\approx70$ for the magnitude relation) are considerably larger than the corresponding estimates using the direct Tully-Fisher relation ($H_0\approx55$). \begin{table} \label{T1} \begin{center} \begin{tabular}{ccc} \hline $\Delta b'(T)$&$a'=0.54$&$a'=-0.115$\\ \hline $\Delta b'(1)$&0.125&0.110\\ $\Delta b'(2)$&0.156&0.124\\ $\Delta b'(3)$&0.129&0.096\\ $\Delta b'(4)$&0.095&0.058\\ $\Delta b'(5)$&0.069&0.030\\ $\Delta b'(6)$&0.0&0.0\\ $\Delta b'(7)$&-0.054&-0.042\\ $\Delta b'(8)$&-0.118&-0.075\\ \hline \end{tabular} \end{center} \caption{The type corrections required for the relevant slopes $a'=0.54$ for the unbiased diameter sample and $a'=-0.115$ for the magnitude sample.} \end{table} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f10}} \caption{The differential $\langle\log H_0\rangle$ as a function of the log of normalized distance $\log d_\mathrm{norm}$ for the plateau galaxies with the adopted relation $\log V_\mathrm{max}=0.54\log D+1.325$. The solid line is the average $\log H_0=1.897$.} \label{F10} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f11}} \caption{The differential $\langle\log H_0\rangle$ as a function of the log of normalized distance modulus $\mu_\mathrm{norm}$ for the plateau galaxies with the adopted relation $\log V_\mathrm{max}=-0.115M-0.235$. The solid line is the average $\log H_0=1.869$.} \label{F11} \end{figure} \section{$H_0$ corrected for a calibrator selection bias} The values of $H_0$ from the direct and inverse relations still disagree {\it even} after we have taken into account the selection in $\log V_\mathrm{max}$, made the type corrections and used the relevant slope. There is -- however -- a serious possibility left to explain the discrepancy. {\it The calibrator sample used may not meet the theoretical requirements of the inverse relation}. In order to transform the relative distance scale into an absolute one a properly chosen sample of calibrating galaxies is needed. What does ``properly chosen" mean? Consider first the direct relation for which it is essential to possess a calibrator sample, which is volume-limited for {\it each} $p_\mathrm{cal}$. This means that for a $p_\mathrm{cal}$ one has $X_\mathrm{cal}$ which is drawn from the complete part of the gaussian distribution function $G(X;X_p,\sigma_{X_p})$, where the average $X_p=ap+b$. If $\sigma_{X_p}$ is constant for all $p$ and the direct slope $a$ has been {\it correctly} derived from the unbiased field sample, it will, when forced onto the calibrator sample, bring about the correct calibrating zero-point. As regards the calibration of the inverse relation the sample mentioned above does not necessarily guarantee a successful calibration. As pointed out by Teerikorpi et al. (\cite{Teerikorpi99}) though the calibrator sample is complete in the {\it direct} sense nothing has been said about how the $p_\mathrm{cal}$'s relate to the corresponding cosmic distribution of $p$'s from which the field sample was drawn. $\langle p\rangle_\mathrm{cal}$ should reflect the cosmic average $p_0$. If not, the relevant field slope when forced to the calibrator sample will bring about a biased estimate for $H_0$. Teerikorpi (\cite{Teerikorpi90}) already recognized that this could be a serious problem. He studied, however, clusters of galaxies where a nearby (calibrator) cluster obeys a different slope than a distant cluster. Teerikorpi et al. (\cite{Teerikorpi99}) developed the ideas further and showed how this problem may be met also when using field galaxies. The mentioned bias when using the relevant slope can be corrected for but is a rather complicated task. For the theoretical background of the ``calibrator selection bias" consult Teerikorpi et al. (\cite{Teerikorpi99}). One may -- as pointed out by Teerikorpi et al. (\cite{Teerikorpi99}) -- use instead of the relevant slope the calibrator slope which also predicts a biased estimate for $H_0$ but which can be corrected for in a rather straightforward manner. For the diameter relation the average correction term reads as \begin{equation} \label{E23} \\ \Delta\log H_0=(3-\alpha)\,\ln 10\,\sigma_D^2 \times\left[\frac{a'_\mathrm{cal}}{a'}-1\right], \end{equation} where $\sigma_D$ is the dispersion of the log linear diameter $\log D_\mathrm{linear}$ and $\alpha$ gives the radial number density gradient : $\alpha=0$ corresponds to a strictly homogeneous distribution of galaxies. For magnitudes the correction term follows from (cf. Teerikorpi \cite{Teerikorpi90}) \begin{equation} \label{E24} \\ \Delta\log H_0= 0.2\,\left[\frac{a'_\mathrm{cal}}{a'}-1\right] \times(\langle M\rangle-M_0). \end{equation} Because $\langle M\rangle-M_0$ simply reflects the classical Malmquist bias one finds: \begin{equation} \label{E25} \\ \Delta\log H_0=\frac{(3-\alpha)\ln 10}{5}\,\sigma_M^2 \times0.2\left[\frac{a'_\mathrm{cal}}{a'}-1\right], \end{equation} Note that one may use the calibrator slope and consequently the correction formulas {\it irrespective} of the nature of the calibrator sample (Teerikorpi et al. \cite{Teerikorpi99}). If the calibrator sample would meet the requirement mentioned, the value corrected with Eqs.~\ref{E23} or~\ref{E25} should equal values obtained from the relevant slopes. Furthermore, our analysis carried out so far would have yielded an unbiased estimate for $H_0$ and thus the problems would be in the direct analysis. However, if the requirement is not met one should prefer the corrective method using the calibrator slope. \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f12}} \caption{Histogram of the $\log V_\mathrm{max}$ values and the individual calibrators (labelled with stars). The vertical solid line gives the median of the plateau $\mathrm{Med}(\log V_\mathrm{max}^\mathrm{plateau})=2.10$ and the dotted line gives the median of the calibrators $\mathrm{Med}(\log V_\mathrm{max}^\mathrm{calib})=2.11$.} \label{F12} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f13}} \caption{The Kolmogorov-Smirnov test for the diameter sample. Pay attention to the rather remarkable similarity between the cumulative distribution functions (cdfs).} \label{F13} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f14}} \caption{The Kolmogorov-Smirnov test for the magnitude sample. Again the cdfs are quite similar.} \label{F14} \end{figure} \subsection{Is the calibrator sample representative?} Is the calibrator bias present in our case? Recall that the calibrators used were sampled from the nearby field to have high quality distance moduli mostly from the HST Cepheid measurements. This means that we have no a priori guarantee that the calibrator sample used will meet the criterium required. We compare the type-corrected diameter and magnitude samples with the calibrator sample. Note that for the diameter sample we use only galaxies residing in the unbiased inverse plateau (i.e. the small selection effect in $\log V_\mathrm{max}$ has been eliminated). In Fig.~\ref{F12} we show the histogram of the $\log V_\mathrm{max}$ values for the diameter sample and the individual calibrators (labelled as stars). The vertical solid line gives the median of the plateau $\mathrm{Med}(\log V_\mathrm{max}^\mathrm{plateau})=2.10$ and the dotted line gives the median of the calibrators $\mathrm{Med}(\log V_\mathrm{max}^\mathrm{calib})=2.11$. In the case of magnitudes both the field and calibrator sample have the same median (2.14). The average values for the diameter case were $\langle\log V_\mathrm{max}^\mathrm{plateau}\rangle=2.09$ and $\langle\log V_\mathrm{max}^\mathrm{calib}\rangle=2.06$, and for the magnitude case $\langle\log V_\mathrm{max}^\mathrm{mag}\rangle=2.12$ and $\langle\log V_\mathrm{max}^\mathrm{calib}\rangle=2.08$. Both the diameter and the magnitude field samples were subjected to strict limits, which means that both inevitably suffer from the classical Malmquist bias. In order to have a representative calibrator sample in the sense described, we would have expected a clear difference between the field and calibrator samples. That the statistics are very close to each other lends credence to the assumption that the calibrator selection bias is present. We also made tests using the Kolmogorov-Smirnov statistics (Figs.~\ref{F13} and~\ref{F14}). In this test a low significance level should be considered as counterevidence for a hypothesis that two samples rise from the same underlying distribution. We found relatively high significance levels (0.89 for the diameter sample and 0.3 for the magnitude sample). Neither these findings corroborate the hypothesis that the calibrator sample is drawn from the cosmic distribution and hence the use of Eqs.~\ref{E23} or~\ref{E25} is warranted. \subsection{The dispersion in $\log D_\mathrm{linear}$} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f15}} \caption{A classical Spaenhauer diagram for normalized distances vs. kinematical distances with a presumed dispersion $\sigma_X=0.28$. } \label{F15} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f16}} \caption{Comparison between average values of $\langle\log d_\mathrm{norm}\rangle$ for different kinematical distances and the theoretical prediction calculated from Eq.~\ref{E26} with $X_0^*=1.37$ and $\sigma_X=0.28$. } \label{F16} \end{figure} In order to find a working value for the dispersion in $\log D_\mathrm{linear}$, we first consider the classical Spaenhauer diagram (cf. Sandage \cite{Sandage94a}, \cite{Sandage94b}). In the Spaenhauer diagram one studies the behaviour of $X$ as a function of the redshift. If the observed redshift could be translated into the corresponding cosmological distance, then $X$ inferred from $x$ and the redshift would genuinely reflect the true size of a galaxy. In practice, the observed redshift cannot be considered as a direct indicator of the cosmological distance because of the inhomogeneity of the Local Universe. Peculiar motions should also be considered. Thus the inferred $X$ suffers from uncertainties in the underlying kinematical model. The Spaenhauer diagram as a diagnostics for the distribution function is always constrained by our knowledge of the form of the true velocity-distance law. Because the normalized distance (cf. Sect.~3.) is proportional to the linear diameter we construct the Spaenhauer diagram as $\log d_\mathrm{norm}$ vs. $\log d_\mathrm{kin}$ thus avoiding the uncertainties in the absolute distance scale. The problems with relative distance scale are -- of course -- still present. The fit shown in Fig.~\ref{F15} is not unacceptable. The dispersion used was $\sigma_X=0.28$, a value inferred from the dispersion in absolute B-band magnitudes $\sigma_M=1.4$ (Fouqu\'e et al. \cite{Fouque90}) based on the expectation that the dispersion in log linear diameter should be one fifth of that of absolute magnitudes. We also looked how the average values $\langle\log d_\mathrm{norm}\rangle$ at different kinematical distances compare to the theoretical prediction which, in a strictly limited sample of $X$'s, at each log distance is formally expressed as \begin{equation} \label{E26} \\ \langle X\rangle_\mathrm{d} = X_0^*+ \frac {2\sigma_X}{\sqrt2\pi} \frac {\exp\bigl[-(X_\mathrm{lim}-X_0^*)^2/(2\sigma_X^2)\bigr]} {\mathrm{erfc}\bigl[(X_\mathrm{lim}-X_0^*)/(\sqrt2\sigma_X)\bigr]}. \end{equation} Here $X$ refers to $\log d_\mathrm{norm}$. The curve in Fig.~\ref{F16} is based on $X_0^*=1.37$ and $\sigma_X=0.28$. The averages from the data are shown as bullets. The data points follow the theoretical prediction reasonably well. \subsection{Corrections and the value of $H_0$} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f17}} \caption{A least squares fit the type corrected calibrator sample yielding $a'=0.73$. The type correction was based on $a'=0.54$.} \label{F17} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics[211,195][460,384]{8156.f18}} \caption{A least squares fit the type corrected calibrator sample yielding $a'=-0.147$. The type correction was based on $a'=-0.115$.} \label{F18} \end{figure} Consider a strictly homogeneous universe, i.e. $\alpha=0$. In Eqs. \ref{E23} and \ref{E25} one needs values for slope $a'_\mathrm{c}$. Least squares fit to the type-corrected calibrator sample yields $a'_\mathrm{c}=0.73$ for the diameter relation and $a'_\mathrm{c}=-0.147$ for the magnitude relation. (cf. Figs. \ref{F17} and \ref{F18}). These slopes correspond to diameter zero-point $b'_\mathrm{c}(6)=1.066\pm0.103$ and to magnitude zero-point $b'_\mathrm{c}=-0.879\pm0.131$ The {\it biased} estimates for average $\log H_0$ are $\langle\log H_0\rangle = 1.910\pm0.188$ for the diameters and $\langle\log H_0\rangle = 1.876\pm0.176$ for the magnitudes. For the zero-points and the averages we have given the $1\sigma$ standard deviations. The {\it mean error} in the averages is estimated from \begin{equation} \label{E27} \\ \epsilon_{\langle\log H_0\rangle}\approx \sqrt{\frac{\sigma_{B'}^2}{N_\mathrm{cal}} +\frac{\sigma_{\log H_0}^2}{N_\mathrm{gal}}}, \end{equation} where $\sigma_{B'}=\sigma_{b'}/a'_\mathrm{cal}$ for diameters and $\sigma_{B'}=0.2\sigma_{b'}/a'_\mathrm{cal}$ for magnitudes. The use of Eq.~\ref{E27} is acceptable because the dispersion in $b'$ does not correlate with the dispersion $\log H_0$. With the given slopes and dispersions we find: \begin{itemize} \item $\langle\log H_0\rangle = 1.910\pm0.037$ for the diameters \item $\langle\log H_0\rangle = 1.876\pm0.046$ for the magnitudes. \end{itemize} Eq. \ref{E23} predicts an average correction term for the slopes $a'_\mathrm{c}=0.73$ and $a'=0.54$ together with $\sigma_X=0.28$ $\Delta\log H_0=0.191$. and Eq. \ref{E25} with $a'_\mathrm{c}=-0.147$,$a'=-0.115$ and $\sigma_M=1.4$ $\Delta\log H_0=0.151$. When applied to the above values we get the corrected, unbiased estimates \begin{itemize} \item $\langle\log H_0\rangle= 1.719\pm0.037$ for the diameters \item $\langle\log H_0\rangle = 1.725\pm0.046$ for the magnitudes. \end{itemize} These values translate into Hubble constants \begin{itemize} \item $H_0=52^{+5}_{-4}\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ for the inverse diameter B-band Tully-Fisher relation, and \item $H_0=53^{+6}_{-5}\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ for the inverse magnitude B-band Tully-Fisher relation. \end{itemize} These corrected values are in good concordance with each other as well as with the estimates established from the direct diameter Tully-Fisher relation (Theureau et al. \cite{Theureau97b}). Note that the errors in the magnitude relation are slightly larger than in the diameter relation. This is expected because for the diameter relation we possess more galaxies. The error is however mainly governed by the uncertainty in the calibrated zero-point. This is expected because though the dispersion in inverse relation as such is large it is compensated by the number galaxies available. Finally, how significant an error do the correction formulae induce? We suspect the error to mainly depend on $\alpha$. The correction above was based on the assumption of homogeneity (i.e. $\alpha=0$). Recently Teerikorpi et al. (\cite{Teerikorpi98}) found evidence that the average density radially decreases around us ($\alpha\approx0.8$) confirming the more general (fractal) analysis by Di Nella et al. (\cite{DiNella96}). Using this value of $\alpha$ we find $\Delta\log H_0=0.140$ for the diameters and $\Delta\log H_0=0.111$ for the magnitudes yielding \begin{itemize} \item $\langle\log H_0\rangle = 1.770\pm0.037$ for the diameters \item $\langle\log H_0\rangle = 1.765\pm0.046$ for the magnitudes. \end{itemize} In terms of the Hubble constant we find \begin{itemize} \item $H_0=59^{+5}_{-4}\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ for the inverse diameter B-band Tully-Fisher relation, and \item $H_0=58^{+6}_{-5}\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ for the inverse magnitude B-band Tully-Fisher relation. \end{itemize} \section{Summary} In the present paper we have examined how to apply the inverse Tully-Fisher relation to the problem of determining the value of the Hubble constant, $H_0$, in the practical context of the large galaxy sample KLUN. We found out that the implementation of the inverse relation is not as simple task as one might expect from the general considerations (in particular the quite famous result of the unbiased nature of the relation). We summarize our main results as follows. \begin{enumerate} \item A straightforward application of the inverse relation consists of finding the average Hubble ratio for each kinematical distance and tranforming the relative distance into an absolute one through calibration. The 15 calibrator galaxies used were drawn from the field with cepheid distance moduli obtained mostly from the HST observations. The inverse diameter relation predicted $H_0\approx 80\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ and the magnitude relation predicted $H_0\approx 70\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ The diameter value for $H_0$ is about 50 percent and the magnitude value about 30 percent larger than those obtained from the direct relation (cf. Theureau et al. \cite{Theureau97b}). \item We examined whether this discrepancy could be resolved in terms of some selection effect in $\log V_\mathrm{max}$ and the type dependence of the zero-points on the Hubble type. One expects these to have some influence on the derived value of $H_0$. Only a minuscule effect was observed. \item There is -- however -- a new kind of bias involved: if the $\log V_\mathrm{max}$-distribution of the calibrators does not reflect the cosmic distribution of the field sample {\it and} the relevant slope for the field galaxies differs from the calibrator slope the average value of $\log H_0$ will be biased if the relevant slope is used (Teerikorpi et al. \cite{Teerikorpi99}). \item We showed for the unbiased inverse plateau galaxies i.e. a sample without galaxies probably suffering from selection in $\log V_\mathrm{max}$, that the calibrators and the field sample obey different inverse diameter slopes, namely $a'_\mathrm{cal}=0.73$ and $a'= 0.54$, Also, the magnitude slopes differed from each other ($a'_\mathrm{cal}=-0.147$ and $a'= -0.115$). For the diameter relation we were able to use 2142 galaxies and for the magnitude relation 1713 galaxies. These sizes are significant. \item We also found evidence that the calibrator sample does {\it not} follow the cosmic distribution of $\log V_\mathrm{max}$ for the field galaxies. This means that if the relevant slopes are used a too large value for $H_0$ is found. Formally, this calibrator selection bias could be corrected for but is a complicated task. \item One may use instead of the relevant slope the calibrator slope which also brings about a biased value of $H_0$. Now, however, the correction for the bias is an easy task. Furthermore, this approach can be used irrespective of the nature of the calibrator sample and should yield an unbiased estimate for $H_0$. \item When we adopted this line of approach we found \begin{itemize} \item $H_0=52^{+5}_{-4}\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ for the inverse diameter B-band Tully-Fisher relation, and \item $H_0=53^{+6}_{-5}\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ for the inverse magnitude B-band Tully-Fisher relation \end{itemize} for a strictly homogeneous distribution of galaxies ($\alpha=0$) and \begin{itemize} \item $H_0=59^{+5}_{-4}\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ for the inverse diameter B-band Tully-Fisher relation, and \item $H_0=58^{+6}_{-5}\mathrm{\ km\,s^{-1}\,Mpc^{-1}}$ for the inverse magnitude B-band Tully-Fisher relation. \end{itemize} for a decreasing radial density gradient ($\alpha=0.8$). \end{enumerate} These values are in good concordance with each other as well as with the values established from the corresponding direct Tully-Fisher relations derived by Theureau et al. (\cite{Theureau97b}), who gave a strong case for the long cosmological distance scale consistently supported by Sandage and his collaborators. Our analysis also establishes a case supporting such a scale. It is worth noting that this is the first time when the {\it inverse} Tully-Fisher relation clearly lends credence to small values of the Hubble constant $H_0$. \begin{acknowledgements} {We have made use of data from the Lyon-Meudon extragalactic Database (LEDA) compiled by the LEDA team at the CRAL-Observatoire de Lyon (France). This work has been supported by the Academy of Finland (projects ``Cosmology in the Local Galaxy Universe'' and ``Galaxy streams and structures in the Local Universe''). T. E. would like to thank G. Paturel and his staff for hospitality during his stay at the Observatory of Lyon in May 1998. Finally, we thank the referee for useful comments and constructive criticism.} \end{acknowledgements}
\section{Introduction} Planetary nebulae (PNe) are extraordinarily useful for probing stellar evolution and cosmology. In extragalactic astronomy, the planetary-nebula luminosity function (PNLF) is one of the most accurate and reliable indicators of relative distance (see the review of Jacoby {\refit et~al.\ } 1992); in stellar astrophysics, PNe allow us to examine the physics of mass loss and the timescales of stellar evolution (e.g., the review of Iben 1995). Since young PNe are bright emission-line sources, they make ideal test particles for dynamical studies, both in the Milky Way and in external galaxies ({\refit e.g.,}\ Ciardullo, Jacoby, \& Dejonghe 1993; Amaral {\refit et~al.\ } 1996). Finally, since the chemical abundances in a PN reflect the chemistry of the interstellar medium at the time of its progenitor's formation (with some light species possibly altered subsequently by stellar nucleosynthesis), these objects can yield unique insights into galactic star-formation histories and chemical and stellar evolution (cf.~Dopita {\refit et~al.\ } 1997). Almost every interesting quantity related to PNe in the Milky Way---their space density, formation rate, Galactic distribution, sizes, ionized and total nebular masses, contribution to chemical evolution, and the luminosities and evolutionary states of their central stars---depends critically upon their distances. But, unfortunately, the distances to PNe within the Galaxy are known only poorly. It is a remarkable irony that while the PNLF can be used to derive relative distances to external galaxies to better than $\sim 10\%$ (cf.~Jacoby {\refit et~al.\ } 1992), the distances to Milky Way PNe are typically known to no better than a factor of $\sim 2$ or worse ({\refit e.g.,}\ Terzian 1993, 1997). In fact, of the $\sim 1100$ known Galactic PNe, only a dozen or so have distances that are reasonably well determined using direct methods. For the rest, it is necessary to use various statistical techniques. One fundamental, but rarely used, method for obtaining distances to Galactic PNe is through the photometric parallaxes of resolved companion stars. Perhaps two-thirds of all stars are members of binary systems, and the evidence suggests that the orbital period distribution of these binaries is a Gaussian in $\log P$ centered at $P \approx 180$~years, with a dispersion of 2.3 in the logarithm of the period (Duquennoy \& Mayor 1991). Main-sequence binaries with periods less than $\sim 1000$~days, but still wide enough for a red giant to form, will eventually evolve through a common-envelope phase and produce systems with dramatically shorter periods or even a coalesced binary (see Bond \& Livio 1990; Yungelson, Tutukov, \& Livio 1993). However, stars with larger initial periods will not interact, and their separations will actually increase with time, as a consequence of stellar mass loss. As a result, it is likely that nearly half of all planetary-nebula nuclei (PNNs) have wide binary companions. Despite this expectation, only a few PNNs are actually known to have resolved visual companions. The best-known case is the nucleus of NGC~246, which has a 14th-mag K-dwarf companion $3\farcs 8$ away (Minkowski 1960; Cudworth 1973). Fitting of this companion to the main sequence provides what is generally accepted as one of the most accurate PN distances (quoted as 430~pc by Cahn, Kaler, \& Stanghellini 1992, and revised to $495^{+145}_{-100}$~pc on the basis of CCD photometry by Bond \& Ciardullo 1999), and one that is almost always used as a primary calibrator of statistical distance methods. Other PNNs reported to have resolved companions include those of NGC 650-1, A~24, A~30, and A~33 (Cudworth 1973), NGC~3132 (Kohoutek \& Laustsen 1977), NGC~6853 (Cudworth 1973, 1977), A~63 (Krzeminski 1976), K~1-14 (Kaler 1981), and PuWe~1 (Purgathofer \& Weinberger 1980), but at the time of this survey, only the companion to NGC~3132 had photometry accurate enough to provide a reliable spectroscopic parallax (Pottasch 1980, 1984). Here we present the results of a {\sl Hubble Space Telescope\/} ({\sl HST\/}\/) snapshot survey of Galactic planetary nebulae, designed to detect and measure resolved binary PNN companions. In \S 2, we describe the survey and the photometric procedures used to measure all the stars present on our CCD frames. In \S 3, we use these data to create a list of PNNs which have visual companions deserving of follow-up observations. In \S 4, we discuss the issue of interstellar reddening, and compare measures of extinction based on two-color photometry with estimates derived from the emission lines of the nebulae. In \S 5, we transform our {\sl HST\/}\/ stellar magnitudes to the standard Landolt (1983, 1992) system, and present $V$ and $I$ magnitudes for 109 central stars. Because of the superior resolving power of the {\sl HST\/}, these magnitudes are generally better than PNN measurements made from the ground, especially for objects with $V > 15.4$. In \S 6, we discuss our candidate binaries individually, and derive distances to those systems that are most likely to be physically associated. Finally, in \S 7, we compare our distances to estimates based on statistical methods, and discuss the implications our observations have for the Galactic PN distance scale. \section{Target Selection, Observations, and Reductions} Our observations were carried out as a ``snapshot survey'' during Cycles~3 and~5 of the {\sl HST\/}\/ General Observer program. {\sl HST\/}\/ snapshots are short exposures taken during occasional gaps that remain in the observing program after as many primary scientific observations as possible have been scheduled. The targets that are observed during these opportunities are selected at random from a list of candidates provided by the observers. The PNe in our target pool were chosen using criteria designed to maximize our chances of finding resolved companion stars. First among these criteria was that the objects in our survey had to be nearby. Our target selection was heavily influenced by the list of very nearby PNe given by Terzian (1993), and nearly all of our targets have statistical distances from Cahn, Kaler, \& Stanghellini (1992; hereafter CKS) and/or Zhang (1995) that are less than $\sim 3$~kpc. Almost as important was our Galactic-latitude criterion: in order to reduce contamination from superposed field stars, most of our targets were chosen to have $|b| > 10^\circ$. A third selection criterion was known or suspected binarity of the central star. We included in our target list the nine PNNs with reported visual companions (as listed above), along with A~34, A~66, and Th~2-A, whose visual companions were noted by H.E.B. during ground-based observations. In addition we included seven central stars that have composite or late-type spectra but are unresolved from the ground, and show no evidence for being extremely close binaries. Finally, to increase the usefulness of our dataset for comparison with other distance techniques, we included six PNe with Very Large Array expansion distances (Hajian, Terzian, \& Bignell 1993, 1995; Hajian \& Terzian 1996), 15 nebulae with distance estimates based on model-atmosphere analyses of their central stars (M\'endez, Kudritzki, \& Herrero 1992), and seven PNNs known to be very close binaries (A~46, A~63, A~65, HFG~1, K~1-2, LoTr~5, and NGC~2346; cf.~Bond 1994), from which distances can potentially be obtained from light-curve solutions. In all, 144~PNNs were included in our input target lists. Our snapshot images were taken with the {\sl HST\/}\/ between 1993 and 1997. The Cycle~3 (1993) observations were obtained with the Planetary Camera of the original Wide Field and Planetary Camera (WF/PC){}. These data consisted of single exposures through the F785LP (``$I$'') filter plus occasional exposures through the F555W (``$V$'') filter, usually with the PN central star positioned near the center of the PC6 CCD{}. The Cycle~5 (1995-97) data were obtained using the F814W (``$I$'') and F555W (``$V$'') filters of the Wide Field Planetary Camera~2 (WFPC2), with the CCD gain set at 14 e$^-$/DN{}. For these observations the PNNs were centered on the Planetary Camera chip and usually imaged twice in each filter, with the second exposure typically being $\sim 2$ times longer than the first. The uneven exposure times optimized our ability to detect binaries. We generally tried to scale our shorter integrations so that the PNN's image fell just short of saturation ({\refit i.e.,}\ $\sim 50,000$~electrons in the central pixel); this insured an accurate measurement of the PNN's magnitude and color, and enabled us to search for companions with $\Delta m \lesssim 4.5$ to within $\rho\sim 0\farcs 05$ of the central star. Conversely, we attempted to scale our longer exposures so that the central pixel of the PNN could receive as much charge as possible ($\sim 100,000$~electrons) without bleeding significantly into its neighboring pixels. By doing this, we could search for companions almost $\sim 7$~mag fainter than the central star with $\rho\gtrsim 0\farcs 3$ separation. Because of the limited time available per snapshot visit, we imposed a maximum integration time of 900~s per filter. The second, deeper exposures in each filter described above were omitted if they would be longer than 500~s in F814W, or 200~s in F555W{}. Exposures longer than 600~s were split into two equal-length integrations to aid in cosmic-ray removal. Because of the uncertainties of the ground-based PNN magnitudes used to calculate our exposure times, we had anticipated that not all of our targets would be optimally exposed. In fact, however, most of our program PNNs had useful exposures, and we were able to perform a careful search for visual binaries around almost every object. There was a guide-star acquisition error for our observation of Lo~4, which resulted in our obtaining only a single F555W exposure under gyro control. A summary of the WF/PC and WFPC2 observations appears in Tables~1 and 2. Of the 144 objects in the target list, 113 were actually observed: 26 with WF/PC, 84 with WFPC2, and 3 with both. Photometric reduction of our CCD images was accomplished using a combination of IRAF and DAOPHOT/ALLSTAR (Stetson 1987). After the standard STScI pipeline de-biasing and flat-fielding, we removed charged-particle events (``cosmic rays'') from the data. For this task, the Cycle~3 and Cycle~5 frames were handled differently. Because the Cycle~3 data consisted only of single exposures through the F785LP and F555W filters, we could not simply ``stack'' these images and remove discrepant pixels. Instead we employed a semi-automatic iterative technique, which took advantage of the fact that stars on the original WF/PC frames had aberrated point-spread-functions (PSFs). First, we filtered our images using a $7 \times 7$ moving window, and masked all points that deviated by more than $3 \, \sigma$ from the median defined by their surroundings. We then compared these filtered images to their parent images, and carefully examined the cores of all stars on the frames. If the filtering algorithm had affected the cores, the original pixel values were manually inserted back into the images. In this way, the bulk of the cosmic rays were removed without affecting the stellar photometry. Reduction of the WFPC2 data proceeded in a more standard manner. All the data were first multiplied by a corrective image, in order to compensate for the geometrical distortions present in the flat-field images. We then used the DAOPHOT routine {\it find\/} to produce a list of candidate objects for each frame. These lists were culled of cosmic-ray events and other spurious detections by first rejecting all objects that fell within the vignetted areas of the frames, and then comparing the object lists derived from the individual frames of a given target. Objects found on the shorter exposures, but not on the longer ones, were discarded as cosmic rays. Photometry of the WFPC2 data was performed via a PSF fitting technique. Using all the frames of our survey, we identified $\sim 20$ bright, isolated stars in each of the four CCD camera fields, and from these data we defined the instrument's position-dependent PSF{}. Once the PSF was defined, we used it to derive preliminary instrumental magnitudes and create star-subtracted images. If any poor subtractions were found, due either to the presence of an undetected close binary, or to a poorly determined nebular background, the process was repeated with a modified star list and/or a model for the nebular emission (cf.~Ciardullo \& Bond 1996). The raw instrumental magnitudes derived by ALLSTAR were then scaled to an 11-pixel-radius aperture using an aperture correction derived from the bright stars, and corrected for the effects of finite charge-transfer efficiency following the prescription of Holtzman {\refit et~al.\ } (1995). Finally, the WFPC2 magnitudes were corrected for a non-linear term related to integration time (Casertano 1997), and put on the photometric system defined by Holtzman {{\null} et al.} Because the aberrated WF/PC images had a brighter limiting magnitude than their WFPC2 counterparts, there were, in general, fewer stars on these frames and saturation in the stellar cores was less of a problem. Thus, our photometric techniques were simpler. In most cases, photometry was accomplished by summing the flux of each star within an aperture of radius 5~pixels; only when the stellar separations were of the order of a few pixels, or when the stellar cores were saturated, were DAOPHOT's PSF fitting routines used. The small-aperture measurements were then converted to total instrumental magnitudes using aperture corrections found from bright stars, and scaled to the {\sl HST\/}\/ standard system using the ground-based WF/PC calibration of Harris {\refit et~al.\ } (1991) and the flight calibration of Hunter {\refit et~al.\ } (1992). Finally, depending on the Julian Date of the observation, an additional correction of up to $0.1$~mag was applied to the F555W magnitudes, in order to account for the WF/PC chips' loss of sensitivity with time due to contamination (cf.~Ritchie \& MacKenty 1993, 1994). \section{Detection of Binaries} Because our data contain no information about proper motion or radial velocity, our assignment of physical binarity has to rely solely on the spatial coincidence of stars. The probability of our identifying a binary PNN depends critically both on the apparent separation of the companion star, and on the field-star surface density. (In other words, in a sufficiently crowded field even a closely separated physical pair would be mistaken for an optical double.) To improve our chances of detecting visual binaries, our target sample was weighted toward objects at high galactic latitude and toward nearby objects, as described above. Still, because the surface density of bright stars is much lower than for faint stars, our analysis is significantly more sensitive to brighter companions than to fainter ones. To estimate whether a given apparent companion star located near a PNN is physically associated with it, we calculate the Poissonian probability $P$ that a random field star as bright as, or brighter than, the companion would be projected by chance within a radius $\rho$ of the PNN{}. Mathematically, \begin{equation} P = 1 - \left(1 - {\pi \rho^2 \over A} \right)^N, \end{equation} where $A$ is the total sky area surveyed by each WF/PC or WFPC2 frame, and $N$ is the number of stars within this area that are at least as bright as the companion. $P$ is a function of the local stellar surface density, which we determined directly from star counts on the frames. If true physical binaries are common in our sample, there should be an excess of pairs with very low values of $P${}. (It was, of course, just such an argument, first made by Michell 1767, that established the existence of physical double stars.) In Figure 1 we plot the distribution of $P$ for our sample of PNNs, using the stellar neighbor of each central star that has the lowest value of $P$ (i.e., the highest probability of being physically associated; generally, but not always, this is also the nearest neighbor of the PNN){}. If all of our frames were populated entirely by randomly distributed field stars, without any true physical pairs, the distribution would be flat; instead, Figure~1 shows a large excess of companions with very small chance probabilities, particularly for $P\le0.05$. This is a clear demonstration that we are detecting true resolved physical doubles. We therefore chose all of the apparent binaries with a chance probability of 5\% or less for further investigation, and list them in Table~3. Most of the columns are self-explanatory. Columns~2 and 3 give the separation, $\rho$, between the PNN and the candidate companion, and the J2000 position angle of the companion with respect to the central star. Column~8 presents the probability $P$ of a chance projection, as calculated from equation~(1). To give an indication of the uncertainty in the probability, we give in the final column of Table~3 the value of $P$ calculated by increasing the local stellar density by $1 \, \sigma$ above that deduced directly from the star counts. Table~3 is {\it not\/} necessarily a list of true binary PNNs: it is merely the complete set of objects that, formally, have more than a 95\% probability of being a physical pair. Since we imaged over a hundred PNNs in the course of this survey, we can expect $\sim 6$ chance superpositions to be contained in the list. In \S 6 we will examine the properties of the PNNs in question and attempt to select those systems that are most likely to be true physical pairs. Returning to Figure~1, we note that in addition to the sharp peak at small probabilities, there is a slight excess of objects with chance probabilities between 0.05 and 0.40. This suggests that a few additional PNNs not included in Table~3 also have physical companions. Unfortunately, because the ratio of true binaries to chance superpositions in this probability range is low, the only way to identify these objects would be through proper-motion and/or radial-velocity measurements of the candidate companions. \section{Interstellar Extinction} The first step in deriving distances from two-color photometry is to obtain an estimate of the amount of interstellar extinction suffered by each object. This can be done in two ways. The first method is to use the observed emission ratios ({\refit i.e.,}\ the relative intensities of H$\alpha$, H$\beta$, and the radio continuum) in the PNNs' surrounding nebulae. The intrinsic {H$\alpha$\ } to {H$\beta$\ } ratio of a typical $10,000$~K nebula is $\sim 2.86$ ({\refit e.g.,}\ Brocklehurst 1971); by comparing this value to the observed line ratio, it is possible to compute the logarithmic extinction at H$\beta$, denoted $c$. Similarly, an estimate of $c$ can be obtained by comparing the nebula's radio continuum (measured at 5~GHz) to its total emission at H$\beta$ (cf.~Milne \& Aller 1975). Measures of $c$ exist in the literature for a large number of objects, and a useful compilation appears in CKS\null. Unfortunately, because most PNe exhibit complex multi-zone structures, the exact values to use for the intrinsic H$\alpha$/{H$\beta$\ } and H$\beta$/radio ratios can be difficult to determine; in addition, observational errors introduced by atmospheric dispersion, background confusion (for radio-faint PNe), and uncertain emission-line photometry (for large and low-surface-brightness PNe) can lead one astray. Finally, as our {\sl HST\/} images of NGC~7027 and IC~4406 in Figure~2 show, the dust within a PN can be extremely patchy. Thus adoption of a global extinction value for the particular line of sight to the central star (or to its companion) may not be appropriate. The second method of determining the extinction to a PNN is to compare the observed color of the central star with that expected from a hot source. Simulations using the WFPC2 efficiency curves (Biretta {\refit et~al.\ } 1996) demonstrate that a star whose optical continuum falls on the Rayleigh-Jeans tail of the Planck function should have an F555W$-$F814W color of $\sim -0.4$. As the histogram of Figure~3 illustrates, this value is confirmed in our survey: the distribution of PNN colors observed with WFPC2 has a hard blue limit of ${\rm F555W} - {\rm F814W} \sim -0.4$, and a long tail to the red caused by interstellar (and/or circumstellar) reddening. Values for the extinction of each PNN can therefore be obtained directly from our measured colors. Of course, some of the same uncertainties that affect the nebula-based extinction ratios apply here: spectroscopy demonstrates that there are real departures from Rayleigh-Jeans continua in some objects ({\refit e.g.,}\ from Wolf-Rayet emission lines; see, for example, Smith \& Aller 1969; Heap 1982; M\'endez {\refit et~al.\ } 1986). Moreover, some PNNs have intrinsically red colors (either from an unresolved late-type companion, or because they themselves have evolved back to a ``born-again'' red-giant phase). Finally, even if all central stars were identical, small-scale non-uniformities in the dust distribution (such as seen in Figure~2) could cause a companion star to have a different reddening value than the hot component. Figure~4 compares the nebular extinction measurements (taken from CKS and Kingsburgh \& Barlow 1994) with reddenings derived from our WFPC2 photometry using the assumption that the intrinsic ${\rm F555W} - {\rm F814W}$ color of the central star is $-0.40$. For the figure, $E$(F555W$-$F814W) has been transformed to $E(B$$-$$V)$ using the extinction table for an O6 star given by Holtzman {\refit et~al.\ } (1995). In the figure, M~2-9 and NGC~2022 have not been plotted, since both of their nuclei are marginally non-stellar on our {\sl HST\/}\ frames, and are probably surrounded by thick circumstellar dust. In addition, PNNs with known composite or late-type spectra have been omitted from the diagram. In those cases where the PN has both a radio-flux and Balmer-decrement value for $c$, we have used the mean of the two measurements (except for Mz~2, whose Balmer-decrement extinction of zero is clearly unphysical given the object's other properties). The line in Figure~4 plots the relation between $E(B-V)$ and $c$ derived by reddening a 100,000~K Planck curve with the Cardelli, Clayton, \& Mathis (1989) extinction law, folding the resultant energy distribution through the F555W and F814W system response curves, and transforming from $E(\rm F555W-F814W)$ to $E(B-V)$ using the table of Holtzman {\refit et~al.\ } (1995). This relation has a slight curvature, due to shifts in effective filter wavelength with increasing extinction ({\refit e.g.,}\ Kaler \& Lutz 1985). An excellent approximation to this curve is \begin{equation} E({\rm F555W-F814W}) = 0.913 \, c - 0.012 \, c^2. \end{equation} As can be seen in Figure~4, this law fits the data reasonably well in the mean, as the average difference between the two reddening estimators is $\Delta E(B$$-$$V) = 0.00 \pm 0.02$. However, it is also obvious that there is a substantial amount of scatter; the dispersion about the relation is $\sigma_{E(B-V)} = 0.16$~mag. Interestingly, this scatter is not confined to highly reddened PNe: even for those objects with $c < 0.2$ the scatter is still substantial, $\sigma_{E(B-V)} = 0.13$. Without additional information, it is impossible to determine whether the scatter in Figure~4 is due principally to errors in the nebular extinction values or uncertainties associated with our PNN colors. In some individual cases, we can test whether the nebular extinction is reasonable by computing the dereddened color of the central star: if the application of $c$ results in a color that is significantly bluer than the Rayleigh-Jeans limit, then the extinction has obviously been overestimated. However, from the data at hand, there is no reason to believe that one extinction estimate is better than the other. For simplicity, we therefore assume that both methods have similar uncertainties, of order $\sigma_{E(V-I)} \simeq 0.11$~mag. If a planetary nebula has both a color-estimated reddening and a reasonable value for the nebular reddening, then we use the mean of these two numbers, and assign an uncertainty of $\sigma_{E(V-I)} = 0.11 / \sqrt{2} = 0.08$~mag. If no measurements of the nebula exist, if the value of $c$ leads to an implausible color for the central star, or if the PNN is known to have a composite or late-type spectrum (or poor photometry), then we use a reddening based solely on the remaining valid method, and assume $\sigma_{E(V-I)} = 0.11$~mag. \section{Transformation to $V$ and $I$} The final step before deriving PNN distances from binaries is to transform the photometric systems defined by the WF/PC and WFPC2 filters to the Johnson-Kron-Cousins $V$ and $I$ system as defined by the standard stars of Landolt (1983, 1992). For the WFPC2 data, this was done using the equations given by Holtzman {\refit et~al.\ } (1995), which should be good to $\sim 2\%$ for colors in the range $-0.3 < V - I < 1.5$. Note that the bluest of our PNNs actually lie slightly outside this range at $V$$-$$I \approx -0.4$, and the transformations for these stars are not formally applicable. However, as the Holtzman {\refit et~al.\ } data for the extremely blue stars of $\omega$~Cen and the blue spectrophotometric standards Grw~$+70^\circ$5824, Feige~110, and AGK~$+81^\circ$266 show, the equations relating F555W and F814W to $V$ and $I$ are well behaved at the blue end, and produce residuals that are no worse than those for stars of moderate color. Hence our mild extrapolation for the central stars should be valid. Transformation equations are also available for the original WF/PC filter set (cf.~Harris {\refit et~al.\ } 1991; Saha {\refit et~al.\ } 1994). However, these relations are only defined for stars with $B$$-$$V$ colors between 0.0 and 1.6. Since many of our nuclei are much bluer than this, we chose not to use these equations directly. Instead, we re-derived the transformations for $V$ and $I$ by combining the red-star data of Harris {\refit et~al.\ } (1991) with observations of his six blue Landolt standards: G~162-66, GD~108, G~163-50, G~93-48, Feige~67, and SA~114-750. The resulting transformation equations, which are valid in the range $-0.5 < {\rm F555W} - {\rm F785LP} < 3$ are \begin{equation} {\rm F785LP} - I = 0.054 - 0.154 \ (V - I) - 0.002 \ (V - I)^2 \label{eq1} \end{equation} \begin{equation} {\rm F555W} - V = 0.001 + 0.061 \ (V - I) - 0.009 \ (V - I)^2. \label{eq2} \end{equation} Both equations have r.m.s.~residuals of 0.023~mag. We note here that ground-based photometry of central stars within high-surface-brightness nebulae can be extremely difficult, due to the bright, irregularly distributed flux from their surrounding nebulae (see Ciardullo \& Bond 1996 for a discussion of this well-known problem). However, the superior resolving power of the {\sl HST\/}\/ reduces this problem enormously. Consequently, the $V$ and $I$ magnitudes obtained in our survey represent a useful new database for future PN analyses, even for the majority of stars that are not resolved binaries. Our measured PNN $V$ magnitudes and $V$$-$$I$ colors are listed in Table~4; for reference, the nebular-based values of $c$ (which are taken from literature measurements of the Balmer-decrement and/or radio flux density) are also tabulated. Table~5 lists the PNN $I$ magnitudes for those objects not observed in $V$; except for the composite and late-type stars A~46, He 1-5, and HFG~1, these data were transformed to the standard system using the nebular-derived value of $c$ and the assumption that the intrinsic $V-I$ color of the objects is $-0.4$. (Abell 46, He 1-5, HFG~1 were transformed using ground-based colors.) In most cases, the magnitudes listed in the tables are accurate to better than $\sim 0.05$~mag, and represent mean values derived from our two exposures. In a few cases, partially saturated stellar images compromised our ability to derive an accurate magnitude for the PNN{}. For the objects where this occurred, we double-checked our measurements by deriving both a PSF magnitude and an aperture-photometry magnitude, and intercomparing the results computed from our long-exposure frame with those obtained from the short exposure. Those PNNs with uncertain magnitudes and colors are noted in Tables~4 and 5 with colons. Two central stars, those of IC~4997 and NGC~5315, were so badly overexposed that we were unable to derive $V$ and $I$ magnitudes; and two more, NGC~6309 and NGC~6369, were too badly overexposed in the $I$ band only. For A~82 we are uncertain of the identity of the central star, as discussed in detail below. Figure~5 compares our {\sl HST\/}\/ $V$ magnitudes to ground-based measurements obtained by Kaler and collaborators (Shaw \& Kaler 1985, 1989; Jacoby \& Kaler 1989) and by Tylenda {\refit et~al.\ } (1991). The figure demonstrates that for PNNs with magnitudes brighter than $V \sim 15.4$, the agreement is generally good. There is no systematic error between our measurements and those of Tylenda {{\null} et al.,} as the mean difference between the two magnitude systems is $0.00 \pm 0.06$~mag. Moreover, although our {\sl HST\/}\/ magnitudes are systematically fainter than the Kaler {\refit et~al.\ } measurements, the offset is small: when the highly discrepant objects IC~3568, NGC~7662, and NGC~7008 (which we resolve as a binary with two comparably bright components) are omitted, the mean difference between the two systems is $0.05 \pm 0.03$~mag. We note that, in general, the scatter between our measurements and those of Kaler {\refit et~al.\ } is significantly smaller ($\sigma \simeq 0.13$~mag) than that for Tylenda {\refit et~al.\ } ($\sigma \simeq 0.32$~mag). But, more importantly, both sets of residuals are larger than the internal errors of the measurements would indicate. Some of this additional scatter undoubtedly comes from intrinsic variability in the PNNs themselves, as Bond \& Ciardullo (1990) and others have shown that $\sim 0.1$~mag variability in PN central stars is not uncommon. For PNNs fainter than $V \simeq 15.4$, the agreement between the {\sl HST\/}\/ data and the ground-based measurements is considerably poorer. This is presumably due to the increased importance of nebular contamination in the ground-based photometry. The effect is particularly noticeable in the Kaler {\refit et~al.\ } data, where $V$-band fluxes derived from aperture photometry are overestimated by as much as $\sim 3$~mag. \section{Physical Associations and Optical Doubles} The goal of this paper is to determine distances to PNe with resolved binary nuclei by fitting the companion stars to the main sequence. To do this, one should, in principle, use a main sequence appropriate for the metallicity of each individual nebula. However, of the candidates listed in Table~3, fewer than half have nebular abundance measurements, and of those, only $\sim 5$ have reliable data on atomic species that can reasonably be assumed to have been unaffected by nuclear processing. We have therefore chosen not to attempt any metallicity correction. Moreover, since most PNe arise from an old-disk population, we have also chosen not to use a main sequence based on observations of young, metal-rich clusters such as the Hyades or Pleiades. Instead for our distance estimates, we use an $M_V$, $V$$-$$I$ main sequence that is derived from a spline-fit to a color-magnitude diagram of old-disk field stars. This relation, which has kindly been provided to us by H.~Harris (1998), is based on U.S. Naval Observatory (USNO) CCD parallaxes of faint field stars (Monet {\refit et~al.\ } 1992; Dahn 1993; plus recent unpublished measurements), USNO photographic parallaxes of bright stars (Dahn {\refit et~al.\ } 1988), and a combination of data on large-parallax stars from other observatories. Since the dataset excludes known halo objects, high-velocity stars, and binaries, the resultant main sequence should be applicable directly to our PN companions. The adopted USNO main sequence is given in Table~6. As mentioned above, our candidate binary PNNs were selected purely on the basis of spatial coincidence. Hence it is likely that a few optical doubles are mixed in with the true physical pairs. Although it is impossible to state with complete certainty whether any individual object is, or is not, a binary, we can make some plausibility arguments based on the inferred properties of the stars and nebulae. Table~7 lists inferred properties of all of our candidate binaries, grouped according to our belief that the physical association is probable, possible, or doubtful. The reasons for our categorization are given case-by-case below. To derive the distance moduli given in column~3, we dereddened the stars using the adopted $E(B$$-$$V)$ of column~2 and the prescriptions given by Harris {\refit et~al.\ } (1991) and Holtzman {\refit et~al.\ } (1995). We then fit the dereddened companion(s) to the USNO main sequence of Table~6. Column~4 lists the formal errors in the derived distance moduli; these were calculated by combining in quadrature our photometric errors, the errors associated with the PNN reddening estimates, and the uncertainty in the main-sequence $M_V$ value. Note that this last term, which we assume to be $\sim 0.3$~mag, usually dominates the error. Due to the spread of field-star metallicities and the effects of stellar evolution, the observed main sequence in the solar neighborhood has at least a spread of 0.3~mag (Perryman {\refit et~al.\ } 1995; Jaschek \& G\'omez 1998). Consequently, although the mean main sequence at any color may be well defined, individual distance determinations which are based only on two-color photometry must have at least this amount of error. In the future it should be possible to refine our distance estimates via metallicities deduced from spectroscopy and/or Str\"omgren photometry of the companion stars, but for the present, our individual PN distance measurements carry this substantial uncertainty. Columns~6 through 9 in Table~7 give various properties of the stars and nebulae under the assumption the companion stars are, indeed, physically associated with their PNNs. Column~6 is the projected PNN-companion star separation in astronomical units, column~7 is the physical size of the PNN's nebula as derived from the angular diameters listed in Acker {\refit et~al.\ } (1992), column~8 is the absolute $V$ magnitude of the PNN, and column~9 is the nebula's absolute [\ion{O}{3}] $\lambda 5007$ magnitude, based on the line fluxes given by Acker {\refit et~al.\ } (1992). Following Jacoby (1989), we define \begin{equation} M_{5007} = -2.5 \log F_{5007} - 13.74, \end{equation} where the monochromatic flux, $F_{5007}$, is given in ergs~cm$^{-2}$~s$^{-1}$. A discussion of the individual objects appears below. \subsection{Probable Physical Pairs} {\it Abell 31:\/} The central star of Abell~31 lies within an extremely large (diameter $970\arcsec$) nebula, and has a very faint, close companion that is detected only on our $I$ frames. The extremely red color of the companion, $(V-I)_0 \gtrsim 3.2$, places an interesting limit on the distance: in order to be on the main sequence, the companion star must be closer than $\sim 440$~pc. This compares to statistical distance estimates that range from 230~pc (CKS) to 1~kpc (Zhang 1995). The small projected separation derived for the pair ($< 115$~A.U.) and the very low probability for chance superposition (0.07\%) argue strongly for additional observations. {\it Abell 33:\/} The companion to Abell~33 was first detected by Cudworth (1973); the separation and position angle of the pair has not changed significantly since then. The low probability of a chance superposition and the reasonable implied parameters of the system suggest that the stars are physically associated. Additional evidence for this conclusion comes from the approximate agreement between our distance of 1.2~kpc and the statistical distance estimates of $\sim 0.7$~kpc (CKS), 1.6~kpc (Maciel 1984), and 2.9~kpc (Zhang 1995). We note here that Abell~33 (and Abell~24) have the largest reddening discrepancies in our sample. According to the observed $V$$-$$I$ color of the central star, the extinction towards Abell~33 should be close to zero. However, Kaler, Shaw, \& Kwitter (1990) have used the nebular emission lines to infer a total {H$\beta$\ } extinction of almost a magnitude. Because of the object's high galactic latitude, the low extinction values derived by earlier emission-line studies (Chopinet \& Lortet-Zuckermann 1976; Kaler 1983), and the negative reddening implied by the ratio of radio to H$\beta$ flux (CKS), we have chosen to ignore the Balmer-line extinction measurement for this object, and have used $E(V$$-$$I) = 0$ in our calculations. {\it K 1-14:\/} Prints of the Palomar Sky Survey (POSS) reveal that a pair of stars separated by $9\farcs1$ is at the center of this nebula. H.E.B. noted this fact $\sim 20$~years ago, as did Kaler (1981), and on this basis we included the object in our program. However, Kaler \& Feibelman (1985) concluded that neither of these stars is the PNN: using both {\it International Ultraviolet Explorer\/} ({\it IUE\/}) observations and a large-scale ground-based plate provided to them by F.~Sabbadin, these authors concluded that the actual central star of K~1-14 is a fainter third object, lying {\it between\/} the two POSS stars. Our {\sl HST\/}\/ images confirm that the Kaler-Feibelman conclusion is correct: an extremely blue object lies $\sim 2\farcs 4$ southwest of the brighter POSS star. (The PNN is invisible on the POSS prints due to the overlapping images of the field stars.) Remarkably, the hot star itself has a very close, even fainter companion $0\farcs 36$ away. The very small separation of {\it this\/} pair, coupled with the low stellar density in the field, argue strongly for a physical association. Our main-sequence-fitting distance of $\sim 3$~kpc is in reasonable agreement with that determined from statistical methods (3.4~kpc, CKS; 5.3~kpc, Maciel 1984). {\it K 1-22:\/} This system has a number of remarkable parallels to K~1-14. The discoverer of the PN (Kohoutek 1971) noted that three stars appear near the center of the nebula, and proposed that the brightest of these was the PNN{}. Smith \& Gull (1975) confirmed that this star is very blue, but Kaler \& Feibelman (1985) noted that the visual luminosity of the star was too large to match the flux distribution extrapolated from the {\sl IUE\/}\/ measurements. Instead Kaler \& Feibelman suggested that a fainter object $4''$ east of Kohoutek's candidate was the PN's true central star. Our {\sl HST\/}\/ frames confirm that the original Kohoutek-Smith-Gull star is indeed the PNN, but its visual flux is augmented by a very close visual companion $0\farcs 35$ away. In fact, at $V=17.13$, the companion is so red ($V-I=1.12$) that it is actually brighter than the PNN in the $I$ band. Given the low stellar density in the field, the pair almost certainly form a bound system. Our derived distance of 1.3~kpc is reasonably consistent with distances based on statistical techniques (1.0~kpc, CKS; 3.4~kpc, Zhang 1995). {\it K 1-27:\/} The faint companion to K~1-27 has the color of a late A~star; consequently if the star is on the main sequence, its faint apparent magnitude ($V \simeq 21.3$) implies an implausibly large distance (55~kpc). There is, however, another possible solution for this system: the companion star could be a white dwarf. Instead of placing the secondary on the main sequence, we obtained the distance given in Table~7 by putting the companion on the white-dwarf cooling sequence defined by the hydrogen-rich white-dwarf models of Bergeron, Wesemael, \& Beauchamp (1995). The locus of points defined by these models is in excellent agreement with that observed for field white dwarfs (Monet {\refit et~al.\ } 1992), and, by adopting the curve, we derive a white-dwarf absolute magnitude of $M_V = 12.77$. If we use this value, and ignore the nebular Balmer-line extinction estimate $c = 0.28$ (which leads to an unphysically blue color for the central star), then we derive a distance to the system of $\sim 470$~pc. There is only one other distance estimate, statistical or otherwise, for K~1-27. By modeling the absorption lines from the hydrogen-deficient central star, Rauch, K\"oppen, \& Werner (1994) derived a distance to the object of $1.29^{+1.05}_{-0.58}$~kpc. This number, however, assumes $c = 0.28$; with a more reasonable value of $c = 0.08$, their distance decreases by a factor of $\sim 1.2$, and comes into marginal agreement with ours. In most other respects, our white-dwarf hypothesis is reasonable. At an absolute magnitude of $M_V = 12.77$, the cooling age of a $0.575 M_\odot$ companion white dwarf is $\sim 1$~Gyr (cf.~Bergeron, Wesemael, \& Beauchamp 1995), and thus the star would not be in a particularly rapid phase of evolution. Similarly, if the distance is indeed 0.47~kpc, then the derived values for the nebular size (0.1~pc) and binary separation (260~A.U.) are also plausible. The only potential problem lies in the extremely small amount of flux radiated by the nebula in [\ion{O}{3}] $\lambda 5007$. At an absolute [\ion{O}{3}] $\lambda 5007$ magnitude of $M_{5007} \sim 8.7$, the nebula would be a full 13.1~mag fainter than the bright end of the [\ion{O}{3}] $\lambda 5007$ planetary nebula luminosity function (Ciardullo {\refit et~al.\ } 1989; Jacoby {\refit et~al.\ } 1992), and the intrinsically faintest [\ion{O}{3}] $\lambda 5007$ source in the Strasbourg-ESO planetary nebula catalog (Acker 1992). Note, however, that the nebular and central-star properties of K~1-27 (Henize \& Fairall 1981; M\'endez, Kudritzki, \& Simon 1985) are very similar to those of Abell~36, an extremely high-excitation object which, from its [\ion{O}{3}] line flux and statistical distance, {\it is\/} the faintest [\ion{O}{3}] $\lambda 5007$ source currently in the catalog ($M_{5007} \sim 7.2$). In fact, both Henize \& Fairall and M\'endez {\refit et~al.\ } have remarked that the properties of the K~1-27 nebula and central star are more extreme than those of Abell~36, and a large fraction of the nebula's oxygen is in O~IV{}. Consequently, the intrinsically small amount of [\ion{O}{3}] $\lambda 5007$ emission is entirely reasonable. We consider this a likely white dwarf-PNN binary system. {\it Mz~2:\/} This PN is 3 degrees from the Galactic plane, and only 31 degrees from the Galactic center. Thus, the field-star density in the region is extremely high: there are over 3400 stars recorded on our Wide Field and Planetary Camera frames. Nevertheless, only five of these stars are as bright as, or brighter than, our putative companion. The relative scarcity of $I = 15.8$ stars, and the small ($0\farcs 28$) separation between the PNN and the companion, leads to a very high probability (99.99\%) of a physical association. Our distance to Mz~2 of $\sim 2.2$~kpc is in good agreement with most statistical distance estimates (2.3~kpc, CKS; 2.4~kpc, Van de Steene \& Zijlstra 1994; 2.7~kpc, Zhang 1995). Kingsburgh \& English (1992) derive a somewhat larger distance ($\sim 5$~kpc) from the PN's location on the [\ion{O}{2}] density-ionized mass relation, but the subsequent classification of the object as a Type~I PN (Perinotto {\refit et~al.\ } 1994) vitiates this analysis. {\it NGC 1535:\/} The companion to this well-studied PNN has the colors of an early G~star. If the pair form a bound system, then the projected separation between the two stars is $\sim 2400$~A.U., and the distance to the binary is $\sim 2.3$~kpc. This value is in excellent agreement not only with most modern statistical distances (2.3~kpc, CKS; 2.0~kpc, Van de Steene \& Zijlstra 1994; 2.1~kpc, Zhang 1995), but with the distance derived from the non-LTE model-atmosphere analysis of its central star (2.0~kpc, M\'endez, Kudritzki, \& Herrero 1992). This consistency, along with the small probability of a chance superposition, supports the interpretation that this is a true physical system. {\it NGC 3132:\/} The binary nature of this PNN was first discovered by Kohoutek \& Laustsen (1977). The companion star has an A0~V spectral type (Lutz 1977), which implies that the PN was ejected from an even more massive progenitor. The hot star, which is only marginally resolved from the ground, is easily measured on our PC frame, and is 5.65~mag fainter in $V$ than its companion; if we place the A~star on the main sequence, we obtain a distance to the system of $\sim 0.8$~kpc. This is consistent with estimates based on the interstellar reddening along the line of sight (Gathier, Pottasch, \& Pel 1986) and the ground-based spectroscopic parallax of the A~star (Pottasch 1980). Unfortunately, there are two caveats that accompany our distance measurement. The first arises out of the definition of our field-star main sequence. For a PN to exist, the age of the system must be at least $\sim 5 \times 10^7$~years (Bressan {\refit et~al.\ } 1993); this is a non-negligible fraction of the main-sequence lifetime of the A~star. Consequently, some main-sequence evolution must have occurred, and the companion could be up to $\sim 0.2$~mag brighter than its zero-age main sequence luminosity. While the effect is partially mitigated by our use of a field-star main sequence rather than a zero-age main sequence, it is still likely that the companion is older than a typical A-type field star. Fortunately, the effect is small: the difference between the expected mean magnitude of a group of A stars of all ages, and that of a sample of stars with ages of at least $\sim 5 \times 10^7$~years is only $\sim 0.05$~mag. The second problem comes from the uncertainty in our reddening estimation. On the lower main sequence, the $V$ vs.~$V$$-$$I$ reddening vector is roughly parallel to the main sequence. Consequently a small error in the extinction makes little difference to a distance derivation: both $V_0$ and $M_V$ are modified in a similar fashion. For A~stars, however, this is not the case, as the slope of the main sequence is significantly steeper. The result is that a small uncertainty in reddening translates into a large uncertainty in distance. In the case of NGC~3132, the uncertainty in our color-based reddening estimate, $\sigma_{E(V-I)} \sim 0.11$~mag, translates into a $\sim 0.64$~mag uncertainty in distance modulus. This dominates the error given in Table~7. Str\"omgren photometry of the A~star would improve the distance estimate considerably. {\it NGC 7008:\/} It is somewhat surprising that the composite nature of the nucleus of NGC~7008 was not detected prior to our survey. Although the angular separation between the PNN and the companion is small ($0\farcs 4$), the magnitude difference between the two stars is only $\sim 0.5$~mag in $V$, and in $I$ the light from the companion star actually dominates. By placing the companion on the main sequence, we obtain a distance of $\sim 0.4$~kpc, and an implied stellar separation of $\sim 160$~A.U.{} This compares to a distance of $\sim 1.1$~kpc estimated from extinction measurements along the line of sight (Pottasch 1983), and the CKS statistical distance of $\sim 0.9$~kpc. Because the probability of a chance superposition of a $V \sim 14$~mag star near the PNN is extremely small, we consider this a good binary-star candidate. There is one caveat to our distance, however. On both of our F814W frames, the point-spread function for NGC 7008's companion star appears slightly broader than expected from a normal single star. The effect is not large and we cannot rule out the existence of an instrumental problem. However, it is possible that the companion is itself a binary, making the PNN a hierarchical triple. If this is the case, our distance to the object is an overestimate. {\it Sp 3:\/} The companion in this system has the color of an F-type star and is only $0\farcs 3$ from the PNN; when we place it on the main sequence, we derive a distance of $\sim 2.4$~kpc, in good agreement with the CKS statistical distance of 1.9~kpc. This consistency, along with the small probability of superposition, indicates that this is a true physical association. \subsection{Possible Physical Pairs} {\it Abell 7:\/} Our 200-s $I$-band Planetary Camera image shows only two stars on the frame: the PNN and a very faint companion only $0\farcs 9$ away. Since the stellar density in this field is low (it is at $b = -30^\circ$), there is a very high probability that the stars are physically associated. Unfortunately, the companion is $\sim 5.3$~mag fainter than the central star in $I$ and so red that it was not detected on our $V$~frames. This implies a $V$$-$$I$ color redder than 1.21 and an absolute magnitude fainter than $M_V = 6.7$. From main-sequence fitting, we can derive only an upper limit to Abell~7's distance of $\sim 13$~kpc. This limit is not particularly useful, as the size of the nebula ($760\arcsec$) and the central star's white-dwarf spectrum (Liebert 1980) demand that the object be quite nearby. If the CKS statistical distance of $\sim 200$~pc is accurate, then the projected separation between the PNN and the companion star is only $\sim 200$~A.U{}. Deeper images are desirable to investigate this interesting object further. In the meantime we can only classify A~7 as a possible physical binary. {\it Abell 30:\/} The companion to Abell~30 was found from the ground by Cudworth (1973). The position angle has not changed significantly over the ensuing $\sim 25$~years, and our main-sequence fitting distance ($\sim 2$~kpc) is consistent with the CKS statistical distance of 1.7~kpc. On the other hand, the relatively large separation ($5\farcs 25$) does allow a $\sim2\%$ probability of a chance superposition. Moreover, at a distance of 2~kpc, the derived physical separation of the pair ($\sim 10,600$~A.U.) is starting to approach the observed 0.1~pc cutoff imposed by the Galactic tidal field (cf.~Bahcall \& Soneira 1981; Latham {\refit et~al.\ } 1984). We therefore categorize A~30 only as a ``possible'' physical association. {\it Abell~63:\/} The central star of Abell~63 is the 11-hour eclipsing binary UU~Sge (Bond, Liller, \& Mannery 1978); the PNN's nearby companion ($2\farcs8$ away) was first noted during photoelectric observations by Krzeminski (1976). In spite of the PN's low galactic latitude, $b=-3^\circ$, we find only a $1.5\%$ probability that the resolved companion is a chance superposition. Based on the nebular Balmer decrement, the extinction to the object is $c=0.71 \pm 0.10$ (Walton, Walsh, \& Pottasch 1993); this number is in excellent agreement with the value obtained by modeling the 2200~\AA\ interstellar absorption feature (Walton {\refit et~al.\ } 1993; Pollacco \& Bell 1993). When we combine this extinction estimate with our {\sl HST\/}\ photometry, we derive a distance to the companion star of $\sim 1.2$~kpc. From the run of stellar reddening versus distance in that part of the sky, a foreground extinction of $c \sim 0.7$ ($E(B$$-$$V) = 0.5$) implies a distance of $\sim 1$~kpc (cf.~Bond {\refit et~al.\ } 1978), in good agreement with our findings. However, analyses of the spectral type and color of the back hemisphere of the extremely close eclipsing companion yield values that are two to three times larger ($\sim 3.6$~kpc, Walton {\refit et~al.\ } 1993; $3.2 \pm 0.6$~kpc, Pollacco \& Bell 1993; $2.4 \pm 0.4$~kpc, Bell, Pollacco, \& Hilditch 1994). These distances could be overestimates, if some of the extreme heating effects on the companion ``leak'' around to the unilluminated back side. Unfortunately, without improved observations, we can only identify this as a possible physical system. {\it IC 4637:\/} This new pair---which remarkably was never noted from the ground---was discovered on our original Cycle~3 F785LP WF/PC frame and followed up with WFPC2 (F555W and F814W) observations. The implied separation between the PNN and its companion, $\sim 1200$~A.U., is large, but not excluded by any means; similarly, the derived nebular size of $\sim 0.05$~pc is small, but again not unreasonable. Interestingly, the statistical distances to this object vary widely, from 0.8~kpc (Amnuel {\refit et~al.\ } 1984) to 2.3~kpc (CKS), but all estimates lie on the high side of our 0.5~kpc value. Furthermore, the object's only individual distance determination, the non-LTE model-atmosphere value of M\'endez, Kudritzki, \& Herrero (1992), is larger still (3.3~kpc). We are therefore led to classify this object only as a possible physical binary. {\it NGC 2392:\/} Like Abell 7, NGC~2392 has a faint companion which is barely detected on our $I$~frames, and is invisible in $V$. Our upper limit to the distance, $\sim 6.4$~kpc, is not particularly useful, since both non-LTE model atmospheres (M\'endez {\refit et~al.\ } 1992) and statistical techniques (CKS; Van de Steene \& Zijlstra 1994; Zhang 1995) place the object closer than $\sim 2$~kpc. Deeper imaging could provide a photometric distance, but for the present, the lack of a $V$ magnitude forces us to categorize the object as a ``possible'' physical association. {\it NGC 2610:\/} Again, there is a close ($0\farcs 6$), extremely faint companion to this object that could not be detected on our $V$ images. A much deeper image is needed to produce a meaningful distance estimate. We again categorize this as a ``possible'' physical system. \subsection{Doubtful Physical Pairs} {\it Abell 24:\/} The visual companion to this star was first noted by Cudworth (1973). We measure a separation of $3\farcs33$, in good agreement with the $3\farcs4$ reported by Cudworth. However, our measurement for the position angle of the binary differs by $5^\circ$ from his value; this offset marginally exceeds the uncertainty in the old measurement (Cudworth 1997). Since for any plausible distance, this change in position angle is larger than is possible from binary orbital motion, the difference argues against the existence of a physical association. Other facts suggestive of a chance superposition include the extremely large diameter derived for the nebula ($\sim 4$~pc) and the factor of five difference between our putative distance ($\sim 2.4$~kpc) and distances derived from statistical techniques ({\refit e.g.,}\ 0.52~kpc, CKS). Despite the fact that the formal probability of a chance superposition is only $\sim 2\%$, we believe that this object is an optical double. {\it NGC 650-1:\/} The companion star of this object was first noticed by Cudworth (1973). However, our {\sl HST\/}\ images reveal that the companion is itself a very close double, with a separation of only $0\farcs 16$. Given the field star density of the region, the two stars of the companion almost certainly form a physical system, as the probability of a $0\farcs 16$ chance superposition is less than 0.1\%. However, associating the companion binary with the PNN is more problematical. The separation and position angle of the companion pair relative to the PNN is the same as it was 25 years ago; thus, despite the fact that the probability of a chance association is $\sim5\%$, this may argue for a physical association. However, if the system is a hierarchical triple, then it is a very peculiar one: although both components of the companion have the same color, $(V-I)_0 \approx 0.87$, one is 0.8~mag brighter than the other. In other words, both stars cannot be on the main sequence. If the PNN is associated with this pair then two of the three stars of this non-interacting trinary happen to be in a phase of rapid evolution. The hypothesis of a bound triple system runs into further problems when one considers the age and metallicity of the stars. NGC~650-1 is a metal-rich Type~I planetary nebula, with a most likely progenitor mass $M \gtrsim 2 M_\odot$ (Peimbert \& Torres-Peimbert 1983). Yet if we assume a solar-like metallicity and place the fainter component of the close pair on the main sequence, then the position of the brighter component in the color-magnitude diagram demands that the system be impossibly old. As the isochrones in Figure~6 show, both of the close components could lie on or close to the same isochrone if the $V-I$ colors were near the blue ends of their error bars, but the age of the pair would still be far in excess of that of the PNN{}. A lower age for the close pair would be possible if a substantial amount of additional reddening were adopted. However, if that were the case, then the reddening would greatly exceed that of the PNN itself, and we would be forced to conclude that the two systems were unrelated. Another alternative is that the fainter component of the close pair is on the main sequence, and the brighter component is itself an unresolved binary containing two equal-brightness main-sequence stars; although not impossible, this option would imply the unlikely conclusion that the close pair is actually a hierarchical triple containing three stars of essentially identical mass. The distance of 4.1~kpc given in Table~7 was derived by fitting the fainter component of the close pair to the main sequence; it exceeds the statistical distances (CKS: 0.7~kpc; Zhang 1995: 1.6~kpc; Van de Steene \& Zijlstra 1994: 1.3~kpc) by a substantial amount. We therefore suspect that the very close pair, whatever its astrophysical explanation, is probably significantly more distant than---and older than---the PNN, and thus not physically associated. {\it PuWe 1:\/} A slightly brighter companion to the central star of this object is visible on the Palomar Sky Survey at a separation of $5''$ (Purgathofer \& Weinberger 1980), but our WF/PC observations reveal that the companion is itself resolved into two stars with a separation of $0\farcs 6$ arcsec. Unlike the NGC~650-1 system, both cool components are apparently on the main sequence, as they produce a consistent set of distances. Since the PN is an extremely large object that extends over $10'$ on the sky (Purgathofer \& Weinberger 1980), we expect it to be relatively nearby, and our derived distance of $\sim 280$~pc does not contradict this hypothesis. Nevertheless, the PuWe~1 PNN is probably not associated with the close pair. Based on the stellar density in the region, and the rather large ($5\arcsec$) separation between the PNN and the companions, the likelihood of a chance superposition in the region is relatively large, $\sim 3\%$. Moreover, PuWe~1 is one of the few PNNs with a reliable trigonometric parallax measurement, $2.3 \pm 0.4$~mas (Harris {\refit et~al.\ } 1997). This is more than $2 \sigma$ smaller than we would predict based on our companion star measurements. Even more telling is that the proper motion of the close pair appears to be significantly different from that of the PNN (Harris 1998). Thus, we believe the stars are not physically associated. \subsection{Noteworthy Optical Doubles and Unresolved Objects} Several other stars in our survey were included because they were identified as candidate visual binaries on the basis of ground-based observations, but we have concluded that they are not physical pairs. In addition, a few other objects are definitely binaries, based on their composite spectra and/or red colors, but we did not resolve them. These objects are discussed in this subsection. {\it Abell~78:\/} In his survey of the nebular features of this PN, Jacoby (1979) noted that an extended patch of H$\alpha$ emission $\sim 10\farcs 2$ from the PNN was spatially coincident with a star, but that there was no other evidence for a physical association. Our star counts in the region suggest that the ``companion'' is most likely an unrelated field star which is merely projected onto the nebula. {\it Abell 82:\/} A~82 has a diameter of about $90''$, and was included in our program as a candidate binary because, as recounted by Kaler \& Feibelman (1985), there is a relatively bright late-type star at the center of the nebula. Kaler \& Feibelman's {\sl IUE\/}\/ spectra did reveal a weak, apparently reddened, UV continuum within the $10''\times 20''$ aperture of the instrument, but the derived color temperature was much too low to explain the nebula's ionization. The central object has a K-type spectrum (Kaler \& Feibelman 1985) and colors of $V=14.90$, $B$$-$$V=1.28$, $V$$-$$I=1.36$ (Kwitter, Jacoby, \& Lydon 1988, and our {\sl HST\/}\/ measurements). Kwitter {\refit et~al.\ } have suggested that the image of the K~star masks the true PNN on ground-based photographs (cf.~the case of K~1-14 above), but we find no evidence on our {\sl HST\/}\/ frames for any blue object near the center of the nebula. Alternatively, Kaler \& Feibelman (1985) have proposed that a faint object $6''$ northwest of the K~star is the PNN, but our photometry ($V=18.15$, $V-I=1.10$) demonstrates that this star is not blue either. In fact, the bluest object on our PC frame is a $V=12.85$, $V$$-$$I=0.35$ star located $18''$ southeast of the K~star. This star might have been within the {\sl IUE\/}\/ aperture if the pointing was slightly inaccurate, and thus could be the reddened source detected by Kaler \& Feibelman. However, it is too far off-center to be the planetary's central star. In view of the above ambiguities, we investigated the {\sl IUE\/}\/ observations in further detail, with the aid of copies of the original observing scripts which were kindly provided by W. A. Feibelman. Two short-wavelength spectra were obtained, SWP~19771 (1983 Apr 20, 25~min), which shows no convincing detection, and SWP~19908 (1983 May 5, 120~min), which detected the reddened continuum described above, and shows that the source was well-centered in the aperture. For both observations, a blind offset was done from a nearby 11th-mag star onto the coordinates of the center of the PN{}. Our measurements on the Digitized Sky Survey show that the resulting pointing would have been about $10''$ directly west of the 14th-mag K-type star, and thus would either have missed it, or at best have had it at the edge of the aperture. However, handwritten notes from the telescope operator indicate that a possible additional telescope movement may have been performed onto a ``second central star,'' which we speculate could have been the 12.8-mag star to the southeast of the K~star. In this case, the weak UV spectrum would be nicely explained. If so, however, it appears that {\sl IUE\/}\/ never actually observed the star at the center of the PN{}. We are thus left with two plausible conclusions: either the true hot central star is a faint, unresolved companion of the K-type star; or a ``born-again'' scenario (see next paragraph) for the K~star itself may have to be invoked. Further ground- and space-based spectroscopic observations of this PNN are urged. {\it He~1-5\/ \rm and \it H~3-75:\/} The central star of He~1-5 is the well-known object FG~Sge, which appears to be a PNN that has undergone a late helium thermal pulse and has become a ``born-again'' red giant. In agreement with spectroscopic observers ({\refit e.g.,}\ Feibelman \& Bruhweiler 1990), and in agreement with the born-again scenario, we find no evidence for a hot companion in our {\sl HST\/}\/ images. H~3-75 is an interesting and possibly related object: according to Sanduleak (1984) the central star has a K-type spectrum. {\it IUE\/} ultraviolet spectra (Bond 1993) show no trace of a hot star, and our {\sl HST\/}\/ frames likewise show no resolved companion. The PNN is therefore a strong candidate for another born-again giant. Spectroscopic observations are needed. {\it He 2-36:\/} The optical central star of He 2-36 has a spectral type of A2~III (M\'endez 1978), and {\it IUE\/} observations suggest the presence of a hot companion (Feibelman 1985). Our frames do not resolve the binary. {\it M 1-2:\/} This object has a G2~Ib spectral type (O'Dell 1966), and strong forbidden and permitted emission lines reminiscent of those seen in a planetary nebula (see Grauer \& Bond 1981 and references therein). Our {\sl HST\/}\/ observations do not resolve any binary companion, nor do they show a resolved nebular component. The system is probably a symbiotic-like binary which is too compact to be resolved by {\sl HST\/}, although Feibelman (1983) has argued from {\it IUE\/} observations that the star is surrounded by a young planetary nebula. {\it NGC 1514:\/} The central star is a well-known composite system, containing a hot sdO star and an A-type companion (Kohoutek \& Hekela 1967; Greenstein 1972). Greenstein's radial velocity measurements indicate that the period of the system must be quite long (or perhaps that the binary is seen nearly pole-on). Nevertheless, our {\sl HST\/}\/ images fail to resolve the system, setting an upper limit of approximately 40 A.U. for the projected separation. {\it NGC 6853:\/} The binarity of the PNN was first suggested by Cudworth (1973), who identified a $V \simeq 17$ companion $6\farcs 5$ from the central star. Astrometric measurements (Cudworth 1977) support this contention, as the proper motion of the companion is similar to that of the central star. Unfortunately, given the density of field stars in the region, our probability calculations cannot confirm this claim of binarity, as there is almost a 90\% chance that a random field star will be projected within $6\farcs 5$ of the PNN{}. In fact, based on the star counts, the best candidate for association with NGC~6853 is a $V \approx 18.7$ star $1\farcs 1$ from the central star, but even this object has a 12\% probability of being a chance superposition. We do note that if we assume Cudworth's star is associated with the PNN, then our {\sl HST\/}\/ photometric values of $V = 16.91$, $V$$-$$I = 1.83$, coupled with the assumption of no foreground reddening, leads to a distance of $430 \pm 62$~pc. This is in agreement with the distance of $380 \pm 64$~pc recently obtain by Harris {\refit et~al.\ } (1997) from USNO parallax measurements. In keeping with the precepts of this paper, we will not discuss this object any further, but we urge radial-velocity measurements for the system. {\it Th 2-A:\/} As noted in \S 2, this object was included in our program because of a nearby companion noted on ground-based CCD frames. However, there are over 2300 field stars present on our {\sl HST\/}\ frames, and the star in question has a $\sim 50\%$ probability of being a chance superposition. We therefore cannot classify it as a possible visual binary. {\it A~34 and A~66:\/} Like Th 2-A, nearby companions were noted during ground-based observations. The companion of A~34 has a 28\% probability of being there by chance, and for A~66 the probability is 21\%. {\it A~46, A~65, HFG~1, K~1-2, LoTr~5, \rm and \it NGC~2346:\/} All of these central stars (along with A~63, discussed above) are known to be extremely close binaries, based on their photometric variability (cf.~Bond \& Livio 1990). Not unexpectedly, none of them were resolved in our survey, and we do not find any nearby resolved companions that have a high probability of being physically associated. \section{Comparison with Statistical Distance Scales} Due to their complexity and vast range in size, luminosity, mass, and excitation, there is no reliable method for obtaining distances to large samples of individual Galactic planetary nebulae. As a result, in order to investigate planetary nebulae as a class, it is necessary to rely on statistical distance estimators. The principle behind these statistical distances is straightforward: the emitted Balmer-line flux from an ionized plasma depends almost exclusively on the total mass of the emitting region and the plasma density. Consequently, if the ionized mass of a nebula can be estimated, then its observed flux and angular size can be used to calculate its distance. The key, of course, is to know the amount of ionized mass contained in the nebula. There are several prescriptions in the literature for estimating this mass, starting with the original assumption by Shklovsky (1956) that the ionized masses of all PNe are the same, and that all PNe are optically thin. Other formulations include adopting an ionized mass that is (a)~linearly proportional to nebular radius (Maciel \& Pottasch 1980), (b)~proportional to a power of the radius (Zhang 1995), (c)~proportional to the radio brightness temperature of the nebula (Van de Steene \& Zijlstra 1995; hereafter VdSZ), (d)~dependent on the [\ion{O}{2}]-derived nebular density (Kingsburgh \& Barlow 1992; Kingsburgh \& English 1992), or (e)~constant for optically thin nebulae, but proportional to an optical-thickness parameter for denser objects (Daub 1982; CKS){}. Once calibrated, each of these relations is capable of producing distance estimates to large numbers of objects. Unfortunately, the number of PNe with independently known distances, which can therefore be used as zero-point calibrators for these methods, is extremely small. Moreover, some of the calibrators have distances that are themselves controversial. For example, CKS used 19~PNe with ``well-determined'' distances to calibrate their distance scale, while VdSZ used 23 calibrators. A comparison of the two samples, however, reveals that only 16~PNe are common to both datasets, and of those, two objects have adopted distances that differ by more than a factor of two! A major reason for this dichotomy is that many of the ``well-determined'' PN distances are based on such methods as reddening of field stars projected near the PN line of sight (Gathier, Pottasch, \& Pel 1986), Galactic \ion{H}{1} absorption measurements (Gathier, Pottasch, \& Goss 1986), nebular expansion parallaxes ({\refit e.g.,}\ Hajian {\refit et~al.\ } 1995), and non-LTE atmospheric modeling of PN central stars ({\refit e.g.,}\ M\'endez {\refit et~al.\ } 1992). None of these methods is unassailable, and each carries its own (possibly substantial) uncertainty. Our new sample of PNe with visual-binary nuclei significantly increases the number of objects with reliable distance measurements, and constitutes a new and important set of data with which to calibrate PN statistical distances. In addition to having quantifiable errors, the PNe in our sample have distinctly different selection criteria from those measured by other methods. PNe with interstellar-medium-based distances are mostly distant objects in the plane of the Milky Way. Similarly, PNe with nebular-expansion distances are objects that are bright and optically thick, while those analyzed with non-LTE model atmospheres have highly evolved central stars. Our wide binary stars, however, are primarily nearby objects and objects at high galactic latitude. Consequently, our sample not only enlarges the PN calibrator database, but also reduces the possibility of a systematic error due to selection biases. The usefulness of our dataset for testing statistical distance techniques is demonstrated in Figure~7, which compares directly measured PN distances with those from four different statistical methods. For the directly measured distances, we use the 10 ``probable'' associations listed in Table~7 (one of which, A~31, is only an upper limit), along with the three additional distances to the ``possible'' associations which are not upper limits. To these we add the new ground-based distance to NGC~246 (based on photometry of its wide binary companion; Bond \& Ciardullo 1999), and trigonometric distances to seven PNe derived from recent $> 3 \sigma$ parallaxes measured by the {\it Hipparcos\/} satellite (ESA 1997) and the U.S. Naval Observatory (Harris {\refit et~al.\ } 1997). These accurate photometric and geometrical distances are plotted against statistical distances computed from the 5~GHz flux measurements and angular diameters given by Zhang \& Kwok (1993) and CKS, using the prescriptions of CKS, VdSZ, Maciel \& Pottasch (1980), and Zhang (1995). For reference, the data of Figure~7 are given in Table~8. It is immediately obvious that distances from all of the statistical methods have considerable dispersion. Compared to the binary and astrometric distances, the CKS and Zhang estimates scatter by $\sigma \sim 1.7$~mag and $\sim 1.8$~mag in distance modulus, respectively. The VdSZ distances exhibit the smallest dispersion, $\sim 1.6$~mag, while the Maciel estimates have the largest, $\sim 2.4$~mag. (This last result is not very surprising, since some of the PNe considered here have radii outside the formal limits of the Maciel calibration.) Interestingly, a significant amount of dispersion is attributable to one object, PHL~932, which has a {\it Hipparcos\/} parallax distance of $110^{+48}_{-26}$~pc, but statistical distances that range from 800~pc (CKS) to 5.0~kpc (Zhang). The central star of PHL~932 is an exceptionally unusual object. With a sdB spectral type (M\'endez {\refit et~al.\ } 1988), it is one of only two known PNNs of this class, lying well off the normal post-AGB evolutionary tracks. It has been suggested that the star may have evolved through a common-envelope binary interaction (Iben \& Tutukov 1993), and thus the ionized mass could differ substantially from that of normal PNe. If PHL~932 is arbitrarily disregarded, then the dispersion in the VdSZ and Zhang errors drops dramatically to $\sim 1.1$~mag and $\sim 1.3$~mag, respectively. Even this, however, is much larger than the errors expected from the techniques. Even more surprising are the zero-point offsets in the scales exhibited in the figure. All four of the statistical methods examined here systematically overestimate the distances to the objects in our sample. CKS come closest to reproducing our distance scale, with estimates that are, on average, only $\sim 25\%$ larger than the photometric and geometrical measurements. Since this is a $1 \, \sigma$ result, their analysis is still consistent with our measurements. The distance scales defined by VdSZ, Maciel, and Zhang, however, are all significantly too long, with mean distance moduli that are 1.6, 1.2, and 2.7~mag larger than our own. The sizes of these offsets are extraordinary, especially when one considers that the VdSZ and Zhang relations work well for samples of PNe in the Galactic bulge. Another way of looking at the problem is to compare the properties of our PNe with those of other PNe with ``well-determined'' distances. Following VdSZ, we plot in Figure~8 the distance-independent PN radio brightness temperature (which is related to the CKS optical depth parameter), against the distance-dependent quantity of PN radius. The filled circles represent planetary nebulae with resolved binary companions, the crosses show PNe with trigonometric-parallax distances, and the open circles are the VdSZ sample of PNe with what they considered to be ``well-determined'' distances. It is clear from the figure that the PNe in our new sample do not obey the rather tight relation defined by the VdSZ calibrators; at a given radius, the binary and astrometric PNe are systematically fainter in the radio and have a larger amount of scatter (although the scatter is dominated by a few outliers). We interpret Figure~8 as revealing a classical selection effect. The distances for the PNe in the VdSZ sample come almost exclusively from reddening and \ion{H}{1} absorption distance determinations; hence these objects are nearly all relatively bright, high-surface-brightness PNe that can be seen at great distances along the plane of the Milky Way. Consequently, a calibration using this sample of objects nicely recovers the distances to Galactic bulge PNe. The planetaries in our sample, however, are primarily nearby, faint, and optically thin. They represent a population of objects that has apparently received little weight in the calibration of statistical distances. Figure 8 also points out the probable cause of a long-standing controversy about the Galactic PN distance scale. For years, there have been two distance scales for Milky Way planetaries. The traditional ``short'' distance scale adopted by Cahn \& Kaler (1971) is supported by the extinction-distance relation of Pottasch (1984), the Magellanic Cloud observations of CKS and Webster (1969), and the Galactic bulge measurements of Stasi\'nska {\refit et~al.\ } (1991). All of these methods use bright PNe, similar to those analyzed in the VdSZ study. The opposing ``long'' distance scale, which is larger by a factor of $\sim 1.5$, is supported by statistical-parallax measurements (Cudworth 1974), stellar-atmosphere models (M\'endez {\refit et~al.\ } 1992), [\ion{O}{2}] line-ratio density estimates (Kingsburgh \& Barlow 1992; Kingsburgh \& English 1992), and number counts of PNe in other galaxies (Peimbert 1990). These techniques study a different sample of PNe, and include local, lower-surface-brightness objects. Based on the results of Figures~7 and 8, it is therefore not surprising that a different distance scale is derived. It is unfortunately less obvious---and beyond the scope of this paper---how one could devise a new ``grand-unification'' calibration that simultaneously handles both the lower-surface-brightness objects that prevail among the nearby nebulae, and the brighter PNe that dominate samples like those in the Galactic bulge and extragalactic systems. We leave this daunting task to future workers. \section{Conclusion} We have successfully used a large-scale {\sl HST\/}\/ snapshot survey to find 19 resolved companions of central stars in planetary nebulae. We consider ten of these systems to be probable physical associations, another six to be possible associations, and the remaining three to be doubtful. By fitting the companions to the main sequence (or in one case the white-dwarf cooling sequence), we have derived reliable distances to the PNe. Comparison with various statistical distance estimates reveals that all of the current statistical methods {\it overestimate\/} the distances to our sample. A more detailed examination suggests that the well-studied nebulae used as calibrators for statistical methods are biased toward high-surface-brightness, low-Galactic-latitude objects. Our sample, on the other hand, contains more objects of lower surface brightness, which may have systematically lower nebular masses. It will be a challenge to future refinements of the statistical methods to include an additional correction for this effect. The primary source of error in our PN distance measurements from resolved binaries is the unknown metallicities of the companion stars. This uncertainty propagates directly into the definition of the $M_V$, $V$$-$$I$ main sequence used to obtain the absolute magnitudes (and distance moduli) of the stars. Our distances could therefore be improved substantially via abundance analyses of the nebulae and/or stellar atmosphere analyses (or intermediate-band photometry) of the companion stars. In addition, since several candidate companion stars were detected only in $I$, deeper multicolor imaging with {\sl HST\/}\/ (possibly with the restored NICMOS camera) could add significantly to the list of PNe with direct distance determinations. Finally, in order to confirm our candidate companion stars and identify additional ones, proper-motion and radial-velocity measurements are needed. \acknowledgments We thank H.~C.~Harris for providing a definition of the field-star $M_V$, $V$$-$$I$ main sequence and for unpublished information. We would also like to thank G.~Jacoby for instructive conversations at the start of this project and on the properties of optically thin and thick nebulae, and for his participation as a Co-Investigator on the Cycle~3 portion of the project. S.~Torres-Peimbert provided support at a critical moment. This work was supported by STScI grants GO-04308.01-92A and GO-06119.02-94A, and NSF NYI grant AST 92-577833. \newpage
\section{Introduction} The rapid expansion of the use of high quality crystaline materials in optical and eletronic devices has strongly stimulated research, both theoretical and experimental, on dynamics of crystalization. Computer simulation has played an important role on the development and understanding of crystal growth. During growth the solid-fluid interface can display several interesting phenomena like segregation, dynamical instabilities and pattern formation \cite{cross}. \\ A crystal can grow from the adjacent fluid by different mechanisms, depending on the structure of the interface (rough or smooth), material purity, growth rates, temperature gradients and so on. For a crystal to grow atoms or molecules must be transported from the fluid towards the interface with a non-zero sticking probability. besides that, the latent heat generated at the interface as well as the solute excess segregated must be carried away. These requirements can be met in a controlled way by puting the sample in an appropriated furnace submmited to a temperature gradient and pulling it with constant speed toward the colder region. results of such experiments have been compared with results of two dimensional models. Nijmeijr and Landau\cite{marco} reported molecular dynamic (MD) simulation on laser heated pedestal growth of fibers. Previous simulations consisted of kinetic models using Monte Carlo techniques and numerical solution of transport equations\cite{weeks}. \\ \section{Simulation} In this work we use MD to simulate the solidification of a two component system of a solvent (particles of type a) and solute (particles of type b) interating via a Lennard-Jones (LJ) potential. By tunning the LJ parameters we set the struture of the interface (rough or smooth) and the segregation coefficient. We also simulate morphological instabilities, the planar inteface becoming cellular and eventually dendritic (Mullins-sekerka instability.). The simulation is carried out with all particles interacting through the LJ potential \begin{equation} \phi_{i,j}(r_{i,j}) = \epsilon_{i,j} \left [ (\frac{\sigma_{i,j}}{r_{i,j}})^{12} - (\frac{\sigma_{i,j}}{r_{i,j}})^{6} . \right ] \end{equation} The indexes $i$ and $j$ stands for particles in the positions $r_i$ and $r_j$ respectively and $r_{i,j} = |r_i - r_j|$. There are three types of interactions, solvent-solvent, solvent-solute and solute-solute. In each case the LJ parameters are labeled as $(a,a)$, $(a,b)$ and $(b,b)$. distance $r$, time $t$ and temperature $T$ are measured in units of $\sigma_{a,a}$, $\sigma_{a,a}(m_a/\epsilon_{a,a})^{\frac{1}{2}}$ and $k_{B}/\epsilon_{a,a}$. initially we distribute $N=n_x \times n_z$ particles over the two dimensional volume $L_x \times L_z$. We assume periodic boundary conditions in the $x$ direction. In the $z$ direction we divide the system in to two distinct regions, a solid and a fluid one. In the solid region particles stand initially in their equilibrium position in a total of $n_x \times n_{0z}$ particles. on the fluid region we distribute the rest of the particles with the density chosen to be $\rho = 0.5\sigma_{a,a}^{-2}$, randomly distributed in a triangular lattice and slightly dislocated from their equilibrium position. We impose a temperature gradient along the $z$ direction using a velocity renormalization approach. The system is divided in two regions, one cold which is set to $T=0$ and the other to $T=T_h$ higher than the melting temperature $T_m$ which is $T_m=0.4$ in our units. we let the system evolve in time for $N_e$ time steps of size $\delta t$ until it reachs equilibrium. Once the equilibrium is reached we start pulling the system in the $+z$ direction, at a pulling velocity $V_p$. The $-z_{max}$ layer works as a particle source, maintaining a constant flow to the material. The basic experimental set up is shown in Fig.~\ref{eps1} \begin{figure} \includegraphics[width=.9\textwidth]{figure1.eps} \caption[]{Basic experimental set up for directional growth. The interface motion is visualized using an optical microscope. $T_m$ is the melting temperature of the mixture and $V_p$ is the pulling velocity.} \label{eps1} \end{figure} In Fig.~\ref{eps2} we show an example of solute segregation during directional growth o fthe binary mixture: caprolactane as solvent and methyl-blue as solute. In the top part of Fig.~\ref{eps2} is shown an image of the crystal (left side) and melt (right side), with maximum concentration of methyl-blue at the melt side of the interface. From the gray level of the image we obtain the methyl-blue concentration profile across the sample, which in the melt, decays exponetially as a function of the distance from the interface (bottom part of Fig.~\ref{eps2}). Fig.~\ref{eps3} shows the simulation results done in a system initially with $27 \times 30$ particles, solute concentration of $5\%$, pulling velocity $V_p = 0.004$, and LJ parameters given by : $\epsilon_{a,b}=0.5$, $\epsilon_{b,b}=0.1$, $\sigma_{a,b}=1.0$, $\sigma_{b,b}=1.0$ and $m_b = 1.0$. \begin{figure} \includegraphics[width=.9\textwidth]{figure2.2.ps} \caption[]{Directional solidification of a binary mixture showing the solute segregation at the interface. The experimental result is for the mixture caprolactane (solvent) and methyl-blue (solute).} \label{eps2} \end{figure} After averaging over many runs to improve statistics one obtains the steady-state solute concentration profile represented as data points with error bars. \begin{figure} \includegraphics[width=.9\textwidth]{figure3.eps} \caption[]{Our simulation results using the Lennard-Jones potential describes in detail the experiment.} \label{eps3} \end{figure} It is shown in Fig.~\ref{eps3} a theoretical result obtained from the diffusion equation for the solute concentration in the fluid phase\cite{coura} \begin{equation} \frac{\partial c_f}{\partial t} = D\frac{\partial^2 c_f}{\partial z^2} + V_f \frac{\partial c_f}{\partial z} \end{equation} where $D$ is the solute diffusion coefficient in the fluid phase and $V_f$ is the velocity of the system of reference. The diffusion equation is supplemented with the boundary conditions: \begin{subeqnarray} c_f & = & c_0 ~~~~~~~~~~~~~~~~ at ~~ z \rightarrow \infty \ts , \label{f1a}\\ (1-K)V_fc_f & = & -D\frac{d c_f}{d z} ~~~~~~~~ at ~~ z = 0 \ts .\label{f1b} \end{subeqnarray} The second condition is just the mass conservation at the interface and $K$ is the ratio between the solute and solvent concentrations. \section{Cellular Instability} With the LJ parameters chosen above the morphological instabilities were inhibited during the cellular growth. We can stimulate instabilities by tunning the LJ parameters, particularly by slightly varying the equilibrium position between particles $a$ and $b$, i.e., changing $\sigma_{a,b}$. Fig.~\ref{eps4} show the effect of varying this parameter. \begin{figure} \includegraphics[width=.9\textwidth]{figure4.eps} \caption[]{The figure shows the concentration profile found in our simulation. It is to be compared to Fig.~\ref{eps2} } \label{eps4} \end{figure} For $\sigma_{a,b} = 1$ the interface is rough, becoming cellular and dendritic for $\sigma_{a,b} > 1$. As shouls be expected a lower $\sigma_{a,b}$ sets the interface smoother than larger values indicating that the system is very sensitive to geometric factores. (We have also simulated some experiments fixing the value of $\sigma_{a,b}$ and varying $\epsilon_{a,b}$ and $\epsilon_{b,b}$, the results did not show any qualitative change in their behavior. Raising $\sigma_{a,b}$ breaks the hexagonal structure o fthe solid. The energy involved in such distortion is so large that the interface segragate solute particles, bonds $(a,a)$ and $(b,b)$ are preferible than $(a,b)$. Once particles of the type $b$ concentrate at the interface its melting point is lowered and channels of $b$ particles are formed inside the crystal. Fig.~\ref{eps5} Shows this effect. \begin{figure} \includegraphics[width=.8\textwidth]{figure5.2.ps} \caption[]{Cellular instability originated by varying $\sigma_{a,b}$. From left to right $\sigma_{a,b}=1.05,1.10$ and $1.20.$} \label{eps5} \end{figure} We are now studying the stability \cite{figueiredo} of those interfaces. We start with a smoth interface then, we introduce a perturbation at the interface. It has been shown that near the bifurcation from planar to cellular the time evolution of the most instable Fourier mode of the perturbation can be described by a third order Landau-amplitude equation. This however is the subject of a new research and will be soon published elsewhere. \section{Acknowledgements} This work was partially supported by CNPq, FAPEMIG and FINEP. Numerical work was done at the CENAPAD-MG/CO and in the LINUX parallel cluster at the {\sl Laborat\'orio de Simula\c{c}\~ao} Departamento de F\'{\i}sica - UFMG.
\section{Introduction} The main goal of this paper is to demonstrate, on the examples of semisimple algebras of second order ($ A_2, B_2, C_2, G_2$), the general construction connecting a semisimple algebra of a given grading to an exactly integrable system. The simplest example is the two-di\-men\-sional Toda lattice considered and integrated in the case of an arbitrary semisimple algebra almost 20 years ago \cite{leznov:LG, leznov:ls0}\footnote{For $A_n$ series this problem was solved more then 150 years ago in Darbouxs papers!}. For main grading, exactly integrable systems were explicitly found and described in the recent papers of the author \cite{leznov:LM} (so called Abelian~case). In the present paper we follow three dif\/ferent and independent aims. The f\/irst is to relate unknown up to now integrable systems to nonabelian gradings\footnote{Note that the zero order subspace is a non-commutative algebra by itself.} (see \cite{leznov:GC} in this con\-nec\-tion). The second one is to get rid of the restriction of nonabelian Toda theory to use only subspaces with zero and $\pm 1$ grading indices. The last, but not the least important one, is to provide the reader with a scheme of how the group representation theory (in the very restricted volume) can be applied to the theory of integrable systems. For our purposes here it is not the shortest and simplest way to the result that is important, but the result by itself. Therefore, in concrete examples we tried to use calculations that can be followed and checked directly using only simplest algebra. The paper is organized as follows. Section~2 contains the background information on representation theory of semisimple algebras and groups (as a rule without proofs). Section~3 describes the general construction, mathematical tricks and methods used in main sections. In Section~4 concrete examples of semisimple algebras of second order are considered in details for all possible gradings. Concluding remarks and perspectives for further investigation are outlined in Section~5. \section{Semisimple algebras and groups} Let ${\cal G}$ be an arbitrary f\/inite-dimensional graded Lie algebra\footnote{We make no distinction between algebras and super-algebras, just keeping in mind that even (odd) elements of super-algebras are always multiplied by even (odd) elements of the Grassman space.}. Then $\cal G$ can be written as a direct sum of subspaces of dif\/ferent grading indices \begin{equation} {\cal G}=\left(\oplus^{N_-}_{k=1} {\cal G}_{-\frac{k}{2}}\right) {\cal G}_0 \left(\oplus^{N_+}_{k=1}{\cal G}_{\frac{k}{2}}\right). \label{GR} \end{equation} Generators with an integer grading index are called bosonic, while those with half-integer grading index are named fermionic. The positive (negative) grading corresponds to upper (lower) triangular matrices. The grading operator $H$ for an arbitrary semisimple algebra can be written as a linear combination of elements of commutative Cartan subalgebra taking unity or zero values on the generators of simple roots \begin{equation} H=\sum^r_{i=1} \left(K^{-1}c\right)_i h_{i}. \label{cartan1} \end{equation} Here $K^{-1}$ is the inverse Cartan matrix $K^{-1}K=KK^{-1}=I$ and $c$ is a column of zeros and unites in an arbitrary order. Under the main grading all $c_i=1$. In this case $(K^{-1}c)_i=\sum\limits_{j=1}^r K^{-1}_{i,j}$, where $r$ is the rank of the algebra. As usually, generators of simple roots $X^{\pm}_i$ (raising/lowering operators) and Cartan elements $h_i$ satisfy the system of commutation relations: \begin{equation} [h_i , h_j]=0, \qquad [h_i,X^{\pm}_j]=\pm K_{j,i}X^{\pm}_j, \qquad [X^{+}_i,X^{-}_j\}={\delta}_{i,j} h_j, \quad (1 \leq i,j \leq r),\label{aa6} \end{equation} where $K_{ij}$ are elements of Cartan matrix and brackets $[,\}$ stand for the graded commutator. The highest vector $\ve{j}$ ($\vc{j} \equiv \ve{j}^{\dagger}$) of the $j$--th fundamental representation has the following properties: \begin{equation} X^{+}_i\ve{j}=0, \qquad h_i\ve{j}={\delta}_{i,j}\ve{j}, \qquad \vc{j}\ve{j}=1. \label{high} \end{equation} The representation is exhibited by applying lowering operators $X^{-}_i$ to the vector $\ve{j}$ repeatedly and extracting all linearly-independent vectors with non-zero norm. The f\/irst few basis vectors are \begin{equation} \ve{j}, \quad X^{-}_j\ve{j}, \quad X^{-}_i X^{-}_j\ve{j}\neq 0,\quad K_{i,j}\neq 0,\quad i\neq j. \label{vectors} \end{equation} } In fundamental representations an important identity for matrix elements of a group element $G$ holds\footnote{Recall that a superdeterminant is def\/ined as $\mbox{sdet} \left(\begin{array}{cc} A, & B \\ C, & D \end{array}\right) \equiv \det (A-BD^{-1}C ) (\det D)^{-1}$.}~\cite{leznov:ls0} \begin{equation} \mbox{sdet} \left(\begin{array}{cc} \vc{j}X_j^+GX_j^-\ve{j}, & \vc{j}X_j^+G\ve{j} \vspace{3mm}\\ \vc{j}GX_j^-\ve{j}, & \vc{j} G \ve{j} \end{array}\right) = \prod^r_{i=1}\vc{i} G \ve{i}^{-K_{ji}}, \label{recrel} \end{equation} The identity (\ref{recrel}) is in fact a generalization (to the case of an arbitrary semisimple Lee super-group) of the famous Jacobi identity that relates determinants of orders $(n-1)$, $n$ and $(n+1)$ of some special matrices. As we will see in the next section, this identity is of such importance in deriving exactly integrable systems that one can even say that it is responsible for their existence. We will still refer to (\ref{recrel}) as to ``the f\/irst Jacobi identity''. In addition to (\ref{recrel}), there exists another independent identity of key importance~\cite{leznov:l} \begin{equation} \hspace*{-5pt}\begin{array}{l} \displaystyle (-1)^P K_{i,j} {\vc{j}X_j^+X_i^+ G \ve{j}\over \vc{j} G \ve{j}}+ K_{j,i} {\vc{i}X_i^+X_j^+ G \ve{i}\over \vc{i} G \ve{i}} \vspace{3mm}\\ \displaystyle \qquad +K_{ij}K_{j,i}(-1)^{jP}{\vc{j}X_j^+ G \ve{j}\over \vc{j} G \ve{j}}{\vc{i}X_i^+ G \ve{i}\over \vc{i} G \ve{i}}=0 ,\qquad i\neq j \end{array}\label{J2} \end{equation} which will be called the second Jacobi identity. This identity is responsible (in the above sense) for the existence of hierarchies of integrable systems invariant with respect to integrable mappings that are connected to every exactly integrable system. Either from (\ref{recrel}) or from (\ref{J2}) it is possible to construct many usefull recurrent relations that are used in further consideration. Taking into account the importance of Jacobi identities (\ref{recrel}) and (\ref{J2}) for further consideration we present below a brief proof of (\ref{recrel}). Let us consider the left hand side of (\ref{recrel}) as a function on the group. The action on an arbitrary group element $G$ in the def\/inite representation $l$ of the operators of the right (left) regular representation is by def\/inition \begin{equation} M_{\mbox{\scriptsize left}}(\tilde M_{\mbox{\scriptsize right}}) G= M_l G (\tilde M_l), \label{AM} \end{equation} where $M_l$, $\tilde M_l$ are the generators (the matrices of corresponding dimension) of shifts on the group in a given $l$ representation. Now let us act with an arbitrary generator of the simple positive root $(X^+_s)_r$ on the left hand side of (\ref{recrel}). This action is equivalent to dif\/ferentiation and therefore should be applied consequently to the f\/irst and second columns of the matrix (\ref{recrel}) adding the results. The action on the second column results in zero as a corollary of the def\/inition of the higest state vector (\ref{high}). Action on the f\/irst column is dif\/ferent from zero only in the case $s=j$. But in this case using the same def\/inition of the highest state vector we conclude that as a result of dif\/ferentiation of the f\/irst column it becomes equal to the second one with the zero f\/inal result. Thus considered as a function on the group the left hand side of (\ref{recrel}) is also proportional to the highest vector (or a linear combination of such vectors) of some other representation. The higest vector of the irreducible representation is uniquely def\/ined by the values that Cartan generators take on it. If Cartan generators take on the highest vector values $V(h_i)=l_i$, the last can be uniquely represented in the form \begin{equation} \vc{l} G \ve{l}=C \prod_{i=1}^r(\vc{i} G \ve{i})^{l_i}.\label{HVD} \end{equation} Calculating the values of Cartan generators on the left hand side of equation (\ref{recrel}) (both left and right with the same result) and using the last comment about the form of the highest vector, we prove (\ref{recrel}) ($C=1$, as can be seen by putting $G=1$ and comparing both sides). The second Jacobi identity can be proven by similar argument \cite{leznov:l}. The following generalization of the f\/irst Jacobi identity will be very important in calculations dealing with nonabelian gradings. Let $\ve{\alpha}$ be basis vectors of some representation in the strict order of increasing the number of lowering generators (see (\ref{high}) and (\ref{vectors})). We also assume that the action of a generator of an arbitrary positive simple root on each basis vector results in a linear combination of the previous ones. Then the principal minors of an arbitrary order of the matrix ($G$ is an arbitrary element of the group): \[ G_{\alpha}=\vc{\alpha} G \ve{\alpha} \] are annihilated from the right (from the left) by generators of positive (negative) roots. Indeed this is equivalent to dif\/ferentiation and therefore it is necessary to act on each column (line) of the minors matrix and add the results. But the action of the generator of a positive simple root on the state vector with a given number of lowering operators transforms it into a state vector with a number of lowering operators on unity less, which according to our assumption is a linear combination of previous columns (lines). Thus in all cases the lines or columns of the resulting determinant are linearly dependent with zero result. The generators of Cartan subalgebra obviously take the def\/inite values on minors of these kind and if the corresponding values are $l^s_i$, it is possible to write the equality in correspondence with (\ref{HVD}) \begin{equation} Min_s= C_s \prod_{i=1}^r \vc{i} G \ve{i}^{l^s_i},\label{GJI} \end{equation} where constants $C_s$ can be determined as described above. \setcounter{equation}{0} \section{General construction and technique of computation} The grading of a semisimple algebra is def\/ined by the values that the grading operator $H$ takes on the simple roots of the algebra. As it was mentioned above, this values can be only zeros and unites in an arbitrary order. \[ [H, X^{\pm}_i]=\pm X^{\pm}_i,\qquad H=\sum_1^r (K^{-1}c)_i h_i,\qquad c_i=1,0. \] On the level of Dynkin's diagrams the grading can be introduced by using two colors for its dots: black for simple roots with $c_i=1$ and red for roots with $c_i=0$. To each consequent sequence of the red (simple) roots the corresponding semisimple algebra (subalgebra of the initial one) is connected. All these algebras are obviously mutually commutative and belong to the zero graded subspace. Cartan elements of the black roots also belong to the zero graded subspace. We will use the usual numeration of the dots of Dynkin diagrams and all red algebras will be distinguished by an index of their f\/irst root~$m_s$. The rank of $m_s$-th red algebra will be denoted as $R_s$. Thus $X^{\pm}_{m_s},X^{\pm}_{m_s+1},\ldots,X^{\pm}_{m_s+R_s-1}$ is the system of simple roots of $m_s$ red algebra. After these preliminary comments turn to the general construction \cite{leznov:l1}. Let two group valued functions $ M^+(y)$, $M^-(x)$ be solutions of $S$-matrix type equations \begin{equation} \hspace*{-5pt}\begin{array}{l} \displaystyle M^+_y=\left(\sum_0^{m_2} B^{(+s}(y)\right) M^+\equiv\left(B^{(0}+L^+\right)M^+, \vspace{3mm}\\ \displaystyle M^-_x=M^-\left(\sum_0^{m_1} A^{(-s}(x)\right)\equiv M^-\left(A^{(0}+L^-\right), \end{array} \label{I} \end{equation} where $B^{(+s}(y)$, $A^{(-s}(x)$ take values in $\pm s$ graded subspaces correspondingly and $s=0, 1, 2,\ldots,m_{1,2}$. In each f\/inite-dimensional representation $B^{(+s}(y)$, $A^{(-s}(x)$ are upper (lower) triangular matrices and therefore equations~(\ref{I}) are integrated in quadratures. The composite group valued function $K$ plays the key role in our construction \begin{equation} K=M^+ M^-. \label{II} \end{equation} It turns out that matrix elements of $K$ in various fundamental representations are related by closed systems of equivalent relations, which can be interpreted as exactly integrable system with known general solution. Bellow we describe calculation methods to prove this proposition. First of all let us calculate the second mixed derivative $(\ln \vc{i} K \ve{i})_{x,y}$, where index $i$ belongs to the black dot of Dynkin diagram. We have \begin{equation} (\ln \vc{i} K \ve{i})_x= {\vc{i} K (A^0+L^-) \ve{i}\over \vc{i} K \ve{i}}= A^0_i(x)+ {\vc{i} K L^- \ve{i}\over \vc{i} K \ve{i}}. \label{MI} \end{equation} Indeed, $ K_x=M^+(y)M^-_x(x)=K (A^0+L^-)$ as a corollary of equation for $M^-$. All red components of $A^0$ under the action on the black highest vector state $ \ve{i}$ lead to zero result in connection with (\ref{vectors}). The action of Cartan elements of the black roots state vector satisf\/ies the condition $h_j \ve{i}=\delta_{i,j} \ve{i}$ and thus only coef\/f\/icient on $h_i$ remains in the f\/inal result (\ref{MI}). Further dif\/ferentiation (\ref{MI}) with respect to $y$, with the help of arguments above, leads to following result: \begin{equation} (\ln \vc{i} K \ve{i})_{x,y}=\vc{i} K \ve{i})^{-2} \pmatrix{ \vc{i} K \ve{i}, & \vc{i} K L^- \ve{i} \cr \vc{i} L^+ K \ve{i}, & \vc{i} L^+ K L^- \ve{i} \cr}. \label{AR} \end{equation} Applying (\ref{AM}) of the previous section to the left hand side of (\ref{AR}), we f\/inally obtain \begin{equation} (\ln \vc{i} K \ve{i})_{x,y}=L^-_rL^+_l \ln \vc{i} K \ve{i})^{-1}. \label{ARR} \end{equation} Thus the problem of calculating the mixed second derivative is reduced to purely algebraic manipulations on the level of representation theory of semisimple algebras and groups. Further evaluation of (\ref{ARR}) is connected with repeated application of the f\/irst~(\ref{recrel}) and second (\ref{J2}) Jacobi identities as it will be clear from the material of the next section. As it was mentioned above, the red algebras of zero order graded subspace in general case are not commutative. This leads to additional computational dif\/f\/iculties. Let us denote by $ \ve{m_i}$ the highest vector of $m_i$th fundamental representation of the initial algebra. Of course, $\ve{m_i}$ is simultaneously the highest vector of the f\/irst fundamental representation of the $m_i$ red algebra. Let $\vc{\alpha_i}, \ve{\beta_i}$ be basis vectors of the f\/irst fundamental representation (this restriction is not essential) of $m_i$-th red algebra and let us consider the matrix elements of element $K$ in this basis. $R_i+1\times R_i+1$ matrix ($R_i+1$ is the dimension of the f\/irst fundamental representation) with matrix elements $\vc{\alpha_i} K \ve{\beta_i}$ will be denoted by a single symbol $u_i$ (index $i$ takes values from one to the number of the red algebras, which is the function of the choosen grading). For derivatives of matrix elements of so constructed matrix we have consequently (index~$i$ we omite for a moment): \begin{equation} \vc{\alpha} u_x \ve{\beta}=\vc{\alpha} K (A^0+L^-) \ve{\beta}=\! \sum_ {\gamma} \vc{\alpha} K \ve{\gamma}\vc{\gamma} I A^0 \ve{\beta} +\vc{\alpha} K L^- \ve{\beta}. \label{MCD} \end{equation} Or equivalently \[ u^{-1} u_x= A^0(x)+u^{-1} \vc{} K L^- \ve{}. \] Further dif\/ferentiation with respect to $y$ variable leads to \begin{equation} \hspace*{-5pt}\begin{array}{l} \displaystyle \vc{}((u^{-1} u_x)_y\ve{}=u^{-1} \vc{} (B^0+L^+) K L^- \ve{}-u^{-1}\vc{} (B^0+L^+) K \ve{} u^{-1}\vc{} K L^-\ve{} \vspace{2mm}\\ \displaystyle \qquad = u^{-1}(\vc{} L^+ K L^- \ve{}-\vc{} L^+ K \ve{} u^{-1} \vc{} K L^-\ve{}). \end{array} \label{MC} \end{equation} The last expression may be brought to the form of the ratio of two determinants of $R_i+2$ and $R_i+1$ orders respectively with the help of standard transformations: \begin{equation} \vc{} u (u^{-1} u_x)_y \ve{}={\mbox{Det}_{N_i+1}\pmatrix{ u & K L^-\ve{} \cr \vc{} L^+ K & \vc{} L^+ K L^- \ve{} \cr}\over \mbox{Det}_{N_i}(u)}. \label{MCC} \end{equation} The generalised Jacobi identity (\ref{GJI}) of the previous section plays the key role for discovery of the last expression and will be exploited many times. \setcounter{equation}{0} \section{The algebras of second rank {\mathversion{bold}$A_2$, $B_2, C_2$, $G_2$}} All elements of these algebras may be constructed by consequent multi-commutation of generators of four simple roots $X^{\pm}_{1,2}$ with the basic system of commutation relations \begin{equation}\label{RS} \hspace*{-5pt}\begin{array}{l} [X^+_1,X^-_1]=h_1,\qquad [X^+_1,X^-_2]=[X^+_2,X^-_1]=0,\qquad[X^+_2,X^-_2]=h_2, \vspace{2mm}\\ \displaystyle {}[h_1,X^{\pm}_1]= {\pm} 2X^{\pm}_1, \qquad [h_2,X^{\pm}_2]= {\pm} 2X^{\pm}_2, \vspace{2mm}\\ \displaystyle {}[h_1,X^{\pm}_2]= {\mp} pX^{\pm}_2, \qquad [h_2,X^{\pm}_1]= {\mp} X^{\pm}_1, \qquad p=1,2,3. \end{array} \end{equation} In all cases there are three possible nontrivial gradings: $(1,1)$ -- the principle one (Abelian case), $(1,0)$ -- the grading of the f\/irst simple root and $(0,1)$ -- of the second simple one. In the case of the principle grading corresponding integrable systems for arbitrary semisimple algebras were found and described in \cite{leznov:LM}. Each further subsections will be devoted to detail consideration of nonabelian gradings $(1,0)$, $(0,1)$, which are equivalent to each other only in the case of $A_2$ algebra. In the end of this mini-introduction we present the second Jacobi identity as applied to the algebras of second rank: \begin{equation} {\vc{2} X^+_2 X^+_1 K \ve{2}\over \vc{2} K \ve{2}}+p {\vc{1} X^+_1 X^+_2 K \ve{1}\over \vc{1} K \ve{1}}=p {\vc{2} X^+_2 K \ve{2}\over \vc{2} K \ve{2}} {\vc{1} X^+_1 K \ve{1}\over \vc{1} K \ve{1}}\label{2JI} \end{equation} or in notation, which will be introduced by the way of consideration \[ \bar \alpha_{21}+p\bar \alpha _{12}=p\bar \alpha _1\bar \alpha _2,\qquad \alpha_{12}+p\alpha _{21}=p\alpha _1 \alpha _2. \] \subsection{Unitary {\mathversion{bold}$A_2$} serie} The root system of this algebra consists of three elements with the generators $X^{\pm}_1,X^{\pm}_1,X^{\pm}_{12}$ $\equiv \pm [X^{\pm}_1,X^{\pm}_2]$. This case corresponds to $p=1$ in (\ref{RS}). For def\/initeness we restrict ourselves by $(1,0)$ grading $[H,X^{\pm}_1]=\mp X^{\pm}_1$, $[H,X^{\pm}_2]=0$. $L^{\pm}$ operators belong to $\pm 1$ graded subspaces and have the form: \[ L^+=\bar c_1 X^+_1+\bar c_2 [X^+_2,X^+_1],\qquad L^-=c_1 X^-_1+c_2 [X^-_1,X^-_2], \] where $c_{1,2}\equiv c_{1,2}(x)$, $\bar c_{1,2}\equiv \bar c_{1,2}(y)$. The object of investigation is $2\times 2$ matrix $u$ in the basis of the second fundamental representation of $A_2$ algebra\footnote{The (bra) basis vectors of the three dimensional (``qwark'') second fundamental representation of $A_2$ algebra are the $\vc{2}$, $\vc{2} X^+_2$, $\vc{2} X^+_2 X^+_1 $.}: \begin{equation} u=\left(\begin{array}{cc} \vc{2} K \ve{2}, & \vc{2} K X_2^-\ve{2} \vspace{1mm}\\ \vc{2}X_2^+ K \ve{2}, & \vc{2}X_2^+ K X_2^-\ve{2} \end{array}\right). \label{recrelI} \end{equation} In correspondence with (\ref{MCC}) we have: \begin{equation} \vc{} u (u^{-1} u_x)_y \ve{}={\mbox{Det}_3\pmatrix{ u & I K L^- \ve{} \cr \vc{} L^+ K I & \vc{} L^+ K L^- \ve{} \cr}\over \mbox{Det}_2(u)}. \label{MCC"} \end{equation} The action of operators $L^{\pm}$ on basis vectors $\ve{2}, X^-_2 \ve{2}$ ($\vc{2},\vc{2} X^+_2$) is the following: \[ \hspace*{-5pt}\begin{array}{l} L^-\ve{2}=c_2 X^-_1 X^-_2 \ve{2}, \qquad L^-X^-_2\ve{2}=c_1 X^-_1 X^-_2 \ve{2}, \vspace{2mm}\\ \vc{2}L^+=\bar c_2 \vc{2} X^+_2 X^+_1,\qquad \vc{2} X^+_2 L^+=\bar c_1 \vc{2} X^+_2 X^+_1. \end{array} \] So in this case the following sequence of basis vectors from generalized Jacobi identity (\ref{GJI}) takes places: \[ \vc{2}, \quad \vc{2} X^+_2, \quad \vc{2} X^+_2 X^+_1. \] The summed values of Cartan generators $h_1$, $h_2$ on this basis take zero values and so $\mbox{Det}_3$ from (\ref{MCC"}) equal to unity (with correct account of the constant). This is a really highest vector of scalar, one-dimensional representation of $A_2$ algebra. Finally (\ref{MCC"}) leads to the system, which matrix function $u$ satisfy: \begin{equation} (u^{-1} u_x)_y =(\mbox{Det}\; u)^{-1} u^{-1}\pmatrix{ c_2 \bar c_2, & c_1 \bar c_2 \cr c_2 \bar c_1, & c_2 \bar c_2 \cr}. \label{A_2} \end{equation} In usual notations the system (\ref{A_2}) is nonabelian $A_2 (1,0)$ Toda chain. The system (\ref{A_2}) is obviously form-invariant with respect to transformation: \[ u\to \bar g(y) u \bar g(x). \] With the help of this transformation the arbitrary up to now functions $c$, $\bar c$ may be evaluated to a constant values. \subsection{Orthogonal {\mathversion{bold}$B_2$} serie equivalent to simplectic one {\mathversion{bold}$C_2$}} This case corresponds to the choise $p=2$ in (\ref{RS}). Both gradings are not equivalent to each other and must be considered separately. First fundamental representation for $B_2$ algebra is the second one for $C_2$ serie and vice versa. \subsubsection{(1,0) grading} Generators $L^{\pm}$ may contain components with $\pm 1$, $\pm2$ graded indexes and have the form: \[ \hspace*{-5pt}\begin{array}{l} L^+=\bar c_1 X^+_1+\bar c_2 [X^+_2,X^+_1]+\bar c^2 [[X^+_2,X^+_1]X^+_1], \vspace{2mm}\\ L^-=c_1 X^-_1+c_2[X^-_1,X^-_2]+c^2 [ X^-_1[X^-_1,X^-_2]]. \end{array} \] The object of investigation is two dimensional matrix $u$ in the basis of the second fundamental representation of $B_2$ algebra. The main equation (\ref{MCC"}) also does not change. The action of $L^{\pm}$ operators on the basis vectors have now the form\footnote{Five basis vectors of the f\/irst fundamental representation of the $B_2$ algebra are the following: $ \ve{2}$, $X^-_2 \ve{2}$, $X^-_1 X^-_2 \ve{2}$, $X^-_1 X^-_1 X^-_2 \ve{2}$, $X^-_2 X^-_1 X^-_1 X^-_2 \ve{2}$.}: \[ \hspace*{-5pt}\begin{array}{l} L^-\ve{2}=(c_2+c^2 X^-_1) X^-_1 X^-_2 \ve{2}, \qquad L^-X^-_2\ve{2}= (c_1 +c^2 X^-_2 X^-_1) X^-_1 X^-_2 \ve{2}, \vspace{2mm}\\ \vc{2}L^+=\vc{2} X^+_2 X^+_1(\bar c_2+\bar c^2 X^+_1),\qquad \vc{2} X^+_2 L^+= \vc{2} X^+_2 X^+_1 (\bar c_1+\bar c^2 X^+_1 X^+_2). \end{array} \] Substituting this expression into (\ref{MCC"}) after some trivial evaluations we come to the following relation: \begin{equation} \hspace*{-5pt}\begin{array}{l} \displaystyle u(u^{-1} u_x)_y =(\mbox{Det}\; u)^{-1} \pmatrix{ \bar c_2+\bar c^2 (X^+_1)_l, & 0 \cr \bar c_1+\bar c^2 (X^+_1 X^+_2)_l , & 0 \cr} \vspace{3mm}\\ \displaystyle \qquad \qquad \qquad \times \pmatrix{ c_2+c^2 (X^-_1)_r , & c_1 +c^2 (X^-_2 X^-_1)_r \cr 0 , & 0 \cr} \mbox{Det}_3. \end{array} \label{F} \end{equation} In the last expression $\mbox{Det}_3$ satisfy all conditions of (\ref{GJI}), with the sequence of bases vectors: \[ \vc{2}, \quad \vc{2} X^+_2, \quad \vc{2} X^+_2 X^+_1. \] In this case the summed value of Cartan element $h_1$ is equal to 2, of $h_2$ -- to 0. So with the correct value of numerical factor we obtain $\mbox{Det}_3=2 \vc{1} K \ve{1}^2 $. The action of the f\/irst line operator in (\ref{F}) on $(\vc{1} K \ve{1})^2$ leads to the line of the form: \begin{equation} 2 (\vc{1} K \ve{1})^2 (c_2+2 c^2 \alpha_1 , c_1 +2 c^2 \alpha_{21}) \label{L}, \end{equation} where following abbreviations are used: \begin{equation} \hspace*{-5pt}\begin{array}{l} \displaystyle \bar \alpha_1={\vc{i} X^+_i K \ve{i}\over \vc{i} K \ve{i}},\qquad \bar \alpha_{12}={\vc{1} X^+_1 X^+_2 K \ve{1}\over \vc{1} K \ve{1}}, \vspace{3mm}\\ \displaystyle \bar \alpha_{21}={\vc{2} X^+_2 X^+_1 K \ve{2}\over \vc{2} K \ve{2}}, \qquad \alpha_i={\vc{i} K X^-_i \ve{i}\over \vc{i} K \ve{i}},\qquad i=1,2, \vspace{3mm}\\ \displaystyle \alpha_{21}={\vc{1} K X^-_2 X^-_1 \ve{1}\over \vc{1} K \ve{1}},\qquad \alpha_{12}={\vc{2} K X^-_1 X^-_2 \ve{2}\over \vc{2} K \ve{2}}. \end{array}\label{NOT} \end{equation} Now it is necessary to act with the help of the column operator (\ref{F}) on the line (\ref{L}). The result of this action on scalar factor may be presented in the form ($\mbox{Det}_2 u=\vc{1} K \ve{1}^2$): \[ 2\pmatrix{ \bar c_2+2 \bar c^2 \bar \alpha_1, & 0 \cr \bar c_1+2 \bar c^2 \bar \alpha_{12} , & 0 \cr} \pmatrix{ c_2+2 c^2 \alpha_1 , & c_1 +2 c^2 \alpha_{21} \cr 0 , & 0 \cr}. \] The action of the column operator (\ref{F}) on the line (\ref{L}) leads to additional matrix: \[ 4c^2 \bar c^2 \pmatrix{ (X^+_1)_l \alpha_1 & (X^+_1)_l \alpha_{21} \cr (X^+_2 X^+_1)_l \alpha_1 & (X^+_2 X^+_1)_l \alpha_{21} \cr}. \] With the help of formulae of Appendix~I the last matrix may be evaluated to the form: \[ 4 c^2 \bar c^2 (\mbox{Det}\; u)^{-1} u . \] Gathering all results together, we obtain f\/inally: \begin{equation} u(u^{-1} u_x)_y =2 \pmatrix{ p_1 \bar p_1, & p_2 \bar p_1 \cr p_1 \bar p_2, & p_2 \bar p_2 \cr}+ 4 c^2 \bar c^2 (\mbox{Det} \; u)^{-1} u, \label{B_2} \end{equation} where \[ p_1=c_2+2 c^2 \alpha_1,\qquad \bar p_1=\bar c_2+2 \bar c^2 \bar \alpha_1,\qquad p_2=c_1+2 c^2 \alpha_{21},\qquad \bar p_2=\bar c_1+2 \bar c^2 \bar \alpha_{12}. \] Now we would like to show that the derivatives $(p_{\alpha})_y$ and $(\bar p_{\alpha})_x$ are functionally dependent on matrix $u$ and themselves, closing in this way the system of equations of equivalence and representing it in the form of closed system of equations for $8$ unknown functions: 4~matrix elements of $u$ and $4$ components of $2$ two-dimensional spinors $p$, $\bar p$. Let us follow now the main steps of the necessary calculations. Using the introduced above technique we have subsequently: \[ (p_1)_y=2 c^2 (\alpha_1)_y= { 2 c^2\over \mbox{Det}\, (u)}\; \mbox{Det}\pmatrix{ \vc{1} K \ve{1} & \vc{1} K X^-_1\ve{1} \cr \vc{1} L^+ K \ve{1} & \vc{1} L^+ K X^-_1 \ve{1} \cr}. \] The action of $L^+$ on the state vector $\vc{1}$ is the following: \[ \vc{1}L^+=\vc{1} X^+_1(\bar c_1-\bar c_2 X^+_2-2 \bar c^2 X^+_2 X^+_1). \] Substituting the last expression in the previous equation and using the f\/irst Jacobi identity for its two f\/irst terms (linear in $\bar c_1$, $\bar c_2$) we obtain: \begin{equation} \hspace*{-5pt}\begin{array}{l} \displaystyle (p_1)_y={ 2 c^2\over \mbox{Det}\,(u)}(\bar c_1 \vc{2} K \ve{2} - \bar c_2 \vc{2} X^+_2 K \ve{2}) \vspace{3mm}\\ \displaystyle \qquad -{ 2 c^2\over \mbox{Det}\, (u)} \; \mbox{Det} \pmatrix{ \vc{1} K \ve{1} & \vc{1} K X^-_1\ve{1} \cr \vc{1}X^+_1X^+_2 X^+_1 K \ve{1} & \vc{1}X^+_1X^+_2 X^+_1 K X^-_1 \ve{1} \cr}. \end{array} \label{AE} \end{equation} Substituting into the second Jacobi identity (\ref{2JI}) ($p=2$) the f\/irst one in the form: \[ \vc{2} K \ve{2}=\mbox{Det}\pmatrix{ \vc{1} K \ve{1} & \vc{1} K X^-_1\ve{1} \cr \vc{1} X^+_1 K \ve{1} & \vc{1} X^+_1 K X^-_1 \ve{1} \cr} \] we obtain after some trivial transformations equality for two second order determinants: \[ \hspace*{-5pt}\begin{array}{l} \displaystyle \pmatrix{ \vc{1} X^+_1 K \ve{1} & \vc{1} X^+_1 K X^-_1\ve{1} \cr \vc{1}X^+_1X^+_2 K \ve{1} & \vc{1}X^+_1X^+_2 K X^-_1 \ve{1} \cr} \vspace{3mm}\\ \qquad = \pmatrix{ \vc{1} K \ve{1} & \vc{1} K X^-_1\ve{1} \cr \vc{1}X^+_1X^+_2 X^+_1 K \ve{1} & \vc{1}X^+_1X^+_2 X^+_1 K X^-_1 \ve{1} \cr}. \end{array} \] Evaluating the last column of the f\/irst determinant with the help of the f\/irst Jacobi identity: \[ \hspace*{-5pt}\begin{array}{l} \displaystyle \vc{1} X^+_1 K X^-_1\ve{1}={\vc{2} K \ve{2}+\vc{1} X^+_1 K \ve{1} \vc{1} K X^-_1\ve{1}\over \vc{1} K \ve{1}}, \vspace{3mm}\\ \displaystyle \vc{1} X^+_1 X^+_2 K X^-_1\ve{1}={\vc{2} X^+_2 K \ve{2}+\vc{1} X^+_1 X^+_2 K \ve{1} \vc{1} K X^-_1\ve{1}\over \vc{1} K \ve{1}} \end{array} \] we obtain for it: \[ \bar \alpha_1 \vc{2} X^+_2 K \ve{2}-\bar \alpha_{12} \vc{2} K \ve{2}. \] Finally we have: \begin{equation} (p_1)_y={ 2 c^2\over \mbox{Det}\,(u)}(u_{11}\bar p_2 -u_{21}\bar p_1),\qquad (p_2)_y={ 2 c^2\over \mbox{Det}\,(u)}(u_{12 }\bar p_2 -u_{22}\bar p_1). \label{UB_2}). \end{equation} So (\ref{B_2}), (\ref{UB_2}) and the same system for derivatives of $(\bar p)_x$ is the closed system of identities or $B_2(1,0;2,2;c^2,\bar c^2)$ exactly integrable system connected with the $B_2$ semisimple serie. To the best of our knowledge this system was not mentioned in literature before. >From the physical point of view the exactly integrable system (\ref{B_2}), (\ref{UB_2}) may be considered as a model of interacting charge ${1\over 2}$ particle $(\bar p, p)$ with scalar-vector neutral f\/ield $u$. Putting $ c^2=\bar c^2=0$, we come back to nonabelian Toda lattice system for single matrix valued unknown function $u$. \subsubsection{(0,1) grading} Generators $L^{\pm}$ contain only the components with $\pm 1$ graded indexes and have the form: \[ \hspace*{-5pt}\begin{array}{l} \displaystyle L^+=\bar d_1 X^+_2+\bar d_2 [X^+_1,X^+_2]+{1\over2} \bar d_3 [X^+_1[X^+_1, X^+_2]], \vspace{3mm}\\ \displaystyle L^-=d_1 X^-_2+d_2 [X^-_2,X^-_1]+{1\over 2} d_3 [ X^-_1[X^-_1,X^-_2]]. \end{array} \] With respect to transformation of $1$ -- red group $A_1$ functions $d_i(x)$, $(\bar d_i(y))$ are components of three dimensional $A_1$ vectors. The object of investigation is two dimensional matrix $u$ in the basis of the second fundamental representation of $B_2$ algebra. The main equation (\ref{MCC"}) conserves its form. The action of $L^{\pm}$ operators on the basis vectors have now the form\footnote{Four basis vectors of the f\/irst fundamental of the $C_2$ algebra are the following: $ \ve{1}$, $X^-_1 \ve{1}$, $X^-_2 X^-_1 \ve{1}$, $X^-_1 X^-_2 X^-_1 \ve{1}$.}: \[ \hspace*{-5pt}\begin{array}{l} L^-\ve{1}=(d_2-d_3 X^-_1) , \qquad L^-X^-_1\ve{1}= (d_1 -d_2 X^-_1) X^-_2 X^-_1 \ve{1}, \vspace{2mm}\\ \vc{2}L^+=\vc{2} X^+_2 X^+_1(\bar d_2-\bar d_3 X^+_1),\qquad \vc{2} X^+_2 L^+= \vc{2} X^+_2 X^+_1 (\bar d_1-\bar d_2 X^+_1 X^+_2). \end{array} \] Substituting this expression into (\ref{MCC"}), keeping in mind that $\mbox{Det}_3$ satisfy all conditions~of (\ref{GJI}), after some trivial evaluations we come to the following relation $(\mbox{Det}_3=\vc{1} K \ve{1})$: \begin{equation} \hspace*{-5pt}\begin{array}{l} \displaystyle u(u^{-1} u_x)_y =(\mbox{Det}\, u)^{-1} \pmatrix{ \bar d_2-\bar d_3 (X^+_1)_l, & 0 \cr \bar d_1-\bar d_2 (X^+_1)_l , & 0 \cr} \vspace{3mm}\\ \displaystyle \qquad \qquad \times \pmatrix{ d_2-d_3 (X^-_1)_r , & d_1 -d_2 ( X^-_1)_r \cr 0 , & 0 \cr} \vc{1} K \ve{1}. \end{array} \label{FF} \end{equation} Not cumbersome transformation leads the last expression to the f\/inally form: \begin{equation} (u^{-1} u_x)_y =(\mbox{Det}\, u)^{-1} u^{-1} \pmatrix{ \bar d_2, & -\bar d_3 \cr \bar d_1, & -\bar d_2 \cr} u \pmatrix{ d_2, & d _1 \cr -d_3, & -d_2 \cr}. \label{B21} \end{equation} (\ref{B21}) is nonabelian Toda chain for $B_2$ algebra with $(0,1)$ grading. To the best of our knowledge it was not considered before. System (\ref{B21}) is form-invariant with respect to transformation $u\to \bar g(y) u g(x)$, with the help of which it is possible to evaluate matrices depending on $x$, $y$ arguments to constant values. We omit here the question about the possible canonical forms of the system $B_2(0,1;1,1;\bar d,d)$ (\ref{B21}). \subsection{The case of {\mathversion{bold}$G_2$} algebra} As it is possible to expect, this case is the most cumbersome. It corresponds to the choise $p=3$ in (\ref{RS}). Firstly, we will consider the case of $(0,1)$ grading as the most simple one. It is connected with the 7-th dimensional f\/irst fundamental representation of $G_2$ algebra (group). The second one connected with $(1,0)$ grading is $14$-th dimensional. \subsubsection{(0,1) grading} In this case $L^{\pm}$ may contain the components $\pm 1$, $\pm 2$ graded subspaces and have the form: \[ \hspace*{-5pt}\begin{array}{l} \displaystyle L^+= \bar d_1 X^+_2+ \bar d_2 [X^+_1,X^+_2]+{1\over 2}\bar d_3 [X^+_1[X^+_1, X^+_2]] \vspace{3mm}\\ \displaystyle \qquad +{1\over 6}\bar d_4 [X^+_1[X^+_1[X^+_1,X^+_2]]]+{1\over 3} \bar d^2 [X^+_2[X^+_1[X^+_1[X^+_1,X^+_2]]]], \end{array} \] $L^-=(L^+)^T$, where $T$ sign of transposition; with simultaneously exchanging all coef\/f\/icients $\bar d\to d$. This operation we will call as ``hermitian conjugation". Four coef\/f\/icient functions $d_i$, $\bar d_i$ on the generators of the $\pm 1$ graded subspaces in $L^{\pm}$ are united to the ${3\over 2}$ multiplate, with respect to gauge transformation initiated by group elements $g_0(x)$, $\bar g_0(y)$ belonging to the f\/irst red group. The f\/irst fundamental representation of $G_2$ algebra is $7$-th dimensional with the basis vectors: \[ \hspace*{-5pt}\begin{array}{l} \ve{1}, \quad X^-_1\ve{1}, \quad X^-_2X^-_1\ve{1}, \quad X^-_1X^-_2X^-_1\ve{1}, \quad X^-_1X^-_1 X^-_2X^-_1\ve{1}, \vspace{2mm}\\ X^-_2X^-_1X^-_1X^-_2X^-_1\ve{1}, \quad X^-_1X^-_2X^-_1X^-_1X^+_2X^-_1\ve{1}. \end{array} \] The action of the operators $L^{\pm}$ on $A_1$ basis of $u$ matrix is as follows: \[ \hspace*{-5pt}\begin{array}{l} \displaystyle \vc{1} L^+=\vc{1} X^+_1X^+_2(\bar d_2 -\bar d_3 X^+_1+{1\over 2} \bar d_4 X^+_1X^+_1- \bar d^2 X^+_1X^+_1X^+_2), \vspace{3mm}\\ \displaystyle \vc{1} X^+_1L^+=\vc{1} X^+_1X^+_2(\bar d_1 - \bar d_2 X^+_1+{1\over 2} \bar d_3 X^+_1X^+_1-\bar d^2 X^+_1X^+_1X^+_2X^+_1). \end{array} \] The action of the operator $L^-$ on $A_1$ basis from the left may be obtained from the last formulae with the help of ``hermitian conjugation'': \[ L^-\ve{1}= (\vc{1} L^+)^T,\qquad L^-X^-_1\ve{1}= (\vc{1} X^+_1L^+)^T,\qquad \bar d\to d. \] As in the previous sections the result of calculation of the main determinant (\ref{MCC"}) it is possible to present in the form of the product of column operator on the line one applied to the highest vector $\vc{1} K \ve{1}^2$ of the $(2,0)$ representation of $G_2$ algebra (see Appendix~II). The line operator form is the following \[ \hspace*{-5pt}\begin{array}{l} \displaystyle \Bigr [d_2-d_3X^-_1+{1\over 2}d_4(X^-_1)^2-{1\over 4} d^2(2X^-_1X^-_2-3X^-_2X^-_1) X^-_1, \vspace{3mm}\\ \displaystyle \qquad d_1-d_2X^-_1+{1\over 2}d_3(X^-_1)^2-{1\over 4}d^2X^-_1(2X^-_1X^-_2-3X^-_2 X^-_1)X^-_1\Bigr], \end{array} \] where $X^-_i\equiv (X^-_i)_r$. Nonusual (compared with the previous examples) form of the coef\/f\/icient on $d^2$ term is the prise for $p=3$ in (\ref{RS}) in the case of $G_2$ algebra. The column operator is obtained from the line one with the help introduced above rules of the ``hermitian conjugation''. For rediscovering of last symbolical expression up to the form of usual $2\times 2$ matrix let us introduce two dimensional column vector $\bar q$ the result of the action of the column operator on the highest vector $\vc{1} K \ve{1}^2$ divided by itself. The same in the line case will be denoted as $q$. Explicit expressions for the line components of $q$ have the form: \begin{equation}\label{Q} \hspace*{-5pt}\begin{array}{l} \displaystyle q_1=\left(d_2+{1\over 3}d^2\alpha_{112}\right)- 2\left(d_3+{2\over 3}d^2\alpha_{12}\right)\alpha_1+\left(d_4+2\alpha_2d^2\right)\alpha_1^2, \vspace{3mm}\\ \displaystyle q_2=\left(d_1+{1\over 3}d^2\alpha_{1112}\right)- 2\left(d_2+{1\over 3}d^2\alpha_{112}\right)\alpha_1+ \left(d_3+{2\over 3}d^2\alpha_{12}\right)\alpha_1^2 \end{array} \end{equation} and with the help of ``hermitian conjugation'' corresponding expressions for the components for the column $\bar q$. The result of the action of line operator on the highest vector in connection with all said above is equal to the numerical line vector $\vc{1} K \ve{1}^2 (q_1, q_2)$. The action of the column operator on it may be devided on two steps: the action on the scalar factor $\vc{1} K \ve{1}^2 $, with the f\/inally matrix $\vc{1} K \ve{1}^2 \bar q q$ (multiplication by the law the column on the line) and the terms with partial mutual dif\/ferentiation of the scalar and the lines factors. All formulae for concrete calculation of such kind the reader can f\/ind in Appendix~II. It is necessary to pay attention to the fact, that $X^+_2 q_i=X^-_2 \bar q_i=0$, which one can check without any dif\/f\/iculties with the help of formulae of the Appendix~I. Gathering all these results we obtain the equation of equivalence for $u$ function: \begin{equation} u(u^{-1}u_x)_y={\det}^{-1}(u)\sum_{i,j,k,l} u_{ij} u_{kl} \epsilon_{ik} \epsilon_{jl}\bar p^{ik} p^{jl}+4 d^2\bar d^2(\mbox{Det}\, (u))^{-1} u,\label{BE} \end{equation} where $u_{ij}$ elements of the matrix $u$, $\epsilon_{ij}$ symmetrical tensor of the second rank with the components $\epsilon_{12}=\epsilon_{21}=-1$, $\epsilon_{11}=\epsilon_{22}=1$, $\bar p^{ij}$, $p^{ij}$ are two-dimensional column and line vectors correspondingly with the components (the law of multiplication is the column on the line): \[ \hspace*{-5pt}\begin{array}{l} \displaystyle p^{11}=\left(d_2+{1\over 3} d^2\alpha_{112},d_1+{1\over 3} d^2\alpha_{1112}\right), \qquad p^{22}=\left(d_4+2d^2\alpha_2,d_3+{2\over 3} d^2\alpha_{12}\right), \vspace{3mm}\\ \displaystyle p^{12}=p^{21}=\left(d_2+{1\over 3} d^2\alpha_{112},d_3+{2\over 3} d^2\alpha_{12}\right). \end{array} \] It remains only to f\/ind the derivatives $(\bar p_{ij})_x$, $(p_{kl})_y$ and convince ourselves that together with the (\ref{BE}) they compose the closed system of equations of equivalence or exactly integrable $G_2(0,1;2,2;\bar d^2,d^2)$ system. Four components of $p^{22}$, $p^{11}$ with respect to transformation of the f\/irst red algebra compose the ${3\over 2}$ spin-multiplet. So it will be suitable to redenote them by single four-dimensional symbol $p_i$. And the same for ``hermitian conjugating'' values $\bar p_i$. Let us follow the calculation of $(\bar p_4)_x=2\bar d^2 (\bar \alpha_2)_x$. The calculation of this the derivative do not dif\/ferent from the corresponding computations of Section~3 (see (\ref{AR}) and (\ref{ARR})). We have consequently: \begin{equation} (\bar \alpha_2)_x=\vc{2} K \ve{2})^{-2} \pmatrix{ \vc{2} K \ve{2}, & \vc{2} K L^- \ve{2} \cr \vc{2} X^+_2 K \ve{2}, & \vc{2} X^+_2 K L^- \ve{2} \cr}. \label{ARm} \end{equation} With the help of the technique used many times before we evaluate the last expression to: \[ \hspace*{-5pt}\begin{array}{l} (\bar \alpha_2)_x=L^-_r(X^+_2)_l \ln \vc{2} K \ve{2} =\left[d_1-d_2 X^-_1+{1\over 3} d_3 (X^-_1)^2\right. \vspace{3mm}\\ \displaystyle \qquad \left. -{1\over 6} d_4 (X^-_1)^3+ d^2([[[X^-_2,X^-_1]X^-_1]X^-_1]-X^-_2(X^-_1)^3)\right] \theta_2. \end{array} \] Using with respect to the last expression formulae of Appendix~I we come to the system of equations of equivalence for $\bar p$ components of ${3\over 2}$ multiplet: \begin{equation} \label{BLE} \hspace*{-5pt}\begin{array}{l} \displaystyle (\bar p_4)_x={2\bar d^2\over \mbox{Det}^2 (u)}\left(p_1 u_{11}^3-3p_2 u^2_{11} u_{12}+ 3p_3 u_{11} u^2_{12}-p_4 u^3_{12}\right), \vspace{3mm}\\ \displaystyle (\bar p_3)_x={2\bar d^2\over \mbox{Det}^2 (u)}\left(p_1 u_{11}^2 u_{21}-p_2 \left(u^2_{11} u_{22}+ 2u_{11} u_{21} u_{12}\right)\right. \vspace{3mm}\\ \displaystyle\phantom{(\bar p_2)_x=} \left. +p_3 \left(2u_{11} u_{12}u_{21}+u^2 _{12}u_{21}\right)-p_4 u^2_{12}u_{22}\right), \vspace{3mm}\\ \displaystyle (\bar p_2)_x={2\bar d^2\over \mbox{Det}^2 (u)} \left(p_1 u_{11} u^2_{21}-p_2 \left(u^2_{21} u_{12}+ 2u_{11} u_{21} u_{22}\right)\right. \vspace{3mm}\\ \displaystyle \phantom{(\bar p_2)_x=} \left. +p_3 \left(2u_{22} u_{12}u_{21}+u^2 _{22}u_{11}\right)-p_4 u^2_{22}u_{12}\right), \vspace{3mm}\\ \displaystyle (\bar p_1)_x={2\bar d^2\over \mbox{Det}^2 (u)}\left(p_1 u_{21}^3-3p_2 u^2_{21} u_{22}+ 3p_3 u_{21} u^2_{22}-p_4 u^3_{22}\right). \end{array} \end{equation} And corresponding system for derivatives $p_y$, which can be obtained from (\ref{BLE}) with the help of ``hermitian conjugation''. The symmetry of the constructed exactly integrable $G_2(0,1;2,2;\bar d^2,d^2)$ system (\ref{BE}), (\ref{BLE}) is higher than any possible espectations. >From the physical point of view this system may be considered as the interuction of charge ${3\over 2}$ spin particle ($p,\bar p$) with neutral scalar-vector f\/ield $u$. \subsubsection{(1,0) grading} In this case $L^{\pm}$ may contain the components $\pm 1$, $\pm 2$, $\pm 3$ graded subspaces and have the form: \[ \hspace*{-5pt}\begin{array}{l} L^+=\bar c_1 X^+_1+ \bar c_2 [X^+_1,X^+_2]+\bar c^2 [X^+_1[X^+_1,X^+_2]] \vspace{2mm}\\ \displaystyle \qquad + \bar c^3_1 [X^+_1[X^+_1[X^+_1,X^+_2]]]+\bar c^3_2 [X^+_2[X^+_1[X^+_1[X^+_1, X^+_2]]]], \end{array} \] $L^-=(L^+)^T$, where $T$ sign of transposition ($(X^+_i)^T=X^-_i)$; with simultaneously exchange of all coef\/f\/icients $\bar c\to c$. This operation was called as ``hermitian conjugation'' in the previous subsection and we conserve here this notation. As always we begin from the equation of equivalence for two dimensional matrix $u$ connected with the second simple root of $G_2$ algebra. For the decoded of universal equation~(\ref{MCC"}) it is necessary the knowledge of the action of $L^{\pm}$ on the basis. We represent below only part of basis vectors of the second fundamental ($14$-th dimensional) representation of $G_2$ algebra: \[ \hspace*{-5pt}\begin{array}{l} \ve{2}, \ X^-_2\ve{2}, \ X^-_1X^-_2\ve{2}, \ X^-_1X^1_2X^-_2\ve{2}, \ X^-_1X^-_1X^-_1X^-_2\ve{2}, \ X^-_2X^-_1X^-_1X^-_2\ve{2}, \vspace{2mm}\\ X^-_2X^-_1X^-_1X^-_1X^-_2\ve{2}, \ X^-_1X^-_2X^-_1X^-_1X^-_2\ve{2}, \ X^-_1X^-_2X^-_1X^-_1X^-_1X^-_2\ve{2}. \end{array} \] The main equations (\ref{I}) are obviously invariant with the respect to the gauge transformation iniciated by $g_0(x)$, $\bar g_0(y)$ elements of the red algebra of the second simple root. With respect to this transformations two coef\/f\/icients of zero $(c^1,\bar c^1)$ and third $(c^3,\bar c^3)$ order graded subspaces are transformed as spinor (anti-) multiplets; $c^2$, $\bar c^2$ are the scalar ones. With the help of such transformation it is always possible to satisfy the condition $c^3_2=\bar c^3_2=0$ (what is essential simplif\/ied the calculation) and reconstruct the general case at the f\/inal step using invariance condition. The action of the $L^{\pm}$ operators on the basis states of the second red algebra has the form: \[ \hspace*{-5pt}\begin{array}{l} \vc{2} L^+=\vc{2} X^+_2X^+_1(-\bar c^1_2 +\bar c^2 X^+_1-\bar c^3_1 X^+_1X^+_1 +\bar c^3_2 (2X^+_1X^+_1X^+_2-3X^+_1X^+_2X^+_1), \vspace{2mm}\\ \vc{2} X^+_2L^+=\vc{2} X^+_2X^+_1(\bar c^1_1 +\bar c^2 X^+_1X^+_2+ (X^+_1X^+_1X^+_2-3X^+_1X^+_2X^+_1)(\bar c^3_1-\bar c^3_2 X^+_2). \end{array} \] The action of the operator $L^-$ on $A_1$ basis from the left may be obtained from the last formulae with the help of ``hermitian conjugation'': \[ L^-\ve{2}= (\vc{2} L^+)^T,\qquad L^-X^-_1\ve{2}= (\vc{2} X^+_1L^+)^T,\qquad \bar c\to c. \] Taking into account arguments of the Appendix~II the result of the calculation of determinant of the third order (\ref{MCC"}) may be presented in the operator column on line form, acting on the highest vector of $(4,0)$ representation ($3 \vc{1} K \ve{1}^4$) of $G_2$ algebra. The line (``hermitian conjugating'' column) operators has the form (in this expression we put $c^3_2=\bar c^3_2=0$): \[ \left(-c^1_2 +c^2 X^-_1-{1\over 2}\bar c^3_1 (X^-_1)^2,\quad c^1_1 +c^2 X^-_2X^-_1+{1\over 8}c^3_1 (X^-_2X^-_1X^-_1-6X^-_1X^-_2X^-_1)\right). \] Further calculations are on the level of accurate application of dif\/ferentiation rules and combination terms of the same nature. Equation of equivalence for $u$ function have the f\/inal form: \begin{equation} u(u^{-1}u_x)y=3 \, {\det}^{{1\over 3}}(u) \bar p^1 p^1+12 \, {\det}^{-{1\over 3}}(u) \bar p^2 p^2 u+18 \,{\det}^{-1}(u) (u\bar c^3) (c^3 u), \label{HV} \end{equation} where $p^1$ is the spinor with the components $p^1=(-c^1_2+4c^2\alpha_1 -6c^3_1\alpha_1^2, c^1_1+4c^2 \alpha_{21}-c^3_1(\alpha_{121}+2\alpha_1\alpha_{21}))$; scalar $p^2= c^2-3c^3_1\alpha_1$ and corresponding expressions for bar values. We present the system of equivalence equations without any further comments: \begin{equation} (\bar p^2)_x=-3 \, {\det}^{-{2\over 3}} \sum_{i,j,k,l}\bar c^3_i u_{ij}\epsilon_{kl} p^1_l,\qquad (\bar p^1_i)_x={\det}^{-{2\over 3}}\bar p^2 \sum_{j,k,l} u_{ij} \epsilon_{k,l}p^1_l,\label{LLL} \end{equation} where $\epsilon_{k,l}=-\epsilon_{l,k}$ antisymmetrical tensor of the second rank $\epsilon_{1,2}=-\epsilon_{2,1}=1$. And, of, course the corresponding system with the derivatives $p^1_y$, $p^2_y$. Physical interpretation of the last system may be connected with spinor particle interacting with charged scalar $(p^2,\bar p^2)$ and neutral scalar-vector f\/ields in two dimensions. \section{Concluding remarks} In some sense in the present paper the initial idea of Sofus Lie to introduce continuous groups as powerful apparatus for solving the dif\/ferential equations is realized. On the examples of semisimple groups of second order we have decoded this idea and described explicitly exactly integrable systems whose general solutions can be obtained with the help and in the terms of group representation theory. We have no doubts (and partially can prove this) that the same construction is applicable to the case of arbitrary Lie groups and hope to prove this statement completely or to see the proof in the literature in the nearest future. \subsection*{Acknowledgements} Author is indebted to the Instituto de Investigaciones en Matem'aticas Aplicadas y en Sistemas, UNAM for beautiful conditions for his work. Author freundly thanks N.~Ata\-ki\-shiyev for permanent discussions in the process of working on this paper and big practical help. This work was done under partial support of Russian Foundation of Fundamental Researches (RFFI) GRANT. \renewcommand{\theequation}{{\rm I}.\arabic{equation}} \setcounter{equation}{0} \section*{Appendix~I} The formulae below are the general ones and have in their foundation the f\/irst Jacobi identity only. Let us def\/ine: \[ \theta_j=\prod_{i=1}^r (\vc{i} G \ve{i})^{-K_{ji}}. \] As a result of dif\/ferentiation of $\ln \theta_i$, we obtain: \begin{equation} (X^-_q)_r \theta_i=-\theta_i K_{iq} \alpha_q,\qquad (X^+_q)_l \theta_i= -\theta_i K_{iq} \bar \alpha_q, \label{AI1} \end{equation} \begin{equation} (X^-_q)_r \bar \alpha_i=\delta_{q,i} \theta_i,\qquad (X^+_q)_l \alpha_i= \delta_{q,i} \theta_i. \label{AI2} \end{equation} In the case of the second order algebras: \begin{equation} \theta_1={\vc{2} G \ve{2}\over \vc{1} G \ve{1}^2}, \qquad \theta_2={\vc{1} G \ve{1}^p\over \vc{1} G \ve{1}^2}.\label{AI3} \end{equation} \renewcommand{\theequation}{{\rm II}.\arabic{equation}} \setcounter{equation}{0} \section*{Appendix II} Let us consider the determinant of the third order the matrix entiries of which are coinsided with the matrix elements of $G_2$ group element $K$ taken between the bra and the ket three dimensional bases: \begin{equation} \vc{1},\ \vc{1} X^+_1,\ \vc{1} X^+_2X^+_1X^+_1X^+_2X^+_1,\ \ve{1}, \ X^-_1\ve{1}, \ X^-_2X^-_1X^-_1X^-_2X^-_1\ve{1}.\label{AII} \end{equation} Acting on such determinant by generator $(X^+_2)_r$ and taking $(X^-_1X^-_1)_r$ out of its sign we come to the following ket basis: \[ \ve{1}, X^-_1\ve{1}, X^-_2X^-_1\ve{1} \] which in connection with the (\ref{GJI}) tell us that the initial $\mbox{Det}_3$ (up to the terms anihilated by generators of the positive simple roots from right and negative ones from the left) belongs to $(2,0)$ ($Vh_1=2$, $Vh_2=0$) representation of $G_2$ group. For initial determinant $Vh_1=1$, $Vh_2=0$. Each basis vector (see (\ref{high})) may be obtained with consequent application of the lowering operators to the higest vector ($\vc{1} K \ve{1}^2$ in the present case). There are two possibility to combination of the lowering operators: \[ ( (AX^-_2X^-_1+BX^-_1X^-_2)X^-_1 )_r \] and the same expression from the left combination of the raising generators. The condition that $\mbox{Det}_3$ is anihilated by generators $(X^+_1)_r(X^-_1)_l$, which is a direct corollary of the structure of the bra and ket basises, allow to f\/ind relation between the constants $3A+2B=0$ and obtain the expression used in the main text (\ref{Q}) and above. We obtain the following value for $\mbox{Det}_3$ in basis (\ref{AII}): \[ \hspace*{-5pt}\begin{array}{l} \displaystyle \mbox{Det}_3={1\over 16}((2X^-_1X^-_2-3X^-_2X^-_1)X^-_1) \vspace{3mm}\\ \displaystyle \phantom{\mbox{Det}_3=} \times (X^+_1(2X^+_2X^+_1- 3X^+_1X^+_2)) \vc{1} K \ve{1}^2+\vc{1} K \ve{1}. \end{array} \] Below we present necessary formulae for calculation of (\ref{BE}). We restrict ourselves by (11) component of it. All ``mixed'' terms may be gothered in the following form: \begin{equation} \hspace*{-5pt}\begin{array}{l} \displaystyle -{\bar d^2\over 4}\left[2(X^+_1X^+_2X^+_1 q_1)-3(X^+_2X^+_1X^+_1 q_1)-8\bar \alpha_1(X^+_2X^+_1 q_1)-8\bar \alpha_1 (X^+_1 q_1)\right] \vspace{3mm}\\ \displaystyle \qquad - \bar d_3 (X^+_1 q_1)+2\bar d_4 \bar \alpha_1(X^+_1 q_1)+{\bar d_4\over 2} ((X^+_1)^2 q_1). \end{array} \label{AII1} \end{equation} Using the def\/inition of vector $q$ (\ref{Q}) and formulae of Appendix~I, we obtain: \[ \hspace*{-5pt}\begin{array}{l} (X^+_1 q_1)=2\theta_1(p^{22}_1\alpha_1-p^{22}_2)\equiv 2\theta_1 P,\quad X^+_2 P=0,\quad (X^+_2X^+_1 q_1)=2\theta_1 \bar \alpha_2 P, \vspace{2mm}\\ (X^+_1X^+_1 q_1)=2\theta_1^2 p^{22}_1-4\theta_1 \bar \alpha_1 P,\quad (X^+_1X^+_2X^+_1 q_1)=2\theta_1(\bar \alpha_{21}-2 \bar \alpha_1 \bar \alpha_2 )P+2\theta_1^2 \bar \alpha_2 p^{22}_1, \vspace{2mm}\\ (X^+_2X^+_1X^+_1 q_1)=4\theta_1^2 \bar \alpha_2 p^{22}_1+4d^2 \theta_1^2 \theta_2-4\theta_1 \bar \alpha_1 \bar \alpha_2 P-4\theta_1 \bar \alpha_{12} P. \end{array} \]
\section{Introduction} It is well-known that crystal fields present in real materials can influence their magnetic behavior considerably.\cite{CFbooks} In this paper we study a ferromagnetic insulator, with localized interacting magnetic moments, which are also subject to a crystal field of cubic symmetry. The finite-temperature critical phenomena in these systems have been investigated previously within the Landau-Ginzburg-Wilson theory,\cite{AharonyKleinert} and the cubic universality class and its critical exponents have been analyzed. However, in such phenomenological theories, the form of the order parameter is simply postulated, using mainly symmetry arguments. To see whether such an order really develops in the low-temperature regime one must start with the microscopic quantum Hamiltonian. In terms of a quantum spin Hamiltonian, the magnetic moments are described by spin operators, and the crystal field takes the form of a single-site term. The lowest value of $S$ for which a cubic anisotropy term has any effect is $S=2$. Using this value our model Hamiltonian takes the form \begin{equation} \label{model} H = - \sum_{\langle ij\rangle} \bbox{S}_{i} \bbox{S}_{j} - D \sum_{i} \left[ (S^{x}_{i})^{4} + (S^{y}_{i})^{4} + (S^{z}_{i})^{4} \right], \end{equation} with $S_{i}^{\alpha}$ ($\alpha=x,y,z$) denoting spin operators for $S=2$ at lattice site $i$. The first term is the isotropic, ferromagnetic exchange interaction between nearest neighbors, and the second term represents the crystal field anisotropy; for $S=2$ this is the most general single-site operator with cubic symmetry. The exchange term alone produces ferromagnetic order in the low-temperature phase; the question we want to address is how this order is affected by the cubic crystal field. The above Hamiltonian has been treated within the mean-field approximation (MFA) by many authors.\cite{SznajdValkova} It has been found that the sign of the crystal-field constant~$D$ defines the easy axes of the magnetic ordering at low temperatures. For $D=0$ the system is isotropic and the spontaneous magnetization can lie along any direction; for $D < 0$ the cube diagonals $[111]$ are preferred; for $0 < D < (4z)/3$, where $z$ is the number of nearest neighbors, the directions parallel with the cube edges $[100]$ are chosen. Moreover, the MFA predicts that for $D > (4z)/3$ the system is disordered, irrespective of the temperature. Recently, two of us (M.D.\ and J.S.) have investigated the above model,\cite{OurRS} applying the Raleyigh-Schr\"{o}dinger perturbation theory in the limit $D \rightarrow \infty$. This calculation is valid for large $D$, and the following picture emerges from it. In agreement with the MFA, for large $D$, there is no magnetic order at any temperature, i.e., the averages of the spin operators vanish: \[ \langle S_{i}^{\alpha} \rangle = 0, \hspace*{1em} \text{for} \hspace*{1em} \alpha=x,y,z. \] However, in contrast to the MFA prediction, the symmetry may still be spontaneously broken by the appearance of a purely quadrupolar order, described by the averages \[ \langle (S_{i}^{x})^2 \rangle \neq \langle (S_{i}^{y})^2 \rangle = \langle (S_{i}^{z})^2 \rangle. \] (One may interchange $x$, $y$, and $z$ in the above relation.) This purely quadrupolar phase was predicted to exist in the model on a two-dimensional lattice at least at zero temperature. In three dimensions\cite{OurCB} and in higher dimensions it should be present both in the ground state and at finite (but low) temperatures. We suppose, that the quadrupolar phase also exists in one dimension at zero temperature (note that the broken symmetry is not continuous but discrete), although in this case the perturbation-theory arguments are weaker. The result that there may appear a purely quadrupolar order in the model in Eq.\ (\ref{model}), which contains only the bilinear, ``magnetic'' exchange and the single-site term, is rather surprising. Usually, higher order exchanges are responsible for quadrupolar ordering.\cite{SivardiereMorin} The appearance of quadrupolar order in the present case is entirely a quantum effect; one would not obtain purely quadrupolar order by simply replacing the spin operators in Eq.\ (\ref{model}) with classical vectors, and then seeking the configuration with minimum energy. The result is also of methodological interest. It is easy to understand that one cannot get purely quadrupolar order in this model by the MFA: only magnetization can act as a mean field when the exchange term contains only first powers of the spin operators. Therefore, if the purely quadrupolar phase exists, we have the example that, in contrast with the general belief, the MFA may give an answer which is qualitatively wrong in high dimensions. Unfortunately, there is no fully reliable method which would enable us to investigate the model in two or three dimensions, for general $D$. Therefore, the aim of the present paper is to check and confirm the existence of the purely quadrupolar phase numerically at least in the one-dimensional situation at zero temperature. We use White's density-matrix renormalization group (DMRG) method,\cite{Whiteoriginal} which has a proven record of extreme efficiency in dealing with such problems. The method allows us to study the model for general $D \geq 0$, not only in the limit of large $D$ as in Ref.\ \onlinecite{OurRS}, and we obtain the zero temperature phase diagram, together with some properties of the (quantum) phase transition separating the magnetic and quadrupolar phases. The paper is organized as follows: we begin with summarizing the information obtained from the MFA and the perturbation theory in Sec.\ \ref{sec2}. Then we discuss how long-range order can be detected using the DMRG in Sec.\ \ref{sec3}. Section \ref{sec4} is devoted to the concrete numerical results, while Sec.\ V is a brief discussion of our conclusions. We also attach an Appendix in which numerical predictions, obtained using identical techniques, are confronted with exact results in the test case of the 1D Ising model in a transverse field. \section{Predictions from MFA \protect\\ and perturbation theory} \label{sec2} Let us first list the eigenvalues $e_{k}$ and eigenstates $| \psi_{k} \rangle$ of the single-site term in the Hamiltonian in Eq.\ (\ref{model}) \begin{eqnarray} e_{1}=-24 D & \hspace*{2.5em} & | \psi_{1} \rangle = \frac{1}{\sqrt{2}} (| 2 \rangle + | -2 \rangle), \nonumber \\ e_{2}=-24 D & \hspace*{2.5em} & | \psi_{2} \rangle = | 0 \rangle, \nonumber \\ e_{3}=-18 D & \hspace*{2.5em} & | \psi_{3} \rangle = | 1 \rangle, \nonumber \\ e_{4}=-18 D & \hspace*{2.5em} & | \psi_{4} \rangle = \frac{1}{\sqrt{2}} (| 2 \rangle - | -2 \rangle), \nonumber \\ e_{5}=-18 D & \hspace*{2.5em} & | \psi_{5} \rangle = -| -1 \rangle, \label{eigenstates} \end{eqnarray} where by $| 2 \rangle, | 1 \rangle, \ldots, | -2 \rangle$ we denote eigenstates of $S^{z}$. The five states are split into a doublet and a triplet, and the doublet energy is lower for $D > 0$. The doublet is {\em nonmagnetic}, i.e., \begin{equation} \label{nonmagneticproperty} \langle \psi_{i} | S^{\alpha} | \psi_{j} \rangle = 0, \end{equation} for any $i,j=1,2$ and $\alpha=x,y,z$. Clearly, the crystal field with $D > 0$ opposes the magnetic ordering, since the nonmagnetic states are preferred in the total wave function of the system. For $D=0$ the model is isotropic and possesses the classical, ferromagnetic ground state with saturated magnetization. As mentioned in the Introduction, the MFA predicts \cite{SznajdValkova} that the magnetic phase exists for $0 < D < 8/3$ ($z=2$ for the chain), and that the directions of the cube edges $[100]$ are the easy axes in this phase. According to the MFA, the ground-state magnetization decreases with $D$ and disappears continuously at $D=8/3$. The MFA result that the magnetic order vanishes at some finite $D$ is a strong argument that this is indeed the case, since the MFA usually overestimates the tendency towards ordering. Nevertheless, the actual value of $D$ at which the magnetic phase ends may be much smaller than $8/3$ (for example, in the Ising chain in a transverse field the MFA critical field is two times larger than the exact critical field). Whether the actual phase transition is continuous or discontinuous is also a matter of question. We believe that the easy axes in the magnetic phase are given correctly by the MFA. This approximation neglects the quantum fluctuations in the ground state, but it seems very unlikely that these could change the easy-axis directions, especially in the vicinity of $D=0$, where the magnetization is almost saturated and the quantum fluctuations are small. Considering the large-$D$ limit now, we make use of the perturbation theory developed in Ref.\ \onlinecite{OurRS}. Since the two nonmagnetic states $| \psi_{1} \rangle$ and $| \psi_{2} \rangle$ dominate for large $D$, the model reduces to an effective two-state Hamiltonian. Restricting our attention to the one-dimensional situation from now on, the effective Hamiltonian $H_{\rm eff}$ can be written in terms of Pauli matrices $\sigma_{i}^{\alpha}$ acting on the states $| \psi_{1} \rangle$ and $| \psi_{2} \rangle$, and has the form\cite{OurRS} \begin{eqnarray} H_{\rm eff} & = & -\tilde{J_{1}} \sum_{i} \left( \sigma_{i}^{x} \sigma_{i+1}^{x} + \sigma_{i}^{z} \sigma_{i+1}^{z} \right) -\tilde{J_{2}} \sum_{i} \left( \sigma_{i}^{y} \sigma_{i+1}^{y} \right) \nonumber\\ && -\tilde{J_{3}} \sum_{i} \left( \sigma_{i}^{x} \sigma_{i+2}^{x} + \sigma_{i}^{z} \sigma_{i+2}^{z} \right) -\tilde{J_{4}} \sum_{i} \left( \sigma_{i}^{z} \sigma_{i+1}^{z} \sigma_{i+2}^{z} \right. \nonumber\\ && \left. \quad - \sigma_{i}^{x} \sigma_{i+1}^{x} \sigma_{i+2}^{z} - \sigma_{i}^{x} \sigma_{i+1}^{z} \sigma_{i+2}^{x} - \sigma_{i}^{z} \sigma_{i+1}^{x} \sigma_{i+2}^{x} \right)\nonumber\\ &&+ {\cal O}(\frac{1}{D^4}), \label{Heff} \end{eqnarray} \narrowtext with \begin{eqnarray*} \tilde{J_{1}} & = & \frac{1}{2 D} -\frac{1}{12 D^{2}} +\frac{1}{72 D^{3}} + {\cal O} \left( \frac{1}{D^{4}} \right), \\ \tilde{J_{2}} & = & \frac{1}{8 D^{2}} +\frac{1}{32 D^{3}} + {\cal O} \left( \frac{1}{D^{4}} \right), \\ \tilde{J_{3}} & = & \frac{1}{18 D^{3}} + {\cal O} \left( \frac{1}{D^{4}} \right), \\ \tilde{J_{4}} & = & \frac{1}{24 D^{3}} + {\cal O} \left( \frac{1}{D^{4}} \right). \end{eqnarray*} (The exchange coupling constant $J$ from Ref.\ \onlinecite{OurRS} is set to $1$ in the present work.) The quadrupole operators $[(S_{i}^{\alpha})^{2}-2]$ of the original model in Eq.\ (\ref{model}) turn out to be represented by linear combinations of Pauli matrices as\cite{OurRS} \begin{eqnarray} [(S_{i}^{x})^{2}-2] \hspace*{0.8em}& \longmapsto &\hspace*{0.8em} \sqrt{3} \sigma_{i}^{x} - \sigma_{i}^{z} + {\cal O}(1/ D^2), \nonumber \\ \protect [(S_{i}^{y})^{2}-2] \hspace*{0.8em}& \longmapsto &\hspace*{0.8em} -\sqrt{3} \sigma_{i}^{x} - \sigma_{i}^{z} + {\cal O}(1/ D^2), \nonumber \\ \protect [(S_{i}^{z})^{2}-2] \hspace*{0.8em}& \longmapsto &\hspace*{0.8em} 2 \sigma_{i}^{z} + {\cal O}(1/ D^2). \label{Ssigma} \end{eqnarray} The analogous representations of $S_{i}^{\alpha}$ and $(S_{i}^{\alpha} S_{i}^{\beta} + S_{i}^{\beta} S_{i}^{\alpha})$, where $ \alpha \neq \beta $, vanish up to ${\cal O}(1/D^{3})$. In the lowest order in $1/ D$, only $\tilde{J_{1}}$ is present in the effective Hamiltonian, and $H_{\rm eff}$ describes the planar model with continuous SO(2) symmetry in the $XZ$ plane. While this model, conventionally written as the $XY$ model, is ordered at zero temperature in two dimensions and above, in one dimension it is known to be critical due to enhanced quantum fluctuations.\cite{XY,Bar-McC} The in-plane correlation function \[ \langle \sigma_{i}^{x} \sigma_{i+l}^{x} \rangle = \langle \sigma_{i}^{z} \sigma_{i+l}^{z} \rangle \] decays asymptotically as a power law $1/\sqrt{l}$, where $l$ is the distance between the two sites. Hence, on the basis of Eq.\ (\ref{Ssigma}), we argue that for $D \rightarrow \infty$ a critical state, in which the quadrupolar correlation function \[ \left\langle \left[(S_{i}^{\alpha})^{2}-2\right] \left[(S_{i+l}^{\alpha})^{2}-2\right] \right\rangle, \qquad \alpha=x,y,z, \] decays asymptotically as $1/\sqrt{l}$, is approached. Taking into account the higher-order terms in $H_{\rm eff}$, we obtain the planar, $XY$-type model with small perturbations. The three-spin term with the coupling constant $\tilde{J_{4}}$ is crucial, because it reduces the symmetry of the effective Hamiltonian from the full rotational symmetry in the $XZ$ plane to the symmetry of discrete rotations through angles $\pm (2 \pi)/3$. In the one-dimensional case the $\tilde{J_{2}}$ and $\tilde{J_{3}}$ terms are expected to be marginally irrelevant,\cite{Nomura} but in the lack of a rigorous study of the off-diagonal three-point symmetry-breaking term with $\tilde{J_{4}}$, we can only speculate (and then eventually check numerically) whether the ${\cal O}(1/D^3)$ contributions are relevant or irrelevant. If they are relevant, they can stabilize the ``magnetic order'' (i.e., the nonvanishing averages of $\sigma_{i}^{\alpha}$) in the effective model $H_{\rm eff}$, since the Mermin-Wagner and Coleman's theorems do not hold when the symmetry is not continuous but discrete. If this is indeed the case, then Eq.\ (\ref{Ssigma}) implies that for large $D$ a quadrupolar order, associated with the operators $[(S_{i}^{\alpha})^{2}-2]$, exists in the original model of Eq.\ (\ref{model}) (see Ref.\ \onlinecite{OurRS} for details). Alternatively, if all the perturbations turn out to be irrelevant, there should be a critical phase for large $D$. To summarize the above, we present the relevant (order) parameters to be analyzed as suggested by the MFA and the perturbation theory: \begin{eqnarray} m^{\alpha} = \langle S^{\alpha}_{i} \rangle, \label{defm} \qquad q^{\alpha} = \left\langle (S^{\alpha}_{i})^{2} - 2 \right\rangle. \label{defq} \end{eqnarray} These ground-state averages are bulk values in the thermodynamic limit. Note, that in a disordered state $m^{\alpha}=q^{\alpha}=0$ for $\alpha=x,y,z$. Based on the above analysis we expect that for $0<D<D_m$ (where $D_m$ is probably smaller than the MFA value $8/3$) a magnetic phase with \begin{equation} \label{magneticpar} m^{x} \neq 0, \hspace*{1em} m^{y} = m^{z} = 0, \hspace*{1.7em} q^{x} > q^{y} = q^{z}, \end{equation} exists, while for $D>D_m$ a purely quadrupolar phase with \begin{equation} \label{quadrupolarpar} m^{x} = m^{y} = m^{z} = 0, \hspace*{1.7em} q^{x} > q^{y} = q^{z} \end{equation} emerges. (The indices $x,y,z$ in the above descriptions of the phases may be interchanged.) In both phases the original cubic symmetry of the Hamiltonian is spontaneously broken. Note however that the subgroup remaining invariant is different in the two cases. As discussed above, for large $D$ a critical (disordered) phase with $q^{\alpha}=0$ cannot be excluded either. In any case, the values of $q^{\alpha}$ must go to zero for $D \rightarrow \infty$. \section{The method} \label{sec3} The density-matrix renormalization group (DMRG) method is one of the most reliable numerical techniques to study one-dimensional quantum lattice problems. For a detailed description of the algorithm see Ref.\ \onlinecite{Whiteoriginal}. There are different implementations of the technique; nevertheless it seems that the most accurate results can be obtained by investigating finite systems with open boundary conditions. In this setup the DMRG provides good approximations for the ground state and low-lying excited states of long, but finite quantum chains. This can then be supplemented by a detailed finite size scaling analysis using appropriate scaling assumptions. Usually for the case of spontaneous symmetry breaking the existence of long-range order is checked by studying appropriate two-point correlation functions $\langle A_n A_{n^\prime}\rangle$. The direct observation of one-point functions $\langle A_n \rangle$ is apparently hindered by the fact that in finite systems, where tunneling is always finite between different vacua, the ground state is a symmetric combination of all possible ordering directions, and thus naively $\langle A_n \rangle=0$. A possible way out, however, is to break the symmetry artificially by a (small) auxiliary field which then forces the system to make a choice between the {\it a priori} undetermined directions, without noticably changing the value of the order parameter. This method has considerable advantages in the DMRG numerics where the measurement of two-point functions is rather awkward due to the open boundary condition. In our implementation the symmetry-breaking auxiliary fields are only applied to the first and last spins of the open chain. These two {\em boundary fields} are chosen to be identical, leaving the system symmetric for $i\to L-i$ reflections ($L$ is the chain length). Then the Hamiltonian which is actually simulated in the numerical calculations is \begin{equation} H \rightarrow H - {\bf h}\cdot \left( {\bf B}_1 +{\bf B}_L\right) \label{bf-setup} \end{equation} where ${\bf B}_i$ and ${\bf h}$ are vectors of appropriate single-site operators and boundary fields. Since the system is not translation invariant, the one-point function $\langle A_n \rangle$ depends on the position $n$ near the chain ends, but approaches its bulk value in the middle of the system. If the system size is much larger than the correlation length the one-point function rapidly saturates and forms a plateau. Alternatively, if the chain length is too small comparing to the correlation length (e.g., the system is critical or it is close to criticality) no plateau appears to develop. There are two ways to estimate the order parameter and the associated correlation length from one-point functions. One can define the value $a_{L/2}\equiv \langle A_{L/2} \rangle$ measured in the middle of the chain and study its dependence on the total chain length $L$. It is expected that $a_{L/2}(L)\to a$ (with $a$ the bulk value of the order parameter) exponentially fast as $L\to\infty$, whenever the system is away from criticality, while the convergence is only algebraic at criticality. A disadvantage of this method is that many independent DMRG runs must be done with different $L$ values, or one is forced to use the ``infinite lattice'' algorithm which is less precise. Alternatively, one can analyze the {\em profiles} $a_n\equiv \langle A_n \rangle$ as a function of $n$ at a fixed length $L$, providing that $L$ exceeds the correlation length $\xi$ reasonably (which can be checked {\it a posteriori}). In this latter case the profile is expected to have the following asymptotic form far from the chain ends \begin{equation} \label{fittingformula} a_n = a + c \frac{e^{-n / \xi}}{n^\chi} + c \frac{e^{-(L+1-n) / \xi}}{(L+1-n)^{\chi}} , \end{equation} where $a$ is the bulk value of the order parameter, $\xi$ is the associated correlation length, $\chi$ is a suitable exponent, and $c$ is a constant. In general one can assume that the order parameter and the correlation length measured this way are identical to their bulk values derived from two-point correlation functions. However, the exponent of the polynomial prefactor $\chi$ is normally different from its bulk value and may depend on the boundary conditions chosen. In practice we analyze our numerical data by dividing the chain into smaller segments with a length of the order of $\xi$, and perform local least-squares fits using the formula in Eq.\ (\ref{fittingformula}). We have found it appropriate to fix $\chi$ first and then look for its value which makes the other fitting parameters the most stable in different segments as $n\to\infty$. Our procedure of determining order parameters and correlation lengths was extensively tested on a simpler problem, the 1D Ising model in a transverse field (ITF), where exact answers to many questions of interest are available. We summarize our experience with this model in the Appendix. In most cases the DMRG is very efficient to calculate spectral gaps. The method is less useful, however, if the ground state is (asymptotically) highly degenerate, since then several states have to be targeted together, resulting in a rapid drop in accuracy. In the cubic model under investigation we expect a 6-fold ground-state degeneracy in the magnetic phase, and a 3-fold degeneracy in the quadrupolar phase, in accordance with the number of easy-axis directions for the two order parameters. The 6-fold degenerate case already exceeded our computational limitations, thus some of our conclusions with respect to that case are based on short chain exact diagonalization results only. Errors in the quantitative predictions we report in the next sections stem from two sources, either from the limited accuracy of the numerical technique we have been using, or from the approximate (asymptotic) nature of the formulas we applied for fitting and extrapolating the data. As for the DMRG errors, our results are somewhat more precise when the ``finite lattice'' algorithm was used. In this case several iterations were done until convergence was reached. In principle, results obtained through the ``infinite lattice'' method inevitably contain a systematic ``environment'' error, and thus should be treated with less confidence. However, in the present problem improvements due to the ``finite lattice'' iterations turned out to be rather small, providing enough ground to assume that even our ``infinite lattice'' results have sufficient accuracy. In general we made various runs keeping different number of states $M$, and extrapolated our results for $M\to\infty$, or at least checked that the results had been converged in this parameter. Typically, we found good convergence in $M$ despite the fact that the maximum number of states kept did not exceed $M=85$. Note that the number of degrees of freedom at a single site is five in the $S=2$ case, and the model in question has no continuous axial symmetry (total $S^z$ is not conserved) which could have been utilized to facilitate the DMRG calculation in the usual way. \section{Numerical results} \label{sec4} In order to obtain the zero temperature phase diagram and critical properties of the model we carried out extensive numerical calculations using exact diagonalization techniques on short chains and the DMRG method on longer systems. In the case of open boundary conditions with boundary fields the actual form of the Hamiltonian used in the simulations was \begin{eqnarray} \label{actmodel} H &=& - \sum_{i=1}^{L-1} \bbox{S}_{i} \bbox{S}_{i+1} - \sum_{i=1}^L D_i \left[ (S^{x}_{i})^{4} + (S^{y}_{i})^{4} + (S^{z}_{i})^{4} \right] \nonumber\\ &&-{\bf h}\cdot {\bf (B_1+B_L)} \end{eqnarray} with a reduced cubic crystal field at the chain ends (to counterbalance the missing bonds for the first and the last spins) \begin{equation} D_i = \left\{ \begin{array}{ll} D & {\rm if}\; i=2,\dots,L-1 \\ D/2 & {\rm if}\; i=1,L \end{array} \right. \end{equation} and a general symmetry-breaking boundary field containing magnetic and quadrupolar terms \begin{equation} {\bf B} = \left( \begin{array}{c} S^x\\S^z\\(S^x)^2\\(S^y)^2\\(S^z)^2 \end{array} \right), \qquad {\bf h} = \left( \begin{array}{c} h_x\\h_z\\h_{xx}\\h_{yy}\\h_{zz} \end{array} \right). \label{Bh} \end{equation} In order to avoid handling complex numbers, we did not include a boundary term proportional to $S^y$. After the ground state had been found, we measured the order parameters. To facilitate further discussion we introduce the notations \begin{eqnarray} m_{n}^{\alpha} = \langle S_{n}^{\alpha} \rangle, \qquad q_{n}^{\alpha} = \langle (S_{n}^{\alpha})^{2}-2 \rangle \label{defmq} \end{eqnarray} with $n=1,\dots,L$; $\alpha=x,y,z$. \subsection{Ground state degeneracies and gaps} Figure \ref{fig:degen}(a) shows the lowest energy levels in a periodic chain with $L=8$ sites obtained using exact diagonalization techniques. Although the chain is short, the 6-fold degeneracy, consisted of a singlet, a doublet, and a triplet is clearly discernible for small values of $D$. In the large $D$ (quadrupolar) region the degeneracy seems to be 3-fold, consisted of the singlet and the doublet only. In both regimes we observe signs of a gap above these states, although the number of data points ($L=4,6,8$) available for a given $D$ did not allow us to perform a detailed finite-size scaling study. Note that since the broken symmetry is discrete we do not expect any massless Goldstone modes in the ordered regimes. Excited states are rather massive domain walls, which separate domains with different ordering directions, and propagate in the system. \begin{figure}[hbt] \epsfxsize=\columnwidth\epsfbox{fig1b.eps} \caption{(a) Low-energy spectrum vs cubic anisotropy $D$ in a short chain with $L=8$ using periodic boundary condition. Shading schematically indicates higher-energy states. (b) Ground state--first excited state (singlet--doublet) gap (1st XS), and ground state--second excited state gap (2nd XS) as a function of $1/L$ (solid lines) and $1/L^2$ (dashed lines) at $D=1.3$. The ticks in the horizontal axes correspond to $L=10,20,\ldots,50$. \label{fig:degen} } \end{figure} As was discussed in the former section, we could only apply the DMRG for the 3-fold degenerate case. We calculated the low-lying four states in an open chain at $D=1.3$, which is above the critical point $D_m$ (see later sections), by targeting four states without any boundary fields. The results, presented in Fig.\ \ref{fig:degen}(b), strongly support the 3-fold asymptotic degeneracy with a finite gap $\Delta\approx 0.08\pm 0.02$ above it in the thermodynamic limit. Unfortunately, this calculation was rather time consuming, which impeded us to repeat it for other $D$ values. \subsection{Magnetic order} \label{sec4B} Let us now investigate the spontaneous magnetization and the corresponding correlation length. In these calculations, purely magnetic boundary fields were used, i.e., $h_{xx}=h_{yy}=h_{zz}=0$ in Eq.\ (\ref{Bh}). \begin{figure}[hbt] \epsfxsize=\columnwidth\epsfbox{fig2b.eps} \caption{(a) The parameter $m^{x}_{L/2}$ vs the chain length $L$, and (b) the profile $m^{x}_n$ vs the position $n$ at $L$ fixed, for some representative values of $D$. Purely magnetic boundary fields applied along the $x$ axis as given by $h_{x}$. \label{fig:magn} } \end{figure} Expecting the magnetic easy axes along the cube edges we started with applying a boundary field along the $x$ axis setting $h_{x}=10$ or $0.1$, $h_{z}=0$. Due to the symmetry, with such a boundary field one must obtain $m_{n}^{x} \neq 0$ and $m_{n}^{y}=m_{n}^{z}=0$. Figure \ref{fig:magn}(a) shows how the parameter $m_{L/2}^{x}$, measured in the center of the chain, changes with the total chain length $L$ for representative values of the crystal-field strength $D$. One can see that for $D=0.5,1,1.2$ the parameter $m_{L/2}^{x}$ converges to finite values independent of $h_{x}$. This clearly indicates a finite magnetic order for these $D$ values. On the other hand, for $D=1.3,1.5,2$ the parameter $m_{L/2}^{x}$ goes to zero as $L\to\infty$, meaning no magnetic order. As characteristic examples the magnetization profiles $m_{n}^{x}$ with $L$ fixed are shown in Fig.\ \ref{fig:magn}(b) for $D=1$ and $2$. For $D=1$ a plateau corresponding to a finite spontaneous magnetization $m^{x}\approx 1.76$, while for $D=2$ a plateau with $m^{x}=0$ can be observed. So far we have assumed that the cube edges give the easy axes in the magnetic phase. To verify this assumption, we also applied boundary fields which do not coincide with the expected easy directions. Choosing $h_{x}=0.04$ and $h_{z}=0.03$ we found that for $L\to\infty$ the parameter $m_{L/2}^{x}$ always tends to the same value as found above, while $m_{L/2}^{z}$ goes to zero, in agreement with Eq.\ (\ref{magneticpar}). \begin{figure}[hbt] \epsfxsize=\columnwidth\epsfbox{fig3b.eps} \caption{The spontaneous magnetization $m$ (left axis, dots) and the corresponding correlation length $\xi_{m}$ (right axis, triangles) as functions of $D$. Notice the large error bars for $\xi_{m}$ in the magnetic phase. \label{fig:magnD} } \end{figure} The above calculations allowed us to determine how the bulk spontaneous magnetization $m$ changes with the crystal-field strength $D$. We assume that monotonic dependencies of $m^{\alpha}_{L/2}$ on $L$, like those in Fig.\ \ref{fig:magn}(a), provide correct bounds for $m$, as was found for the ITF chain in the Appendix. The magnetization $m$ is plotted in Fig.\ \ref{fig:magnD} as a function of $D$, together with the corresponding correlation length $\xi_m$, discussed in details below. Since $\xi_m$ increases around the phase transition, very long chains have to be studied in order to find the large-$L$ limit of $m_{L/2}^{x}$ in this region. The largest $D$ for which we can see the magnetic order is $D=1.23$. For this $D$, taking $L=400$, we obtain a finite $m$ with an uncertainty of $0.003$ (for smaller $D$, shorter chains are sufficient to calculate $m$ much more precisely). Due to the increasing correlation length, and thus computational limitations, we were not able to trace the dependence of $m$ on $D$ further. Starting from $D=1.26$, the phase with no magnetic order is observed. Although it cannot be shown directly that the spontaneous magnetization disappears continuously at the phase-transition point, the sharp increase of the correlation length suggests that the transition is continuous. We therefore attempt to fit the dependence of $m$ on $D$, shown in Fig.\ \ref{fig:magnD}, assuming a power law singularity and a linear regular term \begin{equation} \label{fitm} m(D) = c_1 (D_m-D)^{\beta} + c_2 (D_m-D). \end{equation} The fit is very good and we obtain \begin{eqnarray*} D_m = 1.2374 \pm 0.0004,\qquad \beta = 0.127 \pm 0.004, \end{eqnarray*} $c_1 = 2.15 \pm 0.05$, $c_2 = -0.17 \pm 0.05$. Interestingly, the exponent $\beta$ is very close to that of the two-dimensional Ising model. The obtained parameter $D_m$ is our best estimate of the phase-transition point. To calculate the magnetic correlation length $\xi_{m}$, we analyze the profiles of $m_{n}^{x}$ at fixed $L$, analogous to those presented in Fig.\ \ref{fig:magn}(b). As was described earlier, we divide the chain into short sections and perform local fits using the formula in Eq.\ (\ref{fittingformula}). For each $D>D_m$, i.e., in the phase with no magnetic order, the local values of the fitting parameters obtained with $\chi=0$ are very stable, and the convergence with increasing $n$ (towards the center of the chain) is convincing. The fitted order parameter $m$ is always very close to zero. It seems that the decay of $m_{n}^{x}$ is asymptotically purely exponential, like the decay of $m^{\rm ITF}_n$ in the disordered phase of the ITF chain, discussed in the Appendix. Due to the stability of local $\xi$'s, the correlation length $\xi_{m}$ in the phase with no magnetic order can be calculated very precisely (at $D=1.27$ the error in $\xi_{m}$ is around 0.1\% and it is smaller for larger $D$). In the magnetically ordered phase, for $D<D_m$, the local fitting parameters obtained with $\chi=0$ are not stable. The local $\xi$'s increase towards the center of the chain, like in the ordered phase of the ITF chain, suggesting the existence of an algebraic prefactor. To calculate $\xi_m$, we vary $\chi$ and look for its value where the local $\xi$'s are most stable. We do not find a universal prefactor, and the obtained $\chi$ depends on $D$ and $h_x$. However, the estimates of $\xi_m$ are independent of the boundary field, which justifies the procedure; the uncertainties in $\xi_m$ are around $10$\%. (The same procedure was tested for the ordered phase of the ITF chain and yielded acceptable estimates of the correlation length --- see Appendix.) The dependence of the correlation length $\xi_{m}$ on the crystal-field strength $D$ is plotted in Fig.\ \ref{fig:magnD}. Due to the large error bars for $D<D_m$, the functional dependence can only be analyzed for $D>D_m$. Here, assuming a second order transition again, we fit the dependence of $\xi_{m}$ on $D$ with a power law \begin{equation} \label{fitxim} \xi_{m} (D) = c_3 (D-D_m')^{-\nu}. \end{equation} The fit is again rather good, yielding the values \begin{eqnarray*} D_m' = 1.236 \pm 0.004, \qquad \nu = 1.02 \pm 0.06. \end{eqnarray*} Thus, close to the phase transition both the spontaneous magnetization and the corresponding correlation length seem to be well described by power laws. The two independent estimates of the phase-transition point, $D_m$ and $D_m'$ from Eqs.\ (\ref{fitm}) and (\ref{fitxim}), nicely coincide with each other. (In what follows, we will use $D_m$, which is more precise.) Moreover, the exponents $\beta$ and $\nu$ are close to those of the two-dimensional Ising model. Altogether, we have found strong indications that, as far as the magnetic quantities $m$ and $\xi_{m}$ are concerned, the phase transition is continuous and is described by the critical exponents of the two-dimensional Ising model. \subsection{Quadrupolar order} \label{sec4C} We now investigate the quadrupolar order, associated with the parameter $q^{\alpha}$ defined in Eq.\ (\ref{defq}). While the quadrupolar order must be present in the magnetic phase ($D<D_m$), the question whether it exists in the phase with no magnetic order ($D>D_m$) is of central interest. Expecting the quadrupolar easy axes along the cube edges, as given in Eqs.\ (\ref{magneticpar}) and (\ref{quadrupolarpar}), we began with applying boundary fields along the $x$ axis. These were purely quadrupolar: $h_{\alpha}=0$; $h_{xx} \neq 0$, $h_{yy}=h_{zz}=0$ in Eq.\ (\ref{Bh}); or purely magnetic: $h_{x} \neq 0$, $h_{z}=0$; $h_{\alpha \alpha}=0$. With such boundary fields, due to the symmetry, one must obtain $q_n^x \neq q_n^y=q_n^z$. \begin{figure}[hbt] \epsfxsize=\columnwidth\epsfbox{fig4b.eps} \caption{(a) The parameter $q^{x}_{L/2}$ vs the chain length $L$, and (b) the profile $q^{x}_n$ vs the position $n$ at $L$ fixed, for two values of $D$. Purely magnetic ($h_{x}$) and purely quadrupolar ($h_{xx}$) boundary fields applied along the $x$ axis. \label{fig:quad} } \end{figure} We found that in both phases the order parameter $q_{L/2}^{x}$ measured in the chain center converges with increasing $L$ to finite values which do not depend on the actual boundary fields. Figure \ref{fig:quad}(a) shows these dependencies for $D=1.2<D_m$ and $D=1.3>D_m$. The appropriate profiles of $q^{x}_{n}$ with $L$ fixed are plotted in Fig.\ \ref{fig:quad}(b). In both phases the picture is typical for an ordered state: $q^{x}_{n}$ has plateaus in the middle of the chain, which correspond to finite bulk $q^{\alpha}$. Thus, we find quadrupolar order on both sides of $D_m$, confirming the existence of a purely quadrupolar phase for $D>D_m$. To verify that the cube edges give the easy axes of the quadrupolar ordering, we applied a quadrupolar boundary field having all three components different from zero: $h_{\alpha}=0$, $h_{xx}=0.03$, $h_{yy}=0.02$, and $h_{zz}=0.01$. Then, in both phases, it was observed that $q^{x}_{L/2}>q^{y}_{L/2}=q^{z}_{L/2}$ in the large-$L$ limit, and the limiting values of $q^{\alpha}_{L/2}$ were the same as found above, in accordance with Eqs. (\ref{magneticpar}) and (\ref{quadrupolarpar}). Assuming that monotonic dependencies of $q^{\alpha}_{L/2}$ on $L$ provide correct bounds for the bulk value $q$, we determined $q$ for various values of $D$. ($q$ is defined as the largest of $q^{\alpha}$ in a broken-symmetry state.) The results are depicted as bold points in Fig.\ \ref{fig:quadD}; the precision in $q$ is $10^{-3}$ or better. We observed that, starting from $D \approx 1.3$, the associated correlation length $\xi_q$ increases rapidly with $D$ (see the inset in Fig.\ \ref{fig:quadD}). At the same time, the DMRG precision for $q^{\alpha}_{n}$ fell dramatically. For these reasons, already at $D=1.45$ (and for larger $D$ too) we were unable to obtain precise results for a chain long enough to show us the large-$L$ limit of $q^{\alpha}_{L/2}$ (for $D=1.45$ and $1.5$ we obtained upper bounds on $q$, which are shown as open circles in Fig.\ \ref{fig:quadD}). We had similar problems close to the transition point $D_m$. \begin{figure}[hbt] \epsfxsize=\columnwidth\epsfbox{fig5b.eps} \caption{The quadrupolar order parameter $q$ as a function of $D$. For $D=1.45$ and $1.5$ upper bounds on $q$ are depicted as the open circles. The inset shows very rough estimates of the corresponding correlation length $\xi_{q}$. \label{fig:quadD} } \end{figure} The effective Hamiltonian $H_{\rm eff}$ in Eq.\ (\ref{Heff}) may help to understand the difficulties for $D\agt 1.4$. Notice, that the symmetry-breaking term $\tilde{J_4}$ in $H_{\rm eff}$ is very small, $\tilde{J_4}/\tilde{J_1}={\cal O}(1/D^2)$. Thus even if $\tilde{J_4}$ is relevant, the correlation length is expected to diverge rather rapidly as $D\to\infty$. On the other hand, the observed abrupt increase in $\xi_{q}$ and the emerging numerical difficulties beyond $D\sim 1.4$ may indicate an additional phase transition in which the quadrupolar long-range order vanishes, and gives rise to an extended critical phase with algebraically decaying quadrupolar correlations beyond a critical value $D_q$. A naive linear fit to the data points for $q$ in Fig.\ \ref{fig:quadD} would yield a value $D_q\approx 2.1$, but we are not in a position to give any firm claims in this question. Note however that if $D_q=\infty$, i.e., if the quadrupolar phase extends to $D\to\infty$, the curve $q(D)$ should finally bend upwards, and in the data points in Fig.\ \ref{fig:quadD}, together with the upper bounds at $D=1.45$ and $1.5$, the (small) curvature is consistently downwards. Now we analyze the dependence of $q$ on $D$ in Fig.\ \ref{fig:quadD} close to the phase-transition point $D_m$. For $D<D_m$ we assume a power-law singularity and a linear regular term \begin{equation} \label{fitqleft} q(D) = q_{\rm left} + c_4 (D_m-D)^{\theta} + c_5 (D_m-D). \end{equation} Fitting yields \begin{eqnarray*} q_{\rm left} = 1.838 \pm 0.01, \qquad \theta = 0.5 \pm 0.1, \end{eqnarray*} $c_4 = 0.44 \pm 0.1$, $c_5 = -0.36 \pm 0.1$. Thus, on this side of the phase-transition point, the quadrupolar order parameter appears to have singular behavior, but the exponent $\theta$ differs from $\beta$ describing the spontaneous magnetization. For $D>D_m$, we suppose the behavior is regular, which is suggested by Fig.\ \ref{fig:quadD}. Our fitting formula is now \begin{equation} \label{fitqright} q(D) = q_{\rm right} + c_6 (D-D_m) + c_7 (D-D_m)^{2}, \end{equation} giving \begin{eqnarray*} q_{\rm right} & = & 1.840 \pm 0.002, \end{eqnarray*} $c_6 = -1.83 \pm 0.03$, $c_7 = -0.6 \pm 0.1$. The left and right estimates of the order parameter $q$ at $D_m$ are in agreement, which supports our scaling assumptions. However, we cannot exclude the possibility that a singular derivative in $q$ exists also for $D>D_m$, so close to $D_m$ that we do not observe it in our numerics. To calculate the correlation length $\xi_{q}$, we analyze the profiles of $q^{\alpha}_{n}$, like those in Fig.\ \ref{fig:quad}(b). In our opinion, only a purely quadrupolar boundary field is suitable for this purpose. With a magnetic boundary field, the decay of $q_{n}^{\alpha}$ could be driven by the decay of $m_{n}^{\alpha}$, and so governed by $\xi_{m}$ and not $\xi_{q}$. Using the formula in Eq.\ (\ref{fittingformula}) again, we find that both for $D<D_m$ and $D>D_m$ the prefactor exponent $\chi$ is nonzero and it depends on $D$ and the boundary field. The fitting yields rough estimates to $\xi_q$ (Fig.\ \ref{fig:quadD} inset), the errors are believed to be around $20$\%. The results suggest that $\xi_{m} = \xi_{q}$ for $D<D_m$, which is the expected answer, since the magnetic and quadrupolar order is intimately connected in this phase. For $D>D_m$, the correlation length $\xi_q$ increases with $D$ starting from $D\approx 1.3$, as mentioned above, but, due to the poor precision, we cannot resolve whether $\xi_{q}$ diverges (numerically it increases sharply) for $D \rightarrow D_m$ or it remains finite. \section{Conclusions} In summary, we have studied the ground-state properties of a one-dimensional $S=2$ ferromagnetic spin chain with single-site cubic crystal-field anisotropy $D$. We argued that in contrast with the mean-field prediction, perturbation theory suggests the possibility that a purely quadrupolar phase exists for large values of $D$. This quadrupolar phase was expected to be separated from the magnetic phase by a continuous quantum phase transition at a critical point $D_m$. We verified this conjecture by investigating the model numerically using the density-matrix renormalization group method on open chains with special, symmetry-breaking boundary conditions. In most cases, the method provided precise estimates of the magnetic and quadrupolar order parameters. Very close to the phase-transition point and in the large-$D$ limit, however, our results were less accurate due to the rapidly increasing correlation lengths. For the correlation lengths themselves we could only obtain rather rough estimates. Evidence has been obtained that, in qualitative agreement with the mean-field prediction, the spontaneous magnetization diminishes continuously at the phase-transition point. Regarding the magnetic properties, the transition seems to be characterized by the critical exponents of the two-dimensional Ising model. This could be plausible, since the extra broken symmetry of the magnetic ground state with respect to that of the quadrupolar ground state is just a $Z_2$ subgroup --- the same group as the one breaking down spontaneously in the case of the Ising model. Both in the magnetic and nonmagnetic phases we demonstrated the presence of a quadrupolar order. In the former case the quadrupolar order is just a ``secondary'' effect, inevitably present in any $S\ge 1$ model with magnetic order. In the latter case, however, it is the ``primary'' order, constituting a qualitatively different broken-symmetry phase. At the phase-transition point $D_m$, the quadrupolar order parameter is continuous. For $D\alt D_m$, we could clearly discern a singularity in the derivative of the order parameter, and found that the quadrupolar correlation length diverges together with the magnetic correlation length. For $D\agt D_m$, our observations are much less concrete: the order parameter can be fitted reasonably well by a low-order (regular) polynomial without any singular terms, and, although the quadrupolar correlation length increases as $D_m$ is approached from above, our results do not suggest unambiguously a divergence on this side. It is not clear for us whether the phase transition involving the magnetization should have any precursor in the quadrupolar fluctuations above $D_m$. Mainly due to computational limitations we were unable to resolve convincingly the question whether the quadrupolar phase extends to $D\to \infty$ or there is a finite value $D_q$ where the quadrupolar long-range order disappears. This question is unique to the one-dimensional situation --- in two dimensions and higher we expect a finite quadrupolar order at zero temperature for any $D>D_m$. Finally, the scenario for the magnetic-to-quadrupolar phase transition was supported by the analysis of the degeneracy of the ground state as a function of $D$. We observed a 6-fold and a 3-fold asymptotic degeneracy below and above $D_m$ respectively, in accordance with the the number of possible ordering directions in the two phases. The confirmation of the purely quadrupolar phase in the one-dimensional case gives rise to the belief that such a phase should also exist in higher dimensions at sufficiently low temperatures when the cubic crystal field is strong. Increasing temperature necessarily destroys the quadrupolar order and leads to a finite-temperature phase transition between the quadrupolar and a completely disordered phase. Note that the correct critical theory of this transition cannot be obtained by simply substituting spins with classical vectors in our model, since this approach is unable to account for the purely quadrupolar order. The usual practice of neglecting quantum fluctuations by treating spins classically around a finite-temperature phase transition would confront fundamental difficulties in this case. \acknowledgements The authors thank Steven White for the suggestion of applying symmetry-breaking boundary conditions. Valuable discussions with Enrico Carlon, Andrzej Drzewi\'{n}ski, and Uli Schollw\"{o}ck are also acknowledged. This research was supported by the KBN (Poland) grant no.\ 2P03B07214 and the EPSRC (UK) grant no.\ GR/L55346. \begin{appendix} \section*{A test calculation} Our numerical procedure and extrapolation methods were tested on the zero-temperature 1D Ising model in a transverse field (ITF) where an exact solution for the bulk quantities is available.\cite{ITF,Bar-McC} Our experience with this model provided useful guidelines to refine our procedure in the study of the cubic model of Eq.\ (\ref{model}). The Hamiltonian of the ITF model on an open chain with length $L$ is defined as \begin{eqnarray} \label{HITF} H^{\rm ITF} &=& - \sum_{i=1}^{L-1} \sigma_{i}^{z} \sigma_{i+1}^{z} - \sum_{i=1}^L \Gamma_i \sigma_{i}^{x}\\ \Gamma_i &=& \left\{ \begin{array}{ll} \Gamma & {\rm if}\; i=2,\dots,L-1 \\ \Gamma/2 & {\rm if}\; i=1,L \end{array} \right. \nonumber \end{eqnarray} where $\sigma_{i}^{\alpha}$ denotes Pauli matrices at site $i$, and $\Gamma$ is the transverse magnetic field. Note that in the above definition the spins at the chain ends, being coupled to one neighbor only, are subject to the reduced field $\Gamma/2$. We concentrate on the spontaneous magnetization $ m^{\rm ITF} = \langle \sigma_{i}^{z} \rangle $ and the corresponding correlation length $\xi^{\rm ITF}$. For $\Gamma=0$ the Hamiltonian $H^{\rm ITF}$ has the classical, ferromagnetic ground state with $m^{\rm ITF} = \pm 1$, and the single-site term in $H^{\rm ITF}$ tends to destroy the spontaneous magnetization; in this respect the ITF model is similar to the cubic model in Eq.\ (\ref{model}). In the thermodynamic limit there are two phases in the ITF chain, distinguished by the order parameter $m^{\rm ITF}$. For $0 < \Gamma < 1$ the ground state is ordered due to spontaneous symmetry breaking \begin{equation} m^{\rm ITF} = \pm \left(1-\Gamma^{2}\right)^{1/8}, \hspace*{2em} \xi^{\rm ITF} = - \frac{1}{2 \log \Gamma}, \label{orderedITF} \end{equation} while for $\Gamma>1$ there is a disordered phase with \begin{equation} m^{\rm ITF} = 0, \hspace*{2em} \xi^{\rm ITF} = \frac{1}{\log \Gamma}. \label{disorderedITF} \end{equation} At $\Gamma=1$ the ground state is critical and $\xi^{\rm ITF} = \infty$. In our study the following values of the transverse field were considered: $\Gamma=0.9$ (ordered phase) where $\xi^{\rm ITF} \approx 4.75$; $\Gamma=1.1$ (disordered phase) where $\xi^{\rm ITF} \approx 10.49$; and the critical point $\Gamma=1$ with $\xi^{\rm ITF}=\infty$. We computed the position dependent order parameter $m^{\rm ITF}_{n} = \langle \sigma_{n}^{z} \rangle$ using the DMRG method with boundary fields up to $L=200$. As a symmetry-breaking boundary field we considered \begin{equation} B = \sigma^z \end{equation} in Eq.\ (\ref{bf-setup}), and we used two different values of $h$: a strong $h=10$ and a weak $h=0.1$ boundary field. In the case of the ITF chain the DMRG is extremely precise even close to the critical point.\cite{Gaborandcousin} Keeping a relatively small number of states $M=64$ the truncation errors are already negligible and the fitting procedure can be tested on numerically exact data. We emphasize that although rigorous results are available for quantities in the thermodynamic limit, the behavior of relatively short chains with complicated boundary conditions is not feasible for a study with purely analytical tools even for the ITF model. \begin{figure}[hbt] \epsfxsize=\columnwidth\epsfbox{fig6b.eps} \caption{ITF chain results. (a) Order parameter $m^{\rm ITF}_{L/2}$ measured in the center of the chain vs the chain length $L$. (b) Order parameter profiles $m^{\rm ITF}_{n}$ as a function of position in the chain $n$ at fixed $L=200$. Curves are labeled by the values of the transverse field $\Gamma$ and the boundary-field strength $h$. \label{fig:ITF} } \end{figure} The value of the order parameter at one of the central spins $m^{\rm ITF}_{L/2}$ as a function of the chain length $L$ is shown in Fig.\ \ref{fig:ITF}(a). For these data the ``infinite-lattice'' DMRG was used from $L=4$ to $L=200$, thus a systematic ``environment error'' cannot be excluded.\cite{Gaborandcousin} In the ordered phase, $m^{\rm ITF}_{L/2}$ converges to the same finite value for both values of $h$, and the two dependencies are monotonic. At each $L$, correct bounds for the bulk $m^{\rm ITF}$ in Eq.\ (\ref{orderedITF}) are obtained, and at $L=200$, ten digits of $m^{\rm ITF}$ are recovered. In the disordered phase, $m^{\rm ITF}_{L/2}$ converges to zero for both boundary fields $h$. At the critical point, the limit of $m^{\rm ITF}_{L/2}$ does not show up in Fig.\ \ref{fig:ITF}(a), we can only say that the correlation length must be of the order of $L=200$ or larger. In Fig.\ \ref{fig:ITF}(b) we show the order parameter profiles $m^{\rm ITF}_{n}$ for a chain length fixed at $L=200$. This length exceeds the exact correlation lengths $\xi^{\rm ITF}$ for $\Gamma=0.9$ and $\Gamma=1.1$ many times. There are plateaus for $\Gamma=0.9$ and $\Gamma=1.1$, which correspond to the bulk values of the order parameter. There is no plateau at the critical point where the correlation length is infinite and the dependence of the profile is algebraic. In order to calculate the correlation length $\xi^{\rm ITF}$, we analyze the profiles from Fig.\ \ref{fig:ITF}(b) using the ansatz in Eq.\ (\ref{fittingformula}). As is described in Sec.\ III we first fix a value of the exponent $\chi$ and then fit for the other parameters. Repeating the fits in different sections of the chain we seek the value of $\chi$ which makes the other parameters the most stable as a function of $n$. For $\Gamma=1.1$ the value of the exponent which yields the most stable fitting parameters is $\chi \approx 0$. From positions $n=50$ to $n=100$, and for both $h$, the fitted correlation length $\xi$ agrees with the exact $\xi^{\rm ITF}$ with a precision of 7 digits! It seems that, asymptotically, the decay of $m^{\rm ITF}_{n}$ in the disordered phase is purely exponential. This should be compared with the exponent $\chi_{\rm bulk}=1/2$ characterizing the decay of the bulk two-point correlation function in this phase.\cite{Bar-McC} (For a similar difference in algebraic prefactors of one and two-point correlation functions in an $S=1$ chain see Ref.\ \onlinecite{Sor-Aff}.) \begin{figure}[hbt] \epsfxsize=\columnwidth\epsfbox{fig7b.eps} \caption{ITF chain results. Ordered phase $\Gamma=0.9$. The local values of the correlation length $\xi$ from Eq.\ (\protect\ref{fittingformula}), obtained with different boundary fields $h$ and exponents $\chi$, as a function of the position $n$ at $L=200$. The exact correlation length $\xi^{ITF}$ is shown as a horizontal line. \label{fig:ITF2} } \end{figure} In the ordered phase for $\Gamma=0.9$ we can also start with $\chi = 0$. In this case, however, the fitting parameters are not stable, and $\xi$ increases with $n$ for both values of $h$, as is illustrated by Fig.\ \ref{fig:ITF2}. This suggests that a nonzero prefactor is present. The value of $\chi$ which makes the correlation length $\xi$ the most stable seems to depend on the boundary field applied. We obtain $\chi \approx 1.25 \pm 0.05$ for $h=10$ and $\chi \approx 0.9 \pm 0.05$ for $h=0.1$, we cannot see a universal prefactor. (Note that for the bulk two-point functions the exponent of the prefactor is $\chi_{\rm bulk}=2$ in this phase.) The values of $\xi$ calculated with the above prefactors are shown in Fig.\ \ref{fig:ITF2}. It is seen that using these prefactors we find better estimates for $\xi^{\rm ITF}$ than with $\chi=0$. Anticipating computational limitations in the investigation of the $S=2$ cubic model in Eq.\ (\ref{model}), we also analized shorter chains, such that $L \approx 10 \xi^{\rm ITF}$ during the test calculations. For $\Gamma=1.1$ and $L=100$ four digits of the exact $\xi^{\rm ITF}$ are recovered; for $\Gamma=0.9$ and $L=50$ only rough estimates for $\xi^{\rm ITF}$ are obtained, the error is around 10 \%. Our observations can be summarized as follows: \begin{itemize} \item The order parameter $m^{\rm ITF}$ can be calculated accurately in both phases; it can be decided whether the system is ordered or not. \item In the ordered phase, upper and lower bounds for $m^{\rm ITF}$ are obtained by considering strong and weak boundary fields, respectively. \item The correlation length $\xi^{\rm ITF}$ can be calculated very precisely in the disordered phase. \item In the ordered phase, rather rough estimates of $\xi^{\rm ITF}$ can be found. \end{itemize} \end{appendix}
\section{Introduction} Characterizing the impact of star formation on a galaxy's ISM is a fundamental step towards understanding the interplay between star formation and gas and, ultimately, the mechanisms responsible for galactic star formation. In a large-scale star formation event, stellar winds and supernova explosions from massive stars generate a feedback mechanism by injecting energy into the ISM, which may produce gas outflows and, in more extreme cases, superwinds (Elmegreen 1992, Silk 1997, Shull 1993, Chu \& Kennicutt 1994, Heckman, Armus \& Miley 1990). Outflows/superwinds may act as a self-regulating mechanism for the burst, by removing gas from the site of star formation (Heckman 1997, Kennicutt 1989, Meurer et al. 1997). At a less extreme level, OB associations will drive ionization and shock fronts through the ISM, far away from the site of the massive stars, and in some cases causing the star formation process to propagate spatially (Elmegreen \& Lada 1977, McCray \& Kafatos 1987, Parker et al. 1992, Satyapal et al. 1997, Puxley, Doyon \& Ward 1997). Exploring the interactions between a starburst and its surroundings requires studies of components of the host galaxy which are most directly affected by the presence of the starburst. In the optical, the diffuse ionized gas (DIG) is a prime candidate, and in this paper we investigate the ionized ISM in two nearby starburst galaxies, NGC5253 and NGC5236. The presence of DIG as a general component of the ISM of galaxies is well known (e.g., Monnet, G. 1971). Over the last 15 years, a number of studies have revealed its presence in a large variety of galaxy types: spirals, star-forming irregulars and starbursts (Kennicutt \& Hodge 1986, Kennicutt, Edgar, \& Hodge 1989, Gallagher \& Hunter 1990, Hunter \& Gallagher 1992, Hunter \& Gallagher 1997, Lehnert \& Heckman 1995, Rand 1998, Wang, Heckman, \& Lehnert 1998, 1999). Although generally associated with the presence (or past presence) of massive stars, the DIG can extend over 10 times larger spatial scales than the ionizing stars (Reynolds 1991), and, as a result, the mechanism of ionization of the DIG is still a matter of debate. Photoionization from OB stars appears to account for the Reynold's Layer (Dove \& Shull 1994) and may be a general mechanism for exciting the DIG (Ferguson et al. 1996a, Ferguson, Wyse \& Gallagher 1996b, Hunter \& Gallagher 1997). This requires that more than 20--30\% of the ionizing photons leak out of HII regions. However, the emission line ratios of the DIG are often very different from what is expected from photoionization, and evidence has been accumulating in favor of mixed photoionization/shock, or some other, heating mechanism for the DIG (Sivan, Stasinska, \& Lequeux 1986, Hunter \& Gallagher 1990, Martin 1997, Rand 1998). The DIG may be composed of at least two phases: a quiescent one, which comprises 80\% of the H$\alpha$ emission and has a scale height comparable to that of stars, and a turbulent one, with a scale height about 3 times that of stars (Wang 1998, Wang, Heckman \& Lehnert 1997). These studies demonstrate that the effects of the massive stars extend to galactic scales, and could affect the subsequent evolution of the galaxy. Star formation in starburst galaxies proceeds at a pace that is one-to-two orders of magnitude higher than in ``quiescent'' galaxies and thus has a major impact on the DIG, as seen in its kinematics, spatial distribution, and ionization (e.g., Marlowe et al. 1995). However, the details are not yet clear as to how the properties of the DIG correlate with the star formation rate of the host galaxy. For instance, does the importance of the DIG increase in a starburst relative to a quiescent galaxy or does it simply become brighter and, therefore, easier to observe (Wang et al. 1998)? To date, studies of the physical conditions of the ionized ISM in galaxies, and in starbursts in particular, have been pursued mainly via long-slit spectroscopy of a limited number of regions. However, structurally complex objects like starbursts cannot be unraveled without fully accounting for the morphology of both gas and stars. A complementary approach is to use narrow and broad band imaging to obtain a complete map of the gas emission. While the advantage of imaging is to fully characterize the spatial distribution of the ionized gas, the disavantage is that only the brightest ionized lines can be reasonably imaged with sufficient depth. In addition, the matching of the narrow filter passband with the redshift of the galaxy represents a technical difficulty for large samples of galaxies. Imaging and long-slit spectroscopy provide complementary approaches to the study of the ionized gas in galaxies. Here we present images of the two nearby starburst galaxies NGC5253 (v=404~km/s) and NGC5236 (v=516~km/s) in the light of [OIII], H$\alpha$, and [SII]. Variations of the intensity of ionization lines, expecially the low ionization ones like [SII], are direct indicators of changing physical conditions of the gas. The two galaxies form a binary pair in the Centaurus group at a distance of a few Mpc (4~Mpc for NGC5253, Sandage et al. 1994). They have completely different characteristics: NGC5253 is a peculiar dwarf (Caldwell \& Phillips 1989) and NGC5236 is a massive, grand design spiral, classified as an SABc (Telesco et al. 1993); the two form a metal~poor/metal~rich pair, with NGC5253 at about 1/6~Z$_{\odot}$ and NGC5236 at about 2~Z$_{\odot}$. They are both experiencing a high-intensity burst of star formation in their central regions, possibly triggered by an encounter between the two about 1~Gyr ago. This possibility was first suggested by van den Bergh (1980) on the basis of various evidence, including the warping of the HI disk of NGC5236 (Rogstad, Lockhart \& Wright 1974). Thanks to their closeness, these galaxies are excellent laboratories for studying spatial variations of the gas conditions in starbursts; 1$^{\prime\prime}$ corresponds to a linear scale of 19~pc in NGC5253, the size of a typical HII region. NGC5253 is a ``benchmark starburst'', with centrally concentrated recent star formation superimposed on an older, quiescent stellar population. The central star-forming region is very blue, although it is crossed by dust lanes which produce patchy and heavy obscuration and make this galaxy at the same time an excellent UV- and far-IR emitter (Kinney et al. 1993, Aitken et al. 1982, Telesco et al. 1993, Walsh \& Roy 1989, Calzetti et al. 1997). Radio observations reveal that a large fraction of the most recent star formation is hidden by dust (Turner, Ho \& Beck 1998). The bulk of the ongoing starburst is located in an area 50--60~pc in size, where the stars are about 5~Myr old. The UV emission in this area is dominated by a $\sim$3--4~Myr old stellar cluster, but the ionization is being driven by a $\sim$2~Myr old, dust-buried, central cluster (Calzetti et al. 1997, Crowther et al. 1998). The star formation rate (SFR) density of $\sim$10$^{-4}$~M$_{\odot}$~yr$^{-1}$~pc$^{-2}$ corresponds to the maximum levels observed in star-forming galaxies (Meurer et al. 1997). Extending beyond the starburst is a $\sim$300~pc region where star formation has been active at a roughly constant level for the last $\approx$100~Myr, with a SFR density about 0.01 of the starburst's; a handful of bright stellar clusters with ages between 10 and 60~Myr are contained in this area (Calzetti et al. 1997). The ionized gas around the starburst is slowly expanding, with a velocity $\sim$10~km/s, at least within the inner 200~pc region (Martin \& Kennicutt 1995, see also Strickland \& Stevens 1999). The HII emission extends for $>$1~kpc from the center, with two identified kpc-scale superbubbles in the western periphery, one of them expanding with a velocity of 35~km/s (Marlowe et al. 1995). Moderately high [OIII]/H$\beta$ ratios in the presence of high [SII]/H$\alpha$ ratios up to $\sim$40~arcsec (800~pc) from the center suggest that shocks or some mechanism other than photoionization contribute to the gas excitation (from long slit spectroscopy, see Martin 1997). The peculiar HI kinematic indicates rotation about the major axis of the galaxy (Kobulnicky \& Skillman 1995), but an alternative interpretation, that NGC5253 has accreted/is accreting relatively unprocessed gas along the minor axis, has been suggested to account for the unusually low CO luminosity (Turner, Beck \& Hurt 1997). The starburst in NGC5236 is comparable in intensity to the spectacular event in NGC5253. The central star formation extends for $\sim$20$^{\prime\prime}$ across ($\sim$360~pc at 3.7~Mpc distance), and is bright at all wavelengths, ranging from X-ray (Trinchieri, Fabbiano \& Paulumbo 1985, Ehle et al. 1998), through the UV (Bohlin et al. 1983, Kinney et al. 1993), optical, near-IR (Gallais et al. 1991, Rouan et al. 1996), and mid-IR (Telesco et al. 1993), to the radio (Turner \& Ho 1994). The nucleus proper is luminous in the near-IR owing to the large dust obscuration in the center of the galaxy (Gallais et al. 1991, Rouan et al. 1996). Large amounts of dust are present, and form multiple dark lanes which surround the center and cross it in the N-S direction. Gas inflow along the bar collecting at the inner Lindblad resonance may be fueling the central starburst (Petitpas \& Wilson 1998). An optically visible arc of star formation, possibly the main source of the starburst's UV emission, lies about 6$^{\prime\prime}$ S and SW of the nucleus (Heap et al. 1993, Bohlin et al. 1983). In HST-WFPC1 U-band observations, the arc breaks down into a series of very young star clusters, with ages of $\sim$2--6~Myr. The arc-shaped structure could be part of a ring surrounding the nucleus where the other sections are not currently actively forming stars (Gallais et al. 1991). Perhaps star formation has not been co-eval, but sequential, propagating through the ring. Differences between the Br$\gamma$ EW and CO EW maps support this picture (Puxley et al. 1997). Mid-IR (Telesco et al. 1993) and radio (Turner \& Ho 1994) maps depict a different picture. These show two main sources amid diffuse emission: the northern-most source coincides with an optical clump about 11$^{\prime\prime}$ NW of the nucleus and slightly W of the central dust lane, while the southern one is obscured at shorter wavelengths by the dust lane in which it appears to be embedded (Telesco et al. 1993). The detection of only two ``point-like'' sources at long wavelength led various authors to advocate their extreme youth. Whatever processes triggered the central starbursts in the two galaxies, they were acting on two very different environments: a grand-design, massive spiral galaxy in NGC5236, and a dwarf in NGC5253. One candidate for the triggering perturbation is the encounter between the two galaxies about 1~Gyr ago (van den Bergh 1980); if this is the case, the timing of the trigger is the same. This would reduce by one the number of free parameters in the problem. The difference between the two environments would then be the major variable, making this pair an important test-bed for starburst studies. This paper therefore focuses on the role of the host galaxy environment on the evolution of the starburst by investigating the physical properties and variations of the large-scale structure of the ionized medium associated with each of the two starbursts. Section~2 describes the observations and data reduction; section~3 presents the analysis of the observations, with special emphasis on nebular line emission; the discussion is contained in section~4 and the conclusions in section~5. \section{Observations and Data Reduction} Broad and narrow band images of NGC5253 (Figure~1) and NGC5236 (Figure~2) were obtained at the 2.5-m telescope of the Las Campanas Observatory with the Direct Camera and a 2k$\times$2k CCD during the nights of April 28 -- May 1, 1997. Broad band images were obtained with 3~in.$\times$3~in. filters in the Harris U, V, and R. Narrow band images were obtained using 2~in.$\times$2~in. filters on loan from CTIO, centered at the redshifted wavelengths of [OIII]$\lambda\lambda$4959,5007~\AA, H$\alpha+$[NII]$\lambda\lambda$6548,6584~\AA~ and [SII]$\lambda\lambda$6717,6731~\AA~ (see Table~1). The platescale on the CCD is 0$^{\prime\prime}$.26/pix, implying a total field of view of 8$^{\prime}$.8 for the broad band images. The small size of the narrow band filter introduced vignetting at the edges of the CCD, and the final unvignetted field of view was about 5$^{\prime}$.2; the presence of scattered light from the edges of the filters further reduced the useful field of view of the narrow band images to about 4$^{\prime}$.7. The seeing varied during the course of the four nights in the range 0$^{\prime\prime}$.9--1$^{\prime\prime}$.2. Given the surface brightness variation of more than a factor 100 from the center to the edges of the starburst regions (and more than 10,000 in line emission intensity), the exposure times ranged from 30~s to 600~s in V and R and from 30~s to 1,200~s in U and in the narrow band filters, to achieve suitable exposure levels in different regions of the galaxies. Offsets of a few arcseconds between frames were introduced to remove cosmetic defects (bad pixels and two central bad columns) from the final combined images. Table~2 lists for both galaxies the total exposure time in each filter. Data reduction followed the standard procedure of bias subtraction, flat-fielding, registration, and co-addition of the images. Both dome and twilight exposures were used to remove pixel-to-pixel variations and illumination patterns from the images. Residual scattered light in the [SII] filter was removed by subtracting a surface fit to the background. The background fit for NGC5253 was adopted for both galaxies, since the background of NGC5236 could not be fit as this galaxy completely fills the field of view. The two central bad columns of the chip were linearly interpolated in each image with values from surrounding columns. Cosmic rays were removed from individual frames before co-addition, using an algorithm developed by M. Dickinson (1997, private communication) for the identification of sharp, positive discontinuities over scales of $\sim$1~pix. This technique removed around 80--90\%~ of the cosmic rays; final co-addition of multiple frames removed most of the remaining events. Both galaxies were observed close to their culminating points, and the effects of airmass variations were generally less than 3\%; exception were U and [OIII], where such effects were as large as 12\%~ and 5\%, respectively, and corrections were therefore applied. Absolute calibrations were obtained from observations of two spectrophotometric standards from Hamuy et al. (1994). One of the standards was also observed during the night at different azimuths to derive airmass corrections. Absolute flux calibrations are listed in Table~1 for all filters, together with the internal error (in percentage). \subsection{Emission Line Images} More than one emission line is included in each of the three narrow band filters. Both the redshifted [OIII]$\lambda$4959~\AA~ and [OIII]$\lambda$5007~\AA~ contribute to the emission in the 5000/70 filter; the second line is located almost at the center of the passband, while the [OIII]$\lambda$4959~\AA~ is located on the ramp, making the determination of its contribution rather uncertain. Our estimates give a best value of $\sim$95\% for the filter transmission at the redshifted [OIII]$\lambda$4959~\AA~ relative to the filter transmission at [OIII]$\lambda$5007~\AA, with a range between 75\% and 106\%. Three lines, the redshifted H$\alpha$, [NII]$\lambda$6548~\AA~ and [NII]$\lambda$6584~\AA, contribute to the emission in the 6563/78 filter, with the reddest [NII] line located at 90\% and 72\% of the peak transmission for NGC5253 and NGC5236, respectively. Both the redshifted [SII]$\lambda$6717~\AA~ and [SII]$\lambda$6731~\AA~ are located close to the transmission peak in the 6737/76 filter. The calibrated V and R images were used to subtract the stellar continuum from the [OIII], H$\alpha$+[NII], and [SII] images. After matching the FWHM of the stars in the broad and narrow band frames, the continuum images were recursively rescaled and subtracted from the narrow band images till optimal removal of the galaxy stellar continuum was achieved. The field stars were initially used to obtain a first guess on the scaling factor, but refinements on this factor were necessary due to the bluer stellar continuum of the galaxies' centers relative to the stars. For the weak [SII] emission we first subtracted the H$\alpha$+[NII] nebular emission from the R-band image and then used this nebular~emission-free image to remove the stellar continuum from the [SII] image. There is a marked color gradient in the V-band image of NGC5236 with the center bluer than the external regions; therefore accurate continuum subtraction over the entire [OIII] image of NGC5236 could not be achieved. We paid particular attention to the central region, and obtained a satisfactory [OIII] emission~line image of the inner $\sim$40$^{\prime\prime}$. This region is comparable in size to the area of H$\alpha$ emission detected above 5$\sigma$ (see section~3.3 below), and thus is sufficient for our purposes. The accuracy of the calibration of the narrow band filters was checked against ground-based spectrophotometry of the centers of the galaxies (Storchi-Bergmann, Kinney \& Challis 1995) and, for NGC5253, against H$\alpha$ images obtained with the Hubble Space Telescope Wide Field and Planetary Camera 2 (Calzetti et al. 1997). In all cases, our calibrations gave flux values slightly higher than the spectra and the HST image. For NGC5253, our [SII], H$\alpha+$[NII] and [OIII] fluxes are about 5\%, 7\%~ and 10\%~ larger, respectively, than what is measured from the spectrum. For NGC5236, the [SII] and H$\alpha+$[NII] image fluxes are 8\% and 2\% larger, respectively, than the spectrum. To compare our H$\alpha+$[NII] image of NGC5253 with the HST one, the contribution of the [NII]$\lambda$6584~\AA~ line had to be removed; from the spectrum we estimate that the [NII] flux is on average 13\%~ of the H$\alpha$ flux, although variations are expected as a function of position (Kobulnicky et al. 1997). After the subtraction of this contribution, our narrow band image gave fluxes consistently $\sim$6\% higher than the HST image. The source of this fairly small, but systematic discrepancy is unclear. We have considered under-subtraction of the stellar continuum, filter calibration, and presence of Balmer absorption in the calibration stars, but none of those reproduces all of the observed discrepancy. However, the relative calibration of the three narrow band images is good at the 5\%~ level. For consistency, we rescale our emission line images to the spectroscopic/HST values. For NGC5253, HST WFPC2 images centered on the H$\beta$ line emission (Calzetti et al. 1997) are used here to supplement the ground based images. The HST H$\beta$ image has been rotated, smoothed, resampled, and registered to match the ground-based images. \section{Analysis and Results} \subsection{Line Ratio Maps} Ratios of metal-to-hydrogen lines are common diagnostics of the physical conditions of the ionized gas. We produced maps of [SII]$\lambda\lambda$6717,6731\AA/H$\alpha$ and [OIII]$\lambda$5007\AA/H$\beta$ for NGC5253 (Figures~3c, 3e and 3f), and [SII]$\lambda\lambda$6717,6731\AA/H$\alpha$ and [OIII]$\lambda$5007\AA/H$\alpha$ for NGC5236 (Figure~4e and 4f). The line ratio maps have been created using a 5~$\sigma$ detection threshold for each line, after all the images have been smoothed to the seeing of the photometrically worst night ($\sim$1$^{\prime\prime}$.2) and resampled to 5~pix$\times$5~pix bins (1$^{\prime\prime}$.3$\times$1$^{\prime\prime}$.3). The [OIII] line maps have been divided by a factor 1.3 to remove the contribution from the [OIII]$\lambda$4959\AA. For NGC5236, we don't have an H$\beta$ emission line image, thus the corresponding ratio map [OIII]$\lambda$5007/H$\beta$ could not be constructed. We use the H$\alpha$ image instead, with cautionary remarks about the potentially large effects of reddening variations in the center of the galaxy (see below). We also note that the difficulty of subtracting the continuum from the [OIII] image for this galaxy contributes to the larger uncertainty in the line fluxes; a number of the data bins are below our required 5~$\sigma$ threshold, and the `usable' [OIII]$\lambda$5007/H$\alpha$ map includes the central $\sim$30$^{\prime\prime}$ region only. This is slightly, but possibly significantly, smaller than the extent of the 5~$\sigma$ H$\alpha$ ionized region, which occupies the central $\sim$40$^{\prime\prime}$ (see section~3.3). In NGC5253, regions beyond the central $\sim$30$^{\prime\prime}$ in radius have H$\beta$ flux detections below 5~$\sigma$, and we have used the [OIII]/H$\alpha$ ratio instead, assuming that reddening corrections are small at large distance from the center of the starburst (see discussion below). \subsubsection{Underlying Stellar Absorption} Corrections for the stellar absorption underlying the Balmer lines are important expecially at the faintest surface brightness levels, where the EW of the emission line is correspondingly small. We used the ratio of the H$\alpha$ and the H$\beta$ emission to the corresponding underlying continua to derive maps of the EW of these lines; for H$\alpha$ we used the R-band image as continuum, while for H$\beta$ we used the extrapolated continuum image from the HST V and I images of NGC5253 (Calzetti et al. 1997). The line fluxes were then corrected for the presence of underlying stellar absorption with constant value EW=3~\AA~ (e.g., McCall et al. 1985). Figures~3d and 4d show the H$\alpha$ EW emission maps of the two galaxies, after the correction. Unknown variations of the underlying stellar absorption EW increase the undertainty in the line flux at the detection threshold, where the emission line EWs are generally small (Figures~3d and 4d). In addition, the contribution of an intermediate/old underlying stellar population proportionally increases as the distance from the center of the starburst (the young population) increases, thus gradually changing the underlying stellar absorption from $\sim$3~\AA~ to $\approx$5~\AA. The combination of the two effects implies an uncertainty of $\sim$20\% and $\sim$50\% for the H$\alpha$ measurements at the detection threshold in NGC5253 and NGC5236, respectively. The variable underlying stellar absorption is taken into account in the following sections every time its effect is relevant to the measurements. \subsubsection{Dust Reddening} Corrections for dust reddening are generally small for the [SII]/H$\alpha$ and [OIII]/H$\beta$ maps, due to the closeness in wavelength of each pair of lines, but can be large for the [OIII]/H$\alpha$, because the wavelength difference is large. We discuss reddening corrections separately for each of the two galaxies. Dust extinction in the central $\sim$30$^{\prime\prime}\times$30$^{\prime\prime}$ of NGC5253 is highly variable (Calzetti et al. 1997); there is a E-W dust lane bisecting the central section of the galaxy and the central ionizing stellar cluster is deeply embedded in a highly opaque dust cloud. We thus use the HST reddening map of Calzetti et al. (1997) to remove extinction effects from the line ratios. We assume the reddening is foreground, which should be a reasonable approximation for most regions, since the H$\alpha$/H$\beta$ ratio approaches the unreddened case in the vast majority of the bins. However, we already know that the foreground geometry is altogether wrong in the case of the central cluster, where the amount of extinction is above A$_V$=10~mag and the geometry is known not to be foreground (e.g., Beck et al. 1996); such cases should be statistically insignificant when trends between lines are analyzed, as they include a relatively small number of bins. Regions beyond $\sim$30$^{\prime\prime}$ radius, where the HST reddening map is not available, are corrected with the assumption of a small, constant reddening E(B$-$V)=0.1, of which 0.05 are from our Galaxy (Burstein \& Heiles 1982). This assumption is reasonable, as the H$\alpha$/H$\beta$ map indicates that the reddening decreases to small values beyond a radius of $\sim$20$^{\prime\prime}$ from the center (Calzetti et al. 1997). In the absence of a reddening map for NGC5236, we have adopted the constant value E(B$-$V)=0.35 for the dust extinction correction, which includes both intrinsic and Galactic foreground extinction, as derived from ground-based spectroscopy of the central starburst (Calzetti et al. 1994). Given the small wavelength difference between [SII] and H$\alpha$ the impact of reddening corrections is no larger than 6\% for a reddening variation between E(B$-$V)=0 to 0.7. The impact is of course much larger for the [OIII]$\lambda$5007\AA/H$\alpha$; the intrinsic ratio changes by 40\% if E(B$-$V)=0.7 instead of 0.35. We know from the study of Telesco et al. (1993) that there is a consirable amount of dust with a complex geometry in the center of NGC5236. For this reason, the [OIII]/H$\alpha$ ratio map, which will be briefly discussed in the next sections, should be considered a preliminary substitute for [OIII]/H$\beta$ for this galaxy. \subsubsection{The Contribution of [NII] to the H$\alpha$ maps} The ratio of the HST/ground~based H$\alpha+$[NII] images gives an estimate of the [NII] intensity change across the central region of NGC5253, because the HST image does not contain the [NII]$\lambda$6584~\AA~ emission while the ground~based image does. The image ratio appears constant to within 12\% in the central $\sim$27$^{\prime\prime}$ (radius), which is where the S/N is high. Thus variations of the [NII] intensity are not expected to be more than twice its average value. This is in agreement with results from long-slit spectroscopy (Lehnert \& Heckman 1995, Martin 1997), which indicate that the variation in the [NII]/H$\alpha$ ratio is very small for the central $\sim$25--30$^{\prime\prime}$ of NGC5253 (see, however, Kobulnicky et al. 1997). In the light of this result, we derived a ``pure'' H$\alpha$ image by removing 17\%~ of the flux from the original image. The 17\%~ figure represents the contribution of the two [NII]$\lambda\lambda$6548,6584~\AA~ lines. For NGC5236, the contribution of [NII]$\lambda\lambda$6548,6584\AA~ to the H$\alpha$ map has been estimated from spectroscopic data (Storchi-Bergmann et al. 1995): the [NII] line contributes $\sim$39\% of the total line flux in the 6563/78 filter, in agreement with the high metallicity of the galaxy. In the shocked regions of both galaxies, our derived [SII]/H$\alpha$ is a lower limit to the true value as [NII]/H$\alpha$ is expected to increase for increasing [SII]/H$\alpha$, leading to an underestimate of the [NII] contribution to our H$\alpha$ maps. This will have in general the effect of weakening the shock diagnostics, an opposite effect to that induced by variable underlying stellar absorption. \subsection{NGC 5253} \subsubsection{The Ionized Gas Morphology} The morphology of the nebular gas emission in NGC5253 has been described by a number of authors (e.g., Marlowe et al. 1995, Martin \& Kennicutt 1995, Calzetti et al. 1997). We review here a few basic facts. The ionized gas emission is circularly symmetric around a stellar cluster located almost at the geometric center of the galaxy (Figure~1, bottom panel); this cluster, with an age of $\sim$2~Myr, is also the youngest stellar cluster in the galaxy (Calzetti et al. 1997). The azimuthally-averaged H$\alpha$ emission monothonically decreases in surface brightness from the cluster outward. The regular morphology of the gas emission is in striking contrast with the morphology of the UV and optical stellar continuum (Figures~3a and 3b), which is elongated from NE to SW, along the major axis of the galaxy (e.g. Martin \& Kennicutt 1995). We clearly detect in each of the H$\alpha$, [OIII] and [SII] maps the two western bubbles described in Marlowe et al. (1995): the one closer to the minor axis is the weakest of the two, and we detect the outer shell of the expanding gas; the other, which is almost along the major axis of the galaxy, is well detected and shows a wealth of substructure (Figures~1 and 3). A number of filaments extend outward from the center, both in the North and in the East-South direction. The presence of ionized gas along the dust lane, SE of the center, is detected in both our H$\alpha$ and [OIII] images, with a hint in the 5~$\sigma$ [SII] image. \subsubsection{Photoionization and Shock-Ionization} Figure~5 (panel~a) shows the line ratios measured in each 1$^{\prime\prime}$.3 resolution element and compares those with models for gas photoionization and for shock excitation. The photoionization models give the variation of the line ratios for changing ionization parameter U. One set of models has been taken from Martin (1997), who ran CLOUDY (Ferland 1993) for a range of metallicities and effective temperatures of the ionizing source. Two other models are from Sokolowski (1993), who analyzed the cases of depleted metal abundances and of hardened photoionizing continuum; the models assume cosmic abundances and an ionizing source given by an instantaneous burst of star formation with a Salpeter stellar mass function up to 120~M$_{\odot}$. The last model reproduces the scenario in which the soft ionizing photons are the first to be absorbed by the ISM, thus the ionizing continuum hardens as it travels across the galaxy. Predicted line ratios for shock excitation are from Shull \& McKee (1979), for cosmic abundances and a range of shock velocities and for the special case of depleted metal abundances. The metal-to-hydrogen line ratios change as a function of the ionization parameter and this can potentially explain the observed variation in Figure~5 (e.g, Hunter 1994). The ionization parameter U measures the relative amount of ionizing photons relative to the amount of gas. Increasing the distance from the ionizing source decreases the value of U, lowering the [OIII]/H$\beta$ ratio and increasing the [SII]/H$\alpha$ ratio (Domg\"orgen \& Mathis 1994). The data of NGC5253 appear to follow this trend qualitatively both in Figure~5 and in the ionization map [OIII]/[SII] of Figure~3g. Except along the dust lane (see discussion below), the [OIII]/[SII] ratio decreases from the center to the edges of the ionized region. However, a quantitative comparison (Figure~5, panel~a) shows that the [OIII]/H$\beta$ value decreases less steeply than expected from variations of the ionization parameter, a trend already noted by Martin (1997) for a sample of dwarf galaxies. The photoionization model with T$_e$=50,000~K and [O/H]=0.2~[O/H]$_{\odot}$, which closely matches the metallicity of NGC5253 ($\sim$1/6~[O/H]$_{\odot}$), marks a lower envelope to the data points in Figure~5, while it is in reasonable agreement with the data at the highest values of U, in regions closest to the central ionizing source in the starburst. A somewhat better representation of the data is given by the model with depleted abundances (Sokolowski 1993), but most of the data points are still above the locus of the photoionization lines. The observed line ratios behave as if there is an increasingly important shock component (Martin 1997) or the radiation spectrum is progressively hardened towards the external regions (Wang 1998). Ionization in NGC5253 can be directly compared with the well-studied Large Magellanic Cloud. Figure~5, panel (b), shows the [OIII]/H$\beta$ versus [SII]/H$\alpha$ line ratios of a sample of HII regions, giant and supergiant shells in the LMC from Hunter (1994). Although some of the data points show the same extreme behavior as NGC5253, the majority of the shells agree with photoionization models. The metallicity of the LMC is about twice that of NGC5253, thus the comparison between the two galaxies is not immediate, as the LMC data naturally occupy a locus to the lower left relative to the NGC5253 data. Neverthless, the majority of the LMC line ratios are between the solar metallicity and the depleted abundances models, and the strongest outliers are in giant shells, which seem to require a hardened radiation field. To further discriminate between ionization mechanisms, we have plotted the ratios [SII]/H$\alpha$ and [OIII]/H$\beta$ as a function of the physical distance from the central star cluster in NGC5253 (Figure~6). The mean value of [SII]/H$\alpha$ ([OIII]/H$\beta$) increases (decreases) for increasing distance from the `center of ionization'. Again, photoionization models reproduce qualitatively, but not quantitavely, this trend. The relationship between ionization parameter and distance has been taken from Martin (1997, see her Equation~1); the size of the ionized sphere has been assumed to correspond to the size of the H$\alpha$ emission, around 71$^{\prime\prime}$--81$^{\prime\prime}$ in radius, or 1.4--1.6~kpc. A quantitative test shows that photoionization alone cannot fully explain the observed trend of the line ratios, and, in particular, cannot account for the increasing {\em spread} about the mean values with increasing distance. The increasing spread with distance is quite evident in the [OIII]/H$\beta$ diagram (Figure~6b). We interpret this as an effect of the increasing importance of shock-ionization (or other non-photoionization mechanism, see Haffner, Reynolds \& Tufte 1999) over photoionization further from the center. The physical extent of the starburst and metallicity variations may play a role in the line ratio spread, but we do not expect these to be the dominant effects. We will show in the next section that the starburst population extends over a much smaller area, less than 1/6, than the ionized gas. Both the mean value and the spread of [SII]/H$\alpha$ increase for decreasing H$\alpha$ surface brightnesses (Figure~7, see Wang, Heckman \& Lehnert 1998, Martin 1997, and Ferguson et al. 1996b on a variety of galaxies), supporting what is observed in Figure~6. The reddening-corrected ionizing photon rate from the starburst is $\log$~Q(H$^o$)=52.57--52.78, depending on the dust opacity adopted for the central star cluster (9~mag$\le$A$_V\le$35~mag, Calzetti et al. 1997). These values correspond to a Str\"omgrem radius R$_S$=240--280~pc, for an electron density of 93~cm$^{-3}$ and temperature $\sim$10,000~K (Storchi-Bergman, Kinney \& Challis 1995, CKS94), and for a filling factor of 0.01 (Martin 1997). The calculated Str\"omgren radius is at least a factor $\sim$4.5 smaller than the extent of the H$\alpha$ emission. \subsubsection{The Morphology of `Shocks' and DIG} Adopting arbitrarily the constraints $\log([OIII]/H\beta)>0.2$ and $\log([SII]/H\alpha)>-0.35$ to discriminate between predominance of photoionization and predominance of shock excitation/hardening of radiation (or other mechanism), the location of the `shocked' regions is graphically represented in Figure~3h. The figure shows that the purely photoionized region has a circularly symmetric shape centered almost exactly on the main ionizing cluster, with radius $\simeq$560~pc; it is comparable in size to the stellar population of the starburst, although the morphology of the two is different (see next section). The `shocked' gas has a markedly asymmetric morphology; the majority of the bins with $\log([OIII]/H\beta)>0.2$ and $\log([SII]/H\alpha)>-0.35$ is located in the south-west region, and there appears to be an overlap of filaments and arches extending out of the main starburst area. The `shocked' gas extends in the direction of the major axis of the galaxy; one would expect expanding gas to prefer the direction of the minor axis (e.g., Heckman et al. 1990, De Young \& Heckman 1994, Martin 1998, Meurer, Staveley-Smith, \& Killeen 1998), which does not seem to be true for NGC5253. The constraint $\log([SII]/H\alpha)>-0.35$ corresponds to a H$\alpha$ surface brightness less than 6.39$\times$10$^{-16}$~erg~cm$^{-2}$~s$^{-1}$~arcsec$^{-2}$, or a normalized surface brightness SB(H$\alpha$)/SB$_e<$0.01, in Figure~7. We stress again here that variations in the underlying stellar absorption would have typically no more than a 20\% effect on both the SB(H$\alpha$) and the [SII]/H$\alpha$ ratio (see Figure~3d). The ratio SB(H$\alpha$)/SB$_e<$0.01 marks a sharp increase in the median value of [SII]/H$\alpha$. A similar sharp break is evident also in the histogram of the bins with specific value of H$\alpha$ surface brightness (Figure~8): below SB(H$\alpha$)/SB$_e=$0.01 there is a large gradient in the relative number of bins. Such ``breaks'' have been pointed out by Wang (1998) as the transition between HII regions and DIG. In the case of NGC5253, the DIG surrounds the central starburst up to a distance of at least $\sim$1.4--1.6~kpc in some directions (to our detection limit). \subsubsection{The Morphology of the Starburst Population} In order to compare in detail the morphologies of the ionized gas and of the ionizing stars, we have removed the underlying galaxy from the broad band image, so that the structure of the current starburst would be enhanced. The U band image includes the [OII]$\lambda$3727~\AA~ doublet emission in its passband. In the central ~15$^{\prime\prime}$ [OII] has EW=130~\AA~, thus providing about 20\% contribution to the U emission. Since we did not observe the [OII] emission, we used the [SII] map, which was rescaled to the [OII] intensity observed by Storchi-Bergmann et al. (1995). This procedure relies on the (reasonable) assumption that [OII] and [SII] have the same morphology and intensity distribution; this is supported by the observations of Martin (1997). The H$\alpha$ contribution to the R band image has been removed in a more straightforward manner (see section~2). All three broad band images were corrected for the effects of dust reddening using the HST reddening maps (or the constant value E(B$-$V)=0.1 outside the range of the HST maps) and the prescription of Calzetti et al. (1997). The U-band isophotes external to $\sim$70$^{\prime\prime}$ follow an exponential profile (Caldwell \& Phillips 1989), typical of the old stellar populations in spheroidal and irregular dwarf galaxies. Indeed, NGC5253 would very likely be classified as a dwarf elliptical (Sersic et al. 1972) if it weren't for the central starburst. At smaller radii there is a clear excess relative to the exponential fit; Caldwell \& Phillips attribute this excess to the star formation event which occurred in the galaxy over the last $\approx$1~Gyr. Not all of this excess is composed of ionizing stars; along the galaxy major axis, the region between 50$^{\prime\prime}$ and 70$^{\prime\prime}$ is not associated with H$\alpha$ emission (to our detection limit). The ionizing starburst appears more concentrated than the population excess over the exponential light profile. To remove the non-ionizing stellar population underlying the ionizing starburst, we used the isophotes between 50$^{\prime\prime}$ and 70$^{\prime\prime}$, and attempted both an exponential profile and a r$^{1/4}$ law model. The latter fits the non-ionizing population isophotes to r$\sim$34$^{\prime\prime}$ better than the exponential profile. This isophotal profile was extrapolated to the center and subtracted from the original image. The residual, namely the central starburst, is shown in Figure~3a for the U band. All three continuum images present the same morphology, thus hinting that dust extinction does not affect the global appearance in the optical passbands. The colors of the underlying galaxy are fairly uniform, with values U$-$V=0.5$\pm$0.2 and V$-$R=0.80$\pm$0.25, typical of a stellar population dominated by A7 and later type stars, which will not contribute to the photoionizing luminosity. The comparison between the starburst continuum emission and any of the line emission or line ratio maps shows an obvious characteristic: the line emission is more extended, by more than a factor $\sim$2, than the continuum emission (see Figure~3a with 3b). This is true for both the photoionized and shock-ionized parts of the nebular line emission, confirming that the photoionized gas is displaced relative to the ionizing stars by more than $\sim$1~kpc from the external perimeter of the starburst. A plot of the H$\alpha$ surface brightness as a function of the U$-$V color of the starburst population shows the expected trend that higher SB(H$\alpha$) coincide on average with bluer U$-$V colors (Figure~9). The regions with SB(H$\alpha$)/SB$_e>$0.01 have typical colors U$-$V$\simeq-$0.5,$-$1.3, corresponding to ages between 1 and 100~Myr for constant star formation and between 1 and 30~Myr for an instantaneous burst population (Leitherer \& Heckman 1995); this agrees with the age range found by Calzetti et al. (1997). A few points in this region have U$-$V$<-$1.6, bluer than the bluest models for stellar populations. This reflects uncertainties in the color derivation and, possibly, an imperfect subtraction of the strong [OII] emission from the U-band image. Lower SB(H$\alpha$) correspond to regions with typical colors of nonionizing or mildly ionizing populations. We highlight again that the ionizing stellar population extends over an area which is $<$1/6 of the area of the detected gas emission. \subsubsection{Star Formation in the Dust Lane} The values of [OIII]/H$\alpha$ along the dust lane (see the little `horn' sticking out at the bottom left of Figure~3f) have median $\simeq$6, compatible with the values in the center of the starburst. In addition, the ratio [OIII]/[SII] remains high, around or above 10 (about 1/2 the value of the central cluster, see Figure~3g), along the entire dust lane. Insufficient reddening correction due to the presence of the dust lane would make both ratios even higher. This is one of the areas responsible for the marked spread in the [OIII]/H$\beta$ values at large distance from the center. We can place an upper limit [SII]/H$\alpha<$0.35 in this area. Both line ratios are compatible with this region being almost purely photoionized. However, there are no obvious ionizing stars in this area, although we cannot exclude that star formation is heavily embedded in the dust lane, and thus is not visible. Even if this is the case, star formation in the lane is happening at a relatively low intensity level; the star-formation-sensitive 10~$\mu$m map of Telesco et al. (1993), indeed, does not show emission along the dust lane, and dust obscuration is less effective at 10~$\mu$m than in the optical. \subsection{NGC 5236} \subsubsection{Morphology of the Starburst} The global morphology of the ionized gas emission in NGC5236 is far simpler than in NGC5253, and, likely, easier to interpret. Most of the H$\alpha+$[NII] emission comes from the central $\sim$40$^{\prime\prime}$, where the starburst is located, and along the spiral arms (Figure~2b). Unlike NGC5253, there is little or no evidence for arcs, loops or filaments of ionized gas extending outward from the central starburst. In the center of the galaxy, the brightest part of the H$\alpha+$[NII] emission, above 15$\sigma$, occupies a region $\sim$30$^{\prime\prime}$ across (corresponding to a physical size of 540~pc), comparable in size and morphology to the bright blue stellar emission detected in the U band (above 50$\sigma$, Figures~4a and 4b). The optically brightest part of the starburst is located in the south-western arc of blue stellar clumps, about 15$^{\prime\prime}$ in length. The northern tip of the arc appears to bend in the east direction, but this morphology probably is an effect of the crossing of the dust lane (see Gallais et al. 1991). The arc-shape of the stellar continuum is fairly well mirrored by the H$\alpha+$[NII] emission, with no obvious exceptions. The ionized gas and blue star morphology of the center of NGC5236 is typical of the central starbursts hosted in massive galaxies, where rings, arcs, and ``spirals'' of star formation are common structures (Maoz et al. 1996, Colina et al. 1997). All characteristics of NGC5236, including those described below, are consistent with star formation occuring in a sharply bounded inner nuclear disk, perhaps defined by the inner Lindblad resonance (ILR) as suggested by Telesco et al. (1993). Figure~4c displays the HST-WFPC1 image of the center of the galaxy in the F336W filter (Heap et al. 1993, roughly corresponding to the U-band), where the arc (`A' in Figure~4c) clearly splits into several individual stellar clusters and the northern `bend' of the arc (`B' in Figure~4c) splits into three clusters. We cannot resolve the individual line emission of each of the clusters in `B' from the ground-based image, but their summed flux locates the peak of H$\alpha+$[NII] emission in the galaxy center, with a total observed flux F(H$\alpha+$[NII])=1.40$\times$10$^{-12}$~erg~s$^{-1}$~cm$^{-2}$, measured in an aperture of 2.6$^{\prime\prime}$ diameter. The nucleus (`N' in Figure~4c) is located about 6$^{\prime\prime}$ NE of the arc; it appears as a lump in the ground-based U-band image (Figure~4a), but with a weak H$\alpha+$[NII] emission (Figure~4b). About 11$^{\prime\prime}$ North-West of the arc there are two bright HII knots (`C' and `D' in Figure~4c; clump `D' is not visible in the U-band image); they have comparable intensity in the narrow line emission, with the northern-most of the two (`D') being only 30\% brighter than the other, but very different U brightnesses, with `D' being 5.9 times fainter than the other. We identify `D' as coincident with the mid-IR northern source (Telesco et al. 1993). The difference in line/continuum emission between the two knots is likely an age effect with `D' being younger than `C'; if it were simply an effect of dust reddening `D' would appear in near-IR imaging, but this is not the case (Gallais et al. 1991). The younger age of knot `D' is supported by the value of the H$\alpha$ EW, which is about 180~\AA~ for this knot, while it is only about 90~\AA~ for knot `C'. Larger values of the EW(H$\alpha$) locate comparatively younger regions (Leitherer \& Heckman 1995); in the case of the center of NGC5236, an imaginary line joining region `B' with knots `C' and `D' identifies the youngest part of the starburst, with values EW(H$\alpha$)$\approx$200~\AA~ (Figure~4d), about twice those of surrounding regions (Telesco et al. 1993). The inferred ages from such EWs are less than 10$^7$~yr, for an instantaneous burst of star formation (Leitherer \& Heckman 1995, see Puxley et al. 1997). The three luminous condensations in the arc (`A' in Figure~4c, corresponding to multiple clusters in the HST image) are between 1.61 and 2.05 times brighter in U than clump `B', while they are a factor between 2.4 and 4.2 fainter in H$\alpha+$[NII]. In [OIII], the features in arc `A' are about 2.4 times brighter than `B'. If the metallicity along the arc is roughly constant, these differences are immediately understandable in terms of dust reddening, with `B' being more reddened than `A'. This is reasonable as `B' is located very close to the NS dust lane. \subsubsection{Gas Excitation} The [OIII]$\lambda$5007\AA/H$\alpha$ ratio is plotted as a function of [SII]$\lambda\lambda$6717,6731\AA/H$\alpha$ in Figure~10. Sokolowsky's models, derived for cosmic abundances, are expected to work fairly well for this galaxy, whose center has average metallicity about twice solar. The data are not inconsistent with photoionization models, in the entire range considered. There is little evidence for shocks in NGC5236, although our line ratios cannot be used as the only criterion for deciding the ionization mechanism, because of the potential for heavy dust reddening effects in the [OIII]/H$\alpha$ ratio. The plot of [SII]/H$\alpha$ as a function of the distance from the H$\alpha$ peak (Figure~11a) also shows that photoionization appears to be the main gas excitation mechanism, as the trend of the upper envelope to the points closely follows Sokolowski's model for depleted abundances. Further support to the photoionization picture comes from the range of values covered by the [SII]/H$\alpha$ ratio: it is very close to that measured in NGC5253, despite the fact that NGC5236 is at least one order of magnitude more metal-rich (cf. Figure~11a with 6a). The latter conclusion does not qualitatively change even if there is a 50\% uncertainty in the stellar absorption underlying the H$\alpha$ emission or a similar uncertainty in the [NII] contribution to the H$\alpha$ image. The plot of the [OIII]/H$\alpha$ ratio as a function of the distance from the peak of the H$\alpha$ emission is instead fairly inconclusive (Figure~11b): here the scatter in the data points dominates any trend. The scatter in Figure~11b is likely the superposition of two effects: one is the inhomogeneity of the dust reddening, which we cannot correct for with our data, the other may be the lack of a correlation between the line ratio and the distance from the peak of the H$\alpha$ emission. The presence of the second effect is confirmed by Figure~11a. In this case variations in the reddening induce small changes in the line ratio; neverthless, the plot still shows a fairly large scatter. The most straightforward interpretation is that the peak of the H$\alpha$ emission is not the absolute peak of the ionized gas emission. Unlike the case of NGC5253 (Figure~6 and discussion in Calzetti et al. 1997), the gas morphology in the center of NGC5236 cannot be described as the effect of a main central ionizing source, but is far more complex with multiple emission peaks of almost comparable intensity (see Figure~4b). As in NGC5253, the largest values of the [SII]/H$\alpha$ ratio are reached in the regions of lowest H$\alpha$ surface brightness (Figure~12). Here, however, the scatter is much larger than in the case of NGC5253, probably due to the insufficient extinction correction of the H$\alpha$ surface brightness and uncertain correction for the underlying stellar absorption. Also, the H$\alpha$ surface brightness limit reached for NGC5236 is about 3 times higher than for NGC5253, due to shorter exposure times in both the 6563/78 and the R band filters, only partially compensated by the fact that the red continuum of NGC5236 is about 5 times brighter than that of the other galaxy. The histogram of the number of area bins having a specific value of the H$\alpha$ surface brightness (Figure~13) shows that the two galaxies have similar behavior (slope and upper limit) at the high brightness end, but differ quite substantially in the low surface brightness regime. In particular, NGC5236 does not show the `break' in the power-law trend shown by NGC5253. Thus, the transition between HII regions and DIG is less clear in the spiral galaxy. Such variety of behaviors among galaxies has been previously observed (Wang, Heckman \& Lehnert 1998). The starburst population of NGC5236 is not easily separated from the underlying stellar population, because of the presence of uneven dust/stellar population distribution across the entire galaxy. The direct comparison of the U and H$\alpha$ images, discussed above (Figure~4, panels a-c), shows that the morphology of the blue stars closely follows that of the ionized gas. This suggests that the U-band image of the center of NGC5236 is tracing the optically detectable starburst population. Figure~14 shows the azimuthally-averaged profile of the surface brightness of both H$\alpha$ and U-band in annuli of increasing distance from the center. The two surface brightnesses have similar half-light radii, around 5$^{\prime\prime}$.5, with profiles of almost identical shape. In both cases the assumption is that the emission of the underlying non-starburst stellar population is fairly constant out to $\sim$40$^{\prime\prime}$ (see Figure~14). The almost identical values of the half-light radii also confirms that the gas emission in the center of NGC5236 is as extended as the starburst population, and there is no evidence for `leakage' of ionized photons beyond the starburst region. In summary, the characteristics of the nebular emission in the starbursting center of NGC5236 are typical of gas excited predominantly by photoionization. This conclusion should be regarded as preliminary, for the following three reasons: (1) the continuum-subtracted narrow-band images of NGC5236 are less deep than those of NGC5253, especially the crucial [OIII] image; (2) the contribution of the [NII] lines to the H$\alpha$ emission is almost 4 times higher in NGC5236 than in NGC5253; the impact of variations of the [NII]/H$\alpha$ line ratio on the [SII]/H$\alpha$ map is moderate if the [NII]/H$\alpha$ changes by less than 50\%~ but increases for larger variations; (3) we do not have appropriate information to correct for variations of the dust reddening across the central region. Our [OIII] maps of NGC5236 contain limited information despite the long exposure times (see Figure~4f). This is a consequence of the large metallicity in this galaxy: higher metallicities correspond to lower intensities for the O$^{++}$ ion emission. Incidentally, while [SII]/H$\alpha$ has typical values in the range 0.15--0.4 in the Northern spiral arm of the galaxy, the ratio covers the range 0.2--1.4 and has a more extended cross section in the Southern spiral arm (Figure~2c). A difference between the two arms is evident neither from the stellar continuum morphology, nor from the colors. Since we cannot easily discriminate between dust reddening and age of the stellar population, a difference between the intrinsic stellar populations in the two arms cannot be excluded. \section{Discussion} The investigation of the morphology and physical conditions of the ionized gas in the galaxy pair NGC5253/NGC5236 demonstrates that the ionization structure is different in the two central starbursts, and probably reflects the morphological difference of the host galaxies. Both galaxies responded to a possibly common trigger with a large scale central starburst (size of order 500~pc), but the starburst in NGC5253 was probably more extended in the past (Caldwell \& Phillips 1989). NGC5236 is also experiencing a somewhat milder event in its center, with a SFR about 1/3--1/4 that of NGC5253, although dust reddening corrections are uncertain. The major difference between the two starbursts appears to be, however, in their impact on the surrounding ISM, as discussed below. \subsection{NGC5253} \subsubsection{The Diffuse Ionized Medium} The central concentration of the blue stars relative to the ionized gas seen in NGC5253 is typical of intense star-forming events, ranging from giant HII regions like 30~Dor or NGC604 (Kennicutt \& Chu 1994, Mu\~noz-Tu\~non 1994), to Blue Compact Dwarf galaxies (e.g., Meurer et al. 1992). The structure of the starburst in NGC5253 closely matches these expectations. The extended ionized gas emission is probably the manifestation of the hydrodynamic effects of radiation pressure and stellar winds/supernovae on the ISM surrounding the starburst, in combination with the photoionization effects of luminous sources. This conceptual model accounts for complex structures, like bubbles and filaments, in the low surface brightness H$\alpha$ emission. In terms of luminosity, the H$\alpha$ flux which is not directly associated with massive stars and, thus, can be associated with the DIG is 13\%~ of the total. This fraction is for the projected emission only; we do not attempt to extrapolate it to the entire 3-dimensional distribution as this would require ``guessing'' the gas distribution along the line of sight. The locus of massive stars is defined as the region where the U-band emission from the starburst is detected (Figure~3a). Thus about one-tenth, and probably more, of the H$\alpha$ emission in NGC5253 is spatially separated from the source of ionization. \subsubsection{The Contribution of Shocks and Other Processes} The peripheral regions of the ionized emission in NGC5253 show the presence of a shock or other non-photoionization component in the gas excitation mechanism. Although the candidate shock structures we identify in the previous section (Figure~3h) need spectroscopic confirmation, it is clear that some contribution to the nebular line emission from non-photoionization (`shock') excitation needs to be present to explain the observed line ratios. The morphology of the `shocked' gas (Figure~3h) closely follows that of the bubbles South-West of the starburst center and the filaments in the Northern region. In particular, the bulk of the `shocked' area is located at the position of the major-axis bubble, and extends in the direction of the finger of soft X-ray emission in the map of Martin \& Kennicutt (1995) and along the axis of the X-ray emission detected by Strickland \& Stevens (1999). The X-ray finger of Martin \& Kennicutt extends for about 2$^{\prime}$ away from the galaxy center, not very different from the 1$^{\prime}$.3 scale of the `shocked' region. A possibility therefore exists that the hot gas in overlapping superbubbles, which is most likely responsible for the extended X-ray emission (Strickland \& Stevens 1999), also affects the optical emission line spectrum. Given the relatively low photon luminosity of the X-ray sources in NGC~5253 ($\leq$10$^{44}$~photons~cm$^{-2}$~s${-1}$), the X-rays should make at modest contribution to the level of photoionization in most of NGC~5253. However, in addition to shocks within the hot bubbles (see Martin \& Kennicutt 1995), there is the possibility that transition layers of warm, rapidly cooling gas exist within or on the boundaries of these regions. One example of this type of emission region are `turbulent mixing layers', such as those described by Slavin, Shull, \& Begelman (1993), which could become significant sources of emission in regions with low gas column densities. Emission line ratios from mixing layers can mimic shocks in the diagnostic emission line ratios which we have available, and this possibility therefore merits future examination. However, since hot ejecta from the central starburst can be responsible for shocking the outer regions of the DIG, we discuss below the viability of shocks to explain the observed line ratios. A size scale of about 1$^{\prime}$.3 for the shocked region corresponds to a physical size of 1.45~kpc, which, for an expansion velocity of 35~km~s$^{-1}$ (Marlowe et al. 1995), corresponds to an age of 40~Myr for the bubble. Star formation within the central $\sim$300~pc has been ongoing for $\sim$100~Myr, long enough to drive such a shock (Calzetti et al. 1997). The low velocity values observed by Marlowe et al. represent a potential difficulty for interpreting the ionization of this region as due to shocks. However, the line-of-sight velocity may not be representative of the expansion velocity of the bubble; if we adopt a shock velocity of 100~km/s, the age of the region decreases to $\sim$15~Myr. The H$\alpha$ intensity associated with the shocked component is 2.2\%~ of the total H$\alpha$ emission, after correction for underlying stellar absorption and dust obscuration (accounting only the area for which [SII]/H$\alpha$ is detected above 5~$\sigma$, see Figure~3). This value is an upper limit, as the regions we define as `shocked' can also be partially photoionized; we assume, conservatively, that only half, or 1.1\%, of the ionized gas emission in the `shocked' areas is actually excited by shocks. This fraction, however, does not include the fainter, more extended DIG, since here [SII] is either undetected or is detected below our 5~$\sigma$ cut. Whichever the actual shocked H$\alpha$ emission fraction, it will still be a few percent at most. The H$\alpha$ flux associated with the shock is 3.2~E$-$13~erg~s$^{-1}$~cm$^{-2}$, corresponding to a luminosity of 6.1$\times$10$^{38}$~erg~s$^{-1}$ at the distance of NGC5253. We can compare this value with the amount of mechanical energy input expected from the starburst. The reddening-corrected flux density at 2,600~\AA~ from the star-forming region is F(2600)=4.5~E$-$13~erg~s$^{-1}$~cm$^{-2}$~\AA$^{-1}$, corresponding to a luminosity of $\sim$8.6$\times$10$^{38}$~erg~s$^{-1}$~\AA$^{-1}$, from the data of Calzetti et al. (1997). This estimate does not take into account the fraction of massive stars so deeply buried in dust, e.g., along the dust lane, that their accounting is missing from the UV flux; a comparison between the optical nebular emission and the radio thermal emission (Beck et al. 1996) shows that the missing fraction amounts to $\approx$20\%~ of the reddening-corrected UV flux. Given the star formation history of the galaxy and the above UV flux, the mechanical energy being deposited into the ISM by massive stars is $\sim$8.5$\times$10$^{40}$~erg~s$^{-1}$ (Leitherer \& Heckman 1995), a value very similar to what calculated by Marlowe et al. (1995) and by Martin \& Kennicutt (1995). This energy rate is about 50\% higher than the one derived for an expanding superbubble with the age and size given above and density of 12~cm$^{-3}$ (Martin \& Kennicutt 1995), using the self-similar solution of Weaver et al. (1977), and is more than sufficient to produce the observed X-ray luminosity (Strickland \& Stevens 1999). About 2.5\% of the shock input power is emitted in H$\alpha$ (Binette, Dopita \& Tuohy 1985), implying that the gas shocked by the starburst in NGC5253 can produce a total luminosity L(H$\alpha$)$\sim$2.1$\times$10$^{39}$~erg~s$^{-1}$. Most of the detected `shocked' gas component is located in the south-western quadrant relative to a sphere centered on the main cluster. If the mechanical energy is emitted with bi-polar symmetry from the central starburst (Strickland \& Stevens 1999), this quadrant is likely to receive $\approx$25\%~ of the total energy available to shocks, or L(H$\alpha$)$\sim$5.3$\times$10$^{38}$~erg~s$^{-1}$, comparable to the observed luminosity of 6.1$\times$10$^{38}$~erg~s$^{-1}$. The total mechanical energy available from the starburst can shock-excite gas to produce a total H$\alpha$ luminosity $\sim$3.4~times what observed or about one-fourth of the 13\% of H$\alpha$ emission we associate with the DIG. Observationally photoionization provides between 80\% and 90\% of the excitation of the DIG in NGC5253; this fraction is potentially lower, but not lower than $\sim$70\%, even if all the mechanical energy were available to excite the gas. Notably absent are shocks in the Eastern region; in section~3.2.5 we showed that the ionized gas associated with the dust lane in the E region is more consistent with photoionization rather than shocks (or other mechanism), even though local stellar ionizing sources have not yet been found. The absence of an obvious shock component here has another intepretation, possibly complementary to the previous one. The dust lane coincides with the position of the extremely weak CO detection in this galaxy (Turner et al. 1997). Turner et al. have interpreted the very low CO luminosity as evidence for the presence of extremely metal-poor gas in the area, possibly infalling gas which is feeding the central starburst. If the metallicity along the dust lane is lower than the average in the starburst, the [OIII]/H$\beta$ and the [SII]/H$\alpha$ ratios are expected to be higher and lower, respectively, than the average. Thus, presence of shocks along this region would go undetected by our method, as we are assuming a uniform metallicity for the ionized gas across the entire central region. \subsubsection{The Structure of the Starburst} Shocks in NGC5253 appear to have a preferential direction along the galaxy's major axis. In addition the region along the dust lane, namely along the optical minor axis, appears dominated by photoionization. These facts, together with the HI kinematical data of Kobulnicky \& Skillman (1995) and the CO detection of Turner et al. (1997) suggest the following picture for the starburst in NGC5253: \begin{enumerate} \item The central star formation is being fueled by gas which is either infalling along the minor axis, as suggested by Turner et al., or is located in a `disk' rotating about the major axis, as suggested by Kobulnicky \& Skillman on the basis of the HI rotation. \item Hot ejecta from supernovae explosions and stellar winds drive the expansion of the ISM described in Martin \& Kennicutt (1995) and in Strickland \& Stevens (1999). The expansion is driven mostly in the direction perpendicular to the disk/infalling~gas, where the gas density is lower (or, alternatively, is driven along the gas disk rotation axis). Hot ejecta from the central starburst, thus, shock the gas preferentially along the optical major axis. Also, the shocked gas is detected mainly along the south-western side of the major axis; the propagation of the northward shocks may be prevented by the high density region of the infalling-gas/rotating-gas-disk, which is located in the northern side of the star forming site. \end{enumerate} This general picture is rather different from the one found for other dwarf galaxies by Marlowe et al. (1995), where the location of bubbles is preferentially along the optical minor axis, suggesting that the ionized gas expands mainly in the direction perpendicular to the plane of the galaxy. One can speculate that past interaction with NGC5236 played a role in the geometry of NGC5253; the massive `companion' is located in the NW quadrant, at position angle P.A.$\sim-$20$^o$. This direction is only $\sim$25$^o$ away from the minor axis and almost orthogonal, just $\sim$15--20$^o$ away, to the direction of the shocked gas. Thus, the direction along which the encounter between the two galaxies happened may have determined the initial gas infall and subsequent gas expansion directions in NGC5253. \subsection{NGC5236} For NGC5236, the gas morphology allows a more straightforward interpretation than in NGC5253, although our conclusions are limited by the shallowness of the [OIII] map and by presence of large amount of dust in combination with the lack of an H$\beta$ image to perform dust reddening corrections. The H$\alpha$ emission correlates fairly well with the blue emission from the ionizing stars, and there is no evidence for extended ionized gas emission. The strong spatial overlap between blue stars and ionized gas indicates that we are not seeing any `bonafide' DIG in the center of this galaxy; rather, ionization appears to be a local process. The [SII]/H$\alpha$ values fall into the photoionization range even after allowing for large uncertainties in the underlying stellar absorption and in the [NII] contribution, suggesting very little, if any, contribution from a shock or other non-photoionization component. The central starburst in NGC5236 is a milder perturbation on its giant spiral galaxy host than the one in NGC5253. The past encounter with NGC5253 may have produced a stellar bar and/or triggered the gas inflow towards the center along the bar. The presence of an inner Lindblad resonance (Telesco et al. 1993) is the additional ingredient needed to produce a ring starburst (Shlosman, Begelman \& Frank 1990). The absence of a shocked component in the ionized gas can be interpreted as an effect of the deep potential well in the center of NGC5236. The more massive the galaxy, the harder it is to disrupt the gas disk, especially in a high density center (De Young \& Heckman 1994, MacLow \& Ferrara 1998). In NGC5236, the mechanical energy being deposited into the ISM by the central starburst is $\approx$2$\times$10$^{40}$~erg~s$^{-1}$. This amount of power is probably inadequate to disrupt the ISM in the dense nuclear disk of NGC5236. The higher gas densities limit the growth of superbubbles, while the larger gravitational forces make expansion out of the plane more difficult. The calculations of De Young \& Heckman refer to disruption along the minor axis, while we are looking at ISM expansion parallel to the gas disk (NGC5236 is seen nearly face-on). However, if the ISM is left intact along the minor axis, it is even more likely to remain confined in the center of the disk. Lack of obvious filamentary structures, bubbles and superbubbles in the ionized gas of NGC5236 fits in this picture. \section{Conclusions} The analysis of the starburst galaxy `odd couple' NGC5253 and NGC5236 reveals very different ionized gas morphologies. The metal-poor, dwarf member of the pair, NGC5253, has the DIG emission typical of intense bursts of star formation, that accounts for about 13\% of the projected H$\alpha$ luminosity. A small ($\sim$10--20\%), but not negligible, fraction of the DIG is ionized by shocks or other non-photoionization mechanism; this implies that between 80\% and 90\% of the H$\alpha$ emission from the DIG is due to photoionization from massive stars. The morphology of the `shocked' gas is quite peculiar, as it extends along the optical major axis, orthogonal to the direction from which presumably the gas is feeding the central starburst. If the `shocked' gas corresponds to one or more expanding bubbles driven by the central starburst, the in-plane morphology indicates that the metals ejected from the central region will remain inside the galaxy, and will not be lost in the intergalactic medium (Mac Low \& Ferrara 1998). Photoionization of the DIG from massive stars means that about 10\%~ of the ionizing photons are escaping from the central starburst zone. In the metal-rich, grand-design spiral member of the pair, NGC5236, there is no clear detection of a DIG component in the starbursting nuclear region and the ionized gas does not show an obvious shocked component. This is probably because the gas is confined to the center by the deep potential well of the galaxy, and remains near the massive stars responsible for its photoionization. The fraction of DIG to total ionized gas in both starbursts is much smaller, probably due to projection effects, than the 20--50\% measured in less active star-forming galaxies (Ferguson et al. 1996a, Wang et al. 1997, 1998). Although we can only place a lower limit to the amount of DIG in the two starbursts, it is unlikely that the actual fraction will be higher than what has been observed in other galaxies. \acknowledgments D.C., C.J.C., and A.L.K. thank the Carnegie Observatories for the hospitality and for granting them observing time at the Las Campanas Observatory. D.C. thanks Crystal Martin for useful discussions and suggestions during the analysis of the images. Part of this manuscript was written at the Kitt Peak 2.1-m telescope, during a stormy night. \clearpage \begin{deluxetable}{lrrrr} \large \tablecaption{Summary of Filters, Calibrations, Limiting Sensitivities. \label{tbl-1}} \tablewidth{0pt} \tablehead{ \colhead{Filter} & \colhead{$\lambda_c$\tablenotemark{a}} & \colhead{FWHM} & \colhead{Flux Conversion\tablenotemark{b}} & Detection Limit\tablenotemark{c}\\ \colhead{ } & \colhead{(\AA)} & \colhead{(\AA)} & \colhead{(erg cm$^{-2}$ \AA$^{-1}$ ADU$^{-1}$)} & } \startdata Harris U & & & 2.139~E$-$18 (5\%) & 4.6~E$-$20 \nl Harris V & & & 2.550~E$-$19 (2\%) & 4.3~E$-$20 \nl Harris R & & & 1.160~E$-$19 (1.5\%) & 3.6~E$-$20\nl 5000/70 & 4994 & 77 & 4.697~E$-$18 (4\%) & 7.4~E$-$18 \nl 6563/78 & 6568 & 68 & 2.361~E$-$18 (3\%) & 8.0~E$-$18 \nl 6737/76 & 6747 & 91 & 1.796~E$-$18 (3\%) & 9.3~E$-$18 \nl \enddata \tablenotetext{a}{The central wavelength of the narrow band filters.} \tablenotetext{b}{The flux zeropoint is given with, in parenthesis, the internal uncertainty.} \tablenotetext{c}{The limiting surface brightness is in erg~s~cm$^{-2}$~arcsec$^{-2}$ for the continuum-subtracted narrow band images ([OIII], H$\alpha$, and [SII]) and it is a surface brightness density in erg~s~cm$^{-2}$~arcsec$^{-2}$~\AA$^{-1}$ for the broad band images. The values refer to 1~$\sigma$ detection limits of the deepest images obtained in this project (see Table~2), rebinned to a resolution of 1$^{\prime\prime}$.3, namely 5$\times$5~pix$^2$.} \end{deluxetable} \clearpage \begin{deluxetable}{lrrrr} \large \tablecaption{Summary of Exposure Times. \label{tbl-2}} \tablewidth{0pt} \tablehead{ \colhead{ } & \colhead{NGC5253} & \colhead{NGC5253} & \colhead{NGC5236} & \colhead{NGC5236}\\ \colhead{Filter} & \colhead{Exp. Time\tablenotemark{a}} & \colhead{Exp. Time\tablenotemark{a}} & \colhead{Exp. Time\tablenotemark{a}} & \colhead{Exp. Time\tablenotemark{a}}\\ \colhead{ } & \colhead{(s)} & \colhead{(s)} & \colhead{(s)} & \colhead{(s)} } \startdata Harris U & 900. & 8400. & 210. & 4500. \nl Harris V & 120. & 1200. & 190. & 920. \nl Harris R & 240. & 1520. & 260. & \nodata \nl 5000/70 & 180. & 10200. & 870.& 3000. \nl 6563/78 & 240. & 3900. & 780. & 1440. \nl 6737/76 & 12000. & \nodata & 1560. & 3840. \nl \enddata \tablenotetext{a}{The second and fourth columns are the total exposure times of the final unsaturated image for NGC5253 and NGC5236, respectively; the third and fifth columns are the total exposure times of the final images which have the central $\sim$5$^{\prime\prime}$--20$^{\prime\prime}$ of the galaxies saturated. Each unsaturated/saturated image is the combination of multiple exposures in the range 30--1200~s.} \end{deluxetable} \clearpage
\section{Introduction} \label{sec:intro} \par The hybrid inflationary scenario \cite{hybrid}, which can reproduce the measurements of the cosmic background explorer (COBE) \cite{cobe} with more or less `natural' values of the relevant coupling constants, is almost automatically realized \cite{lyth,dss} in supersymmetric grand unified theories (GUTs). In particular, a moderate extension of the minimal supersymmetric standard model (MSSM) based on a left-right symmetric gauge group provides \cite{lss} a `natural' framework for the implementation of hybrid inflation. The $\mu$ problem of MSSM can be easily resolved \cite{dls} in the context of this model by coupling the inflaton system to the electroweak higgs superfields. \par At the end of inflation, the inflaton (oscillating system) predominantly decays into electroweak higgs superfields, thereby `reheating' the universe. However, its subdominant decay mode to right handed neutrinos leads \cite{atmo}, via their subsequent decay, to the production of a primordial lepton asymmetry in the universe. Nonperturbative electroweak sphaleron effects, which violate baryon and lepton number, then partially transform this asymmetry to the observed baryon asymmetry of the universe (BAU). \par We analyze the consequences of this baryogenesis mechanism on $\nu_{\mu}-\nu_{\tau}$ mixing. We find \cite{atmo} that, for masses of $\nu_{\mu}$, $\nu_{\tau}$ which are consistent with the small angle MSW resolution of the solar neutrino problem and the recent results of the SuperKamiokande experiment \cite{superk}, maximal $\nu_{\mu}-\nu_{\tau}$ mixing can be achieved. The required values of the relevant parameters are, however, quite small. \par In Sec.\ref{sec:hybrid}, we review the nonsupersymmetric (Sec.\ref{subsec:nonsusy}) as well as the supersymmetric (Sec.\ref{subsec:susy}) version of the hybrid inflationary scenario. In Sec.\ref{sec:baryons}, we discuss baryogenesis through a primordial leptogenesis. In particular, Sec.\ref{subsec:leptons} is devoted to the generation of the primordial lepton number. The topologically nontrivial structure of the vacuum in gauge theories and the resulting nonperturbative baryon and lepton number violating phenomena in the standard model are analyzed in Sec.\ref{subsec:sphaleron}. The rate of these phenomena at finite temperatures is calculated by employing electroweak sphalerons and the final BAU is estimated. Finally, in Sec.\ref{sec:lr} the supersymmetric model based on a left-right symmetric gauge group is presented. In particular, the solution of the $\mu$ problem (Sec.\ref{subsec:mu}), inflation (Sec.\ref{subsec:inf}) and leptogenesis (Sec.\ref{subsec:reheatlepto}) are sketched. \section{Hybrid Inflation} \label{sec:hybrid} \subsection{The non Supersymmetric Version} \label{subsec:nonsusy} \par The most important disadvantage of most inflationary scenarios was that they needed extremely small coupling constants in order to reproduce the results of COBE \cite{cobe}. This difficulty was overcome some years ago by Linde \cite{hybrid} who proposed, in the context of nonsupersymmetric GUTs, a clever inflationary scenario known as hybrid inflation. The idea was to use two real scalar fields $\chi$ and $\sigma$ instead of one that was normally used. The field $\chi$ provides the vacuum energy which drives inflation while $\sigma$ is the slowly varying field during inflation. The main advantage of this scenario is that it can reproduce the observed temperature fluctuations of the cosmic background radiation (CBR) with `natural' values of the parameters in contrast to previous realizations of inflation (like the `new' \cite{new} or `chaotic' \cite{chaotic} inflationary scenarios). The potential utilized by Linde is \begin{equation} V ( \chi, \sigma)= \kappa^2 \left( M^2 - \frac {\chi^2}{4}\right)^2 + \frac{\lambda^2 \chi^2 \sigma^2} {4} + \frac {m^2\sigma^2}{2}~~, \label{eq:lindepot} \end{equation} where $\kappa,~\lambda$ are dimensionless positive coupling constants and $M$, $m$ mass parameters. The vacua lie at $\langle \chi\rangle=\pm 2 M$, $\langle \sigma \rangle=0$. Putting $m$=0, for the moment, we observe that the potential possesses an exactly flat direction at $\chi=0$ with $V(\chi=0 ,\sigma)=\kappa^2 M^4$. The mass squared of the field $\chi$ along this flat direction is given by $m^2_\chi=-\kappa^2 M^2+\frac{1}{2} \lambda^2 \sigma^2$ and remains nonnegative for $\sigma \geq \sigma_c = \sqrt{2}\kappa M/\lambda $. This means that, at $\chi=0$ and $\sigma \geq \sigma_c$, we obtain a valley of minima with flat bottom. Reintroducing the mass parameter $m$ in Eq.(\ref{eq:lindepot}), we observe that this valley acquires a nonzero slope. A region of the universe, where $\chi$ and $\sigma$ happen to be almost uniform with negligible kinetic energies and with values close to the bottom of the valley of minima, follows this valley in its subsequent evolution and undergoes inflation. The quadrupole anisotropy of CBR produced during this inflation can be estimated to be \begin{equation} \left(\frac {\delta T}{T}\right)_{Q} \approx \left(\frac {16 \pi}{45}\right)^{1/2} \frac{\lambda \kappa^2 M^5} {M^3_Pm^2}~, \label{eq:lindetemp} \end{equation} where $M_{P}=1.22\times 10^{19}{\rm{GeV}}$ is the Planck scale. The COBE~\cite{cobe} result, $(\delta T/T)_{Q} \approx 6.6 \times 10^{-6}$, can then be reproduced with $M\approx 2.86\times 10^{16}$ GeV, the supersymmetric GUT vacuum expectation value (vev), and $m \approx 1.3~\kappa \sqrt {\lambda}\times 10^{15}$ GeV $\sim 10^{12}$ GeV for $\kappa, \lambda \sim 10^{-2}$. Inflation terminates abruptly at $\sigma=\sigma_{c}$ and is followed by a `waterfall', i.e., a sudden entrance into an oscillatory phase about a global minimum. Since the system can fall into either of the two available global minima with equal probability, topological defects are copiously produced if they are predicted by the particular particle physics model one is considering. \subsection{The Supersymmetric Version} \label{subsec:susy} \par The hybrid inflationary scenario is \cite{lyth} `tailor made' for application to supersymmetric GUTs except that the mass of $\sigma$, $m$, is unacceptably large for supersymmetry, where all scalar fields acquire masses of order $m_{3/2} \sim 1$ TeV (the gravitino mass) from soft supersymmetry breaking. To see this, consider a supersymmetric GUT with a (semi-simple) gauge group $G$ of rank $\geq 5$ with $G \to G_S$ (the standard model gauge group) at a scale $M \sim 10^{16}$ GeV. The spectrum of the theory below $M$ is assumed to coincide with the MSSM spectrum plus standard model singlets so that the successful predictions for $\alpha_{s}$, ${\rm{sin}}^{2} \theta_{W}$ are retained. The theory may also possess global symmetries. The breaking of $G$ is achieved through the superpotential \begin {equation} W =\kappa S( \phi\bar{\phi}- M^2), \label{eq:superpot} \end {equation} where $\phi$, $\bar{\phi}$ is a conjugate pair of standard model singlet left handed superfields which belong to nontrivial representations of $G$ and reduce its rank by their vevs and $S$ is a gauge singlet left handed superfield. The coupling constant $\kappa$ and the mass parameter $M$ can be made positive by phase redefinitions. This superpotential is the most general renormalizable superpotential consistent with a $U(1)$ R-symmetry under which $W \to e^{i\theta} W,~S \to e^{i \theta}S,~\phi \bar {\phi} \to \phi\bar{\phi}$ and gives the potential \begin{equation} V=\kappa^2 \mid M^2-\phi\bar{\phi}\mid^2 +\kappa^2 \mid S \mid^2 (\mid \phi \mid^2 +\mid \bar{\phi}\mid^2) +{\rm{ D-terms}}. \label{eq:hybpot} \end{equation} Restricting ourselves to the D flat direction $\phi= \bar{\phi}^* $ which contains the supersymmetric minima and performing appropriate gauge and R- transformations, we can bring $S$, $\phi$, $\bar{\phi}$ on the real axis, i.e., $S \equiv \sigma/\sqrt{2}$, $\phi=\bar{\phi} \equiv \chi/2$, where $\sigma$, $\chi$ are normalized real scalar fields. The potential then takes the form in Eq.(\ref{eq:lindepot}) with $\kappa = \lambda$ and $m=0$ and, thus, Linde's potential for hybrid inflation is almost obtainable from supersymmetric GUTs but without the mass term of $\sigma$ which is, however, of crucial importance since it provides the slope of the valley of minima necessary for inflation. \par One way to obtain a valley of minima useful for inflation is \cite{lp} to replace the trilinear term in $W$ in Eq.(\ref{eq:superpot}) by the next order nonrenormalizable coupling. Another way, which we will adopt here, is \cite{dss} to keep the renormalizable superpotential in Eq.(\ref{eq:superpot}) and use the radiative corrections along the inflationary valley ($\phi=\bar{\phi}= 0$~, $S > S_{c}\equiv M$). In fact, due to the mass splitting in the supermultiplets $\phi$, $\bar{\phi}$ caused by the supersymmetry breaking `vacuum' energy density $\kappa^{2} M^{4}$ along this valley, there are important radiative corrections. At one-loop, the inflationary potential is given \cite{dss,lss} by $$ V_{\rm{eff}}(S) = \kappa^{2}M^{4} $$ \begin{equation} \left[ 1 +\frac{\kappa^2}{32 \pi^2} \left(2\ln\left( \frac{\kappa^{2}S^{2}}{\Lambda^2}\right)+ \left(\frac{S^2}{S_{c}^2}-1\right)^2 \ln\left(1-\frac{S_{c}^2}{S^2}\right) +\left(\frac{S^2}{S_{c}^2}+1\right)^2 \ln\left(1+\frac{S_{c}^2}{S^2}\right)\right) \right]~, \label{eq:veffexact} \end{equation} where $\Lambda$ is a suitable mass renormalization scale. For $S$ sufficiently larger than $S_{c}$, \begin{equation} V_{{\rm{eff}}} (S) = \kappa^2 M^4 \left[ 1 + \frac{\kappa^2}{16\pi^2} \left( \ln \left(\frac{\kappa^{2} S^{2}}{\Lambda^2}\right) + \frac{3}{2} - \frac{S_c^4}{12S^4} + \cdots \right)\right]~. \label{eq:veff} \end{equation} Using this effective potential, one finds that the cosmic microwave quadrupole anisotropy \begin{equation} \left(\frac{\delta T}{T}\right)_{Q} \approx 8 \pi \left(\frac{N_{Q}}{45}\right)^{1/2} \frac{x_{Q}}{y_{Q}}\left(\frac{M}{M_{P}}\right)^2~. \label{eq:cobe} \end{equation} Here, $N_{Q}$ is the number of e-foldings experienced by the universe between the time the quadrupole scale exited the inflationary horizon and the end of inflation and $y_{Q}=x_{Q}(1-7/(12x_{Q}^2)+\cdots)$ with $x_{Q}=S_{Q}/M$, $S_{Q}$ being the value of the scalar field $S$ when the scale which evolved to the present horizon size crossed outside the de Sitter (inflationary) horizon. Also, from Eq.(\ref{eq:veff}), one finds \begin{equation} \kappa \approx \frac{8\pi^{3/2}}{\sqrt{N_{Q}}} ~ y_{Q}~\frac{M}{M_{P}}~. \label{eq:kappa} \end{equation} \par The inflationary phase ends as $S$ approaches $S_{c}$ from above. Writing $S=xS_{c}$, $x=1$ corresponds to the phase transition from $G$ to $G_{S}$ which, as it turns out, more or less coincides with the end of the inflationary phase as one deduces from the slow roll conditions \cite{dss,lazarides}. Indeed, the $50-60$ e-foldings needed for the inflationary scenario can be realized even with small values of $x_{Q}$. For definiteness, we will take $x_{Q}\approx 2$. From COBE \cite{cobe} one then obtains $M \approx 5.5 \times 10^{15}~{\rm{GeV}}$ and $\kappa\approx 4.5\times 10^{-3}$ for $N_{Q}\approx 56$. Moreover, the primordial density fluctuation spectral index $n \simeq 0.98$. We see that the relevant part of inflation takes place at $S\sim 10^{16}~{\rm{GeV}}$. An interesting consequence of this is \cite{lyth,lss,sugra} that the supergravity corrections can be negligible. \par In conclusion, it is important to note that the superpotential $W$ in Eq.(\ref{eq:superpot}) leads to hybrid inflation in a `natural' way. This means that a) there is no need of very small coupling constants, b) $W$ is the most general renormalizable superpotential allowed by the gauge and R- symmetries, c) supersymmetry guarantees that the radiative corrections do not invalidate inflation, but rather provide a slope along the inflationary trajectory which drives the inflaton towards the supersymmetric vacua, and d) supergravity corrections can be brought under control so as to leave inflation intact. \section{Baryogenesis via Leptogenesis} \label{sec:baryons} \subsection{Primordial Leptogenesis} \label{subsec:leptons} \par In most hybrid inflationary models, it is not convenient to produce the observed BAU in the customary way, i.e., through the decay of color $~3,~\bar{3}~$ fields $(g,~g^{c})$. Some of the reasons are the following: i) For theories where leptons and quarks belong to different representations of the unifying gauge group $G$ (which is the case, for example, for $G=G_{LR}\equiv SU(3)_c\times SU(2)_L\times SU(2)_R \times U(1)_{B-L}$ or $SU(3)_c \times SU(3)_L \times SU(3)_R$), the baryon number can be made almost exactly conserved by imposing an appropriate discrete symmetry. In particular, for $G=G_{LR}$, we can impose \cite{baryonparity} a discrete symmetry under which $q\rightarrow - q$, $q^{c} \rightarrow - q^{c}$, $\bar{q} \rightarrow - \bar{q}$, $\bar{q}^{c} \rightarrow - \bar{q}^{c}$ and all other superfields remain invariant ($q,~q^{c},~\bar{q},~\bar{q}^{c}$ are superfields with the quantum numbers of the quarks, antiquarks and their conjugates respectively). ii) For theories where such a discrete symmetry is absent, we could, in principle, use as inflaton a pair of conjugate standard model singlet superfields $N$, $\bar{N}$ which decay into $g$, $g^{c}$. For $G=SU(3)_c \times SU(3)_L \times SU(3)_R$, $N$ ($\bar{N}$) could be the standard model singlet component of the (1, $\bar{3}$, 3) ( (1, 3, $\bar{3}$) ) superfields with zero $U(1)_{B-L}$ charge. But this is again unacceptable since the breaking of $SU(3)_c \times SU(3)_L \times SU(3)_R$ by the vevs of $N$, $\bar{N}$ predicts \cite{trinifiedmonopoles} magnetic monopoles which can then be copiously produced after inflation. Also the gravitino constraint \cite{gravitino} on the `reheat' temperature, $T_{r} \stackrel{_{<}}{_{\sim }} 10^9$ GeV, implies $m_g \stackrel{_{<}}{_{\sim }} 10^{10}$ GeV (~from the coupling ($~m_g / \langle N\rangle) Ngg^{c}$~) leading to strong deviation from MSSM and possibly proton decay. \par So it is preferable to produce first a primordial lepton asymmetry \cite{lepto} which can then be partially turned into the observed baryon asymmetry of the universe by the nonperturbative sphaleron effects \cite{sphaleron} of the electroweak sector. In the particular model based on $G_{LR}$ which we will consider later, this is the only way to produce the BAU since the inflaton decays into higgs superfields and right handed neutrinos. The subsequent decay of right handed neutrinos into ordinary higgs particles (higgsinos) and light leptons (sleptons) can produce the primordial lepton asymmetry. It is important, though, to ensure that this primordial lepton asymmetry is not erased \cite{turner} by lepton number violating $2 \rightarrow 2$ scattering processes such as $ l l \rightarrow h^{(1)}\,^{*} h^{(1)}\,^{*}$ or $l h^{(1)} \rightarrow \bar{l} h^{(1)}\,^{*}$ ($l$ represents a lepton doublet and $h^{(1)}$ the higgs superfield which couples to the up type quarks) at all temperatures between $T_{r}$ and 100 GeV. This is automatically satisfied since the primordial lepton asymmetry is protected \cite{ibanez} by supersymmetry at temperatures between $T_{r}$ and $T \sim 10^{7}$ GeV, and for $T \stackrel{_{<}}{_{\sim }} 10^{7}$ GeV, these $2 \rightarrow 2$ scattering processes are well out of equilibrium provided \cite{ibanez} $m_{\nu_{\tau}}\stackrel{_<}{_\sim} 10~{\rm{eV}}$, which readily holds in our case (see below). \par The lepton asymmetry produced by the out-of-equilibrium decay ($M_{\nu^{c}_{i}} \gg T_{r})$ of the right handed neutrinos $\nu^{c}_{i}$, which emerged from the inflaton decay, is \cite{lepto} \begin{equation} \frac {n_{L}}{s} \approx - \frac{3}{16 \pi} \frac {T_{r}}{m_{infl}}\sum_{l \neq i} g(r_{li}) \frac{{\rm{Im}}(U~M^{D}\,^{\prime} ~M^{D}\,^{\prime}\,^{\dagger}~U ^{\dagger})^{2}_{il}} {|\langle h^{(1)}\rangle|^{2}(U~M^{D}\,^{\prime} ~M^{D}\,^{\prime}\,^{\dagger}~U ^{\dagger})_{ii}}~, \label{eq:genlept} \end{equation} where $n_L$ and $s$ are the lepton number and entropy densities, $m_{infl}$ in the inflaton mass, $M^{D}\,^{\prime}$ is the diagonal `Dirac' mass matrix, $U$ a unitary transformation so that $UM^{D}\,^{\prime}$ is the `Dirac' mass matrix in the basis where the `Majorana' mass matrix of $\nu^{c}$~'s is diagonal and $|\langle h^{(1)} \rangle| \approx 174~{\rm{GeV}}$ for large ${\rm tan}\beta$. The function \begin{equation} g(r_{li}) = r_{li}~\ln \left(\frac {1 + r^{2}_{li}} {r^{2}_{li}}\right)~,~r_{li} = \frac {M_{l}} {M_{i}}~, \label{eq:gfunction} \end{equation} with $g(r) \sim 1/r$ as $r \rightarrow \infty$. Here we have taken into account the following prefactors: i) At `reheat', $n_{infl} m_{infl} = (\pi^{2}/30) g_{*} T^{4}_{r}$ ($n_{infl}$ is the inflaton number density and $g_{*}$ the effective number of massless degrees of freedom) which together with the relation $s = (2 \pi^{2}/45) g_{*}T^{3}_{r}$ implies that $n_{infl}/s = (3/4)(T_{r}/m_{infl})$. ii) Since each inflaton decays into two $\nu^{c}$~'s, their number density $n_{\nu^{c}} = 2 n_{infl}$ which then gives $n_{\nu^{c}}/s= (3/2)(T_{r}/m_{infl})$. iii) Supersymmetry gives an extra factor of two. \subsection {Sphaleron Effects} \label{subsec:sphaleron} \par To see how the primordial lepton asymmetry partially turns into the observed BAU, we must first discuss the nonperturbative baryon ($B$-) and lepton ($L$-) number violation \cite{thooft} in the standard model. Consider the electroweak gauge symmetry $SU(2)_{L} \times U(1)_{Y}$ in the limit where the Weinberg angle $\theta_W=0$ and concentrate on $SU(2)_{L}$ (inclusion of $\theta_{W}\neq 0$ does not alter the conclusions). Also, for the moment, ignore the fermions and higgs fields so as to have a pure $SU(2)_{L}$ gauge theory. This theory has \cite{vacuum} infinitely many classical vacua which are topologically distinct and are characterized by a `winding number' $n \in Z$. In the `temporal gauge' ($A_{0}=0$), the remaining gauge freedom consists of time independent transformations and the vacuum corresponds to a pure gauge \begin{equation} A_{i} = \frac{i}{g}~\partial _{i}g(\bar{x}) g ^{-1}(\bar{x})~, \label{eq:gauge} \end{equation} where $g$ is the $SU(2)_{L}$ gauge coupling constant, $\bar{x}$ belongs to ordinary 3-space, $i$ =1,2,3, $g(\bar{x}) \in SU(2)_{L}$, and $g(\bar{x}) \rightarrow 1 $ as $ \mid\bar{x}\mid \rightarrow \infty$. Thus, the 3-space compactifies to a sphere $S^{3}$ and $g(\bar{x})$ defines a map: $S^{3} \rightarrow SU(2)_{L}$ (with the $SU(2)_{L}$ group being topologically equivalent to $S^{3}$). These maps are classified into homotopy classes constituting the third homotopy group of $S^{3},~\pi_{3} (S^{3})$, and are characterized by a `winding number' \begin{equation} n = \int d^{3}x~\epsilon^{ijk}~{\rm{tr}} \left(\partial_{i}g(\bar{x}) g^{-1}(\bar{x})\partial_{j}g(\bar{x}) g^{-1}(\bar{x}) \partial_{k}g(\bar{x}) g^{-1}(\bar{x})\right). \label{eq:wind} \end{equation} The corresponding vacua are denoted as $\mid n\rangle$, $n\in Z$. \par The tunneling amplitude from the vacuum $\mid n_{-} \rangle$ at $t=-\infty$ to the vacuum $\mid n_{+} \rangle~$ at $t=+\infty$ is given by the functional integral \begin{equation} \langle n_{+}\mid n_{-}\rangle = \int(dA)~e^{-S(A)} \label{eq:path} \end{equation} over all gauge field configurations satisfying the appropriate boundary conditions at $t=\pm \infty$. Performing a Wick rotation, ~$x_0 \equiv t \rightarrow -i x_{4}$, we can go to Euclidean space-time. Any Euclidean field configuration with finite action is characterized by an integer topological number known as the Pontryagin number \begin{equation} q = \frac {g^{2}}{16\pi^{2}} \int d^{4}x~{\rm{tr}}\left(F^{\mu \nu} \tilde{F}_{ \mu \nu}\right)~, \label{eq:pontryagin} \end{equation} with $\mu$,$\nu$=1,2,3,4 and $\tilde{F}_{\mu \nu}= \frac {1}{2}\epsilon_{\mu \nu \lambda \rho} F^{\lambda \rho}$ being the dual field strength. But ${\rm{tr}} (F^{\mu \nu} \tilde{F}_{\mu \nu}) = \partial ^{\mu} J_{\mu}$, where $J_{\mu}$ is the `Chern-Simons current' given by \begin{equation} J_{\mu}=\epsilon_{\mu \nu \alpha \beta}~{\rm{tr}} \left(A^{\nu} F^{\alpha \beta}-\frac{2}{3}gA^{\nu}A^{\alpha} A^{\beta}\right). \label{eq:csc} \end{equation} In the `temporal gauge' ($A_0=0$), \begin{eqnarray*} q=\frac{g^{2}}{16 \pi^{2}} \int d^{4}x ~\partial^{\mu}J_{\mu}= \frac{g^{2}}{16\pi^{2}} \mathop{\Delta}_{x_{4}=\pm \infty} \int d^{3}x~J_{0} \end{eqnarray*} \begin{equation} =\frac{1}{24\pi^{2}}\mathop{\Delta}_{x_{4}= \pm \infty} \int d^{3}x~\epsilon^{ijk}~ {\rm{tr}}\left(\partial_{i} g g^{-1}\partial_{j} gg^{-1}\partial_{k}gg^{-1}\right)=n_{+}-n_{-}~. \label{eq:interpol} \end{equation} This means that the Euclidean field configurations which interpolate between the vacua $\mid n_{+}\rangle, ~\mid n_{-}\rangle$ at $x_4 =\pm \infty$ have Pontryagin number $q= n_{+}-n_{-}$ and the path integral in Eq.(\ref{eq:path}) should be performed over all these field configurations. \par For a given $q$, there is a lower bound on $S(A)$, \begin{equation} S(A) \geq \frac{8 \pi^{2}}{g^{2}}\mid q \mid~, \label{eq:lbound} \end{equation} which is saturated if and only if $F_{\mu \nu}= \pm \tilde{F} _{\mu \nu}$, i.e, if the configuration is self-dual or self-antidual. For $q$=1, the self-dual classical solution is called instanton \cite{instanton} and is given by (in the `singular' gauge) \begin{equation} A_{a \mu}(x)=\frac{2 \rho^{2}}{g(x-z)^{2}} ~\frac{\eta_{a \mu \nu} (x-z)^{\nu}} {(x-z)^{2} + \rho^{2}}~, \label{eq:instanton} \end{equation} where $\eta_{a \mu \nu}$ ($a$=1,2,3; $\mu$,$\nu$= 1,2,3,4) are the t' Hooft symbols with $\eta_{aij}= \epsilon_{aij}$ ($i$,$j$=1,2,3), $\eta_{a4i}= -\delta_{ai}$, $\eta_{ai4}=\delta_{ai}$ and $\eta_{a44}=0$. The instanton depends on four Euclidean coordinates $z_{\mu}$ (its position) and its scale (or size) $\rho$. Two successive vacua $\mid n\rangle$,~$\mid n+1\rangle$ are separated by a potential barrier of height $\propto \rho^{-1}$. The Euclidean action of the interpolating instanton is always equal to $8 \pi^{2}/g^{2}$, but the height of the barrier can be made arbitrarily small since the size $\rho$ of the instanton can be taken arbitrarily large. \par We now reintroduce the fermions into the theory and observe~\cite{thooft} that the $B$- and $L$- number currents carry anomalies, i.e., \begin{equation} \partial_{\mu} J^{\mu}_{B} = \partial _{\mu} J^{\mu}_{L} = - n _{g} \frac {g^{2}}{16 \pi ^{2}}~{\rm{tr}} (F_{\mu \nu} \tilde{F}^{\mu \nu})~, \label{eq:anomaly} \end{equation} where $n_{g}$ is the number of generations. It is then obvious that the tunneling from $\mid n_{-}\rangle$ to $\mid n_{+}\rangle$ is accompanied by a change of the $B$- and $L$- numbers, $\Delta B=\Delta L=- n_{g}q= - n_{g} (n_{+}-n_{-})$. Note that i) $\Delta (B-L)=0$, and ii) for $q$=1, $\Delta B=\Delta L=-3$ which means that we have the annihilation of one lepton per family and one quark per family and color (12-point function). \par We, finally, reintroduce the Weinberg-Salam higgs doublet $h$ with its vev given by \begin{equation} <h> = \frac {v}{\sqrt{2}}~\left(\matrix{0 \cr 1 \cr} \right)~,~v \approx 246 ~{\rm{GeV}}~. \label{eq:vev} \end{equation} It is then easy to see that the instanton ceases to exist as an exact solution. It is replaced by the so called `restricted instanton' \cite{restricted} which is an approximate solution for $\rho \ll v^{-1}$. For $\mid x-z\mid \ll \rho$, the gauge field configuration of the `restricted instanton' essentially coincides with that of the instanton and the higgs field is \begin{equation} h(x) \approx \frac {v}{\sqrt{2}} ~\left(\frac{(x-z)^{2}} {(x-z)^{2} + \rho^{2}}\right)^{1/2} \left(\matrix {0 \cr 1 \cr } \right)~~. \label{eq:restricted} \end{equation} For $\mid x-z \mid \gg \rho$, the gauge and higgs fields decay to a pure gauge and the vev in Eq.(\ref{eq:vev}) respectively. The action of the `restricted instanton' is $S_{ri}=(8 \pi^{2}/g^{2})+\pi^{2} v^{2} \rho^{2}+ \cdots$, which implies that the contribution of big size `restricted instantons' to the path integral in Eq.(\ref{eq:path}) is suppressed. This justifies {\it a posteriori} the fact that we restricted ourselves to approximate instanton solutions with $\rho \ll v^{-1}$. \par The height of the potential barrier between the vacua $\mid n\rangle,~\mid n+1\rangle$ cannot be now arbitrarily small. This can be understood by observing that the static energy of the `restricted instanton' at $x_{4}=z_{4}$ ($\lambda$ is the higgs self-coupling), \begin{equation} E_{b}(\rho) \approx \frac{3 \pi^{2}} {g^{2}}~\frac{1}{\rho} + \frac {3}{8}\pi^{2} v^{2} \rho^{2} + \frac {\lambda}{4}\pi^{2} v^{4}\rho^{3}~, \label{eq:static} \end{equation} is minimized for \begin{equation} \rho_{{\rm{min}}} = \frac {\sqrt{2}}{gv} \left(\frac {\lambda} {g^{2}}\right)^{-1/2}\left(\left(\frac {1}{64} + \frac {\lambda}{g^{2}}\right)^{1/2} - \frac {1}{8}\right)^{1/2} \sim M^{-1}_{W}~, \label{eq:rhomin} \end{equation} and, thus, the minimal height of the potential barrier turns out to be $E_{{\rm{min}}} \sim M_{W} / \alpha_W$ ($M_{W}$ is the weak mass scale and $\alpha_{W}=g^{2}/4 \pi$). The static solution which corresponds to the top (saddle point) of this potential barrier is called sphaleron \cite{sphaleronsol} and is given by \begin{equation} \bar{A} = v~\frac {f(\xi)}{\xi}~\hat{r} \times \bar{\tau}~, ~h = \frac {v}{\sqrt{2}}~t(\xi)~ \hat {r}\cdot \bar{\tau} \left (\matrix{0 \cr 1 \cr} \right), \label{eq:sphaleron} \end{equation} where $\xi=2M_{W}r$, $\hat{r}$ is the radial unit vector in ordinary 3-space and the 3-vector $ \bar{\tau}$ consists of the Pauli matrices. The functions $f(\xi),~t(\xi)$, which can be determined numerically, tend to zero as $\xi \rightarrow 0$ and to 1 as $\xi \rightarrow \infty$. The mass (static energy) of the sphaleron solution is estimated to be \begin{equation} E_{{\rm{sph}}} = \frac {2M_{W}}{\alpha_{W}}~k, ~1.5 \leq k \leq 2.7~,~ {\rm{for}} ~0 \leq \lambda \leq \infty~, \label{eq:sphmass} \end{equation} and lies between 10 and 15 TeV. \par At zero temperature the tunneling from $\mid n\rangle$ to $\mid n+1\rangle$ is utterly suppressed \cite{thooft} by the factor exp$(-8 \pi^{2} /g ^{2})$. At high temperatures, however, thermal fluctuations over the potential barrier are frequent and this transition can occur \cite{sphaleron} with an appreciable rate. For $M_{W} \stackrel{_{<}}{_{\sim }} T \stackrel{_{<}}{_{\sim }} T_{c}$ ($T_c$ is the critical temperature of the electroweak transition), this rate can be calculated \cite{sphaleron} by expanding around the sphaleron (saddle point) solution and turns out to be \begin{equation} \Gamma \approx 10^{4}~ n_{g}~ \frac {v (T)^{9}}{T^{8}} ~{\rm{exp}} (-E_{{\rm{sph}}}(T)/T)~. \label{eq:sphrate} \end{equation} Assuming that the electroweak phase transition is a second order one, $v(T)$ and $E_{{\rm{sph}}}(T) \propto (1 - T^{2}/T^{2}_{c})^{1/2}$. One can then show that $\Gamma \gg H$ ($H$ is the Hubble parameter) for temperatures $T$ between $\sim 200$ GeV and $\sim T_{c}$. Furthermore, for temperatures above $T_{c}$, where the sphaleron solution ceases to exist, it was argued~\cite{sphaleron} that we still have $\Gamma \gg H$. The overall conclusion is that nonperturbative $B$- and $L$- number violating processes are in equilibrium in the universe for cosmic temperatures $T\stackrel{_{>}}{_{\sim }} 200$ GeV. Remember that $B-L$ is conserved by these processes. \par Given a primordial $L$- number density, one can calculate \cite{turner,ibanez} the resulting $n_{B}/s$ ($n_B$ is the $B$- number density). In MSSM, the $SU(2)_{L}$ instantons produce the effective operator (in symbolic form) \begin{equation} O_{2} = (q q q l)^{n_{g}} ( \tilde {h}^{(1)} \tilde {h}^{(2)}) \tilde{W}^{4}~, \label{eq:woperator} \end{equation} and the $SU(3)_{c}$ instantons the operator \begin{equation} O_3 = ( q q u^{c} d^{c})^{n_{g}} \tilde {g}^{6}~, \label{eq:coperator} \end{equation} where $q$, $l$ are the quark, lepton $SU(2)_{L}$ doublets respectively, $u^{c}$, $d^{c}$ the up, down type antiquark $SU(2)_{L}$ singlets respectively, $h^{(1)}$, $h^{(2)}$ the higgses which couple to up, down type quarks respectively, $g$, $W$ the gluons and $W$ bosons and tilde represents their superpartners. We will assume that these interactions together with the usual MSSM interactions are in equilibrium at high temperatures. The equilibrium number density of ultrarelativistic particles $\Delta n \equiv n_{{\rm{part}}} - n_{{\rm{antipart}}}$ is given by \begin{equation} \frac {\Delta n} {s} = \frac {15 g}{4 \pi^{2} g_{*}} \left (\frac {\mu}{T}\right) \epsilon~, \label{eq:chemical} \end{equation} where $g$ is the number of internal degrees of freedom of the particle under consideration, $\mu$ its chemical potential and $\epsilon=2$ or 1 for bosons or fermions. For each interaction in equilibrium, the algebraic sum of the chemical potentials of the particles involved is zero. Solving these constraints, we end up with only two independent chemical potentials, $\mu_{q} $ and $\mu_{\tilde{g}}$, and the $B$- and $L$- asymmetries are expressed \cite{ibanez} in terms of them: $$ \frac {n_{B}}{s} = \frac {30}{4 \pi^{2}g_{*}T} (6n_{g} \mu_{q} - (4n_{g} - 9) \mu_{\tilde{g}})~, $$ \begin{equation} \frac {n_{L}}{s} = - \frac {45}{4 \pi^{2} g_{*}T} \left(\frac {n_{g}(14 n_{g} +9)} {1+2 n_{g}} \mu_{q} + \Omega (n_{g}) \mu_{\tilde{g}}\right)~, \label{eq:bla} \end{equation} where $\Omega(n_{g})$ is a known \cite{ibanez} function. Now soft supersymmetry breaking couplings come in equilibrium at $T \stackrel{_{<}}{_{\sim }}10^7$ GeV since their rate $\Gamma_{S} \approx m^{2}_{3/2} /T \stackrel{_{>}}{_{\sim }} H \approx 30~T^{2}/M_{P}$. In particular, the nonvanishing gaugino mass implies $\mu _{\tilde{g}} =0$ and Eqs.(\ref{eq:bla}) give \cite{ibanez} \begin{equation} \frac {n_{B}}{s} = \frac {4(1+2n_{g})}{22n_{g}+13} ~\frac {n_{B-L}}{s}~. \label{eq:bbminl} \end{equation} Equating $n_{B-L}/s$ with the primordial $n_{L}/s $, we have $n_{B}/s = (- 28/79) (n_{L}/s)$, for $n_{g}=3$. \section {The `Left-Right' Model} \label{sec:lr} \par We will now study in detail a moderate extension of MSSM based on the left-right symmetric gauge group $G_{LR}$ which provides \cite{lss} a suitable framework for hybrid inflation. The inflaton is associated with the breaking of $SU(2)_{R}$ and consists of a gauge singlet and a pair of $SU(2)_{R}$ doublets. The $\mu$ problem is resolved \cite{dls} by introducing \cite{lss,dls} a trilinear superpotential coupling of the gauge singlet inflaton to the electroweak higgs doublets. In the presence of gravity-mediated supersymmetry breaking, this gauge singet acquires a vev and, thus, generates \cite{dls}, via its coupling to the higgses, the $\mu$ term. \par The inflaton system, after the end of inflation, predominantly decays into higgs superfields and `reheats' the universe. Moreover, its subdominant decay into right handed neutrinos provides \cite{atmo} a mechanism for baryogenesis via leptogenesis. For $\nu_{\mu}$, $\nu_{\tau}$ masses from the small angle MSW resolution of the solar neutrino puzzle and the recent results of the SuperKamiokande experiment \cite{superk}, maximal $\nu_{\mu}-\nu_{\tau}$ mixing can be achieved \cite{atmo}. \subsection {The $\mu$ Problem} \label{subsec:mu} \par The breaking of $SU(2)_R\times U(1)_{B-L}$ is achieved by the renormalizable superpotential \begin{equation} W = \kappa S(l^c\bar l^{c}- M^2)~, \label{W} \end{equation} where $S$ is a gauge singlet chiral superfield and $l^c$, $\bar l^{c}$ is a conjugate pair of $SU(2)_R$ doublet chiral superfields which acquire superheavy vevs of magnitude $M$. The parameters $\kappa$ and $M$ can be made positive by phase redefinitions. \par The $\mu$ problem can be resolved \cite{dls} by introducing the extra superpotential coupling \begin{equation} \delta W = \lambda S h^2 =\lambda S \epsilon^{ij} h_i^{(1)}h_j^{(2)}~, \label{lambda} \end{equation} where the chiral electroweak higgs superfield $h=(h^{(1)}, h^{(2)})$ belongs to a $(1,2,2)_{0}$ representation of $G_{LR}$ and $\lambda$ can again be made positive. The scalar potential which results from the terms in Eqs.(\ref{W}) and (\ref{lambda}) is (for canonical K\"ahler potential): \begin{eqnarray} V= | \kappa l^c\bar l^{c} + \lambda h^2 - \kappa M^2|^2 +(m_{3/2}^2 + \kappa^2 |\bar l^c|^2 + \kappa^2 |l^c|^2 + \lambda^2 |h|^2)|S|^2 + m_{3/2}^2(|\bar l^c|^2 \nonumber\\ + |l^c|^2 + |h|^2) + \left (Am_{3/2}S( \kappa l^c\bar l^{c} + \lambda h^2 - \kappa M^2) + 2\kappa m_{3/2} M^{2}S+ {\rm h. c.} \right )~, \label{V} \end{eqnarray} where $m_{3/2}$ is the universal scalar mass (gravitino mass) and $A$ the universal coefficient of the trilinear soft terms. For exact supersymmetry ($m_{3/2} \rightarrow 0$), the vacua are \cite{dls} at \begin{equation} S = 0,~~~\kappa l^c\bar l^{c} + \lambda h^2 = \kappa M^2,~~~ l^c = e^{i\phi}\bar l^{c*}~~~h^{(1)}_i = e^{i\theta}\epsilon_{ij}h^{(2)j*}, \end{equation} where the last two conditions arise from the requirement of D flatness. We see that there is a twofold degeneracy of the vacuum which is lifted by supersymmetry breaking. We get two degenerate (up to $m_{3/2}^{4}$) ground states ($\kappa\neq\lambda$): the desirable (`good') vacuum at $h = 0$ and $l^c\bar l^{c} = M^2$ and the undesirable (`bad') one at $h \neq 0$ and $l^c\bar l^{c} = 0$. They are separated by a potential barrier of order $M^{2}m_{3/2}^{2}~$. \par To leading order in supersymmetry breaking, the term of the potential $V$ in Eq.(\ref{V}) proportional to $A$ vanishes, but a destabilizing tadpole term for $S$ remains: \begin{equation} 2\kappa m_{3/2}M^{2}S + {\rm h.c.}~. \label{tadpole} \end{equation} This term together with the mass term of $S$ (evaluated at the `good' vacuum) give $\langle S\rangle \approx - m_{3/2}/\kappa$ which, substituted in Eq.(\ref{lambda}), generates \cite{dls} a $\mu$ term with \begin{equation} \mu =\lambda \langle S\rangle \approx - \frac{\lambda}{\kappa}m_{3/2}~. \label{mu} \end{equation} Thus, coupling $S$ to the higgses can lead to the resolution of the $\mu$ problem. \par The model can be extended \cite{dls} to include matter fields too. The superpotential has the most general form respecting the $G_{LR}$ gauge symmetry and a global $U(1)$ R-symmetry. Baryon number is automatically implied by this R-symmetry to all orders in the superpotential, thereby guaranteeing the stability of proton. \subsection{The Inflationary Trajectory} \label{subsec:inf} \par The model has \cite{lss,dls} a built-in inflationary trajectory parametrized by $|S|$, $|S| > S_c=M$ for $\lambda>\kappa$ (see below). All other fields vanish on this trajectory. The $F_S$ term is constant providing a constant tree level vacuum energy density $\kappa^2 M^4$, which is responsible for inflation. One-loop radiative corrections (from the mass splitting in the supermultiplets $l^c$, $\bar l^{c}$ and $h$) generate a logarithmic slope \cite{dss} along the inflationary trajectory which drives the inflaton toward the minimum. For $|S| \leq S_c=M$, the $l^c$, $\bar l^{c}$ components become tachyonic and the system evolves towards the `good' supersymmetric minimum at $h=0$, $l^c=\bar l^c=M$ (for $\kappa > \lambda$, $h$ is destabilized first and the system would have evolved towards the `bad' minimum at $h \neq 0$, $l^c= \bar l^c = 0$). For all values of the parameters considered here, inflation continues at least till $|S|$ approaches the instability at $|S|=S_c$ as one deduces from the slow roll conditions \cite{dss,lazarides}. The cosmic microwave quadrupole anisotropy can be calculated \cite {dss} by standard methods and turns out to be \begin{equation} \left(\frac{\delta T}{T}\right)_{Q} \approx \frac{32 \pi^{5/2}} {3\sqrt{5}}\left(\frac{M}{M_P}\right)^{3} \kappa^{-1}x_{Q}^{-1}\Lambda (x_{Q})^{-1}~, \label{anisotropy} \end{equation} \begin{eqnarray*} \Lambda(x)=\left(\frac{\lambda} {\kappa}\right)^{3}\left[\left(\frac{\lambda} {\kappa}x^2-1\right)\ln \left(1-\frac{\kappa} {\lambda}x^{-2}\right) +\left(\frac{\lambda}{\kappa}x^2+1\right) \ln \left(1+\frac{\kappa} {\lambda}x^{-2}\right)\right] \end{eqnarray*} \begin{equation} +(x^2-1)\ln (1-x^{-2})+(x^2+1)\ln (1+x^{-2})~, \label{temp} \end{equation} with $x=|S|/S_c$ and $S_Q$ being the value of $|S|$ when the present horizon scale crossed outside the inflationary horizon. (Notice that here we had to replace the contribution to the effective potential in Eq.(\ref{eq:veffexact}) from the $\phi$, $\bar{\phi}$ supermultiplets of the model in Sec.\ref{subsec:susy} by the contribution from the $l^c$, $\bar l^{c}$ and $h$ supermultiplets.) The number of e-foldings experienced by the universe between the time the quadrupole scale exited the horizon and the end of inflation is \begin{equation} N_Q \approx 32 \pi^3 \left(\frac{M}{M_P}\right)^2 \kappa^{-2}\int_{1}^{x_{Q}^{2}}\frac{dx^2}{x^2} \Lambda(x)^{-1}~. \label{efoldings} \end{equation} The spectral index of density perturbations turns out to be very close to unity. \subsection{`Reheating' and Leptogenesis} \label{subsec:reheatlepto} \par After reaching the instability at $|S|=S_c$, the system continues \cite{bl} inflating for another e-folding or so reducing its energy density by a factor of about $2-3$~. It then rapidly settles into a regular oscillatory phase about the vacuum. Parametric resonance is safely ignored in this case \cite{bl}. The inflaton (oscillating system) consists of the two complex scalar fields $S$ and $\theta=(\delta \phi + \delta\bar{\phi})/\sqrt{2}$, where $\delta \phi = \phi - M$, $\delta \bar{\phi} = \bar{\phi} - M$, with mass $m_{infl}=\sqrt{2}\kappa M$ ($\phi$, $\bar{\phi}$ are the neutral components of $l^c$, $\bar l^{c}$). \par The scalar fields $S$ and $\theta$ predominantly decay into electroweak higgsinos and higgses respectively with a common decay width $\Gamma_{h}=(1/16\pi)\lambda^{2} m_{infl}$, as one can easily deduce from the couplings in Eqs.(\ref{W}) and (\ref{lambda}). Note, however, that $\theta$ can also decay to right handed neutrinos $\nu^c$ through the nonrenormalizable superpotential term \begin{equation} \frac{M_{\nu^c}}{2M^{2}}\bar{\phi} \bar{\phi} \nu^c \nu^c~, \label{majorana} \end{equation} allowed by the gauge and R- symmetries of the model \cite{lss,dls}. Here, $M_{\nu^c}$ denotes the Majorana mass of the relevant $\nu^c$. The scalar $\theta$ decays preferably into the heaviest $\nu^c$ with $M_{\nu^{c}} \leq m_{infl}/2$. The decay rate is given by \begin{equation} \Gamma_{\nu^c} \approx \frac{1}{16\pi}~\kappa^2 m_{infl}~ \alpha^2 (1-\alpha^2)^{1/2}~, \label{decayneu} \end{equation} where $0\leq \alpha=2M_{\nu^c}/m_{infl} \leq 1$. The subsequent decay of these $\nu^{c}$ 's produces a primordial lepton number \cite{lepto} which is then partially converted to the observed BAU through electroweak sphaleron effects. \par The energy densities $\rho_{S}$, $\rho_{\theta}$, and $\rho_{r}$ of the oscillating fields $S$, $\theta$, and the `new' radiation produced by their decay to higgsinos, higgses and $\nu^c$ 's are controlled by the equations: \begin{equation} \dot{\rho}_{S}=-(3H+\Gamma_{h})\rho_{S}~, ~\rho_{\theta}(t)=\rho_{S}(t) e^{-\Gamma_{\nu^c}(t-t_0)}~, \label{infdensity} \end{equation} \begin{equation} \dot{\rho}_{r}=-4H\rho_{r}+\Gamma_{h}\rho_{S}+ (\Gamma_{h}+\Gamma_{\nu^{c}})\rho_{\theta}~, \label{raddensity} \end{equation} where \begin{equation} H=\frac{\sqrt{8\pi}}{\sqrt{3}M_P}~(\rho_{S}+ \rho_{\theta}+\rho_{r})^{1/2} \label{hubble} \end{equation} is the Hubble parameter and overdots denote derivatives with respect to cosmic time $t$. The cosmic time at the onset of oscillations is taken $t_0\approx 0$. The initial values of the various energy densities are taken to be $\rho_{S}(t_0)=\rho_{\theta}(t_0)\approx \kappa^{2}M^{4}/6$, $\rho_{r}(t_{0})=0$. The `reheat' temperature $T_{r}$ is calculated from the equation \begin{equation} \rho_{S}+\rho_{\theta}=\rho_{r}= \frac{\pi^2}{30}~g_{*}T_{r}^{4}~, \label{reheat} \end{equation} where the effective number of massless degrees of freedom is $g_{*}$=228.75 for MSSM. \par The lepton number density $n_{L}$ produced by the $\nu^{c}$ 's satisfies the evolution equation: \begin{equation} \dot{n}_{L}=-3Hn_{L}+2\epsilon\Gamma_{\nu^c} n_{\theta}~, \label{lepton} \end{equation} where $\epsilon$ is the lepton number produced per decaying right handed neutrino and the factor of 2 in the second term of the rhs comes from the fact that we get two $\nu^c$ 's for each decaying scalar $\theta$ particle. The `asymptotic' ($t\rightarrow 0$) lepton asymmetry turns out to be \begin{equation} \frac{n_{L}(t)}{s(t)}\sim 3 \left(\frac{15}{8}\right)^{1/4}\pi^{-1/2} g_{*}^{-1/4}m_{infl}^{-1}~ \frac{\epsilon\Gamma_{\nu^c}} {\Gamma_{h}+\Gamma_{\nu^c}}~\rho_{r}^{-3/4} \rho_{S}e^{\Gamma_{h}t}~. \label{leptonasymmetry} \end{equation} \par Assuming hierarchical light neutrino masses, we take $m_{\nu_{\mu}}\approx 2.6\times 10^{-3}~\rm{eV}$ which is the central value of the $\mu$-neutrino mass coming from the small angle MSW resolution of the solar neutrino problem \cite{smirnov}. The $\tau$-neutrino mass is taken $m_{\nu _{\tau}}\approx 7\times 10^{-2} ~\rm{eV}$, the central value from SuperKamiokande \cite{superk}. Recent analysis \cite{giunti} of the results of the CHOOZ experiment shows that the oscillations of solar and atmospheric neutrinos decouple. We thus concentrate on the two heaviest families ignoring the first one. Under these circumstances, the lepton number generated per decaying $\nu^c$ is \cite{lazarides,neu} \begin{equation} \epsilon=\frac{1}{8\pi}~g \left(\frac{M_3}{M_2}\right) ~\frac{{\rm c}^{2}{\rm s}^{2}\ \sin 2\delta \ (m_{3}^{D}\,^{2}-m_{2}^{D}\,^{2})^{2}} {|\langle h^{(1)}\rangle|^{2}~(m_{3}^{D}\,^{2}\ {\rm s}^{2}\ + \ m_{2}^{D}\,^{2}{\rm \ c^{2}})}~, \label{epsilon} \end{equation} where $g(r)=r\ln (1 + r^{-2})$~, $|\langle h^{(1)}\rangle|\approx 174~\rm{GeV}$, ${\rm c}=\cos \theta ,\ {\rm s}=\sin \theta $, and $\theta$ ($0\leq \theta \leq \pi /2$) and $\delta$ ($-\pi/2\leq \delta <\pi/2 $) are the rotation angle and phase which diagonalize the Majorana mass matrix of $\nu^{c}$ 's with eigenvalues $M_2$, $M_3$ ($\geq 0$). The `Dirac' mass matrix of the neutrinos is considered diagonal with eigenvalues $m_{2}^{D}$, $m_{3}^{D}$ ($\geq 0$). \par For the range of parameters considered here, the scalar $\theta$ decays into the second heaviest right handed neutrino with mass $M_{2}$ ($<M_{3}$) and, thus, $M_{\nu^{c}}$ in Eqs.(\ref{majorana}) and (\ref{decayneu}) should be identified with $M_{2}$. Moreover, $M_{3}$ turns out to be bigger than $m_{infl}/2$ as it should. We will denote the two positive eigenvalues of the light neutrino mass matrix by $m_{2}$ (=$m_{\nu _{\mu }}$), $m_{3}$ (=$m_{\nu _{\tau }}$) with $m_{2}\leq m_{3}$. All the quantities here (masses, rotation angles and phases) are `asymptotic' (defined at the grand unification scale $M_{GUT}$). \par The determinant and the trace invariance of the light neutrino mass matrix imply\cite{neu} two constraints on the (asymptotic) parameters which take the form: \begin{equation} m_{2}m_{3}\ =\ \frac{\left( m_{2}^{D}m_{3}^{D} \right) ^{2}}{M_{2}\ M_{3}}~, \label{determinant} \end{equation} \begin{eqnarray*} m_{2}\,^{2}+m_{3}\,^{2}\ =\frac{\left( m_{2}^{D}\,\, ^{2}{\rm c}^{2}+m_{3}^{D}\,^{2}{\rm s}^{2}\right) ^{2}}{M_{2}\,^{2}}+ \end{eqnarray*} \begin{equation} \ \frac{\left( m_{3}^{D}\,^{2}{\rm c}^{2}+m_{2}^{D}\, ^{2}{\rm s}^{2}\right)^{2}}{M_{3}\,^{2}}+\ \frac{2(m_{3}^{D}\,^{2}-m_{2}^{D}\,^{2})^{2} {\rm c}^{2}{\rm s}^{2}\,{\cos 2\delta }} {M_{2}\,M_{3}}~\cdot \label{trace} \end{equation} \par The $\mu-\tau$ mixing angle $\theta _{23}$ (=$\theta _{\mu\tau}$) lies \cite{neu} in the range \begin{equation} |\,\varphi -\theta ^{D}|\leq \theta _{23}\leq \varphi +\theta ^{D},\ {\rm {for}\ \varphi + \theta }^{D}\leq \ \pi /2~,~~~~~ \label{mixing} \end{equation} where $\varphi$ ($0\leq \varphi \leq \pi /2$) is the rotation angle which diagonalizes the light neutrino mass matrix, and $\theta ^{D}$ ($0\leq \theta ^{D} \leq \pi /2$) is the `Dirac' (unphysical) mixing angle in the $2-3$ leptonic sector defined in the absence of the Majorana masses of the $\nu^{c}$ 's. \par Assuming approximate $SU(4)_{c}$ symmetry, we get the asymptotic (at $M_{GUT}$) relations: \begin{equation} m_{2}^{D}\approx m_{c}\ , \ m_{3}^{D}\approx \ m_{t}\ , \ \sin\theta ^{D}\approx |V_{cb}|~. \label{asympt} \end{equation} Renormalization effects, for MSSM spectrum and $\tan \beta \approx m_{t}/m_{b}$, are incorporated \cite{neu} by substituting in the above formulas the values: $m_{2}^{D}\approx 0.23~{\rm GeV}$, $\ m_{3}^{D}\approx 116$ GeV and $\sin \theta ^{D} \approx 0.03$. Also, $\tan^{2} 2 \theta _{23}$ increases by about 40\% from $M_{GUT}$ to $M_{Z}$. \par We take a specific MSSM framework \cite{als} where the three Yukawa couplings of the third generation unify `asymptotically' and, thus, $\tan \beta \approx m_{t}/m_{b}$. We choose the universal scalar mass (gravitino mass) $m_{3/2} \approx 290~{\rm{GeV}}$ and the universal gaugino mass $M_{1/2} \approx 470~{\rm{GeV}}$. These values correspond \cite{asw} to $m_{t}(m_{t})\approx 166~ {\rm{GeV}}$ and $m_{A}$ (the tree level CP-odd scalar higgs mass) =$M_{Z}$. The ratio $\lambda/\kappa$ is evaluated \cite{carena} from \begin{equation} \frac{\lambda}{\kappa}=\frac{|\mu|}{m_{3/2}}\approx \frac{M_{1/2}}{m_{3/2}}\left(1-\frac{Y_{t}} {Y_{f}}\right)^{-3/7}\approx 3.95~, \label{renorm} \end{equation} where $Y_{t}=h_{t}^2\approx 0.91$ is the square of the top-quark Yukawa coupling and $Y_{f}\approx 1.04$ is the weak scale value of $Y_{t}$ corresponding to `infinite' value at $M_{GUT}$. \par Eqs.(\ref{anisotropy})-(\ref{efoldings}) can now be solved, for $(\delta T/T)_{Q} \approx 6.6\times 10^{-6}$ from COBE, $N_Q \approx 50$ and any value of $x_{Q}>1$. Eliminating $x_{Q}$, we obtain $M$ as a function of $\kappa$ depicted in Fig.\ref{reheating}. The evolution Eqs.(\ref{infdensity})-(\ref{hubble}) are solved for each value of $\kappa$. The parameter $\alpha^{2}$ in Eq.(\ref{decayneu}) is taken equal to 2/3. This choice maximizes the decay width of the inflaton to $\nu^{c}$ 's and, thus, the subsequently produced lepton asymmetry. The `reheat' temperature is then calculated from Eq.(\ref{reheat}) for each value of $\kappa$. The result is again depicted in Fig.\ref{reheating}. \par The mass of the second heaviest $\nu^c$, into which the scalar $\theta$ decays partially, is given by $M_{2}= M_{\nu^{c}}=\alpha m_{infl}/2$ and $M_{3}$ is found from the `determinant' condition in Eq.(\ref{determinant}). The `trace' condition in Eq.(\ref{trace}) is then solved for $\delta(\theta )$ which is subsequently substituted in Eq.(\ref{epsilon}) for $\epsilon$. The leptonic asymmetry as a function of the angle $\theta$ can be found from Eq.(\ref{leptonasymmetry}). For each value of $\kappa$, there are two values of $\theta$ satisfying the low deuterium abundance constraint $\Omega _{B}h^{2} \approx 0.025$. (These values of $\theta$ turn out to be quite insensitive to the exact value of $n_{B}/s$.) The corresponding $\varphi$ 's are then found and the allowed region of the mixing angle $\theta _{\mu\tau}$ in Eq.(\ref{mixing}) is determined for each $\kappa$. Taking into account renomalization effects and superimposing all the permitted regions, we obtain the allowed range of $\sin^{2} 2 \theta _{\mu\tau}$ as a function of $\kappa$, shown in Fig.\ref{angle}. We observe that $\sin^{2} 2 \theta _{\mu\tau} \stackrel{_{>}}{_{\sim }} 0.8$ (from SuperKamiokande \cite{superk}) corresponds to $1.2\times 10^{-6} \stackrel{_{<}}{_{\sim }}\kappa \stackrel{_{<}}{_{\sim }}3.4\times 10^{-6}$ which is rather small. (Fortunately, supersymmetry protects it from radiative corrections.) \par The corresponding values of $M$ and $T_{r}$ can be read from Fig.\ref{reheating}. One finds that $1.4\times 10^{15} ~{\rm{GeV}}\stackrel{_{<}}{_{\sim }} M \stackrel{_{<}}{_{\sim }}2\times 10^{15}~{\rm{GeV}}$ and $1.8\times 10^{7}~{\rm{GeV}}\stackrel{_{<}}{_{\sim }} T_{r}\stackrel{_{<}}{_{\sim }}8.7\times 10^{7} ~{\rm{GeV}}$. We observe that $M$ turns out to be somewhat smaller than the MSSM unification scale $M_{GUT}$. (It is anticipated that $G_{LR}$ is embedded in a grand unified theory.) The `reheat' temperature, however, satisfies the gravitino constraint ($T_{r}\stackrel{_{<}}{_{\sim}} 10^{9}~{\rm{GeV}}$). Note that, for the values of the parameters chosen here, the lightest supersymmetric particle (LSP) is \cite{asw} an almost pure bino with mass $m_{LSP} \approx 0.43M_{1/2}\approx 200~{\rm{GeV}}$ and can, in principle, provide the cold dark matter of the universe. On the contrary, there is no hot dark matter candidate, in the simplest scheme. \par In conclusion, we have shown that, in a supersymmetric model based on a left-right symmetric gauge group and leading `naturally' to hybrid inflation, the $\mu$ problem can be easily solved . The observed BAU is produced via a primordial leptogenesis. For masses of $\nu_{\mu}$, $\nu_{\tau}$ from the small angle MSW resolution of the solar neutrino puzzle and SuperKamiokande, maximal $\nu_{\mu}-\nu_{\tau}$ mixing can be achieved. The required values of the coupling constant $\kappa$ are, however, quite small ($\sim 10^{-6}$). \vspace{0.5cm} This work is supported by E.U. under TMR contract No. ERBFMRX--CT96--0090. \def\ijmp#1#2#3{{ Int. Jour. Mod. Phys. } {\bf #1~}(19#2)~#3} \def\pl#1#2#3{{ Phys. Lett. }{\bf B#1~}(19#2)~#3} \def\zp#1#2#3{{ Z. Phys. }{\bf C#1~}(19#2)~#3} \def\prl#1#2#3{{ Phys. Rev. Lett. }{\bf #1~}(19#2)~#3} \def\rmp#1#2#3{{ Rev. Mod. Phys. }{\bf #1~}(19#2)~#3} \def\prep#1#2#3{{ Phys. Rep. }{\bf #1~}(19#2)~#3} \def\pr#1#2#3{{ Phys. Rev. }{\bf D#1~}(19#2)~#3} \def\np#1#2#3{{ Nucl. Phys. }{\bf B#1~}(19#2)~#3} \def\mpl#1#2#3{{ Mod. Phys. Lett. }{\bf #1~}(19#2)~#3} \def\arnps#1#2#3{{ Annu. Rev. Nucl. Part. Sci. }{\bf #1~}(19#2)~#3} \def\sjnp#1#2#3{{ Sov. J. Nucl. Phys. }{\bf #1~} (19#2)~#3} \def\jetp#1#2#3{{ JETP Lett. }{\bf #1~}(19#2)~#3} \def\app#1#2#3{{ Acta Phys. Polon. }{\bf #1~}(19#2)~#3} \def\rnc#1#2#3{{ Riv. Nuovo Cim. }{\bf #1~}(19#2)~#3} \def\ap#1#2#3{{ Ann. Phys. }{\bf #1~}(19#2)~#3} \def\ptp#1#2#3{{ Prog. Theor. Phys. }{\bf #1~} (19#2)~#3} \def\plb#1#2#3{{ Phys. Lett. }{\bf#1B~}(19#2)~#3} \def\ibid#1#2#3{{ ibid. }{\bf#1~}(19#2)~#3} \def\apjl#1#2#3{{ Astrophys. J. Lett. }{\bf#1~}(19#2)~#3}
\section{INTRODUCTION} \label{sec_int} In this paper, we study the scattering of surface plasmon polaritons (SPP) by surface defects. SPP are $p$-polarized electromagnetic (EM) waves bound to a dielectric-metal interface and caused by the surface oscillations of the electron plasma of the metal.\cite{sp} They propagate along the metal interface a distance of the order of the SPP mean free path (ranging from microns in the visible to millimeters in the infrared, of course depending also on the metal being considered), undergoing scattering processes due to surface roughness. This constitutes a classical problem of fundamental interest not only in the case of individual defects (cf. Ref. \onlinecite{pmbg94} and references therein), but also for periodically or randomly (or both) distributed defects.\cite{pg96,prb96,kits96,prb97} Furthermore, it is obviously crucial in any light scattering problem involving rough metal surfaces where roughness-induced excitation of SPP occurs. This has been explicitly shown in connection with either single defects \cite{psg94,chema95,alb} or random corrugation,\cite{psg95,wo,mmm95,owm98} the latter configuration being relevant to the phenomenon of (SPP mediated) enhanced backscattering of light. In addition to that, light-SPP coupling plays a central role in other phenomena such as anomalous transmission through metal slabs with hole arrays,\cite{ebbe98,schro98} surface-enhanced Raman scattering, \cite{sers,vp96,jcp98} or biosensing.\cite{bio} In recent years, the advent of near-field optical microscopy \cite{nf} has opened up the possibility to study experimentally SPP in a direct manner. Among the various configurations developed, photon scanning tunneling microscopy \cite{pstm} (PSTM), basically exploiting SPP excitation in the attenuated total reflection arrangement, has made it possible to probe the SPP structure,\cite{marti,adam} localized SPP on randomly rough surfaces,\cite{bozh95} and SPP resonances in fractal colloid clusters \cite{tsai94} and single particles.\cite{nie97,klar98} Moreover, PSTM images have been obtained by surface-enhanced Raman scattering probing single molecules adsorbed on single nanoparticles. \cite{nie97} PSTM in combination with direct-write lithography has made it possible to create sub-micron defects on metal surfaces.\cite{smol95} Particularly relevant to the present work are the recent experimental studies on SPP scattering by surface defects.\cite{smol,bozh,smol99} These studies have shown evidence of drastically distinct scattering properties depending on the defect size. Specifically, surface defects favoring SPP reflection and light coupling, called SPP mirrors and flashlights,\cite{smol} respectively, have been described, as well as SPP microlenses and microcavities;\cite{bozh} SPP Bloch waves have also been imaged in periodic arrays of surface defects.\cite{smol99} Interestingly, the possibilities of artificially creating micro-optical components for SPP have also been noted in these studies. Much has to be done, however, from the theoretical standpoint. Quite recently, calculations for circularly symmetric defects have successfully accounted for the peculiar azimuthal dependence of the radiated pattern;\cite{snm97} in addition, such calculations have been used to retrieve the surface profile. \cite{pasc98} In the case of one-dimensional surface defects, preliminary calculations have focused on the optimization of the defect size to obtain SPP mirrors and so-called light-emitters.\cite{apl98} In this regard, it is our purpose to address in detail the SPP scattering by one-dimensional surface defects, including near-field and far-field calculations (along with energy balance) and their dependence on defect size parameters. Thus we expect not only to shed light on the experimental works mentioned above, but also to find and predict related effects. The physical system we consider here is a planar one-dimensional metal surface with a one-dimensional defect. The surface corrugation is modeled by using a local impedance boundary condition (IBC) on a flat surface. The connection between surface impedance and real surface corrugation has been recently demonstrated,\cite{aamibc} and its validity to give accurate quantitative results has been shown in numerical calculations of grating-induced SPP-photon coupling. \cite{prb96} A scattering-theoretic formulation of the interaction of an SPP with the surface roughness is developed by imposing the IBC on the amplitude of the magnetic field in the vacuum region in the form of a Rayleigh expansion. Upon solving the resulting integral equation for the scattering amplitude, the magnetic field at any point in the vacuum half-space can be calculated. We will focus on the far field angular distribution and the surface field amplitudes to determine, respectively, the total radiated energy $S$, and the SPP reflection $R_{SP}$ and transmission $T_{SP}$ coefficients. By numerical simulation calculations, these quantities are computed. The paper is organized as follows. The theoretical formulation is derived in Sec.~\ref{sec_the}, and some details pertaining to the numerical procedure are given in the Appendix. In Sec.~\ref{sec_res}, we show the results obtained for a single Gaussian defect and the influence of defect width and height. Finally, Section~\ref{sec_con} summarizes the conclusions drawn from this research. \section{THEORY} \label{sec_the} \subsection{Scattering Equations} \label{sec_the_se} We study the scattering of a p-polarized SPP of frequency $\omega$ propagating along a flat vacuum-metal interface ($x_3=0$) by a one-dimensional obstacle (constant along the $x_2$-axis, see Fig.~\ref{fig_plasmon}). Under these circumstances, the three-dimensional electromagnetic problem can be cast into a two-dimensional scalar problem in such a way that the single, nonzero component of the magnetic field amplitude $H_2(x_1,x_3)$ is the solution of the corresponding two-dimensional Helmholtz equation in the upper (vacuum) and lower (metal) half spaces. The magnetic field in vacuum is assumed to be the sum of an incoming SPP and a scattered field as follows \begin{eqnarray} \lefteqn{H_2^>(x_1,x_3) =\exp[ik(\omega)x_1-\beta_o(\omega)x_3]} && \nonumber \\ && \phantom{+++}+\int_{-\infty}^{\infty}\frac{dq}{2\pi} \, R(q,\omega)\exp[iqx_1+i\alpha_o(q,\omega)x_3] , \label{eq_h} \end{eqnarray} where \begin{eqnarray} k(\omega) & \equiv k^R(\omega)+ik^I(\omega)={\displaystyle \frac{\omega}{c}\left(1-\frac{1}{\epsilon(\omega)} \right)^{1/2},} \label{eq_kw} \end{eqnarray} \begin{eqnarray} \beta_0(\omega) & = & \left(k(\omega)^2-\frac{\omega^2}{c^2} \right)^{1/2}=\frac{\omega}{c}[-\epsilon(\omega )]^{-1/2}, \label{eq_bw} \end{eqnarray} and \begin{mathletters}\begin{eqnarray} \alpha_o(q,\omega) & = & \left(\frac{\omega^2}{c^2}-q^2 \right)^{1/2}\;\;\;\; |q|\leq\frac{\omega}{c} \\ & = & i\left(q^2-\frac{\omega^2}{c^2}\right)^{1/2} \;\;\; |q|>\frac{\omega}{c}. \end{eqnarray}\end{mathletters} Note that the expressions for the SPP wavevector components $k(\omega)$ and $\beta_0(\omega)$ in vacuum apply in the limit $|\epsilon(\omega)|\gg 1$. This stems from the fact that the continuity conditions across the interface are mapped onto a local IBC on the planar surface $x_3=0$ in the form \begin{eqnarray} \lefteqn{\left.\frac{\partial}{\partial x_3} H_2^>(x_1,x_3)\right|_{x_3=0}} && \nonumber \\ && =-\frac{\omega}{c}\,\frac{1+s(x_1)}{[-\epsilon (\omega)]^{ 1/2}} \;H_2^>(x_1,x_3)|_{x_3=0} , \label{eq_ibc} \end{eqnarray} where the superscript $>$ indicates the vacuum region, $-(\omega/c)[-\epsilon (\omega )]^{-1/2}s(x_1)$ is the contribution to the surface impedance associated with the obstacle, and $\epsilon(\omega)$ is the isotropic, frequency-dependent dielectric function of the metal. The IBC has been widely used in the past to model the vacuum-metal interface qualitatively, especially in the infrared region of the optical spectrum. Furthermore, it has been recently proven to be quantitatively accurate in calculations of grating-induced photon-SPP coupling \cite{prb96} by using the connection between surface impedance and real corrugation demonstrated in Ref. \onlinecite{aamibc}. In order to calculate the scattering amplitude $R(q,\omega)$, we substitute Eq. (\ref{eq_h}) into Eq. (\ref{eq_ibc}), and obtain the integral equation $R(q,\omega )$ satisfies, \begin{eqnarray} R(q,\omega)=&& G_0(q,\omega)V\bbox(q|k(\omega)\bbox) \nonumber\\ && +G_0(q,\omega)\int_{-\infty}^{\infty}\! \frac{dp}{2\pi} V(q|p)R(p,\omega), \label{eq_ierq} \end{eqnarray} where \begin{eqnarray} G_0(q,\omega) & \equiv & \frac{i\epsilon(\omega)}{\epsilon(\omega)\alpha_0(q,\omega) +i(\omega/c)[-\epsilon(\omega)]^{1/2}} \label{eq_gq} \end{eqnarray} is the Green's function of the SPP on the unperturbed surface [$s(x_1)=0$]. We have also introduced the scattering potential \begin{eqnarray} V(q|p) & \equiv & \beta_0(\omega)\hat{s}(q-p), \label{eq_vq} \end{eqnarray} with \begin{eqnarray} \hat{s}(Q)&=&\int_{-\infty}^{\infty}\! dx_1\; e^{-iQx_1}s(x_1), \label{eq_sq} \end{eqnarray} to simplify the notation. Equation (\ref{eq_ierq}) can be rewritten in a more convenient manner by substituting \begin{eqnarray} R(q,\omega) &=& G_0(q,\omega)T(q,\omega), \label{eq_rtq} \end{eqnarray} into it, so that \begin{eqnarray} T(q,\omega)=&& V\bbox(q|k(\omega)\bbox) \nonumber \\ && +\int_{-\infty}^{\infty}\! \frac{dp}{2\pi}V(q|p)G_0(p,\omega)T(p,\omega). \label{eq_ietq} \end{eqnarray} Equation (\ref{eq_ietq}), along with Eqs. (\ref{eq_h}) and (\ref{eq_rtq}), is the basis of our theoretical formulation. In solving Eq. (\ref{eq_ietq}), it is very important how we deal with the poles appearing in the Green's function (\ref{eq_gq}). First, we rewrite the latter in the form \begin{eqnarray} G_0(q,\omega) & = & C(q,\omega)\left( \frac{1}{q-k(\omega)}-\frac{1}{q+k(\omega)}\right), \end{eqnarray} with \begin{eqnarray} C(q,\omega) & \equiv & \frac{\epsilon(\omega)\alpha_0(q,\omega)- i(\omega/c)[-\epsilon(\omega)]^{1/2}}{2i\epsilon(\omega) k(\omega)}. \end{eqnarray} We now assume that the metal dielectric function is given by Drude's expression \begin{eqnarray} \epsilon(\omega) &=& 1-\frac{\omega_p^2}{\omega^2} , \end{eqnarray} where $\omega_p$ is the plasma frequency, in the absence of absorption losses. Therefore, in light of Eq. (\ref{eq_kw}), we have to take the limit $k^I(\omega)\rightarrow 0$ in Eq. (\ref{eq_gq}), to obtain \begin{eqnarray} G_0(q,\omega) &&= C(q,\omega)\left( \left.\frac{1}{q-k^R(\omega)}\right|_P-\left. \frac{1}{q+k^R(\omega)}\right|_P\right. \nonumber \\ && \left.\phantom{\frac{1}{2}}+\pi i[\delta\bbox(q-k^R(\omega)\bbox) +\delta\bbox(q+k^R(\omega)\bbox)]\right) . \label{eq_gqri} \end{eqnarray} The first two terms on the right-hand side of Eq. (\ref{eq_gqri}) have meanings in the Cauchy's principal value sense, whereas the last two terms are delta functions. Once Eq. (\ref{eq_ietq}) is solved for $T(q,\omega)$ [we will see below how to do so numerically with the help of Eq. (\ref{eq_gqri})], we proceed to calculate the electric and magnetic near fields, the SPP-photon coupling, and the SPP reflection and transmission coefficients in the following manner. \subsection{Near Field} \label{sec_the_nf} The magnetic field at any point in the vacuum half-space can be straightforwardly calculated from Eq. (\ref{eq_h}), upon recalling Eq. (\ref{eq_rtq}), which relates $T(q,\omega)$ with the scattering amplitude $R(q,\omega)$. Then the electric field components in vacuum are easily written also as functions of $R(q,\omega)$ by means of a Maxwell curl equation as follows: \begin{mathletters}\label{eq_e}\begin{eqnarray} E_1^>(x_1,x_3) = &&\frac{c}{\omega}i\beta_0(\omega) \exp[ik(\omega)x_1-\beta_0(\omega)x_3] \nonumber \\ && + \frac{c}{\omega}\int_{-\infty}^{\infty}\frac{dq}{2\pi} \,\alpha_0(q,\omega) R(q,\omega) \nonumber \\ && \times \exp[iqx_1+i\alpha_0(q,\omega)x_3] \\ E_2^>(x_1,x_3) = && 0 \\ E_3^>(x_1,x_3) = &&-\frac{c}{\omega}k(\omega) \exp[ik(\omega)x_1-\beta_0(\omega)x_3] \nonumber \\ && - \frac{c}{\omega}\int_{-\infty}^{\infty} \frac{dq}{2\pi}\, qR(q,\omega) \nonumber \\ && \times \exp[iqx_1+i\alpha_0(q,\omega)x_3]. \end{eqnarray}\end{mathletters} The time-averaged Poynting vector thus reads: \begin{eqnarray} <{\bf S}> = \frac{c}{8\pi} \Re \left({\bf E}\times{\bf H}^*\right) = \frac{c}{8\pi} \Re \left(-E_3 H_2^*,0,E_1 H_2^*\right) , \label{eq_pv} \end{eqnarray} where $\Re$ denotes the real part and the asterisk the complex conjugate. \subsection{Radiated energy} \label{sec_the_s} The total power carried away from the surface in the form of volume electromagnetic waves propagating in the vacuum region above it, per unit length of the system along the $x_2$-axis is \begin{eqnarray} P_{sc}= & & \int_{-\infty}^{\infty}\! dx_1\; <S^{(sc)}_3> \nonumber \\ = &&\frac{c^2}{8\pi\omega} \int_{-\omega/c}^{\omega/c} \frac{dq}{2\pi}\alpha_0(q,\omega)\mid\!R(q,\omega)\!\mid^2 . \label{eq_psc} \end{eqnarray} Note that only the scattered field contribution to the $x_3$-component of the time-averaged Poynting vector is used. Equation (\ref{eq_psc}) must be normalized by the power carried by the incident SPP per unit length along the $x_2$-axis \begin{eqnarray} P_{inc}= & & \int_0^{\infty}\! dx_3\; <S^{(inc)}_1> =\frac{c^2k(\omega)}{16\pi\omega\beta_0(\omega)} , \label{eq_pin} \end{eqnarray} where $<S^{(inc)}_1>$ is the $x_1$-component of the time-averaged Poynting vector of the incident SPP. Then the total, normalized scattered power $S$ is given by \begin{eqnarray} S &= {\displaystyle\frac{P_{sc}}{P_{inc}}} & =\int_{-\pi/2}^{\pi/2} \! d\theta_s \frac{\partial R}{\partial\theta_s}, \label{eq_s} \end{eqnarray} where \begin{eqnarray} \frac{\partial R}{\partial\theta_s} =& &{\displaystyle\frac{1}{2 \pi}\frac{\beta_0(\omega)}{2k(\omega)}\;\alpha_0^2\left( q=\frac{\omega}{c}\sin\theta_s\right)} \nonumber\\ && \times\left| R\left(q=\frac{\omega}{c}\sin\theta_s\right)\right|^2 , \label{eq_drc} \end{eqnarray} is the differential reflection coefficient (DRC), namely, the fraction of the energy of the incident SPP that is scattered into an angular region of width $d\theta_s$ about the scattering direction $\theta_s$ where the scattering angle $\theta_s$ is measured clockwise with respect to the $x_3$-axis (see Fig.~\ref{fig_plasmon}). \subsection{Reflection and transmission coefficients} \label{sec_the_rt} In order to evaluate the amplitude of the reflected and transmitted SPP, we study the behavior of $H_2^>(x_1,x_3)$, Eq. (\ref{eq_h}), with the help of Eqs. (10) and (15), on the surface $x_3=0$. At this point, great care has to be taken when calculating the contribution to the scattered field in Eq. (2) from the Cauchy principal value integrals arising from the first two terms on the right-hand side of Eq. (15). We assume that the obstacle has a finite extent and is centered about $x_1=0$. If we focus on the regions $x_1\ll 0$ and $x_1\gg 0$ far from the obstacle, it can be shown, by working out the contributions from those integrals in the complex $q$-space with the help of Cauchy's theorem,\cite{cauchy} that the magnetic field is given by: \begin{mathletters}\label{eq_hxinf}\begin{eqnarray} H_2^>(x_1,x_3=0) =& &\exp[ik^R(\omega)x_1] \nonumber \\ & & +r(\omega)\exp[-ik^R(\omega)x_1], x_1\ll 0 \\ = && t(\omega)\exp[ik^R(\omega)x_1], \phantom{+-} x_1\gg 0, \end{eqnarray}\end{mathletters} where the amplitudes of the reflected and transmitted SPP, $r(\omega)$ and $t(\omega)$, respectively, are \begin{mathletters}\label{eq_spprt}\begin{eqnarray} r(\omega) &=& iT\bbox(-k^R(\omega),\omega\bbox)\; C\bbox(-k^R(\omega), \omega\bbox) \\ t(\omega) &=& 1+iT\bbox(k^R(\omega),\omega\bbox)\; C\bbox(k^R(\omega), \omega\bbox) . \end{eqnarray}\end{mathletters} Equations (\ref{eq_hxinf}) manifest the fact that, away from the obstacle, only the incident and reflected SPP (on the left-hand side, see Fig. 1) and the transmitted SPP (on the right-hand side) propagate along the interface. The corresponding reflection and transmission coefficients are \begin{mathletters}\label{eq_spprt2}\begin{eqnarray} R(\omega) &=& \mid\! r(\omega)\!\mid^2 \\ T(\omega) &=& \mid\! t(\omega)\!\mid^2 . \end{eqnarray}\end{mathletters} \subsection{Numerical Calculations} \label{sec_the_nc} The integral equation (\ref{eq_ietq}) is numerically solved by converting it into a matrix equation through a quadrature scheme. The details are given in the Appendix. It should be pointed out that the discretization $q$-mesh is chosen in such a way that $q=\pm k^R(\omega)$ are always points on the mesh, as required by Eq. (\ref{eq_spprt}). In addition, the discretization is not regular: the density of $q$-points around the poles at $q=\pm k^R(\omega)$ is considerably larger ($\Delta q\approx 10^{-4}\omega/c$) than it is either in the radiative region $|q|\leq\omega/c$ or in the non-radiative region away from the poles ($\Delta q\approx 10^{-2}\omega/c$). The number $N$ of $q$-points needed in the numerical procedure depends not only on the accuracy required to sample the pole regions, but also on the explicit form of the obstacle, which enters in the calculation through its Fourier-transform in Eq. (\ref{eq_sq}). Throughout this work, typically $N=2600$, except for the larger defects for which up to $N=4000$ points are employed. The convergence of the numerical results with increasing $N$ has been checked in the most unfavorable cases. \section{RESULTS AND DISCUSSION} \label{sec_res} Note that up to now no restrictions have been imposed on the shape of the obstacle apart from its having a finite extent along the $x_1$-axis. Its surface impedance function $s(x_1)$ is connected to the actual surface profile defined by $x_3 = f(x_1)$ through\cite{aamibc} \begin{eqnarray} s(x) & = & -\frac{1-\epsilon(\omega)}{d(\omega)\epsilon(\omega)} [1-d^2(\omega)D^2]^{1/2} f(x_1) + O(f^2), \label{eq_sh} \end{eqnarray} where $d(\omega)=(c/\omega)[-\epsilon(\omega)]^{-1/2}$ is the optical skin depth, and $D\equiv d/dx_1$. In the case of small skin depths and surface slopes $(dD)^2\ll 1$, the square root term on the rhs of Eq. (\ref{eq_sh}) can be expanded as \begin{eqnarray} \lefteqn{[1-(dD)^2]^{1/2}= 1-\frac{1}{2} (dD)^2-\ldots} \nonumber \\ && \phantom{++}-\frac{1.1.3\ldots(2n-3)}{2.4.6\ldots 2n}(dD)^{2n} +O\bbox((dD)^{2n+2}\bbox). \label{eq_ddn} \end{eqnarray} Then the Fourier transform of the surface impedance function, which is needed in the calculation [cf. Eq. (\ref{eq_vq})], is related to the Fourier transform $\hat{f}(Q)$ of the surface profile function through \begin{eqnarray} \lefteqn{\hat{s}(Q)= -\frac{1-\epsilon(\omega)}{d(\omega) \epsilon(\omega)}\left(1-\frac{1}{2} [-i d(\omega)Q]^2\right.} && \nonumber \\ && \left.\phantom{++} -\frac{1}{8}[-i d(\omega)Q]^4 +O\bbox([-i d(\omega)Q]^6\bbox)\right)\hat{f}(Q). \label{eq_sqf} \end{eqnarray} In what follows, we will restrict the analysis to a Gaussian defect of $1/e$ half-width $a$ and height $h$: \begin{eqnarray} f(x_1) & = & h\exp\left(-x_1^2/a^2\right). \end{eqnarray} In addition, unless otherwise stated, we retain in Eq. (\ref{eq_ddn}), and thus in Eq. (\ref{eq_sqf}), only the zeroth order term in the expansion in powers of $[-i d(\omega)Q]^2$, as implicitly done in Ref. \onlinecite{apl98}. Therefore the function $\hat{s}(Q)$ we will use in our calculations is \begin{mathletters} \begin{eqnarray} \hat{s}(Q) &=& \pi^{1/2} s_0 a\exp[-(aQ)^2/4], \end{eqnarray} with \begin{eqnarray} s_0 &=& -\frac{1-\epsilon(\omega)}{\epsilon(\omega)} \frac{h}{d(\omega)}. \end{eqnarray} \label{eq_sqfg} \end{mathletters} It should be emphasized that the approximation involved in retaining only the lowest order term in the expansion (\ref{eq_ddn}) affects only the expression connecting the surface impedance with the real surface profile, the scattering formulation being rigorous and energy conserving (recall that losses are not accounted for) whatever the surface impedance is. Nonetheless, inasmuch as we wish to be able to quantitatively relate our results with real defect sizes, the effect of neglecting the higher order terms in Eq. (\ref{eq_ddn}) has to be determined. We have thus verified in the most unfavorable cases that including the first order term in $[-id(\omega )Q]^2$ in Eq. (\ref{eq_sqf}) barely modifies our calculations. In order to establish the accuracy and efficiency of the numerical calculations based on the formulation above, we first calculate the function $T(q,\omega)$ [cf. Eq. (\ref{eq_ietq})], following the numerical procedure outlined in Sec. \ref{sec_the} and the Appendix, for two Gaussian defects of half-width $a/\lambda=0.1$ and heights $h/\lambda=\pm 0.05$ (protuberance and indentation of equal height/depth), where $\lambda$ is the wavelength of the SPP. From these results, the SPP reflection and transmission coefficients are straightforwardly calculated [cf. Eqs. (\ref{eq_spprt}) and (\ref{eq_spprt2})], along with the DRC [cf. Eqs. (\ref{eq_rtq}) and (\ref{eq_drc})]. Furthermore, the magnetic and electric fields at any point in the vacuum half-space can be calculated from Eqs. (\ref{eq_h}) and (\ref{eq_e}) by using Eq. (\ref{eq_rtq}). In Fig.~\ref{fig_hi} we present the results thus obtained for the magnetic field intensities at the vacuum-metal interface in the vicinity of the Gaussian defects, and for the angular distribution of the scattered field in the far field. From the surface magnetic field in Fig.~\ref{fig_hi}(a), it is evident that both surface defects reflect back part of the incoming SPP, which interferes with the incoming SPP giving rise to the oscillatory pattern to the left of the defect (negative $x_1$-axis). Near the defect the magnetic field is perturbed. The outgoing transmitted SPP is seen to the right of the defect. The SPP reflection and transmission coefficients are: $R_{SP}=0.0025$ and $T_{SP}=0.9825$ for the protuberance, and $R_{SP}=0.0041$ and $T_{SP}=0.9728$ for the indentation. In Fig.~\ref{fig_hi}(b) a fairly uniform angular distribution of the DRC is observed (this will be discussed below). The total scattered power calculated from Eq. (\ref{eq_s}) is $S=0.0149$ for the protuberance, and $S=0.0231$ for the indentation. Energy conservation is thus satisfied within a $0.01\%$ error. We find that away from the vicinity of the defect, the magnetic field is fully described by either the interference between incoming and reflected SPP on the left-hand side, or by merely the transmitted SPP on the right-hand side. This corroborates, as expected, our argument in Sec.~\ref{sec_the_rt} leading to Eqs. (\ref{eq_hxinf}). The reflection and transmission coefficients are thus calculated from Eqs. (\ref{eq_spprt}) and (\ref{eq_spprt2}). \subsection{Energy balance dependence on defect size} \label{sec_res_rts} The question now arises naturally as to how efficient the surface defect is in coupling the incoming SPP into the different outgoing channels (either SPP or photons), or conversely, what the appropriate defect parameters are that maximize or minimize those channels; this is crucial for both an understanding of the scattering process and the design of practical devices. To that end, we have studied the dependence of the scattering coefficients $R_{SP}$, $T_{SP}$, and $S$ on the defect half-width $a$ for both Gaussian protuberances and indentations of different heights $h/\lambda=0.05$ and 0.2. The results are shown in Fig.~\ref{fig_rts}. Several general features are evident from these results. First, SPP reflection is relevant only for very narrow defects, $a<\lambda/5$, for either protuberances or indentations. Indeed there is an optimum defect width for which $R_{SP}$ is maximum.\cite{apl98} These defects are called {\it plasmon mirrors}.\cite{smol,apl98} For increasing defect widths, protuberances and indentations begin to behave differently, except for their negligible contribution to SPP reflection. On the one hand, SPP transmission through protuberances monotonically diminishes at the expense of radiation. The conversion is steeper the higher the defect is. Indentations, however, exhibit an oscillatory pattern with increasing defect width, in such a way that radiation (SPP transmission) increases (decreases), passes through a maximum (minimum), and then tends asymptotically to 0 (1). The oscillation period, the defect width that yields maximum radiation, and the value of this maximum, all depend on the surface height. Note that both protuberances and indentation may behave as {\it light-emitters}\cite{apl98} (high SPP-light conversion efficiency) for an appropriate (and distinct) range of defect parameters. Below we analyze in detail the behavior of SPP mirrors and light-emitters. \subsection{Narrow defects: SPP mirrors} \label{sec_res_mir} Surface defects playing the role of SPP mirrors have been studied experimentally in PSTM configurations.\cite{smol} This phenomenon has been analyzed for different defect shapes in Ref. \onlinecite{apl98}, where in addition a simple analytical prediction is given through a perturbation-theoretic argument. In the case of Gaussian-shaped defects, the predicted half-width that yields maximum reflection is: $a_{mirr}\approx [2^{1/2}k^R(\omega)]^{-1}$. Our numerical results further corroborate this prediction, since it is seen in Fig.~\ref{fig_rts} that $a_{mirr}/\lambda\approx 0.1$ no matter what the defect height is [as long as this height does not exceed the range of validity of Eq. (\ref{eq_sh})]. Nonetheless, the maximum SPP reflection increases with the defect height, and is slightly larger for indentations. The electric and magnetic near-field intensities for Gaussian defects, placed at the origin, of width $a/\lambda=0.1$ and heights $h/\lambda=\pm 0.05$ (protuberance and indentation, respectively) are presented in Figs.~\ref{fig_nfpm_p} and~\ref{fig_nfpm_i}, respectively,. The near-field maps are quite similar in both cases. The oscillations to the left of the obstacles clearly reveal the interference between the incident and backscattered SPP, their period being $T\approx 2\pi/(2k^R)$, as expected, and their contrast being related to $R_{SP}$. A bright region is seen to the right of the defects that is due to the strong SPP transmission. Poynting vector maps superimposed on the electric near-field intensity maps confirm the description of the energy flow given above. The corresponding angular distributions of scattered light (DRC) have been shown in Fig.~\ref{fig_hi}(b). There are no significant qualitative differences between protuberances and indentations, both yielding a fairly structure-less angular dependence; quantitatively, an indentation leads to stronger light coupling. The qualitative behavior is somewhat expected: the same perturbation-theoretic argument predicting maximum SPP reflection for a defect width that maximizes the scattering potential [cf. Eq. (\ref{eq_vq})] at backscattering,\cite{apl98} leads to a mostly uniform SPP coupling into EM waves in the radiative region ($|q|\leq\omega/c$). \subsection{Wide protuberances: Total light-emitters} \label{sec_res_wp} The ability of Gaussian-shaped protuberances to couple SPP into light has been pointed out in Ref. \onlinecite{apl98}. Here we analyze in detail the conditions for protuberances and indentations alike to behave as light-emitters with coupling efficiencies beyond 90$\%$, larger than that reported in Ref. \onlinecite{apl98}. Figure~\ref{fig_rts} above illustrates the discussion. To begin with, let us focus on protuberances. For widths beyond those producing significant SPP reflection (SPP mirrors), SPP-light conversion increases monotonically whereas, as expected from energy conservation, SPP transmission decays. This variation is faster for higher protuberances. Indeed, the curves in Fig. \ref{fig_rts} (bottom) indicate that $\lim_{a\rightarrow\infty}S=1$ even for the small protuberance. We have found coupling efficiencies beyond 90$\%$ in the case of $h/\lambda=0.2$ and $a/\lambda\geq 3.6$. In Fig.~\ref{fig_nfle_p}, the electric and magnetic near-field intensity maps for one such defect are shown. The absence of oscillations to the left of the defect reveals that SPP reflection is small; SPP transmission is small too (though considerably larger than $R_{SP}$), as seen on the right-hand side of the defect. A light beam is observed leaving the surface from the defect at near-grazing scattering angles. This energy flow picture is further corroborated by the angular distribution of the DRC shown in Fig.~\ref{fig_ile}, with a maximum at $\theta_s\approx 74^{\circ}$. Qualitatively, the fact that the metal protuberance enters the vacuum half-space seems to favor the SPP-photon coupling (playing the role of a {\it launching platform}). \subsection{Wide indentations: Light-emitters and SPP total transmission} \label{sec_res_wi} In the case of wide indentations, however, the behavior of the different outgoing channels differs from that for protuberances, and exhibits a richer phenomenology. Upon increasing the width of the indentation beyond the range of significant SPP reflection (see Fig.~\ref{fig_rts}), SPP transmission reaches a minimum value leading to maximum radiation, and then slowly grows towards {\it total transmission} (no radiation) in an oscillatory manner. The defect width that yields maximum radiation, its value, and the oscillations depend on the defect height. To understand such behavior, we plot in Figs.~\ref{fig_nfle_i_3} and~\ref{fig_nfle_i4} the electric near-field intensity maps for the higher defects ($h/\lambda=0.2$) of widths $a/\lambda=0.3$ and $a/\lambda=4$, respectively. These widths correspond to the first absolute and sixth subsidiary, maxima in Fig.~\ref{fig_rts} (bottom), respectively. Both indentations give rise to a negligible amount of SPP reflection (no oscillations to the left of the defect), as expected from Fig.~\ref{fig_rts} (top). SPP transmission (to the right of the defect) is very small for $a/\lambda=0.3$, but a strong light beam at grazing scattering is observed ($S=94\%$). For $a/\lambda=4$, although most of the energy goes into $T_{SP}=86.1\%$, a small amount of radiation also at grazing scattering angles is seen [recall that even though for this width a local maximum occurs in $S$, its value is very small $S=13.3\%$, see Fig~\ref{fig_rts} (bottom)]. But what is very illustrative to the discussion on the behavior of the outgoing channels is the near-field within the indentation (strictly speaking, right on top of the indentation region, since we are using an IBC on a flat surface). Oscillations are found therein, the number of minima (one in Fig.~\ref{fig_nfle_i_3} and six in Fig.~\ref{fig_nfle_i4}) being directly related to the position of the corresponding maxima in the $S$ vs. $a$ curve [see Fig.~\ref{fig_rts}]. Therefore, it can be inferred that the oscillatory behavior of SPP transmission and conversion into light in indentations is governed by a cavity-like effect. In fact, the near field map (not shown here) in the vicinity of any minimum-radiation indentation provides further evidence for this suggestion. As a consequence, the range of defect widths for which high coupling efficiencies are encountered is far more restrictive for indentations. In fact, in contrast with protuberances, only sufficiently deep indentations ($h/\lambda\geq 0.2$) are capable of producing radiation efficiencies $S>90\%$, and only for a narrow range of parameters. It seems as if the indentation geometry would somehow hinder grazing light scattering. Therefore, this kind of {\it light-emitters} might not correlate with any of the reciprocal versions (SPP flashlights) seen in PSTM experiments.\cite{smol,bozh} \subsection{Large width limit} \label{sec_res_lwl} Although the energy conservation criterion is reasonably well satisfied in our calculations, even for defects wider than those used in Fig.~\ref{fig_rts} (we have reached up to $a/\lambda=20$), one has to be careful when interpreting the results in the limit $a/\lambda\rightarrow\infty$. It turns out that the determination of the behavior of defects in this limit is important, since different tendencies have been encountered for protuberances and indentations (total radiation and transmission, respectively). The analysis of the appropriate defect width that yields maximum coupling is not simple even if making use of the Born approximation, since it requires the evaluation of the integral of $\hat{s}(q-k^R)$ for all homogeneous waves $|q|<\omega/c$. And yet such an approximation does not properly describe the formally exact numerical calculations. Alternatively, we have carried out an analytic calculation based on the use of a boundary condition similar to the Kirchhoff approximation. The approach relies on the expression for the scattering amplitude in terms of an integral equation along the surface with the magnetic field and its normal derivative inside the integrand (cf. Refs. \onlinecite{josaa91,ann} for the integral equation formulation, and Ref. \onlinecite{prb96} for its version making use of the IBC on a flat surface). By assuming that the surface magnetic field is given by the incoming SPP, the scattering amplitude reads in the large width limit: \begin{eqnarray} \lim_{a/\lambda\rightarrow\infty} R(q,\omega)= 2\pi\delta\bbox(q-k^R(\omega)\bbox)(1+\pi s_0). \end{eqnarray} Although it does not satisfy energy conservation (not surprisingly, due to the approximation involved), the former result gives an estimation of the limiting behavior shown above (see Fig.~\ref{fig_rts}): SPP transmission saturates for indentations ($s_0>0$), whereas protuberances ($s_0<0$) tend to decrease SPP transmission (in agreement with the numerical calculations, yet the exact limit is not predicted). \section{CONCLUSIONS} \label{sec_con} We have presented a theoretical formulation that describes in a rigorous manner the scattering of a surface plasmon polariton propagating along a planar vacuum-metal interface by a one-dimensional obstacle modeled through an impedance boundary condition. By solving the $k$-space scattering integral equation upon which the formulation is based, the angular spectrum of the scattered electromagnetic field in the vacuum half-space above the metal surface can be calculated, which in turn allows us to obtain the near electric and magnetic fields, the amplitudes of the reflected and transmitted SPP, and the angular distribution of the intensity of radiated waves resulting from the conversion of SPP into volume waves. A numerical method to solve the scattering integral equation has been put forth. We have made use of these calculation methods to study the SPP scattering by one-dimensional Gaussian defects, either protuberances or indentations. In particular, the dependence of the scattering process on the surface defect parameters has been analyzed. Several conclusions can be drawn from our results with respect to the behavior of Gaussian protuberances or indentations. SPP reflection is only significant for very narrow surface defects, with half-widths $a<c/\omega$. Our near field results explicitly show that in this case protuberances and indentations behave alike, the latter reflecting SPP slightly more efficiently. The dependence of the SPP reflection coefficient on the half-width confirms for different defect heights the condition predicted in Ref. \onlinecite{apl98} of maximum SPP reflection, leading to the {\it plasmon mirrors} seen in PSTM experiments.\cite{smol} For wider Gaussian defects, protuberances and indentations yield a entirely different picture, the only common feature being the negligible contribution to SPP reflection. Protuberances, on the one hand, increasingly radiate more light at near grazing scattering angles at the expense of SPP transmission. They behave as {\it light-emitters} with coupling efficiencies approaching 100\% with increasing half-width. The higher the defect is, the larger the SPP-light conversion. On the other hand, indentations tend to total SPP transmission without radiation with increasing half-width. The increase (decrease) of the SPP transmission (light coupling) occurs in an oscillatory manner starting from an absolute minimum (maximum) in transmission (radiation) for small half-widths, the period of the oscillations being related to the defect impedance in a way reminiscent of a cavity-like effect. Interestingly, we have found that for sufficiently deep indentations, this maximum radiation value can be extremely large (even larger than 90\%), so that the Gaussian indentation thus behave as a light-emitter. Our results and discussion provide a thorough picture of the different aspects of SPP scattering by surface defects which, besides being interesting in itself as a scattering process, appears to be useful in a number of related problems. \cite{ebbe98,schro98,nie97,klar98,smol95,smol,bozh,smol99,pasc98,apl98} It is rigorous for one-dimensional defects and indeed sheds light on the two-dimensional case, and can in turn explain and predict experimental results.\cite{smol,bozh} In this regard, it would be interesting to perform experiments on metal surfaces with defects of controled profile. With respect to the 2D case, it should be emphasized that the recent work by Shchegrov {\it et al.}\cite{snm97} for circularly symmetric surface defects reproduces the radiation pattern with peculiar lobes in the azimuthal angle dependence observed experimentally.\cite{smol} However, further theoretical work is needed that could address more complicated geometries used in the experiments and/or unexplained processes involving surface plasmon-polaritons. \cite{ebbe98,nie97,smol99} \acknowledgments This work was supported in part by Army Research Office Grant No. DAAH 0-96-1-0187, and by both the Spanish DGES (Grants No. PB96-0844-C02-02 and PB97-1221) and the Consejo Superior de Investigaciones Cient{\'\i}ficas. J.A.S.-G. acknowledges fruitful discussions with J. M. S{\'a}iz.
\section*{Introduction} This paper is concerned with the local statistics of the simultaneous zeros of $k$ random holomorphic sections $s_1, \dots, s_k \in H^0(M, L^N)$ of the $N^{\rm th}$ power $L^N$ of a positive Hermitian holomorphic line bundle $(L,h)$ over a compact K\"ahler\ manifold $M$ (where $k\le m=\dim M$). The terms `random' and `statistics' are with respect to a natural Gaussian probability measure $d \nu_N$ on $H^0(M, L^N)$ which we define below. In the special case where $M = \C\PP^m$ and $L$ is the hyperplane section bundle $\mathcal{O}(1)$, sections of $L^N$ correspond to holomorphic polynomials of degree $N$, and $(H^0(\C\PP^m, \mathcal{O}(N)), d \nu_N)$ is known as the ensemble of ${\operatorname{SU}}(m+1)$ polynomials in the physics literature. To obtain local statistics, we expand a ball $U$ around a given point $z^0$ by a factor $\sqrt N$ so that the average density of simultaneous scaled zeros is independent of $N$. We then ask whether the simultaneous scaled zeros behave as if thrown independently in $\sqrt{N} U$ or how they are correlated. Correlations between (unscaled) zeros are measured by the so-called {\it $n$-point zero correlation function\/} $K_{nk}^N(z^1,\dots,z^n)$, and those between scaled zeros are measured by the scaled correlation function $K_{nk}^N(\frac{z^1}{ \sqrt{N}},\dots,\frac{z^n}{ \sqrt{N}})$. Our main result is that the large $N$ limits of the scaled $n$-point correlation functions $K_{nk}^N(\frac{z^1}{ \sqrt{N}},\dots,\frac{z^n}{ \sqrt{N}})$ exist and are universal, i.e. are independent of $M,\ L$ and $h$ as well as the point $z^0$. Moreover, the scaling limit correlation functions can be calculated explicitly. We find that the limit correlations are short range, i.e. that simultaneous scaled zeros behave quite independently for large distances. On the other hand, nearby zeros exhibit some degree of repulsion. To state our problems and results more precisely, we begin with provisional definitions of the correlation functions $K_{nk}^N(z^1,\dots,z^n)$ and of the scaling limit. (See \S\S \ref{notation}--\ref{corfuns} for the complete definitions and notation.) In order to provide a standard yardstick for our universality results, we give $M$ the K\"ahler\ metric $\omega$ given by the (positive) curvature form of $h$. The metrics $h$ and $\omega$ then induce a Hilbert space inner product on the space $H^0(M, L^N)$ of holomorphic sections of $L^N$, for each $N\ge 1$. In the spirit of \cite{SZ} we use this $\mathcal{L}^2$-norm to define a Gaussian probability measure $d \nu_N$ on $H^0(M, L^N)$. When we speak of a random section, we mean a section drawn at random from this ensemble. More generally, we can draw $k$ sections $(s_1, \dots, s_k)$ independently and at random from this ensemble. Let $Z_{(s_1, \dots, s_k)}$ denote their simultaneous zero set and let $|Z_{(s_1, \dots, s_k)}|$ denote the ``delta measure" with support on $Z_{(s_1, \dots, s_k)}$ and with density given by the natural Riemannian volume $(2m - 2k)$-form defined by the metric $\omega$. To define the $n$-point zero correlation measure $K_{nk}^N(z^1, \dots, z^n)$ we form the product measure $$|Z_{(s_1, \dots, s_k)}|^n=\big(\underbrace{|Z_{(s_1, \dots, s_k)}|\times\cdots\times|Z_{(s_1, \dots, s_k)}|}_{n}\big)\quad \mbox{on}\quad M^n:= \underbrace{M\times\cdots\times M}_n\,.$$ To avoid trivial self-correlations, we puncture out the generalized diagonal in $M^n$ to get the punctured product space $$M_n=\{(z^1,\dots,z^n)\in M^n: z^p\ne z^q \ \ {\rm for} \ p\ne q\}\,.$$ We then restrict $|Z_{(s_1, \dots, s_k)}|^n$ to $M_n$ and define $K_{nk}^N(z^1, \dots, z^n)$ to be the expected value $E(|Z_{(s_1, \dots, s_k)}|^n)$ of this measure with respect to $\nu_N$. When $k = m$, the simultaneous zeros almost surely form a discrete set of points and so this case is perhaps the most vivid. Roughly speaking, $K_{nk}^N(z^1, \dots, z^n)$ gives the probability density of finding simultaneous zeros at $(z^{1}, \dots, z^n)$. The first correlation function $K_{1kN}$ just gives the expected distribution of simultaneous zeros of $k$ sections. In a previous paper \cite{SZ} by two of the authors, it was shown (among other things) that the expected distribution of zeros is asymptotically uniform; i.e. $$K_{1k}^N(z^0)=c_{mk}N^k+O(N^{k-1})\,,$$ for any positive line bundle (see \cite[Prop.~4.4]{SZ}). The question then arises of determining the higher correlation functions. As was first observed by \cite{BBL} and \cite {H} for ${\operatorname{SU}}(2)$ polynomials and by \cite{BD} for real polynomials in one variable, the zeros of a random polynomial are non-trivially correlated, i.e. the zeros are not thrown down like independent points. We will prove the same for all ${\operatorname{SU}}(m+1)$ polynomials and hence, by universality of the scaling limit, for any $M, L, h$. To introduce the scaling limit, let us return to the case $k = m$ where the simultaneous zeros form a discrete set of points. Since an $m$-tuple of sections of $L^N$ will have $N^m$ times as many zeros as $m$-tuples of sections of $L$, it is natural to expand $U$ by a factor of $\sqrt{N}$ to get a density of zeros that is independent of $N$. That is, we choose coordinates $\{z_q\}$ for which $z^0=0$ and $\omega(z^0)=\frac{i}{2}\sum_q dz_q\wedge d\bar z_q$ and then rescale $z \mapsto \frac{z}{ \sqrt{N}}$. Were the zeros thrown independently and at random on $U$, the conditional probability density of finding a simultaneous zero at a point $w$ given a zero at $z$ would be a constant independent of $(z,w)$. Non-trivial correlations (for any codimension $k\in\{1,\dots,m\}$) are measured by the difference between $1$ and the (normalized) {\it $n$-point scaling limit zero correlation function\/} $$\widetilde K_{nkm}^\infty(z^1,\dots,z^n) =\lim_{N\to\infty}\left(c_{mk}N^k\right)^{-n} {K_{nk}^N\left(\frac{z^1}{ \sqrt{N}},\dots,\frac{z^n}{ \sqrt{N}}\right)}\,,\qquad (z^1,\dots,z^n)\in U_n\,.$$ Our main result (Theorem \ref{uslc}) is universality of the scaling limit correlation functions: \bigskip\begin{quote} {\it The $n$-point scaling limit zero correlation function $\widetilde K_{nkm}^\infty(z^1,\dots,z^n)$ is given by a universal rational function, homogeneous of degree $0$, in the values of the function $e^{i\Im (z\cdot \bar w)-{\frac{1}{2}} |z-w|^2}$ and its first and second derivatives at the points $(z,w)=(z^p,z^{p'})$, $1\le p,p'\le n$. Alternately it is a rational function in $z^p_q, \bar z^p_q, e^{z^p\cdot \bar z^{p'}}$} \end{quote} \bigskip The function $e^{i\Im (z\cdot \bar w)-{\frac{1}{2}} |z-w|^2}$ which appears in the universal scaling limit is (up to a constant factor) the Szeg\"o kernel $\Pi_1^\H(z,w)$ of level one for the reduced Heisenberg group ${\bf H}^n_{\rm red}$ (cf.\ \S \ref{notation}). Its appearance here owes to the fact that the correlation functions can be expressed in terms of the Szeg\"o kernels $\Pi_N(x,y)$ of $L^N$. I.e., let $X$ denote the circle bundle over $M$ consisting of unit vectors in $L^*$; then $\Pi_N(x,y)$ is the kernel of the orthogonal projection $\Pi_N:\mathcal{L}^2(X)\to \mathcal{H}^2_N(X) \approx H^0(M, L^N)$. Indeed we have (Theorem \ref{npointcor}): \bigskip \begin{quote} {\it The $n$-point correlation $\widetilde K_{nk}^N(z^1,\dots,z^n)$ is given by the above universal rational function, applied this time to the values of the Szeg\"o kernel $\Pi_N$ and its first and second derivatives at the points $(z^p,z^{p'})$. } \end{quote} \bigskip In view of this relation between the correlation functions and the Szeg\"o kernel, it suffices for the proof of the universality theorem \ref{uslc} to determine the scaling limit of the Szeg\"o kernel $\Pi_N$ and to show its universality. Indeed we shall show (Theorem \ref{neardiag}) that: \bigskip \begin{quote} {\it Let $(z_1,\dots,z_m,\theta)$ denote local coordinates in a neighborhood $\widetilde U\approx U\times S^1$ of a point $(z^0,\lambda)\in X$ (where $(z_1,\dots,z_m)$ are the above local coordinates about $z_0\in M$). We then have $$N^{-m}\Pi_N\left(\frac{z}{\sqrt{N}},\frac{\theta}{N}; \frac{z'}{\sqrt{N}},\frac{\theta'}{N}\right)= \Pi^\H_1(z,\theta;z',\theta') + O(N^{-1/2})\;.$$ }\end{quote} \bigskip The fact that the correlation functions can be expressed in terms of the Szeg\"o kernel may be explained in (at least) two ways. The first is that the correlation functions may be expressed in terms of the joint probability density $D_{nk}^N(x,\xi;z)dxd\xi$ of the (vector-valued) random variable $$(x,\xi)=\big[x^p,\xi^p\big]_{1\le p\le n}\;,\qquad x^p=(s_1(z^p),\dots, s_k(z^p)),\quad \xi^p= (\nabla s_1(z^p), \dots, \nabla s_k(z^p))$$ given by the values of the $k$ sections and of their covariant derivatives at the $n$ points $\{z^p\}$. Our method of computing the correlation functions is based on the following probabilistic formula (Theorem \ref{density}): \bigskip \begin{quote} {\it For $N$ sufficiently large so that the density $D_{nk}^N (x,\xi;z)$ is given by a continuous function, we have $$ K_{nk}^N(z)= \int d\xi\,D_{nk}^N(0,\xi;z)\prod_{p=1}^n \det \left(\xi^{p}_j\xi^{p*}_{j'}\right)_{1\le j,j'\le k}\,, \quad z=(z^1,\dots,z^n)\in M_n\,,$$ where $\xi=(\xi^1,\dots,\xi^n)$ and $\xi^{p*}_j:L^N_{z^p}\to T_{M,z^p}$ denotes the adjoint to $\xi^p_j:T_{M,z^p}\to L^N_{z^p}$.}\end{quote} \bigskip This formula, which is valid in a more general setting, is based on the approach of Kac \cite{Ka} and Rice \cite{Ri} (see also \cite {EK}) for zeros of functions on ${\mathbb R}^1$, and of \cite{Halp} for zeros of (real) Gaussian vector fields. Since our probability measure $d\nu_N$ (on the space of sections) is Gaussian, it follows that $D_{nk}^N$ is also a Gaussian density. It will be proved in \S \ref{D-meet-S} that the covariance matrix of this Gaussian may be expressed entirely in terms of $\Pi_N$ and its covariant derivatives. This type of formula for the correlation function of zeros was previously used in \cite{BD}, \cite {H} and the works cited above. We believe that this formula will have interesting applications in geometry. A second link between correlation functions and Szeg\"o kernels is given by the Poincar\'e-Lelong formula. In fact, this was our original approach to computing the correlation functions in the codimension 1 case. For the sake of brevity, we will not discuss this approach here; instead we refer the reader to our companion article \cite{BSZ}. {From} the universality of our answers, it follows that the scaling limit pair correlation functions depend only on the distance between points: $$\widetilde K^\infty_{2km}(z^1,z^2)=\kappa_{km}(r)\,,\qquad r=|z^1-z^2|\,,$$ where $\kappa_{km}$ depends only on the dimension $m$ of $M$ and the codimension $k$ of the zero set. In \S \ref{formulas}, we give explicit formulas for the limit pair correlation functions $\kappa_{km}$ in some special cases. Our calculation uses the Heisenberg model, which (although noncompact) is the most natural one since the scaled Szeg\"o kernels are all equal to $\Pi_1$, and there is no need in this case to take a limit. We also discuss the hyperplane section bundle $\mathcal{O} (1) \to \C\PP^m$, which is the most studied, since the sections of its powers are the ${\operatorname{SU}}(m+1)$ polynomials---homogeneous polynomials in $m+1$ variables---and the case $m=1$ (the ${\operatorname{SU}}(2)$ polynomials) appears frequently in the physics literature (e.g., \cite{BBL, FH, H, KMW, PT}). We give expressions for the zero correlations $K^N_{nk}$ for the ${\operatorname{SU}}(m+1)$ polynomials and by letting $N\to\infty$, we obtain an alternate derivation of our universal formula for the scaling limit correlation. We show (Theorem \ref{shortrange}) that $\kappa_{km}(r)= 1+O(r^4e^{-r^2})$ as $r\to +\infty$, and hence these correlations are short range in that they differ from the case of independent random points by an exponentially decaying term. We observe that when $\dim M=1$, there is a strong repulsion between nearby zeros in the sense that $\kappa_{11}(r) \to 0$ as $r \to 0$, as was noted by Hannay \cite{H} and Bogomolny-Bohigas-Leboeuf \cite{BBL} for the case of ${\operatorname{SU}}(2)$ polynomials. These asymptotics are illustrated by the following graph (see also \cite{H}): \vspace{-.5in} \begin{figure}[ht]\label{kappa11}\centering \epsfig{file=k11.eps,height=2.5in} \caption{The 1-dimensional limit pair correlation function $\kappa_{11}$} \end{figure} For $\dim M=2$, the simultaneous scaled zeros of a random pair $(s_1, s_2)$ of sections still exhibit a mild repulsion ($\lim_{r \to 0} \kappa_{22}(r) =\frac{3}{4}$), as illustrated in Figure~2 below. \vspace{-.1in} \begin{figure}[ht]\label{kappa22}\centering \epsfig{file=k22.eps,height=2.5in} \caption{The limit pair correlation function $\kappa_{22}$} \end{figure} The function $\kappa_{mm}(r)$ can be interpreted as the normalized conditional probability of finding a zero near a point $z^1$ given that there is a zero at a second point a scaled distance $r$ from $z^1$ (in the case of discrete zeros in $m$ dimensions). The above graphs show that for dimensions 1 and 2, there is a unique scaled distance where this probability is maximized. It would be interesting to explore the dependence of the correlations on the dimension. To ask one concrete question, do the simultaneous scaled zeros in the point case become more and more independent in the sense that $\kappa_{mm} (r) \to 1$ as the dimension $m \to \infty$? When $k<m$, the zero sets are subvarieties of positive dimension $m-k$; in this case the expected volume of the zero set in a small spherical shell of radius $r$ and thickness $\varepsilon$ about a point in the zero set must be $\sim \varepsilon r^{2m-2k-1}$. Hence we have $\kappa_{km}(r)\sim r^{-2k}$, for small $r$. The graph of the limit correlation function for the case $m=2,k=1$ is given in Figure~3 below. \vspace{-.25in} \begin{figure}[ht]\label{kappa12}\centering \epsfig{file=k12.eps,height=2.5in} \caption{The limit pair correlation function $\kappa_{12}$} \end{figure} To end this introduction, we would like to link our methods and results at least heuristically to a long tradition of (largely heuristic) results on universality and scaling in statistical mechanics (cf. \cite{FFS}). One may view the rescaling transformation on $U$ as generating a renormalization group. The intuitive picture in statistical mechanics is that the renormalization group should carry a given system (read ``$L \to M$'') to the fixed point of the renormalization group, i.e. to the scale invariant situation. We observe that the local rescaling of $U$ is nothing other than the Heisenberg dilations $\delta_{\sqrt{N}}$ on ${\bf H}^m_{\rm red}$. Since these dilations are automorphisms of the (unreduced) Heisenberg group, the Szeg\"o kernel of ${\bf H}^m$ is invariant under these dilations; i.e., it is the fixed point of the renormalization group. As predicted by this intuitive picture, we find that in the scaling limit all the invariants of the line bundle, in particular its zero-point correlation functions, are drawn to their values for the fixed point system (read ``Heisenberg model''). \bigskip \section{Notation}\label{notation} We begin with some notation and basic properties of sections of holomorphic line bundles, their zero sets, Szeg\"o kernels, and Gaussian measures. We also provide two examples that will serve as model cases for studying correlations of zeros of sections of line bundles in the high power limit. \subsection{Sections of holomorphic line bundles}\label{cxgeom} In this section, we introduce the basic complex analytic objects: holomorphic sections and the currents of integration over their zero sets. We also introduce Gaussian probability measures on spaces of holomorphic sections. For background in complex geometry, we refer to \cite{GH}. Let $M$ be a compact complex manifold and let $L\to M$ be a holomorphic line bundle with a smooth Hermitian metric $h$; its curvature 2-form $\Theta_h$ is given locally by \begin{equation}\label{curvature}\Theta_h=-\partial\dbar \log\|e_L\|_h^2\;,\end{equation} where $e_L$ denotes a local holomorphic frame (= nonvanishing section) of $L$ over an open set $U\subset M$, and $\|e_L\|_h=h(e_L,e_L)^{1/2}$ denotes the $h$-norm of $e_L$. We say that $(L,h)$ is positive if the (real) 2-form $\omega=\frac{\sqrt{-1}}{2}\Theta_h$ is positive, i.e., if $\omega$ is a K\"ahler\ form. We henceforth assume that $(L,h)$ is positive, and we give $M$ the Hermitian metric corresponding to the K\"ahler\ form $\omega$ and the induced Riemannian volume form \begin{equation}\label{dV} dV_M= \frac{1}{m!} \omega^m\;.\end{equation} Since $\frac{1}{\pi}\omega$ is a de Rham representative of the Chern class $c_1(L)\in H^2(M,{\mathbb R})$, the volume of $M$ equals $\frac{\pi^m}{m!}c_1(L)^m$. The space $H^0(M, L^{N})$ of global holomorphic sections of $L^N=L\otimes\cdots\otimes L$ is a finite dimensional complex vector space. (Its dimension, given by the Riemann-Roch formula for large $N$, grows like $N^m$. By the Kodaira embedding theorem, the global sections of $L^N$ give an embedding into a projective space for $N\gg 0$, and hence $M$ is a {\it projective algebraic manifold.\/}) The metric $h$ induces Hermitian metrics $h^N$ on $L^N$ given by $\|s^{\otimes N}\|_{h^N}=\|s\|_h^N$. We give $H^0(M,L^N)$ the Hermitian inner product \begin{equation}\label{inner}\langle s_1, s_2 \rangle = \int_M h^N(s_1, s_2)dV_M \quad\quad (s_1, s_2 \in H^0(M,L^N)\,)\;,\end{equation} and we write $|s|=\langle s,s \rangle^{1/2}$. We now explain our concept of a ``random section." We are interested in expected values and correlations of zero sets of $k$-tuples of holomorphic sections of powers $L^N$. Since the zeros do not depend on constant factors, we could suppose our sections lie in the unit sphere in $H^0(M,L^N)$ with respect to the Hermitian inner product (\ref{inner}), and we pick random sections with respect to the spherical measure. Equivalently, we could suppose that $s$ is a random element of the projectivization ${\mathbb P} H^0(M,L^N)$. Another equivalent approach is to use Gaussian measures on the entire space $H^0(M,L^N)$. We shall use the third approach, since Gaussian measures seem the best for calculations. Precisely, we give $H^0(M,L^N)$ the complex Gaussian probability measure \begin{equation}\label{gaussian}d\nu_N(s)=\frac{1}{\pi^m}e^ {-|c|^2}dc\,,\qquad s=\sum_{j=1}^{d_N}c_jS_j^N\,,\end{equation} where $\{S_j^N\}$ is an orthonormal basis for $H^0(M,L^N)$ and $dc$ is $2d_N$-dimensional Lebesgue measure. This Gaussian is characterized by the property that the $2d_N$ real variables $\Re c_j, \Im c_j$ ($j=1,\dots,d_N$) are independent random variables with mean 0 and variance ${\frac{1}{2}}$; i.e., $${\mathbf E}\, c_j = 0,\quad {\mathbf E}\, c_j c_k = 0,\quad {\mathbf E}\, c_j \bar c_k = \delta_{jk}\,.$$ Here and throughout this paper, ${\mathbf E}\,$ denotes expectation. In general, a {\it complex Gaussian measure\/} (with mean 0) on a finite dimensional complex vector space $V$ is a measure $\nu$ of the form (\ref{gaussian}), where the $c_j$ are the coordinates with respect to some basis. Explicitly, the complex Gaussian measures on ${\mathbb C}^m$ are the probability measures of the form \begin{equation}\label{gaussian2} \frac{e^{-\langle \Delta^{-1}z,z\rangle}}{\pi^m\det \Delta}dz \end{equation} where $\Delta=(\Delta^j_k)$ is a positive definite Hermitian matrix and $$\langle \zeta, z\rangle=\zeta\cdot\bar z=\sum_{q=1}^m \zeta_q\bar z_q$$ denotes the standard Hermitian inner product in ${\mathbb C}^m$. For the Gaussian measure (\ref{gaussian2}), we have \begin{equation}\label{gaussian3} {\mathbf E}\,(z_j z_k)=0, \qquad {\mathbf E}\,(z_j \bar z_k)=\Delta^j_k\,.\end{equation} If $\nu$ is a complex Gaussian on $V$ and $\tau:V\to \widetilde V$ is a surjective linear transformation, then $\tau_*\nu$ is a complex Gaussian on $\widetilde V$. In particular, if $\widetilde V={\mathbb C}^m$, then, $\tau_*\nu$ is of the form (\ref{gaussian2}), where the covariance matrix $\Delta$ is given by (\ref{gaussian3}) with $z_j=z_j\circ\tau:V\to{\mathbb C}$. We shall consider the space $\mathcal{S}=H^0(M,L^N)^k$ ($1\le k \le m$) with the probability measure $d\mu=d \nu\times\cdots\times d\nu$, which is also Gaussian. Picking a random element of $\mathcal{S}$ means picking $k$ sections of $H^0(M,L^N)$ independently and at random. For $s=(s_1,\dots,s_k)\in\mathcal{S}$, we let $$Z_s=\{z\in M: s_1(z)=\cdots=s_k(z)=0\}$$ denote the zero set of $s$. Note that if $N$ is sufficiently large so that $L^N$ is base point free, then for $\mu$-a.a. $s\in\mathcal{S}$, we have ${\operatorname{codim\,}} Z_s=k$. (Indeed, the set of $s$ where ${\operatorname{codim\,}} Z_s<k$ is a proper algebraic subvariety of $H^0(M,L^N)^k$. In fact, by Bertini's theorem, the $Z_s$ are smooth submanifolds of complex dimension $m-k$ for almost all $s$, provided $N$ is large enough so that the global sections of $L^N$ give a projective embedding of $M$, but we do not need this fact here.) For these $s$, we let $|Z_s|$ denote Riemannian $(2m-2k)$-volume along the regular points of $Z_s$, regarded as a measure on $M$: \begin{equation}\label{volmeasure} (|Z_s|,\phi)=\int_{Z_s^{\rm reg}}\phi d{\operatorname{Vol}}_{2m-2k}=\frac{1}{(m-k)!}\int_{Z_s^{\rm reg}}\phi \omega^{m-k}\,.\end{equation} It was shown by Lelong \cite{Le} (see also \cite{GH}) that the integral in (\ref{volmeasure}) converges. (In fact, $|Z_s|$ can be regarded as the total variation measure of the closed current of integration over $Z_s$.) We regard $|Z_s|$ as a measure-valued random variable on the probability space $(\mathcal{S},d\mu)$; i.e., for each test function $\phi\in\mathcal{C}^0(M)$, $(|Z_s|,\phi)$ is a complex-valued random variable. \subsection{Szeg\"o kernels}\label{Szegokernels} As in \cite{Z,SZ} we now lift the analysis of holomorphic sections over $M$ to a certain $S^1$ bundle $X \to M$. This is a useful approach to the asymptotics of powers of line bundles and goes back at least to \cite{BG}. We let $L^*$ denote the dual line bundle to $L$, and we consider the circle bundle $X=\{\lambda \in L^* : \|\lambda\|_{h^*}= 1\}$, where $h^*$ is the norm on $L^*$ dual to $h$. Let $\pi:X\to M$ denote the bundle map; if $v\in L_z$, then $\|v\|_h=|(\lambda,v)|$, $\lambda\in X_z=\pi^{-1}(z)$. Note that $X$ is the boundary of the disc bundle $D = \{\lambda \in L^* : \rho(\lambda)>0\}$, where $\rho(\lambda)=1-\|\lambda\|^2_{h^*}$. The disc bundle $D$ is strictly pseudoconvex in $L^*$, since $\Theta_h$ is positive, and hence $X$ inherits the structure of a strictly pseudoconvex CR manifold. Associated to $X$ is the contact form $\alpha= -i\partial\rho|_X=i\bar\partial\rho|_X$. We also give $X$ the volume form \begin{equation}\label{dvx}dV_X=\frac{1}{m!}\alpha\wedge (d\alpha)^m=\alpha\wedge\pi^*dV_M\,.\end{equation} The setting for our analysis of the Szeg\"o kernel is the Hardy space $\mathcal{H}^2(X) \subset \mathcal{L}^2(X)$ of square-integrable CR functions on $X$, i.e., functions that are annihilated by the Cauchy-Riemann operator $\bar\partial_b$ (see \cite[pp.~592--594]{Stein}) and are $\mathcal{L}^2$ with respect to the inner product \begin{equation}\label{unitary} \langle F_1, F_2\rangle =\frac{1}{2\pi}\int_X F_1\overline{F_2}dV_X\,,\quad F_1,F_2\in\mathcal{L}^2(X)\,.\end{equation} Equivalently, $\mathcal{H}^2(X)$ is the space of boundary values of holomorphic functions on $D$ that are in $\mathcal{L}^2(X)$. We let $r_{\theta}x =e^{i\theta} x$ ($x\in X$) denote the $S^1$ action on $X$ and denote its infinitesimal generator by $\frac{\partial}{\partial \theta}$. The $S^1$ action on $X$ commutes with $\bar{\partial}_b$; hence $\mathcal{H}^2(X) = \bigoplus_{N =0}^{\infty} \mathcal{H}^2_N(X)$ where $\mathcal{H}^2_N(X) = \{ F \in \mathcal{H}^2(X): F(r_{\theta}x) = e^{i N \theta} F(x) \}$. A section $s$ of $L$ determines an equivariant function $\hat{s}$ on $L^*$ by the rule $\hat{s}(\lambda) = \left(\lambda, s(z) \right)$ ($\lambda \in L^*_z, z \in M$). It is clear that if $\tau \in {\mathbb C}$ then $\hat{s}(z, \tau \lambda) = \tau \hat{s}$. We henceforth restrict $\hat{s}$ to $X$ and then the equivariance property takes the form $\hat{s}(r_{\theta} x) = e^{i \theta}\hat{s}(x)$. Similarly, a section $s_N$ of $L^{N}$ determines an equivariant function $\hat{s}_N$ on $X$: put $$\hat{s}_N(\lambda) = \left( \lambda^{\otimes N}, s_N(z) \right)\,,\quad \lambda\in X_z\,,$$ where $\lambda^{\otimes N} = \lambda \otimes \cdots\otimes \lambda$; then $\hat s_N(r_\theta x) = e^{iN\theta} \hat s_N(x)$. The map $s\mapsto \hat{s}$ is a unitary equivalence between $H^0(M, L^{ N})$ and $\mathcal{H}^2_N(X)$. (This follows from (\ref{dvx})--(\ref{unitary}) and the fact that $\alpha= d\theta$ along the fibers of $\pi:X\to M$.) We let $\Pi_N : \mathcal{L}^2(X) \rightarrow \mathcal{H}^2_N(X)$ denote the orthogonal projection. The Szeg\"o kernel $\Pi_N(x,y)$ is defined by \begin{equation} \Pi_N F(x) = \int_X \Pi_N(x,y) F(y) dV_X (y)\,, \quad F\in\mathcal{L}^2(X)\,. \end{equation} It can be given as \begin{equation}\label{szego}\Pi_N(x,y)=\sum_{j=1}^{d_N}\widehat S_j^N(x)\overline{\widehat S_j^N(y)}\,,\end{equation} where $S_1^N,\dots,S_{d_N}^N$ form an orthonormal basis of $H^0(M,L^N)$. Pick a local holomorphic frame $e_L$ for $L$ over an open subset $U\subset M$, let $e_L^*$ denote the dual frame, and write $h(z)=h(e_L(z),e_L(z))=\|e_L\|_h^2$. The map $(z,e^{i\theta})\mapsto e^{i\theta}h(z)^{1/2} e_L^*(z)$ gives an isomorphism $U\times S^1\approx \pi^{-1}(U)\subset X$, and we use the coordinates $(z,\theta)$ to identify points of $\pi^{-1}(U)$. For $s\in H^0(M,L^N)$, we have \begin{equation} \label{hats}\hat s(z,\theta)= \left\langle s(z),e^{iN\theta}h(z)^{N/2}e^*_L (z)\right\rangle = e^{iN\theta}h(z)^{N/2} f(z),\quad s=fe_L^{\otimes N}\,. \end{equation} Although the Szeg\"o kernel is defined on $X$, its absolute value is well-defined on $M$ as follows: writing $S_j^N=f^N_j e_L^{\otimes N}$, we have \begin{equation}\label{PiN}\Pi_N(z,\theta;w,\phi) =e^{iN(\theta-\phi)}\Pi_N(z,0;w,0)= e^{iN(\theta-\phi)}h(z)^{N/2} h(w)^{N/2}\sum_{j=1}^{d_N}f^N_j(z)\overline{f^N_j(w)}\,,\end{equation} for $z,w\in U$. (Here we may take $U$ to be the disjoint union of connected neighborhoods of $z$ and $w$, if $z$ is not close to $w$.) Thus we can write $$|\Pi_N(z,w)|= |\Pi_N(z,0;w,0)|\,,$$ which is independent of the choice of local frame $e_L$. On the diagonal we have $$\Pi_N(z,z)= \Pi_N(z,\theta;z,\theta)=\sum_{j=1}^{d_N}\|S_j^N(z)\|_{h^N}\,.$$ The Hermitian connection $\nabla$ on $L$ induces the decomposition $T_X=T_X^{H}\oplus T_X^{V}$ into horizontal and vertical components, and we let $t^H$ denote the horizontal lift (to $X$) of a vector field $t$ in $M$. We consider the horizontal operators on $X$: $$d_{z_q}^H \buildrel {\operatorname{def}}\over = d_{(\d /\d z_q)^H}\,,\quad d_{\bar z_q}^H\buildrel {\operatorname{def}}\over = d_{(\d /\d \bar z_q)^H}\,,$$ where $z_1,\dots,z_m,\theta$ denote local coordinates on $X$. We note that \begin{equation} \label{horizontal} d_{z_q}^H \hat s = (\nabla^N_{z_q} s)\raisebox{2pt}{$\wh{\ }$} \;,\quad s\in H^0(M,L^N)\,,\end{equation} where $\nabla^N$ is the induced connection on $L^N$. We then have \begin{eqnarray}d^H_{z_q}\Pi_N(z,\theta;w,\phi) &=&\sum_{j=1}^{d_N}\left(\nabla^N_{z_q} S_j^N\right)\raisebox{2pt}{$\wh{\ }$}(x)\overline{\widehat S_j^N(y)}\nonumber\\ &=&e^{iN(\theta-\phi)}h(z)^{N/2} h(w)^{N/2}\sum_{j=1}^{d_N}f^N_{j;q}(z)\overline{f^N_j(w)}\,,\nonumber\\ \label{dPiN} d^H_{z_p} d^H_{\bar w_q}\Pi_N(z,\theta;w,\phi)&=& \sum_{j=1}^{d_N}\left(\nabla^N_{z_p} S_j^N\right)\raisebox{2pt}{$\wh{\ }$}(x)\overline{\left(\nabla^N_{w_q} S_j^N\right)\raisebox{2pt}{$\wh{\ }$}(y)}\\ &=&e^{iN(\theta-\phi)}h(z)^{N/2} h(w)^{N/2}\sum_{j=1}^{d_N}f^N_{j;p}(z)\overline{f^N_{j;q}(w)}\,, \nonumber \end{eqnarray} $$ \qquad \nabla^N_{z_q}=\nabla^N_{\d /\d z_q}\,,\quad f^N_{j;q} =\frac{\d f}{\d z_q}+Nf^N_jh^{-1}\frac{\d h}{\d z_q}\,.$$ We can also use (\ref{hats}) and (\ref{horizontal}) to describe the horizontal lift in local coordinates: \begin{equation}\label{horizontal2} d^H_{z_q}= \frac{\d}{\d z_q} -\frac{i}{2} \frac{\d\log h}{\d z_q}\frac{\d}{\d\theta}\,.\end{equation} \subsection{Model examples} In two special cases we can work out the Szeg\"o kernels and their derivatives explicitly, namely for the hyperplane section bundle over $\C\PP^m$ and for the Heisenberg bundle over ${\mathbb C}^m$, i.e. the trivial line bundle with curvature equal to the standard symplectic form on ${\mathbb C}^m$. These cases will be important after we have proven universality, since scaling limits of correlation functions for all line bundles coincide with those of the model cases. In fact, the two models are locally equivalent in the CR sense. In the case of $\C\PP^m$, the circle bundle $X$ is the $2m + 1$ sphere $S^{2m + 1}$, which is the boundary of the unit ball $B^{2m + 2} \subset {\mathbb C}^{m+1}$. In the case of ${\mathbb C}^m$, the circle bundle is the reduced Heisenberg group ${\bf H}^m_{\rm red}$, which is a discrete quotient of the simply connected Heisenberg group ${\mathbb C}^m \times {\mathbb R}$. As is well-known, the latter is equivalent (in the CR and contact sense) to the boundary of $B^{2m + 2} $ (\cite{Stein}). \subsubsection{${\operatorname{SU}}(m+1)$-polynomials}\label{example1} For our first example, we let $M=\C\PP^m$ and take $L$ to be the hyperplane section bundle $\mathcal{O}(1)$. Sections $s\in H^0({\mathbb C}{\mathbb P}^m,\mathcal{O}(1))$ are linear functions on ${\mathbb C}^{m+1}$; the zero divisors $Z_s$ are projective hyperplanes. The line bundle $\mathcal{O}(1)$ carries a natural metric $h_{{\operatorname{FS}}}$ given by \begin{equation}\label{hfs} \|s\|_{h_{{\operatorname{FS}}}}([w])=\frac{|(s,w)|}{|w|}\;, \quad\quad w=(w_0,\dots,w_m)\in{\mathbb C}^{m+1}\;,\end{equation} for $s\in{\mathbb C}^{m+1*}\equiv H^0({\mathbb C}{\mathbb P}^m,\mathcal{O}(1))$, where $|w|^2=\sum_{j=0}^m |w_j|^2$ and $[w]\in{\mathbb C}{\mathbb P}^m$ is the complex line through $w$. The K\"ahler\ form on $\C\PP^m$ is the Fubini-Study form \begin{equation} \omega_{{\operatorname{FS}}}=\frac{\sqrt{-1}}{2}\Theta_{h_{{\operatorname{FS}}}}=\frac{\sqrt{-1}}{2} \partial\dbar \log |w|^2 \,.\end{equation} The dual bundle $L^*=\mathcal{O}(-1)$ is the affine space ${\mathbb C}^{m+1}$ with the origin blown up, and $X=S^{2m+1}\subset{\mathbb C}^{m+1}$. The $N$-th tensor power of $\mathcal{O}(1)$ is denoted $\mathcal{O}(N)$. Elements $s_N\in H^0({\mathbb C}{\mathbb P}^m,\mathcal{O}(N))$ are homogeneous polynomials on ${\mathbb C}^{m+1}$ of degree $N$, and $\hat s_N=s_N|_{S^{2m-1}}$. The monomials \begin{equation}\label{orthonormal}s^N_J = \left[\frac{(N+m)!}{\pi^m j_0!\cdots j_m!}\right]^{\frac{1}{2}} z^J\,,\quad z^J= z_0^{j_0}\cdots z_m^{j_m},\quad\quad J=(j_0,\ldots,j_m),\ |J|=N\end{equation} form an orthonormal basis for $H^0({\mathbb C}{\mathbb P}^m,\mathcal{O}(N))$. (See \cite[\S 4.2]{SZ}; the extra factor $\left(\frac{m!}{\pi^m}\right)^{1/2}$ in (\ref{orthonormal}) comes from the fact that here $\C\PP^m$ has the usual volume $\frac{\pi^m}{m!}$, whereas in \cite{SZ}, the volume of $\C\PP^m$ is normalized to be 1.) Hence the Szeg\"o kernel for $\mathcal{O}(N)$ is given by \begin{equation}\label{Szegosphere} \Pi_N(x,y)=\sum_J \frac{(N+m)!}{\pi^mj_0!\cdots j_m!}x^J \bar y^J = \frac{(N+m)!} {\pi^mN!}\langle x,y\rangle^N\,.\end{equation} Note that $$\Pi(x,y)=\sum_{N=1}^\infty\Pi_N(x,y)=\frac{m!}{\pi^m} (1-\langle x, y\rangle)^{-(m+1)} = 2\pi\times [\mbox{classical Szeg\"o kernel on} \ S^{2m+1}]\,.$$ (The factor $2\pi$ is due to our normalization (\ref{unitary}).) \subsubsection{The Heisenberg model}\label{example2} Our second example is the linear model ${\mathbb C}^m \times {\mathbb C} \to {\mathbb C}^m$ for positive line bundles $L \to M$ over K\"ahler\ manifolds and their associated Szeg\"o kernels. It is most illuminating to consider the associated principal $S^1$ bundle ${\mathbb C}^m \times S^1 \to {\mathbb C}^m$, which may be identified with the boundary of the disc bundle $D \subset L^*$ in the dual line bundle. This $S^1$ bundle is the {\it reduced Heisenberg group} ${\bf H}_{\rm red}^m$ (cf. \cite{F}, p. 23). Let us recall its definition and properties. We start with the usual (simply connected) Heisenberg group ${\bf H}^m$ (cf. \cite{F} \cite{Stein}; note that different authors differ by factors of $2$ and $\pi$ in various definitions). It is the group ${\mathbb C}^m \times {\mathbb R}$ with group law $$(\zeta, t) \cdot (\eta, s) = (\zeta + \eta, t + s + \Im (\zeta \cdot \bar{\eta})).$$ The identity element is $(0, 0)$ and $(\zeta, t)^{-1} = (- \zeta, - t)$. Abstractly, the Lie algebra of ${\bf H}_m$ is spanned by elements $Z_1, \dots, Z_m, \bar{Z}_1, \dots, \bar{Z}_m, T$ satisfying the canonical commutation relations $[Z_j, \bar{Z}_k] = -i \delta_{jk} T$ (all other brackets zero). Below we will select such a basis of left invariant vector fields. ${\bf H}^m$ is a strictly convex CR manifold which may be embedded in ${\mathbb C}^{m + 1}$ as the boundary of a strictly pseudoconvex domain, namely the upper half space $ \mathcal{U} ^m := \{z \in {\mathbb C}^{m + 1}: \Im z_{m + 1} > {\frac{1}{2}} \sum_{j = 1}^m |z_j|^2 \}$. The boundary of $\mathcal{U} ^m$ equals $\partial \mathcal{U} ^m = \{z \in {\mathbb C}^{m + 1}: \Im z_{m + 1} = {\frac{1}{2}} \sum_{j = 1}^m |z_j|^2\}$. ${\bf H}^m$ acts simply transitively on $\partial \mathcal{U} ^m$ (cf. \cite{Stein}, XII), and we get an identification of ${\bf H}^m$ with $\partial \mathcal{U} ^m$ by: $$[\zeta, t] \to (\zeta, t + i |\zeta|^2) \in \partial \mathcal{U} ^m.$$ The Szeg\"o projector of ${\bf H}^m$ is the operator $\Pi: \mathcal{L}^2({\bf H}^m) \to \mathcal{H}^2({\bf H}^m)$ of orthogonal projection onto boundary values of holomorphic functions on $\mathcal{U} ^m$ which lie in $\mathcal{L}^2$. The kernel of $\Pi$ is given by (cf. \cite{Stein}, XII \S 2 (29)) $$\Pi(x,y) = K(y^{-1} x), \;\;\;\;\;\;\; K(x) = - C_m \frac{\partial}{\partial t} [t + i |\zeta|^2]^{-m} \in\mathcal{D}'(\H^m)\,.$$ The linear model for the principal $S^1$ bundle described in \S 1.2 is the so-called reduced Heisenberg group ${\bf H}^m_{\rm red}={\bf H}^m/ \{(0, 2\pi k): k \in {\mathbb Z}\} = {\mathbb C}^m \times S^1$ with group law $$(\zeta, e^{i t}) \cdot (\eta, e^{i s}) = (\zeta + \eta, e^{i(t + s + \Im (\zeta \cdot \bar{\eta}))}).$$ It is the principal $S^1$ bundle over ${\mathbb C}^m$ associated to the line bundle $L_\H={\mathbb C}^m\times{\mathbb C}$. The metric on $L_\H$ with curvature $\Theta=\partial\dbar |z|^2$ is given by setting $h_\H(z)=e^{-|z|^2}$; i.e., $|f|_{h_\H}= |f|e^{-|z|^2/2}$. The reduced Heisenberg group ${\bf H}^m_{\rm red}$ may be viewed as the boundary of the dual disc bundle $D \subset L^*_\H$ and hence is a strictly pseudoconvex CR manifold. It seems most natural to approach the analysis of the Szeg\"o kernels on ${\bf H}^m_{\rm red}$ from the representation-theoretic point of view. Let us begin with the case $N = 1$. We thus consider the space $\mathcal{V}_1 \subset \mathcal{L}^2({\bf H}^m_{\rm red})$ of functions $f$ satisfying $\frac{1}{i}\frac{\partial}{ \partial \theta} f = f$, which forms a (reducible) representation of ${\bf H}^m_{\rm red}$ with central character $e^{i \theta}$. By the Stone-von Neumann theorem there exists a unique (up to equivalence) representation $(V_1, \rho_1)$ with this character and by the Plancherel theorem, $\mathcal{V}_1 \cong V_1 \otimes V_1^*$. The space of CR functions in $\mathcal{V}_1$ is an irreducible invariant subspace. Here, by CR functions we mean the functions satisfying the left-invariant Cauchy-Riemann equations $\bar{Z}^L_q f = 0$ on ${\bf H}^m_{\rm red}$. Here, $\{\bar{Z}^L_q\}$ denotes a basis of the left-invariant anti-holomorphic vector fields on ${\bf H}^m_{\rm red}$. Let us recall their definition: we first equip ${\bf H}^m_{\rm red}$ with its left-invariant contact form $\alpha^L = \sum_q(u_qdv_q-v_qdu_q) + d\theta$ ($\zeta=u+iv$). The left-invariant CR holomorphic (resp.\ anti-holomorphic) vector fields $Z_q^L$ (resp.\ $ \bar{Z}_q^L$) are the horizontal lifts of the vector fields $\frac{\partial}{\partial z_q}$ (resp.\ $\frac{\partial}{\partial \bar{z}_q}$) with respect to $\alpha^L$. They span the left-invariant CR structure of ${\bf H}^m_{\rm red}$ and the $Z_q^L$ obviously have the form $Z_q^L = \frac{\partial}{\partial z_q} + A \frac{\partial}{\partial \theta}$ where the coefficient $A$ is determined by the condition $\alpha^L(Z_q^L ) = 0$. An easy calculation gives: $$Z_q^L = \frac{\partial}{\partial z_q} + \frac{i}{2}\bar{z}_q \frac{\partial}{\partial \theta}, \;\;\;\; \bar{Z}^L_q = \frac{\partial}{\partial \bar{z}_q} - \frac{i}{2} z_q \frac{\partial}{\partial \theta}.$$ The vector fields $\{\frac{\partial}{\partial \theta}, Z_q^L, \bar{Z}_q^L\}$ span the Lie algebra of ${\bf H}^m_{\rm red}$ and satisfy the canonical commutation relations above. We then define the Hardy space $\mathcal{H}^2 ({\bf H}^m_{\rm red})$ of CR holomorphic functions, i.e. solutions of $\bar{Z}_q^L f = 0$, which lie in $\mathcal{L}^2({\bf H}^m_{\rm red})$. We also put $\mathcal{H}^2_1 = \mathcal{V}_1 \cap \mathcal{H}^2({\bf H}^m_{\rm red}).$ The group ${\bf H}^m_{\rm red}$ acts by left translation on $\mathcal{H}^2_1$. The generators of this representation are the right-invariant vector fields $Z_q^R, \bar{Z}_q^R$ together with $\frac{\partial}{\partial \theta}$. They are horizontal with respect to the right-invariant contact form $\alpha^R = \sum_q(u_qdv_q-v_qdu_q) - d\theta$ and are given by: $$Z_q^R = \frac{\partial}{\partial z_q} - \frac{i}{2}\bar{z}_q \frac{\partial}{\partial \theta}, \;\;\;\; \bar{Z}^R_q = \frac{\partial}{\partial \bar{z}_q} + \frac{i}{2} z_q \frac{\partial}{\partial \theta}\,.$$ In physics terminology, $Z_q^R$ is known as an annihilation operator and $\bar{Z}_q^R$ is a creation operator. The representation $\mathcal{H}^2_1$ is irreducible and may be identified with the Bargmann-Fock space of entire holomorphic functions on ${\mathbb C}^m$ which are square integrable relative to $e^{-|z|^2}$ (or equivalently, holomorphic sections of the trivial line bundle $L_\H={\mathbb C}^m\times {\mathbb C}$ mentioned above, with hermitian metric $h_\H=e^{-|z|^2}$). The identification goes as follows: the function $\phi_0(z, \theta) := e^{i \theta} e^{-|z|^2/2}$ is CR holomorphic and is also the ground state for the right invariant ``annihilation operator;'' i.e., it satisfies $$\bar{Z}^L_q \phi_0(z, \theta) = 0 = Z_q^R \phi_0(z, \theta)\,.$$ Any element $F(z,\theta)$ of $\mathcal{ H}^2_1$ may be written in the form $F (z, \theta) = f(z) \phi_0.$ Then $\bar{Z}^L_q F =(\frac{\partial}{\partial \bar{z}_q} f ) \phi_0$, so that $F$ is CR if and only if $f$ is holomorphic. Moreover, $F \in \mathcal{L}^2({\bf H}^m_{\rm red})$ if and only if $f$ is square integrable relative to $e^{-|z|^2}$. The Szeg\"o kernel $\Pi_1^\H(z, \theta, w, \phi)$ of ${\bf H}^m_{\rm red}$ is by definition the orthogonal projection from $\mathcal{L}({\bf H}^m_{\rm red})$ to $H^2_1.$ As will be seen below, $\Pi_1^\H(z, \theta, w, \phi) = \frac{1}{\pi_m}e^{i (\theta -\phi )} e^{ (z\cdot\bar w -{\frac{1}{2}} |z|^2 -{\frac{1}{2}}|w|^2) }$, which is the left translate of $\phi_0$ by $(-w, - \phi)$. In the physics terminology it is the coherent state associated to the phase space point $w.$ So far we have set $N = 1$, but the story is very similar for any $N$. We define $\mathcal{H}^2_N$ as the space of square- integrable CR functions transforming by $e^{i N \theta}$ under the central $S^1$. By the Stone-von Neumann theorem there is a unique irreducible $V_N$ with this central character. The main difference to the case $N = 1$ is that $\mathcal{H}^2_N$ is of multiplicity $N^m$. The Szeg\"o kernel $\Pi_N^\H(x,y)$ is the orthogonal projection to $\mathcal{H}^2_N$ and is given by the dilate of $\Pi^\H_1$. Thus, $$\Pi_N^\H(x,y) =\frac{1}{\pi^m} N^m e^{i N (\theta -\phi )} e^{ N(z\cdot\bar w -{\frac{1}{2}} |z|^2 -{\frac{1}{2}}|w|^2) }.$$ To prove these formulae for the Szeg\"o kernels, we observe that the reduced Szeg\"o kernels are obtained by projecting the Szeg\"o kernel on ${\bf H}^m$ to ${\bf H}_{\rm red}^m$ as an automorphic kernel, i.e. $$\Pi^\H(x, y) = \sum_{n \in {\mathbb Z} } \Pi(x, y\cdot (0, 2\pi n)).$$ Let us write $x = (z, \theta), y = (w, \phi)$. Then the $N$-th Fourier component $\Pi^\H_N(x, y)$ of $\Pi^\H$, i.e. the projection onto square integrable holomorphic sections of $L^N$, is given by: \begin{eqnarray*}\Pi^\H_N(x, y)&=& \int_{{\mathbb R}} e^{- i N t} \Pi( e^{i t} x, y ) dt\ =\ \int_{{\mathbb R}} e^{- i N t} K(e^{ i t} y^{-1} x ) dt\\&=& \int_{{\mathbb R}} e^{- i N t} K(z - w, e^{i (\theta - \phi + t + \Im (z \cdot \bar{w})} ) dt\,.\end{eqnarray*} Here we abbreviated the element $(0, e^{it})$ by $e^{it}$. Change variables $t \mapsto t- \theta + \phi - \Im (z \cdot \bar{w})$ to get $$\begin{array}{l}\Pi^\H_N(x, y) = e^{i N (\theta - \phi)} e^{i N \Im (z \cdot \bar{w}}) \int_{{\mathbb R}} e^{- i N t} K(z - w, t) dt \\ \\ = e^{i N ( \theta - \phi)}e^{i N \Im (z \cdot \bar{w})} \hat{K}_t(z - w, N)\end{array} $$ where $\hat{K}_t$ is the Fourier transform of $K$ with respect to the $t$ variable. By \cite[p.~585]{Stein}, the full ${\mathbb R}^{2m} \times {\mathbb R}$ Fourier transform of $K$ is given by $\hat{K} (z, N) = C'_m e^{-|z|^2/ 2N}$, so by taking the inverse Fourier transform in the $z$ variable we get the Fourier transform just in the $t$ variable: \begin{equation}\label{PiHN} \Pi^\H_N(x, y) = \frac{1}{\pi^m} N^m e^{i N (\theta - \phi)}e^{ iN \Im (z \cdot \bar{w})} e^{- {\frac{1}{2}} N |z - w|^2}. \end{equation} (Our constant factor $\frac{1}{\pi^m}$ in (\ref{PiHN}) is determined by the condition that $\Pi_N^\H$ is an orthogonal projection.) In our study of the correlation functions, we will need explicit formulae for the horizontal derivatives of the Szeg\"o kernel. The left-invariant derivatives are given by \begin{eqnarray} N^{-m} Z^L_q \Pi_N^\H(z,\theta;w,\phi) &=&\, N (\bar{w}_q - \bar{z}_q) \Pi_N^\H(z,\theta;w,\phi)\,, \nonumber\\ \label{dPiNHleft} & & \\ N^{-m} Z^L_q \bar{W}^L_{ q'} \Pi_N^\H(z,\theta;w,\phi) &=& N^2 (z_{q'} - w_{q'})(\bar w_q - \bar{z}_q) \Pi_N^\H(z,\theta;w,\phi) + N\delta_{qq'} \Pi_N^\H(z,\theta;w,\phi)\,. \nonumber \end{eqnarray} Comparing the definitions of the horizontal vector fields with (\ref{horizontal}), using $h_\H=e^{-|z|^2}$, we see that $d^H_{z_q}=Z^L_q$, as expected, since $\alpha^L$ agrees with the contact form $\alpha$ for $L_\H$ (as defined in \S \ref{Szegokernels}). We will see later that our formulas for computing correlations are valid with any connection, and thus it is sometimes useful to also consider the right invariant derivatives: \begin{eqnarray} N^{-m} Z^R_q \Pi_N^\H(z,\theta;w,\phi) &=&\, N \bar{w}_q \Pi_N^\H(z,\theta;w,\phi)\,, \nonumber\\ \label{dPiNH} & & \\ N^{-m} Z^R_q \bar{W}^R_{ q'} \Pi_N^\H(z,\theta;w,\phi) &=& N^2 z_{q'}\bar w_q \Pi_N^\H(z,\theta;w,\phi) + N\delta_{qq'} \Pi_N^\H(z,\theta;w,\phi)\,. \nonumber \end{eqnarray} \begin{rem} Recall that the metric on $\mathcal{O}(N)\to\C\PP^m$ is given by $h^N(z)=(1+|z|^2)^{-N}$ using the coordinates and local frame from Example \ref{example1}. Since $$h^N(z/\sqrt{N})\to h_\H(z)\,,$$ the Heisenberg bundle can be regarded as the scaling limit of $\mathcal{O}(N)$. (Of course, in the same way $L_\H$ is the scaling limit of $L^N$, for any positive line bundle $L\to M$.) \end{rem} \bigskip \section{Correlation functions} \label{corfuns} This section begins with a generalization to arbitrary dimension and codimension a formula of \cite{H} and \cite{BD} for the ``correlation density function'' in the one-dimensional case. In fact, our formula (Theorem~\ref{density}) applies to a general class of probability spaces of $k$-tuples of (real or complex) functions. We then specialize to the case where the space of sections has a Gaussian measure. Finally, we show how the correlations of the zeros of $k$-tuples of sections of the $N$-th power of a holomorphic line bundle are given by a rational function in the Szeg\"o kernel $\Pi_N$ and its derivatives (Theorem \ref{npointcor}). \subsection {General formula for zero correlations} For our general setting, we let $(V,h)$ be a Hermitian holomorphic vector bundle on an $m$-dimensional Hermitian complex manifold $(M,g)$. (Here, we make no curvature assumptions.) Suppose that $\mathcal{S}$ is a finite dimensional subspace of the space $H^0(M,V)$ of global holomorphic sections of $V$, and let $d\mu$ be a probability measure on $\mathcal{S}$ given by a semi-positive $\mathcal{C}^0$ volume form that is strictly positive in a neighborhood of $0\in\mathcal{S}$. (We shall later apply our results to the case where $V=L^N\oplus\cdots\oplus L^N$, for a holomorphic line bundle $L$ over a compact complex manifold $M$, and $\mathcal{S}=H^0(M,V)$ with a Gaussian measure $d\mu$. Our formulation involving general vector bundles allows us to reduce the study of $n$-point correlations to the case $n=1$, i.e., to expected densities of zeros.) As in the introduction, we introduce the punctured product $$M_n=\{(z^1,\dots,z^n)\in \underbrace{M\times\cdots\times M}_n: z^p\ne z^q \ \ {\rm for} \ p\ne q\}\,,$$ and we write $$s(z)=(s(z^1),\dots,s(z^n))\,,\ \nabla s(z)=(\nabla s(z^1),\dots,\nabla s(z^n))\,,\quad z=(z^1,\dots,z^n)\in M_n\,,$$ where $\nabla s(\zeta)\in T^*_\zeta\otimes V_\zeta$ is the covariant derivative with respect to the Hermitian connection on $V$. We define the map $$\mathcal{J}:M_n \times\mathcal{S}\to \big[({\mathbb C}\oplus T^*_M)\otimes V\big]^n\,,\quad \mathcal{J} (z,s)=(s(z),\nabla s(z))\,;$$ i.e., $\mathcal{J} (z,s)$ is the 1-jet of $s$ at $z\in M_n$. We write $g=\Re\sum g_{qq'}dz_q\otimes d\bar z_{q'},h_{jj'}=h(e_j,e_{j,})$, where $\{z_1,\dots,z_m \}$ are local coordinates in $M$ and $\{e_1,\dots,e_k \}$ is a local frame in $V$ ($m=\dim M,\ k={\rm rank}\,V$). We let $G=\det (g_{qq'})$, $H=\det (h_{jj'})$. We let $$d\zeta=\frac{1}{m!}\omega^m_{\zeta}=G(\zeta)\prod_{j=1}^m d\Re \zeta_j d\Im \zeta_j\,,\quad \zeta\in M$$ denote Riemannian volume in $M$, and we write \begin{equation}\label{dbda} x^p=\sum_j b^p_j e_j(z^p),\quad dx^p=H(z^p) \prod_{j} d\Re b^p_j d\Im b^p_j \quad x^p\in V_{z^p}\,,\end{equation} $$\xi^p=\sum_{j,q} a^p_{jq} dz_q\otimes e_j|_{z^p}, \quad d\xi^p= G(z^p)^{-k}H(z^p)^{m} \prod_{j,q} d\Re a^p_{jq} d\Im a^p_{jq} \quad \xi_j\in (T^*_M\otimes V)_{z^p}\,.$$ The quantities $dx^p,\; d\xi^p$ are the intrinsic volume measures on $V_{z^p}$ and $(T^*_M\otimes V)_{z^p}$, respectively, induced by the metrics $g,h$. \begin{defin} Suppose that $\mathcal{J} $ is surjective. We define the {\it $n$-point density\/} $D_n(x,\xi,z)dxd\xi dz$ of $\mu$ by \begin{equation}\label{jointdist} \mathcal{J} _*(dz\times d\mu)=D_n(x,\xi,z)dxd\xi dz\,,\quad x=(x^1,\dots,x^n)\in V_{z^1}\times\dots\times V_{z^n}\,,\end{equation} $$\xi=(\xi^1,\dots,\xi^n)\in (T^*_M\otimes V)_{z^1}\times\dots\times(T^*_M\otimes V)_{z^n}\,,\ z=(z^1,\dots,z^n)\in M_n\,,$$ $$dx=dx^1\cdots dx^n\,,\quad d\xi=d\xi^1\cdots d\xi^n\,,\quad dz=dz^1\cdots dz^n\,.$$ In this case, for each $z\in M_n$, the (vector-valued) random variable $(s(z),\nabla s(z))$ has (joint) probability distribution $D_n(x,\xi,z)dxd\xi$.\end{defin} \begin{rem} If we let $n=1$ and fix a point $z\in M$, then the measure $D(x, \xi, z) dx d\xi$ is intrinsically defined as a measure on the space $J^1_z(M, V)$ of 1-jets of sections of $V$ at $z$. Taking a section to its $1$-jet at $z$ defines a map $\mathcal{J}_z: \mathcal{S} \to J^1_z(M,V)$ and hence induces a measure $\mathcal{J}_{z*} \mu$ on $J^1_z(M,V)$ independently of any choices of connections, coordinates or metrics. Similarly for $n>1$, $D(x, \xi, z) dx d\xi$ is an intrinsic measure on $\prod _{p=1}^n J^1_{z_p}(M, V)$. \end{rem} For a vector-valued 1-form $\xi\in T^*_{M,z}\otimes V_{z}= {\rm Hom}(T_{M,z},V_z)$, we let $\xi^*\in {\rm Hom}(V_{z},T_{M,z})$ denote the adjoint to $\xi$ (i.e., $\langle\xi^*v,t \rangle=\langle v,\xi t\rangle\,$), and we consider the endomorphism $\xi\xi^*\in {\rm Hom}(V_{z},V_{z})$. In terms of local frames, if $$\xi=\sum_j \xi_j \otimes e_j=\sum_{j,q}a_{jq} dz_j\otimes e_j\,,$$ then $$\xi^*=\sum_{j,q}\alpha_{jq}\frac{\d}{\d z_q}\otimes e_j^*\,,\qquad \alpha_{jq}=\sum_{j',q'} h_{jj'}\gamma_{q'q}\bar a_{j'q'}\,,$$ where $\big(\gamma_{qq'}\big)=\big(g_{qq'}\big)^{-1}$; hence we have \begin{equation}\label{endo} \xi\xi^*=\sum_{j,j',j'',q,q'} h_{j'j''}a_{jq}\gamma_{q'q}\bar a_{j''q'}\, e_j\otimes e_{j'}^* \,.\end{equation} Its determinant is given by \begin{equation}\label{detendo}\det(\xi\xi^*)=H\det\left(\sum_{q,q'} a_{jq}\gamma_{q'q}\bar a_{j'q'}\right)_{1\le j,j'\le k}=H\det \langle\xi_j, \xi_{j'}\rangle= H\|\xi_1\wedge\dots\wedge \xi_k\|^2\,.\end{equation} \begin{rem} The measure $\det(\xi \xi^*) D(0, \xi, z) d\xi dz$ will play a fundamental role in our study of correlation functions. We observe here that it depends only on the metric $\omega$ on $M$, and in the case where the zero sets are points ($k=m$), it is independent of the choice of metric on $M$ as well. Indeed, as mentioned in the previous remark, $D(x,\xi, z) dx d\xi$ is well-defined on $J^1_z(M,V)$. The conditional density $D(0, \xi, z) d\xi$ equals $\mathcal{J}_{z*} \mu /dx |_{x = 0}$ and thus depends only on the choice of volume forms $dx^p$ on $V_{z^p}$. Since $dz/dx$ transforms in the opposite way to $\det \xi \xi^*$ it follows that $\det(\xi \xi^*) D(0, \xi, z) d\xi dz$ is an invariantly defined measure on $(T^*_M\otimes V)^n$. \end{rem} Recall that for $s\in\mathcal{S}$ so that ${\operatorname{codim\,}} Z_s=k$, we let $|Z_s|$ denote Riemannian $(2m-2k)$-volume along the regular points of $Z_s$, regarded as a measure on $M$. \begin{defin} For $s\in\mathcal{S}$ so that ${\operatorname{codim\,}} Z_s=k$, we consider the product measure on $M_n$, $$|Z_s|^n=\big(\underbrace{|Z_s|\times\cdots\times|Z_s|}_{n}\big)\,.$$ Its expectation ${\mathbf E}\, |Z_s|^n$ is called the {\it $n$-point zero correlation measure.} \end{defin} We shall use the following general formula to compute the correlations of zeros and to show universality of the scaling limit: \begin{theo}\label{density} Let $M,V,\mathcal{S},d\mu$ be as above, and suppose that $\mathcal{J} $ is surjective and the volumes $|Z_s|$ are locally uniformly bounded above. Then \begin{equation}\label{a8} {\mathbf E}\,|Z_s|^n =K_n(z)dz \,,\quad K_n(z)= \int d\xi\,D_n(0,\xi;z)\prod_{p=1}^n \det \left(\xi^{p}\xi^{p*}\right)\,.\end{equation} \end{theo} The function $K_{n}(z^1,\dots,z^n)$, which is continuous on $M_n$ is called the {\it $n$-point zero correlation function\/}. For $k< m$, (\ref{a8}) holds on all of the $n$-fold product $M\times\cdots\times M$, including the diagonal, and $K_n$ is locally integrable on $M\times\cdots\times M$ (and is infinite on the diagonal). In the case $k=m$, when the zero sets are discrete, the zero correlation measure on $M\times\cdots\times M$ is the sum of the absolutely continuous measure $K_n(z)dz$ plus a measure supported on the diagonal. \medskip\noindent{\em Proof of Theorem \ref{density}:\/} Consider the Hermitian vector bundle $V_n=\bigoplus_{p=1}^n \pi_p^*V\longrightarrow M_n$, where $\pi_p:M_n\to M$ denotes the projection onto the $p$-th factor. By replacing $V\to M$ with $V_n\to M_n$ and $s\in H^0(M,V)$ with $$\tilde s(z^1,\dots,z^n)=(s(z^1),\dots,s(z^n))\in H^0(M_n,V_n)\,,$$ and noting that $T_{M_n,z}=\prod_p T_{M,z^p}$ and $|Z_s|^n=|Z_{\tilde s}|$, we can assume without loss of generality that $n=1$. It follows from the above remarks that $D(0,\xi;z)$ does not depend on the choice of connection on $V$. We can also verify this in terms of local coordinates: write $s=\sum b_j e_j$, $\nabla s = \sum a_{jq} dz_q \otimes e_j$ as in (\ref{jointdist}); we have $a_{jq}=\frac{\d b_j}{\d z_q} + \sum_k b_k\theta^k_{jq}$. Then if we write $a^0_{jq}=\frac{\d b_j}{\d z_q}$, we have $$\frac{\d(a_{jq},b_j)}{\d(a^0_{jq},b_j)} = 1\,.$$ Hence $D(0,\xi;z)$ is unchanged if we substitute the (local) flat connection given by $a^0_{jq}$. We now restrict to a coordinate neighborhood $U\subset M$ where $V$ has a local frame $\{e_j\}$. By hypothesis, we can suppose that the $e_j$ are restrictions of sections in $\mathcal{S}$. We write $s=\sum s_j e_j$, and by the above we may assume that $\nabla s = \sum ds_j \otimes e_j$. We use the notation $$|\!|\!|\xi|\!|\!|=\sqrt{\det(\xi\xi^*)}\,, \quad {\rm for} \ \xi\in T^*_{M,z}\otimes V_{z}= {\rm Hom}(T_{M,z},V_z)\,.$$ Then by (\ref{detendo}), $$|\!|\!|\nabla s|\!|\!|^2 =H\| d s_1\wedge\cdots \wedge d s_k\|^2=\|\Psi\|\,,$$ where $\Psi$ is the $(k,k)$-form on $U$ given by: $$\Psi=H \left(\frac{i}{2} ds_1\wedge \overline{d s_1}\right)\wedge\cdots\wedge \left(\frac{i}{2} ds_k\wedge \overline{d s_k}\right)\,.$$ Thus, by the Leray formula, \begin{equation}\label{Leray} |Z_s|=\|ds_1\wedge\cdots\wedge ds_k\|^2 \left.\frac{dz }{ \frac{i}{2}ds_1\wedge d\bar s_1\cdots\wedge\frac{i}{2}ds_k\wedge d\bar s_k}\right|_{Z_s}=|\!|\!|\nabla s|\!|\!|^2 \left. \frac{dz}{\Psi}\right|_{Z_s}\,,\end{equation} Define the measure $\lambda$ on $M\times\mathcal{S}$ by \begin{equation}\label{lambda} (\lambda,\phi)=\int_\mathcal{S} \left(|Z_s|,\phi(z,s)\right)d\mu(s)\,.\end{equation} Then $$\pi_*\lambda={\mathbf E}\,|Z_s|^n\,,$$ where $\pi:M\times\mathcal{S}\to M$ is the projection. Hence, \begin{equation}\label{Leray2}\lambda=\int_\mathcal{S} d\mu(s)\; |Z_s| =\int_\mathcal{S} d\mu(s)\left.\left(|\!|\!| \nabla s|\!|\!|^2\frac{dz}{ \Psi}\right)\right|_{Z_s}\,. \end{equation} For (almost all) $x\in{\mathbb C}^{k}$, let $I(s,x)$ be the measure on $U$ given by $$(I(s,x),\phi)=\int_{s(z)=\sum x_j e_j(z)}\phi(z) d{\operatorname{Vol}}_{(2m-2k)n}(z)= \int_{s(z)=\sum x_j e_j(z)} |\!|\!| \nabla s|\!|\!|^2\frac{dz}{ \Psi}\phi(z)\,,\ \phi\in\mathcal{C}^0(U) \,,$$ where the second equality is by (\ref{Leray}) applied to $s-\sum x_j e_j(z)$. Then \begin{equation}\label{I} \int I(s,x) dx=|\!|\!| \nabla s(z)|\!|\!|^2 dz\,.\end{equation} Now let $\lambda_x$ be the measure on $U$ given by $$(\lambda_x,\phi)=\int_\mathcal{S} (I(s,x),\phi)d\mu(s) \,.$$ \begin{claim} The map $x\mapsto (\lambda_x,\phi)$ is continuous.\end{claim} \noindent To prove this claim, we first note that the hypothesis that $|Z_s|$ is locally uniformly bounded implies that $(I(s,x),\phi)\le C<+\infty$ uniformly in $s,x$. Thus we can assume without loss of generality that $\mu$ has compact support in $\mathcal{S}$. By hypothesis, the map $$\sigma:U\times \mathcal{S}\to {\mathbb C}^k,\qquad \sigma(z,s)= (s_1(z),\dots,s_k(z))$$ is a submersion. We can now write $\lambda_x$ as a fiber integral of a compactly supported $\mathcal{C}^0$ form: $$\lambda_x=\frac{1}{(m-k)!}\int_{\sigma^{-1}(x)}\phi(z)\omega^{m-k}(z)\wedge d\mu(s)\,,$$ and thus $\lambda_x$ is continuous, verifying the claim. \medskip We note that $\lambda_0=\lambda|_U$. Hence, to complete the proof, we must show that $$\pi_* \lambda_0=K_1(z)dz|_U\,.$$ By (\ref{jointdist}) and (\ref{I}), for a test function $\phi(x,\xi,z)$, \begin{eqnarray*} \int \phi(x,\xi,z)|\!|\!|\xi|\!|\!|^2 D_1(x,\xi,z)dxd\xi dz &=& \int d\mu(s)\int \phi(\mathcal{J} (z,s)) |\!|\!| \nabla s(z)|\!|\!|^2dz\\ &=&\int dx\int (I(s,x),{\phi\circ \mathcal{J} }) d\mu(s)\\ &=&\int (\lambda_x,\phi\circ \mathcal{J} ) dx\,.\end{eqnarray*} By choosing $\phi(x,\xi,z)=\rho_\varepsilon(x)\psi(z)$, where $\rho_\varepsilon$ is an approximate identity, and letting $\varepsilon\to 0$, we conclude that $$\int \psi(z)K_1(z)dz=\int \psi(z)|\!|\!|\xi|\!|\!|^2 D_1(0,\xi,z)d\xi dz =(\lambda_0,\psi(z))\,.$$ \qed\medskip We note the following analogous formula for real manifolds: \begin{theo}\label{density-real} Let $V$ be a $\mathcal{C}^\infty$ real vector bundle over a $\mathcal{C}^\infty$ Riemannian manifold $M$, and let $\mu$ be a probability measure on a finite dimensional vector space $\mathcal{S}$ of $\mathcal{C}^\infty$ sections of $V$ given by a semi-positive volume form that is strictly positive at $0$. Suppose that the volumes $|Z_s|$ are locally uniformly bounded above. Let $D_n(x,\xi,z)dxd\xi dz$ denote the $n$-point density of $\mu$. Then \begin{equation}\label{a8-real} {\mathbf E}\,|Z_s|^n =K_n(z)dz \,,\quad K_n(z)= \int d\xi\,D_k(0,\xi,z)\prod_{p=1}^n \sqrt{\det(\xi^{p}\xi^{p*})}\,. \end{equation} \end{theo} The proof is similar to that of Theorem \ref{density}, except that (\ref{Leray}) is replaced by the Leray formula \begin{equation}\label{Leray-real} |Z_s|=\|ds_1\wedge\cdots\wedge ds_k\| \left.\frac{d\zeta }{ ds_1\wedge\cdots\wedge ds_k}\right|_{Z_s}\end{equation} in the real case. \subsection{Formula for Gaussian densities}\label{formula-gaussian} We now specialize our formula from Theorem \ref{density} to the case where $\mu$ is a Gaussian measure. Fix $z=(z^1,\dots,z^n)\in M_n$ and choose local coordinates $\{z^p_q\}$ and local frames $\{e^p_j\}$ near $z^p$, $p=1,\dots,n$. We consider the random variables $b^p_j,\ a^p_{jq}$ given by \begin{equation}\label{fg1} s(z^p)=\sum_{j=1}^k b^p_j e^p_j,\quad \nabla s(z^p)=\sum_{j=1}^k\sum_{q=1}^m a^p_{jq}dz^p_q\otimes e^p_j,\qquad p=1,\dots,n.\end{equation} By (\ref{gaussian})--(\ref{gaussian2}) and (\ref{dbda})--(\ref{jointdist}) the $n$-point density $$D_n(x,\xi,z)dxd\xi dz=D_n\left[\prod_{p=1}^nG(z^p)^{-k}H(z^p)^{m}\right] dbdadz$$ is given by: \begin{equation}\label{fg2}D_{n}(b,a;z)= \frac{\exp\langle-\Delta_{n}^{-1}v, v\rangle}{\pi^{kn(1+m)}\det\Delta_{n}}\;,\qquad v=\pmatrix b\\ a\endpmatrix\,, \end{equation} where \begin{equation}\label{fg3} \Delta_{n}=\left( \begin{array}{cc} A_{n} & B_{n} \\ B^*_{n} & C_{n} \end{array}\right)\end{equation} $$A_{n}=\big(A^{jp}_{j'p'}\big)= \big (\,{\mathbf E}\, b^p_j \bar b^{p'}_{j'}\,\big),\quad B_{n}=\big(B^{jp}_{j'p'q'}\big)= \big (\,{\mathbf E}\, b^p_j \bar a^{p'}_{j'q'}\,\big),\quad C_{n}=\big(C^{jpq}_{j'p'q'}\big)= \big(\,{\mathbf E}\, a^{p}_{jq}\bar a^{p'}_{j'q'}\,\big);$$ $$j,j'=1,\dots,k;\quad p,p'=1,\dots,n; \quad q,q'=1,\dots,m. $$ (We note that $A_n,\ B_n,\ C_n$ are $kn\times kn,\ kn\times knm,\ knm\times knm$ matrices, respectively; $j,p,q$ index the rows, and $j',p',q'$ index the columns.) The function $D_{n}(0,a;z)$ is a Gaussian function, but it is not normalized as a probability density. It can be represented as \begin{equation}\label{fg4} D_{n}(0,a;z)=Z_{n}(z) D_{\Lambda_n}(a;z), \end{equation} where \begin{equation}\label{fg5} D_{\Lambda_n}(a;z)=\frac{1}{\pi^{knm}\det\Lambda_{n}}\exp\left( -{\langle\Lambda^{-1}_{n}a, a\rangle}\right) \end{equation} is the Gaussian density with covariance matrix \begin{equation}\label{fg6} \Lambda_{n}=C_{n}-B^*_{n}A^{-1}_{n}B_{n} =\left(C^{jpq}_{j'p'q'} -\sum_{j_1,p_1,j_2,p_2}\bar B_{jpq}^{j_1p_1}\Gamma^{j_1p_1}_{j_2p_2}B^{j_2p_2}_{j'p'q'}\right) \qquad (\Gamma=A_n^{-1}) \end{equation} and \begin{equation}\label{fg7} Z_{n}(z)=\frac{\det\Lambda_{n}}{\pi^{kn}\det\Delta_{n}} =\frac{1}{\pi^{kn}\det A_{n}}\,. \end{equation} This reduces formula (\ref{a8}) to \begin{equation}\label{fg8} K_{n}(z)=\frac{1}{\pi^{kn}\det A_{n}}\left\langle \prod_{p=1}^n \det\left(a^{p*}\gamma^pa^p\right)\right\rangle_{\Lambda_{n}} \end{equation} where $\langle\cdot\rangle_{\Lambda_{n}}$ stands for averaging with respect to the Gaussian density $D_{\Lambda_n}(a;z)$, and $(\gamma^p_{qq'})= (g^p_{qq'})^{-1}$, $g^p_{qq'}=g_{qq'}(z^p)$. \subsection{Densities and the Szeg\"o kernel}\label{D-meet-S} We return to our positive Hermitian line bundle $(L,h)$ on a compact complex manifold $M$ with K\"ahler\ form $\omega=\frac{i}{2}\Theta_h$. We now apply formulas (\ref{fg3})--(\ref{fg8}) to the vector bundle $$V=\underbrace{L^N\oplus\cdots\oplus L^N}_k$$ and space of sections $$\mathcal{S}=H^0(M,V)=H^0(M,L^N)^k$$ with the Gaussian measure $\mu=\nu_N\times\cdots\times\nu_N$, where $\nu_N$ is the standard Gaussian measure on $H^0(M,L^N)$ given by (\ref{gaussian}). We denote the resulting $n$-point density by $D_{nk}^N$, and we also write $\Delta_n=\Delta_{nk}^N,\ A_n =A_{nk}^N$, etc. As above, we fix $z=(z^1,\dots,z^n)\in M_n$ and choose local coordinates $\{z^p_q\}$ near $z^p$, $p=1,\dots,n$. We also choose local frames $\{e^p_L\}$ for $L$ near the points $z^p$ so that $$\|e^p_L(z^p)\|_h=1\,.$$ For $s\in\mathcal{S}$, we write \begin{equation}\label{dms1} s(z^p)= \left( \begin{array}{c} s_1(z^p)\\ \vdots\\ s_k(z^p) \end{array} \right)= \left( \begin{array}{c} b^p_1\\ \vdots\\ b^p_k \end{array} \right) (e^p_L(z^p))^{\otimes N} \,,\end{equation} \begin{equation}\label{dms2} \nabla_N s_j(z^p) =\sum_{q=1}^m a^p_{jq}dz^p_q\otimes (e^p_L(z^p))^{\otimes N}\,.\end{equation} Since the $s_j$ are independent and have identical distributions, we have \begin{equation}\label{dms3} A_{nk}^N=\big(A^{jp}_{j'p'}\big)= \big(\delta_{jj'}{\mathbf E}\,(b^p_1 \bar b^{p'}_1)\big)\,,\quad B_{nk}^N=\big(B^{jp}_{j'p'q'}\big)= \big(\delta_{jj'}{\mathbf E}\,(b^p_1 \bar a^{p'}_{1q'})\big)\,, \end{equation} $$C_{nk}^N=\big(C^{jpq}_{j'p'q'}\big)= \big(\delta_{jj'}{\mathbf E}\,(a^p_{1q} \bar a^{p'}_{1q'})\big)\,.$$ We write $$s_1=\sum_{\alpha=1}^{d_N}c_\alpha S^N_\alpha=\left(\sum_{\alpha=1}^{d_N}c_\alpha f^p_\alpha\right)(e^p_L)^{\otimes N}\,,$$ where $\{S^N_\alpha\}$ is an orthonormal basis for $H^0(M,L^N)$. Using the local coordinates $(z^p,\theta)$ in $X$ as described in \S \ref{cxgeom}, we have by (\ref{dms3}) and (\ref{PiN}) (noting that $h(z^p)=0$ by the above choice of local frames), \begin{equation}\label{dms4}A^{jp}_{j'p'} =\delta_{jj'}\sum_{\alpha,\beta=1}^{d_N}{\mathbf E}\,(c_\alpha \bar c_\beta)f^p_\alpha(z^p) \overline{f^{p'}_\beta(z^{p'})}=\delta_{jj'} \sum_{\alpha=1}^{d_N}f^p_\alpha(z^p) \overline{f^{p'}_\alpha(z^{p'})}=\delta_{jj'} \Pi_N(z^p,0;z^{p'},0)\,.\end{equation} Similarly, \begin{equation}\label{dms5} B^{jp}_{j'p'q'}=\delta_{jj'} \sum_{\alpha=1}^{d_N}f^p_\alpha(z^p) \overline{f^{p'}_{\alpha;q'}(z^{p'})}=\delta_{jj'} d^H_{\bar w_{q'}}\Pi_N(z^p,0;z^{p'},0)\,,\end{equation} \begin{equation}\label{dms6} C^{jpq}_{j'p'q'}=\delta_{jj'} \sum_{\alpha=1}^{d_N}f^p_{\alpha;q}(z^p) \overline{f^{p'}_{\alpha;q'}(z^{p'})}=\delta_{jj'} d^H_{z_{q}}d^H_{\bar w_{q'}}\Pi_N(z^p,0;z^{p'},0)\,.\end{equation} \begin{lem}\label{detN} There is a positive integer $N_0=N_0(M,n)$ such that $$\det\big(\Pi_N(z^p,0;z^{p'},0)\big)_{1\le p,p'\le n} \ne 0\,,$$ for distinct points $z^1,\dots,z^n$ of $M$ and for all $N\ge N_0$. \end{lem} \begin{proof} It is a well-known consequence of the Kodaira Vanishing Theorem (see for example, \cite{GH}) that we can find $N_0$ such that if $N\ge N_0$ and $x_1,\dots,x_n\in M$ with $x_p\ne x_1$ for $2\le p\le n$, then there is a section $s\in H^0(M,L^N)$ with $s(x_1)\ne 0$ and $s(x_p)=0$ for $2\le p\le n$. We write $\widetilde A_{pp'}=\Pi_N(z^p,0;z^{p'},0)$. Suppose on the contrary that $\det (\widetilde A_{pp'})=0$, and chose a nonzero vector $v=(v_1,\dots,v_n)$ such that $\sum_p v_p\widetilde A_{pp'}=0$. Then recalling (\ref{szego}), we have \begin{equation}\label{det1} 0=\sum_{p,p'} v_p \widetilde A_{pp'}\bar v_{p'} = \sum_{p,p',\alpha} v_p \widehat S^N_\alpha(z^p,0) \overline{\widehat S^{N}_\alpha(z^{p'},0) v_{p'}} =\sum_{\alpha=1}^{d_N} |x_\alpha|^2\,,\end{equation} where $x_\alpha=\sum_p v_p\widehat S^N_\alpha(z^p,0)$. Since the $S^N_\alpha$ span $H^0(M,L^N)$, it follows that for all $s\in H^0(M,L^N)$, we have $\sum_p v_p \hat s(z^p) =0$. But this contradicts the fact that, choosing $p_0$ with $v_{p_0}\ne 0$, we can find a section $s\in H^0(M,L^N)$ with $s(z^{p_0})\ne 0$ and $s(z^p)=0$ for $p\ne p_0$. \end{proof} Thus we see that the $n$-point correlation functions depend only on the Szeg\"o kernel, as follows: \begin{theo}\label{npointcor} Let $(L,h)$ be a positive Hermitian line bundle on an $m$-dimensional compact complex manifold $M$ with K\"ahler\ form $\omega=\frac{i}{2}\Theta_h$, let $\mathcal{S}=H^0(M,L^N)^k$ ($k\ge 1$), and give $\mathcal{S}$ the standard Gaussian measure $\mu$ described above. Let $n\ge 1$ and suppose that $N$ is sufficiently large so that $\mathcal{J}$ is surjective. Let $z=(z^1,\dots,z^n)\in M_n$ and choose local coordinates $(\zeta_1,\dots,\zeta_m)$ at each point $z^p$ such that $\Theta_h(z^p)=\sum_q d\zeta_q\wedge d\bar\zeta_q$, $1\le p\le n$. Then the $n$-point correlation $K_{nk}^N(z)$ is given by a universal rational function, homogeneous of degree $0$, in the values of $\Pi_N$ and its first and second derivatives at the points $(z^p,z^{p'})$. Specifically, \begin{equation}\label{npointcoreq} K_{nk}^N(z)= \frac{\mathcal{P}_{nkm}\big(\Pi_N(z^p,z^{p'}),d^H_{\bar w_{q}}\Pi_N(z^p,z^{p'}), d^H_{z_{q}}\Pi_N(z^p,z^{p'}), d^H_{z_{q}}d^H_{\bar w_{q'}}\Pi_N(z^p,z^{p'})\big)}{\pi^{kn}\left[\det \big(\Pi_N(z^p,z^{p'})\big)_{1\le p,p'\le n}\right]^{k(n+1)}}\end{equation} ($1\le p,p'\le n,\ 1\le q,q'\le m$), where $\mathcal{P}_{nkm}$ is a universal homogeneous polynomial of degree $kn(n+1)$ with integer coefficients depending only on $n,k,m$. \end{theo} \begin{proof} The $n$-point zero correlation $K_{nk}^N(z)$ is given by equation (\ref{fg8}) with $\gamma^p_{qq'}=\delta_{qq'}$. By the Wick formula (\cite[(I.13)]{Si}), the expectation $$\left\langle \prod_{p=1}^n \det\left(a^{p*}a^p\right)\right\rangle_{\Lambda_{n}}$$ in (\ref{fg8}) is a homogeneous polynomial (over ${\mathbb Z}$) of degree $kn$ in the coefficients of $\Lambda_{n}$. By (\ref{fg6}) and (\ref{dms4}), the coefficients of $\det \big(\Pi_N(z^p,z^{p'})\big)\Lambda_n$ are homogeneous polynomials of degree $n+1$ in the coefficients of $A_n,B_n,C_n$. The conclusion then follows from (\ref{dms4})--(\ref{dms6}). \end{proof} \begin{rem} In the statement of Theorem \ref{npointcor}, we wrote $\Pi_N(z,w)$ for $\Pi_N(z,\theta;w,\phi)$. Since the expression is homogeneous of degree 0, it is independent of $\theta$ and $\phi$. Alternately, we could regard $\Pi_N(z,w)$ as functions on $M\times M$ having values in $L_z\otimes\overline{L_w}$ (replacing the horizontal derivatives with the corresponding covariant derivatives); again the degree 0 homogeneity makes the expression a scalar. Furthermore, since Theorem \ref{density} is valid for all connections, we can replace the horizontal derivatives in (\ref{npointcoreq}) with the derivatives with respect to an arbitrary connection.\end{rem} \subsection {Zero correlation for ${\operatorname{SU}}(m+1)$-polynomials} \label{zcorpoly} In this section, we use our methods to describe the zero correlation functions for ${\operatorname{SU}}(m+1)$-polynomials. We do not carry out the computations in complete detail, since we are primarily interested in the scaling limits, which we shall compute in \S \ref{formulas}. The ${\operatorname{SU}}(m+1)$-polynomials are random homogeneous polynomials of degree $N>0$ on ${\mathbb C}^{m+1}$, \begin{equation}\label{a1} s(z)=s(z_0,z_1,\dots,z_m)=\sum_{|J|=N} \sqrt{N!/J!}\,{c_J}z^J,\quad z^J=z_0^{j_0}\cdots z_m^{j_m},\quad J!=j_0!\cdots j_m!, \end{equation} where the coefficients $c_J$ are complex independent Gaussian random variables with mean 0 and variance 1: \begin{equation}\label{a2} {\mathbf E}\, c_J=0;\qquad {\mathbf E}\, c_J\overline{c_K}= \delta_{JK},\quad \delta_{JK}=\delta_{j_0k_0}\dots\delta_{j_mk_m}; \qquad {\mathbf E}\, c_Jc_K=0. \end{equation} Then $s(z)$ is a Gaussian random polynomial on ${\mathbb C}^{N+1}$ with first and second moments given by \begin{equation}\label{a3} {\mathbf E}\, s(z)=0;\qquad {\mathbf E}\, s(z)\overline{s(w)}=\langle z, w \rangle^N =\left(\sum_{q=0}^m z_q\overline{w_q}\right)^N; \qquad {\mathbf E}\, s(z)s(w)=0. \end{equation} This implies that the probability distribution of $s(z)$ is invariant with respect to the map $s(z)\to s(Uz)$ for all $U\in {\operatorname{SU}}(m+1)$. Let $(\mathcal{S}_N,\mu_N)$ denote the Gaussian probability space of independent $k$-tuples ($k\le m$) of ${\operatorname{SU}}(m+1)$-polynomials of degree $N$. For $s=(s_1,\dots,s_k)\in \mathcal{S}_N$, the zero set \[ Z_s=\{z: s_1(z)=\dots=s_k(z)=0\}\,. \] is an algebraic variety in the complex projective space $\C\PP^m$. We will assume that $\C\PP^m$ is supplied with the Fubini-Study Hermitian metric $\omega$, which is ${\operatorname{SU}}(m+1)$-invariant. In the affine coordinates $z=(1,z_1,\dots,z_m)$, \begin{equation}\label{a3a} \omega=\frac{\sqrt{-1}}{ 2}\partial\dbar\log\left(1+\sum|z_q|^2\right)=\frac {\sqrt{-1}}{ 2}\left[\frac{\sum dz_q\wedge\overline{dz_q}}{1+\sum |z_q|^2} -\frac{\left(\sum \overline{z_q}dz_q\right)\wedge \left(\sum z_q\overline{dz_q}\right)} {\left(1+\sum |z_q|^2\right)^2}\right]\,; \end{equation} i.e., \begin{equation}\label{FSg} \omega= \frac{\sqrt{-1}}{ 2}\sum g_{qq'}dz_q\wedge dz_{q'}\,, \quad g_{qq'}=\frac{(1+|z|^2)\delta_{qq'}-\bar z_qz_{q'}}{(1+|z|^2)^2}\,.\end{equation} To simplify our computations, we consider only points $z^p$ with finite affine coordinates, $z^p=(1,z^p_1,\dots,z^p_m),\; p=1,\dots,n$, and we regard the ${\operatorname{SU}}(m+1)$-polynomials $s_j$ as polynomials of degree $\le N$ on ${\mathbb C}^m$; i.e., we regard the $s_j$ as sections of the trivial line bundle on ${\mathbb C}^m$ with the flat metric $h=1$ (so that the covariant derivatives coincide with the usual derivatives of functions). As above, we consider the random variables $$b^p_j=s_j(z^p),\; a^p_{jq}=\frac{\d s_j }{ \d z_q}(z^p)\,,$$ and we denote their joint distribution by \begin{equation}\label{a11a} \begin{array}{rl} D_{nk}^N(b,a;z)db\,da,\quad & b=\left(b^1,\dots,b^n\right),\quad b^p= (b^p_j)_{j=1,\dots,k};\\ &\\ & a=\left(a^1,\dots,a^n\right),\quad a^p=\left(a^p_{jq}\right)_{j=1,\dots,k;\;q=1,\dots,m}\,. \end{array} \end{equation} (Here, the $n$-point density is with respect to Lebesgue measure $db=\prod db^p_j d\bar b^p_j,\; da=\prod da^p_{jq}d\bar a^p_{jq}$.) We assume that $N>nm$ to ensure that $\mu_N$ possesses a continuous $n$-point density. Since $\mu_N$ is Gaussian, the density $D_{nk}^N(b,a;z)$ is Gaussian as well, and it is described by the covariance matrix \begin{equation}\label{a14} \Delta_{nk}^N=\left( \begin{array}{cc} A_{nk}^N & B_{nk}^N \\ B_{nk}^{N*} & C_{nk}^N \end{array}\right) \end{equation} where \begin{equation}\label{a14a} \begin{array}{l} A_{nk}^N=\left (\,{\mathbf E}\, s_j(z^p)\overline{s_{j'}(z^{p'})}\,\right),\\ B_{nk}^N=\left (\,{\mathbf E}\, s_j(z^p)\overline{\frac{\d s_{j'}}{\d z_{q'}}(z^{p'})}\,\right),\\ C_{nk}^N=\left (\,{\mathbf E}\, \frac{\d s_j}{\d z_q}(z^p) \overline{\frac{\d s_{j'}}{\d z_{q'}}(z^{p'} )}\,\right);\\ j,j'=1,\dots,k;\quad p,p'=1,\dots,n; \quad q,q'=1,\dots,m. \end{array} \end{equation} By (\ref{a14a}) and (\ref{a3}), \begin{equation}\label{a23} \begin{array}{ll} A_{nk}^N=\left(\delta_{jj'}S_N(z^p,z^{p'})\right)\,,&\displaystyle S_N(z,w)=\left(1+\sum_{r=1}^m z_r\overline{w_r}\right)^N,\\ B_{nk}^N=\left(\delta_{jj'}S_{Nq'}(z^p,z^{p'})\right)\,, &\displaystyle S_{Nq'}(z,w)=Nz_{q'}\left(1+\sum_{r=1}^m z_r\overline{w_r}\right)^{N-1}\,, \\ C_{nk}^N=\left(\delta_{jj'} S_{Nqq'}(z^p,z^{p'})\right)\,,&\displaystyle S_{Nqq'}(z,w)=N(N-1)\overline{w_q}z_{q'} \left(1+\sum_{r=1}^m z_r\overline{w_r}\right)^{N-2}\\ &\displaystyle \qquad +\delta_{qq'}N\left(1+\sum_{r=1}^m z_r\overline{w_r}\right)^{N-1} \end{array} \end{equation} The $n$-point zero correlation functions $K_{nk}^N$ for the ${\operatorname{SU}}(m+1)$-polynomial $k$-tuples $\mathcal{S}_{N}$ can be computed by substituting (\ref{a23}) into formulas (\ref{fg6}) and (\ref{fg8}). (Alternately, we can compute the zero correlation functions with respect to the Euclidean volume on ${\mathbb C}^m$ by setting $\gamma^p=$Id in (\ref{fg8}).) \begin{rem} Note that the one-point correlation function, or the zero-density function, is constant, since it is invariant with respect to the group ${\operatorname{SU}}(m+1)$. Indeed, by B\'ezout's theorem and (\ref{volmeasure}), \begin{equation}\label{a4} |Z_s|(1)={\operatorname{Vol}}(V_s)=\int_{V_s}{\textstyle\frac{1}{ (m-k)!}}\omega^{m-k} =\Omega_{2m-2k}\deg V_s=\Omega_{2m-2k}N^k\;, \end{equation} where \begin{equation}\label{a4a}\Omega_{2\ell} = {\operatorname{Vol}}\, \C\PP^\ell = \frac{\pi^\ell}{ \ell !}\,.\end{equation} Hence, \begin{equation}\label{a32b} K_{1k}^N(z)=\frac{{\operatorname{Vol}}\, Z_s}{{\operatorname{Vol}}\, \C\PP^m}=\frac{N^{k}\Omega_{2m-2k}}{\Omega_{2m}} =\frac{N^{k}m!}{(m-k)!\pi^k}\,.\end{equation} We can also use our formulas to compute $K_{1k}^N$ directly: By (\ref{a23}), \begin{equation}\label{a32c} A_{1k}^N=\Big(\delta_{jj'}(1+|z|^2)^N\Big)\,,\quad B_{1k}^N=\Big(\delta_{jj'}Nz_{q'}(1+|z|^2)^{N-1}\Big)\,,\end{equation} $$C_{1k}^N=\Big(\delta_{jj'}N[(N-1)\bar z_qz_{q'}+(1+|z|^2)\delta_{qq'}](1+|z|^2)^{N-2}\Big)\,.$$ Hence by (\ref{fg6}), \begin{equation}\label{a32d} \Lambda_{1k}^N = \Big(\delta_{jj'}N[(1+|z|^2)\delta_{qq'}-\bar z_qz_{q'}](1+|z|^2)^{N-2}\Big)= \Big(\delta_{jj'}N(1+|z|^2)^Ng_{qq'}(z)\Big)\,.\end{equation} In the hypersurface case ($k=1$), we compute $$K_{11}^N=\frac{1}{ \pi (1+|z|^2)^N}\left.\left\langle\sum_{q,q'=1}^m \bar a_{1q}\gamma_{qq'} a_{1q'}\right\rangle\right|_{\Lambda_{11}^N}= \frac{N}{ \pi}\sum_{q,q'=1}^m \gamma_{qq'}g_{q'q}=\frac{Nm}{ \pi}\,,$$ as expected. For $k>1$, we have $\Lambda_{1k}^N(0)=NI$ where $I$ is the unit matrix, and (\ref{fg8}) yields $$ K_{1k}^N(0)=\frac{N^{k}}{\pi^k}\left\langle \det\left(\sum_{q=1}^m \bar a_{jq}{a_{j'q}}\right)_{j,j'=1,\dots,k} \right\rangle_I =\frac{N^{k}m!}{(m-k)!\pi^k}\,,$$ which agrees with (\ref{a32b}). \end{rem} \bigskip \section{Universality and scaling} Our goal is to derive scaling limits of the $n$-point correlations between the zeros of {\it random} $k$-tuples of sections of powers of a positive line bundle over a complex manifold. We expect the scaling limits to exist and to be universal in the sense that they should depend only on the dimensions of the algebraic variety of zeros and the manifold. Our plan is the following. We first describe scaling in the Heisenberg model, which we use to provide the universal scaling limit for the Szeg\"o kernel (Theorem \ref{neardiag}). Together with Theorem \ref{npointcor}, this demonstrates the universality of the scaling-limit zero correlation in the case of powers of any positive line bundle on any complex manifold. \subsection{Scaling of the Szeg\"o kernel in the Heisenberg group} Our model for scaling is the Szeg\"o kernel for the reduced Heisenberg group described in \S \ref{example2}. Recall that for the simply-connected Heisenberg group ${\bf H}^m$, the scaling operators (or Heisenberg dilations) $$ \delta_r (\zeta, t) = (r \zeta, r^2 t)\,,\quad r \in {\mathbb R}^+$$ are automorphisms of ${\bf H}^m$ (\cite{F} \cite{Stein}). Since the Szeg\"o kernel $\Pi$ of ${\bf H}^m$ is the unique self-adjoint holomorphic reproducing kernel, it follows that it must be invariant (up to a multiple) under these automorphisms. In fact, one has (\cite[p.~538]{Stein}): \begin{equation} \label{SI} \Pi(\delta_rx,\delta_ry)=r^{-2m-2}\Pi(x,y)\end{equation} The condition for a dilation $\delta_r$ to descend to the quotient group ${\bf H}_{\rm red}^m$ is that $r^2 {\mathbb Z} \subset {\mathbb Z}$, or equivalently, $r= \sqrt{N}$ with $N \in {\mathbb Z}^+$. Note however that $\delta_{\sqrt{N}}$ is not an automorphism of ${\bf H}^m_{\rm red}$ and there is no well-defined dilation by $\sqrt{N}^{-1}$. The scaling identity (\ref{SI}) descends to ${\bf H}^m_{\rm red}$ in the form \begin{equation}\Pi^\H_N(x, y) = N^m \Pi^\H_1 (\delta_{\sqrt{N}}\, x, \delta_{\sqrt{N}}\, y)\end{equation} with \begin{equation}\label{sksl} \Pi^\H_1(x, y) =\frac{1}{\pi^m} e^{i(\theta - \phi)}e^{ i \Im (z \cdot \bar{w})} e^{- {\frac{1}{2}} |z - w|^2}\,,\quad x=(z,\theta)\,,\ y=(w,\phi)\,. \end{equation} (Recall (\ref{PiHN}).) Informally, we may say that the scaling limit of $\Pi^\H_N$ equals $\Pi^\H_1$. Since scaling by $\sqrt{N}^{-1}$ is not well-defined on ${\bf H}_{\rm red}^m$ it is more correct to say that $\Pi^\H_N$ is the $\sqrt{N}$ scaling of the scaling limit kernel. \subsection{Scaling limit of a general Szeg\"o kernel} We now show that $\Pi^\H_1$ is the scaling limit of the $N$-th Szeg\"o kernel $\Pi_N$ of an arbitrary positive line bundle $L \to M$ in the sense of the following ``near-diagonal asymptotic estimate for the Szeg\"o kernel." \begin{theo} \label{neardiag} Let $z_0\in M$ and choose local coordinates in a neighborhood of $z_0$ so that $\Theta_h(z_0)=\sum dz_j\wedge d\bar z_j$. Then \begin{eqnarray*}N^{-m}\Pi_N(z_0+\frac{u}{\sqrt{N}},\frac{\theta}{N}; z_0+\frac{v}{\sqrt{N}},\frac{\phi}{N})&=& \frac{1}{\pi^m} e^{i(\theta-\phi)+i\Im (u\cdot \bar v)-{\frac{1}{2}} |u-v|^2} + O(N^{-1/2})\\ &=& \Pi^\H_1(u,\theta;v,\phi) + O(N^{-1/2})\;.\end{eqnarray*} \end{theo} To prove Theorem \ref{neardiag}, we need to recall the Boutet de Monvel-Sjostrand parametrix construction: \begin{theo} {\rm \cite [Th.~1.5 and \S 2.c]{B.S}} Let $\Pi(x,y)$ be the Szeg\"o kernel of the boundary $X$ of a bounded strictly pseudo-convex domain $\Omega$ in a complex manifold $L$. Then: there exists a symbol $s \in S^{n}(X \times X \times {\mathbb R}^+)$ of the type $$s(x,y,t) \sim \sum_{k=0}^{\infty} t^{n-k} s_k(x,y)$$ so that $$\Pi (x,y) = \int_0^{\infty} e^{i t\psi(x,y)} s(x,y,t ) dt $$ where the phase $\psi \in C^{\infty}(X \times X)$ is determined by the following properties: $\bullet$ $\psi(x,x) = \frac{1}{i} \rho(x)$ where $\rho$ is the defining function of $X$. $\bullet$ $\bar{\partial}_x \psi$ and $\partial_y \psi$ vanish to infinite order along the diagonal. $\bullet$ $\psi(x,y) = -\bar{\psi}(y,x)$. \end{theo} The integral is defined as a complex oscillatory integral and is regularized by taking the principal value (see \cite{B.S}). The phase is determined only up to a function which vanishes to infinite order at $x = y$ and its Taylor expansion at the diagonal is given by \begin{equation} \psi(x + u, x + v) = \frac{1}{i} \sum \frac{\partial^{\alpha + \beta} \rho} {\partial z^{\alpha}\partial \bar{z}^{\beta}}(x) \frac{u^{\alpha}}{\alpha!}\frac{\bar{v}^{\beta}}{\beta!}.\end{equation} The Szeg\"o kernels $\Pi_N$ are Fourier coefficients of $\Pi$ and hence may be expressed as: \begin{equation}\Pi_N(x,y) = \int_0^{\infty} \int_0^{2\pi} e^{- i N \theta} e^{it \psi( r_{\theta} x,y)} s(r_{\theta} x,y,t) d\theta dt \end{equation} where $r_{\theta}$ denotes the $S^1$ action on $X$. Changing variables $t \mapsto N t$ gives \begin{equation} \Pi_N(x,y) = N \int_0^{\infty} \int_0^{2\pi} e^{ i N ( -\theta + t \psi( r_{\theta} x,y))} s(r_{\theta} x,y, t N) d\theta dt\,.\end{equation} We now fix $z_0$ and consider the asymptotics of \begin{equation}\begin{array}{l}\displaystyle \Pi_N(z_0 + \frac{u}{\sqrt{N}}, 0; z_0 + \frac{v}{\sqrt{N}}, 0) \\[14pt]\displaystyle \quad\quad= N \int_0^{\infty} \int_0^{2\pi} e^{ i N ( -\theta + t\psi( z_0 + \frac{u}{\sqrt{N}}, \theta; z_0 + \frac{v}{\sqrt{N}}, 0))} s(z_0 + \frac{u}{\sqrt{N}}, \theta; z_0 + \frac{v}{\sqrt{N}}, 0), t N) d\theta dt \,.\end{array}\end{equation} In our setting the phase takes the following concrete form: We let $h (z,\bar{w})$ be the almost analytic function on $M \times M$ satisfying $h(z,\bar{z}) = \|e_L\|^{-2}_h(z)$. The function $h(z,\bar{w})$ is defined by \begin{equation}\label{taylorh} h (z_0 + u, \bar z_0 + \bar v) = \sum \frac{\partial^{\alpha + \beta} h(z,\bar{z})} {\partial z^{\alpha} \partial \bar{z}^{\beta}} (z_0) \frac{u^{\alpha}}{\alpha!} \frac{\bar{v}^{\beta}}{\beta!}. \end{equation} We consider the complex manifold $Y=L^*$ and we let $(z,\lambda)$ denote the coordinates of $\xi \in Y$ given by $\xi = \lambda (e_L^*)_z$. In the associated coordinates $(x,y) = (z, \lambda, w, \mu)$ on $Y \times Y$, we have: \begin{equation} \rho(z,\lambda) = 1 - h(z, \bar{z})|\lambda|^2,\;\;\;\; \psi(z, \lambda, w, \mu) = \frac{1}{i}(1 - h(z,\bar{w}) \lambda \bar{\mu})\;. \end{equation} We consider $\Omega=\{\rho<0\}$ and $X=\partial\Omega=\{\rho=0\}$. We may assume without loss of generality that $h(z,\bar{w}) = \bar{h}(w,\bar{z})$ since $h(z, \bar{z})$ is real so we could replace $h$ by $\frac{1}{2}h(z,\bar{w}) + {\frac{1}{2}}\bar{h}(w,\bar{z})$. On $X$ we have $h(z, \bar{z}) |\lambda|^2 = 1$ so we may write $\lambda = h(z, \bar{z})^{-{\frac{1}{2}} } e^{i \phi}$, and similarly for $\mu$. So for $(x,y) = (z, \phi, w, \phi') \in X \times X$ we have \begin{equation} \psi(z, \phi, w, \phi') = \frac{1}{i} \left[1 - \frac{h(z,\bar{w})}{ \sqrt{h(z, \bar{z})} \sqrt{h(w, \bar{w})}} e^{i (\phi - \phi')}\right]\;. \end{equation} It follows that \begin{equation}\begin{array}{l}\displaystyle\psi( z_0 + \frac{u}{\sqrt{N}}, \theta; z_0 + \frac{v}{\sqrt{N}}, 0)\\[14pt] \displaystyle \quad\quad= \frac{1}{i} \left[1 - \frac{h(z_0 + \frac{u}{\sqrt{N}},\bar{z_0} + \frac{\bar{v}}{\sqrt{N}})}{ \sqrt{h(z_0 + \frac{u}{\sqrt{N}}, \bar{z_0} + \frac{\bar u}{\sqrt{N}})} \sqrt{h(z_0 + \frac{v}{\sqrt{N}}, \bar{z_0} + \frac{\bar v}{\sqrt{N}})}} e^{i \theta}\right] . \end{array}\end{equation} We now assume that $e_L$ is a normal frame centered at $z_0$. By definition, this means that \begin{equation} h(z_0)=1,\quad \d h |_{z_0} = \bar\partial h|_{z_0} = 0. \end{equation} We furthermore assume that our coordinates $\{z_j\}$ are chosen so that the Levi form of $h$ is the identity at $z_0$: \begin{equation}\frac{\d^2 h}{\d z^\alpha \d z^\beta}(z_0,\bar z_0) =\delta_{\alpha\beta}\,.\end{equation} (This is equivalent to specifying that $\omega(z_0)=\frac{i}{2}\sum_j dz_j \wedge d\bar z_j$.) Then by (\ref{taylorh}), \begin{equation}\label{taylorh2} h(z_0 + \frac{u}{\sqrt{N}},\bar z_0 + \frac{\bar v}{\sqrt{N}}) = \frac{1}{N}u\cdot \bar v + O(N^{-3/2}). \end{equation} Now let us return to the phase. It is given by \begin{equation}\label{phase} t \left[ 1 - \frac{h(z_0 + \frac{u}{\sqrt{N}},\bar z_0 + \frac{\bar v}{\sqrt{N}})}{h(z_0 + \frac{u}{\sqrt{N}},\bar z_0 + \frac{\bar u}{\sqrt{N}})^{{\frac{1}{2}}} h(z_0 + \frac{v}{\sqrt{N}},\bar z_0 + \frac{\bar v}{\sqrt{N}})^{{\frac{1}{2}}}} e^{i \theta}\right] - i \theta. \end{equation} By (\ref{taylorh2}), the phase (\ref{phase}) has the form: \begin{equation} (t[ 1 - e^{i \theta}] - i \theta) + \frac{t}{N}\left[u\cdot\bar v - {\frac{1}{2}} |u|^2 - {\frac{1}{2}} |v|^2\right]e^{i \theta} + O(N^{-3/2}). \end{equation} It is now evident that $\Pi_N(z_0 + \frac{u}{\sqrt{N}}, 0; z_0 + \frac{v}{\sqrt{N}}, 0)$ is given by an oscillatory integral with phase $(t [ 1 - e^{i \theta}] - i \theta)$; the latter two terms can be absorbed into the amplitude. Thus we have: \begin{equation}\begin{array}{l} \Pi_N(z_0 + \frac{u}{\sqrt{N}}, 0; z_0 + \frac{v}{\sqrt{N}}, 0)\\[14pt]\quad = N \int_0^{\infty} \int_0^{2\pi} e^{ i N (t [ 1 - e^{i \theta}] - i \theta)} e^{t[u\cdot\bar v - {\frac{1}{2}} |u|^2 - {\frac{1}{2}} |v|^2] +O(N^{-1/2})} s(z_0 + \frac{u}{\sqrt{N}}, \theta; z_0 + \frac{v}{\sqrt{N}}, 0), t N) d\theta dt.\end{array}\end{equation} We may then evaluate the integral asymptotically by the stationary phase method as in \cite{Z}. The phase is precisely the same as occurs in $\Pi_N(x,x)$, and as discussed in \cite{Z}, the single critical point occurs at $t = 1, \theta = 0$. We may also Taylor-expand the amplitude to determine its contribution to the asymptote. Precisely as in the calculation of the stationary phase expansion in \cite{Z}, we get: \begin{equation} \Pi_N(z_0 + \frac{u}{\sqrt{N}}, 0; z_0 + \frac{v}{\sqrt{N}}, 0) = \frac{N^m}{\pi^m} e^{u\cdot\bar v - {\frac{1}{2}} |u|^2 - {\frac{1}{2}} |v|^2} + O(N^{m - {\frac{1}{2}}}). \end{equation} Finally, we note that $$u\cdot\bar v - {\frac{1}{2}} |u|^2 - {\frac{1}{2}} |v|^2=-{\frac{1}{2}} |u-v|^2 +i\Im (u\cdot \bar v)\,,$$ which completes the proof of Theorem \ref{neardiag}.\qed \subsection{Universality of the scaling limit of correlations of zeros} We are now ready to pass to the scaling limit as $N\to\infty$ of the correlation functions of sections of powers $L^N$ of our line bundle. To explain this notion, let us consider the case $k=m$ where the zeros are (almost surely) discrete. An $m$-tuple of sections of $L^N$ will have $N^m$ times as many zeros as $m$-tuples of sections of $L$. Hence we must expand our neighborhood (or contract our ``yardstick") by a factor of $N^{m/2}$. Let $z^0\in M$ and choose a coordinate neighborhood $U\in M$ with coordinates $\{z_j\}$ for which $z^0=0$ and $\omega(z^0)=\frac{i}{2}\sum_q dz_q\wedge d\bar z_q$. We define the {\it $n$-point scaling limit zero correlation function\/} $$K_{nkm}^\infty(z)=\lim_{N\to\infty}\frac{1}{ N^{nk}} K_{nk}^N\left(\frac{z}{ \sqrt{N}}\right)\,,\quad z=(z^1,\dots,z^n)\in ({\mathbb C}^m)_n\,.$$ We show below (Theorem \ref{uslc}) that this limit exists and that $K_{nkm}^\infty$ is universal by passing to the limit in Theorem \ref{npointcor}, using Theorem \ref{neardiag}. First, we need the following fact: \begin{lem}\label{detinfty} Let $z^1,\dots,z^n$ be distinct points of ${\mathbb C}^m$. Then $$\det\left(\Pi^\H_1(z^p,0;z^{p'},0)\right)= e^{-\sum|z^p|^2}\det\left( e^{z^p\cdot\bar z^{p'}}\right) \ne 0\;.$$\end{lem} \begin{proof} We consider the first Szeg\"o projector on the reduced Heisenberg group \begin{equation} \Pi^\H_1:\mathcal{L}^2(\H^m_{\rm red})\to \mathcal{H}^2_1(\H^m_{\rm red})\approx \mathcal{L}^2({\mathbb C}^m,e^{-|z|^2})\cap\mathcal{O}({\mathbb C}^m)\,,\end{equation} where $$ \mathcal{L}^2({\mathbb C}^m,e^{-|z|^2})=\left\{f\in\mathcal{L}^2_{\rm loc}({\mathbb C}^m): {\textstyle \int_{{\mathbb C}^m}}|f|^2 e^{-|z|^2}dz <+\infty\right\}\,.$$ (See the remark at the end of \S \ref{example2}.) Its kernel can be written in the form \begin{equation} \Pi^\H_1(z,\theta;w,\phi)=e^{i(\theta-\phi)} \sum_{\alpha=1}^\infty f_\alpha(z)\overline{f_\alpha(w)}\,,\end{equation} where the $f_\alpha$ form a complete orthonormal basis for $\mathcal{L}^2({\mathbb C}^m,e^{-|z|^2})\cap\mathcal{O}({\mathbb C}^m)$. (E.g., $\{f_\alpha\}$ can be taken to be the set of monomials $\left\{c_{j_1\cdots j_m}z_1^{j_1}\cdots z_m^{j_m}\right\}$. In fact, $\Pi^\H_1(z,0;w,0))$ is just a ``weighted Bergman kernel" on ${\mathbb C}^m$.) We now mimic the proof of Lemma \ref{detN}, except this time we have an infinite sum over the index $\alpha$; this sum converges uniformly on bounded sets in ${\mathbb C}^m\times {\mathbb C}^m$ since the sup norm over a bounded set is dominated by the Gaussian-weighted $\mathcal{L}^2$ norm (by the same argument as in the case of the ordinary Bergman kernel on a bounded domain). We then obtain a nonzero vector $(v_1,\dots,v_n)\in{\mathbb C}^m$ such that $\sum_p v_p f_\alpha(z^p)=0$ for all $\alpha$. But then $\sum_p v_p f(z^p)=0$ for all polynomials $f$ on ${\mathbb C}^m$, a contradiction. \end{proof} We can now show the universality of the scaling limit of the zero correlation functions: \begin{theo}\label{uslc} Let $(L,h)$ be a positive Hermitian line bundle on an $m$-dimensional compact complex manifold $M$ with K\"ahler\ form $\omega=\frac{i}{2}\Theta_h$, let $\mathcal{S}=H^0(M,L^N)^k$ ($k\ge 1$), and give $\mathcal{S}$ the standard Gaussian measure $\mu$. Then $$\frac{1}{ N^{nk}} K_{nk}^N\left(\frac{z^1}{\sqrt{N}}, \dots, \frac{z^n}{\sqrt{N}}\right) = K_{nkm}^\infty(z^1,\dots,z^n) + O\left(\frac{1}{\sqrt{N}}\right)\,,$$ where $K_{nkm}^\infty(z^1,\dots,z^n)$ is given by a universal rational function in the quantities $z^p_q, \bar z^p_q, e^{z^p\cdot \bar z^{p'}}$, and the error term has $\ell^{th}$ order derivatives $\le \frac{C_{S,\ell}}{\sqrt{N}}$ on each compact subset $S\subset ({\mathbb C}^m)_n$, for all $\ell\ge 0$. \end{theo} \begin{proof} By taking the scaling limit of (\ref{npointcoreq}), we obtain \begin{equation}\label{uscleq} K_{nkm}^\infty(z)= \frac{\mathcal{P}_{nkm}\big(\Pi^\H_1(z^p,z^{p'}),d^H_{\bar w_{q}}\Pi^\H_1(z^p,z^{p'}), d^H_{z_{q}}\Pi^\H_1(z^p,z^{p'}), d^H_{z_{q}}d^H_{\bar w_{q'}}\Pi^\H_1(z^p,z^{p'})\big)}{\pi^{kn}\left[\det \big(\Pi^\H_1(z^p,z^{p'})\big)_{1\le p,p'\le n}\right]^{k(n+1)}}\,. \end{equation} Indeed, since the coefficients of $\Lambda_n$ are either of degree 1 in the coefficients of $C_n$ or of degree 2 in the coefficients of $B_n$, we see by the proof of Theorem \ref{npointcor}, using (\ref{dPiNH}), (\ref{dms5})--(\ref{dms6}) and Theorem \ref{neardiag}, that the leading term of the asymptotic expansion of $K_{nk}^N$ is $N^{nk}$ times the right side of (\ref{uscleq}). The bound on the error term follows from Theorem \ref{neardiag} and Lemma \ref{detinfty}. Substituting into (\ref{uscleq}) the values of $\Pi^\H_1(z^p,z^{p'})$ and its horizontal derivatives obtained from (\ref{dPiNHleft}) (with $N=1$) and (\ref{sksl}) and canceling out common factors of $e^{-|z^p|^2/2}$ and $\pi$, we obtain \begin{eqnarray} K_{nkm}^\infty(z)&=& \frac{\mathcal{P}_{nkm}\big(e^{z^p\cdot \bar z^{p'}},(z^{p}_q-z^{p'}_q) e^{z^p\cdot \bar z^{p'}}, (\bar z^{p'}_q -\bar z^{p}_q)e^{z^p\cdot\bar z^{p'}}, [(z^p_{q'}-z^{p'}_{q'})(\bar z^{p'}_q-\bar z^{p}_q)+\delta_{qq'}]e^{z^p\cdot\bar z^{p'}}\big)}{\pi^{kn}\left[\det \big(e^{z^p\cdot\bar z^{p'}}\big)_{1\le p,p'\le n}\right]^{k(n+1)}}\nonumber \\ \label{uscleq2left}\\ &=& \frac{\mathcal{Q}_{nkm}\big(z^p_q,\bar z^p_q, e^{z^p\cdot \bar z^{p'}}\big)}{\pi^{kn}\left[\det \big(e^{z^p\cdot \bar z^{p'}}\big)\right]^{k(n+1)}}\,,\nonumber \end{eqnarray} where $\mathcal{Q}_{nkm}$ is a universal polynomial (homogeneous of degree $k(n+1)$ in each of the variables $e^{z^p\cdot \bar z^{p'}}$ and with integer coefficients). \end{proof} \begin{rem} As we remarked previously, formula (\ref{uscleq}) is valid for any connection, so we can replace the left invariant vector fields with their right-invariant counterparts to obtain \begin{equation}\label{uscleq2} K_{nkm}^\infty(z)= \frac{\mathcal{P}_{nkm}\big(e^{z^p\cdot \bar z^{p'}},z^{p}_q e^{z^p\cdot \bar z^{p'}}, \bar z^{p'}_q e^{z^p\cdot\bar z^{p'}}, (z^p_{q'}\bar z^{p'}_q+\delta_{qq'})e^{z^p\cdot\bar z^{p'}}\big)}{\pi^{kn}\left[\det \big(e^{z^p\cdot\bar z^{p'}}\big)_{1\le p,p'\le n}\right]^{k(n+1)}}\,. \end{equation}\end{rem} \bigskip \section {Formulas for the scaling limit zero correlation function}\label{formulas} We now apply the formulas from \S\S \ref{formula-gaussian}--\ref{D-meet-S} to transform (\ref{uscleq2}) into explicit formulas for $K_{nkm}^\infty$. We use the right-invariant connection $\alpha^R$ so that $d^H_{z_q} = Z_q^R.$ Indeed, by the proofs of Theorems \ref{npointcor} and \ref{uslc} (which use formulas (\ref{fg6}), (\ref{fg8}), (\ref{dms4})--(\ref{dms6})), formula (\ref{uscleq2}) becomes \begin{equation}\label{a28} K_{nkm}^\infty(z^1,\dots,z^n)=\frac{1}{\pi^{kn}\det A_{nkm}}\left\langle \prod_{p=1}^n \det (a^pa^{p*})\right\rangle_{\Lambda_{nkm}}\,, \end{equation} where \begin{equation}\label{a29} \Lambda_{nkm}=C_{nkm}-B^*_{nkm}A^{-1}_{nkm}B_{nkm} \end{equation} with \begin{equation}\label{a30} \begin{array}{ll} A_{nkm}=\left(\delta_{jj'}S(z^p,z^{p'})\right)\,,&\displaystyle S(z,w)=\exp\left(\sum_{r=1}^m z_r\overline{w_r}\right),\\ B_{nkm}=\left(\delta_{jj'}S_{q'}(z^p,z^{p'})\right)\,, &\displaystyle S_{q'}(z,w) =z_{q'}\exp\left(\sum_{r=1}^m z_r\overline{w_r}\right)\,, \\ C_{nkm}=\left(\delta_{jj'} S_{qq'}(z^p,z^{p'})\right)\,,&\displaystyle S_{qq'}(z,w)=(\delta_{qq'}+\overline{w_q}z_{q'}) \exp\left(\sum_{r=1}^m z_r\overline{w_r}\right)\\ j,j'=1,\dots,k;\quad p,p'=1,\dots,n;&\displaystyle q,q'=1,\dots,m. \end{array} \end{equation} The metric tensor $g^p$ in (\ref{fg8}) becomes a unit tensor in the scaling limit, so there is no $\gamma^p$ on the right in (\ref{a28}). Because $\Pi_1^\H$ is invariant with respect to unitary transformations and equivariant with respect to translations (i.e., $\Pi_1^\H(z+u,w+u)=e^{i\Im(z\cdot \bar u)} e^{-i\Im(w\cdot \bar u)} \Pi_1^\H(z,w)$), the scaling limit zero correlation $K_{nkm}^\infty$ is invariant with respect to the group of isometric transformations---unitary transformations and translations---of ${\mathbb C}^m$. In particular, the limit one-point zero correlation, or the zero-density function, is constant, since it is invariant under translation. Indeed by (\ref{a30}), $A_{1km}=e^{|z|^2}I_k$ and $\Lambda_{1km}=e^{|z|^2}I_{km}$, where $I_k$, resp.\ $I_{km}$, denotes the unit $k\times k$, resp.\ $(km)\times (km)$, matrix. Thus by (\ref{a28}) and the Wick formula, \begin{equation}\label{a32e} K_{1km}^\infty(z)=\frac{1}{\pi^k e^{k|z|^2}}\left\langle \det\left(\sum_{q=1}^m \bar a_{jq}{a_{j'q}}\right)_{j,j'=1,\dots,k} \right\rangle_{e^{|z|^2}I_{km}} =\frac{m!}{\pi^k(m-k)!}\,.\end{equation} Thus we define the {\it normalized n-point scaling limit zero correlation function} \begin{equation}\label{nslzc} \widetilde K^\infty_{nkm}(z)= (K^\infty_{1km})^{-n}K^\infty_{nkm}(z)=\left(\frac{\pi^k(m-k)!}{m!}\right)^n K^\infty_{nkm}(z)\,.\end{equation} \begin{rem} These formulas also follow from \S \ref{zcorpoly}. For example, equation (\ref{a32e}) is a consequence of (\ref{a32b}) since $$K_{1km}^\infty(z)=\frac{1}{N^k} K_{1k}^N(z)\,.$$ Furthermore, using the notation of \S \ref{zcorpoly}, we observe that \begin{equation}\label{a27} \begin{array}{ll} &\displaystyle\lim_{N\to\infty} S_N\left(\frac{z}{\sqrt{N}}, \frac{w}{\sqrt{N}}\right)= \lim_{N\to\infty}\left(1+N^{-1}\sum_{r=1}^m z_r\overline{w_r}\right)^N =S(z,w)\,,\\ &\displaystyle\lim_{N\to\infty} N^{-1/2}S_{Nq'}\left(\frac{z}{\sqrt{N}}, \frac{w}{\sqrt{N}}\right)= S_{q'}(z,w)\,,\\ &\displaystyle\lim_{N\to\infty} N^{-1}S_{Nqq'}\left(\frac{z}{\sqrt{N}}, \frac{w}{\sqrt{N}}\right)= S_{qq'}(z,w)\,. \end{array} \end{equation} Equations (\ref{a27}) provide an alternate derivation of (\ref{a30}). \end{rem} \subsection{Decay of correlations}\label{decay} Explicit formulas for the correlation functions $\widetilde K^\infty_{nkm}$ can be obtained from (\ref{a28}), (\ref{a30}) and the Wick formula. We shall illustrate these computations for the cases $n=2,\ k=1,2$ in \S\S \ref{explicit1}--\ref{explicit2} below. We now note that the limit correlations are ``short range" in the following sense: \begin{theo}\label{shortrange} The correlation functions satisfy the estimate $$\widetilde K^\infty_{nkm}(z^1,\dots,z^n) = 1 +O(r^4 e^{-r^2}) \quad {\rm as}\ r\to \infty\,, \quad r=\min_{p\ne p'}|z^p-z^{p'}|\,.$$ \end{theo} \begin{proof} We use formula (\ref{uscleq}), which comes from (\ref{a28})--(\ref{a29}) as in the proof of Theorem \ref{uslc}. To determine the matrices $A,B,C$, we let $d^H_{z_q}=Z^L_q,\ d^H_{\bar w_q}=\bar W^L_q$ (instead of the right-invariant vector fields we used above). Recalling (\ref{dPiNHleft}), we have: \begin{eqnarray}\label{ABC} A^{jp}_{j'p'}&=& \delta_{jj'} A^p_{p'}\,,\qquad A^p_{p'} = \pi^m \Pi_1^\H(z^p,0;z^{p'},0)\,,\nonumber \\ B^{jp}_{j'p'q'} &=& \delta_{jj'}(z^{p}_{q'}-z^{p'}_{q'}) A^p_{p'}\,,\\ C^{jpq}_{j'p'q'} &=& \delta_{jj'}\big(\delta_{qq'}+(\bar z^{p'}_q-\bar z^{p}_{q}) ( z^{p}_{q'}-z^{p'}_{q'})\big) A^p_{p'} \,.\nonumber \end{eqnarray} By (\ref{sksl}), \begin{eqnarray*} A^p_{p'}&=&\left\{\begin{array}{ll}1 & p=p'\\ O(e^{-r^2/2})\, \quad & p\ne p'\end{array}\right.\,,\\ B&=&O(re^{-r^2/2})\,,\\ C&=&I+O(r^2 e^{-r^2/2})\,,\qquad C^{jpq}_{jpq}=1 \,.\end{eqnarray*} Recalling (\ref{fg6}), we have \begin{equation}\label{Lambda}\Lambda=I+O(r^2 e^{-r^2/2})\,,\qquad \Lambda^{jpq}_{jpq}=1+O(r^2 e^{-r^2})\,.\end{equation} We now apply formula (\ref{a28}); note that the Wick formula involves terms that are products of diagonal elements of $\Lambda$, and products that contain at least two off-diagonal elements of $\Lambda$. The former terms are of the form $1+O(r^2e^{-r^2})$, and the latter are $O(r^4e^{-r^2})$. Similarly, $\det A=1+O(r^4e^{-r^2})$. The desired estimate then follows from (\ref{nslzc}). \end{proof} We shall see from our computations of the pair correlation below that Theorem \ref{shortrange} is sharp. The theorem can be extended to estimates of the connected correlation functions (called also truncated correlation functions, cluster functions, or cumulants), as follows. The $n$-point connected correlation function is defined as (see, e.g., \cite[p.~286]{GJ}) \begin{equation}\label{cc1}\widetilde T^{\infty}_{nkm}(z^1,\dots,z^n)=\sum_G(-1)^{l+1}(l-1)! \prod_{j=1}^l \widetilde K^{\infty}_{n_jkm}(z^{i_1},\dots,z^{i_{n_j}}), \end{equation} where the sum is taken over all partitions $G=(G_1,\dots,G_l)$ of the set $(1,\dots,n)$ and $G_j=(i_1,\dots,i_{n_j})$. In particular, recalling that $\widetilde K^\infty_{1km}\equiv 1$, \begin{eqnarray*}\label{cc2} \widetilde T^{\infty}_{1km}(z^1)&=&\widetilde K^{\infty}_{1km}(z^1)=1\,,\\ \widetilde T^{\infty}_{2km}(z^1,z^2)&=&\widetilde K^{\infty}_{2km}(z^1,z^2) -\widetilde K^{\infty}_{1km}(z^1)\widetilde K^{\infty}_{1km}(z^2) \ =\ \widetilde K^{\infty}_{2km}(z^1,z^2)-1\,,\\ \widetilde T^{\infty}_{3km}(z^1,z^2,z^3)&=&\widetilde K^{\infty}_{3km}(z^1,z^2,z^3) -\widetilde K^{\infty}_{2km}(z^1,z^2)\widetilde K^{\infty}_{1km}(z^3) -\widetilde K^{\infty}_{2km}(z^1,z^3)\widetilde K^{\infty}_{1km}(z^2)\\ &&\ -\ \widetilde K^{\infty}_{2km}(z^2,z^3)\widetilde K^{\infty}_{1km}(z^1) +2 \widetilde K^{\infty}_{1km}(z^1)\widetilde K^{\infty}_{1km}(z^2) \widetilde K^{\infty}_{1km}(z^3)\\ &=&\widetilde K^{\infty}_{3km}(z^1,z^2,z^3) -\widetilde K^{\infty}_{2km}(z^1,z^2) -\widetilde K^{\infty}_{2km}(z^1,z^3) -\widetilde K^{\infty}_{2km}(z^2,z^3)+2\,, \end{eqnarray*} and so on. The inverse of (\ref{cc1}) is \begin{equation}\label{cc3}\widetilde K^{\infty}_{nkm}(z^1,\dots,z^n)=\sum_G \prod_{j=1}^l \widetilde T^{\infty}_{n_jkm}(z^{i_1},\dots,z^{i_{n_j}}) \end{equation} (Moebius' theorem). The advantage of the connected correlation functions is that they go to zero if at least one of the distances $|z^i-z^j|$ goes to infinity (see Corollary \ref{ccc} below). In our case the connected correlation functions can be estimated as follows. Define \begin{equation}\label{cc4} d(z^1,\dots,z^n)=\max_{\mathcal{G}}\prod_{l\in L} |z^{i(l)}-z^{f(l)}|^2e^{-|z^{i(l)}-z^{f(l)}|^2/2}. \end{equation} where the maximum is taken over all oriented connected graphs $\mathcal{G}=(V,L,i,f)$ such that $V=(z^1,\dots,z^n)$ and for every vertex $z^j\in V$ there exist at least two edges emanating from $z^j$. Here $V$ denotes the set of vertices of $\mathcal{G}$, $L$ the set of edges, and $i(l)$ and $f(l)$ stand for the initial and final vertices of the edge $l$, respectively. Observe that the maximum in (\ref{cc4}) is achieved at some graph $\mathcal{G}$, because $te^{-t/2}\le 2/e<1$ and therefore the product in (\ref{cc4}) is less or equal $(2/e)^{|L|}$ which goes to zero as $|L|\to\infty$. \begin{theo}\label{cc} The connected correlation functions satisfy the estimate $$\widetilde T^\infty_{nkm}(z^1,\dots,z^n) = O(d(z^1,\dots,z^n)) \quad {\rm as}\ \max_{p,q}|z^p-z^q|\to \infty\,,$$ provided that $\min_{p,q}|z^p-z^q|\ge c>0$. \end{theo} This theorem implies Theorem \ref{shortrange} because of the inversion formula (\ref{cc3}). To prove the theorem let us remark that we can rewrite (88) (using the Wick theorem) as a sum over Feynman diagrams. Namely, for the normalized correlation functions $\widetilde K^\infty_{nkm}(z^1,\dots,z^n)$ we have that \begin{equation}\label{cc5} \widetilde K^\infty_{nkm}(z^1,\dots,z^n) ={[(m-k)!/m!]^n\over \det A_{nkm}} \sum_{\mathcal{F}} A_{\mathcal{F}}(z^1,\dots,z^n)\,, \end{equation} where the sum is taken over all graphs $\mathcal{F}=(V,L,i,f)$ (Feynman diagrams) such that $V=(z^1,\dots,z^n)$ and the edges $l\in L$ connect the paired variables $a^{i(l)}_{jq},\; a^{f(l)*}_{j'q'}$ in a given term of the Wick sum for $\widetilde K^\infty_{nkm}(z^1,\dots,z^n)$. The function $A_{\mathcal{F}}(z^1,\dots,z^n)$ is a sum over all terms in the Wick sum with a fixed Feynman diagram $\mathcal{F}$. In other words, to get $A_{\mathcal{F}}(z^1,\dots,z^n)$ we fix pairings $(a^p_{jq},a^{p'*}_{j'q'})$ prescribed by $\mathcal{F}$ and sum up in the Wick formula over all indices $j,q$ at every $a^p$. A remarkable property of the connected correlation functions is that they are represented by the sum over connected Feynman diagrams (see, e.g., \cite{GJ}), \begin{equation}\label{cc6}\widetilde T^\infty_{nkm}(z^1,\dots,z^n)= {[(m-k)!/m!]^n\over \det A_{nkm}}{\sum_{\mathcal{F}}}^{\rm conn} A_{\mathcal{F}}(z^1,\dots,z^n)\,.\end{equation} Since $\det A_{nkm}\ge c_1>0$ and $|\Lambda^{jpq}_{j'p'q'}|\le c_2<+\infty$ when $\min_{p,q}|z^p-z^q|\ge c>0$, we conclude from (\ref{fg6}), (\ref{ABC}) and (\ref{sksl}) that for all connected Feynman diagrams $\mathcal{F}$, \begin{equation}\label{cc7} A_{\mathcal{F}}(z^1,\dots,z^n)=O(d)\,, \end{equation} where $d=d(z^1,\dots,z^n)$ is defined in (\ref{cc4}). Summing up over $\mathcal{F}$, we prove Theorem \ref{cc}.\qed \begin{cor}\label{ccc} The connected correlation functions satisfy the estimate $$\widetilde T^\infty_{nkm}(z^1,\dots,z^n) = O(R^2e^{-R^2/2}) \quad {\rm as}\ R\to\infty\,,\quad R= \max_{p,q}|z^p-z^q|\,,$$ provided that $\min_{p,q}|z^p-z^q|\ge c>0$. \end{cor} \subsection {Hypersurface pair correlation}\label{explicit1} We now give an explicit formula [(\ref{a40})] for pair correlations in codimension 1 ($k=1,\ n=2$). The case $m=1$ of this formula coincides, as it must, with the formula given by \cite{H} and \cite{BBL} for the universal scaling limit pair correlation for ${\operatorname{SU}}(2)$ polynomials. In another paper \cite{BSZ}, we gave a different proof of (\ref{a40}) using the Poincar\'e-Lelong formula. Since the scaling-limit pair correlation function $K_{2km}^\infty(z^1,z^2)$ is invariant with respect to the group of isometries of ${\mathbb C}^m$, it depends only on the distance $r=|z^1-z^2|$, so we can set $z^1=0$ and $z^2=(r,0,\dots,0)$. To simplify notation, we shall henceforth write $A=A_{2km},\; B=B_{2km},\; C=C_{2km},\; \Lambda=\Lambda_{2km}$. In this case, (\ref{a30}) reduces to \begin{equation}\label{a33} \begin{array}{lll} A= \left(\begin{array}{ll} 1 & 1 \\ 1 & e^{r^2} \end{array}\right);\\ B=(B^p_{p'q});&\ (B^p_{p'1})= \left(\begin{array}{ll} 0 & 0 \\ r & re^{r^2} \end{array}\right);&\quad (B^p_{p'q})= \left(\begin{array}{ll} 0 & 0 \\ 0 & 0 \end{array}\right),\quad q\ge 2;\\ C=(C^{pq}_{p'q'});&\ (C^{p1}_{p'1})= \left(\begin{array}{ll} 1 & 1 \\ 1 & (1+r^2)e^{r^2} \end{array}\right); &\quad (C^{pq}_{p'q'})=\delta_{qq'} \left(\begin{array}{ll} 1 & 1 \\ 1 & e^{r^2} \end{array}\right),\quad q,q'\ge 2. \end{array} \end{equation} The matrix \begin{equation}\label{a34} \Lambda=(\Lambda^{pq}_{p'q'})=C-B^*A^{-1}B \end{equation} is given by \begin{equation}\label{a35} \Lambda^{p1}_{p'1}= \left(\begin{array}{ll} \displaystyle\frac{e^u-1-u}{e^u-1} & \displaystyle\frac{e^u-1-ue^u}{e^u-1} \\ \displaystyle\frac{e^u-1-ue^u}{e^u-1} & \displaystyle\frac{e^{2u}-e^u-ue^u}{e^u-1} \end{array}\right);\quad \Lambda^{pq}_{p'q'}=\delta_{qq'} \left(\begin{array}{ll} 1 & 1 \\ 1 & e^u \end{array}\right),\ q,q'\ge 2\,, \end{equation} where $u=r^2=|z^1-z^2|^2$. By (\ref{a28}), (\ref{nslzc}) and the formula for $A$ in (\ref{a33}), we have \begin{equation}\label{a36} \widetilde K^\infty_{21m}(z^1,z^2)=\frac{1}{m^2(e^u-1)}\left\langle \left(\sum_{q=1}^m a^1_q\overline{a^1_q}\right) \left(\sum_{q'=1}^m a^2_{q'}\overline{a^2_{q'}}\right)\right\rangle_{\Lambda} \end{equation} By the Wick formula (see for example, \cite[(I.13)]{Si}), \begin{equation}\label{a37}\begin{array}{lll} \widetilde K^\infty_{21m}(z^1,z^2)&=&\displaystyle\frac{1}{m^2(e^u-1)}\left[ \left(\sum_{q=1}^m\langle a^1_q\overline{a^1_q}\rangle_{\Lambda_2}\right) \left(\sum_{q'=1}^m\langle a^2_{q'}\overline{a^2_{q'}}\rangle_{\Lambda_2}\right) +\sum_{q,q'=1}^m \langle a^1_q\overline{a^2_{q'}}\rangle_{\Lambda_2} \langle \overline{a^1_q}a^2_{q'}\rangle_{\Lambda_2}\right]\\[14pt] &=&\displaystyle\frac{1}{m^2(e^u-1)}\left[ \left(\sum_{q=1}^m \Lambda^{1q}_{1q}\right)\left(\sum_{q'=1}^m \Lambda^{2q'}_{2q'}\right)+\sum_{q,q'=1}^m \Lambda^{1q}_{2q'} \Lambda^{2q'}_{1q}\right]\,.\end{array}\end{equation} Substituting the values of $\Lambda^{pq}_{p'q'}$ given by (\ref{a35}), we obtain \begin{equation}\label{a37a} \begin{array}{lll} \widetilde K^\infty_{21m}(z^1,z^2)&=&\displaystyle\frac{1}{m^2(e^u-1)}\left[ \left(\frac{e^u-1-u}{e^u-1}+m-1\right) \left(\frac{e^{2u}-e^u-ue^u}{e^u-1}+(m-1)e^u\right)\right.\\[14pt] &&\quad\displaystyle\left.+\left(\frac{e^u-1-ue^u}{e^u-1}\right)^2+(m-1)\right]\,, \qquad u=|z^1-z^2|^2\,.\end{array} \end{equation} After simplification, \begin{equation} \widetilde K^\infty_{21m}(z^1,z^2)=\frac{u^2(e^{2u}+e^u)-2u(e^{2u}-e^u) +m^2(e^u-1)^2e^u+m(e^u-1)^2}{m^2(e^u-1)^3}\,. \end{equation} Putting $u=2t$ and writing \begin{equation}\label{a39} \widetilde K^\infty_{21m}(z^1,z^2)=\kappa_{1m}(|z^1-z^2|)\,, \end{equation} we then obtain \begin{equation}\label{a40} \kappa_{1m}(r)=\frac{\left[\frac{1}{2}(m^2+m)\sinh^2t+t^2\right] \cosh t -(m+1)t\sinh t}{m^2\sinh^3t}+\frac{(m-1)}{2m},\quad t=\frac{r^2}{2}\,. \end{equation} The case $m=1$ of formula (\ref{a40}) was obtained by Bogomolny-Bohigas-Leboeuf \cite{BBL} and Hannay \cite{H}. As $r\to\infty$, \begin{equation}\label{a40a} \kappa_{1m}(r)=1 + \frac{r^4-2(m^2+1)r^2+m(3m+1)}{m^2}e^{-r^2} +O(r^4e^{-2r^2})\,. \end{equation} The following expansion of the correlation function was obtained from (\ref{a40}) using Maple$^{\rm TM}$: \begin{eqnarray*} \kappa_{1m}& =&\frac{m-1}{2m}t^{-1}+\frac{m-1}{2m}+ \frac{1}{6}\,{\frac {\left (m+2\right )\left (m+1\right )}{{m}^{2}}}t -{\frac {1}{90}}\,{\frac {\left (m+4\right )\left (m+3\right )} {{m}^{2}}}{t}^{3}\\&&\ + {\frac {1}{945}}\,{\frac {\left (m+6\right )\left (m+5\right )} {{m}^{2}}}{t}^{5} -{\frac {1}{9450}}\,{\frac {\left (m+8\right )\left (m+7\right ) }{{m}^{2}}}{t}^{7}\\&&+ {\frac {1}{93555}}\,{\frac {\left (m+10\right )\left (m+9\right )}{{m}^{2}}}{t}^{9} -{\frac {691}{638512875}}\,{\frac {\left (m+12\right )\left (m+11 \right )}{{m}^{2}}}{t}^{11}\\&&+ {\frac {2}{18243225}}\,{\frac {\left (m+14\right )\left (m+13\right )}{{m}^{2}}}{t}^{13} -\cdots\;. \end{eqnarray*} In particular, in the one-dimensional case we have \begin{equation}\label{a40b} \kappa_{11}(r)=\frac{1}{2}r^2 - \frac{1}{36}r^6 +\frac{1}{720}r^{10} -\frac{1}{16800}r^{14} + \cdots\;.\end{equation} \subsection{Pair correlation in higher codimension}\label{explicit2} Next we compute the two-point correlation functions for the case $k=2$. For $k>1$, we have \begin{equation}\label{b1} A=(A^{jp}_{j'p'})=(\delta_{jj'}A^p_{p'})\,,\quad B=(B^{jp}_{j'p'q'})=(\delta_{jj'}B^{p}_{p'q'})\,,\quad C=(C^{jpq}_{j'p'q'})=(\delta_{jj'}C^{pq}_{p'q'})\,,\end{equation} where $A^p_{p'},B^{p}_{p'q'},C^{pq}_{p'q'}$ are given by (\ref{a33}). It follows that \begin{equation}\label{b2} \Lambda=(\Lambda^{jpq}_{j'p'q'})=(\delta_{jj'}\Lambda^{pq}_{p'q'})\,,\end{equation} where $\Lambda^{pq}_{p'q'}$ is given by (\ref{a35}). By (\ref{a28}), \begin{equation}\label{a41} K^\infty_{2km}(z^1,z^2)=\frac{1}{\pi^{2k}(e^u-1)^k}\left\langle \det\left|a_j^1\overline{a_{j'}^1}\right|_{j,j'=1,\dots,k} \det\left|a_j^2\overline{a_{j'}^2}\right|_{j,j'=1,\dots,k} \right\rangle_{\Lambda},\quad a_j^p\overline{a_{j'}^{p}}=\sum_{q=1}^m a^p_{jq}\overline{a^{p}_{j'q}}\,, \end{equation} where $u=r^2=|z^1-z^2|^2$ as before. Observe that the random variables $a^p_{jq}$ and $\overline{a^{p'}_{j'q'}}$ are independent if either $j\not= j'$ or $q\not= q'$. Recalling (\ref{nslzc}), we write \begin{equation} \widetilde K^\infty_{2km}(z^1,z^2) =\kappa_{km}(|z^1-z^2|)\,. \end{equation} When $k=2$, (\ref{a41}) reduces to the following \begin{equation}\label{a42} \kappa_{2m}(r)=\frac {\left\langle\left[ (a^1_1\overline{a^1_1})( a^1_2\overline{a^1_2}) -(a^1_1\overline{a^1_2})( a^1_2\overline{a^1_1})\right] \left[ (a^2_1\overline{a^2_1})( a^2_2\overline{a^2_2}) -(a^2_1\overline{a^2_2})( a^2_2\overline{a^2_1})\right] \right\rangle_{\Lambda}}{m^2(m-1)^2(e^u-1)^2}\,. \end{equation} By the Wick formula, \begin{equation} \kappa_{2m}(r)=\frac{d_{11}-d_{21}-d_{12}+d_{22}}{ m^2(m-1)^2(e^u-1)^2}\,, \end{equation} where \begin{equation}\label{a43} \begin{array}{rl} d_{11}= &\displaystyle\left\langle (a^1_1\overline{a^1_1})( a^1_2\overline{a^1_2}) (a^2_1\overline{a^2_1})( a^2_2\overline{a^2_2}) \right\rangle_{\Lambda} =\sum_{\alpha,\beta,\gamma,\delta}\left\langle a^1_{1\alpha}\overline{a^1_{1\alpha}} a^1_{2\beta}\overline{a^1_{2\beta}} a^2_{1\gamma}\overline{ a^2_{1\gamma}} a^2_{2\gamma}\overline{ a^2_{2\gamma}} \right\rangle_\Lambda\\ =&\displaystyle \sum_{\alpha,\beta,\gamma,\delta} \Lambda^{1\alpha}_{1\alpha}\Lambda^{1\beta}_{1\beta} \Lambda^{2\gamma}_{2\gamma}\Lambda^{2\delta}_{2\delta} \ + 2\sum_{\alpha,\beta,\gamma} \Lambda^{1\alpha}_{1\alpha}\Lambda^{1\beta}_{2\beta} \Lambda^{2\beta}_{1\beta}\Lambda^{2\gamma}_{2\gamma} \ + \sum_{\alpha,\beta} \Lambda^{1\alpha}_{2\alpha}\Lambda^{2\alpha}_{1\alpha} \Lambda^{1\beta}_{2\beta}\Lambda^{2\beta}_{1\beta}\\ =&\left[\left(\sum_{q}\Lambda^{1q}_{1q}\right) \left(\sum_{q}\Lambda^{2q}_{2q}\right) \ + \sum_q \Lambda^{1q}_{2q}\Lambda^{2q}_{1q}\right]^2\,; \end{array} \end{equation} similarly, \begin{equation}\label{a43a} \begin{array}{rl} d_{12}= &\displaystyle \left\langle (a^1_1\overline{a^1_1})( a^1_2\overline{a^1_2}) (a^2_1\overline{a^2_2})( a^2_2\overline{a^2_1}) \right\rangle_{\Lambda}\\ =&\left(\sum_q \left[\Lambda^{2q}_{2q}\right]^2\right)\left(\sum_q \Lambda^{1q}_{1q}\right)^2 + 2\left(\sum_q\Lambda^{2q}_{2q}\Lambda^{2q}_{1q}\Lambda^{1q}_{2q}\right) \left(\sum_q\Lambda^{1q}_{2q}\right) +\sum_q \left[\Lambda^{2q}_{1q}\Lambda^{1q}_{1q}\right]^2\,, \end{array} \end{equation} \begin{equation}\label{a43b} \begin{array}{rl} d_{21}= &\displaystyle \left\langle (a^1_1\overline{a^1_2})( a^1_2\overline{a^1_1}) (a^2_1\overline{a^2_1})( a^2_2\overline{a^2_2}) \right\rangle_{\Lambda}\\ =&\left(\sum_q \left[\Lambda^{1q}_{1q}\right]^2\right)\left(\sum_q \Lambda^{2q}_{2q}\right)^2 + 2\left(\sum_q\Lambda^{1q}_{1q}\Lambda^{2q}_{1q}\Lambda^{1q}_{2q}\right) \left(\sum_q\Lambda^{2q}_{2q}\right) +\sum_q \left[\Lambda^{2q}_{1q}\Lambda^{1q}_{2q}\right]^2\,, \end{array} \end{equation} \begin{equation}\label{a43c} \begin{array}{rl} d_{22}= &\displaystyle \left\langle (a^1_1\overline{a^1_2})( a^1_2\overline{a^1_1}) (a^2_1\overline{a^2_2})( a^2_2\overline{a^2_1}) \right\rangle_{\Lambda}\\ =&\left(\sum_q \left[\Lambda^{1q}_{1q}\right]^2\right)\left(\sum_q \left[\Lambda^{2q}_{2q}\right]^2\right) + 2\sum_q\Lambda^{1q}_{1q}\Lambda^{2q}_{2q}\Lambda^{2q}_{1q}\Lambda^{1q}_{2q} +\left(\sum_q \Lambda^{2q}_{1q}\Lambda^{1q}_{2q}\right)^2\,. \end{array} \end{equation} Substituting the values of the matrix elements of $\Lambda$ we then obtain \begin{equation} \begin{array}{ll} \kappa_{2m}(r)&= \displaystyle \frac{(m^2-m)e^{2u}+2(m-1)e^u+2}{(e^u-1)^2m(m-1)} -\frac{4ue^u[(m-1)e^u+1](m+1)}{(e^u-1)^3(m-1)m^2}\\ &\\ &\displaystyle +\frac {2u^2e^u[(m-1)e^{2u}+2me^u+1]}{(e^u-1)^4(m-1)m^2},\qquad u=r^2. \end{array} \end{equation} As $r\to\infty$, \begin{equation}\label{z} \kappa_{2m}= 1+\frac{2[r^4-2(m+1)r^2+m(m+1)]e^{-r^2}}{m^2} +O(r^4e^{-2r^2})\,. \end{equation} As $r\to 0$, \begin{equation} \begin{array}{rl} \kappa_{2m}(r)&\displaystyle =\frac{m-2}{m}\,r^{-4}+\frac{m-2}{m}\,r^{-2} +\frac{5m^2-7m+12}{12(m-1)m} +\frac{(m-2)(m+2)(m+1)}{12(m-1)m^2}\,r^2\\ &\\ &\displaystyle +\frac{(m+3)(m+2)}{240(m-1)m}\,r^4 -\frac{(m-2)(m+4)(m+3)}{720(m-1)m^2}\,r^6+\dots \end{array} \end{equation} When $m=2$ the asymptotics reduce to \begin{equation} \kappa_{22}(r)=\frac{3}{4}+\frac{r^4}{24}-\frac{r^8}{288} +\frac{r^{12}}{4800}-\frac{r^{16}}{96768}+\dots, \end{equation} and in this case $\kappa_{22}$ is a series in $r^4$. \bigskip
\section{Introduction} \label{sec:intro} The study involved in the attempt to detect gravitational waves represents one of the most exciting field in modern physics. The detection will provide a decisive test of general relativity and open a new window for astrophysical observation. Unfortunately, in spite a great deal of effort, no one has yet succeeded in the detection of the gravitational waves. However ground-based laser interferometers, such as LIGO,\cite{abramovici1992} VIRGO,\cite{bradaschia1990} TAMA\cite{kuroda1997} and GEO600,\cite{hough1992} are being constructed and will soon enter a stage of data taking. Strong sources of gravitational waves should be regions where gravity is general relativistic and where the velocities of bulk motion are near the speed of light. Many researchers have theoretically investigated coalescing and colliding black holes and neutron stars as candidates for detection. In addition, gravitational waves emitted during stellar collapse to black holes have been studied by perturbative calculations of a spherically symmetric background.\cite{CPM,Seidel} \ Because of the existence of event horizon, the emitted gravitational waves are dominated by the quasi-normal modes of the Schwarzschild black hole in the asymptotic region. Some other realistic candidates exist, e.g., supernovae, rapidly rotating neutron stars, relic gravitational waves of the early universe, stochastic gravitational waves from cosmic strings, and so on. In this paper we attempt to investigate whether a naked singularity, if such exists, is a strong source of gravitational radiation, and, moreover, to understand the dynamics and observational meaning of naked singularity formation. Several researchers have shown that the final fate of gravitational collapse is not always a singularity covered by an event horizon. For example, in the Lema\^{\i}tre-Tolman-Bondi (LTB) spacetime,\cite{Tolman34,Bondi} a naked shell-focusing singularity appears from generic initial data for spherically symmetric configurations of the rest mass density and a specific energy of the dust fluid.\cite{Eardley,Christodoulou,Newman,Joshi-Dwivedi} \ The initial functions in the most general expandable form have been considered.\cite{Jhingan:1997ia} \ The matter content in such a spacetime may satisfy even the dominant energy condition. In this case with a small disturbance of the spacetime, very short wavelength gravitational waves, which are created in the high density region around a singularity, may propagate to the observer outside the dust cloud because of the absence of an event horizon. If this is true, extremely high energy phenomena which cannot be realized by any high energy experiment on Earth can be observed. In this case information regarding the physics of so-called `quantum gravity' may be obtained. Also, these waves may be so intense that they destroy the Cauchy horizon. In this paper the generation of gravitational waves during the collapse of spherical dust ball with small rotational motion is considered. Nakamura, Shibata and Nakao\cite{nsn1993} have suggested that a naked singularity may emit considerable gravitational wave radiation. This was proposed from the estimate of gravitational radiation from a spindle-like naked singularity. They modeled the spindle-like naked singularity formation in gravitational collapse by the Newtonian prolate dust collapse for a sequence of general relativistic, momentarily static initial data. It should be noted that the system they considered is different from that considered in this article and that their result is controversial. There are numerical analyses that both support and do not support the results of Nakamura, Shibata and Nakao for prolate collapse\cite{Shapiro} and for cylindrical collapse.\cite{Echeverria:1993wf,Chiba} Due to the non-linear nature of the problem, it is difficult to analytically solve the Einstein equation. Therefore, numerical methods will provide the final tool. However, its singular behavior makes accurate numerical analysis very difficult at some stage. In this article, we investigate odd-parity linear gravitational waves from the collapse of an inhomogeneous spherically symmetric dust cloud. Even for the linearized Einstein equation, we must perform the numerical integration. However, in contrast to the numerical simulation of the full Einstein equation, high precision is guaranteed for the numerical integration of the linearized Einstein equation, even in the region with extremely large spacetime curvature. Furthermore, the linear stability of known examples of naked singularity formation is necessary as a first step to understand the general dynamics near the ``naked singularity formation.'' Recently, Iguchi, Nakao and Harada \cite{Iguchi} (INH) studied odd-parity metric perturbations around a naked singularity in the LTB spacetime. In INH, it was found that the propagation of odd-parity gravitational waves is not affected by collapse of a dust cloud, even if there appears a central naked singularity. When we consider the generation of gravitational waves from the dust collapse, we should analyze the perturbations including their matter part. Here we investigate the evolution equation of the odd-parity quadrupole mode for metric and matter perturbations. This matter perturbation relates to small rotational motion, and it produces gravitational waves during the collapse. We follow the derivation of the evolution equations of the perturbations and the numerical method to integrate these equations of INH. We investigate the time evolutions of the gauge invariant metric variables at the symmetric center, where a naked singularity appears, and at a constant circumferential radius $R$. We show that the gauge invariant variable diverges only at the center and it does not propagate to the outside. As we know, the LTB spacetime is one candidate as a counterexample of the cosmic censorship hypothesis (CCH), which was introduced by Penrose.\cite{Penrose69,Penrose79} \ The CCH is a very helpful assumption, because various theorems on the properties of black holes were proved under this assumption. Here the precise formulation and validity of cosmic censorship is not our concern. We only consider the situation in which the extremely high-density and large-curvature region can be seen by some observer from the gravitational collapse. Such a situation can be regarded as a naked singularity in a practical sense since we are not yet able to predict phenomena beyond the Planck scale. However, here it should be noted that even from this practical point of view, the stability of the Cauchy horizon with respect to gravitational-wave perturbations still has an important physical meaning. Since in the system considered here, the mode propagating with the speed of light is the gravitational waves only, the instability of the Cauchy horizon implies that the gravitational wave introduces an extremely large spacetime curvature along the null hypersurface near the Cauchy horizon from the region near the central singularity. Hence, if the Cauchy horizon in this system is unstable, naked singularity formation of this type might be a strong source of gravitational radiation. Here we comment on the problem motivated in INH, i.e., ``nakedness of the naked singularity.'' Odd-parity matter perturbations are produced by rotational motion of the dust cloud. Therefore the growth of the matter perturbation would cause the centrifugal force to dominate the radial motion of the cloud. Hence, an angular momentum bounce would occur. If this occurs at the center, the central singularity will disappear. This situation seems to be inevitable for the dipole mode of odd-parity matter perturbations. In the case of a spherically symmetric system composed of counter-rotating particles, the angular-momentum bounce can actually prevent the naked singularity formation.\cite{Harada} \ However, for a system without spherical symmetry, it is still an open question how the non-linear asphericity works in the final stage of singularity formation with rotational matter perturbations because of the difficulty of both the analytic and numerical treatment. Further, if the initial matter perturbation is sufficiently small, it might still be possible that the radius of the spacetime curvature becomes the Planck length. In this case, the behavior of other modes of perturbations is crucial to understand the classical dynamics in the region of the Planck scale. As will be shown below, since there is no gravitational wave of the dipole mode and the quadrupole mode generated in the dust cloud does not propagate to the outside, the Cauchy horizon is marginally stable against odd-parity perturbations originating in the aspherical rotational motion of matter. The paper is organized as follows: In \S \ref{sec:be}, the basic equations are developed; in \S \ref{sec:results}, the numerical results are presented; these results are discussed in \S \ref{sec:discussion}; in \S \ref{sec:summary}, we summarize our results. We adopt the geometrized units, in which $c=G=1$. The signature of the metric tensor and sign convention of the Riemann tensor follow Ref.~\cite{MTW}. \section{Basic equations} \label{sec:be} We consider the evolution of odd-parity perturbations of the LTB spacetime up to linear order. The background spacetime and perturbation method are described in INH. \cite{Iguchi} \ In this section we briefly review them. Using the synchronous coordinate system, the line element of the background LTB spacetime can be expressed in the form \begin{equation} \label{bgmetric} d{\bar s}^{2}= {\bar g}_{\mu\nu}dx^{\mu}dx^{\nu}\equiv -dt^2+A^{2}(t,r)dr^{2} +R^2(t,r)(d\theta^{2}+\sin^{2}\theta d\phi^{2}). \end{equation} The energy-momentum tensor for the dust fluid is \begin{equation} \label{bgmatter} {\bar T}^{\mu\nu} = {\bar \rho}(t,r){\bar u}^{\mu}{\bar u}^{\nu}, \end{equation} where ${\bar\rho}(t,r)$ is the rest mass density and ${\bar u}^{\mu}$ is the 4-velocity of the dust fluid. In the synchronous coordinate system, the unit vector field normal to the spacelike hypersurfaces is a geodesic, and there is a freedom of which timelike geodesic field is adopted as the hypersurface unit normal. Using this freedom, we can always set ${\bar u}^{\mu}=\delta^{\mu}_{0}$, since the 4-velocity of the spherically symmetric dust fluid is tangent to an irrotational timelike geodesic field. Then the Einstein equations and the equation of motion for the dust fluid reduce to the simple equations \begin{eqnarray} A &=& \frac{\partial_{r}R}{\sqrt{1+f(r)}}, \label{eq:A} \\ {\bar \rho}(t,r) &=& \frac{1}{8\pi} \frac{1}{R^2 \partial_{r}R}{dF(r)\over dr},\label{eq:einstein} \\ (\partial_{t}R)^2-\frac{F(r)}{R} &=& f(r),\label{eq:energyeq} \end{eqnarray} where $f(r)$ and $F(r)$ are arbitrary functions of the radial coordinate, $r$. From Eq.\ (\ref{eq:einstein}), $F(r)$ is related to the Misner-Sharp mass function,\cite{Misner} $m(r)$, of the dust cloud in the manner \begin{equation} \label{mass} m(r) = 4\pi \int_0^{R(t,r)}{\bar \rho}(t,r)R^2dR = 4\pi \int_0^r{\bar \rho}(t,r)R^2\partial_{r}R dr =\frac{F(r)}{2}. \end{equation} Hence Eq.\ (\ref{eq:energyeq}) might be regarded as the energy equation per unit mass. This means that the other arbitrary function, $f(r)$, is recognized as the specific energy of the dust fluid. The motion of the dust cloud is completely specified by the function $F(r)$ (or equivalently, the initial distribution of the rest mass density, ${\bar \rho}$) and the specific energy, $f(r)$. When we restrict our calculation to the case that the symmetric center, $r=0$, is initially regular, the central shell focusing singularity is naked if and only if $\partial_{r}^{2}{\bar\rho}|_{r=0}<0$ is initially satisfied for the marginally bound collapse, $f(r)=0$. \cite{Singh-Joshi,Jhingan} \ For a collapse that is not marginally bound, there exists a similar condition as an inequality for a value depending on the functional forms of $F(r)$ and $f(r)$.\cite{Newman,Singh-Joshi,Jhingan} \ Let us now derive the perturbation equations. The perturbed metric tensor is expressed in the form \begin{equation} \label{metric} g_{\mu\nu} = {\bar g}_{\mu\nu}+h_{\mu\nu}, \end{equation} where ${\bar g}_{\mu\nu}$ is a metric tensor of a spherically symmetric background spacetime and $h_{\mu\nu}$ is a perturbation. The energy-momentum tensor is written in the form \begin{equation} \label{ene-mom} T_{\mu\nu} = \bar{T}_{\mu\nu} + \delta T_{\mu\nu}, \end{equation} where $\bar{T}_{\mu\nu}$ is a background quantity and $\delta T_{\mu\nu}$ is a perturbation. By virtue of the spherical symmetry of the background spacetime, ${\bar T}_{\mu\nu}$ can be expressed in the form \begin{eqnarray} \label{GS-matter} \bar{T}_{\mu\nu}dx^{\mu}dx^{\nu} &=& \bar{T}_{ab}dx^adx^b+\frac{1}{2}\bar{T}_B^{~B}R^2(t,r)d\Omega^2, \end{eqnarray} where the subscripts and superscripts $a,b,\cdots$ represent $t$ and $r$, while $A,B,\cdots$ represent $\theta$ and $\phi$. The odd-parity perturbations of $h_{\mu\nu}$ and $\delta T_{\mu\nu}$ are expressed in the form \begin{eqnarray} \label{GS-pmetric} h_{\mu\nu} &=& \left(\begin{array}{ccc} 0 & 0 & h_0(t,r)\Phi^m_{l~B} \\ & 0 & h_1(t,r)\Phi^m_{l~B} \\ \mbox{sym} & & h_2(t,r)\chi^m_{l~AB} \end{array} \right), \end{eqnarray} \begin{eqnarray} \label{GS-pmatter} \delta T_{\mu\nu} &=& \left(\begin{array}{ccc} 0 & 0 & t_0(t,r)\Phi^m_{l~B} \\ & 0 & t_1(t,r)\Phi^m_{l~B} \\ \mbox{sym} & & t_2(t,r)\chi^m_{l~AB} \end{array} \right) , \end{eqnarray} where $\Phi^m_{l~B}$ and $\chi^m_{l~AB}$ are odd-parity vector and tensor harmonics associated with the spherical symmetry of the background spacetime.\cite{Regge} \ We set all the arbitrary constants in the definitions of harmonics to unity. We then introduce the gauge-invariant variables defined by Gerlach and Sengupta.\cite{Gerlach-Sengupta} \ The metric variables are given by \begin{equation} \label{ka} k_a = h_a-\frac{1}{2}R^2\partial_a\left(\frac{h_2}{R^2}\right). \end{equation} The matter variables are given by the combinations \begin{eqnarray} \label{la} L_a &=& t_a-\frac{1}{2}\bar{T}_B^{~B}h_a ,\\ \label{l} L &=& t_2-\frac{1}{2}\bar{T}_B^{~B}h_2 . \end{eqnarray} In the LTB case, the odd-parity gauge-invariant matter variables become \begin{eqnarray} \label{La-L} L_0 = \bar{\rho}(t,r) U(t,r)~~~~~~{\rm and}~~~~~~ L_1 = L = 0, \end{eqnarray} where $U(t,r)$ represents the perturbation of the 4-velocity as $\delta u_\mu = (0,0,U(t,r)\Phi_{l~B}^m)$. The evolution equation for the matter variable (Eq.\ (3$\cdot$19) in INH), \begin{equation} \label{T-Bconserve} \partial_t\left(AR^2L_0\right) = 0, \end{equation} is easily integrated, and we obtain \begin{equation} \label{kakuundouryou} L_0 = \frac{1}{AR^2}{dJ(r)\over dr}, \end{equation} where $J(r)$ is an arbitrary function depending only on $r$. From Eqs.\ (\ref{eq:einstein}), (\ref{La-L}), and (\ref{kakuundouryou}), we obtain the relation \begin{eqnarray} \label{U(r)} U(t,r)&=&8\pi \sqrt{1+f(r)}\frac{dJ(r)/dr}{dF(r)/dr} \\ &\equiv& U(r), \end{eqnarray} so $U(t,r)$ is independent of the time coordinate $t$. We introduce a gauge-invariant variable for the metric as \begin{equation} \label{psis} \psi_s \equiv {1\over A}\left[\partial_t \left(\frac{k_1}{R^2}\right)-\partial_r \left(\frac{k_0}{R^2}\right)\right]. \end{equation} The metric perturbation variables, $k_0$ and $k_1$, are reconstructed from the linearized Einstein equations, \begin{eqnarray} \label{wave2} \partial_r\left(R^4\psi_s\right)+A\left(l-1\right) \left(l+2\right)k_0 &=& 16\pi AR^2L_0, \\ \label{wave3} \partial_t\left(R^4\psi_s\right)+\frac{1}{A} \left(l-1\right)\left(l+2\right)k_1 &=& 0, \end{eqnarray} by substituting $\psi_s$. It was shown in INH that $\psi_s$ is closely connected to the tetrad components of the magnetic part of the Weyl tensor. These components diverge where and only where $\psi_s$ diverges. From the linearized Einstein equations we obtain the linearized evolution equation for the odd-parity perturbation as \begin{eqnarray} \label{wave4} \partial_{t}\left(\frac{A}{R^2}\partial_{t} \left(R^4\psi_s \right)\right)&-&\partial_{r} \left(\frac{1}{AR^2}\partial_{r} \left(R^4\psi_s\right)\right) + \left(l-1\right) \left(l+2\right)A\psi_s \nonumber\\ &=& -16\pi\partial_{r}\left(\frac{1}{AR^2}{dJ\over dr}\right). \end{eqnarray} The regularity conditions at the center are also considered in INH. The result for the matter perturbation $L_0$ is given by \begin{equation} \label{r-0-L} L_0 \longrightarrow L_c(t)r^{l+1}+O(r^{l+3}). \end{equation} Therefore the matter perturbation variables vanish at the regular center independent of the value of $l$. The metric variable $\psi_s$ behaves near the center as \begin{eqnarray} \label{r-0-psis} \psi_{s} &\longrightarrow& \psi_{sc}(t)r^{l-2}+O(r^l) ~~~~~~~~~~~~\mbox{for $l\geq 2$}, \\ \label{r-0-psis1} \psi_{s} &\longrightarrow& \psi_{sc}(t)r+O(r^3) ~~~~~~~~~~~~~~~~\mbox{for $l=1$}. \end{eqnarray} From the above equations, it is shown that only the quadrupole mode, $l=2$, of $\psi_{s}$ does not vanish at the regular center. \section{Results} \label{sec:results} We numerically solve the wave equation (\ref{wave4}) in the case of marginally bound collapse, $f(r)=0$, and the quadrupole mode, $l=2$. We follow numerical techniques employed in INH to integrate the time evolution of the perturbations. The numerical code is essentially the same as that in INH, which is tested by comparison with the analytic solution for the Minkowski background. By virtue of the relation $f(r)=0$, we can easily integrate Eq.\ (\ref{eq:energyeq}) and obtain \begin{equation} \label{f=0} R(t,r) = \left(\frac{9F}{4}\right)^{1/3}[t_0(r)-t]^{2/3}, \end{equation} where $t_0(r)$ is an arbitrary function of $r$. The naked singularity formation time is $t_0=t_0(0)$. Using the freedom for the scaling of $r$, we choose $R(0,r)=r$. This scaling of $r$ corresponds to the following choice of $t_{0}(r)$: \begin{equation} \label{t0} t_0(r) = \frac{2}{3\sqrt{F}}r^{3/2}. \end{equation} Here note that, from Eq.\ (\ref{eq:A}), the background metric variable, $A$, is equal to $\partial_{r}R$. Then, the wave equation (\ref{wave4}) becomes \begin{eqnarray} \label{wave-eq} \frac{\partial^2 \psi_s}{\partial t^2}-\frac{1}{(\partial_{r}R)^2} \frac{\partial^2 \psi_ s}{\partial r^2} &=& \frac{1}{(\partial_{r}R)^{2}}\left(6\frac{\partial_{r}R}{R} -\frac{\partial_{r}^{2}R}{\partial_{r}R}\right)\frac{\partial \psi_s}{\partial r}-\left(6\frac{\partial_{t}R}{R} +\frac{\partial_{t}\partial_{r}R}{\partial_{r}R}\right) \frac{\partial \psi_s}{\partial t} \nonumber \\ & &-4\left[\left(\frac{\partial_{t}\partial_{r}R}{\partial_{r}R}\right) \frac{\partial_{t}R}{R} +\frac{1}{2}\left(\frac{\partial_{t}R}{R}\right)^2\right]\psi_s \nonumber \\ & &-\frac{16\pi}{(\partial_r R)R^2}\partial_{r} \left(\frac{r^2 \rho (r) U(r)}{(\partial_r R)R^2}\right), \end{eqnarray} where $\rho (r) = \bar{\rho}(0,r)$ is the density profile at $t=0$. We solve this partial differential equation numerically. Before getting into the detailed explanation of the numerical techniques, we comment on the behavior of the matter perturbation variable $L_0$ around a naked singularity on the slice $t=t_0$. The regularity conditions of $L_0$ and $\bar{\rho}$ determine the behavior of $U(r)$ near the center as \begin{equation} \label{dJdr-c} U(r) \propto r^{l+1}. \end{equation} This property does not change even if a central singularity appears. However, the $r$ dependence of $R$ and $A$ near the center changes at that time. Assuming a rest mass density profile of the form \begin{equation} \rho(r) = \rho_0 + \rho_n r^n + \cdots , \end{equation} we obtain the relation \begin{equation} t_0(r) \propto t_0 + t_n r^n \end{equation} from Eqs.\ (\ref{eq:einstein}) and (\ref{t0}), where $n$ is a positive even integer. After substituting this relation into Eq.\ (\ref{f=0}), the lowest order term is absent from the square brackets of it. Then we obtain the behavior of $R$ and $A$ around the central singularity as \begin{equation} \label{singr} R(t_0,r) \propto r^{1+ {2\over3} n}, \end{equation} and \begin{equation} \label{singa} A(t_0,r) \propto r^{{2\over3} n} \end{equation} on the slice $t=t_0$. As a result, we obtain the $r$ dependence of $L_0$ around the center when the naked singularity appears as \begin{equation} L_0(t_0,r) \propto r^{l-2n+1}. \end{equation} For example, if $l=2$ and $n=2$, then $L_0$ is inversely proportional to $r$ and diverges toward the central naked singularity. Therefore the source term of the wave equation is expected to have a large magnitude around the naked singularity. Thus the metric perturbation variable $\psi_s$ as well as matter variable $L_0$ may diverge toward the naked singularity. Instead of the $(t,r)$ coordinate system, we introduce a single-null coordinate system, $(u,r')$, where $u$ is an outgoing null coordinate and chosen so that it agrees with $t$ at the symmetric center and we choose $r'=r$. We perform the numerical integration along two characteristic directions. Therefore we use a double null grid in the numerical calculation. A detailed explanation of these coordinates is given in INH. By using this new coordinate system, $(u,r')$, Eq.\ (\ref{wave-eq}) is expressed in the form \begin{eqnarray} \label{dphis/dlambda} \frac{d\phi_s}{du} &=& -\frac{\alpha}{R}\left[3\partial_{r}R +{1\over2}R(\partial_{t}R)\partial_{t}\partial_{r}R -\frac{5}{4}(\partial_{t}R)^{2}\partial_{r}R\right]\psi_s \nonumber \\ & & -\frac{\alpha}{2}\left[\frac{\partial_{r}^{2}R}{(\partial_{r}R)^2} -\frac{2}{R}\left(1-\partial_{t}R\right) \right]\phi_s -\frac{8\pi \alpha}{R}\partial_{r}\left( \frac{r^2 \rho (r) U(r)}{(\partial_rR)R^2}\right),\\ \label{delpsi/delrprime} \partial_{r'}\psi_{s}&=& \frac{1}{R}\phi_s -3\frac{\partial_{r}R}{R} \left(1+\partial_{t}R\right)\psi_s, \end{eqnarray} where the ordinary derivative on the left-hand side of Eq.\ (\ref{dphis/dlambda}) and the partial derivative on the left-hand side of Eq.\ (\ref{delpsi/delrprime}) are given by \begin{eqnarray} \frac{d}{du} &=& \partial_u + \frac{dr'}{du}\partial_{r'} = \partial_u-\frac{\alpha}{2\partial_r R}\partial_{r'} \nonumber \\ &=&\frac{\alpha}{2}\partial_t-\frac{\alpha}{2\partial_r R}\partial_{r},\\ \label{r'a} \partial_{r'} &=& -\frac{(\partial_r u)_t}{(\partial_t u)_r}\partial_{t}+\partial_{r}= (\partial_{r}R)\partial_{t}+\partial_{r}, \end{eqnarray} respectively. Also, $\phi_s$ is defined by Eq.\ (\ref{delpsi/delrprime}) and $\alpha$ is given by \begin{equation} \alpha\equiv {1\over (\partial_{t} u)_r}. \end{equation} We integrate Eq.\ (\ref{dphis/dlambda}) using the scheme of an explicit first order difference equation. We use the trapezoidal rule, \begin{equation} \psi_{s j+1}=\psi_{s j}+\frac{\Delta r'}{2}\left((\partial_{r'}\psi_{s})_{j}+(\partial_{r'}\psi_{s})_{j+1}\right), \end{equation} to integrate Eq.\ (\ref{delpsi/delrprime}). For the boundary condition at the center we demand that $\psi_s$ behaves as $\psi_{sc}(t)+\psi_{s2}(t)r^2$ on a surface of $t=\mbox{const}$. We numerically realize this condition by two-step interpolation. First the values of $\psi_s$ are derived at two points on the surface of $t=\mbox{const}$ from the interpolation on the slices of $u=\mbox{const}$. Next, using these two values, the central value of $\psi_s$ is derived from the interpolation on the slice of $t=\mbox{const}$. Another way to determine the central value of $\psi_s$ is as follows. We first obtain the central value of $\phi_s$ from Eq.\ (\ref{dphis/dlambda}). From Eq.\ (\ref{delpsi/delrprime}) and the boundary conditions, the relation of $\psi_s$ and $\phi_s$ at the center is given by \begin{equation} \label{phi-psi} \phi_s = 3 \partial_{r'}R \psi_s. \end{equation} Using this relation the central value of $\psi_s$ is obtained. In our numerical analyses the results of these two methods agree well. We assume $\psi_s$ vanishes on the initial null hypersurface. Therefore, there exist initial ingoing waves which offset the waves produced by the source term on the initial null hypersurface. In INH, it is confirmed that this type of the initial ingoing waves propagate through the dust cloud without net amplification even when they pass through the cloud just before the appearance of the naked singularity. Therefore those parts of the metric perturbations would not diverge at the center. We adopt the initial rest mass density profile \begin{equation} \label{density} \rho(r)=\rho_0 \frac{1+ \exp\left(-\frac{1}{2}\frac{r_1}{r_2}\right)} {1+ \exp\left(\frac{r^n -r_1^n}{2 r_1^{n-1}r_2}\right)}, \end{equation} where $\rho_0$, $r_1$ and $r_2$ are positive constants and $n$ is a positive even integer. As a result the dust fluid spreads all over the space. However, if $r \gg r_1,r_2$, then $\rho(r)$ decreases exponentially, so that the dust cloud is divided into the core part and the envelope which would be considered as the vacuum region essentially. We define a core radius as \begin{equation} \label{cradious} r_{\mbox{\scriptsize core}}=r_1+\frac{r_2}{2}. \end{equation} If we set $n=2$, there appears a central naked singularity. This singularity becomes locally or globally naked depending on the parameters ($\rho_0,r_1,r_2$). However, if the integer $n$ is greater than $2$, the final state of the dust cloud is a black hole independently of the parameters. Then we consider three different density profiles connected with three types of the final state of the dust cloud, globally and locally naked singularities and a black hole. The outgoing null coordinate $u$ is chosen so that it agrees with the proper time at the symmetric center. Therefore, even if the black hole background is considered, we can analyze the inside of the event horizon. Corresponding parameters are given in Table \ref{tab:parameter}. Using this density profile, we numerically calculate the total gravitational mass of the dust cloud $M$. In our calculation we adopt the total mass $M$ as the unit of the variables. The source term of Eq.\ (\ref{dphis/dlambda}), \begin{equation} \label{souce} S(t,r)=-\frac{8\pi \alpha}{R}\partial_{r}\left( \frac{r^2 \rho (r) U(r)}{(\partial_rR)R^2}\right), \end{equation} is determined by $U(r)$. As mentioned above, the constraints on the functional form of $U(r)$ are given by the regularity condition of $L_0$. From Eq.\ (\ref{dJdr-c}), $U(r)$ should be proportional to $r^{l+1}$ toward the center. We localize the matter perturbation near the center to diminish the effects of the initial ingoing waves. Therefore we define $U(r)$ such that \begin{equation} \label{dJdr} r^2 \rho(r) U(r) = \left\{ \begin{array}{ccc} U_0 \left( \frac{r}{r_b} \right)^5 \left(1-\left ( \frac{r}{r_b} \right)^2 \right)^5 &{\mbox{for}} & 0\leq r\leq r_b,\\ 0 &{\mbox{for}} & r>r_b, \end{array} \right. \end{equation} where $U_0$ and $r_b$ are arbitrary constants. In our numerical calculation we chose $r_b$ as $r_{\mbox{\scriptsize core}}/2$. This choice of $r_b$ has no special meaning, and the results of our numerical calculations are not sensitive to it. First we observe the behavior of $\psi_s$ at the center. The results are plotted in Fig.\ \ref{fig:center}. The initial oscillations correspond to the initial ingoing waves. After these oscillations, $\psi_s$ grows proportional to $(t_0-t)^{- \delta}$ for the naked singularity cases near the formation epoch of the naked singularity. For the case of black hole formation, $\psi_s$ exhibits power-law growth in the early part. Later its slope gradually changes but it grows faster than in the case of the naked singularity. For the naked singularity cases, the power-law indices $\delta$ are determined by $(t_0-t)\dot{\psi_s}/\psi_s$ locally. The results are shown in Fig.\ \ref{fig:index}. From this figure we read the final indices as 5/3 for both naked cases. Therefore the metric perturbations diverge at the central naked singularity. \begin{table} \caption{Parameters of initial density profiles, power law indices, and damped oscillation frequencies.} \label{tab:parameter} \begin{tabular}{cccccccc} & final state & $\rho_0$ & $r_1$ & $r_2$ & $n$ &power index& damped oscillation frequency\\ \hline (a) & globally naked & $1 \times 10^{-2}$ & 0.25 & 0.5 & 2 & 5/3 & --- \\ (b) & locally naked & $1 \times 10^{-1}$ & 0.25 & 0.5 & 2 & 5/3 & 0.37+0.089$i$\\ (c) & black hole & $2 \times 10^{-2}$ & 2 & 0.4 & 4 & --- & 0.37+0.089$i$\\ \end{tabular} \end{table} We also observe the wave form of $\psi_s$ along the line of a constant circumferential radius outside the dust cloud. The results are shown in Figs.\ \ref{fig:outsideg}--\ref{fig:outsideb}. Figure \ref{fig:outsideg} displays the wave form of the globally naked case (a), Fig.\ \ref{fig:outsidel} displays the wave form of the locally naked case (b), and Fig.\ \ref{fig:outsideb} displays the wave form of the black hole case (c). The initial oscillations correspond to the initial ingoing waves. In the case of a locally naked singularity and black hole formation, damped oscillations dominate the gravitational waves. We read the frequencies and damping rates of these damped oscillations from Figs.\ \ref{fig:outsidel} and \ref{fig:outsideb} and give them in terms of complex frequencies as $0.37+0.089i$ for locally naked and black hole cases. These agree well with the fundamental quasi-normal frequency of the quadrupole mode $(2M\omega = 0.74734 + 0.17792 i)$ of a Schwarzschild black hole given by Chandrasekhar and Detweiler.\cite{Chandra} \ In the globally naked singularity case (a), we did not see this damped oscillation because of the existence of the Cauchy horizon. In all cases the gravitational waves generated by matter perturbations are at most quasi-normal modes of a black hole, which is generated outside the dust cloud. Therefore intense odd-parity gravitational waves would not be produced by the inhomogeneous dust cloud collapse. We should not expect that the central extremely high density region can be observed by this mode of gravitational waves. We can calculate the radiated power of the gravitational waves and thereby grasp the physical meaning of the gauge-invariant quantities.\cite{CPM} \ To relate the perturbation of the metric to the radiated gravitational power, it is useful to specialize to the radiation gauge, in which the tetrad components $h_{(\theta)(\theta)}-h_{(\phi)(\phi)}$ and $h_{(\theta)(\phi)}$ fall off as $O(1/R)$, and all other tetrad components fall off as $O(1/R^2)$ or faster. Note that, in vacuum at large distance, the spherically symmetric background metric is given by the Schwarzschild solution, where hereafter we adopt the Schwarzschild coordinates, \begin{equation} \label{Schmetric} ds^2 = -\left( 1-\frac{2M}{R}\right)d\tau ^2 +\left ( 1-\frac{2M}{R}\right)^{-1} dR^2 + R^2 \left( d\theta^{2}+\sin^{2}\theta d\phi^{2}\right) . \end{equation} The relation between the line elements Eq.\ (\ref{bgmetric}) and Eq.\ (\ref{Schmetric}) is given by the transfer matrix \begin{eqnarray} d\tau &=& \frac{1}{1-(\partial_t R)^2} (dt+\partial_r R \partial_t R dr), \\ dR &=& \partial_t Rdt + \partial_r R dr. \end{eqnarray} In this gauge, the metric perturbations in Eq. (\ref{GS-pmetric}) behave as \begin{eqnarray} h_0, h_1 &=& O\left(\frac{1}{R}\right), \\ h_2 &=& w(\tau - R_*) + O(1), \end{eqnarray} where \begin{equation} R_* = R + 2M \ln \left( \frac{R}{2M} - 1\right) + \mbox{const}. \end{equation} Then, the gauge-invariant metric perturbations (\ref{ka}) are calculated as \begin{eqnarray} k_0 &=& -\frac{1}{2}w^{(1)}R + O(1), \\ k_1 &=& \frac{1}{2}w^{(1)}R + O(1), \end{eqnarray} where $w^{(1)}$ denotes the first derivative of $w$ with respect to its argument. In this radiation gauge, the radiated power $P$ per unit solid angle is given by the formula which was derived by Landau and Lifshitz \cite{Landau} from their stress-energy pseudo-tensor: \begin{equation} \frac{dP}{d\Omega}=\frac{R^2}{16\pi}\left[\left(\frac{\partial h_{(\theta)(\phi)}}{\partial \tau}\right)^2 +\frac{1}{4}\left(\frac{\partial h_{(\theta)(\theta)}}{\partial \tau}-\frac{\partial h_{(\phi)(\phi)}}{\partial \tau}\right)^2\right]. \end{equation} For the axisymmetric mode, i.e., $m=0$, the above formula is reduced to \begin{equation} \frac{dP}{d\Omega}=\frac{1}{64\pi}(w^{(1)})^2A_l (\theta), \end{equation} where \begin{equation} A_l (\theta) \equiv \frac{2l+1}{4\pi}\sin^4\theta\left(\frac{d^2P_l(\cos\theta)}{(d\cos\theta)^2}\right)^2. \end{equation} Then, by using the gauge-invariant quantities and integrating over the all solid angles, the formula for the power of gravitational radiation is obtained in the following form: \begin{eqnarray} \frac{dP}{d\Omega}&=&\frac{1}{16\pi}\frac{k_0^2}{R^2}A_l (\theta)=\frac{1}{16\pi}\frac{k_1^2}{R^2}A_l (\theta),\\ P &=& \frac{1}{16\pi} B_l \frac{k_0^2}{R^2} = \frac{1}{16\pi} B_l \frac{k_1^2}{R^2} , \end{eqnarray} where \begin{equation} B_l \equiv \frac{(l+2)!}{(l-2)!}. \end{equation} Using Eq. (\ref{wave3}), the radiated power $P$ of the quadrupole mode is given by \begin{equation} P = \frac{3}{32\pi}R'^2\left[\partial_{\tau}\left(R^3 \psi_s\right)\right]^2. \end{equation} Figure \ref{fig:power} displays the time evolution of the radiated power $P$. The radiated power also has a finite value at the Cauchy horizon. The total energy radiated by odd-parity quadrupole gravitational waves during the dust collapse should not diverge. \section{Discussion} \label{sec:discussion} First we consider the behavior of the source term $S(t,r)$ around the naked singularity. From the regularity conditions and Eqs.\ (\ref{dJdr-c}), (\ref{singr}), and (\ref{singa}), the asymptotic behavior of the source term is obtained as \begin{equation} S(t,r) \propto r^{l-1} \end{equation} for $t<t_0$ and \begin{equation} S(t,r) \propto r^{l-\frac{8}{3}n-1}, \end{equation} at $t=t_0$. For example, in the case $l=2$ and $n=2$, the souce term behaves on $t=t_0$ as \begin{equation} S(t,r) \propto r^{-13/3}, \end{equation} and then it diverges at the center. Thus the divergency of $\psi_s$ at the center originates from the source term. To confirm this, we numerically integrate the source term along the ingoing null lines with respect to $u$ and estimate the central value of $\phi_s$. We define this `estimated' value as \begin{equation} \Phi_s \equiv \int S(t,r) du. \end{equation} Using Eq.\ (\ref{phi-psi}) we can define the estimated value of $\psi_s$ as \begin{equation} \Psi_s \equiv \frac{\Phi_s}{3 \partial_rR}. \end{equation} We plot it in Fig.\ \ref{fig:int} together with the corresponding $\psi_s$. The estimated value has the same power-law index of $\psi_s$. We conclude that the behavior of $\psi_s$ is determined by the source term in the dust cloud. We next consider the stability of the Cauchy horizon. We found that the metric perturbation produced by the source term does not propagate outside the dust cloud, except for quasi-normal ringing. The source term, which controls $\psi_s$, does not diverge at the Cauchy horizon. Therefore $\psi_s$ should not diverge at the Cauchy horizon and should not destroy it. Then, even if odd-parity perturbations are considered, it will not be the case that the LTB spacetime loses its character as a counterexample to CCH due to Cauchy horizon instability. Also, it does not seem that such collapse is a strong source of gravitational waves. In this paper, we have dealt with the marginally bound case. For the case of non-marginally bound collapse, the condition of the appearance of the central naked singularity is slightly different from that in the above case\cite{Singh-Joshi,Jhingan} and hence there is the possibility that the behavior of $\psi_{s}$ in this case is different from that in the marginally bound case. However, it is well known that the limiting behavior of the metric as $t \rightarrow t_0(r)$ is common for all the cases:\cite{Landau} \begin{equation} R \approx \left(\frac{9F}{4}\right)^{1/3}\left(t_0-t\right)^{2/3} ,~~~A \approx \left(\frac{2F}{3}\right)^{1/3} \frac{t_0'}{\sqrt{1+f}} \left(t_0-t\right)^{-1/3}. \end{equation} We conjecture that the results of the perturbed analysis for the non-marginal collapse would be similar to the results for the marginal bound. \section{Summary} \label{sec:summary} We have studied the behavior of the odd-parity perturbation in the LTB spacetime including the matter perturbation. For the quadrupole mode, where gravitational waves exist, we have numerically investigated the wave equation. For the case of naked singularity formation, the gauge-invariant metric variable, $\psi_s$, diverges according to a power law with power index 5/3 at the center. This power index is closely related to the behavior of the matter perturbation around the center. We have also observed $\psi_s$ at a constant circumferential radius. For the globally naked case, we cannot see intense gravitational waves propagated from the center just before the crossing of the Cauchy horizon. For the locally naked case, we have confirmed that there exist quasi-normal oscillations. As a result, we conclude that the type of singularity changes due to the odd-parity perturbation because $\psi_s$ diverges at the center. However, the Cauchy horizon is marginally stable against odd-parity perturbations. At the final stage of the collapse, the effects of the rotational motion are important and the centrifugal force might dominate the radial motion. If this is true, the central singularity would disappear when an odd-parity matter perturbation is introduced. For the dipole mode, such a situation seems to be inevitable. However, we should note that it is a non-trivial and open question how non-linear asphericity affects the final fate of the singularity-formation process. Further, in the case of initially sufficiently small aspherical perturbations, the radius of spacetime curvature at the center might reach the Planck length, and hence there is still the possibility that the naked singularity is formed there in a practical sense. However, as our present analysis has revealed, since the Cauchy horizon is stable with respect to odd-parity linear perturbations, there is little possibility that this collapse is a strong source of odd-parity gravitational waves. There remains important related works to be completed. The first problem is to consider the even-parity mode in which the metric and matter perturbations are essentially coupled with each other. We are now investigating this problem. Finally, we should consider the non-linear effects to complete this analysis. This problem will be analyzed in the future. \section*{Acknowledgements} We would like to thank T. Nakamura for helpful and useful discussions and N. Sugiyama for careful reading of the manuscript. We are also grateful to H. Sato and colleagues in the theoretical astrophysics group at Kyoto University for useful comments and encouragement. This work was partly supported by Grants-in-Aid for Scientific Research (No.\ 9204) and for Creative Basic Research (No.\ 09NP0801) from the Japanese Ministry of Education, Science, Sports and Culture.
\section{Introduction} It has been observed that many galaxies and clusters of galaxies are endowed with a magnetic field with typical strength of order $10^{-6}$ \rm G \cite{kron}. The origin of these large scale magnetic fields is unknown. In order for magnetic fields to have this order of magnitude it is widely believed that an enormous amplification of an initial seed field must have taken place. This amplification is usually explained by dynamo theory which can enhance the magnetic field exponentially \cite{dyn}. However, dynamos cannot create a magnetic field, and so in order for them to act they require a seed. At present it is not clear whether this seed field has its origin from some astrophysical mechanism after recombination, during the epoch of galaxy formation and afterwards or whether the seed field is of primordial origin, produced in the very early universe. In the latter scenario it is believed that the primordial magnetic field would have been frozen into the highly conductive plasma as the universe expanded and cooled. Because of the high conductivity diffusion would be small and magnetic flux conserved. If a magnetic field was produced in the early universe and was present at the time of recombination it may have had a significant effect on many astrophysical processes including the formation of galaxies and stars. There are several ways of obtaining limits on cosmological magnetic fields. Limits have been obtained by Faraday rotation measurements of intergalactic fields \cite{vallee,bla-bur-oli}. Other constraints have been obtained through the consideration of the effects of magnetic fields on primordial nucleosynthesis \cite{kern-sta-vach,grasso-ruben,che oli-sch-tru} and on the distortion in the microwave background due to the presence of a cosmological magnetic field \cite{bar-fer-silk,dur}. Even if a primordial magnetic field was to weak to be of astrophysical significance, it is still of principal interest to study cosmic magnetic fields today because they can provide direct and important information about the kind of physics that must have taken place in the early universe. There have been quite a few mechanisms proposed for ways of producing magnetic fields in the early universe. We will not discuss them here but refer to \cite{hind-ever} for a brief review. In this work we will not be concerned with any particular model for the generation of primordial magnetic fields. Instead we will focus on the universial problem of how a primordial field, whatever its origin, will develop as the universe evolved. In order to do this one needs to consider magnetohydrodynamics (MHD) in an expanding universe. Doing full numerical relativistic MHD simulations of the physics of the early universe is hard and requires extensive computer memory and time. Greater dynamical range can be obtained by resorting to approximate methods. The cascade model of Ref.\ \cite{brand} is one such method, which is thought to reproduce well the flow of energy between wavenumbers of the full MHD equations, at the cost of a severe truncation in the number of degrees of freedom. It was found in \cite{brand} that energy was transferred from small to large scales in an inverse cascade, and that the correlation length of the initially random field increased with what looked like a small power of (conformal) time. This is pleasing if one wants to derive the galactic dynamo seed field from a primordial process, for general arguments of causality and energy conservation indicate that such an inverse cascade is actually necessary \cite{hind-ever}. In this work we will be using a string model approach to relativistic MHD to study the evolution of cosmic magnetic fields. The connection between MHD and string dynamics have previously been studied by Semenov \cite{semenov1} and Olesen \cite{olesen1}. Our approach is simular to that of Semenov \cite{semenov1} but more general since we do not assume that there is a conserved particle number density. We take essentially the opposite direction of Olesen \cite{olesen1}, in that we derive string equations from MHD and not the other way around. Once we have reduced the MHD equations to a string model, the results can be understood in terms of the coarsening dynamics of cosmic string networks \cite{vil-shell,hind-kib}. The rate of increase of the network scale length $\xi$ is given by the characteristic velocity of waves on the string, in this case the Alfv\'en velocity, which decreases as the magnetic field decreases in strength. The string approach indicates that this decrease in strength is primarily due to reconnection on small scales: small flux loops are continually created, transferring energy away from the network of infinitely long flux lines. The transfer of energy from the large-scale field happens in a self-similar manner: the magnetic field power spectrum can be displayed as \begin{equation} |{\mathbf{B}}_{{\mathbf{k}}}|^2 \propto \tau^{-\alpha} P(k\xi(\tau)), \end{equation} where ${\mathbf{k}}$ is wavenumber and $\tau$ conformal time. A powerful scaling argument due to Olesen \cite{olesen2} shows, in the limit of ideal MHD, that $\xi \propto \tau^{2/(n+5)}$, where the initial power spectrum behaves as $k^n$ at low $k$. Causality dictates that $n \ge 2$ \cite{dur} (and not $n\ge 0$ as one of us \cite{hind-ever} and another author \cite{son} has stated). As we violate the ideal condition by allowing reconnection, it is not clear that this is the correct power of $\tau$. This scaling law is our main result. We see no sign of a true inverse cascade, in the sense that power is not transferred from small to large scales. If anything, the transfer is from large to small, and it is only because energy is being lost faster from small scales that we see an increase in the scale length $\xi$. \section{Relativistic MHD and strings} In this work we will concentrate on the ideal limit of MHD. This means that we neglect any viscous effects and treat matter as a perfect fluid. This is a good approximation at sufficiently large scales. During the radiation dominated era, which we are mainly concerned with here, the universe was a very good conductor \cite{turn-widr,aho-enq}. We therefore consider the $\sigma\rightarrow\infty$ limit of MHD where magnetic diffusion can be ignored and the magnetic field can be considered to be frozen into the plasma and thus conserving magnetic flux. The starting point for ideal relativistic MHD is the energy-momentun tensor \begin{equation}T^{\mu\nu}=(\rho c^{2}+p)\frac{U^{\mu}U^{\nu}}{c^{2}}-pg^{\mu\nu}+\frac{1}{4\pi}({F^{\mu}}_{\gamma}F^{\nu\gamma}-\frac{1}{4}g^{\mu\nu}F_{\gamma\delta}F^{\gamma\delta}) \label{ideal_e.m.tensor} \end{equation} consisting of the ideal fluid part and electromagnetic part of the energy-momentum tensor. Here $p$ is the fluid pressure, $\rho c^{2}$ is the energy density of the fluid, $U^{\mu}$ is the four-velocity of the fluid satisfying the normalisation condition $U^{\mu}U_{\mu}=c^{2}$ and $F^{\mu\nu}$ is the electromagnetic field tensor. The evolution equations for the system are given by \begin{eqnarray} {T^{\mu\nu}}_{;\nu}&=&0 \label{cov} \\ {F^{\mu\nu}}_{;\nu}&=&\frac{4\pi}{c}J^{\mu} \label{max1} \\ ^*\!{{F}^{\mu\nu}}_{;\nu}&=&0 \label{max2} \\ \sigma F^{\mu\nu}U_{\nu}&=&J^{\mu}-J^{\nu}U_{\nu}U^{\mu} \label{ohm} \end{eqnarray} where equation (\ref{cov}) expresses covariant energy-momentum conservation, equations (\ref{max1}) and (\ref{max2}) are Maxwell's equations with $J^{\mu}$ being the four-current density. In equation (\ref{max2}) $^*\!{{F}^{\mu\nu}}$ is the dual field tensor defined through the relation \begin{equation}^*\!{{F}_{\mu\nu}}=\frac{1}{2}\epsilon_{\mu\nu\gamma\rho}F^{\mu\nu} \label{dual} \end{equation} where $\epsilon_{\mu\nu\gamma\rho}$ is the Levi-Cevita symbol. Equation (\ref{ohm}) is the relativistic version of Ohm's law where $\sigma$ is the conductivity of the fluid, measured in the fluid rest frame. We now repeat the derivation in Subramanian and Barrow \cite{sub-bar} to show that the evolution equations are conformally invariant and the evolution can therefore be transformed from the expanding universe to a flat (Minkowski) spacetime. In so doing we obtain an equivalent set of equations which are easier to handle. Two metrics $g_{\mu\nu}$ and $\tilde{g}_{\mu\nu}$ are said to be conformally related to each other if $\tilde{g}_{\mu\nu}=\Omega^{2}g_{\mu\nu}$ where $\Omega$ is a non-zero differentiable function. The flat Robertson-Walker line element has the form \begin{equation}ds^{2}=dt^{2}-a^{2}(t)d\mathbf{x}^{2} \label{rw} \end{equation} where $t$ is the comoving proper time and $a(t)$ is the scale factor. This metric describes a isotropic and homogeneous universe with zero curva\-ture. The appearence of a hypothetical primordial magnetic field of some strength need not violate the assumption of isotropy and homogeniety because although the presence of a magnetic field will locally generate bulk motions in the fluid, if we look at sufficiently large scales isotropy and homogeniety will be regained. At large scales the magnetic field, whatever its origin, can be considered as essentially random since the correlation length of the field is bounded from above by not exceeding the causal horizon. This justifies the use of the Robertson-Walker metric. We introduce conformal time $\tau$ defined by $d\tau=a^{-1}dt$ so that equation (\ref{rw}) becomes \begin{equation}ds^{2}=a^{2}(\tau)(d\tau^{2}-d\mathbf{x}^{2}) \end{equation} and hence $\eta_{\mu\nu}=\tilde{g}_{\mu\nu}=\Omega^{2}g_{\mu\nu}$ with $\Omega=a^{-1}(\tau)$. We note that under conformal transformations the ideal energy-momentum tensor $T^{\mu\nu}$ transformation as $T^{\mu\nu}=a^{-6}\tilde{T}^{\mu\nu}$. That this is so can be seen directly from the definition of the energy-momentum tensor \begin{equation}T^{\mu\nu}=\frac{2}{\sqrt{-g}}\frac{\partial}{\partial g_{\mu\nu}}(\sqrt{-g}L_{matter}) \end{equation} The new scaled fields (denoted by tilde) obey ordinary energy-momentum conservation. To see this we note that the ideal energy-momentum tensor is traceless, $T\equiv {T^{\mu}}_{\mu}=0$, provided the perfect fluid has the equation of state $p=\frac{1}{3}\rho c^{2}$. For most of the period before decoupling, the early universe was radiation dominated and one can use the above equation of state. We have \begin{equation}{T^{\mu\nu}}_{;\nu}={T^{\mu\nu}}_{,\nu}+\Gamma^{\mu}_{\nu\rho}T^{\nu\rho}+\Gamma^{\nu}_{\sigma\nu}T^{\sigma\mu} \end{equation} Using \begin{equation}\Gamma^{\mu}_{\sigma\rho}=\frac{\dot{a}}{a}(\delta^{\mu}_{\sigma}\delta^{0}_{\rho}+\delta^{\mu}_{\rho}\delta^{0}_{\sigma}-g_{\sigma\rho}\delta^{\mu}_{0})\quad \textrm{and} \quad \Gamma^{\mu}_{\sigma\rho}=4\frac{\dot{a}}{a}(\delta^{\mu}_{\sigma}) \end{equation} we get \begin{equation}{T^{\mu\nu}}_{;\nu}={T^{\mu\nu}}_{,\nu}+2\frac{\dot{a}}{a}T^{\mu 0}-\frac{\dot{a}}{a}\delta^{\mu}_{0}T+4\frac{\dot{a}}{a}T^{\mu 0} \end{equation} But since $T=0$ then \begin{equation}(a^{-6}\tilde{T}^{\mu 0})_{,\nu}+6\frac{\dot{a}}{a}(a^{-6}\tilde{T}^{\mu 0})=({\tilde {T}}^{\mu\nu}{}_{,\nu})a^{-6}=0 \end{equation} and hence \begin{equation}{\tilde {T}}^{\mu\nu}{}_{,\nu}=0 \end{equation} This means that under conformal transformations our original variables will transform to a set of new scaled variable satisfying the following relations: $\rho=a^{-4}\tilde{\rho}$, $p=a^{-4}\tilde{p}$, $U^{\mu}=a^{-1}\tilde{U}^{\mu}$, $J^{\mu}=a^{-4}\tilde{J}^{\mu}$, $F^{\mu\nu}=a^{-4}\tilde{F}^{\mu\nu}$. Now consider Maxwell's equation (\ref{max1}). Since $F^{\mu\nu}$ is anti symmetric the left hand side simplifies to \begin{equation}{F^{\mu\nu}}_{;\rho}={F^{\mu\nu}}_{,\rho}+\Gamma^{\nu}_{\sigma\nu}F^{\mu\sigma}=a^{-4}{\tilde {F}}^{\mu\nu}{}_{,\rho} \end{equation} So the equation for the scaled fields becomes \begin{equation}{\tilde {F}}^{\mu\nu}{}_{,\nu}=\frac{4\pi}{c}\tilde{J}^{\mu} \end{equation} For the four-velocity we have \begin{equation}U_{\nu}=g_{\nu\rho}U^{\rho}=a^{2}\tilde{g}_{\nu\rho}(a^{-1}\tilde{U}^{\rho})=a\tilde{U}_{\nu} \end{equation} Hence \begin{equation}\sigma F^{\mu\nu}U_{\nu}=\sigma a^{-3}\tilde{F}^{\mu\nu}\tilde{U}_{\nu}=a^{-4}(\tilde{J}^{\mu}+\tilde{J}^{\nu}\tilde{U}_{\nu}\tilde{U}^{\mu}) \end{equation} So Ohm's law remains invariant under conformal transformations if we define the scaled conductivity through $\sigma=\tilde{\sigma}a^{-1}$ So we arrive at the fundamental equations of relativistic MHD, \begin{eqnarray} {\tilde {T}}^{\mu\nu}{}_{,\nu}&=&0 \\ {\tilde {F}}^{\mu\nu}{}_{,\nu}&=&\frac{4\pi}{c}\tilde{J}^{\mu} \\ ^*\!{\tilde {F}}^{\mu\nu}{}_{,\nu}&=&0 \\ \tilde{\sigma}\tilde{F}^{\mu\nu}U_{\nu}&=&\tilde{J}^{\mu}-\tilde{J}^{\nu}\tilde{U}_{\nu}\tilde{U}^{\mu}/c^2 \end{eqnarray} From here on we drop the tilde, on the understanding that we mean scaled fields. We will now introduce a new set of coordinates which will enable us to write the MHD equations as non-linear string equations. We define a magnetic four-vector $b^{\mu}$ through the relation \begin{equation}b^{\mu}=\, ^*\!{F}_{\mu\nu}U^{\nu}/c \end{equation} We also define new coordinates $x'=(\eta,\sigma,\psi,\zeta)$ such that $\eta$ are coordinate lines of fluid elements and $\sigma$ are coordinate lines of magnetic flux. Hence $U'^{\mu}=(cq,0,0,0)$ and $b'^{\mu}=(0,\beta,0,0)$ satisfying $U'^{2}=U^{2}=c^{2}$ and $b'^{2}=b^{2}=-B^{2}$ respectively. Thus we have the metric tensor in the new coordinates \[g'_{\mu\nu}=diag(1/q^{2},-B^{2}/\beta^{2},-h_{AB}) \] where $A,B=2,3$. The new coordinate vectors are \begin{equation}\frac{\partial x^{\mu}}{\partial\eta}=\frac{U^{\mu}}{q} \end{equation} \begin{equation}\frac{\partial x^{\mu}}{\partial\sigma}=\frac{b^{\mu}}{\beta} \end{equation} Since the introduction of these coordinates relies on the frozen in property of the plasma we will refer to them as frozen-in coordinates. In the frozen-in coordinate system we can trace the trajectories of fluid elements by simply varying the value of our time coordinate $\eta$ and keeping the values of the other coordinated fixed. Similarly, we can trace the magnetic field lines in the frozen-in system by varying the value of $\sigma$ and keeping the other three coordinates at fixed values. The analysis of the MHD equations is usually performed in terms of the magnetic field and velocity distributions. However, in the description of MHD phenomena, the concept of a magnetic flux tube is often introduced. The reason is that it is sometimes convenient, and we gain a better physical insight, if we base the description on this concept rather on the magnetic field and velocity distributions. A magnetic flux tube is defined as the volume $V$ enclosed by a closed surface $\Sigma$ which is everywhere parallel to the ambient magnetic field vector, and two cross-sectional surface areas $S_{A}$ and $S_{B}$ at either end. The flux tube therefore consists of a bundle of magnetic field lines which enter and exit the volume through the end of surfaces $S_{A}$ and $S_{B}$. The Cartesian coordinate system, traditionally employed in MHD does not really lend itself to an analysis of the magnetic flux tube bahaviour. The frozen-in coordinate system on the other hand, does. The frozen-in coordinates provide a coordinate system co-moving with a flux tube, and is therefore the more natural choice for the mathematical analysis of flux tube behaviour. We now consider the equation for magnetic evolution \begin{equation}^*\!{F^{\mu\nu}}_{;\nu}=0 \end{equation} Since $^*\!{F}^{\mu\nu}$ is antisymmetric the divergence is given by \begin{equation}\frac{1}{\sqrt{-g}}\partial_{\nu}(\sqrt{-g}\,^*\!{F^{\mu\nu}})=0 \end{equation} Using the fact that we can express the dual field tensor as \begin{equation}^*\!{F}^{\mu\nu}=(b^{\mu}U^{\nu}-U^{\mu}b^{\nu})/c \label{F_b_U}\end{equation} we have \begin{equation}\frac{1}{\sqrt{-g}}\partial_{\nu}\left(\sqrt{-g}q\beta\Big(\frac{b^{\mu}}{\beta}\frac{U^{\nu}}{q}-\frac{U^{\mu}}{q}\frac{b^{\nu}}{\beta}\Big)\right)=0 \end{equation} which gives \begin{eqnarray} \lefteqn{q\beta\left(\Big(\frac{U^{\nu}}{q}\Big)\partial_{\nu}\Big(\frac{b^{\mu}}{\beta}\Big)-\Big(\frac{b^{\nu}}{\beta}\Big)\partial_{\nu}\Big(\frac{U^{\mu}}{q}\Big)\right)+{} }\nonumber\\ & &{}\frac{1}{\sqrt{-g}}\left(\frac{b^{\mu}}{\beta}\partial_{\nu}(\sqrt{-g}\beta U^{\nu})-\frac{U^{\mu}}{q}\partial_{\nu}(\sqrt{-g}q b^{\nu})\right)=0 \end{eqnarray} But first square bracket is the Lie derivative and so vanishes, giving \begin{equation}\frac{b^{\mu}}{\beta}\partial_{\nu}(\sqrt{-g}\beta U^{\nu})-\frac{U^{\mu}}{q}\partial_{\nu}(\sqrt{-g}q b^{\nu})=0 \end{equation} Thus we have \begin{equation}\partial_{\nu}(\sqrt{-g}\beta U^{\nu})=0 \end{equation} and \begin{equation}\partial_{\nu}(\sqrt{-g}q b^{\nu})=0 \end{equation} Hence in our comoving frame we get \begin{equation}\frac{\partial}{\partial\eta}(\sqrt{-g}\beta q)=\frac{\partial}{\partial\eta}(B\det(h_{AB}))=0 \end{equation} \begin{equation}\frac{\partial}{\partial\sigma}(\sqrt{-g}\beta q)=\frac{\partial}{\partial\sigma}(B\det(h_{AB}))=0 \end{equation} And so we see that \[B\det(h_{AB})=F(\psi,\zeta) \] where $F(\psi,\zeta)$ is an arbitrary function of $\psi,\zeta$. We therefore have the freedom to choose $\det(h_{AB})$ such that $F(\psi,\zeta)=\bar{B}$ where $\bar{B}$ is a constant and so \[h_{AB}=\frac{\bar{B}}{B}\delta_{AB} \] which means that we can write \[\sqrt{-g}=\frac{\bar{B}}{\beta q} \] We now study the equations of motion, starting from the energy-momentum tensor given in equation (\ref{ideal_e.m.tensor}). Using the above expression, equation (\ref{F_b_U}), for the dual field tensor and the connection between the field tensor and its dual, equation (\ref{dual}), we can write the energy-momentun tensor in the following form \begin{equation}T^{\mu\nu}=\left(\rho c^{2}+p+\frac{B^{2}}{4\pi}\right)\frac{U^{\mu}U^{\nu}}{c^{2}}-\frac{b^{\mu}b^{\nu}}{4\pi}-g^{\mu\nu}\left(p+\frac{B^{2}}{8\pi}\right) \end{equation} From energy-momentum conservation we have \begin{eqnarray} {T^{\mu\nu}}_{,\nu}&=&\frac{(\rho c^{2}+p+B^{2}/4\pi))}{\beta}\frac{U^{\mu}}{c}\partial_{\nu}(\beta U^{\nu}) \nonumber\\ &+&\beta(U\cdot\partial)\left(\frac{(\rho c^{2}+p+B^{2}/4\pi)}{\beta} \frac{U^{\mu}}{c}\right)- \nonumber\\ & &{}\frac{1}{4 \pi q}\partial_{\nu}(q b^{\nu})-\frac{q}{4\pi} (b\cdot\partial)\left(\frac{b^{\mu}}{q}\right)-g^{\mu\nu}P_{,\nu}=0 \end{eqnarray} Here $P$ is the total pressure from both fluid and electromagnetic field. Note that the first and the third terms in this equation vanish. Writing this equation in our comoving reference frame we find \begin{equation} \beta q\left[\frac{\partial}{\partial\eta} \left(\Big(\rho c^{2}+p+\frac{B^{2}}{4\pi} \Big)\frac{q}{c\beta} \frac{\partial x^{\mu}}{\partial\eta}\right)- \frac{1}{4\pi}\frac{\partial}{\partial\sigma}\left(\frac{\beta}{q}\frac{\partial x^{\mu}} {\partial\sigma}\right)\right]-g^{\mu\nu}P_{,\nu}=0 \label{stringeq} \end{equation} Equation (\ref{stringeq}) is the equation of motion in the frozen-in coordinates. In the frozen-in coordinate system the MHD equation of motion reduces to a set of non-linear string equations. The behaviour of a magnetic flux tube is therefore formally analogous to that of a non-linear string. The last term of the left hand side of equation (\ref{stringeq}) take account of inhomogeneity (i.e.\ pressure gradients) and it describes the coupling between neighbouring flux tubes whilst moving through a non-uniform plasma medium. To summarize, we have shown that the behaviour of a magnetic flux tube is formally analogous to that of a string and one can therefore model a magnetised plasma as a fluid composed of strings. We have relied heavily on the arguments of Semenov and Semenov and Berkinov \cite{semenov1}, with one improvement: we have dropped their assumption that there is a conserved particle number density, which is neither necessary nor generally applicable in the early Universe. Our derivation is also complementary to that of Olesen \cite{olesen1}, who starts from the relativistic string equations and shows that they can be interpreted as describing the motion of narrow flux tubes, providing the total pressure remains constant across the tube. \section{Approximate string equations} Exploiting the freedom to change coordinates in the $\sigma,\eta$ subspace, we choose $\beta$ such that $\beta=qB$ and using this in the equation of motion (\ref{stringeq}) we have \begin{equation} \frac{1}{B}\frac{\partial}{\partial\eta}\left(\frac{c\,B}{v_{A}^{2}} \frac{\partial x^{\mu}}{\partial\eta}\right)-\frac{1}{B} \frac{\partial}{\partial\sigma}\left(B\frac{\partial x^{\mu}} {\partial\sigma}\right)-\frac{4\pi}{B^{2}q^{2}}g^{\mu\nu}P_{,\nu}=0, \label{stringeq2} \end{equation} where we have defined the relativistic Alfv\'en velocity as \begin{equation} v_{A}=\frac{c\,B/\sqrt{4\pi}}{\left(\rho c^{2}+p+B^{2}/4\pi\right)^{1/2}}. \label{alfven} \end{equation} Rearranging equation (\ref{stringeq2}) we can write it as \begin{equation} \frac{1}{v_{A}}\frac{\partial}{\partial\eta}\left(\frac{1}{v_{A}}\frac{\partial x^{\mu}}{\partial\eta}\right)- \frac{\partial^{2}x^{\mu}}{\partial\sigma^{2}}= -\frac{1}{B}\frac{\partial}{\partial\eta}\left(\frac{c\,B}{v_{A}}\right) \frac{\partial x^{\mu}}{\partial\eta}+ \frac{1}{B}\frac{\partial B}{\partial\sigma} \frac{\partial x^{\mu}}{\partial\sigma}+ \frac{4\pi}{B^{2}q^{2}}g^{\mu\nu}P_{,\nu} \label{stringeq3} \end{equation} We now argue that for the particular situation we are interested in, it is justified to neglect the three terms on the right hand side of equation (\ref{stringeq3}). The third term on the RHS of equation (\ref{stringeq3}) can in general not be neglected because in many astrophysical situations pressure gradients are important. However, in the early universe the pressure from the magnetic field should be much smaller than the fluid radiation pressure and since the early universe had a very high degree of homogeniety it follows that the gradients of the total pressure are small. The second term on the RHS of equation (\ref{stringeq3}) does not have have a definite sign and so if averaged over time it will be zero. Dropping this term is equivalent to replacing $B$ by its root mean square value. Again using the fact that the fluid radiation pressure in the early universe was much larger than the pressure from the magnetic field and remembering expression (\ref{alfven}) for the relativistic Alfv\'en velocity it is seen that the first term on the right side of equation (\ref{stringeq3}) is indeed small since the ratio $B/v_{A}$ is approximatly constant, thus justifying our decision to neglect it. We now rescale the time parameter $\eta$ to $\bar{\eta}$ through the relation \begin{equation} \frac{\partial}{\partial\bar{\eta}}=\frac{c}{v_{A}(\eta)} \frac{\partial}{\partial\eta}, \end{equation} where we have called attention to the fact that the Alfv\'en velocity may be time-dependent. Using the above mentioned approximations and our new time parameter $\bar{\eta}$ we are left with the following equation of motion \begin{equation} \frac{1}{c^{2}}\frac{\partial^{2}x^{\mu}}{\partial\bar{\eta}^{2}}- \frac{\partial^{2}x^{\mu}}{\partial\sigma^{2}}=0. \label{e:str_eqn} \end{equation} We also recall that the four-dimensional orthogonality of the coordinates, and the coordinate choice enforced by $\be = qB$, supplements this equation with constraints \begin{equation} \dot x \cdot x' = 0, \qquad \frac{1}{c^2}\dot x^2 + x^{'2} = 0, \label{e:str_con} \end{equation} where dot (prime) denotes derivative with respect to the timelike parameter $\bar{\eta}$ $(\sigma)$. We note that the equation (\ref{e:str_eqn}) and constraint (\ref{e:str_con}) is of exactly the same form as the equation for a Nambu-Goto string in Minkowski spacetime expressed in the conformal gauge \cite{vil-shell,hind-kib}. With ideal Nambu-Goto strings in Minkowski spacetime one can further take $x^0 = c\bar{\eta}$, to obtain the system \begin{eqnarray} \frac{1}{c^{2}}\ddot{\mathbf{x}}-\mathbf{x}''&=&0,\label{codeeq} \\ \dot{\mathbf{x}}\cdot\mathbf{x}' &=& 0 \label{codecon1} \\ \frac{1}{c^{2}}\dot{\mathbf{x}}^{2}+\mathbf{x}^{'2} &=& 1 \label{codecon2} \end{eqnarray} This is not possible in general for MHD strings, for one can easily verify that \begin{eqnarray} \dot{\mathbf{x}}\cdot\mathbf{x}' &=& c\frac{\gamma^2}{q^2}\frac{{\mathbf{v}} \cdot {\mathbf{B}}}{v_A B},\\ \frac{1}{c^{2}}\dot{\mathbf{x}}^{2}+\mathbf{x}^{'2} &=& \frac{\gamma^2}{q^2} \left( \frac{{\mathbf{v}}^2}{v_A^2} + \frac{({\mathbf{v}} \cdot {\mathbf{B}})^2}{c^2B^2}\right). \end{eqnarray} where $\gamma=(1-\mathbf{v}/c)^{-1/2}$ is the usual relativistic gamma factor. However, as long as ${{\mathbf{v}} \cdot {\mathbf{B}}} = 0$ in the initial conditions, the first constraint is preserved by the evolution. Furthermore, we can use our remaining coordinate freedom to define $q$ by \begin{equation} q^2 = \gamma^2 \frac{{\mathbf{v}}^2}{v_A^2}, \label{e:qchoice} \end{equation} in which case we really do reproduce the Nambu-Goto equations. We should bear in mind however that we have made several approximations on the way, which are worth reiterating. \begin{enumerate} \item We have neglected pressure gradients. \item We have made a kind of mean-field approximation in replacing the magnetic field by its root mean square value. \item We have neglected the effect of the time-dependence of $B/v_A$. \end{enumerate} We have argued that all this approximations are reasonable in the context of the early Universe, and find that the Nambu-Goto equations can be used to approximate a class of MHD velocity and magnetic field configurations which satisfy ${\mathbf{v}} \cdot {\mathbf{B}} = 0$. They may also be a reasonable approximation to other configurations, provided we understand that the constraints are satisfied in an average sense. \section{Algorithm, simulations and results} The equation of motion (\ref{codeeq}) can be evolved using the Smith-Vilenkin algorithm \cite{smith-vil}. The Smith-Vilenkin algorithm provides an exact discrete evolution for a string network defined on a face-centered cubic lattice and the evolution equations are \begin{equation}\mathbf{x}(\eta+\delta,\sigma)=\frac{1}{2}[\mathbf{x}(\eta,\sigma+\delta)+\mathbf{x}(\eta,\sigma-\delta)]+\mathbf{v}(\eta,\sigma)\delta \end{equation} \begin{equation}\mathbf{v}(\eta+\delta,\sigma)=\frac{1}{2}[\mathbf{v}(\eta,\sigma+\delta)+\mathbf{v}(\eta,\sigma-\delta))]+\frac{1}{4\delta}[\mathbf{x}(\eta,\sigma+2\delta)-2\mathbf{x}(\eta,\sigma)+\mathbf{x}(\eta,\sigma-2\delta)] \end{equation} By using a leap-frog method, updating the positions and velocities at alternate timesteps, the above equations allow us to calculate the exact future evolution of $\mathbf{x}$ and $\mathbf{v}$ from some appropriately choosen initial conditions. Initial string configurations are generated by a method due to Vachaspati and Vilenkin \cite{vach-vil}. They considered string formation in a global $U(1)$- model. The $U(1)$-manifold is discretized by allowing the phase to take only three possible values. These values are then placed randomly on the sites of the cubic lattice. As we go around the face of a cube in real space, the phase descibes a certain trajectory in the group space. A string passes through the face of a cube if that trajectory has a non-zero winding number. With this method the string segments join up to form either closed loops or else open strings which intersect the boundaries of the cubic lattice. The shape of the strings will be Brownian with step size corresponding to the cell size of the lattice. We have seen how relativistic MHD under the approximations discussed above can be described in terms of magnetic flux tubes satisfying the Nambu-Goto equations of motion. In order to represent the continuous distribution of magnetic flux by a network of string, we gather together a flux $\Phi$ into ideal Nambu-Goto strings at positions $X^i(\bar\eta,\sigma)$, with \cite{olesen1} \begin{equation} B^{i}(\tau,\mathbf{x})= \Phi\int d\sigma\frac{\partial X^i(\bar{\eta},\sigma)}{\partial\sigma} \delta^{3}\left(\mathbf{x}-\mathbf{X}(\bar{\eta},\sigma)\right) \end{equation} A few words need to be said about reconnection. In real fluids, magnetic flux tubes interact and reconnections take place when the field lines cross each other. In ideal MHD there are no dissipative or viscous effects. Physical reconnection between field lines cannot take place without resistive effects and therefore the topology of the magnetic field lines is frozen in the fluid and does not change with time. Reconnection, which is a local process, is difficult to describe and the details are not well understood. The efficiency of reconnection is not known and neither is how this process depend on the local geometry involved, like the relative inclination of the flux tubes. To answer this question, one has to go to numerical solutions of the underlying theory, that is MHD. In order to allow for reconnection to take place in our simulations we will take a more simple approach to reconnection. In our model strings are allowed to reconnect only if they pass through the same lattice point. When two strings meet, they intercommute with a probability $P$ and in this work we put $P=1$. Reconnection here take place instantaneously between the discrete evolution time steps. This is physically reasonable since the reconnection timescale is small compared with the evolution of the system. We use a simple estimate for the characteristic length scale $\xi$ which is defined by \begin{equation} \xi^2=V/L, \end{equation} where $V$ is the volume of the lattice and $L$ is the total length of string in the box, excluding small loops. In the Smith-Vilenkin algorithm it is in fact possible to have a loop of zero spatial extent, occupying just one lattice point, which travels at the wave velocity. If such a loop is formed it does not contribute to the magnetic field, and we do not count it in the calculation of the total length. It is a well-known feature of Nambu-Goto simulations that nearly all the string ends up in this kind of loop \cite{vil-shell,hind-kib}. In the MHD context we should probably not regard these loops as representing magnetic field lines at all: the energy is probably being dissipated in a very small-scale reconnection process. In any case, the fact that loops are generically produced so small underlines the fact that strings are very efficient at transferring power from large to small scales. We have studied simulations of the evolution of the magnetic field on lattices with sizes $(128\delta)^{3}$ and $(256\delta)^{3}$, using a code originally developed by Sakellariadou \cite{SakVil88}, which implements both the Vachaspati-Vilenkin algorithm for the initial conditions and the Smith-Vilenkin alorithm for the evolution. Periodic boundary conditions were used in all simulations and the evolution time were resticted to less than half the box size, since after this time causal influences have had time to propagate around the box. Having a representation for the magnetic field in real space we use a three dimensional Fast Fourier Transform algorithm to get a Fourier mode representation. The power spectrum for the magnetic field $|\mathbf{B}_{k}|^{2}$ can then be calculated at every time step by averaging over the amplitudes of all Fourier modes with a wave number between $k$ and $k+2\pi/\delta$. Previously it has been shown that networks of cosmic strings, modelled as Nambu- Goto strings in Minkowski spacetime tend towords a scaling regime \cite{graham1}. This means that the characteristic length scale of the network grows as $\sim c\,t$ where $t$ is Minkowski time. The characteristic length scale for a primordial magnetic field $\xi$ does not grow with the horizon. Instead we expect the magnetic field to grow as $\sim c\,\bar{\eta}$. It is interesting to see if the magnetic field power spectra show scaling behaviour. In order to investigate this we express the power spectrum in terms of a scaling function $P$ which is defined through the following expression \begin{equation} |B(k,\bar{\eta})|^{2}=V \frac{\Phi^2 P(k\xi)}{\xi}. \label{e:scaling} \end{equation} Here $V$ is the volume of the box and its appearance is just a normalisation convention. This form for the scaling function ensures that that the fluctuations obey the scaling law \begin{equation} B^2 = \int \frac{d^{3}k}{(2\pi)^3} |B_{k}|^{2}\propto\xi^{-4}\Phi^2 \end{equation} which is consistent with our picture of a coherent flux $\Phi$ in a region of size $\xi$. Figure 1 shows the chacteristic length scale of the magnetic field versus $c\,\bar{\eta}$ for a typical ensemble. The measured scaling function $P(k\xi)$ is displayed in Figure 2. The data was taken from 12 runs on a $(256\delta)^{3}$ lattice, for values of $c\,\bar{\eta}$ between $40-60$. It is seen that the power spectrum of the magnetic field does reach a scaling regime. This means that the evolution of the network will be self-simular with respect to $c\,\bar{\eta}$. It is clear that the network does exhibit the property of scaling, with $\xi= x\!_*\cdot c\,\bar{\eta}$. The scaling amplitude $x\!_{*}$ can be obtained by looking at the ratio $\xi/\bar{\eta}$ towards the end of the simulation and it is roughly $x\!_{*}\simeq 0.3$ (see \cite{graham1} for a more accurate determination). Thus we can write \begin{equation} \xi=x\!_{*}\cdot c\,\bar{\eta}=x\!_{*}\int v_{A}(\eta)d\eta. \end{equation} Unfortunately, we do not yet know how $v_A$ depends on the time parameter $\eta$ or the conformal time $\tau$. The Alfv\'en velocity $v_A$ depends on $B$, which in turn depends on $\Phi$ and $\xi$ through (\ref{e:scaling}). All we can infer from the information at hand is a consistency relation: if $\xi \propto \eta^r$ and $\Phi \propto \eta^s$, then $3r = s+1$. The extra information we need comes from the covariance of ideal MHD under the scale transformation \cite{olesen2} \begin{equation} {\mathbf{x}} \to l{\mathbf{x}}, \quad \tau \to l^{1-h} \tau, \quad {\mathbf{v}} \to l^h {\mathbf{v}}, \quad {\mathbf{B}} \to l^h {\mathbf{B}}, \end{equation} where $h$ is arbitrary. One can show that under this transformation, \begin{equation} V^{-1} |\mathbf{B}_{k}|^{2} \to l^{3+h} V^{-1} |\mathbf{B}_{k}|^{2}. \end{equation} If we define a function $\Lambda(k,\tau)$ by \begin{equation} V^{-1} |\mathbf{B}_{k}|^{2} = \tau^{(3+h)/(1-h)} \Lambda(k,\tau), \end{equation} then we see that under the same transformation \begin{equation} \Lambda(k,\tau) \to \Lambda(k/l,\tau l^{1-h}) =\Lambda(k,\tau), \end{equation} from which we immediately infer that \begin{equation} \Lambda(k,\tau) = \Lambda(k\tau^{1/(1-h)}). \end{equation} Furthermore, if $|\mathbf{B}_{k}|^{2}$ behaves as $k^n$ as $k\to 0$, we have \begin{equation} |\mathbf{B}_{k}|^{2} \propto \tau^{(3+h+n)/(1-h)} k^n \end{equation} in that limit. It is often assumed that the large-scale power is not affected by small-scale processes \cite{olesen2,son}, in which case $h=-n-3$, and we find the scaling laws (derived by the same authors) \begin{equation} \xi \propto \tau^{2/(n+5)}, \qquad \Phi \propto \tau^{(1-n)/(n+5)}. \end{equation} In the early universe individual particles move with relativistic velocities. However, we expect that the bulk velocity of the fluid $v$ to be non-relativistic. Hence \begin{equation} U^0 = qc \frac{\partial \tau}{\partial \eta} \simeq c, \end{equation} and given (\ref{e:qchoice}) we see that, on average, \begin{equation} \eta = \sqrt{\langle{\dot{\mathbf{x}}^2}\rangle} \tau /v_A, \end{equation} where $\langle{\dot{\mathbf{x}}^2}\rangle$ is the mean square string velocity. Simulations give this to be $0.36$ \cite{graham1}. \section{Conclusions} We have seen how the relativistic MHD equations, with a few reasonable assumptions, may be recast as string-like equations for the motion of the flux lines. This allows us enormous gains in dynamic range in the simulation of a random magnetic field in the early Universe, without being forced to the ideal limit, for we incorporate diffusivity by allowing reconnections between the strings. The result is that we can understand the evolution of magnetic fields in terms of the evolution of a network of strings, and we find that the power spectrum quickly evolves to a self-similar or scale-invariant form, with scale length $\xi$ increasing in time. What this power law is we are unable to say: ideal MHD predicts $\xi \propto \tau^{2/(n+5)}$, where $n$ is the low $k$ exponent of the power spectrum, but as we have departed from ideality by allowing reconnection, we cannot make a prediction. The increase in scale length comes about by the strings straightening at the Alfv\'en velocity, while forming very small loops which can dissipate energy quickly. This is the new feature that the string formulation brings to light: strings transfer energy from large to small scales in an extremely efficient manner. Thus, although the scale length increases, it is because power is preferentially lost from small scales. Whether it is fair to call this an inverse cascade is a matter of terminology. What is clear is that the dynamics predicted by the string model of MHD is certainly not of the right kind to produce seeds for the galactic dynamo from magnetic fields created in the very early Universe. There are of course many places where this line of argument is vulnerable. The model makes approximations which we have tried to highlight. Furthermore, our string simulations use special string configurations to make gains in computational efficiency: the strings lie on a cubic lattice to start with, and one may be suspicious that this may introduce some artifice into the dynamics. However, the propensity of a string network to scale is firmly believed, so we are confident that the magnetic field power spectrum will also scale. What is probably not well approximated is the actual form of the power spectrum, which betrays the particularly string-like feature of a $k^{-1}$ tail, due to the fact that all the flux is held to be concentrated in a narrow tube. Furthermore, the string model may be deficient in its description of helicity, which is known to be extremely important in the development of true inverse cascades \cite{pouquet,corn,FieCar98}. The helicity is represented by the linking number of the strings, but we are not able to incorporate a local contribution induced, for example, by twisted tubes of flux. It may well be that we are missing some very important dynamics here. We clearly need to check our results against a non-ideal MHD code, to see if the predicted self-similar dynamics emerges, and also to find the correct power law for the scale length. This project is currently in hand. We are extremely grateful to Mairi Sakellariadou for the use of her Minkowski space string code. We have also benefited from conversations with Axel Brandenburg, Carlo Barenghi, Richard Rijnbeek and Vladimir Semenov. MH is supported by PPARC grant no.\ GR/L56305. MC is supported by Centrala Studien\"amden (CSN).
\section*{Microscopic chaos from Brownian motion?} In a recent Letter in {\em Nature}, Gaspard et al.~\cite{G} claimed to present empirical evidence for microscopic chaos on a molecular scale from an ingenious experiment using a time series of the positions of a Brownian particle in a liquid. The Letter was preceded by a lead article~\cite{DS} emphasising the fundamental nature of the experiment. In this note we demonstrate that virtually identical results can be obtained by analysing a corresponding numerical time series of a particle in a manifestly microscopically nonchaotic system. As in Ref.~\cite{G} we analyse the position of a single particle colliding with many others. We use the Ehrenfest wind-tree model~\cite{E} where the pointlike (``wind'') particle moves in a plane colliding with randomly placed fixed square scatterers (``trees'', Fig. 1a). We choose this model because collisions with the flat sides of the squares do not lead to exponential separation of corresponding points on initially nearby trajectories. This means there are no positive Lyapunov exponents which are characteristic of microscopic chaos here. In contrast the Lorentz model used in~\cite{G} as an example similar to Brownian motion is a wind-tree model where the squares are replaced by hard (circular) disks (cf.\cite{G}, Fig.~1) and exhibits exponential separation of nearby trajectories, leading to a positive Lyapunov exponent and hence microscopic chaos. Nevertheless, we now demonstrate that the nonchaotic Ehrenfest model reproduces all the results of Ref.~\cite{G}. A particle trajectory segment shown in Fig.~1b is strikingly similar to that for the Brownian particle, (cf.\cite{G}, Fig.~2). Our subsequent analysis parallels that of Ref.~\cite{G}, where more details may be found. Thus the microscopic chaoticity is determined by estimating the Kolmogorov-Sinai entropy $h_{KS}$, using the method of Procaccia and others~\cite{GP,CP} via the information entropy $K(n,\epsilon,\tau)$ obtained from the frequency with which the partical retraces part of its (previous) trajectory within a distance $\epsilon$ for $n$ measurements spaced at a time interval $\tau$. Since for the systems considered here $h_{KS}$ equals the sum of the positive Lyapunov exponents, the determination of a positive $h_{KS}$ would imply microscopic chaos. As in~\cite{G} we find that $K$ grows linearly with time (Fig 1c and~\cite{G} Fig.~3), giving a positive (non-zero) bound on $h_{KS}$ (Fig 1d and~\cite{G} Fig.~4). Indeed our Figs.~1b-d for a microscopically nonchaotic model are virtually identical with the corresponding figures 2-4 of~\cite{G}. Therefore Gaspard et al. did not prove microscopic chaos for Brownian motion. The algorithm of~\cite{GP,CP} as applied here cannot determine the microscopic chaoticity of Brownian motion since the time interval between measurements, $1/60\;s$ in~\cite{G}, is so much larger than the microscopic time scale determined by the inverse collision frequency in a liquid, approximately $10^{-12}\;s$. A decisive determination of microscopic chaos would involve, it seems at the very least, a time interval $\tau$ of the same order as characteristic microscopic time scales. \noindent \\ {\bf C. P. Dettmann, E. G. D. Cohen}\\ Center for Studies in Physics and Biology,\\ Rockefeller University,\\ New York, NY 10021, USA\\ \\ {\bf H. van Beijeren}\\ Institute for Theoretical Physics,\\ University of Utrecht,\\ 3584 CC Utrecht, The Netherlands
\section{Introduction}\label{intro} The Anderson-type metal-insulator transition (MIT) has been the subject of investigation for decades since Anderson formulated the problem in 1958 \cite{anderson}. He proposed that increasing the strength of a random potential in a three-dimensional (3D) lattice may cause an ``absence of diffusion'' for the electrons. Today, it is widely accepted that near this exclusively-disorder-induced MIT the d.\ c.\ conductivity $\sigma$ behaves as $|E-E_c|^\nu$, where $E_c$ is the critical energy or the mobility edge at which the MIT occurs, and $\nu$ is a universal critical exponent \cite{kramer}. Numerical studies based on the Anderson Hamiltonian of localization have supported this scenario with much evidence \cite{kramer,schreiber,bulka,hofstetter,slevin}. In measurements of $\sigma$ near the MIT in semiconductors and amorphous alloys this behavior was also observed with varying values of $\nu$ ranging from $0.5$--$1.3$ \cite{nu,lauinger,stupp}. It is currently believed that these different exponents are caused by interactions in the system \cite{belitz}. Indeed, an MIT may be induced not only by disorder but also by interactions such as electron-electron and electron-phonon interactions, among others \cite{mott2}. Nevertheless, the experimental confirmation of the critical behavior of $\sigma$ allows the use of the Anderson model in order to describe the transition between the insulating and the metallic states in disordered systems. Besides for the conductivity $\sigma$, experimental investigations can also be done for thermoelectric transport properties such as the thermoelectric power $S$ \cite{lauinger,sherwood,lakner}, the thermal conductivity $K$ and the Lorenz number $L_0$. The behavior of these quantities at low temperature $T$ in disordered systems close to the MIT has so far not been satisfactorily explained. In particular, some authors have argued that $S$ diverges \cite{sherwood,castellani} or that it remains constant \cite{sivan,enderby} as the MIT is approached from the metallic side. In addition, $|S|$ at the MIT has been predicted \cite{enderby} to be of the order of $\sim 200\,\mu$V/K. On the other hand, measurements of $S$ close to the MIT conducted on semiconductors for $T\leq1\,$K \cite{lakner} and on amorphous alloys in the range $5\,$K$\leq{T}\leq350\,$K \cite{lauinger} yield values of the order of 0.1-1$\,\mu$V/K. They also showed that $S$ can either be negative or positive depending on the donor concentration in semiconductors or the chemical composition of the alloy. The large difference between the theoretical and experimental values is still not resolved. The objective of this paper is to study the behavior of the thermoelectric transport properties for the {\em Anderson model}\/ of localization in disordered systems near the MIT at low $T$. We clarify the above mentioned difference in the theoretical calculations for $S$, by showing that the radius of convergence for the Sommerfeld expansion used in Refs.\ \cite{castellani,sivan} is zero at the MIT. We show that $S$ is a finite constant at the MIT as argued in Refs.\ \cite{sivan,enderby}. Besides for $S$, we also compute the $T$ dependence for $\sigma$, $K$, and $L_0$. Our approach is neither restricted to a low- or high-$T$ expansion as in Refs.\ \cite{castellani,sivan}, nor confined to the critical regime as in Ref.\ \cite{enderby}. We shall first introduce the model in Sec.\,\ref{Amodel}. Then in Secs.\,\ref{linear} and \ref{sec-etc} we review the thermoelectric transport properties in the framework of linear response and the present formulations in calculating them. In Sec.\,\ref{method} we shall show how to calculate the $T$ dependence of these properties. Results of these calculations are then presented in Sec.\,\ref{result}. Lastly, in Sec.\,\ref{conclude} we discuss the relevance of our study to the experiments. \section{The Anderson Model of Localization}\label{Amodel} The Anderson model \cite{anderson} is described by the Hamiltonian \begin{equation} H=\sum_{i}\epsilon_{i}|i\rangle\langle{i}|+\sum_{i\neq{j}}t_{ij} |i\rangle\langle{j}| \label{hamilton} \end{equation} where $\epsilon_{i}$ is the potential energy at the site $i$ of a regular cubic lattice and is assumed to be randomly distributed in the range $[-W/2,W/2]$ throughout this work. The hopping parameters $t_{ij}$ are restricted to nearest neighbors. For this system, at strong enough disorder and in the absence of a magnetic field, the one-particle wavefunctions become exponentially localized at $T=0$ and $\sigma$ vanishes \cite{kramer}. Illustrating this, we refer to Fig.\,\ref{dosfig} \begin{figure} \hspace*{0.75cm} \resizebox{0.4\textwidth}{!}{ \includegraphics{density.eps} } \caption{ The density of states of a 3D Anderson model, averaged over many disorder realizations with $W=12$. The solid vertical lines at $-E_{c}$ and $E_c$ denote the mobility edges.} \label{dosfig} \end{figure} where we show the density of states $\rho(E)$ obtained by diagonalizing the Hamiltonian (\ref{hamilton}) with the Lanczos method as in Ref.\,\cite{milde0,milde}. The states in the band tails with energy $|E|>E_c$ are localized within finite regions of space in the system at $T=0$ \cite{kramer}. When the Fermi energy $E_F$ is within these tails at $T=0$ the system is insulating. Otherwise, if $|E_{F}|<E_{c}$ the system is metallic. The critical behavior of $\sigma$ is given by \begin{equation} \sigma(E)=\left\{ \begin{array}{cc} \sigma_{0}\left|1-\frac{E}{E_{c}}\right|^{\nu}, & \quad |E| \leq E_{c}, \\ 0, & \quad |E|>E_{c}, \end{array}\right. \label{dc_cond} \end{equation} where $\sigma_{0}$ is a constant and $\nu$ is the conductivity exponent \cite{kramer}. Thus, $E_{c}$ is called the mobility edge since it separates localized from extended states. At the critical disorder $W_c=16.5$, the mobility edge occurs at $E_c=0$, all states with $|E|>0$ are localized \cite{schreiber,bulka} and states with $E=0$ are multifractal \cite{schreiber,milde0}. The value of $\nu$ has been computed from the non-linear sigma-model \cite{wegner}, transfer-matrix methods \cite{kramer,slevin}, Green functions methods \cite{kramer}, and energy-level statistics \cite{hofstetter,els}. Here we have chosen $\nu= 1.3$, which is in agreement with experimental results in Si:P \cite{stupp} and the numerical data of Ref.\ \cite{hofstetter}. More recent numerical results \cite{kramer,slevin}, computed with higher accuracy, suggest that $\nu = 1.5\pm 0.1$. As we shall show later, this difference only slightly modifies our results. We emphasize that the Hamiltonian (\ref{hamilton}) only incorporates the electronic degrees of freedom of a disordered system and further excitations such as lattice vibrations are not included. For comparison with the experimental results, we measure $\sigma$ in Eq.\,(\ref{dc_cond}) in units of $\mathrm{\Omega}^{-1}\mbox{cm}^{-1}$. We fix the energy scale by setting $t_{ij} = 1$ eV. Hence the band width of Fig.\ \ref{dosfig} is comparable to the band width of amorphous alloys \cite{haussler}. Furthermore, the experimental investigations of the thermoelectric power $S$ in amorphous alloys \cite{lauinger} have been done at high electron filling \cite{private} and thus we will mostly concentrate on the MIT at $E_c$. \section{Linear Thermoelectric Effects}\label{linear} \subsection{Definition of the Transport Properties} Thermoelectric effects in a system are due mainly to the presence of a temperature gradient $\mathbf{\nabla}T$ and an electric field $\mathbf{E}$ \cite{ashcroft}. We recall that in the absence of $\mathbf{\nabla}T$ with $\mathbf{E}\neq{0}$, the electric current density $\langle\mathbf{j}\rangle$ flowing at a point in a conductor is directly proportional to $\mathbf{E}$, \begin{equation} \langle\mathbf{j}\rangle=\sigma\mathbf{E}\;. \label{sigma1} \end{equation} By applying a finite gradient $\mathbf{\nabla}T$ in an open circuit, electrons, the thermal conductors, would flow towards the low-$T$ end as shown in Fig.\,\ref{bar}. This causes a build-up of negative charges at the low-$T$ end and a depletion of negative charges at the high-$T$ end. Consequently, this sets up an electric field $\mathbf{E}$ which opposes the thermal flow of electrons. For small $\mathbf{\nabla}T$, it is given as \begin{equation} \mathbf{E}=S\mathbf{\nabla}{T}\,. \label{tp1} \end{equation} This equation defines the \textit{thermopower} $S$. In the Sommerfeld free electron model of metals, $S$ is found to be directly proportional to $-T$ \cite{ashcroft}. Note that the negative sign is brought about by the charge of the thermal conductors. \begin{figure} \hspace*{0.75cm} \resizebox{0.4\textwidth}{!}{ \includegraphics{bar.eps} } \caption{ In an open circuit, a temperature gradient $\nabla{T}$ induces an electric field $\mathbf{E}$ in the opposite direction which opposes the thermal flow of electrons.} \label{bar} \end{figure} For small $\mathbf{\nabla}T$, the flow of heat in a system is proportional to $\nabla{T}$. Fourier's Law gives this as \begin{equation} \langle{\mathbf{j}_{q}}\rangle=K(-\mathbf{\nabla}T) \label{tc1} \end{equation} where $\langle{\mathbf{j}_{q}}\rangle$ is the heat current density and $K$ is the thermal conductivity \cite{ashcroft}. At low $T$, the phonon contribution to $\sigma$ and $K$ becomes negligible compared to the electronic part \cite{ashcroft}. As $T\rightarrow{0}$, $\sigma$ approaches a constant and $K$ becomes linear in $T$. One can then verify the empirical law of Wiedemann and Franz which says that the ratio of $K$ and $\sigma$ is directly proportional to $T$ \cite{wiedemann,chester}. The proportionality coefficient is known as the Lorenz number $L_0$, \begin{equation} L_{0}=\frac{e^2}{k_{B}^2}\frac{K}{\sigma{T}}\, \label{lo1} \end{equation} where $e$ is the electron charge and $k_{B}$ is the Boltzmann constant. For metals, it takes the universal value $\pi^{2}/3$ \cite{ashcroft,chester}. Strictly speaking, the law of Wiedemann and Franz is valid at very low $T$ ($\lesssim{10}\,$K) and at high (room) $T$. This is because in these regions the electrons are scattered elastically. At $T\sim10-100\,$K deviations from the law are observed which imply that $K/\sigma{T}$ depends on $T$. In summary, Eqs.\,(\ref{sigma1})-(\ref{lo1}) express the phenomenological description of the transport properties. \subsection{The Equations of Linear Response} A more compact and general way of looking at these thermoelectric ``forces'' and effects is as follows: the responses of a system to $\mathbf{E}$ and $\mathbf{\nabla}T$ up to linear order \cite{callen} are \begin{equation} \langle\mathbf{j}\rangle = |e|^{-1}\left(|e|L_{11} \mathbf{E}-L_{12}T^{-1}\mathbf{\nabla}T\right) \label{ecurrent} \end{equation} and \begin{equation} \langle\mathbf{j}_{q}\rangle = |e|^{-2}\left(|e|L_{21} \mathbf{E}-L_{22}T^{-1}\mathbf{\nabla}T\right) \label{hcurrent}. \end{equation} The kinetic coefficients $L_{ij}$ are the keys to calculating the transport properties theoretically. Using Ohm's law (\ref{sigma1}) in Eq.\,(\ref{ecurrent}), we obtain \begin{equation} \sigma = L_{11}\,. \label{sigma2} \end{equation} Also from Eq.\,(\ref{ecurrent}), $S$, measured under the condition of zero electric current, is expressed as \begin{equation} S = \frac{L_{12}}{|e|TL_{11}}\;. \label{tp2} \end{equation} With the same condition, Eq.\,(\ref{hcurrent}) yields \begin{equation} K=\frac{L_{22}L_{11}-L_{21}L_{12}}{|e|^{2}TL_{11}}\,. \label{tc2} \end{equation} From Eq.\ (\ref{lo1}) $L_0$ is given as \begin{equation} L_{0}=\frac{L_{22}L_{11}-L_{21}L_{12}}{(k_{B}TL_{11})^2}\,. \label{lo2} \end{equation} Therefore, we will be able to determine the transport properties once we know the coefficients $L_{ij}$. We note that in the absence of a magnetic field, as considered in this work, the Onsager relation $L_{21}=L_{12}$ holds \cite{callen}. Eliminating the kinetic coefficients in Eqs.\,(\ref{ecurrent}) and (\ref{hcurrent}) in favor of the transport properties, we obtain \begin{equation} \langle\mathbf{j}\rangle = \sigma\mathbf{E}-\sigma{S}\nabla{T}\label{ecurrent2} \end{equation} and \begin{equation} \frac{\langle\mathbf{j}_{q}\rangle}{T} = S\langle\mathbf{j}\rangle -\frac{K\nabla{T}}{T}.\label{entropy} \end{equation} Here, $\langle\mathbf{j}_{q}\rangle/T$ is simply the entropy current density \cite{callen}. Hence, the thermopower is just the entropy transported per Coulomb by the flow of thermal conductors. According to the third law of thermodynamics, the entropy of a system and, thus, also $\langle\mathbf{j}_{q}\rangle/T$ will go to zero as $T\rightarrow{0}$. We can check with Eqs.\,(\ref{ecurrent2}) and (\ref{entropy}) that this is satisfied by our calculations in the 3D Anderson model. \subsection{Application to the Anderson Transition} In general, the linear response coefficients $L_{ij}$ are obtained through the Chester-Thellung-Kubo-Greenwood (CTKG) formulation \cite{chester,kubo}. The kinetic coefficients are expressed as \begin{equation} L_{11}= \int_{-\infty}^{\infty} A(E) \left[ - \frac{\partial f(E,\mu,T)}{\partial E} \right] dE\,, \label{l11} \end{equation} \begin{equation} L_{12}= - \int_{-\infty}^{\infty} A(E)\left[E-\mu(T)\right] \left[- \frac{\partial f(E,\mu,T)}{\partial E} \right] dE\,, \label{l12} \end{equation} and \begin{equation} L_{22}= \int_{-\infty}^{\infty} A(E)\left[E-\mu(T)\right]^2 \left[- \frac{\partial f(E,\mu,T)}{\partial E} \right] dE\,, \label{l22} \end{equation} where $A(E)$ contains all the system-dependent features, $\mu(T)$ is the chemical potential and \begin{equation} f(E,\mu,T)=1/\left\{1+\exp([E-\mu(T)]/k_{B}T)\right\} \label{eq-fermi} \end{equation} is the Fermi function. The CTKG approach inherently assumes that the electrons are noninteracting and that they are scattered elastically by static impurities or by lattice vibrations. A nice feature of this formulation is that all microscopic details of the system such as the dependence on the strength of the disorder enter only in $A(E)$. This function $A(E)$ can be calculated in the context of the relaxation-time approximation \cite{ashcroft}. However, an exact evaluation of $L_{ij}$ is difficult, if not impossible, since it relies on the exact knowledge of the energy and $T$ dependence of the relaxation time. In most instances, these are not known. In order to incorporate the Anderson model and the MIT in the CTKG formulation, a different approach is taken: We have seen in Eq.\,(\ref{sigma2}) that the d.c.\ conductivity is just $L_{11}$. Thus, to take into account the MIT in this formulation, we identify $A(E)$ with $\sigma(E)$ given in Eq.\,(\ref{dc_cond}). The $L_{ij}$ in Eqs.\,(\ref{l11})-(\ref{l22}) can now be easily evaluated close to the MIT without any approximation, once the $T$ dependence of the chemical potential $\mu$ is known. Unfortunately, this is not known for the experimental systems under consideration \cite{nu,lauinger,stupp,sherwood,lakner}, nor for the 3D Anderson model. Thus one has to resort to approximate estimations of $\mu$, as we do next, or to numerical calculations, as we shall do in the next sections. \section{Evaluation of the Transport Coefficients} \label{sec-etc} \subsection{Sommerfeld expansion in the metallic regime} \label{sec-sommerfeld} Circumventing the computation of $\mu(T)$, one can use that $-\partial{f}/\partial{E}$ is appreciable only in an energy range of the order of $k_{B}T$ near $\mu\approx{E_F}$. The lowest non-zero $T$ corrections for the $L_{ij}$ are then accessible by the Sommerfeld expansion \cite{ashcroft}, provided that $A(E)$ is nonsingular and slowly varying in this region. Hence, in the limit $T\rightarrow{0}$, the transport properties are \cite{sommer} \begin{equation} \sigma=A(E_F)+\frac{\pi^2}{6}(k_{B}T)^{2}\left.\frac{d^{2}A(E)} {dE^2}\right|_ {E=E_F}\,,\label{sigma3} \end{equation} \begin{equation} S=- \frac{\pi^{2}k_{B}^{2}T}{3|e| A(E_{F})} \left.\frac{dA(E)}{dE}\right|_{E=E_F}\,, \label{tp3} \end{equation} \begin{equation} K=\frac{\pi^{2}k_{B}^{2}T}{3e^2}\left\{A(E_F) -\frac{\pi^{2}(k_{B}T)^2}{3A(E_F)} \left[\frac{dA(E)}{dE}\right]_{E=E_F}^2 \right\}\,, \label{tc3} \end{equation} and consequently \begin{equation} L_0=\frac{\pi^{2}}{3}\left\{1-\frac{\pi^{2}(k_{B}T)^2}{3[A(E_F)]^2} \left[\frac{dA(E)}{dE}\right]_{E=E_F}^2 \right\}. \label{lo3} \end{equation} In the derivations of $S$, $K$, and $L_0$, the term of order $T^2$ in Eq.\ (\ref{sigma3}) has been ignored as is customary. We remark that the terms of order $T^2$ in Eqs.\,(\ref{tc3}) and (\ref{lo3}) are usually dropped, too. In this case in the metallic regime, $L_0$ reduces to the universal value $\pi^2/3$ \cite{ashcroft}. The above approach was adopted in Refs.\,\cite{castellani} and \cite{sivan} to study thermoelectric transport properties in the metallic regime close to the MIT. From Eq.\ (\ref{tp3}), the authors deduce \begin{equation} S=-\frac{\nu\pi^{2}k_{B}^{2}T}{3|e|(E_{F}-E_{c})}\,. \label{tpCAS} \end{equation} In the metallic regime, this linear $T$ dependence of $S$ agrees with that of the Sommerfeld model of metals \cite{ashcroft}. However, setting $A(E) = \sigma(E)$ at the MIT \cite{castellani} in Eq.\,(\ref{dc_cond}) is in contradiction to the basic assumption of the Sommerfeld expansion, since it is not smoothly varying at $E_{F}=E_c$. Thus identifying $A(E) = \sigma(E)$ in Eqs.\ \ref{sigma3} - \ref{lo3} is only valid in the metallic regime with $k_B T \ll |E_c - E_F|$. \subsection{Exact calculation at $\mu(T)= E_c$} \label{sec-crit} A different approach taken by Enderby and Barnes is to fix $\mu=-E_c$ at finite $T$ and later take the limit $T\rightarrow{0}$ \cite{enderby}. Thus, again without knowing the explicit $T$ dependence of $\mu$, the coefficients $L_{ij}$ can be evaluated at the MIT. For the transport properties they obtain, \begin{equation} \sigma=\frac{\sigma_{o}\nu (k_{B}T)^{\nu}I_{\nu}}{\left|E_{c}\right|^{\nu}}\,, \label{sigmaEB} \end{equation} \begin{equation} S = - \frac{k_{B}}{|e|} \frac{\nu+1}{\nu} \frac{I_{\nu+1}}{I_{\nu}}\,, \label{tpEB} \end{equation} \begin{equation} K=\frac{\sigma_{o} (k_{B}T)^{\nu+2}}{e^{2}T\left|E_{c}\right|^{\nu}}\left[(\nu+2)I_{\nu+2} -\frac{(\nu+1)^2I_{\nu+1}^2}{\nu{I}_{\nu}}\right]\,, \label{kEB} \end{equation} and \begin{equation} L_{0}=\left[\frac{(\nu+2)I_{\nu+2}}{\nu{I}_{\nu}}- \frac{(\nu+1)^{2}I_{\nu+1}^2}{(\nu{I}_{\nu})^2}\right]. \label{loEB} \end{equation} Here $I_{1}=\ln2$, $I_{\nu}= (1-2^{1-\nu})\Gamma(\nu)\zeta(\nu)$ for $\mathrm{Re}(\nu)>0,\;\nu\neq1$, with $\Gamma(\nu)$ and $\zeta(\nu)$ the usual gamma and Riemann zeta functions. We see that at the MIT, $S$ does not diverge nor go to zero but remains a universal constant. Its value depends only on the conductivity exponent $\nu$. This is in contrast to the result (\ref{tpCAS}) of the Sommerfeld expansion. In addition, we find that $\sigma\propto T^{\nu}$ and $K\propto T^{\nu+1}$ as $T\rightarrow{0}$. Hence, $\sigma$ and $K/T$ approach zero in the same way. This signifies that the Wiedemann and Franz law is also valid at the MIT recovering an earlier result in Ref.\ \cite{strinati} obtained via diagrammatic methods. However, at the MIT, $L_0$ does not approach $\pi^2/3$ but again depends on $\nu$. We emphasize that Eqs.\ (\ref{sigmaEB})-(\ref{loEB}) are exact at $T$ values such that $\mu(T)-E_c=0$ \cite{enderby}. Thus the $T$ dependence of $\sigma$, $S$, $K$, and $L_0$ for a given electron density can only be determined if one knows the corresponding $\mu(T)$. \subsection{High-temperature expansion} In this section, we will study the lowest-order corrections to the results obtained before with $\mu(T) = E_c$. We do this by expanding the Fermi function (\ref{eq-fermi}) for $|E_c - \mu(T)| \ll k_B T$. In addition, we assume $\mu(T) \approx E_F$ for the temperature range considered. This procedure gives \begin{equation} \hspace*{-0.15cm} \sigma = L_{11} = \frac{\sigma_{o}\nu (k_{B}T)^{\nu}}{\left|E_{c}\right|^{\nu}} \left[ I_{\nu}-(\nu-1)I_{\nu-1}\frac{E_c - E_F}{k_B T} \right] \label{eq-ht-sigma}. \end{equation} For the thermo\-power, the leading-order correction can be obtained without expanding $f(E,\mu,T)$ in $L_{11}$ and $L_{12}$. This yields a constant for $S$ at the MIT \cite{sivan}. We obtain \begin{equation} S = - \frac{k_{B}}{|e|} \left[ \frac{\nu+1}{\nu} \frac{I_{\nu+1}}{I_{\nu}} + \frac{E_c - E_F}{k_B T} \right]\,. \label{eq-ht-tp} \end{equation} For $K$ and $L_0$, we again have to use the expansion of $f(E,\mu,T)$ as in (\ref{eq-ht-sigma}) in order to get non-trivial terms. The resulting expressions are cumbersome and we thus refrain from showing them here. We remark that the basic ingredients used in the high-$T$ expansion are somewhat contradictory, namely, the expansion is valid for high $T$ such that $|E_c - E_F| \ll k_B T$, whereas $\mu(T)=E_F$ is true only for $T=0$. At present, we thus have various methods of circumventing the explicit computation of $\mu(T)$. However, their ranges of validity are not overlapping and it is a priori not clear whether the assumptions for $\mu(T)$ are justified for $S$ or any of the other transport properties close to the MIT. In order to clarify the situation, we numerically compute $\mu(T)$ in the next section and then use the CTKG formulation to compute the thermal properties without any approximation. \section{The Numerical Method}\label{method} In Eqs.\,(\ref{l11})-(\ref{l22}), the explicit $T$ dependence of the coefficients $L_{ij}$ occurs in $f(E,\mu,T)$ and $\mu(T)$. More precisely, knowing $\mu(T)$, it is straightforward to evaluate the $L_{ij}$. We recall that, for any set of noninteracting particles, the number density of particles $n$ can be determined as \begin{equation} n(\mu,T) = \int_{-\infty}^{\infty}dE \rho(E) f(E,\mu,T)\;, \label{numden} \end{equation} where $\rho(E)$ is again the density of energy levels (in the unit volume) as in Fig.\ \ref{dosfig}. Vice versa, if we know $n$ and $\rho(E)$ we can solve Eq.\ (\ref{numden}) for $\mu(T)$. The density of states $\rho(E)$ for the 3D Anderson model has been obtained for different disorder strengths $W$ as outlined in Sec.\ \ref{Amodel}. We determine $\rho(E)$ with an energy resolution of at least $0.1$ meV ($\sim 1$ K). Using $\rho(E)$, we first numerically calculate $n$ at $T=0$ for the metallic, critical and insulating regimes using the respective Fermi energies $|E_F| < E_c$, $E_F = E_c$, and $|E_F| > E_c$. With $\mu=E_{F}$, we have \begin{equation} n(E_{F})=\int_{-\infty}^{E_F}dE\rho(E)\,.\label{numEF} \end{equation} Next, keeping $n$ fixed at $n(E_F)$, we numerically determine $\mu(T)$ for small $T >0$ such that $|n(E_F) - n(\mu, T)|$ is zero. Then we increase $T$ and record the respective changes in $\mu(T)$. Using this result in Eqs.\ (\ref{l11})--(\ref{l22}) in the CTKG formulation, we compute $L_{ij}$ by numerical integration and subsequently determine the $T$ dependent transport properties (\ref{sigma2})--(\ref{lo2}). We consider the disorders $W=8$, $12$, and $14$ where we do not have large fluctuations in the density of states. These values are not too close to the critical disorder $W_c$, so that we could clearly observe the MIT of Eq.\ (\ref{dc_cond}). The respective values of $E_c$ have been calculated previously \cite{schreiber} to be close to $7.0$, $7.5$, and $8.0$. Within our approach, we choose $E_c$ to be equal to these values. \section{Results and Discussions} \label{result} Here we show the results obtained for $W=12$ with $E_c=7.5$. The results for $\sigma$, $K$, and $L_0$ are the same at $-E_c$ and $E_c$ since they are functions of $L_{11}$, $L_{22}$ and $L_{12}^{2}$, only. On the other hand, this is not true for $S$. \subsection{The Chemical Potential} In Fig.\,\ref{mufig}, we show how $\mu(T)$ behaves for the 3D Anderson model at $E_F - E_c = 0$, and $\pm 0.01$. To compare results from different energy regions we plot the difference of $\mu(T)$ from $E_F$. We find that $\mu(T)$ behaves similarly in the metallic and insulating regions and at the MIT for both mobility edges at low $T$. In all cases we observe $\mu(T)\propto{T}^2$. Furthermore, we see that $\mu(T)$ at $-E_{c}$ equals $-\mu(T)$ at $E_c$. This symmetric behavior with respect to $E_{F}=\mu$ reflects the symmetry of the density of states at $E=0$ as shown in Fig.\,\ref{dosfig}. For comparison and as a check to our numerics, we also compute with our method $\mu(T)$ of a free electron gas. The density of states is \cite{ashcroft} \begin{equation} \rho(E)= \frac{3}{2} \frac{n}{E_F}\left( \frac{E}{E_F} \right)^{1/2} \end{equation} and we again use $E_F=E_c = 7.5$. We remark that this value of the mobility edge is in a region where $\rho(E)$ increases with $E$ in an analogous way as $\rho(E)$ for the Anderson model at $-E_c$ . Thus, as shown in Fig.\ \ref{mufig}, $\mu(T)$ of a free electron gas is concave upwards as in the case of the Anderson model at $-E_c$. We also plot the result for $\mu(T)$ obtained by the usual Sommerfeld expansion for Eq.\,(\ref{numden}), \begin{equation} E_{F}-\mu(T)=\frac{E_F}{3}\left(\frac{\pi k_B T}{2E_F}\right)^2. \label{fegmu} \end{equation} We see that our numerical approach is in perfect agreement with the free electron result. \begin{figure} \hspace*{0.75cm} \resizebox{0.4\textwidth}{!}{% \includegraphics{mu_w12.eps} } \caption{ The temperature dependence of the chemical potential $\mu$ measured with respect to the Fermi energy near both mobility edges. Also shown is $\mu(T)$ for a free electron gas. The solid line denotes $\mu(T)$ of Eq.\ (\protect\ref{fegmu}).} \label{mufig} \end{figure} \subsection{The d.c.\ Conductivity} \label{dcc} In Fig.\,\ref{dccondfig} we show the $T$ dependence of $\sigma$. The values of $E_F$ we consider and the corresponding fillings $n$ are given in Tab.\ \ref{tab:sym}. \begin{table}[b] \caption{ Differences of $E_F$ and $n(E_F)$ with respect to the mobility edge at $E_c=7.5$. The density at $E_c$ corresponds to $n= 97.768\%$.} \label{tab:sym} \begin{center} \begin{tabular}{lr@{}lr@{}lc} \hline\noalign{\smallskip} regime & \multicolumn{2}{c}{$E_F-E_c$} & \multicolumn{2}{c}{$n(E_F)-n(E_c)$} & symbol \\ & \multicolumn{2}{c}{(eV)} & \multicolumn{2}{c}{($\%$)} & \\ \noalign{\smallskip}\hline\noalign{\smallskip} metallic & -0. &010 & -0. &031 & $\circ$ \\ & -0. &007 & -0. &022 & $\bigtriangledown$ \\ & -0. &005 & -0. &015 & $\Box$ \\ & -0. &003 & -0. &009 & $\bigtriangleup$ \\ & -0. &001 & -0. &003 & $\Diamond$ \\ \noalign{\smallskip} critical & 0. &000 & 0. &000 & $\bullet$ \\ \noalign{\smallskip} insulating & 0. &001 & 0. &003 & $+$ \\ & 0. &003 & 0. &009 & $\times$ \\ & 0. &010 & 0. &031 & $*$ \\ \noalign{\smallskip}\hline \end{tabular} \end{center} \end{table} The conductivity at $T=0$ remains finite in the metallic regime with $\sigma/\sigma_{o}= {|1 - E_F/E_c|}^{\nu}$, because $(-\partial f/\partial E) \rightarrow \delta(E-E_F)$ in Eq.\ (\ref{l11}) as $T\rightarrow{0}$. Correspondingly, we find $\sigma=0$ in the insulating regime at $T=0$. In the critical regime, $\sigma(T\rightarrow{0})\sim {T}^{\nu}$, as derived in Ref.\ \cite{enderby}, see Eq.\,(\ref{sigmaEB}). We note that as one moves away from the critical regime towards the metallic regime one finds within the accuracy of our data that $\sigma\sim{T}^2$. We observe that in the metallic regime $\sigma$ increases for increasing $T$. This is different from the behavior in a real metal where $\sigma$ decreases with increasing $T$. However, as explained in Sec.\ \ref{Amodel}, the behavior of $\sigma$ in Fig.\,\ref{dccondfig} is due to the absence of phonons in the present model. We also show in Fig.\,\ref{dccondfig} results of the Sommerfeld expansion (\ref{sigma3}) and the high-$T$ expansion (\ref{eq-ht-sigma}) for $\sigma$. Paradigmatic for what is to follow we see that the radius of convergence of the Sommerfeld expansion decreases for $E_F\rightarrow{E_c}$ and in fact is zero in the critical regime. On the other hand, the high-$T$ expansion is very good in the critical regime down to $T=0$ at $E_c=E_F$. The small systematic differences between our numerical results and the high-$T$ expansion for large $T$ are due to the differences in $\mu(T)$ and $E_F$. The expansion becomes worse both in the metallic and insulating regimes for larger $T$. All of this is in complete agreement with the discussion of the expansions in Sec.~\ref{sec-etc}. \begin{figure} \hspace*{0.75cm} \resizebox{0.4\textwidth}{!}{ \includegraphics{dccondSF.eps}} \caption{ The low temperature behavior of the d.c. conductivity $\sigma$. The symbols are as shown in Tab.\ \protect\ref{tab:sym}. The dashed lines represent the Sommerfeld expansion result for $\sigma(T)$ as given in Eq.\,(\protect\ref{sigma3}). For all $8$ choices of $E_F-E_c$, the corresponding high-$T$ expansion (\protect\ref{eq-ht-sigma}) is indicated by solid lines.} \label{dccondfig} \end{figure} \subsection{The Thermopower} \label{tp} In Fig.\,\ref{tpfig1}, we show the behavior of the thermopower at low $T$ near the MIT. In the metallic regime, we find $S\rightarrow{0}$ as $T\rightarrow{0}$. At very low $T$, $S\propto T$ as predicted by the Sommerfeld expansion (\ref{tpCAS}). We see that the Sommerfeld expansion is valid for not too large values of $T$. But upon approaching the critical regime, the expansion becomes unreliable similar to the case of the d.c.\ conductivity of Sec.\ \ref{dcc}. This behavior persists even if we include higher order terms in the derivation of $S$ such as the ${\rm O}(T^2)$ term of Eq.\ (\protect\ref{sigma3}) as shown in Fig.\,\ref{tpfig1}. \begin{figure} \hspace*{0.75cm} \resizebox{0.4\textwidth}{!}{ \includegraphics{thermopower_BW_W12.eps}} \caption{ The low temperature behavior of the thermopower $S$. The symbols are as shown in Tab.\ \protect\ref{tab:sym}. The dashed lines represent the behavior of $S(T)$ in the metallic regime as given in Eq.\,(\protect\ref{tpCAS}). The dot-dashed lines indicate $S$, calculated with the ${\rm O}(T^2)$ term of Eq.\ (\protect\ref{sigma3}), for $E_F-E_c= -0.01$ eV $(\circ)$ and $-0.001$ eV $(\diamond)$. Solid lines are obtained from the high-$T$ expansion (\protect\ref{eq-ht-tp}). The inset shows the behavior at $E_F=E_c$ on an enlarged scale. } \label{tpfig1} \end{figure} Before discussing the critical regime in detail, let us turn our attention to the insulating regime. Here, $S$ becomes very large as $T\rightarrow{0}$. We have observed that it even appears to approach infinity. A seemingly divergent behavior in the insulating regime has also been observed for Si:P \cite{liu}, where it has been attributed to the thermal activation of charge carriers from $E_F$ to the mobility edge $E_c$. However, there is a simpler way of looking at this phenomenon. We refer again to the open circuit in Fig.\,\ref{bar}. Suppose we adjust $T$ at the cooler end such that $\nabla{T}$ remains constant. As $T\rightarrow{0}$ both $\sigma$ and $K$ vanish in the case of insulators --- for $K$ we show this in the next section. This implies that as $T$ decreases it becomes increasingly difficult to move a charge from $T$ to ${T}+\delta{T}$. We would need to exert a larger amount of force, and hence, a larger $\mathbf{E}$ to do the job. From Eq.\,(\ref{tp1}), this implies a larger $S$ value. In the critical regime, i.e., setting $E_F = E_c$, we observe in Fig.\,\ref{tpfig1} that for $T\rightarrow 0$ the thermopower $S$ approaches a value of $228.4\,\mu$V/K. This is exactly the magnitude predicted \cite{enderby} by Eq.\,(\ref{tpEB}) for $\nu=1.3$. In the inset of Fig.\,\ref{tpfig1}, we show that the $T$ dependence of $S$ is linear. The nondivergent behavior of $S$ clearly separates the metallic from the insulating regime. Furthermore, just as for $\sigma$, the Sommerfeld expansion for $S$ breaks down at $E_{F}={E_c}$, i.e., the radius of convergence is zero. Thus, the divergence of Eq.\,(\ref{tpCAS}) at $E_{F}={E_c}$ reflects this breakdown and is not physically relevant. On the other hand, the high-$T$ expansion \cite{sivan} nicely reflects the behavior of $S$ close to the critical regime as also shown in Fig.\ \ref{tpfig1}. For $E_F = E_c$, the high-$T$ expansion (\ref{eq-ht-tp}) assumes a constant value of $S$ for all $T$ due to setting $\mu(T)=E_F$. This is approximately valid, the differences are fairly small as shown in the inset of Fig.\,\ref{tpfig1}. We stress that there is no contradiction that $S>0$ in our calculations whereas $S<0$ in Ref.\,\cite{enderby}. In Fig.\,\ref{pnfig}, we compare $S$ in energy regions close to $E_{c}$ and to $-E_{c}$ \cite{villa}. Clearly, they have the same magnitude but $S<0$ at $-E_c$ and $S>0$ at $E_c$. The two cases mainly differ in their number density $n$. At $-E_c$ the system is at low filling with $n=2.26\%$ while at $E_c$ the system is at high filling with $n=97.74\%$. The sign of $S$ implies that at low filling the thermoelectric conduction is due to electrons and we obtain the usual picture as in Fig.\,\ref{bar} where the induced field $\mathbf{E}$ is in the direction opposite to that of $\nabla{T}$. At high filling, $S>0$ means that $\mathbf{E}$ is directed parallel to $\nabla{T}$. This can be interpreted as a change in charge transport from electrons to holes. We remark that this sign reversal also occurs in the insulating as well as in the critical regime. \begin{figure} \hspace*{0.75cm} \resizebox{0.4\textwidth}{!}{ \includegraphics{pos_neg.eps}} \caption{ An example that the magnitude of $S(T)$ is the same in metallic regions close to $-E_c$ ($\blacksquare$) and $E_c$ ($\circ$). The $+$-symbols indicate $|S|$ for $-E_c$ and $|E_F-E_c|= 0.01$ eV in all cases.} \label{pnfig} \end{figure} In Fig.\,\ref{tpfig2}, we take the data of Fig.\,\ref{tpfig1} and plot them as a function of $\mu-E_c$. Our data coincides with the isothermal lines which were calculated according to Ref.\ \cite{enderby} by numerically integrating $L_{12}$ and $L_{11}$ for a particular $T$ to get $S$. We observe that all isotherms of the insulating ($\mu > E_c$) and the metallic ($\mu < E_c$) regimes cross at $\mu = E_c$ and $S=228.4\,\mu$V/K. Comparing with Eq.\,(\ref{tpCAS}), we again find that the Sommerfeld expansion does not give the correct behavior of $S$ in the critical regime. \begin{figure} \hspace*{0.75cm} \resizebox{0.4\textwidth}{!}{ \includegraphics{muthermo.eps}} \caption{ The data of $S$ in Fig.\,\protect\ref{tpfig1} shown as a function of $\mu$ measured from $E_c = 7.5$ eV. The horizontal line indicates the fixed point MIT value as given in Eq.\,(\protect\ref{tpEB}). The thin dashed lines represent isotherms of $S$ calculated using the same method as in Ref.\,\protect\cite{enderby}. The solid line is an isotherm of $S$ obtained from Eq.\,(\protect\ref{tpCAS}) for $T=22.3$ K. } \label{tpfig2} \end{figure} \begin{figure} \hspace*{0.75cm} \resizebox{0.4\textwidth}{!}{% \includegraphics{muscale2.eps} } \caption{Scaling plot of the thermopower $S$. The thick dashed line indicates the fixed point value at the MIT, the solid line represents the high-$T$ expansion (\protect\ref{eq-ht-tp}), and the thin dashed line shows the Sommerfeld expansion. The inset shows the difference in the scaling when plotting $S$ for $E_F-E_c= -0.001$ eV as function of $(\mu - E_c)/k_B T$ (open symbols) or $(E_F - E_c)/k_B T$ (filled symbols). } \label{tpfig3} \end{figure} The data presented in Fig.\,\ref{tpfig2} suggest that one can scale them onto a single scaling curve. In Fig.\,\ref{tpfig3}, we show that this is indeed true, when plotting $S$ as a function of $(\mu-E_c)/k_B T$. We emphasize that the scaling is very good and the small width of the scaling curve is only due to the size of the symbols. The result for the high-$T$ expansion is indicated in Fig.\,\ref{tpfig3} by a solid line. It is good close to the MIT. In the metallic regime, the Sommerfeld expansion correctly captures the decrease of $S$ for large negative values of $(\mu -E_c)/k_B T$. We remark that a scaling with $(E_F-E_c)/k_B T$ as predicted in Ref.\ \cite{sivan} is approximately valid. The differences are very small as shown in the inset of Fig.\,\ref{tpfig3}. \subsection{The Thermal Conductivity and the Lorenz Number} In Fig.\,\ref{thcondfig}, we show the $T$ dependence of the thermal conductivity $K$. We see that $K\rightarrow{0}$ as $T\rightarrow{0}$ whether it be in the metallic or insulating regime. We note again that this simple behavior is due to the fact that our model does not incorporate phonon contributions. The $T$ dependence of $K$ varies whether one is in the metallic regime or in the insulating regime and how far one is from the MIT. Directly at the MIT, we find that $K\rightarrow{0}$ as $T^{\nu+1}$ confirming the $T$ dependence of $K$ as given in Eq.\,(\ref{kEB}). Near the localization MIT, the $T$ dependence of $K/T$ is thus the same as for $\sigma$ in agreement with Ref.\,\cite{strinati}. Again, we see that the Sommerfeld expansion (\ref{tc3}) is reasonable only at low $T$ in the metallic regime. As for $\sigma$ and $S$, we see that the high-$T$ expansion is again fairly good in the vicinity of the critical regime. \begin{figure} \hspace*{0.75cm} \resizebox{0.4\textwidth}{!}{ \includegraphics{thcond.eps}} \caption{ The thermal conductivity $K$ as a function of temperature. The symbols are as shown in Tab.\ \protect\ref{tab:sym}. The dashed lines were obtained in ${\rm O}(T)$ from the Sommerfeld expansion (\ref{tc3}) for the metallic regime. The results of the high-$T$ expansion for the $8$ choices of $E_F-E_c$ are indicated by solid lines.} \label{thcondfig} \end{figure} At this point we are able to determine the behavior of the entropy in the system as $T\rightarrow{0}$. In the metallic regime, $S$ and $K$ vanish as $T\rightarrow{0}$, while in the critical and insulating regime, $\sigma$ and $K$ vanish as $T\rightarrow{0}$. Applying these results to Eqs.\,(\ref{ecurrent2}) and (\ref{entropy}) yields that for all regimes the entropy current density $\langle\mathbf{j_q}\rangle/T$ vanishes as $T\rightarrow{0}$. Therefore, we find that the third law of thermodynamics is satisfied for our numerical results of the 3D Anderson model. Next, we present the Lorenz number (\ref{lo1}) as a function of $T$ in Fig.\,\ref{Lonumfig}. In the metallic regime, we obtain the universal value $\pi^2/3$ as $T\rightarrow{0}$. Note that for a metal this value should hold up to room $T$ \cite{ashcroft}. However, our results for the Anderson model show a nontrivial $T$ dependence. One might have hoped that the higher-order terms in Eq.\,(\ref{lo3}) could adequately reflect the $T$ dependence of our $L_0$ data. However, this is not the case as shown in Fig.\,\ref{Lonumfig}. This indicates that even if we incorporate higher order $T$ corrections the Sommerfeld expansion will not give the right behavior of $L_0$ near the MIT. We emphasize that the radius of convergence of Eq.\,(\ref{lo3}) is even smaller than for $\sigma$, $S$ and $K$. Similarly, the high-$T$ expansion is also much worse than previously for $\sigma$, $S$ and $K$. Thus in addition to the results for the critical regime, we only show in Fig.\,\ref{Lonumfig} the results for nearby data sets in the insulating and metallic regimes. The $T$ dependence of $L_0$ is linear as shown in the inset of Fig.\,\ref{Lonumfig}. As before for $S$, the high-$T$ expansion does not reproduce this. At the MIT, $L_{0}=2.4142$. This is again the predicted \cite{enderby} $\nu$-dependent value as given in Eq.\,(\ref{loEB}). \begin{figure} \hspace*{0.75cm} \resizebox{0.4\textwidth}{!}{ \includegraphics{LorenzwSF.eps}} \caption{ The Lorenz number $L_0$ as function of temperature. The symbols are as shown in Tab.\ \protect\ref{tab:sym}. The dashed circles mark the values of $L_0$ at $T=0$ for metallic and insulating regimes. The dashed lines were obtained from Eq.\ (\protect\ref{lo3}). The results of the high-$T$ expansion for $E_F-E_c= 0$ eV, $\pm 0.001$ eV and $0.003$ eV are indicated by solid lines. The inset shows the behavior at $E_F=E_c$ on an enlarged scale.} \label{Lonumfig} \end{figure} In the insulating regime, one can show analytically by taking the appropriate limits that $L_0$ approaches $\nu +1$ as $T\rightarrow 0$. In agreement with this, we find that $L_0=2.3$ at $T=0$ in Fig.\ \ref{Lonumfig}. At first glance, it may appear surprising that a transport property in the insulating regime could be determined by a universal constant of the critical regime such as $\nu$. However, in the evaluation of the coefficients $L_{ij}$, the derivative of the Fermi function for any finite $T$ decays exponentially and thus one will always have a non-zero overlap with the critical regime. In the evaluation of Eq.\ (\ref{lo2}), this $\nu$ dependence survives in the limit $T\rightarrow 0$. In real materials, we expect the relevant high-energy transfer processes to be dominated by other scattering events and thus $L_0$ should be different. Nevertheless, for the present model, this $\nu$ dependence holds. \subsection{Possible Scenarios in the Critical Regime} The results presented in Sec.\ \ref{tp} for the thermopower at the MIT show that $S=228.4\, \mu$V/K for $\nu=1.3$. This value is $2$ orders of magnitude larger than those measured near the MIT \cite{lauinger,sherwood,lakner}. However, as mentioned in the introduction, the conductivity exponents found in many experiments are either close $\nu=0.5$ or to $1$ \cite{nu} and one might hope that this difference may explain the small experimental value of $S$. Also, recent numerical studies of the MIT by transfer-matrix methods together with non-linear finite-size scaling find $\nu=1.57\pm{0.03}$ \cite{slevin}. In Tab.\ \ref{tab:nu} we summarize the values of $S$ and $L_0$ at the MIT for these conductivity exponents. We see that all $S$ values still differ by $2$ orders of magnitude from the experimental results. Furthermore, we note that our results for $S$ and $L_0$ are independent of the unit of energy. Even if, instead of $1\,$ eV, we had used $t_{ij}=1\,$ meV, which is appropriate in the doped semiconductors \cite{nu,stupp,lakner,liu}, we would still obtain the values as in Tab.\ \ref{tab:nu}. \begin{table}[b] \caption{ The thermopower and the Lorenz number at the MIT for a 3D Anderson model evaluated for various $\nu$ at $E_c=7.5$ eV. The values for $\nu=0.5$ and $1$ have already been shown in Ref.\ \protect\cite{enderby}.} \label{tab:nu} \begin{center} \begin{tabular}{r@{}lr@{}lr@{}l} \hline\noalign{\smallskip} \multicolumn{2}{c}{$\nu$} & \multicolumn{2}{c}{$S$} & \multicolumn{2}{c}{$L_0$} \\ & & \multicolumn{2}{c}{($\mu$V/K) } & & \\ \noalign{\smallskip}\hline\noalign{\smallskip} 0. & 5 & 163. & 5 & 1. & 7761 \\ 1. & 0 & 204. & 5 & 2. & 1721 \\ 1. & 3 & 228. & 4 & 2. & 4142 \\ 1. & 57 & 249. & 7 & 2. & 6372 \\ \noalign{\smallskip}\hline \end{tabular} \end{center} \end{table} Thus our numerical results for the thermopower of the Anderson model at the MIT show a large discrepancy from experimental results. This may be due to our assumption of the validity of Eq.\ (\ref{dc_cond}) for a large range of energies, or due to the absence of a true Anderson-type MIT in real materials, or due to problems in the experiments. A different scenario for a disorder driven MIT has been proposed by Mott, who argued that the MIT from the metallic state to the insulating state is discontinuous \cite{mott1}. Results supporting such a behavior have been found experimentally \cite{mott2,moebius}. According to this scenario, $\sigma$ drops from a finite value $\sigma_{min}$ to zero \cite{mott1} for $T=0$ at the MIT. This minimum metallic conductivity $\sigma_{min}$ was estimated by Mott to be \begin{equation} \sigma_{min}\simeq\frac{1}{a}\frac{e^2}{\hbar} \label{dc_min} \end{equation} where $a$ is some microscopic length of the system such as the inverse of the Fermi wave number, $a\approx{k}_{F}^{-1}$. As summarized in Ref.\,\cite{mott2}, experiments in non-crystalline materials seem to indicate that $\sigma_{min}>{300}\;\mathrm{\Omega}^{-1}$cm$^{-1}$. Let us assume the behavior of $\sigma(E)$ close to the MIT to be \begin{equation} \sigma(E)= \left\{ \begin{array}{cc} \sigma_{min}, & \quad |E| \leq E_{c}, \\ 0, & \quad |E|>E_{c}, \end{array}\right. \label{mmc} \end{equation} with $\sigma_{min}={300}\;\mathrm{\Omega}^{-1}$cm$^{-1}$. Using the numerical approach of Sec.\ \ref{method}, we obtain $S=119.5\;\mu$V/K at the MIT. This value is still rather large and thus the assumption of a minimum metallic conductivity as in Eq.\ (\ref{mmc}) cannot explain the discrepancy from the experimental results. We remark that the order of magnitude of $S$ is not changed appreciably, even if we add to the metallic side of Eq.\,(\ref{mmc}) a term as given in Eq.\,(\ref{dc_cond}) with $\sigma_0$ a few hundred $\mathrm{\Omega}^{-1}$cm$^{-1}$ and $\nu=1$. Lastly, we note that the transport properties calculated for $W = 8$ and $14$ do not differ from those obtained for $W = 12$ in both the metallic and insulating regions provided we are at temperatures $T \lesssim 100$K. For $S$ and $L_0$ at the MIT we obtain the same values as for $W = 12$. Again we observe that both $S$ and $L_0$ approach these values linearly with $T$, but with different slopes. Our results show that the higher the disorder strength the smaller the magnitude of the slope. \section{Conclusions} \label{conclude} In this paper, we investigated the thermoelectric effects in the 3D Anderson model near the MIT. The $T$ dependence of the transport properties is determined by $\mu(T)$. We were able to compute $\mu(T)$ by numerically inverting the formula for the number density $n(\mu,T)$ of noninteracting particles. Using the result for $\mu(T)$, we calculated the thermoelectric transport properties within the Chester-Thellung-Kubo-Greenwood formulation of linear response. As $T\rightarrow{0}$ in the metallic regime we verified that $\sigma$ remains finite, $S\rightarrow{0}$, $K\rightarrow{0}$ and $L_{0}\rightarrow\pi^{2}/3$. On the other hand, in the insulating regime, $S\rightarrow\infty$. This we attribute to both $\sigma$ and $K$ going to zero. Thus, it becomes increasingly difficult to achieve equilibrium and, hence, the system requires $\mathbf{E}\rightarrow\infty$. For $L_0$, we obtained a universal value of $\nu+1$ even in the insulating regime. Directly at the MIT, the thermoelectric transport properties agree with those obtained in Ref.\ \cite{enderby}. Namely, as $T\rightarrow{0}$, we found $\sigma\sim T^{\nu}$, $K\sim T^{\nu+1}$, while $L_0\rightarrow{\mbox{const}}$. The thermopower $S$ also remains nearly constant in the critical regime and, in particular, it does not diverge at the MIT in contrast to earlier calculations using the Sommerfeld expansion at low $T$ \cite{castellani}. Here we showed that the difference is not so much due to an order of limits problem, but rather reflects the breakdown of convergence of the Sommerfeld expansion at the MIT \cite{sivan}. Our result is supported by scaling data for $S$ at different values of $T$ and $E_F$ onto a single curve which is continuous across the transition. Some of the experiments \cite{lauinger,sherwood} for $S$ have been influenced by the Sommerfeld expansion such that the authors plot their results as $S/T$. We remark that in such a plot the signature of the MIT is hard to identify, since $S/T$ at the MIT diverges as $T\rightarrow 0$ solely due to the decrease in $T$. Our results suggest that plots as in Figs.\,\ref{tpfig1} and \ref{tpfig2} should show the MIT more clearly. The value of $S$ is at least two orders of magnitude larger than observed in experiments \cite{lauinger,sherwood,lakner}. This large discrepancy may be due to the ingredients of our study, namely, we assumed that a simple power-law behavior of the conductivity $\sigma(E)$ as in Eq.\ (\ref{dc_cond}) was valid even for $E \ll E_c$ and $E \gg E_c$. Furthermore, we assumed that it is enough to consider an averaged density of states $\rho(E)$. While the first assumption is of course crucial, the second assumption is of less importance as we have checked: Local fluctuations in $\rho(E)$ will lead to fluctuations in the thermoelectric properties for finite $T$, but do not lead to a different $T\rightarrow 0$ behavior: $S$ remains finite with values as given in Tab.\ \ref{tab:nu}. Moreover, averaging over many samples yields a suppression of these fluctuations and a recovery of the previous behavior for finite $T$. In this context, we remark that --- naively assuming all other parts of the derivation are unchanged --- implications of many-particle interactions such as a reduced single-particle density of states at $E_F$ \cite{coulombgap}, will only modify the $T$ dependence of $\mu$. Consequently, the $T$ dependencies of $S$, $\sigma$, $K$, and $L_0$ may be different, but their values at the MIT remain the same. Our results also suggest that the critical regime is very small. Namely, as the filling increases slightly from $n= 97.74\%$ to $97.80\%$, the behavior of the system changes from metallic to critical and finally to insulating. Up to the best of our knowledge, such small changes in the electron concentration have not been used in the measurements of $S$ as in Refs.\ \cite{lauinger,sherwood,lakner}. We emphasize that such a fine tuning of $n$ is not essential for measurements of $\sigma$ as is apparent from Fig.\ \ref{dccondfig}. Of course, one may also speculate \cite{enderby} that these results suggest that a true Anderson-type MIT has not yet been observed in the experiments. \begin{acknowledgement} We thank Frank Milde for programming help and Thomas Vojta for useful and stimulating discussions. We gratefully acknowledge stimulating communications from John E.~Enderby and Yoseph Imry. This work has been supported by the DFG as part of SFB393. \end{acknowledgement}
\section{INTRODUCTION} Theoretical modeling of chemical abundance gradients in galaxies provides an important tool for understanding galactic evolution. However, chemical evolutionary models applied for the dynamics of galaxies on a Hubble time scale encounter some principal difficulties (Shore \& Ferrini 1995). The chemical evolution of galaxies is usually considered in the one-zone approximation with a postulated accretion rate and some assumed law of star formation. These models do not take into account global hydrodynamic processes of the redistribution of the matter in the galactic disks. The spiral density waves, viscosity, and the disk-halo-disk circulation work in different ways to redistribute matter within the galactic disks during the galactic evolution, which considerably complicates the problem. We discuss one class of galaxies where the modern theories of the chemical evolution can be successfully applied. These are the starburst galaxies with the large scale rings of an enhanced star formation. These galaxies are believed to be the result of recent galactic collisions, and result from the passage of a compact companion galaxy through the disk of a galaxy along its rotation axis (Lynds \& Toomre 1976). Such collisions produce the wave of enhanced star formation, and are responsible for the nature of the large scale rings of the young massive stars observed in some galaxies. The time-scale of bursts of star formation in the galaxies similar to the Cartwheel is much shorter compared to the Hubble time, making secular hydrodynamic processes unimportant. The chemical evolution in the fast waves of star formation is governed therefore by a few parameters which can be measured, or at least estimated, and ring galaxies provide a possibility to determine the relative role of SNIa and SNII in the enrichment of the interstellar matter by heavy elements. It is known that there are two physically different sources of heavy element production. The $\alpha$-elements and oxygen are mainly produced in massive stars experiencing SNII explosions (Woosley \& Weaver 1995), and iron peak elements are substantially contributed by SNIa which are assumed to be the end product of close binary evolution (Tsujimoto et al. 1995). The relative role of these mechanisms is somewhat unclear due to uncertainties in knowledge of the input parameters, and uncertainties of stellar evolutionary models (Ishimaru \& Arimoto 1997; Gibson, Loewenstein \& Mushotsky 1998). The clock of the SNIa explosions and thus the enrichment due to SNIa is substantially delayed with respect to SNII. Such a delay in iron chemical enrichment compared to the $\alpha$-elements will reveal itself in the star-forming ring galaxies as a relative abundance $[Fe/\alpha]$ radial gradient. The value of this gradient depends only on a few parameters such as the velocity of the wave, the slope of the initial mass function, and on the fraction of close binary systems producing iron via type SNIa explosions. The number of ``free'' parameters is hence drastically reduced, and ring galaxies give us unique opportunity to make observational conclusions about the role of SNIa explosions in chemical evolution. \section{CHEMICAL EVOLUTION MODEL} As in our previous paper (Korchagin et al. 1998), we discuss two possible mechanisms of the formation of large-scale star-forming rings. The conventional scenario assumes that rings originate in the direct collisions of disk galaxies with their companions. Instead, we consider the possibility that galactic collision only plays the role of a ``detonator'' stimulating further self-propagating star formation. This model assumes that the wave is a self-organized phenomenon, and is similar to the ``fire in the forest'' models discussed earlier. The most observationally studied ring galaxy is the Cartwheel, and we adopt therefore parameters of the model obtained for this archetype galaxy. The velocity of the wave was chosen equal to 90 km/s --- the value, which gives the best fit to the optical surface brightness gradients observed in the Cartwheel galaxy (Korchagin et al. 1998). In both models we use the Salpeter IMF ($\alpha = 1.35$) with masses of stars located in the interval 0.5~M$_\odot \leq m_s \leq 50$~M$_\odot$. It is remarkable however, that the ratio of iron to oxygen production in the wave does not depend on the parameters determining the rate or the efficiency of star formation. This ratio is mainly determined by the IMF slope, and by the heavy element output of the SN explosions. This fact makes the theoretical predictions of the radial dependence $[Fe/O]$ ratio quite robust. The mathematical formulation of both models, as well as the parameters prescribing the rate of star formation can be found in Korchagin et al. (1998). The basic ideas of the chemical evolution model which we incorporate here to describe abundance gradients in the starburst ring galaxies were outlined by Tenorio-Tagle (1996). The process of element enrichment in galaxies can be summarized as follows. The freshly formed elements ejected by the SN explosions fill the hot interiors of the bubbles, which were created earlier by the stellar winds, and/or by the release of energy in a solitary, or multiple SN explosions. The new elements are mixed with pre-existing matter through evaporation of a part of the cold shell surrounding the bubble, before they are expelled into the halo of the galaxy in a hot, SN-driven winds (``chimneys''). Hot enriched gas eventually cools and ``rains'' back on the galactic disk completing the circle of the element enrichment in a new event of star formation. Recent observations support this picture. Kobulnicky (1999) did not find any evidence for metal abundance enhancement in the vicinity of young star clusters, concluding that the freshly ejected materials are stored in a hot, $10^6$ gas phase. Elmegreen (1997), considering processes of mixing and contamination of interstellar gas after passage of a spiral density wave, came to a similar conclusion. He finds that there is no element mixing on time scales of a few $\times 10^7$ years. The following processes determine spatial distribution of heavy elements in the wake of a star-forming wave. {\it a) Stellar Evolution:} The new-born stars ``inherit'' heavy element abundances of the pre-existing gaseous disk, and hence their atmospheres are not expected to contain the heavy elements produced in the current starburst. Gas released by the low-mass single stars during their red giant phase and the SN explosions with mass $m<11$ is assumed to have oxygen and iron abundances equal to that at the birth epoch of the stars. We take this abundance as to be one fifth of the solar. The fraction of mass returned to the interstellar medium by stars with $m<11$ during their evolution is taken from K\"oppen \& Arimoto (1991). Stars with masses $m > 11 M_{\odot}$ experience SNII explosions and release freshly formed elements. We use in our simulations the oxygen and iron yields taken from models of Woosley \& Weaver (1995) and Nomoto et al. (1997). The SNIa result from C-deflagration of white dwarfs in binary systems with masses of progenitors $\le 8 M_{\odot}$. The nature of the SNI progenitors is rather unclear, and different scenarios give the lifetimes of SNIa progenitors in the interval $\sim 0.1 - 0.6$ Gyr (Branch 1998). In our simulations, we consider therefore three different mass intervals for the SNIa progenitors. Namely, we assume that SNIa are produced in binaries with primary stars of 2.5--8 $M_{\odot}$, 2.5--6 $M_{\odot}$, and 2.5--3.5 $M_{\odot}$. Hence the first supernovae explode at $4\times10^7$, $7\times10^7$ and $2.66\times10^8$ years after the start of the burst, for the three mass intervals. In the latter case, SNIa become unimportant in the heavy element enrichment by fast ring waves of star formation such as inferred in Cartwheel. We adopt the nucleosynthesis output for SNIa from the updated W7-model of Thielemann, Nomoto \& Hashimoto (1993) with fraction of binaries $f_{SNI}$ equal to 0.04 --- the value which Kobayashi et al. (1998) found to be adjusted to reproduce the chemical evolution in the solar neighborhood. {\it b) Evaporation:} There are two sources of mixing of the chemically unprocessed gas with the freshly formed elements inside the hot bubble. The first mechanism is the classical thermal conductivity, which results typically a few thousand solar masses being evaporated from the shell surrounding the hot bubble (e.g. Shull \& Saken 1995). A larger potential source of mixing, but far more uncertain, is the photo-evaporation or shock ablation of the ambient clumpy medium. We consider therefore the following models of the chemical enrichment in the Cartwheel. In the first model, we assume that the freshly formed elements are instantaneously mixed with the remaining ambient gas, or equally, that all remaining ambient gas is transformed into a hot phase by the star formation process. In the second model, we consider the conductive evaporation of shells to be the source of mass in the bubble. For comparison, we also computed the $[Fe/O]$ gradients for pure SN ejecta without mixing. Using expression for the conductive mass-loss of the shell in the adiabatic bubble (Shull \& Saken 1995) and assuming $10^{51}$ erg of energy released by each SN and the shell expansion time of $10^6$ yrs, one can write the expression for the SN-evaporated mass in units $10^7 M_{\odot}$, $10^6$ yr and 1 kpc as: \begin{equation} M_{evap} = (0.7\times10^{-4}) n_0(r,t)^{-2/35} \kappa^{2/7} \end{equation} Here, $n_0(r,t)$ is the density of the ambient gas, and $\kappa < 1$ is a scaling factor. {\it c) Cooling:} The hot processed gas streams out in the halo, and the subsequent abundance gradients are determined by the concurrent processes of cooling and diffusion. Using Raymond's et al. (1976) cooling coefficient we find that the cooling time of hot X-ray emitting gas with the temperature $\sim$ few $\times 10^6 K$ and the density $10^{-2}-10^{-3}$~cm$^{-3}$ is about a few $\times (10^7 - 10^8)$ yrs which is large, or comparable to the dynamical time of the starburst. At the temperature interval typical for the X-ray emitting halo gas, the heavy elements, particularly iron, dominate the cooling process, accounting from half to three-forth of the total cooling rate (Raymond et al 1976). The heavy element abundances in the Cartwheel are deficient by factor of ten as compared to the Orion Nebula in our galaxy (Fosbury \& Hawarden 1977), which will increase the estimated cooling time. We assumed therefore that cooling is unimportant during the starburst time in the Cartwheel. {\it d) Diffusion:} The heavy ion admixtures diffuse into a hot hydrogen plasma of halo, decreasing possible abundance gradients produced by the SN explosions. The diffusion of light ions is effective on the time scale of the starburst (Tenorio-Tagle 1996), but is ineffective for heavy, highly ionized ions. In plasma with temperature of a few $\times 10^6$ K, the oxygen will be ionized up to OVI, and iron up to FeXI--FeIV. The mean free paths of oxygen and iron in plasma with temperature $5\times 10^6$ K and density $10^{-3}$ will be about 50pc and 12pc correspondingly, which gives values of the dimensionless diffusion coefficients $D_O \approx 5\times 10^{-3}$ and $D_{Fe} \approx 6\times 10^{-4}$. Our simulations show, that with such values, the diffusion of heavy elements is unimportant on the time-scales of passage wave front in the disk. {\it e) Grain Formation:} For the pressure typical in a SN-processed gas, the grain formation is effective for temperatures below 1000 K (Gail \& Sedlmayr 1985). Hot halo gas will preserve therefore the abundance compositions which were formed during the ejecta. However, if the hot gas had enough time to cool, iron will be selectively depleted with respect to oxygen due to grain formation. Assuming that the SN ejecta mix with all the remaining ambient gas, we can write the equation describing the spatial distribution of the heavy elements in the wake of the star-forming wave: \begin{eqnarray*} {\partial M_i(r,t) \over \partial t } =& -Z_i^{0} B(r,t) + D_i {\partial^2 M_i \over \partial r^2} \\ +& \int_{M_{min}}^{M_{max}} B(r,t-\tau_{m}) \phi (m) R_i(m)dm \quad (2) \end{eqnarray*} Here $B(r,t)$ denotes the birth rate of stars as a function of time and radius of the galactic disk, $M_i$ is the surface density of an i-th element, $Z_i^{0}=M_i^0/M_{GAS}$ is the initial abundance of the i-th element in the star-forming gas $\phi (m)$ is the initial mass function with minimum of maximum masses $M_{min}$ and $M_{max}$ respectively. $D_i$ is the diffusion coefficient of i-th element, and $M_{GAS} $ is the initial constant surface density of gas. The coefficient $R_i(m)$ gives ratio of mass of the ejected element to the mass of a star $m$. In a similar way, we can write the equation governing the chemical evolution behind the wave when the stellar ejecta are mixed with gas thermally evaporated from the shells with its mass prescribed by the equation (1). Once the IMF is fixed, the birth and death rates can be determined at each radial grid zone making use of the lifetimes of stars. Then the heavy element distributions can be computed by means of equations (1) and (2). \section{Results } Figure 1 illustrates the $[Fe/O]$ enrichment produced by the wave of star formation for the three different scenarios of the element mixing and three different mass intervals of the pre-SNIa stars. The position of the wave taken at time $t=240 Myr$ corresponding to the present location of the outer ring in the Cartwheel galaxy. The left frames show the abundance gradients for the Nomoto et al. (1997) SNII outputs, and the right frames present the results of simulations for the Woosley \& Weaver (1995) SNII iron and oxygen outputs. The bottom frames of Figure 1 show the $[Fe/O]$ profiles if masses of pre-SNIa stars 2.5--3.5 $M_{\odot}$. In this case, all SNIa progenitors have lifetimes longer than the time of the wave propagation , and the role of SNIa stars in the enrichment is negligible. The solid lines of the Figure 1 show the $[Fe/O]$ radial dependence when hot SN ejecta are mixed with the part of ambient gas evaporated from the SN shells. For comparison, we also plot the abundance gradients when the freshly synthesized elements are instantaneously mixed with the all ambient gas (dashed lines), and the abundance gradients produced by the pure SN ejecta (dotted lines). Figure 1 depicts more sharp growth of the oxygen abundance as compared to the iron one in front of the wave of star formation. Such behavior directly follows from the yields of iron and oxygen by the SNII explosions. In the model of Woosley \& Weaver (1995) the oxygen yield increases with stellar mass until the progenitor mass reaches approximately 40 $M_{\odot}$, and then tends to saturate. Supernova models of Nomoto et al. (1997) give rapid increase of oxygen yield with the mass of progenitor. The iron yield on the contrary decreases with stellar mass in both the models. Therefore, SNII with moderate masses of progenitor are responsible mainly for the iron production, whereas SNII with progenitors of higher mass produce mainly oxygen. Large $[Fe/O]$ gradients in the region of the wave of star formation are changed to slower growth of the $[Fe/O]$ ratio towards the center of galaxy when metal enrichment due to SNIa events start to become important. The top and intermediate frames of Figure 1 show the radial profiles of $[Fe/O]$ for the SNIa progenitors with masses 2.5--8 $M_{\odot}$ and 2.5--6 $M_{\odot}$, respectively. The ``pollution'' by the SNIa hot ejecta causes the increase of the $[Fe/O]$ gradient towards the center of the disk. The effect is obviously most noticeable for the pure SN ejecta (dotted lines). The element mixing diminishes the effect, which however remains about 0.05 dex in the Fe/O ratio in both SNII output models and both models of mixing of newly-formed elements. The radial abundance profiles of oxygen and iron obtained for the density wave are much similar to those obtained for the induced wave of star formation. The radial distributions of relative abundance $[Fe/O ]$ are qualitatively independent of the particular model of star formation and the uncertain parameters such as the rate of star formation and the ``efficiency'' of star formation. This makes our conclusions robust, and the relative abundance gradients $[Fe/\alpha ]$ might be used as a tool for determining the relative role of SNIa in the chemical enrichment. \section{DISCUSSION} The nebular abundances measured by Fosbury \& Hawarden (1977) in the Cartwheel pertain to the heavy element abundances before the passage of the wave. Therefore, observations in the inner ring regions are required to test our predictions. Interstellar abundances can be inferred however using X-ray emission from hot halo gas. The launch of the Advanced X-ray Astrophysics Facility ($\it AXAF$) with its spatial resolution of $\approx 0.^{\prime \prime}5$ would allow to measure the heavy element gradients in the ring galaxies of star formation. If the hot gas has enough time to cool after the passage of the wave of star formation, and the newly-formed elements reside in the warm ionized or the cold phases of interstellar medium, then abundances can be inferred using metallic absorption lines towards background objects. The absorption lines of most atoms found in the interstellar medium occur at ultraviolet wavelengths, and hence abundance determinations require satellite observations. With the launch of {\it Hubble Space Telescope} interstellar abundances are now available for a number of atoms, including iron and the $\alpha$ elements $O, N, C, Mg$ etc. (Savage \& Sembach 1996). In external galaxies, the measurement of interstellar abundances depends on the availability of background sources. The most widely used background sources are quasars, using which abundances in halos of galaxies and in the intervening clouds have been determined (e.g. Morton et al. 1980). In a recent study, Gonz\'alez-Delgado et al. (1998) have detected interstellar absorption lines towards the bright starburst nucleus of NGC\,7714 in the ultraviolet spectrum taken with the {\it Hubble Space Telescope}. Hence, when background sources are available the present technology allows for the measurement of interstellar absorption lines. In ring galaxies, the measurement of interstellar abundance thus depends on the inclination of the ring to the line of sight. In cases, where the ring is favorably aligned, the spectrum of center regions and the ring HII regions are likely to contain interstellar absorption lines. \begin{acknowledgments} VK acknowledges Prof. S. Miyama for hospitality, and National Astronomical Observatory of Tokyo for providing COE fellowship. \end{acknowledgments} \newpage
\section{Introduction} The Short Wavelength Spectrometer (SWS) on board the Infrared Space Observatory (ISO) has allowed us, for the first time, to observe the pure rotational spectrum of H$_2$ and study the OTP ratio in warm (T$\sim$ 100 - 1000 K) regions. Previous to ISO observations, the OTP ratio of H$_2$ had been estimated towards regions heated by UV radiation and shocks to temperatures $>$ 2000 K using the relative strengths of the vibrational lines. In outflow regions where shock excitation is the main heating mechanism, observations of the vibrational lines typically reveal OTP ratios of $\sim$ 3 (Smith, Davis \& Lioure 1997). This is the expected value in thermodynamic equilibrium for gas with temperatures $>$ 200 K. However, towards the regions heated mainly by the UV radiation (PDRs), the vibrational lines give OTP values in the range 1.5 - 2 (see e.g. Chrysostomou et al. 1993). The pure rotational lines of H$_2$ have been detected by ISO toward several Galactic regions (S140: Timmermann et al. 1996; Cepheus A West: Wright et al. 1996; BD+40$\deg$4124: Wesselius et al. 1996; HH54: Neufeld et al. 1998). The observations towards all these regions except HH54 are consistent with the H$_2$ rotational lines arising in gas with temperature $>$ 200 K and with an OTP ratio of 3. This value is not consistent with those derived from the vibrational lines in dense PDRs but is in agreement with theoretical predictions. Recently Sternberg \& Neufeld (1999) have argued that the low OTP ratios measured from vibrational lines do not represent the actual OTP ratio in PDRs but it is simply a consequence of the optical depth effects in the fluorescent pumping of the vibrational lines. The non-equilibrium OTP ratio measured towards HH54 has been interpreted as arising in hot-shocked excited gas that has not reached the equilibrium yet. In this interpretation, the observed OTP ratio would be the legacy of an earlier stage in the thermal history of the gas when the gas temperature was 90~K. In this {\it Letter} we report the observations of the H$_2$ rotational lines towards the prototypical PDR associated with the reflection nebula NGC 7023. An OTP ratio in the range 1.5 - 2 is derived from our observations. This is the first detection of an OTP ratio $<$ 3 in a PDR based on the pure H$_2$ rotational lines. \section{Observations} The observations of the S(0), S(1), S(2), S(3), S(4) and S(5) rotational lines of H$_2$ towards the peak of the PDR associated with NGC 7023 (R.A. (2000):\ra[21 01 32.5], Dec(2000):\dec[68 10 27.5]). were made using the Short-Wavelength Spectrometer (SWS) (de Grauuw et al. 1996) on board the Infrared Space Observatory (ISO) (Kessler et al. 1996) during revolution 514 with a total on-target time of 5149 s. At the spectral resolution of this mode ($\lambda$/$\Delta$$\lambda$ $\sim$ 1000-2000) all the observed lines are unresolved. Data reduction are carried out with version 7.0 of the Off Line Processing routines and the SWS Interactive Analysis at the ISO Spectrometer Data Center at MPE. Further analysis has been made using the ISAP software package. The uncertainities in the calibration are of 15 \%, 25 \%, 25 \%, 25 \%, 20\% and 30\% for the S(5), S(4), S(3), S(2), S(1) and S(0) lines respectively (Salama et al. 1997). \section{H$_2$ rotational lines} In Fig. 1 we present the spectra of the S(0), S(1), S(2), S(3), S(4) and S(5) H$_2$ lines towards the peak of the PDR associated with NGC 7023. The observed intensitites and some interesting observational parameters are presented in Table 1. The data have been corrected for dust attenuation using the extinction curve of Draine \& Lee (1984), and a value of 0.43 for the dust opacity at 0.55 $\mic$. This value has been derived from the ISO LWS01 spectrum (Fuente et al. 1999, hereafter paper II). The extinction for the S(0), S(1), S(2), S(3), S(4) and S(5) lines is very small and amounts to 0.01, 0.02, 0.02, 0.04, 0.02 and 0.01 respectively. Fig. 2 shows the rotational diagram for the H$_2$ corrected for extinction effects and assuming extended emission. (Note that the errors in Fig. 2 are entirely dominated by the calibration uncertainities.) This diagram shows that the ortho-H$_2$ levels have systematically lower N$_u$/g$_u$ values (where N$_u$ and g$_u$ are the column densities and degeneracies of the upper levels of the transitions) than the adjacent J--1 and J+1 para-H$_2$ levels producing a $``$zig-zag" distribution. In fact, the ortho-H$_2$ levels seem to define a curve which is offset from that of the para-H$_2$ levels (see Fig. 2). This is the expected trend if the OTP ratio is lower than 3. The offset between the two set of data is systematic and seems to show the trend of being larger for the low energy levels (S(0) and S(1) transitions), than for the high energy levels (S(2), S(3), S(4) and S(5) transitions). The offset between the ortho- and para- H$_2$ curves is larger than the observational errors and cannot be due to the different apertures for the different lines. For the case of a point-like source we would have to correct by a factor of 1.93 the intensity of the S(0) line and by a factor of 1.35 the intensities of the S(1) and S(2) lines. In this case, the offset between the para- and ortho- curves would increase. Based on the spatial distribution of the HI(21 cm) line (Fuente et al. 1998) and those of the CII (158 $\mic$) and OI (63 $\mic$) lines (Chokshi et al. 1988, paper II), we will assume in the following a beam filling factor of 1 for all the observed H$_2$ lines. Although the exact shape of the rotational diagram depends on the excitation conditions of the region, the $``$zig-zag" features cannot be explained by any model which assumes an equilibrium OTP ratio. In thermodynamic equilibrium, the rotational temperature between an ortho- (para-) H$_2$ level and the next J+1 level should always increase with the energy of the upper level giving rise to a smooth curve in the rotational diagram regardless of the temperature profile of the region. A $``$zig-zag" distribution implies that for some pair of levels the rotational temperature decreases with the energy of the upper level. One can only get this effect with a non-equilibrium OTP ratio. We have compared our data with the models by Burton et al. (1992) in order to estimate the incident UV field, density and OTP ratio. Burton et al. (1992) assumed a constant value of 3 for the OTP ratio in their calculations. The best fit to our data is for an incident UV field of G$_o$ = 10$^4$ in units of the Habing field and a density of n = 10$^6$ cm$^{-3}$, but it systematically underestimates the intensity of the para-H$_2$ transitions and overestimates the intensities of the ortho-H$_2$ transitions (see open triangles in Fig. 3). This is the expected behavior if one assumes an OTP ratio of 3 and the actual OTP ratio in the region is $<$ 3. We have corrected the line intensities predicted by the model for different values of the OTP ratio assuming that the total amount of H$_2$ molecules at a given temperature and the line ratios between levels of the same symmetry are not affected by the OTP ratio. These assumptions are correct for the low-J transitions where the collisional excitation dominates. In Fig. 3 we compare our data with the predicted diagram for an OTP ratio of 3 which correspond to the model without any correction (open triangles), 2 (open circles) and 1 (open squares). The diagram for an OTP ratio of 3 clearly does not fit any of our observational points. We find the best fit to our data with an OTP ratio of 2. In this case, the predicted intensities for the S(1), S(2) and S(3) lines are in agreement with the observed values. Only the predicted intensity for the S(0) line is not consistent with the observations. To fit the intensity of the S(0) line it is necessary to assume an OTP ratio of $\leq$ 1.5 (see the case OTP = 1 in Fig. 3), but in this case, we will have a worse fit for the S(1), S(2) and S(3) lines than for an OTP ratio of 2. As discussed in Section 4, this suggests a possible variation of the OTP ratio with the temperature. The cooler gas emitting only in the S(0) line could have an OTP ratio of $<$ 1.5, while the warmer gas emitting in the S(1), S(2) and S(3) lines could have an OTP ratio of 2. However, to establish this variation unambigously, it is necessary to have a very accurate knowledge of the excitation conditions in the region. To be conservative, we will conclude that the OTP ratio is in the range 1.5 - 2 in this PDR. Martini et al. (1997) found an OTP ratio of 2.5$\pm$0.4 using the near-IR H$_2$ vibrational lines towards their position 1 which is 20$''$ offset from ours. Taking into account that because of the optical depth effects in the vibrational lines this value is just a lower limit to the actual OTP ratio, it proves that the OTP ratio is close to 3 for the gas with kinetic temperatures T$_k$ $>$ 2000 K. This difference between the OTP ratios derived from the rotational and vibrational lines argues in favor of a variaton of the OTP ratio with the kinetic temperature. Since the OTP ratio in this source is different from the equilibrium value, the rotation temperature between an ortho- and a para- level does not represent an estimate of the gas kinetic temperature. To estimate the gas kinetic temperature we have calculated the rotation temperature between levels of the same symmetry. For the para-H$_2$ levels, we have derived a rotation temperature of $\sim$ 290 K from the S(0) and S(2) lines and of $\sim$ 500 K from the S(2) and S(4) lines. For the ortho-H$_2$ levels, the derived temperature is $\sim$ 440 from S(1) and S(3) lines and $\sim$ 700 K from the S(3) and S(5) lines. Based on these calculations we conclude that the OTP ratio is $\sim$ 1.5 - 2 in the gas with kinetic temperatures $\sim$ 300 - 700 K. We derive a total H$_2$ column density of 5 10$^{20}$~cm$^{-2}$ assuming a rotation temperature of 290 K between the J = 0 and J = 2 para-levels and of 440 K between the J = 1 and J = 3 ortho-levels. \section{Discussion} Based on the H$_2$ rotational lines data we have derived an OTP ratio in the range of 1.5 - 2 for the gas with kinetic temperatures $\sim$ 300 - 700 K towards the reflection nebula NGC 7023. This is the second object with a non-equilibrium OTP ratio measured from the H$_2$ pure rotational lines. The first non-equilibrium OTP ratio was detected towards the outflow source HH54, and interpreted as arising in shock-heated gas that has not reached the equilibrium yet. This interpretation is not plausible for NGC 7023. The high dust temperature,T$_d$$\approx$ 40 K, and the detection of the SiII (34.8 $\mic$) and CII (157.7 $\mic$) lines (paper II) proves the existence of a PDR in this region. Although a shock component might also exist (see Martini et al. 1997), the heating is dominated by UV photons. It is then expected that most of the warm H$_2$ arises in the PDR. A non-equilibrium value of the OTP ratio is not expected for the physical conditions prevailing in a dense PDR. The initial OTP ratio after the H$_2$ formation is very uncertain. Because of the large exothermicity of the formation process, it is expected to be 3 . However, this OTP ratio can change if the H$_2$ molecules remain on the grain surface long enough to reach the equilibrium at the dust temperature. In this case, the OTP ratio after H$_2$ formation will be that of the equilibrium at the grain temperature. After the ejection of the H$_2$ molecules to the gas phase, exchange reactions with H and H$^+$ change the OTP ratio until achieving the equilibrium at the gas temperature. For the gas temperatures traced by the H$_2$ rotational lines ($>$ 200 K), the equilibrium OTP ratio is 3. One possibility to explain the low OTP ratio observed in NGC 7023 is to suppose that the OTP ratio in the H$_2$ formation is lower than 3, and once in the gas phase, the H$_2$ molecules are destroyed before attaining the equilibrium value at the gas temperature. The dust temperature in a PDR with G$_o$ = 10$^4$ and n = 10$^6$ cm$^{-3}$ is $<$ 100 K. In particular, the dust temperature measured towards the PDR peak in NGC 7023 using our LWS01 spectrum is 40 K which corresponds to an equilibrium OTP ratio of $\sim$ 0.1. The OTP conversion in the atomic region is dominated by H - H$_2$ collisions with a rate of 10$^{-13}n$ s$^{-1}$ where $n$ is the atomic hydrogen density (see Sternberg \& Neufeld 1999 and references therein). The hydrogen density derived from the HI image published by Fuente et al. (1998) is $\sim$ 5 10$^3$ cm$^{-3}$. A similar value ($\sim$ 0.5 - 1 10$^4$ cm$^{-3}$) was derived by Chokshi et al. (1988) based on the OI and CII lines. However, densities larger than 10$^5$ cm$^{-3}$ are derived from molecular data (Fuente et al. 1996, Lemaire et al. 1996, Gerin et al. 1998). The density of atomic hydrogen in the region with T$_k$ $\sim$ 300 K is expected to be about an order of magnitude lower than the H$_2$ density. Then we assume an atomic hydrogen density of $\sim$ 10$^4$ cm$^{-3}$ in our calculations. With this value, the OTP conversion rate due to H$_2$ - H collisions is 10$^{-9}$ s$^{-1}$. At the cloud surface the unshielded photodissociation rate is $\sim$ 5 10$^{-11} G_\circ$ s$^{-1}$ $\sim$ 5 10$^{-7}$ s$^{-1}$, i.e., 2 orders of magnitude larger than the OTP conversion rate. Then, in the cloud surface, before self-shielding is important (A$_v$ $\leq$ 0.3 mag, T$_k$$>$ 500 K), an OTP ratio lower than 3 can be explained by assuming that the OTP ratio in the H$_2$ formation is the equilibrium value at the grain temperature. Deeper into the molecular cloud, when H$_2$ is self-shielded, one can consider that the H$_2$ destruction rate is similar to the H$_2$ reformation rate which is given by $\sim$ 3 10$^{-17} n$ s$^{-1}$. In this region, a lower limit to the OTP conversion rate is given by the conversion rate due to H$_2$ - H$^+$ collisions which is $\sim$ 10$^{-17} n$ s$^{-1}$ (Sternberg \& Neufeld 1999). The OTP conversion rate is of the same order than the destruction rate and the OTP ratio is expected to be close to the equilibrium. According to these estimates one expects to have an OTP ratio close to 3 at low temperatures and a non-equilibrium OTP ratio at higher temperatures. The contrary trend is derived from our observations. The high energy S(1), S(2) and S(3) lines are fitted with an an OTP ratio of 2 while the low energy S(0) line is better fitted with an OTP ratio of 1.5. Another possibility to explain the non-equilibrium OTP value in NGC 7023 is to consider the case of a dynamic PDR, i.e. the dissociation front is advancing into the molecular cloud. In this case we do not have to assume a non-equilibrium OTP ratio in the H$_2$ formation. The PDR is being fed continuously by the cool gas of the molecular cloud in which the equilibrium value of the OTP ratio is lower than 3. This gas is heated by the stellar UV radiation to temperatures $>$ 200 K but leaves the PDR before attaining the equilibrium OTP ratio at this temperature. In this scenario, the gas which is expected to have an OTP ratio smaller than the equilibrium value is the gas that has more recently been incorporated into the PDR, which is also the gas at lowest temperature. The OTP ratio is expected to increase and reach values close to 3 at high temperatures. This behavior is consistent with the trend observed in our data. As discussed above, the OTP H$_2$ conversion is mainly due to H - H$_2$ collisions with a conversion rate of $\sim$ 10$^{-9}$ s$^{-1}$. To have a significant fraction of the gas with a non-equilibrium OTP ratio the photodissociation front must advance about 1 - 2 mag in the conversion time, i.e., it must penetrate into the molecular cloud at a velocity of $\sim$ 10$^7 n^{-1}$ kms$^{-1}$. Skinner et al. (1993) proposed the existence of an anisotropic ionized wind associated with this star based on radio continuum observations. Later, Fuente et al. (1998) detected an HI outflow with a velocity of $\sim$ 7.5 kms$^{-1}$. They proposed that this outflow is formed when the gas that has been photodissociated by the UV radiation, is accelerated by the stellar winds along the walls of a biconical cavity. The velocity of the HI outflow cannot account for the velocity at which the photodissociation front must advance into the molecular cloud to have an OTP ratio lower than 3 unless the OTP conversion rates are severely overestimated (by a factor of $>$ 10) and/or the density of the outflow is $>$ 10$^5$ cm$^{-3}$. In spite of this problem, a photodissociation front penetrating into the cloud because of the outflow seems to be the most plausible explanation for the low OTP ratio measured in this region. The existence of an outflow can also explain the difference between this region and other PDRs like S140 in which an OTP ratio of 3 has been derived from the H$_2$ rotational lines. \acknowledgments We would like to thank the referee for his/her helpful comments. This work has been partially supported by the Spanish DGES under grant number PB93-0048 and the Spanish PNIE under grant number ESP97-1490-E. N.J.R-F acknowledges Conserjer{\i}a de Educaci\'on y Cultura de la Comunidad de Madrid for a pre-doctoral fellowship. \clearpage \newpage
\section{Introduction} With new instruments such as MEGACAM at CFHT (\cite{Bouetal98}\ 1998) and the VST at the European Southern Observatory (\cite{Arnetal98}\ 1998), wide-field surveys detect the weak lensing of faint galaxies by large scale structure will soon become a reality (see \cite{Mel98}\ 1998 for a recent review). Weak lensing by large-scale structure produces a correlated distortion in the ellipticities of the galaxies on the percent level (\cite{Blaetal91} 1991; \cite{Mir91}\ 1991; \cite{Kai92} 1992) which can be used to measure a two-dimensional projection of the intervening mass distribution (\cite{TysValWen90} 1990; \cite{KaiSqu93} 1993). If the redshift of the source galaxies are known, then more information can be extracted out of weak lensing by tomography, i.e. differencing the two-dimensional projected images to recover the three-dimensional distribution. In the absence of spectroscopy, approximate redshifts for the faint galaxies can be determined through photometric techniques (see e.g. \cite{Hogetal98} 1998 and references therein) and with the large number of galaxies at $R \lesssim 25$ ($\sim 10^5$ deg$^{-2}$) the properties of distribution can be known to good accuracy (\cite{Sel98} 1998). Indeed the weak lensing surveys already plan to use photometric redshift information at least on a small subsample to measure the low order moments of the distribution such as its mean. These are important for determining the cosmological implications of the data (\cite{Smaetal95a} 1995; \cite{ForMelDan96} 1995; \cite{LupKai97} 1997). The potential of tomographic techniques, especially in the wide-field limit where the cosmological information is completely contained in the two-point functions or power spectra, remains largely unexplored. Indeed most studies of weak lensing (e.g. \cite{JaiSel97} 1997; \cite{Kai92} 1998; \cite{HuTeg99} 1999) simply assume a delta-function distribution of galaxies making tomography impossible. In this {\it Letter}, we study the prospects for weak lensing tomography within the framework of the adiabatic cold dark matter (CDM) class of models for structure formation. We begin by defining the power spectrum statistics for an arbitrary set of galaxy redshift distributions. These are the power spectrum of the convergence map for each distribution and the cross-correlation between the maps. We then quantify how much additional information can be extracted by subdividing a single magnitude-limited sample into bins in redshift and analyzing their joint power spectra and cross-correlation. We conclude with a discussion of how errors in photometric redshifts might affect tomographic techniques. \section{Power Spectra} Generalizing the results of \cite{Kai98} (1992, 1998), we define the angular power spectra and cross-correlation of sky maps of the convergence based on a series of galaxy redshift distributions $n_i(z)$ (see also \cite{Sel98} 1998) \begin{eqnarray} {{\bf P}^{\kappa}_{ij}}(\ell) &\equiv& {1 \over 2\ell+1} \left< a^*_{(\ell m)\, i} a_{(\ell m) \, j} \right> \, \nonumber\\ &\approx& 2\pi^2 \ell \int d D \, {{g_i(D) g_j(D)} \over D_A^3(D)} \Delta_\Phi^2 (k_\ell,D) \,, \label{eqn:ppsi} \end{eqnarray} where $a_{(\ell m)\, i}$ are the spherical harmonic coefficients of the maps. Here $D = \int_0^z (H_0/H) dz$ is the dimensionless comoving distance and \begin{equation} D_A(D)= \Omega_K^{-1/2} {\sinh(\Omega_K^{1/2}D)} \end{equation} is the dimensionless angular diameter distance, where $\Omega_K=1 - \sum_i \Omega_i$ is the effective density in spatial curvature in units of the critical density. The efficiency with which gravitational potential fluctuations $\Phi$, as measured by their dimensionless power per logarithmic interval $\Delta_\Phi^2 \equiv k^3 P_\Phi/2\pi^2$, lens the given galaxy distribution $n_i$ is described by \begin{equation} g_i(D) = D_A(D) \int_D^\infty dD'\, \left[ n_i { {dz \over dD} }\right](D') {D_A(D'-D) \over D_A(D')} \,. \end{equation} Note that $n_i$ is normalized so that $\int_0^\infty dz \, n_i(z)=1$. Finally $k_\ell = \ell H_0 / D_A(D)$ is the wavenumber which projects onto the angular scale $\ell$ at distance $D$. For small fields of view, the spherical harmonics of order $\ell$ can be replaced by Fourier modes with angular frequency $\omega$. We use the \cite{PeaDod96} (1996) scaling relations to evalute $\Delta_\Phi^2$ in the non-linear density regime. Equation~(\ref{eqn:ppsi}) assumes that the redshift distributions are sufficiently wide to encompass many wavelengths of the relevant fluctuations ($2\pi/k_\ell$) along the line of sight so that the Limber equation holds even tomographically (see \cite{Kai98} 1998). These power spectra define the cosmic signal. Shot noise in the measurement from the intrinsic ellipticity of the galaxies adds white noise to the cosmic signal making the observed power spectra \begin{equation} {\bf C}_{ij}(\ell) = {{\bf P}^{\kappa}_{ij}}(\ell) + {\left< \gamma_{\rm int}^2\right>\delta_{ij}/\bar n_i} \,, \label{eqn:covariance} \end{equation} where $\left< \gamma_{\rm int}^2 \right>^{1/2}$ is the rms intrinsic shear in each component, and ${\bar n}_i$ is the number density of the galaxies per steradian on the sky in the whole distribution $n_i(z)$. \begin{figure}[t] \centerline{\epsfxsize=3.5truein\epsffile{fig1.eps}} \caption{Subdividing the source population. Partitioning the galaxies by the median redshift (or distance $D$) yields lensing efficiencies with strong overlap.} \label{fig:g} \end{figure} The distributions $n_i(z)$ need not be physically distinct galaxy populations. Consider a total distribution $n(z)$ with \begin{equation} \left[ n {d z \over dD} \right](D) \propto D^\alpha \exp[-(D/D_*)^\beta] \,, \label{eqn:distribution} \end{equation} which roughly approximates that of a magnitude-limited survey, and take $\alpha=1, \beta=4$ for definiteness (assumed throughout unless otherwise stated). One can {\it subdivide} the sample into redshift bins to define the distributions $n_i(z)$. The power spectra for cruder partitions can always be constructed out of finer ones: if the $j$ and $k$ bins are combined, then \begin{eqnarray} \bar n_{j+k}^2 {\bf P}_{(j+k)(j+k)}^\kappa &=& \bar n_{j}^2 {\bf P}_{jj}^\kappa + 2 \bar n_j \bar n_k {\bf P}_{jk}^\kappa + \bar n_k^2 {\bf P}_{kk}^\kappa \,, \nonumber\\ \bar n_{j+k} {\bf P}_{i(j+k)}^\kappa &=& \bar n_j {\bf P}_{ij}^\kappa + \bar n_k {\bf P}_{ik}^\kappa \,. \end{eqnarray} In Fig.~\ref{fig:g}, we show an example where the galaxies with $z< z_{\rm median}$ are binned into $n_1$ and the rest into $n_2$. Here and throughout we will take our fiducial cosmology as an adiabatic CDM model with matter density $\Omega_m=0.35$, dimensionless Hubble constant $h=0.65$, baryon density $\Omega_b=0.05$, cosmological constant $\Omega_\Lambda=0.65$, neutrino mass $m_\nu=0.7$ eV, the initial potential power spectrum amplitude $A$, and tilt $n_S=1$. We also plot in Fig. \ref{fig:g} the lensing efficiency function $g_i(D)$. Notice that despite having non-overlapping source distributions (upper panel), the lensing efficiencies strongly overlap (bottom panel) implying that the resulting convergence maps will have a correspondingly large cross correlation. This is of course because the high and low redshift galaxies alike are lensed by low-redshift structures. Also for this reason, there will be always be a stronger signal in the high redshift bins. This fact will be important for signal-to-noise considerations in choosing the bins. All of these properties can be seen in Fig.~\ref{fig:power}, where we plot the resultant power spectra and their cross correlation for the equal binning of Fig.~\ref{fig:g}. \begin{figure}[t] \centerline{\epsfxsize=3.5truein\epsffile{fig2.eps}} \caption{Power spectra and cross correlation for a subdivision in two across the median redshift $z_{\rm median}=1$ and errors for a survey of $5^{\circ}$ on the side, $\left< \gamma_{\rm int}^2 \right>^{1/2}=0.4$, and $\bar n = 2 \times 10^5$ deg$^{-2}$. Note the strong correlation ${\bf R}_{ij}$ between the two power spectra make the combination of the power spectra less constraining than a naive interpretation of the individual errors would imply.} \label{fig:power} \end{figure} \section{Redshift Binning and Parameter Estimation} While subdividing the sample into finer bins always increases the amount of information, there are two considerations that limit the effectiveness of redshift divisions. The first is set by the shot noise from the intrinsic ellipticities of the galaxies. Once the number density $\bar n_i$ per bin is so small that shot noise surpasses the signal in equation (\ref{eqn:covariance}), further subdivision no longer helps. The point at which this occurs depends on the angular scale of interest. The greater number of galaxies encompassed by the larger angular scales boosts the signal to noise (see Fig.~\ref{fig:power} and \cite{Kai92} 1992). Based on this criterion, one should separately subdivide the data to extract the maximal large and small angle information. However there is a second consideration. If the lensing signal does not change significantly across the redshift range of the whole distribution, then subdivision will not add information. These considerations can be quantified by considering the correlation coefficient between the power spectra of the subdivisions: ${\bf R}_{ij} = {\bf P}^\kappa_{ij}/ ({\bf P}^\kappa_{ii} {\bf P}^\kappa_{jj})^{1/2}$. For the model of Fig.~\ref{fig:power}, the power spectra are highly correlated (${\bf R}_{12} \sim 0.8$) even with only two subdivisions. Thus even though there is enough signal to noise to subdivide the sample further, one gains little information by doing so. One can combine these two considerations by diagonalizing the covariance matrix and considering the signal to noise in the diagonal basis. The appropriate strategy for subdivision depends on the true redshift distribution of the galaxies and the model for structure formation. One should therefore perform this test on the actual data to decide how to subdivide the sample. Nevertheless, to make these considerations more concrete, let us consider the specific goal of measuring the cosmological parameters $p_\alpha$ assuming that the underlying adiabatic CDM cosmology described above is correct. The Fisher information matrix can be used to quantify the effect of subdivision. It is defined as \begin{equation} {\bf F}_{\alpha\beta} = -\left< \partial^2 \ln L \over \partial p_\alpha \partial p_\beta \right>_ {\bf x} \,, \end{equation} where $L$ is the likelihood of observing a data set ${\bf x}$ given the true parameters $p_1 \ldots p_\alpha$. Generalizing the results of \cite{HuTeg99} (1998) to multiple correlated power spectra, we obtain\footnote{When taking these derivatives the redshift distribution $n_i(z)$ is held fixed as opposed to the distance distribution $[n_i(z)dz/dD]$ in equation~(\ref{eqn:distribution}).} \begin{equation} {\bf F}_{\alpha\beta} = \sum_{\ell=2}^{\ell_{\rm max}} ({\ell+1/2}) f_{\rm sky} {\rm tr} [{\bf C}^{-1} {\bf C}_{,\alpha} {\bf C}^{-1} {\bf C}_{,\beta}]\,, \label{eqn:Fisher} \end{equation} under the assumption of Gaussian signal and noise, where $f_{\rm sky}$ is fraction of sky covered by the survey, the covariance matrix ${\bf C}$ was defined in equation~(\ref{eqn:covariance}), and commas denote partial derivatives with respect to the cosmological parameters $p_\alpha$. We take $\ell_{\rm max}=3000$ to approximate the increased covariance due to the nonlinearities producing non-Gaussianity in the signal (\cite{ScoZalHui99} 1999). Since the variance of an unbiased estimator of a parameter $p_\alpha$ cannot be less than $\sigma(p_\alpha)=({\bf F}^{-1})_{\alpha\alpha}$, the Fisher matrix quantifies the best statistical errors on parameters possible with a given data set. For the purposes of this work, the absolute errors on parameters are less relevant than the improvement in errors from subdividing the data (see \cite{HuTeg99} for the former). We therefore test a 4 dimensional subset of the adiabatic CDM parameter space to see how subdivision can help separate initial power ($A$) from the various contributors to the redshift-dependent evolution of power ($\Omega_\Lambda$,$\Omega_K$,$m_\nu$). For reference the standard errors $\sigma_\alpha$ for this parameter space without subdivision are given in Table 1. Errors in the full parameter space would be increased but note that the neglected parameters $(\Omega_m h^2$,$\Omega_b h^2$, and $n_S$) are exactly those that the CMB satellite experiments should constrain precisely (see e.g. \cite{Junetal96} 1996; \cite{EisHuTeg99} 1999). As an example, we take a sample with $z_{\rm median}=1$ and $\bar n = 2 \times 10^5$ deg$^{-2}$ as appropriate for a magnitude limit of $R \sim 25$ (see \cite{Smaetal95b} 1995b). The signal to noise in the full sample is quite high, e.g.\ at $\ell=100$, $S/N \sim 25$. Thus we expect that subdividing the sample should improve parameter estimation. \begin{center} {TABLE 1\\[4pt] \scshape Parameter Estimation for $z_{\rm median}=1$} \\[3pt] \begin{tabular}{llllllll} }%\vspace{2pt}\tableline\tableline}%\vspace{2pt} }%\vspace{2pt}\tskip $p_\alpha$ & $\sigma_\alpha f_{\rm sky}^{1/2}$ & \multicolumn{6}{c}{Error Improvement} \\ & 1 & 2(${1 \over 2}$) & 2(${1 \over 4}$) & 2(${1 \over 8}$) & 3(${1 \over 3}$) & 3(${1 \over 4}$) & 3(${1 \over 8}$) \vphantom{\Big[}\\ }%\vspace{2pt}\tableline}%\vspace{2pt}\tskip $\Omega_\Lambda$ & 0.040 & 6.5 & 6.9 & 5.7 & 7.2 & 7.7 & 6.9\\ $\Omega_K$ & 0.023 & 2.9 & 3.1 & 2.9 & 3.3 & 3.5 & 3.2\\ $m_\nu$ & 0.044 & 1.7 & 2.0 & 2.1 & 2.1 & 2.2 & 2.2\\ $\ln A$ & 0.064 & 1.7 & 2.0 & 2.0 & 2.1 & 2.2 & 2.1 \end{tabular} \end{center} As shown in Table~1, subdividing this sample in equal halves, denoted as 2(1/2), improves the errors $\sigma_\alpha$ by a factor of 2 to 7. Since the signal in the lower redshift bin is smaller than in the higher redshift bin, it suffers comparatively more from the intrinsic noise variance. One can optimize the binning to correct for this effect. Dividing the sample so as to isolate the upper quarter [2(1/4)] improves the errors modestly whereas isolating the upper eighth deproves them. We plot the full range as a function of the fraction of galaxies in the upper bin in Fig.~\ref{fig:improvement}. Notice that though the improvement factor is roughly flat from $0.15-0.5$, it drops rapidly when noise dominates either the upper or lower fraction. If the signal were the same in both bins, this would occur at $0.04$ and $0.96$ for $\ell=100$. The fact that the true improvement is skewed to smaller upper fractions reflects the fact that the signal increases to higher redshifts. \begin{figure}[t] \centerline{\epsfxsize=3.5truein\epsffile{fig3.eps}} \caption{Tomographic error improvements on $\Omega_\Lambda$ for $z_{\rm median}=1$. Upper panel: improvement as a function of the fraction of galaxies in the upper redshift bin for 2 bins versus 3 bins (same fraction in upper two bins). Lower panel: redshift corresponding to the upper division.} \label{fig:improvement} \end{figure} Moving to three divisions makes only a small improvement over two. In Table~1 we give the results of taking 3 bins with an equal number of galaxies in the upper two bins, e.g.\ [3(1/4)] represents a division by number of ($1/2$, $1/4$, $1/4$). In fact the errors for three bins can be higher than those with two if not chosen wisely. We conclude that for a redshift distribution of the form given by equation~(\ref{eqn:distribution}) with $z_{\rm median}=1$, $\alpha=1$ and $\beta=4$, crude partitioning suffices to regain most of the redshift information in adiabatic CDM models where the change in the growth rate across the distribution is slow and controlled by a small number of cosmological parameters How robust are these conclusions against changes in the distribution and model? A wider redshift distribution offers greater opportunities for tomography. For example, let us widen the distribution by taking $\beta=2$ in equation~(\ref{eqn:distribution}). Then the gains by simply halving the distribution are a factor of 9.7 for $\Omega_\Lambda$; going to a 3(1/4) scheme raises this to 12. These considerations are also relevant for deeper surveys. Consider a survey with $z_{\rm median}=2$ and $\bar n=3.6 \times 10^5$ deg$^{-2}$. The parameter estimation results are given in Table 2. Not only is the overall improvement from subdivision larger (up to a factor of 24 for three bins) but the relative improvements between parameters also changes. This is because even within the adiabatic CDM paradigm the importance of the different parameters in determining the growth of structure depends on redshift. Perhaps the most important aspect of weak lensing tomography is that it has the ability to falsify the underlying adiabatic CDM model. For this reason, it is wise to examine the power spectra from the redshift bins directly, since these are the observables, rather than jump directly to modelling the data with parameters under the adiabatic CDM framework. For example, tomography may show that the component that accelerates the expansion of the universe is not the cosmological constant or call into question the fundamental assumption that structure forms through the gravitational instability of cold dark matter. \begin{center} {TABLE 2\\[4pt] \scshape Parameter Estimation for $z_{\rm median}=2$}\\[3pt] \begin{tabular}{lllllll} }%\vspace{2pt}\tableline\tableline}%\vspace{2pt} }%\vspace{2pt}\tskip $p_\alpha$ & $\sigma_\alpha f_{\rm sky}^{1/2}$ & \multicolumn{5}{c}{Error Improvement} \\ & 1 & 2(${1 \over 2}$) & 2(${1 \over 4}$) & 2(${1 \over 8}$) & 3(${1\over 4}$) & 3(${1\over 8}$) \vphantom{\Big[} \\ }%\vspace{2pt}\tskip\tableline}%\vspace{2pt}\tskip $\Omega_\Lambda$ & 0.063 & 19 & 21 & 20 & 24 & 24\\ $\Omega_K$ & 0.030 & 6.7 & 7.7 & 8.0 & 8.9 & 9.1\\ $m_\nu$ & 0.027 & 2.3 & 2.9 & 3.0 & 3.2 & 3.4 \\ $\ln A$ & 0.040 & 2.1 & 2.6 & 2.1 & 3.1 & 3.2 \end{tabular} \end{center} \section{Discussion} We have shown the precision with which cosmological parameters can be measured from a weak-lensing survey can be significantly enhanced by tomographically determining the evolution of the statistical properties of large-scale structure across the finite redshift width of the source distribution. Crude redshift binning of the data can recover most of the statistical information contained in the redshifts. For example, most of the gain for a magnitude limited survey with $z_{\rm median}=1$, under the adiabatic cold dark matter paradigm, comes from separating out the upper and lower redshift halves of the distribution. For wider distributions and stronger rates of change in the growth of structure, more information can be extracted by finer binning, especially of the higher redshift portion of the sample where the signal is greater. The appropriate number of bins can be empirically determined by examining the correlation between bins and the noise properties of the data. We have been assuming that the individual redshifts of the galaxies will be known sufficiently precisely to determine the redshift distribution of the subsamples. Realistically, the redshift information will be limited by the accuracy of photometric redshift techniques which currently show errors of $\Delta z \sim 0.1$ (68\% CL) for $0.4 \lesssim z \lesssim 1.4$ (\cite{Hogetal98} 1998). While statistical errors on the large samples of galaxies considered above are negligible, systematic errors or biases in the technique may cause problems. It is beyond the scope of this letter to test these issues fully. To give some feel for their effect, let us consider the median redshift $z_{\rm median}$ as an additional parameter with a prior uncertainty from photometric redshifts of the full individtual error $0.1$. Including this uncertainty degrades the precision in the parameters by $3\%$ in the worst case. While this effect is negligible, more worrying is a bias that is a function of redshift, especially in the largely untested regime $1.4 \lesssim z \lesssim 2$, as that can shift the difference between the power spectra of the subdivisions. Isolating the few percent of galaxies at $z \gtrsim 2.5$, where the techniques are tested, yields gains that are comparable to the optimal division (see Fig.~\ref{fig:improvement} lower panel), but the compactness of such galaxies poses an obstacle for measuring the lensing distortion from the ground (\cite{Steetal96} 1996). Despite these caveats, this study shows that tomography with weak lensing is both possible and would substantially improve the precision with which we can measure the growth of structure in the universe. {\it Acknowledgements:} I would like to thank D.J. Eisenstein, D.W. Hogg, J. Miralda-Escude, D.N. Spergel, M. Tegmark, J.A. Tyson, M. White, and D. Wittman for useful conversations. W.H.\ is supported by the Keck Foundation, a Sloan Fellowship, and NSF-9513835
\section{Physics motivation} Recent interest in R-parity violating (RPV) SUSY decay modes is motivated by the possible high-$Q^2$ event excess at HERA~\cite{HERA}. When interpretation of the excess through first-generation leptoquarks was excluded by the D\O~\cite{D0-lq} and CDF~\cite{CDF-lq} experiments, it was suggested~\cite{HERA-RPV} that such an effect could be explained via the $s$-channel production of a charm or top squark decaying into the $e + jet$ final state. Both the production and the decay vertices would thereby violate R-parity. Although more recent data has not confirmed the previous event excess, and despite the combined analysis showed that the anomalous events reported by the H1 and ZEUS experiments were unlikely to originate from the production of a single $s$-channel narrow resonance~\cite{Bassler}, interest in RPV signatures has not abatted. The CDF and D\O\ Collaborations have recently performed searches for RPV SUSY~\cite{CDF-RPV,D0-RPV}, and have set new mass limits on the RPV SUSY particles. Both experiments focussed their searches on the $\lambda'$ couplings, as motivated by the high-$Q^2$ HERA event excess. The results of the D\O\ searches are extended to the Run~2 case and the expected sensitivity to the RPV couplings is discussed. \section{D\O\ Search for RPV neutralino decays} The D\O\ search for RPV SUSY considered the case of neutralino LSP which decays into a lepton and two quarks due to a finite RPV $\lambda'$ coupling (see Fig.~\ref{fig:RPV-decay}). Both the electron and muon decay channels were considered, corresponding to what commonly referred to as $\lambda'_{1ij}$ and $\lambda'_{2ij}$ couplings, respectively. The corresponding final states contain either $2e $ or $2\mu $ and at least four accompanying jets. Unlike at HERA, this search is not sensitive to the value of the RPV coupling, as long as it is large enough so that the neutralino decays within the D\O\ detector. That corresponds to $\lambda' \geq 10^{-3}$, which gives a lot of room, given current indirect constraints~\cite{Herbi}. \begin{figure}[thb] \centerline{\protect\psfig{figure=rpv_decay.eps,height=1.8in}} \bigskip \caption{RPV decay of a neutralino LSP into a lepton and two quarks.} \label{fig:RPV-decay} \end{figure} We assume that the neutralino (LSP) pairs are produced in cascade decays of other supersymmetric particles and use all SUSY pair production mechanisms when generating signal events. Signal events were generated within the SUGRA framework with the following values of SUSY parameters: $A_0 = 0$, $\mu < 0$ and $\tan\beta = 2$ (the results are not sensitive to the value of $A_0$ .) Center of mass energy of the colliding beams was taken to be 2 TeV. {\footnotesize ISAJET}~\cite{ISAJET} was used for event generation. The acceptance and resolution of the D\O\ detector were parametrized using the following resolutions: $\delta E/E = 2 \% \oplus 15\%/\sqrt{E}~\mbox{[GeV]}$ (electrons), $\delta (1/p)/(1/p) = 0.018 \oplus 0.008 (1/p)$ (muons), and $\delta E/E = 3\% \oplus 80\%/\sqrt{E}~\mbox{[GeV]}$ (jets) and found consistent with the full detector simulation based on {\footnotesize GEANT}~\cite{GEANT}. \begin{figure}[htb] \vspace*{-0.3in} \centerline{\protect\psfig{figure=rpv_points.eps,width=4in}} \caption{Points in the $(m_0,m_{1/2})$ SUGRA parameter space used to generate RPV events in the $ee + 4$~jets channel.} \label{fig:RPV-points} \end{figure} Figure~\ref{fig:RPV-points} shows the points in the $(m_0,m_{1/2})$ SUGRA parameter space where signal Monte Carlo events were generated for the electron channel. Similar points were studied for the muon-decay channel. \section{Selection criteria for the dielectron channel} A multijet trigger was used for the analysis of Run~1 data. It was found to be nearly 100\% efficient for the typical RPV signal. Since Run~2 trigger list will include a similar trigger, we assume trigger efficiency of 100\% and do not perform any trigger simulations for the Run~2. The following offline selections were used: \begin{itemize} \item At least two good electrons, the leading one with $E_T(e) > 15$ GeV and the other one with $E_T(e) > 10$ GeV; \item Rapidity range $\mid \eta \mid \leq 1.1$ (central calorimeter), or $1.5 \leq \mid \eta \mid \leq 2.5$ (end calorimeters) for all the electrons; \item Energy isolation for the electrons: the EM energy in the R=0.2 cone about the center of gravity of the EM cluster, subtracted from the total energy in R=0.4 cone, should not exceed 15\% of the EM energy in the R-0.2 cone. \item At least four jets with $E_T(j) > 15 $ GeV and $\mid \eta \mid < 2.5$; \item The dielectron invariant mass ($M_{ee}$) should not be in the Z-mass interval, ie, $ \mid M_{ee} - M_Z \mid > 15$ GeV/$c^2$. \end{itemize} In the present analysis we have dropped the requirement on $H_T = \sum E_T(e) + \sum E_T(j)$ , but retained all other offline criteria that were used in the previous analysis of data from Run~I~\cite{D0-RPV}. \section{Selection in the dimuon channel} The following event selection requirements were used for the muon decay channel: \begin{itemize} \item Two muons, the leading one with $p_T >$ 15 GeV, and the other one with $p_T >$ 10 GeV. \item Rapidity range $|\eta| < 2.3$ for both muons. \item Energy isolation requirement for both muons, i.e. the calorimeter energy accompanying the muon in a ($\eta$ $\phi$) cone of 0.4 should be consistent with that from a minimum ionising particle. \item At least four jets with $E_T(j) > 15$ GeV and $|\eta| < 2.5$; \end{itemize} \section{Signal efficiencies} The number of signal events expected can be written as: $\langle N \rangle = \mathcal{L} \cdot \sigma \cdot \epsilon$, where $\langle N \rangle$ is the expected number of events for luminosity $ \mathcal{L} $, $\sigma$ is the cross-section, and $\epsilon$ is the overall efficiency. The efficiency $\epsilon$ can be split into three terms: $\epsilon = {\epsilon}_{\rm trig} \cdot {\epsilon}_{\rm kin} \cdot {\epsilon}_{\rm id}$. Here $\epsilon_{\rm trig}$ is the trigger efficiency for the events that pass the offline cuts ( assumed to be 100\%), $\epsilon_{\rm kin}$ is the efficiency for offline criteria, which includes kinematic, fiducial and topological requirements, and $\epsilon_{\rm id}$ is the electron/jet identification efficiency. The efficiency for identifying jets is very high ($> 95\%$) and is expected to stay the same in Run 2. Electron identification efficiencies in Run 1 were $80 \pm 7\%$ in the central ($|\eta| < 1.1$) and $71 \pm 7\%$ in the forward ($1.5 < |\eta| < 2.5$) regions~\cite{D0-RPV}. These efficiencies were calculated for electrons with $E_T(e) > 25$~GeV, It drops by about 30\% for electrons with $E_T(e) = 10$~GeV. The muon identification efficiencies used in Run 1 were $62 \pm 2\%$ in the central ($|\eta| < 1.0$) and $24 \pm 4\%$ in the forward ($1.0 < |\eta| < 1.7$) regions~\cite{rvmu2}. These were calculated for muons with $p_T > 15$~GeV. For muons with $10~\mbox{GeV} < p_T < 15$~GeV the efficiencies were 80\% smaller on average~\cite{rvmu3}. In the present analysis we have taken the overall particle identification efficiency to be $0.90 \pm 0.09$ in each channel, independent of lepton $E_T$, primarily due to the expectation of a better tracker and muon spectrometer for the upgraded D\O\ experiment. \section{Backgrounds} The main backgrounds are expected to arise from Drell-Yan production in association with four or more jets, dilepton top-quark events, and QCD multijet events. The latter is the dominant background for the electron channel (followed by the Drell-Yan background). In the case of muons, the background is dominated by the Drell-Yan and top pair production. We used Monte Carlo to calculate background from the first two sources, and data to estimate background from QCD jets. Background for the Run 1 analysis was estimated to be $1.8\pm 0.2\pm 0.3$ (with $1.27 \pm 0.24$ from QCD and $0.42\pm 0.15\pm 0.16$ from the other processes) for $\sim 100\ pb^{-1}$ of data. To extrapolate this number to the data set from Run 2, we have simply multiplied it by the ratio of luminosities to obtain $36\pm 4 \pm 6$ events. However, it is expected that due to the central magnetic field in the upgraded D\O\ detector, the probability of jets to be misidentified as electrons will be reduced by a factor of $\sim 2$ in Run 2. We have therefore considered a second scenario with the smaller expected background of $15 \pm 1.5 \pm 1.5$ events. For the muon channel, the expected background has been scaled directly from the Run 1 analysis. We expect $10 \pm 1 \pm 1$ background events in Run 2. \section{Results} In order to obtain the sensitivity of Run 2 in to RPV decays, we calculated the efficiency for signal for all the mass points shown in Fig.~\ref{fig:RPV-points}. Typical efficiencies, the signal cross section in the $ee + 4$~jets channel, and the expected event yield in 2~$fb^{-1}$ of data, for several representative $(m_0,m_{1/2})$ points, are given in Table~\ref{table:RPV-eff}. Similar numbers are obtained for the muon channel. \begin{table}[htb!] \begin{center} \caption{Efficiency $\times$ BR (\%), signal cross section and the expected event yield in 2~$fb^{-1}$ of data, at various $(m_0,m_{1/2})$ parameter space points. } \begin{tabular}{||c|c|c|c|c||} \hline $m_0$ & $m_{1/2}$ & Efficiency $\times$ BR (\%) & Cross section & $\langle N \rangle$ \\ (GeV) & (GeV) & & (pb) & (in 2 $fb^{-1}$) \\ \hline 60 & 235 & $7.9\pm 1.1 $ & 0.16 & $25.2 \pm 3.4 $ \\ 60 & 245 & $8.3\pm 1.1 $ & 0.08 & $12.8 \pm 1.7 $ \\ 60 & 255 & $8.3\pm 1.1 $ & 0.06 & $10.5 \pm 1.4 $ \\ 100 & 220 & $6.1\pm 0.8 $ & 0.10 & $12.2 \pm 1.7 $ \\ 100 & 230 & $7.0\pm 1.0 $ & 0.08 & $11.3 \pm 1.5 $ \\ 180 & 240 & $7.0\pm 0.9 $ & 0.05 & $7.1 \pm 1.0 $ \\ 320 & 240 & $7.1\pm 0.9 $ & 0.05 & $6.9 \pm 1.0 $ \\ \end{tabular} \end{center} \label{table:RPV-eff} \end{table} We use these efficiencies to obtain exclusion limits in the $(m_0,m_{1/2})$ plane at 95\% CL, assuming that no excess of events will be observed above the predicted background. The exclusion contours for the electron and muon channel are shown in Fig.~\ref{fig:RPV-e} and \ref{fig:RPV-mu}, respectively. Numerical values of the limits are summarized in Table~\ref{table:RPV-limits}. \begin{table}[hbt!] \begin{center} \caption{Lower limits on the squark and gluino masses from Run 2.} \begin{tabular}{||c|c|c|c||} \hline ~ & Lower limit on $m_{\tilde{q}}$ & Lower limit on $m_{\tilde{g}}$ & Limit when $m_{\tilde{q}} = m_{\tilde{g}}$ \\ ~ & ( For any $m_{\tilde{g}}$) & ( For any $m_{\tilde{q}}$ ) & \\\hline \multicolumn{4}{||c||}{Electrons}\\ \hline Run 1 & 252 GeV & 232 GeV & 283 GeV \\ Run 2 (Scenario I) & 430 GeV & 490 GeV & 490 GeV \\ Run 2 (Scenario II) & 520 GeV & 575 GeV & 585 GeV \\\hline \multicolumn{4}{||c||}{Muons}\\\hline Run 2 & 560 GeV & 640 GeV & 665 GeV \\ \end{tabular} \end{center} \label{table:RPV-limits} \end{table} \begin{figure}[t] \centerline{\protect\psfig{figure=rpv_electrons.eps,width=\textwidth}} \caption{Estimated exclusion contour for Run 2 in the $(m_0,m_{1/2})$ plane for $tan \beta = 2$, $A_0=0$, $\mu <0$, from the $ee + 4$~jets channel. Scenario I corresponds to a background of $36\pm 4 \pm 6$ events (direct scaling from Run 1); scenario II uses the background of $15 \pm 1.5 \pm 1.5$ events (scaling, but with improvements in the detector taken into account).} \label{fig:RPV-e} \end{figure} It's worth mentioning that our analysis provides a conservative estimate of the sensitivity achievable in Run~2, since no formal optimization of the signal vs. background has been performed. We expect that a formal optimization can improve the sensitivity in the mass reach by 15--20\%. \begin{figure}[t] \vspace*{0.1in} \centerline{\protect\psfig{figure=rpv_muons.eps,width=\textwidth}} \caption{Estimated exclusion contour for Run 2 in the $(m_0,m_{1/2})$ plane for $tan \beta = 2$, $A_0=0$, $\mu <0$, from the $\mu\mu + 4$~jets channel for background of $10 \pm 1.0 \pm 1.0$ (direct scaling from Run 1).} \label{fig:RPV-mu} \vfill \end{figure}
\section{Introduction} It is often claimed that chiral interaction of two-dimensional fermionic gauge models poses an obstruction to gauge symmetry. In this paper we clarify several aspects of this question for different regularizations of the chiral fermionic determinant, including the new Faddeevian regularization case proposed by Mitra\cite{PM}, under the point of view of the Stone's soldering formalism\cite{ms}. It is worth mentioning that understanding the properties of 2D fermionic actions is crucial in several aspects. For instance, the 1-cocycle necessary in recent discussions on smooth functional bosonization \cite{DNS,DS}, which is just the expression of the 2D anomaly, is known to be the origin of higher dimensional anomalies through a set of descent equations\cite{DTMP}. Incidentally, the anomaly phenomenon still defies a complete explanation. This paper is devoted to analyze and explore the restrictions that the soldering mechanism \cite{ms,ADW,W,AW} imposes over the regularization ambiguity of 2D chiral fermionic determinants. The soldering technique that is dimensionally independent and designed to work with dual manifestations of some symmetry is well suited to deal with the chiral character of 2D anomalous gauge theories. Recently \cite{ABW} a new interpretation for the phenomenon of dynamical mass generation known as Schwinger mechanism\cite{ls}, has been proposed which explores the ability of the soldering formalism to embrace interference effects. In that study the interference of right and left gauged Floreanini-Jackiw chiral bosons \cite{FJ} was shown to lead to a massive vectorial mode, for the special case where the Jackiw-Rajaraman (JR) regularization parameter is $a=1$ \cite{JR}\cite{RR}. After the discovery that the $\chi QED_{2}$ could be consistently quantized if the regularization ambiguity were properly taken into account, the investigation on this subject has received considerable attention and emphasis \cite{many}\cite{AAR}. The quantization of the model was considered from different points of view, both canonical and functional and the spectrum and unitarity was analyzed by distinct techniques, including the gauge invariant Wess-Zumino formulation\cite{WZ}, with results consistent with Ref.\cite{JR}. Despite this spate of interest, a surprising new result was reported recently by Mitra \cite{PM} showing that a different regularization prescription was yet possible, leading to new consequences. He proposed a new (faddeevian) regularization class, materialized by a unique and conveniently chosen mass term leading to a canonical description with three constraints. Recall that in \cite{JR} and \cite{RR}, the Hamiltonian framework was structured in terms of two classes with two ($a > 1$) and four ($a=1$) second-class constraints respectively. Mitra's work brings a clear interpretation for the reasons leading the bosonization ambiguity to fit into three instead of two distinct classes, classified according to the number of constraints present in the model. It is the main goal of this paper to study the restrictions posed by the soldering formalism over this new regularization class. Since soldering has ruled out the two-constraint class solution of Jackiw-Rajaraman as being able to dynamically generate mass via right-left interference, we are led to ask if the new Faddeevian class of chiral bosons proposed by Mitra do interfere constructively to produce a massive vectorial mode. To find an answer to this question we review, in Section 2, the procedure of \cite{MM} to obtain the multi-parametric regularization effective action based on the Pauli-Villars regularization proposed in \cite{FS}. This effective action is the point of departure for an extention of the ambit of Ref.\cite{PM} that is needed to our purpose in this paper and to be developed in Section 3. The bosonised theory satisfying Faddeev's structure for the constraint algebra is studied in the canonical approach. The mass of the photon scalar field is computed and its dependence on the ambiguity parameter is shown to be tantamount to that in Ref.\cite{JR}; the massless sector however is more constrained than its counterpart in \cite{JR}, corroborating the results of \cite{PM}. The restrictions imposed by the soldering are worked out in Section 4. We find the striking new result that the interference effects lift the parameter dependence by discriminating the value of the only non ambiguous class. Our results give a clear interpretation for the Schwinger mechanism as a left-right interference phenomenon, as suggested by Jackiw\cite{DTMP}. Our findings are further discussed in the final section. \section{The Effective Action} In a gauge invariant theory, free of anomalies, the canonical description reveals a couple of first-class constraints, with the Gauss law $G(x)$ appearing as the secondary constraint for the momentum $\pi_0(x)$ corresponding to the scalar potential $A_0(x)$. In an anomalous gauge theory, on the contrary, gauge invariance is lost and the constraint algebra for the gauge generator becomes afflicted by the presence of a Schwinger term \begin{eqnarray} \label{G1} \left [G(x) , \pi_0(y)\right ] &=& 0\nonumber\\ \left [ G(x) , G(y)\right ] &=& \imath \, \hbar \,{\cal C} \,\delta '(x-y) , \end{eqnarray} where ${\cal C}$ is some constant. This structure introduces extra degrees of freedom into the quantum theory as argued by Faddeev\cite{LF}. The quantum chiral Schwinger model with the usual regularization ($a \geq 1$) does have more degrees of freedom than its classical counterparts, as expected, but does not fit into Faddeev's scheme above due to the functional dependence of the Gauss generator on the scalar potential, which leads to a different constraint algebra than (\ref{G1}), \begin{eqnarray} \label{G2} \left [G(x) , \pi_0(y)\right ] &\not=& 0\nonumber\\ \left [ G(x) , G(y)\right ] &=& 0 . \end{eqnarray} The second-class nature of the set is then due to the non-commutative character of the primary and secondary constraints. The new regularization class for the fermionic determinant proposed by Mitra has the virtue of fitting perfectly into Faddeev's picture. In this section we shall review the computation of the fermionic determinant leading to this new scheme. Our starting point is the action for fermionic sector of the chiral Schwinger model, \begin{equation} S = \int\,d^{2}x\,\,\bar{\psi}(x)\,\left[ i\partial\!\!\!/ - q\,\sqrt{\pi}\,A\!\!\!/(x)\,\left(1 + i\,\gamma_{5}\right)\right]\, \psi(x) \label{2.1} \end{equation} where $\psi(x)$ is a fermionic field and $A_{\mu}$ is the vector gauge field in a (1 + 1) dimensional spacetime. From this classical action we obtain the following effective action\cite{FS} \begin{equation} \exp\,i\,S_{eff}^{(0)}[A(x)] = \int\,{\cal D}\psi(x)\,{\cal D}{\bar\psi}(x)\, \exp\,i\,S\left[{\bar\psi}(x),\psi(x),A(x)\right]. \label{2.2} \end{equation} In a formal level this is a nonlocal action that reads, \begin{equation} \exp\,i\,S_{eff}^{(0)}[A(x)] = - q^{2}\,\int\,d^{2}x\,A^{\mu}(x)\, \left(\eta_{\mu\alpha} + \epsilon_{\mu\alpha}\right)\, \frac{\partial^{\alpha}\partial^{\beta}}{\partial^{2}}\, \left(\eta_{\beta\nu} - \epsilon_{\beta\nu}\right)\, A^{\nu}(x) , \label{2.3} \end{equation} but there is an ambiguity related to the regularization procedure adopted. Let us discuss the regularization procedure proposed by Frolov and Slavnov\cite{FS}. To this end we add a multi-parametric regularising action \begin{equation} S_{reg}[A(x)] = \sum_{r=1}^{2n-1}\int\,d^{2}x\,\, {\bar \psi}_{r}(x)\,\left[ i\,\partial\!\!\!/ - m_{r} - q\,\sqrt{\pi}\, A_\mu(x)\,\Gamma_{r}^{\mu}\right]\,\psi_{r}(x) \label{2.4} \end{equation} where \begin{equation} \Gamma_{r}^{\mu} = \left[a_{r}\,K^{\mu\nu}\, \left(1 + i\gamma_{5}\right) + b_{r}\,\Sigma^{\mu\nu}\, \left(1 - i\gamma_{5}\right)\right]\,\gamma_{\nu} . \label{2.5} \end{equation} Here $\psi_{r}(x)$ are the regulators fields with mass $m_{r}$ whose couplings $\Gamma_{r}^{\mu}$ (or $K^{\mu\nu}$ and $\Sigma^{\mu\nu}$) are matrices which will be determined later. These regulators bring up the following partition function \begin{equation} \exp\,i\,S_{reg}^{eff}[A] = \int\,\Pi_{r}\,{\cal D}\psi(x)\,{\cal D}{\bar\psi}(x)\, \exp\,i\,S_{reg}\left[{\bar\psi}_{r}(x),\psi(x)_{r},A(x)\right] \label{2.6} \end{equation} which can be solved to \cite{MM} \begin{equation} S_{reg}^{eff}[A] = -q^{2}\,\frac{\pi}{2}\,\int\,d^{2}x\, A_{\mu}(x)\,G^{\mu\nu}(x,y)\,A^{\nu}(y) \label{2.7} \end{equation} with \begin{equation} G^{\mu\nu}(x,y) = \int\,\frac{d^{2}p}{(2\pi)^{2}}\,{\bar G}^{\mu\nu}(p)\, \exp\left[-i \cdot p\, (x - y)\right] . \label{2.8} \end{equation} Now ${\bar G}_{\mu\nu}(p)$ is found to be \[ {\bar G}^{(r)}_{\mu\nu}(p) = \frac{1}{\pi}\,\left\{\left( a_{r}^{2}\,T^{1}_{\mu\nu\lambda\kappa} + b_{r}^{2}\,T^{1}_{\mu\nu\lambda\kappa} \right)\,\left[2\,\left(1 + A_{r}\right)\left(\eta^{\lambda\kappa} - \frac{p^{\lambda}\,p^{\kappa}}{p^{2}}\right) + A_{r}\,\eta^{\lambda\kappa}\right]+ 2A_{r} a_{r} b_{r} M_{\mu\nu}\right\} \] where \begin{equation} A_{r} = 1 - \frac{i}{y_{r}}\,\ln(-1) + {\cal O}(y_{r}) \end{equation} and also \begin{eqnarray} T_{\mu\nu\lambda\kappa}^{1} &=& K_{\mu\rho}\, \left(\delta_{\lambda}^{\rho} + \epsilon_{\kappa}^{\sigma}\right)\, K_{\nu\sigma}\, \left(\delta_{\kappa}^{\sigma} + \epsilon_{\kappa}^{\sigma}\right) \nonumber \\ T_{\mu\nu\lambda\kappa}^{2} &=& \Sigma_{\mu\rho}\, \left(\delta_{\lambda}^{\rho} + \epsilon_{\kappa}^{\sigma}\right)\, \Sigma_{\nu\sigma}\, \left(\delta_{\kappa}^{\sigma} + \epsilon_{\kappa}^{\sigma}\right) \nonumber \\ M_{\mu\nu} &=& \left[ K_{\mu\lambda}\,\left(\eta^{\lambda\kappa} - \epsilon^{\lambda\kappa}\right)\, \Sigma_{\nu\kappa} + \Sigma_{\mu\lambda}\, \left(\eta^{\lambda\kappa} - \epsilon^{\lambda\kappa}\right)\, K_{\nu\kappa}\right] \nonumber \\ y_{r}^{2} &=& \frac{p^{2}}{m_{r}^{2}} . \end{eqnarray} Imposing the conditions \cite{FS} \begin{eqnarray} \sum_r \epsilon_r a_r^2 &=& \sum_r \epsilon_r b_r^2=0\nonumber\\ \sum_r \epsilon_r m_r a_r^2 &=& \sum_r \epsilon_r m_r b_r^2= \sum_r \epsilon_r m_r a_r b_r =0\nonumber\\ 2\sum_r \epsilon_r a_r b_r &=& 1 \end{eqnarray} where $\epsilon_r = (-1)^{r+1}$ is the Grassman parity and then letting $m_{r} \rightarrow \infty$ we get, \begin{equation} S_{reg}^{eff}[A] = q^{2}\,\frac{1}{2}\,\int\,d^{2}x\,A_{\mu}(x)\, M^{\mu\nu}(x,y)\,A^{\nu}(y). \label{2.16} \end{equation} Jackiw and Rajaraman found a regularized solution with a diagonal choice for the matrix \begin{eqnarray} M^{\mu\nu}\,=\,\pmatrix{a & 0 \cr 0 & a \cr}\, \delta(x - y), \label{Matrix0} \end{eqnarray} with $a\geq 0$, corresponding to the cases with two and four-constraint's classes. The physical content of these cases, as disclosed by them, was found to correspond to an $a$-dependent massive photon field and a massless fermion for the former, while in the later the photon field was absent. Mitra noticed that the alternative choice \begin{eqnarray} M^{\mu\nu}\,=\,\pmatrix{1 & -1 \cr -1 & -3 \cr}\, \delta(x - y), \label{Matrix1} \end{eqnarray} leads to a new class of solutions with three second-class constraints and found that the physical spectrum of the model contains a chiral fermion and a photon field with mass $m=4\,q^2$. To work out the soldering formalism and obtain the interference contribution coming from the chiral fermions we need to generalize the regularization dependence of the effective action. This is done in the next section. \section{Hamiltonian Analysis and Spectrum} In their seminal work Jackiw and Rajaraman\cite{JR} showed that the $\chi QED_{2}$ could be consistently quantized by including the bosonization ambiguity parameter satisfying the condition $a \geq 1$ to avoid tachyonic excitations. Later on, working out the canonical structure of the model, Rajaraman\cite{RR} showed that the cases $a>1$ and $a=1$ belonged to distinct classes: the $a=1$ case represents the four-constraints class, while the $a>1$ class presents only two constraints. The latter is a continuous one-parameter class, while the former class is non ambiguous containing only one representative. The consequences of these distinct constraint structures are that the $a>1$ class presents, besides the massless excitation also a massive scalar excitation ($m^2=\frac{e^2a^2}{a-1}$) that is not found on the other case. In the canonical approach the commutator between the primary and the secondary constraints vanishes in the first case. The emergence of two more constraints completes the second-class set. Mitra found the amazing fact that with an appropriated choice of the regularization mass term it is possible to close the second-class algebra with only three constraints. His model is not manifestly Lorentz invariant, but the Poincar\'e generators have been constructed \cite{PM} and shown to close the relativistic algebra on-shell. The main feature of this new regularization is the presence of a Schwinger term in the Poisson bracket algebra of the Gauss law, which limits the set to only three second-class constraints. To see this let us write the CSM Lagrangian, with faddeevian regularization but with Mitra's regulator properly generalized to meet our purposes, \begin{equation} {\cal L} = -\frac 14 \, F_{\mu\nu}\,F^{\mu\nu} + \frac{1}{2}\,\partial_\mu\phi\,\partial^{\mu}\phi + q\,\left(g^{\mu\nu} + b\,\epsilon^{\mu\nu}\right)\, \partial_{\mu}\phi\,A_{\nu} + \frac{1}{2}\,q^2\,A_{\mu} M^{\mu\nu} A_{\nu}\,, \label{Lagrangian} \end{equation} where $F_{\mu\nu} = \partial_{\mu}A_{\nu}-\partial_{\nu}A_{\mu}$; $g_{\mu\nu} = \mbox{diag}(+1,-1)$ and $\epsilon^{01} = -\epsilon^{10} = \epsilon_{10} = 1$. $b$ is a chirality parameter, which can assume the values $b=\pm 1$. The mass-term matrix $M_{\mu\nu}$ is defined as \begin{eqnarray} M^{\mu\nu}\,=\,\pmatrix{1 & \alpha \cr \alpha & \beta \cr}\, \delta(x - y). \label{Matrix} \end{eqnarray} Notice that we have chosen unity coefficient for $A_{0}^{2}$ term. In a sense, this choice resembles Rajaraman's $a=1$ class and is the trademark of the faddeevian regularization. In fact, Rajaraman's class is a singular point in the ``space of parameters''. Its canonical description has the maximum number of constraints with no massive excitation. Such a case is found in (\ref{Lagrangian}) if we make $\alpha = 0$ in (\ref{Matrix}). The appearance of a new class needs a non vanishing value for $\alpha$. With Mitra's choice, $\alpha = -1$ and $\beta = -3$, the photon becomes massive ($m^2=4 \, q^2$), but the remaining massless fermion has a definite chirality, opposite to that entering the interaction with the electromagnetic field. This choice is, however, too restrictive and may be relaxed leading to new and interesting consequences. In this work the coefficients $\alpha$ and $\beta$ are in principle arbitrary, but the mass spectrum will impose a constraint between them. This is best seen in the Hamiltonian formalism. The canonical Hamiltonian is readily computed \begin{eqnarray} H = \int\!\!\! &dx& \!\!\! \left\{ \frac{1}{2}\,\left(\pi^{1}\right)^{2} + \frac{1}{2}\,\pi_{\phi}^{2} + \pi^{1}A_{0}^{'} + \frac{1}{2}\,\phi^{'2} + q\,(\,b\phi'\,-\,\pi_{\phi}\,)\,A_0 + q\left(\phi^{'} - b\,\pi_{\phi}\right)A_{1} + \right. \nonumber \\ & & \left. +\,q^{2}\left(b - \alpha\right)A_{0}A_{1} + \frac{1}{2}q^{2}\left(1-\beta\right)A_{1}^{2} \right\}\, . \label{Hamiltonian} \end{eqnarray} The stationarity algorithm leads a set of three constraints \begin{eqnarray} \label{omega} \Omega_{1} &=& \pi^{0} , \nonumber \\ \Omega_{2} &=& \left(\pi^{1}\right)^{'} + q\left(\pi_{\phi}-\,b\,\phi^{'}\right) - q^{2}\left(b - \alpha\right)A_{1} , \\ \Omega_{3} &=& -\left(b -\alpha\right)\,\pi^{1} + 2\,\alpha\,A_{0}^{'} + \left(1 +\beta\right)A_{1}^{'} ,\nonumber \end{eqnarray} which are easily seen to be second-class, viz. \begin{eqnarray} \left\{\Omega_{1}(x), \Omega_{3}(y)\right\} &=& 2\,\alpha\, \frac{\partial}{\partial x}\,\delta (x-y)\, , \nonumber \\ \left\{\Omega_{2}(x), \Omega_{2}(y)\right\} &=& -2\,q^2\,\alpha\, \frac{\partial}{\partial x}\,\delta (x-y)\, , \nonumber \\ \left\{\Omega_{2}(x), \Omega_{3}(y)\right\} &=& q^{2}\left(b-\alpha\right)^{2}\,\delta (x-y)- \left(1+\beta\right)\,\frac{\partial^{2}}{\partial x \, \partial y}\, \delta (x-y)\, , \nonumber \\ \left\{\Omega_{3}(x), \Omega_{3}(y)\right\} &=& -2\left(b-\alpha\right) \left(1+\beta\right)\frac{\partial}{\partial x}\,\delta (x-y)\,\,, \end{eqnarray} with the other brackets vanishing. This is in sharp contrast with the usual regularization possessing two or four second-class constraints. To perform quantization we compute the Dirac brackets, \begin{eqnarray} \label{db} \left\{\,\phi(x)\,,\,\phi(y)\,\right\}_{D} &=& -\frac{1}{4\,\alpha}\, \theta (x-y)\, , \nonumber \\ \left\{\phi(x),A_{1}(y)\right\}_{D} &=& -\frac{1}{2\,q\,\alpha}\, \delta (x-y)\, , \nonumber \\ \left\{\phi(x),\pi^{1}(y)\right\}_{D} &=& -\frac{q}{4\,\alpha} \left(b - \alpha\right)\, \theta (x-y)\, , \nonumber \\ \left\{A_{1}(x),A_{1}(y)\right\}_{D} &=& \frac{1}{2\,q^{2}\,\alpha}\, \frac{\partial}{\partial x}\, \delta (x-y)\, , \\ \left\{\pi^{1}(x),A_{1}(y)\right\}_{D} &=& -\left(\frac{b+\alpha}{2\,\alpha} \right)\delta (x-y)\, , \nonumber \\ \left\{\pi^{1}(x),\pi^{1}(y)\right\}_{D} &=& -\frac{q^{2}}{4\,\alpha} \left(b-\alpha\right)^{2} \theta (x-y)\,\,. \nonumber \end{eqnarray} The reduced Hamiltonian is obtained by strongly eliminating $\pi^{0}$, $A_{0}^{'}$ and $\pi_{\phi}$ from the constraints (\ref{omega}) and substituting in the canonical Hamiltonian (\ref{Hamiltonian}), \begin{eqnarray} \label{hr} H_{r} = \int\!\!\! &dx& \!\!\! \left\{ \frac{1}{2}\left(\pi^{1}\right)^{2} - \alpha\,\pi^{1}A_{1}^{'} + q\left(1-b\,\alpha\right)A_1\phi^{'} + \phi^{'2}-\frac{b}{q}\,\phi^{'} \left(\pi^{1}\right)^{'} + \right. \nonumber \\ && \left. + \frac{1}{2\,q^{2}} \left(\pi^{1}\right)^{'2} +\frac{1}{2}\,q^{2}\left(\alpha^{2}- \beta\right)A_{1}^{2}\right\}\, . \end{eqnarray} Making use of (\ref{db}) and (\ref{hr}) we get the following equations of motion for the remaining fields, \begin{eqnarray} \dot{\phi} &=& b\,\phi^{'} - \frac{1}{q} \left(\pi^{1}\right)^{'} + \frac{q}{2\,\alpha} \left(1 - 2\alpha^{2}+\beta\right)A_{1} \label{phi_eq}\, ,\label{phi} \\ \dot{\pi}^{1} &=& -b \left(\pi^{1}\right)^{'} +\frac{q^{2}}{2\,\alpha}\left[(b-\alpha) (1-\alpha^{2}) - (b+\alpha)(\alpha^{2}-\beta)\right]A_{1} \label{pi1}\, , \\ \dot{A}_{1} &=& \left(\frac{\alpha + b}{2\alpha}\right)\pi^{1} - \left(\frac{1 + \beta}{2\alpha}\right)A_{1}^{'}\,\, . \label{a1} \end{eqnarray} We are now ready to determine the spectrum of the model. Isolating $\pi^{1}$ from the Eq.(\ref{a1}) and substituting in the equation (\ref{pi1}), we will have \begin{eqnarray} \left(\frac{2\,\alpha}{\alpha + b}\right)\ddot{A}_{1}+ b\left(\frac{1 + \beta}{\alpha + b}\right)A_{1}^{''}\!\!\! &=& \!\!\! -\left(\frac{2\,b\,\alpha}{\alpha + b} + \frac{1+\beta}{\alpha +b}\right) \dot{A}_{1}^{'}\,+\, \nonumber \\ \!\!\! & + & \!\!\! \frac{q^2}{2\alpha}\left[\left(b - \alpha\right) \left(1-\alpha^{2}\right) - \left(b + \alpha\right) \left(\alpha^{2}-\beta\right)\right]A_{1}\, . \label{eq} \end{eqnarray} To get a massive Klein-Gordon equation for the photon field we must set \begin{equation} \label{condicao} \left(1 + \beta\right) + b\,\left(2\alpha\right) = 0\, , \end{equation} which relates $\alpha$ and $\beta$ and shows that the regularization ambiguity adopted in \cite{PM} can be extended to a continuous one-parameter class (for a chosen chirality). We have, using (\ref{eq}) and (\ref{condicao}), the following mass formula for the massive excitation of the spectrum, \begin{equation} m^{2} = q^{2}\,\frac{\left(1 + b\,\alpha\right)^{2}}{b\,\alpha}\, . \end{equation} Note that to avoid tachyonic excitations, $\alpha$ is further restricted to satisfy $b\,\alpha = |\alpha|$, so $\alpha \rightarrow -\alpha$ interchanges from one chirality to another. Observe that in the limit $\alpha \rightarrow 0$ the massive excitation becomes infinitely heavy and decouples from the spectrum. This leads us back to the four-constraints class. It is interesting to see that the redefinition of the parameter as $a=1+ |\alpha|$ leads to, \begin{equation} m^2 = \frac{q^2 a^2}{a-1} \end{equation} which is the celebrate mass formula of the chiral Schwinger model, showing that the parameter dependence of the mass spectrum is tantamount to both the Jackiw-Rajaraman and the faddeevian regularizations. Let us next discuss the massless sector of the spectrum. To disclose the presence of the chiral excitation we need to diagonalize the reduced Hamiltonian (\ref{hr}). This procedure may, at least in principle, impose further restrictions over $\alpha$. This all boils down to find the correct linear combination of the fields leading to the free chiral equation of motion. To this end we substitute $\pi^{1}$ from its definition and $A_{1}$ from the Klein-Gordon equation into equation (\ref{phi_eq}) to obtain \begin{eqnarray} \label{vinte} 0 &=&\frac{\partial}{\partial t}\left\{ \phi + \frac{q}{2\alpha} \left(\frac{2 + 2\,b\,\alpha - \alpha^{2}}{m^{2}}\right) \dot{A}_{1} + \frac{1}{q}\left(\frac{\alpha}{\alpha + b} \right)A_{1}^{'}\right\}\,-\, \nonumber \\ &-& \frac{\partial}{\partial x}\left\{ b\,\phi - \frac{1}{q}\left(\frac{\alpha}{\alpha + b}\right)\dot{A}_{1} + \left[\frac{q}{2\alpha}\left(\frac{2 + 2\,b\,\alpha -\alpha^{2}}{m^{2}}\right) -\frac{1}{q}\left(\frac{2\,b\,\alpha}{\alpha+b}\right)\right]\right\}\,\,. \end{eqnarray} This expression becomes the equation of motion for a self-dual boson $\chi$ \begin{equation} \dot{\chi} - b\,\chi^{'} = 0 \label{quiral} \end{equation} if we identify the coefficients for $\dot A_1$ and $A_1^{'}$ in the two independent terms of (\ref{vinte}) with, \begin{equation} \label{chi} \chi = \phi + \frac 1{q} \left(\frac{\alpha}{\alpha + b}\right)\left(A_{1}^{'}- b \dot{A}_{1}\right). \end{equation} This field redefinition, differently from the case of the massive field whose construction imposed condition (\ref{condicao}), does not restrain the parameter $\alpha$ any further. Using (\ref{omega}) and (\ref{quiral}), all the fields can be expressed as functions of the free massive scalar $A_1$ and the free chiral boson $\chi$, interpreted as the bosonized Weyl fermion. The main result of this section is now complete, i.e., the construction of the one-parameter class regularization generalizing Mitra's proposal. The stage is now set to study the interference of chiral actions with (one-parameter) faddeevian regularization. \section{Effects of Interference} In this section we use the soldering formalism introduced in \cite{ms} to examine the restriction imposed by chiral interference over the regularization ambiguity parameter when the faddeevian approach is adopted. This study, taken in the framework of the usual JR regularization, establishes a strong restriction over the parameter's values and gives rise to a new interpretation for the mechanism of dynamical mass generation occurring in the Schwinger model. This study is meaningful and necessary since a new class of theories with three second-class constraints has emerged: it must be verified if new solutions resulting from interference will lead to a gauge invariant massive excitation. To begin with, let us rewrite explicitly the two chiral actions presented in (\ref{Lagrangian}) in the appropriate light-cone variables, \begin{eqnarray} {\cal L}_+\,&=& \partial_{+}\rho\,\partial_{-}\rho + \frac{1}{2}\, \left(\partial_{-}A_{+} - \partial_{+}A_{-}\right)^{2} + 2\,q\,\partial_{-}\rho\,A_{+} -2\, q^{2}\,|\alpha|\,A_{-}^{2} +\nonumber \\ & &+\;q^{2}\,\left(1\,+\,|\alpha|\right)A_{-}A_{+} \label{mitra+}\\ {\cal L}_-\,&=&\partial_{+}\varphi\,\partial_{-}\varphi + \frac{1}{2}\, \left(\partial_{-}A_{+} - \partial_{+}A_{-}\right)^{2} + 2\,q\,\partial_{+}\varphi\,A_{-} -2\, q^{2}\,|\bar\alpha|\,A_{+}^{2} +\nonumber \\ & &+\;q^{2}\,\left(1\,+\,|\bar\alpha|\right)A_{-}A_{+} \;\;, \label{mitra-} \end{eqnarray} where we have used the convention ${\cal L}_{\pm} = {\cal L}|_{b}$. For clarity, we have used different fields $(\varphi, \rho)$ for opposite chiralities and the corresponding mass-term parameters $(\alpha,\overline{\alpha})$ to make clear that these chiral theories are uncorrelated. However by making use of soldering formalism we will get a meaningful combination of these components. The main point of soldering is to lift the global Nother symmetry of each chiral component to a local symmetry of the system as a whole. Showing only the main parts of the soldering formalism we can see that the axial transformation ($\delta\varphi =\delta\rho = \eta$) leads to \begin{eqnarray} \label{CW10} \delta {\cal L}_+\,&=&\, \partial_-\eta\,J_+(\rho) \nonumber \\ \delta {\cal L}_-\,&=&\, \partial_+\eta\,J_-(\varphi) \end{eqnarray} where $J_{-}(\varphi)=2(\partial_-\varphi\,+\,q A_-)$ and $J_{+}(\rho)=2(\partial_+\rho\,+\,q A_+)$ are the Noether's currents and $\eta$ is the gauge parameter. Next we introduce the soldering field $B_{\pm}$ appropriately coupled to the Noether currents to obtain the once iterated chiral actions as, \begin{eqnarray} {\cal L}_+^{(0)} &\rightarrow& {\cal L}_+^{(1)}\,=\,{\cal L}_+^{(0)}\,+\,B_+\, J_-(\varphi) \nonumber \\ {\cal L}_-^{(0)} &\rightarrow& {\cal L}_-^{(1)}\,=\,{\cal L}_-^{(0)}\,+\,B_-\, J_+(\rho) \;\; . \end{eqnarray} The soldering fields act as partial compensators for the variance (\ref{CW10}), transforming vectorially under the axial symmetry, $\delta B_{\pm}=\partial_{\pm}\eta$. It is now possible to define an effective Lagrangian invariant under the combined transformation of the chiral fields and compensators as, \begin{equation} {\cal L}_{eff}\,=\,{\cal L}_+^1\,+\,{\cal L}_-^1\,+\,2\,B_+\,B_-\;\;. \label{Leff} \end{equation} The soldered action is obtained using the fact that $B_{\pm}$ are auxiliary fields. Their elimination may be done altogether from their field equations but the effects of soldering will persist as a residual symmetry for the remaining fields. This will naturally cohere the otherwise independent chiral fields $\varphi$ and $\rho$ in the form of a soldered Lagrangian for a collective field $\Phi$ as, \begin{eqnarray} \label{solda} {\cal L}_{eff}\,&=&\,\partial_+\,\Phi\,\partial_-\,\Phi\,-\,2\,q\, (A_+\,\partial_-\, \Phi\,-\,A_-\,\partial_+\,\Phi)\,+\, \frac 12 \left(\partial_+ A_-\, -\, \partial_- A_+\right)^2\nonumber\\ \,&+&\,{q^2 \over 2}\left[\alpha\,A_+^2\, - \,\overline{\alpha}\,A_-^2\, - \,\left(\alpha - \,\overline{\alpha}\right) A_+\,A_-\,\right] \end{eqnarray} where $\Phi=\varphi-\rho$. Notice that except for the last term, the soldered action describes the massive gauge invariant bosonised version of the Schwinger model, with the gauge invariant collective field $\Phi$ playing the role the of the photon field. Gauge invariance imposes a strong constraint over the parameters as, \begin{equation} \alpha\,= \,\overline{\alpha} = 0. \end{equation} This value corresponds to the $a=1$, four-constraints regularization class. This is a remarkable result, consistent with \cite{ABW}. A notable feature of the present analysis is the disclosure of a new class of parameterizations and their dependence with the number of constraints. Different aspects of this feature were elaborated and the consequences of interference computed. To discuss further the implications of interference on chiral actions it is best to compare with the existing literature. This also serves to put the present work in a proper perspective. To be precise, it was initialy shown that in the faddeevian approach there are actually three second-class constraints with a real parameter dependence. To disclose this one-parameter dependence of the faddeevian regularization is a new interesting result. The counting of constraints explains the presence of only one chiral excitation in the spectrum (besides the massive mode). This is in contrast with the usual JR regularization where the massless excitation is scalar, and is essentially tied to the fact that this regularization is less constrained. The restrictions of soldering however confine the appearance of a massive vector excitation to the interference of modes belonging to the $a=1$ class that, being more constrained, has only a massless scalar in the spectrum. This might raise questions about the interference of the chiral modes in this class. It should be noticed however that the use of light-cone variables in the soldering constrains even further these chiral actions. Both the two and the four-constraints classes display chiral excitations instead of massless scalars. The original chiral mode of the Mitra's class therefore disappear in the presence of the extra light-cone constraint and there appears to exist an ambiguity challenging the real meaning of the soldering. In fact there are no massless particles in the spectrum of (\ref{mitra+}-\ref{mitra-}) for the light-cone setting. However, what is important to observe in this scenario is that the whole process of soldering is done in the Lagrangian framework, such that the limit $a\rightarrow 1$ is well defined. This is also valid for the JR regularization. The limit leads to the $a=1$ action and the canonical analysis may be done unambiguously. Oppositely, the Hamiltonian formulation has the $a=1$ point as a singularity, as shown in (\ref{db}). \section{Conclusions} In this work we studied the bosonized form of the CSM fermionic determinant adopting the three-constraints regularization parameterized by a single real number. This extends early regularizations proposed by JR and Mitra. Our results display a clear-cut separation of the existing classes shown to depend only on the number of second-class constraints. The new class with faddeevian regularization and three second-class constraints has been worked out in great detail. The spectrum has been shown to consist of a chiral boson and a massive photon field. The mass formula for the scalar excitation was shown to reproduce the JR result. Considerations of unitarity therefore restrain the range of the regularization parameter similarly. The use of the soldering formalism supplemented by gauge invariance restricts the otherwise arbitrary ambiguity parameter to the specific value $a=1$, which corresponds to the four-constraints class. This is new result that discriminates the special character of this unambiguous regularization point and gives a precise interpretation of the Schwinger dynamical mass generation mechanism as a consequence of right and left interference. To conclude we stress that the formalism and analysis proposed here illuminates the close connection among anomalous gauge theories, the interference phenomenon and the mechanism of dynamical mass generation, providing a variety of new possibilities with practical applications. \bigskip \noindent {\bf Acknowledgment:} This work is supported in part by CNPq, FINEP, CAPES, FAPESP and FUJB (Brazilian Research Agencies).
\section{#1}\setcounter{equation}{0}} \def\baselinestretch{1.1}\textheight 23.5cm\textwidth 16cm\parskip 1ex \oddsidemargin 0pt\evensidemargin 0pt\topmargin -40pt\jot = .5ex \setlength{\parskip}{2mm} \newcommand{\tilde{x}_t}{\tilde{x}_t} \newcommand{\rightarrow}{\rightarrow} \newcommand{\mu}{\tilde{M}_p} \newcommand{\tilde{g}_s}{\tilde{g}_s} \newcommand{\tilde{v}_t}{\tilde{v}_t} \newcommand{\tilde{M}_s}{\tilde{M}_s} \newcommand{\begin{equation}}{\begin{equation}} \newcommand{\end{equation}}{\end{equation}} \newcommand{\eel}[1]{\label{#1}\end{equation}} \newcommand{\begin{eqnarray}}{\begin{eqnarray}} \newcommand{\end{eqnarray}}{\end{eqnarray}} \newcommand{\eeal}[1]{\label{#1}\end{eqnarray}} \newcommand{\begin{equation}\begin{array}{rcl}}{\begin{equation}\begin{array}{rcl}} \newcommand{\end{array}\end{equation}}{\end{array}\end{equation}} \newcommand{\eaql}[1]{\end{array}\label{#1}\end{equation}} \newcommand{\begin{equation}\begin{array}{rcl}}{\begin{equation}\begin{array}{rcl}} \newcommand{\eeacn}[1]{\end{array}\label{#1}\end{equation}} \newcommand{\begin{array}}{\begin{array}} \newcommand{\end{array}}{\end{array}} \newcommand{\nonumber \\}{\nonumber \\} \newcommand{\equ}[1]{(\ref{#1})} \renewcommand{\a}{\alpha} \renewcommand{\b}{\beta} \renewcommand{\d}{\delta} \newcommand{\Delta}{\Delta} \newcommand{\varepsilon}{\varepsilon} \newcommand{\lambda} \newcommand{\La}{\Lambda}{\lambda} \newcommand{\La}{\Lambda} \newcommand{\nu}{\nu} \newcommand{\omega} \newcommand{\OM}{\Omega}{\omega} \newcommand{\OM}{\Omega} \newcommand{\psi} \newcommand{\PS}{\Psi}{\psi} \newcommand{\PS}{\Psi} \newcommand{\s}{\sigma} \newcommand{{\phi}} \newcommand{\F}{{\Phi}}{{\phi}} \newcommand{\F}{{\Phi}} \newcommand{{\varphi}}{{\varphi}} \newcommand{{\upsilon}} \newcommand{\Y}{{\Upsilon}}{{\upsilon}} \newcommand{\Y}{{\Upsilon}} \newcommand{\zeta}{\zeta} \renewcommand{\AA}{{\mathcal A}} \newcommand{{\mathcal B}}{{\mathcal B}} \newcommand{{\mathcal C}}{{\mathcal C}} \newcommand{{\mathcal D}}{{\mathcal D}} \newcommand{\bigtriangledown}{\bigtriangledown} \newcommand{{\alpha^{'}}}{{\alpha^{'}}} \newcommand{\begin{eqnarray}}{\begin{eqnarray}} \newcommand{\end{eqnarray}}{\end{eqnarray}} \newcommand{\nonumber}{\nonumber} \newcommand{\,\raisebox{-1.0ex}{\,\raisebox{-1.0ex} {$\stackrel{\textstyle <}{\sim}$}\,} \newcommand{\,\raisebox{-1.0ex}{$\stackrel{\textstyle >}{\,\raisebox{-1.0ex}{$\stackrel{\textstyle >} {\sim}$}\,} \newcommand{Schwarzschild $\:$}{Schwarzschild $\:$} \newcommand{\varphi}{\varphi} \newcommand{\Delta}{\Delta} \newcommand{\tilde{x}}{\tilde{x}} \newcommand{\hspace{-.65cm}}{\hspace{-.65cm}} \newcommand{\journal}[4]{{\rm #1~}{#2}\,(19#3)\,#4} \newcommand{\journal {Ann. Inst. Henri Poincar\'e}}{\journal {Ann. Inst. Henri Poincar\'e}} \newcommand{\journal {Helv. Phys. Acta}}{\journal {Helv. Phys. Acta}} \newcommand{\journal {Sov. J. Part. Nucl.}}{\journal {Sov. J. Part. Nucl.}} \newcommand{\journal {Int. J. Mod. Phys.}}{\journal {Int. J. Mod. Phys.}} \newcommand{\journal {Physica (Utrecht)}}{\journal {Physica (Utrecht)}} \newcommand{\journal {Phys. Rev.}}{\journal {Phys. Rev.}} \newcommand{\journal {JETP Lett.}}{\journal {JETP Lett.}} \newcommand{\journal {Phys. Rev. Lett.}}{\journal {Phys. Rev. Lett.}} \newcommand{\journal {J. Math. Phys.}}{\journal {J. Math. Phys.}} \newcommand{\journal {Rev. Mod. Phys.}}{\journal {Rev. Mod. Phys.}} \newcommand{\journal {J. Diff. Geom.}}{\journal {J. Diff. Geom.}} \newcommand{\journal {Comm. Math. Phys.}}{\journal {Comm. Math. Phys.}} \newcommand{\journal {Class. Quantum Grav.}}{\journal {Class. Quantum Grav.}} \newcommand{\journal {Z. Phys.}}{\journal {Z. Phys.}} \newcommand{\journal {Nucl. Phys.}}{\journal {Nucl. Phys.}} \newcommand{\journal {Phys. Lett.}}{\journal {Phys. Lett.}} \newcommand{\journal {Mod. Phys. Lett.}}{\journal {Mod. Phys. Lett.}} \newcommand{\journal {Phys. Reports}}{\journal {Phys. Reports}} \newcommand{\journal {Progr. Theor. Phys.}}{\journal {Progr. Theor. Phys.}} \newcommand{\journal {Nuovo Cim.}}{\journal {Nuovo Cim.}} \newcommand{\journal {Acta Phys. Pol.}}{\journal {Acta Phys. Pol.}} \newcommand{\journal {Astrophys. Jour.}}{\journal {Astrophys. Jour.}} \newcommand{\journal {Astrophys. Jour. Lett.}}{\journal {Astrophys. Jour. Lett.}} \newcommand{\journal {Ann. Phys. }}{\journal {Ann. Phys. }} \newcommand{{\em Phys. Rev. D}}{{\em Phys. Rev. D}} \newcommand{\journal {Gen. Rel. Grav.}}{\journal {Gen. Rel. Grav.}} \newcommand{g_{YM}}{g_{YM}} \newcommand{g_{eff}}{g_{eff}} \begin{document} \newcommand{\preprint}[1]{\begin{table}[t] \begin{flushright} \begin{large}{#1}\end{large} \end{flushright} \end{table}} \preprint{hep-th/9904035} \begin{center} \Large{\bf A Comment on the Entropy of Strongly\\ Coupled ${\cal N}=4$} \vspace{5mm} \normalsize{N. Itzhaki} \vspace{5mm} { Department of Physics\\ University of California, Santa Barbara, CA 93106}\\ {\it <EMAIL>} \end{center} \begin{abstract} We propose a field theory argument, which rests on the non-renormalization of the two point function of the energy-momentum tensor, why the ratio between the entropies of strongly coupled and weakly coupled ${\cal N}=4$ is of order one. \end{abstract} \baselineskip 18pt The Maldacena conjecture \cite{mal} and the entropy of near-extremal D3-branes \cite{gkp} imply that the ratio between the entropies, at fixed temperature, of strongly coupled and weakly coupled ${\cal N}=4$ is $3/4$. In ${\cal N}=4$, unlike 2D CFT, the entropy is not protected thus it is not surprising that the ratio is not $1$. It is surprising, however, that the ratio is not a function of the 't Hooft coupling, $\lambda= g_{YM}^2 N$, which vanishes when $\lambda \rightarrow \infty$. The reason is the following {\em perturbative} argument.\footnote{Though this argument is widely known we did not find it in the literature. A closely related discussion can be found in \cite{hms,bjsv}.} At finite temperature, $T$, the expectation value of the fields is $\langle \phi^2 \rangle =T^2.$ As a result the potential term in SYM, which has the form $ V \sim g_{YM}^2 [\phi_i, \phi_j ]^2,$ induces a mass, $m^2 \sim \lambda T^2$, for a generic field. At small 't Hooft coupling the induced masses are much smaller then the temperature so to a good approximation the contribution to the entropy is of $N^2$ massless fields with a small correction which reduces the entropy.\footnote{For a rigorous discussion on the weakly coupled region see \cite{ft}.} At large 't Hooft coupling the induced masses are much larger then the temperature. Therefore, the contribution to the entropy from a generic field (not in the Cartan subalgebra of $SU(N)$) is suppressed at the strongly coupled region. Since the argument above rests on perturbation theory it cannot be trusted all the way to the strongly coupled region and hence, strictly speaking, there is no contradiction with the Maldacena conjecture. Still, it is fair to say that it is somewhat disturbing that the only field theory argument available (as far as we know) leads to that conclusion. Especially, when a similar argument for SYM in $1+1$ dimensions \cite{hms,bjsv} leads to results that fit so nicely into the Maldacena conjecture for D1-branes \cite{juan,pp,aki}. The purpose of this short note is to put forward a {\em field theory} argument, which does not rest on the AdS/CFT correspondence, that implies that the entropy at large coupling is of the order of the entropy at weak coupling. The argument rests on the ${\cal N}=4$ non-renormalization theorem for the energy-momentum tensor two point function and therefore it cannot be generalised to two dimensional SYM which is a non-conformal theory and hence the R-symmetry cannot protect the two-point function. We study SYM in a box whose volume is $L_z A$ with $A=L_x L_y$ and we take the limit $L_x, L_y \gg L_z$. Consider the transformation $x_3 \rightarrow x_3(1-\varepsilon)$ with $\varepsilon \ll 1$. The variation of the action under this transformation is \begin{equation} \d S=\varepsilon\int d^3 x \int_0^{L_z} dx_3 T_{33}. \end{equation} Therefore, the variation of the expectation value of $T_{00}$ is \begin{eqnarray} \d \langle T_{00}(0)\rangle =\int {\cal D} \phi ( e^{-(S+\d S)} -e^{-S}) T_{00}(0)=\varepsilon\int d^3x \int_{0}^{L_z}dx_3 \langle T_{00}(0) T_{33}(x)\rangle, \end{eqnarray} where ${\cal D} \phi$ represents integration with respect to all fields. To calculate the integral we need to know the energy-momentum tensor two point function. On $R^4$ non-renormalization theorem protects the energy-momentum tensor two point function. Thus on $R^4$ we can use the free SYM result \begin{equation}\label{11} \langle T(0)T(x) \rangle =\frac{N^2}{x^8}, \end{equation} where we have suppressed numerical factors of order one and the Lorentz indices (for details see \cite{2p}). However, what we need is not the two-point function in $R^4$ but rather in $R^3 \times S^1$. In two dimensions the conformal transformation group contains the transformation from $R^2$ to $R \times S^1$. Therefore, the two-points function in $R \times S^1$ are determined by the two-points function in $R^2$ and the dimensions of the operators. This is an important ingredient in Cardy's proof that the asymptotic growth of the number of state of a 2D CFT depends only on the central charge and not on the details of the CFT \cite{car,car1}. In four dimensions, however, the conformal transformations do not contain the transformation from $R^4$ to $R^3 \times S^1$. Thus, we do not know the exact form of the energy-momentum two point function for strongly coupled SYM on $R^3 \times S^1$.\footnote{For weakly coupled theories one can find directly on $R^3\times S^1$ the mode expansion of the relevant fields. So there is no need to start with the two-points function on $R^4$.} What we do know is that at distances smaller then $L_z$ the boundary condition is irrelevant and so eq.(\ref{11}) is a good approximation at short distances. Therefore, for a given point on $S^1$ we can calculate the contribution to $\d \langle T_{00}(x_3)\rangle$ from the region $|x_3-x_3^{'}| < L_z$. The integral divergent at short distances. The regularized result is $ \d \langle T_{00}(0)\rangle \sim \frac{\varepsilon N^2}{ L_z^4}$. Integrating over the volume we find the variation of the ground state energy, which yields after integration with respect to $\d L_z=\varepsilon L_z$, the ground state energy \begin{equation}\label{12} E_0\sim \frac{N^2 A}{L_z^3}. \end{equation} It is important to emphasis that we have assumed in the calculation of $\d \langle T_{00}\rangle$ that the integration over the whole region does not contain cancellations between the region where eq.(\ref{11}) is a good approximation and the region where it is not. Such cancellations can, in principle, reduce the ground state energy in a significant way to yield $E_0$ which is suppressed at large coupling. Therefore, our argument is not a proof but rather a strong indication that the entropies ratio is of order one. In other words, we estimate the Casimir energy, which is a boundary condition effect, using an approximation which is not sensitive to the details of the boundary condition but only to the distance between the boundaries. Eq.(\ref{12}) implies that the partition function at low temperature (compared to $L_z$) is, \begin{equation} Z\sim \exp \left( \frac{N^2 A \beta}{L_z^3}\right) . \end{equation} Now we can use the standard argument of switching the roles of $\beta $ and $L_z$ \footnote{See \cite{hm} for a related discussion in the context of the AdS/CFT correspondence.} to end up with the partition function of strongly coupled SYM at high temperature (compared to the size of the box) \begin{equation} Z\sim \exp (N^2 V T^3), \end{equation} which agrees, up to a numerical factor in the exponent, with the partition function of weakly coupled SYM. \vspace{.5cm} \hspace{-.65cm} {\bf Acknowledgements} I would like to thank Aki Hashimoto for helpful discussions. Work supported in part by the NSF grant PHY97-22022.
\section{Introduction} \label{sec:intro} Error-correction is required whenever information has to be reliably transmitted through a noisy environment. The theoretical grounds for classical error-correcting codes were first presented in 1948 by Shannon \cite{shannon}. He showed that it is possible to transmit information trough a noisy channel with a vanishing error probability by encoding up to a given critical rate $R_c$ equivalent to the {\it channel capacity}. However, Shannon's arguments were non-constructive and devising such codes turned out to be a major practical problem in the area of information transmission. In 1989 Sourlas \cite{sourlas89,sourlas94} proposed that, due to the equivalence between addition over the field $\{0,1\}$ and multiplication over $\{{\pm 1}\}$, many error-correcting codes can be mapped onto many-body spin-glasses with appropriately defined couplings. This observation opened the possibility of applying techniques from statistical physics to study coding systems, in particular, these ideas were applied to the study of parity check codes. These linear block codes can be represented by matrices of $N$ columns and $M$ rows that transform $N$-bit messages to $M$ ($>N$) parity checks. Each row represents bits involved in a particular check and each column represents checks involving the particular bit. The number of bits used in each check and the number of checks per bit depends on the code construction. We concentrate on the case where exactly $C$ checks are performed for each bit and exactly $K$ bits compose each check. The {\it code rate} $R$ is defined as the information conveyed per channel use $R=H_2(f_s)N/M=H_2(f_s)K/C$, where $H_2(f_s)= -(1-f_s)\;\mbox {log}_2 (1-f_s)\;- \;f_s\;\mbox {log}_2 (f_s) $ is the binary entropy of the message with bias $f_s$. \begin{figure} \hspace*{.4cm} \epsfxsize=150mm \epsfbox{figure1.ps} \vspace{0.5cm} \caption{The encoding, message corruption in the noisy channel and decoding can be represented as a Markovian process. The aim is to obtain a good estimative $\mbox{\boldmath $\widehat {\xi}$}$ for the original message $\xi$.} \label{encode} \end{figure} In the mapping proposed by Sourlas a message is represented by a binary vector $\mbox{\boldmath $\xi$} \in\{\pm 1\}^N$ encoded to a higher dimensional vector $\mbox{\boldmath $J^0$}\in\{\pm 1\}^M$ defined as $J^{0}_{\langle i_{1}, i_{2} \ldots i_{K}\rangle} = \xi_{i_{1}} \xi_{i_{2}} \ldots \xi_{i_{K}}$, where $M$ sets of $K$ indices are randomly chosen. A corrupted version $\mbox{\boldmath $J$}$ of the encoded message $\mbox{\boldmath $J^0$}$ has to be decoded for retrieving the original message. The decoding process can be viewed as a statistical Bayesian process \cite{iba98} (see Fig.\ref{encode}). Decoding focuses on producing an estimate $\widehat{\mbox{\boldmath $\xi$}}$ to the original message that minimizes a given expected loss $\langle\langle {\cal L}(\xi,\widehat{\xi})\rangle_{p(J\mid\xi)}\rangle_ {p(\xi)}$ averaged over the indicated probability distributions. The definition of the loss depends on the particular task; the simple Hamming distance ${\cal L}(\xi,\widehat{\xi})=\sum_j \xi_j \widehat{\xi}_j$ can be used for decoding binary messages. An optimal estimator for this particular loss function is $\widehat{\xi}_j=\mbox{sign}\langle S_j \rangle_{p(S\mid J)}$ \cite{iba98}, where $\mbox{\boldmath $S$}$ is a $N$ dimensional binary vector representing outcomes of the decoding process. Using Bayes' theorem, the posterior probability can be written as $\mbox{ln }p(\mbox{\boldmath $S$}\mid\mbox{\boldmath $J$} )=\mbox{ln }p(\mbox{\boldmath $J$}\mid \mbox{\boldmath $S$}) + \mbox{ln }p(\mbox{\boldmath $S$}) + \mbox{const}$. Sourlas has shown \cite{sourlas94} that for parity check codes this posterior can be written as a many-body Hamiltonian: \begin{eqnarray} \label{eq:Hamiltonian} \mbox{ln }p(\mbox{\boldmath $S$}\mid\mbox{\boldmath $J$} )&=&-\beta\; {\cal H}(\mbox{\boldmath $S$})\nonumber\\ &=&\beta \sum_{\mu} {\cal A}_{\mu} \ J_{\mu} \ \prod_{i\in\mu} S_{i} + \beta{\cal H}_{\mbox{\scriptsize prior}} (\mbox{\boldmath $S$}), \end{eqnarray} where $\mu=\left\langle i_{1},\ldots i_{K} \right\rangle$ is a set of indices and ${\cal A}$ is a tensor with the properties ${\cal A}_\mu\in\{0,1\}$ and $\sum_{\mu\setminus i}{\cal A}_\mu=C$ $\forall i$, which determines the $M$ components of the codeword $\mbox{\boldmath $J$}^{0}$. The second term ${\cal H}_{\mbox{\scriptsize prior}} (\mbox{\boldmath $S$})$ stands for the prior knowledge on the actual messages; it can be chosen as ${\cal H}_{\mbox{\scriptsize prior}}(\mbox{\boldmath $S$})=F\sum_{j=1}^{N} S_j$ to represent the expected bias in the message bits. For the simple case of a memoryless binary symmetric channel (BSC), $\mbox{\boldmath $J$}$ is a corrupted version of the transmitted message $\mbox{\boldmath $J$}^{0}$ where each bit is independently flipped with probability $p$ during transmission. The hyper-parameter $\beta$, that reaches an optimal value at Nishimori's temperature \cite{iba98,rujan93,nishi}, is related to the channel corruption rate. The decoding procedure translates to finding the thermodynamical spin averages for the system defined by the Hamiltonian (\ref{eq:Hamiltonian}) at a certain temperature (Nishimori's temperature for optimal decoding); as the original message is binary, the retrieved message bits are given by the signs of the corresponding averages. In the statistical physics framework the performance of the error-correcting process can be measured by the overlap between actual message and estimate for a given scenario characterized by a code rate, corruption process and information content of the message. To asses the typical properties we average this overlap over all possible codes $\cal A$ and noise realizations (possible corrupted vectors $\mbox{\boldmath $J$}$) given the message $\mbox{\boldmath $\xi$}$ and then over all possible messages: \begin{equation} \label{eq:mag} m=\frac {1}{N}\left \langle \sum_{i=1}^N {\xi}_i \;\left \langle \mbox{sign}\langle S_i \rangle \right\rangle_{{\cal A},J|\xi}\right \rangle_{\xi} \end{equation} Here $\mbox{sign}\langle S_i \rangle$ is the sign of the spins thermal average corresponding to the Bayesian optimal decoding. The average error per bit is then given by $p_e = (1-m)/2 $. Although this performance measure is not the usual physical magnetization (it can be better described as a measure of misalignment of the decoded message), for brevity, we will refer to it as {\it magnetization}. From the statistical physics point of view, the number of checks per bit is analogous to the spin system connectivity and the number of bits in each check is analogous to the number of spins per interaction. Sourlas' code has been studied in the case of extensive connectivity , where the number of bonds $C \!\sim\!$ \scriptsize $\left( \begin{array}{c} N-1 \\ K-1 \end{array} \right)$ \normalsize scales with the system size. In this case it can be mapped onto known problems in statistical physics such as the SK \cite{SK} ($K\!\!=\!\!2$) and Random Energy (REM) \cite{Derrida_REM} ($K \!\!\rightarrow\!\! \infty$) models. It has been shown that the REM saturates Shannon's bound \cite{sourlas89}. However, it has a rather limited practical relevance as the choice of extensive connectivity corresponds to a vanishingly small code rate. Here we present an analysis of Sourlas' code for the case of finite connectivity where the code rate is non-vanishing, detailing and extending our previous brief reports \cite{ks98a,ks98b}. We show that Shannon's bound can also be attained at finite code rates. We study the decoding dynamics and discuss the connections between statistical physics and belief propagation methods. This paper is organized as follows: in Section II we introduce a naive mean-field model that contains all the necessary ingredients to understand the system qualitatively. Section III describes the statistical physics treatment of Sourlas' code showing that Shannon's bound can be attained for finite code rates if $K\rightarrow\infty$. The finite $K$ case and the Gaussian noise are also discussed in Section III. The decoding dynamics is analyzed in Section IV. Concluding remarks are given in Section V. Appendices with detailed calculations are also provided. \section{Naive Mean Field Theory} \label{sec:naive} \subsection {Equilibrium} \begin{figure} \hspace*{.4cm} \epsfxsize=120mm \epsfbox{figure2.eps} \caption{Code performance measured by the magnetization $m$ as a function of the noise level $p$ as given by the naive mean-field theory at code rate $R=1/2$ and $K=2,3,4$ respectively from the bottom. The long-dashed line indicates PARA-FERRO coexistence. Insets: Maximum initial deviation $\lambda$ for convergence at a noise level $p=0.1$. Top inset: $K=3$ and increasing $C$. Bottom inset: Code rate $R=1/2$ and increasing $K$.} \label{naive} \end{figure} To gain some insight into the code behavior one can start by considering that the original message is $\xi_j=1$ for all $j$ (so $m=1$ will correspond to perfect decoding) and use Weiss' mean-field theory as a first (naive) approximation. The idea is to consider an effective field given by (for unbiased messages with $F=0$): \begin {equation} \label{eq:naiveeff} h^{\mbox{\scriptsize eff}}_j =\sum_{\{\mu:j\in \mu\}}J_\mu \prod_{i\in\mu\setminus j} S_i \end{equation} acting in every site. The first strong approximation here consists in disregarding the reaction fields that describe the influence of site $j$ back over the system. The local magnetization can then be calculated: \begin {equation} \label{eq:naivemag} m_j=\left\langle\mbox{tanh}\left(\beta h^{\mbox{\scriptsize eff}}_j \right)\right\rangle_{J,S}\simeq \mbox{tanh}\beta \left\langle h^{\mbox{\scriptsize eff}}_j \right\rangle_{J,S}, \end{equation} where we introduced a further approximation taking averages inside the function that can be seen as a high temperature approximation. Disregarding correlations among spins and computing the proper averages one can write: \begin {equation} \label{eq:naivemagav} m=\mbox{tanh} \left( \beta \; C (1-2p)\; m^{K-1}\right), \end{equation} where $p$ is the noise level in the channel. An alternative way to derive the above equation is by considering the free-energy: \begin {equation} \label{eq:naivefree} f(m)=-(1-2p)\frac{C}{K}m^{K}-\frac{s(m)}{\beta}. \end{equation} The entropic term $s(m)$ is: \begin {equation} s(m)=-\frac{1+m}{2}\mbox{ ln} \left(\frac{1+m}{2}\right)-\frac{1-m}{2}\mbox{ ln} \left(\frac{1-m}{2}\right). \end{equation} Minimizing this free-energy one can obtain Eq.(\ref{eq:naivemagav}) whose solutions give the possible phases after the decoding process. In Fig. \ref{naive} we show the maximum magnetization solutions $m$ for Eq.(\ref{eq:naivemagav}) as a function of the flip rate $p$ at code rate $R=1/2$ and $K=2,3,4$. For $K=2$ the performance degrades faster with the noise level than in the $K>2$ case. The dashed line indicates coexistence between paramagnetic (PARA) $m=0$ and ferromagnetic (FERRO) $m>0$ phases. \subsection{Decoding Dynamics} \begin{figure} \hspace*{3cm} \epsfxsize=50mm \epsfbox{figure3.ps} \vspace{0.5cm} \caption{Graph representing a code.} \label{node} \end{figure} In a naive mean-field framework the decoding process can be seen as an iterative solution for (\ref{eq:naivemagav}) starting from a magnetization value that depends on the prior knowledge about the original message. The fixed points of this dynamics correspond to the minima of the free-energy; a specific minimum is reached depending on the initial condition. In the insets of Fig.\ref{naive} we show, as a measure for the basin of attraction, the maximal deviation between the initial condition and the original message $\lambda=1-m_0$ that allows convergence to a FERRO solution. At the bottom inset we show the deviation $\lambda$ at code rate $R=1/2$, increasing values of $K$ and noise level $p=0.1$ . An increasing initial magnetization is needed when $K$ increases, decoding without prior knowledge is only possible for $K=2$. The top inset shows $\lambda$ for $K=3$, $p=0.1$; as $C$ increases (code rate decreases), the basin of attraction increases. One can understand intuitively how the basin of attraction depends on the connectivities by representing the code in a graph with bit and check nodes and looking at the mean-field behavior of a single bit node (see Fig.\ref{node}). The corrupted checks contribute wrong ($-1$ for the ``all ones'' message case) values to the bit nodes ($m<1$ in the mean field). Since check node values correspond to a product of $K-1$ bit values, the probability of updating these nodes to the wrong values increases with $K$, degrading the overall performance. On the other hand, if $C$ increases for a fixed $K$ the bit nodes gather more information and are less sensitive to the presence of (a limited amount of ) wrong bits . Although this naive picture indicates some of the qualitative features of real codes, one certainly cannot rely in its numerical predictions. In the following sections we will study Sourlas' codes using more sophisticated techniques that will substantially refine the analysis. \section{Equilibrium} \subsection{Replica Theory} \label{sec:replica} In the following subsections we will develop the replica symmetric theory for Sourlas' codes and show that, in addition to providing a good description of the equilibrium, it describes the typical decoding dynamics using belief propagation methods. The previous naive ``all ones'' messages assumption can be formally translated to the gauge transformation \cite{frad} $S_{i} \!\!\mapsto\!\! S_{i} \xi_{i}$ and $J_{\mu}\!\!\mapsto\!\! J_{\mu}\prod_{i\in \mu} \xi_{i} $ that maps any general message to the FERRO configuration defined as $\xi_i^{*}=1$ $\forall i$. One can then rewrite the Hamiltonian in the form: \begin{equation} \label{eq:Hamiltonian_gauge} {\cal H}(\mbox{\boldmath $S$})=- \sum_{\mu} {\cal A}_{\mu} \ J_{\mu} \ \prod_{i\in\mu} S_{i} - F \sum_{k}\xi_k S_{k} \ , \end{equation} With this transformation, the bits of the uncorrupted encoded message are $J^0_i=1$ $\forall i$ and, for a BSC, the corrupted bits are random variables with probability: \begin{equation} \label{eq:xi_J_prob_dist} {\cal P}\left(J_{\mu}\right) = (1\!-\!p)\ \delta \left(J_{\mu} \!-\! 1 \right) + p \ \delta\left(J_{\mu} \!+\! 1 \right), \end{equation} where $p$ is the channel flip rate. For deriving typical properties of these codes one has obtain an expression for the free-energy by invoking the replica approach where the free-energy is defined as: \begin{equation} \label{eq:freenergy} f= -\frac{1}{\beta}\lim_{N\rightarrow\infty} \frac{1}{N} \left.\frac{\partial} {\partial {\mathit n}}\right |_{{\mathit n}=0} \langle {\cal Z}^{ \mathit n}\rangle_{{\cal A},\xi,J}, \end{equation} where $\langle {\cal Z}^{ \mathit n}\rangle_{{\cal A},\xi,J}$ represents an analytical continuation in the interval $n\in[0,1]$ of the replicated partition function defined as: \begin{equation} \label{eq:partit} \langle {\cal Z}^{n}\rangle_{{\cal A},\xi,J} = \mbox{Tr}_{\{S_j^\alpha\}} \left[\left \langle e^{ \beta F \sum_{\alpha,k}\xi_k S^\alpha_{k}}\right \rangle_{\xi}\left\langle \exp\left(\beta\sum_{\alpha,\mu} {\cal A}_{\mu} \ J_{\mu} \ \prod_{i\in\mu} S^\alpha_{i} \right) \right\rangle_{{\cal A},J} \right]. \end{equation} The magnetization can be rewritten in the gauged variables as : \begin{equation} \label{eq:mag_gauged} m= \left \langle\left \langle \mbox{sign}\langle S_i \rangle \right\rangle_{{\cal A},J|\xi^*}\right \rangle_{\xi}, \end{equation} where $\xi^*$ denotes the transformation of a message $\xi$ into the FERRO configuration. The usual magnetization per site can be easily obtained by calculating \begin{equation} \label{eq:fundamental} \left \langle\left \langle S_i \right \rangle\right\rangle_{{\cal A},J,\xi}=- \left( \frac {\partial f} { \partial(\xi F)} \right). \end{equation} From this derivative one can find the distribution of the effective local fields $h_j$ that can be used to asses the magnetization $m$, since $\mbox{sign}\left(\langle S_j\rangle\right)=\mbox{sign}(h_j)$ . To compute the replicated partition function we closely follow Ref. \cite{wong_a}. We average uniformly over all codes ${\cal A}$ such that $\sum_{\mu\setminus i}{\cal A}_{\mu}= C$ $\forall i$ to find: \begin{eqnarray} \label{eq:partit_2} \langle {\cal Z}^{n}\rangle_{{\cal A},\xi,J}& =&\exp \left\{ N \;Extr_ {q,\widehat{q}}\left[C-\frac{C}{K}+\frac{C}{K}\left(\sum_{l=0}^{n}{\cal T}_l\sum_{\langle \alpha_1 \ldots \alpha_l\rangle} q_{\alpha_1 \ldots \alpha_l}^{K} \right)\right.\right.\nonumber\\ & -&\left.\left. C \left(\sum_{l=0}^{n}\sum_{\langle \alpha_1 \ldots \alpha_l\rangle}q_{\alpha_1 \ldots \alpha_l}\widehat{q} _{\alpha_1 \ldots \alpha_l}\right) \nonumber \right. \right.\\ &+& \left.\left.\ln \mbox{Tr}_{\{S^{\alpha}\}}\left \langle e^{\beta F\xi\sum _{\alpha}S^\alpha}\right\rangle_{\xi}\left(\sum_{l=0}^{n} \sum_{\langle \alpha_1 \ldots \alpha_l\rangle}\widehat{q} _{\alpha_1 \ldots \alpha_l}S^{\alpha_1}\ldots S^{\alpha_l} \right)^C \right]\right\}, \end{eqnarray} where ${\cal T}_l=\langle \tanh^l(\beta J) \rangle_J$, as in \cite{viana}, and $q_0=1$. We give details of this calculation in the Appendix A. At the extremum the order parameters acquire expressions similar to those of Ref. \cite{wong_a}: \begin{eqnarray} \label{order-param} \widehat{q}_{\alpha_1,...,\alpha_l}&=& {\cal T}_l\; q^{K-1}_ {\alpha_1,...,\alpha_l}\nonumber \\ q_{\alpha_1,...,\alpha_l}&=&\left \langle \left (\prod_{i=1}^l S^{\alpha_i} \right) \left(\sum_{l=0}^{n}\sum_{\langle \alpha_1 \ldots \alpha_l\rangle}\widehat{q}_{\alpha_1 \ldots \alpha_l}S^ {\alpha_1}\ldots S^{\alpha_l}\right)^{-1}\right\rangle_{\cal X}. \end{eqnarray} where \begin{equation} {\cal X}=\left \langle e^{\beta F\xi\sum_{\alpha}S^\alpha}\right\rangle_ {\xi}\left(\sum_{l=0}^{n}\sum_{\langle \alpha_1 \ldots\alpha_l\rangle}\widehat{q}_{\alpha_1 \ldots \alpha_l} S^{\alpha_1}\ldots S^{\alpha_l} \right)^{C}, \end{equation} and $\langle...\rangle_{\cal X}=\mbox{Tr}_{\{S^{\alpha}\}}\left[(...) {\cal X}\right]/\mbox{Tr}_{\{S^{\alpha}\}}\left[(...)\right]$. The term $\widehat{p}(\underline {S})=\sum_{l=0}^{n}\sum_{\langle \alpha_1 \ldots\alpha_l\rangle} \widehat{q}_{\alpha_1 \ldots \alpha_l} S^{\alpha_1}\ldots S^{\alpha_l}$ represents a probability distribution over the space of replicas and $p_0(\underline{S})=\left \langle e^{\beta F\xi\sum_{\alpha}S^\alpha}\right\rangle_{\xi}$ is a prior distribution over the same space. For reasons that will become clear in Section \ref{sec:decoding}, $q_{\alpha_1,...,\alpha_l}$ represents one $l$-th momentum of the equilibrium distribution of a bit-check edge in a belief network during the decoding process and $\widehat{q}_{\alpha_1 \ldots \alpha_l}$ represents $l$-th moments of a check-bit edge equilibrium distribution . The distribution ${\cal X}$ represents the probability of a certain site (bit node) configuration subjected to exactly $C$ interactions and with prior probability given by $p_0$. \subsection{Replica Symmetric Solution} \label{sec:symmetric} The replica symmetric (RS) ansatz can be introduced via the auxiliary fields $\pi(x)$ and $\widehat{\pi}(y)$ in the following way (see also \cite{wong_a}): \begin{eqnarray} \label{eq:auxfields} \widehat{q}_{\alpha_1 ... \alpha_l}&=&\int \: dy \; \widehat{\pi}(y) \tanh^l(\beta y) ,\nonumber\\ q_{\alpha_1 ... \alpha_l}&=&\int \: dx \; \pi(x) \tanh^l(\beta x) \end{eqnarray} for $l=1,2,\ldots$. Plugging it into the replicated partition function (\ref{eq:partit_2}), performing the limit $n\rightarrow 0$ and using Eq.(\ref{eq:freenergy}) (see Appendix \ref{app:B} for details) one obtains: \begin{eqnarray} \label{eq:freesym} f&=&-\frac{1}{\beta}\: Extr_{\pi,\widehat{\pi}}\left \{\alpha \ln \cosh \beta \right. \\ &+& \alpha \int \left[ \prod_{l=1}^{K} dx_{l} \ \pi(x_{l}) \right] \left\langle \ln \left[ 1 + \tanh \beta J \ \prod_{j=1}^{K} \tanh \beta x_{j} \right] \right\rangle_{J} \nonumber \\ &-& C \int dx \ dy \ \pi(x) \ \widehat{\pi}(y) \ \ln \left[ 1 + \tanh \beta x \ \tanh \beta y \right] \nonumber\\ &-& C \int dy \ \widehat{\pi}(y) \ \ln \cosh \beta y \nonumber\\ &+&\left. \int \left[ \prod_{l=1}^{C} dy_{l} \ \widehat{\pi}(y_{l}) \right] \left\langle \ln \left[ 2 \cosh \beta \left(\sum_{j=1}^{C} y_{j} + F \xi \right) \right]\right\rangle_{\xi} \right \}\nonumber, \end{eqnarray} where $\alpha=C/K$. The saddle-point equations, obtained by varying Eq.(\ref{eq:freesym}) with respect to the probability distributions, provide a set of relations between $\pi(x)$ and $\widehat{\pi}(y)$ \begin{eqnarray} \label{eq:saddle_point} \fl\pi(x) &=& \int \left[ \prod_{l=1}^{C-1} dy_{l} \ \widehat{\pi}(y_{l}) \right] \ \left\langle \delta \left[ x - \sum_{j=1}^{C-1} y_{j} - F \xi \right]\right\rangle_{\xi} \\ \fl \widehat{\pi}(y) &=& \int \left[ \prod_{l=1}^{K-1} dx_{l} \ \pi(x_{l}) \right] \ \left\langle \delta \left[ y - \frac{1}{\beta} \tanh^{-1} \left( \tanh\beta J \ \prod_{j=1}^{K-1} \tanh \beta x_{j} \right) \right] \right\rangle_{J} \ . \nonumber \end{eqnarray} Later we will show that this self-consistent pair of equations can be seen as a mean-field version for the belief propagation decoding. Using Eq.(\ref{eq:fundamental}) one finds that the local field distribution is : \begin{equation} \label{eq:local_field} P(h)=\int \left[ \prod_{l=1}^{C} dy_{l} \ \widehat{\pi}(y_{l}) \right] \ \left\langle \delta \left[ h - \sum_{j=1}^{C} y_{j} - F \xi \right]\right\rangle_{\xi}, \end{equation} where $\widehat{\pi}(y)$ is given by the saddle point equations above. The magnetization (\ref{eq:mag}) can then be calculated using: \begin{equation} \label{eq:mag_sym} m = \int d h \ \mbox{sign} (h) \, P(h). \end{equation} The code performance can be assessed by assuming a particular prior distribution for the message bits, solving the saddle-point equations (\ref{eq:saddle_point}) numerically and then computing the magnetization. Instabilities in the solution within the space of symmetric replicas can be probed looking at second derivatives of the functional whose extremum defines the free-energy (\ref{eq:freesym}). The simplest necessary condition for stability is having non-negative second functional derivatives in relation to $\pi(x)$ (and $\widehat{\pi}(y)$) : \begin{equation} \label{eq:stability} \fl \frac{1}{\beta} \int \left[\prod_{l=1}^{K-2} dx_{l} \ \pi(x_{l}) \right] \left\langle \ln \left[ 1+ \tanh \beta J \ \tanh^2 \beta x\prod_{j=1}^{K-2} \tanh \beta x_{j} \right] \right\rangle_{J} \geq 0, \end{equation} for all $x$. The replica symmetric solution is expected to be unstable for sufficiently low temperatures (large $\beta$). For high temperatures we can expand the above expression around small $\beta$ to find the stability condition: \begin{equation} \label{eq:stabhigh} \langle J\rangle_{J} \langle x \rangle_{\pi}^{K-2}\geq 0 \; \end{equation} We expect the average $\langle x \rangle_{\pi}=\int dx\, \pi(x)\, x$ to be zero in PARA phase and positive in FERRO phase, satisfying the stability condition. This result is still generally inconclusive, but provides some evidence that can be examined numerically. In Section \ref{sec:bound} we will test the stability of our solutions using condition (\ref{eq:stability}). In the next sections we restrict our study to the unbiased case ($F=0$), which is of practical relevance, since it is always possible to compress a biased message to an unbiased one. \subsection {Case $K\rightarrow\infty$, $C=\alpha K$ } \label{sec:Kinfty} For this case one can obtain solutions to the saddle-point equations for arbitrary temperatures. In the first saddle-point equation (\ref{eq:saddle_point}) one can write: \begin{equation} \label{central_limit} x=\sum_{l=1}^{C-1} y_l \approx (C-1)\langle y \rangle_{\widehat{\pi}} = (C-1)\int dy \; y\; \widehat{\pi}(y). \end{equation} It means that if $\langle y \rangle_{\widehat{\pi}}=0$ (as it is the in PARA and spin glass (SG) phases) then $\pi(x)$ must be concentrated at $x=0$ implying that ${\pi}(x)=\delta(x)$ and $\widehat{\pi}(y)=\delta(y)$ are the only possible solutions. Moreover, Eq.(\ref{central_limit}) implies that in FERRO phase one can expect $x\approx{\cal O}(K)$. Using Eq.(\ref{central_limit}) and the second saddle-point equation (\ref{eq:saddle_point}) one can find a self-consistent equation for the mean-field $\langle y \rangle_{\widehat{\pi}}$: \begin{equation} \label{mean_field} \langle y \rangle_{\widehat{\pi}} = \left \langle\frac{1}{\beta}\, \mbox{tanh}^{-1} \left[\mbox{tanh}(\beta J)\left(\mbox{tanh}(\beta (C-1)\langle y \rangle_{\widehat{\pi}}) \right)^{K-1} \right]\right\rangle_J. \end{equation} For a BSC the above average is over distribution (\ref{eq:xi_J_prob_dist}). Computing the average, using $C=\alpha K$ and rescaling the temperature as $\beta = \tilde{\beta} (\mbox {ln}K)/K $, in the limit $K\rightarrow\infty$ one obtains: \begin{equation} \label{mean_field2} \langle y \rangle_{\widehat{\pi}} = (1-2p)\left[\mbox{tanh} (\tilde{\beta}\alpha\langle y\rangle_{\widehat{\pi}} \mbox{ ln}(K)) \right]^{K}, \end{equation} where $p$ is the channel flip probability. The mean-field $\langle y \rangle_{\widehat{\pi}} = 0 $ is always a solution to this equation (either PARA or SG); at $\beta_c =\mbox{ln}(K) /( 2\alpha K(1-2p))$ an extra non-trivial FERRO solution emerges with $\langle y \rangle_{\widehat{\pi}}=1-2p$. As the connection with the magnetization $m$ is given by Eq. (\ref{eq:local_field}) and Eq. (\ref{eq:mag_sym}); it is not difficult to see that it implies $m=1$ for FERRO solution. One remarkable point is that the temperature were the FERRO solution emerges is $\beta_c \sim {\cal O}(\mbox{ln}(K)/K)$; it means that in a simulated annealing process PARA-FERRO barriers emerge quite early for large $K$ values implying metastability and, consequently, a very slow convergence. It seems to advocate the use of small $K$ values in practical applications. This case is analyzed in Section \ref{sec:finite}. For $\beta>\beta_c$ both PARA and FERRO solutions exist. \begin{figure} \hspace*{2cm} \epsfxsize=120mm \epsfbox{figure4.eps} \caption{Phase diagram in the plane temperature $T$ versus noise level $p$ for $K\rightarrow\infty$ and $C=\alpha K$, with $\alpha=4$. The dotted line indicates Nishimori's temperature $T_N$ . Full lines represent coexistence. The critical noise level is $p_c$. The necessary condition for stability in the FERRO phase is satisfied above the dashed line.} \label{phase} \end{figure} The FERRO free-energy can be obtained from Eq.(\ref{eq:freesym}) using Eq.(\ref{central_limit}), being $f_{\mbox{\scriptsize FERRO}}=-\alpha (1-2p)$. The corresponding entropy is $s_{\mbox{\scriptsize FERRO}}=0$ indicating a single solution. The PARA free-energy is obtained by plugging $\pi(x)=\delta(x)$ and $\widehat{\pi}(y)=\delta(y)$ into Eq. (\ref{eq:freesym}): \begin{eqnarray} \label{eq:freepara} f_{\mbox{\scriptsize PARA}}&=& -\frac{1}{\beta}(\alpha \mbox{ ln}(\mbox{cosh }\beta) + \mbox{ln }2) ,\\ s_{\mbox{\scriptsize PARA}}&=&\alpha(\mbox{ln}(\mbox{cosh }\beta) -\beta\mbox{ tanh }\beta) + \mbox{ln }2. \end{eqnarray} PARA solutions are unphysical for $\alpha > (\mbox {ln } 2)/(\beta \mbox { tanh }\beta - \mbox{ln ch }\beta)$, since the corresponding entropy is negative. To complete the phase diagram picture we have to assess the spin-glass free-energy and entropy. We have seen in the beginning of this section that replica symmetric SG and PARA solutions consist of the same field distributions for $K\rightarrow\infty$, implying unphysical behavior. In order to produce a solution with non-negative entropy one has to break the replica symmetry. We use here a pragmatic way to build this solution, using the simplest one-step replica symmetry breaking known as {\it frozen spins}. It was observed in Ref. \cite{gross} that for the REM a one-step symmetry breaking scheme gives the exact solution. In this scheme the $n$ replicas' space is divided to groups of $m$ identical solutions. It was shown that an abrupt transition in the order parameter from a unique solution (Edwards-Anderson parameter $q=1$, SG phase) to a completely uncorrelated set of solutions ($q=0$, PARA phase) occurs. This transition takes place at a critical temperature $\beta_g$ that can be found by solving the appropriate saddle-point equations; this temperature is given by the root of the replica symmetric entropy ($s_{\scriptsize RS}=0$) meaning that the RS-RSB transition occurs at the same point as the PARA-SG in this model. The symmetry breaking parameter was found to be $m_g=\beta_g/\beta$, indicating that this kind of solution is physical only for $\beta>\beta_g$, since $m_g \leq 1$ \cite{parisi}, indicating a PARA-SG phase transition. The free-energy can be computed by plugging the order parameters in the effective Hamiltonian, obtained after averaging over the disorder and taking the proper limits. It shows no dependence on the temperature, since for $\beta>\beta_g$ the system is completely frozen in a single configuration. For the Sourlas' code, in the regime we are interested in, SG solutions to the saddle-point equations are given by $\pi(x)=\delta(x)$ and $\widehat{\pi}(y)=\delta(y)$. The RSB-SG free-energy that guaranties continuity in the SG-PARA transition is identical to $f_{\mbox{\scriptsize PARA}}$, since the SG and PARA solutions have exactly the same structure, to say: \begin{equation} \label{eq:free_rsb_sg} f_{\mbox{\scriptsize RSB-SG}}=-\frac{1}{\beta_g}\;(\alpha \;\mbox{ln}\,(\mbox{cosh }\,\beta_g) + \mbox{ln }\,2), \end{equation} where $\beta_g$ is a solution for $s_{\mbox{\scriptsize RS-SG}}= \alpha\;(\mbox{ln}\,(\mbox{cosh }\,\beta) -\beta\;\mbox{tanh }\,\beta) +\mbox{ln }\,2=0$. In Fig.\ref{phase} we show the phase diagram for a given code rate $R$ in the temperature $T$ versus noise level $p$ plane. \subsection{Shannon's Limit} \label{sec:bound} Shannon's analysis shows that up to a critical code rate $R_c$, which equals the channel capacity, it is possible to recover information with arbitrarily small error probability for a given noise level. For the BSC : \begin{equation} \label{eq:shannon} R_c=\frac{1}{\alpha_c}=1+p\mbox{ log}_2 \;p + (1-p)\mbox{ log}_2\; (1-p). \end{equation} Sourlas' code, in the case where $K\rightarrow\infty$ and $C \sim {\cal O}(N^K)$ can be mapped onto the REM and has been shown to be capable of saturating Shannon's bound in the limit $R\rightarrow 0$ \cite{sourlas89}. In this section we extend the analysis to show that Shannon's bound can be attained by Sourlas' code at zero temperature also for $K\rightarrow\infty$ limit but with connectivity $C=\alpha K$. In this limit the model is analogous to the diluted REM analyzed by Saakian in \cite {saakian}. The errorless phase is manifested in a FERRO phase with perfect alignment ($m=1$) (condition that is only possible for infinite $K$) up to a certain critical noise level; a further noise level increase produces frustration leading to a SG phase where the misalignment is maximal ($m=0$). The FERRO-SG transition is analogous to the transition from errorless decoding to decoding with errors described by Shannon. A PARA phase is also present when the transmitted information is insufficient to recover the original message ($R>1$). \begin{figure} \hspace*{2cm} \epsfxsize=120mm \epsfbox{figure5.eps} \caption{Histogram representing the mean-field distribution $\widehat{\pi}(y)$ obtained by Monte-carlo integration at low temperature ($\beta=10$, $K=3$,$C=6$ and $p=0.1$). Dotted lines represent solutions obtained by iterating self-consistent equations both with five peak and three peak ans\"atze. Inset: detailed view of the weak regular part arising in the Monte-carlo integration. } \label{fields} \end{figure} At zero temperature saddle-point equations (\ref{eq:saddle_point}) can be rewritten as: \begin{eqnarray} \label{eq:sp_infty} \fl \pi(x) &=& \int \left[ \prod_{l=1}^{C-1} dy_{l} \ \widehat{\pi}(y_{l})\right] \ \delta \left[ x - \sum_{j=1}^{C-1} y_{j} \right] \\ \fl \widehat{\pi}(y) &=& \int \left[ \prod_{l=1}^{K-1} dx_{l} \ \pi(x_{l}) \right] \ \left\langle \delta \left[ y - \mbox{sign}(J \prod_{l=1}^{K-1} x_{l}) \mbox{min}(\mid J\mid, ... , \mid x_{K-1} \mid)\right] \right\rangle_{J} \ , \nonumber \end{eqnarray} The solutions for these saddle-point equations may, in general, result in probability distributions with singular and regular parts. As a first approximation we choose the simplest self-consistent family of solutions which are, since $J=\pm 1$, given by: \begin{eqnarray} \widehat{\pi}(y)&=&p_+\delta(y-1)+p_0\delta(y)+p_-\delta(y+1)\\ \pi(x)&=&\sum_{l=1-C}^{C-1} T_{[p_{\pm},p_0;C-1]}(l) \,\delta(x-l), \end{eqnarray} with \begin{equation} T_{ \left[ p_{+}, p_{0}, p_{-}; C-1 \right]} (l) = \sum^{\prime}_{ \{k,h,m\}} \frac{(C-1)!}{k! \ h! \ m!} \ p_{+}^{k} \ p_{0}^{h} \ p_{-}^{m}, \end{equation} where the prime indicates that $k,h,m$ are such that $k-h=l; \ k+h+m=C-1$. Evidence for this simple ansatz comes from Monte-carlo integration of Equation (\ref{eq:saddle_point}) at very low temperatures, that shows solutions comprising three dominant peaks and a relatively weak regular part. Inside FERRO and PARA phases a more complex singular solution comprising five peaks $\widehat{\pi}(y)=p_{+2}\delta(y-1)+p_{+}\delta(y-0.5)+p_0\delta(y)+p_-\delta(y+0.5)+p_{-2}\delta(y+1)$ collapses back to the simpler three peak solution. In Fig.\ref{fields} we show a typical result of a Monte-carlo integration for the field $\widehat{\pi}(y)$. The two peak that emerge by using either the three peak ansatz or the five peak ansatz are shown as dotted lines. In the inset we show the weak regular part of the Monte-carlo solution. Plugging the above ansatz in the saddle-point equations one can write a closed set of equations in $p_{\pm}$ and $p_{0}$ that can be solved numerically (see appendix D for details). \begin{figure} \hspace*{2cm} \epsfxsize=120mm \epsfbox{figure6.eps} \caption{Phase diagram in the plane code rate $R$ versus noise level $p$ for $K\rightarrow\infty$ and $C=\alpha K$ at zero temperature. The FERRO-SG coexistence line corresponds to the Shannon's bound.} \label{bound} \end{figure} The three peak solution can be of three types: FERRO ($p_+>p_-$), PARA ($p_0=1$) or SG ($p_-=p_+$). Computing free-energies and entropies enables one to construct the phase diagram. At zero temperature the PARA free-energy is $f_{\mbox{\scriptsize PARA}}=-\alpha$ and the entropy is $s_{\mbox{\scriptsize PARA}}=(1-\alpha)\mbox{ ln }2$, this phase is physical only for $\alpha<1$, what is expected since it corresponds exactly to the regime where the transmitted information is not sufficient to recover the actual message ($R>1$). The FERRO free-energy does not depend on the temperature, having the form $f_{\mbox{\scriptsize FERRO}}=-\alpha(1-2p)$ with entropy $s_{\mbox {\scriptsize FERRO}}=0$. One can find the FERRO-SG coexistence line that corresponds to the maximum performance of a Sourlas' code by equating Eq.(\ref{eq:free_rsb_sg}) and $f_{\mbox{\scriptsize FERRO}}$. Observing that $\beta_g=\beta_N(p_c)$ (as seen in Fig.\ref{phase} ) we found that this transition coincides with Shannon's bound Eq.(\ref{eq:shannon}). It is interesting to note that in the large $K$ regime both RS-FERRO and RSB-SG free-energies (for $T<T_g$) do not depend on the temperature, it means that Shannon's bound is valid also for finite temperatures up to $T_g$. In Fig.\ref{bound} we give the complete zero temperature phase diagram. The stability of replica symmetric FERRO and PARA solutions used to obtain Shannon's bound can be checked using Eq.(\ref{eq:stability}) at zero temperature: \begin{equation} \label{eq:stability0temp} \fl \lim_{\beta \rightarrow \infty} \frac{1}{\beta} \int \left[\prod_{l=1}^{K-2} dx_{l} \ \pi(x_{l}) \right] \left\langle \ln \left[ 1+ \tanh \beta J \ \tanh^2 \beta x\prod_{j=1}^{K-2} \tanh \beta x_{j} \right] \right\rangle_{J} \ge 0, \end{equation} for all $x$. For PARA solutions the above integral vanishes, trivially satisfying the condition, while for FERRO solution in the $K$ large regime, $x_l\approx{\cal O}(K)$ and the integral becomes \begin{equation} -2p \left [ \left (1- \Theta \left( x+1 \right ) \right ) + |x| \left (\Theta \left( x+1 \right ) -\Theta \left( x- 1 \right ) \right ) + \Theta \left( x- 1 \right ) \right ], \end{equation} where $\Theta(x)=1$ for $x\geq 0$ and $0$ otherwise, indicating instability for $p>0$. For the noiseless case $p=0$ the stability condition is satisfied. The instability of FERRO phase opens the possibility that Sourlas' code does not saturate Shannon's bound, since a correction to the FERRO solution could change FERRO-SG transition line. However, it was shown in Section \ref{sec:symmetric} that this instability vanishes for large temperatures, what supports, to some extent, the FERRO-SG line obtained and the saturation of Shannon's bound in some region, as long as the temperature is lower than Nishimori's temperature. For finite temperatures the stability condition for FERRO solution can be rewritten as: \begin{equation} \left(1+\mbox{tanh}(\beta)\mbox{tanh}^2(\beta x)\right)^{(1-p)} \left(1-\mbox{tanh}(\beta)\mbox{tanh}^2(\beta x)\right)^p \ge 1 \; \forall x. \end{equation} For $p=0$ the condition is clearly satisfied. For finite $p$ a critical temperature above which the stability condition is fulfilled can be found numerically. In Fig.\ref{phase} we show this critical temperature in the phase diagram; one can see that there is a considerable region in which our result that Sourlas' code can saturate Shannon's bound is supported. Conclusive evidence to that will be given by simulations presented in Section \ref{sec:decoding}. \subsection{Finite K Case} \label{sec:finite} Although Shannon's bound only can be attained in the limit $K\rightarrow\infty$, it was shown in the Section \ref{sec:Kinfty} that there are some possible drawbacks, mainly in the decoding of messages encoded by large $K$ codes, due to large barriers which are expected to occur between PARA and FERRO states. In this section we consider the finite $K$ case, for which we can solve the RS saddle-point equations (\ref{eq:saddle_point}) for arbitrary temperatures using Monte-carlo integration. We can also obtain solutions for the zero temperature case using the simple iterative method described in Section \ref{sec:bound}. \begin{figure} \hspace*{2cm} \epsfxsize=120mm \epsfbox{figure7.eps} \caption{Top: zero temperature magnetization $m$ as a function of the noise level $p$ for various $K$ values at code rate $R=1/2$, as obtained by the iterative method . Notice that the RS theory predicts a transition of second order for $K=2$ and first order for $K>2$. Bottom: RS-FERRO free-energies (white circles for $K=2$ and from the left: $K\,=\,3,4,5$ and $6$) and RSB-SG free-energy (dotted line) as functions of the noise level $p$. The arrow indicates the region where the RSB-SG phase starts to dominate. Inset: a detailed view of the RS-RSB transition region.} \label{kfinite} \end{figure} At the top of Fig.\ref{kfinite} we show the zero temperature magnetization $m$ as a function of the noise level $p$ at code rate $R=1/2$. These curves were obtained by using the three peak ansatz of the Section \ref{sec:bound}. It can be seen that the transition is of second order for $K=2$ and first order for $K>3$ similarly to extensively connected models. The transition as described by the RS solution tends to $p=0.5$ as $K$ grows. Note that this does not correspond to perfect retrieval since the RSB spin glass phase dominates for $p>p_c$ (see bottom of Fig.\ref{kfinite}). In the bottom figure we plot RS free-energies and RSB frozen spins free-energy, from which we determine the critical probability $p_c$ where the transition occurs (pointed by an arrow). After the transition, free-energies for $K=3,4,5$ and $6$ acquire values that are lower than the SG free-energy; nevertheless, the entropy is negative and these free-energies are therefore unphysical. It is remarkable that this critical value does not change significantly for finite $K$ in comparison to infinite $K$. Observe that Shannon's bound cannot be attained for finite $K$, since $m=1$ exactly only if $K\rightarrow\infty$. The $K=2$ model with extensive connectivity (SK) is known to be somewhat special, a full Parisi solution is needed to recover the concavity of the free-energy and the Parisi order function has a continuous behavior \cite{mezard}. No stable solution is known for the intensively connected model (Viana-Bray model). In order to check the theoretical result obtained one relies on simulations of the decoding process at low temperatures. In Section VIII we show that the simulations are in good agreement with the theoretical results. \subsection {Gaussian Noise} \label{sec:gauss} Using the replica symmetric free-energy (\ref{eq:freesym}) and the frozen spins RSB free-energy (\ref{eq:free_rsb_sg}) one can easily extend the analysis to other noise types. The general PARA free-energy and entropy can be written: \begin{eqnarray} \label{eq:fparagen} f_{\mbox{\scriptsize PARA}}&=&-\frac{1}{\beta}\;\left(\alpha\;\langle \mbox{ln}\,(\mbox{ch }\beta J)\rangle_J +\mbox {ln }\,2\right) \nonumber \\ s_{\mbox{\scriptsize PARA}}&=&\alpha\;\left(\langle \mbox{ln}\,(\mbox{ch }\beta J)\rangle_J - \beta \langle J\;\mbox{tanh}\,(\beta J)\rangle_J\right)+\mbox {ln }2. \end{eqnarray} The SG-RSB free-energy is given by : \begin{equation} \label{eq:fsggen} f_{\mbox{\scriptsize SG-RSB}}=-\frac{1}{\beta_g}\;\left(\alpha\;\langle \mbox{ln }(\mbox{ch }\beta_g J)\rangle_J +\mbox {ln }2 \right), \end{equation} with $\beta_g$ defined as the solution of \begin{equation} \label{eq:entropy} \alpha\;\left(\langle \mbox{ln }(\mbox{ch }\beta_g J)\rangle_J - \beta_g \langle J\;\mbox{tanh }(\beta_g J)\rangle_J\right)+\mbox {ln }2 =0 . \end{equation} The FERRO free-energy is in general given by $f_{\mbox{\scriptsize FERRO}}=-\alpha\;\langle J\rangle_J=-\alpha\;\langle J \; \mbox{tanh }(\beta_N J)\rangle_J$ (see Appendix \ref{app:nishifree}). The maximum performance of the code is defined by the critical line : \begin{equation} \label{eq:line} \alpha\left(\langle \mbox{ln}(\mbox{ch }\beta_g J)\rangle_J - \beta_g \langle J\;\mbox{tanh}(\beta_N J)\rangle_J\right)+\mbox {ln }2 =0, \end{equation} obtained by equating free-energies in PARA and FERRO phases. Comparing this expression with entropy (\ref{eq:entropy}) it can be seen that $\beta_g=\beta_N$ at the critical line; the same behavior observed in the BSC case. From Eq.(\ref{eq:line}) one can write: \begin{equation} \label{eq:capacity} R_c=\beta_N^2 \frac{\partial}{\partial \beta} \left[\frac{1}{\beta}\langle \mbox{log}_2 \mbox{ cosh}(\beta J)\rangle_J\right]_{\beta=\beta_N}, \end{equation} that can be used to compute the performance of the code for arbitrary symmetric noise. \begin{figure} \hspace*{2cm} \epsfxsize=120mm \epsfbox{figure8.eps} \vspace{1cm} \caption{Critical code rate $R_c$ and channel capacity for a binary Gaussian channel as a function of the signal to noise rate $S/N$ (solid line). Sourlas' code saturates Shannon's bound. Channel capacity of the unconstrained Gaussian channel (dashed line).} \label{gaussian} \end{figure} Supposing that the encoded bits can acquire totally unconstrained values Shannon's bound for Gaussian noise is given by $R_c=\frac{1}{2}\mbox{ log}_2(1 +S/N)$, where $S/N$ is the signal to noise ratio, defined as the ratio of source energy per bit (squared amplitude) over the spectral density of the noise (variance). If one constrains the encoded bits to binary values $\{\pm 1\}$ the capacity of a Gaussian channel is: \begin{equation} \label{eq:capacity_gauss} R_c=\int dJ\;P(J\mid 1) \mbox{ log}_2 P(J\mid 1) - \int dJ \; P(J) \mbox{ log}_2 P(J), \end{equation} where $P(J\mid J^0)=\frac{1}{\sqrt{2\pi \sigma^2}}\mbox{ exp}(-\frac{(J-J^0)^2}{2\sigma^2})$. In Fig.\ref{gaussian} we show the performance of Sourlas' code in a Gaussian channel together with the capacities of the unconstrained and binary Gaussian channels. We show that $K\rightarrow\infty$, $C=\alpha K$ Sourlas' code saturates Shannon's bound for the binary Gaussian channel as well. The significantly lower performance in the unconstrained Gaussian channel can be trivially explained by the binary coding scheme while signal and noise are allowed to acquire real values. \section {Decoding Dynamics} \label{sec:decoding} \subsection{Belief Propagation} \label{sec:belief} The decoding process of an error-correcting code relies on computing averages over the marginal posterior probability $P(S_j\mid \mbox{\boldmath $J$})$ for each one of the $N$ message bits $S_j$ given the corrupted encoded bits $J_\mu$ (checks), where $\mu=\langle i_1 \ldots i_K \rangle$ is one of the $M$ sets chosen by the tensor ${\cal A}_\mu$. The probabilistic dependencies existing in the code can be represented as a bipartite graph known as a {\it belief network} where nodes in one layer correspond to the $M$ checks $J_\mu$ while nodes in the other to the $N$ bits $S_j$. Each check is connected to exactly $K$ bits and each bit is connected exactly to $C$ checks (see Fig.\ref{belief}a). Pearl \cite{pearl} proposed an iterative algorithm for computation of marginal probabilities in belief networks. These algorithms operate by updating beliefs (conditional probabilities) locally and propagating them. Generally the convergence of these iterations depends on the absence of loops in the graph. As can be seen in Fig.\ref{belief}a, networks that define error-correcting codes may include loops and convergence problems may occur. Recently it was shown that in some cases Pearl's algorithm works even in the presence of loops \cite{weiss}. \begin{figure} \hspace*{4cm} \epsfxsize=80mm \epsfbox{figure9.ps} \vspace{1cm} \caption{(a) Belief network representing an error-correcting code. Each bit $S_j$ (white circles) is linked to exactly $C$ checks and each check (black circles) $J_{\mu}$ is linked to exactly $K$ bits. (b) Graphical representation of the field $r_{\mu j}$. The grey box represents the mean field contribution $\prod_{l\in {\cal L}(\mu)\setminus j}q_{\mu l}$ of the other bits on the check $J_\mu$. (c) Representation of one of the fields $q_{\mu l}$. } \label{belief} \end{figure} The particular use of belief networks as decoding algorithms for error-correcting codes based on sparse matrices was discussed by MacKay in \cite{mackay95b}. In this work a loop-free approximation for the graph in Fig.\ref{belief}a was proposed (see \cite{pearl} for a general discussion on such approximations). In fact, it was shown in \cite{urbanke} that the probability of finite length loops in these graphs vanishes with the system size. In this framework the network is decomposed in a way to avoid loops and the conditional probabilities $q^{(S)}_{\mu j}$ and $r^{(S)}_{\mu j}$ are computed. The set of bits in a check $\mu$ is defined as ${\cal L}(\mu)$ and the set of checks over the bit $j$ as ${\cal M}(j)$. The probability that $S_j=S$ given information on all checks other than $\mu$ is denoted $q^{(S)}_{\mu j} =P(S_j=S\mid \{J_{\nu}:\nu \in {\cal M}(j)\setminus\mu\})$ and $r^{(S)}_{\mu j} = \mbox{Tr}_{\{S_l:l\in{\cal L}(\mu)\setminus j\}} P(J_{\mu}\mid S_j=S, \{S_l:l \in {\cal L}(\mu)\setminus j \}) \prod _{l\in {\cal L}(\mu)\setminus l} q^{(S_l)}_{\mu l}$ is the probability of the check $J_{\mu}$ if the bit $j$ is fixed to $S_j=S$ and the other bits involved are supposed to have distributions given by $q^{(S_i)}_{\mu i}$ . In Fig.\ref{belief}b one can see a graphical representation of $r^{(S)}_{\mu j}$ that can be interpreted as the influence of the bit $S_j$ and the mean-field $\prod _{l\in {\cal L}(\mu)\setminus l} q^{(S_l)}_{\mu l}$ (representing bits in ${\cal L}(\mu)$ over than $l$) over the check $J_\mu$. In the Fig.\ref{belief}c we see that each field $q^{(S)}_{\mu l}$ represents the influence of the checks in ${\cal M}(l)$, excluding $\mu$, over each bit $S_l$, this setup excludes the loops that may exist in the actual network. Employing Bayes theorem, $q^{(S)}_{\mu j}$ can be rewritten as: \begin{equation} q^{(S)}_{\mu j} = a_{\mu j}\;P(\{J_{\nu}:\nu \in {\cal M}(j)\setminus\mu\}\mid S_j)\;p^{(S)}_{j}, \end{equation} where $a_{\mu j}$ is a normalization constant such that $q^{(+1)}_{\mu j}+q^{(-1)}_{\mu j}=1$ and $p^{(S)}_{j}$ is the prior probability over the bit $j$. The distribution $P(\{J_{\nu}:\nu \in {\cal M}(j)\setminus\mu\}\mid S_j)$ can be replaced by a mean-field approximation by factorizing dependencies using fields $r^{(S)}_{\mu j}$: \begin{eqnarray} \label{eq:belief} q^{(S)}_{\mu j}&=&a_{\mu j}p^{(S)}_{j}\prod_{\nu\in{\cal M}(j)\setminus\mu} r^{(S)}_{\nu j} \nonumber \\ r^{(S)}_{\mu j} &=& \mbox{Tr}_{\{S_l:l\in{\cal L}(\mu)\setminus j\}} P(J_{\mu}\mid S_j=S, \{S_i:i \in {\cal L}(\mu)\setminus j \}) \prod _{i\in {\cal L}(\mu)\setminus j} q^{(S_i)}_{\mu i}. \end{eqnarray} A message estimate $\widehat\xi_j =\mbox{sign}\left(\langle S_j \rangle_{q^{(S)}_j}\right)$ can be obtained by solving the above equations and computing the pseudo-posterior: \begin{equation} q^{(S)}_{j}=a_{j}p^{(S)}_j\prod_{\nu\in{\cal M}(j)}r^{(S)}_{\nu j}, \label{eq:pseudo} \end{equation} where $a_{j}$ is a normalization constant. By taking advantage of the normalization conditions for the distributions $q^{(+1)}_{\mu j}+q^{(-1)}_{\mu j}=1$ and $r^{(+1)}_{\mu j}+r^{(-1)}_{\mu j}=1$ one can change variables and reduce the number of equations (\ref{eq:belief}) to the couple $\delta q_{\mu j} = q^{(+1)}_{\mu j}-q^{(-1)}_{\mu j}$ and $\delta r_{\mu j} = r^{(+1)}_{\mu j}-r^{(-1)}_{\mu j}$. Solving these equations, one can find back $r^{(S)}_{\mu j}=\frac{1}{2} \left(1\;+\;\delta r_{\mu j}S_j \right)$ and the pseudo-posterior can be calculated to obtain the estimate. \subsection{Connection with Statistical Physics} \label{sec:connection} The belief propagation algorithm was shown in \cite{mackay95b} to outperform other methods such as simulated annealing. In \cite{ks98a} it was proposed that this framework can be reinterpreted using statistical physics. The main ideas behind the approximations contained in (\ref {eq:belief}) are somewhat similar to the Bethe \cite{bethe} approximation to diluted two-body spin glasses. Actually, for systems involving two-body interactions it is known that the Bethe approximation is equivalent to solving exactly a model defined on a Cayley tree and that this is a good approximation for finitely connected systems in the thermodynamical limit \cite{wong_d}. In fact, loops in the connections become rare as the system size grows and can be neglected without introducing significant errors. The belief propagation can be seen as a Bethe-like approximation for multiple bodies interaction systems. The mean-field approximations used here are also quite similar to the TAP approach \cite{tap}. The fields $q^{(S)}_{\mu j}$ correspond to the mean influence of other sites other the site $j$ and the fields $r^{(S)}_{\nu j}$ represent the influence of $j$ back over the system (reaction fields). The analogy can be exposed by observing that the likelihood $p(J_\mu\mid \mbox{\boldmath $S$})$ is proportional to the Boltzmann weight: \begin{equation} w_B(J_\mu \mid \{S_j:j \in {\cal L}(\mu) \}) = \mbox{exp}\left(-\beta J_\mu \; \prod_{i\in\mu} S_{i}\right). \end{equation} That can be also written in the more convenient form: \begin{equation} \label{eq:likelihood} w_B(J_\mu \mid \{S_j :j \in {\cal L}(\mu) \}) = \frac{1}{2} \mbox{cosh}(\beta J_\mu)\left( 1\; + \;\mbox{tanh}(\beta J_{\mu})\; \prod_{j\in{\cal L}(\mu)}S_j \right). \end{equation} The variable $r_{\mu j}^{(S_j)}$ can then be seen as proportional to the effective Boltzmann weight obtained by fixing the bit $S_j$: \begin{equation} \label{eq:effective} w_{\mbox{\scriptsize eff}}(J_\mu \mid S_j) = \mbox{Tr}_{\{S_l\; :\;l \in {\cal L}(\mu)\setminus j \}}\; w_B(J_\mu \mid \{S_l\; :\;l \in {\cal L}(\mu) \})\prod _{l\in {\cal L}(\mu)\setminus j} q^{(S_l)}_{\mu l}. \end{equation} Plugging Eq.(\ref{eq:likelihood}) for the likelihood in equations (\ref{eq:belief}), using the fact that the prior probability is given by $p_j^{(S)}=\frac{1}{2}\left(1+\mbox{tanh}(\beta S F)\right)$ and computing $\delta q_{\mu j}$ and $\delta r_{\mu j}$: \begin{eqnarray} \label{eq:tap} \delta r_{\mu j}&=&\mbox{tanh}(\beta J_\mu) \prod_{l\in{\cal L}(\mu)\setminus j} \delta q_{\mu l}\nonumber \\ \delta q_{\mu j}&=&\mbox{tanh}\left(\sum_{\nu\in{\cal M}(l)\setminus \mu} \mbox{tanh}^{-1}( \delta r_{\nu j}) +\beta F \right). \end{eqnarray} \begin{figure} \hspace*{2cm} \epsfxsize=120mm \epsfbox{figure10.eps} \caption{Magnetization as a function of the flip probability $p$ for decoding using TAP equations for $K=2$. From the bottom: Monte-carlo solution of the RS saddle-point equations for unbiased messages ($f_s=0.5$) at $T=0.26$ (line) and $10$ independent runs of TAP decoding for each flip probability (plus signs), $T=0.26$ and biased messages ($f_s=0.1$) at Nishimori's temperature $T_N$.} \label{TAP1} \end{figure} The pseudo-posterior can then be calculated: \begin{equation} \label{eq:pseudoposterior} \delta q_{j}=\mbox{tanh}\left(\sum_{\nu\in{\cal M}(l)} \mbox{tanh}^{-1}( \delta r_{\nu j}) +\beta F \right), \end{equation} providing Bayes' optimal decoding $\widehat{\xi}_j=\mbox{sign}(\delta q_{j})$. It is important at this point to support the mean-field assumptions used here by methods of statistical physics \cite{ks98a}. The factorizability of the probability distributions can be explained by weak correlations between connections (checks) and by the cluster property: \begin{equation} \lim_{N\rightarrow \infty} \frac{1}{N^2}\sum_{i\neq j} \left(\langle S_i S_j\rangle_{p(S\mid J)} - \langle S_i \rangle_{p(S\mid J)} \langle S_j \rangle_{p(S\mid J)}\right)^2 \rightarrow 0 \end{equation} that bits $S_j$ obey within a pure state \cite{mezard}. \begin{figure} \hspace*{2cm} \epsfxsize=120mm \epsfbox{figure11.eps} \caption{Magnetization as a function of the flip probability $p$ for decoding using TAP equations for $K=5$. The dotted line is the replica symmetric saddle-point equations Monte-carlo integration for unbiased messages ($f_s=0.5$) at the Nishimori's temperature $T_N$. The bottom error bars correspond to $10$ simulations using the TAP decoding. The decoding performs badly on average in this scenario. The upper curves are for biased messages ($f_s=0.1$) at the Nishimori's temperature $T_N$. The simulations agree with results obtained using the replica symmetric ansatz and Monte-carlo integration.} \label{TAP2} \end{figure} One can push the above connections even further. Eqs.(\ref{eq:tap}), of course, depend on the particular received message $\mbox{\boldmath $J$}$. In order to make the analysis message independent, one can use a gauge transformation $\delta r_{\mu j} \mapsto \xi_j \delta r_{\mu j}$ and $\delta q_{\mu j} \mapsto \xi_j \delta q_{\mu j}$ to write: \begin{eqnarray} \label{eq:gaugetap} \delta r_{\mu j}&=&\mbox{tanh}(\beta J) \prod_{l\in{\cal L}(\mu)\setminus j} \delta q_{\mu l}\nonumber \\ \delta q_{\mu j}&=&\mbox{tanh}\left(\sum_{\nu\in{\cal M}(l)\setminus \mu} \mbox{tanh}^{-1}( \delta r_{\nu j}) +\beta \xi_j F \right). \end{eqnarray} In this form a success in the decoding process correspond to $\delta r_{\mu j}>0$ and $\delta q_{\mu j}=1$ for all $\mu$ and $j$. For a large number of iterations, one can expect the ensemble of belief networks to converge to an equilibrium distribution where $\delta r$ and $\delta q$ are random variables sampled from distributions $\widehat{\rho}(y)$ and $\rho(x)$ respectively. By transforming these variables as $\delta r=\mbox{tanh}(\beta y)$ and $\delta q=\mbox{tanh}(\beta x)$ and considering the actual message and noise as quenched disorder, Eqs.(\ref{eq:gaugetap}) can be rewritten as: \begin{eqnarray} \label{eq:newgaugetap} y&=&\frac{1}{\beta}\left\langle \mbox{tanh}^{-1}\left(\mbox{tanh}(\beta J) \prod_{j=1}^{K-1} \mbox{tanh}(\beta x_j)\right)\right\rangle_J\nonumber \\ x&=&\left\langle \sum_{j=1}^{C-1} y_j + \xi F\right\rangle_\xi. \end{eqnarray} The above relations lead to a dynamics on the distributions $\widehat{\rho}(y)$ and $\rho(x)$, that is exactly the same obtained when solving iteratively RS saddle-point equations (\ref{eq:saddle_point}). The probability distributions $\widehat{\rho}(y)$ and $\rho(x)$ can be ,therefore, identified with $\widehat{\pi}(y)$ and $\pi(x)$ respectively and the RS solutions correspond to decoding a generic message using belief propagation averaged over an ensemble of different codes, noise and signals. Eqs.(\ref{eq:tap}) are now used to show the agreement between the simulated decoding and analytical calculations. For each run, a fixed code is used to generate 20000 bit codewords from 10000 bit messages, corrupted versions of the codewords are then decoded using (\ref{eq:tap}). Numerical solutions for 10 individual runs are presented in Figs.\ref{TAP1} and \ref{TAP2}, initial conditions are chosen as $\delta r_{\mu l}=0$ and $\delta q_{\mu l}=\mbox{tanh}(\beta F)$ reflecting prior beliefs. In Fig.\ref{TAP1} we show results for $K=2$ and $C=4$ in the unbiased case, at code rate $R=1/2$ (prior probability $p_j^{(1)}=f=0.5$) at a low temperature $T=0.26$ (we avoided $T=0$ due to numerical difficulties). Solving saddle-point equations (\ref{eq:saddle_point}) numerically using Monte-carlo integration methods we obtain solutions with good agreement to simulated decoding. In the same figure we show the performance for the case of biased messages ($p_j^{(1)}=f_s=0.1$), at code rate $R=1/4$. Also here the agreement with Monte-carlo integrations is rather convincing. The third curve in Fig.\ref{TAP1} shows the performance for biased messages at Nishimori's temperature $T_N$, as expected, it is far superior compared to low temperature performance and the agreement with Monte-carlo results is even better. In Fig.\ref{TAP2} we show the results obtained for $K=5$ and $C=10$. For unbiased messages the system is extremely sensitive to the choice of initial conditions and does not perform well in average even at Nishimori's temperature. For biased messages ($f_s=0.1$, $R=1/4$) results are far better and in agreement with Monte-carlo integration of the RS saddle-point equations. The experiments show that belief propagation methods may be used successfully for decoding Sourlas-type codes in practice, and provide solutions that are well described by RS analytical solutions. \subsection{Basin of Attraction} \label{sec:basin} \begin{figure} \hspace*{2cm} \epsfxsize=120mm \epsfbox{figure12.eps} \caption{Top: Maximum initial deviation $\lambda$ for decoding. Top: $\lambda$ as function of the number of interactions $K$. Circles are averages over $10$ different codes with $N=300$, $R=1/3$ and noise level $p=0.1$. Bottom: $\lambda$ as function of the connectivity $C$. Circles are averages over $10$ codes with $N=300$, $K=3$ and noise level $p=0.1$. Lines and $\times$'s correspond to the RS dynamics described by the saddle-point equations. } \label{basins} \end{figure} To asses the size of the basin of attraction we consider the decoding process as a dynamics in the graphs space where edges $\delta q_{\mu j}$ are considered as dynamical variables. In gauged transformed equations (\ref{eq:gaugetap}) , the perfect decoding of a message correspond to $\delta q_{\mu j}=1$ . To analyse the basin of attraction we start with random initial values with a given normalized deviation from the perfect decoding $\lambda=\frac{1}{NC}\sum_{\mu j}(1-\delta q_{\mu j}^{0})$. It is analogous to the finite magnetizations used in the naive mean-field of Section II, since a given $\delta q_{\mu j}^{0}$ corresponds to a given magnetization value by using Eq.(\ref{eq:pseudoposterior}). In Fig.\ref{basins} we show the maximal deviation in initial conditions required for successful decoding. Top figure shows an average over $10$ different codes with $N=300$ (circles) for a fixed code rate $R=1/3$, fixed noise level $p=0.1$ and increasing $K$. Bottom figure shows the maximal deviation in initial conditions for a fixed number of spins per interaction $K=3$, noise level $p=0.1$ and increasing $C$. We confirm the fidelity of the RS description by comparing the experimental results with the basin of attraction predicted by saddle-point equations (\ref{eq:saddle_point}). One can interpret these equations as a dynamics in the space of distributions $\pi(x)$. Performing the transformation $X=\mbox{tanh}(\beta x)$, one can move to the space of distributions $\Pi(X)$ with support over $[-1,+1]$. The initial conditions can then be described simply as $\Pi^0(X)=(1-\frac{\lambda}{2})\delta(X-1)+\frac{\lambda}{2}\delta(X+1)$. In Fig.\ref{basins} we show the basin of attraction of this dynamics as lines and $\times$'s. The $K=2$ case is the only practical code from a dynamical point of view, since it has the largest basin of attraction and no prior knowledge on the message is necessary for decoding. Nevertheless, this code's performance degrades faster than the $K>2$ case as shown in Section III, which points to a compromise between good dynamical properties in one side and good performance in the other. One idea could be having a code with changing $K$, starting with $K=2$ to guarantee convergence and progressively increasing its values to improve the performance \cite{idosaad}. On the other hand, the basin of attraction increases with $C$. Again it points to a trade off between good equilibrium properties (small $C$ and large code rates) and good dynamical properties (large $C$, large basin of attraction). Mixing small and large $C$ values in the same code seems to be a way to take advantage of this trade-off \cite{LMSS,davey,VSK}. \section{Concluding Remarks} In this paper we studied, using the replica approach, a finite connectivity many-body spin glass that corresponds to Sourlas' codes for finite code rates. We have shown, using a simplified one step RSB solution for spin glass phase, that for $K\rightarrow\infty$ and $C=\alpha K$ regime at low temperatures the system exhibits a FERRO-SG phase transition that corresponds to Shannon's bound. However, we have also shown that the decoding problem for large $K$ has bad convergence properties when simulated annealing strategies are used. We were able to find replica symmetric solutions for finite $K$ and found good agreement with practical decoding performance using belief networks. Moreover, we have shown that RS saddle-point equations actually describe the mean behavior of belief propagation algorithms. We studied the dynamical properties of belief propagation and compared to statistical physics predictions, confirming the validity of the description. The basin of attraction was shown to depend on $K$ and $C$. Strategies for improving the performance were discussed. The same methodology has been recently employed successfully \cite{kms99} to state-of-the-art algorithms as the recent rediscovered Gallager codes \cite{mackay95} and its variations \cite{idosaad,VSK}. We believe that the connections found between belief networks and statistical physics can be further developed to provide deeper insights into the typical performance of general error-correcting codes. \label{sec:conclusions} \ack This work was partially supported by the program ``Research For The Future'' (RFTF) of the Japanese Society for the Promotion of Science (YK) and by EPSRC grant GR/L52093 and a Royal Society travel grant (DS and RV).
\section{Introduction} Some five years ago, Seiberg and Witten gave an ansatz for the dominant piece of the effective action governing the light degrees of freedom of $SU(2)$ ${\cal N} = 2$ super Yang-Mills theory at low energy\cite{SeiWitt}. It is given in terms of an auxiliary complex algebraic curve ${\cal C}$ (whose moduli space is identified with the quantum moduli space of the low-energy theory ${\cal M}_\Lambda$) and a given meromorphic differential, $dS_{SW}$, that induces a special geometry on ${\cal M}_\Lambda$ (see Refs.\cite{reviews} for excellent reviews). The solution was soon extended to other gauge groups and matter content by determining both the appropriate complex curve and meromorphic differential \cite{sun,klemm,sunmat,hananoz,son}, thus leading to a substantial progress in our understanding of ${\cal N}=2$ supersymmetric gauge theories. In particular, the appearance of an auxiliary Riemann surface made it possible to identify remarkable connections with string theory, singularity theory of differentiable maps and integrable systems. The framework that will be used in this paper is strongly inspired by the latter one. It is well-known that, as long as ${\cal N}=2$ supersymmetry is unbroken, the low energy effective action is given in terms of a holomorphic prepotential ${\cal F}$. The solution proposed by Seiberg and Witten embodies a prescription to compute this prepotential. However, its explicit evaluation for a given gauge group and matter content is technically involved and it requires to integrate an expression for the B-periods of $dS_{SW}$ as functions of its A-periods. The complexity of this procedure increases rapidly with the rank of the gauge group, even without matter hypermultiplets. Whereas perturbative contributions to ${\cal F}$ are exhausted by one-loop diagrams \cite{Sei}, the non-perturbative part is given by an infinite series of instanton corrections. The importance of instanton calculus lies precisely in the fact that it provides one of the few non-perturbative links between the Seiberg--Witten solution and the microscopic non-abelian field theory that it is supposed to describe effectively at low energies. From the microscopic point of view, the first few instanton contributions to the asymptotic semiclassical expansion of the effective prepotential have been computed for gauge group $SU(N_c)$, and a remarkable agreement with the Seiberg--Witten solution has been found \cite{hunter}. From the side of the effective theory, several methods for determining the instanton corrections have been developed in the last few years by using the Picard--Fuchs equations \cite{klemm,itoyang1}, holomorphicity arguments \cite{itosas1}, analytic continuation \cite{dhoker} (also for non-hyperelliptic curves \cite{isa}), modular anomaly equations \cite{MNW}, etc. Among them, we would like to distinguish those methods that lead to recursion relations for the $k$-instanton corrections, as long as they give an implicit expression for the {\em exact} solution. For the case of $SU(2)$, recursion relations determining the whole instanton expansion have been obtained both in the pure gauge theory \cite{Matone} and when matter is included \cite{itoyang1}. These recursion relations were obtained by combining the renormalization group and Picard-Fuchs equations. Also one-instanton corrections for the whole ADE series where obtained along the same lines in \cite{itoyang}. $SU(N_c)$ with additonal matter in the adjoint representation was considered in \cite{DHPh} from the point of view of the Calogero-Moser model, and in \cite{MNW} (for ADE groups with dual Coxeter number $k_D\leq 6$) where the modular anomaly equations of softly broken ${\cal N}=4$ supersymmetric gauge theories were invoked. In a recent paper, \cite{emm}, a new strategy was observed to work very well for the case of pure $SU(N_c)$. It is based on a set of first- and second-order equations for the logarithmic derivatives of the prepotential with respect to the dynamical scale $\Lambda$, evaluated all over the moduli space \begin{eqnarray} \pder{{\cal F}}{\log\Lambda\,} & = & {\beta\over 2\pi i } u_{2}~, \label{lder} \\ \dpder{{\cal F}}{(\log\Lambda)} & = & -{\beta^2\over 2\pi i} \pder{u_{2}}{ a^i}\pder{u_{2}}{ a^j}{1\over i\pi} \partial_{\tau_{ij}}\log\Theta_E(0|\tau) ~, \label{llder} \end{eqnarray} where the different quantities entering these expressions will be explained below. Each one of these equations was obtained separately in the last few years by many authors \cite{Matone,marimoore,losev,EY,rge}. It was not until very recently that a unifying approach based on the Whitham hierarchy was shown to be useful to obtain both \cite{ITEP}. In Ref.\cite{emm}, the ansatz for the semiclassical expansion of the prepotential for gauge group $SU(N_c)$ was inserted in both sides of equations (\ref{lder})--(\ref{llder}) with the result of an elegant and systematic procedure that allowed us to compute instanton corrections {\em up to any desired order} with relatively little effort. In particular, this method does not require knowledge of the actual solution for the periods $a$ and $a_D$ of $dS_{SW}$, a fact which spares a considerable amount of work. In this paper we will exhibit the strengh of this method by extending the results of \cite{emm} to any classical gauge group with and without matter content. We shall limit ourselves to asymptotically free theories. The obvious question arises, about the validity of equations like (\ref{llder}) for all these situations. It turns out that, for our present purposes, the only constraint which seems to be unavoidable within this method is that massive hypermultiplets have to be introduced in pairs, degenerated in mass. This can be understood either from a purely theoretical study, which we shall leave for a separate paper \cite{egrmm}, or else, ``a posteriori", for the consistency of the results. Our aim is that this paper could be useful to anyone interested in finding explicit expressions for the instanton corrections to the effective prepotential of ${\cal N}=2$ supersymmetric gauge theories. To this end, general formulas for up to three-instanton corrections will be given and the method to obtain arbitrary higher corrections will be clearly explained. A number of particular examples will be also worked out for the lower rank groups and small number of flavour-pairs, in order to make them easily available for futher comparison even to other methods or, actually, to the results obtained in the microscopic non-Abelian field theory. For the one- and two-instanton corrections, our results coincide with those in the literature while, for most of the three-instanton contributions, our results are new. \section{Recursive Evaluation of the Effective Prepotential} \setcounter{equation}{0} \indent \subsection{ Review and Notation} For completeness, we shall start by reviewing the r\^ ole of the different ingredients that enter the formulas (\ref{lder}) and (\ref{llder}). The low-energy dynamics of ${\cal N}=2$ supersymmetric theories with classical gauge group $G$ corresponding to $N_c$ colors, and $N_f$ hypermultiplets of mass $m_f = m_{(f+ N_f/2)}$ ({\it i.e.} degenerated in masses) in the fundamental representation of $G$, can be described in terms of an auxiliary hyperelliptic curve ${\cal C}$ given by \begin{equation} y^2 = (P(\lambda,e_p)+T(\lambda,m_f,\Lambda))^2 - 4\Lambda^{\beta} F(\lambda,m_f) ~, \label{curveone} \end{equation} where $P$ is the characteristic polynomial of $G$, $\Lambda$ is the quantum generated dynamical scale, $\beta$ is the coefficient of the one-loop ${\cal N}=2$ beta function and $e_p$ are the eigenvalues of the complex scalar field $\langle\phi\rangle = \sum_p e_p h_p$ (see Ref.\cite{Lie} for the conventions followed in the notation of Lie group and Lie algebra objects) that belongs to the ${\cal N}=2$ vector supermultiplet in the adjoint of $G$. $T$ and $F$ are polynomials that do not depend on the {\em moduli} $e_p$ and $T$ is different from zero only when $N_f>N_c$. As pointed out in Ref.\cite{dhoker}, when the gauge group is $SU(N_c)$ or $SO(N_c)$ all dependence on $T$ can be absorbed in a redefinition of $e_p$, the effective prepotential remaining untouched. Thus, we can set $T=0$, and write the hyperelliptic curve ${\cal C}$ as \begin{equation} y^2 = P^2(\lambda,e_p) - 4\Lambda^{\beta} F(\lambda,m_f) ~. \label{curve} \end{equation} In the case of $Sp(N_c)$, there is a residual value for $T$, $T = \Lambda^{N_c-N_f+2}(\prod_{f=1}^{N_f/2}m_f^2)$. In order to consider all the different cases within a unified framework we will neglect this contribution by setting the mass of one of the degenerated hypermultiplets to zero (say for example $m_{N_f/ 2}=0$). Following Ref.\cite{dhoker} this case will be denoted $Sp(N_c)''$, and the corresponding hyperelliptic curve has the form (\ref{curve}) with the proviso that, according to our previous remark, the number of hypermultiplets will be at least two, $N_f=N_f'+2\geq 2$. Consequently, the case of pure $Sp(N_c)$ will not be attainable from our results though it is quite simple to make the appropriate modifications, as will be explained below. It is convenient, for later use, to list the form of $P$, $F$ and $\beta$ for all classical groups: \vskip7mm \centerline {\begin{tabular}{|c||c|c|c|c|c|c|c|c|} \hline $G$ & $P(\lambda,e_p)$ & $F(\lambda)$ & $\beta/2$ & $l_G$ & $\xi$ & $\xi^+$ & $\xi^-$ & $k_D$ \\ \hline\hline $SU(r+1)$ & $\displaystyle\prod_{p=1}^{r+1}(\lambda-e_p)$ & $\displaystyle\prod_{f=1}^{N_f}(\lambda+m_f)$ & $r+1-N_f/2$ & r+1 & 1 & 1 & 0 & r+1 \\ \hline $SO(2r)$ & $\displaystyle\prod_{p=1}^{r}(\lambda^2-e_p^2)$ & $\displaystyle\lambda^4\prod_{f=1}^{N_f}(\lambda^2-m_f^2)$ & $ 2r-N_f-2$ & r & 1 & 1 & 1 & 2r-2 \\ \hline $SO(2r+1)$ & $\displaystyle\prod_{p=1}^{r}(\lambda^2-e_p^2)$ & $\displaystyle\lambda^2\prod_{f=1}^{N_f}(\lambda^2-m_f^2)$ & $2r-N_f-1$ & r & 1 & 1 & 1 & 2r-1 \\ \hline $Sp(2r)''$ & $\displaystyle\prod_{p=1}^{r}(\lambda^2-e_p^2)$ & $\displaystyle\prod_{f=1}^{N_f'}(\lambda^2-m_f^2)$ & $2r-N_f'$ & r & 2 & 2 & 2 & r+1 \\ \hline \end{tabular}} \vskip5mm \centerline{ {\em Table 1}} \vskip7mm\noindent The symmetric polynomials $\bar u_k(e_p)$ and $ t_i(m_f)$ are defined through the expansions \begin{eqnarray} P(\lambda,e_p) &\equiv & \lambda^{h} - \sum_{k=2}^{h} \lambda^{h-k} \bar u_k(e_p) ~, \label{casimirs} \\ F(\lambda,m_f) &\equiv &\lambda^{h} + \sum_{k=1}^{h} \lambda^{h-k} t_k(m_f) ~, \label{lastes} \end{eqnarray} where $h$ stands, in each case, for the highest power in $P(\lambda,e_p)$ or $F(\lambda,m_f)$. The moduli for each curve (\ref{curve}) can be taken to be either the independent roots $e_p$ or the $a^i$ defined as the coefficients of $\langle \phi\rangle = a^i H_i$ in the Chevalley basis, hence linearly related to $e_p$ (see Ref.\cite{Lie}). Neither of these parameters are invariant under Weyl transformations which, in particular, act by permutation on the $e_p$. On the contrary, the symmetric polynomials (Casimirs) $\bar u_k$ provide faithfull coordinates for the moduli space of vacua. In particular, $\bar u_2 ={1\over 2}{\rm tr} \phi^2= (\frac{\xi^+ +\xi^-}{2\xi^+})\sum_{p=1}^{l_G} e_p^2 $, and in general: $\bar u_k = {1\over k} {\rm tr} \phi^k+$ ..., the dots standing for homogeneous powers of lower Casimir operators. The Seiberg--Witten meromorphic differential can be written as \begin{equation} dS_{SW}(u_k,m_f) = \left({P}' - \frac{P F'}{2 F}\right) {\lambda d\lambda \over y} ~, \label{diff} \end{equation} and the quantum relations between the low-energy coordinates of the moduli space $a^i, a_{D\,j}$ and the ``mean field" order parameters $u_k(a) = {1\over k} {\rm tr} \langle \phi^k\rangle + \cdots = \bar u_k(a^i) + {\cal O}(\Lambda)$ are implicitely given by the period integrals \begin{equation} a^i(u_k,m_f) = \oint_{A^i} dS_{SW}(u_k,m_f) ~~~~;~~~~ a_{D\,j}(u_k,m_f) = \oint_{B_j} dS_{SW}(u_k,m_f) ~, \label{theper} \end{equation} where $A^i$ and $B_j$ constitute a symplectic basis of homology cycles with canonical intersections of the hyperelliptic curve (\ref{curve}); the effective prepotential ${\cal F}$ is implicitely defined by the equation \begin{equation} a_{D\,i} = \pder{{\cal F}(a)}{a^i} ~, \label{implicit} \end{equation} so that its exact determination involves the integration of functions $a_{D\,i}(a)$ for which there is not a closed form available. In this context, the existence of an algorithm that let us determine the exact form of ${\cal F}$ without going through the actual computation of $a^i(u_k,m_f)$ and $a_{D\,j}(u_k,m_f)$ is welcome. As mentioned above, our first ingredient is the set of RG equations (\ref{lder})--(\ref{llder}). Strictly speaking, as they stand, they are valid for the pure gauge theory. In presence of matter, the first equation receives an additional term which depends only on the masses and the second one has to be modified for $\beta=2$ where it receives an additional constant contribution \cite{egrmm}. In summary\footnote{ Although the addition of an $a^i$-independent terms to ${\cal F}$ is unphysical from the point of view of the effective theory, the embedding of the Seiberg-Witten solution into the Whitham dynamics fixes them as a function of the bare coupling $\tau_0$. A similar behaviour was observed in the study of ${\cal F}$ near the strong coupling singularities of $SU(N_c)$ ${\cal N}=2$ super Yang--Mills theory \cite{edemas}.} \begin{eqnarray} \pder{{\cal F}}{\log\Lambda\,} & = & {\beta\over 2\pi i } u_{2} + \varphi(m)~, \label{lderm} \\ \dpder{{\cal F}}{(\log\Lambda)\,} & = & -{\beta^2\over 2\pi i} \pder{u_2}{a^i} \pder{u_2}{a^j} {1\over i\pi}\partial_{\tau_{ij}}\log\Theta_E(0|\tau) + {\beta^2\over 2\pi i}\Lambda^2 \delta_{\beta,2} ~. \label{formula} \end{eqnarray} In Eq.(\ref{formula}), $\Theta_E(0|\tau)$ is Riemann's theta function associated to the hyperelliptic curve ${\cal C}$ \begin{equation} \Theta \left[ {\vec\alpha\atop \vec\beta}\right] (\xi|\tau)= \sum_{n_k \in {\bf Z} }\displaystyle{ e^{i \pi \bigl[ \tau_{ij}(n_i+ \alpha_i) (n_j+ \alpha_j) + 2 (n_i+ \alpha_i)(\xi + \beta_i) \bigr]} }~, \label{thetaf} \end{equation} where $E = \left[ {\vec\alpha\atop \vec\beta}\right]$ stands for an even half integer characteristic. In almost all the cases this characteristic will be $E=\left[{0,\dots,0\atop {1\over 2},\dots,{1\over 2}}\right]$, as it is in the case of pure $SU(N_c)$ \cite{emm}. The only exception is $Sp(2r)''$ for which the characteristic gets modified to the value $E=\left[{0,\dots,0,0\atop 0,\dots,0,{1\over 2}}\right]$. The second ingredient is an ansatz for the instanton expansion of the prepotential valid for any classical gauge group $G$ and $N_f$ massive hypermultiplets with paired masses \begin{eqnarray} {\cal F}_{G,N_f} & = & \frac{\tau_0^G}{4\pi i}\sum_{\alpha_+} Z_{\alpha_+}^2 + {i\over 4 \pi}\xi \sum_{\alpha_+} Z_{\alpha_+}^2 \log \,{Z_{\alpha_+}^2\over \Lambda^2}-{i\over 4\pi}\xi^+ \sum_{p=1}^{l_G}\sum_{f=1}^{N_f/2}(e_p+m_f)^2\log\,{{(e_p+m_f)^2}\over {\Lambda^2}} \nonumber\\ && -{i\over 4\pi}\xi^- \sum_{p=1}^{l_G}\sum_{f=1}^{N_f/2}(e_p-m_f)^2\log\,{{(e_p-m_f)^2}\over {\Lambda^2}}+{1\over 2 \pi i}\sum_{k=1}^\infty {\cal F}_{k}(Z)\Lambda^{k\beta} ~, \label{elprep} \end{eqnarray} where $\alpha_+$ denotes a positive root and $\sum_{\alpha_+}$ is the sum over all positive roots. The set $\{ \alpha_i \}_{i=1,...,r}$ stands for the simple roots of the corresponding classical Lie algebra, they generate the root lattice $\Delta = \{\alpha = n^i \alpha_i| n^i\in {\mathbb{Z}}\}$. The dot product $(\cdot)$ of two simple roots $\alpha_i$ and $\alpha_j$ gives an element of the Cartan matrix, $A_{ij}=\alpha_i\!\cdot\!\alpha_j$, and extends bilinearly to arbitrary linear combinations of simple roots. So, for example, for any root $\alpha = n^j\alpha_j\in\Delta$, the quantities $Z_\alpha$ are defined by $Z_\alpha = a\!\cdot\! \alpha \equiv a^i A_{ij} n^j$ where $a= a^i\alpha_i$. For non-simply laced Lie algebras this product is not symmetric. Simple roots can be written in terms of the orthogonal set of unit vectors $\{\epsilon_p\}_{p=1,\cdots,l_G}$. The order parameters $a^i$ and $e_p$ are related by $e_p = a\!\cdot\!\epsilon_p$. The exact relations and the actual values of $\alpha_i\!\cdot\!\epsilon_p$ for each classical gauge group can be found in Ref.\cite{Lie}. Finally, we also have that $\lambda^k \!\cdot\! \alpha_j= \delta^k{_j}$ define the fundamental weights. In particular, this means that $\alpha_i = \sum_k A_{ik}\lambda^k$. We have also introduced three parameters $\xi,~\xi^+$ and $\xi^-$, so as to deal with all classical Lie algebras within one single ansatz; $k_D$ denotes the dual Coxeter number. The particular values of these variables for each classical gauge group are shown in Table 1. The coefficient of the one-loop beta function turns out to be given by \begin{equation} \beta/2 = \xi k_D - \frac{N_f}{2}\left( 1 ~+~ \frac{\xi^-}{\xi^+}\right) ~. \label{bfun} \end{equation} By expressing the roots $\alpha_+$ in terms of $\epsilon_p$, ({\it i.e.}, $Z_{\alpha_+}$ in terms of $e_p$) one can also check that the following relation holds \begin{equation} \xi^+~\sum_{p=1}^{l_G} e_p^2 = {1\over k_D}~\sum_{\alpha_+} Z_{\alpha_+}^2 ~. \label{rel} \end{equation} \subsection{The procedure} It is important to notice that we may shift $\tau_0^G$ in (\ref{elprep}) to any value by appropriately rescaling $\Lambda$, and this will be reflected in our choice for the normalization of the ${\cal F}_{k}(Z)$. We have fixed it in all cases to be $\tau_0^G=3\beta/2k_D$ so that quadratic terms in $Z_\alpha$ do not contribute to the coupling constant $\tau_{ij}$ (see (\ref{latau}) below). The l.h.s. of Eq.(\ref{formula}) can be easily computed from the expansion of the effective prepotential (\ref{elprep}) to be \begin{equation} \dpder{{\cal F}}{(\log \Lambda)} = {1\over 2\pi i} \sum_{k=1}^\infty (k \beta)^2~{\cal F}_k(Z) ~\Lambda^{k\beta} ~, \label{makdj} \end{equation} then comparing (\ref{makdj}) and (\ref{formula}) we get \begin{equation} \sum_{k=1}^{\infty}k^2{\cal F}_k\Lambda^{k \beta}=-\pder{u_2}{a^i} \pder{u_2} {a^j} {1\over i\pi}\partial_{\tau_{ij}}\log \Theta_E(0|\tau)~\label{ic} + \Lambda^2 \delta_{\beta,2} ~, \end{equation} such that the instanton correction ${\cal F}_k$ can be obtained through a set of recursive relations after expanding the r.h.s. of Eq.(\ref{ic}) in powers of $\Lambda^{\beta}$. The expansion of the derivative of the quadratic Casimir in powers of $\Lambda$ can be obtained from the RG equation (\ref{lderm}) and (\ref{elprep}), \begin{eqnarray} \pder{u_2}{a^i} &=& {2\pi i\over \beta} \ppder{{\cal F}}{a^i}{\log\Lambda\,} \cr &=& {2\xi\over \beta}\sum_{\alpha_+}Z_{\alpha_+}\partial_iZ_{\alpha_+} -{N_f\over\beta } (\xi^++\xi^-) \sum_p e_p\partial_i e_p+ \sum_{k=1}^\infty k {\cal F}_{k,i} ~ \Lambda^{k \beta}\cr &=& \xi^+\sum_pe_p\partial_ie_p + \sum_{k=1}^\infty k {\cal F}_{k,i} ~ \Lambda^{k \beta}\equiv \sum_{k=0}^{\infty}{{\cal H}_{2,i}^{(k)}}\Lambda^{k \beta} ~, \label{expanh} \end{eqnarray} where ${\cal F}_{k,i}= \partial{\cal F}_k/\partial a^i $ and use has been made of (\ref{bfun}) and (\ref{rel}). To expand the theta function we need to compute from (\ref{elprep}) the couplings in the semiclassical region, \begin{eqnarray} \tau_{ij} &=&\ppder{{\cal F}}{a^i}{a^j} \nonumber \\ &=& {i\over 2\pi}\xi \sum_{\alpha_+} \pder{Z_{\alpha_+}}{a^i} \pder{Z_{\alpha_+}}{a^j} \log \left({ Z^2_{\alpha_+} \over \Lambda^2} \right) -{i\over 2\pi}\xi^+ \sum_{p=1}^{l_G}\sum_{f=1}^{N_f/2}\pder{e_p}{a_i}\pder{e_p}{a_j} \log~{{(e_p+m_f)^2} \over {\Lambda^2}}\nonumber\\ &&-{i\over 2\pi}\xi^- \sum_{p=1}^{l_G}\sum_{f=1}^{N_f/2}\pder{e_p}{a_i}\pder{e_p}{a_j} \log~{{(e_p-m_f)^2} \over {\Lambda^2}}+{1\over 2\pi i} \sum_{k=1}^\infty {\cal F}_{k,ij} \Lambda^{k \beta} ~, \label{latau} \end{eqnarray} with ${\cal F}_{k,ij}= \ppder{{\cal F}_k}{a^i}{a^j}$. So the term involving the couplings that appear in the theta function $\Theta_E$ can be written as \begin{eqnarray} i\pi \, n^i\tau_{ij} n^j &=& \sum_{\alpha_+} \log \left({Z_{\alpha_+}\over \Lambda}\right)^{-\xi(\alpha\,\!\cdot\alpha_+)^2} +~\sum_{p,f}\log~\left(\frac{e_p+m_f}{\Lambda}\right)^{\xi^+(\alpha\,\!\cdot \epsilon_p)^2}+ \nonumber\\ && + ~\sum_{p,f} \log~\left(\frac{e_p-m_f}{\Lambda}\right)^{\xi^-(\alpha\,\!\cdot\epsilon_p)^2} +~ {1\over 2}\sum_{k=1}^\infty~ (\alpha\!\cdot\!\!\F''_k\!\!\cdot\! \alpha) ~\Lambda^{k \beta} ~, \label{tauroot} \end{eqnarray} where $\alpha\!\cdot\!\!\F''_k\!\!\cdot\! \alpha\equiv\sum_{i,j}n^i{\cal F}_{k,ij} n^j$. Also $\alpha= n^i\alpha_i$, and we have set \begin{equation} n^i \left(\pder{Z_{\alpha_+}}{a^i}\right) = n^i~(\alpha_i\!\cdot\!\alpha_+) = \alpha\!\cdot\!\alpha_+ ~~~~~;~~~~~~~ n^i \left(\pder{e_p}{a^i}\right) = n^i~(\alpha_i\!\cdot\!\epsilon_p) = \alpha\cdot\epsilon_p ~. \end{equation} Inserting (\ref{tauroot}) in the Theta function (\ref{thetaf}) with a characteristic $E=\left[{0,\dots,0\atop {1\over 2},\dots,{1\over 2}}\right]$, we obtain \begin{eqnarray} \Theta_E(0|\tau) &=& \sum_{\vec n} \exp\biggl[ i\pi n^i\tau_{ij} n^j + i \pi \sum_{k} n_k\biggr] \cr &=& \sum_{\alpha\in\Delta}(-1)^{\rho\cdot \alpha} \prod_{\alpha_+}\left( {Z_{\alpha_+}\over \Lambda}\right)^{-\xi(\alpha\,\!\cdot\alpha_+)^2} \prod_{p,f}\left({{e_p+m_f}\over{\Lambda}}\right)^{\xi^+(\alpha\,\!\cdot\epsilon_p)^2} \prod_{p,f}\left({{e_p-m_f}\over{\Lambda}}\right)^{\xi^-(\alpha\,\!\cdot\epsilon_p)^2} \cr &&~~~~~~~\times\prod_{k=1}^\infty\exp\left({{1\over 2} (\alpha\!\cdot\!\!\F''_k\!\!\cdot\! \alpha) \Lambda^{k \beta}}\right) \cr &=& \sum_{s=0}^\infty \sum_{\alpha\in\Delta_s} (-1)^{\rho\cdot\alpha} \prod_{\alpha_+} Z_{\alpha_+}^{-\xi(\alpha\,\!\cdot\alpha_+)^2} \prod_{p}\left[ R(e_p)\right]^{(\alpha\,\!\cdot\epsilon_p)^2} \nonumber\\ &&~~~~~~~\times\prod_{k=1}^\infty\left(\sum_{m=0}^\infty {1\over 2^m m!}\left(\alpha\!\cdot\!\!\F''_k\!\!\cdot\! \alpha\right)^m\, \Lambda^{k\beta m }\right) \Lambda^{s\beta } \cr &\equiv& \sum_{l=0}^\infty \Theta^{(l)} \Lambda^{l \beta} ~. \label{expantheta} \end{eqnarray} In the previous expression, $\rho$ is the maximal weight $\rho = \sum_{i=1}^{r} \lambda^i$. In the case of $Sp(N_c)''$, due to its peculiar characteristic, the dot product ${\rho\cdot \alpha}$ needs to be replaced by~ ${\rho_r\!\cdot\alpha= n^r}$. On the other hand, $\Delta_s\subset\Delta$ is a subset of the root lattice composed of those lattice vectors $\alpha\in\Delta_s$ that fulfill the constraint $ \frac{1}{2}\xi^+ \sum_p(\alpha\!\cdot \epsilon_p)^2 = s ~. $ The Weyl group permutes the $\epsilon_p$, hence the previous statement is Weyl invariant. In other words, the sets $\Delta_s$ are unions of Weyl orbits. This fact guarantees that the final result will recombine into Weyl invariant expressions. In (\ref{expantheta}) we have also introduced the polynomial \begin{eqnarray} R(\lambda,m_f) &=& \prod_{f=1}^{N_f/2}(\lambda+m_f)^{\xi^+} \prod_{f=1}^{N_f/2}(\lambda-m_f)^{\xi^-} \label{erres}\\ &\equiv& \left(\lambda^h + \sum_{i=1}^{h} q_i(m_f)\lambda^{h-i}\right)^{\xi^+} ~, \label{lasquses} \end{eqnarray} where $h$, again, stands for the appropriate highest power. It is at this point where the need for pairwise equal masses enters. Otherwise, we would be dealing with square root factors of the form $(\lambda \pm m_f)^{1/2}$. Now we can collect the first few terms in the expansion (\ref{expantheta}), \[ \Theta^{(0)} = 1 ~, ~~~~~~~~~~ \Theta^{(1)} = \sum_{\alpha\in\Delta_1}(-1)^{\rho\cdot\alpha} \prod_{\alpha^+} Z_{\alpha_+}^{-\xi(\alpha\cdot\alpha_+)^2} \prod_p\left[R(e_p)\right]^{(\alpha\cdot \epsilon_p)^2} ~, \] \begin{eqnarray} \Theta^{(2)} & = & \sum_{\alpha\in\Delta_1}(-1)^{\rho\cdot\alpha}\,{1\over 2} (\alpha\!\cdot\!\!\F''_1\!\!\cdot\! \alpha ) \prod_{\alpha^+} Z_{\alpha_+}^{-\xi(\alpha\cdot\alpha_+)^2} \prod_p\left[R(e_p)\right]^{(\alpha\cdot \epsilon_p)^2} \nonumber\\ & & + \sum_{\beta\in\Delta_2}(-1)^{\rho\cdot\beta} \prod_{\alpha^+} Z_{\alpha_+}^{-\xi(\beta\cdot\alpha_+)^2} \prod_p\left[R(e_p)\right]^{(\alpha\cdot \epsilon_p)^2} \nonumber ~. \end{eqnarray} However, in the logarithmic derivative, the Theta function appears in the denominator so we shall need the expansion of $\Theta(0|\tau)^{-1}$ in terms of $\Lambda$. We can write this expansion as \begin{equation} \Theta(0|\tau)^{-1} = \sum_{l=0}^{\infty} \Xi^{(l)}(\Theta) \, \Lambda^{2Nl} ~. \label{tauinv} \end{equation} Here $\Xi^{(0)}(\Theta) = 1$ and for $\Xi^{(l)}(\Theta)$ we can write in general \begin{equation} \Xi^{(l)}(\Theta) = \sum_{(p_1,...,p_k) \in {\bf N}^k}^{p_1 + 2p_2 + ... + kp_k = l} \chi_{(p_1,...,p_k)} \prod_{i=1}^l (\Theta^{(i)})^{p_i} ~, \label{tauinvcoef} \end{equation} where the coefficients $\chi$ are parametrized by the partition elements $(p_1,...,p_k)$. The first few values for these parameters are, for example, \[ \chi_{(1)} = -1 ~, ~~~ \chi_{(2,0)} = 1 ~, ~~~ \chi_{(0,1)} = -1 ~, ~~~ \chi_{(3,0,0)} = -1 ~, ~~~ \chi_{(1,1,0)} = 2 ~, ~~~ \chi_{(0,0,1)} = -1 ~, \] and using these values we can immediately obtain the lower $\Xi^{(l)}(\Theta)$. Next, we compute the derivative of the theta function with respect to the period matrix \begin{eqnarray} {1\over i\pi}\partial_{\tau_{ij}}\Theta_E(0,\tau) & = & \sum_{ n} n^i n^j \exp\biggl[ i\pi n^k\tau_{kl} n^l + i \pi \sum_{k} n_k\biggr] \cr & = & \sum_{s=1}^\infty \sum_{\alpha\in\Delta_s} (-1)^{\rho\cdot\alpha} (\lambda^i\!\cdot\alpha)(\lambda^j\!\cdot\alpha) \prod_{\alpha_+} Z_{\alpha_+}^{-\xi(\alpha\,\!\cdot\alpha_+)^2} \prod_{p}\left[ R(e_p)\right]^{(\alpha\,\!\cdot\epsilon_p)^2} \nonumber\\ & & ~~~~~~\times\prod_{k=1}^\infty\exp\left({{1\over 2} (\alpha\!\cdot\!\!\F''_k\!\!\cdot\! \alpha) \Lambda^{k\beta }}\right) \Lambda^{s \beta}\nonumber\\ &\equiv& \sum_{p=1}^\infty \Theta_{ij}^{(p)} \Lambda^{p\beta} ~. \label{expanthij} \end{eqnarray} Now, collecting all the pieces and inserting them back into (\ref{formula}), we find for ${\cal F}_k(Z)$ the following expression: \begin{equation} {\cal F}_k(Z) = - k^{-2} \sum_{p, q, l=0}^{p+q+l = k-1}\sum_{ij} {\cal H}_{2,i}^{(p)} {\cal H}_{2,j}^{(q)} \Theta_{ij}^{(k-p-q-l)} \Xi^{(l)} ~, \label{elresul} \end{equation} in terms of previously defined coefficients. If we look at the factors on the r.h.s. of Eq.(\ref{elresul}), it is easy to see that they involve ${\cal F}_1,{\cal F}_2,...$ up to ${\cal F}_{k-1}$. In fact, although both ${\cal H}_2^{(p)}$ and $\Theta^{(p)}$ depend on ${\cal F}_1,....{\cal F}_p$, the indices within parenthesis reach at most the value $k-1$ as $\Theta_{ij}^{(0)}=0$. Moreover $\Theta_{ij}^{(k)}$ depends on ${\cal F}_1,...,{\cal F}_{k-1}$ since the vector $\alpha=0$ is missing from the lattice sum. This ``lucky accident" has its origin in the particular form of the characteristic in the semiclassical (duality) frame, and seems to be an essential feature in order to build up a recursive procedure to {\em compute all the instanton coefficients by starting just from the perturbative contribution to} ${\cal F}(a)$ \cite{edemas}. For the first few cases we may develop (\ref{elresul}) to find \begin{eqnarray} {\cal F}_1 &=& - \sum_{ij} {\cal H}^{(0)}_{2,i} {\cal H}^{(0)}_{2,j} \Theta_{ij}^{(1)} \nonumber \\ {\cal F}_2 & = & - \frac{1}{4} \sum_{ij} \left(\Theta_{ij}^{(2)} {\cal H}_{2,i}^{(0)} {\cal H}_{2,j}^{(0)} + \Theta_{ij}^{(1)}(2 {\cal H}_{2,i}^{(1)}{\cal H}_{2,j}^{(0)} - {\cal H}_{2,i}^{(0)}{\cal H}_{2,j}^{(0)}\Theta^{(1)})\right) \nonumber\\ &\vdots& \nonumber \end{eqnarray} After some algebraic manipulations (\ref{elresul}) admits the following general form \begin{equation} {\cal F}_k=-\frac{1}{k^2} \sum_{s=1}^k\sum_{ \alpha\in\Delta_{s}} (-1)^{\rho\cdot\alpha} \prod_{\alpha_+} Z_{\alpha_+}^{ -\xi (\alpha\,\!\cdot\alpha_+)^2} \prod_{p}\left[R(e_p)\right]^{ (\alpha\,\!\cdot\epsilon_p)^2} \Phi_{k+1-s}(\alpha) ~, \label{ff} \end{equation} where the functions $\Phi_{k} (\alpha)$ depend on ${\cal F}_1\cdots, {\cal F}_{k-1}$ and have to be evaluated case by case. For the first few we have \begin{eqnarray} \Phi_1(\alpha) &= & Z_{\alpha(G)}^2 ~, \label{phi0} \\ \Phi_2({\cal F}_1,\alpha) & = & {\cal F}_1 + 2 (\alpha\!\cdot\!\!\F'_1 ) Z_{\alpha(G)} + {1\over 2} (\alpha\!\cdot\!\!\F''_1\!\!\cdot\! \alpha) Z_{\alpha(G)}^2 ~, \label{phi1} \\ \Phi_3({\cal F}_1,{\cal F}_2,\alpha) & = & 4{\cal F}_2 + 4 (\alpha\!\cdot\!\!\F'_2) Z_{\alpha(G)} + (\alpha\!\cdot\!\!\F'_1)^2 + {1\over 2} (\alpha\!\cdot\!\!\F''_1\!\!\cdot\! \alpha)\left( {\cal F}_1 + 2 (\alpha\!\cdot\!\!\F'_1) Z_{\alpha(G)} \right) \nonumber \\ & & + {1\over 8} (\alpha\!\cdot\!\!\F''_1\!\!\cdot\! \alpha)^2 Z_{\alpha(G)}^2 + {1\over 2} (\alpha\!\cdot\!\!\F''_2\!\!\cdot\! \alpha) \, Z_{\alpha(G)}^2 ~, \label{phi2} \end{eqnarray} where $\alpha\!\cdot\!\!{\cal F}'_k = n^i{\cal F}_{k,i}$. Expressions (\ref{ff})-(\ref{phi2}) make patent the iterative character of the procedure. ~$Z_{\alpha(G)}$ stands for $n^i {\cal H}_{2,i}^{(0)}$~, and for simply laced groups, $Z_{\alpha(G)}=Z_\alpha$ while for non-simply laced, the exact form will be given below. In the case $\beta =2$, as we can see from Eq.(\ref{ic}) the first instanton correction acquires a shift $\tilde{{\cal F}}_1={\cal F}_1+1~.$ It soon becomes clear that, except for the simplest cases, the concrete evaluation of the ${\cal F}_k$ has to be carried out by symbolic computation. In the next sections we illustrate our procedure with explicit examples in several cases for the lower rank groups. A last word concerning the possibility to split the masses is in order. Generically, all resulting expressions involve powers of the degenerated masses $\{m_f,~f=1,...,N_f/2\}$. They must be recovered from the exact result for arbitrary masses in the coincidence limit $m_f = m_{f+N_f}$. The possibility to go back unambiguously only happens for low powers of $m_f$. As a thumb rule, we have checked in several cases that the following prescription does the job: for $SU(N_c)$ and when powers of $m_f$ are not higher than 2, $m_f\to {1\over 2}(m_f + m_{f+N_f})$ and $m_f^2 \to m_f m_{f+N_f}$; while for the rest of the groups and for (even) powers of $m_f$ not higher than 4, $m_f^2\to {1\over 2}(m_f^2 + m^2_{f+N_f})$ and $m_f^4 \to m^2_f m^2_{f+N_f}$. Typically these cases only occur in ${\cal F}_1$. Another way to see this is to observe that if we write down ${\cal F}_1$ in terms of the $q_i(m_f),~ f=1,..N_f/2$ as given in (\ref{lasquses}), these factors always appear precisely in those combinations that build up the $t_k= t_k(q_i)$ as given in (\ref{lastes}), which are valid for arbitrary masses. \section{Results for Simply-Laced Lie Algebras} \setcounter{equation}{0} \indent We start by giving the concrete expression for $Z_{\alpha(G)} = n^i{\cal H}_{2,i}^{(0)}$. In the case of $A_r$ and $D_r$ Lie algebras, we have that $\sum_p e_p^2 = a^i a^j A_{ij} $ and also that $\xi^+=1$. Hence $$ Z_{\alpha(G)} = n^i{\cal H}_{2,i}^0 =n^i\sum_p e_p\partial_i e_p=n^i a^j A_{ij} = a\!\cdot\! \alpha = Z_\alpha $$ We shall define $\Delta_0=\prod_{\alpha_+}Z_{\alpha+}^2~. $ To avoid confusion we must mention that, although we have expressed all results in terms of Weyl invariant polynomial combinations $\bar u_k(a^i)$, for notational clearness we shall drop the bar. \subsection{{\boldmath $SU(r+1)$} with {\boldmath $N_f$} hypermultiplets} The only asymptotically free theories that we can consider within our approach, for these groups, are $N_f=2,\ldots,2r$. Let us list some of the results that we have obtained by using our formulas (we omit the case of $SU(N_c)$ whithout matter which can be found in \cite{emm}): \subsubsection{\boldmath $SU(2)$} \noindent\underline{$N_f=2$}~ Using $u_2=a^2$, we found the following corrections: \begin{eqnarray} {\cal F}_1 &=& \rule{0mm}{10mm}\frac{u_2+m^2}{2u_2} ~, \label{su2f1}\\ {\cal F}_2 &=& \rule{0mm}{10mm}\frac{u^2_2-6u_2 m^2 + 5 m^4}{64 u^3_2} ~, \label{su2f2} \\ {\cal F}_3 &=& \rule{0mm}{10mm}\frac{5u^2_2 m^2-14u_2 m^4+9m^6}{192u_2^5} ~. \label{su2f3} \end{eqnarray} The one- and two-instanton contributions coincide with those computed in \cite{dhoker} (after adjusting $\Lambda$ to $\Lambda/2$). In the one-instanton correction, as discussed before, it is possible to split the masses, $m^2 \to m_1m_2$, so that the result for non-degenerated $(ND)$ matter hypermultiplets is \begin{equation} {\cal F}_{1,ND} = \rule{0mm}{10mm}\frac{u_2+m_1m_2}{2u_2} ~, \label{su2non} \end{equation} in agreement with the result in Ref.\cite{dhoker}. Also, the case of one hypermultiplet can then be considered by letting $m_2\to\infty$ while keeping $\Lambda m_2$ finite an equal to the square of the dynamical scale $\Lambda^2$ that corresponds to $N_f=1$. \subsubsection{\boldmath $SU(3)$} To express our results, we introduce Weyl invariant combinations in terms of the $a_i$-variables, $u_2=a_1^2+a_2^2-a_1a_2$ and $u_3=a_1a_2 (a_1-a_2)$. For the case $N_f=4$, we will denote $q_1=m_1+m_2$ and $q_2=m_1m_2$. We obtained: ~ \noindent\underline{$N_f=2$} \begin{eqnarray} {\cal F}_1 &=&\rule{0mm}{10mm} {(2u_2^2+6m^2u_2-18mu_3)}/{\Delta_0} ~, \label{su3f1} \\ {\cal F}_2 &=&\rule{0mm}{10mm} (5u_2^6+153m^4u_2^4+162m^2u_2^5-1998m^3u_3u_2^3-414mu_3 u_2^4+1701m^4u_3^2u_2 \nonumber\\ &&+4374m^2u_3^2u_2^2+162u_3^2u_2^3+729u_3^4-2916m^3u_3^3-2673 mu_3^3u_2)/{\Delta_0^3} ~, \label{su3f2} \\ {\cal F}_3 &=&\rule{0mm}{10mm}(48u_2^{10}+12320m^6u_2^7+31792m^4u_2^8+4992m^2u_2^9 -366624m^5u_3u_2^6-12032mu_3u_2^8\nonumber\\ &&-253088m^3u_3u_2^7+478116m^6u_3^2u_2^4+2276856m^4u_3^2 u_2^5+529236m^2u_3^2u_2^6\nonumber\\ &&+5600u_3^2u_2^7-3684852m^5u_3^3u_2^3-4654800m^3u_3^3u_2^4 -394524mu_3^3u_2^5+994356m^6u_3^4u_2\nonumber\\ &&+7097544m^4u_3^4u_2^2+3969648m^2u_3^4u_2^3+105192u_3^4u_2^4 -1469664m^5u_3^5\nonumber\\ &&-4878468m^3u_3^5u_2-1571724mu_3^5u_2^2+1364688m^2u_3^6+ 215784u_3^6u_2)/{\Delta_0^{5}} ~. \label{su3f3} \end{eqnarray} Again, the splitting of the one-instanton correction can be done by letting $m^2 \to m_1m_2$ and $m \to (m_1+m_2)/2$, \begin{equation} {\cal F}_{1,ND} = \rule{0mm}{8mm}(2u_2^2+6m_1m_2u_2-9(m_1+m_2)u_3)/{\Delta_0} ~, \label{su3non1} \end{equation} and the reduction of hypermultiplets mentioned above is immediate. Our results (\ref{su3f1}), (\ref{su3f2}) and (\ref{su3non1}) agree with those obtained in \cite{dhoker}. ~ \noindent\underline{$N_f=4$} \begin{equation} {\cal F}_1 = \rule{0mm}{8mm}(2u_2^3-9u_3^2-6q_1u_2u_3+2(q_1^2+2q_2)u_2^2+6q_2^2u_2- 18q_1q_2u_3)/{\Delta_0} ~, \label{su3bf1} \end{equation} and the length of the expressions growths rapidly. Again, having in mind that ${\cal F}_{1,ND}$ should be linear in the polynomials of the masses it is possible to carefully split the masses, $m_1^2 \to m_1m_3$, $m_2^{\,2} \to m_2m_4$, $m_1 \to (m_1+m_3)/2$ and $m_2 \to (m_2+m_4)/2$ (or, in terms of the polynomials of the masses, $2q_1 \to t_1$, $q_1^2+2q_2 \to t_2$, $2q_1q_2 \to t_3$ and $q_2^2 \to t_4$) so that \begin{equation} {\cal F}_{1,ND} = \rule{0mm}{8mm}(2u_2^3-9u_3^2-3t_1u_2u_3+2t_2u_2^2-9t_3u_3 +6t_4u_2)/{\Delta_0} ~, \label{su3non2} \end{equation} and then one can reduce it to an odd number of matter hypermultiplets in the way mentioned before \cite{dhoker}. \subsubsection{\boldmath $SU(4)$} As the expressions become too long, we will only display the one-instanton correction. To express our results, we introduce the classical values of the Weyl invariant Casimirs $u_2=a_1^2+a_2^2+a_3^2-a_1a_2-a_2a_3$, $u_3=a_1a_2(a_1-a_2)+a_2a_3(a_2-a_3)$ and $u_4=a_1^2a_2a_3-a_1a_2^2a_3 -a_1^2a_3^2+a_1a_2a_3^2$. For the case $N_f=4$, we will denote $q_1=m_1+m_2$ and $q_2=m_1m_2$. \vskip2mm \noindent\underline{$N_f=2$} \begin{eqnarray} {\cal F}_1 &=& (8m^2u_2^3+6u_3^2u_2-8mu_3u_2^2-36u_3^2m^2+32m^2u_2u_4 \nonumber\\ && -16u_2^2u_4+96mu_3u_4-64u_4^2)/{\Delta_0} ~. \label{su4f1} \end{eqnarray} \noindent\underline{$N_f=4$} \begin{eqnarray} {\cal F}_1&=&(2u_3^2u_2^2-8u_4u_2^3+12u_3^2u_4-32u_4^2u_2-18q_1u_3^3 +64q_1u_4u_3u_2\nonumber\\ &&+6(q_1^2+2q_2)u_3^2u_2-16(q_1^2+2q_2)u_4u_2^2-64(q_1^2+2q_2) u_4^2-8q_1q_2u_3u_2^2\nonumber\\ &&+96q_1q_2u_4u_3+8q_2^2u_2^3-36q_2^2u_3^2+32q_2^2u_4u_2)/ {\Delta_0} ~. \label{su4bf1} \end{eqnarray} \subsection{{\boldmath $SO(2r)$} with {\boldmath $N_f$} hypermultiplets} The only asymptotically free theories that we can consider within our approach, for these groups, are $N_f=0,2,\ldots,2(r-2)$. Notice that the case $N_f=0$ corresponds to vanishing values of $\xi^+$ and $\xi^-$ in (\ref{elprep}), which in the formulas of the instanton corrections implies $R=1$. Let us list some of the result one can easily obtain by using our formulas: \subsubsection{\boldmath $SO(4)$} For this group the classical values of the Casimir operators in terms of the $a_i$ are given by $u_2=2a_1^2+2a_2^2$, $u_4=-(a_1^4+a_2^4-2a_1^2a_2^2)$, and we can only consider the pure case ($N_f=0$). ~ \noindent\underline{$N_f=0$} \begin{eqnarray} {\cal F}_1&=&\rule{0mm}{5mm}2^2u_2/{\Delta_0} ~, \\ {\cal F}_2&=&\rule{0mm}{10mm}2(5u_2^3-60u_2u_4)/{\Delta_0^3} ~, \\ {\cal F}_3&=&\rule{0mm}{10mm}2^5(3u_2^5-120u_2^3u_4+240u_4^2u_2)/{\Delta_0^5} ~. \end{eqnarray} \subsubsection{\boldmath $SO(6)$} For this group we have $u_2=2a_1^2+2a_2^2+2a_3^2-2a_1a_2-2a_1a_3$, $u_4=-(a_1^4+a_2^4+a_3^4-2a_1^3a_2-2a_1a_2^3-2a_1^3a_3-2a_1a_3^3+3a_1^2a_2^2 +3a_1^2a_3^2-2a_2^2a_3^2-2a_1^2a_2a_3+2a_1a_2^2a_3+2a_1a_2a_3^2)$ , $u_6=a_1^4a_2^2-2a_1^3a_2^3+a_1^2a_2^4-2a_1^4a_2a_3+2a_1^3a_2^2a_3+a_1^4a_3^2 +2a_1^3a_2a_3^2-2a_1^2a_2^2a_3^2-2a_1^3a_3^3+a_1^2a_3^4$, and we can consider the cases $N_f=0$ and $N_f=2$. ~ \noindent\underline{$N_f=0$} \begin{eqnarray} {\cal F}_1&=&\rule{0mm}{5mm}2^2(-u_2u_4-9u_6)/{\Delta_0} ~,\\ {\cal F}_2&=&\rule{0mm}{10mm}2(-5u_2^5u_4^3-43u_2^3u_4^4-172u_2u_4^5- 60u_2^6u_4u_6-647u_2^4u_4^2u_6-1701u_2^2u_4^3u_6+36u_2^5u_6^2 \nonumber\\ &&-1827u_2^3u_4u_6^2-3276u_4^4 u_6+1323u_2u_4^2u_6^2-14337u_2^2u_6^3-21627u_4u_6^3)/{\Delta_0^3} ~. \end{eqnarray} \noindent\underline{$N_f=2$} \begin{equation} {\cal F}_1=\rule{0mm}{5mm}2^2(-m^4u_2u_4-9m^4u_6-4m^2u_4^2+u_4^2u_2+12m^2u_2u_6 -4u_6u_2^2-3u_4u_6)/{\Delta_0} ~. \end{equation} Again, the splitting of the masses is possible for the one-instanton correction. This happens for all the classical groups. \subsubsection{\boldmath $SO(8)$} In this case the expressions of the Casimir operators in terms of the $a_i$ gets too long so we are not going to list them here. We have found the following expressions: \noindent\underline{$N_f=0$} \begin{equation} {\cal F}_1=2^2(9u_2^3u_8-u_2^2u_4u_6+3u_2u_6^2+32u_2u_4u_8+ 48u_6u_8-4u_4^2u_6)/{\Delta_0} ~. \end{equation} \noindent\underline{$N_f=2$} \begin{eqnarray} {\cal F}_1&=&2^2(9m^4u_2^3u_8-m^4u_2^2u_4u_6+3m^4u_2u_6^2+32m^4u_2 u_4u_8+48m^4u_6u_8-4m^4u_4^2u_6\nonumber\\ && -4m^2u_6^2u_2^2-12m^2u_4u_6^2-u_2u_4u_6^2-9u_6^3+12m^2u_4u_8u_2^2+32m^2u_8u_4^2 +4u_8u_4^2u_2\nonumber\\ &&+16m^2u_8u_6u_2-3u_8u_6u_2^2+32u_8u_6u_4+128m^2u_8^2-48u_8^2u_2)/{\Delta_0} ~. \end{eqnarray} The one-instanton corrections agree with those computed in Ref.\cite{dhoker}. \section{Results for non Simply-Laced Lie Algebras} \setcounter{equation}{0} \indent \subsection{{\boldmath $SO(2r+1)$} with {\boldmath $N_f$} hypermultiplets} In this case the form of $Z_{\alpha(G)}$ is different from the simply-laced cases. Indeed, using the fact that \[ \alpha_i\!\cdot\!\epsilon_q = \delta_{i,q}-\delta_{i+1,q} ~~~~~~~~~~~ \alpha_r\!\cdot\!\epsilon_q = 2\delta_{r,q} ~, \] and setting $\xi^+=1$, we have \begin{eqnarray} Z_{\alpha(G)} &=& n^i{\cal H}_{2,i}^{(0)} ~= ~n^i\sum_pe_p\partial_ie_p \nonumber\\& = &n^i\sum_pe_p(\alpha_i\!\cdot\epsilon_p) = \sum_{i=1}^{r-1}~\left(e_i-e_{i+1}\right)n^i+2 e_r n^r \nonumber\\ & = & Z_\alpha + Z_{\alpha_r} n^r ~. \label{zetagso} \end{eqnarray} The only asymptotically free theories that we can consider within our approach, for $SO(2r+1)$, are $N_f=0,2,\ldots,2r-2$. Notice that the case $N_f=0$ corresponds, as in $SO(2r)$, to take in (\ref{elprep}) $\xi^+=\xi^-=0$ which in the formulas of the instanton corrections means to set $R=1$. Let us list some of the results that we have obtained: \subsubsection{\boldmath $SO(5)$} We can consider within our approach the cases $N_f=0$ and $N_f=2$. For this group we have $u_2=2a_1^2+4a_2^2-4a_1a_2$, $u_4=-(a_1^4-4a_1^3a_2+4a_1^2a_2^2)$. We found: ~ \noindent\underline{$N_f=0$} \begin{eqnarray} {\cal F}_1 &=& \rule{0mm}{5mm}-2^3u_4/{\Delta_0} ~, \label{so5f1} \\ {\cal F}_2 &=&\rule{0mm}{10mm} 2(u_2^3u_4^2-76u_2u_4^3)/{\Delta_0^3} ~, \label{so5f2} \\ {\cal F}_3 &=& \rule{0mm}{10mm}2^{7}(3u_2^4u_4^4-232u_2^2u_4^5+176u_4^6)/{3\Delta_0^5} ~. \label{so5f3} \end{eqnarray} \noindent\underline{$N_f=2$} \begin{eqnarray} {\cal F}_1 &=&\rule{0mm}{6mm} 2^2 (-2m^4u_4+2m^2u_2u_4-u_2^2u_4-2u_4^2)/{\Delta_0} ~, \label{so5bf1} \\ {\cal F}_2&=&\rule{0mm}{10mm}2(m^8u_2^3u_4^2-76m^8u_2u_4^3+152m^6u_2^2u_4^3 -32m^6u_4^4 -78m^4u_2^3u_4^3+168m^4u_2u_4^4\nonumber\\ &&+12m^2u_2^4u_4^3-88m^2u_2^2u_4^4+96m^2u_4^5-u_2^5u_4^3 -60u_2u_4^5+u_2^3u_4^4)/{\Delta_0^3} ~, \label{so5bf2} \\ {\cal F}_3 &=& \rule{0mm}{10mm}2^6(6m^{12}u_2^4u_4^4-464m^{12}u_2^2u_4^5 +352m^{12}u_4^6-9m^{10}u_2^5u_4^4+1256m^{10}u_2^3u_4^5\nonumber\\ &&-1744m^{10}u_2u_4^6+3504m^8u_2^2u_4^6+3m^8u_2^6u_4^4-1212m^8u_2^4u_4^5 -960m^8u_4^7\nonumber\\ &&+2976m^6u_2u_4^7-3024m^6u_2^3u_4^6+498m^6u_2^5u_4^5-2864m^4u_2^2u_4^7 +1054m^4u_2^4u_4^6\nonumber\\ &&-86m^4u_2^6u_4^5+736m^4u_4^8-1104m^2u_2u_4^8+824m^2u_2^3u_4^7 -137m^2u_2^5u_4^6\nonumber\\ &&+5m^2u_2^7u_4^5+240u_2^2u_4^8-128u_4^9-56u_2^4u_4^7+5u_2^6u_4^6)/{3\Delta_0^5} ~. \end{eqnarray} \subsubsection{\boldmath $SO(7)$} Here we have $u_2=2a_1^2+2a_2^2+4a_3^2-2a_1a_2-4a_2a_3$~, $u_4= -(a_1^4+a_2^4-2a_1^3a_2-2a_1a_2^3-4a_2^3a_3+3a_1^2a_2^2+8a_1^2a_3^2+4a_2^2a_3^2 -8a_1^2a_2a_3+8a_1a_2^2a_3-8a_1a_2a_3^2)$, $u_6=a_1^4a_2^2+a_1^2a_2^4-2a_1^3a_2^3-4a_1^4a_2a_3+8a_1^3a_2^2a_3 -4a_1^2a_2^3a_3+4a_1^4a_3^2-8a_1^3a_2a_3^2+4a_1^2a_2^2a_3^2$~, and for $N_f=4$ we also denote $q_2=m_1^2+m_2^{2}$ and $q_4=m_1^2m_2^{\,2}$. We can consider the cases $N_f=0$, $N_f=2$ and $N_f=4$. ~ \noindent\underline{$N_f=0$} \begin{eqnarray} {\cal F}_1&=&\rule{0mm}{5mm}2^3(u_2^2u_6+3u_4u_6)/{\Delta_0} ~, \\ {\cal F}_2&=&\rule{0mm}{10mm}2\, u_6^2(-u_2^6u_4^3-12u_2^4u_4^4-48u_2^2u_4^5- 64u_4^6-76u_2^7u_4u_6\nonumber\\ &&-839u_2^5u_4^2u_6-3057u_2^3u_4^3u_6-3588u_2u_4^4u_6-44u_2^6u_6^2 -1695u_2^4u_4u_6^2\nonumber\\ &&-10827u_2^2u_4^2u_6^2-16308u_4^3u_6^2+1863u_2^3u_6^3-567u_2u_4u_6^3 -18225u_6^4)/{\Delta_0^3} ~. \end{eqnarray} \noindent\underline{$N_f=2$} \begin{equation} {\cal F}_1=2^3u_6(m^4u_2^2+3m^4u_4+m^2u_2u_4+9m^2u_6-3u_2u_6)/{\Delta_0} ~. \end{equation} \noindent\underline{$N_f=4$} \begin{eqnarray} {\cal F}_1&=&2^3u_6(q_4^2u_2^2+3q_4^2u_4-3(q_2^2+2q_4)u_2u_6 +(q_2^2+2q_4)u_4^2+q_2q_4u_2u_4+9q_2q_4u_6+3q_2u_4u_6\nonumber\\ &&+4q_2u_2^2u_6-q_2u_2u_4^2-45u_6^2-32u_2u_4u_6 -8u_2^3u_6+7u_4^3+2u_2^2u_4^2)/{\Delta_0} ~. \end{eqnarray} Again, the one-instanton corrections agree with previous results \cite{dhoker}. \subsection{{\boldmath $Sp(2r)$} with {\boldmath $N_f$} hypermultiplets} In this subsection we are going to consider the case of $Sp(2r)''$, {\it i.e.}, the case of $Sp(2r)$ with two massless hypermultiplets and $N_f'=N_f-2$ matter hypermultiplets. In this case the form of $Z_{\alpha(G)}$ is, as in the case of $SO(2r+1)$, different from the simply-laced cases. Now, using \[ \alpha_i\!\cdot\!\epsilon_q = \delta_{i,q}-\delta_{i+1,q} ~~~~~~~~~~~~~~~~ \alpha_r\!\cdot\!\epsilon_q = \delta_{r,q} ~, \] we see that \begin{eqnarray} Z_{\alpha(G)} &=& n^i{\cal H}_{2,i}^{(0)}\nonumber\\ & \equiv & 2n^i\sum_pe_p\partial_ie_p =2 n^i \sum_p e_p(\alpha_i\!\cdot\epsilon_p) = 2\left[\sum_{i=1}^{r-1}\left(e_i-e_{i+1}\right)n^i+ e_r n^r\right] \nonumber\\ & = & 2 Z_{\alpha} - Z_{\alpha_r}n^r ~. \label{zetagsp} \end{eqnarray} The only asymptotically free theories that we can consider within our approach, for $Sp(2r)''$, are $N_f'=0,2,\ldots,2r-2$. Notice that the case $N_f'=0$ now means to put $R(e_p)=e_p^4$ cause we are considering two massless hypermultiplets. Let us list some of the results that we have obtained: \subsubsection{\boldmath $Sp(4)$} For this group we can consider the cases $N_f'=0$ and $N_f'=2$. We have $u_2=2a_1^2+a_2^2-2a_1a_2$, $u_4= -(a_1^4-2a_1^3a_2+a_1^2a_2^2)$. ~ \noindent\underline{$N_f'=0$} \begin{eqnarray} {\cal F}_1&=\rule{0mm}{5mm}&u_2/{2\Delta_0} ~, \\ {\cal F}_2&=&\rule{0mm}{10mm}(5u_2^5+43u_2^3u_4+172u_2u_4^2)/4{\Delta_0^3} ~, \\ {\cal F}_3&=&\rule{0mm}{10mm}2^2(9u_2^9+143u_2^7u_4+927u_2^5u_4^2+2840u_2^3u_4^3 +5680u_2u_4^4)/3{\Delta_0^5} ~. \end{eqnarray} \noindent\underline{$N_f'=2$} \begin{eqnarray} {\cal F}_1&=&\rule{0mm}{5mm}(m^4u_2+4m^2u_4-u_2u_4)/2{\Delta_0} ~, \\ {\cal F}_2&=&\rule{0mm}{10mm}(5m^8u_2^5+43m^8u_2^3u_4+172m^8u_2u_4^2+12m^6u_2^4u_4 -24m^6u_2^2u_4^2+352m^6u_4^3+54m^4u_2^3u_4^2\nonumber\\ &&-264m^4u_2u_4^3+152m^2u_2^2u_4^3-32m^2u_4^4-5u_2^3u_4^3 +60u_2u_4^4)/4{\Delta_0^3} ~, \\ {\cal F}_3&=&\rule{0mm}{10mm}2^2(9m^{12}u_2^9+143m^{12}u_2^7u_4+927m^{12}u_2^5u_4^2 +2840m^{12}u_2^3u_4^3\nonumber\\ &&+5680m^{12}u_2u_4^4+28m^{10}u_2^8u_4+304m^{10}u_2^6u_4^2+1536m^{10}u_2^4u_4^3 -1408m^{10}u_2^2u_4^4\nonumber\\ &&+9728m^{10}u_4^5-11952m^8u_2u_4^5+6600m^8u_2^3u_4^4+75m^8u_2^7u_4^2 +537m^8u_2^5u_4^3\nonumber\\ &&+15872m^6u_2^2u_4^5-1200m^6u_2^4u_4^4+120m^6u_2^6u_4^3-1792m^6u_4^6 +7760m^4u_2u_4^6\nonumber\\ &&+321m^4u_2^5u_4^4-5416m^4u_2^3u_4^5-3584m^2u_2^2u_4^6+1280m^2u_4^7 +752m^2u_2^4u_4^5\nonumber\\ &&-9u_2^5u_4^5+360u_2^3u_4^6-720u_2u_4^7)/3{\Delta_0^5} ~. \end{eqnarray} \subsubsection{\boldmath $Sp(6)$} For this group we can consider the cases $N_f'=0$, $N_f'=2$ and $N_f'=4$. We let $u_2=2a_1^2+2a_2^2+a_3^2-2a_1a_2-2a_2a_3$, $u_4= -(a_1^4+a_2^4-2a_1^3a_2-2a_2^3a_3-2a_1a_2^3+3a_1^2a_2^2+2a_1^2a_3^2+a_2^2a_3^2 -4a_1^2a_2a_3+4a_1a_2^2a_3-2a_1a_2a_3^2)$ , $u_6=a_1^4a_2^2-2a_1^3a_2^3+a_1^2a_2^4-2a_1^4a_2a_3+4a_1^3a_2^2a_3 -2a_1^2a_2^3a_3+a_1^4a_3^2-2a_1^3a_2a_3^2+a_1^2a_2^2a_3^2$. For $N_f'=4$, and we also have $q_2=m_1^2+m_2^{\,2}$ and $q_4=m_1^2m_2^{2}$. ~ \noindent\underline{$N_f'=0$} \begin{eqnarray} {\cal F}_1&=&\rule{0mm}{5mm}2(3u_2u_6-u_2^2u_4-4u_4^2)/{\Delta_0} ~, \\ {\cal F}_2&=&\rule{0mm}{10mm}2^4(-5u_2^6u_4^5-60u_2^4u_4^6-240u_2^2u_4^7-320u_4^8 +43u_2^7u_4^3u_6+546u_2^5u_4^4u_6+2304u_2^3u_4^5u_6\nonumber\\ &&+3232u_2u_4^6u_6-172u_2^8u_4u_6^2-2255u_2^6u_4^2u_6^2 -9715u_2^4u_4^3u_6^2-13272u_2^2u_4^4u_6^2+2064u_4^5u_6^2\nonumber\\ &&+180u_2^7u_6^3-1107u_2^5u_4u_6^3-18975u_2^3u_4^2u_6^3-46908u_2u_4^3u_6^3 +2439u_2^4u_6^4\nonumber\\ &&-3240u_2^2u_4u_6^4-57672u_4^2u_6^4+19197u_2u_6^5)/{\Delta_0^3} ~. \end{eqnarray} \noindent\underline{$N_f'=2$} \begin{eqnarray} {\cal F}_1=\rule{0mm}{7mm}2(3m^4u_2u_6-m^4u_2^2u_4-4m^4u_4^2-4m^2u_2^2u_6 -12m^2u_4u_6-u_2u_4u_6-9u_6^2)/{\Delta_0} ~. \end{eqnarray} \noindent\underline{$N_f'=4$} \begin{eqnarray} {\cal F}_1&=&\rule{0mm}{7mm}2(3q_4^2u_2u_6-q_4^2u_2^2u_4-4q_4^2u_4^2-4q_2q_4u_2^2u_6 -12q_2q_4u_4u_6-(q_2^2+2q_4) u_2u_4u_6\nonumber\\ &&-9(q_2^2+2q_4)u_6^2+12q_2u_2u_6^2-4q_2u_4^2u_6-3u_4u_6^2-4u_2^2u_6^2 +u_2u_4^2u_6)/{\Delta_0} ~. \end{eqnarray} \subsection{The case of pure {\boldmath $Sp(2r)$}} As we discussed above, the case of $Sp(2r)$ without matter hypermultiplets cannot be obtained from our previous formulas, as long as we are considering at least two massless hypermultiplets. Nevertheless, one can treat this case separately in an analogous way. In fact, we can fix our ansatz for the effective prepotential (\ref{elprep}) to the one first considered by Ito and Sasakura \cite{itosak} by setting \begin{equation} \xi=1~~~~~~~~~\xi^+=\xi^-=0 ~~~~~~~~\mbox{and}~~~~~~~~~~ \tau_0 = 3 ~. \label{settings} \end{equation} Now, we can introduce the effective prepotential into Eq.(\ref{formula}) and the same kind of formulas for the instanton correction would be obtained, provided we have for this case a characteristic $E=\left[{0,\dots,0\atop {1\over 2},\dots,{1\over 2}}\right]$. Note that, being $N_f=0$, we must set $R=1$ in our formulas. We also need the value of $Z_{\alpha(G)}$ which turns out to be the same as that in $Sp(2r)''$, \ie \ $Z_{\alpha(G)}=2Z_\alpha-n_rZ_{\alpha_r}$. \subsubsection{\boldmath $Sp(4)$} For this group we have, as we saw before, $u_2=2a_1^2+a_2^2-2a_1a_2$, $u_4= -(a_1^4-2a_1^3a_2+a_1^2a_2^2)$. In terms of them the first instanton corrections are \begin{eqnarray} {\cal F}_1&=&\rule{0mm}{5mm}2^3(u_2^2+4u_4)/{\Delta_0} ~, \\ {\cal F}_2&=&\rule{0mm}{10mm}2^6(5u_2^7+59u_2^5u_4+232u_2^3u_4^2 +304u_2u_4^3)/{\Delta_0^3} ~, \\ {\cal F}_3&=&\rule{0mm}{10mm}2^{14}(9u_2^{12}+184u_2^{10}u_4+1526u_2^8u_4^2 +6496u_2^6u_4^3+14656u_2^4u_4^4\nonumber\\ &&+15872u_2^2u_4^5+5632u_4^6)/{3\Delta_0^5} ~. \end{eqnarray} \subsubsection{\boldmath $Sp(6)$} For this group we have $u_2=2a_1^2+2a_2^2+a_3^2-2a_1a_2-2a_2a_3$, $u_4= -(a_1^4+a_2^4-2a_1^3a_2-2a_2^3a_3-2a_1a_2^3+3a_1^2a_2^2+2a_1^2a_3^2+a_2^2a_3^2 -4a_1^2a_2a_3+4a_1a_2^2a_3-2a_1a_2a_3^2)$ , $u_6=a_1^4a_2^2-2a_1^3a_2^3+a_1^2a_2^4-2a_1^4a_2a_3+4a_1^3a_2^2a_3 -2a_1^2a_2^3a_3+a_1^4a_3^2-2a_1^3a_2a_3^2+a_1^2a_2^2a_3^2$, and the first instanton corrections are \begin{eqnarray} {\cal F}_1&=&\rule{0mm}{5mm}2^5(u_2^2u_4^2+4u_4^3-4u_2^3u_6-18u_2u_4u_6 -27u_6^2)/{\Delta_0} ~, \\ {\cal F}_2&=&\rule{0mm}{10mm} 2^{12}(-5u_2^6u_4^7-60u_2^4u_4^8-240u_2^2u_4^9-320u_4^{10}+59u_2^7u_4^5u_6 +738u_2^5u_4^6u_6\nonumber\\ &&+3072u_2^3u_4^7u_6+4256u_2u_4^8u_6-232u_2^8u_4^3u_6^2-3021u_2^6u_4^4u_6^2 -12699u_2^4u_4^5u_6^2\nonumber\\ &&-15736u_2^2u_4^6u_6^2+6288u_4^7u_6^2+304u_2^9u_4u_6^3+4120u_2^7u_4^2u_6^3 +15518u_2^5u_4^3u_6^3+1716u_2^3u_4^4u_6^3\nonumber\\ &&-55728u_2u_4^5u_6^3-16u_2^8u_6^4+5928u_2^6u_4u_6^4+54486u_2^4u_4^2u_6^4 +113373u_2^2u_4^3u_6^4-216u_2^5u_6^5\nonumber\\ &&-41148u_4^4u_6^4+39447u_2^3u_4u_6^5+182250u_2u_4^2u_6^5-729u_2^2u_6^6 +89667u_4u_6^6)/{\Delta_0^3} ~. \end{eqnarray} The one-instanton corrections agree with those computed in Ref.\cite{itosak}. \section{Concluding Remarks} \setcounter{equation}{0} \indent In the present paper, we have shown how instanton corrections to the effective prepotential of ${\cal N}=2$ supersymmetric theories can be computed in a variety of cases including all classical gauge groups and even number of degenerated fundamental matter hypermultiplets, {\em up to arbitrary order}. As compared to other approaches developed in the literature, we should stress that the one presented in this paper has an important feature in that it does not require an explicit knowledge of the BPS spectrum as a function of the moduli, at the same time that it allows to consider a huge variety of cases within a unified framework. Also, being recursive, it admits an easy implementation on a computer. We have illustrated the remarkable simplicity of our procedure by displaying many explicit expressions which should be quite useful for further comparison with the results obtained by other means. Conversely, our results admit a second reading: They could be thought of as a highly non-trivial test of the connection between the Seiberg--Witten solution to the low energy dynamics of ${\cal N}=2$ supersymmetric gauge theories, and the theory of Whitham (adiabatic) deformations of a given integrable system \cite{emm,ITEP}. In this sense, it is important to remind that the {\em new} equation (\ref{formula}), which is a key ingredient of our procedure, is originated in the latter framework, as it is shown in detail in Ref.\cite{egrmm}. Aside from being an interesting mathematical problem by itself, the embedding of the Seiberg-Witten solution within a Whitham hierarchy seems to be the appropriate framework for the study of many physical phenomena. For example, the so-called {\em slow Whitham times} can be consistently thought of as spurion vector supermultiplets that can be used to break ${\cal N}=2$ supersymmetry down to ${\cal N}=0$ with non-quadratic Casimir operators \cite{emm}. In this way, the Whitham hierarchy can be interpreted as a family of supersymmetry breaking deformations of the original theory associated with the higher Casimir operators of the gauge group. This issue generalizes to the ${\cal N}=0$ case the family of ${\cal N}=1$ supersymmetry breaking terms considered, for instance, in Ref.\cite{ad}. The key feature of the Whitham formalism lies on the fact that, as the dependence on the slow times is encoded in the prepotential, it is possible to obtain the {\em exact} effective potential of the theory, in the spirit of \cite{emm,soft}. This allows to perform a detailed study, both qualitative and quantitative, of the vacuum state of the theory once supersymmetry is broken, as well as of the appearance of monopole condensates, mass gaps, etc. Preliminary results on this program were published in Ref.\cite{emm}. The formalism is even useful near the Argyres--Douglas singularities, where non-local degrees of freedom become simultaneously massless, provided one approaches them along any of the submanifolds where a unique monopole gets massless \cite{emm}. There is another place where deformations of the prepotential by means of the Whitham times are relevant. It is the study of contact terms in the twisted version of ${\cal N}=2$ gauge theories, where these new variables play the r\^ole of sources for insertions of certain class of operators in the generating functional \cite{marimoore,losev,wm,takasaki}. Another interesting point is given by the uses of our starting equations (\ref{lderm})--(\ref{formula}) to study the strong coupling expansion of the prepotential near the singularities of the quantum moduli space, as it was done in Ref.\cite{edemas} for the case of pure $SU(N_c)$. In particular, these equations provide us with a set of non-trivial constraints that facilitate the study of the couplings between different {\em magnetic photons}, originally found in Ref.\cite{ds}, that take place at such points. The expansion of the prepotential near the maximal points by other methods, as the deformations of the auxiliary singular Riemann manifold \cite{dHP}, is not sensitive to such kind of terms. Several interesting questions remain open aside from the ones just mentioned. For example, the case of arbitrary masses cannot be treated within our approach, except for the one-instanton correction (which, on the other hand, is enough for leading order comparison purposes). In fact, from the Whitham hierarchy side, one can show that indeed the formulas used in this paper are insufficient to tackle the generic scenario, though it seems to be possible to refine the formalism in order to extend its applicability to some cases of unpaired masses \cite{egrmm}. The additional corrections that appear in the generic case are, nevertheless, quite difficult to manage with. Finally, another avenue for future research is, certainly, the connection of this formalism with the string theory and D-brane approach to supersymmetric gauge theories, where some steps has already been given in the last few years \cite{gorsky}. We believe that these matters deserve further study. \section*{Acknowledgements} We are pleased to thank Marcos Mari\~no for helpful comments and a careful reading of the manuscript. The work of J.D.E. is supported by a fellowship of the Ministry of Education and Culture of Spain. The work of J.M. was partially supported by DGCIYT under contract PB96-0960.
\section{Introduction} The possibility that massive neutron stars might be driven unstable to collapse to black holes when placed in a close binary orbit was first suggested by Wilson, Mathews and Maronetti (hereafter WMM; ~\cite{WWM}) on the basis of their approximate relativistic numerical simulations. This finding was quite unexpected, partly because it disagreed with earlier Newtonian calculations ~\cite{LRS} which showed that tidal fields stabilize binary stars against radial collapse. In fact, none of the follow-up post-Newtonian (PN)~\cite{LAI2} ~\cite{LOM} or approximate analytic analyses ~\cite{Kip} indicate the presence of any relativistic radial instability in fluid binaries, nor does an independent dynamical simulation ~\cite{Shib}. No evidence is found for the WMM ``crushing effect'' in the relativistic numerical calculations of either corotational ~\cite{Baum} or irrotational ~\cite{Bon} binaries in quasi-equilibrium circular orbits. These later calculations are particularly careful to adopt the same simplifications as WMM (e.g, a conformally flat 3-metric). All of the calculations except those of WMM suggest that the maximum allowed rest mass of a fluid star in a binary is in fact slightly larger than the value in isolation. While it would appear that the WMM effect may not be present for fluid stars, the work of WMM raises the interesting question as to whether the ``crushing instability'' might exist for a different type of binary system. Specifically, are there {\it any} relativistic binary systems for which tidal fields can trigger the collapse of a compact object known to be stable in isolation? To address this question, Shapiro ~\cite{Us} offered a simple candidate compact object consisting of a test particle orbiting a Schwarzschild black hole outside the innermost-stable circular orbit (ISCO). Such a ``compact object'' is obviously stable in isolation, but when it is placed in a binary orbit about a distant mass, an integration of the test-particle equations of motion reveals that the tidal field of the distant mass can cause a test particle to plunge inside the hole. Shapiro's model is very promising as a simple illustration of the crushing instability in action, but the original analysis, as he emphasized, was highly simplified and somewhat heuristic. First, the study was confined to two special test-particle orbits, one coplanar with and the other perpendicular to the companion orbital plane. Moreover, the motion of the particle in the perpendicular case was artificially constrained to remain in a plane, precluding any precessional motion or wobbling. Most important, the analysis was based on a post-Newtonian treatment of the 3-body problem in which the tidal piece of the equation for the relative motion of the test particle about the black hole was treated to lowest (Newtonian) order and the nontidal piece replaced by the exact, fully relativistic expression for geodesic motion in Schwarzschild geometry. While such a hybrid approach, similar in spirit to one proposed by Kidder, Will and Wiseman~\cite{LEK} for the 2-body problem, should capture the essential dynamics, it is far from rigorous. In particular, it is not clear {\it a priori} how reliable it is to use a Newtonian tidal term when the test particle is moving at high velocity in a very strong gravitational field close to a black hole. In this paper we improve and extend the simple model presented in ~\cite{Us} of a compact object subject to binary-induced collapse. We use the equations of Regge and Wheeler ~\cite{RW} and Zerilli ~\cite{Z} to derive the perturbation to the spacetime close to a Schwarzschild black hole due to a distant mass. We note in passing that the resulting perturbed spacetime provides another example supporting the recent conjecture of Bekenstein ~\cite{Bek} that the horizon area of a perturbed black hole is an adiabatic invariant. We obtain the geodesic equations of motion in the perturbed spacetime and solve them to track the dynamical evolution of a dense spherical swarm of 20,000 test particles placed around the hole outside the ISCO. The `black hole $+$ swarm' constitutes a compact collisionless cluster whose stability against tidally-induced collapse we determine. We compare our refined treatment with the simpler version presented in ~\cite{Us} and show that the original equations track the behavior reasonably well. We confirm that the crushing instability can occur in a binary system containing simple compact collisionless clusters like the ones we construct. \section{HYBRID -- PN TREATMENT} \subsection{Basic Equations} Consider first a system of three bodies. Assume that two of the bodies, 1 and 2, say, are much closer to each other than they are to the third, so that the influence of body 3 on the relative orbit of 1 and 2 may be treated as a small tidal perturbation. Specialize to the case in which body 1 is a Schwarzschild black hole of mass $M$ and body 2 is a test particle ($m_t \ll M $). Since it is far away from the 1-2 pair (the ``compact object''), body 3 can be treated as a point particle, with mass $m$. Let {\bf r} be the coordinate position of the test particle relative to the hole and {\bf R} be the coordinate position of the distant point mass relative to the hole. Following Shapiro ~\cite{Us}, we may write the equation for the relative motion of the test particle about the hole as \begin{equation} \label{six} \ddot{\bf r} =\ -\ \left({M \over r^2}\right)\ [A{\bf e_r} + B{\bf v}] + {m \over R^3} \ \left[ {3{\bf R} \cdot {\bf r} \over R^2} \ \ {\bf R}\ \ -\ \ {\bf r} \right]\, . \end{equation} where ${\bf e_r} = {\partial\over\partial r}$. In the Newtonian limit, $A=1$ and $B=0$ in Eq.~(\ref{six}). For isolated binaries, Lincoln and Will~\cite{CWL} derive post-Newtonian expressions for $A$ and $B$ for arbitrary masses, correct through 2.5PN order. Kidder, Will and Wiseman~\cite{LEK} provide a ``hybrid'' set of equations in which the sum of the terms in $A$ and $B$ that are independent of the ratio $\eta = \mu /(M +m_t), \mu = M m_t/(M +m_t) ,$ is replaced by the exact expression for geodesic motion in the Schwarzschild geometry around a body of mass $M$, while the terms dependent on $\eta$ are left unaffected. Their resulting equation of motion is therefore exact in the test-body limit $(\eta \rightarrow 0)$ and is valid to 2.5PN order when appropriately expanded for arbitrary masses. We shall utilize these same hybrid expressions for $A$ and $B$ and, for simplicity, work in the test-body limit by taking one member of our close pair to have a mass much smaller than the other. Adopting harmonic (de Donder) coordinates, the resulting (Schwarzschild) expressions for $A$ and $B$ are given by \begin{equation} \label{twelve} A \ =\ \left[{1-M /r\over (1\ +\ M /r)^3} \right]\ \ -\ \ \left[ {2-M /r \over 1-(M /r)^2}\right]\ {M \over r}\ \dot{r}^2\ +\ v^2 ,\end{equation} \begin{equation} \label{thirteen} B\ =\ -\ \left[{4-2M /r \over 1-(M /r)^2}\right]\ \dot{r}\, , \end{equation} where \begin{equation} \label{fourteen} v^2\ =\ \dot{r}^2\ +\ r^2 (\dot{\theta}^2 + \sin^2\theta\dot{\phi}^2). \end{equation} To model the three-body system, we must also know the position of the companion (body 3) relative to the compact object. The leading Newtonian piece of the equation of motion for {\bf R} gives \begin{equation} \label{five} \ddot {\bf R}\ =\ -{M+m \over R^3}{\bf R} \, . \end{equation} We are interested in the dynamical behavior of the close pair, regarded as a single ``compact object'', as it inspirals toward the distant mass $m$ due to gravitational radiation emission. To treat the inspiral of this ``binary'' ($m$ in orbit about the ``compact object''), we must include radiation reaction terms in the lowest order (Newtonian) orbit Eq.~(\ref{five}). Formally, such a treatment requires a consistent expansion up to 2.5PN order. In lieu of this, we shall analyze the inspiral by assuming that the binary is in a nearly circular, Keplerian orbit, which undergoes a slow inspiral due to gravitational radiation loss in the quadrupole limit. This assumption is equivalent to inserting a quadrupole radiation reaction potential in the binary orbit Eq.~(\ref{five}) and neglecting the lower-order, (non-dissipative) PN corrections in that particular equation. While such an expression is not formally consistent to 2.5PN order, it faithfully tracks the secular inspiral of the binary in the limit treated here in which the binary system is at wide (nonrelativistic) separation. The details of the inspiral are not important here, only that the inspiral serves to bring a tidal perturber slowly in from infinity toward our compact object~\cite{OUR}. The resulting equations for the binary inspiral are then ~\cite{Us} \begin{equation} \label{fifteen} R(t)/R(0) = (1 -t/T)^{1/4} \end{equation} and \begin{equation} \label{sixteen} \theta(t) -\theta(0) ={8 \over 5}\left({ M + m \over R(0)^3} \right)^{1/2}T [1 - (1-t/T)^{5/8}] \, , \end{equation} (Note the typo in ~\cite{Us}.) The binary inspiral timescale $T$ is given by \begin{equation} \label{seventeen} T/M = {5 \over 256} {(R(0)/M )^4 \over {((m+M)/M)(m/M)}} \, . \end{equation} We will use Eqs.~(\ref{fifteen})--(\ref{seventeen}) in analyzing the relative orbit Eq.~(\ref{six}). \subsection{Numerical Implementation} To assess the fate of the compact object, we followed the motion of a spherical swarm of 20,000 test particles about the black hole. At $t=0$, the test particles were placed randomly about the hole at a radius $r/M=5.9$, well outside the ISCO of an isolated black hole at $r/M = 5$, and set in circular orbits of arbitrary orientation. We set $m = M$ and started the simulation with the companion at $R/M = 60$. At this initial separation, the tidal field of the companion is negligible at the compact object, and the test particles begin in stable equilibrium orbits. We integrate (\ref{six}) until the time at which $R/M = 26.30$. Cartesian coordinates were used throughout. The results of the simulation are summarized in Figure 1, where we plot the mean cluster radius as a function of companion separation. Clearly, the compact object remains stable until the companion comes within $R/M \approx 30$, at which point the tidal field causes many of the particles to plunge into the hole. We find that 17,919 particles (89.6 \%) have fallen into the black hole by $t/M = 121,893$, the time at which the companion reaches $R/M = 26.30$. The simulation confirms the existence of a crushing instability. \begin{figure} \epsfxsize=3in \begin{center} \leavevmode \epsffile{fig1a.ps} \end{center} \epsfxsize=3in \begin{center} \leavevmode \epsffile{fig1b.ps} \end{center} \caption{Evolution of the cluster in the hybrid-PN approximation. The mean radius of the 20,000 test-particles in the initially spherical swarm is plotted as a function of the tidal companion radius in the top figure and as a function of coordinate time in the bottom figure.} \end{figure} Several checks were performed to ensure the accuracy of our code, which integrates the ordinary differential orbit equations by a standard fourth-order Runge Kutta algorithm with an adaptive stepsize ~\cite{NR}. For example, in the absence of the companion, the code conserves particle energy and angular momentum over the full time period of integration and reliably locates the Schwarzschild ISCO with a controllable precision. These tests are not so trivial when the integrations are performed in Cartesian coordinates. \section{Schwarzschild Perturbation Treatment} \subsection{Basic Equations} \begin{figure} \epsfxsize=1.5in \begin{center} \leavevmode \epsffile{fig2.ps} \end{center} \caption{The adopted coordinate system. The black hole $M$ is at the origin and the companion $m$ is on the $z$ axis.} \end{figure} We will now improve the previous treatment by incorporating the tidal field of the companion star in a fully relativistic fashion. To do this, we treat the effect of the companion as a small perturbation on a Schwarzschild background. By assuming that the companion orbits at a much greater distance from the black hole than the test particles, and hence moves with a much smaller angular velocity, we can work in the quasi-static approximation. We place the companion on the $z$ axis of our spherical polar coordinate system centered on the black hole (see Figure 2). We follow Regge and Wheeler~\cite{RW} and divide the linear metric perturbations into independent ``even'' and ``odd'' components according to \begin{equation} \label{2one} g_{\mu\nu}\ =\ g_{\mu\nu}(S) + h_{\mu\nu}(\mbox{odd}) + h_{\mu\nu}(\mbox{even}) \, , \end{equation} where $g_{\mu\nu}(S)$ is the usual Schwarzschild metric. In the appropriate gauge, Regge and Wheeler found that the odd perturbations may be written as \begin{equation} \label{2two} \begin{array}{l} \mathflushleft{h_{\mu\nu}(\mbox{odd})\ =\ \pmatrix{ 0 & 0 & 0 & h_0(r) \cr 0 & 0 & 0 & h_1(r) \cr 0 & 0 & 0 & 0 \cr h_0(r) & h_1(r) & 0 & 0 \cr}\times}\cr \mathstrutter{15pt} \mathflushright{\sin\theta {\partial \over \partial\theta}Y_{L0}(\theta)\times e^{(i\omega t)} \, .}\cr \end{array} \end{equation} Likewise, the even perturbations may be written, after simplifying gauge transformations, as \begin{equation} \label{2three} \begin{array}{l} \mathflushleft{h_{\mu\nu}(\mbox{even}) =}\\ \mathstrutter{35pt}\mathcenter{\pmatrix{\scriptstyle H_0(r)(1-2M/r) &\scriptstyle H_1(r) &\scriptstyle 0 &\scriptstyle 0 \cr \scriptstyle H_1(r) & \mathstrutter{12pt}{\scriptstyle H_2(r) \over (1-2M/r)} & \scriptstyle 0 & \scriptstyle 0 \cr \scriptstyle 0 & \scriptstyle 0 & \scriptstyle r^2 K(r) & \scriptstyle 0 \cr \scriptstyle 0 & \scriptstyle 0 & \scriptstyle 0 & \scriptstyle r^2 \sin^2\theta K(r) \cr }\times}\\ \mathstrutter{15pt}\mathflushright{Y_{L0}(\theta)\times e^{(i\omega t)} \, .}\\ \end{array} \end{equation} We now proceed to solve Einstein's equations to first order in the perturbation functions as in Zerilli ~\cite{Z}, using a delta function (point) source for the companion. We first find solutions to the homogeneous equations away from the companion and then use the inhomogeneous source terms to match these solutions. Setting $\omega = 0$ to comply with our quasi-static approximation, Zerilli's equations reduce to \begin{equation} \label{2four} h_1 = H_1 = 0; \ \ \ \ \ H_0\ =\ H_2 \equiv H \, \end{equation} \begin{equation} \label{2five} \begin{array}{l} \displaystyle \mathflushleft{{d^2 h_0 \over dr^2} + \left[{4M \over r^2}-{L(L+1) \over r}\right]{h_0 \over r-2M} = }\\ \mathstrutter{15pt}\mathflushright{-8\pi r\left[{1 \over 2}L(L+1)\right]^{-1/2}Q_{L}^{(0)}\, ,}\\ \end{array} \end{equation} \begin{equation} \label{2six} \begin{array}{l} \mathflushleft{\bigl(1 - {2M \over r}\bigr)^2 {d^2 K \over dr^2} + \bigl(1-{2M \over r}\bigr)\bigl(3- {5M \over r}\bigr){1 \over r}{dK \over dr}}\\ \mathstrutter{15pt}\mathonefourth{- \bigl(1-{2M \over r}\bigr)^2\ {1 \over r}\ {dH \over dr} - \bigl(1-{2M \over r}\bigr){1 \over r^2}(H-K)}\\ \mathstrutter{15pt}\mathflushright{- \bigl(1-{2M \over r}\bigr) {1 \over 2r^2}L(L+1)(H+K) = 8\pi A_L^{(0)} \, ,}\\ \end{array} \end{equation} \begin{equation} \label{2seven} \mathcenter{-{1-M/r \over r-2M}{dK \over dr}+{1 \over r}{dH \over dr} + {2-L(L+1) \over 2r(r-2M)}(H-K) = 8\pi A_L \, .} \end{equation} In the above equations, $Q_L$, $A_L$, and $A_L^{(0)}$ represent components of the stress-energy tensor. For a point source companion moving along a Schwarzschild geodesic, these terms are given in Appendix E of Zerilli ~\cite{Z}. From Zerilli's tabulation of the source terms, one sees that all the source terms go to zero in the static limit except for $A_L^{(0)}$, i.e. $T_{00}$. Homogeneous solutions may be found analytically for both the even and odd perturbations. Regge and Wheeler show that, in the large $r$ limit (where the companion resides), there is one solution for $h_0$ varying like $r^{-L}$, and another varying like $r^{L+1}$. Since there is no source term for $h_0$ ($Q_L = 0$), the only way that both $h_0$ and its first derivative can be continuous and regular everywhere is to require $h_0 = 0$. Therefore, there is no contribution from odd parity perturbations. The static solutions for $H$ and $K$ are found both by Regge and Wheeler~\cite{RW} and by Zerilli~\cite{Z} in terms of hypergeometric functions. We are interested in the lowest order nontrivial contribution, the quadrupole piece, generated by the companion, so we restrict our attention to the $L=2$ perturbation. Inside the orbit of the companion, the solution which is regular at the horizon is \begin{equation} \label{2eight} \begin{array}{ll} \displaystyle H \ =\ \kappa_1 r(r-2M) & \approx\kappa_1 r^2 \\ \displaystyle K \ =\ \kappa_1 (r^2-2M^2) & \approx\kappa_1 r^2 \, , \end{array} \end{equation} where the last equality in each equation above holds at large $r \gg M$ and $\kappa_1$ is a constant to be determined by matching at the companion. These equations may be verified by direct substitution into Eq.~(\ref{2six}) or (\ref{2seven}). Outside the orbit of the companion, the perturbation is given by the solution regular at infinity, \begin{equation} \label{2nine} \begin{array}{ll} H\ &=\ \kappa_2 \Bigl[{2M (2M^3+4M^2r-9M r^2+3r^3) \over (2M -r)r} \\ & \hskip90pt +{3r^2(r-2M )^2\ln(1-2M /r) \over (2M -r)r}\Bigr] \\ & \approx \kappa_2 r^{-3} \\ & \\ K\ &=\ \kappa_2 M^2\bigl[-6+{{4M}\over{r}}-{{6r}\over{M}}-{{3r^2\ln(1-2M /r)}\over{M^2}} \\ & \mathstrutter{15pt} \hskip 120pt + 6\ln(1-2M /r)\bigr] \\ & \approx \kappa_2 r^{-3} \, , \end{array} \end{equation} where the last equality in each equation again holds at large $r \gg M$ and $\kappa_2$ is another constant to be determined by matching. Now we determine the two constants by matching the solutions at the radius of orbit of the companion, $R$. Because $A_L^{(0)}$, the source term of (\ref{2six}), does not go to zero in the quasi-static case, $\kappa_1$ and $\kappa_2$ do not vanish, as was the case for the odd parity solutions. Taking $R \gg M$ as required by the quasi-static approximation, we use the asymptotic expressions in equations (\ref{2eight}) and (\ref{2nine}) for the matching. Requiring that $K$ be continuous across $R$ then yields our first condition \begin{equation} \label{2ten} \kappa_1 R^2 = \kappa_2 R^{-3} ~~\Rightarrow~~ \kappa_2 = R^5 \kappa_1 \, . \end{equation} The second condition may be found in either of two ways. First, one may integrate Equation (\ref{2six}) across the source to obtain a jump condition relating difference in the first derivative of $K$ on either side of $r=R$ to the source strength. Alternatively, one may compute $\kappa_1$ directly by taking the Newtonian limit of $H$ and matching it to the Newtonian tidal potential. Either way, one obtains $\kappa_1 = {{m}\over{R^3}} 4\sqrt{\pi/5}$. Thus, $h_{00}$, for example, is \begin{equation} \label{2thirteen} \begin{array}{l} \mathflushleft{h_{00} = (1-2M /r)(r^2-2M r)\kappa_1Y_{20}(\theta)}\\ \mathstrutter{15pt}\mathflushright{= (1-2M /r)(r^2 - 2M r){m \over R^3}(3\cos^2\theta - 1)\, .}\\ \end{array} \end{equation} From (\ref{2thirteen}), one can immediately verify that the metric derived above has the correct Newtonian limit---the potential reduces to the classic Hill potential. This solution for the even parity solutions was first discovered by Moeckel~\cite{Mo}. The entire metric may now be written out, defining $P \equiv P_2(\cos\theta) = 3\cos^2\theta - 1$: \begin{equation} \label{2fourteen} \begin{array}{l} \mathflushleft{ds^2 = \Bigl[-\bigl(1-{2M \over r}\bigr)+{m \over R^3}P(r-2M )^2\Bigr]dt^2}\\ \mathstrutter{15pt}\mathcenter{+ \Bigl[\bigl(1-{2M \over r}\bigr)^{-1}+{m \over R^3}Pr^2\Bigr]dr^2}\\ \mathstrutter{15pt}\mathflushright{+ \Bigl[r^2 + {m \over R^3}Pr^2(r^2-2M^2)\Bigr]d\Omega^2 \, .} \end{array} \end{equation} We note that the perturbed spacetime (\ref{2fourteen}) furnishes another example supporting the conjecture of Bekenstein ~\cite{Bek} that the horizon area of a near-equilibrium black hole is an adiabatic invariant (see Appendix). To compare with the hybrid-PN treatment, we transform to harmonic coordinates by replacing the areal radial coordinate, $r$, with the harmonic radius, $r_h = r - M$. Dropping the subscript, we obtain in harmonic coordinates \begin{equation} \label{2fifteen} \begin{array}{l} \mathflushleft{ds^2 \ = \displaystyle -{r-M \over r+M }dt^2 + {r+M \over r-M }dr^2 + (r+M)^2d\Omega^2}\\ \maththreefourths{\mathstrutter{15pt} \displaystyle + {m \over R^3}P\bigl[(r-M)^2dt^2 + (r+M)^2dr^2} \\ \mathflushright{\displaystyle + (r+M )^2(r^2+2rM -M^2)d\Omega^2\bigr] \, .}\\ \end{array} \end{equation} Obtaining the geodesic equations in this spacetime up to first order in the tidal expansion parameter ${m r^2 \over R^3}$, we find the equations of motion for a test particle near the black hole to be \begin{equation} \label{2nineteen} \begin{array}{l} \displaystyle \mathflushleft{\ddot r = {3M \dot r^2 \over r^2-M^2}-{M (r-M ) \over (r+M )^3} + (r-M )\Omega^2}\\ \mathonefourth{\mathstrutter{15pt} + {m \over R^3} P \Big[-3r\dot r^2+(r+2M ){(r-M )^2 \over (r+M )^2}}\\ \mathcenter{\mathstrutter{15pt} + (r-M)(r^2+4rM+M^2)\Omega^2\Big] }\\ \mathflushright{\mathstrutter{15pt} - {m \over R^3} 2(r^2-M^2){dP \over d\theta} \dot\theta\dot r}\\ \\ \mathflushleft{\ddot\theta = {4M-2r \over r^2-M^2}\dot\theta\dot r + \cos\theta\sin\theta \dot\phi^2}\\ \mathonefourth{\mathstrutter{15pt} + {m \over R^3} P [-2(M+2r)\dot\theta\dot r]}\\ \mathcenter{\mathstrutter{15pt} + {m \over R^3} {dP \over d\theta} {1 \over 2} \Big[-(3r^2+2rM-3M^2)\dot\theta^2}\\ \mathflushright{\mathstrutter{15pt} + (r^2+2rM-M^2)\sin^2\theta \ \dot\phi^2 + {(r-M)^2 \over (r+M)^2} + \dot r^2\Big]}\\ \\ \mathflushleft{\ddot\phi \ =\ {4M-2r \over r^2-M^2} \dot\phi\dot r - 2 \cot\theta \ \dot\theta\dot\phi - 2{m \over R^3} P(M+2r)\dot\phi\dot r}\\ \mathflushright{\mathstrutter{15pt} -2{m \over R^3} {dP \over d\theta} (r^2+rM-M^2)\dot\theta\dot\phi \, ,}\\ \end{array} \end{equation} where dot denotes differentiation with respect to Schwarzschild coordinate time, and $\Omega^2 = \dot\theta^2 + \sin^2\theta\dot\phi^2$. To avoid coordinate singularities, it is convenient to integrate (\ref{2nineteen}) in Cartesian coordinates. To facilitate an otherwise tedious transformation, we first rewrite the equations of motion in 3-vector form. We make the identification \begin{equation} \label{defacc} \begin{array}{lll} {\bf \ddot r} = & &( \ddot r-r\dot\phi^2\sin^2\theta-r\dot\theta^2){\partial\over\partial r}\\ \mathstrutter{15pt} &+ &( \ddot\theta+(2 \dot r\dot\theta-r\dot\phi^2\sin\theta \cos\theta)/r ){\partial \over \partial\theta}\\ \mathstrutter{15pt} &+ &( \ddot\phi +(2 \dot r\dot\phi \sin\theta+2r\dot\theta\dot\phi \cos\theta)/(r\sin\theta) ){\partial\over\partial\phi} \, ,\\ \end{array} \end{equation} a vector which we have constructed to be the same as the Newtonian 3-acceleration in spherical coordinates and we know transforms to $(\ddot x,\ddot y,\ddot z)$ in Cartesian coordinates. Identifying the velocity 3-vector ${\bf v} = (\dot r,\dot\theta,\dot\phi)$ in the spherical coordinate basis and setting ${\bf r} = r{\partial\over\partial r}$ allows us to write (\ref{2nineteen}) in the compact form 3-vector form \begin{equation} \label{rvector} \begin{array}{l} \displaystyle \mathflushleft{{\bf \ddot r} = -{M \over r^2} [A{\bf e_r} + B{\bf v}]} \\ \displaystyle \mathflushright{\mathstrutter{15pt} + {m \over R^3 } \left[\alpha \left({ 3{\bf r}\cdot {\bf R} \over R^2 }{\bf R} - {\bf r}\right) + \beta {\bf v} + \gamma {\bf e_r} \right] \, ,}\\ \end{array} \end{equation} where $A$ and $B$ are again given by Eqs.~(\ref{twelve}) and ~(\ref{thirteen}). The quantities $\alpha$, $\beta$, and $\gamma$ are given by \begin{equation} \label{2twentyone} \begin{array}{l} \displaystyle \mathflushleft{\alpha = (r^2+2rM-M^2){v^2-\dot r^2 \over r^2} + \dot r^2 + {(r-M)^2 \over (r+M)^2}} \\ \\ \displaystyle \mathflushleft{\beta = -2(3\cos^2\Theta-1)(M+2r)\dot r} \\ \displaystyle \mathflushright{\mathstrutter{15pt} + 12\cos\Theta\sin\Theta(r^2+rM-M^2)\dot\Theta}\\ \\ \displaystyle \mathflushleft{\gamma = (3\cos^2\Theta-1)\Biggl[2M\bigl(\dot r^2+{(r-M)^2 \over (r+M)^2} \bigr)}\\ \displaystyle \maththreefourths{\mathstrutter{15pt} + M(r^2-2rM-M^2) {v^2-\dot r^2 \over r^2} \Biggr]}\\ \displaystyle \mathflushright{\mathstrutter{15pt} - 12rM \dot r\dot\Theta\cos\Theta\sin\Theta \, ,}\\ \end{array} \end{equation} where $\Theta$ is the angle between $\bf r$ and ${\bf R}$. It is now straightforward to write out the Cartesian components of Eq.~(\ref{rvector}). To incorporate the (slow) inspiral of the companion in the context of our quasi-static approximation, we treat ${\bf R} = {\bf R(t)}$ in Eq. (\ref{rvector}) as a parameter, which is slowly evolved in accord with Eqs. (\ref{fifteen}), (\ref{sixteen}) and (\ref{seventeen}). The leading terms independent of the companion $m$ are identical in both (\ref{rvector}) and (\ref{six}), a result which is consistent with the adoption of hybrid equations in the PN analysis. The tidal term in (\ref{rvector}) reduces to the lowest-order Newtonian expression used in (\ref{six}) in the weak-field, slow-velocity limit where $ v \ll 1$, $ M/r \ll 1$. Since the test particles orbit close to the hole, the relativistic tidal expression will indeed cause departures from the motion predicted by the Newtonian tidal term. \subsection{Numerical Implementation} A spherical swarm identical to the one evolved with the hybrid-PN equations of motion was evolved with Eq.(\ref{rvector}). Because collapse occurs later in this simulation, the swarm was evolved until $R/M=26.30$. The results are summarized in Figures 3 -- 5. In Figure 3, the mean cluster radius is computed once again and compared to the hybrid-PN result. The qualitative nature of the evolution is the same, and the existence of a crushing instability is evident. The main effect of the new relativistic terms is to delay the collapse until the perturber gets somewhat closer to the hole. In particular, of 20,000 particles, 16,324 fell into the hole (81.6 \%) by the time the companion reached $R/M = 26.30$, a slightly smaller percentage than in the hybrid-PN simulation. Once again, the compact cluster is observed to be stable in isolation, but driven to collapse by the presence of a sufficiently strong tidal perturbation. \begin{figure} \epsfxsize=3in \begin{center} \leavevmode \epsffile{fig3a.ps} \end{center} \epsfxsize=3in \begin{center} \leavevmode \epsffile{fig3b.ps} \end{center} \caption{Evolution of the cluster in the Schwarzschild perturbation treatment. The mean radius of the 20,000 test-particles in the initially spherical swarm is plotted as a function of the tidal companion radius in the top figure and as a function of coordinate time in the bottom figure. In the top figure we compare the results found for our two treatments; the behavior is qualitatively similar, but the crushing effect occurs slightly later in the fully relativistic, Schwarzschild perturbation treatment.} \end{figure} \onecolumn \begin{figure}[p] \begin{center} \leavevmode \hbox{\vbox{ \hbox{\hbox to 3in{\hfil \Huge \bf A \hfil} \hbox to 1in{} \hbox to 3in{\hfil \Huge \bf B \hfil}} \hbox{\vrule width 0pt height 5pt} \hbox{\fbox{\epsfxsize=3in\epsffile{fig4a.ps} \hbox to 1in{} \epsfxsize=3in\epsffile{fig4b.ps}} \hskip-7in \vbox to 2.25in{\vskip5pt \hbox to 7in{\hfil \LARGE $t/M = 0$ \hfil} \vfil} } \hbox{\fbox{\epsfxsize=3in\epsffile{fig4c.ps} \hbox to 1in{} \epsfxsize=3in\epsffile{fig4d.ps}} \hskip-7in \vbox to 2.25in{\vskip5pt \hbox to 7in{\hfil \LARGE $t/M = 1.18\times10^5$ \hfil} \vfil} } \hbox{\fbox{\epsfxsize=3in\epsffile{fig4e.ps} \hbox to 1in{} \epsfxsize=3in\epsffile{fig4f.ps}} \hskip-7in \vbox to 2.25in{\vskip5pt \hbox to 7in{\hfil \LARGE $t/M = 1.22\times10^5$ \hfil} \vfil} } \hbox{\vrule width 0pt height 5pt} }} \end{center} \caption{Snapshots of the cluster swarm at selected times during the binary inspiral. In $A$ we show the view looking down along the $z$-axis, which is perpendicular to the orbital plane of the companion; in $B$ we show the view looking along the $x$-axis, which lies in the orbital plane of the companion. For the top frame the companion is at $R/M = 60$; for the middle frame the companion is at $R/M = 30.7$; for the bottom frame the companion is at $R/M = 26.30$.} \end{figure} \twocolumn \begin{figure} \begin{center} \leavevmode \hbox to 246pt{\hskip40pt \hbox{\vbox{ \hbox{\fbox{\epsfxsize=2in\epsffile{fig5a.ps}} \hskip-2in \vbox to 1.5in{\vskip5pt \hbox{\LARGE a} \vfil}} \hbox{\fbox{\epsfxsize=2in\epsffile{fig5b.ps}} \hskip-2in \vbox to 1.5in{\vskip5pt \hbox{\LARGE b} \vfil}} \hbox{\fbox{\epsfxsize=2in\epsffile{fig5c.ps}} \hskip-2in \vbox to 1.5in{\vskip5pt \hbox{\LARGE c} \vfil}} \hbox{\vrule width 0pt height 5pt} }} \hfil} \end{center} \caption{ Test-particle trajectories about the black hole during the inspiral of the binary companion from $R/M=60$ to $R/M=26.30$. Frame (a) shows the orbit of a particle moving in the orbital plane of the companion; it is not captured by the time the integrations terminate. Frame (b) shows the capture of a particle initially orbiting in a plane perpendicular to the companion plane. Frame (c) shows the precession and capture of a particle initially orbiting at an angle of $45^o$ to the companion plane.} \end{figure} As the geodesic equations in the perturbative relativistic treatment are derived from a self-consistent Lagrangian, they satisfy strict conservation laws even in the presence of a companion, assuming it is stationary. These conservation laws provide a means of testing our code and particle integration scheme. For example, if the companion is fixed at an arbitrary position on the z-axis, the perturbed Schwarzschild spacetime (\ref{2fifteen}) admits two Killing vectors, $\partial\over\partial t $ and $\partial\over\partial \phi$, yielding conservation of particle energy $p_t$ and angular momentum $p_{\phi}$, even for large tidal fields. Given that we linearize the tidal field and retain only the lowest order terms in $m$ in our equations of motion (\ref{rvector}), energy and angular momentum conservation are no longer exact. However, we have tested our code and have shown that it reliably obeys these conservation laws to the required order. The evolution of the swarm is depicted in Figure 4, where snapshots of the cluster are shown at three different times from two different viewing angles. By the end of the simulation, the original spherical swarm is reduced to a sparse cylindrical band of particles whose axis is perpendicular to the orbital plane of the companion. Apparently, particle orbits at small inclination angles to this plane are more stable than those which are perpendicular, a result already noted in ~\cite{Us}. This feature is evident in Figure 5, which plots the trajectories of three representative particles. The simulation confirms that gravitational collapse can be induced in a collisionless cluster by the tidal field of a binary companion.
\section{Introduction} Classical continuum Heisenberg ferromagnet models (CCHFM) exhibit a rich variety of nonlinear behaviour. In particular, over the past two decades, several integrable nonlinear ferromagnetic models have been idenfied [1-3]. So, some number integrable spin systems in 2+1 dimensions are found [4-8]. The underlying nonlinear spin excitations in such spin systems represented by solitons, domain walls, vortecies, lumps and dromions [4-11]. Theirs study are of considerable intrinsic interest, especially from the points of view of both mathematics and physics. Integrable spin systems in 2+1 dimensions as their (1+1)-dimensional counterparts, display fascinating geometrical aspects: they are gauge and L-equivalent to the nonlinear Schr${\ddot o}$dinger - type equations (the Davey-Stewartson equation, the Zakharov equations (ZE), the Strachan equation and so on) [5, 9-10, 12]. Generally speaking, between spin systems and the differential geometry take places the deep connection [12, 13-17, 20]. An important feature of two dimensional spin systems is the existence the topological invariant $$ Q = \frac{1}{4\pi} \int\int dxdy {\bf S}\dot ({\bf S}_{x} \wedge {\bf S}_{y}) \eqno (1) $$ such that the solutions of them are classified by the topological charge (1). In (9) ${\bf S} = (S_{1}, S_{2}, S_{3})$ is the three-dimensional unit vector (the spin vector). The class of exact solutions of (2+1)-dimensional spin systems is a very rich one. The solitons, vortices, dromions, lumps are among them. These solutions have the important physical significance. So, for example, vortecies play an active role in the dynamics and termodynamics of quasi-two-dimensional magnets [18-20]. The present paper is devoted to the study of the following CCHFM in two space and one time dimensions $$ {\bf S}_t + {\bf S}\wedge \{(b+1) {\bf S}_{\xi \xi} -b{\bf S}_{\eta \eta}\} + bu_{\eta} {\bf S}_{\eta} + (b+1)u_{\xi}{\bf S}_{\xi} = 0 \eqno(2a) $$ $$ u_{\xi \eta} = {\bf S}({\bf S}_{\xi}\wedge {\bf S}_{\eta}) \eqno(2b) $$ with $$ \xi = \frac{x}{2} + \frac{a+1}{\alpha}y, \quad \eta = -\frac{x}{2} - \frac{a}{\alpha}y. \eqno(3) $$ where $a, b $ are real constants, $\alpha^{2} = \pm 1$. It is the Myrzakulov XX (M-XX) equation [5]. Hereafter, for convenience, we use the conditional notations, e.g. equation (2) we denote by the M-XX equation. Equation (2) is integrable. We will distinguish the two integrable cases: the M-XXA equation as $\alpha^{2} = 1$ and the M-XXB equation as $\alpha^{2} = -1$. Also, equation (2) contains several integrable cases: \\ (i) $b = 0$, yields the M-VIII equation $$ {\bf S}_t + {\bf S}\wedge {\bf S}_{\xi \xi} + w{\bf S}_{\xi} = 0 \eqno(4a) $$ $$ w_{ \eta} = {\bf S}({\bf S}_{\xi}\wedge {\bf S}_{\eta}) \eqno(4b) $$ where $w=u_{\xi}$. (ii) $a = b = - \frac{1}{2}$, yields the celebrated Ishimori equation [4] $$ {\bf S}_t + \frac{1}{2}{\bf S}\wedge \{ {\bf S}_{\xi \xi} + {\bf S}_{\eta \eta}\} -\frac{1}{2}u_{\eta} {\bf S}_{\eta} + \frac{1}{2}u_{\xi}{\bf S}_{\xi} = 0 \eqno(5a) $$ $$ u_{\xi \eta} = {\bf S}({\bf S}_{\xi}\wedge {\bf S}_{\eta}) \eqno(5b) $$ where $\xi = \frac{1}{2}(x + \frac{1}{\alpha}y), \quad \eta = -\frac{1}{2}(x -\frac{1}{\alpha}y).$ The Ishimori equation (5) is the first integrable spin system in plan, which can be solved by the inverse scattering method (IST). This equation were studied by the different authors from variety points of view (see, e.g. [4,6-7,9,11,21]). (iii) Let now we put $b = 0, \eta =t$, then equation (2) reduces to the (1+1)-dimensional M-XXXIV equation $$ {\bf S}_t + {\bf S}\wedge {\bf S}_{\xi \xi} + w{\bf S}_{\xi} = 0 \eqno(6a) $$ $$ w_{ t} + \frac{1}{2} ({\bf S}^{2}_{\xi})_{\xi} = 0 \eqno(6b) $$ This integrable equation was introduced in [5] to describe nonlinear dynamics of compressible magnets. Equation (2) is the (2+1)-dimensional integrable generalization of the (1+1)-dimensional CCHFM or the isotropic Landau-Lifshitz equation (LLE) $$ {\bf S}_t = {\bf S} \wedge {\bf S}_{xx} \eqno(7) $$ and in 1+1 dimensions reduces to it. Here, it should be mentioned that the M-XX equation (2) is not the only integrable generalization of the LLE (7) in 2+1 dimensions. There exist several another integrable generalizations, e.g the following one, $$ {\bf S}_t = ( {\bf S} \wedge {\bf S}_{y} + u {\bf S})_x \eqno(8a) $$ $$ u_{x} = - {\bf S}({\bf S}_x \wedge {\bf S}_y) \eqno(8b) $$ This equation, which is known as the Myrzakulov I (M-I) equation, is again completely integrable [5,10,12]. As integrable, equation (2) can be solved by the IST method. The applicability of the IST method to the M-XX equation (2) is based on its equivalence to the compatibility condition of the following linear equations [5] $$ \Phi_{Z^{+}} = S\Phi_{Z^{-}} \eqno(9a) $$ $$ \Phi_{t} = 2i[S+(2b+1)I]\Phi_{Z^{-}Z^{-}} + W\Phi_{Z^{-}} \eqno(9b) $$ where $Z^{\pm} = \xi \pm \eta$ and $$ W = 2i\{(2b+1)(F^{+} + F^{-} S) +(F^{+}S + F^{-}) + (2b+1)SS_{Z^{-}}+\frac{1}{2}S_{Z^{-}} + \frac{1}{2} SS_{Z^{+}} \}, \quad $$ $$ S= \pmatrix{ S_3 & rS^- \cr rS^+ & -S_3 },\quad S^{\pm}=S_{1}\pm iS_{2} \quad S^2 = EI,\quad E = \pm 1, \quad r^{2}=\pm 1, $$ $$ F^{+} = 2iu_{Z^{-}}, \quad F^{-}=2iu_{Z^{+}}. $$ In fact, from the condition $\Phi_{Z^{+}t} = \Phi_{tZ^{+}}$, we get the equation $$ iS_t + \frac{1}{2}[S,(b+1) S_{\xi \xi} -bS_{\eta \eta}] + ibu_{\eta} S_{\eta} + i(b+1)u_{\xi}S_{\xi} = 0 \eqno(10a) $$ $$ u_{\xi \eta} = \frac{1}{4i}tr(S[S_{\xi},S_{\eta}]) \eqno(10b) $$ which is the matrix form of equation (2). \section{Bilinearization} It could be of interest to study the equation (2) by the IST method. But to look for the some special solutions, it is convenient use the other may be more practical method - the Hirota bilinear method. Remaining the use of the IST method in future, in this paper, we work with the Hirota method. To this purpose, we construct the bilinear form of (2) as $E=r=1$. Let us we now introduce the following transformation for the components of spin vector ${\bf S}$ and for the derivatives of scalar potential $u$ $$ S^{+} = \frac{2\bar f g}{\bar f f + \bar g g}, \quad S_3 = \frac{\bar f f - \bar g g}{\bar f f + \bar g g}, \eqno (11a) $$ $$ u_{\xi} =- 2i\frac{D_{\xi}(\bar f\circ f + \bar g \circ g)}{\bar f f + \bar g g}, \quad u_{\eta}= 2i \frac{D_{\eta}(\bar f \circ f + \bar g \circ g)}{\bar f f + \bar g g} \eqno (11b) $$ Here the Hirota operators $D_{x}, D_{y}$ and $D_{t}$ are defined by $$ D^{l}_{\xi}D^{m}_{\eta}D^{n}_{t} f(\xi, \eta, t)\circ g(\xi, \eta, t) = (\partial_{\xi}-\partial_{\xi^{\prime}})^{l} (\partial_{\eta}-\partial_{\eta^{\prime}})^{m} (\partial_{t}-\partial_{t^{\prime}})^{n} f(\xi, \eta, t)\circ g(\xi^{\prime}, \eta^{\prime}, t^{\prime})\mid_{\xi=\xi^{\prime}, \eta = \eta^{\prime}, t=t^{\prime}} . \eqno(12) $$ Substituting the formulae (11) into the M-XX equation (2), we obtain the bilinear equations $$ [iD_{t} - (b+1)D_{\xi}^{2} +bD_{\eta}^{2}](\bar f \circ g) = 0 \eqno (13a) $$ $$ [iD_{t} - (b+1)D_{\xi}^{2} +bD_{\eta}^{2}](\bar f \circ f - \bar g \circ g) = 0 \eqno (13b) $$ $$ \{D_{\xi} D_{\eta}+D_{\eta} D_{\xi}\} (\bar f f + \bar g g) \circ (\bar f f + \bar g g) = 0 \eqno (13c) $$ Note that equation (13c) coincide with the compatibility condition $ u_{\xi\eta} = u_{\eta\xi} $. Equations (13) is the desired Hirota bilinear form of equation (2). \section{Solutions} Now we can construct some special solutions of equation (2). As examples, we find simplest soliton, domain wall and vortex solutions. \subsection{Soliton solution} \begin{center} {\bf FIND THE SOLITON SOLUTIONS.} \end{center} \subsection{Domain wall solution} In order to obtain a domain wall solutions, we make the choice $$ f=1. \eqno (14) $$ Then, equations (13a,b) reduce to $$ ig_{t}+(b+1)g_{\xi\xi} - b g_{\eta\eta} = 0 \eqno(15a) $$ $$ (b+1)\bar g_{\xi}g_{\xi} - b\bar g_{\eta}g_{\eta} = 0 \eqno(15b) $$ Let us we consider the case, when $\alpha^{2} = -1$, i.e. the M-XXB equation. In this case, equation (13c) is identically satisfied by any analytical function $g = g(\xi, \eta, t)$. For example, the simplest non-trivial solution of equation (2) is $$ g = \exp \chi_{1} \eqno (16) $$ where $$ \chi_{1} = m_{1}\xi + n_{1}\eta + i [(b+1)m^{2}-bn^{2}]t + \chi_{10} = \chi_{1R}+i\chi_{1I} \eqno (17) $$ Thus, the spin components and the potential field are given by $$ S^{+} = e^{i\chi_{1I}} sech \chi_{1R},\quad S_{3} = - th\chi_{1R},\quad u = 2\ln (1+e^{2\chi_{1R}}). \eqno (18) $$ \subsection{ Vortex solution}. To construct vortex solution, we use the equation (13) and assume that its solution has the form $$ f=f(\xi, t), \quad g = g(\xi, t) \eqno(19) $$ Then the condition (13c) is satisfied automatically. At the same time, equations (13a,b) are satisfy if $$ if_{t}+ (b+1) f_{\xi\xi} = 0 \quad ig_{t} + (b+1) g_{\xi \xi} = 0 \eqno(20) $$ Hence, we obtain the following multi-vortex solutions of the M-XXB equation (2) $$ g_{N} = \sum_{j=0}^{N}\sum_{m+2n=j}\frac{a_{j}}{m!n!}(\frac{2}{b+1})^{\frac{m}{2}}\xi^{m}(2it)^{n} \eqno(21a) $$ $$ f_{N} = \sum_{j=0}^{N-1}\sum_{m+2n=j}\frac{b_{j}}{m!n!}(\frac{2}{b+1})^{\frac{m}{2}}\xi^{m}(2it)^{n} \eqno(21b) $$ where $a_{j}$ and $b_{j}$ are arbitrary complex constants, $m,n$ are the non-negative integer numbers. In particular, the 1-vortex solution isgiven by $$ f=b_{0}, \quad g=a^{\prime}_{1}\xi + a_{0} \eqno (22) $$ where $a^{\prime}_{1}=a_{1}(\frac{2}{b+1})^{\frac{1}{2}}$. The corresponding solution of equation (2) is given by $$ S^{+} = \frac{2b_{0}(a^{\prime}_{1}\xi + a_{0})}{\mid b_{0}\mid^{2} + \mid a^{\prime}_{1}\xi + a_{0}\mid^{2}} \eqno (23a) $$ $$ S_{3} = \frac{ \mid b_{0}\mid^{2} - \mid a^{\prime}_{1}\xi + a_{0}\mid^{2}} {\mid b_{0}\mid^{2} + \mid a^{\prime}_{1}\xi + a_{0}\mid^{2}} \eqno (23b) $$ $$ u = 2\ln (\mid b_{0}\mid^{2} + \mid a^{\prime}_{1}\xi + a_{0}\mid^{2}) \eqno (23c) $$ So, the 1-vortex solution is static. To find the dynamic solution, we consider the 2-vortex solution, which has the form $$ f=b^{\prime}_{0}\xi + b_{0}, \quad g=\frac{a_{2}}{b+1}\xi^{2} + \frac{a_{2}}{2} 2it+ a^{\prime}_{1}\xi + a_{0}, \quad b^{\prime}_{1}=b_{1}(\frac{2}{b+1})^{\frac{1}{2}} \eqno (24) $$ The interesting question is the dynamics of these vortices. Let us rewrite the solution (21) in the following factorized form $$ f(\xi, t) = b_{0} \prod^{N}_{j=1}[\xi - p_{j}(t)] \eqno(25a) $$ $$ g(\xi, t) = a_{0} \prod^{N}_{j=1}[\xi - q_{j}(t)] \eqno(25b) $$ where $p_{j}$ and $q_{j}$ denote the positions of the zeros for $f$ and $g$, and $a_{0}, b_{0}$ are constants. Substituting (25) into (20), we get the evolutions of $p_{j}$ and $q_{j}$ as $$ p_{jt} = -i(b+1)\sum^{N}_{k\not= j}\frac{1}{p_{j}-p_{k}} \eqno(26a) $$ $$ q_{jt} = -i(b+1)\sum^{N}_{k\not= j}\frac{1}{q_{j}-q_{k}} \eqno(26b) $$ where $j,k=1,2,...,N$. Hence, we get the Calogero-Moser type system $$ p_{jtt} = 2(b+1)^{2}\sum^{N}_{k\not= j}\frac{1}{(p_{j}-p_{k})^{3}} \eqno(27a) $$ $$ q_{jtt} = 2(b+1)^{2}\sum^{N}_{k\not= j}\frac{1}{(q_{j}-q_{k})^{3}} \eqno(27b) $$ with the following Hamiltonian $$ H=\frac{1}{2} \sum (p_{jt}^{2}+q^{2}_{jt}) + (b+1)^{2}\sum [ (p_{j}-p_{k})^{-2} +(q_{j}-q_{k})^{-2}]. \eqno(28) $$ \subsection{ Dromion solution} In this subsection we would like get the dromion [23] solution of equation (2), but please \begin{center} {\bf FIND THE DROMION SOLUTIONS.} \end{center} \section{Gauge equivalent equation} Finally, let us we present the gauge equivalent counterpart of equation (2). It has the form $$ iq_t+ (1 + b)q_{\xi \xi } - b q_{\eta \eta } + vq = 0 \eqno(29a) $$ $$ ip_t - (1 + b)p_{\xi \xi } + b p_{\eta \eta } - vp = 0 \eqno(29b) $$ $$ v_{\xi \eta } = -2\{(1+ b) (pq)_{\xi \xi} - b(pq)_{\eta \eta}\} \eqno(29c) $$ where $p,q$ are some complex functions. Equation (29) is related with the Zakharov equations [22]. To prove gauge equivalence between equations (2) and (29), let us perform the gauge transformation $\Psi = g \Phi$, where the function $\Phi$ is the solution of equations (9) and $g$ is a 2x2 matrix such that $$ S = g^{-1}\sigma_{3}g \eqno (30) $$ and $$ g_{Z^{+}}g^{-1} - \sigma_{3} g_{Z^{-}}g^{-1} = \left ( \begin{array}{cc} 0 & q \\ p & 0 \end{array} \right) \eqno(31) $$ Under this transformation the function $\Psi$ satisfies the following set of linear equations $$ \Psi_{Z^{+}} = \sigma_{3} \Psi_{Z^{-}} + B_{0}\Psi \eqno(32a) $$ $$ \Psi_{t} = 4i C_{2} \Psi_{Z^{-}Z^{-}} + 2 C_{1} \Psi_{Z^{-}} + C_{0}\Psi. \eqno(32b) $$ where $B_{0},\quad C_{j}$ are given by $$ B_{0}= \pmatrix{ 0 & q \cr p & 0 }, \quad C_{2}= \pmatrix{ b+1 & 0 \cr 0 & b },\quad C_{1}= \pmatrix{ 0 & iq \cr ip & 0 },\quad C_{0}= \pmatrix{ c_{11} & c_{12} \cr c_{21} & c_{22} } $$ $$ c_{12}=i[(4b+3)q_{Z^{-}}+q_{Z^{+}}] \quad c_{21}=-i[(4b+1)p_{Z^{-}}+ p_{Z^{+}}] $$ and $v=i(c_{22}-c_{11})$. Here $c_{jj}$ are the solution of the following equations $$ c_{11Z^{-}}- c_{11Z^{+}} = i[(4b+3)(pq)_{Z^{-}} + (pq)_{Z^{+}}] $$ $$ c_{22Z^{-}}+ c_{22Z^{+}} = i[(4b+1)(pq)_{Z^{-}} + (pq)_{Z^{+}}]. $$ The compatibility condition of equations (32) gives the equation (29). Therefore the M-XX equation (2) and the equation (29) are gauge equivalent to each other. Now let us proceed to the M-XX equation (10). It is not difficult to check that if $g$ obeys equations (31) then the $S$ in the form (30) obeys the M-XX equation (10) with $$ u = -2i\alpha \ln \det g \eqno(33) $$ \section{Conclusion} To conclude, in this paper we have found some exact solutions, namely domain wall and vortex solutions of the (2+1)-dimensional CCHFM - the M-XX equation. We have shown that the dynamics of vortices are governed by the Calogero-Moser type systems. Also we have presented the gauge equivalent counterpart of this equation. \section{Particular open problems} Finally, we also would like to pose the following particular problems: \\ {\bf Problem N1:} Find solutions of the M-XX equation by the $\bar \partial$-dressing method. \\ {\bf Problem N2:} Find solutions of the M-XX equation using the nonlocal Riemann-Hilbert problems method. \\ {\bf Problem N3:} Find solutions of the M-XX equation by means of the Darboux transformation as in [21]. \\ {\bf Problem N4:} Find the other solutions of the M-XX equation (solitons, dromions and so on) by the Hirota bilinear method.\\ If you have some results in these directions, please, inform me by E-mail: <EMAIL>. We are ready to interaction. \section{Acknowledgments} The author would like to thank Prof. M.Lakshmanan for hospitality during his visit to Bharathidasan University and for stimulating discussions. He is grateful for helpful conversations with A.Kundu, Radha Balakrishnan, M.Daniel, R.Radha and R.Amuda. Also he would like to thank Dr. Radha Balakrishnan for hospitality during his visit to the Institute of Mathematical Sciences.
\section{Introduction} The compact non-thermal radio source Sgr~A* has long been recognized as a massive black hole candidate. Recent results from stellar proper motion studies indicate that there is a dark mass of $\sim 2.6 \times 10^6 M_{\sun}$ enclosed within 0.01 pc (Genzel {\it et al.\ } \markcite{genze97} 1997, Ghez {\it et al.\ } \markcite{ghez98} 1998). Very long baseline interferometry studies at millimeter wavelengths have shown that the intrinsic radio source coincident with the dark mass has a size that is less than 1 AU and a brightness temperature greater than $10^9$ K (Rogers {\it et al.\ } \markcite{roger94} 1994, Bower \& Backer \markcite{bower98} 1998, Lo {\it et al.\ } \markcite{lo98} 1998). Together these points are compelling evidence that Sgr~A* is a cyclo-synchrotron emitting region surrounding a massive black hole. Nevertheless, specific details of the excitation of high energy electrons, their distribution and the accretion of infalling matter onto Sgr~A* are unknown (e.g., Falcke, Mannheim \& Biermann \markcite{falck93} 1993, Melia \markcite{melia94} 1994, Narayan {\it et al.\ } \markcite{naray98} 1998, Mahadevan \markcite{mahad98} 1998). Linear polarization stands as one of the few observables of Sgr~A* not extensively investigated observationally or theoretically. However, we expect linear polarization to arise from the cyclo-synchrotron radiation that is responsible for the radio to millimeter wavelength spectrum. A homogeneous, optically-thin, synchrotron source with a uniform magnetic field has a fractional polarization of 70\%. Measured fractional polarizations in AGN are typically a few percent at wavelengths shorter than 6 cm where the compact cores dominate the highly-polarized radio lobes in the total flux (e.g., Aller, Aller \& Hughes \markcite{aller92} 1992). However, polarization VLBI images sometimes show regions of significantly enhanced polarization (Brown, Roberts \& Wardle \markcite{brown94} 1994). The polarization of Sgr~A* may prove to be as important a diagnostic of models for the radio to millimeter spectrum as it has been for AGN. Detection of linear polarization in AGN has firmly established synchrotron emission as the radiation mechanism. Comparison of the evolution of linear polarization to the evolution of total intensity has provided a strong argument for the existence of shocks in the relativistic jets of AGN (e.g., Hughes, Aller and Aller \markcite{hughe85} 1985). Detection of similar correlations in polarized and total intensity variations in Sgr~A* would be convincing evidence for a jet. Other models may have unique signatures for polarized intensity variations. Sgr~A* is located in a region with strong magnetic fields and high electron density. The image of Sgr~A* is significantly scatter-broadened by intervening thermal plasma (e.g., Lo {\it et al.\ } \markcite{lo98} 1998), as are the images of many masers in the Galactic Center region (Frail {\it et al.\ } \markcite{frail94} 1994). Furthermore, nonthermal filaments in the Galactic Center region show RMs which vary on the arcsecond scale and are as large as 4000 ${\rm\ rad\ m^{-2}}$ (Yusef-Zadeh, Wardle \& Parastaran \markcite{yusef97} 1997). Such large RMs can effectively depolarize a signal detected with a large bandwidth. In \S 2, we discuss the effect of large RMs on a polarized signal and our Fourier transform technique for detecting large RMs. In \S 3.1, we present VLA continuum observations at 4.8 GHz. In \S 3.2 and \S 3.3, we present VLA spectro-polarimetric observations at 4.8 GHz and 8.4 GHz. These observations are sensitive to a wide-range of RMs. In \S 4, we consider other effects of interstellar matter on a polarized signal from Sgr~A*. And in \S 5, we discuss the consequences of our upper limits for the polarization on models for Sgr~A*. In a future paper, we will address millimeter polarization observations of Sgr~A*. \section{Searching for Large RMs} In an ionized and magnetized region right and left circularly polarized waves will have different indices of refraction. This leads to a wavelength-dependent delay between circular polarizations which is equivalent to a rotation of the position angle $\chi$ of a linearly polarized signal \begin{equation} \chi_F = {\rm RM}\ \lambda^2, \end{equation} where RM is the rotation measure. This rotation of $\chi$ is equivalent to a rotation in the two-dimensional Stokes $Q$ and $U$ space. A linearly polarized signal will be significantly depolarized in an observing bandwidth $\Delta\nu$ if $\chi$ rotates by more than one radian, or if the RM exceeds \begin{equation} {\rm RM_{max}}={1\over 2 } {1 \over \lambda^2}{\nu \over \Delta\nu}. \label{eqn:rmmax} \end{equation} If the bandwidth $\Delta\nu$ is split into $n \times \delta\nu$ channels, a search can be made for RMs larger than ${\rm RM_{max}}$. When $\chi$ wraps through more than one turn, $n\pi$ ambiguities make it impossible through a linear least squares fit to detect RMs. Fourier transforming the complex visibility $P=Q + i U$ with respect to $\lambda^2$ searches for large RMs without loss of sensitivity. The maximum RM detectable in this scheme can be found by replacing $\Delta\nu$ with $\delta\nu$ in Equation~\ref{eqn:rmmax}. In addition to detecting RMs that exceed ${\rm RM_{max}}$, the technique is sensitive to multiple RMs from the same object. A more detailed analysis of this technique can be found in Killeen {\it et al.\ } \markcite{kille99} (1999). A continuum observation with the VLA at 4.8 GHz with 50 MHz has ${\rm RM_{max}} \approx 10^4 {\rm\ rad\ m^{-2}}$. Splitting the band into 256 channels increases ${\rm RM_{max}}$ by two orders of magnitude to $3.5\times 10^6 {\rm\ rad\ m^{-2}}$. The minimum fully-sampled RM detectable in a spectro-polarimetric data set, ${\rm RM_{min}}$, is approximately equal to ${\rm RM_{max}}$ for a continuum data set with the same total bandwidth. The RM can be found to better accuracy than ${\rm RM_{min}}$. We estimate the error to be \begin{equation} \sigma_{\rm RM} = { {\rm RM_{min} } \over {\rm SNR} } . \end{equation} SNR is the ratio of the peak amplitude in Fourier space to the off-peak root-mean-square noise. \section{Observations and Data Reduction} \subsection{VLA Continuum Polarimetry at 4.8 GHz} The VLA of the National Radio Astronomy Observatory\footnote{The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc.} observed Sgr~A* on 10 and 18 April 1998 in the A array at 4.8 GHz with a bandwidth of 50 MHz. Instrumental calibration was performed with the compact sources 1741-038 and 1748-253. The right-left phase difference was set with observations of 3C~286. Only baselines longer than 100 $k\lambda$ were used for Sgr~A*. Several nearby calibrator sources, GC 441, W56 and W109, were also observed (Backer \& Sramek \markcite{backe99} 1999). All sources were self-calibrated and imaged in Stokes $I, Q$ and $U$. We summarize the measured polarized and total intensities of Sgr~A* and the calibrators in Table~\ref{tab:vla6cm}. The rms noise in the Sgr~A* map is 74 $\ \mu {\rm Jy}$. Consistency between the results on the two days indicates the accuracy of the results. Polarization was reliably detected from all sources but Sgr~A* and GC 441. The measured polarization at the position of Sgr~A* is 0.1\%. This value is equal to the average off source fractional polarization in the map and is, therefore, an upper limit. The maximum RM detectable with this bandwidth is $\sim 10^4 {\rm\ rad\ m^{-2}}$. \subsection{VLA Spectro-polarimetry at 4.8 GHz} The VLA observed Sgr~A* in the A array in a spectro-polarimetric mode at 4.8 GHz on 27 November 1992. Observations were carried out in 8 consecutively-spaced frequency bands of 6.25 MHz each. Each band was divided into 32 separate frequency channels. The bands covered the frequency range from 4832 MHz to 4882 MHz. Five scans of 2.5 minutes apiece on Sgr~A* were interleaved with six scans of 2.5 minutes apiece on NRAO~530 in each frequency band. Amplitude, phase and polarization calibration were performed separately for each band. Polarization calibration was performed with NRAO~530 alone and with NRAO~530 and Sgr~A*, producing similar final results. The right-left phase difference was set for each band with an observation of 3C~286. For each source, the spectral data were time-averaged and exported from AIPS for further processing. A bandpass correction was applied. The complex polarization was then Fourier-transformed with respect to $\lambda^2$. Sampling effects were removed through a one-dimensional CLEAN method. The CLEAN method permits a better estimate of the RM peak and of the noise level. The sampling sidelobes are readily visible for 3C~286 and NRAO~530 in Figure~\ref{fig:fouramp6cm}. Our tests with noise data and with synthetic signals indicate that the CLEAN method does not generate false signals and improves the accuracy of peak determination. Applying CLEAN to the 4.8 GHz NRAO~530 data reduced the noise in the spectrum from 1.6 mJy to 0.26 mJy. The range of fully-sampled RM is $10^4 {\rm\ rad\ m^{-2}}$ to $3.5\times 10^6 {\rm\ rad\ m^{-2}}$. The Fourier amplitude for each source is shown in Figure~\ref{fig:fouramp6cm} and the results are summarized in Table~\ref{tab:rm6cm}. These images are without bandpass correction and dirty-beam removal. We also calculate and plot the Fourier transform for a distribution of Gaussian noise. Strong peaks at low RM are apparent for both 3C~286 and NRAO~530, as expected. The measured values are consistent with the known RMs of these sources: $1\pm 2 {\rm\ rad\ m^{-2}}$ for 3C~286 and $-63 \pm 5 {\rm\ rad\ m^{-2}}$ for NRAO~530 (Rusk \markcite{rusk88} 1988). No strong peak is apparent for Sgr~A* at any RM. The maximum Fourier amplitude for Sgr A* is $0.15\%$ at ${\rm RM}=2.1 \times 10^6 {\rm\ rad\ m^{-2}}$. Imaging Sgr A* with and without a RM correction produced a peak polarization of $0.2\%$. This is equal to the fractional polarization of thermal ionized gas in the vicinity of Sgr A*, indicating that we are limited by residual instrumental polarization. \subsection{VLA Spectro-polarimetry at 8.4 GHz} The VLA observed Sgr~A* in the A array in a spectro-polarimetric mode at 8.4 GHz, also on 27 November 1992. Observations were carried out in 7 frequency bands of 6.25 MHz each. Each band was divided into 32 separate frequency channels. Five bands covered the frequency range from 8405 MHz to 8437 MHz. Two other bands were centered at 8150 MHz and 8700 MHz. Five scans of 2.5 minutes apiece on Sgr~A* were interleaved with six scans of 2.5 minutes apiece on NRAO~530 in each frequency band. Amplitude, phase and polarization calibration were performed separately for each band. Polarization calibration was performed with NRAO~530 alone and with NRAO~530 and Sgr~A* together, producing similar final results. The right-left phase difference was set for each band with an observation of 3C~286. The sources W56, 1741-312, GC 441, W109 and 1748-253 were observed for two minutes in the three 6.25 MHz bands centered at 8150 MHz, 8420 MHz and 8700 MHz. The results for all sources were the same using all frequency bands or only the inner 5 bands. The same reduction steps were taken for the 8.4 GHz data as for the 4.8 GHz data. The Fourier amplitudes for all sources are shown in Figures~\ref{fig:fouramp4cm1} and \ref{fig:fouramp4cm2}. These images are without bandpass correction and dirty-beam removal. The results are summarized in Table~\ref{tab:rm3cm}. These data are sensitive to $3.5 \times 10^5 < | {\rm RM} | < 1.5 \times10^7 {\rm\ rad\ m^{-2}}$. There are strong detections of linear polarization in 3C~286 and NRAO~530 at RMs consistent with zero. Significant detections were also made for W56, W109, 1741-312 and 1748-253, also at RMs consistent with zero. No polarization was detected in GC 441. The errors in RM for these secondary calibrators are larger due to the sparser frequency coverage and shorter observing time. For Sgr A*, we detect a peak in the Fourier spectrum of 0.17\% at ${\rm RM}=24000 \pm 37000 {\rm\ rad\ m^{-2}}$. Imaging Sgr A* with and without RM corrections, we find a fractional polarization of 0.1\%. Off-source fractional polarizations are typically 0.1\%, again implying that we are limited by residual instrumental polarization. We tested noise models to see if we could reproduce a weak signal at non-zero RM. We used an input signal with ${\rm RM} = 0$ at 0.15\% of the peak intensity of Sgr A* and noise that matched that of Sgr A*. This is the model plotted in Figure~\ref{fig:fouramp4cm1}. The measured RM peak wandered within the error range. \section{Interstellar Propagation Effects} The interstellar medium may depolarize a linearly polarized radio wave in two ways: significant rotation of the polarization position angle through the observing bandwidth; and, differential Faraday rotation along the many paths that contribute to the scatter-broadened image of Sgr A*. We have already addressed the first effect in \S 2 and found in \S 3 that Sgr A* is not depolarized by RMs less than $1.5\times 10^7 {\rm\ rad\ m^{-2}}$. We now consider the second effect. The scattering region will depolarize the signal if $\delta\chi_F\approx\pi$. For our observing wavelengths, $\delta{\rm RM}=900 {\rm\ rad\ m^{-2}}$ and $\delta{\rm RM}=2400 {\rm\ rad\ m^{-2}}$. Over the scattering size of 50 mas at 4.8 GHz, this corresponds to $\delta{\rm RM}/\delta{\theta}=18000 {\rm\ rad\ m^{-2} arcsec^{-1}}$. Observed variations in RM in the GC region are many orders of magnitude below those necessary to depolarize Sgr A*. Observations on the arcsecond to arcminute scale of a nonthermal filament within 1 degree of Sgr A* find a maximum $\delta{\rm RM}/\delta{\theta}=250 {\rm\ rad\ m^{-2}\ arcsec^{-1}}$ (Yusef-Zadeh, Wardle \& Parastaran \markcite{yusef97} 1997). Extrapolation of the RM structure function to the scattering size implies $\delta{\rm RM}\approx 50 {\rm\ rad\ m^{-2}}$. However, these observations are made on a much larger scale than the scattering disk of Sgr A* and the scattering medium is believed to be inhomogeneous. Could the more extreme conditions necessary to depolarize Sgr A* exist in the Galactic Center scattering region? The RM is expressed as \begin{equation} {\rm RM} = 0.8 n_e B L {\rm\ rad\ m^{-2}}, \end{equation} where $n_e$ is the electron number density in cm$^{-3}$, $B$ is the magnetic field parallel to the line of sight in $\mu{\rm G}$ and $L$ is the size scale in pc. Since $L$ must be a fraction of the scattering diameter, we find $L \sim 0.1 \theta_{Sgr A*} D_{Sgr A*} \sim 10^{-4} {\rm\ pc}$. Yusef-Zadeh {\it et al.\ } \markcite{yusef94} (1994) argued that the photo-ionized skins of molecular clouds in the GC region have a similar length scale, milliGauss fields and $n_e \sim 10^4 {\rm cm^{-3}}$. This matches the depolarization condition if the regions are fully turbulent. However, if the constraints on the outer scale of turbulence derived by Lazio \& Cordes \markcite{lazio98} (1998) are correct, then $L \sim 10^{-7}{\rm\ pc}$. In this case, the RM condition and pressure balance between the magnetic and thermal components can only be satisfied if $B\sim 10 {\rm\ mG}$ and $n_e \sim 10^6 {\rm\ cm^{-3}}$. These conditions are extreme even for the GC region. The largest magnetic fields as measured for OH masers are on the order of a few milliGauss (e.g., Yusef-Zadeh {\it et al.\ } \markcite{yusef96} 1996). Ionized densities measured for H II regions on the arcsecond scale ($\la 0.1$ pc) are significantly less than $10^5 {\rm\ cm^{-3}}$ (Mehringer {\it et al.\ } \markcite{mehri93} 1993). No depolarization is predicted for the higher temperature and lower density model of Lazio \& Cordes \markcite{lazio98} (1998) for $B < 1 {\rm\ G}$. The conditions necessary to depolarize at 8.4 GHz are even more extreme. We conclude, therefore, that the conditions necessary to depolarize Sgr A* are unlikely to occur in the scattering region. We consider now whether depolarization may occur in the accretion region of Sgr A*, where the electron density and magnetic field strength are large but the length scale is smaller. If we consider the simplest model of spherical infall with $\dot{M}=10^{-4} M_{\sun} y^{-1}$ and equipartition between particle, magnetic and gravitational energy (Melia \markcite{melia94} 1994), then the change $\delta{\rm RM}$ over an interval $\delta r$ at a radius $r$ from Sgr A* is \begin{equation} \delta {\rm RM} = 1.2 \times 10^{14} r^{-11/4} \delta r {\rm\ rad\ m^{-2}}, \end{equation} where we have expressed $r$ and $\delta r$ in units of the gravitational radius $r_g =2 G M/c^2=7.8\times 10^{11} {\rm\ cm}$ for a $2.6 \times 10^6 M_{\sun}$ black hole. This relation only holds for $r \ga 10^{3}$, where the temperature falls below $10^9 {\rm\ K}$. The RM inside of this radius is negligible unless there is a separate population of cold electrons. We consider the effect of cold electrons in more detail in the following Section. If the scattering screen is at a distance of 100 pc from Sgr A*, then the image will be an average of ray paths over a tangential length scale $l \sim 2\times 10^{-5} r$. Fluctuations in the RM will depolarize Sgr A* at a given radius if $\delta{\rm RM} > 900 {\rm\ rad\ m^{-2}}$ and $l > l_0$, where $l_0$ is the outer scale of turbulence. Assuming $\delta r \sim r$, we find that depolarization will occur only if $l_0 \la 10^{-5} {\rm\ pc}$. Although this scale is much smaller than the outer scale in the local ISM (Armstrong, Rickett \& Spangler \markcite{armst95} 1995), it is a scale that may be pertinent to the dense, energetic environment of the accretion region. \section{An Intrinsically Weakly Polarized Sgr A*} A polarization fraction less than 1\% is uncommon in compact radio sources at wavelengths shorter than 6 cm (Aller, Aller \& Hughes \markcite{aller92} 1992). However, optically thick quasar cores observed with VLBI are frequently weakly polarized (Cawthorne et al. \markcite{cawth93} 1993). Such cores may be analogous to the radio source in Sgr A*, which, due to its low power, may not produce the strong shocks in the jet that are the source for higher polarization regions in quasars. Weak polarization is more common in radio galaxies than quasars or blazars and it is also more common in compact-double sources or sources with irregular morphologies (Aller, Aller \& Hughes \markcite{aller92} 1992). A notable source with a very low polarization fraction ($<0.1\%$) is the radio galaxy 3C 84, which has a very irregular morphology. If the radio to millimeter spectrum of Sgr A* does arise in a jet, the low power of this jet or environmental effects in the Galactic Center region may limit the magnetic field order. For the case of a spherically symmetric emitting region, an ordered magnetic field may depolarize the source, as well. Alternatively, low energy electrons in the synchrotron environment of Sgr A* may Faraday depolarize the source. The ADAF model and the Bondi-Hoyle accretion model predict the presence of non-relativistic electrons in the accretion region. Observations at millimeter wavelengths may resolve many of the questions raised in this paper. Interstellar effects are reduced such that depolarization in the scattering region is extremely unlikely and depolarization in the accretion region must occur at radii less than 0.01 pc. Furthermore, the synchrotron emission arises from a more compact and presumably more homogeneous region. The source may also have less synchrotron self-absorption at millimeter wavelengths, although this is not required by all models. We will report in a future paper on millimeter polarimetric observations of Sgr A*. \acknowledgements We thank Alok Patnaik for enlightening discussions. HF is supported by DFG grant Fa 358/1-1\&2.
\section{Introduction} It has long been recognized that measurement of the profile of the Fe K$\alpha$ fluorescence line found in many AGNs at $\sim$6.4 keV (Mushotzky 1995, Tanaka et al. 1995, Yaqoob et al. 1995, Fabian et al. 1995, Nandra et al. 1997a and references therein) can provide an important tracer of matter in the vicinity of the postulated supermassive black hole. Doppler and gravitational shifts would imprint characteristic signatures on the line profile which map the geometric and dynamical distributions of matter surround the black hole. Additional information concerning the geometry of matter in the active nucleus could in principle be derived by studying the rapid variability of the line profile, intensity and their relationship with the continuum variations. So far, rapid variability in Fe K line has been detected only in two Seyfert galaxies. Yaqoob el al. (1996) presented the evidence for rapid variability of the Fe K line profile in the narrow-line Seyfert galaxy NGC 7314, which is consistent with a diskline of constant equivalent width superposed on a constant flux narrow line (presumably from the torus). They found while the X-ray continuum flux varied by a factor of 2 on a time scale of hundreds of seconds, the emission in the red wing of the Fe K line below $\sim$6 keV responded to those variations on the time scales of less than $\sim3\times10^4$ s, and the response becomes slower and slower towards the line peak near 6.4 keV. Rapid variability of the Fe K line profile was also found in the bright Seyfert 1 galaxy MCG --6-30-15 (Iwasawa et al. 1996). On time scales of less than $\sim10^4$ s, the variability pattern is the same as seen in NGC 7314. But on long time scales of about several 10$^4$ s, it shows a different behavior. When the source is bright, the Fe K line is weak and dominated by the narrow core, whilst during a deep minimum, a huge red tail appears. The intensity of broad Fe K line correlates inversely with the continuum flux, so does the equivalent width. In this Letter we present an evidence of rapid variability of Fe K line in another nearby (Z = 0.0023) low luminosity narrow-line Seyfert 1 galaxy NGC 4051, which is well known for its rapid variability in the X-ray band. Nandra \& Pounds (1994) found an equivalent width of Fe K line $140\pm70$ eV from the Ginga data by assuming a single narrow line. A narrow Fe K emission line was also detected with an upper limit on the FWHM of $\sim460$ eV and equivalent width EW $\sim170$ eV during the first ASCA observation in 1993 (Mihara el al. 1994, hereafter M94) for this object. When the target was re-observed by ASCA in 1994 (Guainazzi el al. 1996, hereafter G96), the Fe K line was found to be stronger (EW = $350^{+170}_{-150}$ eV) and broad($\sigma>0.2 keV$). \section{The ASCA data} NGC 4051 was observed twice by ASCA from 1993 April 25 22:30 to April 26 21:30 and from 1994 July 6 15:28 to July 9 10:40. In this Letter, we concentrated on the Solid-state Imaging Spectrometer (SIS) data because of the better energy resolution it provides (Inoue 1993). The data reductions were performed with the ASCA standard software XSELECT according to the following criteria: satellite not passing through the South Atlantic Anomaly, geomagnetic cutoff rigidity greater than 6 GeVc$^{-1}$, and minimum elevation angle above Earth's limb of $10^{\rm o}$ and $20^{\rm o}$ for nighttime and daytime observations, respectively. Source counts were extracted from a circular area of radius 4' for the SIS0 and SIS1. Because of an error in satellite pointing during the first observation, the image fell into the dead region between two chips in SIS1 (M94). Therefore we only used SIS0 data from this observation for spectral analysis. To avoid the complexity caused by the "warm absorber" and "soft excess" (M94, G96), we use the 2.5--10.0 keV data for spectral analysis. Spectral fits were carried out using the XSPEC, and the background was taken from the blank sky data. \section{Variability of Iron K Line} \subsection{Comparisons between two observations} Time average spectra of Fe K$\alpha$ line profiles for the two observations were given by Nandra el al. (1997a) \& G96. Due to the poor statistics, the line profile cannot be measured exactly, therefore we model the line with a single Gaussian function. The underlying continuum is fitted with a single power-law absorbed by the Galactic column density ($N_H\approx$1.3$\times$$10^{20} cm^{-2}$). We did not take into account of the possible Compton reflection component here because of its small impact on the ASCA spectra. Results of the single Gaussian fits are given in Table 1. For comparison, the results of diskline (Fabian el al. 1989, George \& Fabian 1991) fits can be found in Nandra el al. (1997a) \& G96. The variability of Fe K line is evident (EW changes from 166 eV to 330 eV and FWHM from 4500 km/s to 50000 km/s), however, there is no obvious variability in the photon index of X-ray continuum. Because of the pointing error of the satellite during the first observation, part of the photons fell into the gaps between the CCD chips. The continuum flux of Obs1 in Table 1 is only a lower limit. However, since the gaps occupy a fairly small fraction of the selected area that includes our target ($\sim18\%$), the continuum flux obtained cannot be biased too much. Thus we conclude that there is no significant difference in the continuum flux between two observations. \subsection{Rapid FeK Line variability in the second observation} During the two ASCA observations, NGC 4051 shows large amplitude X-ray continuum flux changes on time scales of $\sim$100 s. In order to see if there are short time scale variations of Fe K line within a single observation, we choose Obs2 which last longer time ($\sim$3 days) for analysis in detail. Including a large flare in the beginning and a deep minimum near the end, the whole observation has been divided into five time intervals (from i1 to i5) with a similar exposure time ($\sim$10 ks) as shown in Fig. 1 (different from G96). Among the five intervals, i-2, i-3, i-4 have similar continuum fluxes. Results of single Gaussian fittings are shown in Table 1 (We also give the results of diskline fits to the five intervals, respectively). The photon index is quite similar between i2 to i4 at about $\Gamma\sim$1.80. A steeper continuum is suggested for i-1 and a flatter one for i-5. The single Gaussian line dispersions and equivalent widths are also similar between i2 to i4. A stronger and broader line is detected in i-1, and a weaker narrower one in i-5 (Fig. 2). In order to show the variability of Fe K line more clearly, we present a contour plot of Gaussian line width versus the equivalent width for i-1, i-5 and i-2+3+4 (a summed dataset of i-2,i-3,i-4) obtained from the single Gaussian fits in Figure 3. For i-1, the equivalent width of Fe K line is 733 eV, much larger than that for the time average spectrum, and the Fe K line is much broader ($\sigma$ = 0.96 keV). While for i-5, the EW of Fe K line is only 165 eV, and much narrower (only a narrow line has been detected, $\sigma<$0.09 keV). Due to the poorer statistics in i-5, there is concerning that a weak broad Fe K component might be un-noticeable. To exclude this possibility, we add a broad Gaussian line (E$_{G} = 6.4$ keV, $\sigma = $0.46 keV, the same with the time average spectrum) to the fit. The upper limit for the EW of this component is 80 eV, far smaller than those in i-1 to i-4. These results clearly demonstrate the large variability of Fe K line during the second ASCA observation. G96 also analyzed the Fe K line, but failed to detect the variability of Fe K line profiles and equivalent widths. We would like to point out here that the time sequence selection used in G96, different to what we did in this paper, tends to smear out the line variations. \section{Discussions} \subsection{The X-ray continuum flux versus the equivalent width of Fe K line} We detect significant change not only in line equivalent width but also in line profile through single Gaussian fitting to the two observations. From 1993.4.25 to 1994.7.6, Fe K line in NGC 4051 became broader and stronger, in spite of the fact the continuum flux remains almost constantly. This shows that the long time scale variability of Fe K line can be independent of the continuum variation. Such kind of behavior can be induced by the change in structure of the line emission region or the geometry of the hard X-ray source in the context of disk line model. The broader and stronger line emission during the 1994 epoch indicates that the average line emission region is much closer to the putative accretion disk. We also detect rapid variability ($\sim10^4$s) of Fe K line within the second observation. During the bright flare (i-1), the Fe K line is broad and strong, while during the deep minimum (i-5), Fe K line is narrower and weaker. Using the ASCA data of 39 AGNs, Nandra el al. (1997b) found a clear decrease in the strength of the Fe K$\alpha$ line with the increase in the luminosity (X-ray Baldwin effect, Iwasawa \& Taniguchi 1993), which was thought due to the fact that high-luminosity AGNs have high accretion rates, causing the accretion disk to become ionized (Matt et al. 1996). An inverse correlation between the Fe K line EW and the X-ray continuum flux was also found in MCG --6-30-15. The line EWs (sum of the broad and narrow component, calculated at the centroid energy 6.4 keV, see Table 2, Iwasawa et al. 1996) versus continuum flux is plotted here in Fig. 4b. The case for the NGC 4051 shows that the situation is more complicated than this. We find an increase in the equivalent width with no apparent change in the continuum flux when the two ASCA observations are considered. Moreover, an increase in the line EW with the increase X-ray continuum flux (Fig. 4a), completely opposite to the Baldwin effect and the variability in MCG --6-30-15, is found during the second ASCA observation. It indicates a complicate mechanism should be considered. \subsection{rapid variability on time scale $\sim10^4$s} It is well known that the X-ray source in many AGNs is highly variable on short time scales $\sim$100 s. As Fe K line is produced via fluorescent process, its strength and profile should response to the continuum variability. Obviously, the measurement of the temporal response is the best way to map the matter distribution in innermost regions around the black hole. In principle, the distance from the X-ray continuum source to the line emission region could be determined through observing reverberation effect (Reynolds et al. 1998). With the Fe K line profile, we could derive the mass of the black hole in the heart of nucleus. The characteristic time scales on which reverberation effects occur is the light crossing time of one gravitational radius, which is $\sim$50 s for a 10$^7$ M$_{\sun}$ black hole. But, all current available instruments are not able to measure such a fast time response. Even for a bright Seyfert galaxy, ASCA can only obtain 1$\sim$2 Fe K line photons in 100 s, too few to define the line flux and profile. However, the Fe K line variability of time scales $\sim10^4$ s can provide constraints on the long time scale ($\sim10^4$ s) variation of the X-ray source or the line emission region. So far, such variability has been detected in only three Seyfert galaxies, NGC 7314, MCG --6-30-15 and NGC 4051. For NGC 7314, a simple disk-plus-torus model could explain its rapid variability fairly well. But for MCG --6-30-15 and NGC 4051, the situation is more complicated. According to the disk model, the fluorescent iron lines is produced from the hard X-ray irradiation of the disk composed of cold gas around a black hole. When the X-ray source is bright, iron will start to be ionized, and in the intermediate ionization states, the resonance scattering can cause a reduction in the line flux (Matt et al. 1996). It is a possible reason causing the anti-correlation of the EW of broad components to the X-ray continuum flux in MCG --6-30-15. We consider that the ionized disk is also a possible reason for the positive correlation in NGC 4051 because a factor $\sim$2 larger EW for FeXXV than cold iron could occur (Matt et al. 1996). Though not well constrained, the diskline energies for i1 to i5 in Table 1. show an trend to correlate with the continuum flux, which gives some support to this opinion. There is also another possible explanation to the the rapid variability of Fe K line. Fabian (1997) suggested that the X-ray continuum might be generated by magnetic flares above the accretion disk. At any given time there are only a few flares otherwise the rapid variability will be average out. When the dominated flares move around on the disk, changes in the Fe K line profile would be seen. Such model could explain the odd variability of Fe K line in MCG --6-30-15 well. We think it is also appropriate for the rapid variability in NGC 4051 discovered in this paper. For the sequence i-5, the Fe K line profile is similar to the bright flare in MCG --6-30-15. The small FWHM might due to a dominate flare located far away from the central black hole, or a succession of flares on the approaching side of the disk. The broad line during i-1 might due to a bright dominate flare very close ($\sim$6$R_g$) to the black hole, and the large EW might due to a overabundance of iron. \acknowledgments This work is supported by Chinese National Natural Science Foundation, PanDeng Project and Foundation of Ministry of Education. The authors would also like to thank S.A. Huang for great help with ASCA data reduction.
\section{Introduction} The current interest for dwarf spheroidal galaxies (hereafter DSphs) as satellites of the Milky Way is raised by two fundamental questions concerning their evolution: the real content of dark matter (DM) in these objects (see, for example, Mateo \cite{Mateo94},\cite{Mateo97},\cite{Mateo98b}; Piatek \& Pryor \cite{Piatek95}; Burkert \cite{Burkert97} and references therein) and their implication in the hierarchical formation of the galactic halo (e.g., Johnston et al. \cite{Johnston96}). The Sagittarius (Sgr) dwarf galaxy is the closest known satellite galaxy to the center of the Milky Way, $R_{GC} \sim 16$ kpc (Ibata et al. \cite{Ibata95}, \cite{Ibata97}). Due to its proximity we can expect from its study an additional contribution to our understanding of DSphs in general. Since the announcement of its discovery by Ibata et al. (\cite{Ibata94}) the structure and evolution of Sgr have been extensively discussed and simulated by various authors: Ibata et al. (\cite{Ibata95}), Johnston et al. (\cite{Johnston95}), Vel\'azquez \& White (\cite{Velazquez95}), Whitelock et al. (\cite{Whitelock96}), Mateo et al. (\cite{Mateo96}), Alard (\cite{Alard96}), Ibata et al. (\cite{Ibata97}), Edelsohn \& Elmegreen (\cite{Edelsohn97}), Layden \& Saragedini (\cite{Layden97}), Ibata \& Lewis (\cite{Ibata98}), Mateo et al. (\cite{Mateo98}). Important problems concerning this system are: 1) its possible lifetime before dissolution, 2) the possible presence of DM in it. A critical point in the studies mentioned above is the question whether Sgr is in virial equilibrium or not. \begin{table} \caption{Observational parameters of Sagittarius dwarf galaxy (Ibata et al. \cite{Ibata97})} \label{Sgr_tbl} \begin{minipage}{5cm} \begin{tabular}{ll} \hline Parameter& \\ \hline $r_{hb}$ & 0.55 kpc\\ $\sigma_o$ & 11.4 km/s \\ $\mu_{oV}$ & 25.4 mag/arcsec$^2$ \\ $L_{t}$ & $\geq 10^7$ L$_{\odot}$ \\ $(M/L)_o$ & 50 M$_{\odot}/$L$_{\odot}$ \\ $M_{t}$ & $> 10^9$ M$_{\odot}$ \\ $(l,b)$\footnote{Galactic coordinates} & (5.6$^o$, -14$^o$) \\ $(U,V,W)$\footnote{Galactic velocities} & (232,0:,194)$\pm$ 60 km/s \\ $d_{\odot}$\footnote{Heliocentric distance} & 25 kpc \\ $R_{GC}$\footnote{Galactocentric distance} & 16 kpc \\ $v_r$\footnote{Radial velocity} & 171 km s$^{-1}$\\ $(dv/db)$\footnote{Gradient of the radial velocity} & -3 km s$^{-1}$/degree \\ \end{tabular} \end{minipage} \end{table} Usually, the total inferred mass of the DSph is calculated by assuming it is in virial equilibrium\footnote{In this paper, the {\it inferred mass} is always obtained by assuming a virial equilibrium for the satellite.}. In this case, the central mass-to-light ratio depends on the velocity dispersion through the equation (Richstone \& Tremaine \cite{Richstone86}) \begin{equation} \left(\frac{M}{L}\right)_o= \eta \frac{9 \sigma^2_o}{2 \pi G \mu_o r_{hb}} \label{ML} \end{equation} \noindent where $\eta$ is near unity for a wide variety of models, $\sigma_o$ the central velocity dispersion, $\mu_o$ the central surface brightness and $r_{hb}$ the half-brightness radius. The analysis of the validity of Eq. (\ref{ML}) for evolving DSphs has been studied in detail in the present context by Kroupa (\cite{Kroupa97}, hereafter K97). In Table \ref{Sgr_tbl}, we have summarized the parameters of Sgr DSph measured by Ibata et al. (\cite{Ibata97}). For the values of $\sigma_o$, $r_{hb}$ and $\mu_{oV}$ given in Table \ref{Sgr_tbl}, those authors obtain a high central mass-to-luminosity ratio $(M/L)_o= 50$ M$_{\odot}/$L$_{\odot}$ by using the Eq. (\ref{ML}). Therefore, the total inferred mass assuming virial equilibrium, $M_{t}$, is DM dominated. The values of total luminosity, $L_{t}$, and the total inferred mass, $M_{t}$, are also listed in Table \ref{Sgr_tbl}. However, the values of $(M/L)_o$ and $L_t$ suggested by Mateo et al. (\cite{Mateo98}) are different. These authors have discovered a tidal extension of Sgr, which implies $L_t \leq 5.8 \times 10^7$ L$_{\odot}$ and, as a consequence, the $(M/L)_o$ ratio could decrease to 10 M$_{\odot}/$L$_{\odot}$. Adopting the structural and orbital characteristics usually assumed (given in Table \ref{Sgr_tbl}), Ibata et al. (\cite{Ibata97}) recently suggested that such a satellite galaxy is expected to be tidally disrupted and destroyed after several pericentric passages, unless a significant quantity of DM is present inside it, as inferred assuming virial equilibrium. The observations of the Sgr globular clusters give an age spread of its constituents $> 4$ Gyrs and the youngest globular cluster (Terzan 7) is 9-13 Gyrs old (Montegriffo et al. \cite{Montegriffo98}). The fast disruption obtained by the numerical methods raise the question about the age and the dynamical history of Sgr. However, it must be emphasized that the initial time $t=0$ of the simulations does not necessarily coincide with the time of formation of the oldest constituents of the satellite. The partial formation of the galactic halo by hierarchical processes (accretion of small galaxies) has recently received an increase of interest, stimulated for instance by the investigation by Lynden-Bell \& Lynden-Bell (1995) on the reality of streams (Lynden-Bell \cite{Lynden82}) in the close environment of the Milky Way, among them the well known Magellanic Stream. DSphs seem to belong to one or another of these streams and their evolution clearly depends on their environment. This scenario of satellite formation involves a low DM content for the DSphs (Barnes \& Hernquist \cite{Barnes92}; Kroupa \cite{Kroupa98b}). However, the only simulations of no-dark matter satellite galaxies able to survive in a tidal field are those of K97 and Klessen \& Kroupa (\cite{Klessen98}, hereafter KK98). These authors have studied a region in the parameter space ($M_{sat}, r_{sat}$), where $M_{sat}$ and $r_{sat}$ are the mass and a typical radius of the satellite, and they have obtained a residual satellite from a more massive one. Nevertheless, their remnant satellite galaxies are fainter than a typical DSph, unless the true $(M/L)_{\mbox{real}} < 3$. These circumstances make questionable the maintenance of dynamical equilibrium and consequently the virial estimation of mass for these systems. In this paper we present self-consistent N-body simulations of Sgr and a more accurate and plausible scenario of its evolution, which suggests that Sgr is likely to be able to survive for a long time ($6-10$ Gyr) by orbiting in the Galaxy without being dominated by DM. In Sect. 2, we present the model of the Milky Way used in our simulations. In Sect. 3, we describe the different models of the satellite galaxy and corresponding scenarios of interaction with the Galaxy. In Sect. 4 some numerical considerations are given. In Sect. 5, we present the results of our simulations for the different chosen scenarios. These results are discussed in Sect. 6 in connection with recent conjectures by Kroupa (\cite{Kroupa98}, hereafter K98) on the parameter space possible for progenitors of surviving DSphs without DM. Finally, a summary is given in Sect. 7. \section{The model of the Milky Way} \label{galaxy} The initial model of the primary galaxy, representing the Milky~Way, is based on the axisymmetric initial conditions adopted by Fux (\cite{Fux97}) in his N-body modeling of the Galaxy. The mass distribution is divided in three components: (i) an oblate stellar nucleus-spheroid with a spatial density $\rho\propto r^{-1.8}$ in the central region, in agreement with the near-infrared observations of the inner bulge (Becklin \& Neugebauer \cite{Becklin68}; Matsumoto et al. \cite{Matsumoto82}), and $\propto r^{-3.3}$ outside the bulge, as the number density counts of stellar halo objects (e.g. Preston et al. \cite{Preston91} for RR Lyrae; Zinn \cite{Zinn85} for globular clusters), (ii) a double exponential stellar disc, with radial and vertical scale lengths $h_R=2.5$~kpc and $h_z=250$~pc and (iii) an oblate dark halo with an exponential profile ensuring a roughly flat rotation curve out to $R=40$~kpc. Both oblate components have a flattening $c/a=0.5$. The total mass of the luminous component is $M_L=8.25 \times 10^{10}$ M$_{\odot}$. The model is identical to Fux's model m04t0000, except that the dark halo is more extended, with a scale length $b=13$~kpc and a total (untruncated) mass $M_{\rm DH}=2.4\times 10^{11}$~M$_{\odot}$, and that the truncation radius is moved further away at $R_{\rm c}=70$~kpc. The chosen truncation radius ensures a non-vanishing density everywhere along the orbit of the satellite galaxy. The resulting rotation curve of the model is shown in Fig.~\ref{f1}. \begin{figure}[t] \centerline{\psfig{figure=f1.ps,width=8.8cm}} \caption[]{Rotation curve of the initial Milky Way model (full line), with the contributions of each component} \label{f1} \end{figure} \par The initial kinematics is obtained by solving the hydrodynamical Jeans equations, assuming an isotropic velocity dispersion for the disc component and a anisotropic but centrally oriented velocity ellipsoid for the other components. The velocity distribution of the nucleus-spheroid and dark halo is generated from a 3D generalization of the Beta distribution which limits the number of escaped particles and thus greatly improves the equilibrium of the isolated dynamical model. More details are given in Fux~(\cite{Fux97}). \section{Satellite models} The effects that a DSph suffers when it is accreted on a primary galaxy strongly depend on its structure and its orbit. We have considered two kinds of satellite models. They correspond to two scenarios of interaction between the Galaxy and the satellite. In the first models (f-models), we represent the satellite as a {\it standard} King's sphere (King \cite{King66}), which matches the observational constraints for the total virial mass, $M_t$, the core radius, $r_o$, and the central velocity dispersion, $\sigma_o$, of the Sgr DSph. In this case, the effects on the DSph are maximal because the satellite, originally in an isolated situation, suddenly undergoes strong tidal perturbations. Our alternative scenario (s-models) assumes that either the DSph is formed in the tidal tail of another major accretion event and therefore it is built in equilibrium with the environment, or the DSph falls slowly from a quasi-isolated situation to a tidal field region and it has enough time to readjust itself to the environmental forces. In both situations, we begin our simulations when the satellite has already reached the equilibrium with the dense environment and, therefore, the life time of the DSph orbiting in the galaxy is expected to be longer than in our first scenario. In the present case, the satellite is modelled by a {\it modified} King's sphere (G\'omez-Flechoso \& Dom\'{\i}nguez-Tenreiro \cite{Gomez97}), which takes into account the tidal potential produced by the primary galaxy. We have checked several orbits in order to compare the observational features of the Sagittarius dwarf and the numerical results in both fast and slow accreting scenarios. The orbits we have chosen reproduce the present position, $(l,b)$, and galactocentric velocity, $(U,V,W)$, of Sagittarius DSph in the observational range (Table \ref{Sgr_tbl}), and, therefore, the simulated orbits are polar, as suggested by the void component of the proper motion parallel to the Galactic Plane. The orbits are either low eccentricity orbits in the central part of the Galaxy or high eccentricity orbits. The central mass-to luminosity ratio of the models has been calculated by using the Eq. (\ref{ML}). The central velocity dispersion $\sigma_o$ involved in this equation has been measured along the line-of-sight. We have removed the particles with the largest radial velocities (relative to the radial velocity of the center of mass of the satellite) to prevent contamination by outlying particles. We have only considered those particles with projected distance to the center of the satellite smaller than 0.5 kpc. \subsection{f-models} The distribution function of an isolated galaxy fulfills the collisionless Boltzmann equation. We can represent an isolated DSph as a solution of this equation. King's spheres are an example of that. If an initially isolated DSph reaches the inner regions of a galaxy within a short timescale, it has no time to modify its internal structure. In this case the satellite maintains its isolated King's sphere distribution function at the beginning of the simulation. Once in the inner orbit, the satellite will evolve quickly, because tidal forces are strong in these regions. This scenario is unrealistic, because, in reality, a satellite which reaches the denser parts of the galaxy has suffered the influence of the galaxy potential for a long period of time. However, we have run simulations in such a case because they correspond to the common initial conditions assumed in the literature (Johnston et al. \cite{Johnston95}, Oh et al. \cite{Oh95}, Vel\'azquez \& White \cite{Velazquez95}, Johnston et al. \cite{Johnston96}, Edelsohn \& Elmegreen \cite{Edelsohn97}, K97, KK98, Ibata \& Lewis \cite{Ibata98}) and in order to compare the results with those of the more realistic s-model satellites. The parameters of the King's model we have selected for the satellite model in a fast accreting scenario are given in Table \ref{tbl1} and they have been chosen to reproduce the present characteristics of Sagittarius, as was done in other simulations (Vel\'azquez \& White \cite{Velazquez95}). The total mass of the satellite model corresponds to the virial mass inferred for Sgr. Therefore, if the observational constraints on the luminous mass are considered, the satellite model could represent a DM dominated DSph. For these models, we have assumed $(M/L)_{\mbox{real}}=10$ M$_{\odot}/$L$_{\odot}$ for the calculations of surface brightness, which is the lower limit of the mass-to-luminosity ratio for DM dominated satellites. A higher $(M/L)_{\mbox{real}}$ would represent fainter satellite galaxies (for the same DM content) and an apparently faster dissolution process. \begin{table} \caption{Physical parameters of the isolated King's model of Sgr (f-models): core radius, $r_o$, central velocity dispersion, $\sigma_o$, total mass, $M_t$, dimensionless potential, $W_o$ and tidal radius, $r_t$} \label{tbl1} \begin{tabular}{lllll} \hline $r_o$ & $\sigma_o$ & $M_t$ & $W_o$ & $r_t$ \\ (kpc) & (km/s) & ($10^7$ M$_{\odot}$)& & (kpc) \\ \hline 0.527 & 15.0 & 12.0 & 3.26 & 2.736\\ \hline \end{tabular} \end{table} The initial apocenter and pericenter and the period of the low eccentricity orbit (f-A) and the high eccentricity orbits (f-B1 and f-B2) are listed in Table \ref{tbl1-2}. \begin{table} \caption{Parameters of the orbits of the f-models} \label{tbl1-2} \begin{tabular}{llll} \hline Model & $r_{min}$ & $r_{max}$ & Period\\ & (kpc) & (kpc) & (Gyr)\\ \hline f-A & 12 & 18 & 0.23\\ f-B1 & 8 & 38 & 0.45\\ f-B2 & 10 & 70 & 0.95\\ \hline \end{tabular} \end{table} \subsection{s-models} In the other possible scenario we assume that either the satellite has been formed inside the tidal tail of a major merger (e.g. numerical simulations by Barnes \& Hernquist 1992, and the observational counterpart by Duc \& Mirabel \cite{Duc97}) or it has been slowly accreted. In the first case, if the dwarf galaxy is formed in equilibrium with the tidal force of the environment, it does not contain a significant amount of DM (Barnes \& Hernquist \cite{Barnes92}). \begin{table} \caption{Galaxy parameters of the s-model satellites} \label{tbl2} \begin{tabular}{llllll} \\ \hline Model&$r_o$&$\sigma_o$& $M_t$ &$W_o$&$r_t$ \\ &(kpc)& (km/s) &($10^7$ M$_{\odot}$)& & (kpc) \\ \hline s-A & 0.06 & 11.0 & 0.93 &3.78 & 0.55 \\ s-B1 & 0.1 & 11.0 & 1.66 &3.95 & 0.99 \\ s-B2a& 0.1 & 15.0 & 5.27 &5.42 & 1.93 \\ s-B2b& 0.1 & 11.0 & 2.33 &4.87 & 1.48 \\ s-B2c& 0.3 & 15.0 & 6.04 &2.98 & 2.02 \\ \hline \end{tabular} \end{table} \begin{table} \caption{Orbital parameters of the s-model satellites} \label{tbl2-2} \begin{tabular}{lllll} \\ \hline Model&$r_{min}$&$r_{max}$&$r_{ave}$ & Period\\ & (kpc) & (kpc) & (kpc) & (Gyr)\\ \hline s-A & 15 & 20 & 20 & 0.25\\ s-B1 & 8 & 40 & 35 & 0.45\\ s-B2 & 10 & 74 & 50 & 1.25\\ \hline \end{tabular} \end{table} In the second case, according to cosmological models of hierarchical structure formation, satellite systems are produced around massive galaxies. These satellites could contain DM halos (e.g. Cole et al. \cite{Cole94}). Tidal forces could be negligible in the outer regions of the galaxy where the satellite is formed. However, if we assume that the satellite does not go through the denser and central regions of the main galaxy in the first perigalacticon, we could expect that an initially massive satellite falls slowly on the center of the galaxy by dynamical friction, loses part of its mass and reaches a central orbit. In this way, the satellite has time to modify its internal structure and to reach the equilibrium with the environment. DSphs described in the two last scenarios have to be in equilibrium with the tidal forces of the environment. In the paper by G\'omez-Flechoso \& Dom\'{\i}nguez-Tenreiro (\cite{Gomez97}), the structural parameters (total mass, $M_t$, velocity dispersion, $\sigma_o$, core radius, $r_o$, etc.) of a satellite in the tidal field of the primary galaxy have been estimated. Those authors have proved that, in general, a galaxy in a tidal field can be described by a two parameter distribution function. They have solved altogether the Poisson equation and the collisionless Boltzmann equation for a galaxy, taking into account the potential of the galaxy and the tidal potential of the environment. Only spherical terms of the tidal field were considered in this theory. However, the obtained {\it equilibrium} solutions are better representations of the system than {\it isolated} models. This result suggests that in our problem we could represent a DSph in equilibrium with the tidal field of the primary galaxy as a {\it modified King's sphere} (see G\'omez-Flechoso \& Dom\'{\i}nguez-Tenreiro \cite{Gomez97}) with two free parameters. We have chosen the central velocity dispersion, $\sigma_o$, and the core radius, $r_o$, as input parameters, because both can be determined from observations. In our simulations, the values for these two parameters are $r_o \sim 0.06-0.3$~kpc and $\sigma_o \sim 11-15$~km/s, which reproduce the characteristics of the Milky Way satellites. Thus, we will try to explain the Sagittarius satellite as a typical DSph which has evolved orbiting for a long time in the Galaxy potential. The other parameters of the model (total mass, tidal radius and dimensionless central potential) are automatically determined by the tidal potential at each position of the orbit. The satellite parameters of the models which have been performed are listed in Table \ref{tbl2}. The second column is the core radius, $r_o$, and the third column is the central velocity dispersion, $\sigma_o$, which are input parameters of the modified King's spheres. The total mass, $M_t$, the dimensionless central potential, $W_o$, and the tidal radius, $r_t$, (columns 4, 5 and 6) are output parameters obtained by solving the collisionless Boltzmann equation with the tidal potential at the averaged distance of the orbit to the Galaxy center (parameter $r_{ave}$ in Table \ref{tbl2-2}). As it can be seen in Table \ref{tbl2}, the dimensionless central potential and the mass of the satellite decrease for inner positions of the equilibrium satellite for models with the same $r_o$ and $\sigma_o$. Furthermore, the mass of the models is smaller than the mass inferred from observations using kinematic arguments (Ibata et al. \cite{Ibata97}, Mateo \cite{Mateo94}) and it is in agreement with the observed luminous mass, assuming $(M/L)_{\mbox{real}} \sim 2-5$ M$_{\odot}/$L$_{\odot}$. Therefore, we have assumed $(M/L)_{\mbox{real}}=2$ M$_{\odot}/$L$_{\odot}$ for all the s-models, that is a typical value for the stellar population of a DSph. The low value of the mass-to-luminosity ratio is in agreement with satellites either formed in tidal tails of major accretion events or tidally modified by orbiting for a long time in a tidal potential. The apocenter, $r_{max}$, the pericenter, $r_{min}$, and the period of the orbits are listed in Table \ref{tbl2-2}. The s-A orbit is an orbit of low eccentricity in the inner region of the Galaxy and s-B1 and s-B2 orbits have higher eccentricity. \section{Numerical details} \label{numerical} The models have been evolved using the treecode algorithm kindly provided by Barnes \& Hut (\cite{Barnes86}) with a tolerance parameter $\theta=0.7$ and a time-step $\Delta t = 1$ Myr. The number of particles of the luminous and dark halo components of the primary galaxy are 15671 and 29648, respectively, and the mass of the dark matter particles is 3 times larger that for the luminous particles. All the satellite models have 4000 equal-mass particles, except s-B2a and s-B2c which have 8000 particles. We use a softening length varying proportionally to the cubic root of the particle mass of the component, in order to avoid well-known usual numerical effects in the simulations (e.g. Merritt \cite{Merritt96}; Theis \& Spurzem \cite{Theis99}). For the luminous particles of the Galaxy it is $\epsilon_L =0.23$ kpc and for the dark matter halo $\epsilon_{DH} = 0.33$ kpc. The f-models have $\epsilon_S=0.06$ kpc, but this value is changed to $\epsilon_S=0.05$ kpc for the s-A, s-B1 and s-B2b satellites and $\epsilon_S=0.04$ kpc for the s-B2a and s-B2c satellites. \section{Results} \subsection{The main galaxy} The main galaxy develops a bar-like structure, described by Fux (\cite{Fux97}) for an isolated model of our Galaxy. The global effects of the satellite on the primary galaxy are weak, since the mass ratio of both objects is huge. Besides, the poor resolution of the Galaxy model prevents a detailed description of the local effects of the Sagittarius accretion on the Milky Way. Therefore, we only deal with dynamical effects felt by the satellite galaxy. \subsection{f-models} \label{fastsat} For our f-models, we begin the simulations when the satellite has already reached the inner regions of the primary galaxy. A DSph galaxy on a low eccentricity orbit in these inner dense regions of the Galaxy undergoes stronger disruptions than on more eccentric orbits, because the tidal forces are stronger at all positions on the trajectory. \subsubsection{f-A satellite} In Fig. \ref{f2}a (b), we plot the angular distribution of the f-A satellite along (perpendicular to) the orbit (which has eccentricity $e=0.2$) as seen from 24 kpc away (it corresponds to the Solar neighbourhood viewpoint), for four snapshots. In this case, the lifetime of the dwarf galaxy, before significant disruption, is nearly 0.4 Gyrs (Fig. \ref{f3}a). \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f2.ps}} \caption{f-models: mass distribution parallel ({\bf a}, {\bf c} and {\bf e}) and perpendicular ({\bf b}, {\bf d} and {\bf f}) to the orbit as function of the angle $\phi$ to the satellite center, as seen from 24 kpc away, which is the present distance between the Sun and the Sagittarius DSph} \label{f2} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f3.ps}} \caption{Radii which enclose 70\%, 60\%, 50\%, 40\%, 30\%, 20\% and 10\% of the initial mass of the satellite, for f-models} \label{f3} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f4.ps}} \caption{Aitoff projection of the f-A and f-B1 models at the end of the simulation (2.1 Gyr)} \label{f4} \end{figure} The material from the satellite galaxy is tidally stripped, forming a long stream and then, after 1.2 Gyrs, an almost close great circle in an Aitoff projection, which survives for a long time (Fig. \ref{f4}a). Moreover, as we can see in Fig. \ref{f2}b, the mean width perpendicular to the orbit on the sky is $7^o$ (it was only $1.9^o$ at the initial time) and the projected surface brightness (Fig. \ref{f5}a) is 5 mag fainter at the end of the simulation (after 2.1 Gyrs). \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f5.ps}} \caption{Surface brightness $\mu_V$ of the f-models, assuming $(M/L)_{\mbox{real},V}=10$ M$_{\odot}/$L$_{\odot}$} \label{f5} \end{figure} \subsubsection{f-B1 and f-B2 satellites} For both f-B1 and f-B2 orbits ($e=0.64$ and 0.75, respectively), disruption mainly occurs at perigalacticon, because tidal forces are more efficient at small galactocentric radii. The lifetime of a satellite depends strongly on the orbit shape. The f-B1 satellite is destroyed after 0.5 Gyr (Fig. \ref{f3}b) whereas the f-B2 satellite survives for 1 Gyr (Fig. \ref{f3}c) due to the longer period of its orbit, at this time the f-B2 satellite undergoes a close {\it interaction} ($r \sim 2$ kpc) with the center of the primary galaxy and it is tidally destroyed. The final destruction of dwarf galaxies is more efficient in our more eccentric orbits and they present fainter surface brightness at the end of the simulation than the satellite on a low eccentric orbit (Fig. \ref{f5}). The satellite particles of f-B models are spread over all directions and no predominant streams are formed, as it can be seen in Fig. \ref{f4}b for the f-B1 model. In Figs. \ref{f2}d and \ref{f2}f, we have represented the width (perpendicular to the orbit) of the stream. At the end of the simulations, the f-B2 models do not have any predominant peak in the mass distribution perpendicularly to the initial orbit. As a general result, due to the tidal field on the satellite we observe: i) a modification of the internal structure of the satellite for both cases of orbits (eccentric and quasi-circular), with increase of the projected velocity dispersion, and ii) a continuous loss of satellite mass and luminosity. Anyone tempted to infer a $(M/L)_o$ ratio from such experiments by assuming virial conditions would find values $100-200$ times higher than the true ones at the disruption time (Fig. \ref{f6}), confirming the results of K97. \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f6.ps}} \caption{Evolution of the log of the ratio between the mass-to-luminosity relation, $(M/L)_o$, inferred from the velocity dispersion values under dynamical equilibrium conditions, and the real mass-to-luminosity relation, $(M/L)_{\mbox{real}}$, for the f-models} \label{f6} \end{figure} \subsection{s-models} \label{slowsat} In this subsection, we analyze the interaction effects on a DSph in equilibrium with the galaxy potential. Either it could have fallen down slowly from an intermediate region in the denser parts of the primary galaxy, loosing part of its mass and becoming a low DM satellite, or it could have been formed in a tidal tail of a major accretion event. \subsubsection{s-A model} DSphs which are theoretical equilibrium solutions to Sgr at a circular inner orbit are small and low mass galaxies. The tidal field is almost constant along this quasi-circular orbit and, therefore, interaction effects on the satellites in equilibrium are not important. For the s-A orbit, there is a continuous loss of mass, but the satellite has still $\sim 20 \%$ of the initial mass (Fig. \ref{f7}c) after 2 Gyrs and it is still detectable (Fig. \ref{f7}a). The central surface brightness has only changed by 2.5 mag (Fig. \ref{f7}a), evolving from 21.3 mag arcsec$^{-2}$ to 24.8 mag arcsec$^{-2}$. The line-of-sight velocity dispersion grows in the outer parts of the satellite in agreement with findings by K97, remaining almost constant in average in the inner parts ($<0.2-0.3$ kpc) (Fig. \ref{f7}b). We have calculated the mass inferred from the line-of-sight velocity dispersion, measured inside a radius of 0.5 kpc from the satellite center, using Eq. (\ref{ML}). Since the central velocity dispersion and the projected density (Fig. \ref{f7}a) evolve slowly in the central region of the satellite at the beginning of the simulation, the inferred mass-to-luminosity ratio remains almost constant, contrary to the f-models. However, as we have explained above, the real mass of the satellite decreases and it is only $\sim 20 \%$ of the initial value after 2 Gyrs. The structural evolution is slow (Fig.~\ref{f7}c) and it leads to a lower central surface brightness at the end of the simulation and a $(M/L)_o$ ratio calculated from the velocity dispersion (inside 0.5 kpc) which is 5 times higher than the real one (Fig.~\ref{f7}d). This effect could be dramatically increased if the velocity dispersion is measured inside a larger radius, because of the velocity dispersion increase in the outer region of the satellite, leading to a calculated $(M/L)_o$ ratio up to 10 times larger than the real one. Very faint tidal streams are formed, which are spread along $\sim 75-100 \%$ of the orbit with a spatial width $\sim 6^o$ (Fig. \ref{f8}). \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f7.ps}} \caption{s-A model: {\bf a} surface brightness $\mu_V$ assuming $(M/L)_{\mbox{real},V}=2$ M$_{\odot}/$L$_{\odot}$, {\bf b} velocity dispersion of the satellite as a function of the radius as same timesteps as in a), {\bf c} evolution of the radii which enclose from 70 to 10 \% of mass of the satellite from the top to the bottom, and {\bf d} evolution of the mass-to-luminosity ratio measured from the velocity dispersion. The observational values of $\mu$, $\sigma$ are given in Table \ref{Sgr_tbl}} \label{f7} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f8.ps}} \caption{{\bf a} Aitoff projection of the s-A model at the end of the simulation (2.1 Gyr), and mass distribution {\bf b} parallel and {\bf c} perpendicular to the orbit as function of the angle $\phi$ to the satellite center} \label{f8} \end{figure} \subsubsection{s-B models} Perigalacticon and apogalacticon of high eccentric s-B1 and s-B2 orbits decrease with the time. The most dramatic example is the {\bf s-B1 orbit}. The DSph passes 1 kpc from the galaxy center at its third passage at perigalacticon (Fig. \ref{f9}c), because of the orbital energy loss. During this passage the satellite is strongly tidally stripped and it loses most of its mass (Fig. \ref{f9}d). The surface brightness of the satellite decreases with time (Fig. \ref{f9}a) and the velocity dispersion and the mass-to-luminosity ratio increase in the outer parts (Figs. \ref{f9}b and \ref{f9}e). The satellite is finally completely destroyed after nearly 1 Gyr. \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f9.ps}} \caption{s-B1 model: {\bf a} surface brightness $\mu_V$ assuming $(M/L)_{\mbox{real},V}=2$ M$_{\odot}/$L$_{\odot}$, {\bf b} velocity dispersion of the satellite as a function of the radius, {\bf c} distance to the satellite the Galactic Center, {\bf d} evolution of the radii which enclose 70, 60, ..., 10\% of the satellite mass (lines from left to right), and {\bf e} evolution of the mass-to-luminosity ratio measured from the velocity dispersion. The observational values of $\mu$ and $\sigma$ are given in Table \ref{Sgr_tbl}} \label{f9} \end{figure} s-B2 is the most external and eccentric orbit which we have chosen. As the other models, it fulfills at certain times the position and velocity constraints for the Sagittarius DSph orbit in the observational range (Ibata et al. \cite{Ibata97}). In order to investigate this case in more details, we have selected three different satellite models for orbiting on this trajectory (s-B2a, s-B2b and s-B2c). The concentration (defined as $c=\log(r_t/r_o)$) varies from 1.28 (s-B2a) to 0.83 (s-B2c). The concentration determines the fate of the satellite, as well as the evolution of the $(M/L)_o$ ratio. The more concentrated the satellite is, the less variation of the inferred $(M/L)_o$ ratio it suffers (Fig. \ref{f10}). \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f10.ps}} \caption{Evolution of the log of the ratio between the mass-to-luminosity relation, $(M/L)_o$, inferred from the velocity dispersion values, and the real mass-to-luminosity relation, $(M/L)_{\mbox{real}}$, for s-B2 satellites. The parameter $c=\log(r_t/r_o)$ is the concentration of the model} \label{f10} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f11.ps}} \caption{Velocity dispersion, $\sigma(r)$, at four snapshots, for s-B2 satellites} \label{f11} \end{figure} The $(M/L)_o$ ratio in Fig. \ref{f10} has been calculated for the central region of the satellite ($< 0.5$ kpc), where the tidal effects on the mass and the velocity dispersion are weaker. However, even in the most concentrated satellite (s-B2a), we obtain $(M/L)_o$ ratios which are $\sim 10 (M/L)_{\mbox{real}}$, these values rise up to several hundreds for the least concentrated model (s-B2c). It is interesting to notice the strong variations of the inferred $(M/L)_o$ ratio with time. This behaviour is caused by the evolution of the surface mass density when the satellite suffers a close encounter with the core of the primary galaxy. It leads to a successive rearrangement of the internal structure of the satellite. Another important parameter in the $(M/L)_o$ calculations is the angle between the observer-satellite line and the main axis of the satellite. The strong anisotropies of the satellite (tails along the orbit, anisotropic velocity dispersion, etc) could produce various $(M/L)_o$ values (as already suggested by K97 and KK98). The line-of-sight velocity dispersion evolves as shown in Fig.~\ref{f11}. The projected velocity dispersion decreases in the central part of the satellite ($r<0.5$ kpc) and it increases for $r>0.5$ kpc. Thus, the $(M/L)_o$ ratio calculated from Eq.~(\ref{ML}) varies, depending on the limit radius used for the measurement of the {\it central} velocity dispersion (this limit radius is related to the observational resolution). \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f12.ps}} \caption{Radii which enclose 70\%, 60\%, 50\%, 40\%, 30\%, 20\% and 10\% of the initial mass of the satellite, for s-B2 models} \label{f12} \end{figure} {\it s-B2a model:} The {\bf s-B2a model} undergoes a long term evolution. It is the most concentrated model that we have chosen and it survives for at least 10 Gyrs. A mild loss of mass is produced, mainly at perigalacticon: the satellite looses the outer layers of mass, but a core enclosing 30\% of the initial mass subsists (Fig. \ref{f12}a). The limit radius is smaller than the initial one, but the half-brightness radius increases (the final satellite is less concentrated). The central surface brightness decreases (Fig. \ref{f13}a), changing from $\sim 21.0$ mag arcsec$^2$ to $\sim 22.0$ mag arcsec$^2$ (assuming $(M/L)_{\mbox{real}}=2$ M$_{\odot}/$L$_{\odot}$). The final s-B2a satellite is smaller ($r_{hb}=0.13$ kpc) than the observed Sgr DSph ($r_{hb}=0.55$ kpc) and the $(M/L)_o$ ratio is not large enough to reproduce the inferred $(M/L)_o = 50$ M$_{\odot}/$L$_{\odot}$ by Ibata et al. (\cite{Ibata97}). {\it s-B2b model:} The mass loss of the {\bf s-B2b satellite} (Fig. \ref{f12}b) is stronger than that of the s-B2a satellite. The particles of the satellite outer region are stripped, mainly at perigalacticon. The tidal stripped material develops a stream forward and backward from the satellite along the orbit. In Fig. \ref{f13}b, a low surface brightness tidal extension of the satellite can be seen at intermediate epochs. At the final snapshot ($t=10.5$ Gyr), the residual core of the satellite is still detectable, it has $\sim 15 \%$ of the initial mass (Fig. \ref{f12}b) and a central projected surface brightness $\mu_o \sim 23.8$ mag arcsec$^{-2}$ assuming $(M/L)_{\mbox{real}}=2$ M$_{\odot}/$L$_{\odot}$. The initial central surface brightness for the same $(M/L)_{\mbox{real}}$ ratio was $21.3$ mag arcsec$^{-2}$. The stream formed close to the satellite has a surface brightness which is 5.0-6.5 mag fainter than the center of the satellite. This stream is similar to those formed by extra-tidal stars observed close to some DSph satellites. \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f13.ps}} \caption{Surface brightness $\mu_V$ of the s-B2 satellites, assuming $(M/L)_{\mbox{real},V}=2$ M$_{\odot}/$L$_{\odot}$} \label{f13} \end{figure} At some snapshots (i.e. 3.7 Gyr, 4.83 Gyr, 6.2 Gyr, 7.3 Gyr, 8.45 Gyr and 9.6 Gyr), when the tidal streams of the satellite are already formed, the radial velocity, $v_r$, and the gradient of the radial velocity, $|dv/db|$, of the s-B2b satellite and of the stream around the satellite reproduce the observations of Sgr DSph region. The observational values of $v_r$ and $|dv/db|$ of Sgr are given in Table \ref{Sgr_tbl}. However, the values of both quantities in the models are strongly dependent on the position and orientation of the orbit. Therefore, small perturbations in the orbit of the models could lead to differences in $v_r$ of several 10 km s$^{-1}$. In general, the sign of the variation of the radial velocity depends on the position of the satellite along the orbit. For some particular positions, the radial velocity gradient is almost 0 km s$^{-1}$/degree. The s-B2b satellite at time 8.47 Gyr is an example of a model which has a good agreement in position with the observations of Sgr DSph (Fig. \ref{f14}c). From the kinematical point of view, the particles in the stream show a velocity gradient $|dv/db| \sim 4$ km s$^{-1}$/degree (Fig. \ref{f14}a), which is slightly higher than the observed value ($|dv/db| =3$ km s$^{-1}$/degree), whereas the radial velocity is slightly lower. Moreover, this satellite model presents a velocity dispersion ($\sigma \sim 6$ km/s, Fig. \ref{f14}b) lower than the observed value ($\sigma \sim 11.4$ km/s). However, the projected velocity dispersion of the stream remains almost constant and equal to the velocity dispersion in the inner region of the system. Observational data show that regions close to the Sagittarius globular clusters (Terzan 8, Terzan 7, Arp 2) have a similar projected velocity dispersion (Ibata et al. \cite{Ibata97}). \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f14.ps}} \caption{s-B2b satellite, at 8.47 Gyr: {\bf a} Radial velocity, $v_r$, along the orbit (constant galactic latitude), {\bf b} velocity dispersion, $\sigma$, along the orbit and {\bf c} contour density map at the same scale as the a) and b) plots (the star represents the mass center of the satellite, located at $b=-11.1^o$), the contours correspond to $\mu=$ 23.2, 23.6, 24.0, 24.7, 26.8 and 28.7~mag arcsec$^2$ from the center} \label{f14} \end{figure} The density contour map of the s-B2b model at 8.47 Gyr (Fig. \ref{f14}c) resembles the observational map for the region around the center of the satellite (see for example Fig. 1 in Ibata et al. \cite{Ibata97}), in spite of the low resolution (few mass points) of the simulation. However, the center of our satellite has a steeper density profile, the half-brightness radius of the model is $r_{hb}=0.12$ kpc and that of Sagittarius DSph is $0.55$ kpc. Even at the end of the simulation ($t=10.5$ Gyr) the satellite remains too concentrated ($r_{hb}=0.17$ kpc). This discrepancy could mean that either the real Sagittarius DSph has undergone a stronger tidal field, which has caused an effective destruction of the satellite core, or it has suffered tidal disruption for a longer time, or the initial density profile of the Sagittarius satellite was more extended than that of our model. In order to test the latter hypothesis, we have built the s-B2c model, which is less concentrated than s-B2b but more massive. s-B2c follows the same orbit as s-B2b. {\it s-B2c model:} The {\bf s-B2c satellite} survives 5.5 Gyr before disruption. The effects at perigalacticon (Fig. \ref{f12}c) are stronger than those of the other s-B2 models, because of the lower concentration. Low surface brightness trailing and leading streams are formed. At 5.88 Gyr, the residual satellite looks like a long tidal extension with a maximum projected surface brightness $\mu \sim 27.0$ mag arcsec$^{-2}$, assuming $(M/L)_{\mbox{real}} = 2$ M$_{\odot}/$L$_{\odot}$ (Fig. \ref{f13}c). The line-of-sight velocity dispersion evolves as shown in Fig. \ref{f11}c. In the outer region of the satellite it increases up to an almost constant velocity dispersion along the orbit. At the last snapshot of Fig. \ref{f11}c (5.88 Gyr) the satellite is already disrupted and the velocity dispersion of the remnant stream has increased with respect to the value before destruction. The central $(M/L)_o$ ratio obtained from Eq. (\ref{ML}) depends on the position of the satellite along the orbit (Fig. \ref{f10}). The most important variations in the $(M/L)_o$ ratio are observed at the perigalactic passage. At this time the satellite is compressed and stretched, its internal distribution is modified and the tidal extensions are more pronounced. At the final steps of the evolution, the more stripped the satellite is, the higher measured the $(M/L)_o$ is. \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f15.ps}} \caption{s-B2c satellite, at 5.38 Gyr: {\bf a} Radial velocity, $v_r$, along the orbit (constant galactic latitude), {\bf b} velocity dispersion, $\sigma$, along the orbit and {\bf c} contour density map (the star represents the mass center of the satellite, located at $b=-10.7^o$), the solid contours correspond to $\mu=24.3, 25.0, 25.5 26.7, 27.5$ mag arcsec$^{-2}$ from the center and the dashed contour corresponds to $\mu=30$ mag arcsec$^{-2}$} \label{f15} \end{figure} We have analyzed the satellite characteristics at 5.38 Gyr from the beginning of the simulation. It corresponds to the last approach to the galactic disc and after the last perigalacticon, which produces total disruption. At this time, the position ($b=-10.7^o$) and the Galactic velocity ($(U,V,W)\sim(214,0:,170)$ km s$^{-1}$) of the satellite agree with the observational data. In Fig. \ref{f15}, the kinematical distribution (radial velocity and line-of-sight velocity dispersion) along the orbit and the contour density maps are shown. The half-brightness radius, $r_{hb}$, and central surface brightness, $\mu_o$, of the model roughly fit the observations (Table \ref{Sgr_tbl}). The values we obtain in the \hbox{s-B2c} simulation at 5.38 Gyr are $r_{hb} \sim 0.4$ kpc and $\mu_o \sim 24.3$ mag arcsec$^{-2}$, which are slightly lower than observational values, but they evolve to larger half-brightness radius and lower surface brightness as the satellite approaches the Galactic disc and gets more disrupted. For example, at $t=5.565$ Gyr we obtain the best accord between the simulations and the observations for $r_{hb}$ and $\mu_o$ (the values of the model are 0.5 kpc and 25.4 mag arcsec$^{-2}$, respectively), but the position and the velocity do not reproduce the observations ($b\sim 20^o$ and $(U,V,W) \sim (103,0,86)$ km/s). The radial velocity gradient along the orbit of s-B2c satellite depends on the orientation of the trajectory, as in the s-B2b model. At 5.38 Gyr we obtain $|dv/db| \sim 1.5$ km s$^{-1}$/degree for the trailing stream, which is lower than the observed value. However, the radial velocity $v_r$ is in agreement with the observations ($v_r=171$ km s$^{-1}$). The satellite velocity dispersion measured inside 0.5 kpc has decreased ($\sigma \sim 7$ km/s), but it is still large enough to give $(M/L)_o \sim 10 (M/L)_{\mbox{real}}$. \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f16.ps}} \caption{Surface brightness $\mu_V$ of the tidal tail close to the Sgr satellite. Lines represent the s-B2c model at various timesteps. Filled squares are the data (with the errorbars) by Mateo et al. (\cite{Mateo98})} \label{f16} \end{figure} \begin{figure} \resizebox{\hsize}{!}{\includegraphics{f17.ps}} \caption{Tidal tail of the s-B2c satellite: {\bf a} Radial velocity, $v_r$, along the orbit (constant galactic latitude), {\bf b} velocity dispersion, $\sigma$, along the orbit and {\bf c} contour density map (the star represents the mass center of the satellite, located at $b=5.9^o$) and the contours correspond to $\mu = 29.6, 29.9, 30.3, 30.8, 32.0$ mag arcsec$^2$} \label{f17} \end{figure} It is interesting to analyze not only the region around the center of mass of the satellite but also the tidal tails which are formed along the orbit. We have done it for several tidal tails close to the satellite. In Fig. \ref{f16}, we plot the surface brightness of the trailing tail of the s-B2c model at various timesteps, assuming $(M/L)_{\mbox{real}} = 2$ M$_{\odot}/$L$_{\odot}$ (a larger value of $(M/L)_{\mbox{real}}$ moves the lines down in the figure and vice versa). The simulations are compared with the trailing tail observed by Mateo et al. (\cite{Mateo98}) recently. As we can see, except for the region closest to the satellite center ($<12^o$), the simulations reproduce the observations, giving a better result when the satellite is more disrupted. This supports the hypothesis that Sgr DSph is close to its destruction. In our simulations, we also obtain a leading tail similar to the trailing one. Unfortunately, the proximity of the Galactic disc to Sgr center makes difficult to test this result of symmetric tails due to still poor observational data. To give another example of the complexity of the satellite structure, we plot in Fig. \ref{f17} the contour density levels and the kinematic behaviour of a tidal tail, at 5.25 Gyrs from the beginning of the simulation. It looks like the Sgr DSph from the kinematical point of view, having radial velocity (Fig. \ref{f17}a) and velocity dispersion (Fig. \ref{f17}b) similar to Sagittarius observed values ($v_r = 171$ km s$^{-1}$ and $\sigma = 11.4$ km s$^{-1}$). The shape of the tail (Fig. \ref{f17}c) also looks like a tidally disrupted satellite, however, the central surface brightness is $\mu_o \sim 29.5$ mag arcsec$^{-2}$ (fainter than a typical DSph) and the computed $(M/L)_o$ ratio is $200 (M/L)_{\mbox{real}}$ (higher than the estimation for DSph galaxies). Nevertheless, this example illustrates, on one hand, the difficulty of distinguishing between a tidal tails and a DSph galaxy when the latter is close to the disruption and, on the other hand, the possibility of identifying objets (globular clusters, etc) which have been tidally disrupted from a satellite galaxy by measuring their kinematic characteristics. \section{Discussion} Summing up the results obtained in our simulations and comparing them with other published works, we agree with other authors (Vel\'azquez \& White \cite{Velazquez95}; Ibata \& Lewis \cite{Ibata98}) that Sgr DSph has a short period orbit ($T \leq 1$ Gyr). This constraint comes from measurement of the position and the radial velocity of Sgr DSph in a realistic model of the Galaxy. However, we disagree about the DM content. These previous works claim that Sgr DSph is a DM dominated satellite, because, contrary to us, they could not obtain a low mass satellite which survives by orbiting in the potential of the Galaxy. Therefore, they cannot explain the complex evolutionary pattern inferred for the Sagittarius DSph, which has suffered a chemical enrichment and evolution and the age interval of its globular clusters which suggests a long life orbiting in the Galaxy. Ibata \& Lewis (\cite{Ibata98}) have tested low massive models in order to reproduce the Sgr DSph characteristics, reaching an unsuccessful result on the matter and concluding that Sgr DSph must have a large DM content. There are two differences between their study and ours. Firstly, they use a rigid potential to model the Galaxy, so no energy interchange is allowed between the satellite and the Galaxy. This restriction could eventually prevent a readjusting of the internal energy of the satellite in order to reach an equilibrium with the environment. Even if the energy transfer from the satellite to the halo is not important in all cases, we emphasize that our treatment is {\it a priori} adaptable to satellites of various masses. Secondly (and mainly), Ibata \& Lewis (\cite{Ibata98}) do not take into account the tidal potential of the Galaxy as they build their satellite model. We have confirmed that the fate of such satellites is different from that of the satellites in equilibrium with the tidal potential as considered here. Consequently, a more accurate model of the initial satellite, that reflects the true dynamical situation of the DSph, is required in order to avoid spurious effects in the simulations. These differences in the models are responsible for the opposite conclusions reached by Ibata \& Lewis, compared with ours concerning DM. Furthermore, the recent observations of Sgr by Mateo et al. (\cite{Mateo98}) prove the existence of a long trailing tail, which supports the hypothesis that Sgr DSph has suffered strong tidal forces, it is close to disruption and, therefore, it is not in virial equilibrium. The existence of this tail also reduces the inferred $(M/L)_o$ ratio. Our low DM satellites ({\it modified King's models}) establish important restrictions on the satellite formation theory. Thus, the satellite galaxies grow either in a quasi-isolated region (falling slowly on the center of the primary galaxy and having time for readjusting their internal structure to the tidal forces), or inside the main galaxy, in the tidal tails of a major accretion event, which automatically implies equilibrium with the tidal forces at the satellite formation epoch. Furthermore, the survival time of the {\it equilibrium} satellites in the potential of the Galaxy depend on the initial concentration of the DSph (the larger concentration, the longer the life-time). We have obtain models (i.e., s-B2a) which survive more than 10 Gyr. The evolution of these satellites gives rise to tidal streams and modifications on the outer velocity dispersion of the satellite and it leads to high observed mass-to-light ratios if dynamical equilibrium is assumed. We have obtained a rather good qualitative agreement with the observational constraints for the model s-B2c in a time interval (from $5.38$ to $5.7$ Gyr), although this agreement is not achieved simultaneously on all the constraints. We remind that the simulation time does not necessarily correspond with the age of the oldest globular cluster of Sgr and, therefore, the s-B2c model is not necessarily in disagreement with the observations. In this case, two possibilities could be invoked to explain the age of the oldest Sgr globular clusters: (i) either the satellite has been orbiting for a long time in a more external region (where tidal forces are smaller), it suffers dynamical friction or/and a deflection with a dense structure of the MW and it reaches the present orbit where, eventually, it is disrupted, or (ii) it has formed in a tidal tail which contained an old stellar population (Kroupa \cite{Kroupa98c}). The tidal tail models illustrate that a DSph without DM orbiting through central regions of the Galactic potential could preserve its structural parameters for a long time. Later on, the DSph could show high inferred $(M/L)_o$ values as it becomes disrupted. The only restriction of the model is that the satellite must reach the central regions of the Galaxy in equilibrium with the tidal forces. That could be possible whether the satellite has been formed in a tidal tail of an accretion event or it has been slowly accreted from the external parts of the Galaxy. The latter possibility could increased the estimated life-time of the satellite before destruction. Recently, K98 has discussed the parameter space of the DSphs which could survive in a tidal field. In spite of the different initial distribution functions, our s-models seem to be located in the available region of the ($M_{sat}$, $r_{sat}$) space for surviving satellites. However, the main difference between the approaches to the problem is that we build the model of the satellite initially in equilibrium with the tidal forces by solving the Boltzmann equation (G\'omez-Flechoso \& Dom\'{\i}nguez-Tenreiro \cite{Gomez97}) and that assures {\it a priori} the survival of the satellite for a long time. \section{Summary} We have performed simulations of Sagittarius-like satellites to follow the evolution of these systems. The satellite and the primary galaxy are modelled as N-body systems, that prevents possible purely numerical effects which could appear in a non-selfgraviting scheme (i.e. if one of the interacting system is represented by a rigid potential). Two satellite models have been tested: \begin{enumerate} \item First of all, we have used {\it isolated King's models} (three free parameter models) for the satellite, but they are tidally stripped and destroyed in a short interval of time, undergoing a fast evolution. \item In order to correctly model the observed situation we had to introduce the {\it modified King's models} (two free parameters-models), which allow us to build up the distribution function of the satellite, taking into account the tidal force of the environment. These s-models could correspond to no DM dominated satellites. In this framework, the initial concentration of the satellite and the trajectory determine its life-time. Central circular orbits lead to more effective tidal destruction events than highly eccentric orbits. For a given kind of orbit, the most concentrated satellites live longer. The model which better reproduces the observations is the s-B2c model after orbiting 5.38-5.6 Gyr in the potential of the Milky Way. For the same orbits, the s-models survive a longer time orbiting in the Galactic potential than the f-models, despite the s-models having smaller masses (smaller binding energies). \end{enumerate} According to the results presented here, our main conclusions about the evolution of Sgr are: \begin{enumerate} \item Sagittarius may not be a dominated DM satellite. \item It follows an eccentric orbit (perigalacticon and apogalacticon are approximately 15 and 70 kpc). \item It was more concentrated in the past than at the present epoch. \item It could have been orbiting in the Galactic halo for a long time (minimum 5 Gyr). \end{enumerate} As general conclusions about the DSph satellites, we can remark that: \begin{enumerate} \item High values of $(M/L)_o$ found in the literature (Ibata et al. \cite{Ibata95}; Irwin \& Hatzidimitriou \cite{Irwin95}; Mateo \cite{Mateo98b} for a review of the Galactic satellites) could be due to an erroneous use of the virial theorem and not to the presence of DM (as already suggested by K97; KK98). \item The progenitor of these satellite could be either lumps formed in a major accretion event or other dwarf galaxies (maybe dwarf irregulars or dwarf spirals) which fall in the potential well of the Galaxy, lose their gas content and evolve to the equilibrium configuration in the tidal field of the Milky Way, giving rise to the population of DSphs. The finding of intermediate states of this evolution could be a support to the proposed scenario. \item According to some of the present results, some DSphs are able to survive for long times orbiting in the Galactic halo. That should be taken into account for the calculations of the dissolution rate of dwarf galaxies in the halo of the primary galaxy. \end{enumerate} \begin{acknowledgements} We would like to thank D. Pfenniger and D. Friedli for critically reading the paper, as well as the referee for remarks which have led to clarify various points of the paper. M.A. G\'omez-Flechoso was supported by the Fundaci\'on Ram\'on Areces through a fellowship. This work has been partially supported by the Swiss National Science Foundation. \end{acknowledgements}
\section{Introduction} Brane constructions in string theory provide powerful tools for analyzing field theories in diverse dimensions and with varying amounts of supersymmetry~\cite{branes1,branes2}. For a review and references see \cite{review}. More recently the Maldacena conjecture \cite{m1,gubs,w1} added a new relation between the large $N$ limit of conformal field theories on branes and the near horizon geometry of the corresponding black brane solutions. The original conjecture was stated for ${\cal N}=4$ SYM realized on $N$ 3-branes, but subsequently more general examples were discovered. One class of such examples are orbifolds of the ${\cal N} = 4$ configuration \cite{ks,oz} and another includes theories on 3-branes in nontrivial F-theory backgrounds \cite{f1,f2,kap}. All of these constructions give rise to conformal theories with varying amounts of supersymmetry. A third class of theories arises on 3-branes at a conifold singularity \cite{kw,ur,nek,unge,josh}. These ${\cal N}=1$ theories are not conformal at all scales, but flow to a line of conformal fixed points in the infrared. For all these theories the correspondence between the large $N$ field theory and supergravity was studied in some detail. For branes on a conifold it turned out to be useful to have a type IIA description which is related to the IIB configuration via T-duality~\cite{ur}. In this paper we study a superconformal ${\cal N} = 1$ $Sp(N)\times Sp(N)$ gauge theory with matter in the fundamental, bifundamental, and antisymmetric representations. We also discuss a specific deformation which preserves ${\cal N}=1$ SUSY but breaks conformal invariance. The resulting theory has a running gauge coupling and flows to a line of superconformal fixed points in the infrared. For both of these theories we give a IIA brane construction as well as a IIB orientifold construction. The latter description allows us to obtain the supergravity solution that is dual to the large $N$ limit of the conformal field theory. The type IIA description, on the other hand, provides a simple way to determine the gauge group, the matter content, and the superpotential of the theories in question. In most ${\cal N}=1$ theories discussed in the AdS/CFT literature (see e.g. \cite{ks,kw}) the R-current which is the superpartner of the stress-energy tensor can be fixed uniquely by field theory considerations. For the theories we discuss here this is not the case. There is a one parameter family of candidate R-currents, both in the theory with vanishing beta function, and in its deformation which flows to a line of fixed points. Since the R-charges of the fields are not uniquely determined, there is no field theory prediction for the dimensions of the chiral primary operators. On the other hand, once we have a supergravity dual of the large $N$ field theory, we know which gauge boson on AdS is the superpartner of the graviton. If we are able to match field theory operators with supergravity states, we can determine the R-charges of all fields and therefore the dimensions of all chiral primary operators. Although there is no firm field-theoretical prediction for the dimensions of fields in the infrared, for the theory with vanishing beta function the most natural assumption is that all fields have canonical dimensions, i.e., that the theory is finite. This will be born out by the supergravity analysis. In the other case, the theory with a running coupling constant, the correct charge assignment in the infrared is harder to guess. Unfortunately the supergravity analysis in this case is on a considerably less solid footing and depends on circumstantial evidence. Nonetheless our analysis suggests a definite R-charge assignment. It would be interesting to find a field theory explanation for it. The type IIA construction involves D4-branes compactified on a circle as well as NS5-branes, D6-branes, and O6-planes. The gauge theory lives on the D4-branes. Our construction is very similar to the brane configurations that give rise to elliptic ${\cal N}=2$ models \cite{branes2,angfinite,jaemo,kap}. One advantage of the IIA description is that the moduli space of the gauge theory is realized geometrically. The flat directions correspond to motions of the 4-branes. Similarly, relevant perturbations of the field theory, such as masses for the matter fields, are also realized geometrically as motions of the 6-branes. This allows us to identify a 6-brane configuration that gives rise to a superconformal ${\cal N}=1$ theory on the 4-branes with an exactly marginal parameter. We can also identify relevant perturbations of the superconformal 4-brane theory that lead to theories with running coupling constants. There is one particular perturbation that gives rise to a theory that flows to a line of conformal fixed points in the infrared. The moduli space of the perturbed theory has a Coulomb branch. A generic ${\cal N} = 1$ theory with a Coulomb branch has a low energy effective gauge coupling that varies over the moduli space. The theory we are considering in this paper has the special feature that the low-energy effective gauge coupling does not depend on the moduli. This will be relevant when we discuss the supergravity description of these theories. In order to construct the supergravity duals we T-dualize the IIA configuration along the compact direction. This operation turns the D6-branes and the O6-planes into D7-branes and O7-planes. The D4-branes turn into D3-branes probing this background. Similar probe theories were studied in \cite{bds,jhs,aks,ah1}, and their relation to supergravity is described in \cite{f1,f2,kap}. Our IIA configuration turns out to be T-dual to 3-branes probing a local piece of an F-theory compactification \cite{sen1,sen2} which is related to the Gimon-Polchinski model \cite{gp}. The simplest such configuration, consisting of two intersecting O7-planes with four coincident 7-branes on top of each, corresponds to the IIA construction of the superconformal 4-brane theory. In the type IIB construction the Ramond-Ramond (RR) charges of the 7-branes are cancelled locally by the charges of the orientifold planes, so the string coupling is constant. Since the type IIB description is a perturbative orientifold, we can find the supergravity dual of the large $N$ limit of the field theory along the lines of \cite{f1,f2}. Matching the spectrum of primary operators with the KK modes allows us to determine the $U(1)_R$ charges of all fields in the conformal theory unambiguously. The matching of non-chiral primaries exhibits a new interesting feature: we find a short supergravity multiplet whose field theory counterpart becomes short only when $N\to\infty$. We interpret this as the evidence that at higher orders in $1/N$ supersymmetry mixes one-particle and two-particle supergravity states. It should also be possible to find a supergravity description of the infrared limit of the deformed theory. Although this theory is not conformal, it has a constant low-energy effective coupling along the Coulomb branch, so the supergravity dual will have a constant dilaton. To find this dual we need to study the deformations of the backgrounds in IIA and IIB and find an explicit map between them. As mentioned before, this is straightforward on the IIA side, since the deformations correspond to motions of the 6-branes. On the IIB side the situation is more involved. We can analyze the deformations on the IIB side by studying the theory on the 7-branes. The eight-dimensional theory on the 7-branes has six-dimensional matter localized at the intersection of orthogonal 7-branes. We analyze the moduli space of this impurity theory following \cite{kap2}, and find an explicit map between the type IIB and type IIA deformations. Among supersymmetric type IIB deformations there is one that maps to a new IIA brane configuration which involves curving D6-branes in the background of an NS5-brane. The map between deformations also allows us to identify the IIB configuration that gives rise to the non-conformal probe theory with moduli-independent effective coupling. We do not have a complete supergravity description of this theory, but a partial description is possible. It supplies enough information to determine the dimension of all chiral operators in the infrared if we use field theory considerations as well. In section \ref{IIA} we discuss the type IIA construction of the probe theory and list some field theory results that we need in subsequent sections. Section \ref{IIB} contains the T-duality, the analysis of the 7-brane impurity theory, and the map between IIA and IIB deformations. We also briefly discuss the exotic IIA deformation that appears as the counterpart of an ordinary deformation in IIB. In section \ref{sugra} we analyze the large $N$ limit of our field theories and their supergravity duals. We discuss the matching of operators with Kaluza-Klein modes in the conformal case and present a partial analysis in the non-conformal case. \section{The IIA construction of the field theory} \label{IIA} \subsection{The IIA brane configuration} A configuration consisting of D4-branes extending in 01236, D6-branes and O6-planes extending in 0123789, and NS5-branes extending in 012389 preserves four supercharges. We obtain an ${\cal N}=1$ supersymmetric field theory in four dimensions after compactifying $X_6$ on a circle with circumference $2\pi R_6$. Specifically we consider configurations with $N$ D4-branes wrapping the compact $X_6$ direction. We put two O6$^-$-planes at $X_6=0,\pi R_6$ and an NS5-brane and its image at $X_6 = R_6\pi/2,3R_6\pi/2$. In order to cancel the total RR charge, we place four physical D6-branes on the circle. An example of such a configuration is shown in Fig.~\ref{fig1}. These brane configurations are very similar to the configurations that give rise to finite ${\cal N} =2$ theories in four dimensions \cite{angfinite,jaemo,kap}. In fact, the configuration we study here can be obtained from one of the ${\cal N} =2$ configurations in \cite{angfinite} by rotating the NS5-branes from the 45 directions into the 89 directions. This breaks half of the supersymmetries, giving an ${\cal N}=1$ theory in $d=4$. \begin{figure} \centerline{\epsfxsize = 11truecm \epsfbox{fig1.eps} } \tighten{ \caption{ Brane configuration: The vertical dashed lines are the O6-planes, the solid lines are the D6-branes, the horizontal line are the D4-branes, and the point represents the NS5-brane. Only half of the $X_6$ circle is shown.} \label{fig1} } \end{figure} Using standard techniques \cite{review}, we can determine the matter content and the superpotential of the field theory on the 4-branes. Unlike the ${\cal N} =2$ case, the $X_6$ position of the D6-branes will play an important role in our analysis. We need to distinguish two cases that are of interest for the analysis in this paper. Either all 6-branes intersect the NS5-brane, or the four 6-branes are split into two groups of two to the left and right of the NS5-brane (as shown in Fig.~\ref{fig1}). These two choices give rise to physically inequivalent theories. The former configuration yields a line of fixed points (parametrized by the dilaton expectation value) that passes through zero coupling, while the latter corresponds to a non-conformal gauge theory which flows to a line of strongly coupled fixed points. \subsection{The conformal case: a field theory analysis}\label{conformal} \label{iib} The theory on the 4-branes turns out to be an $Sp(2N)_1\times Sp(2N)_2$ gauge theory with matter fields $A_i, i=1,2$ in the antisymmetric representation of each of the gauge groups, two bifundamentals ${\cal Q}, \tilde{\cal Q}$, and fundamentals from the 4-6 strings. The brane configuration, and consequently the field theory, admit a symmetry which exchanges the two $Sp$ factors. To determine the number and flavor representations of the fundamentals we need to understand the classical gauge theory on the 6-branes. Note that the worldvolume of the NS5-brane lies within the worldvolume of the D6-branes. It was argued in \cite{karch} that the 6-branes can break on the NS5-brane (see also \cite{nscft}). The gauge group on the four 6-branes turns out to be $U(4)_u\times U(4)_d$, where the two $U(4)$ factors act on the upper and lower halfs of the 6-branes respectively. One-loop effects break the $U(4)_u\times U(4)_d$ symmetry to $SU(4)_u\times SU(4)_d$~\cite{berkoozetal}. The matter content of the 6-brane theory includes a bifundamental hypermultiplet from strings connecting upper and lower halfs of the 6-branes. We will have more to say about the 6-brane theory when we discuss the deformations of this background. For our present purposes we only need to know that the gauge group of the 6-brane theory is the flavor group of the probe theory. The matter content and the superpotential for a 4-brane probe in this background were worked out in \cite{karch}. The fundamentals transform as $q=(\Yfund,1,{\bf 4},1)$, $\tilde{q}=(\Yfund,1,1,{\bf 4})$, $p=(1,\Yfund,{\bar {\bf 4}},1)$, and $\tilde{p}=(1,\Yfund,1,{\bar {\bf 4}})$ under $Sp(2N)_1\times Sp(2N)_2\times SU(4)_u \times SU(4)_d$. The superpotential reads \begin{equation} W = h_1\tilde{\cal Q}A_1 J_1{\cal Q} - h_1{\cal Q}A_2 J_2\tilde{\cal Q} + h_2q{\cal Q} p + h_2\tilde{p}\tilde{\cal Q}\tilde{q}. \end{equation} Here $J_1$ ($J_2$) is the invariant antisymmetric tensor of $Sp(2N)_1$ ($Sp(2N)_2$). Following \cite{line} it is easy to check that this theory has a line of fixed points passing through weak coupling. The one-loop beta function vanishes and the symmetry between the gauge factors implies that both antisymmetric tensors have the same anomalous dimension $\gamma_A$, both bifundamentals have $\gamma_{\cal Q}$ and all fundamentals have $\gamma_q$. Therefore the beta functions of the gauge coupling and the Yukawa couplings in the superpotential are \begin{eqnarray}\label{beta1} \beta_g &\sim& 2(N-1)\gamma_A+4N\gamma_{\cal Q}+8\gamma_q \nonumber \\ \beta_{h_1}&\sim& 2\gamma_{\cal Q}+\gamma_A\\ \beta_{h_2}&\sim& \gamma_{\cal Q}+2\gamma_q. \nonumber \end{eqnarray} Setting all beta functions to zero gives two independent constraints on the three coupling constants. The remaining coupling constant parametrizes a line of superconformal fixed points. Since setting all anomalous dimensions to zero satisfies the constraints, this line passes through the free point $g=h_1=h_2=0$. Note that requiring the beta functions to vanish does not fix anomalous dimensions unambiguously. The most natural assumption is that the dimensions of the fields are unchanged as one moves along the fixed line. This would mean that the theory is finite. The supergravity computation in the last section supports this conjecture by showing that this is true in the large $N$ limit. The moduli space of this theory includes subspaces where it flows to theories with more supersymmetry. For example, giving an expectation value to either ${\cal Q}$ or $\tilde{\cal Q}$ proportional to a unit matrix gives a mass to half of the fundamentals and breaks the gauge group to the diagonal $Sp(2N)_D$. It is a simple matter to show that the resulting theory flows to an ${\cal N}=2$ superconformal theory with gauge group $Sp(2N)$, one antisymmetric hypermultiplet, and four hypermultiplets in the fundamental. Giving such expectation values to both ${\cal Q}$ and $\tilde{\cal Q}$ makes all flavors massive and breaks the gauge group to $SU(N)$. Part of the bifundamentals are eaten by gauge bosons, and the rest give rise to three chiral superfields in the adjoint of $SU(N)$. This theory flows to ${\cal N} = 4$ SYM in the infrared. These field theory results are reproduced in the brane construction if we identify the positions of the D4-branes with the field theory moduli in the following way: \begin{eqnarray} &X_7 \sim {\cal Q Q}^\dagger - \tilde{\cal Q}^\dagger \tilde{\cal Q} & \\ &X_4+iX_5 \sim {\cal Q}\tilde{\cal Q}.& \end{eqnarray} Giving an expectation value to either of the bifundamentals while keeping the other expectation value zero corresponds to moving the 4-branes in the positive or negative $X_7$ direction. Turning on both bifundamentals corresponds to moving the 4-branes in the $X_4$ and $X_5$ directions as well as $X_7$. The effect of these motions on the 4-brane theory agrees with the field theory expectations. If we move the 4-branes off the NS5-branes in the $X_7$ direction, we can ignore the NS5-brane. The remaining branes preserve eight supercharges, and standard techniques \cite{review} confirm the matter content and gauge group stated above for the ${\cal N}=2$ case. Moving the 4-branes in $X_4$ and $X_5$ amounts to separating them from all other branes. The theory on the 4-branes is then ${\cal N} = 4$ SYM as expected. \subsection{The non-conformal case}\label{iic} We can deform the background for the 4-brane theory by moving the 6-branes in the $X_6$ and $X_{4,5}$ directions. These brane motions are parametrized by expectation values of the two complex scalars, ${\cal M}, \tilde{\cal M}$ in the bifundamental hypermultiplet of the $(1,0)$ theory on the intersection of the 6-branes and the NS5-branes \cite{karch}. More precisely, we relate the positions of the 6-branes to ${\cal M}, \tilde{\cal M}$ as follows: \begin{eqnarray} &X_6 \sim {\cal M M}^\dagger - \tilde{\cal M}^\dagger \tilde{\cal M} & \\ &X_4+iX_5 \sim {\cal M}\tilde{\cal M}.& \end{eqnarray} To obtain the configuration shown in Fig.~\ref{fig1} we have to set ${\cal M}={\rm diag}(m_1,m_2,0,0)$ and $\tilde{\cal M}={\rm diag}(0,0,\tilde{m}_3,\tilde{m}_4)$. These bifundamental expectation values act as mass terms in the 4-brane theory. The corresponding terms in the field theory superpotential are \begin{equation} W = \tilde{\cal M}q\tilde{q} + {\cal M}p\tilde{p}. \end{equation} We will be particularly interested in the case $m_1=m_2=\tilde{m}_3=\tilde{m_4}$. In this case the bifundamental expectation values break the $SU(4)_u\times SU(4)_d$ 6-brane gauge group to $SU(2)_1\times SU(2)_2\times U(1)$. After integrating out the massive components of the fundamentals, the superpotential of the 4-brane theory reads \begin{equation}\label{defW} W = h_1\tilde{\cal Q}A_1 J_1{\cal Q} - h_1{\cal Q}A_2 J_2\tilde{\cal Q} + h_3 q{\cal Q}\tilde{\cal Q}\tilde{q} + h_3 \tilde{p}\tilde{\cal Q}{\cal Q}p. \end{equation} The fundamentals now transform as $q=(\Yfund,{\bf 1},{\bf 2},{\bf 1})$, $\tilde{q}=(\Yfund,{\bf 1},{\bf 2},{\bf 1})$, $p=({\bf 1},\Yfund,{\bf 1},{\bf 2})$, and $\tilde{p}=({\bf 1},\Yfund,{\bf 1},{\bf 2})$ under $Sp(2N)_1\times Sp(2N)_2\times SU(2)_1 \times SU(2)_2$. Actually, the superpotential, \eq{defW}, has an accidental $SO(4)_1\times SO(4)_2$ global symmetry under which $q$ and $\tilde{q}$ transform as a $({\bf 4}, {\bf 1})$ while $p$ and $\tilde{p}$ transform as $( {\bf 1},{\bf 4})$. An analysis along the lines of \cite{line} shows that this theory also has a line of superconformal fixed points. The beta functions are given by \begin{eqnarray}\label{beta2} \beta_g &\sim& 4+2(N-1)\gamma_A+4N\gamma_{\cal Q}+4\gamma_q \nonumber \\ \beta_{h_1} &\sim& 2\gamma_{\cal Q}+\gamma_A\\ \beta_{h_3} &\sim& 1+\frac{1}{2}\gamma_{\cal Q}+\gamma_q. \nonumber \end{eqnarray} Demanding that the beta functions vanish, we again find that two out of the three constraints are independent, leaving us with a line of fixed points. In this case, however, the line does not pass through weak coupling, since at least one of the anomalous dimensions must be nonzero. Again the vanishing of the beta functions alone does not determine the values of anomalous dimensions. In the last section we will argue that supergravity considerations allow us to fix this ambiguity for large $N$ and find $\gamma_A=\gamma_Q=0,\gamma_q=-1$. As in the conformal case we can analyze the RG flows both in field theory and using the brane picture. From the brane construction it is clear that we flow to the same ${\cal N}=2$ theory as in the conformal case if we move the 4-branes off the NS5-brane in the positive or negative $X_7$ direction. Moving the 4-branes in $X_{4,5}$ again yields ${\cal N}=4$ SYM. The analysis in the field theory is a little more involved in this case because the one-loop beta function does not vanish. This implies that there will be threshold effects in the matching of the running gauge coupling. On general grounds one would expect the low-energy effective coupling to depend on the size of the bifundamental expectation values in the field theory. However, if we give arbitrary (non zero) expectation values to ${\cal Q}$ and $\tilde{\cal Q}$, fields get integrated out at a variety of scales. Assuming that the expectation value of ${\cal Q}$ is larger than that of $\tilde{\cal Q}$, the $Sp(2N)\times Sp(2N)$ gauge group is broken to the diagonal group at a scale set by ${\cal Q}$. The diagonal $Sp(2N)_D$ is broken to $SU(N)$ at a scale set by $\tilde{\cal Q}$, and finally the fundamentals are integrated out at scale $h_3 {\cal Q}\tilde{\cal Q}$. Matching the gauge couplings at each of these scales we find that the low-energy effective coupling does not depend on the bifundamental expectation values. This is a special feature of this theory that will be important later on. \section{The type IIB description} \label{IIB} \subsection{T-duality}\label{iia} In this section we describe the IIB configuration which is obtained by T-dualizing the IIA brane configuration of section II along $X_6$. Since $\partial /\partial X_6$ is not a Killing vector, performing this T-duality is not completely trivial. Similar T-dualities on IIA configurations that preserve ${\cal N}=2$ supersymmetry on the 4-branes have appeared in the literature \cite{kap,jaemo}. In the ${\cal N}=2$ case the T-duality maps the two O6$^-$-planes and the four D6-branes to an orientifold 7-plane and four D7-branes. The D4-branes become D3-branes probing this background. The NS5-brane and its mirror image turn into a ${\bf Z}_2$ orbifold acting on the 7-brane coordinates transverse to the D3-brane. The T-dual of the IIA configuration without NS5-branes was analyzed in \cite{bds,ah1}. Our configuration differs from the ${\cal N}=2$ case by the orientation of the NS5-branes. Since this modifies the T-duality considerably we discuss it in some detail here. Our first goal is to T-dualize the NS5-branes and the pair of O6$^-$-planes. The other branes can be added later. We begin by separating the NS5-brane and its image in the $X_{4,5}$ directions. The T-dual of the two NS5-branes is a two-center Taub-NUT space. Recall that the two-center Taub-NUT space can be thought of as a circle fibered over ${\bf R}^3$ so that its radius vanishes at two points on ${\bf R}^3$ (the centers). In the present case ${\bf R}^3$ is parametrized by $X_4,X_5,X_7$, while the coordinate along the circle is T-dual to $X_6$. The positions of the centers correspond to the positions of the NS5-branes in $X_4,X_5,X_7$. In the IIA configuration the orientifold projection ensures that position of the physical NS5-brane and its image are related by a reflection of the $X_{4,5}$ coordinates. The T-dual orientifold projection should therefore impose a similar constraint on the location of the centers of the Taub-NUT. The Taub-NUT metric has the following form \begin{eqnarray} ds^2 &=& \left(\frac{4}{b^2}+\frac{1}{R_+}+\frac{1}{R_-}\right)^{-1} \left[ d\sigma+ \left(\frac{Z_+}{R_+}+\frac{Z_-}{R_-}\right)d\arctan\left(\frac{Y}{X}\right) \right]^2 \nonumber \\ &&+\left(\frac{4}{b^2}+\frac{1}{R_+}+\frac{1}{R_-}\right) \left[ dX^2+dY^2+dZ^2\right], \end{eqnarray} where \begin{equation} Z_\pm = Z \pm Z_0 \qquad R_\pm^2 = X^2+Y^2+Z^2_\pm. \end{equation} The $\bf{R}^3$ base is parametrized by $X,Y,Z$, the two centers are located at $(0,0,\pm Z_0)$, and $\sigma$ is the $4\pi$-periodic coordinate on the circle fiber. The parameter $b$ is the asymptotic radius of the fiber. The reflection of $X_{4,5}$ in the IIA picture map into reflections of $Z$ and one other coordinate of $\bf{R}^3$, say $Y$. We will be interested in the limit when the asymptotic radius of the circle fiber, $b$, becomes infinitely large, while the T-dual circle parametrized by $X_6$ shrinks to zero. In this limit the two-center Taub-NUT space becomes an $A_1$ ALE space, also known as Eguchi-Hanson space. It is useful to change coordinates \cite{prasad} to transform the metric above into the Eguchi-Hanson form: \begin{eqnarray} X &=& \frac{1}{8} \sqrt{r^4-a^4} \sin(\theta)\cos(\psi) \\ Y &=& \frac{1}{8} \sqrt{r^4-a^4} \sin(\theta)\sin(\psi) \\ Z &=& \frac{1}{8} r^2 \cos(\theta)\\ \sigma &=& 2\phi, \end{eqnarray} where $a^2=8Z_0$ and $\psi$ has period $2\pi$. The orientifold-induced projection $(Y,Z)\sim (-Y,-Z)$, implies the identification $(\theta,\psi)\sim (\pi-\theta,-\psi)$ for the angular coordinates. The fixed locus of this identification is a two-dimensional submanifold of the Eguchi-Hanson space which has the topology of a cylinder. Next we want to bring the NS5-brane and its image back to the origin of the $X_{4,5}$ plane in the IIA description, which corresponds to setting $a=0$. For $a=0$ the Eguchi-Hanson metric becomes an orbifold metric on ${\bf C}^2/{\bf Z}_2$. To make this explicit we can introduce two complex coordinates \begin{equation}\label{z12} z_{1,2} = r\exp(i\phi/2)\left(\cos(\theta/2)\exp(i\psi/2)\pm i\sin(\theta/2)\exp(-i\psi/2)\right). \end{equation} In these coordinates the $a=0$ Eguchi-Hanson metric becomes flat. The identification $\psi\to \psi+2\pi$ requires that we identify $(z_1,z_2) \to (-z_1,-z_2)$ as expected for ${\bf C}^2/{\bf Z}_2$. The additional orientifold identification acts on the new coordinates as $(z_1,z_2)\to(z_1,-z_2)$, and acting with both orientifold and orbifold identifications flips the sign of $z_1$. The orientifold projections have two fixed planes, $z_{1,2}=0$, which we identify with two O7$^-$-planes. To summarize, the NS5-brane together with two O6$^-$-planes become, under T-duality, a pair of intersecting O7$^-$-planes with six common directions. Now let us put in D-branes. The four physical D6-branes in IIA are located at $X_4=X_5=0$. Under T-duality they become D7-branes wrapping the circle fiber of the Taub-NUT and located at $Y=Z=0$. In other words, they are wrapped on the invariant cylinder of the orientifold projection. Taking the limit $b\to \infty, a\to 0$ we find that the invariant cylinder develops a neck and becomes a pair of planes $z_1=0$ and $z_2=0$ in ${\bf C}^2/{\bf Z}_2$. Thus the four physical D7-branes must be located on these planes. Recall that these planes are the O7$^-$-planes and therefore have 7-brane charge $-4$. It follows that the 7-brane charge is cancelled between the D7-branes and the orientifold planes, and the IIB dilaton is constant. Finally, T-duality turns the D4-branes into D3-branes extending in 0123. To summarize, the T-dual of the IIA configuration in the limit when the radius of $X_6$ goes to zero consists of an O7$^-$-plane with four coincident D7-branes in 01236789, another O7$^-$-plane with four coincident D7-branes in 01234589 and 3-branes in 0123. We will refer to the 7-branes extending in 01234589 as 7$'$-branes. The orientifold group for this configuration is \begin{equation}\label{ogroup} G = \{ 1, (-1)^{F_L}R_{45}\Omega,(-1)^{F_L} R_{67}\Omega,R_{4567} \}, \end{equation} where $R$ reflects the coordinates indicated and $\Omega$ is the worldsheet parity. The splitting of the D6-branes into half-D6-branes discussed in \cite{karch} becomes obvious after T-duality. Indeed, it follows easily from the above formulas that the location of the upper half 6-branes, $X_4=X_5=0, X_7>0$ in the type IIA configuration maps to the locus $z_2=0$ in IIB. Similarly, the lower halfs of the 6-branes, $X_4=X_5=0, X_7<0$, map to $z_1=0$. Thus the upper halfs of D6-branes map to {\it whole} D7-branes located at $z_2=0$, while the lower halfs map to {\it whole} D7-branes at $z_1=0$. To specify the theory on the 7-branes completely we need to make a consistent choice for the action of the orientifolds on the Chan-Paton factors of the 7-7, 7-7$'$, and 7$'$-7$'$ strings. There are at least two such choices. One gives rise to an $SO(8)\times SO(8)$ gauge symmetry \cite{bz}, and classically the other yields a $U(4)\times U(4)$ gauge group on the 7-branes \cite{sen1,sen2}, which is broken to $SU(4)\times SU(4)$ by one-loop effects~\cite{berkoozetal}. The second case is related to the Gimon-Polchinski \cite{gp} orientifold via T-duality. We will be mainly interested in the second orientifold, which we will refer to as the Sen model. Both of these orientifolds were constructed as compact models with a total of four orientifolds and sixteen physical 7-branes of each kind. The 7-brane gauge groups listed here are the parts of the total 7-brane group that are visible to a 3-brane probe near one of the intersections. The theory on a 3-brane probe in the Sen model background was analyzed in \cite{ah1}. The gauge group, matter content, and the superpotential are in complete agreement with the theory we discussed in section \ref{conformal}. Thus we conclude that the IIA configuration with all 6-branes on top of the NS5-brane is T-dual to a local piece of the Sen model \cite{sen1,sen2}. As in the IIA description the flat directions of the field theory correspond to motions of the 3-branes in the 7-brane background. Moving the 3-branes off the intersection point along either of the O7-planes corresponds to giving en expectation value to one of the bifundamentals ${\cal Q},\tilde{\cal Q}$, and moving the 3-branes off both orientifolds gives an expectation value to both ${\cal Q}$ and $\tilde{\cal Q}$. Separating the 3-branes in the direction which the 7- and 7$'$-branes share corresponds to giving expectation values to the antisymmetric tensors $A_1,A_2$. It is instructive to study the deformations of the Sen model and compare these to the deformations of the corresponding IIA construction. The IIA construction has the advantage that all deformations of the background correspond to moving the 6-branes or the NS5-branes. In the IIB picture only some of the deformations are geometric, others correspond to Wilson lines. Once the map between IIA and IIB deformations is established, we can also find the IIB description of the second (non-conformal) IIA configuration discussed in section \ref{iic}. Sen \cite{sen1,sen2} has studied the deformations of the compact model in great detail. In the compact case the field theory on the 7-branes turns out to be a $(1,0)$ theory in six dimensions. Since our IIB configuration is non-compact, we cannot simply use Sen's results. In fact, in our case the theory on the 7-branes is not even six-dimensional, instead it is an eight-dimensional theory with six-dimensional impurities. Such theories have been discussed previously~\cite{kap2,kapdn}. Before we launch into an analysis of the impurity theory we need to discuss the matter content of the 7-brane theory. A single O7$^-$-plane with four coincident 7-branes gives rise to an ${\cal N}=1$ $SO(8)$ theory in eight dimensions. The bosonic degrees of freedom in the eight-dimensional vector multiplet consist of a vector field and a complex scalar, both in the adjoint of the gauge group. The second O7$^-$-plane in our configuration breaks half of the supersymmetries and imposes projections on fields in the vector multiplet. With the projection matrices for the Sen model \cite{gp,sen1}, the surviving constant modes of the fields are a vector and a complex scalar in the ${\bf 6}+ \bar{\bf 6}$. These fields account for the 7-7 strings and there are similar fields on the 7$'$-branes from 7$'$-7$'$ strings. The 7-7$'$ strings are localized at the intersection of 7- and 7$'$-branes. They yield a single hypermultiplet of the six-dimensional $(1,0)$ theory on the intersection, which transforms as a $({\bf 4},{\bf 4})$ under the (classical) $U(4)_7\times U(4)_{7'}$ gauge group. \subsection{The seven-brane impurity theory} In this section we analyze the supersymmetric vacua of the impurity theory on the 7-branes and compare them with the vacua of the T-dual IIA configuration. We expect the vacuum field configurations to be translationally invariant in the six directions common to the 7- and 7$'$-branes. Focusing now on the 7-branes, we see that we can capture the physics by studying the dependence of the 7-brane fields on the remaining two directions transverse to the 7$'$-branes. The 7$'$-branes and the O7$'$-plane intersect this two-dimensional plane in a point. To set up the impurity theory we use a complex affine coordinate $z$ on the plane and define $A_{\bar z} = \frac{1}{2}(A_1+iA_2)$, where $A_i$ are the two components of the $SO(8)$ gauge field living on the 7-branes. The 7-brane theory also contains a complex scalar, $\Phi$, in the adjoint of $SO(8)$ that describes the transverse fluctuations of the 7-branes. The bifundamental $({\cal M},\tilde{\cal M})$ from the 7-7$'$ strings is localized at the point $z=0$. A very similar theory (without orientifold projections) was described in \cite{kap2}. The moduli space of the impurity theory is given by the solution of the equations \begin{eqnarray}\label{imp} &F_{z\bar{z}}-[\Phi,\Phi^\dagger] = \delta(z)\left({\cal MM}^\dagger -\tilde{\cal M}^\dagger\tilde{\cal M}\right)&\nonumber \\ &\bar{D}\Phi=-\delta(z) {\cal M}\tilde{\cal M},& \end{eqnarray} where $F_{z\bar{z}}= \partial A_{\bar z} - {\bar \partial}A_z-[A_z,A_{\bar z}]$ and ${\bar D}={\bar\partial}-A_{\bar z}$. These equations are known as Hitchin equations with sources. They are analogous to the $D$ and $F$ flatness conditions in ordinary supersymmetric field theories. A similar set of equations describes the impurity theory on the 7$'$-branes. To make contact with the notation in \cite{sen1,sen2} we write all 7-brane fields as antisymmetric $8\times 8$ matrices with certain constraints on the entries. This reflects the origin of the fields in the impurity theory. Without the O7$'$-plane, both $A_{\bar z}$ and $\Phi$ would transform in the adjoint of $SO(8)$. Orientifolding with O7$'$ puts additional constraints on these fields \begin{eqnarray}\label{constr} \Phi(z) &=& P\Phi^T(-z)P^{-1} \nonumber \\ A(z) &=& PA^T(-z)P^{-1}, \end{eqnarray} where \begin{equation} P = \left( \begin{array}{cc} P_4 & 0 \\ 0 & -P_4 \end{array} \right), \qquad P_4 = \left( \begin{array}{cc} 0 & {\bf 1}_{2\times 2} \\ -{\bf 1}_{2\times 2} & 0 \end{array} \right). \end{equation} Orientifolding also breaks the gauge group from $SO(8)$ down to the group of all continuous $SO(8)$-valued functions satisfying $g(z)=P g(-z) P^{-1}$. In particular, at $z=0$ the gauge group reduces to $U(4)$. The orientifold projections allow the bifundamentals to be arbitrary complex $8\times 8$ matrices that commute with $P$ \cite{gp}. The impurity equations are consistent if the products of the bifundamentals on the r.h.s.~of \eq{imp} are antisymmetrized in the gauge indices. We need to find all, possibly $z$-dependent, field configurations that satisfy the impurity equations, \eq{imp}, modulo gauge transformations. To this end we make the following ansatz \begin{equation} A_{\bar z} = \frac{T}{z}, \qquad \Phi(z) = \Phi_0 + \frac{\Phi_s}{z}. \end{equation} Here $T, \Phi_0$ and $\Phi_s$ are constant antisymmetric $8\times 8$ matrices. Imposing the constraints, \eq{constr}, determines that $\Phi_0$ transforms in the ${\bf 6}+\bar{\bf 6}$ of $U(4)$ while $T$ and $\Phi_s$ transform as adjoints. The background gauge field, $A_{\bar z}$, can be interpreted as a flat connection that gives rise to a Wilson line around the intersection point at $z=0$. The constant part of the scalar field, $\Phi_0$, corresponds to the asymptotic (i.e. $z\to\infty$) positions of the 7-branes in the directions transverse to the O7-plane, while the singular part, $\Phi_s$, parametrizes a deformation of the shape of the 7-branes. The moduli space of the impurity equations, \eq{imp}, has several branches with rather different physics. The simplest situation arises if all bifundamental expectation values and all singular parts of $A_{\bar z}$ and $\Phi$ vanish. In that case \eq{imp} reduces to the condition \begin{equation} [\Phi_0,\Phi_0^\dagger] = 0, \end{equation} which is solved by \begin{equation}\label{six} \Phi_0 = \left( \begin{array}{cc} 0 & \phi \\ -\phi & 0 \end{array}\right), \qquad \phi = {\rm diag}(\phi_1,\phi_2,\phi_1,\phi_2). \end{equation} As in Ref.~\cite{sen1}, the two complex parameters, $\phi_{1,2}$, parametrize the transverse position of two pairs of 7-branes. We discuss the corresponding IIA deformation in the next section. For the remainder of this section we set $\Phi_0=0$. The impurity equations, \eq{imp}, become inhomogeneous once we turn on an expectation value for the bifundamental fields. Since ${\bar\partial}(1/z) \sim \delta(z)$, and the r.h.s.~of \eq{imp} is proportional to $\delta(z)$, the singular fields above have the right form to satisfy the impurity equations with nonzero bifundamental expectation values. The most generic expectation value of the bifundamentals for which the impurity equations have solutions reads \begin{eqnarray} &{\cal M} = \left(\begin{array}{cc} M_1 & 0 \\ 0 & M_2 \end{array} \right),&\\ & M_1 = \left( \begin{array}{cccc} m_1 & 0 & -im_1 & 0 \\ 0 & m_2 & 0 & -im_2 \\ im_1 & 0 & m_1 & 0 \\ 0 & im_2 & 0 & m_2 \end{array} \right), \qquad M_2 = \left( \begin{array}{cccc} m_3 & 0 & im_3 & 0 \\ 0 & m_4 & 0 & im_4 \\ -im_3 & 0 & m_3 & 0 \\ 0 & -im_4 & 0 & m_4 \end{array} \right),&\nonumber \end{eqnarray} and an expectation value of the same form, but with $m_i$ replaced by $\tilde{m}_i$, for $\tilde{\cal M}$. The impurity equations determine the expectation values of the other fields in terms of ${\cal M}$ and $\tilde{\cal M}$. The residue of $\Phi$ is given by \begin{equation}\label{x45} \Phi_s = {\rm diag}(\Phi_1,\Phi_2), \end{equation} where \begin{equation} \Phi_1 = \left( \begin{array}{cccc} 0 & 0 & -\phi_1 & 0 \\ 0 & 0 & 0 & -\phi_2 \\ \phi_1 & 0 & 0 & 0 \\ 0 & \phi_2 & 0 & 0 \end{array} \right), \qquad \Phi_2 = \Phi_1(\phi_1\to-\phi_3,\phi_2\to -\phi_4), \end{equation} with $\phi_i \sim m_i\tilde{m}_i$. The matrix $T$ in the gauge connection has the same structure as $\Phi_s$, except that $\phi_i$ is replaced by $t_i \sim |m_i|^2 - |\tilde{m}_i|^2$. Before discussing this general solution, we will focus on two special cases. If we set $m_i = \tilde{m}_i$, the r.h.s.~of the first impurity equation vanishes and only the residue of $\Phi$ is turned on. This expectation value of the bifundamentals breaks the $U(4)\times U(4)$ gauge group to a diagonal subgroup. If all $m_i$ are equal this subgroup is $U(4)_D$, and for generic values of $m_i$ we find $U(1)^4$. Since the 7-brane group is broken to a diagonal subgroup, the impurity theory, \eq{imp}, on the 7-branes and the corresponding impurity theory on the 7$'$-branes contain the same information. Therefore it is sufficient to consider only the 7-brane theory. The field $\Phi(z)$ describes the shape of the 7-branes. For large $z$ the 7-branes asymptote to the O7-plane as in the unperturbed case, while they approach the O7$'$-plane for small $z$. Thus we conclude that turning on this bifundamental expectation value deforms pairs of intersecting 7- and 7$'$-branes into a single smooth 7-brane that interpolates between the 7- and 7$'$-branes. This result agrees with the F-theory analysis in \cite{sen2}, where this behavior was interpreted as fusing the 7- and 7$'$-branes together. There are also solutions of the impurity equations with non-zero gauge connection and $\Phi_s=0$. We find one such solution if we set $m_1 = m_2=\tilde{m}_3=\tilde{m}_4$, and all other components of the bifundamentals vanish. For this choice the r.h.s.~of the second equation in \eq{imp} vanishes, which implies $\Phi_s=0$, and $t_1=t_2 \sim |m_1|^2$, $t_3=t_4 \sim -|m_1|^2$. This bifundamental expectation value breaks the $U(4)_7\times U(4)_{7'}$ 7-brane gauge group to a diagonally embedded $U(2)\times U(2)$. Note that this deformation is purely non-geometric. Since $\Phi(z)=0$, the 7-branes have the same shape as in the case without any bifundamental expectation values. It is now a simple matter to identify these two singular solutions with the corresponding deformations in the IIA construction. The first solution with $T=0$, $\Phi_s \neq 0$ corresponds to moving the 6-branes off the NS5-brane in the $X_{4,5}$ direction. If none of the 6-branes coincide, the $U(4)\times U(4)$ gauge symmetry on the 6-branes is broken to $U(1)^4$. This is in complete agreement with the impurity analysis. Note that a deformation that corresponds to fusing 7 and 7$'$-branes together in the IIB description maps into a simple brane motion in the IIA construction, which involves reconnecting the upper and lower halfs of the 6-branes. The second singular solution with $T\neq 0$, $\Phi_s = 0$ also corresponds to a simple brane motion in the IIA description. We identify turning on $m_1$ with the motion of two pairs of 6-branes in the $X_6$ direction. The classical gauge group on the 6-branes is $U(2)\times U(2)$ as expected from the IIB analysis. This brane motion also requires that we reconnect the upper and lower halfs of the 6-branes, so that the resulting 6-brane group is a diagonal subgroup of the original $U(4)\times U(4)$ gauge symmetry. This is in perfect agreement with the analysis of the 7-brane impurity theory. It is straightforward to discuss more general choices for the bifundamental expectation values. The bifundamental expectation values are parametrized by eight complex numbers, $m_i$ and $\tilde{m}_i$, which determine the matrices $T$ and $\Phi_s$ completely. The four parameters in $T$ map into the $X_6$ position of the 6-branes in the IIA description and the entries in $\Phi_s$ correspond to the $X_{4,5}$ positions. Thus we find complete agreement between the brane motions in the IIA description and the moduli corresponding to singular fields in the impurity theory. \subsection{A supersymmetric IIA configuration with curving six-branes} The deformations we discussed so far are rather complicated in the IIB picture and correspond to simple brane motions in the IIA description. In fact, all simple brane motions in the IIA description are accounted for. However, there is a very simple brane motion in IIB, namely the constant solution of the impurity equations given in \eq{six}, that should have a counterpart in the IIA description. Since this deformation corresponds to moving pairs of 7-branes off the orientifold, we can find an explicit equation describing the position of these branes. In terms of the coordinates in \eq{z12} this equation reads $z_2 = const$. Starting from this expression we can reverse the coordinate transformations that took us from the Taub-NUT space to the flat coordinates on ${\bf C}^2/{\bf Z}_2$. This provides an expression for the world volume of the 7-brane in the Taub-NUT coordinates. Since the 7-branes wrap the fiber of the Taub-NUT and the fiber T-dualizes to the compact $X_6$ direction, it is straightforward to find the equation for the worldvolume of the corresponding 6-brane. The result is $X_4^2-c X_7-c^2/4=0$, i.e., a parabola in the $X_4-X_7$ plane. Fig.~\ref{fig2} shows the IIA configuration which is T-dual to the following IIB situation: all 7$'$-branes are coincident with the O7$'$-plane, and one pair of 7-branes is displaced from the O7-plane. \begin{figure} \centerline{\epsfxsize = 5truecm \epsfbox{fig2.eps} } \tighten{ \caption{ Type IIA configuration for non-zero expectation value of the {\bf 6}. The dot represents the NS5-brane, the thick line corresponds to four half 6-branes, the thin line corresponds to two half 6-branes and the curving line is another 6-brane. } \label{fig2} } \end{figure} From this picture one can see that turning on the constant complex scalar on the 7-brane corresponds to fusing two upper halfs of the 6-branes together and moving them off the NS5-brane as shown in the figure. On the IIB side it is obvious that this deformation preserves all supersymmetries. This is somewhat harder to see on the IIA side. Presumably the $H$-field produced by the NS5-brane stabilizes the curved worldvolume of the D6-brane. The effect of this deformation on the probe theory is what we expect from the IIB picture. There we move two 7-branes away from the 3-branes sitting at the intersection point of the orientifold planes. This gives a mass to half of the fundamentals from 7-3 strings. In the IIA picture the deformation accomplishes the same. In the IIB picture moving the 3-branes along the O7$'$-plane and transverse to the O7-plane corresponds to giving the bifundamental field ${\cal Q}$ in the probe theory an expectation value \cite{ah1}. Thus it is possible to move the 3-branes away from the intersection of the orientifolds towards the intersection of the pair of 7-branes with the O7$'$-plane by giving an expectation value to one of the bifundamentals. This is also reflected in the IIA description. We can move the 4-branes in the negative $X_7$ direction by giving an expectation value to one of the bifundamentals (see section \ref{iib}). This moves the 4-branes off the NS5-brane and towards the intersection of the lower half-6-branes with the curving 6-brane. In the IIB description moving a pair of 7-branes away from the O7-plane breaks the 7-brane gauge group from $SU(4)$ down to $SU(2)\times SU(2)$~\cite{sen1,sen2}. Moving all four 7-branes together breaks $SU(4)$ down to $Sp(4)$. This implies that the unbroken gauge group on a single curving 6-brane should be $SU(2)$, while for two coincident curving branes it should be enhanced to $Sp(4)$. It is not at all clear how to see this from the IIA description. \subsection{Comparison with F-theory} Sen argued \cite{sen1,sen2} that the T-dual version of the GP model \cite{gp} is related to an F-theory compactification with certain fluxes through collapsed 2-cycles. The naive candidate for such an F-theory compactification would be a pair of intersecting $D_4$ singularities. However, this cannot be directly related to the GP orientifold, since it would give rise to an $SO(8)\times SO(8)$ gauge symmetry and contain tensionless strings, while the GP model has $SU(4)\times SU(4)$ symmetry and no tensionless strings. The difference is due to NS (and possibly RR) 2-form fluxes through the collapsed 2-cycle at the intersection of the two $D_4$ singularities. These fluxes give a mass to 3-branes wrapping this cycle, thereby preventing the appearance of tensionless strings. These fluxes are not quantized \cite{sen1,sen2}, so we should be able to identify moduli in our IIA description that correspond to turning them off. The NS flux is conventionally identified with the position of the NS5-branes on the $X_6$ circle and the RR flux parametrizes the location of the NS5-branes on the M-theory circle. From the IIB point of view, they are both part of a massless hypermultiplet living at the intersection of the $D_4$ singularities. In order to turn off the NS flux, we move the NS5-brane and its image as well as all D6-branes to coincide with one of the O6-planes. This configuration has an $SO(8)\times SO(8)$ gauge symmetry from the eight upper and eight lower halfs of the 6-branes, as well as tensionless strings from the NS5-brane coincident with its image \cite{hz}. In addition to the hypermultiplet that corresponds to moving the NS5-brane off the orientifold in the $X_4,X_5,X_6,X_{10}$ there is now a tensor multiplet whose scalar expectation value corresponds to separating the two NS5-branes in the $X_7$ direction. All this agrees with the expectations from F-theory. \section{The large $N$ limit} \label{sugra} When the number of D3-branes, $N$, is large there is a dual description of ${\cal N}=1$ superconformal theory on the D3-branes in terms of a supergravity on $AdS_5\times X$, where $X$ is an Einstein manifold (or orbifold)~\cite{m1}. This dual description is valid when the t'Hooft gauge coupling, $g_{YM}^2 N$, is large. In this section we will show how the AdS/CFT correspondence works for the conformal gauge theory with $SU(4)\times SU(4)$ flavor symmetry discussed in section \ref{iib}, and provide evidence that this theory is finite. We will also provide a partial analysis of the non-conformal theory of section \ref{iic} in the large $N$ limit and argue that supergravity suggests a definite R-charge assignment for all the fields in the infrared. \subsection{The conformal case} In the conformal case, $X$ is an orientifold of ${\bf S}^5$. As explained in the previous section, the IIA configuration with $SU(4)\times SU(4)$ gauge symmetry on the 6-branes is T-dual to a local piece of the Sen model. At the $SU(4)\times SU(4)$ point, the Sen model is a perturbative type IIB orientifold with constant string coupling, $\tau$ \cite{sen1,sen2}. Thus the near-horizon geometry of the 3-branes is obtained by orientifolding $AdS_5\times {\bf S}^5$. Similar theories were analyzed in \cite{f1,f2,kap}. Let us denote the orientifolded five-sphere by ${\tilde{\bf S}}^5$. The metric on ${\tilde{\bf S}}^5$ is the angular part of \begin{equation} ds^2 = |dz_1|^2 + |dz_2|^2 + |dw|^2, \end{equation} where $w=X_8+iX_9$ and the variables $z_1,z_2$ are subject to the identifications $z_i \to -z_i$. A $U(1)^3$ subgroup of the $SO(6)$ isometry group of ${\bf S}^5$ commutes with these identifications. It is convenient to take the generators that rotate $z_1$, $z_2$,and $w$ separately as a basis in the Lie algebra of $U(1)^3$. Explicitly, the metric on ${\tilde{\bf S}}^5$ can be written as \begin{equation} ds^2_{{\tilde{\bf S}}^5} = d\theta_1^2+\sin^2(\theta_1)d\phi_1^2 + \cos^2(\theta_1) \left( d\theta_2^2+\sin^2(\theta_2)d\phi_2^2+\cos^2(\theta_2)d\phi_3^3\right), \end{equation} where $\phi_{1,2} \in [0,\pi]$, $\phi_3 \in [0,2\pi]$, and $\theta_{1,2} \in [0,\pi]$. The three angles $\phi_i$ parametrize rotations in the $z_{1,2}$ and $w$ planes respectively. The periodicity of $\phi_{1,2}$ reflects the identifications on $z_{1,2}$. Since this periodicity of $\phi_{1,2}$ is the only thing which distinguishes ${\tilde{\bf S}}^5$ from ${\bf S}^5$, the eigenvalues of the scalar Laplacian on the former can be obtained from those on the latter. The eigenvalue of the scalar Laplacian on ${\bf S}^5$ is $k(k+4)$, where $k=0,1,\ldots$. In terms of the angular momenta, $m_i$, associated with the angles $\phi_i$, we have $k = |m_1|+|m_2|+|m_3|+2l_1+2l_2$, where $l_i$ are non-negative integers. The orientifold projection on the bulk supergravity states amounts to keeping modes with even $m_1$ and $m_2$. In the ${\cal N}=4$ case, the supergravity states with lowest mass squared come from the KK reduction of $h^a_a$, the dilaton mode of the ${\bf S}^5$. The AdS masses of these states are given by $m^2 = k(k-4)$ \cite{romans}, where $k$ is given above. According to \cite{w1}, the AdS mass of a KK state is related to the dimension of the corresponding boundary operator by $\Delta(\Delta-4)=m^2$, which implies $\Delta=k$ for this tower of KK modes. The decomposition of the other supergravity fields yield towers of KK states for which $\Delta = k+n$, where $n$ is a positive integer \cite{w1}. We will see below that only for $n=0$ the KK states couple to chiral primary operators. Therefore we will restrict our analysis to the KK modes from the decomposition of $h^a_a$. The simplest way to identify chiral primaries is to find all states for which $\Delta=\frac{3}{2}R$, where $R$ is the R-charge which is part of the superconformal algebra. The R-current is a certain linear combination of the three $U(1)$ currents. To find this linear combination we first need to determine which supercharges survive the orientifold projection. The orientifold group, ${\bf Z}_2\times {\bf Z}_2$, is generated by $\gamma_1=R_{z_2}\Omega (-1)^{F_L}$ and $\gamma_2=R_{z_1z_2}$. Orientifolding by the first generator breaks $SO(6)$ down to $SU(2)_L\times SU(2)_R \times U(1)_N$ where $U(1)_N$ acts on $z_2$ while $SU(2)_L\times SU(2)_R$ acts on $z_1,w$. The surviving supercharges $(Q_+,Q_-)$ transform as $({\bf 1},{\bf 2})_1$ with respect to this group. Orientifolding by $\gamma_2$ breaks $SU(2)_L\times SU(2)_R$ down to $U(1)_L\times U(1)_R$. We will denote the sum of the $U(1)_L$ and $U(1)_R$ charges by $U(1)_2$, the difference by $U(1)_3$, and refer to $U(1)_N$ as $U(1)_1$. The charges of $z_2, z_1,$ and $w$ under these three $U(1)$'s are given by $(2,0,0),(0,2,0),$ and $(0,0,2)$, respectively. The supercharge $Q_+$ which survives the second orientifolding has $U(1)$ charges $(1,1,-1)$. It follows that the R-charge which is in the same superconformal multiplet as the stress-energy tensor is $\frac{1}{3}(2m_1+2m_2-2m_3)$. Here $2m_1$ is the $U(1)_1$ charge, $2m_2$ is the $U(1)_2$ charge, and $2m_3$ is the $U(1)_3$ charge. The normalization is chosen so that $Q_+$ has R-charge $1$. It follows that any KK mode with $l_1=l_2=0$, $m_1,m_2 \ge 0$ and $m_3 \le 0$ should couple to a chiral primary operator in the boundary field theory. We discussed the identification of geometric motions of 3-branes with flat directions in the 3-brane field theory in the previous section (see also \cite{ah1}). This allows us to determine the $U(1)$ charges of the fields $A_1,A_2,{\cal Q},\tilde{\cal Q}$. The field theory superpotential then fixes the R-charges of the fundamentals $q,{\tilde q},p,{\tilde p}$. The results are summarized in the table below. \begin{equation}\label{table} \begin{array}{c|ccc} & U(1)_1 & U(1)_2 & U(1)_3 \\ \hline A_{1,2} & 0 & 0 & -2 \\ {\cal Q} & 2 & 0 & 0 \\ \tilde{\cal Q} & 0 & 2 & 0 \\ q,p & 0 & 1 & -1 \\ \tilde{q},\tilde{p} & 1 & 0 & -1 \end{array} \end{equation} With these charge assignments in hand it is now a simple matter to match the bulk KK modes and the chiral primary operators in the field theory. Let us give some examples. The supergravity spectrum contains a singleton chiral primary with $U(1)_3$ charge $-2$ and $\Delta =k=1$. This state corresponds to a chiral primary ${\rm Tr} (A_1 J_1) + {\rm Tr} (A_2 J_2)$ in the field theory.\footnote{The antisymmetric representation of $Sp(N)$ is reducible and contains a singlet.} Since $\Delta=1$, this is a free field. For $\Delta = 2$ there are three chiral primary states with geometric $U(1)$ charges $(4,0,0)$, $(0,4,0)$, and $(0,0,-4)$. We identify them with ${\rm Tr}{\cal Q}^T J_1 {\cal Q} J_2$, ${\rm Tr}\tilde{\cal Q}^T J_2 \tilde{\cal Q} J_1$, and ${\rm Tr} (A_1 J_1)^2 + {\rm Tr} (A_2 J_2)^2$. The chiral primary operators with $\Delta = 3$ are ${\rm Tr}[{\cal Q} A_2 {\cal Q}^T J_1+{\cal Q}^T J_1 A_1 J_1 {\cal Q} J_2]$, ${\rm Tr}[\tilde{\cal Q} A_1 \tilde{\cal Q}^T J_2+ \tilde{\cal Q}^T J_2 A_2 J_2 \tilde{\cal Q} J_1]$, and ${\rm Tr} (A_1 J_1)^3 + {\rm Tr} (A_2 J_2)^3$. They correspond to the KK states with charges $(4,0,-2)$, $(0,4,-2)$, and $(0,0,-6)$ respectively. The field theory also contains operators that carry charges under the 7-brane gauge groups. It was pointed out in \cite{f2} that these operators couple to the AdS modes coming from the KK reduction of the 7-brane fields. Our configuration includes an O7-plane with four coincident D7-branes wrapping an ${\bf S}^3$ defined by $|z_1|^2+|w|^2=const$, and similarly an O7$'$-plane with four D7$'$-branes wrapped on $|z_2|^2+|w|^2=const$. The two 3-spheres intersect over a circle. We can focus on the KK modes from the first ${\bf S}^3$. These modes couple to operators that are charged under the $SU(4)_7$ subgroup of the $SU(4)_7\times SU(4)_{7'}$ global symmetry group of the probe theory. The modes living on the other ${\bf S}^3$ couple to similar operators in the field theory that transform under $SU(4)_{7'}$. The KK reduction of the theory on an O7-plane with four coincident 7-branes was discussed in \cite{f2}. In that case there were twice as many supersymmetries as in ours. The simplest way to compute the KK spectrum in our case is to use the results of \cite{f2} and impose the additional projection from the O7$'$-plane. Ref.~\cite{f2} contains a detailed discussion of the 7-brane states and their multiplet structure. The lowest component of the multiplet is a real field in the $({\bf k}, {\bf k+2})_0$ representation of $SU(2)_L\times SU(2)_R\times U(1)_N$, where $k=1,2,\ldots$. This mode comes from KK reduction of the components of the 7-brane gauge field along the ${\bf S}^3$, \begin{equation} A_a = \sum_k a_k Y_a^k, \end{equation} where $Y_a^k$ is the $k$-th vector spherical harmonic on ${\bf S}^3$. These modes couple to operators of dimension $\Delta = k+1$ in the boundary field theory. For simplicity we will only consider operators with $\Delta = 2,3$. The state with $\Delta=2$ transforms in the $({\bf 1},{\bf 3})_0$ and decomposes into modes with $U(1)^3$ quantum numbers $(0,0,0)$ and $(0,\pm 2, \mp 2)$. The $(0,0,0)$ mode has no $U(1)_R$ charge and does not correspond to a chiral primary. The states with $U(1)^3$ charges $(0,2,-2)$ and $(0,-2,2)$ are complex conjugates of each other, so it is sufficient to consider only one of them, e.g., the first. It has R-charge $4/3$ and is, therefore, a chiral primary. This state starts out in the adjoint of the $SO(8)_7$ gauge group on the 7-brane. Since it has $m_2=1$, it is odd under the additional orientifold projection $\gamma_2$. This projection breaks $SO(8)_7$ down to $SU(4)_7$. As explained in~\cite{sen1,sen2}, states in the adjoint of $SO(8)_7$ which are odd under $\gamma_1$ yield ${\bf 6}+\bar{\bf 6}$ of $SU(4)_7$, while even states give adjoints of $SU(4)_7$. It follows that the $(0,2,-2)$ state yields one complex state in ${\bf 6}$ and one complex state in $\bar{\bf 6}$. These KK states correspond to operators $qJ_1q$ and $pJ_2p$, which transform in the $\bf 6$ and $\bar{\bf 6}$ of the 7-brane group respectively. The $\Delta=3$ mode is in the $({\bf 2},{\bf 4})_0$ representation and decomposes into even modes with $U(1)^3$ charges $(0,0,-2)$ and $(0,4,-2)$ and their complex conjugates, as well as odd modes with $U(1)^3$ charges $(0,2,0)$ and $(0,2,-4)$ and their complex conjugates. The even $(0,0,-2)$ mode and the odd $(0,2,0)$ mode do not couple to chiral primary operators, because the R-charge does not match the dimension. The even $(0,4,-2)$ mode couples to a chiral primary operator in the adjoint of $SU(4)$ which we identify as $pJ_2\tilde{\cal Q}J_1q$. The odd $(0,2,-4)$ mode couples to a chiral primary in the ${\bf 6}+\bar{\bf 6}$. The corresponding operators are given by $qA_1q$ and $pJ_2A_2J_2p$. Other scalars on AdS come from the decomposition of the complex scalar field on the 7-branes. These KK modes are in the $({\bf k},{\bf k})_2$ representation of $SU(2)_L\times SU(2)_R\times U(1)_N$ and couple to operators of dimension $k+2$~\cite{f2}. It is straightforward to decompose and project these modes as we did for the KK modes of the vector field. The $\Delta=3$ case is especially simple, since this mode carries only $U(1)_1$ charge. Since the R-charge and the dimension do not satisfy $\Delta=\frac{3}{2} R$, this KK mode does not couple to a chiral primary operator. The same is true for the higher KK modes of the complex scalar field. Finally, there are also states living on the intersection of the 7-branes and 7$'$-branes which is an ${\bf S}^1$ embedded in ${\bf S}^5$. The KK reduction of these states is straightforward, and we will not discuss it. In the above analysis we have focused on chiral primaries. It is also interesting to ask whether non-chiral states match between field theory and supergravity. Some of the non-chiral scalars we have seen, namely the ones coming from the reduction of complex scalars living on the 7-branes, are descendants of the chiral primaries and therefore match automatically. On the other hand, the non-chiral scalars which come from the KK reduction of the gauge field on the 7-branes are primary. One may ask whether the superconformal multiplet they live in is long or short. To answer this question we need to recall some facts about unitary representations of the ${\cal N}=1$ superconformal algebra~\cite{conf}. For our purposes it is sufficient to consider multiplets whose primary states have zero spin. Let the $R$ and $\Delta$ be the R-charge and the dimension of the primary. Unitarity puts restrictions on which values of $R$ and $\Delta$ may occur; the allowed possibilities are (i) $\Delta=R=0$ (the trivial representation), (ii) $\Delta=\frac{3}{2}|R|$ (chiral and anti-chiral representations), (iii) $\Delta\geq\frac{3}{2}|R|+2$. Representations of type (iii) with $\Delta>\frac{3}{2}|R|+2$ contain no null states and therefore are termed long multiplets. Chiral and anti-chiral representations contain null states at level one, i.e., their primaries are annihilated by half of the supercharges. These representations are called short. Representations of type (iii) which saturate the inequality are also short; the null states occur at level two. A well-known example of a short multiplet is a linear multiplet which contains a conserved current. It corresponds to the case $R=0,\Delta=2$. One can check that all non-chiral primaries coming from the reduction of the gauge field on the 7-branes satisfy $\Delta=\frac{3}{2}|R|+2$ and therefore are in short multiplets of type (iii). In particular, the $(0,0,0)$ mode with $\Delta=2$ we have found above is in fact the lowest component of a linear multiplet. It couples to a field theory operator $q^\dagger q - pp^\dagger$ in the adjoint of $SU(4)_7$. The corresponding current is simply the $SU(4)_7$ flavor current. The matching of non-chiral primaries with $\Delta=3$ is a bit more involved. The $(0,2,0)$ mode transforms in ${\bf 6}+\bar{\bf 6}$ of $SU(4)_7$. Its field theory counterparts are $h_1 p^\dagger\tilde{\cal Q}J_1 q + h_2 q J_1 A_1^\dagger J_1 q$ and $h_1 pJ_2\tilde{\cal Q} q^\dagger+h_2 p A_2^\dagger p$, where the flavor indices are antisymmetrized. The $U(1)^3$ charges of these operators match those of the $(0,2,0)$ mode. To show that these operators live in short multiplets, i.e., are annihilated by $\bar D^2$, one needs to use the classical equations of motion. The manipulations one has to go through are very similar to those in~\cite{berkooz}, and are subject to the same caveats. The use of the classical equations of motion is presumably justified in the weakly coupled regime where $g^2_{YM} N$ is small. The supergravity analysis indicates that the operators in question belong to short multiplets even for large $g^2_{YM} N$. An even more interesting situation arises when one tries to match the non-chiral primary with $U(1)^3$ charges $(0,0,-2)$ and $\Delta=3$. This mode lives in the adjoint of $SU(4)_7$. We claim that it corresponds to the field theory operator $h_1 q A_1 J_1q^\dagger-h_1 p^\dagger A_2 J_2p-h_2 q \tilde{\cal Q}^\dagger p$. Evaluating the ${\bar D}^2$ descendant of this operator using the classical equations of motion, one finds that it does not vanish. Instead, the descendant has the form $(q\tilde{q})(\tilde{p}p)$, i.e., it factorizes into a product of two gauge-invariant operators and is therefore subleading at large $N$. It follows that this field theory operator lives in a long multiplet for finite $N$, but is ``close'' to being in a short multiplet in the sense that its dimension approaches the unitarity bound as $N\to\infty$. On the supergravity side this means that the $(0,0,-2)$ one-particle state is in a short multiplet only for $N=\infty$. For finite $N$ the multiplet absorbs another short multiplet made of two-particle states and becomes long. This concludes our analysis of the AdS/CFT correspondence for the Sen model. There is complete agreement between the spectrum of primary operators in the field theory and the scalar Kaluza-Klein states on AdS as required by the AdS/CFT correspondence~\cite{m1,gubs,w1}. The charge assignments in Table~\ref{table} together with the formula $R=\frac{1}{3}(2m_1+2m_2-2m_3)$ imply that all chiral fields have canonical dimensions in the infrared. This is the most natural assumption for a theory with vanishing beta function, but as we pointed out in the introduction there is no field theory proof of this. The supergravity computation is only valid for large $N$ and large $g^2_{YM} N$. However, given that for $g^2_{YM}\ll 1$ and $N$ of order $1$ the dimensions are also canonical, it appears likely that the theory is finite for all $N$. \subsection{The non-conformal case} Next we discuss the deformed ${\cal N}=1$ theory which flows to a line of conformal fixed points in the infrared (section \ref{iic}). We have already pointed out that although the Wilsonian gauge coupling in this theory depends on the scale, the low-energy effective gauge coupling does not vary over the moduli space. This implies that the corresponding IIB background should have constant $\tau$. Indeed, in section \ref{IIB} we showed that the 7-brane background for this configuration is very similar to the background for the conformal theory. As in the conformal case, the 7-branes do not bend and are coincident with the O7-planes. The RR charge of the 7-brane is cancelled locally by the O7-planes, so we expect that the type IIB string coupling is constant. Similarly, the gravitational field of the 7-branes cancels against that of the orientifold planes. Thus it appears that the closed string sector is not affected by this deformation. The only difference between the conformal and the non-conformal case is in the open string sector, namely in the gauge connection on the 7-branes. In the conformal case it is trivial, while in the non-conformal case it is a flat connection which breaks the $SU(4)_7\times SU(4)_{7'}$ group to a diagonally embedded $SU(2)\times SU(2)$. To summarize, the deformation of the 7-brane background that leads to the non-conformal theory changes the properties of the theory on the 7-branes, but it appears not to change the closed string sector. To find a supergravity dual for this non-conformal theory, we need to repeat the analysis above with the new 7-brane background. Since the closed string sector is unchanged, the spectrum of the bulk modes should be the same as before. The matter content of the conformal and the non-conformal theory differ only in the number of flavors and their coupling to the bifundamentals. Therefore both theories have the same spectrum of operators that do not transform under the 7-brane groups. Thus it appears that the dimensions of all chiral primaries uncharged with respect to the flavor group are the same as in the conformal case, i.e., canonical. If antisymmetric tensors and bifundamentals have zero anomalous dimensions then the vanishing of the beta-functions, \eq{beta2}, requires that the fundamentals have dimension $1/2$. This is actually the lowest dimension for the fundamental allowed by unitarity. To show that this assignment of dimensions, or equivalently of R-charges, agrees with supergravity we would have to show that the KK reduction of the 7-brane theory with the singular flat connection switched on, reproduces the expected dimensions of the chiral primaries that involve the fundamentals. Unfortunately we do not know how to analyze the excitations of the impurity theory around nontrivial vacua, so we cannot check that our solution is consistent. Nevertheless, we get a definite prediction for the infrared dimensions of all fields. It would be interesting to confirm the answer by directly analyzing the perturbative expansion of the non-conformal theory at large $N$. \acknowledgements It is a pleasure to thank O.~Aharony, E.~Gimon, J.~Maldacena, G.~Moore, E.~Katz, and E.~Witten for helpful discussions. M.G.~would like to thank the Institute for Advanced Study for hospitality while this work was in progress. The work of M.G~was supported in part by DOE grants \#DF-FC02-94ER40818 and \#DE-FC-02-91ER40671, while that of A.K. by DOE grant \#DE-FG02-90ER40542. {\tighten
\section{Introduction} The equivalence principle was postulated by Einstein as a foundation stone for general relativity. The equivalence principle stipulates that the long-range gravitational interaction is entirely described by a universal coupling of ``matter'' (leptons, quarks, gauge fields and Higgs fields) to a (dynamical) second-rank symmetric tensor field $g_{\mu \nu} (x^{\lambda})$, replacing everywhere in the matter Lagrangian the usual, kinematical, special relativistic (Minkowski) metric $\eta_{\mu \nu}$. This principle assumes that all the non-gravitational (dimensionless) coupling constants of matter (gauge couplings, CKM mixing angles, mass ratios,$\ldots$) are non-dynamical, i.e. take (at least at large distances) some fixed (vacuum expectation) values, independently of where and when, in spacetime, they are measured. Two of the best experimental tests of the equivalence principle are: (i) tests of the universality of free fall, i.e. the fact that all bodies fall with the same acceleration in an external gravitational field; and (ii) tests of the ``constancy of the constants''. Laboratory experiments (due notably, in our century, to E\"otv\"os, Dicke, Braginsky and Adelberger) have verified the universality of free fall to the $10^{-12}$ level. For instance, the fractional difference in free fall acceleration of Beryllium and Copper samples was found to be~\cite{Su94} \begin{equation} \left( \frac{\Delta a}{a} \right)_{\rm Be \, Cu} = (-1.9 \pm 2.5) \times 10^{-12} \, . \label{eq:adel} \end{equation} The Lunar Laser Ranging experiment~\cite{LLR} has also verified that the Moon and the Earth fall with the same acceleration toward the Sun to better than one part in $10^{12}$ \begin{equation} \left( \frac{\Delta a}{a} \right)_{\rm Moon \, Earth} = (-3.2 \pm 4.6) \times 10^{-13} \, . \label{eq:llr} \end{equation} On the other hand, a recent reanalysis of the Oklo phenomenon (a natural fission reactor which operated two billion years ago in Gabon, Africa) gave a very tight limit on a possible time variation of the fine-structure ``constant'', namely~\cite{DD96} \begin{equation} -0.9 \times 10^{-7} < \frac{e_{\rm Oklo}^2 - e_{\rm now}^2}{e^2} < 1.2 \times 10^{-7} \, , \label{eq:oklo1} \end{equation} \begin{equation} -6.7 \times 10^{-17} \, {\rm yr}^{-1} < \frac{d}{dt} \ {\rm ln} \ e^2 < 5.0 \times 10^{-17} \, {\rm yr}^{-1} \, . \label{eq:oklo2} \end{equation} The tightness of the experimental limits (\ref{eq:adel})--(\ref{eq:oklo2}) might suggest to apply Occam's razor and to declare that the equivalence principle must be exactly enforced. However, the theoretical framework of modern unification theories, and notably string theory, suggest that the equivalence principle must be violated. Even more, the type of violation of the equivalence principle suggested by string theory is deeply woven into the basic fabric of this theory. Indeed, string theory is a very ambitious attempt at unifying all interactions within a consistent quantum framework. A deep consequence of string theory is that gravitational and gauge couplings are unified. In intuitive terms, while Einstein proposed a framework where geometry and gravitation were becoming united as a dynamical field $g_{\mu \nu} (x)$, i.e. a soft structure influenced by the presence of matter, string theory extends this idea by proposing a framework where geometry, gravitation, gauge couplings, and gravitational couplings all become soft structures described by interrelated dynamical fields. A symbolic equation expressing this softened, unified structure is \begin{equation} g_{\mu \nu} (x) \sim g^2 (x) \sim G(x) \, . \label{eq:ggg} \end{equation} It is conceptually pleasing to note that string theory proposes to render dynamical the structures left rigid (or kinematical) by general relativity. Technically, Eq.~(\ref{eq:ggg}) refers to the fact that string theory (as well as Kaluza-Klein theories) predicts the existence, at a fundamental level, of scalar partners of Einstein's tensor field $g_{\mu \nu}$, the model-independent ``dilaton'' field $\Phi (x)$, and various ``moduli fields''. The dilaton field, notably, plays a crucial role in string theory in that it determines the basic ``string coupling constant'' $g_s = e^{\Phi (x)}$, which determines in turn the (unified) gauge and gravitational coupling constants $g \sim g_s$, $G \propto g_s^2$, as exemplified by the low-energy effective action \begin{equation} L_{\rm eff} = e^{-2\Phi} \left[ \frac{R(g)}{\alpha'} + \frac{4}{\alpha'} \, (\nabla \Phi)^2 - \frac{1}{4} \, F_{\mu \nu}^2 - i \overline{\psi} \, D \psi - \ldots \right] \, . \label{eq:eff} \end{equation} A softened structure of the type of Eq.~(\ref{eq:ggg}), embodied in the effective action (\ref{eq:eff}), implies a deep violation of Einstein's equivalence principle. Bodies of different nuclear compositions fall with different accelerations because, for instance, the part of the mass of nucleus $A$ linked to the Coulomb interaction of the protons depends on the space-variable fine-structure constant $e^2 (x)$ in a non-universal, composition-dependent manner. This raises the problem of the compatibility of the generic string prediction (\ref{eq:ggg}) with experimental tests of the equivalence principle, such as Eqs.~(\ref{eq:adel}), (\ref{eq:llr}) or (\ref{eq:oklo2}). It is often assumed that the softness (\ref{eq:ggg}) applies only at short distances, because the dilaton and moduli fields are likely to acquire a non zero mass after supersymmetry breaking. However, a mechanism has been proposed~\cite{DP} to reconcile in a natural manner the existence of a {\it massless} dilaton (or moduli) field as a fundamental partner of the graviton field $g_{\mu \nu}$ with the current level of precision $(\sim 10^{-12})$ of experimental tests of the equivalence principle. In the mechanism of~\cite{DP} (see also~\cite{DN} for metrically-coupled scalars) the very small couplings necessary to ensure a near universality of free fall, $\Delta a/a < 10^{-12}$, are dynamically generated by the expansion of the universe, and are compatible with couplings ``of order unity'' at a fundamental level. The aim of the present paper is to emphasize the rich phenomenological consequences of long-range dilaton-like fields, and the fact that high-precision clock experiments might contribute to searching for, or constraining, their existence. More precisely, the basic question we wish to address here is the following: given the existing experimental tests of gravity, and given the currently favored theoretical framework, can high-precision clock experiments probe interesting theoretical possibilities which remain yet unconstrained ? In addressing this question we wish to assume, as theoretical framework, the class of effective field theories suggested by string theory. For historical completeness, let us mention that the theoretical framework which has been most considered in the phenomenology of gravitation, i.e. the class of ``metric'' theories of gravity~\cite{W81}, which includes most notably the ``Brans-Dicke''-type tensor-scalar theories, appears, from a modern perspective, as being rather artificial. This is good news because the phenomenology of ``non metric'' theories is richer and offers new possibilities for clock experiments. Historically, the restricted class of ``metric'' theories was introduced in 1956 by Fierz~\cite{F56} to prevent, in an {\it ad hoc} way, too violent a conflict between experimental tests of the equivalence principle and the existence of a scalar contribution to gravity as suggested by the theories of Kaluza-Klein~\cite{KK} and Jordan~\cite{J}. Indeed, Fierz was the first one to notice that a Kaluza-Klein scalar would generically strongly violate the equivalence principle. He then proposed to restrict artificially the couplings of the scalar field to matter so as to satisfy the equivalence principle. The restricted class of equivalence-principle-preserving couplings introduced by Fierz is now called ``metric'' couplings. Under the aegis of Dicke, Nordtvedt, Thorne and Will a lot of attention has been given to ``metric'' theories of gravity\footnote{Note, however, that Nordtvedt, Will, Haugan and others (for references see~\cite{W81}) studied conceivable phenomenological consequences of generic ``non metric'' couplings, without using a motivated field-theory framework to describe such couplings.}, and notably to their quasi-stationary-weak-field phenomenology (``PPN framework'', see, e.g.,~\cite{W81}). \section{Generic effective theory of a long-range dilaton} Motivated by string theory, we consider the generic class of theories containing a long-range dilaton-like scalar field $\varphi$. The effective Lagrangian describing these theories has the form (after a conformal transformation to the ``Einstein frame''): \begin{eqnarray} L_{\rm eff} &=& \frac{1}{4q} R(g_{\mu\nu}) - \frac{1}{2q} \ (\nabla \varphi)^2 - \frac{1}{4e^2 (\varphi)} \ (\nabla_{\mu} A_{\nu} - \nabla_{\nu} A_{\mu})^2 \nonumber \\ &-& \sum_A \ \left[\overline{\psi}_A \, \gamma^{\mu} (\nabla_{\mu} -iA_{\mu}) \psi_A + m_A (\varphi) \, \overline{\psi}_A \psi_A \right] + \cdots \label{eq:01} \end{eqnarray} Here, $q\equiv 4\pi \, \overline G$ where $\overline G$ denotes a bare Newton's constant, $A_{\mu}$ is the electromagnetic field, and $\psi_A$ a Dirac field describing some fermionic matter. At the low-energy, effective level (after the breaking of $SU(2)$ and the confinement of colour), the coupling of the dilaton $\varphi$ to matter is described by the $\varphi$-dependence of the fine-structure ``constant'' $e^2 (\varphi)$ and of the various masses $m_A (\varphi)$. Here, $A$ is a label to distinguish various particles. [A deeper description would include more coupling functions, e.g. describing the $\varphi$-dependences of the $U(1)_Y$, $SU(2)_L$ and $SU(3)_c$ gauge coupling ``constants''.] The strength of the coupling of the dilaton $\varphi$ to the mass $m_A (\varphi)$ is given by the quantity \begin{equation} \alpha_A \equiv \frac{\partial \ {\rm ln} \ m_A (\varphi_0)}{\partial \ \varphi_0} \, , \label{eq:02} \end{equation} where $\varphi_0$ denotes the ambient value of $\varphi (x)$ (vacuum expectation value of $\varphi (x)$ around the mass $m_A$, as generated by external masses and cosmological history). For instance, the usual PPN parameter $\gamma -1$ measuring the existence of a (scalar) deviation from the pure tensor interaction of general relativity is given by~\cite{DEF},~\cite{DP} \begin{equation} \gamma -1 = -2 \ \frac{\alpha_{\rm had}^2}{1+\alpha_{\rm had}^2} \, , \label{eq:03} \end{equation} where $\alpha_{\rm had}$ is the (approximately universal) coupling (\ref{eq:02}) when $A$ denotes any (mainly) hadronic object. The Lagrangian (\ref{eq:01}) also predicts (as discussed in~\cite{DP}) a link between the coupling strength (\ref{eq:02}) and the violation of the universality of free fall: \begin{equation} \frac{a_A -a_B}{\frac{1}{2} (a_A + a_B)} \simeq (\alpha_A -\alpha_B) \alpha_E \sim -5\times 10^{-5} \, \alpha_{\rm had}^2 \, . \label{eq:04} \end{equation} Here, $A$ and $B$ denote two masses falling toward an external mass $E$ (e.g. the Earth), and the numerical factor $-5 \times 10^{-5}$ corresponds to $A= {\rm Be}$ and $B= {\rm Cu}$. The experimental limit Eq.~(\ref{eq:adel}) shows that the (mean hadronic) dilaton coupling strength is already known to be very small: \begin{equation} \alpha_{\rm had}^2 \lsim 10^{-7} \, . \label{eq:06} \end{equation} Free fall experiments, such as Eq.~(\ref{eq:adel}) or the comparable Lunar Laser Ranging constraint Eq.~(\ref{eq:llr}), give the tightest constraints on any long-range dilaton-like coupling. Let us mention, for comparison, that solar-system measurements of the PPN parameters (as well as binary pulsar measurements) constrain the dilaton-hadron coupling to $\alpha_{\rm had}^2 < 10^{-3}$ (recently announced VLBI measurements improve this constraint to the $2 \times 10^{-4}$ level), while the best current constraint on the time variation of the fine-structure ``constant'' (deduced from the Oklo phenomenon), namely Eq.~(\ref{eq:oklo2}), yields from Eq. (\ref{eq:16}) below, $\alpha_{\rm had}^2 \lsim 3 \times 10^{-4}$. [For an updated review of experimental tests of gravity, see the chapter 14 of the Review of Particle Physics, available on http://pdg.lbl.gov/] To discuss the probing power of clock experiments, we need also to introduce other coupling strengths, such as \begin{equation} \alpha_{\rm EM} \equiv \frac{\partial \ {\rm ln} \ e^2 (\varphi_0)}{\partial \ \varphi_0} \, , \label{eq:07} \end{equation} measuring the $\varphi$-variation of the electromagnetic (EM) coupling constant\footnote{Note that we do not use the traditional notation $\alpha$ for the fine-structure constant $e^2 / 4\pi \hbar c$. We reserve the letter $\alpha$ for denoting various dilaton-matter coupling strengths. Actually, the latter coupling strengths are analogue to $e$ (rather than to $e^2$), as witnessed by the fact that observable deviations from Einsteinian predictions are proportional to products of $\alpha$'s, such as $\alpha_A \alpha_E$, $\alpha_{\rm had}^2$, etc$\ldots$}, and \begin{equation} \alpha_A^{A^*} \equiv \frac{\partial \ {\rm ln} \ E_A^{A^*} (\varphi_0)}{\partial \ \varphi_0} \, , \label{eq:08} \end{equation} where $E_A^{A^*}$ is the energy difference between two atomic energy levels. In principle, the quantity $\alpha_A^{A^*}$ can be expressed in terms of more fundamental quantities such as the ones defined in Eqs. (\ref{eq:02}) and (\ref{eq:07}). For instance, in an hyperfine transition \begin{equation} E_A^{A^*} \propto (m_e \, e^4) \ g_I \ \frac{m_e}{m_p} \ e^4 \ F_{\rm rel} (Z e^2) \, , \label{eq:09} \end{equation} so that \begin{equation} \alpha_A^{A^*} \simeq 2 \, \alpha_e -\alpha_p + \alpha_{\rm EM} \left( 4+\frac{d \ {\rm ln} \ F_{\rm rel}}{d \ {\rm ln} \ e^2} \right) \, . \label{eq:10} \end{equation} Here, the term $F_{\rm rel} (Z e^2)$ denotes the relativistic (Casimir) correction factor~\cite{Casimir}. Moreover, in any theory incorporating gauge unification one expects to have the approximate link~\cite{DP} \begin{equation} \alpha_A \simeq \left( 40.75 - {\rm ln} \ \frac{m_A}{1 \ {\rm GeV}} \right) \ \alpha_{\rm EM} \, , \label{eq:11} \end{equation} at least if $m_A$ is mainly hadronic. \section{Clock experiments and dilaton couplings} The coupling parameters introduced above allow one to describe the deviations from general relativistic predictions in most clock experiments~\cite{TD}. Let us only mention some simple cases. First, it is useful to distinguish between ``global'' clock experiments where one compares spatially distant clocks, and ``local'' clock experiments where the clocks being compared are next to each other. The simplest global clock experiment is a static redshift experiment comparing (after transfer by electromagnetic links) the frequencies of the same transition $A^* \rightarrow A$ generated in two different locations ${\bf r}_1$ and ${\bf r}_2$. The theory of Section 2 predicts a redshift of the form (we use units in which $c=1$) \begin{equation} \frac{\nu_A^{A^*} ({\bf r}_1)}{\nu_A^{A^*} ({\bf r}_2)} \simeq 1 + (1 + \alpha_A^{A^*} \, \alpha_E) \ (\overline{U}_E ({\bf r}_2) - \overline{U}_E ({\bf r}_1)) \, , \label{eq:12} \end{equation} where \begin{equation} \overline{U}_E \, ({\bf r}) = \frac{\overline G \, m_E}{r} \label{eq:13} \end{equation} is the {\it bare} Newtonian potential generated by the external mass $m_E$ (say, the Earth). Such a result has the theoretical disadvantage of depending on other experiments for its interpretation. Indeed, the {\it bare} potential $\overline{U}_E$ is not directly measurable. The measurement of the Earth potential by the motion of a certain mass $m_B$ gives access to $(1+\alpha_B \, \alpha_E) \ \overline{U}_E \, ({\bf r})$. The theoretical significance of a global clock experiment such as (\ref{eq:12}) is therefore fairly indirect, and involves other experiments and other dilaton couplings. One can generalize (\ref{eq:12}) to a more general, non static experiment in which different clocks in relative motion are compared. Many different ``gravitational potentials'' will enter the result, making the theoretical significance even more involved. A conceptually simpler (and, probably, technologically less demanding) type of experiment is a differential, ``local'' clock experiment. Such ``null'' clock experiments have been proposed by Will~\cite{W81} and first performed by Turneaure et al.~\cite{T83}. The theoretical significance of such experiments within the context of dilaton theories is much simpler than that of global experiments. For instance if (following the suggestion of~\cite{PTM}) one locally compares two clocks based on hyperfine transitions in alkali atoms with different atomic number $Z$, one expects to find a ratio of frequencies \begin{equation} \frac{\nu_A^{A^*} ({\bf r})}{\nu_B^{B^*} ({\bf r})} \simeq \frac{F_{\rm rel} (Z_A \, e^2 (\varphi_{\rm loc}))}{F_{\rm rel} (Z_B \, e^2 (\varphi_{\rm loc}))} \, , \label{eq:14} \end{equation} where the local, ambient value of the dilaton field $\varphi_{\rm loc}$ might vary because of the (relative) motion of external masses with respect to the clocks (including the effect of the cosmological expansion). The directly observable fractional variation of the ratio (\ref{eq:14}) will consist of two factors: \begin{equation} \delta \ {\rm ln} \ \frac{\nu_A^{A^*}}{\nu_B^{B^*}} = \left[ \frac{\partial \ {\rm ln} \ F_{\rm rel} (Z_A \, e^2)}{\partial \ {\rm ln} \ e^2} - \frac{\partial \ {\rm ln} \ F_{\rm rel} (Z_B \, e^2)}{\partial \ {\rm ln} \ e^2} \right] \times \delta \ {\rm ln} \ e^2 \, . \label{eq:15} \end{equation} The ``sensitivity'' factor in brackets due to the $Z$-dependence of the Casimir term can be made of order unity~\cite{PTM}, while the fractional variation of the fine-structure constant is expected in dilaton theories to be of order~\cite{DP},~\cite{TD} \begin{eqnarray} \delta \ {\rm ln} \ e^2 (t) &=& -2.5 \times 10^{-2} \ \alpha_{\rm had}^2 \ U(t) \nonumber \\ &-& 4.7 \times 10^{-3} \ \kappa^{-1/2} ({\rm tan} \ \theta_0) \ \alpha_{\rm had}^2 \ H_0 (t-t_0) \, . \label{eq:16} \end{eqnarray} Here, $U(t)$ is the value of the externally generated gravitational potential at the location of the clocks, and $H_0 \simeq 0.5 \times 10^{-10} \ {\rm yr}^{-1}$ is the Hubble rate of expansion. [The factor $\kappa^{-1/2} \ {\rm tan} \ \theta_0$ is expected to be $\sim 1$.] The (rough) theoretical prediction (\ref{eq:16}) allows one to compare quantitatively the probing power of clock experiments to that of equivalence principle tests. Let us (optimistically) assume that clock stabilities of order $\delta \nu / \nu \sim 10^{-17}$ (for the relevant time scale) can be achieved. A differential {\it ground} experiment (using the variation of the Sun's potential due to the Earth eccentricity) would probe the level $\alpha_{\rm had}^2 \sim 3\times 10^{-6}$. A geocentric satellite differential experiment could probe $\alpha_{\rm had}^2 \sim 5\times 10^{-7}$. These levels are impressive (compared to present solar-system tests of the PPN parameter $\gamma$ giving the constraint $\alpha_{\rm had}^2 \simeq (1-\gamma) / 2 < 2 \times 10^{-4}$), but are not as good as the present equivalence-principle limit (\ref{eq:06}). To beat the level (\ref{eq:06}) one needs to envisage an heliocentric differential clock experiment (a few solar radii probe within which two hyper-stable clocks are compared). Such an experiment could, according to Eq. (\ref{eq:16}), reach the level $\alpha_{\rm had}^2 \sim 10^{-9}$. [Let us also note that a gravitational time delay global experiment using clocks beyond the Sun as proposed by C. Veillet (SORT concept) might (optimistically) probe the level $\alpha_{\rm had}^2 \sim 10^{-7}$.] It is, however, to be noted that a much refined test of the equivalence principle such as STEP (Satellite Test of the Equivalence Principle) aims at measuring $\Delta a/a \sim 10^{-18}$ which corresponds to the level $\alpha_{\rm had}^2 \sim 10^{-14}$, i.e. five orders of magnitude better than any conceivable clock experiment. \section{Conclusions} In summary, the main points of the present contribution are: \begin{enumerate} \item[$\bullet$] Independently of any theory, the result (\ref{eq:oklo2}) of a recent reanalysis of the Oklo phenomenon~\cite{DD96} gives a motivation, and a target, for improving laboratory clock tests of the time variation of the fine-structure constant $e^2$ (which are at the $3.7 \times 10^{-14} \ {\rm yr}^{-1}$ level~\cite{PTM}). \item[$\bullet$] Modern unification theories, and especially string theory, suggest the existence of new gravitational-strength fields, notably scalar ones (``dilaton'' or ``moduli''), whose couplings to matter violate the equivalence principle. These fields would induce a spacetime variability of the coupling constants of physics (such as the fine-structure constant). High-precision clock experiments are excellent probes of such a possibility. \item[$\bullet$] The generic class of dilaton theories defined in Section 2 provides a well-defined theoretical framework in which one can discuss the phenomenological consequences of the existence of a dilaton-like field. Such a theoretical framework (together with some assumptions, e.g. about gauge unification and the origin of mass hierarchy) allows one to compare and contrast the probing power of clock experiments to that of other experiments. \item[$\bullet$] Local, differential clock experiments (of the ``null'' type of~\cite{T83}) appear as conceptually cleaner, and technologically less demanding, probes of dilaton-motivated violations of the equivalence principle than global, absolute clock experiments (of the Gravity Probe A type). \item[$\bullet$] If we use the theoretical assumptions of Section 2 to compare clock experiments to free-fall experiments, one finds that one needs to send and intercompare two ultra-high-stability clocks in near-solar orbit in order to probe dilaton-like theories more deeply than {\it present} free-fall experiments. Currently proposed improved satellite tests of the equivalence principle would, however, beat any clock experiment in probing even more deeply such theories. \item[$\bullet$] At the qualitative level, it is, however, important to note that clock experiments (especially of the ``global'', GPA type) probe different combinations of basic coupling parameters than free-fall experiments. This is visible in Eq. (\ref{eq:10}) which shows that $\alpha_A^{A^*}$ contains the leptonic quantity $\alpha_e = \partial \ {\rm ln} \ m_{\rm electron} / \partial \ \varphi_0$ without any small factor\footnote{Free-fall experiments couple predominantly to hadronic quantities such as $\alpha_p = \partial \ {\rm ln} \ m_{\rm proton} / \partial \ \varphi_0$, and to Coulomb-energy effects proportional to $\alpha_{\rm EM}$. The effect of the leptonic quantity $\alpha_e$ is down by a small factor $\sim m_e / m_p \sim 1/1836$.}. \end{enumerate} \section*{References}
\section{Introduction} Several years after the first demonstration \cite{R1}, optical homodyne tomography has become a well established tool for measuring quantum statistical properties of optical radiation. What is particularly fascinating, this technique provides practical means to visualise the measured quantum state in the form of the Wigner function. This success is a result of combining a complete quantum mechanical measurement of field quadratures with a filtered back-projection algorithm used in medical imaging. The purpose of this contribution is to trace some other analogies between quantum state reconstruction and classical image processing, with the motivation to develop novel numerical methods for quantum tomography. Our interest will be focused on imperfect detection \cite{R2,R3}, which has deleterious effects on quantum interference phenomena exhibited by non-classical states \cite{R4}. As we will discuss in Sec.~2, such effects can be related in the phase space representation to image blurring. Restoration of blurred images is a well known problem in classical imaging, and a number of methods has been developed for this purpose. Specifically, we shall briefly describe in Sec.~3 the Richardson algorithm \cite{R5} (known also in statistics as the expectation-maximization algorithm \cite{R6}), which provides an effective iterative procedure to perform image deblurring. An interesting question is, whether classical deblurring methods can be transferred to quantum tomography. We discuss this, in the case of the Richardson algorithm, in Sec.~4. The answer is not straightforward: the Richardson algorithm assumes positive definiteness of the original undegraded image, and this condition is not satisfied by the quantum mechanical Wigner function. We will show that this difficulty can be overcome by expressing the Wigner function in terms of the phase space displaced photon distribution. This yields an iterative algorithm for reconstructing the Wigner function, which incorporates compensation for detection losses in a numerically stable way \cite{R7}. \section{Imperfect detection and image blurring} Homodyne detection is a realization of the quantum mechanical measurement of field quadratures only in the idealized loss-free limit. In practice, a fraction of the signal field is always lost due to the mode-mismatch and the non-unit efficiency of photodiodes. The homodyne statistics collected using a realistic setup is described by the distribution \cite{R8}: \begin{equation} \label{Eq:hxtheta} h(x;\theta) = \int_{-\infty}^{\infty} \mbox{d}x' \; \frac{1}{\sqrt{\pi(1-\eta)}} \exp \left( - \frac{(x-\sqrt{\eta}x')^{2}}{1-\eta} \right) \langle x'_{\theta} | \hat{\rho} | x'_{\theta} \rangle, \end{equation} where $\theta$ is the local oscillator phase, $\eta$ characterizes the overall detector efficiency, and $\langle x'_{\theta} | \hat{\rho} | x'_{\theta} \rangle$ denote diagonal elements of the density matrix $\hat{\rho}$ in the quadrature basis. Application of the back-projection transformation to $h(x;\theta)$ yields, instead of the Wigner function, a generalized phase space quasidistribution function $P(q,p; s)$ with the ordering parameter $s=-(1-\eta)/\eta$. This function can be expressed as a convolution of the Wigner function $W(q,p)$ with a gaussian function: \begin{equation} \label{Eq:Palphas} P(q,p; -|s|) = \int \mbox{d}q'\,\mbox{d}p' \; \frac{1}{\pi |s|} \exp\left( - \frac{1}{|s|}[(q-q')^2+(p-p')^2 ]\right) W(q',p'). \end{equation} Thus what we reconstruct from imperfect homodyne data, is a blurred version of the Wigner function. The question is, whether we can get rid of this blurring in numerical processing of experimental data. A similar problem appears in the following classical context: suppose we observe an image using an imperfect apparatus (for example ill-matched glasses), which generates some blurring. Such blurring can be described by a so-called point spread function specifying the shape generated by a single point of the original image. The observed degraded image is consequently given by a convolution of the original image with the point spread function. Using this language, we can assign the following names to the terms of Eq.~(\ref{Eq:Palphas}): \begin{center} \begin{tabular}{rcl} $W(q',p')$ & --- & original image \\ $ \displaystyle \frac{1}{\pi |s|} \exp\left( - \frac{1}{|s|} [(q-q')^2+(p-p')^2]\right)$ & --- & point spread function \\ $P(q,p; -|s|)$ & --- & degraded image \end{tabular} \end{center} The common problem now is the restoration of the original image from the degraded one, assuming that the point spread function is known. \section{Image restoration} An analytical way to deconvolve Eq.~(\ref{Eq:Palphas}) is to apply the Fourier transform, which maps a convolution onto a direct product. Dividing both the sides by the Fourier-transformed point spread function and evaluating the inverse Fourier transform thus yields an explicit expression for the original image. However, this procedure is very sensitive to statistical fluctuations and numerical truncation errors, which makes its practical application a very delicate matter. These problems have been noted also in the context of quantum tomography \cite{R9}. The numerical instability of the Fourier deconvolution has led to the development of techniques dedicated for image restoration. The basic observation is that statistical noise does not allow us to specify precisely the original image that was 'behind' the blurred data. In principle, the measured degraded image could could be generated by a variety of original images. However, comparing various original images we intuitively expect that some of them were {\em more likely} to generate the measured data than other ones. The maximum-likelihood methodology quantifies this intuition, and selects as an estimate the original image which maximizes the likelihood. In order to discuss this idea in detail, we shall consider a discretized version of Eq.~(\ref{Eq:Palphas}): \begin{equation} \label{Eq:LININPOS} p_{\nu} = \sum_{n} A_{\nu n} w_{n}, \end{equation} where $w_n$ is the original image, $A_{\nu n}$ is the point spread function, and $p_{\nu}$ is the degraded image. Note that this formulation is more general compared to Eq.~(\ref{Eq:Palphas}), because it allows the point spread function to be of different form for each 'element' of the original image indexed with $n$. The likelihood can be quantified using the function \begin{equation} {\cal L} = \sum_{\nu} p_{\nu} \ln \left( \sum_{n} A_{\nu n} w_{n} \right) \end{equation} which has a rigorous derivation when the degraded image is observed as a histogram of events governed by Poissonian statistics. The likelihood function for quantum measurement has been discussed in Ref.~\cite{R10}, where its maximization has been proposed as a method for quantum state estimation. In classical imaging, it is natural to assume that $w_{n}$, as well as $A_{\nu n}$ as a function of $\nu$ for each $n$, are positive definite distributions with sum equal to one. Under these assumptions, it is possible to find the maximum of the likelihood function ${\cal L}$ via simple iteration: \begin{equation} w_{n}^{(i+1)} = \sum_{\nu} p_{\nu} \frac{A_{\nu n} w_{n}^{(i)}}{\sum_{m} A_{\nu m} w_{m}^{(i)}}, \end{equation} which is the essence of the Richardson algorithm for image restoration~\cite{R5}. A simple heuristic derivation of this algorithm can be found in Ref.~\cite{R7}. Of course, the maximum-likelihood approach is not a magic wand solving unconditionally the problem of image restoration. With increasing blurring, the quality of the reconstructed image worsens, and the convergence of the iterative algorithm may be very slow. In many cases, however, it offers superior performance compared to the Fourier deconvolution technique. \section{Quantum tomography} An obvious difficulty with applying the iterative restoration algorithm to quantum tomography is that the object to be reconstructed in the quantum case, i.e.\ the Wigner function, is not positive definite. Nevertheless, there are some other quantum mechanical reconstruction problems, where positivity constraints appear in a natural way. An interesting and nontrivial example is determination of the photon number distribution from random phase homodyne statistics \cite{R11}. The relation between the phase-averaged homodyne statistics and the photon number distribution $\varrho_n$ is given by \begin{equation} \frac{1}{2\pi}\int_{-\pi}^{\pi} \mbox{d}\theta \; h(x;\theta) = \sum_{n} A_n(x) \varrho_n, \end{equation} where $A_{n}(x)$ describe contributions to the homodyne statistics generated by different Fock states $|n\rangle$. This formula, after discretization of $x$, is exactly of the form assumed in Eq.~(\ref{Eq:LININPOS}). Thus we arrive at the following formal analogy: \begin{center} \begin{tabular}{rcl} $\varrho_n$ & --- & original image \\ $A_{n}(x)$ & --- & point spread function \\ $\displaystyle \frac{1}{2\pi} \int_{-\pi}^{\pi} \mbox{d}\theta \; h(x; \theta)$ & --- & degraded image \end{tabular} \end{center} which allows us to apply directly the iterative reconstruction algorithm \cite{R12}. In this procedure, there is one {\em a priori} parameter: it is the cut-off of the distribution $\varrho_n$ specifying the maximum number of photons. Reconstruction of the photon distribution may seem to be quite distant from the starting point of our considerations, which was deblurring of the Wigner function. However, let us recall that the Wigner function can be represented as an alternating sum of the photon distribution $\varrho_{n}(q,p)$ corresponding to the phase space displaced state $\hat{D}^{\dagger}(q,p)\hat{\varrho}\hat{D}(q,p)$: \begin{equation} \label{Eq:Wqp} W(q,p) = \frac{1}{\pi} \sum_{n=0}^{\infty} (-1)^{n} \varrho_{n}(q,p) \end{equation} Obviously, we can apply this formula to evaluate the Wigner function at $q=p=0$. What would be of interest, is the generalization of the maximum-likelihood algorithm to determination of an arbitrarily displaced photon distribution $\varrho_n(q,p)$. This would yield a numerically stable procedure for reconstructing the Wigner function from homodyne statistics, even in the case of the non-unit detection efficiency. Surprisingly, this generalization is quite straightforward. The basic observation is that the displacement transformation has a simple effect on the homodyne statistics, shifting it by $\sqrt{\eta}(q\cos\theta+p\sin\theta)$ for a given local oscillator phase $\theta$. Consequently, we have the relation \begin{equation} \label{Eq:tomolininpos} \frac{1}{2\pi}\int_{-\infty}^{\infty} \mbox{d}\theta \; h(x+\sqrt{\eta} q \cos\theta + \sqrt{\eta}p \sin\theta ; \theta) = \sum_{n} A_{n}(x) \varrho_{n}(q,p), \end{equation} which can be readily implemented in the iterative algorithm. Thus, we have arrived at the following two-step algorithm for quantum tomography: for a given phase space point $(q,p)$, construct the phase-averaged histogram according to the right-hand side of Eq.~(\ref{Eq:tomolininpos}), and iteratively reconstruct $\varrho_n(q,p)$. Then, calculate the value of the Wigner function according to Eq.~(\ref{Eq:Wqp}). In Fig.~1 we present Monte Carlo simulated reconstruction of the Wigner function for a Schr\"{o}dinger cat state detected using a homodyne setup with the efficiency $\eta=90\%$. \begin{figure} \centerline{\setlength{\unitlength}{1pt} \begin{picture}(325,250) \put(0,0){\epsfig{file=kot.eps}} \put(20,225){\small $W(q,p)$} \put(120,0){\small $q$} \put(235,25){\small $p$} \end{picture}} \bigskip \caption{Fig.~1. Reconstruction of the Schr\"{o}dinger cat state $|\Psi\rangle\propto|2i\rangle -|-2i\rangle$ from Monte Carlo simulated homodyne experiment. The homodyne data consisted of $10^5$ events generated for each of $64$ phases spaced uniformly between $0$ and $\pi$. At each point of the grid, the displaced photon statistics was obtained from $10^4$ iterations, starting from a flat distribution for $0\le n \le 39$. In the simulations, the homodyne variable $x$ has been discretized into 16000 bins over the range $-8\le x \le 8$.} \end{figure} The standard filtered back-projection algorithm used in quantum tomography is based on the inverse Radon transform, whose integral kernel is singular. Therefore, a regularization scheme is necessary in processing experimental data. This aspect has a counterpart in the maximum-likelihood algorithm. In this approach, we have the cut-off for the photon distribution which can be regarded as a regularization parameter. Its proper choice is an important matter. Setting it too small perturbes the reconstructed photon distribution. On the other hand, the larger number of $\varrho_n$s, the slower iterations converge. The expected shape of the photon distribution can be quite easily predicted, if we roughly know the region of the phase space where the Wigner function is localized. For this purpose it is useful to recall the semiclassical picture of projections on Fock states as rings in the phase space characterized by the radius $\sqrt{2n}$. The photon distribution is nonzero over the range of $n$ for which the corresponding rings overlap with the localization region for the Wigner function. Truncation of the photon distribution can be introduced as a regularization scheme also in the standard linear reconstruction approach. In such a scheme, the Wigner function would be evaluated from a finite part of the photon distribution reconstructed using the pattern function technique. However, properties of the reconstructed photon distribution make this method very sensitive to statistical noise. This can be straightforwardly seen in the most regular case of $\eta=1$. For large $n$, the error of $\varrho_n$ tends to a fixed nonzero value \cite{R13}, and moreover deviations of consecutive $\varrho_n$s are strongly anticorrelated. Consequently, the alternating sum defined in Eq.~(\ref{Eq:Wqp}) accumulates the statistical uncertainty of the photon distribution \cite{R14}. Let us also note that in principle we could apply the restoration algorithm to homodyne histograms described by Eq.~(\ref{Eq:hxtheta}), in order to obtain deblurred quadrature distributions $\langle x_\theta | \hat{\varrho} | x_\theta \rangle$. In this case statistical fluctuations would play a much more significant role. The advantage of using Eq.~(\ref{Eq:tomolininpos}) is that we use the whole available sample of experimental data to determine a relatively small number of parameters $\varrho_{n}(q,p)$. \medskip \noindent{\bf Acknowledgements} I would like to thank Prof.\ G.~Mauro D'Ariano for his hospitality during my stay in Pavia supported by INFM, and Prof.\ Krzysztof W\'{o}dkiewicz for comments on the manuscript. I have benefited a lot from discussions with Zdenek Hradil on maximum-likelihood methods in quantum state reconstruction. This research is supported by Komitet Bada\'{n} Naukowych, grant 2P03B~013~15.
\section{Introduction.} One of the most popular approaches used to determine the small amplitude envelope soliton solutions in non linear models is the well-known Multiple Scale Expansion (MSE) technique \cite{rem}. This technique amounts to expanding the equations of motion on different time and space scales looking for wave-packet-like solutions; a wave packet is a superposition of plane waves whose frequencies and wave vectors lie in a narrow band, and it can be conveniently described by a plane wave with an amplitude that varies slowly in space and time. Increasing progressively the time and space scales one determines in a first step the carrier wave as a phonon mode of the linearized system, then deduces the partial differential equation that identifies the envelope phase velocity with the wave packet group velocity, and finally derives the NLS equation for the envelope whose diffusion coefficient is in fact the wave packet group velocity dispersion. An alternative method, commonly used in optics, consists in expanding the dispersion relations with respect to the carrier frequency and then in building at each order of this expansion an operator that acts on the envelope function \cite{opt}. MSE has been successfully applied to various non linear systems with scalar fields and the corresponding method adopted in optics has permitted the study of optical solitons in fiber. In this latter case one deals with eletric field components which are not coupled at the linear order of the Maxwell equations. However, many non linear models of interest involve vectorial fields with coupled components at the linear order that give rise to dispersion relations with more than one branch: classical examples are given by multi-atomic lattices, or by lattices in which the mass at each site can move in a multidimensional space. In this work, we show how to find small amplitude envelope soliton solutions in such vectorial lattices problems. The main difficulty with respect to the scalar case is to determine the relative amplitudes of the different components of the field. We will perform a perturbative expansion, around one wavenumber, of the linear system that gives the dispersion relations and the linear eigenmodes; then we will introduce an operator formalism analogous to that used in optics to obtain, from this expansion and from the nonlinear terms, the MSE equations, up to the NLS one. An application of the method presented in this work can be found in a forthcoming paper \cite{noi}, where the envelope soliton solutions of an helicoidal DNA model described by a radial and an angular degree of freedom for each site are derived. \section {Wave-packet in linear vectorial lattices} The NLS equation is obtained when a weak dispersion is balanced by a weak nonlinearity. In order to characterize the dispersion, let us first restrict our attention to the linear part of the system of interest. We start with a one dimensional vectorial linear lattice model given by the equations of motion: \begin{equation} \label{emoto} \frac{\partial^2 E(n,\alpha,t)}{\partial t^2 }= - \sum_{n',\alpha'} J (n-n',\alpha,\alpha') E(n',\alpha',t) \end{equation} where $n,n'$ are the site indices, $\alpha,\alpha'$ are the indices that label the components of the vectorial field $ E(n,\alpha,t)$, and $J (n-n',\alpha,\alpha')$ are the force constants depending on $ n-n'$ for translationally invariant systems. Looking for plane wave solutions of the form \begin{equation} A \vec{V}_l(q) e^{i(q n-\omega_l(q)t)} + c.c. \end{equation} where $A$ is the wave amplitude, the equation of motion is mapped to the operator equation in the wave numbers space: \begin{equation} \label{d} (\hat{J}(q)- \omega^{2}_{l}(q))\vec{V}_{l}(q) =0 \end{equation} where $\hat{J}(q)$ is the Fourier transform of the matrix $\hat{J}(n-n')$. The index $l$ runs from 1 to the number of components of the vectorial field $E(n,\alpha,t)$; the eigenvalues functions $\omega^{2}_{l}(q)$ give the branches of the dispersion relation; the normal modes $\vec{V}_l(q)$ are the orthonormalized eigenvectors of the matrix $\hat{J}(q)- \omega^{2}_{l}(q) $. In order to investigate the dispersion, we now consider a wave-packet-like solution, {\em i.e.} a superposition of plane waves with wave numbers in a small interval: \begin{equation} \label{wpuno} \vec{E}_l(n,t)=\int_{q_0 - \Delta q}^{q_0 + \Delta q} A(q) \, \vec{V}_{l}(q) \, e^{i(q n-\omega_l(q)t)} \; dq +c.c. \end{equation} For each $q$ contributing to the wave packet the system of equations (\ref{d}) must be fulfilled. The weakly dispersive case is obtained by considering only small deviations of $q$ with respect to the wavector $q_0$ corresponding to the center of the wavepacket. To measure this deviation, the wavevector $q$ is written $ q = q_0 + \epsilon q_1$, where $\epsilon \ll 1$. Eq.~(\ref{d}) is solved, $\forall q_1$ in the integration range, by a perturbative technique. The operator $\hat{J}(q_0+\epsilon q_1)$ can be expanded in Taylor series as $\hat{J}(q_0)+\epsilon \hat{J}'(q_0)q_1+\epsilon^2 \hat{J}''(q_0)q_{1}^{2}/2 + \ldots$. The quantities $\epsilon \hat{J'}(q_0)q_1$, $\epsilon^2 \hat{J''}(q_0)q_1^2/2$ are small perturbations with respect to unperturbed operator $\hat{J}(q_0)$, whose eigenvalues are $\omega^2_l=\omega^2_l(q_0)$ and whose eigenvectors $\vec{V}_{l}=\vec{V}_{l}(q_0)$ constitute a complete basis. According to standard perturbation theory \cite{coh} we write the expansions of the eigenvectors and eigenvalues, \begin{eqnarray} \label{v} \vec{V}_{l}(q_0+\epsilon q_1) &=&\vec{V}_{l}+ \epsilon \vec{V}^{(1)}_{l}q_1 + \epsilon^2 \vec{V}^{(2)}_{l}q_1^2/2 +\ldots \\ \label{omega} \omega_l(q_0+\epsilon q_1) &=&\omega_l+\epsilon \omega^{(1)}_l q_1 + \epsilon^2 \omega^{(2)}_l q_1^2/2+ \ldots \end{eqnarray} Equation (\ref{d}) has to be solved at each expansion order: \begin{eqnarray} \label{o1} \text{at order $\epsilon^0$ : }\quad \hat{J}\vec{V}_l &=& \omega^2_l \vec{V}_l \\ \label{o2} \text{at order $\epsilon^1$ : }\quad (\hat{J} \vec{V}^{(1)}_l+\hat{J}' \vec{V}_l)\;q_1 &=& (2\omega_l \omega^{(1)}_l \vec{V}_l+ \omega^2_l \vec{V}^{(1)}_l)\;q_1 \\ \label{o3} \text{at order $\epsilon^2$ : }\quad (\hat{J}' \vec{V}^{(1)}_l + \frac{1}{2} \hat{J}'' \vec{V}_l+ \frac{1}{2} \hat{J} \vec{V}^{(2)}_l)\;q_1^2 &=& {(\omega^{(1)}}^2_l \vec{V}_l +\omega_l \omega^{(2)}_l\vec{V}_l +2\omega_l \omega^{(1)}_l \vec{V}^{(1)}_l+ \frac{\omega^2_l}{2} \vec{V}^{(2)}_l)\;q_1^2 \,. \end{eqnarray} At order $\epsilon^0$, one solves the unperturbed problem determining $\vec{V}_l$ and $\omega_l$. At order $\epsilon$, one determines $\vec{V}^{(1)}_l$, $\omega^{(1)}_l$: imposing to $\vec{V}^{(1)}_l$ to be orthogonal to $\vec{V}_l$ to guarantee the normalization of $\vec{V}_l(q_0+\epsilon q_1)$ the scalar product of (\ref{o2}) with $\vec{V}^*_m \;\; (m\neq l)$ gives \begin{eqnarray} \label{vp} \vec{V}^{(1)}_l= \sum_{m\neq l} \alpha_m \vec{V}_m \\ \label{am} \alpha_m= \frac{ {\vec{V}_m}^* \hat{J}' \vec{V}_l}{ \omega^2_l-\omega^2_m} \;\;\;\;\;\;\; m \neq l \end{eqnarray} and that with $\vec{V}^*_l$ gives \begin{equation} \label{op} \omega^{(1)}_l= \frac{ {\vec{V}_l}^* \hat{J}' \vec{V}_l}{2 \omega_l}\,. \end{equation} At order $\epsilon^2$, one determines $\omega^{(2)}_l$ by multiplying (\ref{o3}) by $\vec{V}^*_l$ \begin{equation} \label{os} \omega^{(2)}_l= \frac{1}{\omega_l} \left({\vec{V}_l}^* \frac{\hat{J}''}{2} \vec{V}_l -{\omega^{(1)}}^2_l+ \sum_{m\neq l} \frac{| {\vec{V}_m}^* \hat{J}' \vec{V}_l|^2} { \omega^2_l-\omega^2_m} \right)\,. \end{equation} We assume, for sake of simplicity, that the eigenmodes of J (see (\ref{d})) are not degenerate, but the generalization of the degenerate case is straightforward with the standard perturbation theory. The phase of each component of (\ref{wpuno}) can be expanded around the central wave number $q_0$, up to second order in $\epsilon q_1=q-q_0$ using the values of $\omega^{(1)}_l$ and $\omega^{(2)}_l$ determined above \begin{equation} \label{wp} \vec{E}_l(n,t)= e^{i(q_0 n-\omega_l(q_0)t)} \epsilon \int_{-\Delta q/\epsilon}^{+\Delta q/\epsilon} A(q_0+ \epsilon q_1) \vec{V}_{l}(q_0+ \epsilon q_1) \; \exp{\left\{i \epsilon q_1(n-\omega^{(1)}_l(q_0)t) -i \epsilon^2 \frac{q_1^2}{2}\omega^{(2)}_l(q_0)t\right\}} \; dq_1+ c.c. \end{equation} Under this form, $\vec{E}_l(n,t)$ appears as a plane wave, henceforth called the carrier wave, with an amplitude that depends on space and time and which corresponds to the integral of equation (\ref{wp}), $\vec{E}_l(n,t) = \vec{F}(n,t) \; \exp[i(q_0 n-\omega_l(q_0)t)] +c.c.$. The fact that $\omega^{(1)}_l(q_0)$, $\omega^{(2)}_l(q_0)$ obey relations (\ref{op}) and (\ref{os}) ensures that this wave packet is a solution of the original equation (\ref{emoto}), up to the order of the various expansions. In order to extend the study to the nonlinear case it is useful to express these conditions under the form of an equation in the space time coordinates for the amplitude. Let us introduce the quantity \begin{equation} \label{ap} A(n,t)= \int_{-\Delta q/\epsilon}^{+\Delta q/\epsilon} A(q_0+ \epsilon q_1) \; \exp{\left\{i \epsilon q_1(n-\omega^{(1)}_l(q_0)t) -i \epsilon^2 \frac{q_1^2}{2}\omega^{(2)}_l(q_0)t\right\}} \; dq_1 \end{equation} Equation (\ref{ap}) shows that $A(n,t)$ slowly varies in space and time. In the spirit of the multiple scale expansion, it is natural to introduce the slow variables $x_1 = \epsilon n$, $t_1 = \epsilon t$ and $t_2 = \epsilon^2 t$ so that $A(n,t)$ can be written as \begin{equation} \label{ap1} A(n,t) = A(x_1,t_1,t_2) = \int_{-\Delta q/\epsilon}^{+\Delta q/\epsilon} A(q_0+ \epsilon q_1) \; \exp{\left\{i q_1( x_1-\omega^{(1)}_l(q_0)t_1) -i \frac{q_1^2}{2}\omega^{(2)}_l(q_0)t_2\right\}} \; dq_1 \end{equation} or \begin{equation} \label{ap2} A(x_1,t_1,t_2) = A(s_1,t_2) = \int_{-\Delta q/\epsilon}^{+\Delta q/\epsilon} A(q_0+ \epsilon q_1) \; \exp{\left\{i q_1 s_1 -i \frac{q_1^2}{2}\omega^{(2)}_l(q_0)t_2\right\}} \; dq_1 \end{equation} with the introduction of the variable $s_1 = x_1 - \omega^{(1)}_l(q_0)t_1$ to switch to the frame moving at the group velocity of the carrier wave. Using the relation $(\partial A / \partial x_1) = (\partial A / \partial s_1) = i<q_1> \equiv \int_{-\Delta q/\epsilon}^{+\Delta q/\epsilon} i q_1\; A(q_0+\epsilon q_1) \; \exp{\left\{i q_1 s_1 -i \frac{q_1^2}{2}\omega^{(2)}_l(q_0)t_2\right\}} \; dq_1$ that derives directly from Eqs.~(\ref{ap1}) and (\ref{ap2}), and the expansion (\ref{v}) of $V(q_0 + \epsilon q_1)$, the amplitude $\vec{F}$ of the wave can be expressed as a function of $A(s_1,t_2)$ by the relation \begin{equation} \label{f} \vec{F}(x_1,t_1,t_2)= \epsilon (\vec{V}_{l}-i \epsilon \vec {V'_l} \frac{\partial }{\partial x_1}) A(x_1,t_1,t_2) \; . \end{equation} {From} (\ref{ap1}) and (\ref{ap2}), we directly derive the equations of motions of $A$ as a function of the slow space--time variables: \begin{equation} \label{wp1} \left( \frac{\partial A}{\partial t_1}+ \omega^{(1)}_l\frac{\partial A}{\partial x_1} \right) =0 \end{equation} and \begin{equation} \label{ls} i \frac{\partial A }{\partial t_2}+ \frac{\omega^{(2)}_l}{2}\frac{\partial^2 A}{\partial s_1^2}=0 \end{equation} where $\omega^{(1)}_l$ and $\omega^{(2)}_l$ are then the group velocity and the group velocity dispersion of the wave packet and determine the peak velocity and the spread out of the envelope function. Equation (\ref{f}) shows that $\vec{V'}_{l}$ determines the first order correction to the direction of the vectorial field solution. \section{Non Linear Vectorial Lattice } We now consider the full equation of motion, including nonlinear on-site potential terms. Extra nonlinear terms depending on time derivatives may also appear in the case of non Cartesian coordinates systems: \begin{eqnarray} \label{emnl} \frac{\partial^2 E(n,\alpha,t)}{\partial t^2 } = - \sum_{n',\alpha'} J (n-n',\alpha,\alpha') E(n',\alpha',t)+ \nonumber\\ \sum_{d=0}^{2} \sum_{k=0}^{d}\sum_{\alpha'',\alpha'\leq \alpha''} c_{d,k}^{\alpha}(\alpha',\alpha'') E^{(k)}(n,\alpha',t)E^{(d-k)}(n,\alpha'',t) + \nonumber\\ \sum_{d=0}^{2} \sum_{k=0}^{d}\sum_{j=0}^{k} \sum_{\alpha''',\alpha''\leq \alpha''', \alpha'\leq \alpha'' } C_{d,k,j}^{\alpha}(\alpha',\alpha'',\alpha''') E^{(j)}(n,\alpha',t)E^{(k-j)}(n,\alpha'',t)E^{(d-k)}(n,\alpha''',t) \end{eqnarray} where $E^{(j)}(n,\alpha,t)$ indicates the $j-$th time derivative of $E(n,\alpha,t)$, and $c_{d,k}^{\alpha}(\alpha',\alpha'')$, $C_{d,k,j}^{\alpha}(\alpha',\alpha'',\alpha''')$ are the quadratic and cubic nonlinear terms numerical coefficients. Index $d$ is the total time derivative order of each term. The terms with $d=0$ are the nonlinear potential force terms; the others derive from the kinetic energy in the case of non Cartesian coordinates so that $d \leq 2$. The quadratic terms in (\ref{emnl}) give rise to second harmonic and constant terms that, if we look for small amplitude solution, have to be included as additional smaller corrections: \begin{eqnarray} \label{wpc} \vec{E}_l(n,t) &=& \epsilon \, e^{i(q_0n_0-\omega_l t_0)}\, (\vec{V}_{l}-i \epsilon \vec {V^{(1)}_l} \frac{\partial }{\partial x_1}) A(x_1,t_1,t_2) \nonumber \\ &+& \epsilon^2 \, e^{2i(q_0n_0-\omega_l t_0)}\, \vec\gamma_l \, A^2(x_1,t_1,t_2) +\epsilon^2 \, \vec\mu_l \, |A(x_1,t_1,t_2)|^2 + O(\epsilon^3) \,. \end{eqnarray} We are interested in situations where dispersion can balance nonlinearity, and therefore have to be measured by the same scaling parameter $\epsilon$. While the overall $\epsilon$ factor was not important in the linear case, it must be explicitly included in the nonlinear case. We solve the equation of motion (\ref{emnl}) on the three characteristic magnitude scales of the wave packet. At order $\epsilon$ we get $\omega_l$ ,$\vec{V}_{l}$ from equation (\ref{o1}). At order $\epsilon^2$ we get for the wave packet term expressed in the form (\ref{wpuno}) the system of equations (\ref{o2}) for each $q_1$ in the integration range. After integration on the envelope distribution this gives rise to the equation in the anti-transformed Fourier space \begin{equation} \label{oe} \left( 2\omega_l \vec{V}_l \frac{\partial }{\partial t_1} + (\hat{J} - \omega^2_l) \vec{V}^{(1)}_l \frac{\partial }{\partial x_1} + \hat{J}' \vec{V}_l \frac{\partial }{\partial x_1} \right) A(x_1,t_1,t_2)=0 \; . \end{equation} In fact from (\ref{ap1}), the average wave numbers deviation is $ <q_1>= -i \partial A / \partial x_1$ and the averaged frequency deviation is in the same way $<\Delta \omega_l>=<\omega_l^{(1)}q_1>= i \partial A / \partial t_1$. By scalar product of (\ref{oe}) and $V^*_m$ $ \forall m\neq l$ we obtain the components $\alpha_m$ (\ref{am}) of $\vec{V}^{(1)}_l$ (\ref{vp}) on the base $\{\vec{V}_m\}$ and by scalar product of (\ref{oe}) and $V^*_l$ we obtain the equation (\ref{wp1}) with $\omega^{(1)}_l$ defined by (\ref{op}). At the same order of expansion one determines the vectors $\vec{\gamma}_l$, $\vec{\mu}_l$ collecting the terms of corresponding order $(\epsilon^{2})$ and phase in the equation of motion (\ref{emnl}) in which the solution form (\ref{wpc}) has been inserted \cite{nota}. One then obtains $\vec{\gamma}_l$ by solving the algebraic system \begin{equation} \label{ga} (\hat{J}(2q_0)-4 \omega^2_l(q_0))\vec{\gamma}_l= \sum_{d=0}^{2}\sum_{k=0}^{d} \sum_{\alpha'',\alpha'\leq \alpha''} \vec{c}_{d,k}(\alpha',\alpha'') (-i\omega_l)^d V_l(\alpha')V_l(\alpha'') \end{equation} where $\vec{c}_{d,k}(\alpha',\alpha'')$ is the vector of components ${c}^{\alpha}_{d,k}(\alpha',\alpha'')$, each derivative with respect to $t_0$ giving a factor $(-i\omega_l)$. And $\vec{\mu}_l$ is obtained from the system \begin{equation} \label{mu} \hat{J}(0)\vec{\mu}_l= \sum_{d=0}^{2}\sum_{k=0}^{d} \sum_{\alpha'',\alpha'\leq \alpha''} \vec{c}_{d,k}(\alpha',\alpha'')[(i\omega_l)^k (-i\omega_l)^{d-k} V^*_l(\alpha')V_l(\alpha'')+(-i\omega_l)^k (i\omega_l)^{d-k} V^*_l(\alpha'')V_l(\alpha')] \,. \end{equation} At order $\epsilon^3$, from (\ref{emnl}) and (\ref{wpc}) for the terms in $ e^{i(q_0n_0-\omega_l(q_0)t_0)}$, one obtains the system of equations: \begin{eqnarray} \label{2r} \left( [(\frac{1}{2} \hat{J}''-{\omega^{(1)}_l}^2)\vec{V}_l+ (\hat{J}'-2\omega_l\omega^{(1)}_l)\vec{V}^{(1)}_l+ (\frac{1}{2}(\hat{J}-\omega^2_l) \vec{V}^{(2)})] \frac{\partial^2}{\partial s_1^2}+ 2 i\omega_l \vec{V}_l \frac{\partial}{\partial t_2}\right)\, A(s_1,t_2)+ \nonumber \\ \vec{\cal{Q}} |A(s_1,t_2)|^2 A(s_1,t_2) =0 \end{eqnarray} where \begin{eqnarray} \label{calq} \vec{\cal{Q}}= \sum_{d=0}^{2}\sum_{k=0}^{d} \sum_{\alpha'',\alpha'\leq \alpha''} \vec{c}_{d,k}(\alpha',\alpha'')[ (i\omega_l)^k (-2i\omega_l)^{d-k} V^*_l(\alpha')\gamma_l(\alpha'')+(-2i\omega_l)^k (i\omega_l)^{d-k} V^*_l(\alpha'')\gamma_l(\alpha')]+ \nonumber\\ 2\sum_{d=0}^{2}\sum_{\alpha'',\alpha'\leq \alpha''} \vec{c}_{d,0}(\alpha',\alpha'')(-i\omega_l)^d \mu_l(\alpha')V_l(\alpha'') +\nonumber \\ \sum_{d=0}^{2} \sum_{k=0}^{d}\sum_{j=0}^{k} \sum_{\alpha''',\alpha''\leq \alpha''', \alpha'\leq \alpha'' } \vec{C}_{d,k,j}(\alpha',\alpha'',\alpha''') [(-i\omega_l)^j(-i\omega_l)^{k-j} (i\omega_l)^{d-k} V_l(\alpha')V_l(\alpha'')V_l^*(\alpha''')+ \nonumber \\ (-i\omega_l)^j(i\omega_l)^{k-j} (-i\omega_l)^{d-k} V_l(\alpha')V^*_l(\alpha'')V_l(\alpha''')+ (i\omega_l)^j(-i\omega_l)^{k-j} (-i\omega_l)^{d-k} V^*_l(\alpha')V_l(\alpha'')V_l(\alpha''')] \end{eqnarray} The first line of the equation (\ref{2r}) corresponds to the third order expansion (\ref{o3}) of the linear operator equation (\ref{d}) applied to the wave packet (terms in $e^{i(q_0n_0-\omega_l(q_0)t_0)}$ in (\ref{wpc})), in the moving reference frame used in (\ref{ls}), $<q_1>= -i ({\partial }/{\partial s_1}) A(s_1,t_2)$, and $<\omega^{(2)}_l {q_1^2}/{2}> = <\Delta(\Delta(\omega_l))>= i (\partial / \partial t_2 ) A(s_1,t_2)$. The first two lines of the non linear coefficients vector $\vec{\cal{Q}}$ arise from the double product, in the nonlinear quadratic force terms in (\ref{emnl}), between $O(\epsilon)$ and $O(\epsilon^2)$ components of (\ref{wpc}); the last two lines come from the nonlinear cubic force terms in (\ref{emnl}) when considering just the $O(\epsilon)$ terms in $\vec{E}_l(n,t)$. Multiplying Eq.~(\ref{2r}) by $\vec{V}_l^*$ we obtain the NLS equation: \begin{equation} \label{nls} \left( P \frac{\partial^2}{\partial s_1^2}+ i \frac{\partial}{\partial t_2} \right) A(s_1,t_2)+ Q |A(s_1,t_2)|^2 A(s_1,t_2) =0 \end{equation} where \begin{equation} \label{p} P=\frac{1}{2\omega_l} \left( \vec{V}^*_l \frac{\hat{J}''}{2}\vec{V}_l - (\frac{ {\vec{V}_l}^* \hat{J}' \vec{V}_l}{2 \omega_l})^2 + \sum_{m\neq l} \frac{| {\vec{V}_m}^* \hat{J}' \vec{V}_l|^2} { \omega^2_l-\omega^2_m} \right) \end{equation} and \begin{equation} \label{q} Q=\frac{\vec{V}^*_l \vec{\cal{Q}}}{2\omega_l} \end{equation} The first part of equation (\ref{nls}) is equation (\ref{ls}) for the wave packet with $2P=\omega^{(2)}_l$ given by the (\ref{os}) to which is now added the nonlinear part with coefficient $Q$. If $PQ > 0$ then the effect of the amplitude dependent non linear potential well in (\ref{nls}) balances the wave packet group velocity dispersion giving rise to the stable envelope soliton solution \cite{rem}. \section{Summary.} The main outlines of the approach we introduce are the following. The envelope soliton like solutions arise in systems with weak dispersion and weak nonlinearity by two parallel series expansions driven by a common expansion parameter ($\epsilon$): on one hand the weakness of the diffusion, for a wave packet like solution, allows an expansion of the equations that regulate the space time behaviour of the solution on different scales; on the other hand the weak nonlinearity, for small amplitude solutions, allows to write the equations of motion at increasing orders of accuracy introducing the nonlinear terms in a progressive way. For scalar fields the Taylor series expansion of dispersion relations gives directly the diffusive part for the envelope equations of motion. Vectorial fields are instead characterized by a linear part which gives rise, in the $q$ space, to an eigenvalue (dispersion relations) and eigenvector (relative amplitude of the different components) problem (\ref{d}) ; to obtain the correct expansion in multiple scales it is then necessary to apply the perturbation theory (\ref{v}),(\ref{omega}). Finally, to combine this perturbative expansion with the nonlinear one, we antitransform the equations (\ref{o2}), (\ref{o3}) as done in (\ref{oe}), (\ref{2r}). Following the approach introduced in this paper it is straightforward to derive the NLS equation for every non linear vectorial lattice with on-site nonlinearities and with an arbitrary number of components. For more complex systems this could even be programmed in symbolic languages to provide a fully automatic method. After having identified the nonlinear coefficients $c_{d,k}^{\alpha}(\alpha',\alpha'')$ and $C_{d,k,j}^{\alpha}(\alpha',\alpha'',\alpha''')$ in the equation of motion (\ref{emnl}), there are only algebraic systems to solve: one has to solve the eigenvectors $V_l(q_0)$ and the eigenvalues $\omega_l(q_0)$ of the matrix $\hat{J}(q_0)$, then the systems (\ref{ga}) and (\ref{mu}) for $\gamma_l$ and $\mu_l$, and to derive $P$ from (\ref{p}) and $Q$ from (\ref{calq}) and (\ref{q}). {From} (\ref{nls}) one then obtains, if $PQ \geq 0$, the envelope function $A(x_1,t_1,t_2)$ that, inserted into (\ref{wpc}) together with the eigenvector correction (\ref{vp}), (\ref{am}), gives the complete $O(\epsilon^2)$ solution we are looking for. \section*{Acknowledgments} We thank R. Monasson for a critical reading of the manuscript.
\section{Introduction} Over the past few years our understanding of string theory has developed considerably. We now know that the five superstring theories as well as low-energy 11-dimensional supergravity are related through an intricate series of dualities and it has been argued that all these theories are limits of an underlying 11-dimensional theory called ``M-theory'' \cite{Witten-various} whose microscopic description is as yet unknown. It has been found that in addition to one-dimensional stringlike excitations there are higher-dimensional branes in each of these theories which may in some regimes be considered to be just as fundamental as the strings. In the five superstring theories there are D-branes of various dimensions \cite{Polchinski} as well as the fundamental string and NS5-branes. In M-theory there are M2-branes and M5-branes which are related to the branes of the superstring theories through various duality transformations. A fundamental class of problems is the identification of the world-volume action for the various branes appearing in the six theories of interest. This problem can be posed in a number of contexts of differing degrees of complexity. The simplest problem is to find the low-energy action for a single brane in a flat background metric with no nontrivial background supergravity fields. A more difficult problem is to find the action for a single brane in an arbitrary background metric and field configuration which satisfies the supergravity equations of motion. The problem can be made still harder by considering systems of many branes, either in a flat or general background. Even for a single fundamental superstring the action in a general background including arbitrary R-R fields is not yet well understood; for recent work in this direction see, for example, \cite{bvw} and references therein. For single D-branes the situation is somewhat better. The action for a single D-brane moving in a general background is the Born-Infeld action \cite{Leigh}, which reduces to the maximally supersymmetric $U(1)$ super Yang-Mills theory on the world volume in the case where the brane is almost flat and has only low-energy excitations. This action is supplemented by Chern-Simons type couplings to background R-R fields \cite{Douglas,Li-bound}. Even for the single D-brane, however, there are subtle issues involved in giving a world-volume supersymmetric description of the Born-Infeld theory. For systems of $N$ D-branes, it is known that the low-energy action for parallel branes in a flat background is given by the supersymmetric $U(N)$ super Yang-Mills theory found by dimensional reduction from 10D \cite{Witten-bound}. So far there has been little progress in describing the action governing systems of many D-branes in a general background. This problem is due in part to the absence of a nonabelian generalization of the Born-Infeld action (although one proposal for such an action was made in \cite{Tseytlin}), and in part to ordering problems which arise even in the low-energy theory in the presence of general backgrounds. In this paper we consider the simplest system of many D-branes in a general background: low-energy configurations of many D0-branes moving in an arbitrary but weak background of type IIA supergravity. According to an argument of Seiberg \cite{Seiberg-DLCQ} (see also \cite{Sen-DLCQ}), the action for such a system of D0-branes should be related to the DLCQ description of M-theory in an associated 11-dimensional supergravity background. This generalizes the BFSS matrix theory conjecture \cite{BFSS,Susskind-DLCQ}, which states that supersymmetric matrix quantum mechanics (the low-energy theory of $N$ D0-branes in flat space) gives a light-front description of M-theory in a flat background. In a previous paper \cite{Mark-Wati-3} we used a matrix theory formulation of the multipole moments of the supercurrent components in 11D supergravity (derived in \cite{Mark-Wati,Dan-Wati-2,Mark-Wati-3}) to propose an explicit description of the matrix theory action up to terms linear in the background fields, as well as an algorithm for using higher-loop calculations in matrix theory to find the higher order terms in the matrix theory action in general backgrounds. In this paper, we use our proposal for the general background Matrix theory action and follow the arguments of Seiberg to deduce the leading terms in the action for multiple D0-branes in weak type IIA supergravity backgrounds. We then perform some simple tests of the Matrix theory action and the related multiple D0-brane action. In the D0-brane case, we show that our prescription satisfies a constraint originally suggested by Douglas \cite{Douglas-curved} that the masses of off-diagonal matrix elements between a pair of separated D0-branes agree with the minimal geodesic length between the D0-branes. This property holds also in the Matrix theory case where the separated objects are a pair of gravitons, and we use it to show that the leading order potential between a pair of gravitons in a weakly curved Ricci-flat background is correctly reproduced by the proposed general background Matrix theory action. The paper is organized as follows. In Section 2 we review our proposal for the linear terms in the general background Matrix theory action. Then, using this action, we follow the arguments of Seiberg to deduce leading terms in the action for multiple D0 branes in an arbitrary weak type IIA supergravity background. In section 3 we describe tests of the IIA and matrix theory actions. We conclude in section 4 with a discussion of related issues and comments on further directions. \section{Linear coupling to backgrounds} In subsection \ref{sec:matrix-background} we recall the proposal made in \cite{Mark-Wati-3} for the terms in the action of matrix theory which are linear in the background fields. In subsection \ref{sec:0-backgrounds} we use the approach of Seiberg to relate this matrix theory action to an action for multiple D0-branes in IIA background fields. This allows us to deduce the leading terms in the multiple D0-brane action, which to the best of our knowledge have not been previously described. \subsection{Backgrounds in matrix theory} \label{sec:matrix-background} In \cite{Mark-Wati-3} we proposed that the linear effects of a general matrix theory background with metric $g_{IJ} = \eta_{IJ} + h_{IJ}$ and 3-form field $A_{IJK}$ could be described by supplementing the flat space matrix theory action \begin{equation} S_{{\rm flat}} = -{1 \over 2 R} \int d\tau \; {\rm Tr} \, \biggl\lbrace -D_{\tau} X_i D_{\tau} X_i + \frac{1}{2} [X_i,X_j] [X_i,X_j] + \Theta_{\alpha} D_{\tau} \Theta_{\alpha} - \Theta_{\alpha} \gamma^i_{\alpha \beta} [X_i,\Theta_{\beta}] \biggr\rbrace \label{eq:flat-action} \end{equation} with additional terms of the form \begin{eqnarray} S_{\rm weak} & = & \int d \tau \; \sum_{n = 0}^{\infty} \sum_{i_1, \ldots, i_n}\frac{1}{n!} \left( \frac{1}{2} T^{IJ (i_1 \cdots i_n)} \partial_{i_1} \cdots \partial_{i_n} h_{IJ} (0) + J^{IJK (i_1 \cdots i_n)} \partial_{i_1} \cdots \partial_{i_n} A_{IJK} (0) \right. \nonumber \\ & &\hspace{1.2in} \left. +M^{IJKLMN (i_1 \cdots i_n)} \partial_{i_1} \cdots \partial_{i_n} A^D_{IJKLMN} (0) + {\rm fermion \; terms} \right) \label{eq:general-background} \end{eqnarray} where $A^D$ is the dual 6-form field which satisfies at linear order \[ dA^D ={}^* (dA). \] In (\ref{eq:general-background}) the matrix expressions $T^{IJ (i_1 \cdots i_n)}, J^{IJK (i_1 \cdots i_n)}, M^{IJKLMN (i_1 \cdots i_n)} $ are the matrix theory forms of the multipole moments of the stress-energy tensor, membrane current and 5-brane current of 11D supergravity. Explicit forms for the parts of these moments depending only on the 9 bosonic transverse matrices $X^i$ were given in \cite{Dan-Wati-2}, and the terms quadratic in the fermions were given in \cite{Mark-Wati-3}, as well as some terms quartic in the fermions. For example, the zeroeth moments of the components of the stress-energy tensor are given by \begin{eqnarray} T^{++} &=& {1 \over R}{\rm STr} \,\left({\rlap{1} \hskip 1.6pt \hbox{1}}\right)\nonumber\\ T^{+i} &=& {1 \over R}{\rm STr} \,\left(D_t X_i\right)\nonumber\\ T^{+-} &=& {1 \over R}{\rm STr} \,\left({1 \over 2} D_t X_i D_t X_i + {1 \over 4} F_{ij} F_{ij} + {1 \over 2} \Theta\gamma^i[X^i,\Theta]\right)\nonumber\\ T^{ij} &=& {1 \over R}{\rm STr} \,\left( D_t X_i D_t X_j + F_{ik} F_{kj} - {1 \over 4} \Theta\gamma^i[X_j,\Theta] - {1 \over 4} \Theta\gamma^j[X_i,\Theta]\right)\nonumber\\ T^{-i} &=& {1 \over R} {\rm STr} \,\left({1 \over 2}D_t X_iD_t X_jD_t X_j + {1 \over 4} D_t X_i F_{jk} F_{jk} + F_{ij} F_{jk} D_t X_k\right) \label{eq:stress-tensor}\\ & & - {1 \over 4R} {\rm STr} \,\left(\Theta_\alpha D_t X_k[X_m,\Theta_\beta]\right)\{\gamma^k\delta_{im} +\gamma^i\delta_{mk} -2\gamma^m\delta_{ki} \}_{\alpha \beta}\nonumber\\ & & - {1 \over 8R} {\rm STr} \,\left(\Theta_{\alpha} F_{kl}[X_m,\Theta_{\beta}]\right)\{ \gamma^{[iklm]} + 2 \gamma^{[lm]} \delta_{ki} + 4\delta_{ki}\delta_{lm} \}_{\alpha \beta}\nonumber\\ & & + {i \over 8R} {\rm Tr} \,(\Theta \gamma^{[ki]} \Theta \; \Theta \gamma^k \Theta)\nonumber\\ T_f^{--} &=& {1 \over 4R} {\rm STr} \,\left(F_{ab}F_{bc}F_{cd}F_{da} - {1 \over 4}F_{ab} F_{ab} F_{cd} F_{cd} + {\Theta} \Gamma^b \Gamma^c \Gamma^d F_{ab} F_{cd} D_a\Theta + {\cal O} ({\Theta^4})\right)\nonumber \end{eqnarray} where ${\rm STr}$ indicates a trace which is symmetrized over all orderings of terms of the forms $F_{ab}, \Theta$ and $[X^i, \Theta]$, indices $i (a)$ run from 1 (0) through 9, and we have defined $F_{0i} = D_t{X}^i, F_{ij} = i[X^i, X^j]$. There are two types of terms which contribute to higher moments of these components of the stress-energy tensor \begin{equation} T^{IJ (i_1 i_2 \cdots i_n)} = {\rm Sym} \, (T^{IJ}; X^{i_1}, X^{i_2}, \ldots, X^{i_n}) + T_{\rm fermion}^{IJ(i_1 i_2 \cdots i_n)} \end{equation} The contributions ${\rm Sym} \, ({\rm STr}\; (Y); X^{i_1}, \ldots, X^{i_n})$ are defined as the symmetrized average over all possible orderings when the matrices $X^{i_k}$ are inserted into the trace of any product $Y$ of the forms $F_{ab}, \Theta, [X^i, \Theta]$. In general there are additional fermionic contributions of arbitrary order to the higher multipole moments, of which the simplest example is the spin contribution to the angular momentum \begin{equation} T_{\rm fermion}^{+i(j)} = {1 \over 8R} {\rm Tr} \,(\Theta \gamma^{[ij]} \Theta) \end{equation} The precise form of these fermionic contributions will not be important to us in this paper, for reasons which will be discussed in section \ref{sec:oscillator-masses}. The results of \cite{Mark-Wati-3} for the matrix membrane and 5-brane currents are reproduced in the Appendix for convenience. With these definitions, (\ref{eq:general-background}) gives a formulation of matrix theory in a weak background metric to first order in the metric $h_{IJ}$, the 3-form $A_{IJK}$, and all their higher derivatives. It was argued in \cite{Mark-Wati-3} that if the matrix theory conjecture is true in flat space, this formulation must be correct at least to order $\partial^4h, \partial^4A$ for a class of backgrounds which can be produced as the long range fields around supergravity sources described by matrix theory objects. We conjectured further that this form may work to all orders and in a general background. It should be emphasized, however, that this formulation can only be given for M-theory backgrounds with a global $U(1)$ symmetry around a compact direction, as we do not know how to incorporate dependence of the background on the compact coordinate $x^-$. We only expect this action to be part of a consistent all-orders matrix theory action in a general background when the background satisfies the equations of motion. The derivation of this action also depended upon a particular choice of gauge for the graviton, so that it may be necessary to restrict attention to background fields satisfying the linearized gauge \begin{equation} \partial^I \bar{h}_{IJ} =\partial^I (h_{IJ}-\frac{1}{2} \eta_{IJ}h_{K}^{\; K}) = 0. \end{equation} \subsection{Backgrounds for D0-branes} \label{sec:0-backgrounds} We now investigate how the Matrix theory action described in the previous section is related to the action for multiple D0-branes in background type IIA supergravity fields. In the case of a flat background, the Matrix theory action may be derived by showing an equivalence between the DLCQ limit of M-theory in a flat background with N units of momentum around the circle and a particular limit of type IIA string theory with N D0-branes \cite{Seiberg-DLCQ}. In this limit, the only remaining degrees of freedom are the lowest energy modes of open strings ending on the N D0-branes. The dynamics of these modes are in general described by a non-abelian generalization of the Born-Infeld action whose complete form is not known. However, in the appropriate limit of type IIA string theory, most of the terms in this action vanish, and we find that the dynamics of DLCQ M-theory in a flat background are described by an action equivalent to the dimensional reduction of D=10 super Yang-Mills theory to 0+1 dimensions. The action for Matrix theory with background supergravity fields given in the previous section has been derived completely within the context of Matrix theory. However, in principle, one should be able to apply Seiberg's arguments to this case also and derive the same action as a limit of the action for D0-branes in type IIA string theory with background supergravity fields. Again, only particular terms in the D-brane action will survive in the appropriate limit, but unlike the flat space case, not even these terms are known except in the case of a single brane. Hence, in the case $N=1$, we should be able to rederive our result from previously known facts about D-branes, but more importantly, we will be able to apply the arguments in reverse for $N>1$ to derive previously unknown leading terms in the action for multiple D0-branes in an arbitrary weak type IIA supergravity background. Using T-duality, our result may be extended to give leading terms in the actions for all other types of D-branes. \subsubsection{Relationship between DLCQ and type IIA backgrounds} We now review the steps taken in \cite{Seiberg-DLCQ} as they apply in the case of weak backgrounds to make precise the relationship between the matrix theory action and the multiple D0-brane actions. In particular, we must determine the relationship between the D=11 supergravity fields appearing in the Matrix theory action (\ref{eq:general-background}) and the type IIA supergravity fields appearing in the related D0-brane action. We start by considering M-theory with background metric \[ {g}_{IJ} = \eta_{IJ} + {h}_{IJ} \] in a frame with a compact coordinate $x^-$ of size $R$ which is light-like in the flat space limit $h_{IJ} = 0$. This theory can be described as a limit of a family of space-like compactified theories. Defining an $\hat{M}$-theory with background metric \[ \hat{g}_{IJ} = \eta_{IJ} + \hat{h}_{IJ} \] in a frame with a spacelike compact direction $x^{10}$ of size $R_s$, the DLCQ limit of the original M-theory can be found by boosting the $\hat{M}$-theory in the $x^{10}$ direction with boost parameter \[ \gamma = \sqrt{{R^2 \over 2 R_s^2} + 1} \] and then taking a limit $R_s \rightarrow 0$. The metric $\hat{g}_{IJ}$ in the $\hat{M}$-theory is related to that of the original M-theory by \begin{eqnarray*} \hat{h}_{ij} &=& h_{ij}\\ \hat{h}_{0\,i} &=& {\alpha \over \sqrt{2}}h_{+i} + {1 \over \alpha\sqrt{2}} h_{-i}\\ \hat{h}_{10\,i} &=& {\alpha \over \sqrt{2}}h_{+i} - {1 \over \alpha\sqrt{2}} h_{-i}\\ \hat{h}_{0\,0} &=& h_{+-} + {\alpha^2 \over 2}h_{++} + {1 \over 2 \alpha^2}h_{--}\\ \hat{h}_{10 \, 10} &=& -h_{+-} + {\alpha^2 \over 2}h_{++} + {1 \over 2 \alpha^2}h_{--}\\ \hat{h}_{0\,10} &=& {\alpha^2 \over 2}h_{++} - {1 \over 2 \alpha^2}h_{--}\\ \end{eqnarray*} where we have defined \begin{eqnarray*} \alpha &=& \gamma(1-v) = \gamma - \sqrt{\gamma^2 - 1}\\ &=& {R_s \over R \sqrt{2}} + {\cal O}((R_s/R)^3) \end{eqnarray*} M-theory on a small spacelike circle of radius $R_s$ is equivalent to type IIA string theory with background fields given to leading order by: \begin{eqnarray*} h^{IIA}_{\mu \nu} &=& \hat{h}_{\mu \nu} + {1 \over 2} \eta_{\mu \nu} \hat{h}_{10\,10}\\ C_{\mu} &=& \hat{h}_{10 \, \mu}\\ \phi &=& {3 \over 4} \hat{h}_{10 \, 10} \end{eqnarray*} and parameters \[ g_s = (R_s M_p)^{3/2}, \; \; \; M_s = R_s^{1/2}M_p^{3/2} \] where $M_p$ is the eleven-dimensional Planck mass. Here we have defined $g_s$ to be a constant and chosen the dilaton $\phi$ so that $\phi=0$ in the case of a circle of constant size $R_s$ with $h_{10 \; 10} = 0$. Thus the effective string coupling is given locally by the combination \[ g_s(\vec{x}) = g_s e^{\phi} . \] \\ Combining the two equivalences, we conclude that DLCQ M-theory with N units of momentum on a lightlike circle of size $R$ and background metric $g_{IJ}$ is equivalent to the $R_s \rightarrow 0$ limit of type IIA string theory with N D0-branes, parameters \[ g_s = (R_s M_p)^{3/2}, \; \; \; M_s = R_s^{1/2}M_p^{3/2} \] and background fields \begin{eqnarray} h^{IIA}_{0 0} &=& {3 \over 2}h_{+-} + {\alpha^2 \over 4}h_{++} + {1 \over 4 \alpha^2} h_{--} \nonumber\\ h^{IIA}_{0 i} &=& {\alpha \over \sqrt{2}} h_{+i} + {1 \over \alpha \sqrt{2}} h_{-i} \nonumber\\ h^{IIA}_{i j} &=& h_{ij} + {1 \over 2} \delta_{ij} (-h_{+-} + {\alpha^2 \over 2}h_{++} + {1 \over 2 \alpha^2} h_{--}) \label{eq:relations}\\ \phi &=& -{3 \over 4}h_{+-} + {3 \alpha^2 \over 8}h_{++} + {3 \over 8 \alpha^2} h_{--}\nonumber\\ C_0 &=& {\alpha^2 \over 2}h_{++} - {1 \over 2 \alpha^2} h_{--} \nonumber\\ C_i &=& {\alpha \over \sqrt{2}} h_{+i} - {1 \over \alpha \sqrt{2}} h_{-i} \nonumber \end{eqnarray} At first glance, such a limit seems problematic. In particular, it appears that for fixed finite values of the DLCQ metric components, the background fields of the equivalent type IIA theory diverge in the limit $R_s \rightarrow 0$ since $1/\alpha \rightarrow \infty$. However, recall from the flat space case that without a further rescaling of the parameters in the type IIA picture, the energies of the states we are interested in go to 0 like $R_s$. As we shall see, the appropriate rescaling of parameters which makes the energies we are interested in finite without changing the physics also ensures that the apparent divergences of background field components do not lead to divergent terms in the final action. Another feature of this action is that after the appropriate rescaling the characteristic length scale $L$ associated with the structure of the metric becomes much smaller than the string length $1/M_s$. While this may seem unusual, it is precisely what is needed for the physics of the system to be completely captured by the open string theory describing the D0-brane theory at substring scales studied in \cite{DKPS}. Indeed, for compact manifolds such as tori, it is this effect which makes it possible for the wrapped string modes corresponding to momentum excitations on the dual space to become physically relevant \cite{WT-compact,Seiberg-DLCQ}. \subsubsection{$N=1$ actions} We now use the correspondence just discussed to make an explicit comparison between the matrix theory and IIA descriptions of a system of $N$ 0-branes in a weak background field. We begin with the case $N=1$. Here, both the Matrix theory and D0-brane actions are known, so we would like to check that the Matrix theory action may be derived from the D0-brane action before proceeding to the case $N>1$ where the D0-brane action is not known. For the case $N=1$, the Matrix theory action (\ref{eq:flat-action},\ref{eq:general-background}) reduces to \begin{eqnarray} S & = & {1 \over R} \int dt \left\{ \frac{\dot{x}^2}{2} + \frac{1}{2} h_{++}(\vec{x}) + h_{+i}(\vec{x}) \dot{x}^i + {1 \over 2} h_{ij}(\vec{x}) \dot{x}^i \dot{x}^j \label{graviton} \right.\\ & &\hspace{1in} + \left.\frac{1}{2} h_{+-}(\vec{x})\dot{x}^2 + \frac{1}{2} h_{-i}(\vec{x}) \dot{x}^2 \dot{x}^i + \frac{1}{8} h_{--}(\vec{x}) \dot{x}^4 \right\}. \nonumber \end{eqnarray} In this case we expect the action to describe a single graviton in curved space with unit momentum along the lightlike circle. Such an action was derived from supergravity in \cite{Mark-Wati}; expression (\ref{graviton}) is indeed identical to the supergravity result given by equation (13) in that paper. The world-volume action for a single D0-brane moving in a general type IIA background supergravity fields is given by \begin{equation} S_{IIA} = - \tau_0\int d \xi e^{-\phi} \sqrt{g_{\mu \nu} (d{x}^{\mu}/d \xi ) (d{x}^{\nu}/d \xi)} + \tau_0 \int C_{\mu} dx^{\mu} \label{eq:single-0} \end{equation} where $\phi$, $g_{\mu \nu}$, and $C_{\mu}$ are the background dilaton, metric, and R-R one-form fields, and the parameter $\tau_0$ is the D0 mass, given by \[ \tau_0 = {M_s \over g_s} . \] One can also consider background R-R three form $C_{\mu \nu \lambda}$ and NS-NS antisymmetric tensor $B_{\mu \nu}$ fields, but these do not couple to a single zero-brane. According to the equivalence presented in the previous section, the Matrix theory action (\ref{graviton}) should arise from the D0-brane action (\ref{eq:single-0}) by rewriting the type IIA background fields in terms of the desired DLCQ supergravity background using the relations (\ref{eq:relations}), then rescaling parameters and taking the limit $R_s \rightarrow 0$. We will now verify this explicitly. Choosing a gauge in which the coordinate time $x^0$ is identified with the worldvolume time $\xi$ we first expand the D0-brane action to leading order in the background fields, giving \begin{equation} S = -\tau_0 \int d\xi \left\{ (1-v^2)^{1/2} (1- \phi) - {1 \over 2} (1-v^2)^{-1/2}(h^{IIA}_{00} + 2h_{0i}v^i + h_{ij}v^iv^j) - C_0 - C_iv^i \right\} \label{eq:expand-Born-Infeld} \end{equation} where $v^i \equiv \dot{x}^i$. We now write the IIA background fields in terms of the background fields in the equivalent DLCQ M-theory using (\ref{eq:relations}). Keeping only the leading term in $R_s/R$ for each of the components of the metric $h_{IJ}$, we find \begin{eqnarray} S & = & {1 \over R_s} \int d \xi \left\{ -1 + {1 \over 2} \frac{R^2_s}{R^2} h_{++}(\vec{x}) + \frac{R_s}{ R} h_{+i}(\vec{x}) v^i\right. \label{eq:replaced}\\ & &\hspace{1in} \left. + {1 \over 2} h_{ij}(\vec{x}) v^i v^j + \frac{1}{2} h_{+-}(\vec{x})v^2 + \frac{v^2}{2}+\frac{1}{2} \frac{R}{R_s} h_{-i}(\vec{x}) v^2 v^i + {1 \over 8} \frac{R^2}{R_s^2} h_{--}(\vec{x}) v^4 \right\}. \nonumber \end{eqnarray} Many of these terms seem to diverge in the $R_s \rightarrow 0$ limit we are interested in. However, as mentioned above, this scaling is deceptive, since we must rescale parameters in the theory so that the energies of the states we are interested in remain finite rather than going to zero in the limit. Indeed, from the fact that the conjugate momentum has a leading term of order $v/R_s$ it can be seen that all the terms in (\ref{eq:replaced}) which are linear in the background contribute to the Hamiltonian at order $R_s$. Thus, as we need for the Seiberg limit, the energy of the states of interest scale as $R_s$. We may now perform the rescaling of \cite{Seiberg-DLCQ} by replacing \[ R \rightarrow ({R_s \over R})^{1/2}R , \;\;\;\;\; \, \, \vec{x} \rightarrow ({R_s \over R})^{1/2}\vec{x}, \;\;\;\;\; \, \, h(\vec{x}) \rightarrow h(\vec{x}). \] Note that the change of variables in the second replacement combines with the rescaling of dimensionful coefficients in the expansion of $h$ to leave $h(\vec{x})$ unchanged, as suggested by the final replacement. With these redefinitions the action (\ref{eq:replaced}) becomes \begin{eqnarray} S & = & \int d \xi \left\{ -\frac{1}{R_s} + {1 \over R}\left( {1 \over 2} h_{++}(\vec{x}) + h_{+i}(\vec{x}) v^i+ {1 \over 2} h_{ij}(\vec{x}) v^i v^j\right. \right.\nonumber\\ & &\left.\left.\hspace{1.5in} + \frac{1}{2} h_{+-}(\vec{x})v^2 + \frac{v^2}{2} +\frac{1}{2} h_{-i}(\vec{x}) v^2 v^i + {1 \over 8} h_{--}(\vec{x}) v^4 \right)\right\}. \nonumber \end{eqnarray} The first term is divergent in the $R_s \rightarrow 0$ limit and arises from the BPS energy of the single 0-brane; this term also appears in the flat space theory and is discounted in the matrix theory limit. Dropping this term gives precisely the matrix theory action described by (\ref{graviton}) in the $N = 1$ case. Thus, we have shown that the known Born-Infeld action for a single D0-brane correctly reproduces the matrix theory action in a weak background when the proper limit is taken. \subsubsection{$N>1$ actions} We now turn to the case $N>1$. Here, the appropriate action for multiple D0-branes is not known, but by requiring that it reproduces the general background Matrix theory action in the Seiberg limit, we will be able to deduce its leading terms. We first write down the D0-brane action in terms of the unknown quantities coupling to the background fields. We define quantities $I_x$ coupling linearly to each of the background fields, so that to leading order in the background fields, the action for $N$ D0 branes is \begin{eqnarray} S &= & S_{{\rm flat}} + \int dt \sum_{n=0}^{\infty} {1 \over n!} \left[ \frac{1}{2} (\partial_{k_1}\cdots \partial_{k_n} h^{IIA}_{\mu \nu}) \; I_h^{\mu \nu (k_1 \cdots k_n)} + (\partial_{k_1}\cdots \partial_{k_n} \phi) \; I_{\phi}^{(k_1 \cdots k_n)}\right.\label{eq:IIA-general}\\ & & \hspace{1in}+ (\partial_{k_1}\cdots \partial_{k_n} C_{\mu }) \; I_0^{\mu (k_1 \cdots k_n)} + (\partial_{k_1}\cdots \partial_{k_n} \tilde{C}_{\mu \nu \lambda \rho \sigma \tau \zeta }) \; I_6^{\mu \nu \lambda \rho \sigma \tau \zeta (k_1 \cdots k_n)} \nonumber\\ & & \hspace{1in}+ (\partial_{k_1}\cdots \partial_{k_n} B_{\mu \nu}) \; I_s^{\mu \nu (k_1 \cdots k_n)} + (\partial_{k_1}\cdots \partial_{k_n} \tilde{B}_{\mu \nu \lambda \rho \sigma \tau}) \; I_5^{\mu \nu \lambda \rho \sigma \tau (k_1 \cdots k_n)} \nonumber\\ & &\hspace{1in} \left.+ (\partial_{k_1}\cdots \partial_{k_n} C^{(3)}_{\mu \nu \lambda }) \; I_2^{\mu \nu \lambda (k_1 \cdots k_n)} + (\partial_{k_1}\cdots \partial_{k_n} \tilde{C}^{(3)}_{\mu \nu \lambda \rho \sigma }) \; I_4^{\mu \nu \lambda \rho \sigma (k_1 \cdots k_n)}\right] \nonumber \end{eqnarray} Here, $S_{{\rm flat}}$ is the flat space action for N D0-branes, whose leading terms are the dimensional reduction of D=10 SYM theory to 0+1 dimensions. The complete form of the higher order terms in the flat space action is not known, but these terms vanish in the Matrix theory limit. We assume that the background satisfies the source-free IIA supergravity equations of motion so that the dual fields $\tilde{C}, \tilde{B}, \tilde{C}^{(3)}$ are well-defined 7-, 6- and 5-form fields given (at linear order) by \begin{equation} d \tilde{C} ={}^* dC, \;\;\;\;\; d \tilde{B} ={}^* dB, \;\;\;\;\; d \tilde{C}^{(3)} ={}^* dC^{(3)}. \end{equation} The sources $I_{2n}$ are associated with Dirichlet $2n-$brane currents, while the sources $I_{s}$ and $I_{5}$ are associated with fundamental string and NS5-brane currents respectively. It may seem surprising that a system of D0-branes can give rise to a nonzero D2-brane, D4-brane or D6-brane charge. Indeed, the integrated higher brane charges must vanish for a system containing a finite number $N$ of D0-branes. Even for $N = 2$, however, a D0-brane configuration can have nonvanishing multipole moments of higher D-brane charges. This essentially arises as the T-dual of the result that the $n$th power of the curvature form $F$ on a Dirichlet D$p$-brane carries $(p-2n)$-brane charge \cite{Witten-small,Douglas}; see \cite{WT-Trieste} and references therein for a further discussion of this issue. The problem we address in this subsection is the determination of the IIA currents $I_x$ under the assumption that this action reproduces the matrix theory action (\ref{eq:general-background}) in the Seiberg limit. As we will see, the leading terms in all the currents other than $I_5$ can be determined and are related to the matrix theory supercurrent components tabulated in the Appendix. For the case $N=1$ we see from (\ref{eq:expand-Born-Infeld}) that the nonvanishing source components $I_x$ are \begin{eqnarray} I_h^{00(k_1\cdots k_n)} &=& {1 \over R_s}(1-\dot{x}^2)^{-1/2} x^{k_1} \cdots x^{k_n} \nonumber\\ I_h^{0i(k_1\cdots k_n)} &=& {1 \over R_s}(1-\dot{x}^2)^{-1/2} \dot{x}^i x^{k_1} \cdots x^{k_n} \nonumber\\ I_h^{ij(k_1\cdots k_n)} &=& {1 \over R_s}(1-\dot{x}^2)^{-1/2} \dot{x}^i \dot{x}^j x^{k_1} \cdots x^{k_n} \label{eq:n1}\\ I_\phi^{(k_1\cdots k_n)} &=& {1 \over R_s}(1-\dot{x}^2)^{1/2} x^{k_1} \cdots x^{k_n} \nonumber\\ I_0^{0(k_1\cdots k_n)} &=& {1 \over R_s} x^{k_1} \cdots x^{k_n} \nonumber\\ I_0^{i(k_1\cdots k_n)} &=& {1 \over R_s}\dot{x}^i x^{k_1} \cdots x^{k_n} \nonumber \end{eqnarray} In the nonabelian case $N > 1$, the quantities $I_x$ will be some complicated functions of the $N \times N$ hermitian matrices $X^i$ as well as the fermionic matrices $\Theta$. For each $I$, we can make an expansion analogous to expanding in velocities for the $N=1$ case. We write \[ I_x = \sum I_{x[n]} \] where $n$ counts the dimension of a function of the matrices $X, \Theta$, giving $X$ dimension 1, $\dot{X}$ dimension 2 and $\Theta$ dimension 3/2. If we do a similar expansion for the flat space action $S_0$, we find that it is precisely the $n=4$ terms that remain in the Matrix theory limit, the higher order terms being scaled to zero. In the general background case, the Matrix theory action will arise from terms $I_n$ with $n \le 8$ (and their higher moments), so it is these terms that our analysis will determine. We now proceed just as in the $N=1$ case. We begin by working through the details of the analysis for those terms coupling to the IIA graviton, dilaton and R-R 1-form field. These terms are the most complicated. The analysis for the remaining bosonic fields is described at the end of this section. By the Seiberg equivalences, the Matrix theory action with background supergravity fields should result from replacing the IIA background fields in (\ref{eq:IIA-general}) with their DLCQ counterparts (\ref{eq:relations}), rescaling parameters as above, and taking the limit $R_s \rightarrow 0$. Before rescaling, we find that the D0-brane action becomes \begin{eqnarray} S &= & S_{\rm flat} - \frac{1}{2}\int dt \left[ (\sqrt{2} \alpha)^{-2} h_{--} \{ {1 \over 2} I_h^{00} + {1 \over 2} I_h^{ii} + {3 \over 2} I_\phi - 2 I_0^0 \} \right. \nonumber\\ & &\hspace{1in} + (\sqrt{2} \alpha)^{-1} h_{-i} \{2 I_h^{0i} - 2 I_0^i \}\nonumber\\ & &\hspace{1in} + h_{ij} \{ I_h^{ij} \}\nonumber\\ & &\hspace{1in} + h_{+-} \{ {3 \over 2} I_h^{00} - {1 \over 2} I_h^{ii} - {3 \over 2} I_\phi \}\label{eq:IIA-rewritten}\\ & &\hspace{1in} + (\sqrt{2} \alpha) h_{+i} \{ I_h^{0i} + I_0^i \}\nonumber\\ & &\hspace{1in} + (\sqrt{2} \alpha)^{2} h_{++} \{ {1 \over 8} I_h^{00} + {1 \over 8} I_h^{ii} + {3 \over 8} I_\phi + {1 \over 2} I_0^0 \} \nonumber\\ & &\hspace{1in} + \left. \{ {\rm higher \; moment \; terms} \}\right] \nonumber \end{eqnarray} The higher moment terms have exactly the same form as the terms written, for example the full set of terms linear in $h_{-i}$ would be \[ \sum_{n=0}^{\infty} {R \over R_s} \partial_{k_1} \cdots \partial_{k_n} h_{-i} \{2 I_h^{0i(k_1 \cdots k_n)} - 2I_0^{i(k_1 \cdots k_n)} \} \] Because the distance scale associated with the metric is rescaled along with the transverse coordinates in the rescaling of \cite{Seiberg-DLCQ}, the rescaling of the partial derivatives in these expressions cancel the scaling of the moment indices. Thus, the higher moment terms which remain in the $R_s \rightarrow 0$ limit are precisely those corresponding to 0th moments which remain in the limit. The only terms in (\ref{eq:IIA-rewritten}) which remain in the limit other than the leading divergent D0-brane energy term should be finite terms corresponding to the matrix theory action (\ref{eq:general-background}). All terms which are linear in the background and carry positive powers of $R/R_s$ in the limit must cancel for the IIA action to agree with the matrix theory action. This gives a number of restrictions on the parts of the IIA currents with particular scaling dimensions. The constraints arising in this fashion for the integrated (monopole) currents are \junk{ In order to see what happens to this action under the appropriate rescaling and change of variables, we must understand how the various terms $I_[n]$ are affected. In the $N=1$ case considered above, each $I_[n]$ consists of a single term and it is easy to see that the general rule is \begin{equation} (\partial_{k_1} \cdots \partial_{k_l} h) I_n \rightarrow ({R_s \over R})^{(l+n)/2} (\partial_{k_1} \cdots \partial_{k_l} h) I_n \label{scaling} \end{equation} For $N>1$, the $I_n$ will involve multiple terms involving both bosonic and fermionic fields. However, in order that the flat space Matrix theory arise correctly from the terms $(L_{flat})_2$ in the D0-brane action, it must be that all terms in a given $I_n$ transform in the same way so that the relations (\ref{scaling}) are still valid. The expansion coefficients of the metric \[ \partial_{k_1} \cdots \partial_{k_l} h \] have mass dimension $l$, so they are multiplied by $(R/R_s)^{l/2}$ under the rescaling. Also, the explicit factors of $R$ in the action (*) will be rescaled by \[ R \rightarrow ({R_s \over R})^{1/2} R \] Applying these transformations to the action (*), we find that the resulting overall power of $R_s$ multiplying a term \[ (\partial_{k_1} \cdots \partial_{k_l} h) I_n \] in the final action is $(R_s)^{(n+k-2)/2}$ where $2 \ge k \ge -2$ counts the number of $+$ indices minus the number of $-$ indices on $h$. Thus, for a given component of $h$, the terms which are finite in the limit have $n=2-k$. It is these terms that will remain to give the Matrix theory action, and the sum of all such terms for the component $h_{IJ}$ must therefore equal $T^{IJ}/2$. Terms with $n<2-k$ become infinite in the Matrix theory limit, so the sum of such terms coupling to a given component of $h$ must be required to cancel. T$n>2-k$ vanish in the Matrix theory limit, so knowledge of the Matrix theory action imposes no further constraints here. With this in mind, we may now compare the action (*) with the matrix theory action to find the following relations:} \begin{eqnarray} ({1 \over 2} I_h^{00} + {1 \over 2} I_h^{ii} + {3 \over 2} I_\phi - 2I_0^0)_0 &=& 0 \nonumber\\ ({1 \over 2} I_h^{00} + {1 \over 2} I_h^{ii} + {3 \over 2} I_\phi - 2I_0^0)_4 &=& 0 \nonumber\\ ({1 \over 2} I_h^{00} + {1 \over 2} I_h^{ii} + {3 \over 2} I_\phi - 2I_0^0)_8 &=&T^{--} \nonumber\\ (I_h^{0i} - I_0^i)_2 &=& 0\nonumber\\ (I_h^{0i} - I_0^i)_6 &=& T^{-i}\nonumber\\ (I_h^{ij})_0 &=& 0\label{eq:constraints}\\ (I_h^{ij})_4 &=& T^{ij}\nonumber\\ ({3 \over 2} I_h^{00} - {1 \over 2} I_h^{ii} - {3 \over 2} I_\phi )_0 &=& 0\nonumber\\ ({3 \over 2} I_h^{00} - {1 \over 2} I_h^{ii} - {3 \over 2} I_\phi )_4 &=& 2T^{+-}\nonumber\\ ( I_h^{0i} + I_0^i)_2 &=& 2T^{+i}\nonumber\\ ({1 \over 8} I_h^{00} + {1 \over 8} I_h^{ii} + {3 \over 8} I_\phi + {1 \over 2} I_0^0 )_0 &=& T^{++} \nonumber \end{eqnarray} We will assume that the degrees at which a given current $I_x$ has nonvanishing contributions are the same as in the $N = 1$ case. These are the terms for which we have explicitly written constraints in (\ref{eq:constraints}). This assumption agrees with what we know about the nonabelian Born-Infeld action. If this assumption is incorrect, there may be additional terms appearing in the IIA currents at other orders which do not contribute to the matrix theory action. Identical relations to (\ref{eq:constraints}) must hold for the quantities coupling to higher order terms in the Taylor expansion of the metric. Solving the constraints (\ref{eq:constraints}), we find \begin{eqnarray} I_h^{00} &=& T^{++} + T^{+-} + (I_h^{00})_8 + {\cal O} (v^6) \nonumber\\ I_h^{0i} &=&T^{+i} + T^{-i} + {\cal O} (v^5) \nonumber\\ I_h^{ij} &=& T^{ij} + (I_h^{ij})_8 + {\cal O} (v^6) \label{eq:result-h}\\ I_\phi &=& T^{++} - {1 \over 3} T^{+-} - {1 \over 3} T^{ii} + (I_\phi)_8 + {\cal O} (v^6) \nonumber\\ I_0^0 &=& T^{++} \nonumber\\ I_0^i &=& T^{+i} \nonumber \end{eqnarray} Here, we have assumed that the one-form field components $C_0$ and $C_i$ should couple to the D0-brane charge $N/R=T^{++}$ and spatial current ${\rm Tr}\;(\dot{X}^i)/R=T^{+i}$ so that the last two expressions are exact. The fourth order quantities $(I_h^{00})_8$, $(I_h^{ij})_8$, and $(I_\phi)_8$ are not completely determined by our analysis, but they must obey the relation \[ ({1 \over 2} I_h^{00} + {1 \over 2} I_h^{ii} + {3 \over 2} I_\phi)_8 = T^{--} \] From the $N = 1$ results (\ref{eq:n1}) we expect that \begin{eqnarray*} (I_h^{00})_8 & = & \frac{3}{2} T^{--}+ (I_h^{00})_{8c} \nonumber\\ (I_h^{ii})_8 & = & 2 T^{--}+ (I_h^{ii})_{8c} \nonumber\\ (I_\phi)_8 & = & -\frac{1}{2} T^{--}+ (I_\phi)_{8c}\nonumber \end{eqnarray*} where $(I)_{8c}$ are quantities of order $v^4$ which contain commutators or fermions and which vanish in the $N = 1$ case of a single spinless graviton considered in the previous subsection. Additional information about the currents should follow from the conservation of the D0 brane stress-energy tensor $I_h^{\mu \nu}(\vec{x})$ which is defined in terms of the moments $I_h^{\mu \nu (k_1 \cdots k_n)}$. As discussed in \cite{mvr}, the relation $D_\mu I_h^{\mu \nu} = 0$ implies \[ \partial_t I_h^{0 \mu (k_1 \cdots k_n)} = I_h^{k_1 \mu (k_2 \cdots k_n)} + \dots + I_h^{k_n \mu (k_1 \cdots k_{n-1})} \] In particular, $(I_h^{ij})_8$ should be precisely determined by \[ (I_h^{ij})_8 = (\partial_t T^{+i(j)} + \partial_t T^{-i(j)})_8. \] So far, we have only dealt with the case of a background metric $h$, dilaton field $\phi$ and R-R 1-form field $C$. The same sort of analysis can be applied for backgrounds having nonvanishing 2-form $B$ or 3-form $C^{(3)}$ fields and their duals, and in fact the analysis is simpler in these cases. In order to describe nontrivial background antisymmetric tensor fields, we must generalize the relations (\ref{eq:relations}) which connect the IIA background fields to the 11D background 3-form field through the Seiberg limit. The $B$ and $C^{(3)}$ fields are related to components of the 3-form field through \begin{eqnarray*} B_{0i} & = & \hat{A}_{10 \; 0 i} = A_{+ -i}\\ B_{ij} & = & \hat{A}_{10 \; ij} = \frac{\alpha}{\sqrt{2}} A_{+ ij} - \frac{1}{ \alpha \sqrt{2}} A_{-ij}\\ C^{(3)}_{0ij} & = & \hat{A}_{0 \; ij} = \frac{\alpha}{\sqrt{2}} A_{+ ij} + \frac{1}{ \alpha \sqrt{2}} A_{-ij}\\ C^{(3)}_{ijk} & = & \hat{A}_{ijk} = A_{ijk} \end{eqnarray*} The constraints on the string and D2-brane currents analogous to (\ref{eq:constraints}) are then \begin{eqnarray} (I_2^{ijk})_0 = (I_s^{0i})_0 = (3I_2^{0ij} -I_s^{ij})_2 & = & 0 \nonumber\\ (I_s^{0i})_4 & = & 3J^{+ -i}\nonumber \\ (I_2^{ijk})_4 & = & J^{ijk}\label{eq:constraints-a}\\ (3 I_2^{0ij} + I_s^{ij})_2 & = & 6 J^{+ ij} \nonumber\\ (3I_2^{0ij} -I_s^{ij})_6 & = & 3 J^{-ij} \nonumber \end{eqnarray} from which we can determine \begin{eqnarray} I_s^{0i} & = & 3J^{+ -i} +{\cal O} (v^4)\nonumber \\ I_2^{ijk} & = & J^{ijk}+{\cal O} (v^4)\label{eq:solution-a}\\ I_2^{0ij} & = & J^{+ ij} + (I_2^{0ij})_6 +{\cal O} (v^5) \nonumber\\ I_s^{ij} & = & 3 J^{+ ij} + (I_s^{ij})_6 +{\cal O} (v^5)\nonumber \end{eqnarray} where the terms $(I)_6$ on the last two lines must satisfy the final relation in (\ref{eq:constraints-a}). In addition, conservation relations for $I_s$ suggest that \[ \partial_t I_s^{0i(j)} = I_s^{ji} = - I_s^{ij} \] from which it follows that \[ (I_s^{ij})_6 = -3\partial_t J^{+-i(j)} = -3 J^{-ij}, \; \; \; \; \; (I_2^{0ij})_6 = 0 \] There are a number of comments worth making about the identifications (\ref{eq:solution-a}). First, note that the factor of 3 appearing in the currents $I_s$ arises from our somewhat unconventional choice of normalization for the couplings $A_p J^p$ between a $p$-form field and its associated current. Often, a factor of $1/p!$ is included in the definition of this coupling. With that redefinition of the currents, the factors of $3$ in our relations would disappear. We have chosen our convention for the coupling to conform with previous literature on the subject. Next, we briefly discuss the physical interpretation of the leading terms in (\ref{eq:solution-a}). The leading term $J^{+ ij}$ in $I_2^{0ij}$ is the total membrane charge of the D0-brane system. This result is the T-dual of the statement that $\int F$ on a $p$-brane is the total $(p-2)$-brane charge coupling to the $(p -1)$-form R-R field \cite{Douglas}. Although one might think that this should be the only contribution to the D2-brane charge of the system, additional contributions may arise from the geometry of the brane embedding \cite{bvs,ghm,Cheung-Yin,Minasian-Moore}. Note that while for finite $N$ the integrated membrane charge $J^{+ ij} = {\rm Tr}\;[X^i, X^j]$ vanishes identically (since a finite size system can have no net membrane charge) the higher moments of the D2-brane charge can be nonvanishing and will couple to the derivatives of the $C^{(3)}$ field. The leading term in $I_s^{0i}$ is the net string winding charge in direction $i$; this is simply the T-dual of the Poynting vector giving momentum on a dual D-brane. The leading term in the current $I_s^{ij}$ is perhaps somewhat surprising. Although this term itself vanishes for finite size matrices, as mentioned above, the first moment is nonvanishing. The existence of this term indicates that there will be a coupling in the multiple D0-brane action of the form \begin{equation} \partial_{[i} B_{jk]} {\rm Tr}\; (X^{[i} X^j X^{k]} + {\rm fermions}). \end{equation} We are rather confused as to the physical origin of this term. Indeed, this term plays a puzzling role in several related situations. For example, after compactification on $T^3$, the term $J^{+ ij}$ should be related by a duality transformation to the NS5-brane charge of the IIA theory \cite{grt}, which we discuss below. It may be possible to understand the role of this term in the theory by studying a T-dual system such as a dual multiple D3-brane configuration. We discuss the connection with the dual theory briefly in the last section of this paper. Now let us consider the currents coupling to the dual fields $\tilde{B}$ and $\tilde{C}^{(3)}$. These currents can be derived in a fashion precisely analogous to the above argument by considering the fields related to the dual 6-form $\tilde{A}$ of the 11D theory. We find \begin{eqnarray} I_4^{0i jkl} & = & 6 M^{+ -i jkl}+{\cal O} (v^4)\nonumber \\ I_5^{ijklmn} & = & M^{ijklmn}+{\cal O} (v^4)\label{eq:solution-ad}\\ I_5^{0ij klm} & = & M^{+ ij klm} + (I_5^{0ij klm})_6 +{\cal O} (v^5) \nonumber\\ I_4^{ij klm} & = & 6 M^{+ ij klm} + (I_4^{ij klm})_6 +{\cal O} (v^5)\nonumber \end{eqnarray} where \[ 6 (I_5^{0ij klm})_6 - (I_4^{ij klm})_6= 6M^{-ijklm}. \] Just as for $I_s$, conservation relations for $I_4$ suggest that \[ \partial_t I_4^{0ijkl(m)} = I_4^{ijklm} \] from which it follows that \[ (I_4^{ijklm})_6 = 6\partial_t M^{+-ijkl(m)} = -6 M^{-ijklm}, \; \; \; \; \; (I_5^{0ijklm})_6 = 0. \] The leading term in $I_4^{0ijkl}$ is the net D4-brane charge \cite{bss,grt}. This is the dual of the instanton number on a D$p$-brane. Unfortunately, we only know from matrix theory the components of the 5-brane current $M^{-IJKLM}$ with one $-$ index. Thus, we can only determine the leading term in the components $I_4^{0ijkl}$ of the D4-brane current, and we have no information about the leading terms in the remaining components of $I_4$ or any components of the NS5-brane current $I_5$. The absence of any known operator for the transverse 5-brane charge $M^{+ ijklm}$ in matrix theory is an infamous problem. No operator of this form appears in the supersymmetry algebra \cite{bss} or in the one-loop effective potential \cite{Dan-Wati-2}. Nonetheless, we should expect higher moments of this operator to appear, corresponding to local transverse 5-brane charge. It has been argued that in a T-dual 3-brane picture the desired operator is S-dual to the charge of a D5-brane perpendicular to the 3-brane \cite{grt} (described by the operator $J^{+ ij}$ mentioned above), although no explicit description of this dual operator has been given. It would be nice to have a better understanding of these terms in the multiple D0-brane action. Finally, we consider the currents coupling to the dual field $\tilde{C}$. We expect the IIA 6-brane current to couple to this field. Unlike the other branes whose currents we have considered, the IIA 6-brane does not arise in a simple fashion from the dimensional reduction of the membrane or the 5-brane of 11D M-theory. Rather, the IIA 6-brane represents a nontrivial metric of Kaluza-Klein monopole form in the 11D theory. Nonetheless, in \cite{Mark-Wati-3} we found a matrix theory description of interactions between such metrics and 0-branes. This appeared in the form of a term in the 2-body matrix theory potential which coupled the 0-brane stress tensor to a 10D 6-brane current $S^{\mu \nu \rho \lambda \sigma \tau \upsilon}$. Since this current already has an essentially 10D form, it is natural to map it directly to the 6-brane current we expect in the IIA theory. Thus, we believe that the 6-brane current of a system of many 0-branes which couples to the background $\tilde{C}$ field will be given by \begin{eqnarray} I_6^{0ijklmn} & = & S^{+ijklmn} +{\cal O} (v^5) \label{eq:solution-6}\\ I_6^{ijklmn p} & = & S^{ijklmn p}+{\cal O} (v^6)\nonumber \end{eqnarray} where the matrix theory form of the 6-brane current is given in the Appendix. \subsection{Summary of results for multiple D0-brane action in IIA} We summarize here our results for the terms in the multiple D0-brane action which couple linearly to the background fields of type IIA supergravity and their derivatives. The full action including all linear terms is given by (\ref{eq:IIA-general}) \begin{eqnarray} S &= & S_{{\rm flat}} + \int dt \sum_{n=0}^{\infty} {1 \over n!} \left[ \frac{1}{2} (\partial_{k_1}\cdots \partial_{k_n} h^{IIA}_{\mu \nu}) \; I_h^{\mu \nu (k_1 \cdots k_n)} + (\partial_{k_1}\cdots \partial_{k_n} \phi) \; I_{\phi}^{(k_1 \cdots k_n)}\right.\nonumber\\ & & \hspace{1in}+ (\partial_{k_1}\cdots \partial_{k_n} C_{\mu }) \; I_0^{\mu (k_1 \cdots k_n)} + (\partial_{k_1}\cdots \partial_{k_n} \tilde{C}_{\mu \nu \lambda \rho \sigma \tau \zeta }) \; I_6^{\mu \nu \lambda \rho \sigma \tau \zeta (k_1 \cdots k_n)} \nonumber\\ & & \hspace{1in}+ (\partial_{k_1}\cdots \partial_{k_n} B_{\mu \nu}) \; I_s^{\mu \nu (k_1 \cdots k_n)} + (\partial_{k_1}\cdots \partial_{k_n} \tilde{B}_{\mu \nu \lambda \rho \sigma \tau}) \; I_5^{\mu \nu \lambda \rho \sigma \tau (k_1 \cdots k_n)} \nonumber\\ & &\hspace{1in} \left.+ (\partial_{k_1}\cdots \partial_{k_n} C^{(3)}_{\mu \nu \lambda }) \; I_2^{\mu \nu \lambda (k_1 \cdots k_n)} + (\partial_{k_1}\cdots \partial_{k_n} \tilde{C}^{(3)}_{\mu \nu \lambda \rho \sigma }) \; I_4^{\mu \nu \lambda \rho \sigma (k_1 \cdots k_n)}\right] \nonumber \end{eqnarray} The multipole moments of the stress tensor $I_h$ and currents coupling to the background dilaton and R-R 1-form field have leading terms given by (\ref{eq:result-h}). The currents coupling to the NS-NS antisymmetric 2-form field and the R-R 3-form field have leading terms given by (\ref{eq:solution-a}). Of the currents coupling to the duals of these two fields, we have only been able to identify leading term in the the moments of the 4-brane current component $I_4^{0ijkl}$ as described in (\ref{eq:solution-ad}). We believe that the leading terms in the components of the 6-brane current coupling to the dual of the 1-form field are as given in (\ref{eq:solution-6}). We have derived these results based on our proposal in \cite{Mark-Wati-3} for the form of the matrix theory action in weak background fields and Seiberg's scaling argument. Our results agree with the known terms in the $N = 1$ Born-Infeld action, and with the known BPS charges of the multiple D0-brane system. In the following section we give a simple test of the results for the terms coupling to the background metric. Further possible tests, applications, and extensions of these results are discussed in the concluding section. \section{Tests of the action} In this section, we test our proposals for the general background actions through two related calculations. First, we consider the D0-brane action in a background describing two seperated branes in a curved space ($h_{ij} \ne 0$). We determine the masses of off diagonal bosonic and fermionic fields and show that these exactly match the geodesic distance to leading order in $h$ in agreement with the constraint suggested by Douglas. Next, we consider the analogous background in matrix theory and compute the leading order one-loop potential between two gravitons in a curved transverse space, showing that curved-space supergravity predictions are reproduced. \subsection{The geodesic length criterion} \label{sec:masses} One of the earliest discussions of the problem of formulating a low-energy theory for many D0-branes moving in a curved space was given in \cite{Douglas-curved}. In that paper, Douglas argued that one of the most basic conditions which such an action must satisfy is that in a background corresponding to a pair of D0-branes living at points $x$ and $y$ there should be light fields with masses equal to the geodesic length between these points. This condition, together with additional axiomatic assumptions, was used by Douglas, Kato and Ooguri in \cite{dko} to give the first few terms in the D0-brane action on a general Calabi-Yau 3-fold which preserves some supersymmetry. In this section we show that our formulation of the linearized coupling to a weak background in the multi-D0-brane action satisfies Douglas's geodesic length criterion. \subsubsection{Setup} We wish to consider a pair of D0-branes at separated points in a weakly curved space. We assume that the transverse metric is described by a small perturbation about a flat background, $g_{ij} = \delta_{ij} + h_{ij}$, while the other components of the metric and the other background fields are trivial. Without loss of generality, we choose coordinates so that one brane is at the origin, while the other has transverse coordinates $r^i$. The situation is described by the multi-D0-brane action (\ref{eq:IIA-general}) with non-zero $h_{ij}$ and fields expanded about background matrices as \begin{eqnarray} X^i &= &\left(\begin{array}{cc} r^i & 0\\ 0 & 0 \end{array} \right)+ \left(\begin{array}{cc} \zeta^i & z^i\\ \bar{z}^i & \tilde{\zeta}^i \end{array} \right) \label{eq:background} \\ \Theta & = &\left(\begin{array}{cc} 0 & 0\\ 0 & 0 \end{array} \right)+ \left(\begin{array}{cc} \eta^i & \chi^i\\ \bar{\chi}^i & \tilde{\eta}^i \end{array} \right), \nonumber \end{eqnarray} Here, the fields $\zeta$, $z$, $\eta$, and $\chi$ represent fluctuations about the background. The geodesic length condition formulated by Douglas states that the masses of the off-diagonal fields $z$ and $\chi$, which arise from strings stretched between the separated branes, should precisely match the geodesic length measured by the metric $h_{ij}$ between the points 0 and $r^i$. We will now compute both the geodesic length and the oscillator masses and show that they agree. \subsubsection{Geodesic distance} We begin with the geodesic length between points $0$ and $r^i$. This geodesic length is the minimum value of the length \begin{equation} \int_\gamma ds = \int_0^{r^i} \sqrt{g_{ij}dx^i dx^j} \label{eq:geodesic-length} \end{equation} taken over all paths $\gamma$ between the two points. Because the geodesic path is an extremum of this functional, the variation of the length under a small variation $\delta \gamma$ of the path is of order $(\delta \gamma)^2$. Since we are interested in changes in the length which are linear in the background metric, we can therefore neglect effects from the change of the geodesic path and simply evaluate the change in the geodesic length by integrating (\ref{eq:geodesic-length}) along the straight line which is the geodesic in the flat metric. Thus, we take \[ x^i(\lambda) = r^i \lambda \] and find \begin{eqnarray*} d(0,r^i) &=& \int_0^1 d\lambda \sqrt{g_{ij}(\vec{x}(\lambda))\dot{x}^i \dot{x}^j}\\ &=& \int_0^1 d\lambda \sqrt{r^2+h_{ij}(\vec{r}\lambda)r^i r^j}\\ &=& \int_0^1 d\lambda \{r + {1 \over 2r} h_{ij}(\vec{r}\lambda)r^i r^j +{\cal O} (h^2)\}\\ &=& r + {1 \over 2r} r^i r^j H_{ij} \end{eqnarray*} where \begin{eqnarray*} H_{ij} &=& \int_0^1 d \lambda \, h_{ij}(\lambda \vec{r})\\ &=& \sum_{n=0}^\infty{1 \over (n+1)!} (r \cdot \partial)^n h_{ij}(0) \end{eqnarray*} This gives us the geodesic length between the two points to linear order in the background metric. In the following section, it will be most convenient to compare squared oscillator masses with the squared geodesic length, given by \begin{equation} \label{eq:length} d^2 = r^2 + r^i r^j H_{ij} +{\cal O} (h^2) \end{equation} \subsubsection{Oscillator masses} \label{sec:oscillator-masses} We now calculate the masses of the off-diagonal fields. From (\ref{eq:IIA-general}), we find that the $N=2$ D0-brane action in the case of a transverse background metric has leading terms \begin{eqnarray} S &=& {1 \over 2R} \int dt \; {\rm Tr}( D_t X^i D_t X^i + {1 \over 2} [X^i, X^j] [X^i, X^j] + i \Theta D_t \Theta - \Theta[\slash{X}, \Theta]) \nonumber\\ & & + {1 \over 2} \int dt \; \sum_{n=0}^\infty {1 \over n!} \partial_{k_1} \cdots \partial_{k_n} h_{ij} T^{ij(k_1 \cdots k_n)} \label{eq:Trans-action} \end{eqnarray} Here, \begin{eqnarray} T^{ij(k_1 \cdots k_n)} &=& {1 \over R} {\rm STr}\; \left( \left\{ D_t X^i D_t X^j - [X^i, X^k] [X^k, X^j] \right.\right. \label{eq:tij}\\ & &\left.\left.\hspace{1in} - {1 \over 4} \Theta \gamma^i [X^j, \Theta] - {1 \over 4} \Theta \gamma^j [X^i, \Theta] \right\} X^{k_1} \cdots X^{k_n}\right) \nonumber\\ & & \hspace{0.3in} + \tilde{T}^{ij(k_1 \cdots k_n)}\nonumber \end{eqnarray} We note here that this is precisely the action for Matrix theory in a background $h_{ij}$ to leading order in the metric. In the next section we will use exactly this action to calculate the one-loop potential between two gravitons, taking the same background (\ref{eq:background}), though allowing $\vec{r}$ to be a function of time. Such a calculation is simplest using a gauge fixed version of the action in which we choose the background field gauge, adding a term \[ S_{fix} = {1 \over R} \int (-D_t X^0 + i[B^i, X^i])^2 \] plus the appropriate ghost terms. For later convenience, we will analyze this gauge fixed version, keeping in mind both the Matrix theory interpretation and D0-brane interpretations. Unlike the Matrix theory action, the complete D0-brane action contains additional terms both in the background intependent part and coupled to $h_{ij}$. However these cannot contribute to the quadratic action since they contain more than two matrices (eg $\dot{X}^i$, $F_{ij}$, $\Theta$) in which there are no entries depending only on background fields. Similarly, the terms $\tilde{T}^{ij(k_1 \cdots k_n)}$, whose form has not been worked out for $n>1$ involve at least two fermions $\Theta$ and one power of $\dot{X}$ or $[X^i, X^j]$ and so contribute only cubic and higher order terms to the action, irrelevant for determining the oscillator masses or computing the one loop potential. We now replace $X$ and $\Theta$ in the action with the matrices given in (\ref{eq:background}), and write down the terms in the action quadratic in the off diagonal fields $z$ and $\chi$. It turns out that the symmetrization prescription for ordering the matrices in $T^{ij(k_1 \cdots k_n)}$ is very important here, since most of the orderings give no contribution to the quadratic terms we are interested in. For example, the first term in (\ref{eq:tij}) contains a term \begin{eqnarray*} \lefteqn{{1 \over 2 n!} (\partial_{k_1} \cdots \partial_{k_n} h_{ij}) \; \; {\rm STr}\;(\dot{X^i} \dot{X^j} \; X^{k_1} \cdots X^{k_n})} \\ &=& {1 \over 2 n!} (\partial_{k_1} \cdots \partial_{k_n} h_{ij})\; \; {1 \over n+1} \sum_{m=0}^n {\rm Tr}\;(\dot{X}^i X^{k_1} \cdots X^{k_m} \dot{X}^j X^{k_{m+1}} \cdots X^{k_n}) \end{eqnarray*} for which only the $m=0$ and $m=n$ terms contribute to the quadratic action in the off diagonal field $z$. Summing over $n$, this contribution gives \begin{eqnarray*} & &\sum_{n=0}^\infty {1 \over (n+1)!} (\partial_{k_1} \cdots \partial_{k_n} h_{ij}) (\dot{\bar{z}}^i \dot{z}^j ) r^{k_1} \cdots r^{k_n}\\ &=&\dot{\bar{z}}^i \dot{z}^j \sum_{n=0}^\infty {1 \over (n+1)!} (r \cdot \partial)^n h_{ij} \\ &\equiv&H_{ij} \dot{\bar{z}}^i \dot{z}^j \end{eqnarray*} Note that the quantity $H_{ij}$ is simply the function $h_{ij}$ integrated over the straight line trajectory between $0$ and $\vec{r}$, \[ H_{ij} = \int_0^1 h_{ij}(\lambda \vec{r}) d \lambda \] which also appeared at first order in the geodesic distance formula (\ref{eq:length}). Using this definition, it is straightforward to write down the remaining terms in the quadratic actions for each of the off diagonal fields. \vspace{0.15in} \noindent {\bf Bosonic Terms} \vspace{0.15in} In exactly the same way as for the terms just derived, we find that the complete set of terms to leading order in $h$ for the nine transverse bosonic fields is \[ S_B = -\bar{z}^i\{(\partial_t^2 + r^2) (\delta_{ij} + H_{ij}) + r^k r^l H_{kl} \delta_{ij} - r^i r^k H_{kj} - H_{ik} r^k r^j \}z^j \] Note that the matrix $(\delta_{ij} + H_{ij})$ multiplies all terms not containing $h$. Since the remaining terms are already of order $h$, if we redefine $z^i$ to eliminate this factor in the first terms, the remaining terms will only be changed at second order in $h$. After such a field redefinition, the kinetic term is proportional to the identity, and the squared oscillator masses are therefore given by the eigenvalues of the constant matrix \begin{eqnarray*} {\bf M} &=& r^2 \left\{ (1+H_{rr}){\rlap{1} \hskip 1.6pt \hbox{1}}_{9 \times 9} + \left( \begin{array}{cccc} 0 & \cdots & 0 & H_{1r}\\ \vdots & & & \vdots \\ 0 & & 0 & H_{8r} \\ H_{1r} & \cdots & H_{8r} & -2H_{rr} \end{array} \right) + {\cal O} (h^2) \right\} \end{eqnarray*} Here, to simplify the formulae, we have made a rotation so that $\vec{r}$ lies in the $9$ direction, which we refer to using the index $r$. It is straightforward to solve directly for the eigenvalues and eigenvectors of this mass matrix, and one finds to this order that the masses are \[ m_1^2 = \cdots = m_7^2 = r^2(1 + H_{rr}), \; \; m_8^2 = r^2(1 + \sqrt{H_{rr}^2 + H_{ri} H_{ir}}), \; \; m_9^2 = r^2(1 - \sqrt{H_{rr}^2 + H_{ri} H_{ir}}) \] where the index $i$ is summed from 1 to 8. The oscillators with masses $m_9$ and $m_8$ correspond to directions lying in the plane defined by the $r$ direction and the perpendicular vector $H_{ir}$. The remaining oscillators correspond to the directions perpendicular to these and have masses which precisely match the geodesic distance (\ref{eq:length}) to leading order in $h$. The agreement between the masses of these perpendicular oscillators and the geodesic distance is precisely the criterion used by Douglas et al. in \cite{Douglas-curved,dko} to constrain the leading terms in multiple D0-brane action on certain classes of manifolds. The fact that the oscillators corresponding to fields not perpendicular to the separation have different masses is also expected. In the non-gauge fixed theory, the off-diagonal fields in the direction of the separation between the branes simply give a combination of a gauge rotation of the system and a relative motion of the D0-branes along a flat direction. This effect explains the failure of the masses $m_8, m_9$ to satisfy Douglas's criterion. \vspace{0.15in} \noindent {\bf Gauge Field} \vspace{0.15in} For a time independent $\vec{r}$, there is no mixing between gauge field and the other bosonic oscillators in the quadratic action, and we find that the quadratic terms involving the off diagonal field $z_0$ are simply \[ S_A =-\bar{z}_0 \{ \partial_t^2 + r^2 + r^2H_{rr} \} z_0 \] Hence, the off diagonal gauge field also has mass equal to the geodesic distance to leading order in $h$, \[ m_0^2 = r^2 + H_{rr} \,. \] \vspace{0.15in} \noindent {\bf Ghost Fields} \vspace{0.15in} Since our gauge fixing term does not depend on the background metric, the off diagonal ghost fields will have a mass given by $m_g^2 = r^2$, as in the flat space case, with action \[ S_G = -\bar{c} \{ \partial_t^2 + r^2 \} c \] \vspace{0.15in} \noindent {\bf Fermionic Fields} \vspace{0.15in} Proceeding in the same way for the quadratic fermion action, we find that the action quadratic in the off diagonal field $\chi$ is \[ S_F =-\bar{\chi}_\alpha \{i\partial_t + \gamma^i_{\alpha \beta} (r^i + {1 \over 2} H_{ij} r^j) \} \chi_\beta \] Thus, in the presence of a background metric, the quadratic fermion action is only changed by a shift \[ r^i \rightarrow r^i + {1 \over 2} H_{ij} r^j \] so the sixteen fermion fields $\chi$ have a mass squared matrix given by \[ M_f^2 = {\rlap{1} \hskip 1.6pt \hbox{1}} (r^2 + H_{ij} r^ir^j + {\cal O} (h^2)) = {\rlap{1} \hskip 1.6pt \hbox{1}} r^2(1 + H_{rr} + {\cal O} (h^2)) \] We see that all the fermionic oscillators have a mass which reproduces the geodesic distance to leading order in $h$. \vspace{0.15in} \noindent {\bf Summary} \vspace{0.15in} To summarize, we have found eight complex bosons (including the gauge field) and sixteen real fermions with masses equal to the geodesic distance between $0$ and $\vec{r}$. Additionally, there are complex bosons with $m^2 = r^2(1 + \sqrt{H_{rr}^2 + H_{ri} H_{ir}})$ and $m^2 = r^2(1 - \sqrt{H_{rr}^2 + H_{ri} H_{ir}})$ and two complex ghosts with $m^2 = r^2$. Thus, the sum of the squared masses weighted by the number of degrees of freedom is identical for the fermions and the bosons (including the ghosts with negative weight), and for both sets of fields, the average mass squared per degree of freedom is exactly the geodesic distance (\ref{eq:length}). We have now seen that the geodesic distance criterion is precisely realized in the proposed multi-D0-brane action. \subsection{Graviton interactions} In this section, we use the general background Matrix theory action to study the interactions between two gravitons with unit momentum around a lightlike circle in a weakly curved space. In various cases, we will compute the one-loop effective action to first order in the metric and compare with the interactions expected from DLCQ supergravity with a curved background. The relevant matrix theory action and background are exactly the same as those considered for D0-branes in the previous section. In this case, however, we do not wish to restrict to gravitons which are fixed in the transverse space, so we allow $\vec{r}$ to be a function of time. As for the case of flat space Matrix theory, we should require that our background matrices satisfy the equations of motion. For block diagonal backgrounds, the equations of motion decouple for each block, so for our case, we require that $\vec{r}(t)$ satisfy the equations of motion derived from the $U(1)$ action (\ref{graviton}). For a metric which is non-trivial only in the transverse directions, the equations of motion are \begin{equation} \label{eq:eom} \ddot{r}^i = \dot{r}^k \dot{r}^l g^{im} (\vec{r}) \{ {1 \over 2} \partial_m g_{kl} (\vec{r}) - \partial_k g_{ml} (\vec{r}) \} \end{equation} This is just the equation for a free non-relativistic particle moving in a curved space. We will consider two simple cases of trajectories which trivially satisfy these equations of motion. First, we may consider the static case $\vec{r}(t) = \vec{r}(0)$. The second case is one for which the metric has a flat direction $i$ (so that $h_{ij} = \partial_ig_{jk} = 0$) and we take the particle to have some velocity in this direction. In this case, the right hand side of (\ref{eq:eom}) vanishes, so that $r^i(t) = r^i + v^it$. \subsubsection{Supergravity predictions} Before proceeding with the matrix theory calculation, we would like to see what supergravity predicts for the interaction potential between two gravitons in a weakly curved space. As above, we start with a static metric $g_{ij}(\vec{x}) = \delta_{ij} + h_{ij}(\vec{x})$ which is assumed to satisfy the source-free Einstein equations \begin{equation} \label{eq:Ricci} 0 = R_{ij} ={1 \over 2}(\Delta h_{ij} - \partial_i \partial_k h_{kj} - \partial_j \partial_k h_{ki} + \partial_i \partial_j h_{kk}) + {\cal O}(h^2) \end{equation} We may find the potential between a pair of gravitons in this space by treating one as a source for a perturbation about the metric $g$ and reading off the potential from the probe action (\ref{graviton}), keeping only terms arising from the perturbation in the original metric due to the presence of the source graviton. For our source, we choose the graviton which sits at the origin of the transverse space with unit momentum in the compact direction. The stress-energy tensor for this particle still has only a single non-vanishing component, \[ T^{++} = T_{--} = {1 \over 2 \pi R^2} \delta(\vec{x}), \] The presence of this graviton will result in a perturbation of the metric $g_{ij}$ which we denote by $\gamma_{ij}$. The fact that $T_{--}$ is the only non-vanishing component of the stress-energy tensor simplifies things considerably, and as with the flat space case, we may solve the Einstein equation taking only the component $\gamma_{--}$ to be non-zero. In this case, the condition that the perturbed metric $g + \gamma$ should continue to satisfy the Einstein equation with source $T$ reduces at leading order in $\gamma$ to the covariant Laplace equation, \begin{equation} g^{ij}\nabla_i \nabla_j \gamma_{--} = g^{ij} (\partial_i \partial_j - \Gamma^k_{ij} \partial_k ) \gamma_{--} = 2\kappa_{11}^2 T_{--} \label{eq:laplace} \end{equation} (see, for example \cite{Weinberg}). We are only interested in the solution at leading order in $h$ (the original background metric), so we expand \[ \gamma_{--} \equiv \gamma_0 + \gamma_1 + {\cal O}(h^2) \] Here $\gamma_0$ is the part independent of $h$, equal to the flat space solution (ignoring non-numerical constants) \begin{equation} \label{eq:flat} \gamma_0 = {15 \over 2 r^7} \end{equation} while $\gamma_1$ is the part linear in $h$. In the case where $g_{ij}$ is the metric corresponding to some choice of coordinates on flat space, the exact solution to (\ref{eq:laplace}) must be given by a covariant version of (\ref{eq:flat}), replacing $r$ with the geodesic distance $d$ between $0$ and $r$. In this case, using (\ref{eq:length}), we have \begin{equation} \label{eq:hlinear} \gamma = {15 \over 2} \{ {1 \over r^7} - {7 r^i r^j \over 2r^9} \int_0^1 d \lambda h_{ij}(\lambda \vec{r})\} + {\cal O}(h^2) \end{equation} In fact, this solves (\ref{eq:laplace}) to leading order in $h$ in any case where the metric $g$ is Ricci-flat as may be verified explicitly by substitution. The $h$ independent part of (\ref{eq:laplace}) reads \[ \partial^2 \gamma_0 = 2\kappa_{11}^2 T_{--}, \] and is just the statement that $\gamma_0$ is the solution for flat space. The linear terms in $h$ in equation (\ref{eq:laplace}) read \[ \partial^2\gamma_1 = h_{ij}\partial_i \partial_j \gamma_0 + ( \partial_i h_{ik} - {1 \over 2} \partial_k h_{ii}) \partial_k \gamma_0 \] and substituting for $\gamma_0$ and $\gamma_1$ from (\ref{eq:hlinear}) it is not hard to check that this holds, making use of the Ricci-flatness condition (\ref{eq:Ricci}), the identity \[ \partial_\lambda h(\lambda \vec{r}) = (r \cdot \partial) h(\lambda \vec{r}), \] and various integrations by parts. This is done most easily by choosing coordinates so that $h$ satisfies the harmonic gauge condition in which \[ \partial_i h_{ij} = { 1 \over 2} \partial_j h_{ii}, \;\;\;\;R_{ij} = \partial^2 h_{ij} + {\cal O}(h^2) \]. Thus, to leading order in the backgrounds we are considering, the metric perturbation due to the presence of a graviton at the origin is simply \[ \gamma_{--}(\vec{x}) = {15 \over 2 d^7(\vec{x})} \] where $d$ is the geodesic distance between 0 and $\vec{x}$. Recalling the action (\ref{graviton}) for the probe graviton moving in the metric produced by this source, we see that to leading order in the background, the curved space graviton-graviton potential is simply \[ V = -{15 \over 16} {v^4 \over d^7} \] In the next sections, we will carry out the graviton potential calculation in Matrix theory to compare with this supergravity prediction. \subsubsection{Static case} First, we consider the static case in which both gravitons have zero transverse velocity. In this case, as in flat space, we expect the potential to vanish at leading order in the transverse metric, as shown above. For this case, we may directly apply the results of section \ref{sec:masses}. The complete action quadratic in the off diagonal fields is \begin{eqnarray} S&=&S_B+S_A+S_F+S_G \nonumber \\ S_B &=& -\bar{z}^i\{(\partial_t^2 + r^2) (\delta_{ij} + H_{ij}) \nonumber \\ & &\hspace{0.5in} + (r^k r^l H_{kl})\delta_{ij} - r^i r^k H_{kj} - H_{ik} r^k r^j \}z^j \nonumber \\ S_A &=& -\bar{z}_0 \{ \partial_t^2 + r^2 + r^i r^j H_{ij} \} z_0 \label{eq:stat-action}\\ S_F &=& -\bar{\chi}_\alpha \{i\partial_t + \gamma^i_{\alpha \beta} (r^i + {1 \over 2} H_{ij} r^j \} \chi_\beta \nonumber \\ S_G &=& -\bar{c} \{ \partial_t^2 + r^2 \} c \nonumber \end{eqnarray} As argued above, this leads to eight complex bosons (including the gauge field) and sixteen real fermions with masses equal to the geodesic distance between $0$ and $\vec{r}$ as well as complex bosons with $m^2 = r^2(1 + \sqrt{H_{rr}^2 + H_{ri} H_{ir}})$ and $m^2 = r^2(1 - \sqrt{H_{rr}^2 + H_{ri} H_{ir}})$ and two complex ghosts with $m^2 = r^2$. The vanishing of the one loop potential to leading order in $h$ is ensured by the fact that the sum of the squared masses weighted by number of degrees of freedom is identical for the fermions and bosons (including ghosts weighted by -1). This follows since the one loop effective action depends only on the oscillator masses and is given by: \begin{eqnarray*} e^{i\Gamma_{1 loop}}&=&\det{}^{-8}(\partial_t^2 + r^2(1 + H_{rr})) \det{}^{8}(\partial_t^2 + r^2(1 + H_{rr})) \det{}^{2}(\partial_t^2 + r^2)\\ & &\det{}^{-1}(\partial_t^2 + r^2(1 + \sqrt{H_{rr}^2 + H_{ri} H_{ir}}))\det{}^{-1}(\partial_t^2 + r^2(1-\sqrt{H_{rr}^2 + H_{ri} H_{ir}}))\\ &=&1 + {\cal O} (h^2) \end{eqnarray*} Hence, in the static case, we find agreement with our expectations from supergravity. \subsubsection{Velocity dependent potential} We now consider the case of two gravitons with relative velocity. Here, we assume that the particle with initial position $\vec{r}$ moves in a direction $\vec{v}$ which is perpendicular to $\vec{r}$ and in which the metric is flat. In this case, we have shown above that supergravity predicts a potential \[ -{15v^4 \over 16 d^7} \] where $d$ is the geodesic separation distance. To simplify our calculations, we rotate coordinates so that $\vec{r}$ and $\vec{v}$ lie in coordinate directions which we denote by indices $r$ and $v$. Thus \[ g_{vv}(\vec{x}) = 1, \; g_{vr}(\vec{x})= g_{vi}(\vec{x}) = \partial_vg_{ij}(\vec{x}) = 0 \] These equations ensure that the matrix theory equations of motion (\ref{eq:eom}) are satisfied for the trajectory $\vec{r}(t) = \vec{r} + \vec{v}t$ that we are considering. Expanding the action (\ref{eq:stat-action}) about this background, we find that the quadratic action for the off diagonal fields is equal to the action for the static case (where $r^i$ is interpreted as $r^i(t)=r^i + v^it$) plus extra terms: \[ S_v = 2i\bar{z}^iv^iz^0 - 2i\bar{z}^0v^iz^i \] Note that these terms (in which the background appears as $\dot{\vec{r}}$ rather than just $\vec{r}$) come only from the flat space action, since in the metric dependent terms, $\dot{X}^i$ only appears coupled to $h_{ij}$ while $\dot{r}^i h_{ij} = 0$. As a result, the remaining calculation is almost identical to the flat space calculation, performed in \cite{DKPS}. To see this, we note that the complete action in this case may be written (eliminating a factor $(\delta_{ij} + H_{ij})$ as above to diagonalize the boson kinetic term) \begin{eqnarray*} S&=&S_B+S_A+S_v+S_F+S_G\\ S_B &=& -\bar{z}^i\{(\partial_t^2 + d^2 + v^2t^2) \} z^i\\ & &\hspace{0.5in} + \bar{z}^r\{r^2H_{ri}\}z^i + \bar{z}^i\{r^2H_{ir}\}z^r + \bar{z}^v\{ H_{ri}rvt\}z^i + \bar{z}^i\{H_{ir}rvt\}z^v \\ S_A &=& -\bar{z}_0 \{ \partial_t^2 + d^2 + v^2t^2 \} z_0\\ S_v &=& 2i\bar{z}^iv^iz^0 - 2i\bar{z}^0v^iz^i\\ S_F &=& -\bar{\chi}_\alpha \{i\partial_t + \gamma^i_{\alpha \beta} (d^i + v^it ) \} \chi_\beta \\ S_G &=& -\bar{c} \{ \partial_t^2 + d^2 \} c\\ & &\hspace{0.5in} + \bar{c} \{H_{rr}\} c \end{eqnarray*} where we have defined \[ d^i = r^i + {1 \over 2} H_{ij} r^j \] so that $d^2$ is the squared geodesic distance. Apart from the terms with explicit factors of $H$ in the boson and ghost actions, this is exactly the flat space action with $r^i$ replaced by $d^i$. Recalling the flat space calculation, we see that at zeroth order in $H$, the oscillators $z^i$, $i=1,\cdots, 7$, and $z^r$ are degenerate with mass $d^2+v^2t^2$ while $z^v$ and $z^0$ combine into oscillators with non-degenerate masses $(d^2 + v^2t^2 \pm 2v)$. Adding the perturbation \begin{equation} \label{pert} \bar{z}^r\{r^2H_{ri}\}z^i + \bar{z}^i\{r^2H_{ir}\}z^r + \bar{z}^v\{ H_{ri}rvt\}z^i + \bar{z}^i\{H_{ir}rvt\}z^v \end{equation} we note that the last two terms do not affect the masses at leading order in $h$, since if we change coordinates to diagonalize the leading order mass matrix, these contribute only to non-diagonal matrix elements connecting eigenvectors of different mass. (Recall from basic perturbation theory that for eigenvectors $|A_1\rangle, \cdots |A_n\rangle$ with degenerate zeroth order eigenvalues and eigenvectors $|B_1\rangle, \cdots |B_m\rangle$ with non-degenerate zeroth order eigenvalues that the first order shift in the eigenvalues for $|B_i\rangle$ come only from the matrix element $\langle B_i|M|B_i\rangle$, while the first order eigenvalues for the space spanned by $|A_i\rangle$ are determined only by the submatrix $\langle A_i|M|A_j\rangle$). The first two terms in (\ref{pert}) have the same effect as they did in the static case, to shift two of the degenerate boson masses by \begin{eqnarray*} \Delta m^2 &=& - H_{rr} \pm \sqrt{H_{rr}^2 + H_{ri}H_{ir}}\\ &\equiv&-H_{rr} \pm G \end{eqnarray*} Meanwhile, the perturbation in the ghost action shifts the two ghost masses by \[ \Delta m_g^2 = - H_{rr} \] Defining \[ F(x) = det(\partial_t^2 + (d^2 - x) +v^2t^2) \] we find that since all of the mass shifts are time independent, the one loop effective action is simply \begin{eqnarray*} e^{i\Gamma} &=& e^{i \Gamma_d} {F(H_{rr})F(H_{rr}) \over F(H_{rr} + G)F(H_{rr} - G)}\\ &=& e^{i \Gamma_d}(1 + {\cal O} (h^2)) \end{eqnarray*} where $\Gamma_d$ is the flat-space potential with $r$ replaced by the geodesic length $d$. We conclude that the leading order one loop potential is simply given by \[ V=-{15 \over 16} {v^4 \over d^7} \] as predicted by supergravity. \section{Discussion} In this paper we have derived the leading terms in the multiple D0-brane action in a weakly curved background metric, dilaton, and antisymmetric tensor background fields. This action was derived by using Seiberg's scaling arguments on an action we recently proposed for the M(atrix) model of M-theory in weak background fields. We found explicit forms for the IIA stress-energy tensor of a multiple D0-brane system, as well as the components of D2-brane, D4-brane, D6-brane and fundamental string currents which couple to the background R-R and B fields of the IIA theory. We tested our action by verifying that it satisfies Douglas's geodesic length criterion. We also showed that the corrections to the one-loop effective potential between a pair of individual 0-branes correctly reproduce curved space supergravity results. The results presented in this paper give for the first time a systematic description of the linear coupling between a system of multiple D-branes and background supergravity fields. There are a number of previous discussions of this sort of action in the literature to which our results can be related. In \cite{dko}, Douglas, Kato and Ooguri used Douglas's geodesic length criterion and other axioms including an assumption of supersymmetry to constrain the form of the multiple D3-brane action on a transverse K\"ahler manifold. They compute the first few correction terms in terms of the curvature tensor $R_{i \bar{j}k \bar{l}}$ of the background. They show that the first term, corresponding to our coupling \begin{equation} (\partial_{k_1 k_2} h_{ij}) I_h^{ij(k_1 k_2)} \end{equation} is uniquely determined by the geodesic length criterion. The linear part of the term they find indeed has the same form $\partial^2 h {\rm Tr}\; X^2 (F^2 + \dot{X}^2)$ as our result for this term. Their approach is not able to uniquely determine the higher order terms, but our results are compatible with the general structure of the linear parts of the structure they find at higher order. It is interesting that they are able to determine the form of some of the quadratic couplings in their work as well as the linear terms. It is natural to try to extend the results of this paper to determine the quadratic and higher order couplings between a system of many D-branes and the supergravity background fields. In \cite{Mark-Wati-3} it was suggested that the coupling to $n$th order terms in the background could be determined by a $n$-loop calculation in matrix theory. The results of Okawa and Yoneya on 3-graviton scattering in matrix theory \cite{Okawa-Yoneya,Okawa-Yoneya-2} seem to indicate that this may work at least to quadratic order in general backgrounds, although there are indications \cite{deg2,Sethi-Stern-2,Lowe-constraints} that there may be problems with extending this approach to higher order. In any case, even the two-loop calculation would be quite challenging to work out for completely general background configurations, so it would be nice to find a simpler approach. The terms coupling the open string fields of the D-brane system to any number of bulk supergravity fields can of course in principle be determined by a perturbative string calculation. These calculations are quite complicated, however, even for the linear coupling terms discussed in this paper. (For recent work computing the curvature squared terms in the single D-brane action see \cite{bbg}). Furthermore, the results we have given here extend to arbitrary derivatives in the background fields and contain a correspondingly arbitrary number of D-brane fields. It is difficult to imagine reproducing such results from perturbative string theory. Despite these difficulties inherent in a systematic derivation of the higher order coupling terms, it may be that the symmetries of the theory and the geodesic length criterion are sufficient to determine the structure of some of the higher-order terms. The results of \cite{dko} on K\"ahler manifolds indicate that it is indeed possible to learn something about about the higher order terms using this approach. In this paper we have found that the combinatorial structure of the higher moment terms is crucial in fixing the masses of off-diagonal strings in accord with the geodesic length criterion. This indicates that this condition will place strong constraints on the possible form of the higher order couplings to the background. Another approach which might help extend the results here to higher order is to find the symmetry principle which corresponds to general coordinate invariance for the multiple D0-brane system. Because the coordinates enter the theory as matrices, there are ordering ambiguities in determining how the operators describing the D0-brane system transform. If this symmetry could be understood in a systematic fashion, it would quite possibly uniquely determine the higher order couplings of the multiple D-brane system to general backgrounds. The approach we have taken here to describing matrix theory and multiple D0-brane systems in curved backgrounds is to assume that the action in a weak background can be written as a matrix quantum mechanics theory with a systematic expansion in powers of the background field strength. This approach certainly seems valid for linear couplings to the background. It is not clear, however, that such an approach can be extended to all orders. In \cite{dos,Douglas-Ooguri}, in fact, it was argued that even on simple manifolds like K3 or ALE spaces it may not be possible to describe DLCQ M-theory using a finite size matrix quantum mechanics theory. We are not sure at this point how the approach we have taken here fits with the results of those papers. One possibility is that the perturbation expansion we are considering in powers of the background field will not converge to a well-defined theory when higher order terms are taken into account and the background is of the form considered in \cite{dos,Douglas-Ooguri}. Indeed, the results of those papers indicate that the expansion in weak backgrounds may break down at quadratic order in the curvature of the background\footnote{Thanks to H.\ Ooguri for correspondence on this point.}. Another possibility is that either the restriction to light-front coordinates in M-theory or the fact that in the Seiberg limit the scale of the metric structures of interest becomes smaller than the string length may have a subtle effect on the relationship between the supergravity and open string descriptions of graviton interactions, leading to some modification in the conclusions of \cite{dos,Douglas-Ooguri}. It is clearly a very important question whether a good low-energy description of D0-branes can be given which maps to M-theory in the Seiberg limit, but we leave a further resolution of this issue to further work. In this paper we have focused on the action for a system of D0-branes in a weak supergravity background. It is possible, however, to T-dualize the action we have given here to get an action for D$p$-branes of arbitrary dimension in a weak background. One particularly interesting case is that of D3-branes. The T-duals of the currents $I_x$ we have determined in this paper are linear combinations of the operators in the ${\cal N} = 4,$ $D = 3 + 1$ super Yang-Mills theory on the world volume of a system of many D3-branes which lie in the short representations of the $SU (2, 2 | 4)$ superconformal symmetry group of the theory. These operators, which couple linearly to the supergravity background fields, play a crucial role in the simplest version of Maldacena's AdS/CFT correspondence \cite{Maldacena-conjecture}. For the fields associated with the lowest partial waves of bulk fields in the AdS space, some of the corresponding operators were found from the Born-Infeld action in \cite{Klebanov-absorption,gkt,flz,Das-Trivedi}. All the other lowest partial wave operators are in principle determined by supersymmetry and group theory from the weight 2 operator $ {\rm Tr}\; X^{(i} X^{j)}$, following \cite{krn,Gunaydin-Marcus,gkp-2,Witten-AdS1} and related work. Although these operators have played a fundamental role in understanding the detailed structure of the AdS/CFT correspondence, only a few of these operators have been described explicitly in terms of the component fields in the D3-brane world-volume theory. In addition, to date no general method for understanding the structure of the higher partial wave operators (which correspond to the higher moments of our currents $I_x^{\cdots (\cdots)}$ through T-duality and which are related through supersymmetry to the higher weight chiral primary operators ${\rm Tr}\; X^{\{i_1} \cdots X^{i_k\}}$) has been presented in the literature, although some discussion of particular operators of this type was given in, for example, \cite{gkt,dfs2}. In a separate paper we will discuss more details of the connection between the results presented here and the operators which are used in the AdS/CFT correspondence for 3-branes. A feature which emerges from the results of \cite{Mark-Wati,Dan-Wati-2,Mark-Wati-3} and the extension of this work in the present paper is the characteristic form of the higher moments of the currents $I_x$ which couple to the derivatives of the background supergravity fields. In general, we find that the $n$th moment of the current $I_x$ has contributions of the form \begin{equation} I_x^{(i_1 \cdots i_n)} = {\rm Sym} \; (I_x; X^{i_1}, X^{i_2}, \ldots X^{i_n}) + I_{x ({\rm fermion})}^{(i_1 \cdots i_n)} \end{equation} where the first term on the RHS is a bosonic contribution to the higher moment given by a symmetrized trace of the polynomial giving the monopole moment of $I_x$ with a product of $n$ $X$'s, and the second term contains fermionic contributions to the higher moment. The fact that all the monopole moments of the currents as well as their higher moments can be expressed in a symmetrized trace form is reminiscent of Tseytlin's suggestion \cite{Tseytlin} for using the symmetrized trace to resolve ordering ambiguities in the nonabelian Born-Infeld action. Indeed, the components $T^{--}$ of the 11D stress tensor are precisely the symmetrized $F^4$ terms appearing at fourth order in the nonabelian Born-Infeld action proposed by Tseytlin. This structure may be helpful in trying to predict the form of higher-order terms in the action without doing explicit matrix theory or string theory calculations. This structure is also helpful in understanding previous work in which higher partial waves of operators on the D3-brane play a role. In particular, in \cite{gkt} the rate of absorption of higher partial waves of minimally coupled scalars in an extremal D3-brane background was computed in supergravity and compared to a D3-brane world-volume calculation. While the analogous calculations for $s$-wave absorption can be matched precisely including numerical coefficients, for partial waves with $l > 1$ a discrepancy was found in \cite{gkt} between the results of these two calculations. The authors suggested that this discrepancy might arise because the higher partial wave operators were not correctly normalized. The results we have given here suggest by T-duality that they indeed used the correct normalization, but that the operators they used should have a symmetrized trace with respect to orderings of the fields. This additional information seems to help resolve the discrepancy found in \cite{gkt}\footnote{Thanks to Steve Gubser and Igor Klebanov for discussions on this point}. \junk{For example, the operators of the form $ O ={\rm Tr}\; F F X X$ appearing as part of the second partial wave operator ${\rm Tr}\; F^2 (X^i X^j -1/6 \delta^{ij} X^2)$ should be replaced by the symmetrized operator \begin{equation} O ={\rm STr}\; FF X X = \frac{2}{3} {\rm Tr}\; FF X X + \frac{1}{3} {\rm Tr}\; FXF X . \end{equation} Since the outgoing particles from the vertices associated with the two terms on the RHS of this expression can be distinguished at large $N$, the single term $O$ carrying a factor of 1 should be replaced by two vertices carrying factors of $(2/3)^2 = 4/9$ and $(1/3)^2 = 1/9$ respectively. Together, these two vertices seem to cancel the factor of 9/5 discrepancy found in \cite{gkt} for $l = 2$, although there are several recalcitrant factors of 2 appearing in the analysis which complicate this story. A similar argument gives combinatorial factors which seem to fix the discrepancy in the case $l = 3$, again up to some rogue factors of 2.} A more detailed discussion of the resolution of this problem will be described in a future publication. \section*{Acknowledgments} We would like to thank Lorenzo Cornalba, Steve Gubser, Dan Kabat, Igor Klebanov, Morten Krogh and Ricardo Schiappa for helpful conversations. The work of MVR is supported in part by the Natural Sciences and Engineering Research Council of Canada (NSERC). The work of WT is supported in part by the A.\ P.\ Sloan Foundation and in part by the DOE through contract \#DE-FC02-94ER40818.
\section{Introduction} \setcounter{equation}{0} The problem of integrability and nonintegrability of Hamiltonian system with $g$ degrees of freedom has been a subject of considerable interest for many years. Recently remarkable progress has been achieved in connection of study of stationary soliton type equations. The method of restricted flows was introduced in \cite{Mos1,20,Mos2} as a non-linearization of the Korteweg-de Vries (KdV) spectral problem and was generalized in \cite{5,2}. The coupled Neumann system, the Neumann system, the Garnier type systems, the Rosochatius-Wojciechowski, Rosochatius and the H\'enon-Heiles type systems are examples of this type. In this relation important results about the algebro-geometrical interpretaion of these systems are obtained in \cite {F}. General approach to completely integrable dynamical systems of Neumann type is discussed in \cite{S, AHP, AHH}. Quasi-periodic solutions and spectral interpretation using both algebro-geometrical and spectral methods are given in \cite{K}. Quasi-periodic solutions ($g=2$) and periodic solutions associated to Lam\'e and Treibich-Verdier potentials are obtained in \cite{EK, CEEK}. New method of constructing elliptic finite-gap solutions of the stationary KdV hierarchy, based on a theorem due to Picard, is proposed in \cite{GW95a,GW95b,GW95c,GW95d,GW96,GW98}. In this paper we are concerned with the following different approaches \begin{itemize} \item integrable dynamical systems related to Hill's equation in the case of finite-gap potential \cite{20, AM80, F, AHP, AHH, S, K, CC} \item method of stationary flows and restricted flows \cite{1,2} \item method of non-linearization of the KdV spectral problem \cite{5} \item method of separation of variables \cite{KKM, KRT, EEKT} \item algebro-geometrical construction \cite{F,S,K}. \end{itemize} We give unified construction based on algebro-geometrical approach. New solutions in terms of Novikov and Gelfand-Dickey (GD) polynomials are given in explicit form. The paper is organized as follows. In section 2 we construct the stationary flows associated to the KdV hierarchy strictly following \cite{1,7,8,9}. Some maps between completely integrable dynamical systems are presented. In section 3 we analyse the relation between restricted KdV flows and Garnier type systems following \cite{8}. The map between stationary and restricted KdV flows are given in section 4 \cite{8}. The list of known dynamical systems related to stationary KdV equations are presented in section 3. In section 4 we formulate Baker-Akhiezer function approach from algebro-geometrical and spectral point of view respectively. New solutions of integrable dynamical system associated to stationary KdV equations in terms of Novikov polynomials are presented . In particular these polynomials associated to Lam\'e potentials coincide with Hermite polynomials and Lam\'e polynomials. In section 5 ($2\times 2$) Lax representations of Garnier, Rosochtius I and stationary second flow of vector nonlinear Schr\"{o}dinger equation are given. Following \cite {EEKT,KRT,T1} $r$-matrix approach of these systems is discussed. \section{KdV hierarchy and Gelfand-Dickey polynomials} \setcounter{equation}{0} In this section we follow the geometrical construction of \cite{1,7,8,9,v98} with only little changes of signs and different spectral parameter. Let $M$ be a bi-Hamiltonian manifold: if the associated Poisson operator $P^{\lambda}:=P_{1}+4\lambda P_{0}$ admits as a Casimir a formal Laurent series $h(\lambda)$ \begin{equation} h(\lambda):=\sum_{j\geq0}\,h_{j}\lambda^{-j} \end{equation} then $h_{0}$ is a Casimir of $P_{0}$ and the coefficients $h_{j}$ ($j\geq0$) are the Hamiltonian functions of a hierarchy of bi-Hamiltonian vector fields $X_{j}$: \begin{equation} X_{j}=P_{1}dh_{j}=P_{0}dh_{j+1}, \quad (j\geq0) \label{VecF} \end{equation} At any point $u\in M$, the bi-Hamiltonian flows are given by $\frac{d u}{% dt_{j}} =X_{j}(u)$, $t_{j}$ being the evolution parameter of the $j$th flow. The vector field (\ref{VecF}) are Hamiltonian also with respect to the Poisson operator $P^{\lambda}$. In fact the recursion relation (\ref{VecF}) can be written as \begin{equation} X_{j}=P^{\lambda}\,dh^{(j)}(\lambda),\quad h^{(j)}(\lambda):=(\lambda^{j} h(\lambda))_{+} \label{X} \end{equation} where the index $+$ means the projection of a Laurent series onto the purely polynomial part. Let $M$ be the algebra of polynomials in $u, u_{x},u_{xx},\ldots$ ($u=u(x)$ is a $C^{\infty}$ function of $x$ and the subscript $x$ means the derivative with respect to $x$), and le $P_{0}$ and $P_{1}$ be the two Poisson operators of the KdV hierarchy \cite{11}: \begin{equation} P_{0}:=\frac{d}{dx}, \qquad P_{1}:=\frac{d^{3}}{dx^{3}} - 4 u \frac{d}{dx} - 2 u_{x} . \label{Opers} \end{equation} The gradients of the Casimirs of the associated Poisson operator $% P^{\lambda} $ can be obtained searching the functions $v(\lambda):=\sum_{j% \geq0}v_{j}\lambda^{-j}$ which are solutions of the following equation: \begin{equation} B^{\lambda}(v(\lambda),v(\lambda))=a(\lambda) \label{GDg} \end{equation} where $a(\lambda)=\sum_{j\geq0}a_{j}\lambda^{-j}$, $a_{j}$ are constant parameters and $B^{\lambda}$ is the bilinear function \begin{equation} B^{\lambda}(w_{1},w_{2}):=w_{1xx}w_{2}+w_{1}w_{2xx}-w_{1x}w_{2x}- 4(u-\lambda)w_{1}w_{2} . \label{Brel} \end{equation} In fact $B^{\lambda}$ is related to the Poisson operator through the relation \begin{equation} \frac{d}{dx}B^{\lambda}(w_{1},w_{2})=w_{1} P^{\lambda} w_{2} +w_{2} P^{\lambda} w_{1} . \end{equation} Equation (\ref{GDg}) can be solved developing the left-hand side as a Laurent series \begin{equation} B^{\lambda}(v(\lambda),v(\lambda)) = \sum_{k\geq -1} B_{k}\lambda^{-k} \end{equation} so that, for each $a(\lambda)$, it furnishes the coefficient of the solution $v(\lambda)$ (unique up to a sign). The solution corresponding to $\bar{a}% (\lambda) =-\lambda$ is the so-called basis solution $\bar{v}(\lambda)$; its first coefficients are \begin{eqnarray} \bar{v}_{0}=1,\quad \bar{v}_{1}=1, \quad \bar{v}=2(u_{xx}+3 u^{2}), \\ \bar{v}_{3}=2(u_{xxxx}+5u_{x}^{2}+10u_{xx}u+10u^{3}) , \end{eqnarray} and so on, namely the gradients of the first KdV Hamiltonians. In what follows we shall consider also the function $v(\lambda)=c(\lambda)\bar{v}% (\lambda)$, which is a solution of (\ref{GDg}) for \begin{equation} a(\lambda)=-\lambda c^{2}(\lambda), \quad c(\lambda)=1 + \sum_{j\geq 1}\,c_{j}\lambda^{-j} , \end{equation} where the coefficients $c_{j}$ are free parameters. In this case the first 1-forms of the hierarchy are $v_{0}=1, v_{1}=\bar{v}_{1}+c_{1}, v_{2}=\bar{v}% _{2}+c_{1}\bar{v}_{1}+c_{2}$, and so on. The coefficient $B_{k}$ can be expressed through the GD polynomials. For each Laurent series $v(\lambda)$ let us consider the functions $% B^{(k)}(\lambda):=B^{\lambda}(v(\lambda),v^{(k)}(\lambda))$, where $% v^{(k)}(\lambda):=\left(\lambda^{k}v(\lambda)\right)_{+}$; these functions have the form \begin{equation} B^{(k)}(\lambda)=\lambda^{k+1}v_{0}^{2}+ \sum_{j=1}^{k-1}\lambda^{k-j}(p_{0j} +v_{0}v_{j+1}) +\sum_{j\geq 0}\lambda^{-j}p_{jk} . \label{BB} \end{equation} It can be shown that \begin{equation} B_{-1}=-v_{0}^{2}, \quad B_{k}=p_{0k}-v_{0}v_{k+1} . \end{equation} Furthermore, if $v(\lambda)$ is a solution of (\ref{GDg}), the coefficients $% p_{jk}$ in (\ref{BB}) are polynomials in $u$ and its $x$-derivatives. They will be referred to as Gelfand-Dickey (GD) polynomials and the function $% B^{\lambda}$ as their generating function. The fundamental property of the GD polynomials, stemming from (\ref{GDg}), (% \ref{BB}) is the following relation with the gradients $v_{j}=dh_{j}$ and the bi-Hamiltonian vector field $X_{k}$: \begin{equation} \frac{d}{dx}p_{jk}=v_{j}X_{k} . \end{equation} We report some GD polynomials to be used in what follows $(v_{0}=1)$: \begin{eqnarray} \label{GD} p_{00}&=&4 u - v_{1} , \nonumber \\ p_{01}&=&8 u v_{1}-v_{1}^{2}-v_{2}+2 v_{1xx} , \nonumber \\ p_{02}&=&4uv_{1}^{2}+8uv_{2}-2v_{1}v_{2}-v_{3}-v_{1x}^{2}+ 2v_{1}v_{1xx}+2v_{2xx}, \nonumber \\ \\ p_{12}&=&8uv_{1}v_{2}-v_{2}^{2}+4uv_{3}-v_{1}v_{3}-v_{4}+ 2v_{1x}v_{2x}+2v_{2}v_{1xx}+2v_{1}v_{2xx}+v_{3xx} , \nonumber \\ p_{kk}&=&2v_{kxx}v_{k}-v_{k}^{2}+4uv_{k}^{2} . \nonumber \end{eqnarray} The GD polynomials corresponding to the basis solution $\bar{v}(\lambda)$ are the polynomials defined in \cite{1}, proposition 12.1.12. \subsection{The method of stationary flows} The method of stationary flows was developed in order to reduce the flows of the KdV hierarchy onto the set $M_{g}$ of fixed points of the $g$th flow $X_{g}$ of the hierarchy \begin{equation} M_{g}:=\left\{u|X_{g}(u,u_{x},\ldots,u^{(2g+1)})=0\right\}. \end{equation} As $M_{g}$ is odd-dimensional it can not be a symplectic manifold; nevertheless we will show that it is a bi-Hamiltonian manifold; it will be referred to as extended phase space. Moreover, $M_{g}$ is naturally foliated, on account of (\ref{VecF}) and (\ref{Opers}), by a one-parameter family of $2g$-dimensional submanifolds $S_{g}$ given by \begin{equation} S_{g}:=\left\{u|v_{g+1}(u,u_{x},\ldots,u^{(2g)}=c\right\} \end{equation} ($c$ being a constant parameter), which are invariant manifolds with respect to each vector field of the KdV hierarchy, due to the invariance of the functions $v_{k}$. So $M_{g}$ can be parametrized naturally by $% v_{1},\ldots, v_{g+1}$ and by their $x$-derivatives $v_{1x},\ldots,v_{gx}$. We shall use these coordinates in what follows. From the computational point of view, one proceeds as follows. (i) Due to (\ref{X}) and (\ref{GDg}), the manifold $M_{g}$ is defined by the solutions $u$ of the equation \begin{equation} B^{\lambda}(v(\lambda),v^{(g)}(\lambda))=\lambda^{g}a(\lambda) \label{BL} \end{equation} where $v(\lambda)=\sum_{j=0}^{g}v_{j}\lambda^{-j}$, $a(\lambda)=% \sum_{j=-1}^{2g} a_{j}\lambda^{-j}$. In particular, if $a(\lambda)=-\lambda c^{2}(\lambda)$, $M_{g}$ is given by \begin{equation} M_{n}=\left\{u|\bar{X}_{n}+\sum_{j=1}^{n}c_{j}\bar{X}_{n-j} = 0\right\} \end{equation} i.e. by the solutions of the Lax-Novikov equations. Taking into acount (\ref {BB}) and choosing $a_{-1}=-1$, by equation (\ref{BL}) the coefficients of $% \lambda^{g+1}$ we get $v_{0}^{2}=1$; from now on we put $v_{0}=1$. Moreover, equating the coefficients of the other powers of $\lambda$ we get the following system: \begin{equation} p_{0k} - v_{k+1} =a_{k},\quad (k=0\ldots, n-1)\quad p_{jn}=a_{n+j},\, (j=0,\ldots,n). \label{PP0} \end{equation} (ii) In order to obtain the first Poisson tensor $P_{0}$, we eliminate $% u=v_{1}/2+a_{0}/4$ from (\ref{PP0}) using the first equation ($k=0$) and we extract the system of $n$ second-order ODEs in the $v_{j}$ ($j=1,\ldots,n$): \begin{equation} p_{0k} - v_{k+1} =a_{k},\quad (k=1\ldots, g-1)\quad p_{0g}=a_{g} \label{P0} \end{equation} which will be referred as $P_{0}$-system. The remaining equations (\ref{PP0}% ) will furnish a set of $g$ independent integrals of motion. In order to obtain a second Poisson structure, we consider the following system: ($P_{1}$% -system) \begin{equation} p_{0k} - v_{k+1} =a_{k},\quad (k=1\ldots, g-1)\quad p_{gg}=a_{2g}, \label{P1} \end{equation} with $u$ as above. (iii) The $P_{0}$-system (\ref{P0}) and the $P_{1}$-system (\ref{P1}) can be written as canonical Hamiltonian systems \begin{equation} r_{kx}=\frac{\partial H_{g}^{(0)}}{\partial s_{k}}, \qquad s_{kx}=-\frac{% \partial H_{g}^{(0)}}{\partial r_{k}}, \end{equation} \begin{equation} q_{kx}=\frac{\partial H_{g}^{(1)}}{\partial p_{k}}, \qquad p_{kx}=-\frac{% \partial H_{g}^{(1)}}{\partial q_{k}} \label{Deq} \end{equation} have $n$ integrals of motion given by \begin{eqnarray} &&K_{j}=-\frac{1}{8}p_{jg}|_{r}=a_{g+j}, (j=1,\ldots,g), \\ &&H_{j}=-\frac{1}{8}p_{jg}|_{x}=a_{g+j}, (j=0,\ldots,g-1), \end{eqnarray} Moreover, the map $\Phi:M_{g}\rightarrow M_{g}$ in the extended phase space generates a second Poisson structure. \section{Integrable dynamical systems related to hierarchy of stationary KdV equations} \setcounter{equation}{0} In the recent years, remarkable progress has been achieved in the description of those quasi(periodic) potentials which belong to a given spectrum. Many integrable systems of differential equations are shown to be closely connected with Hill's equation in the case of a finite gap potential. The coupled Neumann system, The Neumann system, and the Rosochatius systems are examples of this type. In this paragraph we are concerned with the following completely integrable systems The Garnier system \begin{eqnarray} \xi_{ixx} = \left( 2\sum_{j=1}^{g} \xi_{j} \eta_{j} + \tilde{a}_{i}\right) \xi_{i} , \quad \eta_{ixx} = \left( 2\sum_{j=1}^{g} \xi_{j} \eta_{j} + \tilde{a}_{i}\right) \eta_{i} . \label{Garn} \end{eqnarray} The $g$-dimensional anisotropic harmonic oscillator in radial quartic potential, is obtained when $\xi_{i}=\eta_{i}$ $i=1,\ldots, g$ \begin{eqnarray} \xi_{ixx} = \left( 2\sum_{j=1}^{g} \xi_{j}^2 + \tilde{a}_{i}\right) \xi_{i} . \label{oscil} \end{eqnarray} Another interesting integrable system was proposed recently \cite{W}, we call it the Rosochatius-Wojciechowski system. In our context, this system is obtained by the Deift elimination procedure. Let \begin{equation} \xi_{i}=\psi_{i}\exp(\theta_{i}),\, \eta_{i}=\psi_{i}\exp(-\theta_{i}) ,\, \sqrt{f}_{i} = \psi_{i}^{2}\theta_{ix} , \label{Deift} \end{equation} then equations (\ref{Garn}) transform to the Rosochatius-Wojciechowski system \begin{eqnarray} \psi_{ixx} = \left( 2\sum_{j=1}^{g} \psi_{j}^2 \right) \psi_{i} +\tilde{a}_{i}\psi_{i}-f_{i}/\psi_{i}^3 , \,i=1,\ldots ,g \label{RosI} \end{eqnarray} and the Hamiltonian is given by \begin{eqnarray} H=\sum_{j=1}^{g}\,\chi_{j}^{2} - \left(\sum_{k=1}^{g}\,\psi_{k}^{2}\right)^{2} - \sum_{j=1}^{n} \tilde{a}_{j}\psi_{j}^{2} -\sum_{j=1}^{g} \frac{f_{j}}{\psi_{j}^{2}} , \end{eqnarray} where $\chi_{j}=\psi_{jx}$ are canonical momenta. (ii) The coupled Neumann system \begin{eqnarray} \tilde{\xi}_{ixx} + \left( 2\sum_{j=0}^{g} b_{j}\tilde{\xi}_{j} \tilde{\eta}% _{j} +\tilde{\xi}_{jx}\tilde{\eta}_{jx}\right)\tilde{\psi}_{i}= b_{i} \tilde{% \xi}_{i} , \\ \tilde{\eta}_{ixx} + \left( 2\sum_{j=0}^{g} b_{j}\tilde{\xi}_{j} \tilde{\eta}% _{j} +\tilde{\xi}_{jx}\tilde{\eta}_{jx}\right)\tilde{\eta}_{i}= b_{i} \tilde{% \eta}_{i} . \label{CouplNeum} \end{eqnarray} with constraint $\sum_{i=0}^{g}\tilde{\xi}_{i}\tilde{\eta}_{i} =1$, where $% b_{0}< b_{1}< \ldots < b_{g}$ are fixed real numbers. The Neumann system is obtained when $\xi_{i}=\eta_{i}$ \begin{eqnarray} \tilde{\xi}_{ixx} + \left( 2\sum_{j=0}^{g} b_{j}\tilde{\xi}_{j}^2 +\tilde{\xi% }_{jx}^2 \right)\tilde{\xi}_{i}= b_{i} \tilde{\xi}_{i} . \label{Neum} \end{eqnarray} This system describes the motion of uncoupled harmonic oscillators $\tilde{% \xi}_{ixx} = b_{i} \tilde{\xi}_{i}$, constrained by the force $% \sum_{i=0}^{g}(b_{i}\tilde{\xi}_{i}^2 +\tilde{\xi}_{ix}^2)$ to move on the unit sphere $\sum_{i=0}^{g}\tilde{\xi}_{i}^2=1$. Let \begin{equation} \tilde{\xi}_{i}=\tilde{\psi}_{i}\exp(\tilde{\theta}_{i}),\, \tilde{\eta}_{i}=\tilde{\psi}_{i}\exp(-\tilde{\theta}_{i}) ,\, \sqrt{\tilde{f}}_{i} = \tilde{\psi}_{i}^{2}\tilde{\theta}_{ix} , \label{Deift1} \end{equation} then by Deift procedure (\ref{Deift1}), equations (\ref{CouplNeum}) transform to Rosochatius system \begin{eqnarray} \tilde{r}_{ixx} = -\left( \sum_{j=0}^{g} b_{j}\tilde{r}_{j}^2 + \ \tilde{r}_{jx}^2- \frac{\tilde{f}_{j}}{\tilde{r}_{j}^{2}} \right)\tilde{r}_{i}-\frac{\tilde{f}_{j}}{\tilde{r}_{j}^{3}}+ b_{i} \tilde{r}_{i} . \label{Ros} \end{eqnarray} where $\sum_{i=0}^{g}\tilde{r}_{i}^2=1$. \section{Baker-Akhiezer function} \setcounter{equation}{0} We review in this section some basic facts about Baker--Akhieser function which will be used in the sequel. Let $K$ be the hyperelliptic Riemann surface $\mu^2=\prod_{i=0}^{2g} (\lambda-\lambda_{i})=R(\lambda ) $. The points of $K$ are pairs $% P=(\lambda,R)$ and $\lambda(P)$ is the value of the natural projection $% P\rightarrow\lambda(P)$ of $K$ to the complex projective line $CP^{1}$. For given nonspecial divisor $D$, there is an unuque Baker-Akhiezer(BA) function $\Psi(t,\lambda)$, such that (i) the divisor of the poles of $\Psi$ is $D$, (ii) $\Psi$ is meromorphic on $K\backslash\infty$ (iii) when $P\rightarrow \infty$ \begin{eqnarray} \Psi(x,P)\exp(-kx) = 1 +\sum_{s=1}^{\infty} m_{s}(x) k^{-s} , \label{expan1} \end{eqnarray} is holomorphic and $k=\sqrt{\lambda(P)}$ is a local parameter near $P=\infty$% . There is a unique function $u(x)$ such that \begin{equation} \Psi_{xx} - u(x)\Psi= \lambda(P)\Psi , \label{Hill} \end{equation} where $\Psi$ is a BA function. Inserting expansion (\ref{expan1}) into (\ref {Hill}), we obtain \begin{eqnarray} \Psi_{xx} - 2 m_{1x}(x)\Psi -\lambda(P)\Psi =\exp(kx) O(k^{-1}) , \end{eqnarray} and due to the uniqueness of $\Psi$, we prove (\ref{Hill}), with $u(x)=2 m_{1x}(x)$. By the Riemann-Roch theorem, there exists a unique differential $\tilde{% \Omega}$ and a nonspecial divisor $D^{\tau}$ of degree $g$ such that the zeros of $\tilde{\Omega}$ are $D + D^{\tau}$ and the expansion at $P=\infty$% , $\tilde{\Omega}(P)=(1+O(k^{-2})) dk$. For given nonspecial divisor $D^{\tau}$, there exists a unuque dual Baker-Akhiezer (BA) function such that (i) the divisor of the poles of $\Psi$ is $D^{\tau}$, (ii) $\Psi$ is meromorphic on $K\backslash\infty$ (iii) when $P\rightarrow \infty$ \begin{eqnarray} \Psi^{\tau}(x,P)\exp(-kx) = 1 +\sum_{s=1}^{\infty}\tilde{m}_{s}(x) k^{-s}, \label{expan} \end{eqnarray} Fix $\tau$ to be the hyperelliptic involution $P=(\lambda,R)\rightarrow(% \lambda,-R)$, then we have $D^{\tau}=\tau D$, $\Psi^{\tau}(x,P)=\Psi(x,\tau P)$. Let $\sum_{i=1}^{g}\mu_{i}(0)$ be the $\lambda$-projection of $D$, and $% \sum_{i=1}^{g}\mu_{i}(x)$ be the $\lambda$-projection of the zero divisor of $\Psi(x,P)$. The function $\Psi(t,P)\Psi^{\tau}(t,P)$ is meromorphic on $% {\bf CP^{1}}$ and the following identity takes place \begin{equation} \Psi(x,P)\Psi^{\tau}(x,P)=\frac{F(x,\lambda)}{F(0,\lambda)} \label{sqfun} \end{equation} where $F(x,\lambda)=\prod_{i=1}^{g}(\lambda - \mu_{i}(x))$. Introduce the Wronskian \begin{eqnarray} \left\{\Psi(x,P),\Psi^{\tau}(x,P)\right\}=&&\Psi_{x}(x,P)\Psi^{\tau}(x,P) -\Psi(x,P)\Psi_{x}^{\tau}(x,P) = \\ &&\frac{2\sqrt{R(\lambda)}}{\prod_{i=1}^{g}(\lambda -\mu_{i}(0)} , \nonumber \end{eqnarray} and the differential $\tilde{\Omega}$ is given explicitly by \begin{equation} \tilde{\Omega}(P)=\frac{1}{2}\prod_{i=1}^{g} (\lambda - \mu_{i}(0))/\sqrt{% R(\lambda)}d\lambda . \label{diff} \end{equation} We assume that $E(P)$ is a meromorphic function on $K$ with $g+1$ simple poles $\infty,p_{1},\ldots,p_{g}$ and at $P\rightarrow\infty$, $% E(P)=k+\ldots $, and $\tilde{E}(P)$ is meromorphic function with $g+1$ simple poles $q_{0},q_{1},\ldots ,q_{g}$ and at $P\rightarrow\infty$, $% \tilde{E}(P)=k^{-1} +\ldots$. We also suppose that the divisors of poles of $% E(P)$ and $\tilde{E}(P)$ are different from $D$, $D^{\tau}$. Let \begin{eqnarray} \tilde{\xi}_{i}=\tilde{\xi}_{i}^{0}\Psi(x,q_{i}),\quad \tilde{\eta}_{i}=% \tilde{\eta}_{i}^{0}\Psi^{\tau}(x,q_{i}),\quad \nonumber \\ \tilde{\xi}^{0}_{i}\tilde{\eta}_{i}^{0}=Res_{P=q_i}E\tilde{\Omega}, \quad b_{i}=\lambda(q_{i}),\,i=0,\ldots ,g \label{solu1} \\ \xi_{i}=\xi_{i}^{0}\Psi(x,p_{i}),\quad \eta_{i}=\eta_{i}^{0}\Psi^{\tau}(x,p_{i}),\quad \nonumber \\ \xi^{0}_{i}\eta_{i}^{0}=Res_{P=q_i}\tilde{E}\tilde{\Omega}, \quad a_{i}=\lambda(p_{i}),\,i=1,\ldots ,g \label{solu2} \end{eqnarray} then \begin{eqnarray} & &u(x)=-\left(\sum_{i=0}^{g}b_{i}\tilde{\xi}_{i}\tilde{\eta}_{i} +\tilde{\xi% }_{ix}\tilde{\eta}_{ix}\right),\quad \sum_{i=0}^{g}\tilde{\xi}_{i}\tilde{\eta% }_{i}=1 \nonumber \\ & &u(x)=2\sum_{i=1}^{g}\xi_{i}\eta_{i}+ const. \label{poten} \end{eqnarray} Let us construct the meromorphic differential $\tilde{E}\Psi\Psi^{\tau}% \tilde{\Omega}$. By direct computations we have \begin{equation} \sum_{i=0}^{g}\,Res_{P=q_{i}}\,\tilde{E}\Psi\Psi^{\tau}\tilde{\Omega} +Res_{P=\infty}\,\tilde{E}\Psi\Psi^{\tau}\tilde{\Omega} =\sum_{i=0}^{g}% \tilde{\xi}_{i}\tilde{\eta}_{i} - 1 = 0, \end{equation} where $\tilde{\xi}_{i}^{0}\tilde{\eta}_{i}^{0}=Res_{P=q_{i}}\,\tilde{E} \tilde{\Omega}$. Differentiating $\sum_{i=0}^{g}\tilde{\xi}_{i}\tilde{\eta}% _{i} = 1$ twice and using Eq. (\ref{solu1}), we obtain first expression in (% \ref{poten}). The eigenvalue equations \begin{eqnarray} &&\tilde{\xi}_{ixx}=(\lambda(q_{i}) + u(x))\tilde{\xi}_{i}, \\ &&\tilde{\eta}_{ixx}=(\lambda(q_{i}) + u(x))\tilde{\eta}_{i}, \end{eqnarray} by replacing $u(x)$ from (\ref{poten}) are the coupled Neumann system (\ref {CouplNeum}). By computations of the same kind, we have \begin{equation} \sum_{i=1}^{g}\,Res_{P=p_{i}}\,E\Psi\Psi^{\tau}\tilde{\Omega} +Res_{P=\infty}\,E\Psi\Psi^{\tau}\tilde{\Omega} =\sum_{i=1}^{g} \xi_{i} \eta_{i} - u(x)+\frac{1}{2} const. = 0, \end{equation} where $\xi_{i}^{0} \eta_{i}^{0}=Res_{P=p_{i}}\, E\tilde{\Omega}$. The corresponding eigenvalue equations are the Garnier system (\ref{Garn}). \subsection{Spectral interpretation} Let $p$ be a positive real divisor of degree $g$ on a real hyperelliptic curve \begin{equation} \mu^{2}=R(\lambda)=\prod_{i=0}^{2g}(\lambda-\lambda_{i}),\quad \lambda_{0} < \lambda_{1} <,\ldots ,<\lambda_{2g} . \end{equation} The projection $\lambda(p_{i})$ lie in the closed lacunae $% [\lambda_{2i-1},\lambda_{2i}]$. The following considerations, due to Jacobi, can be used to construct such a divisor. Each divisor $p$ determines and is determined by a system of polynomials \begin{eqnarray} &&\tilde{A}(\lambda)=\prod_{i=1}^{g}(\lambda-\lambda(p_{i})),\quad \tilde{C}% (\lambda)=\tilde{A}(\lambda)\, \sum_{i=1}^{g}\,\frac{\sqrt{R(p_{i})}} {% \tilde{A}^{\prime}(p_{i})(\lambda-\lambda(p_{i}))} , \nonumber \\ &&\tilde{B}(\lambda)=\lambda^{g+1}+\ldots , \end{eqnarray} of degrees $g$, $g-1$, $g+1$, respectively, with $R=\tilde{C}^2-\tilde{A}% \tilde{B}$. The complementary divisor $q$ is also determied by this construction. This is the content of the following step. For a given spectral data \begin{equation} \lambda_{0}=0 < \lambda_{1} < ,\ldots ,< \lambda_{2g},\quad \lambda(p_{i})\in [\lambda_{2i-1},\lambda_{2i}],\,i=1,\ldots,g \end{equation} thete exists \begin{equation} \lambda(q_{i}),\,\,i=0,\ldots,g\quad\lambda(q_{0})\in (-\infty,\lambda_{0}],\quad \lambda(q_{i})\in[\lambda_{2i-1},\lambda_{2i}] , \end{equation} such that $R=\tilde{C}^2-\tilde{A}\tilde{B}$, and the projections $% \lambda(q_{i})$ are the roots of $\tilde{B}$. Note that the functions $E(P)$, $\tilde{E}(P)$ are meromorphic on $K$ and the following formulas are immediate \begin{equation} E(P)=(\sqrt{R(\lambda)} +\tilde{C}(\lambda))/\tilde{A}(\lambda), \qquad \tilde{C}(p_{i})=\sqrt{R(p_{i})} \label{E} \end{equation} \begin{equation} \tilde{E}(P)=(\sqrt{R(\lambda)} +\tilde{C}(\lambda))/\tilde{B}(\lambda), \qquad \tilde{C}(q_{i})=-\sqrt{R(q_{i})} \label{tE} \end{equation} Now we recall some facts from the periodic theory of Hill's equation. We suppose that $u(x)$ is a real finite-gap potential, i.e. the operator $L$ has only $2g+1$ simple eigenvalues $\lambda_{0} < \lambda_{1} < ,\ldots ,<\lambda_{2g}$ and the rest of the spectrum consists of double eigenvalues. The periodic spectra of $L$ is determined by the combined eigenvalues of the periodic \begin{equation} L\,f_{2i} = \lambda_{2i}\,f_{2i},\quad f(x+1)=f(x),\,i=0,\ldots,g \end{equation} and the antiperiodic \begin{equation} L\,f_{2i-1} = \lambda_{2i-1}\,f_{2i-1},\quad f(x+1)=-f(x),\,i=1,\ldots,g \end{equation} eigenvalue equations. The intervals $(-\infty,\lambda_{0}],[\lambda_{2i-1},% \lambda_{2i}]$ are termed lacunae. The Floquet solutions(periodic BA function) and the corresponding Floquet multipliers, are given by \begin{eqnarray} &&\Psi(x,\lambda) =\left[ F(x,\lambda)/F(0,\lambda)\right]^{1/2}\, \exp\left(\int_{0}^{x}\,\sqrt{R(\lambda)}/F(x^{\prime},\lambda)dx^{\prime}% \right) , \label{BA1} \\ &&\Psi(x+1,\lambda)=\rho_{+}(\lambda)\Psi(x,\lambda) , \\ &&\Psi^{\tau}(x,\lambda) =\left[ F(x,\lambda)/F(0,\lambda)\right]^{1/2}\, \exp\left(-\int_{0}^{x}\,\sqrt{R(\lambda)}/F(x^{\prime},\lambda)dx^{\prime}% \right) , \label{BA2} \\ &&\Psi^{\tau}(x+1,\lambda)=\rho_{-}(\lambda)\Psi^{\tau}(x,\lambda) , \qquad \rho_{\pm}=\exp(\pm\tilde{p}(\lambda)) , \end{eqnarray} where \begin{equation} \tilde{p}(\lambda)=\int_{0}^{1}\,\sqrt{R(\lambda)}/F(x,\lambda)dx \end{equation} Note that if $\lambda$ is in the periodic spectrum, $\Psi(x,% \lambda_{2i})=f_{2i}$, $i=0,\ldots,g$ is a periodic eigenfunction, and $% \Psi(x,\lambda_{2i-1})=f_{2i-1}$,$i=1,\ldots ,g$ is an antiperiodic eigenfunction. It is well known that the projections of the zeros of the Floquet solution define the auxiliary spectrum of $L$. The following expressions hold \begin{eqnarray} && \xi_{i} \eta_{i}=\prod_{j=1}^{g}\,(\lambda(p_{i})-\mu_{j}(x))/\tilde{A}% ^{\prime}(p_{i}), \label{Sol1} \\ && \xi_{i}^{0} \eta_{i}^{0}=\prod_{j=1}^{g}\,(\lambda(p_{i})-\mu_{j}(0))/ \tilde{A}^{\prime}(p_{i}), \nonumber \\ && \tilde{\xi}_{i}\tilde{\eta}_{i}=\prod_{j=1}^{g}\,(\lambda(q_{i})-% \mu_{j}(x))/ \tilde{B}^{\prime}(q_{i}), \label{Sol2} \\ && \tilde{\xi}_{i}^{0} \tilde{\eta}_{i}^{0}=\prod_{j=1}^{g}\,(% \lambda(q_{i})-\mu_{j}(0))/ \tilde{B}^{\prime}(q_{i}). \nonumber \end{eqnarray} Using (\ref{E}),(\ref{tE}) we obtain \begin{eqnarray} Res_{P=p_{i}}\,E(P)\Psi\Psi^{\tau}\tilde{\Omega}& & = x_{i}^{0}y_{i}^{0}\Psi(x,p_{i})\Psi^{\tau}(x,p_{i}) \nonumber \\ & &=\prod_{j=1}^{g}\,(\lambda(p_{i})-\mu_{j}(x))/A^{\prime}(p_{i}) , \nonumber \end{eqnarray} where $\xi_{i}^{0}\eta_{i}^{0}$ is given by (\ref{Sol1}). Let \begin{eqnarray} e_{2i}^2=\frac{\prod_{i\neq j}^{g}\,(\lambda_{2i} -\lambda_{2j})} {\prod_{j\neq j=0}^{g}\,(\lambda_{2i}-\mu_{j}(0))} ,\quad f_{2i}^2=\frac{% \prod_{j=1}^{g}\,(\lambda_{2i} -\mu_{j}(t))} {\prod_{i\neq j=0}^{g}\,(\lambda_{2i}-\mu_{j}(0)) } = \Psi^2(\lambda_{2i}) , \end{eqnarray} $i=0,\ldots ,g$, then the expressions (\ref{poten}), (\ref{Sol2}) are the famous McKean-Moerbeke expansion of the potential $u(x)$ in terms of squares of the eigenfunctions \begin{equation} u(x)=-\left( 2\sum_{i=0}^{g}\,\lambda_{2i} f_{2i}^2/e_{2i} +\sum_{i=1}^{g}\lambda_{2i-1}-\sum_{i=1}^{g}\lambda_{2i} +\lambda_{0}\right), \end{equation} where the following identity among the squares of eigenfunctions hold on \begin{equation} \sum_{i=0}^{g}\,e_{2i}^{-2} f_{2i}^{2} =1. \end{equation} The results of this section, may be summarized by following: Let $u(x)$ be a real nonsingular finite-gap potential. There exists $g$ eigenfunctions $\Psi(p_{1}),\ldots ,\Psi(p_{g})$ and $g+1$ eigenfunctions $% \Psi(q_{0}),\ldots ,\Psi(q_{g})$ of Hill's equation, corresponding to the eigenvalues $\lambda(p_{1}),\ldots,\lambda(p_{g})$ and $\lambda(q_{0}),% \ldots ,\lambda(q_{g})$, respectively, such that (i) \begin{eqnarray} u(x) = 2\sum_{i=1}^{g}\,\Psi(p_{i})\Psi^{\tau}(p_{i}) e_{i}^{-2} + 2\sum_{i=1}^{g}\lambda(p_{i})-\sum_{i=0}^{2g}\lambda_{i} , \\ e_{i}^{-2}=\prod_{j=1}^{g}(\lambda(p_{i})-\mu_{j}(0))/\tilde{A}% ^{\prime}\quad \Psi(p_{i})\equiv\Psi(x,\lambda)|_{\lambda =p_{i}} ,\,\, i=1,\ldots,g . \nonumber \end{eqnarray} (ii) \begin{eqnarray} u(x) = 2\sum_{i=0}^{g}\,\lambda(q_{i})\Psi(q_{i})\Psi^{\tau}(q_{i}) e_{i}^{-2} - 2\sum_{i=0}^{g}\lambda(q_{i})+\sum_{i=0}^{2g}\lambda_{i} , \label{Expan1} \\ \tilde{e}_{i}^{-2}=\prod_{j=1}^{g}(\lambda(q_{i})-\mu_{j}(0))/\tilde{B}% ^{\prime}, \,\, \Psi(q_{i})\equiv\Psi(t,\lambda)|_{\lambda =q_{i}} ,\,\, i=0,\ldots,g , \nonumber \\ \sum_{i=0}^{g}\tilde{e}^{-2}\Psi(q_{i})\Psi^{\tau}(q_{i}) =1 . \end{eqnarray} The corresponding eigenvalue equations are the Garnier and coupled Neumann system. Let \begin{eqnarray} & &e_{2i-1}^{2}=\prod_{i\neq j}^{g}(\lambda_{2i-1} - \lambda_{2j-1})/ \prod_{j=1}^{g}(\lambda_{2i-1} -\mu_{j}(0)) , \nonumber \\ & &f_{2i-1}^{2}=\prod_{j=1}^{g}(\lambda_{2i-1} - \mu_{j}(x))/ \prod_{j=1}^{g}(\lambda_{2i-1} -\mu_{j}(0)) ,\, i=1,\ldots,g \nonumber \end{eqnarray} then we have the following expansion of the potential $u(x)$ in terms of squares of antiperiodic eigenfunctions \begin{equation} u(x)=2\sum_{i=1}^{g}\,f_{2i-1}^2 e_{2i-1}^{-2} +2 \sum_{i=1}^{g}\lambda_{2i-1} -\sum_{i=0}^{2g}\lambda_{i} . \end{equation} We call the dynamical systems such in (i) , (ii), complementary dynamical systems. \subsection{Solutions in terms of auxiliary spectrum of Hill's equation} The solutions of the Garnier system in terms of auxiliary spectrum $% \mu_{j}(x)$, $j=1,\ldots,g$ are \begin{eqnarray} &&\xi_{i} =\xi_{i}^{0}[ F(x,a_{i})/F(0,a_{i})]^{1/2}\, \exp\left(\int_{0}^{x}\,\sqrt{R(a_{i})}/F(x^{\prime},a_{i})dx^{\prime}% \right) , \label{soluG1} \\ &&\eta_{i}=\eta_{i}^{0}[ F(x,a_{i})/F(0,a_{i}]^{1/2}\, \exp\left(-\int_{0}^{x}\,\sqrt{R(a_{i})}/F(x^{\prime},a_{i})dx^{\prime}% \right) , \label{soluG2} \end{eqnarray} where $\mu_{j}(x)$ satisfies the following system of differential equations \begin{equation} \frac{d}{dx}\mu_{j}(x) = 2\sqrt{R(\mu_{j})}/\prod_{j\neq k}^{g}(\mu_{j}(x) - \mu_{k}(x)) , \label{muSys} \end{equation} with initial conditions \begin{equation} \mu_{j}(0)\in [\lambda_{2i-1},\lambda_{2i}],\quad \xi_{i}^{0}\eta_{i}^{0}=F(0,a_{i})/\prod_{i\neq j}^{g}(a_{i}-a_{j}). \end{equation} Differentiating expressions \begin{eqnarray} \xi_{i}\eta_{i}=\xi_{i}^{0}\eta_{i}^{0}\Psi(x,p_{i})\Psi^{\tau}(x,p_{i}) = \prod_{j=1}^{g}(\lambda(p_{i})-\mu_{j}(x))/\tilde{A}^{\prime}(p_{i}) , \end{eqnarray} and \begin{equation} \xi_{ix}\eta_{i}-\xi_{i}\eta_{ix} = \left\{\Psi(x,p_{i}),\Psi^{\tau}(x,p_{i})\right\}\xi_{i}^{0}\eta_{i}^{0} , \end{equation} we have \begin{eqnarray} \Upsilon(x,P)|_{\lambda=\lambda(p_{i})} = \frac{d}{dx} \log \xi_{i}(t) \nonumber \\ =[\frac{1}{2}\frac{d}{dx}\,\prod_{j=1}^{g}\,(\lambda -\mu_{j}(x)) +\sqrt{% R(\lambda)}]/\prod_{j=1}^{g}\,(\lambda -\mu_{j}(x)) |_{\lambda=\lambda(p_{i})} \label{chi1} \end{eqnarray} \begin{eqnarray} \Upsilon^{\tau}(x,P)|_{\lambda=\lambda(p_{i})} = \frac{d}{dx} \log \eta_{i}(t) \nonumber \\ =[\frac{1}{2}\frac{d}{dx}\,\prod_{j=1}^{g}\,(\lambda -\mu_{j}(x)) -\sqrt{% R(\lambda)}]/\prod_{j=1}^{g}\,(\lambda -\mu_{j}(x)) |_{\lambda=\lambda(p_{i})} \label{chi2} \end{eqnarray} direct integration of (\ref{chi1}), (\ref{chi2}) gives the solutions (\ref {soluG1}), (\ref{soluG2}). The function $\Upsilon(x,P)$ has $g$ poles at $% \mu_{j}(x)$, then the numerator of (\ref{chi1}) is zero when $% \lambda=\mu_{j}(x)$ and the following system takes place \begin{equation} \frac{d}{dx}\left(\prod_{j=1}^{g}\,(\lambda -\mu_{j}(x))\right) |_{\lambda=\mu_{j}(x)} = 2\sqrt{R( \mu_{j}(x)) } . \end{equation} This is another form of the system (\ref{muSys}). In the same way, we can obtain the solutions of the coupled Neumann system by replacing $% \lambda(p_{i})$, $i=1,\ldots,g$ with $\lambda(q_{i})$, $i=0,\ldots,g$ in (% \ref{soluG1}), (\ref{soluG2}). Let $\lambda(p_{i})$ be the antiperiodic eigenvalues $\lambda_{2i-1}$, $% i=1,\ldots,g$. Then the exponential function in (\ref{soluG1}), (\ref{soluG2}% ) cancel, $\xi_{i}^{0}=\eta_{i}^{0}$ and the solutions of the $g$% -dimensional oscillator are \begin{equation} \xi_{i}^{2}=\prod_{j=1}^{g}\,(\lambda_{2i-1} - \mu_{j}(x))/ \prod_{i\neq j}^{g}\,(\lambda_{2i-1}-\lambda_{2j-1}) . \end{equation} Let $\lambda(p_{i})=a_{i}$ be in a general position, i.e. $\lambda(p_{i})\in [\lambda_{2i-1},\lambda_{2i}] $ and by Deift elimination procedure we may identify $\xi_{i}$ with $\xi_{i}^{0}[F(x,a_{i})/F(0,a_{i})]^{1/2}$ and \begin{equation} \theta_{i}=\int_{0}^{x}\,\sqrt{R(a_{i})}/\prod_{i=1}^{g}(a_{i}-\mu_{j}(x^{% \prime}))dx^{\prime}, \quad \xi_{i}^{0}=\eta_{i}^{0} , \end{equation} and, hence, the solutions of the Rosochatius-Wojciechowski system are \begin{equation} \xi_{i}^2=\prod_{j=1}^{g}\,(a_{i}-\mu_{j}(x))/ \prod_{i\neq j}^{g}(a_{i}-a_{j}) . \label{GarnT} \end{equation} Inserting explicit expression of BA-function given by (\ref{BA1}) in Hill's equation we have \begin{eqnarray} \frac{1}{2} F_{xx}(x,\lambda)F(x,\lambda) - \frac{1}{4} F_{x}^{2}(x,% \lambda)-(u(x)+\lambda)F^{2}(x,\lambda)= -R(\lambda) . \label{qEQ} \end{eqnarray} The polynomial solution of (\ref{qEQ}) below we will call Novikov polynomial \cite{Nov}. Assuming that Novikov polynomial dependts on time $t$, the zero curvature representation for KdV hierarchy of equations have the following form \begin{equation} M_{t}(\lambda')-L_x(\lambda^{\prime})+\left[M(\lambda`),L(\lambda^{\prime}) \right ] , \label{ZCeq} \end{equation} where matrices $L$ and $M$ are given by \begin{eqnarray} L(\lambda^{\prime})&=&\left( \begin{array}{cc} -F_{x}(x,t,\lambda^{\prime})/2 & F(x,t,\lambda^{\prime}) \\ -F_{xx}(x,t,\lambda^{\prime})/2 +Q(x,t,\lambda')F(x,t,\lambda') & F_{x}(x,t,\lambda^{\prime})/2 \end{array} \right) \nonumber \\ M^{\prime}(\lambda^{\prime})&=&\left( \begin{array}{cc} 0 & 1 \\ Q(x,t,\lambda') & 0 \end{array} \right)\,. \nonumber \end{eqnarray} The equation (\ref{ZCeq}) is equivalent to \begin{equation} \frac{\partial Q}{\partial t}= -2\left[\frac{1}{4}\partial^3_x -Q(x,t,\lambda^{\prime})\partial_x -\frac{1}{2} Q_{x}(x,t,\lambda^{\prime}) \,\right]\cdot F(x,\lambda^{\prime})\,. \label{Geq} \end{equation} where $Q(x,t,\lambda^{\prime})=u(x)+\lambda^{\prime}$ in the case of KdV hierarchy. Equation (\ref{Geq}) is called the generating equation. For a different choices of the form of $F(x,t,\lambda^{\prime})$ and $Q(x,t,\lambda^{\prime})$, this procedure leads to different hierarchies of integrable equations, as an example to the KdV, nonlinear Shr\"{o}dinger and sine-Gordon hierarchies or to the Dym hierarchy \cite{ACHM}. The Lax representation $L_{x}=[M,L]$ yields the hyperelliptic curve $K=(\mu^{\prime},\lambda^{\prime})$ \begin{eqnarray} &&\mbox{Det}(L(\lambda^{\prime})-\mu^{\prime}\mbox{I})=0 , \\ &&\mu^{2}=-\frac{1}{2}F F_{xx}+\frac{1}{4}F_{x}^{2}+(\lambda'+u)F=R(\lambda') , \nonumber \end{eqnarray} generating the integrals of motion for stationary KdV hierarchy. Using the equation (\ref{qEQ}) and the following expansion of potential $u(x)$ in terms of squares of eigenvalue functions $\xi_{k}^{2}(x)=\beta_{k}(x)$ \begin{equation} u(x)=2\sum_{i=1}^{g}\xi_{k}^2(x)+2\sum_{i=1}^{g} a_{i} -\sum_{k=0}^{2g}\lambda_{i} , \end{equation} have the form \begin{eqnarray} \frac{1}{2}\beta_{kxx}\beta_{k}-\frac{1}{4}\beta_{k}^{2}- (u(x)+a_{k})\beta_{k}^{2}=-\frac{R(a_{k})}{a_{kj}}=-f_{k} \label{SolGarnT} \end{eqnarray} where we use the solutions $F(x,a_{k})/a_{kj}$ of Rosochatius-Wojciechowski system and $a_{kj}\equiv\prod_{k\neq j}^{g}(a_{k}-a_{j})$. Denoting $% f_{k}=R(a_{k})/a_{kj}$ and $d=\sum_{i=1}^{g}a_{i}-\frac{1}{2}% \sum_{k=0}^{2g}\lambda_{k}$ for the original variable $\beta_{k}=\xi_{k}^{2}$ we have the following equation \begin{eqnarray} \xi_{kxx}=2\left(\sum_{i=1}^{g}\xi_{k}^{2} +d\right)\xi_{k}-\frac{f_{k}}{% \xi_{k}^{3}} . \end{eqnarray} To understand the role of GD polynomials and of their generating function in the construction of a map between stationary and restricted flows of KdV equation and exact solution of completely integrable systems related to Hill's equation let us consider the following system: \begin{eqnarray} &&p_{00}-v_{1}=a_{0}, \quad P_{0}\left(v_{1}-\sum_{j=1}^{n}\beta_{j}\right), \\ &&P^{\lambda_{k}}\beta_{k}=0, \qquad (k=1,\ldots,n) \end{eqnarray} where$\lambda_{1},\ldots,\lambda_{n}$ are fixed parameters, $% P^{\lambda_{k}}:=P_{1}+4\lambda_{k} P_{0}$ ($P_{0}$ and $P_{1}$ being the two KdV Poisson operators). This is a system of $(g+2)$ equations in $u, v_{1},\beta_{1},\ldots,\beta_{g}$. The second equation will be referred to as the $P_{0}$-restriction of the first KdV flow $X_{0}=P_{0}v_{1}=v_{1x}$, and the last $n$ equations define the kernel of $g$ Poisson operators extracted from the Poisson operator. On account of (\ref{GD}), (\ref{Opers}) and (\ref{Brel}) this system is equivalent to the following one: \begin{eqnarray} u=\frac{v_{1}}{2}+\frac{a_{0}}{4},\quad v_{1}=\sum_{j=1}^{n}\beta_{j} + c,\quad B^{\lambda_{k}}(\beta_{k},\beta_{k})=f_{k} , \end{eqnarray} where $c$ and $f_{k}$ are free parameters and $B^{\lambda}$ is just the generating function of the GD polynomials. Using the first two equations to eliminate $u$ and $v_{1}$ from the last $g$ equations, one gets a system of $n$ second-order ODEs for $% \beta_{1},\ldots,\beta_{g}$ : \begin{eqnarray} 2\beta_{kxx}\beta_{k}-\beta_{kx}^{2}+2\beta_{k}^{2} \left(\sum_{j=1}^{n}\beta_{j} +d\right) -\lambda_{k}\beta_{k}^{2} = f_{k}, \quad k=1,\ldots,g \label{QRos} \end{eqnarray} where $d:=c+a_{0}/2$. Introducing the so-called eigenfunction variables $% \psi_{j}^{2}=\beta_{j}$ and the momenta $\chi_{j}=\psi_{jx}$, equations (\ref {QRos}) can be written in canonical Hamiltonian form \begin{eqnarray} \psi_{jx}=\frac{\partial K_{G}}{\partial \chi_{j}}, \quad \chi_{jx}=-\frac{% \partial K_{G}}{\partial \psi_{j}}, \quad j=1,\ldots,n \end{eqnarray} with Hamiltonian \begin{eqnarray} K_{G}=\sum_{j=1}^{n}\chi_{j}^{2} -\left(\sum_{k=1}^{n}\psi_{j}^{2}\right)^{2}- \sum_{j=1}^{n} a_{j}\psi_{j}^{2}- \sum_{j=1}^{n}\frac{f_{j}}{\psi_{j}^{2}} . \end{eqnarray} A set of integrals of motion is \begin{eqnarray} I_{j}=&&\chi_{j}^{2}-\psi_{j}^{2}\left(a_{j}+ \sum_{k=1}^{g}\psi_{k}^{2}\right)- \frac{f_{j}}{\psi_{j}^{2}}+ \nonumber \\ && \sum_{k\neq j}^{n}\frac{1}{a_{jk}} \left(-\frac{f_{j}\psi_{k}^{2}}{% \psi_{j}^{2}} - \frac{f_{k}\psi_{j}^{2}}{\psi_{k}^{2}}+ (\psi_{j}\chi_{k}-\psi_{k}\chi_{j})^{2}\right) . \label{IntRosI} \end{eqnarray} where we denote $a_{jk}=a_{j}-a_{k}$. Now we shall construct a map between the $g$-th stationary flow and the previous restricted flow of the KdV hierarchy. To this end we extend the corresponding phase spaces, regarding some free parameters in the Hamiltonian functions as addiitional dynamical variables. As for the $P_{1}$-formulation of the stationary flow we extend its phase space to a $(3g+1)$-dimensional space, $\tilde{M}_{n}$, with coordinates $% (q_{k},p_{k};a_{0},\ldots,a_{g-1},a_{2g})$; analogously we consider the $% P_{0}$-formulation of the first restricted flow in the extended space $% \tilde{{\it M}}_{g}$ with coordinates $(\psi_{k},\chi_{k};f_{1},% \ldots,f_{k},d)$. Let us consider the solutions $q_{k}$ of the dynamical equations (\ref{Deq}% ); then $v^{(g)}(\lambda)$ given by \begin{equation} v^{(g)}(\lambda)=\lambda(q^{2}(\lambda))^{(g-1)}-q_{g}^{2} , \end{equation} with $q(\lambda)=1+\sum_{j=1}^{g}\,q_{j}\lambda^{-j}$, satisfies (2.17), and consequently satisfies the following equation: \begin{equation} B^{\lambda}(v^{(g)},v^{(n)})=\lambda^{2g} \mbox{d}(\lambda) , \end{equation} where, as above, we put $u=v_{1}/2+a_{0}/4$. So, for each $g$-tuple of distinct complex parameters $a_{j}$, any solution $v^{(g)}(\lambda)$ fulfills the system \begin{equation} B^{a_{k}}(v^{(g)}(a_{k}),v^{(g)}(a_{k})=a_{k}^{2g} % \mbox{d}(a_{k}) , \quad (k=1,\ldots,g) \end{equation} where $v^{(g)}(a_{k}):=v^{(\lambda)}|_{\lambda=a_{k}}$. In order to have a solution also satisfying the second equation $v_{1}= \sum_{j=1}^{g}\beta_{j} +c$, the Lagrange interpolation formula can be used. It allows us to represent the polynomial $v^{n}(\lambda)$ by \begin{equation} v^{(n)}(\lambda)=a(\lambda)\left(1+\sum_{j=1}^{g}\frac{\beta_{j} } {\lambda-a_{j} }\right) , \label{LagIF} \end{equation} where $a(\lambda)=\prod_{j=1}^{g}(\lambda-a_{j})$, and \begin{equation} \beta_{j}=\frac{v^{(g)}(a_{k})}{a^{\prime}(a_{k})}, (k=1,\ldots,g) . \end{equation} ($a^{\prime}(\lambda)$ means the derivative of $a(\lambda)$ with respect to $% \lambda$. Obviously the $g$ functions $\beta_{k}$ are solutions of the following system \begin{eqnarray} 2\beta_{kxx}\beta_{k}-\beta_{kx}^{2}+2\beta_{k}^{2} \left(\sum_{j=1}^{g}\beta_{j} +d\right) -\lambda_{k}\beta_{k}^{2} = \frac{% \lambda_{k}^{2g}\mbox{d}(a_{k})}{(a^{\prime}(a_{k}))^{2}} , \quad k=1,\ldots,g \label{QRos1} \end{eqnarray} Furthermore, $\beta_{k}$ satisfy the so-called Bargmann constraint \begin{equation} \sum_{j=1}^{g} (\beta_{j}-a_{j}) = v_{1} , \end{equation} as one can verify by means of (\ref{LagIF}). The function $B^{\lambda}$ is also a generating function of integrals of motion for Garnier system. Indeed evaluating the function $B^{\lambda}$ by means of (\ref{LagIF}) and eliminating the first $x$-derivatives of $% \chi_{k} $ by means of Hamilton equations (\ref{Deq}), one gets \begin{eqnarray} 4\sum_{j=1}^{g}\frac{I_{j}}{\lambda-\lambda_{j}} + \sum_{j=1}^{g} \frac{f_{j}% }{(\lambda-\lambda_{j})^{2}} + 2d-\lambda=\frac{\lambda^{2g}\hat{a}(\lambda)% }{(a(\lambda))^{2}} , \end{eqnarray} where $I_{j}$ are the functions. Taking in this equation the residues at $% \lambda=a_{j}$ it follows that the functions $I_{j}$ are integrals of motion along the flow (\ref{Deq}). Let $\lambda(q_{i})=b_{i}$ be in a general position, i.e. $\lambda(q_{i})=b_{i}\in (-\infty,\lambda_{0}], [\lambda_{2i-1},\lambda_{2i}] $ and by Deift elimination procedure we may identify $\tilde{\psi}_{i}$ with $\tilde{\xi}% _{i}^{0}[F(t,b_{i})/F(0,b_{i})]^{1/2}$ and \begin{equation} \tilde{\theta}_{i}=\int_{0}^{x}\,\sqrt{R(b_{i})}/\prod_{i=1}^{g} (b_{i}-\mu_{j}(x^{\prime}))dx^{\prime}, \end{equation} and, hence, the solutions of the Rosochatius system are \begin{equation} \tilde{\psi}_{i}^2=\prod_{j=0}^{g}\,(b_{i}-\mu_{j}(x))/ \prod_{i\neq j}^{g}(b_{i}-b_{j}) . \end{equation} where $i,j=0,\ldots g$. Now we illustrate the general aproach with some simple examples {\bf Example 1} Let $u(x)=6\wp(x+\omega^{\prime})$ be the two-gap Lam\'e potential with simple periodic spectrum (see for example \cite{EK}) \begin{equation} \lambda_{0}=-\sqrt{3g_{2}},\quad \lambda_{1}=-3e_{0}\quad , \lambda_{2}=-3e_{1}\quad , \lambda_{3}=-3e_{2}\quad , \lambda_{4}=\sqrt{% 3g_{2}}. \end{equation} and the corresponding Hermite polynomial have the form \begin{equation} F(\wp(x+\omega^{\prime}),\lambda)=\lambda^{2}- 3\wp(x+\omega^{\prime})\lambda+ 9\wp^{2}(x+\omega^{\prime})-\frac{9}{4}g_{2} . \label{HerPol} \end{equation} Consider the following genus $2$ nonlinear anisotropic oscillator with Hamiltonian \begin{equation} H=\frac{1}{2}(p_{1}^{2}+p_{2}^{2})+\frac{1}{4}(q_{1}^{2}+q_{2}^{2})^{2}- \frac{1}{2}(a_{1}q_{1}^{2}+a_{2}q_{2}^{2}), \end{equation} where $(q_{i},p_{i})$, $i=1,2$ are canonical variables with $p_{i}=q_{ix}$ and $a_{1},a_{2}$ are arbitrary constants. The simple solutions of these system are given in terms of Hermite polynomial \begin{equation} q_1^2=2\frac{F(x,\tilde{\lambda}_{1})} {\tilde{\lambda}_{2}-\tilde{\lambda}% _{1}} ,\quad q_2^2=2\frac{F(x,\tilde{\lambda}_{2})} {\tilde{\lambda}_{1}-% \tilde{\lambda}_{2}} , \end{equation} Let us list the corresponding solutions \begin{itemize} \item Periodic solutions expressed in terms of single Jacobian elliptic functions \end{itemize} The nonlinear anisotropic oscillator admit the following solutions: \begin{eqnarray} q_1 & = & C_1 \mbox{sn}(\alpha x, k), \\ q_2 & = & C_2 \mbox{cn}(\alpha x, k), \label{onegap} \end{eqnarray} where amplitudes $C_1$ ,$C_2$ and temporal pulsewidth $1/\alpha$ of are defined by parameters $a_{1}$ and $a_{2}$ as \begin{eqnarray} \alpha^2 k^2 & = & a_{2}-a_{1} \nonumber \\ C^{2}_{1} & = & a_{2} + \alpha^2 - \alpha^2 k^2 , \nonumber \\ C^{2}_{2} & = & a_{1} + \alpha^{2} , \nonumber \end{eqnarray} where $0 < k < 1$. Following our spectral method it is clear, that the solutions \ref{onegap} are associated with eigenvalues $\lambda_2 = - e_2$ and $\lambda_3 = - e_3$ of one -- gap Lam\'e potential. \begin{itemize} \item Periodic solutions expressed in terms of products of Jacobian elliptic functions \end{itemize} \begin{eqnarray} q_1 & = & C \mbox{dn}(\alpha x, k) \mbox{sn}(\alpha x, k), \\ q_2 & = & C \mbox{dn}(\alpha x ,k ) \mbox{cn}(\alpha x, k), \label{Flor} \end{eqnarray} where $\mbox{sn}$,$\mbox{cn}$, $\mbox{dn}$ are the standard Jacobian elliptic functions \cite{BE}, $k$ is the modulus of the elliptic functions $% 0 < k < 1$, an the wave characteristic parameters: amplitude $C$, temporal pulsewidth $1/\alpha$ and $k$ are related to the physical parameters and, $k$ through the following dispersion relations \begin{eqnarray} C^{2} & = & \frac{2 (4a_{2} - a_{1})}{5} , \nonumber \\ k^{2} & = & \frac{(4a_{2}-a_{1})} {15} , \nonumber \\ \alpha^{2} & = & \frac {5 (a_{2}-a_{1})} {4a_{2}-a_{1}} . \nonumber \end{eqnarray} We have found the following solutions of the nonlinear oscillator \begin{eqnarray} q_1 & = & C\alpha^2 k^2 \mbox{cn}(\alpha x,k)\mbox{sn}(\alpha x,k) \\ q_2 & = & C\alpha^2 \mbox{dn}^2 (\alpha x, k) + C_{1} \label{UzKos} \end{eqnarray} where $C$, $C_1$, $\alpha$ and $k$ are expressed through parameters $a_{1}$ and $a_{2}$ by the following relations \begin{eqnarray} C^2 & = & \frac {18} {a_{2}-a_{1}} , \nonumber \\ C_1 & = & \frac { C (4a_{1}-a_{2})}{5},\quad \nonumber \\ k^2 & = & \frac {2 \sqrt { \frac{5}{3} (a_{2}^{2}-a_{1}^{2})}} {2 \sqrt{\frac{5}{3} (a_{2}^{2}-a_{1}^{2})}+aa_{2}-3a_{1}}, \nonumber \\ \alpha^2 & = & \frac {1}{10} ( 2 a_{2}-3a_{1}+ \sqrt{\frac{5}{3}(a_{2}^{2}-a_{1}^2) } ). \end{eqnarray} \begin{itemize} \item Periodic solutions associated with the two-gap Treibich-Verdier potentials \end{itemize} Below we construct the two periodic solutions associated with the Treibich-Verdier potential. Let us consider the potential \begin{equation} u(x)=6\wp(x+\omega^{\prime})+2{\frac{(e_1-e_2)(e_1-e_3)}{\wp(x+\omega^{% \prime})-e_1}} \label{tv4} \end{equation} and construct the solution in terms of Lam\'e polynomials associated with the eigenvalues $\tilde{\lambda}_1,\tilde{\lambda}_2$, $\tilde{\lambda}_1 > \tilde{\lambda}_2$ \begin{eqnarray} \tilde{\lambda}_1&=&e_2+2e_1+2\sqrt{(e_1-e_2)(7e_1+2e_2)}, \nonumber \\ \tilde{\lambda}_2&=&e_3+2e_1+2\sqrt{(e_1-e_3)(7e_1+2e_3)}. \label{zz} \end{eqnarray} The finite and real solutions $q_1,q_2$ have the form \begin{eqnarray} q_1&=& C_{1}\mbox{sn}(z,k)\mbox{dn}(z,k) +C_{2}\mbox{sd}(z,k) , \nonumber \\ q_2&=& C_{3}\mbox{cn}(z,k)\mbox{dn}(z,k) +C_{4}\mbox{cd}(z,k), \label{ee2} \end{eqnarray} where $C_{i}$, $i=1,\ldots 4$ are constants and have important geometrical interpretation \cite{EK}. The concrete expressions in terms of $k,\tilde{% \lambda}_{1},\tilde{\lambda_{2}}$ are given in \cite{CEEK} Analogously we can find the elliptic solution associated with the eigenvalues \begin{eqnarray} \tilde{\lambda}_1&=&e_2+2e_1+2\sqrt{(e_1-e_2)(7e_1+2e_2)},\quad \tilde{% \lambda}_2=-6e_1, \label{zz2} \end{eqnarray} We have \begin{eqnarray} q_1&=&\tilde{C}_{1}\mbox{dn}^{2}(z,k) \\ q_2&=& C_{1}\mbox{sn}(z,k)\mbox{dn}(z,k) +C_{2}\mbox{sd}(z,k) , \label{ee3} \end{eqnarray} where $C$ is given in \cite{CEEK}. The general formula for elliptic solutions of genus $2$ nonlinear anisotropic oscillator is given in \cite{CEEK} \begin{eqnarray} q_1^2&=&{\frac{1}{\tilde{\lambda}_2-\tilde{\lambda}_1}} \left(2\tilde{\lambda}_1^2+2\tilde{\lambda}_1\sum_{i=1}^N \wp(x-x_i) \right. \nonumber \\ &&+\left.6\sum_{1\leq i< j\leq N}\wp(x-x_i)\wp(x-x_j)-{\frac{Ng_2}{4}}+ \sum_{1\leq i< j\leq 5}\lambda_i\lambda_j\right), \label{q1} \\ q_2^2&=&{\frac{1}{\tilde{\lambda}_1-\tilde{\lambda}_2}} \left(2\tilde{\lambda}_2^2+2\tilde{\lambda}_2\sum_{i=1}^N \wp(x-x_i) \right. \nonumber \\ &&+\left.6\sum_{1\leq i< j\leq N}\wp(x-x_i)\wp(x-x_j)-{\frac{Ng_2}{4}}+ \sum_{1\leq i< j\leq 5}\lambda_i\lambda_j\right), \label{q2} \end{eqnarray} where $x_{i}$ are solutions of equations $\sum_{i\neq j}\wp'(x_{i}-x_{j})=0, j=1,\ldots, N$ and $N$ is positive integer. {\bf Example 2} Garnier type system. Consider the following genus $2$ Garnier type system with Hamiltonian \begin{equation} H=\frac{1}{2}(p_{1}^{2}+p_{2}^{2})+\frac{1}{4}(q_{1}^{2}+q_{2}^{2})^{2}- \frac{1}{2}(\gamma_{1} q_{1}^{2}+\gamma_{2} q_{2}^{2})+\frac{f_{1}}{q_{1}^{2}}+ \frac{f_{2}}{q_{2}^{2}}, \end{equation} where $(q_{i},p_{i})$, $i=1,2$ are canonical variables with $p_{i}=q_{ix}$ and $a_{1},a_{2}$ are arbitrary constants. The simple solutions of these system are given in terms of Hermite polynomial in the following form (\ref{GarnT}) \begin{equation} q_1^2=2\frac{F(x,a_{1})} {a_{1}-a_{2}} , \quad q_2^2=2\frac{F(x,a_{2})} {a_{2}-a_{1}} , \end{equation} the same settings of periodic spectra, Lam\'e potential and Hermite polynomials (\ref{HerPol}) as in example 2. The main results from the general theory are the following: i) the parameters $f_{1}$ and $f_{2}$ are expressed in terms of algebraic curve by (\ref{SolGarnT}) \begin{eqnarray} f_{1}=\frac{R(a_{1})}{a_{1}-a_{2}}, \quad f_{2}=\frac{R(a_{2})}{a_{2}-a_{1}}, \nonumber \\ R(\lambda)=(\lambda^{2}-3g_{2})(\lambda+3e_{1})(\lambda+3e_{2})(\lambda+3e_{3}) . \nonumber \end{eqnarray} ii) the parameters $a_{1}$ and $a_{2}$ must lie in one or other of the intervals \begin{equation} [-\sqrt{3g_{2}}, 3e_{1}], [3e_{2},3e_{3}], [\sqrt{3g_{2}},\infty) \end{equation} and $\gamma_{1}=3a_{1}+2a_{2}, \gamma_{2}=2a_{1}+3a_{2}$. These results are in complete agreement with solutions obtained by different method in recent paper \cite{pp98}. {\bf Example 3} Simple solutions of the H\'enon-Heiles type system. We consider a generalized H\'enon-Heiles type system with two-degrees of freedom. Its Hamiltonian is \begin{equation} H_{0}=\frac{1}{2}(p_{1}^{2}+p_{2}^{2}) + q_{1}^{3}+\frac{1}{2}q_{1} q_{2}^{2} + \frac{a_{4}}{8 q_{2}^{2}} +\frac{a_{0}}{2}\left(q_{1}^{2} +\frac{% 1}{4} q_{2}^{2}\right) -\frac{a_{1}}{4} q_{1} , \label{HH} \end{equation} where $q_{1}, q_{2}, p_{1}, p_{2}$ are the canonical coordinates and momenta and $a_{0}, a_{1}, a_{4}$ are free constant parameters. This Hamiltonian encompasses the two cases $a_{0}=a_{4}=0$ and $a_{0}=a_{1}=0$ introduced in \cite{18}. Moreover $H_{0}$ is related with the Hamiltonian \begin{equation} H_{H}= \frac{1}{2}(p_{1}^{2}+p_{2}^{2}) + \bar{q}_{1}^{3}+\frac{1}{2}\bar{q}% _{1} \bar{q}_{2}^{2} + \frac{\bar{a_{4}}}{{8 \bar{q}_{2}^{2}}} +\frac{1}{2}\left(A\bar{q}_{1}^{2} + B \bar{q}_{2}^{2}\right) , \end{equation} through the map \begin{equation} q_{1}=\bar{q}_{1} + \frac{A}{2} -2B,\quad q_{2}=\bar{q}_{2}, \quad a_{0}=-2 A+12 B, \quad a_{1}=-A^{2}+16 AB-48 B^{2} . \end{equation} The function $H_{H}$ is the Hamiltonian of a classical integrable H\'enon-Heiles system with the additional term $a_{4}/8\bar{q}% _{2}^{2}$. The function (\ref{HH}) is the Hamiltonian of the vector field obtained reducing $X_{0}(u)=u_{x}$ to the stationary manifold $M_{2}$ given by the fixed points of the flow $X_{2}+c_{1}X_{1}+c_{2}X_{0}$ \begin{equation} M_{2}={u|u^{(5)}+10u_{xxx}u+20u_{xx}u_{x}+30u_{x}u^{2}+ c_{1}(u_{xxx}+6u_{x}u) +c_{2}u_{x}=0} \end{equation} where $c_{1}=-a_{0}/2$, $c_{2}=-a_{1}/2+a_{0}^{2}/4$. It can be obtained specializing to the case $g=2$ the Hamiltonian of the $% P_{1}$-system. In this case $H_{2}^{(1)}=H_{0}$ and the canonical coordinates and momenta are, respectively, $q_{1}=v_{1}/2$, $% q_{2}^{2}=-v_{2} $, $p_{1}=q_{1x}$, $p_{2}=q_{2x}$. The integrals of motion obtained by the reduction of the GD polynomials are \begin{eqnarray} &&H_{0}\equiv -\frac{1}{8}p_{02}|_{x} \\ &&H_{2}\equiv -\frac{1}{8}p_{22}|_{x} =-\frac{a_{4}}{8} \\ &&H_{1}\equiv -\frac{1}{8}p_{12}|_{x}= \end{eqnarray} where $-\frac{1}{8}p_{12}|_{x}$ is given by \begin{eqnarray} p_{2}q_{1}-p_{1}p_{2}q_{2}-\frac{1}{2}q_{1}^{2}q_{2}^{2}-\frac{1}{8}% q_{2}^{4}+ \frac{a_{4}q_{1}}{4q_{2}^{2}}-\frac{a_{0}}{4}q_{1}q_{2}^{2}+ \frac{a_{1}}{8}q_{2}^{2} . \end{eqnarray} Next we will derive $(2\times 2)$ matrix Lax representation for generalized H\'enon-Heiles system (\ref{HH}). Using Lax representation $L_{x}=[M,L]$, particular case of eq. (\ref{ZCeq}) i.e. when there is no time dependence, we have \begin{eqnarray} &&F(x,\lambda)=\lambda^2+\frac{1}{2} q_{1}\lambda-\frac{1}{16}q_{2}^{2}, \quad \quad V=-F_{x}/2=-\frac{1}{4}p_{1}\lambda + \frac{1}{16} q_{2} p_{2} , \nonumber \\ &&W=-F_{xx}/2+Q F=\lambda^{3}-(\frac{1}{2}q_{1}+\frac{1}{4} a_{0})\lambda^2+ \nonumber \\ &&(\frac{1}{4} q_{1}^{2}+\frac{1}{16} q_{2}^{2}-\frac{1}{16}a_{1}+ \frac{1}{8} a_{0} q_{1})\lambda + \frac{1}{16}p_{2}^{2} + \frac{1}{64} \frac{a_{4}}{q_{2}^{2}} , \quad Q(x,\lambda)=\lambda-q_{1}-\frac{1}{4}a_{0} . \nonumber \end{eqnarray} The corresponding algebraic curve have the form \begin{eqnarray} \mu^2=\lambda^{5}-\frac{1}{4}a_{0}\lambda^{4}-\frac{1}{16}a_{1}\lambda^{3} +8 H_{0}\lambda^{2}+32\,H_{1}\lambda-\frac{1}{1024}a_{4} . \end{eqnarray} Using explicit expression for Hermite polinomial (\ref{HerPol}) we obtain the following simple solutions for the system (\ref{HH}): \begin{eqnarray} q_{1}=-6\wp(x+\omega'), \qquad q_{2}^2=-16\,(9\wp(x+\omega')^{2} -\frac{9}{4} g_{2}) . \end{eqnarray} where $a_{0}=0, a_{1}=3.4.7 g_{2}, A_{4}=-3^{4}.4^{4} g_{2} g_{3}$. \section{$2\times 2$ Lax representation and $r$-matrix approach} \setcounter{equation}{0} The Lax equation for completely integrable systems discussed in the previous section \begin{equation} L_x(\lambda^{\prime})=\left[M(\lambda^{\prime}),L(\lambda^{\prime})\right]% \,, \label{Lax} \end{equation} with matrices $L$ and $M$ given by \begin{eqnarray} L(\lambda^{\prime})&=&\left( \begin{array}{cc} V(x,\lambda^{\prime}) & U(x,\lambda^{\prime}) \\ W(x,\lambda^{\prime}) & -V(x,\lambda^{\prime}) \end{array} \right) \label{Llm1} \\ M(\lambda^{\prime})&=&\left( \begin{array}{cc} 0 & 1 \\ Q(x,\lambda^{\prime}) & 0 \end{array} \right)\,. \label{Llm2} \end{eqnarray} is equivalent to the Garnier system, where $U(x,\lambda^{\prime}), V(x,\lambda^{\prime}), W(x,\lambda^{\prime}), Q(x,\lambda^{\prime})$ have the form \begin{eqnarray} U(x,\lambda^{\prime})=a(\lambda^{\prime})\left(1-\sum_{i=1}^{g} \frac{% \xi_{i}\eta_{i}}{\lambda^{\prime}- a_{i}}\right), \, V(x,\lambda^{\prime})=-\frac{1}{2}U_{x}(x,\lambda^{\prime}) \\ W(x,\lambda^{\prime})=a(\lambda^{\prime})\left(\lambda^{\prime}+% \sum_{i=1}^{g}\,\xi_{i}\eta_{i}+ \sum_{i=1}^{g}\frac{\xi_{ix}\eta_{ix}}{% \lambda^{\prime}- a_{i}}\right), \quad Q(x,\lambda^{\prime})=\lambda^{\prime}+ 2\sum_{i=1}^{g}\xi_{i}\eta_{i} . \end{eqnarray} Finally we point out one usefull expression, which is easy to derive from Lax representation (\ref{Lax}) \begin{equation} W(x,\lambda^{\prime})=U(x,\lambda^{\prime})Q(x,\lambda^{\prime})-\frac{1}{2} U(x,\lambda^{\prime})_{xx}. \label{W} \end{equation} The Lax representation yields the hyperelliptic curve $K=(\mu^{\prime},% \lambda^{\prime})$ \begin{equation} \mbox{Det}(L(\lambda^{\prime})-\mu^{\prime}\mbox{I})=0 , \end{equation} generating the integrals of motion $H, F^{(i)}, i=1,\ldots,g$. We have \begin{equation} \mu^{2}=V^{2}(x,\lambda^{\prime})+U(x,\lambda^{\prime}) W(x,\lambda^{\prime}) , \label{Lcurve} \end{equation} From (\ref{Lcurve}) and explicit expressions of $U(x,\lambda^{\prime}), V(x,\lambda^{\prime}), W(x,\lambda^{\prime})$ we obtain \begin{equation} \mu^{2}=a(\lambda^{\prime})^{2}\left(\lambda^{\prime}+ \sum_{i=1}^{g}\frac{% H_{i}}{\lambda^{\prime}-a_{i}} + \frac{1}{4}\sum_{i=1}^{g}\frac{J_{i}^{2}}{% (\lambda^{\prime}-a_{i})^{2}} +\frac{1}{2}\sum_{i=1}^{g}\frac{I_{i}}{% \lambda^{\prime}- a_{i}}\right) , \end{equation} where \begin{eqnarray} I_{i}&=&\sum_{k\neq i}\frac{ (\xi_{k}\eta_{ix}-\eta_{i}\xi_{kx}) (\eta_{k}\xi_{ix}-\xi_{i}\eta_{kx})}{% a_{k}-a_{i}} , \nonumber \\ &+&\sum_{k\neq i}\frac{ (\xi_{i}\xi_{kx}-\xi_{k}\xi_{ix}) (\eta_{i}\eta_{kx}-\eta_{k}\eta_{ix})}{% a_{k}-a_{i}} , \nonumber \\ &&H_{i}=\xi_{ix}\eta_{ix}-a_{i}\xi_{i}\eta_{i}-\xi_{i}\eta_{i} \left(\sum_{k=1}^{g}\xi_{k}\eta_{k}\right) , \, J_{i}=\xi_{ix}\eta_{i}-\xi_{i}\eta_{ix}, \end{eqnarray} and $\sum_{i=1}^{g}H_{i}$ is the Hamiltonian for Garnier system. Simple reduction $\eta_{i}= \xi_{i}^{*}$ gives us the second flow of stationary vector nonlinear Schr\"{o}dinger equation, where by $*$ we denote complex conjugation. The complementary to the last system is complex Neumann system (see for example \cite{IA}) \begin{eqnarray} \tilde{\xi}_{ixx}+2\left(\sum_{j=0}^{g} b_{j}|\tilde{\xi}_{j}|^{2} +|\tilde{% \xi}_{jx}|^{2}\right)\tilde{\xi}_{i}=b_{i}\tilde{\xi}_{i} . \end{eqnarray} with $\sum_{i=0}^{g}|\xi_{i}|^{2}=1$. Using the Deift elimination procedure we obtain new $2\times 2$ Lax pair for Rosochatius-Wojciechowski system. Bellow we list only the final results for Lax pair elements of considered in this paper dynamical systems: \begin{itemize} \item Rosochatius-Wojciechowski system \begin{eqnarray} &&U(x,\lambda^{\prime})=a(\lambda^{\prime})\left(1-\sum_{i=1}^{g} \frac{% \psi_{i}^{2}}{\lambda^{\prime}- a_{i}}\right), \, V(x,\lambda^{\prime})=-\frac{1}{2}U_{x}(x,\lambda^{\prime}) \nonumber \\ &&W(x,\lambda^{\prime})=a(\lambda^{\prime})\left(\lambda^{\prime}+% \sum_{i=1}^{g}\,\psi_{i}^{2} + \sum_{i=1}^{g}\frac{1}{% \lambda^{\prime}- a_{i}} (\psi_{ix}^{2}-\frac{f_{i}}{\psi_{i}^{2}})\right) , \\ && Q(x,\lambda^{\prime})=\lambda^{\prime}+ 2\sum_{i=1}^{g}\psi_{i}^{2} . \nonumber \end{eqnarray} \item Rosochatius system \begin{eqnarray} &&U(x,\lambda^{\prime})=b(\lambda^{\prime})\left(\sum_{i=1}^{g} \frac{% \psi_{i}^{2}}{\lambda^{\prime}- b_{i}}\right), \, V(x,\lambda^{\prime})=-\frac{1}{2}U_{x}(x,\lambda^{\prime}) \nonumber \\ &&W(x,\lambda^{\prime})=b(\lambda^{\prime})\left(1-% + \sum_{i=1}^{g}\frac{1}{% \lambda^{\prime}- b_{i}} (\psi_{ix}^{2}-\frac{f_{i}}{\psi_{i}^{2}})\right) , \\ && Q(x,\lambda^{\prime})=\lambda^{\prime}+ 2\sum_{i=1}^{g}\psi_{i}^{2} , \nonumber \end{eqnarray} where $b(\lambda^{\prime})=\prod_{i=0}^{g}(\lambda^{\prime}-b_{i})$. \item second stationary flow of vector NLSE \begin{eqnarray} &&U(x,\lambda^{\prime})=a(\lambda^{\prime})\left(1-\sum_{i=1}^{g} \frac{% |\xi_{i}|^{2}}{\lambda^{\prime}- a_{i}}\right), \, V(x,\lambda^{\prime})=-\frac{1}{2}U_{x}(x,\lambda^{\prime})\nonumber \\ &&W(x,\lambda^{\prime})=a(\lambda^{\prime})\left(\lambda^{\prime}+% \sum_{i=1}^{g}\,|\xi_{i}|^{2}+ \sum_{i=1}^{g}\frac{|\xi_{ix}|^{2}}{% \lambda^{\prime}- a_{i}}\right), \\ && Q(x,\lambda^{\prime})=\lambda^{\prime}+ 2\sum_{i=1}^{g}|\xi_{i}|^{2} . \nonumber \end{eqnarray} \item complex Neumann system \begin{eqnarray} &&U(x,\lambda^{\prime})=b(\lambda^{\prime})\left(\sum_{i=1}^{g} \frac{% |\xi_{i}|^{2}}{\lambda^{\prime}- b_{i}}\right), \, V(x,\lambda^{\prime})=-\frac{1}{2}U_{x}(x,\lambda^{\prime})\nonumber \\ &&W(x,\lambda^{\prime})=a(\lambda^{\prime})\left(1-% \sum_{i=1}^{g}\frac{|\xi_{ix}|^{2}}{% \lambda^{\prime}- b_{i}}\right), \\ && Q(x,\lambda^{\prime})=\lambda^{\prime}+ 2\sum_{i=1}^{g}|\xi_{i}|^{2} . \nonumber \end{eqnarray} \end{itemize} Finally we want to point out that $I_{i}$ for Rosochatius-Wojciechowski system coincide with expression given in (\ref{IntRosI}). Another Lax equation have the following form \[ L_x(\lambda^{\prime})=\left[M(\lambda`),L(\lambda^{\prime})\right]\,, \] where matrices $L$ and $M$ are given by \begin{eqnarray} L(\lambda^{\prime})&=&\left( \begin{array}{cc} -F_{x}(x,\lambda^{\prime})/2 & F(x,\lambda^{\prime}) \\ -F_{xx}(x,\lambda^{\prime})/2 & F_{x}(x,\lambda^{\prime})/2 \end{array} \right) \equiv \nonumber \\ && \left( \begin{array}{cc} V(x,\lambda^{\prime}) & U(x,\lambda^{\prime}) \\ W^{\prime}(x,\lambda^{\prime}) & -V(x,\lambda^{\prime}) \end{array} \right) \, , \label{KdV1} \\ M^{\prime}(\lambda^{\prime})&=&\left( \begin{array}{cc} 0 & 1 \\ 0 & 0 \end{array} \right)\,. \label{KdV2} \end{eqnarray} where we made the following identification $U(x,\lambda^{\prime})=F(x,% \lambda^{\prime})$. The Poisson bracket relations for the matrix $% L(\lambda^{\prime})$ \cite{KRT,T1} are closed into the following $r$-matrix algebra \begin{equation} \left\{{L_{1}}(\lambda^{\prime}),{L_{2}}(\mu^{\prime})\right\}= [r_{12}(\lambda^{\prime},\mu^{\prime}),{L_{1}}(\lambda^{\prime})]-[r_{21}(% \lambda^{\prime},\mu^{\prime}), {L_{2}}(\mu^{\prime})\,]\, . \label{Lalg} \end{equation} Here the standard notations are introduced: \begin{eqnarray} \label{r1} &&L_{1}(\lambda^{\prime})= L(\lambda^{\prime})\otimes I\,,\qquad L_{2}(\mu^{\prime})=I\otimes L(\mu^{\prime})\,, \nonumber \\ \\ &&r_{12}(\lambda,\mu)=\frac{\Pi}{\lambda-\mu}\,\qquad r_{21}(\lambda,\mu)=\Pi r_{12}(\mu,\lambda)\Pi\,, \label{r2} \end{eqnarray} and $\Pi$ is the permutation operator of auxiliary spaces. The Poisson bracket relations for the Lax matrix $L(\lambda^{\prime})$ are preassigned by the initial symplectic structure. It is necessary to calculate only two brackets \begin{equation} \left\{F(x,\lambda^{\prime}),F(x,\mu^{\prime})\right\}=0\,, \label{Br1} \end{equation} and \begin{eqnarray} \label{Br2} \left\{V(x,\lambda^{\prime}),F(x,\mu^{\prime})\right\}&=& \nonumber \\ &&\left\{ F(x,\lambda^{\prime})\sum_{j=1}^g \frac{\mbox{g}_{jj}\,p_j(x)} {\lambda^{\prime}-\mu_{j}(x)}, \prod_{j=1}^g(\mu^{\prime}-\mu_{j}(x))\right\} \nonumber \\ &=& -F(x,\lambda^{\prime})F(x,\mu^{\prime})\sum_{j=1}^n \frac{\mbox{g}_{jj}}{% (\lambda^{\prime}-\mu_{j}(x))(\mu^{\prime}-\mu_{j}(x))} \nonumber \\ &=&\frac{F(x,\lambda^{\prime})F(x,\mu^{\prime})}{\lambda^{\prime}-\mu^{% \prime}} \sum_{j=1}^g \left(\frac{\mbox{g}_{jj}}{\lambda^{\prime}-\mu_{j}(x)}% - \frac{\mbox{g}_{jj}}{\mu^{\prime}-\mu^{\prime}_{j}(x)}\right) \nonumber \\ &=&\frac{1}{\lambda^{\prime}-\mu^{\prime}}\left[F(x,\mu^{\prime})- F(x,\lambda^{\prime})\right]\,, \end{eqnarray} where we used a standard decomposition of rational function \[ F(x,\lambda^{\prime})^{-1}= \sum_{j=1}^n\frac{\mbox{g}_{jj}}{% \lambda-\mu_{j}(x)}\,,\qquad \mbox{g}_{jj}=\left.\mbox{Res}% \right|_{\lambda^{\prime}=\mu_{j}(x)} F(x,\lambda^{\prime})^{-1}(\lambda)\,. \] and the following definitions \begin{eqnarray} &&\mbox{g}_{jj}=\mbox{Res}_{|\lambda=\mu_{j}(x)}F^{-1}(x,\lambda)= \frac{1}{% \prod_{k\neq j}(\mu_{j}(x)-\mu_{k}(x))}, \\ &&p_{j}(x)=V(x,\lambda)_{|\lambda=\mu_{j}(x)}=\sqrt{R(\mu_{j}(x))}. \end{eqnarray} Another Poisson brackets may be directly derived from these brackets and by definition of the entries of the Lax matrix $L(\lambda)$ via derivative of the single function $F(x,\lambda^{\prime})$ \begin{eqnarray} \label{Br3} \left\{V(x,\lambda^{\prime}),V(x,\mu^{\prime})\right\}&=&0\, \nonumber \\ \left\{W(x,\lambda^{\prime}),F(x,\mu^{\prime})\right\}&=&\frac{d}{dx} \left\{V(x,\lambda^{\prime}),F(x,\mu^{\prime})\right\}= \nonumber \\ &&\frac{2}{\lambda^{\prime}-\mu^{\prime}}\left[V(x,\lambda^{\prime})-V(x,% \mu^{\prime}) \right]\, , \nonumber \\ \\ \left\{W(x,\lambda^{\prime}),F(x,\mu^{\prime})\right\}&=&-\frac{1}{2}\frac{% d^2}{dx^{2}} \left\{V(x,\lambda^{\prime}),F(x,\mu^{\prime})\right\}= \nonumber \\ &&\frac{1}{\lambda^{\prime}-\mu^{\prime}} \left[W(x,\lambda^{\prime})-W(x,% \mu^{\prime})\right]\, , \nonumber \\ \left\{W(x,\lambda^{\prime}),W(x,\mu^{\prime})\right\}&=&-\frac{1}{2}\frac{% d^3}{dx^{3}} \left\{V(x,\lambda^{\prime}),F(x,\mu^{\prime})\right\}=0 \, . \nonumber \end{eqnarray} Applying the following transformation directly to the Lax representation $% L(\lambda^{\prime})$ we obtain a family of the new Lax pairs \cite{KRT,T1} \begin{eqnarray} \label{LnLax} &&L^{\prime}(\lambda^{\prime})=L(\lambda^{\prime})-\sigma_-\cdot \left[\phi(x,\lambda^{\prime})F(x,\lambda^{\prime})^{-1}\right]_{N}\,, \qquad \sigma_-=\left( \begin{array}{cc} 0 & 0 \\ 1 & 0 \end{array} \right)\,, \nonumber \\ \\ &&M^{\prime}(\lambda')=M-\sigma_{-}\cdot\left[\phi(x,\lambda')F(x,\lambda')^{-2 } \right]_{N} = \left( \begin{array}{cc} 0 & 1 \\ Q(x,\lambda') & 0 \end{array} \right)\,. \nonumber \end{eqnarray} Here $\phi(x,\lambda^{\prime})$ is a function on spectral parameter and $[z]_{N}$ means restriction of $z$ onto the ${\rm ad}^*_R$-invariant Poisson subspace of the initial $r$-bracket. For the rational $r$-matrix we can use the linear combinations of the following Taylor projections \begin{equation} {[ z ]_{N}}=\left[\sum_{k=-\infty}^{+\infty} z_k\lambda^k\, \right]_{N}\equiv \sum_{k=0}^{N} z_k\lambda^k\,, \label{Cut} \end{equation} or the Laurent projections. New Lax matrix $L^{\prime}(\lambda^{\prime})$ \cite{KRT,T1} obeys the linear $r$-bracket, where constant $r_{ij}$-matrices substituted by $% r_{ij}^{\prime} $-matrices depending on dynamical variables. \begin{eqnarray} &&r_{12}(\lambda^{\prime},\mu^{\prime})\rightarrow r^{\prime}_{12}=r_{12} \nonumber \\ && -\frac{\left( [\phi(\lambda) F(x,\lambda^{\prime})^{-2}(\lambda^{\prime})]_{N}- [\phi(\mu) F(x,\mu')^{-2}]_{N}\,\right)} {(\lambda^{\prime}- \mu^{\prime})}\cdot\sigma_{-}\otimes\sigma_{-} \,. \label{dpr} \end{eqnarray} \section{Conclusions} In this paper we have given new exact solutions of the physically significant completely integrable dynamical systems. These solutions can be interpreted as eigenfunctions of suitable differential operator. New example of complementary dynamical systems (stationary second flow of the vector nonlinear Schr\"odinger equation and complex Neumann system) is presented. \vspace{2cm} {\bf\Large Acknowledgements} \\*\\*I wish to thank to Prof. V. Gerdjikov, who poined out me the papers \cite{7,8,9}, book \cite{v98} and for valuable discussions.
\section{Introduction} The Sagittarius dwarf galaxy discovered \cite{ib94,ib95} as a comoving group ($V_{hel}\approx 140 {\rm kms^{-1}}$) with a velocity dispersion as small as $2 \rm kms^{-1}$, is the nearest external galaxy and will be amenable to detailed abundance studies with VLT and other 8m class telescopes. These shall be important for two distinct reasons: on the one hand, we shall be able to study chemical evolution in an environment different from our Galaxy; on the other hand, if Sagittarius shows any chemical signature, this should allow us to identify Sagittarius debris, currently populating the Galactic Halo. The colour-magnitude diagram of Sgr shows a wide Red Giant Branch (RGB), which is interpreted as evidence of a spread in metallicity of Sgr. Marconi et al \cite{mara} conclude that the RGB of Sgr lies between that of 47 Tuc ([Fe/H]$=-0.71$) and M2 ([Fe/H]$=-1.58$). In the following I shall describe the work done with the ESO NTT to determine spectroscopic metallicities of giants in Sgr and shall address the potentiality of VLT on this issue. \section{NTT Observations} We selected, from the Marconi et al \cite{mara} sample, stars on the RGB which ought to display the spread in metallicity present in Sgr. We used the ESO NTT telescope with the EMMI instrument in Multi Object Spectrocopy (MOS) mode. The dispersing element was grism \# 5 providing a resolution of about 1500. We have obtained spectra for a total of 57 objects in field Sgr1 of \cite{mara} on June 19th 1996. In addition, we obtained long slit spectra of one of the stars observed with the MOS and of star HD 190287 on September 18th 1998. \begin{figure} \centering \includegraphics[]{pl_histo_fe.epsi} \caption[]{Metallicity distribution for 22 radial velocity members of Sgr } \label{eps1} \end{figure} Of the 57 stars observed 23 matched the criterion 100 $\rm kms^{-1}\le v_{hel}\le 180 ~ kms^{-1}$ \cite{ib97}, which we adopted, to ascribe membership to Sagittarius. For these 23 stars the average radial velocity is $136\rm ~kms^{-1}\pm 18 ~kms^{-1}$, in good agreement with previous determinations; the r.m.s. is dominated by the error in the measure of the radial velocity and not by the velocity dispersion of Sgr. \section{Abundance estimates from low resolution spectra} Our approach has been to define spectral indices which measure some prominent spectral features which may be used for abundance estimates. We defined six spectral indices, two of which measure the Mg I b triplet and the remaining four measure essentially iron and iron-peak elements. The location of the indices may be found in Fig.1 of reference\cite{marb} . The indices are all measured with respect to a common pseudo-continuum, defined by six quasi-continuum windows. The aim is to be able to determine both [Fe/H] and [Mg/Fe]. We computed the values of the indices for a small grid of synthetic spectra (T=4750 K, 5000 K, 5250 K; log g = 2.50 [Fe/H]$=-1.5,-1.0,-0.5,0.0,+0.5$ ; $\rm \xi=2kms^{-1}$) computed with the SYNTHE code \cite{Kur93}. The spectra where broadened with a gaussian profile of 210 $\rm kms^{-1}$ to match the resolution of the observed spectra. The input model-atmospheres were computed with version 9 of the ATLAS code\cite{Kur93} switching off the overshooting option. Models with [Fe/H]$=-1.5$ and $-1.0$ were computed using $\alpha-$enhanced opacity distribution functions. For each star the temperature was determined from the $(V-I)_0$ colour\cite{mara} through the calibration of\cite{alo} . Strictly speaking this calibration is valid only for dwarf stars, however the $V-I$ colour has only a weak dependence on gravity, especially in the colour range we are interested in. We excluded from analysis SgrM 172 ($V-I=1.81$), because it is much cooler than all the other stars. The metallicity was determined from the four iron-sensitive indices by interpolating in the table of computed indices. \subsection{Zero-Points} We made the same analysis for the Sun using the solar atlas\cite{Kur84}, degraded at the same resolution as our grism spectra, as an observed spectrum and a small grid of synthetic spectra appropriate for the Sun (T=5777, log g =4.44, [Fe/H]$=-0.5,0.0,+0.5$). The difference between these derived metallicities and the meteoritic iron abundance, is small (around 0.1--0.2 dex). On the assumption that these differences reflect inadequacies of our atomic data and model atmosphere and that the systematics is the same for the Sun and for our program stars, we treated these as zero-point shifts. \subsection{Cross-checks} We determined the metallicity of the Sun from a twilight spectrum taken with the MOS. The deduced [Fe/H] was $-0.32,-0.04,-0.01,+0.10$ for the four indices respectively. These are not 0.00 because of the noise in the observed spectrum. One of the indices (F1) yields an abundance considerably different from the others. To obtain the final metallicity we decided to use a central estimate, defined as the mean of the two central values, which is akin to the median and more robust than the mean. In this case the central estimate is [Fe/H]$=-0.02$ , in good agreement with the expected value of 0.00. Our program stars are considerably cooler and more luminous than the Sun. We analysed the spectrum of the cool metal-poor giant HD 190287 ([Fe/H]$=-1.34$, $\rm T_{eff}=5300$), taken with EMMI, grism \# 5 and a long-slit. We obtain [Fe/H] $=-1.52,-1.22,-1.23,-1.37$, for the four indices, which yields a central estimate [Fe/H]$=-1.30$. We compared a spectrum of one of our program stars (SgrM 139) through the EMMI long-slit with its MOS spectrum. The derived metallicities were the same ($\rm[Fe/H]_{MOS}=-0.05$ ; $\rm [Fe/H]_{LS} = +0.02$) and also the radial velocities were consistent within errors ($v_{MOS}=+147 {\rm kms^{-1}}$ ; $v_{LS}=+124 {\rm kms^{-1}}$). \subsection{Error estimates} In order to estimate the random errors in our metallicity estimates we used a Montecarlo simulation. We took a synthetic spectrum ($\rm T_{eff} = 5000 K$ [Fe/H]$=-0.5$) to which we added Poisson noise so that S/N = 10 (i.e. the S/N of our spectra) and estimated the metallicity from this noisy spectrum. The mean of 10000 realizations was $\rm <[Fe/H]>= -0.56$ and the standard deviation was 0.25 dex, we take this as our estimate of the random error. In addition one should consider the errors arising from uncertainties in the atmospheric parameters. The largest error comes from the microturbulent velocity. At this resolution we are forced to rely only on strong lines which are very sensitive to microturbulence. The value of $\xi$ must be assumed since there is no way to fix it from the spectrum. The associated error is larger for the more metal-rich stars, if the microturbulence is 1 $\rm kms^{-1}$ the metallicities increase by $\rm \Delta([Fe/H]) = 0.24\times[Fe/H]+0.38 $. A change in effective temperature by 150 K brings about a change of 0.2 dex in metallicity as does a change of surface gravity by 0.5 dex. \subsection{Previous results and future work} Preliminary results of this work have been already presented\cite{marb,bon}. There are two main differences with the results presented here. The first is that here we use pre-tabulated synthetic indices rather than iteratively performing spectrum synthesis for each star and each index. The second is that in the present work the indices are measured with respect to the pseudo-continuum, rather than with respect to an estimate of the true continuum, obtained by iteratively re-normalizing the observed spectrum using the current estimate of the best-matching synthetic spectrum, as done previously. The re-normalization procedure was abandoned because we realized that it is strongly dependent on the initial metallicity estimate. In fact it tends to relax the observed spectrum onto the first guess. Differences on the order of 0.2 dex are obtained from the same observed spectrum but initial metallicity guesses differing by 0.5 dex. Our present results are in agreement with our previous estimates at the level of 0.1 dex except for the two most metal-poor stars (SgrM 124 and SgrM 115) for which we find a metallicity which is 0.32 and 0.41 dex lower. We have so far not determined [Mg/Fe] because the indices M1 and M2 depend on both [Mg/H] and [Fe/H]. Our grid of synthetic spectra includes only ``standard'' values of [Mg/Fe] (0.0 down to [Fe/H]$=-0.5$ and +0.4 below). On the other hand, our preliminary results suggest different [Mg/Fe] ratios, especially for $\rm -0.5 < [Fe/H] < 0.0 $ . On the contrary the F indices are essentially independent of the abundance of $\alpha$ elements. Determining [Mg/Fe] is still possible, once [Fe/H] is fixed through the F indices, but requires a grid of synthetic spectra computed for a range of [Mg/Fe] values or direct spectrum synthesis on a star by star basis. This shall be the object of our future work \section{Results on Sagittarius} The metallicity distribution of our 22 candidate Sagittarus members is shown in Figure 1. The peak at metallicity +0.5 reflects our choice to avoid extrapolation at metallicities higher than our most metal-rich grid point; when a star showed indices indicative of higher metallicity we assigned to it the value +0.5. The distribution appears to be distinctly tri-modal: there is a metal-rich population whose mean metallicity is yet undefined and there are two metal-poor populations which peak around $-1.2$ and $-0.5$ respectively. The existence of the two metal-poor populations is in substantial agreement with previous findings\cite{mara}. The presence of a metal-rich population is something new and unexpected. It is not yet clear whether the metal-rich stars belong to Sagittarius or are Bulge interlopers, although in the latter case their number is uncomfortably large and may require a revision of our understanding of Bulge kinematics. An independent study based on Keck High-Res spectra\cite{sme}, finds that 2 out of 7 stars, with radial velocity compatible with Sagittarius membership, are in fact metal-rich. Yet another photometric survey\cite{bel} suggests the possible existence of a RGB sequence considerably more metal-rich than 47 Tuc. Although we are at an early stage to draw definitive conclusions, circumstantial evidence is arising for a metal-rich population in Sagittarius. This does not fit with our present understanding of chemical evolution of dwarf spheroidals, however we should keep an open mind and stick to the observations. \section{Perspectives for the use of VLT} VLT will quite likely play a fundamental role in the study of Local Group Galaxies and Sagittarius in particular. With UVES it shall be possible to do a detailed line by line analysis for stars along the RGB to below the Horizontal Branch ($V_{HB}\approx 18.2$). The Turn-off $21\le V \le 21.5$ will still be out of reach of UVES though. New perspectives are opened by FORS. The resolution shall be at most 1/2 of that used by us at NTT, however the indices we have defined measure such strong features that they will be usable even at this lower resolution, as shown by simulations we carried out. Moreover the MOS slitlets of FORS have sharp edges and it will be possible to flux-calibrate the spectra, thus allowing to measure spectral indices in absolute flux. This will recover all the information contained the continuum. The spectral coverage (350 - 590 nm) will allow to use the Ca II H and K lines and probably other iron-related indices in the range 400 - 500 nm . Finally with VLT+FORS a S/N of about 100 can be reached in one hour for stars of $V\approx 18.5$, with this S/N our Montecarlo simulations predict the random error in metallicity determinations to be around 0.02 dex. The use of a 3m class telescope, such as NTT, for preliminary observations aimed at the determination of radial velocities has proved to be too time-consuming to be worth the effort. FORS will be able to provide good quality spectra from which radial velocities, metallicities and key abundance ratios, such as [Mg/Fe] and [Ca/Fe], can be determined. The real killer in the studies of Local Group galaxies shall be Giraffe fed by 130 fibres in MEDUSA mode. The highest resolution achieved by Giraffe (15000) is high enough to perform a classical line by line analysis. However the number of spectra we shall get (several hundreds per night) will preclude the use of this approach. We will have to find new ways to estimate abundances from these spectra which will be able to cope with the data flow provided by Flames+Giraffe. One possibility is suggested by our low-resolution experience: define spectral indices that can be related to abundances of particular elements and tabulate their value from synthetic spectra. The estimation of an abundance requires then only an interpolation in an appropriate table. Different indices will be appropriate for different temperature and luminosity regimes. Now is the time to investigate both theoretically and observationally which features will provide useful indices. For example some indices may be defined to measure essentially strong iron lines others essentially weak iron lines, the balance of the two will allow to determine microturbulence. The index approach is not the only possibility. Other very promising techniques such as autocorrelation and cross-correlation should be investigated. A whole range of methods could complement each other to extract information from the spectra, however if any particular star seems deserving of special attention, all the spectra will be nicely archived and one can go back and do the good old line by line analysis. \section{Acknowledgements} I wish to thank G. Marconi, P. Molaro and L. Pasquini for allowing me to present results of common work in advance of publication. This work is based on observations collected at the European Southern Observatory, Chile, ESO N$^\circ$ 57.E-0586. \addcontentsline{toc}{section}{References}
\section{Introduction} Quantum field theory is undoubtedly a very successful theory except for its mathematically rigorous foundation. The basic quantities of quantum field theory are quantum fields, generically denoted by $\varphi(x)$, which are operator-valued singular functions of $N$-dimensional space-time $x^\mu$. In the standard formulation, their properties are governed by the action $S$, which is an $N$-dimensional integral of the Lagrangian density ${\cal L}$ (a function of quantum fields at the same space-time point). Field equations and canonical (anti)commutation relations are derived from $S$ by the standard procedure. Equal-time (anti)commutation relations follow from the canonical ones. \par Field equations and equal-time (anti)commutation relations define the operator properties of quantum fields, that is, it is supposed to determine the operator algebra of quantum fields. Next, this operator algebra is represented in terms of state vectors, whose totality forms an infinite dimensional complex linear space equipped with (indefinite) inner product. This way of formulating the theory is operator-formalism approach. Evidently, this way of thinking is most natural as the formulation of quantum fields. Nevertheless, this approach has never been seriously considered as the standard approach to quantum field theory. The main reasons for this are as follows. First, it has been unknown how to find the solution in this approach. Second, it has been unknown how to deal with the divergence problem which arises from the singular nature of quantum fields. \par The present-day's standard approach is covariant perturbation theory. It is based on the action $S$, rather than field equations. Decomposing it artificially into its free part and its interaction part, one obtains all Green's functions explicitly in terms of Feynman diagrams. Furthermore, renormalization theory tells us how to deal with the divergence problem: divergent contributions are absorbed into counter terms. \par Path-integral formalism gives us the generating functional, $Z$, of the Green's functions. If path integral is defined as the infinite multiple integrals with respect to the expansion coefficients of quantum fields in terms of their free wave functions, $Z$ reproduces the perturbative expressions. If path integral is defined by using an expansion in terms of the functions other than free wave functions, it is not clear whether or not $Z$ yields the Green's functions which reproduce the perturbative ones (if expanded), but it is usually supposed that there exists an appropriate path-integral measure giving the expected results. Although we cannot accept this assumption without reservation,\footnote{This problem is particularly relevant in quantum gravity because gravity's free wave functions are rather artificial quantities. For example, the sum over all possible manifolds having a particular metric signature cannot be realized by a simple infinite multiple integral with respect to expansion coefficients because a definite metric signature requires to insert a product of $\theta$ functions of the determinant and principal minors of $g_{\mu\nu}$ into the path integral.} we do not strictly distinguish perturbative approach and path-integral one in the present paper. \par By taking the logarithm of $Z$, one obtains the generating functional, $W$, of connected Green's functions. Moreover, the effective action $\Gamma$, which is the generating functional of amputated proper Feynman diagrams, is obtained as the functional Legendre transform of $W$. The renormalization is neatly carried out in $\Gamma$ and therefore the anomaly problem is usually discussed also in $\Gamma$. It should be noted here that perturbation series is not unique because the decomposition of the action into its free part and its interaction part is generally altered by a nonlinear redefinition of fields. Accordingly, the effective action $\Gamma$ is a quantity which generally changes under the redefinition of fields. Thus, in the path-integral approach, neither renormalization procedure nor the anomaly problem are quite independent of the choice of quantum fields. This point becomes crucial when unphysical fields, whose natural definitions are not necessarily unique, play important roles as in quantum gravity. \par Now, we have recently succeeded in formulating the method of finding the solution in the operator-formalism approach.\cite{AN1} Our method is as follows. From the field equations and equal-time (anti)commutation relations, we explicitly construct all independent $N$-dimensional (anti)commutation relations, by expanding them, if necessary, into the power series with respect to the parameters involved. We then calculate independent $N$-dimensional multiple (anti)commutators. The representation of the field algebra in terms of state vectors is constructed by giving all $n$-point Wightman functions ($n=1,\,2,\,\ldots\,$), i.e., vacuum expectation values of simple products of $n$ quantum fields, $\varphi_1(x_1)\varphi_2(x_2)\cdots \varphi_n(x_n)$, so as to be consistent with the $(n-1)$ple (anti)commutators and with the energy positivity conditions.\footnote{The Wightman function is a boundary value of an analytic function of the variables $x_i{}^0-x_j{}^0$ $\ (i<j)$ from the lower half-planes.} Here, in contrast with the axiomatic field theory,\cite{SW} we need the Wightman functions involving composite fields, where a composite field is a product of fields at the same space-time point. When we set some of space-time points coincident in a higher-point Wightman function, we generally encounter divergent terms, which must be simply discarded in such a way that the resultant be independent of the ordering of the constituent fields of the composite field (``generalized normal product rule''). In this procedure, we do not introduce anything like a counter term. This is because a well-defined representation of the field algebra should be free of divergence; our standpoint is similar to that of the Lehmann-Symanzik-Zimmermann formalism,\cite{LSZ} in which they developed the renormalized perturbation theory without encountering any unrenormalized quantities. \par Of course, it is extremely difficult to carry out our way of finding the solution in realistic models. But, fortunately, we can explicitly construct the exact solutions in some two-dimensional models by our method.\cite{AN2,AN3,AN4,AN5,Ikeda} Our results are seen to be quite satisfactory, but we encounter an anomalous phenomenon, which we call ``field-equation anomaly'', in quantum-gravity models\cite{AN2,AN3,AN4,AN5} (but not in gauge-theory models\cite{Ikeda}): By construction, our Wightman functions are consistent with all two-dimensional (anti)commutators but not necessarily consistent with nonlinear field equations because there we encounter products of fields at the same space-time point. In any of the quantum-gravity models which we have exactly solved so far, one of field equations is {\it slightly\/} violated at the level of representation.\ifundefined{abstsize}\setcounter{footn}{0}\fi\footnote{The violation is {\it slight\/} in the sense that an anomaly-free equation can be obtained by differentiating the original field equation once or twice.} This is the field-equation anomaly. It is different from the conventional anomalies which arise in connection with particular symmetries. Rather, as clarified in our previous work,\cite{AN6} various conventional anomalies\cite{AN7} are systematically explained on the basis of the field-equation anomaly and their ambiguities are shown to be caused by the nonuniqueness of perturbation theory. In this sense, we regard the field-equation anomaly as a more fundamental concept. \par The purpose of the present paper is to make comparison between the solution obtained by the perturbative approach and the one obtained by the operator-formalism approach explicitly in the conformal-gauge two-dimensional quantum gravity. The exact solution of this model obtained previously\cite{AN4} can be written in terms of {\it tree diagrams only}. If the same were true also in the perturbative solution, no anomaly could be present. As is well known, however, this model has the conformal anomaly except for $D=26$, where $D$ denotes the number of the scalar fields which can be interpreted as the string coordinates. We trace the cause of this paradox and find that the perturbative approach induces some one-loop Feynman diagrams, which would not exist unless a nonzero contribution arose from a zero field. The cause of this strange phenomenon is found to be the use of T$^*$-product (covariantized T-product) of quantum fields\ifundefined{abstsize}\setcounter{footn}{1}\fi\footnote{T$^*$-product is a T-product modified in such a way tha \ifundefined{abstsize} \hbox{$\,\partial_1{}\!^0\hbox{T}^{\hskip-.5pt*}\hskip-.5pt\varphi_1(x_1) \cdots\varphi_n(x_n) \!=\!\hbox{T}^{\hskip-.5pt*}\hskip-.5pt\partial_1{}\!^0\varphi_1(x_1) \cdots\varphi_n(x_n)$.} \else {} $\partial_1{}^0\hbox{T}^*\varphi_1(x_1)\cdots\varphi_n(x_n) =\hbox{T}^*\partial_1{}^0\varphi_1(x_1)\cdots\varphi_n(x_n)$. \fi } in the perturbative or path-integral approach. More generally, in the present paper, we clarify that various anomalous behaviors of this model found in the perturbative approach are caused by the use of T$^*$-product. \par In the present paper, we compare the perturbative or path-integral approach with the operator-formalism approach in the conformal-gauge two-dimensional quantum gravity,\ifundefined{abstsize}\break\else{} \fi whose Lagrangian density is given in \S2. In \S3, \S4 and \S5, respectively, we discuss this model by the operator-formalism approach, by the perturbative approach and by the path-integral approach. In \S6, we criticize the so-called ``FP-ghost number current anomaly''. The final section is devoted to discussions. \ifundefined{abstsize} \newpage \fi \section{Preliminaries} Throughout the present paper, we consider the BRS formalism of the conformal-gauge two-dimensional quantum gravity, in which the conformal degree of freedom is already eliminated.\cite{Yang} Quantum fields are contravariant tensor-density gravitational field $\tg^{\mu\nu}$, FP ghost $c^\mu$, FP anti-ghost ${\skew3\bar c}\hspace*{1pt}_{\mu\nu}$, B field $\,\tilde{\! b}{}_{\mu\nu}$ and scalar fields $\phi_M\; (M=0,\,1,\,\ldots,\,D-1)$. Their BRS transforms are as follows: \begin{eqnarray} {\mbf\delta_*}\, \tg^{\mu\nu}&=& \tg^{\mu\sigma}\partial_\sigma c^\nu +\tg^{\nu\sigma}\partial_\sigma c^\mu -\partial_\sigma(\tg^{\mu\nu}c^\sigma), \\ {\mbf\delta_*}\, c^\mu &=&-c^\sigma\partial_\sigma c^\mu, \\ {\mbf\delta_*}\, {\skew3\bar c}\hspace*{1pt}_{\mu\nu} &=& i\,\tilde{\! b}{}_{\mu\nu}, \\ {\mbf\delta_*}\, \,\tilde{\! b}{}_{\mu\nu} &=&0,\\ {\mbf\delta_*}\, \phi_M &=& -c^\sigma\partial_\sigma\phi_M. \end{eqnarray} Since $\tg^{\mu\nu}$ has only two degrees of freedom because det$\tg^{\mu\nu} =-1$ , it is parametrized as \begin{eqnarray} \tg^{\mu\nu}&=&(\eta^{\mu\nu}+h^{\mu\nu})(1-\det{h^{\sigma\tau}})^{-1/2}, \label{eq:2,6} \end{eqnarray} where $h^{\mu\nu}$ is symmetric and traceless ($\eta_{\mu\nu}h^{\mu\nu}=0$). Correspondingly, ${\skew3\bar c}\hspace*{1pt}_{\mu\nu}$ and $\,\tilde{\! b}{}_{\mu\nu}$ are also symmetric and traceless. It is convenient to rewrite any traceless symmetric tensor $X_{\mu\nu}$ into a vector-like quantity $X^\lambda$ by \begin{eqnarray} X^\lambda &=& {1\over\sqrt{2}}\xi^{\lambda\mu\nu}X_{\mu\nu}, \label{eq:2,7} \end{eqnarray} where $\xi^{\lambda\mu\nu}=1$ for $\lambda+\mu+\nu=$even, $=0$ otherwise. According to \eqno(2,7), we introduce $h_\lambda$, ${\skew3\bar c}\hspace*{1pt}^\lambda$ and $\,\tilde{\! b}{}^\lambda$. \par The BRS-invariant action $S=\int d^2x\,{\cal L}$ is given by the Lagrangian density \begin{eqnarray} {\cal L}&=&{1\over\,2\,}\tg^{\mu\nu}\partial_\mu\phi_M\cdot\partial_\nu\phi^M -{1\over\,2\,}\,\tilde{\! b}{}^\lambda h_\lambda -{i\over\,2\,}{\skew3\bar c}\hspace*{1pt}^\lambda{\mbf\delta_*}\,(h_\lambda), \label{eq:2,8} \end{eqnarray} where $\tg^{\mu\nu}$ is given by \eqno(2,6), that is, \begin{eqnarray} {\cal L}&=&{\cal L}_0+\lag_{\hbox{\tiny I}}, \label{eq:2,9} \end{eqnarray} with \begin{eqnarray} {\cal L}_0&=& {1\over\,2\,}\eta^{\mu\nu}\partial_\mu\phi_M\cdot\partial_\nu\phi^M -{1\over\,2\,}\,\tilde{\! b}{}^\lambda h_\lambda -{i\over\sqrt{2}}\xi_{\lambda\mu\nu}{\skew3\bar c}\hspace*{1pt}^\lambda\partial^\mu c^\nu, \label{eq:2,10} \\ \lag_{\hbox{\tiny I}}&=& h_\lambda \bigg({1\over2\sqrt{2}}\xi^{\lambda\mu\nu}\partial_\mu\phi_M\cdot \partial_\nu\phi^M \if@preprint -{i\over\,2\,}\xi^{\lambda\mu\nu}\xi_{\rho\mu\sigma}{\skew3\bar c}\hspace*{1pt}^\rho \partial_\nu c^\sigma -{i\over\,2\,}\partial_\sigma{\skew3\bar c}\hspace*{1pt}^\lambda\cdot c^\sigma\bigg) +O(h^2). \label{eq:2,11} \hspace*{20pt} \else \!-\!{i\over\,2\,}\xi^{\lambda\mu\nu}\xi_{\rho\mu\sigma}{\skew3\bar c}\hspace*{1pt}^\rho \partial_\nu c^\sigma \!-\!{i\over\,2\,}\partial_\sigma{\skew3\bar c}\hspace*{1pt}^\lambda\cdot c^\sigma\bigg) \!+\!O(h^2). \label{eq:2,11} \hspace*{40pt} \fi \end{eqnarray} The higher-order terms $O(h^2)$ are unnecessary to be specified because they contribute neither to field equations nor to canonical (anti)commutation relations. Furthermore, they give no contribution to perturbation theory. Thus, we may discard them. \par It should be noted that the action is invariant under the FP-ghost conjugation\footnote{So far, this fact has been overlooked because the FP antighost was treated as a tensor.} \begin{eqnarray} &&c^\lambda \ \longrightarrow \ {\skew3\bar c}\hspace*{1pt}^\lambda, \nonumber \\ &&{\skew3\bar c}\hspace*{1pt}^\lambda \ \longrightarrow \ c^\lambda, \nonumber \\ &&\,\tilde{\! b}{}^\lambda \ \longrightarrow \ \,\tilde{\! b}{}^\lambda - i\xi^{\lambda\mu\nu}\xi_{\rho\mu\sigma}{\skew3\bar c}\hspace*{1pt}^\rho \partial_\nu c^\sigma - i\partial_\sigma{\skew3\bar c}\hspace*{1pt}^\lambda\cdot c^\sigma \nonumber \\ &&\hspace*{55pt} + i\xi^{\lambda\mu\nu}\xi_{\rho\mu\sigma}c^\rho \partial_\nu{\skew3\bar c}\hspace*{1pt}^\sigma + i\partial_\sigma c^\lambda\cdot {\skew3\bar c}\hspace*{1pt}^\sigma \ + \ O(h), \end{eqnarray} as in the de Donder-gauge case. \par Analysis can be much simplified by introducing light-cone coordinates $x^\pm=$ $(x^0\pm x^1)/\sqrt{2}$, because then $\xi_{\mu\nu\lambda}=0$ except $\xi_{+++}=\xi_{---}=\sqrt{2}$. From \eqno(2,10) and \eqno(2,11) ($O(h^2)$ is omitted), we have \begin{eqnarray} &&{\cal L}_0=\partial_+\phi_M\cdot\partial_-\phi^M +\Big[-{1\over\,2\,}\,\tilde{\! b}{}^+ h_+ -i{\skew3\bar c}\hspace*{1pt}^+\partial_-c^+ + \quad (\ +\ \longleftrightarrow\ -\ )\Big], \label{eq:2,13} \\ &&\lag_{\hbox{\tiny I}}=h_+\Big[{1\over\,2\,}\partial_+\phi_M\cdot\partial_+\phi^M - i{\skew3\bar c}\hspace*{1pt}^+\partial_+c^+ -{i\over\,2\,}\Big(\partial_+{\skew3\bar c}\hspace*{1pt}^+\cdot c^+ +\partial_-{\skew3\bar c}\hspace*{1pt}^+\cdot c^- \Big) \Big] \ \nonumber \\ &&\hspace*{50pt} + \quad (\ +\ \longleftrightarrow\ -\ ). \label{eq:2,14} \end{eqnarray} In subsequent sections, we start with \eqno(2,9) together with \eqno(2,13) and \eqno(2,14). For later convenience, we introduce the following notation. \begin{eqnarray} \skew3\widetilde{\calT}^\pm &\equiv& \partial_\pm\phi_M\cdot\partial_\pm\phi^M -2i{\skew3\bar c}\hspace*{1pt}^\pm\partial_\pm c^\pm -i\partial_\pm{\skew3\bar c}\hspace*{1pt}^\pm\cdot c^\pm, \\ \skew3\widetilde{\tcalT}^\pm &\equiv& \skew3\widetilde{\calT}^\pm - i\partial_\mp{\skew3\bar c}\hspace*{1pt}^\pm\cdot c^\mp. \end{eqnarray} Then we have \begin{eqnarray} {\partial \lag_{\hbox{\tiny I}} \over \partial h_\pm} &=& {1\over\,2\,}\skew3\widetilde{\tcalT}^\pm. \end{eqnarray} \par The Noether currents of the BRS invariance and the FP-ghost number conservation are given by \begin{eqnarray} j_b{}^\mp &=& -c^\pm \partial_\pm \phi_M\cdot \partial_\pm \phi^M -i{\skew3\bar c}\hspace*{1pt}^\pm c^\pm \partial_\pm c^\pm, \label{eq:2,18} \\ j_c{}^\mp &=& -i{\skew3\bar c}\hspace*{1pt}^\pm c^\pm, \end{eqnarray} respectively. They are of course conserved. \ifundefined{abstsize} \newpage \fi \section{Operator-formalism Approach} For the sake of comparison, we briefly review our previous results of the exact solution obtained by the operator-formalism approach.\cite{AN4} \par The field equations are as follows: \begin{eqnarray} &&h_\pm=0, \label{eq:3,1} \\ &&\,\tilde{\! b}{}^\pm=\skew3\widetilde{\calT}^\pm, \label{eq:3,2} \\ &&\partial_\mp X^\pm=0 \quad \hbox{for } X^\pm=c^\pm,\ {\skew3\bar c}\hspace*{1pt}^\pm,\ \,\tilde{\! b}{}^\pm, \label{eq:3,3} \\ &&\partial_+\partial_-\phi_M=0, \label{eq:3,4} \end{eqnarray} where \eqno(3,3) for $X^\pm=\,\tilde{\! b}{}^\pm$ is derived by differentiating \eqno(3,2). Note that $c^\pm,\ {\skew3\bar c}\hspace*{1pt}^\pm,\ \,\tilde{\! b}{}^\pm$ and $\partial_\pm\phi_M$ are the functions of a single variable $x^\pm$ only. \par Canonical quantization is carried out by taking $\phi_M$ and $c^\pm$ only as the canonical variables. Since they are free fields, their nonvanishing % two-dimensional (anti)\-commutators are easily obtained; we have \begin{eqnarray} &&[\partial_\pm\phi_M(x),\;\phi^N(y)] =-{i\over\,2\,}\delta_M{}^N \delta(x^\pm-y^\pm), \label{eq:3,5} \\ &&\{c^\pm(x),\;{\skew3\bar c}\hspace*{1pt}^\pm(y)\}=-\delta(x^\pm-y^\pm). \end{eqnarray} The commutation relations involving $\,\tilde{\! b}{}^\pm$ are calculated by using \eqno(3,2): \begin{eqnarray} {[} \,\tilde{\! b}{}^\pm(x),\;\phi_M(y) ] &=&-i\partial_\pm\phi_M(x)\cdot\delta(x^\pm-y^\pm), \\ {[} \,\tilde{\! b}{}^\pm(x),\;c^\pm(y) ] &=&-ic^\pm(x)\delta'(x^\pm-y^\pm) -2i\partial_\pm c^\pm(x)\cdot\delta(x^\pm-y^\pm), \\ {[} \,\tilde{\! b}{}^\pm(x),\;{\skew3\bar c}\hspace*{1pt}^\pm(y) ] &=& 2i{\skew3\bar c}\hspace*{1pt}^\pm(x)\delta'(x^\pm-y^\pm) +i\partial_\pm{\skew3\bar c}\hspace*{1pt}^\pm(x)\cdot\delta(x^\pm-y^\pm), \label{eq:3,9} \\ {[} \,\tilde{\! b}{}^\pm(x),\;\,\tilde{\! b}{}^\pm(y) ] &=&i[\,\tilde{\! b}{}^\pm(x)+\,\tilde{\! b}{}^\pm(y)]\delta'(x^\pm-y^\pm). \label{eq:3,10} \end{eqnarray} Evidently, \eqno(3,10) is the BRS transform of \eqno(3,9). \par Since no new operators are encountered in the right-hand sides of \eqno(3,5)-\eqno(3,10), we can easily calculate all multiple (anti)commutators explicitly. We then construct all truncated\footnote{Truncation means to drop the contributions from vacuum intermediate states. The {\it truncated\/} Wightman function corresponds to the {\it connected\/} Green's function. In the present model, the distinction between truncated and nontruncated appears only for $n\geqq4$.} Wightman functions so as to be consistent with all multiple (anti)commutators under the energy positivity condition. \par The 1-point functions are, in principle, completely arbitrary. But we set all of them equal to zero because we should not deliberately violate any of \eqno(3,1), FP-ghost number conservation, BRS invariance and $O(D)$ symmetry. \par The nonvanishing {\it truncated\/} $n$-point Wightman functions are those which consist of $(n-2)$ $\,\tilde{\! b}{}^\pm$'s and of either $c^\pm$ and ${\skew3\bar c}\hspace*{1pt}^\pm$ or two $\phi_M$'s. Diagrammatically, they are represented by tree diagrams. Although we have explicitly constructed all of them,\cite{AN4} we here quote 2-point and 3-point ones only. \par Nonvanishing 2-point Wightman functions are \begin{eqnarray} &&\wightman{\phi_M(x_1)\phi^N(x_2)} =\delta_M{}^N D^{\hbox{\tiny (+)}}(x_1-x_2),\\ &&\wightman{c^\pm(x_1){\skew3\bar c}\hspace*{1pt}^\pm(x_2)} =\wightman{{\skew3\bar c}\hspace*{1pt}^\pm(x_1)c^\pm(x_2)} =-2i\partial_\pmD^{\hbox{\tiny (+)}}(x_1-x_2), \end{eqnarray} where\footnote{$D^{\hbox{\tiny (+)}}(x)$ itself is infrared divergent and therefore requires the introduction of infrared \linebreak[3] cutoff.~\cite{Nakanishi}} \begin{eqnarray} && \partial_\pmD^{\hbox{\tiny (+)}}(x)\equiv -{1\over4\pi}\cdot{1\over x^\pm-i0}. \end{eqnarray} Nonvanishing 3-point ones are \begin{eqnarray} &&\wightman{\phi_M(x_1)\,\tilde{\! b}{}^\pm(x_2)\phi^N(x_3)} =-2\delta_M{}^N \partial_\pmD^{\hbox{\tiny (+)}}(x_1-x_2)\cdot\partial_\pmD^{\hbox{\tiny (+)}}(x_2-x_3), \hspace*{30pt} \label{eq:3,14} \\ &&\wightman{c^\pm(x_1)\,\tilde{\! b}{}^\pm(x_2){\skew3\bar c}\hspace*{1pt}^\pm(x_3)} =8i\partial_\pmD^{\hbox{\tiny (+)}}(x_1-x_2)\cdot\partial_\pm{}^2D^{\hbox{\tiny (+)}}(x_2-x_3) \nonumber\\ &&\hspace*{120pt} -4i\partial_\pm{}^2D^{\hbox{\tiny (+)}}(x_1-x_2)\cdot\partial_\pmD^{\hbox{\tiny (+)}}(x_2-x_3) \label{eq:3,15} \end{eqnarray} and their permutated ones, whose expressions are obtained from the above by changing some of $D^{\hbox{\tiny (+)}}(x)$'s into $-[D^{\hbox{\tiny (+)}}(x)]^*$ so as to become consistent with the energy-positivity condition (and by changing the overall sign if the order of $c^\pm$ and ${\skew3\bar c}\hspace*{1pt}^\pm$ is reversed). \par Our system of Wightman functions is, of course, consistent with the field algebra defined by \eqno(3,5)-\eqno(3,10). It is also consistent with the BRS invariance and the FP-ghost number conservation. It should be noted that we need the use of the generalized normal-product rule to check the BRS invariance. Our system of Wightman functions is also consistent with all {\it linear\/} field equations \eqno(3,1), \eqno(3,3) and \eqno(3,4), but {\it not\/} with the {\it nonlinear\/} field equation \eqno(3,2). Indeed, by using the generalized normal-product rule, we can show that \begin{eqnarray} \wightman{\,\tilde{\! b}{}^\pm(x_1)\,\tilde{\! b}{}^\pm(x_2)}&=&0, \\ \wightman{\,\tilde{\! b}{}^\pm(x_1)\skew3\widetilde{\calT}^\pm(x_2)} &=&\wightman{\skew3\widetilde{\calT}^\pm(x_1)\skew3\widetilde{\calT}^\pm(x_2)} \nonumber\\ &=&2(D-26)[\partial_\pm{}^2D^{\hbox{\tiny (+)}}(x_1-x_2)]^2 \end{eqnarray} in contradiction with \eqno(3,2). Thus the field equation \eqno(3,2), {\it modulo\/} \eqno(3,3) for $X^\pm=\,\tilde{\! b}{}^\pm$, is violated at the level of the representation in terms of state vectors. We call this matter ``field-equation anomaly''. This phenomenon is encountered also in several two-dimensional quantum-gravity models.\cite{AN2,AN3,AN4,AN5} \par The BRS Noether current \eqno(2,18) can be rewritten as \begin{eqnarray} &&j_b{}^\mp=j_b'{}^\mp + (\,\tilde{\! b}{}^\pm-\skew3\widetilde{\calT}^\pm)c^\pm, \\ \noalign{\noindent with} &&j_b'{}^\mp \equiv -\,\tilde{\! b}{}^\pm c^\pm + i{\skew3\bar c}\hspace*{1pt}^\pm c^\pm \partial_\pm c^\pm. \end{eqnarray} At the operator level, $j_b'{}^\mp$ strictly equals $j_b{}^\mp$. But this equality no longer holds at the representation level because of the appearance of the field-equation anomaly. Indeed, $j_b{}^\mp$ is anomalous for $D\not=26$, while $j_b'{}^\mp$ is free of anomaly for any value of $D$.\cite{AN4,AN5} On the other hand, $j_c{}^\mp$ is free of anomaly without making any modification. \section{Perturbative approach} The perturbative approach is so familiar to everybody that no explanation about it is necessary. Nevertheless, when compared with the operator-formalism approach, the perturbative approach is seen to yield some surprising results. \par The Lagrangian density ${\cal L}$ is decomposed into the free one ${\cal L}_0$, which is quadratic with respect to the fields adopted as the basic ones, and the remainder $\lag_{\hbox{\tiny I}}$, called the interaction Lagrangian density. \par The Feynman propagators are obtained by taking the inverse of the differential operator sandwiched by the fields in ${\cal L}_0$. Thus, from \eqno(2,13), we have the following nonvanishing Feynman propagators:\footnote{The subscript 0 indicates that the propagators are free ones.} \begin{eqnarray} &&\taustar{\phi_M(x_1)\phi^N(x_2)}_0=\delta_M{}^ND_{\hbox{\tiny F}}(x_1-x_2), \label{eq:4,1}\\ &&\taustar{\,\tilde{\! b}{}^\pm(x_1)h_\pm(x_2)}_0=-2i\delta^2(x_1-x_2), \label{eq:4,2}\\ &&\taustar{{\skew3\bar c}\hspace*{1pt}^\pm(x_1)c^\pm(x_2)}_0=-2i\partial_\pmD_{\hbox{\tiny F}}(x_1-x_2), \label{eq:4,3} \end{eqnarray} where \begin{eqnarray} \partial_\pmD_{\hbox{\tiny F}}(x)&\equiv&\theta(x^0)\partial_\pmD^{\hbox{\tiny (+)}}(x) +\theta(-x^0)\partial_\pm[D^{\hbox{\tiny (+)}}(x)]^* \nonumber\\ &=&-{1\over 4\pi}\bigg[ {\theta(x^\mp+x^\pm)\over x^\pm-i0} +{\theta(-x^\mp-x^\pm)\over x^\pm+i0} \bigg]. \label{eq:4,4} \end{eqnarray} \par It is quite remarkable that $\taustar{\,\tilde{\! b}{}^\pm\,h_\pm}_0$ is {\it nonvanishing\/} in spite of the fact that $h_\pm$ is a {\it zero\/} operator as is seen from \eqno(3,1). In contrast with the Wightman functions, the T$^*$-product does not respect the validity of the field equations. As is seen from \eqno(4,3), it is also inadmissible to set $\partial_\mp c^\pm=\partial_\mp{\skew3\bar c}\hspace*{1pt}^\pm=0$ in the perturbative approach. Hence we cannot discard the terms involving $\partial_\mp{\skew3\bar c}\hspace*{1pt}^\pm$ in $\lag_{\hbox{\tiny I}}$, that is, we have to distinguish $\skew3\widetilde{\tcalT}^\pm$ from $\skew3\widetilde{\calT}^\pm$. Thus the beautiful result of the operator formalism that $c^\pm$, ${\skew3\bar c}\hspace*{1pt}^\pm$, $\,\tilde{\! b}{}^\pm$ and $\partial_\pm\phi_M$ are irrelevant to $x^\mp$ is no longer valid in the perturbative approach. This fact makes the perturbative calculation complicated and sometimes misleading, as we shall see later. \par By using $\lag_{\hbox{\tiny I}}$ given by \eqno(2,14), we can easily calculate the $n$-point Green's functions. For example, we have \begin{eqnarray} &&\taustar{\phi_M(x_1)\,\tilde{\! b}{}^\pm(x_2)\phi^N(x_3)} =-2\delta_M{}^N\partial_\pmD_{\hbox{\tiny F}}(x_1-x_2)\cdot\partial_\pmD_{\hbox{\tiny F}}(x_2-x_3), \label{eq:4,5} \\ &&\taustar{c^\pm(x_1)\,\tilde{\! b}{}^\pm(x_2){\skew3\bar c}\hspace*{1pt}^\pm(x_3)} =8i\partial_\pmD_{\hbox{\tiny F}}(x_1-x_2)\cdot\partial_\pm{}^2D_{\hbox{\tiny F}}(x_2-x_3) \nonumber \\ &&\hspace*{135pt} -4i\partial_\pm{}^2D_{\hbox{\tiny F}}(x_1-x_2)\cdot\partial_\pmD_{\hbox{\tiny F}}(x_2-x_3), \label{eq:4,6}\\ &&\taustar{c^\pm(x_1)\,\tilde{\! b}{}^\pm(x_2){\skew3\bar c}\hspace*{1pt}^\mp(x_3)} =-2\delta^2(x_1-x_2)\partial_\mpD_{\hbox{\tiny F}}(x_2-x_3). \label{eq:4,7} \end{eqnarray} Evidently, \eqno(4,5) and \eqno(4,6) correspond to \eqno(3,14) and to \eqno(3,15), respectively. However, \eqno(4,7) is a result peculiar to the T$^*$-product. This result is seen to be consistent with the Ward-Takahashi identity \begin{eqnarray} &&\taustar{{\mbf\delta_*}\,(c^\pm(x_1){\skew3\bar c}\hspace*{1pt}^\pm(x_2){\skew3\bar c}\hspace*{1pt}^\mp(x_3))}=0, \end{eqnarray} because the second term of ${\mbf\delta_*}\, c^\pm=-c^\pm\partial_\pm c^\pm-c^\mp\partial_\mp c^\pm$ contributes. \par Now, we come to the crucial point. In sharp contrast with the case of the operator-formalism approach, the perturbative approach yields quite a nontrivial result for the $n$-point Green's function consisting of B-fields only. Indeed, its connected part is given by a sum over one-loop Feynman diagrams. For example, we consider $\taustar{\,\tilde{\! b}{}^\lambda(x_1)\,\tilde{\! b}{}^\rho(x_2)}$. Because of the nonvanishing of \eqno(4,2), the second-order perturbation term yields \begin{eqnarray} &&\taustar{\,\tilde{\! b}{}^\lambda(x_1)\,\tilde{\! b}{}^\rho(x_2)} =\taustar{\skew3\widetilde{\tcalT}^\lambda(x_1)\skew3\widetilde{\tcalT}^\rho(x_2)}_0. \label{eq:4,9} \end{eqnarray} Therefore, we have \begin{eqnarray} &&\taustar{\,\tilde{\! b}{}^\pm(x_1)\,\tilde{\! b}{}^\pm(x_2)} =2(D-26)[\partial_\pm{}^2D_{\hbox{\tiny F}}(x_1-x_2)]^2, \label{eq:4,10} \\ &&\taustar{\,\tilde{\! b}{}^\pm(x_1)\,\tilde{\! b}{}^\mp(x_2)}=-{D-2\over2}[\delta^2(x_1-x_2)]^2. \label{eq:4,11} \end{eqnarray} They are divergent and therefore require the introduction of counter terms. Note that the use of the T$^*$-product is responsible for the appearance of these divergences. \par The nonvanishing of the Green's functions consisting of B-fields only implies the violation of the BRS invariance. In the de Donder gauge case, Takahashi\cite{Takahashi} proposed to convert the violation of the BRS invariance for $D\not=26$ in the two-point B-field Green's function into the conformal anomaly.\footnote{He made no mention about how to remove the BRS violation in the {\it higher-point\/} functions.} We apply his line of thought to the present model. In addition to \eqno(4,10) and \eqno(4,11), we must take it into account the following exact two-point Green's functions: \begin{eqnarray} &&\taustar{h_\lambda(x_1)h_\rho(x_2)}=0, \\ &&\taustar{\,\tilde{\! b}{}^\lambda(x_1)h_\rho(x_2)} =-2i\delta_\rho{}^\lambda\delta^2(x_1-x_2). \label{eq:4,13} \end{eqnarray} The two-point functions of the effective action $\Gamma$ is obtained by taking the matrix inverse of \eqno(4,10)-\eqno(4,13). Accordingly, we have \begin{eqnarray} \Gamma&=&\int d^2x_1\int d^2x_2\bigg[ -{1\over\,2\,}\delta^2(x_1-x_2)\,\tilde{\! b}{}^\lambda(x_1)h_\lambda(x_2) \nonumber\\ && \qquad +{D-26\over2}i\sum_{\alpha=\pm} ([\partial_\alpha{}^2D_{\hbox{\tiny F}}(x_1-x_2)]^2)_{\hbox{\tiny R}} h_\alpha(x_1)h_\alpha(x_2) + \ \cdots\ \bigg], \label{eq:4,14} \ifundefined{abstsize} \else \hspace*{30pt}\fi \end{eqnarray} where a subscript R indicates regularization. The BRS-violating term in \eqno(4,14) is converted into the conformal-anomaly term by adding the conformal degree of freedom. We do not work out this procedure in detail because it is not our aim to do so. \par The important point is the violation of the BRS invariance in the B-field Green's functions. In the de Donder gauge case, the BRS violation has arisen by applying the dimensional regularization only to internal lines but {\it not to external lines}.\cite{AN6,AN8} In the present model, external lines are absent because \eqno(4,2) is local. Instead, as is seen from \eqno(4,9), perturbative approach makes use of the field-equation anomaly without being aware of this fact. \section{Path-integral approach} The path integral directly gives us the generating function of the Green's functions, that is, it deals with the T$^*$-product quantities only. The path integral $Z$ is formally expressed as \begin{eqnarray} &&Z(J)=\int \big(\prod_i{\cal D}\varphi_i\big)\, \exp\, i\!\int d^N\!{}x({\cal L}+\sum_iJ_i\varphi_i) \end{eqnarray} with $Z(0)=1$, where $J_i$ denotes the source function corresponding to the field $\varphi_i$. \par It is possible to derive the path-integral formula [corrected by the Lee-Yang term proportional to $\delta^N(0)$] from the canonical operator formalism.\cite{Buchbinder} In this sense, the path-integral formalism can be regarded as the one equivalent to the operator formalism. But, one should note that, in this derivation, one must use the {\it field equations at the representation level}. This fact implies that the path-integral formalism cannot take care of the existence of the field-equation anomaly. \par From the successful experience of discussing the anomaly problem in gauge theories, it has been customary to believe that any anomaly always arises from the non-invariance of the path-integral measure under the symmetry which leaves the action $S$ invariant. But we point out that anomalies can arise also from the field-equation anomaly which is beyond the scope of the path-integral formalism. \par Let $F(\varphi)$ be an arbitrary function of $\varphi_i$'s. The path-integral measure is supposed to be invariant under the functional translation $\varphi_i \longrightarrow \varphi_i +\delta\varphi_i$. Hence, by considering a variation of a field $\varphi_i$ in $F(i^{-1}\partial/\partial J)Z|_{J=0}$, we obtain \begin{eqnarray} &&i\taustar{F(\varphi){\delta S\over \delta \varphi_i}} +\taustar{{\delta F(\varphi)\over \delta \varphi_i}} =0. \label{eq:5,2} \end{eqnarray} This equation corresponds to the field equation $\delta S/\delta \varphi_i=0$ of the operator formalism. The second term of \eqno(5,2) is a field-equation violating term due to the use of the T$^*$-product. One should never confuse it with the field-equation anomaly. For example, in the conformal-gauge two-dimensional quantum gravity, \eqno(4,14) is reproduced from \eqno(5,2) by setting $F=\,\tilde{\! b}{}^\lambda$ and $\varphi_i=\,\tilde{\! b}{}^\rho$. Likewise, if we set $F=\,\tilde{\! b}{}^\lambda$ and $\varphi_i=h_\rho$ in \eqno(5,2), we obtain \begin{eqnarray} &&\taustar{\,\tilde{\! b}{}^\lambda(\,\tilde{\! b}{}^\rho-\skew3\widetilde{\tcalT}^\rho)}=0, \label{eq:5,3} \end{eqnarray} that is, we do not encounter the field-equation anomaly. Instead, as is shown in the perturbative approach, \eqno(5,3) induces the violation of the BRS invariance in the path-integral approach. \par Historically, the anomaly problem in the conformal-gauge two-dimensional quantum gravity was discussed first by Fujikawa\cite{Fujikawa} in the path-integral formalism. His formulation is not, however, understandable in the framework stated in \S4. \par First, he takes all {\it three\/} degrees of freedom of $g_{\mu\nu}$ as path-integration variables; nevertheless, each of his ghost fields has only {\it two\/} degrees of freedom. The extra one degree of freedom is the conformal one, denoted by $\rho$, is {\it not allowed\/}\footnote{If one carries out the integration over $\rho$ after introducing the conformal ghosts, then it becomes impossible to work out his analysis.} to be integrated (until the Liouville action is derived) in spite of the fact that it is an independent path-integration variable. \par Second, by introducing tilde fields $\tilde\varphi{}_i=\rho^{n_i}\varphi_i$, $n_i$ being a certain fractional number, he claims that the path-integral measure becomes BRS invariant if it is expressed in terms of the tilde fields. He then derives the Liouville action expressed in terms of $\rho$ alone by calculating the variation of the path-integral measure under the conformal transformation. That is, according to his theory, the conformal anomaly is directly obtained {\it without passing through the BRS anomaly\/} in contradiction to the consideration presented in \S4. \par We are thus unable to reproduce his analysis in terms of the explicit solution. \section{FP-ghost number current anomaly} As we emphasized previously,\cite{AN4} there is no FP-ghost number anomaly in the conformal-gauge two-dimensional quantum gravity: The exact solution is completely consistent with the FP-ghost number conservation. The conservation of the FP-ghost number current $j_c{}^\mu$ is a simple consequence of the fact that $c^\pm(x)$ and ${\skew3\bar c}\hspace*{1pt}^\pm(x)$ are independent of $x^\mp$. This property is never violated at the representation level. Nevertheless, many authors have claimed that the FP-ghost number current has anomaly. The reasons for the occurrence of this belief are its correspondence to the Riemann-Roch theorem and the field-equation-violating property of the T$^*$-product. \par Fujikawa\cite{Fujikawa} was the first to claim the existence of the FP-ghost number current anomaly. He derived it by making the FP-ghost number transformation\footnote{This must be done in the original tilde-variable expression. No anomaly is derived by the naive calculation if this transformation is made in the equivalent expression having the Liouville action explicitly.} in his path-integral formalism described at the end of \S5. His result is written as \begin{eqnarray} &&\partial_\lambda\langle\!\langle\,j_c{}^\lambda\,\rangle\!\rangle ={3\over 4\pi}\langle\!\langle\, \partial^2 \log\rho \,\rangle\!\rangle, \label{eq:6,1} \end{eqnarray} where we denote the path integration by $\langle\!\langle\;\cdots\;\rangle\!\rangle$. If the degrees of the reparametrization freedom is suppressed, one may write $-\partial^2\log\rho=\sqrt{g}R$ (Euclidean metric is used). Here we must note that Fujikawa's theory is formulated in the {\it flat\/} background metric and that $\rho$ is the path-integration variable. \par Shortly later, Friedan, Martinec and Shenker,\cite{FMS} who formulated conformal field theory, quoted \eqno(6,1) in the disguised form. They consider a completely {\it curved\/} background metric $\hat{g}{}_{\mu\nu}$. The right-hand side of their equation is const.$\sqrt{\hat{g}{}}\hat{R}$, {\it a function of\/} $\hat{g}{}_{\mu\nu}$, which is nothing but the quantity required by the Riemann-Roch theorem under the prerequisite of the conformal covariance. It may be certainly {\it analogous\/} to \eqno(6,1), but we cannot find any {\it logical connection\/} between them.\footnote{According to Fujikawa (private communication), the possible existence of the zero points of $\rho$ can take care of the effect of the topological number $\int d^2x\,\sqrt{\hat{g}{}} \hat{R}$. We do not see how this idea can be formulated.} \par In the perturbative approach, the Friedan-Martinec-Shenker version of \eqno(6,1) is interpreted, through the consideration based on the effective action, as the matter that\cite{Dusedau,KR} \begin{eqnarray} &&J^\lambda{}_{\mu\nu}\equiv\taustar{j_c{}^\lambda {\delta S\over \delta \hat{g}{}^{\mu\nu}}}|_{\hat{g}{}^{\mu\nu}=\eta^{\mu\nu}} \label{eq:6,2} \end{eqnarray} has a nonvanishing nonlocal term, where $\hat{g}{}_{\mu\nu}$ is a background metric introduced in such a way that the gauge-fixing plus FP-ghost Lagrangian density becomes background covariant. Note in \eqno(6,2) that the background metric is taken to be {\it flat\/} in the Feynman-diagram calculation. \par In the conformal-gauge case, $J^\lambda{}_{\mu\nu}$ is essentially equal to \begin{eqnarray} &&\taustar{j_c{}^\pm(x_1)\skew3\widetilde{\calT}^\pm(x_2)} =-12\partial_\pmD_{\hbox{\tiny F}}(x_1-x_2)\cdot\partial_\pm{}^2D_{\hbox{\tiny F}}(x_1-x_2). \label{eq:6,3} \end{eqnarray} It is in this sense that the FP-ghost number current anomaly is claimed to be obtained in the perturbative approach. It should be noted, however, that {\it if the T$^*$-product is not taken, that is, if $D_{\hbox{\tiny F}}$ is replaced by $D^{\hbox{\tiny (+)}}$, \eqno(6,3) becomes consistent with\/} $\partial_\lambda j_c{}^\lambda=0$. \par D\"usedau\cite{Dusedau} found that $J^\lambda{}_{\mu\nu}$ has no nonlocal term in the de Donder-gauge case. Kraemmer and Rebhan\cite{KR} discussed the gauge dependence of $J^\lambda{}_{\mu\nu}$ and claimed that the gauge independence can be recovered if one adds a contribution from the ``Lagrange-multiplier (or B-field) current''.\footnote{Although we cannot regard their proof as adequate, their claim itself can be verified by explicit calculation.\cite{AN9}} But the relation between this fact and the Riemann-Roch theorem was not discussed. \par Recently, Takahashi\cite{Takahashi} has reconsidered D\"usedau's analysis from his perturbative approach described in \S4. In discussing $j_c{}^\lambda$, he regards the quantum gravitational field as the background metric, just as Fujikawa did. Rederiving D\"usedau's result, he asserts that the vanishing of the FP-ghost number current anomaly can be explained by the existence of the FP-ghost conjugation invariance of the de Donder gauge two-dimensional quantum gravity. One should note, however, that, as pointed out in \S2, the {\it FP-ghost conjugation invariance exists also in the conformal-gauge case}. Therefore, his standpoint would imply the absence of the FP-ghost number current anomaly also in the conformal-gauge case. \par Finally, we note that the nonexistence of the FP-ghost number current anomaly can be shown even if the gauge-fixing background metric is a nonflat one given by $\hat\rho(x)\eta_{\mu\nu}$, where $\hat\rho{}^{-1}$ {\it is assumed to exist}. In this case, \eqno(2,6) is replaced by \begin{eqnarray} &&\tg^{\mu\nu}=(\eta^{\mu\nu}+\hat\rho^{-1} h^{\mu\nu}) (1-\hat\rho{}^{-2}\det{h^{\sigma\tau}})^{-1/2}. \end{eqnarray} The Lagrangian density of this case is obtained from \eqno(2,8) by simply replacing $h_\lambda$ by $\hat\rho^{-1}h_\lambda$. Since ${\mbf\delta_*}\,(\hat\rho^{-1}h_\lambda)=\hat\rho^{-1}{\mbf\delta_*}\,(h_\lambda)$, we can absorb the factor $\hat\rho^{-1}$ into $\,\tilde{\! b}{}^\lambda$ and ${\skew3\bar c}\hspace*{1pt}^\lambda$ by redefining them. Thus the ghost part of the Lagrangian density of the nonflat case becomes completely the same as that of the flat case. Thus nothing new can happen about the FP-ghost number current. \section{Discussion} Nowadays, the path-integral approach and the counter-term business have become so fashionable that many physicists preclude the consideration based on other approaches from the outset. Certainly, the path-integral approach is convenient and successful in gauge theories, but we wish to emphasize that the same is not necessarily true in quantum gravity. \par The path-integral formalism directly deals with the solution at the representation level. Accordingly, if one adopts the path-integral approach, one can no longer perceive what happens in the transition from the operator level to the representation level. Indeed, one cannot describe the existence of the field-equation anomaly in the path-integral approach. \par The quantities describable by the path-integral formalism are those which can be written in terms of the T$^*$-product. The T$^*$-product is certainly a very convenient notion because we need not take care of the ordering problem even for the timelikely separated field operators. On the other hand, as emphasized in the present paper, the T$^*$-product has a demerit of violating the field equations explicitly. As demonstrated in the present paper, this fact induces unpleasant complications and misleading expressions. Furthermore, since the T$^*$-product contains $\theta$-functions, the Green's function is more singular than the corresponding Wightman functions, that is, some singularities found in the perturbative or path-integral approach may be superficial. When this fact is combined with the counter-term business, one is led to introducing counter terms which are purely of the T$^*$-product origin. In the present paper, we have demonstrated that ``anomalies'' also can be of the T$^*$-product origin. \par We hope that more physicists reinvestigate the anomaly problem in quantum gravity without adhering to the path-integral approach. \section*{Acknowledgements} One of the present authors (N. N.) would like to express his sincere thanks to Professor K. Fujikawa and Dr. H. Kanno for the discussions concerning the FP-ghost number current anomaly. \ifundefined{abstsize} \newpage \fi
\section{Introduction} Some Calabi-Yau three-folds(=CY 3-folds) with base, \relax {\bf \rm C}{\rm I\kern-.18em P}$^1(1,s)$ and fiber, \newline K3$= \relax {\bf \rm C}{\rm I\kern-.18em P}^3(u_1, u_2, u_3, u_4)[d]$ are represented in hypersurface in weighted projective 4-space, \relax {\bf \rm C}{\rm I\kern-.18em P}$^4$. \begin{equation} \CY3 {\rm -fold} = \relax {\bf \rm C}{\rm I\kern-.18em P}^4(u_1, s u_1, (s+1)u_2, (s+1)u_3, (s+1)u_4))[(s+1)d], \end {equation} \noindent with $ d= \sum_{i=1}^4 u_i$. \par The type IIA string on CY3-folds which have K3 fiber and T$^2$ fiber with at least one section is dual to the heterotic string on K3 $\times$ T$^2$ as pointed out in~\cite{vafa1,vafa2}. Much research has been explored on CY3-folds with $s=1$ and $u_1$=1 cases and F$_0$ based case~\cite {vafa1,vafa2,aldazabal,bershadsky,candelas1,candelas2,candelas3}. $s=1$ and $u_1$=1 cases are given by the extremal transitions or by the conifold transitions from F$_0$ based CY3-folds \cite{candelas1,candelas2,candelas3}. Thus far, CY3-folds which have been studied are constructed by using dual polyhedra. Some of them may be related to the hypersurface representations of eq. (1) with $s \geq 2$. However, identification between them has some ambiguities and is complicated~\cite {candelas4,candelas0,skarke,avram}. Therefore, the relation between these results and the duality between the type IIA string on CY3-folds with $s \geq 2$ to heterotic string on K3 $\times$ T$^2$ is not clear yet. There have been three types of web sequences of heterotic - type IIA string duality from the terminal CY3-fold in A series \cite{aldazabal,candelas1,candelas2}. (iii) in \cite{candelas2} may be the subset of (i)$^\dag$ in \cite{candelas3}, though the properties of dual polyhedron are slightly different \footnote{ The difference between dual polyhedron of (i)$^\dag $ and (iii) is as follows \cite{candelas1,candelas2,candelas3}. The dual polyhedra of case (i)$^\dag$ have the modified dual polyhedra of K3 fiber. For the case (iii) in A series, the dual polyhedron of K3 fiber is not modified. The highest point in the additional points is represented by the weights of the K3 fiber of the terminal A series in this base. One point such as $(0,\ast,\ast,\ast)$ is also represented by the part of the weight of the terminal K3 fiber. The following element in SL(4,{\bf Z}) can transform these polyhedra into the dual polyhedron given by the ref. of \cite{candelas2}. \[ \left( \begin{array}{cccc} 1 & 0&1&2 \\ 0&1& 2& 3 \\ 0&0 & 1 &2\\ 0&0 & -1& -1\\ \end{array} \right). \] In the base of \cite{candelas2}, the additional points make a line with $x_4=-1$. }. (i)$^\dag$ means the modified (i) in \cite {candelas1} with extra tensor multiplets. \begin{eqnarray} &{\rm (i)}& G_2^0=I \mid_{G_1^0, {n_T}^0} ~{\rm with}~ n^0 = \{ 0,1, 2 \} \rightarrow G_2^0=\hat G_2 \mid_ {G_1^0,{n_T}^0}~ {\rm with}~ n^0 \geq 3 \rightarrow \cdots, \nonumber \\ & \big\{ {\rm (i)}^\dag & (G_2^0, {n_T}^0) \mid_{G_1^0 } \rightarrow (G_2, {n_T}) \mid_ {G_1=G_1^0} \rightarrow \cdots ,\big\} \nonumber \\ &{\rm (ii)}& G_1^0=I \mid_{G_2^0,n_T^0}~{\rm with}~ n^0=j \rightarrow {G_1^0}=\hat G_1 \mid_ {G_2^0, n_T^0}~ {\rm with}~ n^0=j \rightarrow \cdots, \nonumber \\ & \big\{ {\rm (ii )}^\dag & (G_1^0,n_T^0)\mid_{G_2^0} \rightarrow ( {G_1}, n_T)\mid_ {G_2=G_2^0} \rightarrow \cdots, \big\} \nonumber \\ &{\rm (iii)}& (G_2^0=I, n_T^0 )\mid _{G_1^0=I}~{\rm with}~ n^0=0 \rightarrow (G_2,n_T) \mid_{G_1=I} ~ \rightarrow \cdots, \end{eqnarray} \noindent where $\hat G_1$ and $\hat G_2$ are non-Abelian gauge symmetries and $ 12 \geq j \geq 0$, see Fig. 1 and Fig. 2. In this paper, we investigate the possibility of following sequence where we picked up CY3-folds and Hodge numbers from the list by Hosono et al. \cite{yau}. \begin{equation} {\rm (iv)} ~ (G_2^0=I, s=1,n_T^0)\mid_{ G_1^0=I} ~{\rm with}~ n^0=2 \rightarrow (G_2,s=2,n_T)\mid _{G_1=I} \rightarrow \cdots \end{equation} \begin{eqnarray} &G_1^0~ ({\rm or}~G_1)&:{\rm the ~gauge ~symmetry~with ~charged~ matter,~whose~information} \nonumber \\ &~&{\rm ~may~come~ from ~ the~ bottom~ of~dual~polyhedron,} ~\nabla ~ \nonumber \\ &~&{\rm of~ appropriate~ K3~ fiber~ for~ the~ series~ (ii)~(or~(ii}^\dag)) ~\cite{candelas1}, \nonumber \\ &G_2^0~({\rm or}~G_2) &:{\rm the~terminal ~gauge ~symmetry~with~ no~ charged~ matter, ~whose} \nonumber \\ & &{\rm ~information~may~come~ from ~ the~ top~ of~dual~polyhedron,} \nonumber \\ && ~\nabla~ {\rm of~ appropriate~K3~ fiber~ for~ the~ series~ (i)~( or~ (i }^\dag)) ~\cite{candelas1,skarke}. \end{eqnarray} \noindent For A series, the non-Abelian instanton numbers of the vector bundles on K3 are denoted as $(k_1^0,k_2^0);~ k_1^0+k_2^0=24$~ \footnote{ Let $m_i^0=m_i$, $m_{iA}^0=m_{iA}$ and $m_{iB}^0 =m_{iB}~ (i=1,2)$ be Abelian instanton numbers. For B series, non-Abelian instanton numbers and Abelian instanton numbers are denoted as $(k_1^0,m_1^0:k_2^0,m_2^0);~ k_1^0+k_2^0=18,~ m_1^0=m_2^0=3$. For C series, they are denoted as $(k_1^0,m_{1A}^0,m_{2A}^0:k_2^0,m_{1B}^0,m_{2B}^0), ~ k_1^0+k_2^0=14,~ m_{1A}^0=m_{1B}^0=3,~ m_{2A}^0=m_{2B}^0=2$ .}. \noindent The integer of $n^0$ is defined by $k_1^0=12+n^0$ and ~$k_2^0=12-n^0$~ \footnote{ The number of non-Abelian instantons should be greater than three except zero. The terminal group should change to avoid this situation, which causes a correction or a modification \cite{aldazabal,candelas2,candelas3}. For C series, the non-Abelian and Abelian instanton numbers are modified as follows \cite{aldazabal}. In $n^0=5$ case, $ (k_1^0,m_{1A}^0,m_{1B}^0~;~ k_2^0,m_{2A}^0,m_{2B}^0)=(12,3,2;0,3,3)$. In $n^0=6$ case, $(k_1^0,m_{1A}^0,m_{1B}^0~;~ k_2^0,m_{2A}^0,m_{2B}^0)=(13,3,2;0,3,3)$.}. The suffix 0 denotes terminal case with $ n_T=1 $ in each series, where ``terminal case'' means $G_1^0=I$ or $G_1=I$. \noindent When $\Delta n_T$ E$_8$ small instantons shrink, in the terminal gauge symmetry side, the non-Abelian instanton numbers should be modified as follows \cite{aldazabal,candelas3}: $(k_1,k_2)$,~$ k_1+k_2+\Delta n_T=24$,~$k_1=k_1^0 = 12$,~ $k_2=k_2^0-\Delta n_T=12- \Delta n_T$ \footnote {Similarly to A, for B and C, the modifications may follow \newline \noindent B-chain~: $k_1 + m_1=k_1^0 +m_1^0 = 12$,~ $k_2+m_2 =12 -\Delta n_T $, \newline C-chain ~: $ k_1+m_{1A}+m_{1B} = k_1^0+m_{1A}^0+m_{1B}^0 = 12 $,~$k_2+m_{2A}+m_{2B} =12 -\Delta n_T $, \newline which we will discuss in this article.}. The additional $\Delta n_T$ numbers of tensor multiplets are created and 29 $\Delta n_T$ appears in the anomaly cancellation condition \footnote{ 29 is the dual coxetor number of E$_8 - $one \cite{aldazabal,candelas3}. The subtraction of one means the freedom of fixing the place where a tensor multiplet is created. The dual coxetor number of E$_7 -$one = 17 works for B-chain case when a tensor multiplet is created and that of E$_6 -$one =11 for the C-chain case \cite{aldazabal,candelas3}.}. $n_T=n_T^0+\Delta n_T$ with $ n_T^0=1$ is the number of the tensor multiplets in D=6 N=1 compactification \footnote{In D=4 and N=2 case, these tensor multiplets become vector multiplets. }. \vspace{8pt} The puzzles which we would like to consider are as follows: \newline I:~Why (iii) and (iv) have the same Hodge numbers for A series ? Do (iv) satisfy the duality ? Though K3 fibers of CY3-folds in (iv) are different from those in (iii), is there any gauge enhancement for (iv) ? \newline II:~Is it possible to obtain B or C versions of (iii) which satisfy the duality by means of conifold transitions ? Already, the method to construct dual polyhedron was given by \cite{candelas2,candelas3} ~about this question. They pointed out that their Hodge numbers are obtained by shrinking of tensor multiplets \cite{candelas3}. \newline III:~Is it possible to obtain B or C versions of (iv) which satisfy the duality by means of conifold transitions ? \newline IV:~ Do the Hodge numbers in B or C series of (iii) coincide with those of (iv)? \newline V:~ Are there any hypersurface representing CY3-folds with $s \geq 2$ which satisfy the duality from $n^0 \geq 4$ ? \newline In section 2, we investigate prob. I. We also discuss prob. III by using the results of \cite{aldazabal} and \cite{candelas3} in section 3. \newline \section{The duality in K3 fibered Calabi-Yau 3-fold} \par ~~~~~ We treat the extension from A-chain $n^0=2$ terminal case. $\relax{\rm CY}_3=\relax {\bf \rm C}{\rm I\kern-.18em P}^4(1,s,(s+1)(1,4,6))[12(s+1)]$ with $\relax{\rm K3}=\relax {\bf \rm C}{\rm I\kern-.18em P}^3(1,1,4,6)[12]$ fiber have the same Hodge numbers as those examples given by \cite{candelas2,yau}. To see why they coincide, we compare dual polyhedron of CY 3-folds. They are not SL(4,{\bf Z}) equivalent. We represent them in the base where one of vertices of dual polyhedron in the bottom denotes the part of its weight for the hypersurface representation. It is ~$(-s,-s-1,-4(s+1),-6(s+1))$ in this case. The base manifold under the elliptic fibration is F$_2$ for $s=1$ case. F$_i$ denotes Hirzebruch surface. The dual polyhedron of F$_i$ has three vertices~:~$ \{ \vec v_1,\vec v_2, -\vec v_1 -i\vec v_2 \}$. An example of integral points in the dual polyhedron of F$_i$ is given by $\{\vec v_1=(1,0) ,(0,-1), \vec v_2=(0,1),(0,0),(-1,-i)\}$. The Hodge numbers and the dualities in case (iii) are derived by investigating the extremal transition of the dual polyhedron of F$_0$ based CY3-fold in \cite{candelas2}. By using dual polyhedra, we can find a fibrations and base manifolds in some cases \cite{candelas1,candelas3,skarke,avram}. The upper suffix in $\nabla$ denotes the dimension of the lattice of a polyhedron. $\{(x_1,x_2,x_3,x_4)\} ~\subset {}^4\nabla $~ forms the dual polyhedron of CY3-fold. $\{(x_1,x_2)\}~\subset {}^2\nabla~ \subset {}^4 \nabla $ represents the dual polyhedron of the base under the elliptic fibration of CY3-folds. In this paper, they are the blown up Hirzebruch surfaces. $\{(x_2,x_3,x_4)\} \mid_{x_1=0} ~\subset {}^3\nabla ~ \subset {}^4\nabla $ represents the dual polyhedron of K3 fiber of CY3-fold. $ \{ (x_3,x_4) \} \mid_{x_1=x_2=0} ~\subset {}^2\nabla $~$~ \subset {}^4\nabla~$ are the dual polyhedron of common elliptic fiber of CY3-folds and K3fiber. However, this is not a sufficient condition of having elliptic fibration. The result of fibrations in CY3-folds is given in table 1. \begin{table} \[ \begin{array} {|l|l|l|l|l|l|} \hline \multicolumn{3}{|c|} {\rm CY3-folds ~of~ \cite{candelas2}} & \multicolumn{3}{|c|} {\rm CY3-folds ~ of~ hyper ~ surface~ rep.}\\ \hline \Delta n _T & {\rm the~ base~ under~}T^2 & {\rm K3~ fiber} & s & {\rm the~ base~ under~}T^2 & {\rm K3~ fiber} \\ \hline 0 & {\rm F}_0 & \relax {\bf \rm C}{\rm I\kern-.18em P} ^3 (1,1,4,6)[12]&1&{\rm F} _2&CP ^3 (1,1,4,6)[12] \\ \hline 2 & {\rm F} _2~ {\rm blown ~ up} & \relax {\bf \rm C}{\rm I\kern-.18em P} ^3 (1,1,4,6)[12]&2&{\rm F} _2~ {\rm blown~up}&CP ^3 (1,1,4,6)[12] \\ \hline 3 & {\rm F} _3 ~{\rm blown~ up} & \relax {\bf \rm C}{\rm I\kern-.18em P} ^3 (1,2,6,9)[18]&3&{\rm F} _2~ {\rm blown~ up}&\relax {\bf \rm C}{\rm I\kern-.18em P} ^3 (1,1,4,6)[12] \\ \hline 4 & {\rm F} _4~{\rm blown ~up} & \relax {\bf \rm C}{\rm I\kern-.18em P} ^3 (1,2,6,9)[18]&4& {\rm F} _2~ {\rm blown~ up}&\relax {\bf \rm C}{\rm I\kern-.18em P} ^3 (1,1,4,6)[12] \\ \hline 6 & {\rm F} _6~{\rm blown ~up} & \relax {\bf \rm C}{\rm I\kern-.18em P} ^3 (1,3,8,12)[24]&6& {\rm F} _2~{\rm blown ~up}&CP ^3 (1,1,4,6)[12] \\ \hline 8 & {\rm F} _8~{\rm blown ~up} & \relax {\bf \rm C}{\rm I\kern-.18em P} ^3 (1,4,10,15)[30]&8&{\rm F }_2~{\rm blown~ up}& \relax {\bf \rm C}{\rm I\kern-.18em P} ^3 (1,1,4,6)[12] \\ \hline 12 & {\rm F} _{12}~{\rm blown ~up} & \relax {\bf \rm C}{\rm I\kern-.18em P}^3(1,5,12,18)[36]&12&{\rm F} _2~{\rm blown~ up} & \relax {\bf \rm C}{\rm I\kern-.18em P} ^3 (1,1,4,6)[12] \\ \hline \end{array} \] \caption{ The kinds of fibrations of CY3-folds} \normalmarginpar{N.B.~;~ The dual polyhedron of \relax {\bf \rm C}{\rm I\kern-.18em P}$^3$(1,5,12,18)[36] coincides with that of \relax {\bf \rm C}{\rm I\kern-.18em P}$^3$(1,6,14,21)[42].} \end{table} In $s \geq2$ case, the additional integral points to F$_2$ are $\{(-i,-i-1) ~ {\rm with}~ \newline 2 \leq i \leq s,~(-1,-1) \} $ within the base. Thus, for $s \geq2$ case, the base manifolds remain blown up F$_2$ since ${ s+1 \over s} \leq 2$. We can see the Hodge number of the base under the elliptic fibration and $n_T$ from dual polyhedron, $n_T+1 = $ $h^{1,1}({\rm F}_2~{\rm blown ~up})= d_1-2d_0=s+2 ~(s \geq 2),$ where $d_i$ is the number of i-dimensional cones \cite{fulton}. $n_T$ coincides with case (iii) \cite{candelas2}. On the other side, K3 fibrations of case (iii) change so that the base manifolds under elliptic fibrations also alter to F$_i$ blown up $(i \geq 2)$ with including F$_0$ and F$_2$. In any case, the bases under elliptic fibrations in case (iv) are birationally equivalent to those in case (iii), which leads to the following identification. (iii) and (iv) are connected by the extremal transitions within each sequences. $ \Delta n_T=0$ cases are deformed to each other by the change of base manifolds, i.e., F$_0 \leftrightarrow$ F$_2$ by using a non-polynomial freedom of deformation \cite{candelas2,gross}. Both CY 3-folds with $\Delta n_T=0 $ are the same manifold with double K3 fibrations \cite{gross}. \begin{eqnarray} \hat \nabla_{s=12, \Delta n_T=12} \supset \cdots \hat \nabla_{s=3,\Delta n_T=3} \supset \hat \nabla_{s=2,\Delta n_T=2} \supset &\hat \nabla_{s=1,\Delta n_T=0}& \nonumber \\ &\updownarrow & \nonumber \\ \nabla_{\Delta n_T=12} \supset \cdots \supset \nabla_{\Delta n_T=3} \supset \nabla_{\Delta n_T=2} \supset &\nabla_{\Delta n_T=0}&. \end{eqnarray} \noindent For $\Delta n_T > 0$, blown up F$_2$ based CY3-folds in case (iv) also can be deformed to those of case (iii) in \cite{candelas2} by using a non-polynomial freedom of deformation. $ \hat \nabla_{s=i, \Delta n_T =i} \leftrightarrow \nabla_{\Delta n_T=i}, $ \noindent where $\hat \nabla$ denotes dual polyhedron of case (iv) and $\nabla$ denotes those of case (iii) \cite{candelas2}. The CY 3-folds constructed by \cite{candelas2} and those with $s$ are the same manifolds with double K3 fibrations \footnote{They are represented by the same dual polyhedra in \cite{candelas3}.}. Each CY 3-fold with $\Delta n_T=i$ can be represented as a K3 fibration in two inequivalent ways as table 1. The CY3-fold with $s=1$ is the terminal case of A-chains with $n^0=2$ in the duality web \cite{aldazabal}. If duality exists, then their Hodge numbers must satisfy the following conditions which comes from D=6 and D=4 theories as the anomaly cancellation \cite{vafa1,candelas1,candelas2}. \begin{eqnarray} h_{2,1}\mid _{ G_2~ {\rm in~ (iv)} }&=&h_{2,1}^0 \mid_{G_2^0=I~{\rm in ~(i)}} -( a-b\Delta n_T) + {\rm dim.} G_2-29 \Delta n_T, \\ \nonumber h_{1,1} \mid_ {G_2 ~{\rm in~(iv)}} &= & h_{1,1}^0 \mid_{G_2^0=I~{\rm in ~(i)}} + {\rm rank}G_2 + \Delta n_T, \\ \nonumber h_{2,1}^0 \mid_{G_2^0=I~{\rm in ~(i)}} &=&243, ~h_{1,1}^0 \mid_{G_2^0=I~{\rm ~in~ (i)}}=3. \end {eqnarray} \noindent where $G_1^0=G_1 =I$ up to U(1) in these cases. \noindent This duality between heterotic string and Type IIA string is summarized in the table 4, which coincides with case (iii) \cite{candelas2}\footnote{ This may imply heterotic - heterotic string duality.}. For CY3-folds side, $G_1$ and $G_2$ symmetries are due to the quotient singularities of K3 fiber in the first column of table 1 rather than the quotient singularities of CY 3-folds with s in table 2 \footnote{ We compared some superpotentials of type IIB side in case (iii) and those in case (iv) and examined the possibility to derive the terminal gauge symmetry by the method of \cite{lerche}. We will give dual polyhedra of case (iv) in the next paper to report about superpotentials more precisely.} . \begin{table}[h] \[ \begin{array}{|l|l|l|l|l|l|l|l|} \hline s & s=1 & s=2 & s=3 & s=4 & s=6 & s=8 & s=12 \\ \hline {\rm quot. sing.} & A_1 &A_1A_2 & A_1^2 A_3 & A_1^4 A_4 & A_1^3 A_6 & A_1^4 A_8 & A_1^6 A_{12} \\ \hline \end{array} \] \caption{The quotient singularities in K3=\relax {\bf \rm C}{\rm I\kern-.18em P}$^3$(1,1,4,6)[12] fibered CY3-fold } \end{table} For K3$ \times$ T$^2$ side, they are the subgroups of the E$_8 \times $ E$_8$ group for case (iii) \footnote{ They are perturbative gauge symmetries for case (iii) and may be non-perturbative ones for case (iv) as $ \Delta n_T=0 $ case \cite{gross}.}. $a+ bn $ is calculated by the index theorem and denotes the number of $G_1$ (or $G_2$) charged hyper multiplet fields, which is summarized in table 3~\cite{bershadsky,aldazabal,candelas1,candelas3}. $n=n^0$ for $ G_1^0= \hat G_1$ case. The number of the charged hyper multiplets of $G_2$ vanishes. We can see this by substituting $n=-n^0$ for $G_2^0= \hat G_2$ case and $n=-\Delta n_T$ for $G_2=\hat G_2$ for A series ~\cite{candelas2, candelas3}. \begin{table}[h] \[ \begin{array} {|l|l|l|l|}\hline G_1 & {\rm A~ series} &{\rm B~ series} &{\rm C~ series} \\ \hline A_1 & 12n+32 &6k_1-39 & 4k_1-17 \\ \hline A_2 & 18n+54 & 10k_1-64 & 6k_1-26 \\ \hline A_3 & 22n+76 &12k_1-75 & 8k_1-33 \\ \hline A_4 & 25n+100 &14k_1-84 & 9k_1-35 \\ \hline D_4 & 24n+96 & & \\ \hline D_5 & 26n+124 &15k_1-90 & 10k_1-39 \\ \hline E_6 & 27n+162 &16k_1-94 & \\ \hline E_7 & 28n+224 & & \\ \hline \end{array} \] \caption{ The review of the number of the charged hyper multiplets.} \end{table} \begin{table}[h] \[ \begin{array}{|l|l|l|l|l|l|l|l|l|l|}\hline s & U(1)^4 \times G_2 & h^{1,1}& h^{1,2}& k_1& k_2 & n_ T^0&\Delta n_T& n_T &n^0 \\ \hline s=1 & U(1)^4 \times (G_2^0=I) &3 &243 & 12 & 12 & 1&0&1 & 2 \\ \hline s=2 & U(1)^4 \times I &5 &185 & 12 & 12-2& 1&2&3& \\ \hline s=3 & U(1)^4 \times A_2 &8 &164 & 12 &12-3 & 1&3&4 &\\ \hline s=4 & U(1)^4 \times D_4 &11 &155 & 12 & 12-4 & 1&4&5 &\\ \hline s=6 & U(1)^4 \times E_6 &15 &147 & 12 & 12-6 & 1&6&7 &\\ \hline s=8 & U(1)^4 \times E_7 &18 &144 & 12 & 12-8 & 1&8&9 &\\ \hline s=12 & U(1)^4 \times E_8 &23 &143 & 12 & 12-12 & 1&12&13 &\\ \hline \end{array} \] \caption{The duality of (iv) about K3=\relax {\bf \rm C}{\rm I\kern-.18em P}$^3$(1,1,4,6)[12] fibered CY3-fold } \end{table} \section{Discussion and Conclusion} \par In this article, we have studied the property of CY3-folds with $ s\geq 2$ case from A series. They are the same CY 3-folds as those constructed in \cite{candelas2}, which have double K3 fibrations. \newpage For the puzzle III, the conclusion is that the hypersurface representations with $s \geq 2$ cases from B series and C series in $n^0=2$ satisfy the duality of terminal B and C series with extra tensor multiplets \footnote{ For B and C sequences, the modification of eq. (6) is necessary. The change in the Hodge numbers, $h_{2,1}$ according to the change of the terminal group is not only the difference of dim. $G_2^0$~( or dim. $G_2$)~\cite{aldazabal,candelas3}. Thus we compare the Hodge numbers of CY3-folds which have the same terminal gauge symmetry.}. For example, B and C series in s=2, their Hodge numbers can be interpreted as those with shrinking of two instantons from $n^0=2$ case as the ref. of \cite{candelas3}. $s \geq 3$ cases are also explained by shrinking $\Delta n_T$ instantons from the terminal case with the same $\hat G_2$ symmetry. \begin{eqnarray} &\Delta h_{2,1}&= -h_{2,1}^0 \mid_{G_2^0=\hat G_2~{ \rm terminal~in~(i)}} +h_{2,1} \mid_{G_2=\hat G_2~{ \rm terminal~in~(iv)}}, \nonumber \\ &\Delta h_{1,1}&= -h_{1,1}^0 \mid_{G_2^0=\hat G_2~{ \rm terminal~in~(i)}} +h_{1,1} \mid_{G_2=\hat G_2~{ \rm terminal~in~(iv)}}, \nonumber \\ &{\rm B~ series}&~ : -\Delta h_{2,1}=17 \Delta h_{1,1}, ~{\rm C~ series} : -\Delta h_{2,1}=11 \Delta h_{1,1}. \end{eqnarray} \noindent This situation is quite similar to that of A series. They are interpreted as the dualities obtained by unhiggsing of $U(1)$ and $U(1)^2$ from (iv) in A series. For puzzle IV, Candelas, Perevalov and Rajesh seemed to construct B and C versions of (iii) and derived the Hodge numbers already, though they do not write them explicitly \cite{candelas3}. According to the description, Hodge numbers of them are the same as table 5 and table 7. \begin{table} \[ \begin{array}{|l|l|l|l|l|l|l|l|l|l|}\hline s & U(1)^4 \times U(1) \times G_2 & h^{1,1}&h^{1,2} & k_1&k_2&{n_T}^0&\Delta n_T& n_T &n^0\\ \hline s=1 & U(1)^4 \times U(1) \times (G_2^0=I) & 4 & 148 & 9 & 9 &1 & 0 & 1& 2\\ \hline s=2 & U(1)^4 \times U(1) \times I & 6 & 114 & 9 & 9-2 & 1&2&3 & \\ \hline s=3&U(1)^4 \times U(1)\times A_2 & 9 & 101 & 9 & 9-3 & 1&3&4 &\\ \hline s=4&U(1)^4 \times U(1) \times D_4 & 12 & 96 & 9 & 9 -4 & 1&4&5& \\ \hline s=6&U(1)^4 \times U(1)\times E_6 & 16 & 92 & 9 & 9 -6 & 1&6&7& \\ \hline \end{array} \] \caption{ The duality of (iv) about K3=\relax {\bf \rm C}{\rm I\kern-.18em P}$^3$(1,1,2,4)[10] fiberd CY3-fold } \end{table} \begin{table} \[ \begin{array}{|l|l|l|l|l|l|l|l|l|}\hline n^0 & U(1)^4 \times U(1) \times G_2^0 & h^{1,1}&h^{1,2} & k_1&k_2&{n_T}^0&\Delta n_T& n_T \\ \hline n^0=0&U(1)^4\times U(1) \times I & 4 & 148 &9 & 9 &1 & 0&1 \\ \hline n^0=3&U(1)^4 \times U(1) \times A_2 & 6 & 152 & 9+3 & 9-3 & 1&0&1 \\ \hline n^0=4&U(1)^4 \times U(1)\times D_4 & 8 & 164 & 9+4 & 9 -4 & 1&0 &1 \\ \hline n^0=6&U(1)^4 \times U(1) \times E_6 & 10 & 194 & 9+6 & 9 -6 & 1&0&1 \\ \hline \end{array} \] \caption{ The duality of (i) about terminal CY3-folds in B series } \end{table} For the puzzle V, we examine K3 fiberd CY3-folds whose s=1 case is the terminal case in the A-chains with $n^0=4$. Their K3 fiber is \newline K3=\relax {\bf \rm C}{\rm I\kern-.18em P}$^3$(1,2,6,9)[18]. The dual polyhedron of $s=2$ case has following additional points $(-2, -6,-18, -27)$, $(-1,-3,-10,-15)$ and $(-1, -3,-9,-14)$, which imply $\Delta n_T$=1. $ h_{1,1} \mid _{s=2}-h_{1,1}^0 \mid _{s=1}=3$. The terminal group should change from $D_4$ to $G_2$ with rank 6. We can not find $G_2$ which satisfies the extension of eq. (6). It seems that it needs another idea to see the duality in this case. \newline {\bf Acknowledgement} ~~~~~We gratefully acknowledge fruitful conversations with M. Kobayashi, S. Hosono and N. Sakai. We would like to thank K. Mohri mostly for useful discussions. \begin{table} \[ \begin{array} {|l|l|l|l|l|l|l|l|l|l|} \hline s& U(1) \times U(1)^2 \times G_2 & h^{1,1}& h^{1,2}&k_1&k_2&{ n_T }^0& \Delta n_T & n_T &n^0 \\ \hline s=1 & U(1)^4\times U(1)^2 \times (G_2^0=I) &5 & 101&7&7&1&0 &1&2\\ \hline s=2 & U(1)^4 \times U(1)^2 \times I & 7& 79& 7& 7-2 &1&2&3 & \\ \hline s=3 & U(1)^4 \times U(1)^2 \times A_2 & 10 &70& 7 & 7-3 &1&3&4 & \\ \hline s=4 & U(1)^4 \times U(1)^2 \times D_5 & 13 &67& 7 & 7-4 &1&4&5 & \\ \hline s=5 & U(1)^4 \times U(1)^2\times E_6 & 17 &65& 7 & 0 &1&5&6 & \\ \hline s=6 & U(1)^4 \times U(1)^2\times E_6 & 17 &65& 7 & 0 &1&6&7 & \\ \hline \end{array} \] \caption{The duality of (iv) about K3=\relax {\bf \rm C}{\rm I\kern-.18em P}$^3$(1,1,2,2)[6] fiberd CY3-fold } \end{table} \begin{table} \[ \begin{array} {|l|l|l|l|l|l|l|l|l|} \hline n^0& U(1) \times U(1)^2 \times G_2^0 & h^{1,1}& h^{1,2}&k_1&k_2&{ n_T }^0& \Delta n_T & n_T \\ \hline n^0=0 & U(1)^4\times U(1)^2\times I &5 & 101&7&7&1&0 &1\\ \hline n^0=3 & U(1)^4 \times U(1)^2\times A_2 & 7 &103& 7+3& 7-3 &1&0&1 \\ \hline n^0=4 & U(1)^4 \times U(1)^2 \times D_5 & 10 &110& 7+4 & 7-4 &1&0&1 \\ \hline n^0=5 & U(1)^4 \times U(1)^2 \times E_6 & 12 &120& 7+5 & 0 &1&0&1 \\ \hline n^0=6 & U(1)^4 \times U(1)^2 \times E_6 & 11 &131& 7+6 & 0 &1&0&1 \\ \hline \end{array} \] \caption{The duality of (i) about terminal CY3-folds in C series } \end{table}
\section{Introduction} A recent great development of nonperturbative aspects of superstring theory has begun with following two events. One is a proposal of M-theory \cite{Witten1}, which is a hypothetical theory describing strongly coupled region of type IIA superstring theory and reduces to eleven-dimensional supergravity in the low energy limit. The relation among parameters of string theory (the string length $l_s$ and the string coupling $g_s$) and of eleven-dimensional supergravity (the Planck length $l_{11}$ and the radius of compactified the 11-th direction $R_{11}$) is given by \begin{equation} R_{11}=g_sl_s=g_s^{2/3}l_{11}. \end{equation} M-theory gives a systematic view point for string theory --- in considering the M-theory, all known perturbative string theories are unified via various connections from duality. The other is a discovery of D-branes \cite{Polchinski}, which are solitonic solutions of string theory and can be constructed by using conformal field theory in a well-defined manner. Thus they are powerful tools for investigating nonperturbative phenomena in string theory. Further, it was followed by a new interpretation of dimensional reductions of ten-dimensional supersymmetric Yang-Mills (SYM) theory as low energy effective theories for dynamics of the D-branes \cite{Witten2}. Those movements culminated in the conjecture by Banks, Fischler, Shenker and Susskind (BFSS) \cite{BFSS}, that ten-dimensional SYM theory dimensionally reduced to one dimension which describes the low energy dynamics of D0-branes so far, in the infinite momentum frame, gives a constructive definition of the M-theory. This one-dimensional SYM quantum mechanical model is called M(atrix) theory. Moreover, by considering toroidal compactification on a circle and on a two-torus in the manner of Taylor \cite{Taylor}, it leads to a proposal for a nonperturbative definition of type IIA superstring theory \cite{DVV,Banks-Seiberg} and that of type IIB superstring theory \cite{Banks-Seiberg,Sethi-Susskind}, respectively. In this paper, we call them IIA and IIB matrix string theories. The IIA matrix string theory is given by ten-dimensional SYM theory dimensionally reduced to two-dimensions, and the string coupling constant corresponds to the inverse of the SYM coupling. Also, the IIB matrix string theory is a dimensional reduction of the ten-dimensional SYM theory to three dimensions. The string coupling is given by a ratio of lengths of two spatial dimensions of the three-dimensional SYM theory. In addition, with respect to another nonperturbative definition of type IIB superstrings \cite{IKKT}, which is in form of the ten-dimensional SYM theory reduced to a point (zero dimension), proposed by Ishibashi, Kawai, Kitazawa and Tsuchiya, it is referred to IKKT model. Those matrix string theories successfully reproduce some of known results obtained from an analysis of BPS saturated states consisting of fundamental strings, D-branes and their bound states \cite{Banks-Seiberg-Shenker}. This fact is considered as one of evidences that the matrix string theories truely are constructive definitions of string theories. However, with respect to nonperturbative dynamics of the string theories, in order to investigate it we cannot help treating nonperturbative dynamics of the SYM theory sides. Though understanding of this area is now in development \cite{Seiberg,PSS}, unfortunately at present it seems to be not powerful enough for handling the problem. In this paper, we consider the partition function of the matrix string theories --- it is one of the most fundamental quantities reflecting dynamical property of vacua of the theories. The partition function of the IIA matrix string theory was computed in strongly coupled limit of the SYM theory by Kostov and Vanhove \cite{Kostov-Vanhove}. For the partition function of the IKKT model, the exact result was obtained by interpreting the theory as a cohomological field theory by Moore, Nekrasov and Shatashvili \cite{Moore-Nekrasov-Shatashvili}. Here, we exactly calculate the partition function of both of IIA and IIB matrix string theories by mapping the theories to cohomological field theories. In the IIA case our result coincides with the result obtained in the strongly coupled limit, which thus gives a proof of {\it exact quasi classics} discussed in \cite{Kostov-Vanhove}. Also in the IIB case, our result agrees with the exact calculation for the IKKT model, which seems to suggest the equivalence between those two different nonperturbative formulations of IIB string theory. The paper is organized as follows. In the next section, we introduce the IIA and IIB matrix string theories via toroidal compactification of the M(atrix) theory. We briefly explain how string interactions emerge from the SYM theory and show how spinors with correct chirality appear. In section 3, as a preparation for computing the partition function of the matrix strings, we evaluate the partition function of four-dimensional ${\cal N}=4$ $SU(N)$ SYM theory defined on a four-torus by mapping the theory to a cohomological field theory via a twisting procedure. Then in section 4, we exactly compute the partition function of the IIA matrix string theory, and see that our result coincides with the result in the infra-red limit by Kostov and Vanhove. In section 5, for the partition function of the IIB matrix string theory we perform the calculation in the ten-dimensional IIB limit, and find the identical result with the partition function of the IKKT model by Moore, Nekrasov and Shatashvili. Finally, section 6 is devoted to conclusions. In Appendix A, we clarify the connection between two-dimensional ${\cal N}=2$ SYM theory and a cohomological field theory used in section 4. \section{Matrix Strings} In this section we review some basic properties of the IIA and IIB matrix string theories, which are derived from the M(atrix) theory by considering toroidal compactification on $S^1$ and on $T^2$ respectively. In addition, we give an argument that two spinors of the same space-time chirality emerge in the IIB matrix string theory, which has not been found in the literatures. We start with the M(atrix) theory \cite{BFSS}, whose action has the same form as low energy effective action of $N$ D0-branes, that is one-dimensional $U(N)$ SYM theory with 16 supercharges. The BFSS conjecture is that this action exactly describes M-theory in the decompactified limit $R_{11} \rightarrow \infty$ by going to the infinite momentum frame. The infinite momentum frame means infinite amount of boosting along the 11th direction, i.e. momentum of the 11th direction becomes \begin{equation} p_{11}=\frac{N}{R_{11}}\rightarrow \infty. \end{equation} At the same time we must take the limit $N,\; R_{11} \rightarrow \infty,$ in order that the resulting theory represents the strongly coupled limit of type IIA superstring theory. \subsection{IIA Matrix String Theory} Let us consider the compactification of one of the transverse directions, say the 9th, to a circle of the circumference $L,$ denoted by $S^1(L).$ Since in the M(atrix) theory we consider the limit that the 11th direction is decompactified, we have ten-dimensional type IIA string theory with the string length \begin{equation} l_s^2=\frac{2\pi l_{11}^3}{L}. \end{equation} According to the prescription of Taylor \cite{Taylor}, this IIA string theory is described by two-dimensional ${\cal N} =8 \; U(N)$ SYM theory, which is a dimensional reduction of ten-dimensional ${\cal N}=1$ SYM theory \cite{DVV}, \begin{eqnarray} S_{{\rm IIA-MS}} & = & \frac{1}{g^2}\int dt \int_0^Rd\sigma\;{\rm tr} \left[-\frac 14F_{\mu\nu}F^{\mu\nu}-\frac 12(D_{\mu}X^I)^2 +i\theta^T(D_t+\Gamma^9D_{\sigma})\theta \right.\nonumber \\ & & \left. +\frac 14 [X^I, X^J]^2 +\theta^T\Gamma^I[X_I,\theta]\right]. \label{actionIIA-MS} \end{eqnarray} Here, $F_{\mu\nu}=\partial_{\mu}A_{\nu}-\partial_{\nu}A_{\mu} -i[A_{\mu},A_{\nu}]$ is a field strength made from two-dimensional gauge field $A_{\mu},$ $X^I$ ($I=1,\cdots,8$) are Higgs fields and $\theta^{\alpha}$ ($\alpha=1,\cdots,16$) are fermions. We use the convention that the $\Gamma$-matrices are $16\times 16$ real symmetric matrices satisfying \begin{equation} \{\Gamma^I,\Gamma^J\}=2\delta^{IJ}, \hspace{1cm} \Gamma^9=\Gamma^1\cdots\Gamma^8. \end{equation} $\theta^{\alpha}$'s are decomposed into the spinor and conjugate spinor representations (${\bf 8_s}\oplus {\bf 8_c}$) of the rotational group in the transverse directions $SO(8),$ which are characterized by the eigenvalues of $\Gamma^9,$ so called chirality: \begin{equation} \theta^{\alpha}=\theta^{\alpha}_++\theta^{\alpha}_-, \hspace{1cm} \Gamma^9\theta^{\alpha}_{\pm}=\pm\theta^{\alpha}_{\pm}, \label{chiraldecomposition} \end{equation} and $X^I$'s transform as the vector ${\bf 8_v}.$ The spatial coordinate $\sigma,$ which arises as performing the compactification, takes a value on the circle dual to $S^1(L)$: $0\leq\sigma\leq R,$ where\footnote{ The relation between parameters in the matrix string theory and those in SYM theory, eqs. (\ref{RLrelation}) and (\ref{IIAggs}) in the IIA case as well as eqs. (\ref{*}), (\ref{**}) and (\ref{IIBcoupling}) in the IIB case, can be derived either by tracing the procedure by Taylor \cite{Taylor} or by employing another argument in ref. \cite{FHRS}.} \begin{equation} R=(2\pi)^2\frac{l_s^2}{L}. \label{RLrelation} \end{equation} The coupling constant of the SYM theory $g$ is related to the string coupling $g_s$ as \begin{equation} g^2=\frac{2\pi}{(Rg_s)^2}. \label{IIAggs} \end{equation} Weakly coupled strings are recovered by considering the limit $g_s\rightarrow 0$ or equivalently the infra-red limit of the SYM theory \cite{DVV}. In this situation, the theory is described by the eigenvalues of the simultaneously diagonalizable configurations of $X^I$ and $\theta^{\alpha}$: \begin{equation} \tilde{X^I} = {\rm diag}(x^I_1,\cdots,x^I_N), \hspace{1cm} \tilde{\theta^{\alpha}} = {\rm diag}(\theta^{\alpha}_1,\cdots, \theta^{\alpha}_N) \end{equation} where \begin{equation} X^I=V\tilde{X^I}V^{\dagger}, \hspace{1cm} \theta^{\alpha}=V\tilde{\theta^{\alpha}}V^{\dagger}. \label{diagonalpart} \end{equation} The angular variables $V\in U(N)$ and nontrivial configurations of the gauge field yielding non-zero curvature $F_{\mu\nu}$ are energetically decoupled from the theory in the infra-red limit, and the action (\ref{actionIIA-MS}) reduces to the action of multiple Green-Schwarz superstrings in the light-cone gauge. Then the gauge field can take the pure gauge configuration \begin{equation} A_{\mu}=iV\partial_{\mu}V^{\dagger}. \end{equation} Also, the $x^I_i$ and $\theta^{\alpha}_i$ can represent strings of various lengths by considering the multi-valued configuration: \begin{eqnarray} & & \tilde{X^I}(t,\sigma+R) = g^{\dagger}\tilde{X^I}(t,\sigma)g, \hspace{1cm} \tilde{\theta^{\alpha}}(t,\sigma+R) = g^{\dagger}\tilde{\theta^{\alpha}}(t,\sigma)g, \nonumber \\ & & V(t,\sigma+R) = V(t,\sigma)g, \label{BCofdiagonalpart} \end{eqnarray} where $g$ is an element of the Weyl group of $U(N),$ i.e. the permutation group $S_N,$ which permutes the $N$ eigenvalues. In going around the $\sigma$-direction, the eigenvalues are interchanged by the action of $g$ in eqs. (\ref{BCofdiagonalpart}), and as a result they form cycles of various lengths corresponding to permutation cycles in $g.$ Each cycle is interpreted as a single closed string, and thus for $g\in S_N$ consisting of $N_n$ $n$-cycles (satisfying $N=\sum_nnN_n$) one has $N_n$ strings with length $n.$ Note that the total matrices $X^I,$ $\theta^{\alpha}$ in eqs. (\ref{diagonalpart}) and the gauge field $A_{\mu},$ which appear in the SYM theory, remain single-valued although we consider the case that the variables of string coordinates $\tilde{X^I}$ and $\tilde{\theta^{\alpha}}$ are multi-valued as in eqs. (\ref{BCofdiagonalpart}), which is argued in ref. \cite{Wynter}. For the simplest $N=2$ case, the multi-valued configuration is represented by \begin{equation} \tilde{X^I}=\left(\begin{array}{cc} x^I_1 & 0 \\ 0 & x^I_2 \end{array}\right), \hspace{1cm} \tilde{\theta^{\alpha}}=\left(\begin{array}{cc} \theta^{\alpha}_1 & 0 \\ 0 & \theta^{\alpha}_2 \end{array}\right), \hspace{1cm} V=\frac{1}{\sqrt{2}}\left(\begin{array}{cc} e^{i\pi\sigma/R} & -e^{i\pi\sigma/R} \\ 1 & 1 \end{array}\right), \end{equation} where $$ x^I_1(t,\sigma+R)=x^I_2(t,\sigma), \hspace{1cm} x^I_2(t,\sigma+R)=x^I_1(t,\sigma), $$ and $\theta^{\alpha}_i$ satisfies the same boundary condition. The matrices $X^I,$ $\theta^{\alpha}$ and $A_{\mu}$ are single-valued. In particular, the gauge field is \begin{equation} A_t=0, \hspace{1cm} A_{\sigma}=\frac 12 \left(\begin{array}{cc} 1 & 0 \\ 0 & 0 \end{array}\right), \end{equation} which cannot be eliminated by any single-valued gauge transformation. As discussed in \cite {DVV}, relaxing the strict limit $g_s=0,$ the number of the strings are no longer conserved. We can interpret this as a result of string interactions (splitting or joining of one or two strings) which occur when the two sets of eigenvalues $\{x^I_i\}_{I=1,\cdots,8}$ and $\{x^I_j\}_{I=1,\cdots,8}$ coincide. Whenever the interaction occurs, the boundary condition of $\tilde{X^I},$ $\tilde{\theta^{\alpha}}$ and $V$ is changed, i.e. a branch point appears on a cylinder coordinated by $t$ and $\sigma.$ We should remark that as discussed in \cite{GHV}, in spite of this singular configuration of the diagonal variables, for a neighborhood of the branch point the corresponding total matrices exist as a smooth and single-valued configuration of the SYM theory. In the section 4 of ref. \cite{GHV} such a configuration is constructed in the $N=2$ case, which is given by a solution of the dimensionally reduced version of the self-dual equation of four-dimensional Yang-Mills theory to two dimensions: \begin{eqnarray} & & [X_1,X_2]=\frac{i}{g^2}F_{t\sigma}, \label{selfdualeq} \\ & & D_tX_1=-D_{\sigma}X_2, \nonumber \\ & & D_tX_2=D_{\sigma}X_1, \end{eqnarray} and the all other fields are set to zero. Compared to the configuration in the strict limit ($g^2=\infty$) that is $[X_1,X_2]=0,$ the $O(1/g^2)$-correction in eq. (\ref{selfdualeq}) works well to make the configuration in the SYM theory smooth\footnote{An extension of this argument in the case of general $N$ is discussed in refs. \cite{BBN}. I thank L. Bonora for informing me of those literatures.}. \subsection{IIB Matrix String Theory} When compactifying IIA string theory to a circle and taking T-dual to the circle, we have IIB string theory on $S^1.$ In the limit of shrinking the circle of the IIA theory, the $S^1$ of the IIB theory is decompactified, thus we have IIB string theory in ten dimensions. From the view point of M-theory, IIA theory on $S^1$ means M-theory on $T^2.$ We denote the size of the two-torus by $L_1$ and $L_2,$ i.e. $T^2=S^1(L_1)\times S^1(L_2).$ Since there is no distinctive meaning between the two $S^1$'s, now we have two ways to obtain IIB theory corresponding to shrinking either $L_1$ or $L_2.$ From the analysis based on low energy effective theories, two IIB theories we obtain as the result are believed to be equivalent and connected by S-duality \cite{Aspinwall,Schwarz}. In this way, the M-theory perspective yields a geometrical interpretation to the S-duality in type IIB string theory. Let us consider this operation in the M(atrix) theory. Then we obtain three-dimensional ${\cal N}=8$ $U(N)$ SYM theory, which is dimensionally reduced from the ten-dimensional theory, as a matrix model corresponding to IIB theory on $S^1$ \cite{Banks-Seiberg,Sethi-Susskind}: \begin{eqnarray} S_{{\rm IIB-MS}} & = & \frac{1}{g^2}\int dt \int_0^{R_1}d\sigma_1 \int_0^{R_2}d\sigma_2\;{\rm tr} \left[-\frac 14F_{\mu\nu}F^{\mu\nu}-\frac 12(D_{\mu}X^I)^2 \right.\nonumber \\ & & \left.+i \theta^T(D_t+\Gamma^8D_{\sigma_1}+\Gamma^9D_{\sigma_2}) \theta +\frac 14 [X^I, X^J]^2 +\theta^T\Gamma^I[X_I,\theta]\right]. \label{actionIIB-MS} \end{eqnarray} Here we compactified the 8th and 9th transverse directions to the above mentioned (rectangular) two-torus. The size of the spatial directions ($\sigma_1$ and $\sigma_2$) which the SYM theory is defined on is related by T-duality to the two-torus, \begin{equation} R_i=(2\pi)^2\frac{l_s^2}{L_i} \hspace{1cm} (i=1,2), \label{*} \end{equation} and the SYM coupling are given by \begin{equation} g^2=\frac{R_1R_2R_{11}}{(2\pi)^2l_s^4}. \label{**} \end{equation} It is remarkable that the string coupling is given by a ratio of the two lengths of the torus: \begin{equation} g_s=\frac{R_1}{R_2}=\frac{L_2}{L_1}, \label{IIBcoupling} \end{equation} which gives the geometrical understanding of S-duality also in the matrix string level. In the vanishing torus limit $L_1, L_2\rightarrow 0,$ a new dimension opens up and becomes decompactified, so this limit with the ratio $L_2/L_1$ fixed is considered to give type IIB theory of the coupling $g_s$ determined by eq. (\ref{IIBcoupling}) \cite{Sethi-Susskind,FHRS}. Note that the abelian part of the field strength $F_{\mu\nu}$ made from three-dimensional gauge field $A_{\mu}$ reduces to a single scalar field via duality transformation. Although the manifest symmetry of the action (\ref{actionIIB-MS}) is $SO(7),$ it is considered that this scalar and the seven Higgs fields $X^I$ ($I=1,\cdots,7$) together belong to ${\bf 8_v}$ in $SO(8)$ in the ten-dimensional IIB limit \begin{equation} R_1R_2\rightarrow \infty, \hspace{1cm} g_s :{\rm fixed} \label{10DIIBlimit} \end{equation} by the argument for BPS states \cite{FHRS} and by the analysis of the moduli space of the three-dimensional SYM theory \cite{Banks-Seiberg,Seiberg}. The fermions $\theta^{\alpha}$ ($\alpha=1,\cdots,16$) are to represent two space-time spinors of the same chirality in the IIB theory. In fact, after decomposing by the eigenvalues of $\Gamma^9$ as in eq. (\ref{chiraldecomposition}), we put \begin{equation} \psi^{\alpha}_+=\Gamma^8\theta^{\alpha}_-. \label{psi+} \end{equation} Then fermion part of the lagrangian density takes the form \begin{eqnarray} & & {\rm tr} \;[i\theta^T_+(D_t+D_{\sigma_2})\theta_+ +i\psi^T_+(D_t-D_{\sigma_2})\psi_+ +i\theta^T_+D_{\sigma_1}\psi_+ +i\psi^T_+D_{\sigma_1}\theta_+ \nonumber \\ & & \; +\theta^T_+\Gamma^I\Gamma^8[X^I,\psi_+] -\psi^T_+\Gamma^I\Gamma^8[X^I,\theta_+]]. \label{IIBfermionpart} \end{eqnarray} Both of $\theta_+$ and $\psi_+$ are spinors with respect to the manifest symmetry $SO(7)$ of the same eigenvalue (+1) of $\Gamma^9.$ In the weakly coupled limit $g_s\ll 1,$ which means $R_1\ll R_2,$ nonzero modes of $\partial_{\sigma_1}$ energetically decouple, and $\theta_+$ and $\psi_+$ represent the two ${\bf 8_s}$ spinors in IIB theory. Also, in the strongly coupled limit $R_2\ll R_1$, it turns out that $\xi_+=(\theta_++\psi_+)/\sqrt 2$ and $\eta_+=(\theta_+-\psi_+)/\sqrt 2$ become spinors desired in IIB theory\footnote{ In IIA matrix string theory, by the same replacement (\ref{psi+}), we seem to have chiral IIB theory, but it is not correct. The IIA matrix string theory has the manifest symmetry $SO(8).$ Note that $\psi_+$ is not a spinor under the $SO(8)$ due to the $\Gamma^8$ factor in eq. (\ref{psi+}).}. In this way, we can see that in both of the two limits related via S-duality chiral spinors are correctly reproduced in the IIB matrix string theory. \section{Partition Function of four-dimensional ${\cal N}=4$ $SU(N)$ Super Yang-Mills Theory} Here we calculate the partition function of ${\cal N}=4$ $SU(N)$ SYM theory on four-torus $T^4$ by mapping the theory to a cohomological field theory. The dimensional reduced version of this argument is used in later computations of the partition function of IIA and IIB matrix string theories. ${\cal N}=4$ $SU(N)$ SYM theory on a flat four-dimensional space-time is given by the following lagrangian density defined on the ${\cal N}=1$ superspace $(x,\theta,\bar{\theta})$: \begin{eqnarray} {\cal L}_{{\cal N}=4} & = & \frac{1}{16g^2}{\rm tr} \left(W^{\alpha}W_{\alpha}|_{\theta\theta}+ \bar{W}_{\dot{\alpha}} \bar{W}^{\dot{\alpha}}|_{\bar{\theta}\bar{\theta}}\right) \nonumber \\ & & +\frac{1}{g^2}{\rm tr} \left(\Phi_1^{\dagger}e^V\Phi_1+ \Phi_2^{\dagger}e^V\Phi_2+ \Phi_3^{\dagger}e^V\Phi_3\right) |_{\theta\theta\bar{\theta}\bar{\theta}} \nonumber \\ & & +\frac{1}{\sqrt{2}g^2}{\rm tr} \left( \Phi_1[\Phi_2,\Phi_3]|_{\theta\theta}+ \Phi_1^{\dagger}[\Phi_3^{\dagger},\Phi_2^{\dagger}] |_{\bar{\theta}\bar{\theta}}\right). \label{4DN=4} \end{eqnarray} In this section, we use the notation in ref. \cite{Wess-Bagger}. The vector superfield $V$ represents a multiplet containing a gauge field $A_m$ and a complex two-component gauge fermion $\lambda,$ and the chiral superfield $\Phi_s$ ($s=1,2,3$) contains a multiplet of a complex Higgs scalar $B_s$ and a Higgsino $\psi_s.$ All the fields belong to the adjoint representation of $SU(N),$ ``${\rm tr}$'' denotes the trace in the fundamental representation. The theory has the internal symmetry group $SU(4)_I,$ under which the fermions ($\lambda$ and $\psi_s$'s) transform together as {\bf 4} and $B_s$'s as {\bf 6}. Also, the supercharges $Q_{\alpha}^v,$ $\bar{Q}_{v \dot{\alpha}}$ as {\bf 4}, ${\bf \bar{4}}$ respectively, where $v\;(=1,\cdots,4)$ is the $SU(4)_I$ index and $\alpha,$ $\dot{\alpha}$ are Lorentz indices belonging to $SU(2)_L,$ $SU(2)_R.$ In terms of the component fields, the lagrangian takes the form \begin{eqnarray} {\cal L}_{{\cal N}=4} & = & \frac{1}{g^2}{\rm tr} \left[ -\frac 14 F^{mn}F_{mn}-i\bar{\lambda}\bar{\sigma}^mD_m\lambda \frac 12 D^2\right] \nonumber \\ & & \left.+\frac{1}{g^2}\sum_{s=1}^3{\rm tr}\right[ F_s^{\dagger}F_s-(D^mB_s)^{\dagger}(D_mB_s) -i\bar{\psi}_s\bar{\sigma}^mD_m\psi_s \nonumber \\ & & \hspace{15mm} \left. -\frac{i}{\sqrt 2}B_s^{\dagger}[\lambda,\psi_s] +\frac{i}{\sqrt 2}B_s[\bar{\lambda},\bar{\psi}_s] +\frac 12 D[B_s,B_s^{\dagger}]\right] \nonumber \\ & & +\frac{1}{\sqrt 2 g^2}{\rm tr} (F_1[B_2,B_3]+F_2[B_3,B_1]+F_3[B_1,B_2] \nonumber \\ & & \hspace{15mm} -B_1[\psi_2,\psi_3]-B_2[\psi_3,\psi_1]-B_3[\psi_1,\psi_2] \nonumber \\ & & \hspace{15mm} +{\rm h. c.}), \end{eqnarray} where $F_s$ and $D$ are auxiliary fields appearing in $\Phi_s$ and $V.$ \subsection{Twisting of ${\cal N}=4$ Super Yang-Mills Theory} The twisting procedure, which was first introduced by Witten \cite{Witten3}, gives a systematic tool for constructing a cohomological field theory from the original physical theory. Here we briefly explain a twisting procedure of $N=4$ SYM theory adopted by Vafa and Witten \cite{Vafa-Witten}. First, we consider a general four-manifold on which the ${\cal N}=4$ SYM theory is defined. The holonomy group of this manifold is $SO(4)=SU(2)_L\times SU(2)_R,$ that is a gauged symmetry with the spin connection being the (external) gauge field. Adding the global $SU(4)_I,$ we concentrate the symmetry group of the theory $$ H=SU(2)_L\times SU(2)_R\times SU(4)_I. $$ The twisting by Vafa and Witten is performed as follows. Consider the subgroup of the internal symmetry group $SU(4)_I$: $$ SU(4)_I\supset SO(4)=SU(2)_F\times SU(2)_{F'}, $$ then replace the action of $SU(2)_L$ by the diagonal sum $SU(2)'_L=SU(2)_L\oplus SU(2)_{F'}.$ Thus the twisted symmetry group becomes $$ H'=SU(2)'_L\times SU(2)_R\times SU(2)_F. $$ Under $H',$ the supercharges split up as \begin{eqnarray*} & & Q_{\alpha}^v=Q_{\alpha}^{ij}\rightarrow Q_{\alpha}^{i\beta}=\left\{\begin{array}{l} Q_{\alpha}^{i\alpha}\equiv Q^i \\ \frac 12 \varepsilon_{\gamma (\beta}Q_{\alpha)}^{i\gamma} \equiv Q_{\alpha\beta}^i \end{array}\right. \\ & & \bar{Q}_{v \dot{\alpha}}=\bar{Q}_{ij\dot{\alpha}}\rightarrow \bar{Q}_{i\beta\dot{\alpha}}, \end{eqnarray*} where $i$ and $j$ are $SU(2)_F$ and $SU(2)_{F'}$ indices respectively, also $j$ is converted to $\beta$ by the twisting. Here we have the two scalar supercharges $Q^i$ ($i=1,2$), which are nilpotent \begin{equation} \{Q^i, Q^j\}=0 \end{equation} as seen from the supersymmetry algebra of the original supercharges \begin{eqnarray} & & \{Q_{\alpha}^v,\bar{Q}_{w \dot{\alpha}}\}= 2\sigma^m_{\alpha\dot{\alpha}}\delta^v_w P_m, \nonumber \\ & & \{Q_{\alpha}^v,Q_{\beta}^w\}= \{\bar{Q}_{v\dot{\alpha}},\bar{Q}_{w\dot{\beta}}\}=0. \label{***} \end{eqnarray} We can construct a cohomological field theory based on each $Q^i$'s. The $Q^i$'s are related to the original supercharges as \begin{eqnarray} Q^1 & = & Q_1^{11}+Q_2^{12}=Q_{\alpha=1}^{v=1}+Q_{\alpha=2}^{v=2}, \nonumber \\ Q^2 & = & Q_1^{21}+Q_2^{22}=Q_{\alpha=1}^{v=3}+Q_{\alpha=2}^{v=4}. \label{Qrelation} \end{eqnarray} Next, in the case of four-dimensional K\"{a}hler manifolds, the number of the nilpotent scalar supercharges become doubled as we will see\footnote{This property was explored first by Witten \cite{Witten4} in the case of ${\cal N}=2$ SYM theory.}. A four-dimensional K\"{a}hler manifold has the reduced holonomy group $U(1)_L\times SU(2)_R,$ where $U(1)_L$ is a certain subgroup of $SU(2)_L.$ In this case, twisting is done by replacing the $U(1)_L$ by the diagonal sum $U(1)'_L=U(1)_L\oplus U(1)_{F'}.$ Here, $U(1)_{F'}$ is a subgroup of $SU(2)_{F'}.$ We assign the $U(1)_L$ charge $+1$ to $Q_{\alpha=1}^v$ and $-1$ to $Q_{\alpha=2}^v,$ and give a similar assignment of the $U(1)_{F'}$ charge ($+1$ to the $j=1$ index and $-1$ to the $j=2$). Then by twisting, the four $Q_{\alpha}^v$'s in eqs. (\ref{Qrelation}) are to have zero $U(1)'_L$ charge, that is, they all become scalar supercharges, since $Q_1^{i1}=Q_{12}^i$ and $Q_2^{i2}=-Q_{21}^i.$ It is clear that they are nilpotent. Note that one of the four charges is a (chiral) generator of the manifest ${\cal N}=1$ supersymmetry in eq. (\ref{4DN=4}). We will use that charge, denoted by $Q,$ to construct a cohomological field theory in the next subsection. Since the twisting changes the coupling of the fields with the spin connection \cite{Lab-Marino1}, for a general four-manifold the twisted theory differs from the original theory. However, on a manifold with the flat metric ($T^4$) or more generally on a hyper-K\"{a}hler manifold for which the twisting is a trivial operation owing to its holonomy group $SU(2)_R,$ the twisted theory coincides with the original theory. Now we will investigate the $T^4$ case. \subsection{Partition function of ${\cal N}=4$ SYM on $T^4$} The ${\cal N}=4$ $SU(N)$ SYM theory defined on $T^4$ can be seen as a cohomological field theory by (trivial) twisting, because of the reason mentioned above. In the action $S_{{\cal N}=4}$ (\ref{4DN=4}), since $Q$ acts as $\frac{\partial}{\partial\theta^{\alpha}}=\int d\theta^{\alpha}$ (up to total derivative terms) for suitable $\alpha,$ the term with $|_{\theta\theta}$ are written in the $Q$-exact form $\{Q,\cdots\}.$ Also, the term ${\rm tr}\bar{W}_{\dot{\alpha}} \bar{W}^{\dot{\alpha}}|_{\bar{\theta}\bar{\theta}}$ is equal to ${\rm tr} W^{\alpha}W_{\alpha}|_{\theta\theta}$ modulo the total derivative term ${\rm tr} F\wedge F,$ so it is of the $Q$-exact form. Here we consider the periodic boundary condition case on $T^4,$ or equivalently the sector of zero 't Hooft discrete magnetic flux, those total derivative terms can be discarded. The term $ {\rm tr} \;\Phi_1^{\dagger}[\Phi_3^{\dagger},\Phi_2^{\dagger}] |_{\bar{\theta}\bar{\theta}} $ can not be written as the $Q$-exact form, but it is $Q$-invariant because $Q$ is a part of the manifest ${\cal N}=1$ supersymmetric transformation. Thus the action can be written as \begin{equation} S_{{\cal N}=4}=\{Q,\cdots\}+{\cal O}^{(0)}, \end{equation} where \begin{equation} {\cal O}^{(0)}= \int d^4x \frac{1}{\sqrt 2g^2}{\rm tr} \; \Phi_1^{\dagger}[\Phi_3^{\dagger},\Phi_2^{\dagger}] |_{\bar{\theta}\bar{\theta}} \end{equation} is a $Q$-invariant operator and the superscript 0 stands for the $U(1)$-charge defined by the rotation \begin{eqnarray} \Phi_s(\theta,\bar{\theta},x) & \rightarrow & e^{\frac 23i\beta} \Phi_s(e^{-i\beta}\theta,e^{i\beta}\bar{\theta},x) \nonumber \\ W^{\alpha}(\theta,\bar{\theta},x) & \rightarrow & e^{i\beta}W^{\alpha}(e^{-i\beta}\theta,e^{i\beta}\bar{\theta},x) \label{U(1)charge} \end{eqnarray} with $\beta$ being a real parameter. The charge of the component fields can be read off from eqs. (\ref{U(1)charge}) as $-\frac 23,\;+\frac 13,\;+\frac 43,\;0,\;-1,\;0$ for $B_s,\;\psi_s,\;F_s,\;A_m,\;\lambda,\;D,$ respectively. Note that the action has the symmetry under this rotation. Now the problem can be made more tractable by considering the following mass perturbation \begin{equation} \Delta S = \int d^4x \frac{1}{\sqrt 2 g^2} \left(-\frac m2\right) {\rm tr} \; [(\Phi_1^2+\Phi_2^2+\Phi_3^2)|_{\theta\theta} +(\Phi_1^{\dagger 2}+\Phi_2^{\dagger 2}+\Phi_3^{\dagger 2}) |_{\bar{\theta}\bar{\theta}}]. \label{masspert} \end{equation} Using the same argument as above, the first term is written in the form $\{Q,\cdots\},$ and the second term is a $Q$-invariant operator with the $U(1)$-charge $-\frac 23$ denoted by ${\cal O}^{(-2/3)}.$ In cohomological field theories, the perturbation by the $Q$-exact operators does not alter the theory, and the partition function is invariant under the perturbation of ${\cal O}^{(-2/3)}$ if the $U(1)$-symmetry remains left even in the quantum level. In fact, it is so since there is no anomaly in ${\cal N}=4$ theory. Thus we can obtain the answer by computing the partition function of the mass perturbed system. Here, it is remarked that the $U(1)$-symmetry is different from the ghost number symmetry usually used in cohomological field theories. In fact, it is seen that in the cohomological field theory made from ${\cal N}=4$ SYM theory the ghost number symmetry corresponds to the following rotation \cite{Vafa-Witten,Lab-Marino2} \begin{eqnarray*} \Phi_1(\theta,\bar{\theta},x) & \rightarrow & \Phi_1(e^{-i\beta}\theta,e^{i\beta},x) \\ \Phi_2(\theta,\bar{\theta},x) & \rightarrow & \Phi_2(e^{-i\beta}\theta,e^{i\beta},x) \\ \Phi_3(\theta,\bar{\theta},x) & \rightarrow & e^{2i\beta}\Phi_3(e^{-i\beta}\theta,e^{i\beta},x) \\ W^{\alpha}(\theta,\bar{\theta},x) & \rightarrow & e^{i\beta}W^{\alpha}(e^{-i\beta}\theta,e^{i\beta}\bar{\theta},x). \end{eqnarray*} Vafa and Witten \cite{Vafa-Witten} discussed the invariance of the partition function under the mass perturbation based on this ghost number symmetry. Our argument presents another proof of the invariance. We should remark that cohomological field theories have an important feature that the contribution of the path integration localizes in configurations of the classical vacua. In the mass perturbed system, the classical vacua are given by the solutions of these equations: \begin{eqnarray} & & [ B_1, B_2]=m B_3, \nonumber \\ & & [ B_2, B_3]=m B_1, \nonumber \\ & & [ B_3, B_1]=m B_2 \label{classicalvacua1} \end{eqnarray} with \begin{equation} \sum_{s=1}^3[B_s,B_s^{\dagger}]=0. \label{classicalvacua2} \end{equation} A special property of the perturbation is that three equations in (\ref{classicalvacua1}) have the form of $SU(2)$ algebra. So, the solution $\frac im B_s$ is given by the generator of $N$-dimensional representation of $SU(2).$ At the same time, this solution satisfies eq. (\ref{classicalvacua2}) also, because $B_s$'s are anti-hermitian. Since we must consider the various $N$-dimensional representations, both of reducible and irreducible, in general the solution takes the form \begin{equation} \frac im B_s = \left[ \begin{array}{cccc} L_s^{(a_1)} & & & \\ & L_s^{(a_2)} & & \\ & & \ddots & \\ & & & L_s^{(a_l)} \end{array} \right], \label{generalsol} \end{equation} where $L_s^{(a)}$ is a generator of $a$-dimensional irreducible representation of $SU(2),$ and $a_1+a_2+\cdots+a_l=N.$ We remark that all the solutions of the form (\ref{generalsol}) do not contribute to the partition function, but only the $a_1=a_2=\cdots=a_l$ case does. In fact, considering the $l=2$ case, it is easy to see that the unbroken gauge group of the solution (\ref{generalsol}) contains a following $U(1)$-generator \begin{equation} \left[ \begin{array}{cc} e_1{\bf 1}_{a_1} & 0 \\ 0 & e_2{\bf 1}_{a_2} \end{array} \right], \end{equation} where $e_1$ and $e_2$ are real parameters satisfying $e_1a_1+e_2a_2=0.$ Let us consider the $a_1\neq a_2$ case. The interactions are all in the form of commutators, so this $U(1)$ mode is free and massless, which implies that there exists a fermion zero-mode corresponding to the $U(1).$ Thus its contribution to the partition function vanishes. However, when $a_1=a_2,$ the $B_s$ is written as \begin{equation} \frac im B_s={\bf 1}_2\otimes L_s^{(a)}, \end{equation} where we put $a_1=a_2\equiv a.$ Then it is noted that the unbroken gauge group is enhanced to $SU(2)\otimes {\bf Z}_a.$ This mode is not free and thus can give a certain contribution to the partition function, which amounts to a ${\cal N}=1$ $SU(2)\otimes {\bf Z}_a$ SYM theory. For the case of generic $l,$ a similar argument goes on and it can be seen that the vacuum which can contribute to the partition function is a ${\cal N}=1$ $SU(l)\otimes {\bf Z}_a$ SYM theory with $al=N.$ In the consequence the partition function of the ${\cal N}=4$ theory is represented as a sum of the partition function of the various ${\cal N}=1$ theories: \begin{equation} Z_{SU(N)}^{D=4,\; {\cal N}=4}(T^4)= \sum_{al=N}Z_{SU(l)\otimes {\bf Z}_a}^{D=4,\;{\cal N}=1}(T^4). \label{N=4partitionfunction1} \end{equation} Further, since the ${\bf Z}_a$ factor in the gauge group $SU(l)\otimes {\bf Z}_a$ implies a summation of the flat ${\bf Z}_a$ bundle, it yields a factor $a^{4-1}=a^3,$ where the power 4 comes from summing up the flat bundle for each $A_m$'s and the ($-1$) from dividing by the gauge group ${\bf Z}_a.$ Thus we have \begin{equation} Z_{SU(N)}^{D=4,\; {\cal N}=4}(T^4)= \sum_{al=N}a^3Z_{SU(l)}^{D=4,\; {\cal N}=1}(T^4). \label{N=4partitionfunction2} \end{equation} Here we should note that the partition function of the ${\cal N}=1$ theory in the right hand side is given by its Witten index. In the Hamiltonian formalism, the partition function is written as \begin{equation} {\rm Tr}(-1)^Fe^{-\beta H}, \label{trform} \end{equation} where the $(-1)^F$-factor is included in order to impose the periodic boundary condition on the fermion fields. Recall that there is no Higgs field in the ${\cal N}=1$ theory, which implies that there exists no continuous zero-mode in the theory on $T^4.$ Spectra appearing in the theory are all discrete. Then in eq. (\ref{trform}), the contribution of every supersymmetric pair with non-zero energy is precisely cancelled, and only the zero-energy states can contribute. So eq. (\ref{trform}) is independent of $\beta,$ and it coincides with the Witten index \begin{equation} I\equiv\lim_{\beta\rightarrow\infty}{\rm Tr}(-1)^Fe^{-\beta H}, \end{equation} whose value of the ${\cal N}=1$ $SU(l)$ SYM theory is known to be $l$ \cite{Witten5}. Plugging the above formulas, we obtain the answer \begin{equation} Z_{SU(N)}^{D=4,\; {\cal N}=4}(T^4)=\sum_{al=N}a^3l=N\sum_{a|N}a^2, \end{equation} where the summation of $a$ in the right hand side is taken over the divisors of $N.$ Also, for the gauge group $SU(N)/{\bf Z}_N$ we can obtain the partition function of the sector of zero 't Hooft magnetic flux applying the above argument to the ${\bf Z}_N$-factor \begin{equation} Z_{SU(N)/{\bf Z}_N}^{D=4,\; {\cal N}=4}(T^4)= \frac{1}{N^3}Z_{SU(N)}^{D=4,\; {\cal N}=4}(T^4)= \sum_{a|N}\frac{1}{a^2}. \end{equation} \section{IIA Matrix String Partition Function} Here we calculate the partition function of IIA matrix string theory and compare the result of Kostov and Vanhove \cite{Kostov-Vanhove} which has been derived in the case of the strong coupling limit $g\rightarrow\infty.$ We have seen in section 2 that the SYM fields in periodic boundary condition can reproduce the second quantized superstrings with interactions. Thus, considering the two-dimensional SYM theory with periodic boundary condition as the IIA matrix string theory, we evaluate the partition function of this theory. Now we have to take care of the $U(1)$ part of the gauge group $U(N)$ before doing the calculation. We consider the gauge group $U(N)$ in the factorized form \begin{equation} U(N)=U(1)\times (SU(N)/{\bf Z}_N). \label{decomposeU(1)} \end{equation} The meaning of the $U(1)$ part in the IIA matrix string theory, whose field contents are $(A_{\mu},X,\theta),$ is as follows. The $U(1)$ part of $X$ and $\theta$ represents the center of mass coordinates of the strings in transverse directions. We fix it from the translational invariance. On the other hand, the $U(1)$ part of $A_{\mu}$ is related to the number of D-particles, so we have to take into account this. More precisely, as is discussed in \cite{DVV}, the $U(1)$ electric flux corresponds to the number of D-particles: \begin{equation} q=\frac{1}{2\pi}\int_0^Rd\sigma E^{U(1)}\in {\bf Z}. \end{equation} Corresponding to eq. (\ref{decomposeU(1)}), the gauge field $A_{\mu}$ is decomposed as \begin{equation} A_{\mu}=A^{U(1)}_{\mu}T^{U(1)}+A_{\mu}^aT^a, \end{equation} where $T^{U(1)}=\frac 1N{\bf 1}_N,$ and $T^a$ ($a=1,\cdots,N^2-1$) is a generator of $SU(N).$ Since the interaction in the SYM theory appears in the form of commutators, the $U(1)$ gauge part decouples and thus the partition function of the IIA matrix string theory becomes \begin{equation} Z_{{\rm IIA-MS}}=\left(\int \frac{{\cal D}A^{U(1)}_{\mu}}{{\rm Vol}(U(1))} e^{-S^{U(1)}}\right)Z_{SU(N)/{\bf Z}_N}^{D=2,\;{\cal N}=8}(T^2), \label{ZIIAMS} \end{equation} where the Wick rotation was performed. The two-torus, where the SYM theory is defined on, is a rectangular one with the size $T\times R.$ The action of the $U(1)$ gauge part is given by $$ S^{U(1)}=\frac{1}{Ng^2}\int d^2\sigma\frac 14 F^{U(1)}_{\mu\nu}F^{U(1)}_{\mu\nu}. $$ Let us evaluate the first factor in eq. (\ref{ZIIAMS}) by taking the $A^{U(1)}_0=0$ gauge. Then, the Gauss law constraint means that $A^{U(1)}_{\sigma}$ is independent of $\sigma.$ Also, considering the Wilson loop wrapping around the $\sigma$ direction, we see that $\theta(\tau)\equiv RA^{U(1)}_{\sigma}$ is an angular variable whose conjugate momentum $p$ is quantized to an integer. Employing the variable $\theta(\tau)$ and translating to the Hamiltonian form, we can compute the first factor as \begin{eqnarray} \int \frac{{\cal D}A^{U(1)}_{\mu}}{{\rm Vol}(U(1))}\;e^{-S^{U(1)}} & = & \int_{\theta(T)=\theta(0)}{\cal D} \theta (\tau) \;e^{-\frac 12 \frac{1}{Ng^2R}\int_0^Td\tau \dot{\theta}(\tau)^2} \nonumber \\ & = & {\rm Tr} \left( e^{-T\frac 12 Ng^2Rp^2}\right) \nonumber \\ & = & \sum_{p\in {\bf Z}}e^{-\frac{RT}{2}Ng^2p^2}. \label{firstfactor} \end{eqnarray} \subsection{$Z_{SU(N)/{\bf Z}_N}^{D=2,\;{\cal N}=8}(T^2)$} Now our remaining task is evaluating the second factor of eq. (\ref{ZIIAMS}), which is performed by considering the dimensionally reduced version of the analysis in section 3\footnote{ Since in section 3 the supersymmetry algebra (\ref{***}) has no central charges, here we are to consider the ${\cal N}=8$ supersymmetry algebra without central charges as the result of the dimensional reduction. In fact, the central charges are written in the form of total derivatives, so they do not appear under the periodic boundary condition. In the infinite two-dimensional space, central charges exist and represent the topological charges which characterize stable solitonic modes in the theory. On the contrary, in the two-torus with the finite size, such solitonic modes do not exist stably. In this case, it can be considered that with respect to the sector of the zero total charge, dynamical degrees of freedom of the modes are contained in the theory, i.e. in the configurations which the path integration is performed over, and that they arise as metastable states when the sizes of the two-torus becomes large enough. Of course, a similar consideration is possible in the three-torus case in section 5. }. Since the $Q$-exact structure as well as the $Q$-invariant one is preserved even after the dimensional reduction, the dimensionally reduced theory of a cohomological field theory becomes also a cohomological field theory. Further, the $U(1)$-symmetry in (\ref{U(1)charge}) remains nonanomalous in the dimensional reduction to two-dimensions in the case of the $SU(N)$ gauge group, because it contains no $U(1)$ factor. Thus, the arguments of the mass perturbation in section 3.2 can be applied also to the dimensionally reduced case. Then, the mass perturbation breaks the ${\cal N}=8$ supersymmetry to ${\cal N}=2,$ and the partition function takes the form \begin{equation} Z_{SU(N)}^{D=2, \;{\cal N}=8}(T^2)= \sum_{al=N}Z_{SU(l)\otimes {\bf Z}_a}^{D=2, \;{\cal N}=2}(T^2). \label{D=2N=8partitionfunction1} \end{equation} In two-dimensions, by the argument similar as in the four-dimensional case, the ${\bf Z}_a$ factor of the gauge group yields $a^{2-1}=a.$ Hence we have \begin{equation} Z_{SU(N)}^{D=2,\; {\cal N}=8}(T^2)= \sum_{al=N}aZ_{SU(l)}^{D=2,\; {\cal N}=2}(T^2). \label{D=2N=8partitionfunction2} \end{equation} Here the ${\cal N}=2$ theory in two dimensions contains Higgs fields, whose zero momentum modes form continuous spectrum beginning with zero-energy because the field space of the Higgs fields is noncompact. This situation makes ambiguous the relation between the partition function and the Witten index, so we cannot go along the same line as in the four-dimensional case. However, owing to the fact that ${\cal N}=2$ $SU(l)$ SYM theory in two dimensions is a cohomological field theory, the partition function $Z_{SU(l)}^{D=2,\; {\cal N}=2}(T^2)$ can be evaluated. Let us see it from now. After an appropriate field redefinition (see Appendix A), the classical action of the ${\cal N}=2$ SYM theory is written as the BRST exact form \begin{equation} S=Q\int d^2\sigma\;{\rm tr}\left(\frac{1}{8g^2}\eta[\phi,\bar{\phi}] -i\chi\Phi+2g^2\chi H+\frac{1}{2g^2}\psi_{\mu}D_{\mu}\bar{\phi} \right), \label{N=2TFT} \end{equation} where the BRST transformation is defined by \begin{equation} \begin{array}{lll} QA_{\mu}=\psi_{\mu}, & Q\psi_{\mu}=-iD_{\mu}\phi, & Q\phi=0,\\ Q\chi=H, & QH=[\phi, \chi], & \\ Q\bar{\phi}=\eta,& Q\eta=[\phi, \bar{\phi}]. & \end{array} \end{equation} Note that the $Q$ is nilpotent up to the gauge transformation with the parameter $\phi,$ which gives a cohomology to equivalent classes with respect to the gauge transformation. The field contents are as follows. $A_{\mu}$ is a two-dimensional gauge field, and $\psi_{\mu},$ $\eta,$ $\chi$ together stand for fermions. $\phi$ and $\bar{\phi}$ are complex Higgs fields, and $H$ is a bosonic auxiliary field. Ghost number is assigned as $-2$ to $\bar{\phi},$ $-1$ to $\eta$ and $\chi,$ $0$ to $A_{\mu}$ and $H,$ +1 to $\psi_{\mu},$ +2 to $\phi.$ The ghost number conservation is nonanomalous for the same reason as in the case of the dimensional reduction of the $U(1)$-symmetry. The contribution of the path integration of the gauge field localizes in the configurations determined by \begin{equation} \Phi\equiv -2F_{12}=0. \label{Phiequation} \end{equation} The addition of a $Q$-exact term to the action does not change the theory, if it behaves well at infinity in the field space. Thus, we may discard the first term in eq. (\ref{N=2TFT}) \begin{equation} Q\int d^2\sigma\;{\rm tr}\frac{1}{8g^2}\eta[\phi,\bar{\phi}]. \label{firstterm} \end{equation} Also, it can be seen that the partition function is independent of the coupling $g$ for the same reason. The localization (\ref{Phiequation}) can be shown by integrating out $H$ and $\chi$ fields in the $g\rightarrow 0$ limit after the integrals of $\bar{\phi}$ and $\eta.$ Then the partition function becomes \begin{equation} \int\frac{{\cal D}A_{\mu}}{{\rm Vol}(SU(l))} {\cal D}\phi{\cal D}\psi_{\mu} \;\delta(D_{\mu}D_{\mu}\phi+\{\psi_{\mu},\psi_{\mu}\}) \;\delta(D_{\mu}\psi_{\mu}) \;\delta(D_1\psi_2-D_2\psi_1)\;\delta(F_{12}), \label{localization} \end{equation} which indicates the declared localization. It should be noted that the localization is determined by the BRST fixed point $Q\chi=0$ after using the equation of motion of $H.$ Eq. (\ref{localization}) is not in a suitable form for our purpose, so we will deform the theory judiciously as in the section 3 in ref. \cite{Witten6}. We consider the action with the addition of the $Q$-exact term \begin{equation} S(t)=S'+tQ\int d^2\sigma\;{\rm tr}\chi\bar{\phi}, \end{equation} where $S'$ stands for the action $S$ with the term (\ref{firstterm}) eliminated and $t$ is a parameter. If new BRST fixed points that flow in from infinity when $t$ turns on do not contribute, the deformed theory coincides with the original one. After integrating out $H,$ $\eta,$ $\chi$ and $\bar{\phi},$ we end up with the action in the large $t$ case \begin{equation} S(t)= \frac{1}{2g^2t}\int d^2\sigma\;{\rm tr} [F_{12}(D_{\mu}D_{\mu}\phi+\{\psi_{\mu},\psi_{\mu}\}) -iD_{\mu}\psi_{\mu}(D_1\psi_2-D_2\psi_1)]+O(t^{-2}), \label{S(t)1} \end{equation} which can be again written in the $Q$-exact form \begin{equation} S(t)=Q\left[\frac{i}{2g^2t} \int d^2\sigma\;{\rm tr} F_{12}D_{\mu}\psi_{\mu}+O(t^{-2})\right]. \label{S(t)2} \end{equation} Considering the case $t=-iu$ with $u$ large real positive, the $\phi$-integration yields $\delta (D_{\mu}D_{\mu}F_{12}).$ Using the normalizability of $F_{12},$ it means that the localization realizes at the solutions of $D_{\mu}F_{12}=0,$ which contain extra components adding to the localization point of the original theory $F_{12}=0.$ Arising of the extra components is a signal of the flow of new BRST fixed points from infinity. In fact, in the above process, the contribution of the $\chi$ and $\bar{\phi}$ integrals localizes the points \begin{eqnarray*} \chi & = & \frac{1}{2t}D_{\mu}\psi_{\mu}, \\ \bar{\phi} & = & -\frac 1t F_{12}+\frac{1}{2t^2}(D_{\mu}D_{\mu}\phi +\{\psi_{\mu},\psi_{\mu}\}), \end{eqnarray*} which appear first when $t$ turns on. So they are the new fixed points flowing in from infinity. Thus in general, the deformed cohomological field theory does not equivalent to the original one. However, there is a possibility that the BRST invariant operators with the following feature exist --- in calculation of their correlators the extra components do not contribute. If there are such operators, the deformed theory coincides to the original one with respect to the restricted set of the operators. Indeed, we can find such operators. For the following BRST invariant operators \begin{eqnarray} & & \omega\equiv \int d^2\sigma \; {\rm tr} \;(-i\phi F_{12}+\psi_1\psi_2), \\ & & \beta (\phi) = {\rm tr}\;(\;{\rm polynomial\; of \;}\phi), \nonumber \end{eqnarray} we consider the unnormalized expectation value in the deformed theory \begin{equation} \left\langle e^{\omega}\beta(\phi)\right\rangle'\equiv \int\frac{{\cal D}A_{\mu}}{{\rm Vol}(SU(l))}{\cal D}\phi{\cal D}\psi_{\mu} \;\beta(\phi)\;e^{-S(-iu)+\omega}. \label{expectationvalue} \end{equation} Here due to the $e^{\omega}$ factor, we can take the limit $u=\infty$ without changing the behavior of the fields at the infinity, and thus the limit does not change the value of (\ref{expectationvalue}). Further integrating out $\phi$ and $\psi_{\mu},$ we end up with \begin{equation} \left\langle e^{\omega}\beta(\phi)\right\rangle'= \int\frac{{\cal D}A_{\mu}}{{\rm Vol}(SU(l))}\; \beta\left(-i\frac{\delta}{\delta F_{12}}\right)\;\delta(F_{12}), \label{expectationvalue2} \end{equation} where the factor $\delta(F_{12})$ (with a finite degree of the derivative $\frac{\delta}{\delta F_{12}}$) indicates no contributions of the extra components. Therefore, $\left\langle e^{\omega}\beta(\phi)\right\rangle'$ coincides with the unnormalized expectation value in the original theory which we denote by $\left\langle e^{\omega}\beta(\phi)\right\rangle.$ Now we can manage to compute the partition function $Z_{SU(l)}^{D=2,\; {\cal N}=2}(T^2)=\langle 1\rangle.$ Notice that $\omega$ has the ghost number $+2,$ and thus from the ghost number conservation we can show $\langle 1\rangle=\langle e^{\omega}\rangle.$ This is equal to the $\beta=1$ case of eq. (\ref{expectationvalue2}), so we find \begin{equation} Z_{SU(l)}^{D=2,\; {\cal N}=2}(T^2)=\int \frac{{\cal D}A_{\mu}}{{\rm Vol}(SU(l))}\; \delta(F_{12}), \label{partitionfunction3} \end{equation} which counts the number of the small gauge inequivalent configurations satisfying $F_{12}=0.$ The number is unity, since the two-dimensional $SU(l)$ gauge theory has no nontrivial winding number. In the consequence we have \begin{equation} Z_{SU(l)}^{D=2,\; {\cal N}=2}(T^2)=1. \end{equation} It can be confirmed also by doing the concrete calculation, for example, employing the $A_1=0$ gauge fixing in eq. (\ref{partitionfunction3}). Plugging this into eq.(\ref{D=2N=8partitionfunction2}) we get the result \begin{equation} Z_{SU(N)}^{D=2,\; {\cal N}=8}(T^2)=\sum_{al=N}a. \end{equation} Finally by taking account into the ${\bf Z}_N$ factor as in the four dimensional case, the second factor $Z_{SU(N)/{\bf Z}_N}^{D=2,\;{\cal N}=8}(T^2)$ of eq. (\ref{ZIIAMS}) is obtained as \begin{equation} Z_{SU(N)/{\bf Z}_N}^{D=2,\;{\cal N}=8}(T^2)= \frac 1N Z_{SU(N)}^{D=2,\; {\cal N}=8}(T^2) =\sum_{a|N}\frac 1a. \label{resultsecondfactor} \end{equation} \subsection{Result of IIA Matrix String Partition Function} Now we can write down the result of the partition function of the IIA matrix string theory. Substituting eqs. (\ref{firstfactor}) and (\ref{resultsecondfactor}) into eq. (\ref{ZIIAMS}), we have \begin{equation} Z_{{\rm IIA-MS}}=\left(\sum_{a|N}\frac 1a\right)\sum_{p\in{\bf Z}} e^{-\frac{RT}{2}Ng^2p^2}, \label{IIAresult1} \end{equation} which coincides the result obtained in the strongly coupled limit ($g^2\rightarrow \infty$) by Kostov and Vanhove \cite{Kostov-Vanhove}. They have conjectured that their result holds irrespectively of the strength of the coupling, and they called this property {\it exact quasi classics.} Since our calculation has been exactly performed without any approximation, it gives a proof of the {\it exact quasi classics.} Recalling the relations (\ref{RLrelation}) and (\ref{IIAggs}), we rewrite the result (\ref{IIAresult1}) in variables in string theory \begin{equation} Z_{{\rm IIA-MS}}=\left(\sum_{a|N}\frac 1a\right)\sum_{p\in{\bf Z}} e^{-T\frac{NL}{4\pi}\frac{p^2}{(g_sl_s)^2}}. \label{IIAresult2} \end{equation} There are two comments in order. First, the second factor in (\ref{IIAresult2}) represents a certain nonperturbative effect, which corresponds to creation and annihilation of D-particle and anti-D-particle pairs. Such phenomena as creation/annihilation of D- and anti-D- objects cannot be seen in the M(atrix) theory, because in the infinite momentum frame anti-D-particles in the M(atrix) theory are integrated out and do not appear. It can be seen first after compactified to the IIA matrix string theory. Second, there is no perturbative correction in the formula (\ref{IIAresult2}). It agrees to nonrenormalization theorems in perturbative superstring theory by Martinec \cite{Martinec}, which tells that the 0-, 1-, 2-, and 3-point functions of massless string vertex operators receive no perturbative corrections in flat ten-dimensional backgrounds. \section{IIB Matrix String Partition Function} Here, we compute the partition function of IIB matrix string theory in the ten-dimensional IIB limit, and compare the exact result of the IKKT model by Moore-Nekrasov-Shatashvili \cite{Moore-Nekrasov-Shatashvili}. We consider the three-dimensional SYM theory with periodic boundary condition as the IIB matrix string theory for the same reason as in the IIA case. With respect to the gauge group, the $U(1)$ part of the gauge group $U(N)=U(1)\times (SU(N)/{\bf Z}_N)$ for the field contents $(A_{\mu}, X, \theta)$ corresponds to the center of mass coordinates of the strings in transverse directions in the ten-dimensional IIB limit ``$R_1R_2\rightarrow\infty$ with $g_s$ fixed,'' where the $U(1)$ part of the gauge field together with that of $X$ become the transverse coordinates. Thus we fix the $U(1)$ part of the gauge field as well as the Higgs fields, and consider the partition function of the three-dimensional ${\cal N}=8,$ $SU(N)/{\bf Z}_N$ SYM theory as that of the IIB matrix string theory \begin{equation} Z_{{\rm IIB-MS}}=Z_{SU(N)/{\bf Z}_N}^{D=3, \; {\cal N}=8}(T^3). \label{ZIIBMS} \end{equation} The three-torus, where the SYM theory is defined on, is taken to be rectangular of the size $T\times R_1 \times R_2,$ with the Euclidean signature. (We performed the Wick rotation as in the IIA case.) In this case, we can also use the dimensionally reduced version of the arguments of the mass perturbation in the four dimensions, due to the following two reasons. One is that the dimensional reduction of a cohomological field theory is also a cohomological field theory. The other is that in odd dimensions the dimensionally reduced version of the $U(1)$-symmetry in eq. (\ref{U(1)charge}) is anomaly free. Going on along the same line as in the IIA case, we have \begin{eqnarray} Z_{SU(N)}^{D=3,\; {\cal N}=8}(T^3) & = & \sum_{al=N}Z_{SU(l)\otimes{\bf Z}_a}^{D=3, \; {\cal N}=2}(T^3) \nonumber \\ & = & \sum_{al=N}a^2Z_{SU(l)}^{D=3, \; {\cal N}=2}(T^3). \label{ZD=3N=8SYM} \end{eqnarray} Here, there exists a Higgs field in the three-dimensional ${\cal N}=2$ theory. For the same reason as before, we cannot relate the partition function directly to the Witten index. However, if considering the ten-dimensional IIB limit (\ref{10DIIBlimit}) which in fact we are interested in, we can proceed further. Note that in the large volume limit physics becomes independent of the boundary condition. So we can evaluate $Z_{SU(l)}^{D=3, \; {\cal N}=2}(T^3)$ by adopting a twisted boundary condition instead of the periodic boundary condition. \subsection{${\cal N}=2$ partition function with a twisted boundary condition} Here we consider the three-dimensional ${\cal N}=2$ $SU(l)$ SYM theory with the twisted boundary condition \begin{eqnarray} A_i(t,\sigma_1,\sigma_2) & = & A_i(t+T, \sigma_1,\sigma_2) \nonumber \\ & = & PA_i(t,\sigma_1+R_1, \sigma_2)P^{-1} \nonumber \\ & = & QA_i(t,\sigma_1, \sigma_2+R_2)Q^{-1}, \label{twistedBC} \end{eqnarray} where $i=1,2,$ we took the $A_0=0$ gauge fixing, and $P$ and $Q$ are $SU(l)$ matrices satisfying $PQ=QPe^{2\pi i/l}.$ For example, $P$ and $Q$ can be represented as \begin{equation} P=e^{-\pi i\frac{l+1}{l}}\left[\begin{array}{ccccc} 0 & 1 & & & \\ & 0 & 1 & & \\ & &\ddots & \ddots & \\ & & & \ddots & 1 \\ 1 & & & & 0 \end{array}\right], \hspace{1cm} Q=e^{-\pi i\frac{l-1}{l}}\left[\begin{array}{ccccc} 1 & & & & \\ & e^{2\pi i/l}& & & \\ & & e^{4\pi i/l} & & \\ & & & \ddots& \\ & & & & e^{2\pi i(l-1)/l} \end{array}\right]. \end{equation} Supersymmetry requires that the other fields (Higgs $\phi$ and complex fermion $\lambda$) satisfy the same boundary condition as eq. (\ref{twistedBC}). Under this boundary condition, zero momentum modes become trivial \begin{equation} A_i^{(0)}=\phi^{(0)}=\lambda^{(0)}=0, \label{****} \end{equation} because the constant traceless hermitian matrices commuting with $P$ and $Q$ simultaneously do not exist except the trivial case (\ref{****}). In this situation, spectra appearing in the theory are discrete, and thus the partition function coincides the Witten index. Let us consider the Witten index. The argument below is a three-dimensional analogue of Witten's consideration in the four-dimensional case \cite{Witten6}. Also, though it is briefly reported in \cite{Affleck-Harvey-Witten}, we will discuss it in order to make this paper more self-contained. Our problem is now reduced to counting the number of vacua, i.e. the number of (small) gauge inequivalent classes of the classical zero energy states. The gauge transformation preserving both of the boundary condition (\ref{twistedBC}) and the gauge condition $A_0=0$ is generated by the time-independent $SU(l)$ matrix of the following boundary condition \begin{eqnarray} U(\sigma_1,\sigma_2) & = & e^{2\pi ik_1/l}PU(\sigma_1+R_1,\sigma_2)P^{-1} \nonumber \\ & = & e^{2\pi ik_2/l}QU(\sigma_1,\sigma_2+R_2)Q^{-1} \label{UBC} \end{eqnarray} where $k_i=0,1,\cdots,l-1.$ The classical zero energy state is given by the configuration $$ A_i=-i(\partial_iU)U^{-1} $$ with the other fields nil. Here, if this $U$ can be continuously deformed to the identity, there are no nontrivial sectors, and thus the vacuum is unique modulo small gauge transformations. We will see that it is in fact so. The $U$ can be written by the $SU(l)$ matrix $\tilde{U}$ satisfying the simpler boundary condition \begin{equation} U=(Q^{-1})^{k_1}P^{k_2}\tilde{U}, \label{UUtilde} \end{equation} where \begin{eqnarray} \tilde{U}(\sigma_1,\sigma_2) & = & P\tilde{U}(\sigma_1+R_1,\sigma_2)P^{-1} \nonumber \\ & = & Q\tilde{U}(\sigma_1,\sigma_2+R_2)Q^{-1}. \label{UtildeBC} \end{eqnarray} We consider a topological classification of gauge transformations with the boundary condition (\ref{UtildeBC}). Because of $\pi_0(SU(l))=0,$ by a suitable continuous translation in a group manifold of $SU(l),$ we can always start with $\tilde{U}(0,0)=1.$ Then, using eq. (\ref{UtildeBC}) we see that $$ \tilde{U}(0,0)=\tilde{U}(R_1,0)=\tilde{U}(0,R_2)=\tilde{U}(R_1,R_2)=1, $$ i.e. $\tilde{U}$'s on the vertices of a square with the size $R_1\times R_2$ in $(\sigma_1, \sigma_2)$-space are all identity. From the fact $\pi_1(SU(l))=0,$ $\tilde{U}$ on the edges of the square can be taken identity by a continuous deformation. Further, using $\pi_2(SU(l))=0,$ we continuously deform to $\tilde{U}=1$ everywhere in the square. Using $\pi_0(SU(l))=0$ again, by a continuous translation the $U$ in eq. (\ref{UUtilde}) can be taken identity on the square. Thus, it is confirmed that there is no nontrivial topological vacuum sector. We conclude that the Witten index is unity, which leads to \begin{equation} Z_{SU(l)}^{D=3, \; {\cal N}=2}(T^3)|_{{\rm twisted \;B.C.}}=1. \label{ZtwistedBC} \end{equation} \subsection{Result of IIB Matrix String Partition Function} As discussed before, in the ten-dimensional IIB limit we can replace the value of the partition function $Z_{SU(l)}^{D=3, \; {\cal N}=2}(T^3)$ with that of eq. (\ref{ZtwistedBC}) \begin{equation} Z_{SU(l)}^{D=3, \; {\cal N}=2}(T^3)=1. \label{ZperiodicBC} \end{equation} Substituting this into eq. (\ref{ZD=3N=8SYM}), we have \begin{equation} Z_{SU(N)}^{D=3,\; {\cal N}=8}(T^3) = \sum_{al=N}a^2, \end{equation} and thus the partition function of the IIB matrix string theory is obtained as \begin{eqnarray} Z_{{\rm IIB-MS}} & = & Z_{SU(N)/{\bf Z}_N}^{D=3, \; {\cal N}=8}(T^3) \nonumber \\ & = & \frac{1}{N^2}Z_{SU(N)}^{D=3,\; {\cal N}=8}(T^3) \nonumber \\ & = & \sum_{a|N}\frac{1}{a^2}, \label{ZIIBMS2} \end{eqnarray} which is valid in the ten-dimensional IIB limit (\ref{10DIIBlimit}). We should remark that the result (\ref{ZIIBMS2}) coincides the result of the IKKT model \cite{IKKT} by Moore, Nekrasov and Shatashvili \cite{Moore-Nekrasov-Shatashvili}. It is quite nontrivial and might be a signal of the equivalence between the two IIB matrix models arising from the different roots --- one is a compactification of the BFSS M(atrix) theory, and the other is a matrix regularization of the worldsheet action of type IIB superstring in Schild gauge. Also, the result (\ref{ZIIBMS2}) has neither perturbative nor nonperturbative correction. With respect to the former it agrees again with the nonrenormalization theorems in perturbative superstring theory \cite{Martinec}. \section{Conclusions} We have considered the IIA and IIB matrix string theories derived from the M(atrix) theory via toroidal compactification. We have summarized that string interactions emerge as the Yang-Mills instantons and have shown that in the IIB matrix string theory the chiral spinors are correctly reproduced. As a preparation for computation of the matrix string partition functions, we have calculated the partition function of four-dimensional SYM theory on $T^4,$ by mapping the theory to a cohomological field theory. Here, considering the mass perturbation (\ref{masspert}) has made the calculation easier. We have shown the invariance of the partition function under the mass perturbation in a different fashion from the argument by Vafa and Witten \cite{Vafa-Witten}. We have exactly computed the partition function of the IIA matrix string theory by mapping the theory into a cohomological field theory. Our result coincides with the result obtained in the infra-red limit by Kostov and Vanhove \cite{Kostov-Vanhove}, and thus gives a proof of the {\it exact quasi classics} of the theory conjectured by them. The formula for the partition function receives no perturbative correction, which is in conformity with nonrenormalization theorems in perturbative superstring theory by Martinec \cite{Martinec}. Also, there exist some nonperturbative corrections which come from the $U(1)$ electric flux in the SYM theory and which are interpreted as creation and annihilation of D-particle and anti-D-particle pairs. Such phenomena have been reported in high energy scattering of strings in ref. \cite{GHV}. It may be interesting to deepen the meaning of our result from the line of the high energy scattering. Further, we have evaluated the partition function of the IIB matrix string theory in the ten-dimensional IIB limit by a similar method as in the IIA case. Our result receives neither perturbative and nonperturbative corrections, which with respect to the former agrees with the nonrenormalization theorems again. Also, our result coincides with the exact result of the partition function of the IKKT model by Moore, Nekrasov and Shatashvili \cite{Moore-Nekrasov-Shatashvili}. Although both of the IIB matrix string theory and the IKKT model are considered to give type IIB string theory, they have arised from the distinctive origins, and the relation between them have not been clarified yet. Thus our result is quite nontrivial, and may suggest the equivalence of those two models. In the IIB matrix string theory, the S-duality is well understood from a geometry of the two-torus, but the ten-dimensional Lorentz symmetry is not manifest. On the other hand, in the IKKT model side, while there is a manifest Lorentz symmetry, we have not been able to see the S-dual structure. In this situation, it seems to be an important step to explore the equivalence and to establish the precise correspondence between them toward constructing the nonperturbative definition which manifestly realizes both of the Lorentz symmetry and the S-dual structure. \vspace{3cm} \begin{large} {\bf Acknowledgements} \end{large} \vspace{7mm} The preliminary version of this work was presented in KEK theory workshop '99. The author would like to thank the organizers and participants of the workshop, and especially N. Ishibashi, S. Iso, H. Kawai, T. Kuroki, Y. Okada, K. Okuyama and A. Tsuchiya for valuable conversations and encouragements. The research of the author is supported by the Japan Society for the Promotion of Science under the Postdoctral Research Program. \vspace{3cm} \begin{large} {\bf Note Added} \end{large} \vspace{7mm} While writing up the manuscript, I received the papers \cite{GS,P} which discuss issues close to this work. In \cite{GS} the thermodynamic partition function of the IIA matrix string theory is calculated in the $g_s\rightarrow 0$ limit, and in \cite{P} some thermodynamic properties of the IIA and IIB matrix string theories are discussed. \newpage \begin{large} {\bf Appendix} \end{large}
\section*{Acknowledgements} We warmly acknowledge W.~Vogelsang for providing us with the NLO-evolved parton densities in the photon, and G.~Ridolfi for discussions. This work was supported in part by the EU Fourth Framework Programme `Training and Mobility of Researchers', Network `Quantum Chromodynamics and the Deep Structure of Elementary Particles', contract FMRX-CT98-0194 (DG 12-MIHT).
\section{INTRODUCTION} The concept of coherent states (CS) is widely used in different fields of physics and mathematics (see for example Refs. \cite{Per} - \cite{KS}). In particularly, it plays an important role in the Berezin quantization scheme \cite{Ber}, in the analysis of growth of functions holomorphic in a complex domain \cite{Vourd}, in a general theory of phase space quasiprobability distributions \cite{BM}, and in a quantum state engineering \cite{Vo}. It is necessary to note that in present no a unified definition of such states exists in the literature and different authors mean different things when speaking about them. Nevertheless, a careful analysis (see for example Ref. \cite{Klaud}) shows that almost all definitions have some common points that can be taken as a general definition. Following Klauder \cite{Klaud} I mean by coherent states such states that satisfy the following defining properties: (1) CS are defined by vectors $\psi _z(x,t)$ which belong to a Hilbert space $H$ of the states of a quantum system with scalar product $\langle \cdot |\cdot \rangle $; (2) The parameter $z$ takes continuous values from a domain ${\cal D}$ of an $n$-dimensional complex space; (3) There exists a measure $\mu =\mu (z,\bar z)$ (the bar over a symbol indicates complex conjugation) that realizes the resolution of the identity operator $\Bbb I$ acting in $H$ in terms of the projectors on the vectors $\psi _z$ \begin{equation} \int _{\cal D} d\mu |\psi _z\rangle \langle \psi _z|= \Bbb I\,; \label{Id} \end{equation} (4)CS have to prove the property of a temporal stability. By temporal stability I mean that the vectors $\psi _z(x,t)$ remain coherent at all times i.e. satisfy the properties 1.-3. at all times. To satisfy this condition I shall assume that the functions $\psi _z(x,t)$ are solutions of the Schr\"o\-din\-ger equation \[(i\partial _t-h_0)\psi _z(x,t) =0 \] where $h_0$ is the Hamiltonian of a given quantum system which in general can depend on time. Operator $h_0$ is supposed to be Hermitian in $H$ and to have a unique self-adjoint extension. The Eq. (\ref{Id}) should be understood in a weak sense. This means that it is equivalent to the following relation \[ \int _{\cal D} d\mu \langle \psi _a|\psi _z\rangle \langle \psi _z|\psi _b \rangle = \langle \psi _a|\psi _b\rangle \] which should hold for all $\psi _{a,b}$ from a dense set in $H$. Transparent potentials have many remarkable properties. For instance, a quantum particle prove no reflection in the scattering process on such a potential. Another remarkable property is that each level in the discrete spectrum of such a potential occupies a preassigned position, which is controlled by values of the parameters the potential depend on. The discrete spectrum levels may even be situated in the middle of the continuous spectrum. In the latter case we have {\it completely transparent potentials} \cite{Stahlh}. Transparent potentials find a more significant application in soliton theory. There is a marvelous vast literature on this subject. I cite here only a monograph \cite{MatvSal}. Because of their remarkable properties transparent potentials would find an application in pseudopotential theories. Recently they have been used to describe relaxation processes in Fermi liquid \cite{VVT}. CS for transparent potentials are very far from being explored. It may be explained by the fact that up to now no systematic way is known for their investigation. No a clear algebraic structure related to these potentials is known and therefore well known algebraic methods \cite{Per} prove to be a little suitable in this context. No simple ladder operators for the discrete spectrum eigenfunctions of transparent potentials are known as well and therefor we can not use the approach of Ref. \cite{MM} for this purpose. An approach based on the uncertainty relation \cite{NiSai0} is not consistent with the property 3. mentioned above and therefor it should be rejected. A conjecture has been advanced recently \cite{BSjetp,NF} to use Darboux transformation operator approach for investigating the CS of those system that is related by Darboux transformation with that for which the CS are known. Let us have an exactly soluble Hamiltonian $h_0=-\partial _x^2 +V_0(x,t)$ for which the CS $\psi _z(x,t)$ are known and we want to obtain the CS for another Hamiltonian $h_1=-\partial _x^2 +V_1(x,t)$ related with $h_0$ by the Darboux transformation operator that I shall denote by $L$. In general it should be a {\it nonstationary} Darboux transformation operator defined by the following intertwining relation \cite{BSPL} \[L(i\partial _t-h_0)=(i\partial _t-h_1)L\, .\] If such an operator $L$ is known then solutions of the transformed Schr\"o\-din\-ger equation (i.e. the Schr\"o\-din\-ger equation with the Hamiltonian $h_1$) can easily be obtained by the action of the operator $L$ on solutions of the initial Schr\"o\-din\-ger equation (i.e. the Schr\"o\-din\-ger equation with the Hamiltonian $h_0$). It is clear that the functions $\varphi _z(x,t) =L\psi _z(x,t)$ will satisfy all the properties of the CS enumerated above except may be for the property 3. One of the main goal of this paper is to prove that in the case of soliton potentials this property is fulfilled. I would like to mention that this approach has been successfully applied to study CS of anharmonic oscillator Hamiltonians with equidistant and quasiequidistant spectra \cite{BSjetp} and CS of nonstationary soliton potentials \cite{jetp98E} that are related with soliton solutions of the Kadomtsev-Petviashvili equation. With the help of this approach a classical counterpart of the Darboux transformation has been formulated and shown that at classical level this transformation leads to a distortion of a phase space \cite{SamJMP}. CS of a one-soliton potential have been investigated as well and their supercoherent structure has been revealed \cite{NF}. In this paper I generalize these results to a multisoliton case. The paper is organized as follows. In the Section 2 I give well-known results for CS of the free particle in preparation for their application in the following sections. In the Section 3 the Darboux transformation operator from the solutions of the Schr\"o\-din\-ger equation with zero potential to the solutions of the same equation with solitons potential is analyzed as an operator acting in the Hilbert space of the states of a free particle. It is shown that it can not realize a mapping of Hilbert spaces since it is not defined in the whole Hilbert space and can not be extended to the whole Hilbert space. Isomeric operators expressed in terms of continuous bases similar to these previously proposed by L.D. ~Faddeev \cite{Fadd} and analyzed by D.L. ~Pursey \cite{Pur} for the case of purely discrete basis sets are introduced. These operators realize a polar decomposition of Darboux transformation operators. A quazispectral representation of the Darboux transformation operator and its inverse in terms of continuous bases are obtained. In the Section 4. different systems of CS are introduced for soliton potentials. It is established that the resolution of the identity operator exists in every case. Explicit expressions for measures that realize this equality are found. A brief conclusion brings a paper to a close. \section{FREE PARTICLE COHERENT STATES} In this section I give a brief overview of well-known properties of the Hilbert space of states a free particle (see Ref. \cite{Mil} and references therein) and corresponding CS \cite{MM} we need for subsequent analysis. Annihilation $a$ and creation $a^+$ operators \[a=(i-t)\partial _x+ix/2,\quad a^+=(i+t)\partial _x-ix/2\] form the Heisenberg-Weil subalgebra of the six-dimensional Schr\"o\-din\-ger algebra which is a symmetry algebra of the equation with zero potential. Solutions of the free particle Schr\"o\-din\-ger equation which are square integrable over full real axis $\Bbb R =(-\infty , +\infty )$ with respect to the Lebesgue measure are the eigenstates of the symmetry operator $K_0=aa^++a^+a$, $K_0\psi _n(x,t) =(2n+1)\psi _n(x,t) $. Their coordinate representation is as follows $$ \begin{array}{rl} \psi _n(x,t) = & (-i)^n(n!2^n\sqrt {2\pi })^{-1/2} (1+it)^{-1/2} \\ \times & \exp \! \left[-in\arctan t+{\textstyle \frac {y^2}2}(it-1)\right] H_n\left( y \right)\, ,\\ y= & {\displaystyle \frac{x}{\sqrt{2+2t^2}}}\, .\vphantom{\left) {\displaystyle \frac AB} \right)} \end{array} $$ Operators $a$ and $a^+$ are the ladder operators for the basis functions $\psi _n$: $a\psi _n=\sqrt n\psi _{n-1}$, $a^+\psi _n=\sqrt{n+1}\psi _{n+1}$, and $a\psi _0=0$. By ${\cal L}_0 $ I denote the lineal of the functions $\psi_n $, $n=0,1,\ldots $ which is the space of all finite linear combinations of the functions $\psi_n $ with the coefficients from the field $\Bbb C$. The operators $a$ and $a^+$ being linear are defined for all elements from ${\cal L}_0 $ and ${\cal L}_0 $ is invariant with respect to the action of these operators. Since the momentum operator $p_x=-i\partial _x$ and the initial Hamiltonian $h_0$ are expressed in terms of $a$ and $a^+$: $p_x=-(a+a^+)/2$, $h_0=p_x^2$, these operators are defined in ${\cal L}_0 $ and map this space into itself. The Hilbert space of the states of the free particle, $H$, is defined as a closure of the lineal ${\cal L}_0 $ with respect to the measure generated by the scalar product $\langle \psi _a|\psi _b\rangle $, $\psi _{a,b}\in {\cal L}_0 $, which is defined by the Lebesgue integral. The functions $\psi _n$ form an orthonormal basis in $H$, $\langle \psi _n|\psi _k\rangle =\delta _{nk}$. It is well-known \cite{Smirn,RS} that the operators $p_x$ and $h_0$ initially defined on ${\cal L}_0 $ have unique self-adjoint extensions and consequently they are essentially self-adjoint in $H$. The spectrum of $h_0$ and $p_x$ is purely continuous. They have common eigenfunctions $\psi _p=\psi _p(x,t)$: $p_x\psi _p=p\psi _p$, $h_0\psi _p=p^2\psi _p$, $p\in \Bbb R$, which do not belong to $H$ but belong to a more wide space $H_-$ of the linear functionals over $H_+$, $H_+\subset H\subset H_-$ (so called Gelfand triplet). We can choose the Hilbert-Schmidt equipment of the space $H$ by letting $H_+=K_0^{-1}H$ since $K_0^{-1}$ is a Hilbert-Schmidt operator. We refer a reader to Refs. \cite{GSh,GV,BerezShub} for more details on the nested Hilbert space. The coordinate representation of the functions $\psi _p(x,t)$ is well-known and I omit it here. The functions $\psi _p$ form an orthonormal and complete (in the sense of generalized functions) basis in $H$, $\langle \psi _p|\psi _q\rangle =\delta (p-q)$. The completeness condition is expressed symbolically in terms of the projectors onto these functions as follows \begin{equation} \int dp|\psi _p\rangle \langle \psi _p |=\Bbb I\, . \label{PI} \end{equation} I do not indicate the limits of integration in the integrals along the whole real axis. This equality should be understood in a weak sense. This means that it is equivalent to \[ \int dp\langle \psi _j|\psi _p\rangle \langle \psi _p |\psi _k\rangle =\delta _{jk}\, , \quad j,k=0,1,\ldots \] where $\psi _k$, $k=0,1,\ldots $ are orthonormal basis functions in the space $H$. The free particle CS may be obtained by applying a displacement operator in the Heisenberg-Weil group to the vacuum vector $\psi _0$: \[ \psi _z(x,t) =\exp (za^+-\bar z a)\psi _0(x,t)\,, \quad z\in \Bbb C\,. \] These vectors are also the eigenvectors of the annihilation operator $a\psi _z =z\psi _z$. The vectors $\psi _z\in H$ belong to a more wide set then ${\cal L}_0 $. Their Fourier decomposition in terms of the basis $\psi_n$ has the form \begin{equation} \begin{array}{rl} \psi _z= & \Phi \sum \nolimits _n a_n z^n \psi _n\,, \label {psz} \end{array} \end{equation} \vspace{-4ex} \[ \begin{array}{rl} \Phi = & \Phi (z,\bar z)=\exp (-z\bar z /2)\,, \\ a_n= & (n!)^{-1/2},\quad z\in \Bbb C\,. \end{array} \] The vectors $\psi _z(x,t) $ satisfy all the properties enumerated in the Introduction. In particular, the measure $d\mu =d\mu (z,\bar z)$ from the relation (\ref{Id}) is well-% known: $d\mu =dxdy/\pi $, $z=x+iy$ and the domain of integration ${\cal D}$ is the whole complex plane $\Bbb C$. In what follows I will not indicate the domain of integration in the integrals over the measures. Integration will be always extended over the whole complex plane. Finally I give a coordinate representation of the free particle CS $$ \begin{array}{rl} \psi _z(x,t) =& \hspace{-4pt} (2\pi )^{-1/4}(1+it)^{-1/2}\\ & \hspace{-14pt} \times \exp \left[ -{\textstyle \frac 14}(z+\bar z)^2+ {\displaystyle \frac{(x+2iz)^2(it-1)}{4(1+t^2)}} \right] . \end{array} $$ I use the notation $x$ as the spatial coordinate and as the real part of a complex number $z$. I hope that it will not cause a confusion since these quantities will never appear in the same formula. \section{DARBOUX TRANSFORMATIONS AND ISOMETRIC OPERATORS} In this section I give an analysis of Darboux transformation operator $L$ as an operator defined in the Hilbert space $H$. I would like to stress that this operator is unbounded and can not be defined over the whole space $H$. It has a domain of definition which is a subset of $H$ and will be specified. Moreover, it domain of values does not coincide with $H$. Therefor this operator can not realize shifting between Hilbert spaces contrary to published assertion \cite{Montem}. To obtain $N$-soliton potential I use the Darboux transformation operator approach elaborated in details in Ref. \cite{MatvSal}. The action of this operator on a sufficiently smooth function is defined by the formula \[L\psi =W^{-1}(u_1,\ldots ,u_N) W(u_1,\ldots ,u_N,\psi )\] where $W$ stands for the usual symbol of a Wronskian. In the case when the initial potential $V_0$ does not depend on time, the functions $u_k=u_k(x,t)$ being solutions of the initial Schr\"odinger equation may be eigenfunctions of the initial Hamiltonian as well $h_0u_k=a _ku_k$ and in general are not supposed to satisfy any boundary conditions. In this case the transformation operator $L$ does not depend on time and transforms solutions of the initial Schr\"o\-din\-ger equation onto solutions of the Schr\"o\-din\-ger equation with the potential \[V_1=V_0-2\partial _x^2\log W(u_1,\ldots ,u_N)\] which is independent on time. In this paper we need not to use time dependent Darboux transformation which was proposed by V. ~Matveev and M. ~Salle (see Ref. \cite{MatvSal}) and advanced by V. ~Bagrov and B. ~Samsonov \cite{Rev}. To obtain an $N$-soliton potential we should take $V_0=0$ and specify the transformation functions $u_k$ as follows \cite{MatvSal}: \[ \begin{array}{rl} u_{2k-1}= & \cosh (a_{2k-1}x+b_{2k-1})\,,\\ u_{2k}= & \sinh (a_{2k}x+b_{2k})\,,\\ h_0u_k= & -a_k^2u_k\,, \ k=1,2,\ldots ,N\,,\\ & a_1<a_2<\ldots <a_N. \end{array} \] The time dependent phase factors are omitted from these functions since they do not affect all the results. In general the Wronsky determinant contains $N!$ summands. I would like to stress that in a special case of soliton potentials this determinant may be substantially simplified and presented as a sum of $2^{N-1}$ hyperbolic cosines \cite{CSh} \[ W(u_1,\ldots ,u_N)=2^{1-N}\!\!\!\sum_{(\varepsilon_1,\ldots ,\varepsilon_N)}^{2^{N-1}} \!\!\!\varepsilon_2\varepsilon_4\cdots\varepsilon_{p}\] \vspace{-3ex} \[\times \prod_{j>i}^{N}( \varepsilon_{j}a_j- \varepsilon_{i}a_i)\cosh [\sum_{l=1}^N\varepsilon_l (a_{l}x+b_l)]\,, \] where $\varepsilon_i=\pm 1$, the value of the subscript $p$ at $\varepsilon _p$ should be taken equal to $N$ for even $N$ values and to $N-1$ for odd $N$ values, the summation runs over all ordered and nonidentical sets $(\varepsilon_1,\ldots ,\varepsilon_N)$ (the sets $(\varepsilon_1, \ldots ,\varepsilon_N)$ and $(-\varepsilon_1, \ldots ,-\varepsilon_N)$ are declared to be identical). It can be shown \cite{MatvSal} that the potential so obtained is regular and bounded from below. This implies that the Hamiltonian $h_1=-\partial _x^2+V_1$ is essentially self-% adjoint in $H$. It has a mixed spectrum. The position of the discrete spectrum levels is defined by the values of the parameters $a_k$: $E_k=-a_k^2$. Corresponding eigenfunctions have the form \cite{BSTMF} \[ \begin{array}{rl} \varphi _k= & N_kW^{(k)}(u_1,\ldots ,u_N)/W(u_1,\ldots ,u_N)\,,\\ N_k=& ({\textstyle \frac 12}a_k\prod_{j=1(j\ne k)}^N|a_k^2- a_j^2|)^{1/2} \,,\\ h_1\varphi _k= & -a_k^2\varphi _k, \quad k=1,\ldots ,N \end{array} \] where $W^{(k)}(u_1,\ldots ,u_N)$ is the Wronskian of the functions $ u_1,\ldots ,u_N $ except for the function $u_k$ and the factor $N_k$ is introduced to ensure the normalization of the functions $\varphi _k$, $\langle \varphi _k|\varphi _j\rangle =\delta _{kj}$, $k,j=1,\ldots N$. The continuous spectrum corresponds to the semiaxis $E>0$. Continuous spectrum eigenfunctions, $\varphi _p=\varphi _p(x,t) $ of the Hamiltonian $h_1$ may be obtained with the aid of the operator $L$: $\varphi _p= N_p^{-1}L\psi _p$ where the factor $N_p^{-1}$ such that \[N_p^2=(p^2+a_1^2)\ldots (p^2+a_N^2)\] is introduced to ensure the normalization of the functions $\varphi _p$: $\langle \varphi _p|\varphi _q\rangle =\delta (p-q)$, $h_1\varphi _p=p^2\varphi _p$. The set $\left\{\varphi _j,\ j=1,\ldots ,N;\ \varphi _p,\ p\in \Bbb R\right\}$ is complete in $H$. Since the operator $L$ is linear, the relation $L\psi _p=N_p\varphi _p$ defines the action of this operator on every $\psi $ of the form \begin{equation} \psi (x,t)=\int C(p)\psi _p(x,t)dp \label{cp} \end{equation} where $C(p)$ is a finite continuous function over $\Bbb R$. The set of functions of the form (\ref{cp}) is a linear space that I shall denote by ${\cal L}_{0p}$ and it is dense (in the sense of generalized functions) in $H$. (More precisely it is dense in $H_-$ since these are functionals.) Hence, the action of the operator $L$ is defined for every element from ${\cal L}_{0p}$. The image of the space ${\cal L}_{0p}$, that I shall denote by ${\cal L}_{1p}$ consists of the functions \[ \varphi (x,t)=\int C(p)N_p\varphi _p(x,t)dp\, . \] The Darboux transformation operator $L$ together with its Laplace adjoint $L^+$ has remarkable factorization properties \cite{BSTMF,AIS} \begin{equation} g_0=L^+L=(h_0+a_1^2)\ldots (h_0+a_N^2)\, , \label{fac0} \end{equation} \vspace{-3ex} \begin{equation} g_1=LL^+=(h_1+a_1^2)\ldots (h_1+a_N^2)\, . \label{fac} \end{equation} The functions $\psi _p$ are eigenfunctions of $g_0$, $g_0\psi _p=N_p^2\psi _p$. This imply that the functions $\varphi _p$ are eigenfunctions of the operator $g_1$, $g_1\varphi _p=N_p^2\varphi _p$. The discrete spectrum eigefunctions of the operator $h_1$, $\varphi _k$, $k=1,\ldots ,N$ belong to the kernel of the operator $g_1$, $g_1\varphi _k=0$, $k=1,\ldots ,N$. This means that the operator $g_1$ is nonnegative in $H$. Therefor, consider the orthogonal decomposition of the space $H$: $H=H_0\oplus H_1$ where $H_0$ is an $N$-dimensional space with the basis $\varphi _k$, $k=1,\ldots ,N$. The functions $\varphi _p$, $p\in \Bbb R$ form a basis (in the sense of generalized functions) in $H_1$. In what follows I shall not consider the space $H_0$ and restrict the consideration only by the space $H_1$. The operators $h_1$ and $g_1$ being restricted to this space have only a continuous spectrum and the operator $g_1$ is strictly positive. I conserve the same notations for these operators as operators acting in $H_1$ Taking into account the spectral decomposition of these operators \[h_1=\int dpp|\varphi _p\rangle \langle \varphi _p|\, ,\] \vspace{-4ex} \[g_1=\int dpN_p^2|\varphi _p\rangle \langle \varphi _p| \] we can specify their domain of definitions. For the operator $h_1$ it consists of all $\varphi \in H_1$ such that the integral \[ \|h_1\varphi \|^2= \int dpp^2|\langle \varphi|\varphi _p\rangle |^2 \] converges and for the operator $g_1$ we should demand the convergence of the integral \[ \|g_1\varphi \|^2= \int dpN_p^4|\langle \varphi|\varphi _p\rangle |^2\, .\] It is clear that the operator $g_1$ is defined on ${\cal L}_{1p}$ and maps this space into itself. Using this fact and the factorization property (\ref{fac}) we can define the action of the operator $L^+$ onto the functions $\varphi _p$, $L^+\varphi _p=N_p^{-1}L^+L\psi _p=N_p\psi _p$, and extend this operator by linearity on the whole space ${\cal L}_{1p}$. It is not difficult to see that the following equality \[\langle L\psi _p|\varphi _q\rangle = \langle \psi _p|L^+\varphi _q\rangle \] holds for all $\psi _p\in {\cal L}_{0p}$ and $\varphi _q\in {\cal L}_{1p}$. Nevertheless, this fact does not mean that $L^+$ is an operator conjugate with respect to the scalar product to $L$ which domain of definition is ${\cal L}_{0p}$. To find such an operator we have to specify correctly its domain of definition. I shall not look for this domain. Instead I shall give a closed extension $\bar L$ of the operator $L$ and then find its conjugate $\bar L^+$. Once we know the bases $\psi _p$ and $\varphi _p$ in $H$ and $H_1$ respectively we can consider isometric operators \[U=\int dp |\varphi _p\rangle \langle \psi _p| \,,\] \vspace{-4ex} \[U^{-1}=U^+=\int dp |\psi _p\rangle \langle \varphi _p| \, .\] Similar operators have been introduced by L.D. Faddeev \cite{Fadd} and considered by L. Pursey \cite{Pur} for purely discrete bases. These operators are bounded and defined for all elements from $H$ and $H_1$ respectively. Consider now the following operators \begin{equation} \bar L=\int dp N_p|\varphi _p \rangle \langle \psi _p|\, , \label{Lb} \end{equation} \vspace{-4ex} \begin{equation} \bar L^+=\int dp N_p|\psi _p \rangle \langle \varphi _p|\,. \label{Lbk} \end{equation} It is not difficult to specify their domains of definition. For this purpose I use the spectral decompositions of the operator $g_0$ and its square root \[g_0=\int dpN_p^2 |\psi _p\rangle \langle \psi _p|\, ,\] \vspace{-3ex} \begin{equation} g_0^{1/2}=\int dpN_p |\psi _p\rangle \langle \psi _p|\, . \label {g012} \end{equation} It follows that \[\|\bar L\psi \|^2=\|g_0^{1/2}\psi \|^2= \int dp N_p^2|\langle \psi |\psi _p\rangle |^2\,.\] This means that the domain of definition of $\bar L$ coincides with that of $g_0^{1/2}$ and consists of all $\psi \in H$ such that the integral in the right hand side of this equation converges. The domain of definition of $\bar L^+$ coincides with that of the operator $g_1^{1/2}$. It is worthwhile to mention that these domains may be described in terms of conditions imposed on functions that are comprised in these domains (see for example \cite{KosSarg}) since $h_0$ and $h_1$ are operators bounded from below and essentially self-adjoint. It is clear from the formulae (\ref{Lb}) and (\ref{Lbk}) that the operator $\bar L^+$ is conjugate to $\bar L$ with respect to the scalar product and their domains of definition are well specified. Moreover, $\bar L^{++}=\bar L$. This imply \cite{Smirn,RS} that the operator $\bar L$ is closed. The formulae (\ref{Lb}) and (\ref{Lbk}) give quasispectral representation of the closed operators $\bar L$ and $\bar L^+$. It follows from the formulae (\ref{Lb}) and (\ref{Lbk}) that $\bar L\psi _p=L\psi _p=N_p\psi _p$ and $\bar L^+\varphi _p=L^+\varphi _p=N_p\varphi _p$. This means that $\bar L$ is a closed extension of the operator $L$ and $\bar L^+$ is a similar extension of the operator $L^+$ when the domains ${\cal L}_{0p}$ and ${\cal L}_{1p}$ are taken as their initial domains of definitions. From the spectral decomposition of the operators $g_0^{1/2}$ (\ref{g012}) and $g_1^{1/2}$, \[g_1^{1/2}=\int dp N_p|\varphi _p\rangle \langle \varphi _p| \,,\] we obtain the following representations for $\bar L$ and $\bar L^+$: \[\bar L=Ug_0^{1/2}=g_1^{1/2}U \, ,\quad \bar L^+=g_0^{1/2}U^+=U^+g_1^{1/2}\, .\] Such representations are known as {\it polar decompositions} or {\it canonical representations} of closed operators (see for example Refs. \cite{RS,DS}). The action of the operator $U$ on the basis $\psi _n$ gives an orthonormal basis in $H_1$: $\zeta _n=U\psi _n$, $\langle \zeta _n|\zeta_k\rangle =\delta _{nk}$. The functions $\varphi _n=g_1^{1/2}\zeta _n=\bar L\psi _n=L\psi _n$, hence, form a basis in $H_1$ equivalent to an orthonormal (so called Riesz basis, see for example Ref. \cite{GK}). The operator $U$ is nonlocal and rather complicated. Therefor there is no way in which simple explicit expressions can be derived for the functions $\zeta _N$. The functions $\varphi _n(x,t) =L\psi _n(x,t)$ are much simpler but they are not orthogonal to each other: $\langle \varphi _n|\varphi _k\rangle =S_{nk}$. I shall denote by $S$ the infinite matrix with the entries $S_{nk}$. The elements of this matrix can easily be expressed in terms of the elements of another matrix $S^0(a)$ with the entries $S_{nk}^0(a)=\langle \psi _n|h_0+a^2|\psi _k\rangle $: \[S_{nk}=\left[ S^0(a_1)S^0(a_2)\ldots S^0(a_N) \right]_{nk}\] where the use of the factorization property (\ref{fac0}) has been made. Taking into account that $h_0$ is expressed in terms of the ladder operators $a$ and $a^+$ for the basis functions $\psi _n$, $h_0={\textstyle \frac 14} (a+a^+)^2$, we derive the nonzero elements of the matrix $S^0(a)$: $S_{nn}^0(a)=n/2+1/4+a^2$, $S_{nn+2}^0(a)={\textstyle \frac 14}\sqrt {(n+1)(n+2)}$. All the other matrix elements are zero. We see, hence, that the number of nonzero elements in each row and column of the matrix $S$ is finite. Consider now bounded operators \[M=\int dp N_p^{-1}|\varphi _p\rangle \langle \psi _p| \,,\] \vspace{-3ex} \[M^+=\int dp N_p^{-1}|\psi _p \rangle \langle \varphi _p|\] defined in $H$ and $H_1$ respectively. It is not difficult to see that $M\bar L^+=\Bbb I$ is the unit operator in $H_1$ and $M^+\bar L=\Bbb I$ is the unit operator in $H$. Using the spectral resolutions of the operators $g_0^{-1/2}$ and $g_1^{- 1/2}$ \[g_0^{-1/2}= \int dpN_p^{-1}|\psi _p\rangle \langle \psi _p| \,,\] \vspace{-3ex} \[g_1^{-1/2}= \int dpN_p^{-1}|\varphi _p\rangle \langle \varphi _p|\] we derive the polar decompositions of the operators $M$ and $M^+$: \[M=Ug_0^{-1/2}=g_1^{-1/2}U\,, \] \vspace{-3ex} \[M^+=g_0^{-1/2}U^+=U^+g_1^{-1/2}\,.\] It is easily seen that these operators factorise the operators inverse to $g_0$ and $g_1$: $M^+M=g_0^{-1}$, $MM^+=g_1^{-1}$. The functions $\eta _n=g_1^{-1/2}\zeta _n=M\psi _n$ form another basis in $H_1$ equivalent to an orthonormal. This basis is biorthogonal to $\varphi _n$, $\langle \varphi _n|\eta _k\rangle =\delta _{nk}$. It follows the representation for the elements $S_{nk}^{-1}$ $$ \begin{array}{rl} S_{nk}^{-1}= & \langle \eta _n|\eta _k \rangle = \langle \psi _n|g_0^{-1}|\psi _k \rangle \\ = & {\displaystyle \int }dpN_p^{-2}\langle \psi _n|\psi _p\rangle \langle \psi _p|\psi _k\rangle \end{array} $$ As a final remark of this section I would like to notice the following. The space $H_1$ can be obtained as a closure of the lineal ${\cal L}_1$ of all finite linear combinations of the functions $\varphi _n=L\psi _n$ with respect to the norm generated by the scalar product which is a restriction of the given scalar product in $H$ to the lineal ${\cal L}_1$. The set of functions of the form $\varphi =\bar L\psi $ when $\psi $ run through the whole domain of definition of the operator $\bar L$ (i.e. the domain $D_{\sqrt {g_0}}$ of definition of the operator $\sqrt {g_0}$) can not give the whole space $H_1$. Nevertheless, if we define a new scalar product in ${\cal L}_1$, $\langle \varphi _a|\varphi _b\rangle _1\equiv \langle L\psi _a|L\psi _b\rangle = \langle \psi _a|g_0|\psi _b\rangle $, $\psi _{a,b}\in {\cal L}_0$, $\varphi _{a,b}\in {\cal L}_1$ then the closure of ${\cal L}_1$ with respect to the norm generated by this scalar product will coincide with the set $\varphi =\bar L\psi $, $\psi \in D_{\sqrt {g_0}}$. This space is embedded in $H_1$. \section{COHERENT STATES OF SOLITON POTENTIALS} The operator $g_0$ is a symmetry operator for the Schr\"o\-din\-ger equation. Therefor it commutes with the Schr\"o\-din\-ger operator $i\partial _t-h_0$ when applied to the solutions of the Schr\"o\-din\-ger equation. It follows that the operator $U=\bar Lg_0^{-1/2}$ is an intertwining operator for the Schr\"o\-din\-ger operators $U(i\partial _t- h_0)=(i\partial _t-h_1)U$ and therefor it is a transformation operator. Hence, being applied to a solution of the initial Schr\"o\-din\-ger equation (in our case the free particle Schr\"o\-din\-ger equation) it gives a solution of the transformed equation (in our case the Schr\"o\-din\-ger equation with the $N$-soliton potential). The functions $\zeta _n=U\psi _n $ and $\zeta _z=U\psi _z$ are then solutions of the Schr\"o\-din\-ger equation with soliton potential. The Fourier decomposition of the function $\zeta _z$ in terms of the basis $\left\{\zeta _n\right\}$ has the same form as that of the function $\psi _z$ in terms of $\left\{\psi _n\right\}$ \[\zeta _z=\Phi \sum\nolimits _n a_n\zeta _n\, .\] The vectors $\zeta _z$, $z\in \Bbb C$ satisfy all the conditions formulated for CS in the Introduction because of the isometric nature of the operator $U$. The resolution of the identity operator (\ref{Id}) in the space $H_1$ in terms of the projectors on $\zeta _z$ takes place with the same measure $d\mu =dxdy/\pi $, $z=x+iy$. One of the deficiencies of these coherent states is that no a simple explicit expression for such vectors exists. This deficiency may be cured by acting to them by a symmetry operator for the Schr\"o\-din\-ger equation with soliton potential. Consider the vectors \[\varphi _z =g_1^{1/2}\zeta _z=\bar L\psi _z= \Phi \sum \nolimits _na_n\varphi _n\, .\] It is not difficult to see that the value $\langle \psi _z|g_0|\psi _z\rangle $ is finite. This means that $\psi _z $ belong to the domain of definition of the operator $\bar L$ and the above equality has a sense. Moreover, these functions are sufficiently smooth and we can apply to them directly the differential operator $L$. Thus we obtain a coordinate representation of $\varphi _z$. For instance, in the case of the one-soliton potential this representation reads $$ \begin{array}{rl} \varphi _z(x,t)= & -{\textstyle \frac 12} (2\pi )^{-1/4}(1+it)^{-3/2} \\ \times & [x+2iz+2a(1+it)\tanh (ax)] \\ \times & \exp \left[ -{\displaystyle \frac {(x+2iz)^2}{4+4it}}- {\textstyle \frac 14} (z+\bar z)^2 \right]. \end{array} $$ We see thus that these functions are much simpler then $\zeta _z$ and may be analyzed without difficulties. For example it is easily seen that \cite{NF} $|\varphi _z(x,t) |^2=|\varphi _z(-x,-t)|^2$. This property reflects a transparent nature of the one-soliton potential. Another system of states may be obtained with the help of the transformation operator $M$. Consider the vectors \[ \eta _z=g_1^{-1/2}\zeta _z=M\psi _z=\Phi \sum\nolimits _na_n \eta _n\, . \] The operator $M$ being inverse to $L$ has an integral nature. For the case of the one-soliton potential the integration may be carried out analytically \cite{NF}. This yields $$ \begin{array}{rl} \eta _z(x,t)= & {\textstyle \frac {-i}{4}}\sqrt \pi (2\pi )^{-1/4}{\rm sech}(ax)\\ \times & \exp \! \left[ -{\textstyle \frac 14}(z+\bar z)^2+a^2(1+it) \right] \\ \times & \left[ \exp (2iaz){\rm erfc}\! \left( a\sqrt {1+it}+{\displaystyle \frac {x/2+iz}{\sqrt {1+it}}} \right) \right. \\ - & \left. \exp (-2iaz){\rm erfc}\! \left( a\sqrt {1+it}-{\displaystyle \frac {x/2+iz}{\sqrt {1+it}}} \right) \right]\! . \end{array} $$ Where the parameter $b$ is taken to be zero. It is worthwhile to mention that all the states $\psi _z(x,t)$, $\varphi _z(x,t)$, $\eta _z(x,t)$, and $\zeta _z(x,t)$ can not represent nonspreading in time wave packets. Nevertheless, we can interpret them as coherent states since they satisfy all the properties of such states enumerated in the Introduction. I shall show now that for the vectors $\varphi _z$ and $\eta _z$ there exist measures $\mu _\varphi =\mu _\varphi (z,\bar z)$ and $\mu _\eta =\mu _\eta (z,\bar z)$ that realize the resolution of the identity operator in $H_1$ in terms of the projectors on these vectors. First consider another continuous basis in $H_1$: $\eta _p=N_pM\psi _p$, $\langle \eta _p|\eta _q\rangle =\delta (p-q)$, $p,q\in \Bbb R$. Since $\left\{\varphi _p\right\}$ and $\left\{\eta _p\right\}$ are bases in $H_1$, the resolutions of the identity operator of the type (\ref{Id}) in terms of the vectors $\eta _z$ and $\varphi _z $ are equivalent to the equations $$ \begin{array}{c} {\displaystyle\int }d\mu _\eta (z,\bar z)\langle \eta _p|\eta _z\rangle \langle \eta _z|\eta _q\rangle =\delta (p-q)\, ,\\ {\displaystyle \int }d\mu _\varphi (z,\bar z)\langle \varphi _p|\varphi _z\rangle \langle \varphi _z|\varphi _q\rangle =\delta (p-q)\, . \end{array} $$ Taking into account that the functions $\psi _p$ are the eigenfunctions of $g_0$ and $g_0^{-1}$, $g_0\psi _p=N_p^2\psi _p$, $g_0^{-1}\psi _p=N_p^{-2}\psi _p$ we arrive at equations for the measures $\mu _\eta $ and $\mu _\varphi $ \hspace{-6em}\begin{equation} (N_pN_q)^{-1}\int d\mu _\eta \langle \psi _p|\psi _z\rangle \langle \psi _z|\psi _q\rangle =\delta (p-q)\, , \label{mueta} \end{equation} \vspace{-3ex} \hspace{-6em}\begin{equation} N_pN_q \int d\mu _\varphi \langle \psi _p|\psi _z\rangle \langle \psi _z|\psi _q\rangle =\delta (p-q)\, . \label{mufi} \end{equation} Note that the integrals involved in these equations are time-% independent and hence can by calculated at $t=0$. Therefor in what follows I let $t=0$ and look for the measures independent on time. The momentum representation of the CS $\psi _z$ is well-known \[ \begin{array}{rl} \langle \psi _p|\psi _z\rangle = & (2/\pi )^{1/4}\Phi \psi _p(z)\,, \\ \psi _p(z)= & \exp (-p^2+2zp-z^2/2)\,,\ z=x+iy\,. \end{array} \] Let us look for the measure $\mu _\eta $ in the form $d\mu _\eta =\omega _\eta (x)dxdy$, $z=x+iy$. After performing the integration with respect to $y$ in the Eq. (\ref{mueta}) we arrive at an equation for $\omega _\eta (x)$ \[ (2\pi )^{1/2}\int dx \omega _\eta(x)F_p(x)=N_p^2\exp (2p^2)\,, \] \vspace{-3ex} \[ F_p(x)=\exp (4px-2x^2)\,. \] The function $N_p^2$ is a polynomial in $p$ which is known. We conclude then that $\omega _\eta (x)$ is a polynomial in $x$ whose coefficients are uniquely defined by the coefficients of the polynomial $N_p^2$. For instance, for the one-soliton potential we have \[ \omega _\eta (x)=(x^2+a^2-1/4)/\pi\,. \] This proves that the states $\eta _z$ may be interpreted as CS. We note that the states $\eta _z$ are defined with the help of the bounded operator $g_0^{-1/2}$. This is the reason for which the measure $\mu _\eta $ is expressed in terms of ordinary (non generalized) functions. An other case takes place for the states $\varphi _z $ which are defined by the semibounded operator $g_1^{1/2}$. I shall show now that the measure $\mu _\varphi $ is expressed in terms of generalized functions. Let us look for the measure $\mu _\varphi $ in the form $d\mu _\varphi =dyd\omega _\varphi (x)$. The integration in the equation (\ref{mufi}) with respect to $y$ leads us to an equation for the measure $d\omega _\varphi (x)$ \begin{equation}\label{Fp} (2\pi )^{1/2}\int d\omega _\varphi (x)F_p(x)= N_p^{-2} \exp (2p^2)\,. \end{equation} First we note that $|F_p(x+iy)|\le \exp (-dx^2+by^2)$ where $2\le d\le b$. This means that $F_p(x)$ belongs to a subspace of the space $S_{1/2}^{1/2}$ of entire functions $F$ such that $|F(x+iy)|\le \exp (-dx^2+by^2)$, $0\le d\le b$ \cite{GV}. We look for $\omega _\varphi $ as a functional (i.e. a generalized function) over $S_{1/2}^{1/2}$. (We will see that really this is a functional over a subspace ${\stackrel{\circ}{{S}}}{}_{1/2}^{1/2}\subset S_{1/2}^{1/2}$.) As it is known \cite{GV} positive definite functionals (we look for just such a functional) over $S_{1/2}^{1/2}$ are specified by their Fourier transforms. Let $\tilde \omega _\varphi $ be the Fourier transform of the measure $\omega _\varphi (x)$. This means that an integration of a function $F(x)\in S_{1/2}^{1/2}$ with respect to the measure $\omega _\varphi (x)$ should be replaced by the integration of the Fourier transform $\tilde F(t)$ of this function with respect to the measure $\tilde \omega _\varphi $. In particularly \begin{equation}\label{Fpx} \int d\omega _\varphi (x)F_p(x)= \int d\tilde \omega _\varphi (t)\tilde F_p(t) \end{equation} where $\tilde F_p(t)$ is the Fourier image of the function $F_p(x)$ which in our case can easily be found \[\tilde F_p(t)=\sqrt {\pi /2}\exp (2p^2+ipt-t^2/8)\,.\] As a result the Eq. (\ref{Fp}) yields the equation for $\tilde \omega _\varphi (x)$ \[ \pi \int d\tilde \omega _\varphi (t)\exp (-t^2/8+ipt)= N_p^{-2}\,. \] It is an easy exercise to see that $\tilde \omega _\varphi (t)$ may be expressed in terms of elementary functions. For this purpose we look for $\tilde \omega _\varphi (t)$ in the form $d\tilde \omega _\varphi (t)=\rho _\varphi (t)dt$ and use the following representation for the function $N_p^{-2}$: \begin{equation}\label{Np} \begin{array}{rl} N_p^{-2}= & {\displaystyle \sum }_{k=1}^N {\displaystyle \frac {A_k}{\tau +a _k^2}}\,, \quad \tau=p^2\,,\\ A_k= & \left[(dN_p^2/d\tau )_{\tau =-a _k^2}\right]^{-1}. \end{array} \end{equation} After some algebra we obtain a formula for $\rho _\varphi (t)$ \hspace{-6em}\begin{equation}\label{rofi} \rho _\varphi (t)= (2\pi )^{-1}\sum\nolimits _{k=1}^{N}\frac {A_k}{a _k} \exp (t^2/8-a _k|t|)\,. \end{equation} Note that for the function $\rho _\varphi (t)$ of the form (\ref{rofi}) there exist in $S_{1/2}^{1/2}$ such functions $F(p)$ that the integral in the right hand side of the Eq. (\ref{Fpx}) diverges. The convergence condition for this integral imposes a restriction on the decrease of the integrand function $F(x)$ in the left hand side of the Eq. (\ref{Fpx}) as $|x|\to \infty $. This function should satisfy an inequality $|F(x)|\ge \exp (-2x^2-Ax)$ where $A$ is a nonnegative constant own to every function $F(x)\in S_{1/2}^{1/2}$. I denote the set of functions satisfying this condition by ${\stackrel{\circ }{{S}}}{}_{1/2}^{1/2}(\subset S_{1/2}^{1/2})$ which obviously is a linear space. Thus, we have found the measure $\mu _\varphi $ in terms of the generalized function $\omega _\varphi (x)$ over the space ${\stackrel{\circ }{{S}}}{}_{1/2}^{1/2}$, $d\mu_\varphi =dyd\omega _\varphi (x)$, $z=x+iy$ which is defined by its Fourier transform $\tilde \omega _\varphi $. The integrals with respect to this measure should be calculated as follows \[ \int d\mu _\varphi \langle \varphi _a|\varphi _z\rangle \langle \varphi _z|\varphi _b\rangle \equiv \int dt \tilde \rho _\varphi (t)\tilde F_{ab}(t) \] where $ \tilde F_{ab}(t)$ is the Fourier transform of the function \[ F_{ab}(x)=\int dy \langle \varphi _a|\varphi _z\rangle \langle \varphi _z|\varphi _b\rangle \,,\ z=x+iy\,.\] Finally I give comments on the calculation of the norms of the functions $\eta _z$ and $\varphi _z$. The square of the norm of $\eta _z$ may be calculated with the aid of the formula (\ref{Np}) for the function $N_p^{-2}$ and the factorization property of the operator $g_0^{-1}$ in terms of the operators $M$ and $M^+$ \[ \langle \eta _z|\eta _z\rangle = \langle \psi _z|g_0^{-1}|\psi _z\rangle = \int dp N_p^{-2}|\langle \psi _z|\psi _p \rangle |^2\,. \] After some algebra we obtain $$ \begin{array}{rl} \langle \eta _z|\eta _z\rangle = & \sum \nolimits _{k=1}^NA_kF_k\,, \quad z=x+iy\,,\\ F_k= & {\displaystyle \frac {\sqrt {2\pi }}{a_k}}\exp [2(a_k ^2-x^2)] \vphantom {\left({\displaystyle \frac {\sqrt {2\pi }}{a_k}}\right)}\\ \times & {\rm Re}\left[\exp (4ia_kx) {\rm erfc}(a_k \sqrt 2+i\sqrt 2x)\right]\,. \end{array} $$ Similarly, the square of the norm of the function $\varphi _z$ coincides with the expectation value of the operator $g_0$ in the state $\psi _z$. For instance, for the one-soliton potential we obtain $\langle \varphi _z|\varphi _z\rangle = \langle \psi _z|g_0|\psi _z\rangle = 1/4+a^2+x^2$, $z=x+iy$. \section{CONCLUSION} A classical particle proves no reflection in the scattering process on a potential well. For a quantum particle in general this is not the case. Nevertheless, there exists a wide class of potentials called transparent potentials for which the scattering process of the quantum particle comes in some sense about in a similar way that those of the classical particle i.e. without reflection. In my opinion this mysterious phenomena up to now has no any perspicuous explanation. From a practical point of view the answer to this question is rather important. If at quantum level we would be able to force a signal to propagate without reflection we could decrease the output of the emitted signal. All transparent potentials known at present have a remarkable property. They are related with zero potential (free particle) by Darboux transformations. Up to recent times it was believed that such potentials have a finite number of discrete spectrum levels. Nevertheless a method based on an infinite chain of Darboux transformations with the help of which one can create transparent potentials with infinite number of discrete spectrum levels has been proposed recently \cite{Shab}. To understand better the nature of transparent potentials we should investigate them in all details. As it is well known the quantum theory gives a more detailed description of the nature then the classical one. Therefor different quantum systems may correspond to the same classical system. Furthermore, the quantization procedure is not unique (canonical quantization, Berezin quantization, geometric quantization, etc.). In this respect the following question is of interest. What are common points between two classical systems a quantization of which gives the quantum systems that are related to each other by a Darboux transformation operator? In particularly, what are common points between the classical free particle and the particle that moves in a potential quantization of which gives a transparent potential? The CS approach make it possible to formulate clear steps in the direction of obtaining an answer to this question. It permits one to construct a classical mechanics counterpart of a given quantum system and analyze properties of such a system. This approach has been realized recently for the potential of the form $x^2+gx^{-2}$ \cite{SamJMP}. It was established that at classical level the Darboux transformation consists in a distortion of a phase space of the classical system. Moreover, this distortion is consistent with the transformation of the Hamilton function in such a way that the equations of motion remain unchanged. Up to now no any approach for analysis of CS of transparent potentials has been proposed. In this paper I show that the Darboux transformation operator approach is suitable for this purpose. A next step in this direction would be an analysis of the classical counterpart of the quantum system that moves in a transparent potential. \vspace{4ex} \begin{center} {\bf ACKNOWLEDGMENTS} \end{center} It is a pleasure to thank Dr. V.P. ~Spiridonov for many helpful discussions. This work was supported in part by the Russian Fund for Fundamental Research and the Russian Ministry of Education.
\section{Introduction} The understanding of phase transitions on non-crystalline structures has been recently improved by exact results connecting general geometrical features of networks to the existence of spontaneous symmetry breaking \cite{panza1,panza2,mwg2,fss}. The situation is more complex when dealing with the critical behaviour of model systems. For continuos symmetry models the singularity of the free energy, which determines the critical behavior, appears to be related to the infrared spectrum of the Laplacian operator on the network \cite{sfer,on}, while this is not the case for discrete symmetries (e.g. the Ising model). There, all known results suggest that the link between critical behaviour and geometry should involve some other topological features \cite{panza1,panza2}. An interesting result concerns the Sierpinski Gasket, a typical and widely studied fractal, where the Ising model is exactly solved. On this structure, although continuous symmetry models exhibit a power law behaviour for $T\to 0$, the Ising model has an exponential low temperature behaviour which coincides with that found on the linear chain \cite{panza1,panza2}. To analyze the critical regime from an {\it ab initio} point of view, an interesting picture is provided by the study of the singularities of thermodynamic potentials. In 1952 Lee and Yang in two famous papers \cite{leeyang1,leeyang2} first proposed their fundamental approach to phase transitions, consisting in studying the zeros of the partition function of a statistical system, considered as a function of a complex parameter. The partition function on a finite volume is a polynomial in complex activity or fugacity, so that the complete knowledge of the zeros distribution is equivalent to the knowledge of the partition function itself and all thermodynamic quantities can be obtained from it. On a finite volume there are no real zeros, the coefficient being all real and positive. However in the thermodynamic limit the zeros can pinch the real axis (the region of physical interest) producing a singularity in the free energy (or grand-canonical potential)\cite{ruelle}. The pinching points are phase transitions points on the parameter axis and the zeros distribution in their neighbourhood can be connected with the critical properties of the system \cite{ipz}. Unfortunately, a complete knowledge of the zeros is very difficult to obtain except for a few exactly solvable cases. General theorems hold for the zeros distribution in the complex magnetic field plane in a class of ferromagnetic lattice systems, including the Ising model \cite{leeyang2,griff}. On the other hand, very little is known rigorously about the behaviour of the zeros of the partition function in the complex temperature plane, the so called Fisher zeros \cite{fisher}. In general, it is not clear if Fisher zeros arrange on smooth curves even if this is the case in some exactly solvable models. For the Ising model on regular two dimensional lattices \cite{fisher,shrock}, the Fisher zeros arrange on curves that cross the positive real axis at the transition point. In the one dimensional case only two zeros (with infinite multiplicity) are found and these have a nonzero imaginary part, so that there is no singular point for the free energy. For statistical models defined on non periodic discrete structures, Fisher zeros show some peculiar features, making the analysis of their density and location extremely subtle. In particular, on some hierarchical lattices (i. e. q-potts model on diamond hierarchical lattices \cite{derrida}) the zeros have been show to form a fractal set (Julia sets). In this case, while the general approach for identifying the singularity points and the critical behaviour still holds, the widely used arguments concerning scaling of singularities and zeros density with the volume must be handled carefully, as will be shown in the following. In this paper we will study the Ising model on the Sierpinski gasket, obtaining a recursive relation for the partition function, from which the zeros of the $n$-th stage gasket can be obtained from those of the $(n-1)$th. The distribution we obtain is fractal and pinches the real axis at $T=0$, so that a singular point with a power law critical behaviour could be expected. However, since the zeros density is found out to vanish exponentially in the neighbourhood of $T=0$, these zeros don't produce any singularity of the free energy: although the zeros pinch the real axis the `critical behaviour' is the same as the one-dimensional case. \section{Ising model on the Sierpinski gasket} The Sierpinski gasket is a fractal graph which can be built recursively with the following procedure: the initial stage (${\cal G}_0$) is a triangle (3 sites with 3 edges) and the $n$th stage (${\cal G}_n$) is obtained joining 3 ${\cal G}_{n-1}$ at their external corners, to form a bigger triangle (fig. \ref{gask}). \begin{figure}[t] \begin{center} \begin{picture}(290,140)(-20,-20) \put(0,0){\begin{picture}(30,30) \multiput(0,0)(30,0){2}{\circle*{6}} \put(15,30){\circle*{6}} \put(0,0){\line(1,0){30}} \put(0,0){\line(1,2){15}} \put(15,30){\line(1,-2){15}} \end{picture}} \put(10,-20){${\cal G}_0$} \put(50,0){\begin{picture}(60,60) \multiput(0,0)(30,0){2}{\begin{picture}(30,30) \multiput(0,0)(30,0){2}{\circle*{6}} \put(15,30){\circle*{6}} \put(0,0){\line(1,0){30}} \put(0,0){\line(1,2){15}} \put(15,30){\line(1,-2){15}} \end{picture} } \put(15,30){\begin{picture}(30,30) \multiput(0,0)(30,0){2}{\circle*{6}} \put(15,30){\circle*{6}} \put(0,0){\line(1,0){30}} \put(0,0){\line(1,2){15}} \put(15,30){\line(1,-2){15}} \end{picture}} \end{picture}} \put(75,-20){${\cal G}_1$} \multiput(130,0)(60,0){2}{\begin{picture}(60,60) \multiput(0,0)(30,0){2}{\begin{picture}(30,30) \multiput(0,0)(30,0){2}{\circle*{6}} \put(15,30){\circle*{6}} \put(0,0){\line(1,0){30}} \put(0,0){\line(1,2){15}} \put(15,30){\line(1,-2){15}} \end{picture}} \put(15,30){\begin{picture}(30,30) \multiput(0,0)(30,0){2}{\circle*{6}} \put(15,30){\circle*{6}} \put(0,0){\line(1,0){30}} \put(0,0){\line(1,2){15}} \put(15,30){\line(1,-2){15}} \end{picture}} \end{picture}} \put(160,60){\begin{picture}(60,60) \multiput(0,0)(30,0){2}{\begin{picture}(30,30) \multiput(0,0)(30,0){2}{\circle*{6}} \put(15,30){\circle*{6}} \put(0,0){\line(1,0){30}} \put(0,0){\line(1,2){15}} \put(15,30){\line(1,-2){15}} \end{picture}} \put(15,30){\begin{picture}(30,30) \multiput(0,0)(30,0){2}{\circle*{6}} \put(15,30){\circle*{6}} \put(0,0){\line(1,0){30}} \put(0,0){\line(1,2){15}} \put(15,30){\line(1,-2){15}} \end{picture}} \end{picture}} \put(185,-20){${\cal G}_2$} \end{picture} \end{center} \caption{First iterations of gasket's construction} \label{gask} \end{figure} In this way ${\cal G}_n$ has $\frac{3}{2}(3^n-1)$ sites, $3^{n+1}$ edges and its side contains $2^n$ edges. The gasket is obtained as the limit for $n\rightarrow\infty$ of this procedure. The Ising model on the gasket is defined associating the spin variable $\sigma_i=\pm1$ to every site $i$ of the graph, and considering a nearest-neighbours interaction between points joined by an edge (link). The Hamiltonian is therefore: \begin{equation} E= -J\sum_{<i,j>}\sigma_i \sigma_j \end{equation} where the sum runs over the couples of sites joined by a link and $J$ is a positive constant (ferromagnetic coupling). \section{Recursive relation for the partition function} For ${\cal G}_0$ the partition function \begin{equation} Z=\sum_{\{\sigma_i \}} e^{- \beta E} \end{equation} can be seen as a sum of the elements of the rank 3 tensor $M_0$ \begin{equation} Z_0=\sum_{\sigma_1,\sigma_2,\sigma_3=\pm 1} M_0^{\sigma_1\sigma_2\sigma_3} \end{equation} where \begin{equation} M_0^{\sigma_1\sigma_2\sigma_3}=\exp[-\beta E(\sigma_1,\sigma_2,\sigma_3)] \end{equation} $M_0^{\sigma_1\sigma_2\sigma_3}$ can take only 2 values because there are only 2 classes of spin configurations with different energy: \begin{equation} \left\{ \begin{array}{l} M_0^{\sigma\sigma\sigma}=e^{3 \beta J}=y^3 \\ M_0^{\sigma\sigma(-\sigma)}=M_0^{\sigma(-\sigma)\sigma} M_0^{(-\sigma)\sigma\sigma}=e^{- \beta J}=y^{-1} \end{array} \right. \end{equation} where $y= e^{\beta J}$. In terms of $y$ the partition function is: \begin{equation} Z_0=2y^3+6y^{-1} \end{equation} \begin{figure}[t] \begin{center} \begin{picture}(220,100)(-20,-20) \put(0,0){\begin{picture}(60,60) \multiput(0,0)(60,0){2}{\circle*{6}} \put(30,60){\circle*{6}} \put(0,0){\line(1,0){60}} \put(0,0){\line(1,2){30}} \put(30,60){\line(1,-2){30}} \end{picture}} \put(28,70){1} \put(-12,-12){2} \put(68,-12){3} \put(120,0){\begin{picture}(60,60) \multiput(0,0)(30,0){2}{\begin{picture}(30,30) \multiput(0,0)(30,0){2}{\circle*{6}} \put(15,30){\circle*{6}} \put(0,0){\line(1,0){30}} \put(0,0){\line(1,2){15}} \put(15,30){\line(1,-2){15}} \end{picture}} \put(15,30){\begin{picture}(30,30) \multiput(0,0)(30,0){2}{\circle*{6}} \put(15,30){\circle*{6}} \put(0,0){\line(1,0){30}} \put(0,0){\line(1,2){15}} \put(15,30){\line(1,-2){15}} \end{picture}} \end{picture}} \put(148,70){1} \put(108,-12){2} \put(188,-12){3} \put(123,32){a} \put(173,32){b} \put(148,-12){c} \end{picture} \end{center} \caption{Labeling of sites used for ${\cal G}_0$ and ${\cal G}_1$} \label{gasklab} \end{figure} For ${\cal G}_1$ the partition function can be expressed in the same way separating the sum over the states of the inner sites using the tensor $M_1$ whose indices correspond to the spins on the external vertices (fig. \ref{gasklab}): \begin{equation} M_1^{\sigma_1\sigma_2\sigma_3}= \sum_{\sigma_a,\sigma_b,\sigma_c=\pm 1} \exp[-\beta E(\sigma_i)] = \sum_{\sigma_a,\sigma_b,\sigma_c=\pm 1} M_0^{\sigma_1\sigma_a\sigma_b} M_0^{\sigma_a\sigma_2\sigma_c} M_0^{\sigma_b\sigma_c\sigma_3} \end{equation} Now $M_1$ has the same structure as $M_0$ since the possible values are: \begin{equation} \left\{ \begin{array}{l} M_1^{\sigma\sigma\sigma}=4y^{-3}+3y+y^9 \\ M_1^{\sigma\sigma(-\sigma)}=M_1^{\sigma(-\sigma)\sigma} M_1^{(-\sigma)\sigma\sigma}=3y^{-3}+4y+y^5 \end{array} \right. \end{equation} One can obtain $M_1$ from $M_0$ simply by a transformation mapping $y^3$ in $4y^{-3}+3y+y^9$ and $y^{-1}$ in $3y^{-3}+4y+y^5$. This is done by the substitution \begin{equation} y \to f(y)=\left( \frac{y^8-y^4+4}{y^4+3} \right)^{\frac{1}{4}} \label{tr1} \end{equation} followed by the multiplication by \begin{equation} c(y)=\frac{y^4+1}{y^3} \left[ (y^4+3)^3(y^8-y^4+4)\right]^{\frac{1}{4}} \label{tr2} \end{equation} The transformation also gives the new partition function: \begin{equation} Z_1(y)=Z_0(f(y))\ c(y) \end{equation} Following the same argument one can obtain for the $(n+1)$th stage of the gasket ${\cal G}_{n+1}$: \begin{equation} Z_{n+1}(y)=Z_n(f(y))\cdot \left[c(y)\right]^{3^n} \end{equation} Using this recursion relation we get: \begin{equation} Z_n(y)=\frac{2}{y^{3^n}} P_n(y^4) \end{equation} where $P_n(t)$ is a polynomial in $t$ of degree $3^n$ in which the coefficient of $t^{3^n}$ is 1; for $n=0$ one has $P_0(t)=t+3$ while the general case $n>0$ can be proven by induction. \section{Zeros of the partition function} Introducing the variable $x = y^4$ the transformation (\ref{tr1}), (\ref{tr2}) is given by: \begin{equation} \left\{ \begin{array}{l} x \rightarrow \tilde{f}(x)=\frac{x^2-x+4}{x+3} \\ \tilde{c}(x)=(x+1)x^{-\frac{3}{4}} \left[(x+3)^3(x^2-x+4)\right]^{\frac{1}{4}} \end{array} \right. \end{equation} Denoting by $x_n^i$ the zeros of $P_n(x)$ the partition function reads \begin{equation} Z_n=\frac{2}{x^{3^n/4}} \prod_{i=1}^{3^n}(x-x_n^i) \end{equation} and using the recurrence one finds \begin{equation} 2 x^{-\frac{3^{n+1}}{4}} \prod_{i=1}^{3^{n+1}}(x-x_{n+1}^i)= 2 x^{-\frac{3^{n+1}}{4}} \prod_{i=1}^{3^n} \left\{ [(x^2-x+4)-x_n^i(x+3)] (x+1) \right\} \end{equation} This equation shows that for every root $x_n^i$ of $Z_n$, $Z_{n+1}$ has the root $x=-1$ and the 2 solutions of \begin{equation} x_n^i=\tilde{f}(x_{n+1}^i) \end{equation} namely the preimages of $x_n^i$ by the transformation $\tilde{f}$ . Starting from $x_0^1=-3$ one obtains all the zeros of the partition function for the $n$-th stage gasket as shown in table~\ref{zerij}, where $h(x)$ denotes the set of the preimages of $x$ (and $h^k(x)$ is a set of $2^k$ zeros). \begin{table}[ht] \[ \begin{array}{|c|l|} \hline n & \multicolumn{1}{c|}{\mbox{zeros}} \\ \hline \hline 0& -3 \\ \hline 1& -1\hspace{.5cm} h(-3) \\ \hline 2& -1\hspace{.5cm} h(-1)\hspace{.5cm} h(h(-3)) \\ \hline 3& -1\hspace{.5cm} h(-1)\hspace{.5cm} h(h(-1))\hspace{.5cm} h(h(h(-3))) \\ \hline n& -1\hspace{.5cm} h(-1)\hspace{.5cm} \ldots \hspace{.5cm} h^{n-1}(-1)\hspace{.5cm} h^n(-3) \\ \hline \end{array} \] \caption{Zeros of partition function in y} \label{zerij} \end{table} From this table one can see that the preimages of $-1$ by the $j$th iterate of $\tilde{f}$ appear at the $(j-1)$th stage and are zeros of all the following stages, while the preimages of $-3$ are `temporary' zeros. It is also possible to find the multiplicity of these zeros: in fact since every root generates the root $-1$ at the next stage, this value appears $3^{n-1}$ times among the zeros of $n$-th stage, the roots $h(-1)$ appear as many times as $-1$ in the previous stage (their multiplicity is $3^{n-2}$) and in general the multiplicity of the roots belonging to $h^j(-1)$ is $3^{n-j-1}$. In this way one can, in principle, calculate all the zeros of the partition function at any stage. In practice this is possible only for small $n$ because of their exponential growth. An alternative approach \cite{derrida} is to start from a root $x_0$ (for example `$-1$' whose preimages are `permanent') then choose at random one of its two preimages by the transformation $\tilde{f}$ (denoted by $x_1$), then choose one of the preimages of $x_1$ and so on; the set of points obtained in this way is a representative of the set of all roots and has the advantage to contain zeros relative to large $n$. By plotting in the complex temperature plane the roots obtained with both methods it can be seen that very few of them fall in the neighborhood of the real axis and no information can be obtained about the critical behaviour (see fig. \ref{rand} where the zeros are plotted in the plane of the variable $t=e^{-\beta J}$). \begin{figure} \begin{center} \mbox{\epsfig{file=fig3.eps,height=8cm,angle=90}} \end{center} \caption{20000 zeros obtained by the random method (in the plane of $t=e^{-\beta J}$)} \label{rand} \end{figure} A good technique to obtain more zeros near the real axis consists in changing the choice probability of the two preimages \cite{derrida}; the two solutions of $x=\tilde{f}(x^{\prime})$ are \begin{equation} x^{\prime}=h_1(x)=\frac{1 + x - {\sqrt{-15 + 14\,x + {x^2}}}}{2} \end{equation} \begin{equation} x^{\prime}=h_2(x)=\frac{1 + x + {\sqrt{-15 + 14\,x + {x^2}}}}{2} \end{equation} and one can see that the repeated application of $h_2$ gives a sequence of points approaching the real axis. The set of roots obtained by increasing the probability of choosing the second preimage is not a representative set af all roots (it doesn't show their density not even approximately) but gives us a chance to observe their behaviour in the interesting area (see fig. \ref{rand98}). \begin{figure} \begin{center} \mbox{\epsfig{file=fig4.eps,height=8cm,angle=90}} \end{center} \caption{20000 zeros obtained with probability 0.98 of choosing $h_2$ (in the $t$ plane)} \label{rand98} \end{figure} A plot of these roots in the plane of $w=e^{-4\beta J}$ ($T=0$ corresponds to $w=0$) with a {\it log-log} scale (fig. \ref{loglog}) shows that the real and imaginary part are related by a power law: the curve can intersect the real axis in the thermodynamic limit only at the origin. \begin{figure} \begin{center} \mbox{\epsfig{file=fig5.eps,height=8cm,angle=90}} \end{center} \caption{Log-log plot of zeros (in the variable $w=e^{-4\beta J}$)} \label{loglog} \end{figure} Analytically one can verify this power behaviour by studying the transformation of the variable $w=e^{-4\beta J}=x^{-1}$ \begin{equation} g(w)=\left. \frac{1}{f(x)} \right|_{x=\frac{1}{w}}=\frac{w(3w+1)}{4w^2-w+1} \label{g} \end{equation} Assuming that $\Im(w)=A\ \Re(w)^b$, that is \begin{equation} w=\xi+i A \xi^b \label{a} \end{equation} and inserting (\ref{a}) in (\ref{g}) one obtains \begin{equation} \Im(g(\xi+i A \xi^b))= A \xi^b \left(1+8\xi+ O(\xi^{\min\{3,2b-1\}})\right) \end{equation} and \begin{equation} A\ \left(\Re(g(\xi+i A \xi^b))\right)^b=A \xi^b \left(1+4b\xi O(\xi^{\min\{3,2b-1\}}))\right) \end{equation} Choosing $b=2$, one sees that the curve $w=\xi+i A \xi^2$ is `conserved' by transformation $g$ except for higher order terms in $\xi$. \section{Density of zeros} We have seen that the zeros pinch the real axis only at $T=0$ and we proceed by studying their density in the neighbourhood of this point to establish the critical behaviour; it is important to see whether this density (which is quite small, as we have seen) goes to zero or remains finite. A numerical estimate can be obtained by simply counting the zeros (with their multiplicity). This has been done in two ways: \begin{itemize} \item considering only the zeros in the neighbourhood of the real axis and grouping them with regard to their real part (one-dimensional density); \item dividing the complex plane in equal rectangles and counting the zeros contained (two-dimensional density). \end{itemize} \begin{figure} \begin{center} \mbox{\epsfig{file=fig6.eps,height=8cm,angle=90}} \end{center} \caption{Density of zeros vs. their real part in the $t$ plane (I method)} \label{dens1} \end{figure} \begin{figure} \begin{center} \mbox{\epsfig{file=fig7.ps,height=11cm,angle=270}} \end{center} \caption{Density of zeros in the $t$ plane (log scale)} \label{dens2} \end{figure} \begin{figure} \begin{center} \mbox{\epsfig{file=fig8.ps,height=11cm,angle=270}} \end{center} \caption{3-dimensional view of the zeros density in the $t$ plane} \label{dens3} \end{figure} In the first case one obtains plots like fig. \ref{dens1} which shows the density of the zeros of the partition function with $\Im(t)<.3$ for gaskets from ${\cal G}_{10}$ to ${\cal G}_{20}$; in the second case obtains the result shown in figures \ref{dens2} and \ref{dens3} (which refers to ${\cal G}_{18}$). From fig. \ref{dens1} one can see that going from one stage to the next the density does not change appreciably except for the tail towards $0$ that grows longer but is strongly decreasing: the density at $T=0$ appears to vanish exponentially. This behaviour can also be verified by an analytical estimate. First we notice that the zeros near $T=0$ are those obtained by the repeated application of $h_2$. Indeed for $|x|\rightarrow \infty$ we have \begin{equation} h_1(x) \to k \end{equation} while \begin{equation} h_2(x) \to x+4-\frac{16}{x}+O(x^{-2}) \end{equation} and, in terms of real and imaginary part, \begin{equation} h_2(u+i v)\approx \left( u+4-\frac{16 u}{u^2+v^2}\right)+ i \left(v-\frac{16 v}{u^2+v^2}\right) \end{equation} Applying $h_2$ to $z=u+i~v$ for large $u$ one obtains $h_2(z) \simeq 4u+i~v$. In this limit the density of zeros in the $x$ plane becomes the product of two factors, one depending only on $u$ and the other on $v$: \begin{equation} d(u,v) \approx d_1(u) d_2(v) \end{equation} where $d_2(v)$ is bounded. The asymptotic behaviour of $d_1(u)$ for $u\to\infty$ can be obtained by noting that, for each set $h^k(-1)$, the zeros with real part $u$ are those obtained by a sequence ending with the application of $h_1$ followed by $h_2^n$, where $n=\frac{u}{4}+c$ and $c$ is a constant independent of $u$. So this fraction of zeros is $\frac{1}{2}$ to the power $\frac{u}{4}+c$, that is proportional to $\exp (-u/U)$ with $U=4 \ln 2$. Since the total density $d_1$ is a weighted sum of the partial densities that have the same behaviour we have: \begin{equation} d_1(u)\propto \exp(-u/U) \end{equation} To find the density in the $t$ plane we must now divide $d$ by the Jacobian of the transformation: \begin{equation} t=x^{-\frac{1}{4}} \end{equation} This Jacobian turns out to be \begin{equation} \frac{1}{16} (u^2+v^2)^{-\frac{5}{4}} \end{equation} and finally \begin{equation} \tilde{d}(t_r,t_i)\propto\left. d_2(v) e^{-u/U} (u^2+v^2)^{\frac{5}{4}} \right|_{t_r,t_i} \end{equation} where $t_r$ e $t_i$ are the real and imaginary part of $t$. For $t\rightarrow0$ (that is $u\rightarrow +\infty$) the density vanishes exponentially, as we could infer from numerical calculation, and this behaviour has the same effect as a gap near the real axis. Therefore the low temperature regime is not affected by the zeros contained in this region and one observe a situation analogous to the one dimensional case. This can be seen, for example, by comparing the behaviour of thermodynamical quantities: figure \ref{comp} shows that, even if the zeros distributions seem to be quite different, the behaviour of the specific heat for the Sierpinski gasket and the linear chain is essentially the same. \begin{figure}[p] \begin{center} \mbox{\epsfig{file=fig9.eps,height=12cm}} \end{center} \caption{Comparison between the zeros distribution in the $t$ plane and the specific heat for the linear chain (on the left) and the gasket (on the right)} \label{comp} \end{figure} \section{Conclusions} The anomalous behaviour of the density of zeros for the Ising model on the Sierpinski gasket is to be deeply related to the its self-similar geometry. This strongly suggests a careful approach to the analysis of scaling of zeros density on fractals. In particular a stimulating open problem is the relation of this scaling with the geometry of a generic self-similar structure and with known anomalous dimensions. An important step in this direction would be the study of Fisher zeros on the more complex case of a Sierpinski carpet, where an exact solution is still lacking but the Ising model is expected to have a phase transition at finite temperature.
\section{Introduction: Distinguishability of superoperators} The exciting recent developments in the theory of quantum information and computation have already established an enduring legacy. The two most far-reaching results --- that a quantum computer (apparently) can solve problems that will forever be beyond the reach of classical computers \cite{qc}, and that quantum information can be protected from errors if properly encoded \cite{qec} --- have surely earned a prominent place at the foundations of computer science. The implications of these ideas for the future of physics are less clear, but we expect them to be profound. In particular, we anticipate that our deepening understanding of quantum information will lead to new strategies for pushing back the boundaries of quantum-limited measurements. Quantum entanglement, quantum error correction, and quantum information processing can all be exploited to improve the information-gathering capability of physics experiments. In a typical quantum-limited measurement, a classical signal is conveyed over a quantum channel \cite{mabuchi}. Nature sends us a message, such as the value of a weak force, that can be regarded as a classical parameter appearing in the Hamiltonian of the apparatus (or more properly, if there is noise, its master equation). The apparatus undergoes a quantum operation $\$(a)$, and we are to extract as much information as we can about the parameter(s) $a$ by choosing an initial preparation of the apparatus, and a positive-operator-valued measure (POVM) to read it out. Quantum information theory should be able to provide a theory of the {\sl distinguishability of superoperators}, a measure of how much information we can extract that distinguishes one superoperator from another, given some specified resources that are available for the purpose. This distinguishability measure would characterize the inviolable limits on measurement precision that can be achieved with fixed resources. Many applications of quantum information theory involve the problem of distinguishing nonorthogonal quantum {\sl states}. For example, a density operator $\rho_a $ is chosen at random from an ensemble ${\cal E}=\{\rho_a,p_a\}$ (where $p_a$ is an {\it a priori} probability), and a measurement is performed to extract information about which $\rho_a$ was chosen. The problem of distinguishing {\sl superoperators} is rather different, but the two problems are related. For example, let us at first ignore noise, and also suppose that the classical force we are trying to detect is static. Then we are trying to identify a particular time-independent Hamiltonian $H_a$ that has been drawn from an ensemble $\{H_a,p_a\}$. We may choose a particular initial pure state $|\psi_0\rangle$, and then allow the state to evolve, as governed by the unknown Hamiltonian, for a time $t$; our ensemble of possible Hamiltonians generates an ensemble of pure states \begin{equation} \{|\psi_a(t)\rangle = e^{-itH_a}|\psi_0\rangle, p_a\}~. \end{equation} Since our goal is to gain as much information as possible about the applied Hamiltonian, we should choose the initial state $|\psi_0\rangle$ so that the resulting final states are maximally distinguishable. There are many variations on the problem, distinguished in part by the resources we regard as most valuable. We might have the freedom to chose the elapsed time as we please, or we might impose constraints on $t$. We might have the freedom to modify the Hamiltonian by adding an additional ``driving'' term that is under our control. We might use an {\sl adaptive} strategy, where we make repeated (possibly weak) measurements, and our choice of initial state or driving term in later measurements takes into account the information already collected in earlier measurements \cite{wiseman_adaptive}. Imposing an appropriate cost function on resources is an important aspect of the formulation of the problem, particularly in the case of the detection of a static (DC) signal. For example, we could in principle repeat the measurement procedure many times to continually improve the accuracy of our estimate. In this respect, the problem of distinguishing superoperators does not have quite so fundamental a character as the problem of distinguishing states, as in the latter case the no-cloning principle \cite{no_clone} prevents us from making repeated measurements on multiple copies of the unknown state. But for a time-dependent signal that stays ``on'' for a finite duration, there will be a well-defined notion of the optimal strategy for distinguishing one possible signal from another, once our apparatus and its coupling to the classical signal have been specified. Still, for the sake of simplicity, we will mostly confine our attention here to the case of DC signals. We don't know exactly what shape this nascent theory of the distinguishability of superoperators should take, but we hope that further research can promote the development of new strategies for performing high-precision measurements. On the one hand we envision a program of research that will be relevant to real laboratory situations. On the other hand, we seek results that are to some degree robust and general (not tied to some particular model of decoherence, or to a particular type of coupling between quantum probe and classical signal). Naturally, there is some tension between these two central desiderata; rather than focus on a specific experimental context, we lean here toward more abstract formulations of the problem. Our discussion is far from definitive; its goal is to invite a broader community to consider these issues. We will mostly be content to observe that some familiar concepts from the theory of quantum information and computation can be translated into tools for the measurement of classical forces. Some examples include superdense coding, fast database search, and the quantum Fourier transform. Naturally, the connections between quantum information theory and precision measurement have been recognized previously by many authors. Especially relevant is the work by Wootters \cite{wootters}, by Braunstein\cite{braun}, and by Braunstein and Caves \cite{braunstein} on state distinguishability and parameter estimation, and by Braginsky and others \cite{braginsky} on quantum nondemolition measurement. Though what we have to add may be relatively modest, we hope that it may lead to further progress. \section{Superdense coding: improved distinguishability through entanglement} \label{sec:superdense} Recurring themes of quantum information theory are that entanglement can be a valuable resource, and that entangled measurements sometimes can collect more information than unentangled measurements. It should not be surprising, then, if the experimental physicist finds that the best strategies for detecting a weak classical signal involve the preparation of entangled states and the measurement of entangled observables. Suppose, for example, that our apparatus is a single-qubit, whose time-independent Hamiltonian (aside from an irrelevant additive constant), can be expressed as \begin{equation} H_{\vec a}=\vec a\cdot \vec \sigma~; \end{equation} here $\vec a=(a_1,a_2,a_3)$ is an unknown three-vector, and $\sigma_{1,2,3}$ are the Pauli matrices. (We may imagine that a spin-${1\over 2}$ particle with a magnetic moment is employed to measure a static magnetic field.) By preparing an initial state of the qubit, allowing the qubit to evolve, and then performing a single measurement, we can extract at best one bit of information about the magnetic field (as Holevo's theorem \cite{holevo} ensures that the optimal POVM in a two-dimensional Hilbert space can acquire at most one bit of information about a quantum state). If we have two qubits, and measure them one at a time, we can collect at best two bits of information about the magnetic field. In principle, this could be enough to distinguish perfectly among four possible values of the field. In practice, for a generic choice of four Hamiltonians labeled by vectors $\vec a^{(1,2,3,4)}$, the optimal information gain cannot be achieved by measuring the qubits one at a time. Rather a better strategy exploits quantum entanglement. An improved strategy can be formulated by following the paradigm of superdense coding \cite{wiesner}, whereby shared entanglement is exploited to enhance classical communication between two parties. To implement superdense coding, the sender (Alice) and the receiver (Bob) use a shared Bell state \begin{equation} |\phi^+\rangle= {1\over\sqrt{2}}\left(|00\rangle + |11\rangle\right) \end{equation} that they have prepared previously. Alice applies one of the four unitary operators $\{I,\sigma_1,\sigma_2,\sigma_3\}$ to her member of the entangled pair, and then sends it to Bob. Upon receipt, Bob possesses one of the four mutually orthogonal Bell states\begin{eqnarray} \label{bellbasis} |\phi^+\rangle & = & {1\over\sqrt{2}}\left(|00\rangle+|11\rangle\right) = I \otimes I |\phi^+\rangle ~,\cr |\psi^+\rangle & = & {1\over\sqrt{2}}\left(|01\rangle+|10\rangle\right) = \sigma_1 \otimes I |\phi^+\rangle ~,\cr -i|\psi^-\rangle & = & {-i\over\sqrt{2}}\left(|01\rangle-|10\rangle\right) = \sigma_2 \otimes I |\phi^+\rangle ~,\cr |\phi^-\rangle & = & {1\over\sqrt{2}}\left(|00\rangle-|11\rangle\right) = \sigma_3 \otimes I |\phi^+\rangle~;\end{eqnarray} by performing an entangled Bell measurement (simultaneous measurements of the commuting collective observables $\sigma_1\otimes\sigma_1$ and $\sigma_3\otimes\sigma_3$), Bob can perfectly distinguish the states. Although only one qubit passes from Alice to Bob, two classical bits of information are transmitted and successfully decoded. In fact, this enhancement of the transmission rate is optimal -- with shared entanglement, no more than two classical bits can be carried by each transmitted qubit \cite{hausladen}. The lesson of superdense coding is that entanglement can allow us to better distinguish operations on quantum states, and we may apply this method to the problem of distinguishing Hamiltonians.\footnote{This idea was suggested to us by Chris Fuchs \cite{fuchs_private}.} Let us imagine that the magnitude of the magnetic field is known, but not its direction -- then we can choose our unit of time so that $|\vec a|=1$. We may prepare a pair of qubits in the entangled state $|\phi^+\rangle$, and expose only one member of the pair to the magnetic field while the other remains well shielded. In time $t$, the state evolves to \begin{eqnarray} |\psi_{\hat a}(t)\rangle & \equiv &\exp\left(-itH_{\hat a}\otimes I\right)|\phi^+\rangle \cr & = & \left[\cos t(I\otimes I) -i \sin t(\hat a\cdot\vec\sigma\otimes I) \right]|\phi^+\rangle\cr & = & \cos t |\phi^+\rangle \cr &&-i\sin t \left[a_1|\psi^+\rangle -ia_2|\psi^-\rangle + a_3|\phi^+\rangle\right]~; \end{eqnarray} the inner product between the states arising from Hamiltonians $H_{\hat a}$ and $H_{\hat b}$ becomes \begin{equation} \label{superip} \langle\psi_{\hat a}(t)|\psi_{\hat b}(t)\rangle = \cos^2 t + (\hat a \cdot \hat b) \sin^2 t~. \end{equation} For these states to be orthogonal, we require \begin{equation} \hat a \cdot \hat b = -\cot^2 t~. \end{equation} Since $\cot^2 t \ge 0$, the states are not orthogonal for any value of $t$ unless the two magnetic field directions $\hat a$ and $\hat b$ are separated by at least $90^\circ$. Now suppose that the magnetic field (of known magnitude) points in one of three directions that are related by three-fold rotational symmetry. These directions could form a planar trine with $\hat a\cdot \hat b=\hat a\cdot \hat c= \hat b\cdot \hat c = -1/2$, or a ``lifted trine'' with angle $\theta$ between each pair of directions, where $-1/2\le \cos\theta \le 0$. For any such trine of field directions, we may evolve for a time $t$ such that \begin{equation} \cot^2 t = -\cos\theta ~, \end{equation} and perform an (entangled) orthogonal measurement to determine the field. At the point of tetrahedral symmetry, $\cos\theta=-1/3$, we may add a fourth field direction such that the inner product for each pair of field directions is $-1/3$; then all four directions can be perfectly distinguished by Bell measurement. In this case of four field directions with tetrahedral symmetry, the two-bit measurement outcome achieves a two-bit information gain, if the four directions were equally likely {\it a priori}. In contrast, no adaptive strategy in which single qubits are measured one at a time can attain a two-bit information gain. This separation between the information gain attainable through entangled measurement and that attainable through adaptive nonentangled measurement, for the problem of distinguishing Hamiltonians, recalls the analogous separation noted by Peres and Wootters \cite{peres} for the problem of distinguishing nonorthogonal states. \section{Grover's database search: improved distinguishability through driving} Another instructive example is Grover's method \cite{grover} for searching an unsorted database, which (as formulated by Farhi and Gutmann \cite{farhi}) we may interpret as a method for improving the distinguishability of a set of Hamiltonians by adding a controlled driving term. Consider an $N$-dimensional Hilbert space with orthonormal basis $\{|x\rangle\}, ~x=0,1,2,\dots,N-1$, and suppose that the Hamiltonian for this system is known to be one of the $N$ operators \begin{equation} H_x=E|x\rangle\langle x| ~. \end{equation} We are to perform an experiment that will allow us to estimate the value of $x$. We could, for example, prepare the initial state ${1\over \sqrt{2}}(|y\rangle + |y'\rangle)$, allow the system to evolve for a time $T=\pi/E$, and then perform an orthogonal measurement in the basis $|\pm\rangle={1\over \sqrt{2}}(|y\rangle \pm |y'\rangle)$. Then we will obtain the outcome $|-\rangle$ if and only if one of $y,y'$ is $x$. Searching for $x$ by this method, we would have to repeat the experiment for O($N$) distinct initial states to have any reasonable chance of successfully inferring the value of $x$. Our task becomes easier if we are able to modify the Hamiltonian by adding a term that we control to drive the system. We choose the driving term to be \begin{equation} \label{grover_drive} H_D=E|s\rangle\langle s|~, \end{equation} where $|s\rangle$ denotes the state \begin{equation} |s\rangle={1\over \sqrt{N}}\sum_{y=0}^{N-1}|y\rangle~. \end{equation} Then the full Hamiltonian is \begin{equation} H'_x=H_x+H_D=E(|x\rangle\langle x| + |s\rangle\langle s|)~, \end{equation} and we can readily verify that the vectors \begin{equation} |E_{\pm}\rangle\equiv |s\rangle \pm |x\rangle \end{equation} are (unconventionally normalized!) eigenstates of $H$ with the eigenvalues \begin{equation} E_{\pm}=E\left(1\pm {1\over\sqrt{N}}\right)~. \end{equation} We may prepare the initial state \begin{equation} |s\rangle= {1\over 2}(|E_+\rangle + |E_-\rangle)~; \end{equation} since the energy splitting is $\Delta E=2E/\sqrt{N}$, after a time \begin{equation} T=\pi/\Delta E= \pi\sqrt{N}/2E~, \end{equation} this state flops to the state \begin{equation} {1\over 2}(|E+\rangle - |E-\rangle)=|x\rangle~. \end{equation} Thus, by performing an orthogonal measurement, we can learn the value of $x$ with certainty \cite{farhi}. The driving term we have chosen is the continuous time analog of the iteration employed by Grover \cite{grover} for rapid searching. And as the Grover search algorithm can be seen to be optimal, in the sense that a marked state can be identified with high probability with the minimal number of oracle calls \cite{bbbv}, so the driving term we have chosen is optimal in the sense that it enables us to identify the value of the classical parameter labeling the Hamiltonian in the minimal time, at least asymptotically for $N$ large. (In a physics experiment, the ``oracle'' is Nature, whose secrets we are eager to expose.) For this Grover-Farhi-Gutmann problem, we can make a definite statement about how to optimize expenditure of a valuable resource (time) in the identification of a system Hamiltonian. We also note that adding a driving term can sometimes improve the efficacy of the superdense coding method described in \S\ref{sec:superdense}. For example, in the case of three magnetic fields of equal magnitude with threefold symmetry, but with an angle between fields of less than $90^\circ$, applying a driving field along the line of symmetry can make the resultant field directions perfectly distinguishable. In fact, Beckman \cite{beckman} has shown that for any three field vectors forming a triangle that is isosceles or nearly isosceles, a suitable driving field can always by found such that the field directions can be distinguished perfectly. \section{Distinguishing two alternatives} Let's consider the special case in which our apparatus is known to be governed by one of two possible Hamiltonians $H_1$ or $H_2$. If the system is two dimensional, we are trying to distinguish two possible values $\vec a,\vec b$ of the magnetic field with a spin-${1\over 2}$ probe. Suppose for simplicity that the two fields have the same magnitude (normalized to unity), but differing directions. Assuming that we are unable to modify the Hamiltonian by adding a driving term, the optimal strategy is to choose an initial polarization vector that bisects the two field directions $\hat a, \hat b$. Depending on the actual value of the field, the polarization will precess on one of two possible cones. If the angle between $\hat a$ and $\hat b$ is $\theta\ge 90^\circ$, then the two possible polarizations will eventually be back-to-back; an orthogonal measurement performed at that time will distinguish $\hat a$ and $\hat b$ perfectly. But if $\theta < 90^\circ$, the two polarizations are never back-to-back; the best strategy is to wait until the angle between the polarizations is maximal, and to then perform the orthogonal measurement that best distinguishes them. We cannot perfectly distinguish the two field directions by this method. On the other hand, if we are able to apply a known driving magnetic field in addition to the unknown field that is to be determined, then two fields $\vec a$ and $\vec b$ can always be perfectly distinguished. If we apply the field $-\vec b$, then the problem is one of distinguishing the trivial Hamiltonian from \begin{equation} H_{\rm diff}=(\vec a - \vec b)\cdot \vec\sigma~. \end{equation} We can choose an initial polarization orthogonal to $\vec a - \vec b$, and wait just long enough for $H_{\rm diff}$ to rotate the polarization by $\pi$. Then an orthogonal measurement perfectly distinguishes $H_{\rm diff}$ from the trivial Hamiltonian. Evidently, the same strategy can be applied to distinguish two Hamiltonians $H_1$ and $H_2$ in a Hilbert space of arbitrary dimension. We drive the system with $-H_2$; then to distinguish the trivial Hamiltonian from $H_1-H_2$, we chose the initial state \begin{equation} {1\over\sqrt{2}}\left(|E_{\rm min}\rangle + |E_{\rm max}\rangle\right)~, \end{equation} where $E_{\rm min},E_{\rm max}$ are the minimal and maximal eigenvalues of $H_1-H_2$. After a time $t$ with \begin{equation} t(E_{\rm max}-E_{\rm min})=\pi~, \end{equation} this state evolves to the orthogonal state ${1\over\sqrt{2}}\left(|E_{\rm min}\rangle - |E_{\rm max}\rangle\right)$, so that the trivial and nontrivial Hamiltonians can be perfectly distinguished. In the case of the two-dimensional version of the ``Grover problem'' with $H_1= |0\rangle\langle 0|$ and $H_2=|1\rangle\langle 1 |$, this choice for the driving Hamiltonian actually outperforms the Grover driving term of Eq.~(\ref{grover_drive}) --- the two Hamiltonians can be distinguished in a time that is shorter by a factor of $\sqrt{2}$. So while the Grover strategy is optimal for asymptotically large $N$, it is not actually optimal for $N=2$. \section{Distinguishing two alternatives in a fixed time} Let us now suppose that we are to distinguish between two time-independent Hamiltonians $H_1$ and $H_2$, and that a {\sl fixed duration} $t$ has been allotted to perform the experiment. Is the driving strategy described above (in which $-H_2$ is added to the Hamiltonian) always the best possible? If we have the freedom to add a driving term of our choice, then we may assume without loss of generality that we are to distinguish the nontrivial Hamiltonian $H$ from the trivial Hamiltonian $0$. As already noted, if the largest difference $\Delta E=E_{\rm max}-E_{\rm min}$ of eigenvalues of $H$ satisfies $t\Delta E \ge \pi$, then $H$ can be perfectly distinguished from $0$; let us therefore suppose that $t \Delta E < \pi$. If we add a {\sl time-independent} driving term $K$ to the Hamiltonian, and choose an initial state $|\psi_0\rangle$, then after a time t, we will need to distinguish the two states \begin{equation} \label{two_states} e^{-i t K}|\psi_0\rangle~, \quad e^{-it(H + K)}|\psi_0\rangle~. \end{equation} Two pure states will be more distinguishable when their inner product is smaller. Therefore, to best distinguish $H+K$ from $K$, we should choose $|\psi_0\rangle$ to minimize the inner product \begin{equation} \left|\langle\psi_0|e^{it K} e^{-it(H + K)}|\psi_0\rangle\right|~. \end{equation} If we expand $|\psi_0\rangle$ in terms of the eigenstates $\{|a\rangle\}$ of $e^{it K} e^{-it(H + K)}$ with eigenvalues $\{e^{-itE_a}\}$, \begin{equation} |\psi_0\rangle=\sum_a \alpha_a|a\rangle~, \end{equation} this inner product becomes \begin{equation} \label{you_tee} \left|\langle\psi_0|e^{it K} e^{-it(H + K)}|\psi_0\rangle\right|= \left|\sum_a |\alpha_a|^2 e^{-itE_a}\right|~. \end{equation} The right-hand side of Eq.~(\ref{you_tee}) is the modulus of a convex sum of points on the unit circle. Assuming the modulus is bounded from zero, it attains its minimum when $|\psi_0\rangle$ is the equally weighted superposition of the extremal eigenstates of $e^{it K} e^{-it(H + K)}$ -- those whose eigenvalues are maximally separated on the unit circle. For $K=0$, the minimum is $\cos\left( t\Delta E/2\right)$, where $\Delta E$ is the difference of the maximal and minimal eigenvalues of $H$. We prove in Appendix A that turning on a nonzero driving term $K$ can never cause the extremal eigenvalues to separate further, and therefore can never improve the distinguishability of the two states in Eq.~{\ref{two_states}.\footnote{That this might be the case was suggested to us by Chris Fuchs \cite{fuchs_private}.} Therefore, $K=0$ is the optimal driving term for distinguishing two Hamiltonians. In other words, if we wish to distinguish between two Hamiltonians $H_1$ and $H_2$, it is always best to turn on a driving term that precisely cancels one of the two. The above discussion encompasses the strategy of introducing an ancilla entangled with the probe (which proved effective for the problem of distinguishing three or more alternatives). If we wish to distinguish two Hamiltonians $H_1\otimes I$ and $H_2\otimes I$ that both act trivially on the ancilla, the optimal driving term exactly cancels one of them ({\it e.g.}, $K= - H_2\otimes I$), and so it too acts trivially on the ancilla. We derive no benefit from the ancilla when there are only two alternatives. Similarly, if we are trying to distinguish only two time-independent signals in an allotted time, it seems likely there is no advantage to performing a sequence of weak measurements, and adapting the driving field in response to the incoming stream of measurement data. \section{More alternatives: adaptive driving} Now suppose that there are $N$ possible Hamiltonians ${H_1, H_2, \ldots, H_N}$. If there is no time limitation, we can distinguish them perfectly by implementing an adaptive procedure; we make a series of measurements, modifying our driving term and initial state in response to the stream of measurement outcomes. The correct Hamiltonian can be identified by pairwise elimination. First, assume that either $H_1$ or $H_2$ is the actual Hamiltonian, and apply a driving term to perfectly distinguish them, say $H_{D}=-H_1$. After preparing the appropriate initial state and waiting the appropriate time, we make an orthogonal measurement with two outcomes --- the result indicates that either $H_1$ or $H_2$ is the actual Hamiltonian.\footnote{Actually, in a Hilbert space of high dimension, we can make a more complete measurement that will typically return the result that neither $H_1$ nor $H_2$ is the actual Hamiltonian.} If the result is $H_1$, there are two possibilities: either $H_1$ really is the Hamiltonian, or the assumption that one of $H_1$ or $H_2$ is the Hamiltonian was wrong. Either way, $H_2$ has been eliminated. Similarly, if $H_2$ is found, $H_1$ is eliminated. This procedure can then be repeated, eliminating one Hamiltonian per measurement, thereby perfectly distinguishing among the $N$ Hamiltonians in a total of $N-1$ measurements. This algorithm is quite inefficient, however. The measurement record is $N-1$ bits long, while the information gain is only $\log N$ bits. \section{Adaptive phase measurement and the semiclassical quantum Fourier transform} Far more efficient adaptive procedures can be formulated in some cases. Consider, for example, a single qubit in a magnetic field of known direction but unknown magnitude, so that \begin{equation} H_\omega= {\omega\over 2}\sigma_3~, \end{equation} and let us imagine that the value of the frequency $\omega$ is chosen equiprobably from among $N=2^n$ equally spaced possible values. Without loss of generality, we may normalize the field so that the possible values range from 0 to $1-2^{-n}$; then $\omega$ has a binary expansion \begin{equation} \omega = .\omega_1 \omega_2 \ldots \omega_n \end{equation} that terminates after at most $n$ bits. The initial state $|\psi_0\rangle={1\over \sqrt{2}}(|0\rangle + |1\rangle)$ evolves in time $t$ to \begin{equation} |\psi(t)\rangle_\omega= e^{-itH_\omega}|\psi_0\rangle ={1\over \sqrt{2}}(|0\rangle + e^{- i \omega t} |1\rangle) \end{equation} (up to an overall phase). If we wait for a time $t_n=\pi 2^n$, the final state is \begin{equation} |\psi(t_n)\rangle_\omega={1\over \sqrt{2}}(|0\rangle + e^{-i \pi \omega_n } |1\rangle) ~. \end{equation} Now measurement in the $\{{1\over\sqrt{2}}(|0\rangle \pm |1\rangle\})$ basis indicates (with certainty) whether the bit $\omega_n$ is 0 or 1. This outcome divides the set of possible Hamiltonians in half, providing one bit of classical information. The set of remaining possible Hamiltonians is still evenly spaced, but it may have a constant offset, depending on the value of $\omega_n$. However, the value of $\omega_n$ is now known, so the offset can be eliminated. Specifically, if we again prepare $|\psi_0\rangle$ and now evolve for a time $t_{n-1}=\pi 2^{n-1}$, we obtain the final state \begin{equation} |\psi(t_{n-1})\rangle_\omega={1\over \sqrt{2}}(|0\rangle + e^{-i \pi \> (\omega_{n-1} . \omega_n)} |1\rangle)~. \end{equation} Since $\omega_n$ is known, we can perform a phase transformation (perhaps by applying an additional driving magnetic field) to eliminate the phase $e^{-i \pi \> (. \omega_n)}$; Measuring again in the $\{{1\over\sqrt{2}}(|0\rangle \pm |1\rangle\})$ basis determines the value of $\omega_{n-1}$. By continuing this procedure until all bits of $\omega$ are known, we perfectly distinguish the $2^n$ possible Hamiltonians in just $n$ measurements. The procedure is optimal in the sense that we gain one full bit of information about the Hamiltonian in each measurement. Up until now we have imagined that the frequency $\omega$ takes one of $2^n$ equally spaced discrete values, but no such restriction is really necessary. Indeed, what we have described is precisely the implementation of the $n$-qubit semiclassical quantum Fourier transform as formulated by Griffiths and Niu \cite{griffiths} (whose relevance to phase estimation was emphasized by Cleve {\it et al.} \cite{cleve}). Thus the same procedure can be applied to obtain an estimate of the frequency to $n$-bit precision, even if the frequency is permitted to take an arbitrary real value in the interval $[0,1)$. Suppose that we attach to $n$ spins the labels $\{0,1,\ldots,n-2,n-1\}$, and expose the $k$th spin to the field for time $\pi 2^{k+1}$; we thus prepare the $n$-qubit state \begin{equation} \prod_{k=0}^{n-1} {1\over\sqrt{2}}\left(|0\rangle + e^{-i\pi\omega\cdot 2^{k+1}}|1\rangle\right)={1\over 2^{n/2}}\sum_{y=0}^{2^n-1} e^{-2\pi i \omega \cdot y}|y\rangle~. \end{equation} The adaptive algorithm is equivalent to the quantum Fourier transform followed by measurement; hence the $n$-bit measurement outcome $\tilde \omega$ occurs with probability \begin{equation} {\rm Prob}_\omega(\tilde \omega) = \left| {1 \over 2^n} \sum_{y=0}^{2^n - 1} \exp[-2 \pi i y (\omega-\tilde \omega)] \right|^2. \end{equation} If $\omega$ really does terminate in $n$ bits, then the outcome $\tilde\omega$ is guaranteed to be its correct binary expansion. But even if the binary expansion of $\omega$ does not terminate, the probability that our estimate $\tilde \omega$ is correct to $n$ bits of precision is still of order one.\footnote{We might also use the QFT to {\sl compute} eigenvalues of a known many-body Hamiltonian, rather than {\sl measure} eigenvalues of an unknown one \cite{lloyd}.} Of course, to measure the frequency to a precision $\Delta \omega$ of order $2^{-n}$, we need to expose our probe spins to the unknown Hamiltonian for a total time $T$ of order $2\pi\cdot 2^{n}$. The accuracy is limited by an energy-time uncertainty relation of the form $T\Delta\omega\sim 1$. The semiclassical quantum Fourier transform provides an elegant solution to the problem of performing an ideal ``phase measurement'' in the Hilbert space of $n$ qubits. More broadly, any $N$-dimensional Hilbert space with a preferred basis $\{|k\rangle, ~k=0,1,\dots, N-1\}$ has a complementary basis of {\sl phase states} \begin{equation} |\varphi\rangle={1\over\sqrt{N}} \sum_{k=o}^{N-1}e^{ik\varphi}|k\rangle~, \end{equation} with \begin{equation} \varphi = 2\pi j/N~,\quad j=0,1,\dots,N-1~. \end{equation} For example, the Hilbert space could be the truncated space of a harmonic oscillator like a mode of the electromagnetic field, with the occupation number restricted to be less than $N$; then the states $|\varphi\rangle$ are the ``phase squeezed'' states of the oscillator that have minimal phase uncertainty. Since a POVM in an $N$-dimensional Hilbert space can acquire no more than $\log N$ bits of information about the preparation of the quantum state, the phase of an oscillator with occupation number less than $N$ can be measured to at best $\log N$ bits of accuracy. While it is easy to do an orthogonal measurement in the occupation number basis with an efficient photodetector, an orthogonal measurement in the $|\varphi\rangle$ basis is quite difficult to realize in the laboratory \cite{wiseman}. But if the standard basis is the computational basis in the $2^n$-dimensional Hilbert space of $n$ qubits, then an ideal phase measurement is simple to realize. Since the phase eigenstates are actually not entangled states, we can carry out the measurement -- {\sl adaptively} -- one qubit at a time. Note that if we had an arbitrarily powerful quantum computer with an arbitrarily large amount of quantum memory, then adaptive measurement strategies might seem superfluous. We could achieve the same effect by introducing a large ancilla and a driving Hamiltonian that acts on probe and ancilla, with all measurements postponed to the very end. But the semiclassical quantum Fourier transform illustrates that adaptive techniques can reduce the complexity of the quantum information processing required to perform the measurement. In many cases, an adaptive strategy may be realizable in practice, while the equivalent unitary strategy is completely infeasible. \section{Distinguishability and decoherence} In all of our examples so far, we have ignored noise and decoherence. In practice, decoherence may compromise our ability to decipher the classical signal with high confidence. Finding ways to improve measurement accuracy by effectively coping with decoherence is an important challenge faced by quantum information theory. If there is decoherence, our aim is to gain information about the value of a parameter in a master equation rather than a Hamiltonian. To be concrete, consider a single qubit governed by an unknown Hamiltonian $H$, and also subject to decoherence described by the ``depolarizing channel;'' the density matrix $\rho$ of the qubit obeys the master equation \begin{equation} \dot \rho= -i[H,\rho] - \Gamma\left(\rho-{1\over 2} I\right)~, \end{equation} where $\Gamma$ is the (known) damping rate. If we express $\rho$ in terms of the polarization vector $\vec P$, \begin{equation} \rho={1\over 2}(I+\vec P\cdot\vec\sigma)~, \end{equation} and the Hamiltonian as \begin{equation} H={\omega\over 2}~\hat a\cdot \sigma~, \end{equation} then the master equation becomes \begin{equation} \dot{\vec P}=\omega(\hat a \times \vec P) - \Gamma \vec P~.\ \end{equation} The polarization precesses uniformly with circular frequency $\omega$ about the $\hat a$-axis as it contracts with lifetime $\Gamma^{-1}$. Suppose that we are to distinguish among two possible Hamiltonians, which are assumed to be equiprobable. If we are able to add a driving term, we may assume that the two are the trivial Hamiltonian and \begin{equation} H={\omega\over 2}~\sigma_3~. \end{equation} We choose the initial polarization vector $P_0=(1,0,0)$. Then if the Hamiltonian is trivial, the polarization contracts as \begin{equation} \vec P(t)_{\rm triv}= e^{-\Gamma t}(1,0,0)~, \end{equation} while under the nontrivial Hamiltonian it contracts and rotates as \begin{equation} \vec P(t)_{\rm nontriv}=e^{-\Gamma t}(\cos \omega t,\sin \omega t, 0)~. \end{equation} When is the best time to measure the polarization? We should wait until $\vec P_{\rm triv}$ and $\vec P_{\rm nontriv}$ point in distinguishable directions, but if we wait too long, the states will depolarize. The optimal measurement to distinguish the two is an orthogonal measurement of the polarization along the axis normal to the bisector of the vectors $\vec P(t)_{\rm triv}$ and $\vec P(t)_{\rm nontriv}$. At time $t$ the probability that this measurement identifies the Hamiltonian incorrectly is \begin{equation} P_{\rm error}= {1\over 2} - {1\over 2} e^{-\Gamma t}\left|\sin\left({\omega t\over 2}\right)\right|~. \end{equation} This error probability is minimized, and the information gain from the measurement is maximized, at a time $t$ such that \begin{equation} \tan\left({\omega t\over 2}\right)= {\omega\over 2\Gamma}~. \end{equation} If $\Gamma/\omega<<1$, this time is close to $\pi/\omega$, the time we would measure to perfectly distinguish the Hamiltonians in the absence of decoherence. But if $\Gamma/\omega >>1$, then we should measure after a time $t\sim \Gamma^{-1}$ comparable to the lifetime. More generally, consider an ensemble of two density operators $\rho_1$ and $\rho_2$ with {\it a priori}} probabilities $p_1$ and $p_2$ (where $p_1 + p_2=1$), and imagine that an unknown state has been drawn from this ensemble. A procedure for deciding whether the unknown state is $\rho_1$ or $\rho_2$ can be modeled as a POVM with two outcomes. The two-outcome POVM that minimizes the probability of making an incorrect decision is a measurement of the orthogonal projection onto the space spanned by the eigenstates of $p_1\rho_1-p_2 \rho_2$ with positive eigenvalues \cite{helstrom,fuchs_thesis}. The minimal error probability achieved by this measurement is \begin{equation} P_{\rm error}={1\over 2}- {1\over 2}{\rm tr}\left|p_1\rho_1-p_2\rho_2\right|~. \end{equation} Correspondingly, if we are to identify an unknown superoperator as one of $\$_1$ and $\$_2$ (with {\it a priori} probabilities $p_1$ and $p_2$), then the way to distinguish $\$_1,\$_2$ with minimal probability of error is to choose our initial state $\rho_0=|\psi_0\rangle\langle \psi_0|$ to minimize\footnote{We thank Chris Fuchs for a helpful discussion of this point.} \begin{equation} \label{super_error} P_{\rm error}={1\over 2}- {1\over 2}{\rm tr}\left|\left(p_1\$_1-p_2\$_2\right)\rho_0\right|~. \end{equation} In the case of interest to us, the superoperators $\$_1$ and $\$_2$ are obtained my integrating, for time $t$, master equations with Hamiltonians $H_1$ and $H_2$ respectively. We minimize the error probability in Eq.~(\ref{super_error}) with respect to $t$ to complete the optimization. \section{Entanglement and frequency measurement} Consider again the case in which the Hamiltonian is known to be of the form \begin{equation} H_{\omega}={\omega\over 2}~\sigma_3~, \end{equation} but where the frequency $\omega$ is unknown. For the moment, let us neglect decoherence, but suppose that we have been provided with a large number $n$ of qubits that we may use to perform an experiment to determine $\omega$ in a {\sl fixed total time} $t$. What is the most effective way to employ our qubits? Consider two strategies. In the first, we prepare $n$ identical qubits polarized along the $x$-axis. They precess in the field described by $H_\omega$ for time $t$, and then the spin along the $x$-axis is measured. Each spin will be found to be pointing ``up'' with probability \begin{equation} P = {1\over 2}(1+ \cos\omega t) \end{equation} Because the measurement is repeated many times, we will be able to estimate the probability $P$ to an accuracy \begin{equation} \label{n_qubits} \Delta P=\sqrt{P(1-P)/n}={|\sin\omega t|\over 2\sqrt{n}}~. \end{equation} and so determine the value of $\omega$ to accuracy \begin{equation} \label{shot_noise} \Delta \omega = {\Delta P\over t |dP/d(\omega t)|}={1\over t \sqrt n}~. \end{equation} The accuracy improves like $1/\sqrt{n}$ as we increase the number of available qubits with the time $t$ fixed. The second strategy is to prepare an entangled ``cat'' state of $n$ ions \begin{equation} |\psi_0\rangle = {1\over \sqrt{2}}(|000\dots0\rangle + |111\dots 1\rangle)~. \end{equation} The advantage of the entangled state is that it precesses $n$ times faster than a single qubit; in time $t$ it evolves to \begin{equation} |\psi(t)\rangle= {1\over \sqrt{2}}(|000\dots0\rangle + e^{i n\omega t}|111\dots 1\rangle) \end{equation} (up to an overall phase). If we now perform an orthogonal measurement that projects onto the basis ${1\over \sqrt{2}}(|000\dots0\rangle \pm |111\dots 1\rangle)$ ({\it e.g.} a measurement of the entangled observable $\sigma_1\otimes\sigma_1\otimes\cdots\otimes\sigma_1$) then we will obtain the ``+'' outcome with probability \begin{equation} P={1\over 2}(1+\cos n\omega t)~. \end{equation} By this method, $n\omega t$ can be measured to order one accuracy, so that \begin{equation}\ \label{linear_noise} \Delta\omega\simeq { 1\over tn}~, \end{equation} a more favorable scaling with $n$ than in Eq.~(\ref{shot_noise}). This idea of exploiting the rapid precession of entangled states to achieve a precision beyond the shot-noise limit has been proposed in both frequency measurement \cite{wineland} and optical interferometry \cite{yurke}. (One realization of this idea is the proposal by Caves \cite{caves} to allow a squeezed vacuum state to enter the dark port of an interferometer; the squeezing induces the $n$ photons entering the other port to make correlated ``decisions'' about which arm of the interferometer to follow.) \section{Entanglement versus decoherence} In both Eq.~(\ref{shot_noise}) and Eq.~(\ref{linear_noise}), the accuracy of the frequency measurement improves with the elapsed time $t$ as $1/t$. But so far we have neglected decoherence. If the single-qubit state decays at a rate $\Gamma$, then we have seen that the optimal time at which to perform a measurement will be of order $\Gamma^{-1}$. The entangled strategy will still be better if we are constrained to perform the measurement in a time $t<<\Gamma^{-1}$, but further analysis is needed to determine which method is better if we are free to choose the time $t$ to optimize the accuracy. In fact, as Huelga {\it et al.} \cite{huelga} have emphasized, an entangled state is fragile, and its faster precession can be offset by its faster decay rate. Suppose that two qubits are available, both independently subjected to the depolarizing channel with decay rate $\Gamma$. If we prepare the unentangled state, each qubit has the initial pure-state density matrix \begin{equation} \rho_0={1\over 2}(I+\sigma_1)~ \end{equation} polarized along the $x$-axis, and evolves in time $t$ to \begin{equation} \rho(t)={1\over 2}[I+e^{-\Gamma t}(\sigma_1~\cos\omega t+ \sigma_2~\sin\omega t)]~. \end{equation} If we now measure $\sigma_1$, we obtain the $+$ result with probability \begin{equation} \label{nonentangle_prob} \label{single_decohere} P={\rm tr}\left({1\over 2}(I+\sigma_1)\rho(t)\right)={1\over 2}(1+e^{-\Gamma t}\cos\omega t)~. \end{equation} Now suppose that the initial state is the Bell state $|\phi^+\rangle$ of two qubits, with density matrix \begin{equation} \rho_0={1\over 4}\left(I\otimes I+ \sigma_3\otimes\sigma_3 + \sigma_1\otimes\sigma_1-\sigma_2\otimes\sigma_2\right)~. \end{equation} If both spins precess and depolarize independently, this state evolves to \begin{eqnarray} \rho(t)& = & {1\over 4} [I\otimes I+ e^{-2\Gamma t}\big(\sigma_3\otimes\sigma_3 \nonumber\\ & + &\cos 2\omega t(\sigma_1\otimes\sigma_1-\sigma_2\otimes\sigma_2) \nonumber\\ & + & \sin 2\omega t(\sigma_1\otimes\sigma_2+\sigma_2\otimes\sigma_1)\big)]~; \end{eqnarray} if we measure the observable $\sigma_1\otimes\sigma_1$, we find the + outcome with probability \begin{eqnarray} \label{entangle_prob} P & = &{\rm tr}\left({1\over 2}(I\otimes I +\sigma_1\otimes\sigma_1)\rho(t)\right)\nonumber \\ & = & {1\over 2}(1+e^{-2\Gamma t}\cos2 \omega t)~. \end{eqnarray} Note that Eq.~(\ref{entangle_prob}) has exactly the same functional form as Eq.~(\ref{nonentangle_prob}), but with $t$ replaced by $2t$. Therefore, the entangled measurement performed in time $t/2$ collects exactly as much information about the frequency $\omega$ as the measurement of a single ion performed in time $t$. If we have two qubits and total time $t$ available, we can either perform the entangled measurement twice (taking time $t/2$ each time), or perform measurements on each qubit independently (taking time $t$). Either way, we obtain two outcomes and collect exactly the same amount of information on the average. More generally, suppose that we have $n$ qubits and a total time $T>> 1/ \Gamma$ available. We can use these qubits to perform altogether $nT/t$ independent single-qubit measurements, where each measurement requires time $t$. Plugging Eq.~(\ref{single_decohere}) into Eq.~(\ref{n_qubits}) and Eq.~(\ref{shot_noise}) (with $n$ replaced by $nT/t$), and choosing $\cos\omega t\sim 0$ to optimize the precision, we find that the frequency can be determined to accuracy \begin{equation} \Delta\omega=\left({1\over t}\right)\cdot {e^{\Gamma t}\over \sqrt{nT/t}}={1\over \sqrt{nT}}\cdot {e^{\Gamma t}\over \sqrt{t}}~. \end{equation} This precision is optimized if we choose $\Gamma t=1/2$, where we obtain \cite{huelga} \begin{equation} \Delta\omega = \sqrt{2e\Gamma\over nT}~. \end{equation} On the other hand, we could repeat the experiment $T/t$ times using the $n$-qubit entangled state. Then we would obtain a precision \begin{equation} \Delta\omega = \left({1\over nt}\right)\cdot {e^{n\Gamma t}\over \sqrt{T/t}}={1\over \sqrt{nT}}\cdot {e^{n\Gamma t}\over \sqrt{nt}}~, \end{equation} the same function as for uncorrelated qubits, but with $t$ replaced by $nt$. Thus the optimal precision is the same in both cases, but is attained in the uncorrelated case by performing experiments that take $n$ times longer than in the correlated case. That the entangled states offer no advantage in the determination of $\omega$ was one of the main conclusions of Huelga {\it et al.} \cite{huelga}. A similar conclusion applies to estimating the difference in path length between two arms of an interferometer using a specified optical power, if we take into account losses and optimize with respect to the number of times the light bounces inside the interferometer before it escapes and is detected. We would like to make the (rather obvious) point that this conclusion can change if we adopt a different model of decoherence, and in particular if the qubits do not decohere independently. As a simple example of correlated decoherence, consider the case of two qubits with $4 \times 4$ density matrix $\rho$ evolving according to the master equation \begin{equation} \dot \rho = -i[H,\rho] - \Gamma \left(\rho-I/4\right)~. \end{equation} This master equation exhibits the analog, in the four-dimensional Hilbert space, of the uniform contraction of the Bloch sphere described by the depolarizing channel in the case of a qubit. Because the decoherence picks out no preferred direction in the Hilbert space (or any preferred tensor-product decomposition), we call this model ``symmetric decoherence.'' Under this master equation, with both qubits subjected to $H_\omega$ and to symmetric decoherence, the Bell state $\rho_0=|\phi^+\rangle\langle\phi^+|$ evolves in time $t$ to the state \begin{eqnarray} \rho(t)& = & {1\over 4} [I\otimes I+ e^{-\Gamma t}\big(\sigma_3\otimes\sigma_3 \nonumber\\ & + &\cos 2\omega t(\sigma_1\otimes\sigma_1-\sigma_2\otimes\sigma_2) \nonumber\\ & + & \sin 2\omega t(\sigma_1\otimes\sigma_2+\sigma_2\otimes\sigma_1)\big)]~, \end{eqnarray} so that a measurement of $\sigma_1\otimes\sigma_1$ yields the + outcome with probability \begin{equation} \label{ent_symmetric} P ={1\over 2}(1+e^{-\Gamma t}\cos2 \omega t)~. \end{equation} On the other thing, the initial product state \begin{equation} \rho_0={1\over 4} (I+\sigma_1)\otimes(I+\sigma_1) \end{equation} becomes entangled as a result of symmetric decoherence. Were the Hamiltonian trivial, it would evolve to \begin{equation} \rho(t)={1\over 4} I\otimes I + {1\over 4}e^{-\Gamma t}(\sigma_1\otimes I+ I\otimes\sigma_1 +\sigma_1\otimes\sigma_1)~. \end{equation} Including the precession \begin{equation} \sigma_1\to \sigma_1 \cos\omega t + \sigma_2\sin\omega t~, \end{equation} we obtain \begin{equation} \rho(t)={1\over 4} I\otimes I + {1\over 4}e^{-\Gamma t}( \sigma_1\otimes I~\cos\omega t + \cdots~)~, \end{equation} so that measurement of the single-qubit observable $\sigma_1\otimes I$ yields the + outcome with probability \begin{equation} \label{single_symmetric} P={\rm tr}\left({1\over 2}(I\otimes I +\sigma_1\otimes I)\rho(t)\right)={1\over 2}(1+e^{-\Gamma t}\cos\omega t)~. \end{equation} Comparing Eq.~(\ref{single_symmetric}) and Eq.~(\ref{ent_symmetric}), the important thing to notice is that with symmetric decoherence, entangled states decay no faster than product states; therefore, we can enjoy the benefit of entanglement (faster precession) without paying the price (faster decay). To establish more firmly that entangled strategies outperform nonentangled strategies in the symmetric decoherence model, we should consider more closely what are the optimal final measurements for these two types of initial states. To give the problem a precise information-theoretic formulation, we return to the problem of distinguishing two cases, the trivial Hamiltonian and $H_\omega$, which are assumed to be equiprobable. For either the product initial state or the entangled initial state, we evolve for time $t$, and then perform the best measurement that distinguishes between evolution governed by $H_\omega$ and trivial evolution. In both cases, the measurement is permitted to be an entangled measurement; that is, we optimize with respect to all POVM's in the four-dimensional Hilbert space. In either case (initial product state or initial entangled state), we can find the two-outcome POVM that identifies the Hamiltonian with minimal probability of error. When there is no decoherence, this POVM (when restricted to the two-dimensional subspace containing the two pure states to be distinguished) is the familiar orthogonal measurement that best distinguishes two pure states of a qubit. In fact, for symmetric decoherence, this same measurement minimizes the error probability for any value of the damping rate $\Gamma$. It is thus the two-outcome measurement with the maximal information gain (the measurement outcome has maximal mutual information with the choice of the Hamiltonian). Although we don't have a proof, we can make a reasonable guess that, for symmetric decoherence, this two-outcome measurement has the maximal information gain of any measurement, including POVM's with more outcomes. If either initial state evolves for time $t$, and then this optimal POVM is performed, the error probability can be expressed as \begin{equation} P_{\rm error} = {1\over 2} - {1\over 2}e^{-\Gamma t}\left|\sin\theta(t)\right|~; \end{equation} here $\theta(t)$ is the angle between the states --- that is, $\cos \theta(t)$ is the inner product of the evolving and static states, in the limit of no damping ($\Gamma=0$). For the entangled initial state, we have \begin{equation} \theta_{\rm entangled}=\omega t~, \end{equation} and for the product initial state, we have \begin{equation} \cos\theta_{\rm product}=\cos^2 \left({\omega t\over 2}\right)~. \end{equation} Since \begin{equation} |\cos \theta_{\rm entangled}| = |\cos\omega t| \le {1\over 2}(1+\cos \omega t) =|\cos\theta_{\rm product}| \end{equation} for $\cos\theta_{\rm entangled}\ge 0$, the error probability achieved by the entangled initial state is smaller than that achieved by the product state for $0 < \omega t< \pi/2$, which is sufficient to ensure that the error probability optimized with respect to $t$ is always smaller in the entangled case for any nonzero value of $\Gamma$. Similarly, if we optimize the information gain with respect to $t$, the entangled strategy has the higher information gain for all $\Gamma>0$. The improvement in information gain (in bits) achieved using an entangled initial state rather than a product initial state is plotted in Fig.~1 as a function of $\Gamma/\omega$. The maximum improvement of about .136 bits occurs for $\Gamma/\omega\sim .379$. \begin{figure}[t] \begin{center} \leavevmode \epsfxsize=3in \epsfbox{entdec.eps} \end{center} \caption{Improvement in information gain (in bits) achieved by using an entangled initial state, as a function of the ratio of decoherence rate $\Gamma$ to precession frequency $\omega$.} \end{figure} We have already seen in \S II that, even in the absence of decoherence, an entangled strategy may outperform an unentangled strategy if we are trying to distinguish more than two alternatives. This advantage will persist when sufficiently weak decoherence is included, whether correlated or uncorrelated. In that event, since only one member of an entangled pair is exposed to the unknown Hamiltonian, we may be able to shelter the other member of the pair from the ravages of the environment, slowing the decay of the state and strengthening the signal. \section{Conclusions} We feel that quantum information theory, having already secured a central position at the foundations of computer science, will eventually erect bridges connecting with many subfields of physics. The results reported here (and other related examples) give strong hints that ideas emerging from the theory of quantum information and computation are destined to profoundly influence the experimental physics techniques of the future. We have only scratched the surface of this important subject. Among the many issues that deserve further elaboration are the connections between superoperator distinguishability and superoperator norms, the efficacy of the quantum Fourier transform in the presence of decoherence, the measurement of continuous quantum variables, the applications of quantum error correction, and the detection of time-dependent signals. \acknowledgments We thank Constantin Brif, Jon Dowling, Steven van Enk, Jeff Kimble, Alesha Kitaev, and Kip Thorne for instructive discussions about quantum measurement. We are especially grateful to Hideo Mabuchi for introducing us to this fascinating subject, to Chris Fuchs for sharing his insights into state distinguishability, and to Dave Beckman for discussions on improving the superdense coding method by applying a driving field. Thanks to Barry Simon for useful comments on the theorem in Appendix A, and for persuading us that it is not completely trivial. We also thank C. Woodward for helpful correspondence. A.~M.~C. and J.~R. received support from Caltech's Summer Undergraduate Research Fellowship (SURF) program, and A.~M.~C. received a fellowship endowed by Arthur R. Adams. This work has been supported in part by the Department of Energy under Grant No. DE-FG03-92-ER40701, and by DARPA through the Quantum Information and Computation (QUIC) project administered by the Army Research Office under Grant No. DAAH04-96-1-0386. \section*{Appendix A: Fixed-time-driving theorem} In this appendix, we sketch the proof of the theorem stated in \S V. For a unitary $N\times N$ matrix $U$, we define ${\rm maxarg}(U)$ to be the largest argument of an eigenvalue of $U$, where the argument takes values in the interval $(-\pi,\pi]$. Similarly, ${\rm minarg}(U)$ is the minimum argument of an eigenvalue of $U$. Our theorem is:\smallskip {\bf Theorem 1.} {\sl If $H$ and $K$ are finite-dimensional Hermitian matrices, and $\parallel H\parallel_{\rm sup} <\pi$, then \begin{eqnarray} \label{thm1} {\rm maxarg}\left(e^{iK} e^{-i(H+K)}\right) & \le & {\rm maxarg}(e^{-iH}) ~,\\ \label{thm2} {\rm minarg}\left(e^{iK} e^{-i(H+K)}\right) & \ge & {\rm minarg}(e^{-iH})~. \end{eqnarray} } To prove the theorem, we begin with:\smallskip {\bf Lemma 2}. {\sl For unitary $U$ with ${\rm maxarg}(U) \ne \pi$, and Hermitian $A$, \begin{eqnarray} \label{maxlemma2} & &{\rm maxarg} (Ue^{i\varepsilon A})\le {\rm maxarg}(U) + {\rm maxarg}(e^{i\varepsilon A})+ O(\varepsilon^2)~,\nonumber\\ & & \\ \label{minlemma2} & &{\rm minarg} (Ue^{i\varepsilon A})\ge {\rm minarg}(U) + {\rm minarg}(e^{i\varepsilon A})- O(\varepsilon^2)~.\nonumber\\ & & \end{eqnarray} } \noindent {\sl Proof}: Write $U=e^{iB}$, where $B$ is Hermitian and \newline $\parallel B \parallel_{\rm sup} < \pi$; then maxarg$(e^{iB})={\rm max}(B)$, where ${\rm max}(B)$ denotes the maximum eigenvalue of $B$. From the Baker-Campbell-Hausdorff formula, we have \begin{equation} e^{iB}e^{i\varepsilon A}= \exp i\left(B+\varepsilon A + {i\over 2}\varepsilon [C,B] + O(\varepsilon^2)\right) ~, \end{equation} where $C$ is linear in $A$. Then lowest-order eigenvalue perturbation theory tells us that \begin{eqnarray} & &{\rm max}\left(B+\varepsilon A + {i\over 2}\varepsilon [C,B]\right)\nonumber\\ &=& {\rm max}(B) +\langle \psi|\left(\varepsilon A + {i\over 2}\varepsilon [C,B]\right)|\psi\rangle +O(\varepsilon^2)\nonumber\\ &=&{\rm max}(B)+\langle\psi|\left(\varepsilon A \right)|\psi\rangle+O(\varepsilon^2)\nonumber\\ &\le&{\rm max}(B) +{\rm max}(\varepsilon A)+O(\varepsilon^2) \end{eqnarray} (where $|\psi\rangle$ is in the eigenspace of $B$ with maximal eigenvalue). This proves Eq.~(\ref{maxlemma2}). Eq.~(\ref{minlemma2}) is proved similarly. Note that the condition ${\rm maxarg(U)}\ne\pi$ is necessary so that the singularity of the maxarg function can be avoided for $\varepsilon$ sufficiently small. Lemma 2 is all we will need for the proof of Theorem 1. But it is useful to note that Lemma 2 may be invoked to prove:\smallskip {\bf Lemma 3}.\footnote{Strangely, we could find only one reference to this proposition in the literature; it is a special case of Eq.~(8) in \cite{woodward}.} {\sl For unitary $U_1$ and $U_2$, such that \begin{eqnarray} \label{maxargcond} {\rm maxarg}(U_1) + {\rm maxarg}(U_2) &<& \pi~,\\ \label{minargcond} {\rm minarg}(U_1) + {\rm minarg}(U_2) &>& -\pi ~, \end{eqnarray} we have \begin{eqnarray} \label{maxarg} {\rm maxarg}(U_1 U_2) & \le & {\rm maxarg}(U_1) + {\rm maxarg}(U_2) ~,\\ \label{minarg} {\rm minarg}(U_1 U_2) & \ge & {\rm minarg}(U_1) + {\rm minarg}(U_2)~. \end{eqnarray} } \noindent{\sl Proof}: We write \begin{equation} U_1U_2 = U_1 e^{iA}= U_1 \left(e^{iA/n}\right)^n~, \end{equation} where the eigenvalues of A lie in the interval $(-\pi,\pi)$, and apply Lemma 2 repeatedly, obtaining \begin{eqnarray} {\rm maxarg}\left(U_1e^{iA}\right)&\le& {\rm maxarg}(U_1)\nonumber\\ &+& n\left[{\rm maxarg}(e^{iA/n})+ O(n^{-2})\right]~. \end{eqnarray} Taking the $n\to\infty$ limit proves Eq.~(\ref{maxarg}). Eq.~(\ref{minarg}) is proved similarly. Note that because of the conditions Eq.~(\ref{maxargcond}) and Eq.~(\ref{minargcond}), Lemma 2 can be safely applied $n$ times in succession; the accumulated maxarg and minarg of the product never approach $\pi$. To complete the proof of Theorem 1, we invoke the Lie product formula \begin{equation} \lim_{n \to \infty} (e^{A/n} e^{B/n})^n = e^{A+B}~, \end{equation} to write \begin{eqnarray} \label{expexpand} &&e^{iK}e^{-i(H+K)} = \lim_{n \to \infty} (e^{iK/n})^n (e^{-iH/n} e^{-iK/n})^n \nonumber \\ & = & \lim_{n \to \infty} e^{iK/n} \cdots e^{iK/n} e^{-iH/n} e^{-iK/n} \cdots e^{-iH/n} e^{-iK/n}~. \end{eqnarray} Since $e^{iK/n} e^{-iH/n} e^{-iK/n}$ and $e^{-iH/n}$ have the same eigenvalues, Lemma 3 implies that \begin{eqnarray} &{\rm maxarg}&(e^{iK/n} e^{-iH/n} e^{-iK/n} e^{-iH/n}) \nonumber\\ & \le & 2 \cdot {\rm maxarg}(e^{-iH/n}) ~. \end{eqnarray} Similarly, we have \begin{eqnarray} &{\rm maxarg}&\left(e^{iK/n}\left(e^{iK/n} e^{-iH/n} e^{-iK/n} e^{-iH/n}\right) e^{-iK/n}e^{-iH/n}\right) \nonumber\\ & \le & 3 \cdot {\rm maxarg}(e^{-iH/n}) ~, \end{eqnarray} and so on. Hence, applying Lemma 3 altogether $n$ times to the right-hand side of Eq.~(\ref{expexpand}), we find that \begin{eqnarray} {\rm maxarg}\left(e^{iK} \left(e^{-iH/n} e^{-iK/n}\right)^n\right) & \le & n \cdot {\rm maxarg}\left(e^{-iH/n}\right) \nonumber\\ &= &{\rm maxarg}(e^{-iH}) \\ \end{eqnarray} Taking the $n\to\infty$ limit completes the proof of Eq.~(\ref{thm1}). Eq.~(\ref{thm2}) is proved similarly. The upper bound on $\parallel H\parallel_{\rm sup}$ is a key feature of the formulation of Theorem 1. This bound ensures that the conditions Eq.~(\ref{maxargcond}) and Eq.~(\ref{minargcond}) are satisfied each time that Lemma 3 is invoked in the proof. If $\parallel H\parallel_{\rm sup}$ is too large, then counterexamples can be constructed. In any event, for the discussion in \S V, we are interested in the case where the maximal and minimal eigenvalues of $H$ differ by less than $\pi$, and by shifting $H$ by a constant we can ensure that $\parallel H\parallel_{\rm sup}< \pi/2$. Therefore, the theorem enforces the conclusion that if we are to distinguish a nontrivial Hamiltonian from the trivial Hamiltonian in an experiment conducted in a fixed elapsed time, turning on a nonzero time-independent ``driving term'' $K$ provides no advantage.
\section{INTRODUCTION} This is the fifth part of our eight presentations in which we consider applications of methods from wavelet analysis to nonlinear accelerator physics problems. This is a continuation of our results from [1]-[8], in which we considered the applications of a number of analytical methods from nonlinear (local) Fourier analysis, or wavelet analysis, to nonlinear accelerator physics problems both general and with additional structures (Hamiltonian, symplectic or quasicomplex), chaotic, quasiclassical, quantum. Wavelet analysis is a relatively novel set of mathematical methods, which gives us a possibility to work with well-localized bases in functional spaces and with the general type of operators (differential, integral, pseudodifferential) in such bases. In contrast with parts 1--4 in parts 5--8 we try to take into account before using power analytical approaches underlying algebraical, geometrical, topological structures related to kinematical, dynamical and hidden symmetry of physical problems. In this paper we consider the applications of discrete wavelet analysis technique to maps which come from discretization of continuous nonlinear polynomial problems in accelerator physics. Our main point is generalization of wavelet analysis which can be applied for both discrete and continuous cases. We give explicit multiresolution representation for solutions of discrete problems, which is correct discretization of our representation of solutions of the corresponding continuous cases. In part 2 we consider symplectic and Lagrangian structures for the case of discretization of flows by corresponding maps and in part 3 construction of corresponding solutions by applications of generalized wavelet approach which is based on generalization of multiresolution analysis for the case of maps. \section{Veselov-Marsden Discretization} Discrete variational principles lead to evolution dynamics analogous to the Euler-Lagrange equations [9]. Let $Q$ be a configuration space, then a discrete Lagrangian is a map $L: Q\times Q\to{\bf R}$. usually $L$ is obtained by approximating the given Lagrangian. For $N\in N_+$ the action sum is the map $S: Q^{N+1}\to{\bf R}$ defined by \begin{equation} S=\sum_{k=0}^{N-1}L(q_{k+1}, q_k), \end{equation} where $q_k\in Q$, $k\ge 0$. The action sum is the discrete analog of the action integral in continuous case. Extremizing $S$ over $q_1,...,q_{N-1}$ with fixing $q_0, q_N$ we have the discrete Euler-Lagrange equations (DEL): \begin{equation} D_2L(q_{k+1}, q_k)+D_1(q_k, q_{q-1})=0, \end{equation} for $k=1,...,N-1$. Let \begin{equation} \Phi: Q\times Q\to Q\times Q \end{equation} and \begin{equation} \Phi(q_k, q_{k-1})=(q_{k+1}, q_k) \end{equation} is a discrete function (map), then we have for DEL: \begin{equation} D_2L\circ\Phi+D_1L=0 \end{equation} or in coordinates $q^i$ on $Q$ we have DEL \begin{equation} \frac{\partial L}{\partial q^i_k}\circ\Phi(q_{k+1},q_k)+ \frac{\partial L}{\partial q_{k+1}^i}(q_{k+1},q_k)=0. \end{equation} It is very important that the map $\Phi$ exactly preserves the symplectic form $\omega$: \begin{equation} \omega=\frac{\partial^2 L}{\partial q_k^i\partial q_{k+1}^j}(q_{k+1},q_k){\rm d}q_k^i\wedge{\rm d}q^j_{k+1} \end{equation} \section{Generalized Wavelet Approach} Our approach to solutions of equations (6) is based on applications of general and very efficient methods developed by A.~Harten [10], who produced a "General Framework" for multiresolution representation of discrete data. It is based on consideration of basic operators, decimation and prediction, which connect adjacent resolution levels. These operators are constructed from two basic blocks: the discretization and reconstruction operators. The former obtains discrete information from a given continuous functions (flows), and the latter produces an approximation to those functions, from discrete values, in the same function space to which the original function belongs. A "new scale" is defined as the information on a given resolution level which cannot be predicted from discrete information at lower levels. If the discretization and reconstruction are local operators, the concept of "new scale" is also local. The scale coefficients are directly related to the prediction errors, and thus to the reconstruction procedure. If scale coefficients are small at a certain location on a given scale, it means that the reconstruction procedure on that scale gives a proper approximation of the original function at that particular location. This approach may be considered as some generalization of standard wavelet analysis approach. It allows to consider multiresolution decomposition when usual approach is impossible ($\delta$-functions case). We demonstrated the discretization of Dirac function by wavelet packets on Fig.~1 and Fig.~2. Let $F$ be a linear space of mappings \begin{equation}\label{eq:Fin} F\subset \{f|f: X\to Y\}, \end{equation} where $X,Y$ are linear spaces. Let also $D_k$ be a linear operator \begin{eqnarray} &&D_k: f\to\{v^k\},\quad v^k=D_kf,\nonumber \\ &&v^k=\{v_i^k\}, \quad v_i^k\in Y. \end{eqnarray} This sequence corresponds to $k$ level discretization of $X$. Let \begin{equation}\label{eq:Dk} D_k(F)=V^k={\rm span}\{\eta^k_i\} \end{equation} and the coordinates of $v^k\in V^k$ in this basis are $\hat{v}^k=\{\hat{v}^k_i\}$, $\hat{v}^k\in S^k$: \begin{equation} v^k=\sum_i\hat{v}^k_i\eta^k_i, \end{equation}\label{eq:vk} $D_k$ is a discretization operator. Main goal is to design a multiresolution scheme (MR) [10] that applies to all sequences $s\in S^L$, but corresponds for those sequences $\hat{v}^L\in S^L$, which are obtained by the discretization (\ref{eq:Fin}). Since $D_k$ maps $F$ onto $V^k$ then for any $v^k\subset V^k$ there is at least one $f$ in $F$ such that $D_kf=v^k$. Such correspondence from $f\in F$ to $v^k\in V^k$ is reconstruction and the corresponding operator is the reconstruction operator $R_k$: \begin{equation} R_k: V_k\to F, \qquad D_kR_k=I_k, \end{equation} where $I_k$ is the identity operator in $V^k$ ($R^k$ is right inverse of $D^k$ in $V^k$). Given a sequence of discretization $\{D_k\}$ and sequence of the corresponding reconstruction operators $\{R_k\}$, we define the operators $D_k^{k-1}$ and $P^k_{k-1}$ \begin{eqnarray} D_k^{k-1}&=&D_{k-1}R_k: V_k\to V_{k-1}\\ P^k_{k-1}&=&D_kR_{k-1}: V_{k-1}\to V_k\nonumber \end{eqnarray} If the set ${D_k}$ in nested [10], then \begin{equation} D_k^{k-1}P^k_{k-1}=I_{k-1} \end{equation} and we have for any $f\in F$ and any $p\in F$ for which the reconstruction $R_{k-1}$ is exact: \begin{eqnarray}\label{eq:DP} D_k^{k-1}(D_kf)&=&D_{k-1}f\\ P^k_{k-1}(D_{k-1}p)&=&D_kp\nonumber \end{eqnarray} Let us consider any $v^L\in V^L$, Then there is $f\in F$ such that \begin{equation} v^L=D_Lf, \end{equation} and it follows from (\ref{eq:DP}) that the process of successive decimation [10] \begin{equation} v^{k-1}=D_k^{k-1}v^k, \qquad k=L,...,1 \end{equation} yields for all $k$ \begin{equation} v^k=D_kf \end{equation} Thus the problem of prediction, which is associated with the corresponding MR scheme, can be stated as a problem of approximation: knowing $D_{k-1}f$, $f\in F$, find a "good approximation" for $D_k f$. It is very important that each space $V^L$ has a multiresolution basis \begin{equation} \bar{B}_M=\{\bar{\phi}_i^{0,L}\}_i, \{\{\bar{\psi}^{k,L}_j\}_j\}^L_{k=1} \end{equation} and that any $v^L\in V^L$ can be written as \begin{equation}\label{eq:vL} v^L=\sum_i\hat{v}_i^0\bar{\phi}_i^{0,L}+\sum^L_{k=1}\sum_j d_j^k\bar{\psi}_j^{k,L}, \end{equation} where $\{d_j^k\}$ are the $k$ scale coefficients of the associated MR, $\{\hat{v}_i^0\}$ is defined by (11) with $k=0$. If $\{D_k\}$ is a nested sequence of discretization [10] and $\{R_k\}$ is any corresponding sequence of linear reconstruction operators, then we have from (\ref{eq:vL}) for $v^L=D_Lf$ applying $R_L$: \begin{equation}\label{eq:RlDl} R_LD_Lf=\sum_i \hat{f}^0_i\phi^{0,L}_i+\sum_{k=1}^L\sum_jd_j^k\psi_j^{k,L}, \end{equation} where \begin{eqnarray} &&\phi_i^{0,L}=R_L\bar{\phi}_i^{0,L}\in F,\quad \psi_j^{k,L}=R_L\bar{\psi}_j^{k,L}\in F,\nonumber\\ &&D_0 f=\sum\hat{f}^0_i\eta^0_i. \end{eqnarray} When $L\to\infty$ we have sufficient conditions which ensure that the limiting process $L\to\infty$ in (\ref{eq:RlDl}, 22) yields a multiresolution basis for $F$. Then, according to (19), (20) we have very useful representation for solutions of equations (6) or for different maps construction in the form which are a counterparts for discrete (difference) cases of constructions from parts 1-4. \begin{figure}[ht] \centering \epsfig{file=tha138a.eps, width=82.5mm, bb=0 200 599 590, clip} \caption{Wavelet packets.} \end{figure} \begin{figure}[ht] \centering \epsfig{file=tha138b.eps, width=82.5mm, bb=0 200 599 590, clip} \caption{The discretization of Dirac function.} \end{figure} We are very grateful to M.~Cornacchia (SLAC), W.~Her\-r\-man\-nsfeldt (SLAC) Mrs. J.~Kono (LBL) and M.~Laraneta (UCLA) for their permanent encouragement.
\section{Introduction} Max Planck discovered the eponymous constant $\hbar$ when studying black body radiation in 1900 \cite{weav}. He realized immediately that the constants $\hbar$ and $c$ and $G$ determine a natural scale, now called the Planck scale, which is easily gotten by dimensional analysis \cite{mtw}. The Planck distance, time, mass, and energy are \begin{eqnarray} L_p \equiv \sqrt{\frac{G \hbar}{c^3}} \simeq 1.6 \times 10^{-35}\text{ m} \, , \ \ \ T_p \equiv \frac{L_p}{c} = \sqrt{\frac{G \hbar}{c^5}} \simeq 0.54 \times 10^{-43}\text{ sec} \nonumber \\ M_p \equiv \frac{\hbar}{c L_p} = \sqrt{\frac{\hbar c}{G}} \simeq 2.2 \times 10^{-8} \text{ kg}\, , \ \ \ E_p \equiv \sqrt{\frac{\hbar c^5}{G}} \simeq 2.0 \times 10^9 \text{ J} = 1.2 \times 10^{19} \text{ GeV} \, .\label{planck} \end{eqnarray} From its construction the Planck scale should be relevant when the system considered is quantum mechanical ($\hbar$), involves high velocities and high energies ($c$), and gravity is important ($G$). One such system is the very early universe. Another is the collision of elementary particles such as quarks at about the Planck energy, but to achieve the Planck energy in a laboratory would probably require an accelerator about the size of a galaxy. Yet another system in which the Planck scale is relevant is the cloud of virtual particles surrounding any real particle, since the virtual particles may in principle have arbitrarily high energies. Much work has gone into constructing a quantum theory of gravity appropriate to the Planck scale, but with little practical success. The only theory thus far that seems to be a plausible candidate is superstring theory \cite{str}. Motivation for the present work originated at a talk by John Schwarz at the Stanford Linear Accelerator Center in 1996, in which he presented a modified uncertainty principle as a result of superstring theory and scale inversion symmetry. He asked if, in view of its simplicity, it might be more general than superstring theory or any particular quantum gravity theory, and perhaps derivable by simpler means. We hope this work partially answers that question. We do not consider any specific quantum gravity theories here, but instead show that the Heisenberg uncertainty principle is modified when we combine quantum theory and some basic concepts of gravity. We give four separate derivations of the modified uncertainty principle: dimensional analysis in Newtonian theory, an approximate calculation in Newtonian theory, dimensional analysis in general relativity theory, and an approximate calculation in general relativity theory \cite{oha}. All four derivations are heuristic and somewhat rough, as befits a discussion of the uncertainty principle. Moreover the derivations based on Newtonian theory should not be taken too seriously since the Newtonian theory is action-at-a-distance, which is certainly inappropriate for a photon moving at $c$; our purpose in including the Newtonian derivation is to show that the modified uncertainty principle appears to follow from rather general considerations on gravity, in particular that the additional gravitational term is linear in the energy or momentum of the photon. A minimum position uncertainty arises immediately from the modified uncertainty principle, of order of the Planck distance. One might consider this to be a minimum physically meaningful distance, and thereby question whether any theory based on a a smaller scale, eg. a spacetime continuum, really makes operational sense. Additionally, since all measurements dealing with small distances in particle physics are really large momentum scattering experiments, the very concept of a spacetime continuum is doubly suspect. Further speculations of this nature are contained in the conclusions. \section{The Uncertainty Principle a la Heisenberg} Heisenberg in 1923 obtained the uncertainty principle on very general grounds, based only on general principles of optics and the quantization of electromagnetic radiation in the form of photons \cite{sch,weav}. We recall his approach briefly. Consider a wave scattering from an electron into a microscope and thereby giving a measurement of the position of the electron. According to optics and intuition, with an electromagnetic wave of wavelength $\lambda$ we cannot obtain better precision than \begin{equation} \Delta x_H \approx \lambda \, . \label{lamb} \end{equation} Such a wave is quantized in the form of photons, each with a momentum \begin{equation} p=\frac{h}{\lambda} \, . \label{p} \end{equation} In order to interact with the electron an entire photon in the wave must scatter and thereby impart to the electron a significant part of its momentum, which produces an uncertainty in the electron momentum of about $\Delta p \approx p$. Thus we obtain the standard Heisenberg position-momentum uncertainty relation \begin{equation} \Delta x_H \Delta p \approx \lambda\left( \frac{h}{\lambda} \right) \approx h \approx \hbar \, , \label{up} \end{equation} No mention has been made here of the gravitational interaction between the photon and the electron, which we consider below. \section{Newtonian Theory, Dimensional Estimate} We first estimate the effects of gravity in a very rough and heuristic way using Newtonian gravitational theory, with the assumption that the photon behaves as a classical particle with an effective mass equal to its energy divided by $c^2$ \cite{oha}. Suppose the electron is in an experimental region of characteristic size $L$, inside of which it interacts with the photon. It will experience an acceleration due to gravity, \begin{equation} \ddot{\vec{r}} = - \frac{G (E /c^2)}{r^2} \hat{r} \, , \label{acc} \end{equation} where $r$ is the distance between electron and photon. During the interaction, which occurs in characteristic time $L/c$, the electron will acquire, due to gravity, a velocity and move a distance, given respectively by \begin{equation} \Delta v \approx \frac{GE}{c^2 r^2} \left( \frac{L}{c} \right) \, , \ \ \ \ \Delta x_G \approx \frac{GE}{c^2 r^2} \left( \frac{L}{c} \right)^2 \, . \end{equation} These will be uncertain since the photon scatters electromagnetically from the electron at some indeterminate time during the interaction. The electron may be anywhere in the interaction region so the electron-photon distance should be of order $r \approx L$, which is the only distance scale in the problem. Since the photon energy is related to the momentum by $E=pc$ we may also express this as \begin{equation} \Delta x_G \approx \frac{Gp}{c^3} \, . \label{xG} \end{equation} Noting that the electron momentum uncertainty must be of order of the photon momentum, and using the Planck length $L_{p}^{2} \equiv G \hbar/ c^3$ as a parameter, we have \begin{equation} \Delta x_G \approx \frac{G \Delta p}{c^3} = \left( \frac{G \hbar}{c^3}\right) \frac{\Delta p}{\hbar}= L_{p}^{2} \frac{\Delta p}{\hbar} \, .\label{xG2} \end{equation} This is our main result. We add this uncertainty to the Heisenberg relation (3) to obtain the modified uncertainty relation \begin{equation} \Delta x \approx \frac{\hbar}{\Delta p}+ L_{p}^{2} \frac{\Delta p}{\hbar} \, . \label{gup} \end{equation} We refer to this as the extended uncertainty principle - or more descriptively as the gravitational uncertainty principle (GUP). Note that it is invariant under \begin{equation} \frac{\Delta p L_p}{\hbar} \longleftrightarrow \frac{\hbar}{\Delta p L_p} \, . \label{inv} \end{equation} That is, it has a kind of momentum inversion symmetry \cite{str}. \section{Newtonian Theory, Approximate Calculation} We may also make a more explicit estimate than the above using Newtonian theory. As before we suppose the electron is in an experimental region of characteristic size $L$ and interacts with the photon as it crosses the region. The photon scatters electromagnetically from the electron at some uncertain time and at some uncertain position inside the experimental region. Consider first the transverse impulse, that is the motion imparted to the electron perpendicular to the photon direction. We take the photon direction to be $x$ and the transverse direction to be $y$ as in figure 2. The photon passes very rapidly so the electron moves very little in the time it takes the photon to cross the experimental region, and we may thus take $y \approx y_0$ that is we use an impulse approximation. The acceleration is then \begin{equation} \ddot{y}=\frac{G(E/c^2)}{r^2} \left( \frac{y}{r} \right)= \frac{G (p/c)}{r^2} \left( \frac{y}{r} \right) \approx \frac{G(p/c)y_0}{(y_{0}^{2} + c^2 t^2)^{3/2}} \, . \label{ay} \end{equation} We integrate this to get the transverse velocity impulse, \begin{equation} \Delta \dot{y} \approx \frac{2Gp}{c^2 y_0}\left( \frac{cT}{\sqrt{y_{0}^{2} + c^2 T^2}}\right) \, ,\label{vy} \end{equation} where $T=L/c \ge y_0$ is the characteristic interaction time. We thus have roughly \begin{equation} \Delta \dot{y} \approx \frac{2Gp}{c^2 y_0} \, .\label{vy2} \end{equation} Due to this gravitational velocity impulse there will be a change in the position of the electron, which is intrinsically uncertain, given approximately by $\Delta y \approx (\Delta \dot{y} / 2) /T$ or \begin{equation} \Delta y_G \approx \frac{GpT}{c^2 y_0} \approx \frac{Gp}{c^3} \, ,\label{yG} \end{equation} where we have taken $y_0$ to be of order of but less than $L=cT$. This is the same result (\ref{xG}) as we obtained by somewhat more crude dimensional arguments in section III. A similar analysis can be done for motion in the longitudinal direction, with the result \begin{equation} \Delta x_G \approx \frac{Gp}{c^3} \left[ \ln \left( \frac{2L}{x_0} \right) - 1 \right] \, , \end{equation} where $x_0$ is the initial position of the electron. Since $x_0$ must be of order but less than $L=cT$ and the log is a very slowly varying function we obtain about the same result as (\ref{yG}) for the longitudinal uncertainty, \begin{equation} \Delta x_G \approx \frac{Gp}{c^3} \, . \label{xG3} \end{equation} One should be justifiably suspicious of the Newtonian derivations since Newtonian theory treats the gravitational field in front of the radiation as action-at-a-distance, whereas the gravitational field actually propagates at $c$ and cannot extend in front of the radiation. The nature of the gravitational field will become clear when we discuss the calculation using linearized general relativity. \section{General Relativity Theory, Dimensional Estimate} The arguments in the preceding two sections are only marginally convincing as heuristic arguments since they are based on action-at-a-distance Newtonian gravitational theory, with the ad hoc assumption that the energy of the photon produces a gravitational field. In this section and the following we give a dimensional estimate and an approximate calculation based on general relativity theory, free of such drawbacks \cite{abs,mtw,oha}. The field equations of general relativity are \begin{equation} G_{\mu \nu} = - \left( \frac{8 \pi G}{c^4} \right)T_{\mu \nu} \, . \label{fldeq} \end{equation} The left side has the units of inverse distance squared, since it is constructed from second derivatives and squares of first derivatives of the metric. Thus on dimensional grounds we may write the left hand side in terms of deviations of the metric from flat, in schematic order of magnitude dimensional form, as \begin{equation} LHS \approx \frac{\delta g _{\mu \nu}}{L^2} \, , \label{lhs} \end{equation} where $\delta g _{\mu \nu}$ denotes the deviation of the metric from Lorentzian, and $L$ is the same characteristic size as used in sec. III. Similarly the energy-momentum tensor has the units of an energy density, so its components must be roughly equal to the photon energy over $L^3$. Thus we can write the right side of the field equations schematically as \begin{equation} RHS \approx \left( \frac{8 \pi G}{c^4} \right) \frac{E}{L^3} \approx \frac{Gp}{ c^3 L^3} \, . \label{rhs} \end{equation} Equating the dimensional estimates in (\ref{lhs}) and (\ref{rhs}) we get an estimate for the deviation of the metric, \begin{equation} \delta g_{\mu \nu} \approx \frac{Gp}{c^3 L} \, . \label{dg} \end{equation} This deviation corresponds to a fractional uncertainty in all positions in the region $L$, which we identify with a fractional uncertainty in position, $\Delta x_G / L$. Thus we have an uncertainty in position due to the gravitational interaction given by \begin{equation} \frac{\Delta x_G }{L} \approx \delta g_{\mu \nu} \approx \frac{Gp}{c^3 L}\, , \ \ \ \Delta x_G \approx \frac{Gp}{c^3} \, . \label{xG4} \end{equation} As should be expected the characteristic size $L$ has canceled out of the relation. Finally the uncertainty in momentum of the electron must be comparable to the photon momentum, $\Delta p \approx p$, and we obtain the same relation (\ref{xG2}) as before for $\Delta x_G$. \section{General Relativity Theory, Approximate Calculation} To make an approximate calculation we use linearized general relativity theory. From the general energy momentum tensor of the electromagnetic field it is easy to show that for radiation moving in the $x$ direction the specific form of the energy- momentum tensor is \cite{tol,abs} \begin{equation} T_{mu \nu} = F_{\mu \alpha} F^{\alpha}\; _{\nu} + \frac{1}{4}F_{\alpha \beta} F^{\alpha \beta} = \rho \left( \begin{array}{rrrr} 1 \ & -1 \ \ & 0 \ \ & 0 \ \ \\ -1 \ & 1 \ \ & 0 \ \ & 0 \ \ \\ 0 \ & 0 \ \ & 0 \ \ & 0 \ \ \\ 0 \ & 0 \ \ & 0 \ \ & 0 \ \ \end{array} \right) \, , \label{emt} \end{equation} where $\rho=(E^2 + B^2 )/2$ is the energy density of the radiation field, and may be a function of $x -ct \, , \ y \, , \text{ and } z$, corresponding to a truncated plane wave. The equations of linearized general relativity theory follow from (\ref{fldeq}), with the metric taken to be Lorentz plus a small perturbation, $g _{\mu \nu} = \eta_{\mu \nu} + h_{\mu \nu}$. They are \begin{equation} \Box \left [ h_{\mu \nu} - \frac{1}{2} \eta_{\mu \nu} h \right] = - \left( \frac{8 \pi G}{c^4} \right) T_{\mu \nu} \, , \ \ h \equiv \eta^{\mu \nu} h_{ \mu \nu} = h^{\alpha}_{\alpha}\, ,\label{lfldeq} \end{equation} $$ \Box = \frac{\partial^2}{\partial (ct)^2} - \frac{\partial^2}{\partial x^2} - \frac{\partial^2}{\partial y^2} - \frac{\partial^2}{\partial z^2} \, , $$ $$ \text{Lorentz gauge condition } \left[ h_{\mu}^{ \nu} - \frac{1}{2} \eta_{\mu}^{ \nu} h \right]_{| \nu} =0 \, . $$ It is straight-forward to solve this system with the energy momentum tensor given in (\ref{emt}). We are interested only in the inhomogeneous solution, and the system then reduces to a form involving only one unknown function, \begin{equation} h_{\mu \nu} = f(x-ct, y, z) \left( \begin{array}{rrrr} 1 \ & -1 \ \ & 0 \ \ & 0 \ \ \\ -1 \ & 1 \ \ & 0 \ \ & 0 \ \ \\ 0 \ & 0 \ \ & 0 \ \ & 0 \ \ \\ 0 \ & 0 \ \ & 0 \ \ & 0 \ \ \end{array} \right) \, , \ \ \ \Box f = - \left( \frac{8 \pi G}{c^4} \right) \rho \, . \label{h} \end{equation} For convenience we choose the energy density to be a product, \begin{equation} \rho (x-ct,y,z) = \rho_{\parallel}(x-ct) \rho_{\perp}(y,z) \, , \label{rho} \end{equation} and it then follows that the metric function $f$ is also a product of the form $f(x-ct,y,z) = f_{\parallel}(x-ct) f_{\perp}(y,z)$, with \begin{equation} \left( \frac{\partial^2}{\partial y^2} + \frac{\partial^2}{\partial z^2} \right) f_{\perp}(y,z) = \left( \frac{8 \pi G}{c^4} \right) \rho_{\perp}(y,z) \ , , \ \ f_{\parallel}(x-ct) = \rho_{\parallel}(x-ct) \, . \label{wveq} \end{equation} Until now the energy density function has been arbitrary. We now choose a specific function which is convenient for our purposes. For this we take a cylinder as the envelope of the radiation, inside of which the energy density oscillates at twice the frequency of the radiation field; the oscillations however may be ignored since we are interested in an average uncertainty in position. Thus we take the energy density to be a constant $\rho_{0}$ inside a cylinder of length $L$ and radius $R$, with $R$ and $L$ comparable in magnitude. Then the solution to (\ref{wveq}), in cylindrical coordinates $x-ct\, , \ r \, , \ \varphi$, is \begin{equation} f = \frac{4G(E/c^2)}{L} g(r) \theta_{L}(x-ct) = \frac{4Gp}{c^3 L} g(r) \theta_{L}(x-ct) \, , \label{sol} \end{equation} $$ g(r) = \left\{ \begin{array}{lr} r^2/R^2 \, , & r<R \\ 1+ \ln(r^2/R^2) \, , & r>R \end{array} \right\} \, , \ \ \theta_{L}(x-ct) \equiv \theta(x-ct) \theta(ct-x -L) $$ Here $E= \pi R^2 L \rho_{0}$ is the total energy of the radiation, and $\theta_{L}$ is the ``double'' theta function defined above - which is equal to $0$ ahead of and behind the radiation cylinder. Notice that the gravitational field does not extend ahead of the radiation field, unlike in the Newtonian theory. Notice also that the apparent logarithmic divergence of the external field is not a problem since physical effects involve the derivative of $f$ which falls off like $1/r$. According to (\ref{h}) and (\ref{sol}) the metric takes the rather elegant form \begin{equation} ds^2=c^2 dt^2 - (dx^2 + dy^2 + dz^2) + f(cdt - dx)^2 \label{g} \end{equation} It is clear that since $g(r)$ is of order 1 inside and near the cylinder of radiation the deviation of the metric from Lorentz is there of order $4Gp / c^3 L$ and we thereby obtain, using the same idea as in section V, \begin{equation} \frac{\Delta x_G }{L} \approx \frac{Gp}{c^3 L} \approx \frac{G \Delta p}{c^3 L} \, \ \ x_G \approx \frac{4G \Delta p}{c^3} \, . \label{xG5} \end{equation} We may also estimate the gravitationally induced motion of the electron using the above metric and the geodesic equation of motion, to obtain an alternative derivation of the position uncertainty. The geodesic equation of motion is \begin{equation} \frac{d^2 x^i}{ds^2} + \left\{^{\ i}_{\alpha \beta} \right\} \frac{dx^{\alpha}}{ds} \frac{dx^{\beta}}{ds}=0 \, , \ \ \alpha,\beta=0,1,2,3 \, , \ \ i=1,2,3 \, . \label{geo} \end{equation} For an electron moving reasonably slowly the line element is about $ds=cdt$ and the $\alpha=\beta=0$ term dominates, so the equations of motion are the usual Newtonian limit equations \begin{equation} \frac{d^2 x^i}{dt^2}= - \left\{^{\ i}_{0 0} \right\} c^2 \, . \label{geo2} \end{equation} From the metric (\ref{g}) the Christoffel symbols are easily found and we obtain for motion in the longitudinal $x$ and transverse $r$ directions \begin{equation} \frac{d^2 x}{dt^2} = \frac{1}{2} \frac{\partial f}{\partial x} c^2 \, , \ \ \frac{d^2 r}{dt^2} = - \frac{1}{2} \frac{\partial f}{\partial r} c^2 \, . \label{geo3} \end{equation} where we assume motion in the $x$ and $r$ direction. Longitudinal motion is easy to analyze and rather informative. The derivative of $f$ gives two delta functions. As the front of the radiation cylinder passes the electron it first gives the electron a velocity impulse of \begin{equation} \Delta \dot{x} = \frac{2Gp}{c^2 L} g(r) \, , \label{impx} \end{equation} and then as the back of the radiation cylinder passes the electron receives an equal and opposite velocity impulse and stops. In the time of passage it has moved \begin{equation} \Delta x_G = \frac{2Gp}{c^2 L} g(r) T = \frac{2Gp}{c^3 } g(r) \approx \frac{2Gp}{c^3 } \, , \label{xG6} \end{equation} since $g(r)$ is of order 1 in and near the radiation cylinder. Thus we obtain the same result (\ref{xG2}) as previously. Notice that the gravitational field of the radiation only acts as it passes over the electron, not before and not after. For the transverse motion we differentiate $f$ with respect to $r$ to find \begin{equation} \frac{d^2 r}{dt^2} = -\frac{4Gp}{c^2 L} \left\{ \begin{array}{lr} r/R^2 \, , & r<R \\ 1/r \, , & r>R \end{array} \right\} \theta_{L}(x-ct) \end{equation} so that in the region of the cylinder we have very roughly a velocity impulse and a corresponding position change $\Delta r \approx (\Delta \dot{r} /2) T$, given by \begin{equation} \Delta \dot{r} = \frac{4Gp}{c^2 LR}T \, , \ \ \Delta r= \frac{2Gp}{c^2 LR}T^2 \approx \frac{2Gp}{c^3} \end{equation} That is once again we find that the transverse motion corresponding to an uncertainty in position due to gravity is the same as in (\ref{xG}) or (\ref{xG2}). \section{The Minimal Distance Uncertainty} The GUP has a remarkable consequence. If the photon momentum and $\Delta p$ are chosen to be very small then the electron position is imprecise because the long photon wavelength gives poor resolution. If the photon momentum and $\Delta p$ are chosen to be very large, then the gravitational field of the photon makes the electron position very imprecise. Between the two extremes there is a minimum position uncertainty, which we find from (\ref{gup}) to be \begin{equation} \Delta x_{\text{min}} \approx 2\sqrt{\frac{G \hbar}{c^3}} = 2 L_p \, , \ \text{ for } \ \Delta p \approx \sqrt{\frac{\hbar c^3}{G}}\frac{E_p}{c} \,. \end{equation} This means that we can never localize the position of a particle such as an electron to better than about the Planck distance. Similar analyses of spacetime using the path integral formalism lead to analogous conclusions; spacetime at small distances and times undergoes quantum fluctuations, and at the Planck scale the fluctuations are of the same order as the distances involved \cite{mtw,amel}. \section{conclusions} We have shown, using Newtonian and general relativistic gravity, that the position-momentum uncertainty principle of quantum mechanics is modified by an additional term. In both theories it is clear that the extra term must be proportional to the energy or momentum of the photon, so on purely dimensional grounds the order of magnitude of the extra term is uniquely determined. As a consequence there is an absolute minimum uncertainty in the position of any particle such as an electron. Not surprisingly the minimum is of order of the Planck distance. In view of the absolute minimum position uncertainty one may plausibly question whether any theory based on shorter distances, such as a spacetime continuum, really makes sense. Indeed in light of the fact that laboratory experiments which probe small distance properties of particles are all high energy scattering experiments, one might conclude that spacetime at such small scales may not be a useful concept, and that spacetime at the Planck scale may not even exist in any meaningful operational sense. Such ideas are not new and were espoused in the era of S-matrix theory in the 1960s, that is that the scattering amplitude expressed n terms of input and output momenta may be the fundamental reality of high energy physics, and not point-like or string-like particles in a spacetime continuum \cite{sm}. One might even speculate that the spacetime continuum concept actually impedes physics in the same way that the concept of an ether impeded physics in the 19th century. As such, a theoretical structure based entirely on momenta, such as a modern version of S-matrix theory, might be desirable and interesting. \section*{Acknowledgements} This work was supported by NASA grant NAS 8-39225 to Gravity Probe B. John Schwarz posed the question studied here in a talk at the Stanford Linear Accelerator Center in fall 1996, but is of course not responsible for our interpretation of it. Finally the Gravity Probe B theory group at Stanford provided many stimulating discussions.
\section{Introduction\label{introduction}} Since about a decade it has been established that dwarf galaxies are the most numerous type of galaxies in the universe (e.g.\ Mateo 1998)\markcite{mateo98}. Zwicky (1956)\markcite{zwicky56} was among the first to propose that at least some dwarf galaxies might have been created from tidal debris strewn about in intergalactic space by gravitational encounters between massive spiral galaxies. The relatively nearby M\,81 triplet, for example, is one of the most stunning examples of how gravitational interactions can redistribute the neutral atomic gas, originally belonging to individual galaxies, over a huge volume in the form of bridges and tidal tails (van der Hulst 1979\markcite{vanderhulst79}, Yun et al.\ 1993\markcite{yun:etal93}, Yun, Ho \& Lo\ 1994)\markcite{yun:etal94}. The newly born galaxies which may form in these types of interactions are variously referred to as `protogalaxies', i.e., those who have not yet started to form stars, `tidal dwarfs' or `intergalactic molecular complexes'. Recently, they have also been coined `Phoenix galaxies' (Hernquist 1992\markcite{hernquist92}) since they are presumably born phoenix--like out of the `ashes' of galaxy--galaxy collisions. The recent discovery of star forming regions in tidal tails of interacting galaxies (Mirabel, Lutz \& Maza\ 1991\markcite{mirabel:etal91}, Mirabel, Dottori \& Lutz 1992\markcite{mirabel:etal92}; Duc et al.\ 1997\markcite{duc:etal97}; Duc \& Mirabel 1998\markcite{duc:mirabel98}) provides support for the concept of Phoenix--galaxies. This is corroborated by numerical simulations which show that tails produced by tidal interactions are subject to fragmentation (Barnes \& Hernquist 1992\markcite{barnes:hernquist96}; Elmegreen, Kaufman \& Thomasson 1993\markcite{elmegreen:etal93}) which may lead to star formation. Despite the fact that recent star formation is taking place in tidal dwarfs, searches for molecular gas in these objects have been rather unsuccessful thus far. This is surprising as, according to our current understanding of the star formation process, neutral gas needs to get dense enough, become self gravitating, and turn largely from the atomic to a denser molecular phase before star formation can proceed. Because molecular hydrogen, the main constituent of a molecular cloud, is difficult to detect at the temperatures and densities found in molecular clouds, emission from carbon monoxide (CO) is commonly used as a tracer. The only example to date of a molecular complex associated with tidal debris is the one detected by Brouillet et al.\ (1992) within the M\,81 triplet, close to, but outside the optical image of M\,81. Although it is presumably located within an \ion{H}{1} arm torn out of M\,81, no optical emission or star formation appears to be associated with it (Henkel et al.\ 1993\markcite{henkel:etal93}). Since this discovery, a lot of observational effort has been undertaken to detect similar molecular complexes in tidal arms of interacting groups of galaxies -- but without success (e.g., Brouillet, Henkel \& Baudry 1992\markcite{brouillet:etal92}, Smith \& Higdon 1994\markcite{smith:higdon94}). In this paper, we present the detection of the second molecular complex of this kind. \section{Observations\label{observations}} We searched for molecular gas in the tidal arms near NGC\,3077 which show their presence in observations of the 21\,cm line of neutral hydrogen (\ion{H}{1}). NGC\,3077 is a member of the nearby (D=3.2\,Mpc) interacting M\,81 galaxy triplet (see Fig.\ 1, left). Our CO observations were guided by archival \ion{H}{1} data of NGC\,3077 (Fig.~1, right, greyscale) obtained with the NRAO Very Large Array (VLA)\footnote{The National Radio Astronomy Observatory (NRAO) is operated by Associated Universities, Inc., under cooperative agreement with the National Science Foundation.} with an improved angular resolution (10$''$) of a factor of 6 as compared to published \ion{H}{1} observations (van der Hulst 1979\markcite{vanderhulst79}; Yun et al.\ 1993\markcite{yun:etal93}). The increased resolution makes it easier to identify regions of high \ion{H}{1} column density and hence provides better clues as to where to search for molecular clouds. The CO observations were carried out in December 1998 with the IRAM 30\, m radio telescope. We observed the $^{12}$CO (${J=1\to0}$) and (${J=2\! \to \! 1}$) transitions at 115\, GHz and 230\, GHz simultaneously using two dual channel receivers with a wobbling subreflector. The wobbler throw was $\pm4'$ in azimuth. The angular resolution of the telescope is 11$''$ at 230\, GHz and 22$''$ at 115\, GHz. In total, 13 independent positions were observed within 6 hours observing time on a $20''$ grid, which is approximately the size of the telescope beam at 115\,GHz. We used the 1 MHz filter spectrometer and the autocorrelator spectrometer simultaneously. Spectra for individual positions were finally degraded to a velocity resolution of 2.6\,\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi\ at both frequencies. The different receiver setups were combined to one single spectrum per position and transition. Only linear baselines were removed from the spectra. The final rms ($T_{\rm mb}$) values for individual positions are listed in Tab.~1. \begin{deluxetable}{cccccc} \tablewidth{0pc} \tablecaption{Gaussian components of the CO ($J=1\!\to\! 0$) spectra.} \tablehead{ \colhead{$\Delta\alpha$\tablenotemark{a}} & \colhead{$\Delta\delta$\tablenotemark{a}} & \colhead{$T_{\rm mb}$} & \colhead{rms} & \colhead{$v_{\rm hel}$} & \colhead{$\Delta v$} \nl \colhead{$('')$} & \colhead{$('')$} & \colhead{(mK)} & \colhead{(mK)} & \colhead{(\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi)}& \colhead{(\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi)} } \startdata 0& 60 & -- & 14.0 & $-$ & $-$ \nl 0& 40 & -- & 15.9 & $-$ & $-$ \nl --20& 20 & -- & 9.6 & $-$ & $-$ \nl 0& 20 & 72.8& 15.1 & $13.9\pm0.9$ & $15.3\pm2.1$ \nl 20& 20 & 47.8& 13.1 & $14.4\pm1.1$ & $12.5\pm3.0$ \nl --40& 0 & 29.7& 14.8 & $ 9.2\pm2.8$ & $21.7\pm5.5$ \nl --20& 0 & 43.9& 10.8 & $11.9\pm1.4$ & $23.9\pm3.3$ \nl 0& 0 & 59.4& 8.1 & $15.0\pm0.7$ & $15.1\pm1.6$ \nl 20& 0 & -- & 13.5 & $-$ & $-$ \nl 40& 0 & -- & 13.4 & $-$ & $-$ \nl --20&--20& -- & 10.9 & $-$ & $-$ \nl 0&--20& 46.0& 14.4 & $14.2\pm1.0$ & $8.5\pm1.9$ \nl 0&--60& -- & 13.9 & $-$ & $-$ \nl \enddata \tablenotetext{a}{Positions are offsets relative to $\alpha_{1950}=9^h59^m57.\!^s3$, $\delta_{1950}=68^\circ57'13.\!''8$.} \label{gauss} \end{deluxetable} \begin{figure} \plotone{walter.fig1.gif.ps} \caption{ {\em Left:} Optical view of the M\,81--M\,82--NGC\,3077 triplet (taken from the Digitized Sky Survey). A linear scale (adopting a distance to the triplet of D=3.2\,Mpc) is indicated in the upper left. The box marks the region around NGC\,3077 which is blown up on the right. {\em Right:}~Close--up view of the region around NGC\,3077. The greyscale represents the distribution of the neutral hydrogen (\ion{H}{1}) as observed with the VLA. The contours represent the optical image (shown in greyscale on the left). The box indicates the region where molecular gas was discovered (see Fig.~2 for a close--up view of that region). The linear scale is indicated in the lower right. Coordinates are given in the B1950.0 epoch. \label{overview}} \end{figure} \section{Results} Our CO spectra are presented in Figs.\ 2 and 3. Fig.\ 2 gives an overview of our limited mapping result towards the newly detected molecular complex. CO ($J=1\to0$) emission was clearly detected from the central five positions towards the region with the highest \ion{H}{1} column densities. There is evidence of more emission in neighboring spectra, but at a lower than 3\,$\sigma$ confidence level. The properties of the CO ($J=1\to0$) lines as derived from a Gaussian analysis are listed in Tab.~1. Brouillet et al.\ (1992)\markcite{brouillet:etal92} obtained two spectra close to the region we mapped. They are ($-2''$, $-73''$) and (240$''$, 16$''$) offset from our reference position $\alpha_{1950}=9^h59^m57.\!^s3$, $\delta_{1950}=68^\circ57'13.\!''8$. However, Brouillet et al.\ did not detect CO emission in their spectra at a 1\,$\sigma$--rms of about 30\, mK in their 5.2\,\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi\ wide channels. Due to the lower signal-to-noise ratio, emission from the CO ($J=2\to1$) was only unambiguously detected from our (0,0) position, where we integrated longest, and for the average spectrum of the five positions where we detected the lower ($J=1\to0$) transition. For two other positions where we detected the ($1\to0$) transition there is indication for ($2\to1$) emission also, however with less than 3$\sigma$. Fig.\ 3 presents a comparison between the spectra of neutral hydrogen (\ion{H}{1}) and the two lowest rotational CO transitions towards our reference position. The velocities of the molecular gas in the newly detected molecular complex are in excellent agreement with the atomic component. As is also found in other galaxies, the \ion{H}{1} line is broader than the CO line, indicating that the molecular material is clumped and present over a smaller range along the line of sight than the \ion{H}{1}. The ratio of the two integrated CO lines is (${W_{2\to1}/W_{1\to0}) =0.75\pm0.12}$ for the average of the central five positions; because the source is clearly extended (s. Fig. 2) no correction for the different beam areas for the two transitions was made. This ratio is similar to that found in our Galaxy for quiescent cool (10--20\,K) molecular clouds (Falgarone et al.\ 1998)\markcite{falgarone:etal98}. \begin{figure} \epsscale{0.5} \plotone{walter.fig2.ps} \label{spectramap} \caption{$^{12}$CO (${J=1\!\to\!0}$) spectra obtained towards the newly detected molecular complex. Positions are offsets relative to $\alpha_{1950}=9^h59^m57.\!^s3$, $\delta_{1950}=68^\circ57'13.\!''8$. The center of each square corresponds to the location where each displayed spectrum has been obtained. Note that the spacing between two individual spectra (20$''$) is approximately the size of the IRAM 30\,m telescope at 115 GHz (22$''$). The small inserted box indicates the velocity and temperature scale for each spectrum.} \end{figure} \begin{figure} \epsscale{0.5} \plotone{walter.fig3.ps} \label{spectramap} \caption{$^{12}$CO (${J=1\!\to\!0}$) spectra obtained towards the newly detected molecular complex. Positions are offsets relative to $\alpha_{1950}=9^h59^m57.\!^s3$, $\delta_{1950}=68^\circ57'13.\!''8$. The center of each square corresponds to the location where each displayed spectrum has been obtained. Note that the spacing between two individual spectra (20$''$) is approximately the size of the IRAM 30\,m telescope at 115 GHz (22$''$). The small inserted box indicates the velocity and temperature scale for each spectrum.} \end{figure} \section{Discussion\label{discussion}} \subsection{The properties of the molecular complex} Because the CO detection coincides with the peak in the \ion{H}{1} distribution of the tidal gas near NGC\,3077, and strengthened by the fact that the velocities of the \ion{H}{1} and CO are in excellent agreement, it is natural to assume that the newly detected molecular gas is located within the M\,81--M\,82--NGC\,3077 triplet. However, it is important to further rule out a possible Galactic origin. The molecular gas is located towards a direction where the Galaxy is rich in infrared cirrus (de Vries, Heithausen \& Thaddeus\ 1987\markcite{devries:etal87}). A significant part of these cirrus clouds is molecular. However, these clouds are distinct from the cloud we detected for two reasons. First, molecular cirrus clouds in this region of the sky have more negative velocities (de Vries et al.\ 1987\markcite{devries:etal87}), the closest cirrus cloud is at $v_{\rm hel}=-4.5$\,\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi\ thus 18.5\,\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi\ off our CO detection. Secondly, the line width of our molecular cloud ($15\pm1$\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi) is significantly broader than those of the cirrus clouds ($\Delta v\le4.4$\,\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi, cf.\ Heithausen 1996\markcite{heithausen96}) and is more similar to that of giant molecular complexes in our Galaxy. We thus conclude that our object is not connected to foreground cirrus clouds but extragalactic in nature. We therefore assume that the molecular complex is indeed associated with the tidal debris of NGC\,3077. We adopt a distance of 3.2\, Mpc (cf.\ Henkel et al.\ 1993\markcite{henkel:etal93}). As shown in Fig.\ 1, the region where we detected CO is clearly outside the optical extent of NGC\,3077. The radius ($R_{25}$) of NGC\,3077 is $2.\!'7\times2.\!'2$ or 2.5\, kpc $\times$ 2.1\, kpc (de Vaucouleurs et al.\ 1991\markcite{devaucouleurs:etal91}). Molecular gas of NGC\,3077 itself is concentrated to its nucleus with a half width at half maximum (HWHM) of about 0.16\,kpc (Becker, Schilke, \& Henkel 1989)\markcite{becker:etal89}, atomic gas of NGC\,3077 extends further out (HWHM$\,\approx\,$0.9\,kpc). The projected distance of the newly detected molecular complex to the center of NGC\,3077 is $3.\!'9$ corresponding to 3.7\, kpc at the distance of NGC\,3077. If we assume our limited mapping to be complete, molecular gas in the complex covers an area of 0.4\, kpc$^2$, which corresponds to an effective radius of about 350\, pc. This is about twice the value found by Brouillet et al.\ (1992)\markcite{brouillet:etal92} for their intergalactic complex near M\,81. We note, however, that due to the low line--intensities observed, we expect beam dilution to play a major role. This means that the molecular gas in the complex is likely clumped. For comparison purpose we calculate a virial mass of the molecular complex of $M_{\rm vir}\approx 1.7\times10^7$\,M$_{\odot}$, assuming a constant density throughout the molecular gas. This value is similar to the object found by Brouillet et al.\ (1992)\markcite{brouillet:etal92} and also similar to the molecular complex near the nucleus of NGC\,3077 (Becker et al.\ 1989\markcite{becker:etal89}). Adopting a Galactic $X$--factor, the ratio of H$_2$ column density to integrated CO intensity, ${X_{\rm G}=(1.56\pm0.05)\times10^{20}}$\, cm$^{-2}$\,(K\,\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi)$^{-1}$ (Hunter et al.\ 1997\markcite{hunter:etal97}), we derive a molecular mass which is a factor of 20 lower than that based on the assumption of virialisation. This rather large discrepancy in the mass estimates could have two reasons: i) the assumption of virialisation does not hold for this complex; ii) the true $X-$factor in the molecular complex is comparable to values found in low metallicity galaxies (Dettmar \& Heithausen 1989\markcite{dettmar:heithausen89}) and in the Brouillet et al.\ complex. Observations of, e.g., rarer CO isotopomers are needed to investigate this issue further. \subsection{The formation of the molecular complex} The metallicity of interstellar gas can provide important clues as to the origin of that gas. From the mere fact that we see CO we can already conclude that the material in the molecular complex near NGC\,3077 is not primordial, but already pre--processed. Yun et al.\ (1993)\markcite{yun:etal93} performed detailed numerical simulations to explain the impressive tidal \ion{H}{1}--structure in the M\,81 triplet. These simulations suggest that the huge \ion{H}{1} complex east of NGC\,3077 (total \ion{H}{1} mass: $M_{\rm HI}=3\times10^8$M$_{\odot}$) where the molecular complex is situated was formed out of material stripped off the outer parts of NGC\,3077 during the closest encounter with the most massive galaxy of the triplet, M\,81, about $3\times 10^8$ years ago. Optical studies show that NGC\,3077's metallicity is around solar (Martin 1997)\markcite{martin97}. This means that even the gas in the outskirts of NGC\,3077 was already chemically enriched before the encounter with M\,81 which explains why heavy elements are present. \subsection{Signs of star formation?} At this point it is interesting to investigate whether stars have already formed in the newly detected molecular complex near NGC\,3077 or not. In the case of the intergalactic complex detected by Brouillet et al.\ (1992)\markcite{brouillet:etal92}, deep optical follow-up studies showed that no stellar counterpart was visible\markcite{henkel:etal93}. In our case, an optical study of this region using the Russian 6\,m telescope has been undertaken by Karachentsev, Karachentseva \& B\"orngen\ (1985a)\markcite{karachentsev:etal85a} who obtained deep images of an irregular dwarf object south--east of NGC\,3077 which they named the `Garland'. As the name indicates, the Garland is a chain of faint optical knots which occupy an area of $6'\times 4'$ on the sky. Some diffuse emission is also present in and near the region we mapped. Karachentsev, Karachenseva \& B\"orngen\ (1985b)\markcite{karachentsev:etal85b} also performed optical spectroscopy of the four brightest knots in the Garland and determined a velocity of $(55\pm20)$\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi\ for the knot which is nearest to the center of our molecular complex ($30''$ to the southwest). The offset in velocity relative to the complex suggests that the two systems are probably not associated with each other. The observations therefore suggest that star formation did not proceed as yet in the newly detected molecular complex. However, even deeper optical follow up observations are needed to answer this question unambiguously. \section{Conclusions\label{conclusions}} We have established the presence of a giant cold and quiescent intergalactic molecular complex with no evidence of on--going star formation. The newly detected molecular complex is embedded in a large region containing \ion{H}{1} gas and thus has all the ingredients to form a dwarf galaxy in the future. This object may therefore be classified as a `Phoenix' galaxy and might resemble the young dwarf galaxies in the M\,81 group in the future. Our discovery shows that molecular complexes in tidal arms of interacting galaxies are not necessarily rare objects --- although they are difficult to find. \acknowledgments We thank an anonymous referee for valuable comments which helped to improve this Letter. FW acknowledges the 'Deutsche Forschungsgemeinschaft (DFG)' for the award of a stipendium in the Graduate School "The Magellanic Clouds and other Dwarf Galaxies". We thank Christian Henkel, Elias Brinks, and Klaas de Boer for fruitful discussions.
\section{Introduction} \paragraph{} The black hole solutions of four-dimensional (4d) heterotic string theory are very interesting from several points of view \cite{RD}. They give a general description of the whole set of charged black hole solutions of string theory (and general relativity) in four dimensions. Heterotic, type $II A$ and type $II B$ strings in four dimensions are connected one with the other by a web of dualities \cite{DLR,CA}, so that one can use the black hole solutions of 4d heterotic string theory as representative for the whole set of 4d black hole solutions in string theory. Moreover, using truncated models, it has been realized that there are only four classes of black holes, which can be described by effective 4d dilaton gravity models with dilaton coupling $a=\sqrt 3, 1, 1/\sqrt 3, 0$ \cite{DLP,RA}. The universal character of these solutions has found support both from the compositeness idea, according to which the $a= 1, 1/\sqrt 3, 0$ solutions can be considered as bound states of the $a=\sqrt 3$ elementary solution \cite {RA} and from the interpretation of them as intersection of D-branes \cite{PT}. Each class of 4d heterotic string black holes is characterized by its spacetime structure, its singularities and its thermodynamical behavior. Particularly remarkable is the behavior of the $a=0$ black hole (essentially the Reissner-Nordstrom black hole of general relativity) compared with that of the $a= 1, \sqrt 3, 1/\sqrt 3,$ cases. The extremal limit of the $a=0$ black hole is a zero-temperature ground state with non vanishing entropy, whereas for $a= 1, \sqrt 3, 1/\sqrt 3$ at the extremal limit we have zero temperature and entropy. The previous features can be understood using different approaches, even though a complete and satisfactory picture of the all subject is still missing. For instance, the existence of a ground state with the previously described behavior can be traced back to the presence of a mass gap. This mass gap can be explained both using string theory \cite{MS1} or just general features of the 4d effective dilaton gravity theory \cite{HW}. Until now the various attempts to give a general description and to clarify the subject have mainly approached it from ``above'' , i.e using higher dimensional, $d\ge 4$, models. The recently proposed AdS$_{d}$ (anti-de Sitter)/ CFT$_{d-1}$ (conformal field theory) correspondence \cite{Wm} from one side and the discovery of dualities between four- and two-dimensional black holes \cite{Hy} from the other side, have changed slightly the perspective. For $d=2,3$ the AdS/CFT duality has been used to compute the statistical entropy of 2d and 3d black holes \cite{St,CM}. On the other hand, it is known that near the horizon the geometry of the 4d, magnetically charged, $a=0,1,1/{\sqrt 3}$ black hole factorizes as the product of two spaces of constant curvature, ${\cal M}^{2}\times S^{2}$, where $S^{2}$ is the two-sphere and ${\cal M}^{2}$ is 2d Minkowski space for $a=1$ and ${\cal M}^{2}=AdS_{2}$ for $a=0,1/{\sqrt 3}$ \cite{CA}. It is therefore natural to try to extend the computation of Ref. \cite {CM} of the statistical entropy to the 4d case. The hope is that the explanation of the entropy in terms of microstates will also help to explain the different behavior of the various black hole solutions. Motivated by the previous arguments, in this paper we perform a generic $4d\rightarrow 2d$, spherical symmetric dimensional reduction of 4d effective heterotic string theory. We find that the near-horizon, near-extremal behavior of 4d heterotic string black holes are described by a class of 2d dilaton gravity models. We show that these 2d models encode all the relevant information about the 4d theory but in a much simpler form. In particular, we show that the duality group of the 4d theory is realized in the 2d theory in terms of Weyl transformations of the metric and we use the 2d dilaton gravity model to compute the statistical entropy of the near-extremal 4d, $a=1/\sqrt3$, black hole. In Sect. 2 we describe the dimensional reduction $4d\rightarrow 2d$ of heterotic string theory and the single-scalar field truncation that produces the effective 2d dilaton gravity models. In Sect. 3 we study the realization of the 4d duality symmetry in the 2d context, showing that it corresponds to Weyl transformations of the metric. In Sect. 4 we use the 2d dilaton gravity model together with the results of Ref. \cite {CM} to compute the statistical entropy of the near-extremal, 4d, $a=1/\sqrt3$, black hole. Finally in Sect. 5 we draw our conclusions. \section{ Dimensional reduction of 4d heterotic string theory} \paragraph{} In the string frame the bosonic action for heterotic string theory compactified on a six-torus \cite{MS,SE} can be written as follows: \begin{eqnarray} A_{H}&=&{1\over 16\pi}\int d^4x\ \sqrt{-g}\ e^{-2 \f}\left\{ R+4(\partial\phi)^{2}- 2 \left[\left( \partial \sigma\right)^2+\left( \partial \rho\right)^2\right]\right.\nonumber\\ &-& \left.{1\over 4}\left[ e^{-2 \sigma- 2\rho}F_{1}^{2}+e^{-2 \sigma+ 2\rho}F_{2}^{2}+ e^{2 \sigma+2 \rho}F_{3}^{2} + e^{2 \sigma-2 \rho}F_{4}^{2}\right]\right\}. \label{e1} \end{eqnarray} In the bosonic action of heterotic string theory toroidally compactified to $d=4$, we have set to zero the axion fields and all the $U(1)$ field strengths but four, two Kaluza-Klein fields $F_{1}, F_{2}$ and two winding modes $F_{3}, F_{4}$. In the action (\ref{e1}) and throughout this paper we set the 4d Newton constant $G=1$. The scalar fields $\phi,\rho, \sigma$ are related to the standard definitions of the string coupling, K\"ahler form and complex structure of the torus, through the equations \begin{equation} e^{-2 \f}={\rm Im}S,\quad e^{-2\sigma}={\rm Im}T,\quad e^{-2\rho}={\rm Im}U. \label{e2} \end{equation} The extremal, Bogomol'ny-Prasad-Sommerfield (BPS), solutions of the action (\ref{e1}) are given by (see for instance Ref. \cite{FK}) \begin{eqnarray}\label{e1a} ds^{2}&=& - \psi_{1} \psi_{3} \, dt^{2}+\left(\chi_{2}\chi_{4}\right)^{-1} (dr^2+r^2d\Omega^2_{2}), \nonumber\\ e^{4\phi}&=&{\psi_{1}\psi_{3}\over \chi_{2}\chi_{4}},\quad e^{4\sigma}={\psi_{1}\chi_{4}\over \chi_{2}\psi_{3}},\quad e^{4\rho}={\psi_{1}\chi_{2}\over \psi_{3}\chi_{4}},\nonumber\\ F_{1}&=& \pm d\psi_{1}\wedge dt,\quad \tilde F_{2}=\pm d\chi_{2} \wedge dt, \quad \quad F_{3}= \pm d\psi_{3}\wedge dt, \quad \tilde F_{4}= \pm d\chi_{4} \wedge dt, \end{eqnarray} where $(\psi_{1})^{-1}, (\psi_{3})^{-1}, (\chi_{2})^{-1},(\chi_{4})^{-1}$ are harmonic functions, $d\Omega^2_{2}$ is the metric of the two-sphere and $\tilde{F}_{2}= {e^{-2(\phi+\sigma-\rho)}}{^*}F_{2}$, $\tilde{F}_{4}= {e^{-2(\phi-\sigma+\rho)}}{^*}F_{4}$ ($^{*}$ denotes the Hodge dual). Motivated by the fact that the action (\ref{e1}) admits solutions that are the direct product of two two-dimensional spaces of constant curvature ${\cal M}^{2}\times S^{2 }$ \cite{CA}, we study the general, spherical symmetric, dimensional reduction $4d \rightarrow 2d$ of the theory. Let us first fix our notation. Greek letters from the middle of the alphabet denote 4d spacetime indices, $\mu ,\nu \ldots =0,1, 2,3$. Greek letters from the beginning of the alphabet denote 2d spacetime indices, $\alpha,\beta \ldots =0,1$. The capital latin letters $I,J..= 2,3 $ label the coordinates of the internal two-sphere, $ S^{2 }$. The lower-case latin letters $i,j\ldots=1,2,3,4$ are used to label the four $U(1)$ field strengths $F_{(i)}$ whereas $a,b\ldots= 1,2,3$ label the moduli, $\eta_{a}= (\phi,\sigma,\rho)$. The dimensional reduction of the 4d theory is performed by splitting the metric and the $U(1)$ field strengths into their 2d parts, using the ansatz \begin{eqnarray} \label{e3} ds^{2}_{(4)}&=&ds^{2}_{(2)}+ Q^{2}e^{2\psi} d\Omega^2_{2},\nonumber\\ F_{(i) \mu\nu}&=& \left\{ F_{(i) \alpha\beta}, F_{(i) IJ}\right\}, \end{eqnarray} where $Qe^{\psi}$ is the radius of the two-sphere, $Q$ is a constant that is related to the $U(1)$ charges (see Eq. (\ref{e6}) below) and the scalar fields $\eta_{a},\psi$ depend only on the 2d spacetime coordinates. The 4d field equations together with the ansatz (\ref{e3}) constrain strongly the form of the fields strengths $F_{(i)}$, we find \begin{equation}\label{e4} F_{(i)IJ}= p_{i}\epsilon_{IJ}, \quad F_{(i) \alpha\beta}= q_{i} e^{2 \eta\cdot b_{i}-2\psi}\epsilon_{\alpha\beta}. \end{equation} In the previous equations, and in the following, the dot means scalar product in the moduli space, $ b_{i}$ are the vectors: \begin{equation}\label{e5} b_{1}=(1,1,1), \quad b_{2}=(1,1,-1), \quad b_{3}=(1,-1,-1), \quad b_{4}=(1,-1,1), \end{equation} whereas $ p_{i}, q_{i}$ are respectively the magnetic and electric charge-vectors that characterize the particular dimensional reduction. For sake of simplicity, we will consider only vectors of the form, \begin{equation} \label{e6} p_{i}=Q\hat{p_{i}}, \qquad q_{i}=Q\hat{q_{i}}, \end{equation} where $\hat{p_{i}}$ and $\hat{q_{i}}$ are vectors with entries $0$ or $1$. The form of $ \hat {p_{i}}$ and $\hat { q_{i}}$ determines the 4d background on which we are performing the dimensional reduction. In general different charge-vectors will give rise to different 2d models. The 4d solutions are connected one with the other by O(3,Z) duality transformations, some of them change the spacetime structure of the solutions, others leave it invariant \cite {CA}. The most efficient way to organize the spectrum of the 4d solutions is to use the O(3,Z) duality symmetries together with the compositeness idea \cite{CA}. It looks therefore very natural to use the same procedure of Ref. \cite{CA} to classify the 2d models deriving from the dimensional reduction of the action (\ref {e1}). The 4d solutions can be classified in multiplets, labeled by $N$, of a given number (up to four) of elementary constituent, on which the duality symmetry $O(3, Z)$ acts in a natural way \cite{CA} . Moreover, the multiplets with $N=1,2,3,4$ can be put in correspondence with the solutions of the single-scalar, single $U(1)$ field strength model \cite {DLP,RA}, \begin{equation}\label{e6a} A={1\over 16\pi}\int d^4x\ \sqrt{-g}\ \left\{ R -2(\partial\hat \phi)^{2} - {1\over 4} e^{-2 a\hat \phi} F^{2}\right\}, \end{equation} with coupling $a$ given respectively by $a=\sqrt 3, 1, 1/\sqrt 3, 0$. It turns out that 4d solutions characterized by the same number of magnetic and electric elementary constituents produce after dimensional reduction the same 2d model. For this reason the various dimensional reductions (or equivalently the corresponding 2d models) can be classified by giving, apart form $N$, the numbers $n,m$ of electric, respectively, magnetic elementary constituents, with the obvious constrain $N=n+m$. In this way one obtains 8 different 2d models, which can be put in correspondence with BPS states of the 4d model (\ref{e1}), \centerline{ } \noindent \leftline{$N=1$ multiplet, $a=\sqrt 3$,} \begin{eqnarray}\label{e7} n&=&1,\quad m=0,\quad \hat q_{i}=(1,0,0,0), \quad \hat p_{i}=0;\nonumber\\ n&=&0,\quad m=1,\quad \hat q_{i}=0,\quad \hat p_{i}=(0,1,0,0). \end{eqnarray} \noindent \leftline{$N=2$ multiplet, $a=1$,} \begin{eqnarray}\label{e8} n&=&2,\quad m=0,\quad \hat q_{i}=(1,0,1,0), \quad \hat p_{i}=0;\nonumber\\ n&=&0,\quad m=2,\quad \hat q_{i}=0,\quad \hat p_{i}=(0,1,0,1);\nonumber\\ n&=&1,\quad m=1,\quad \hat q_{i}=(1,0,0,0),\quad \hat p_{i}=(0,0,0,1). \end{eqnarray} \noindent \leftline{$N=3$ multiplet, $a={1\over \sqrt 3}$,} \begin{eqnarray}\label{e9} n&=&2,\quad m=1,\quad \hat q_{i}=(1,0,1,0), \quad \hat p_{i}=(0,1,0,0); \nonumber\\ n&=&1,\quad m=2,\quad \hat q_{i}=(1,0,0,0),\quad \hat p_{i}=(0,1,0,1). \end{eqnarray} \noindent \leftline{$N=4$ multiplet, $a=0$,} \begin{equation}\label{f1} n=2,\quad m=2,\quad \hat q_{i}=(1,0,1,0), \quad \hat p_{i}=(0,1,0,1). \end{equation} In the previous equations we have given only one representative element for each model, characterized by $\hat p_{i}$ and $\hat q_{i}$. There are, of course, different values of the U(1) charge-vectors that give rise to the same 2d model. For instance, the model with $N=1, n=1, m=0$ can be obtained not only with the values of $\hat q_{i}$ given in Eq. (\ref{e7}), but also when $\hat p_{i}=0 $ and $\hat q_{i}$ has one entry equal to one, the others being zero. This degeneracy is related with the duality symmetries of the theory and will be discussed in detail in the next section. In the following we will use $S(N,m)$ to denote a dimensional reduction associated with a state with $N$ elementary constituents, of which $m$ are magnetic ($n=N-m$). We can now perform explicitly the dimensional reduction of the model (\ref{e1}) defined by the ansatz (\ref{e3}). The duality symmetry group $O(3,Z)$ contains off-shell dualities, for this reason it is convenient to perform the dimensional reduction at the level of the equation of motion instead of directly dimensionally reducing the action (\ref{e1}). Using the ansatz (\ref{e3}) and the equations (\ref{e4}) in the 4d field equations steming from the action (\ref{e1}), we get the following 2d field equations, \begin{eqnarray}\label{f2} R&+& 4\nabla^{2}\Phi -2 (\partial \Phi)^{2}-2 (\partial \psi)^{2} -2 (\partial \eta)\cdot(\partial \eta)+\nonumber\\ &+&{1\over 2Q^{4}}e^{2\Phi}\left\{ \sum_{i}q_{i}^{2} e^{2 \eta\cdot b_{i}}- e^{-4\psi}\sum_{i}p_{i}^{2} e^{-2 \eta\cdot b_{i}}\right\}+ {2\over Q^{2}}e^{-2\psi}=0, \end{eqnarray} \begin{equation}\label{f3} 4\nabla_{\alpha}\left(e^{-2 \F}\nabla^{\alpha}\eta_{a}\right)-{1\over Q^{4}}\left\{ \sum_{i}b_{ia}q_{i}^{2} e^{2 \eta\cdot b_{i}}- e^{-4\psi}\sum_{i}b_{ia} p_{i}^{2} e^{-2 \eta\cdot b_{i}}\right\}=0, \quad a=2,3, \end{equation} \begin{equation}\label{f4} 4\nabla_{\alpha}\left(e^{-2 \F}\nabla^{\alpha}\psi\right)+{2\over Q^{4}}e^{-4\p} \sum_{i} p_{i}^{2} e^{-2 \eta\cdot b_{i}}- {4\over Q^{4}}e^{-2(\psi+\Phi)}=0, \end{equation} \begin{eqnarray}\label{f5} &&\nabla_{a}\nabla_\beta\Phi +\partial_{a}\Phi\partial_\beta\Phi - (\partial_{\alpha} \eta)\cdot(\partial_{\alpha}\eta)- \partial_{a}\psi\partial_\beta\psi+\nonumber\\ && - g_{\alpha\beta}\left\{ \nabla^{2}\Phi -{1\over 2} \left[(\partial \Phi)^{2}+ (\partial \psi)^{2} +(\partial \eta)\cdot(\partial \eta)\right] -{1\over 8Q^{4}}e^{2\Phi}\left( \sum_{i}q_{i}^{2} e^{2 \eta\cdot b_{i}}+\right.\right.\nonumber\\ &&\left.\left.e^{-4\psi}\sum_{i}p_{i}^{2} e^{-2 \eta\cdot b_{i}}\right)+ {1\over 2 Q^{2}}e^{-2\psi}\right\}=0, \end{eqnarray} where the curvature $R$ and all the differential operators are 2d quantities, the field $\Phi$ is the 2d dilaton related, as usual, to the 4d dilaton $\phi$ by \begin{equation}\label{f6} \phi=\Phi-\psi, \end{equation} and the vector $\eta_{a}$ is now defined in terms of $\Phi$, $\eta_{a}=(\Phi,\sigma,\rho)$. The field equations (\ref{f2})-(\ref{f5}) are rather complicated. They assume a much simpler form by considering a consistent, single-scalar field truncation that reduces them to those of 2d gravity coupled to the dilaton $\Phi$. These models have been widely investigated in recent years under the name of 2d dilaton gravity. The single-scalar field truncation is obtained using an ansatz expressing the fields $\psi,\sigma,\rho$ in terms of the dilaton $\Phi$ in a way that is consistent with the field equations (\ref{f2})-(\ref{f5}), \begin{eqnarray}\label{f7} e^{2\psi}&=&A^{2} \exp\left( {{4(m-2)\over a^{2}N}\,\Phi}\right),\nonumber\\ e^{2\sigma}&=&A^{-1} \exp\left( {{2C_{\sigma}\over a^{2}N}\,\Phi}\right) ,\nonumber\\ e^{2\rho}&=&A^{-1} \exp\left( {{2C_{\rho}\over a^{2}N}\,\Phi}\right), \end{eqnarray} where $A=\sqrt{3/2}$ for $N=1$, $A=1$ for $N=2,3,4$. $C_{\sigma}=-1$ for the states $S(1,1), S(2,1),$ $S(3,2)$; $C_{\sigma}=0$ for $S(2,2), S(2,0), S(4,2)$; $C_{\sigma}=1$ for $S(1,0), S(3,1)$. $C_{\rho}=-1$ for the states $S(1,1), S(1,0), S(3,2),$ $ S(3,1)$; $C_{\rho}=0$ for $S(2,2), S(2,0), S(2,1),$ $ S(4,2)$. After inserting Eqs. (\ref{e7})-(\ref{f1}) and (\ref{f7}) into the field equations (\ref{f2})-(\ref{f5}), we find that, for $N=1,2,3$, they are equivalent to those derived from a class of 2d dilaton gravity models whose action has the form, \begin{equation} \label{f8} A_{2}={1\over 2}\int d^2x\ \sqrt{-g}\ e^{-2 \F}\left[ R+\Omega (\partial\Phi)^{2}+\l^2 V(\Phi)\right], \end{equation} where $\l^2=1/Q^{2}$ for $N=1,2$, and $\l^2=1/(2Q^{2})$ for $N=3$. The kinetic coefficient $\Omega$ and the dilaton potential $V$ are completely determined in terms of $n, m,a, N,$ \begin{equation}\label{f8a} \Omega = 8\left( {1-n\over a^{2 }N}\right), \qquad V(\Phi)= \exp\left[ 4\left( {2-m\over a^{2}N}\right)\Phi\right]. \end{equation} For $N=4$ the field equations (\ref{f2})-(\ref{f5}) become equivalent to those derived from the action, \begin{equation}\label{f9} A={1\over 2}\int d^2x\ \sqrt{-g}\ e^{-2 \F}\left[ R+4(\partial\Phi)^{2}- {1\over 2} F^{2}+\l^2 \right], \end{equation} which describes the heterotic string in 2d target-space \cite{NY}. The class of dilaton gravity models defined by the action (\ref{f8}) contains, as particular cases, models that have been already investigated in the past. For $N=2, m=2$ the action describes the Callan-Giddings-Harvey-Strominger (CGHS) model \cite{CGHS}. For $N=2, m=1$ we get the Weyl rescaled CGHS model investigate in Ref. \cite {CM3}. Finally, for $N=3, m=2$ we obtain the Jackiw-Teitelboim (JT) model \cite{JT}. In general the 2d dilaton gravity model (\ref{f8}) give a near-horizon description of the, near-extremal, 4d black hole solution of the action (\ref{e1}). This near-horizon description corresponds to setting to zero the constant term in the harmonic functions appearing in Eq. (\ref{e1a}). On the other hand, one can easily verify that this is exactly the way to make the ansatz (\ref {f7}) consistent the 4d extremal solutions (\ref{e1a}). \section{Dualities and Weyl transformations} \paragraph{} In the canonical frame $g_{C}= e^{-2\phi}g_{S}$, the 4d field equations following from the action (\ref{e1}) are invariant under the action the $O(3,Z)$ duality group that acts on the moduli $\phi,\sigma,\rho$ and on the $U(1)$ field strengths but leaves the 4d metric unchanged \cite{CA}. The $O(3,Z)$ duality group can be generated using three $S-T-U$ duality transformations $\tau_{S},\tau_{T},\tau_{U}$ together with the permutation group $P_{3}$ acting on the moduli $\phi,\sigma,\rho$ and on the $U(1)$ field strengths $F_{i}$ (see Ref.\cite{CA}). After passing to the string metric $g_{S}$ and performing the dimensional reduction described in the previous section, the duality group $O(3,Z)$ becomes a symmetry of the 2d field equations (\ref{f2})-(\ref{f5}). Because the 4d action (\ref{e1}) is written in terms of the string metric the duality transformation will act also on the 2d metric $g_{\alpha\beta}$ and, owing to Eq. (\ref{e4}), also on the charge-vectors $p_{i}, q_{i}$. One can easily verify that the O(3,Z) duality transformations leave invariant the 2d dilaton field $\Phi$, whereas in terms of the remaining 2d fields $\psi,\rho,\sigma,g_{\alpha\beta}$ and of the charge vectors $p_{i}, q_{i}$ it is realized as follows: \begin{eqnarray}\label{g1} \tau_{S}&:& \psi\rightarrow -\psi -2\Phi,\quad g_{\alpha\beta}\rightarrow e^{-4 (\Phi+\psi)}g_{\alpha\beta}, \quad q_{1}\rightarrow p_{3},\quad q_{3}\rightarrow p_{1},\quad q _{2}\rightarrow p_{4},\nonumber\\ & & q_{4}\rightarrow p_{2};\nonumber\\ \tau_{T}&:& \sigma\rightarrow -\sigma,\quad q_{1} \leftrightarrow q_{4},\quad q_{2}\leftrightarrow q_{3};\nonumber\\ \tau_{U}&:& \rho\rightarrow -\rho,\quad q_{1}\leftrightarrow q_{2},\quad q_{3}\leftrightarrow q_{4}; \end{eqnarray} \begin{eqnarray}\label{g2} P_{3}: & &\sigma \leftrightarrow \rho,\quad q_{2}\leftrightarrow q_{4}; \nonumber\\ & &\psi\rightarrow \sigma-\Phi,\quad \sigma\rightarrow \psi +\Phi,\quad g_{\alpha\beta}\rightarrow e^{2 ( \sigma -\Phi-\psi)}g_{\alpha\beta}, \quad q_{3}\rightarrow p_{4},\quad q_{4}\rightarrow p_{3}; \nonumber\\ & &\psi\rightarrow \sigma-\Phi,\quad\sigma\rightarrow \rho,\quad \rho\rightarrow \Phi+\psi,\quad g_{\alpha\beta}\rightarrow e^{2 ( \sigma -\Phi-\psi)}g_{\alpha\beta}, \quad q_{4}\rightarrow q_{2}, \nonumber\\ & &q_{2}\rightarrow p_{3},\quad q_{3}\rightarrow p_{4}; \nonumber\\ & &\psi\rightarrow \rho-\Phi,\quad \rho\rightarrow \Phi+\psi,\quad g_{\alpha\beta}\rightarrow e^{2 ( \rho -\Phi-\psi)}g_{\alpha\beta}, \quad q_{2}\rightarrow p_{3}, \quad q_{3}\rightarrow p_{2}; \nonumber\\ & &\psi\rightarrow \rho-\Phi,\quad\sigma\rightarrow\Phi+\psi,\quad \rho\rightarrow \sigma,\quad g_{\alpha\beta}\rightarrow e^{2 ( \rho -\Phi-\psi)}g_{\alpha\beta}, \quad q_{2}\rightarrow q_{4}, \nonumber\\ & &q_{3}\rightarrow p_{2},\quad q_{4}\rightarrow p_{3}. \end{eqnarray} The previous equations describe the action of the duality group on electric states ($p_{i}=0$), the action on magnetic states ($q_{i}=0$) can be easily obtained from Eqs. (\ref{g1}),(\ref{g2}), by interchanging $q_{i}\leftrightarrow p_{i}$. The duality group generated by the transformations (\ref{g1}),(\ref{g2}) becomes extremely simple once we perform the single-scalar field truncation described in Sect. 2. One can easily verify that the transformations $\tau_{T},\tau_{U}$ and the first transformation in Eq. (\ref{g2}) change the ansatz (\ref{f7}) but not the resulting 2d action (\ref{f8}), (\ref{f9}). This fact has a natural explanation. $\tau_{T},\tau_{U}$ and the first duality in Eq. (\ref{g2}) do not change the number $n$ of electric (or magnetic $m$) elementary constituents of a state, so that they cannot change the 2d action because the latter is parametrized in terms of $n$ and $m$ only. On the other hand, using Eqs. (\ref{f7}) into Eqs. (\ref{g1}),(\ref{g2}), one finds that the remaining duality transformations (in particular the $\tau_{S}$ duality) act on the 2d dilaton gravity models (\ref{f8}) and (\ref{f9}) as Weyl transformations of the 2d metric, \begin{equation} \label{g3} g_{\alpha\beta}\rightarrow e^{P\Phi} g_{\alpha\beta}, \quad P={4\over a^{2}N} \left( m-m'\right), \end{equation} which map one into the other models in Eqs. (\ref{f8}), (\ref{f9}) with the same value of $N$ but with a number $m$ and $m'$ of magnetic elementary constituents. Acting at $N$ fixed, the Weyl transformations (\ref{g3}) connect one with the other models within a given multiplet. For instance, taking $N=2$ we have three models: $S(2,2)$ ($\Omega = 4$, $V(\phi)=1$, the CGHS model), $S(2,1)$ ($\Omega = 0$, $V(\phi)=\exp (2\Phi)$, the Weyl-rescaled CGHS model of Ref. \cite{CM3} and $S(2,0)$ ($\Omega = -4$, $V(\phi)=\exp (4\Phi)$). These models are obtained one from the other using the Weyl transformations (\ref{g3}). $g_{\alpha\beta}\rightarrow e^{2\Phi} g_{\alpha\beta}$, maps $S(2,2)\rightarrow S(2,1)$ and $S(2,1)\rightarrow S(2,0)$ whereas $g_{\alpha\beta}\rightarrow e^{4\Phi} g_{\alpha\beta}$ maps $S(2,2)\rightarrow S(2,0)$. The $\tau_{S}$ duality is essentially an electro/magnetic duality, so that the strong/weak coupling duality of the 4d theory (\ref{e1}) is translated, after dimensional reduction to two dimensions, at least for the 4d model under consideration, into a dilaton-dependent Weyl rescaling of the 2d metric. It has been shown that dilaton-dependent Weyl transformations leave invariant the physical parameters (the mass and for 2d black hole solutions, the Hawking temperature and radiation flux) associated with the solutions of 2d dilaton gravity \cite{CA2}. Also other physical parameters that can be expressed in terms of the mass and the temperature (e.g the entropy) are invariant under such transformations. The dimensional reduction seems to wash out most of the information about the magnetic or electric character of the 4d solution. At the 2d level the relevant information is encoded in the number $N$, the number of elementary constituents, the actual number of magnetic and electric elementary constituents being irrelevant for the physical parameters of the solution. At the level of the 4d theory this implies the equivalence of electrically and magnetically charged configurations, as long as only excitations near extremality are concerned. Moreover, in the 2d context, the duality implies the equivalence of spacetime structure that behave rather differently. 2d spacetimes of constant curvature ( e.g the solutions of the 2d model $S(3,2)$) are dual (i.e connected by Weyl transformations) to spacetimes with singularities (e.g the solutions of the 2d model $S(3,1)$). Presently, we do not know if this is a peculiarity of the 4d heterotic string theory (\ref{e1}) or a general feature of 4d effective string theory. However, our results indicate that the dimensionally reduced 2d theory takes care only of the relevant physical properties of the 4d model and can therefore be used to give a universal classification of the near-extremal behavior of the 4d black hole solutions of string theory. \section{ Statistical entropy of the a=1/${\bf\sqrt 3}$ 4d black hole} \paragraph{} The results of the previous sections together with those of Ref. \cite{CM} can be used to calculate, microscopically, the entropy of the near-extremal 4d, $a=1/\sqrt3$, black hole. The near-extremal, near-horizon behavior of this black hole is described by the 2d model of Eq. (\ref{f8}) with $N=3, m=2$, i.e by the JT dilaton gravity model. For the JT black hole a derivation of the statistical entropy has been given in Ref. \cite{CM}, one can, therefore, use it to compute, microscopically, the entropy of the 4d, $a=1/\sqrt3$, black hole. For $a=1/\sqrt3$ the single-scalar field model of Eq.(\ref {e6a}) takes the form, \begin{equation}\label{h1} A={1\over 16\pi}\int d^4x\ \sqrt{-g}\ \left\{ R-2(\partial\hat \phi)^{2}- {1\over 4} e^{-{2\over \sqrt 3}\hat \phi}F^{2}\right\}, \end{equation} where the scalar field $\hat \phi$ is connected to the 4d dilaton $\phi$ trough $ \hat \phi= \sqrt 3 \,\phi$. As mentioned above, this model arises as single-scalar field, single $U(1)$ field strength truncation of the $N=3$ composite solutions of the action (\ref{e1}) and as compactification of the five-dimensional (5d) Einstein-Maxwell theory. The general (non extremal) black hole solution of the model has the form \cite{DLP1}, \begin{eqnarray}\label{h2} ds^{2}&=&- H^{-{3\over2}}\left( 1-{\mu\over r}\right) dt^{2}+ H^{3\over2}\left( 1-{\mu\over r}\right)^{-1} dr^2+ H^{3\over2} r^2 d\Omega^2_{2}, \nonumber\\ e^{2\phi}&=& H^{1\over 2},\quad H=1+{\mu \sinh ^{2} \alpha \over r}, \end{eqnarray} where for simplicity we have set the constant mode of the 4d dilaton $\phi_{0}=0$. The integration constants $\mu$ and $\alpha$ are related to the mass and charge of the solution by \begin{equation}\label{h3} M= {1\over 2} \mu \left( 1+{3\over 2}\sinh ^{2} \alpha\right), \quad Q={1\over 2} \sinh 2 \alpha. \end{equation} To be more precise, $Q$ is the (common) charge of the three $U(1)$ fields in the action (\ref{e1}). $Q$ is related to the charge $\hat Q$ of the single $U(1)$ field appearing in the action (\ref{h1}) by $Q= (2/\sqrt 3)\hat Q $. Using the area law we find that the Bekenstein-Hawking entropy of the hole is given by \begin{equation}\label{h4} S={{\cal A} \over 4}= \sqrt \mu\left( \mu+ \mu\sinh ^{2} \alpha\right)^{3/2}, \end{equation} where ${\cal A}$ is the area of the event horizon. The extremal black hole is obtained in the limit $\mu\rightarrow 0,\, \alpha \rightarrow \infty$, keeping $\mu \sinh ^{2} \alpha=Q$. In this limit the solution is given by Eq. (\ref {h2}) with $\mu=0$ and $H=(1+ Q/r)$ whereas the mass is $M_{ex}= {3\over 4}Q$. Let us now consider small excitations near the extremal solution, the entropy of these configurations is given by \begin{equation}\label {h4a} S_{(4)}= \pi Q^{3/2} \sqrt \mu +\ord{\mu^{3/2}}. \end{equation} We know that, when expressed in terms of the string metric, the near-horizon, near-extremal, magnetically charged, black hole solution (\ref{h2}) factorizes as ${\cal M}^{2}\times S^{2}$, where ${\cal M}^{2}$ is the solution of the 2d dilaton gravity model (\ref{f8}) with $N=3,\, m=2, a=1/\sqrt 3$ , \begin{equation}\label{h5} A={1\over 2}\int d^2x\ \sqrt{-g}\ e^{-2 \F}\left( R+2\l^2 \right). \end{equation} (We have rescaled $\l^2 \rightarrow 2\l^2$ in order to mach the conventions of Ref. \cite{CM}). The dilaton gravity model (\ref{h5}) admits solutions that can be interpreted as black holes in 2d AdS space \cite{CM2}, \begin{equation}\label{h6} ds^2=-(\lambda^2x^2-b^2)dt^2+(\lambda^2x^2-b^2)^{-1}dx^2,\qquad e^{-2 \F}=e^{-2\Phi_{0}} \lambda x. \end{equation} The mass $M_{(2)}$ and the entropy $S_{(2)}$ of the 2d black hole are given in terms of the integration constants $b, \Phi_{0}$, by \begin{equation}\label{h7} M_{(2)}={1\over 2} e^{-2\Phi_{0}} b^2\lambda, \quad S_{(2)}=4\pi \sqrt{ e^{-2\Phi_{0}} M_{(2)}\over 2\lambda}. \end{equation} If the 2d model (\ref{h5}) has to describe the near-extremal, near-horizon regime of the 4d black hole solutions (\ref{h2}) then the 2d expression for the entropy (\ref{h7}) should match the leading order of the corresponding 4d quantity in Eq. (\ref{h4a}). To show this, we first write the 4d solution (\ref{h2}) in terms of the string metric, we expand the solution near extremality and near the horizon. After some manipulations we get \begin{eqnarray}\label{h7a} ds^{2}&=&-\left(\l^2 x^{2} -\lambda \mu \right) dt^{2}+ \left(\l^2 x^{2} -\lambda \mu \right)^{-1} dx^2+ e^{2\phi_{0}}Q^{2} d\Omega^2_{2}, \nonumber\\ e^{-2\phi}&=&\sqrt 2 \lambda x. \end{eqnarray} where $\lambda = 1/(2Q)$. As expected the 4d solution factorizes as the product of a 2d spacetime and a two-sphere of constant radius. Taking into account that the dimensional reduction $4d\rightarrow 2d$ implies the following relation between the 2d dilaton $\Phi$ and the 4d one $\phi$: \begin{equation}\label{h8} e^{-2 \F}= {1\over 2} Q^{2} e^{-2 \f}, \end{equation} and comparing Eq. (\ref{h7a}) with Eq. (\ref{h6}), we get \begin{equation}\label{h8a} \mu={2M_{(2)}\over \l^2} e^{2\Phi_{0}},\quad e^{-2\Phi_{0}}={\sqrt 2\over 8\l^2}. \end{equation} Using Eqs. (\ref{h8a}) into the expression (\ref{h4a}) for the 4d entropy and taking into account only the leading term, we obtain a complete agreement with the 2d results, i.e $S_{(4)}=S_{(2)}$. Until now we have considered only 4d, magnetically charged, solutions, i.e the state $S(3,2)$. The 4d electrically charged solution $S(3,1)$ does not factorize as direct product of two 2d spaces. Near extremality it is described by the 2d model with $N=3, m=1$, which is dual to the model (\ref {h5}). Because the 2d entropy does not change under Weyl rescaling of the metric, it follows that the Eqs. (\ref{h7}) and the equality $S_{(4)}=S_{(2)}$ hold also for excitations near 4d extremal, electrically charged, solutions. We have shown that the semiclassical dynamics of small excitations near extremality of the 4d black hole can be described by the 2d model (\ref{h5}) and that at the leading order, the 2d and 4d thermodynamical entropy is the same. One can therefore use the results of Ref. \cite{CM} as an indirect calculation of the statistical entropy of the 4d black hole (\ref{h2}) in the near-extremal regime. In Ref. \cite{CM} we have found a mismatch of a $\sqrt 2$ factor between the thermodynamical and the statistical entropy of the 2d black hole. This implies that also the statistical entropy of the 4d $a=1/\sqrt 3$ black hole agrees only up to a $\sqrt 2$ factor with the thermodynamical result. A simple explanation of this $\sqrt 2$ factor could be found when the 2d AdS black hole arises as compactification of 3d one \cite{CM}. The same arguments of Ref. \cite{CM} apply also in the case under consideration because the 4d black hole solution (\ref{h2}) arises as compactification of the solutions of 5d Einstein-Maxwell gravity that behave, near the horizon, as AdS$_{3}\times S^{2}$. Let us consider the 5d Einstein-Maxwell action, \begin{equation}\label{h9} A=\int d^5x\ \sqrt{-g}\ \left\{ R- {1\over 4} F^{2}\right\}. \end{equation} Compactifying the fifth dimension $x_{4}$, using the ansatz \begin{eqnarray}\label{l1} ds^{2}_{(5)}&=&e^{-{4\over \sqrt 3} \hat \phi}dx_{4}^{2}+e^{{2\over \sqrt 3} \hat \phi}ds^{2}_{(4)},\nonumber\\ F_{\hat \mu, \hat \nu}&=&\left\{ F_{\mu \nu}, F_{4 \nu}\right), F_{4 \nu}=0, \quad \hat \mu, \hat \nu=0\ldots 4, \end{eqnarray} we get the 4d action (\ref{h1}). The extremal 5d solutions (\ref{l1}) behave near the horizon as AdS$_{3}\times S^{2}$, \begin{equation}\label{l2} ds^{2}_{(5)}={r\over Q}\left( -dt^{2}+ dx_{4}^{2}\right) + \left({Q\over r}\right)^{2}dr^{2} +Q^{2} d\Omega^2_{2}. \end{equation} Hence, the explanation of the $\sqrt 2$ factor of Ref. \cite{CM} can be immediately translated to the case under consideration. Moreover, the expression (\ref{l2}) suggests that the discrepancy between thermodynamical and statistical entropy of the 4d, $a=1/\sqrt3$, black hole could also have a geometrical explanation. The 3d part of the metric (\ref{l2}) describes a spacetime that is AdS$_{3}$ with a conical singularity. In fact, if in Eq. (\ref{l2}) $ x_{4}= Q\varphi$, with $0\le\varphi\le 2\pi$, changing coordinates $r\rightarrow r^{2}/4, \varphi \rightarrow 2 \varphi$, the 3d part of the metric (\ref{l2}) becomes \begin{equation}\label{l3} ds^{2}_{(3)}= - {r^{2}\over 4 Q^{2}} dt^{2}+{4 Q^{2}\over r^{2}} dr^{2}+r^{2}d \varphi^{2}, \end{equation} but with $0\le\varphi\le \pi$. \section {Conclusions} The dimensional reduction of 4d heterotic string theory presented in this paper has shown once again that 2d dilaton gravity models can be used as a simplified description that retain the relevant information about the 4d physics. The class of 2d dilaton gravity models we have derived gives a general description of excitations near the extremal 4d heterotic black hole. The geometrical structures, thermodynamical features and the duality symmetries of the 4d theory become much simpler when translated in the 2d context. Particularly interesting are those 2d models that admit $AdS_{2}$ as solution. In this case one can use the AdS/CFT duality to compute the statistical entropy of the near-extreme 4d black hole. We have performed this calculation for the $a=1/\sqrt 3$ black hole but in principle the same should be possible for the $a=0$ black hole. The near-horizon geometry factorizes also for $a=0$ as AdS$_{2} \times S^{2}$ and the excitations near extremality are now described by the model (\ref{f9}). Differently from the $a=1/\sqrt 3$ case, where we have a linear varying $\exp(-2\Phi)$, for $a=0$ the dilaton is constant. A constant dilaton makes a black hole interpretation of the solutions very difficult, at least from the 2d point of view. One cannot use the arguments of Ref. \cite{CM} to compute the statistical entropy of the black hole. This difficulties of the $a=0$ case (actually the most interesting case from the string point of view) are connected with a peculiarity, mentioned in the introduction, of the $a=0$ case, namely the existence of a mass gap separating the extremal configuration from the continuous part of the spectrum. This implies that the finite-energy excitations near extremality are suppressed \cite{MMS}. Probably, this behavior is related with other puzzling features of the AdS$_{2}$/CFT$_{1}$ correspondence \cite{MMS,st1, CM}. .
\section{Introduction} \setcounter{equation}{0} A deeper understanding of flavor mixing and $CP$ violation, observed in the weak interactions, remains one of the major challenges in particle physics. In the standard electroweak theory with three quark families the phenomenon of flavor mixing is described by a $3\times 3$ unitary matrix, which can be expressed in terms of four independent parameters, usually taken to be three rotation angles and one complex phase. There seems no way to obtain any further information about these parameters within the standard model. Any attempt to do so would require new physical inputs which are beyond the standard model. At the present time it seems hopeless to find a complete solution to the fermion mass and flavor mixing problem by theoretical insight alone. One can hope, however, to detect a specific order in the tower of fermion masses and the four parameters of quark flavor mixing, especially in observing links between the parameters of the flavor mixing and the mass eigenvalues. That such links should exist, seems obvious to us. Like in any quantum mechanical system the mixing pattern of the states will influence the pattern of the mass eigenvalues, and vice versa. One possible way to make these links more transparent is to look for specific symmetry limits, e.g., by setting parameters, which are observed to be small, to zero and to study the situation in the symmetry limit first. Following such an approach, we shall demonstrate that (a) a specific description of quark flavor mixing can be derived, (b) two of the three flavor mixing angles are related directly to the quark mass ratios $m_u/m_c$ and $m_d/m_s$, and (c) the unitarity triangle of quark mixing related to $CP$ violation in $B$-meson decays is fixed in terms of these mass ratios and the modulus of the Cabibbo transition element $|V_{us}|$. Furthermore we shall give arguments why an inner angle of the unitarity triangle (angle $\alpha$) should be equal to $90^{\circ}$ or close to $90^{\circ}$. The ``standard'' parametrization of the flavor mixing matrix (advocated by the Particle Data Group \cite{PDG98}) and the original Kobayashi-Maskawa parametrization \cite{KM} were introduced without taking possible links between the quark masses and the flavor mixing parameters into account. The parametrization introduced by us some time ago \cite{FX97,FX98} is based on such a connection, although the specific relations between flavor mixing angles and quark masses might be more complicated than commonly envisaged. It is a parametrization which allows to interpret the phenomenon of flavor mixing as an evolutionary or tumbling process. In the limit in which the masses of the light quarks $(u,d)$ and the medially light quarks $(c,s)$ are set to zero, while the heavy quarks $(t,b)$ acquire their masses, there is no flavor mixing \cite{F87}. Once the masses of the $(c,s)$ quarks are introduced, while the $(u,d)$ quarks remain massless, the flavor mixing is reduced to an admixture between two families, described by one angle $\theta$. As soon as the $u$- and $d$-quark masses are introduced as small perturbations, the full flavor mixing matrix involving a complex phase parameter and two more mixing angles $(\theta_{\rm u}, \theta_{\rm d})$ appears. These angles can be interpreted as rotations between the states $(u,c)$ and $(d,s)$, respectively. In either the ``standard'' parametrization or the Kobayashi-Maskawa representation, however, such specific limits are difficult to consider. For this reason we proceed to describe the flavor mixing by use of the parametrization given in Ref. \cite{FX97}. \section{The flavor mixing matrix} \setcounter{equation}{0} In the standard model or those extensions which have no flavor-changing right-handed currents, it is always possible to choose a basis of flavor space in which the up- and down-type quark mass matrices are hermitian. Without loss of any generality the (1,3) and (3,1) elements of both mass matrices can further be arranged, through a common unitary transformation, to be zero \cite{FX97}. Then one is left with hermitian quark mass matrices of the form \begin{equation} M_{\rm q} \; =\; \left (\matrix{ E_{\rm q} & D_{\rm q} & {\bf 0} \cr D^*_{\rm q} & C_{\rm q} & B_{\rm q} \cr {\bf 0} & B^*_{\rm q} & A_{\rm q} \cr} \right ) \; , \end{equation} where q = u (up) or d (down), and the hierarchy $|A_{\rm q}| \gg |B_{\rm q}|, |C_{\rm q}| \gg |D_{\rm q}|, |E_{\rm q}|$ is generally expected. In this basis, there is no direct mixing between the heavy $t$ (or $b$) quark and the light $u$ (or $d$) quark in $M_{\rm u}$ (or $M_{\rm d}$), i.e., the quark mass matrix is close to the well-known form of ``nearest-neighbour'' interactions \cite{F78}. A mass matrix of the type (2.1) can in the absence of complex phases be diagonalized by a $3\times 3$ orthogonal matrix, described only by two rotation angles in the hierarchy limit of quark masses \cite{F79}. First, the off-diagonal element $B_{\rm q}$ is rotated away by a rotation matrix $R_{23}$ between the second and third families. Then the element $D_{\rm q}$ is rotated away by a transformation $R_{12}$ between the first and second families. No rotation between the first and third families is necessary in either the limit $m_u\rightarrow 0$, $m_d\rightarrow 0$ or the limit $m_t\rightarrow \infty$, $m_b\rightarrow \infty$. Lifting such a hierarchy limit, which is not far from the reality, one needs an additional transformation $R_{31}$ with a tiny rotation angle to fully diagonalize $M_{\rm q}$. Note, however, that the rotation sequence $(R^{\rm u}_{12} R^{\rm u}_{23}) (R^{\rm d }_{12} R^{\rm d }_{23})^{\rm T}$ is enough to describe the $3\times 3$ real flavor mixing matrix, as the effects of $R^{\rm u}_{31}$ and $R^{\rm d}_{31}$ can always be absorbed into this sequence through redefining the relevant rotation angles. By introducing a complex phase angle into the rotation combination $(R^{\rm u}_{23}) (R^{\rm d}_{23})^{\rm T}$, we finally arrive at the following representation of quark flavor mixing \cite{FX97}: \begin{eqnarray} V & = & \left ( \matrix{ c_{\rm u} & s_{\rm u} & 0 \cr -s_{\rm u} & c_{\rm u} & 0 \cr 0 & 0 & 1 \cr } \right ) \left ( \matrix{ e^{-{\rm i}\varphi} & 0 & 0 \cr 0 & c & s \cr 0 & -s & c \cr } \right ) \left ( \matrix{ c_{\rm d} & -s_{\rm d} & 0 \cr s_{\rm d} & c_{\rm d} & 0 \cr 0 & 0 & 1 \cr } \right ) \nonumber \\ \nonumber \\ & = & \left ( \matrix{ s_{\rm u} s_{\rm d} c + c_{\rm u} c_{\rm d} e^{-{\rm i}\varphi} & s_{\rm u} c_{\rm d} c - c_{\rm u} s_{\rm d} e^{-{\rm i}\varphi} & s_{\rm u} s \cr c_{\rm u} s_{\rm d} c - s_{\rm u} c_{\rm d} e^{-{\rm i}\varphi} & c_{\rm u} c_{\rm d} c + s_{\rm u} s_{\rm d} e^{-{\rm i}\varphi} & c_{\rm u} s \cr - s_{\rm d} s & - c_{\rm d} s & c \cr } \right ) \; , \end{eqnarray} where $s_{\rm u} \equiv \sin\theta_{\rm u}$, $c_{\rm u} \equiv \cos\theta_{\rm u}$, etc. The three mixing angles can all be arranged to lie in the first quadrant, i.e., all $s_{\rm u}$, $s_{\rm d}$, $s$ and $c_{\rm u}$, $c_{\rm d}$, $c$ are positive. The phase $\varphi$ may in general take all values between 0 and $2\pi$. Clearly $CP$ violation is present, if $\varphi \neq 0$ or $\pi$. Although we have derived in a heuristic way the particular description of the flavor mixing matrix (2.2) from the hierarchical mass matrix (2.1), we should like to emphasize that (2.2) is a possible way to describe any mixing matrix, one out of nine inequivalent representations classified in Ref. \cite{FX98}. If the phase $\varphi$ in $V$ is disregarded, the resulting rotation matrix (obtained from (2.2) for $\varphi =0$) is just the one used originally by Euler; i.e., the angles $\theta$, $\theta_{\rm u}$ and $\theta_{\rm d}$ correspond to the usual Euler angles \cite{EUL}. Note that this is not the case for other representations of the flavor mixing matrix given in the literature \cite{KM,Others}. The representation given in (2.2) can be interpreted as follows. First, a rotation by the angle $\theta_{\rm d}$ takes place in the plane defined by the $d$ and $s$ quarks. It is followed by a rotation (angle $\theta$) in the $b$--$s'$ plane, where $s'$ denotes the superposition $s' = d \sin\theta_{\rm d} + s \cos\theta_{\rm d}$. At the same time the orthogonal state $d' = d \cos\theta_{\rm d} - s \sin\theta_{\rm d}$ is multiplied by the phase factor $e^{-{\rm i}\varphi}$. Finally a rotation (angle $\theta_{\rm u}$) is applied in the $1$--$2$ plane (about the new third axis). The sequence of rotations corresponds just to the Euler sequence \cite{EUL}: $R_{12} R_{23} R^{\rm T}_{12}$. On the other hand, the original Kobayashi-Maskawa representation \cite{KM} corresponds to the sequence $R_{23} R_{12} R^{\rm T}_{23}$, while the ``standard'' representation \cite{PDG98} corresponds to the sequence $R_{23} R_{31} R_{12}$ (see also the classifications given in Ref. \cite{FX98}). Although all descriptions of the flavor mixing matrix are mathematically equivalent, we emphasize that the Euler sequence $R_{12} R_{23} R^{\rm T}_{12}$ is physically of particular interest, as it involves the rotation matrices $R_{12}$ and $R^{\rm T}_{12}$, which describe the rotations in the light quark sector, in a symmetric way. Since the flavor mixing matrix acts between the quark mass eigenstates ${\cal U} = (u, c, t)$ and ${\cal D} = (d, s, b)$, one could absorb the two $R_{12}$ rotations in a redefinition of the quark fields. The charged weak transition term can be rewritten as follows: \begin{equation} \overline{\cal U}_{\rm L} ~ V ~ {\cal D}_{\rm L} \; =\; \overline{(u, ~ c, ~ t)}^{~}_{\rm L} ~ V ~ \left ( \matrix{ d \cr s \cr b \cr} \right )_{\rm L} \; =\; \overline{(u', ~ c', ~ t)}^{~}_{\rm L} \left ( \matrix{ e^{-{\rm i}\varphi} & 0 & 0 \cr 0 & c & s \cr 0 & -s & c \cr} \right ) \left ( \matrix{ d' \cr s' \cr b \cr} \right )_{\rm L} \; , \end{equation} where $u' = u \cos\theta_{\rm u} - c \sin\theta_{\rm u}$ and $c' = c \cos\theta_{\rm u} + u \sin\theta_{\rm u}$. Thus the angles $\theta_{\rm u}$ and $\theta_{\rm d}$ describe the corresponding rotations in the $(u, c)$ and $(d, s)$ systems. We should like to emphasize that the angles $\theta_{\rm u}$ and $\theta_{\rm d}$ can directly be measured from weak decays of $B$ mesons and from $B^0$-$\bar{B}^0$ mixing. An analysis of the present experimental data yields \cite{PRS98}: $\theta_{\rm u} = 4.87^{\circ} \pm 0.86^{\circ}$ and $\theta_{\rm d} = 11.71^{\circ} \pm 1.09^{\circ}$. Taking the central values for illustration, one has \begin{eqnarray} d' & = & d \cos\theta_{\rm d} ~ - ~ s \sin\theta_{\rm d} \; \approx \; 0.979 d ~ - ~ 0.203 s \; , \nonumber \\ s' & = & d \sin\theta_{\rm d} ~ + ~ s \cos\theta_{\rm d} \; \approx \; 0.203 d ~ + ~ 0.979 s \; ; \nonumber \\ u' & = & u \cos\theta_{\rm u} ~ - ~ c \sin\theta_{\rm u} \; \approx \; 0.996 u ~ - ~ 0.085 c \; , \nonumber \\ c' & = & u \sin\theta_{\rm u} ~ + ~ c \cos\theta_{\rm u} \; \approx \; 0.085 u ~ + ~ 0.996 c \; . \end{eqnarray} The question, about whether these mixtures of mass eigenstates have a specific physical meaning, arises. This will be discussed in some more detail below. Due to the symmetric structure of our mixing matrix (2.2), we are able to interpret the $\theta_{\rm d}$ and $\theta_{\rm u}$ rotations as specific transformations of the corresponding mass eigenstates. Such an interpretation is not possible for the third rotation given by $\theta$, measured to be $2.30^{\circ} \pm 0.09^{\circ}$ \cite{PRS98}. This rotation takes place between the third family of the massive quarks and the $c'$ and $s'$ states. One interpretation would be to associate the rotation of $\theta$ with a transformation among $b$ and $s'$. Another possibility is to describe the effect as a rotation among $t$ and $c'$. However, one could also write $\theta$ as a difference of two other angles, and describe the mixing effect as a combination of a rotation in the $(b, s')$ system and a rotation in the $(t, c')$ system. Thus a unique interpretation does not exist. We remark that the asymmetry between the $\theta$ rotation on the one hand and the $\theta_{\rm u}$ and $\theta_{\rm d}$ rotations on the other hand is a direct consequence of our flavor mixing matrix (which is in turn related to the hierarchical structure of the mass spectrum) and is linked to the fact that there exist three different quark families. It is worthwhile to point out the similarity and difference between our new parametrization and the Kobayashi-Maskawa parametrization, which both result from rotations in the 1--2 and 2--3 planes (i.e., $R_{12}$ and $R_{23}$), in the description of quark flavor mixing. To make a comparison, we write out the latter as follows: \begin{eqnarray} V_{\rm KM} & = & \left (\matrix{ 1 & 0 & 0 \cr 0 & c_2 & s_2 \cr 0 & -s_2 & c_2 \cr} \right ) \left (\matrix{ c_1 & s_1 & 0 \cr -s_1 & c_1 & 0 \cr 0 & 0 & e^{-{\rm i}\delta} \cr} \right ) \left (\matrix{ 1 & 0 & 0 \cr 0 & c_3 & -s_3 \cr 0 & s_3 & c_3 \cr} \right ) \; \nonumber \\ \nonumber \\ & = & \left ( \matrix{ c_1 & s_1 c_3 & -s_1 s_3 \cr -s_1 c_2 & c_1 c_2 c_3 + s_2 s_3 e^{-{\rm i} \delta} & -c_1 c_2 s_3 + s_2 c_3 e^{-{\rm i}\delta } \cr s_1 s_2 & -c_1 s_2 c_3 + c_2 s_3 e^{-{\rm i}\delta} & c_1 s_2 s_3 + c_2 c_3 e^{-{\rm i}\delta} \cr } \right ) \; , \end{eqnarray} where $s_1 \equiv \sin\theta_1$, $c_1 \equiv \cos\theta_1$, etc. The mixing angles $(\theta_{\rm u}, \theta_{\rm d}, \theta)$ are related to $(\theta_1, \theta_2, \theta_3)$ simply through the product of $|V_{ub}|$ and $|V_{td}|$, i.e., $s_{\rm u} s_{\rm d} s^2 = s^2_1 s_2 s_3$ holds. One can also link the phase parameter $\varphi$ to $\delta$ with the help of the common rephasing-invariant measure of $CP$ violation \cite{J85}; i.e., \begin{eqnarray} {\cal J} & = & s_{\rm u} c_{\rm u} s_{\rm d} c_{\rm d} s^2 c \sin\varphi \nonumber \\ & = & s^2_1 c_1 s_2 c_2 s_3 c_3 \sin\delta \; . \end{eqnarray} With no fine-tuning of the relevant mixing angles, we arrive at the equality between $\varphi$ and $\delta$ to an excellent degree of accuracy: \begin{equation} \frac{\sin \varphi}{\sin \delta} \; =\; \frac{c_1 ~ c_2 ~ c_3}{c_{\rm u} ~ c_{\rm d} ~ c} \; =\; 1 ~ - ~ O(\lambda^2) \; , \end{equation} where $\lambda \approx s_{\rm d} \approx s_1 \approx 0.2$. Therefore a large $CP$-violating phase (close to $90^{\circ}$), as required either phenomenologically \cite{FX97} or in a specific dynamical scheme \cite{FX95,Weyers}, must manifest itself in both (2.2) and (2.5). The difference between these two representations is however significant. For example, the Kobayashi-Maskawa parametrization starts from the second largest matrix element of $V$ (i.e., $|V_{ud}|$ instead of $|V_{tb}|$) and leads to quite complicated results for the ratios $|V_{ub}/V_{cb}|$ and $|V_{td}/V_{ts}|$. As summarized in Refs. \cite{FX97,FX98}, the new parametrization (2.2) has a number of advantages over all the others in the study of heavy flavor decays and quark mass matrices. Its usefulness will be seen more clearly in the present work. As an example we explore the interesting connection between our parametrization (2.2) and the unitarity triangle of quark mixing defined by the orthogonality relation \begin{equation} V^*_{ub}V_{ud} ~ + ~ V^*_{cb}V_{cd} ~ + ~ V^*_{tb}V_{td} \; =\; 0 \end{equation} in the complex plane. The inner angles of this triangle, usually denoted as \begin{eqnarray} \alpha & = & \arg \left ( - \frac{V^*_{tb}V_{td}}{V^*_{ub}V_{ud}} \right ) \; , \nonumber \\ \beta & = & \arg \left (- \frac{V^*_{cb}V_{cd}}{V^*_{tb}V_{td}} \right ) \; , \nonumber \\ \gamma & = & \arg \left (- \frac{V^*_{ub}V_{ud}}{V^*_{cb}V_{cd}} \right ) \; , \end{eqnarray} can be determined from some $CP$-violating asymmetries in $B$-meson decays \cite{BB}. Current data indicate that the unitarity triangle (2.8) is congruent, to a good degree of accuracy, with another unitarity triangle defined by the orthogonality relation $V^*_{td}V_{ud} + V^*_{ts}V_{us} + V^*_{tb}V_{ub} =0$ in the complex plane \cite{Xing96}. In view of the approximate congruency between two unitarity triangles and the smallness of three mixing angles, we find that the parametrization (2.2) takes an instructive leading-order form: \begin{equation} V \; \approx \; \left (\matrix{ e^{-{\rm i}\alpha} & s^{~}_{\rm C} e^{{\rm i}\gamma} & s_{\rm u} s \cr s^{~}_{\rm C} e^{{\rm i}\beta} & 1 & s \cr -s_{\rm d} s & -s & 1 \cr} \right ) \; , \end{equation} where $s^{~}_{\rm C} \equiv \sin \theta_{\rm C} \approx |s_{\rm u} - s_{\rm d} e^{-{\rm i}\varphi}|$ with $\theta_{\rm C}$ denoting the Cabibbo rotation angle \cite{Cabibbo}. Clearly $\alpha \approx \varphi$ holds as a straightforward result of (2.10). In this approximation $|V^*_{ub}V_{ud}|$, $|V^*_{cb}V_{cd}|$ and $|V^*_{tb}V_{td}|$, three sides of the unitarity triangle (2.8), are rescaled to $s_{\rm u}$, $s_{\rm d}$ and $s^{~}_{\rm C}$ respectively. The latter are three sides of a new triangle with smaller area ($\approx s_{\rm u}s_{\rm d} \sin\alpha/2$), which will subsequently be referred to as the ``light-quark triangle'' in the heavy quark limit ($m_t\rightarrow \infty$, $m_b\rightarrow \infty$). The values of $\alpha$, $\beta$ and $\gamma$ can therefore be given in terms of $s_{\rm u}$, $s_{\rm d}$ and $s^{~}_{\rm C}$ with the help of the cosine theorem. In particular, relations like \cite{FX97} \begin{equation} \sin\alpha ~ : ~ \sin\beta ~ : ~ \sin\gamma \; \approx \; s^{~}_{\rm C} ~ : ~ s_{\rm u} ~ : ~ s_{\rm d} \; \end{equation} may directly be confronted with the upcoming data on $CP$ asymmetries in $B$ decays \cite{CDF}. Motivated by these interesting results, we shall investigate the role that the light quark sector plays in $CP$ violation for a variety of realistic textures of quark mass matrices. \section{Symmetry limits of quark masses} \setcounter{equation}{0} Going farther from the previous discussions \cite{FX97,FX98}, we remark two useful limits of quark masses and analyze their corresponding consequences on flavor mixing. \subsection{The chiral limit of quark masses} In the limit $m_u \rightarrow 0$, $m_d \rightarrow 0$ (``chiral limit''), where both the up and down quark mass matrices have zeros in the positions $(1,1)$, $(1,2)$, $(2,1)$, $(1,3)$ and $(3,1)$ (see also Ref. \cite{F87}), the flavor mixing angles $\theta_{\rm u}$ and $\theta_{\rm d}$ vanish. Only the $\theta$ rotation affecting the heavy quark sector remains, i.e., the flavor mixing matrix effectively takes the form \begin{equation} \hat{V} \; =\; \left ( \matrix{ \cos\hat{\theta} & \sin\hat{\theta} \cr - \sin\hat{\theta} & \cos\hat{\theta} \cr } \right ) \; , \end{equation} where $\hat{\theta}$ denotes the value of $\theta$ which one obtains in the limit $\theta_{\rm u} \rightarrow 0$, $\theta_{\rm d} \rightarrow 0$. We see that $\hat{V}$ is a real orthogonal matrix, arising naturally from $V$ in the chiral limit. The flavor mixing angle $\hat{\theta}$ can be derived from hermitian quark mass matrices of the following general form (in the limit $m_u \rightarrow 0$, $m_d \rightarrow 0$): \begin{equation} \hat{M}_{\rm q} \; =\; \left ( \matrix{ \hat{C}_{\rm q} & \hat{B}_{\rm q} \cr \hat{B}^*_{\rm q} & \hat{A}_{\rm q} \cr } \right ) \; , \end{equation} where $|\hat{A}_{\rm q}| \gg |\hat{B}_{\rm q}| , |\hat{C}_{\rm q}|$; and q = u (up) or d (down). Note that the phase difference between $\hat{B}_{\rm u}$ and $\hat{B}_{\rm d}$, denoted as $\kappa \equiv \arg (\hat{B}_{\rm u}) - \arg (\hat{B}_{\rm d})$, has no effect on $CP$ symmetry in the chiral limit, but it may affect the magnitude of $\hat{\theta}$. It is known that current data on the top-quark mass and the $B$-meson lifetime disfavor the special case $\hat{C}_{\rm u} = \hat{C}_{\rm d} =0$ for $\hat{M}_{\rm u}$ and $\hat{M}_{\rm d}$ (see, e.g., Ref. \cite{FX95}), hence we take $\hat{C}_{\rm q} \neq 0$ and define a ratio $\hat{r}_{\rm q} \equiv |\hat{B}_{\rm q}|/\hat{C}_{\rm q}$ for convenience. Then we can obtain the flavor mixing angle $\hat{\theta}$, in terms of the quark mass ratios $m_c/m_t$, $m_s/m_b$ and the parameters $\hat{r}_{\rm u}$, $\hat{r}_{\rm d}$, by diagonalizing the mass matrices in (3.2). In the next-to-leading order approximation, $\sin\hat{\theta}$ reads \begin{equation} \sin\hat{\theta} \; = \; \left | \hat{r}_{\rm d} \frac{m_s}{m_b} \left (1 - \hat{\delta}_{\rm d} \right ) ~ - ~ \hat{r}_{\rm u} \frac{m_c}{m_t} \left (1 - \hat{\delta}_{\rm u} \right ) e^{{\rm i} \kappa} \right | \; , \end{equation} where two correction terms are given by \begin{eqnarray} \hat{\delta}_{\rm u} & = & \left (1 + \hat{r}^2_{\rm u} \right ) \frac{m_c}{m_t} \; , \nonumber \\ \hat{\delta}_{\rm d} & = & \left (1 + \hat{r}^2_{\rm d} \right ) \frac{m_s}{m_b} \; . \end{eqnarray} In view of the fact $m_s/m_b \sim O(10) ~ m_c/m_t$ from current data \cite{PDG98,Leut}, we find that the flavor mixing angle $\hat{\theta}$ is primarily linked to $m_s/m_b$ provided $|\hat{r}_{\rm u}| \approx |\hat{r}_{\rm d}|$. Note that in specific models, e.g., those describing the mixing between the second and third families as an effect related to the breaking of an underlying ``democratic symmetry'' \cite{Democracy,New}, the ratios $\hat{r}_{\rm u}$ and $\hat{r}_{\rm d}$ are purely algebraic numbers (such as $|\hat{r}_{\rm u}| = |\hat{r}_{\rm d}| = 1/\sqrt{2}$ or $\sqrt{2}$). For illustration, we take $\hat{r}_{\rm u} = \hat{r}_{\rm d} \equiv \hat{r}$ to fit the experimental result $\sin\hat{\theta} = 0.040 \pm 0.002$ with the typical inputs $m_b/m_s = 26 - 36$ and $m_t/m_c \sim 250$. It is found that the favored value of $|\hat{r}|$ varies in the range 1.0 -- 2.5, dependent weakly on the phase parameter $\kappa$. Note that both $m_s/m_b$ and $m_c/m_t$ evolve with the energy scale (e.g., from the weak scale $\mu \sim 10^2$ GeV to a superhigh scale $\mu \sim 10^{16}$ GeV, or vice versa), therefore $\tilde{\theta}$ itself is also scale-dependent. \subsection{The heavy quark limit} The limit $m_t \rightarrow \infty$, $m_b \rightarrow \infty$ is subsequently referred to as the ``heavy quark limit''. In this limit, in which the $(3,3)$ elements of the up and down mass matrices formally approach infinity but all other matrix elements are fixed, the angle $\theta$ vanishes. The flavor mixing matrix, which is nontrivial only in the light quark sector, takes the form: \begin{eqnarray} \tilde{V} & = & \left ( \matrix{ \tilde{c}_{\rm u} & \tilde{s}_{\rm u} \cr -\tilde{s}_{\rm u} & \tilde{c}_{\rm u} \cr } \right ) \left ( \matrix{ e^{-{\rm i}\tilde{\varphi}} & 0 \cr 0 & 1 \cr } \right ) \left ( \matrix{ \tilde{c}_{\rm d} & -\tilde{s}_{\rm d} \cr \tilde{s}_{\rm d} & \tilde{c}_{\rm d} \cr } \right ) \nonumber \\ \nonumber \\ & = & \left ( \matrix{ \tilde{s}_{\rm u} \tilde{s}_{\rm d} + \tilde{c}_{\rm u} \tilde{c}_{\rm d} e^{-{\rm i}\tilde{\varphi}} & \tilde{s}_{\rm u} \tilde{c}_{\rm d} - \tilde{c}_{\rm u} \tilde{s}_{\rm d} e^{-{\rm i}\tilde{\varphi}} \cr \tilde{c}_{\rm u} \tilde{s}_{\rm d} - \tilde{s}_{\rm u} \tilde{c}_{\rm d} e^{-{\rm i}\tilde{\varphi}} & \tilde{c}_{\rm u} \tilde{c}_{\rm d} + \tilde{s}_{\rm u} \tilde{s}_{\rm d} e^{-{\rm i}\tilde{\varphi}} \cr } \right ) \; . \end{eqnarray} where $\tilde{s}_{\rm u} = {\rm sin} \tilde{\theta }_{\rm u}, \tilde{c}_{\rm u} = {\rm cos} \tilde{\theta}_{\rm u}$, etc. The angles $\tilde{\theta}_{\rm u}$ and $\tilde{\theta}_{\rm d}$ are the values for $\theta_{\rm u}$ and $\theta_{\rm d}$ obtained in the heavy quark limit. Since the $(t, b)$ system is decoupled from the $(c, s)$ and $(u, d)$ systems, the flavor mixing can be described as in the case of two families. Therefore the mixing matrix $\tilde{V}$ is effectively given in terms of only a single rotation angle, the Cabbibo angle $\theta_{\rm C}$ \cite{Cabibbo}: \begin{equation} \sin \theta_{\rm C} = \mid \tilde{s}_{\rm u} \tilde{c}_{\rm d} ~ - ~ \tilde{c}_{\rm u} \tilde{s}_{\rm d} ~ e^{-{\rm i} \tilde{\varphi}} \mid \; . \end{equation} Of course $\tilde{V}(\theta_{\rm C})$ is essentially a real matrix, because its complex phases can always be rotated away by redefining the quark fields. We should like to stress that the heavy quark limit, which carries the flavor mixing matrix $V$ to its simplified form $\tilde{V}$, is not far from the reality, since $1 - c \approx 0.1 \%$ holds \cite{FX97}. Therefore $\theta_{\rm u}$, $\theta_{\rm d}$ and $\varphi $ are expected to approach $\tilde{\theta}_{\rm u}$, $\tilde{\theta}_{\rm d}$ and $\tilde{\varphi}$ rapidly, as $\theta \rightarrow 0$, corresponding to $m_t \rightarrow \infty $ and $m_b \rightarrow \infty$. However, the concrete limiting behavior depends on the specific algebraic structure of the up and down mass matrices. If two hermitian mass matrices have the parallel hierarchy with texture zeros in the (1,1) (2,2), (1,3) and (3,1) elements, for example, the magnitude of $\theta $ is suppressed by the terms proportional to $m_t^{-1/2}$ and $m_b^{-1/2}$ \cite{F79}; and if the (2,2) elements are kept nonvanishing and comparable in magnitude with the (2,3) and (3,2) elements, then $\theta $ is dependent on $m_t^{-1}$ and $m_b^{-1}$ \cite{Democracy,New}. The angles $\tilde{\theta}_{\rm u}$ and $\tilde{\theta}_{\rm d}$ as well as the phase $\tilde{\varphi }$ are well-defined quantities in the heavy quark limit. The physical meaning of these quantities can be seen more clearly, if we take into account a specific and realistic model for the Cabibbo-type mixing in the light quark sector. It is well known that in the absence of the $u$-quark mass a relation between the Cabibbo angle $\theta_{\rm C}$ and the mass ration $m_d/m_s$ follows, if the quark mass matrices have the structure: \begin{eqnarray} \tilde{M}_{\rm u} & = & \left( \matrix{ ~ {\bf 0} & ~ {\bf 0} ~ \cr ~ {\bf 0} & ~ m_c \cr } \right) \; , \nonumber \\ \tilde{M}_{\rm d} & = & \left( \matrix{ {\bf 0} & \tilde{B}_{\rm d} \cr \tilde{B}^*_{\rm d} & \tilde{A}_{\rm d} \cr } \right) \; . \end{eqnarray} The diagonalization of $\tilde{M}_{\rm d}$ leads to the relation ${\rm tan } \theta_{\rm C} = \sqrt{m_d/m_s}$ . The texture-zero pattern of $\tilde{M}_{\rm d}$, i.e., the vanishing of its (1,1) element, is already present in certain classes of models (see, e.g., Refs. \cite{F77,Weinb}). The relation for the Cabibbo angle is known to agree very well with the experimental observation. For numerical discussions, we make use of the quark masses $m_u = (5.1 \pm 0.9)$ MeV, $m_d = (9.3 \pm 1.4)$ MeV, $m_s = (175 \pm 25)$ MeV and $m_c = (1.35 \pm 0.05)$ GeV at the scale $\mu =1$ GeV \cite{Leut}. Then one finds $\theta_{\rm C} = 13.0^{\circ } \pm 1.8^{\circ}$ or $\sin \theta_{\rm C} = 0.225 \pm 0.031$, consistent with the observed value of $|V_{us}|$ (i.e., $0.217 \leq |V_{us}| \leq 0.224$ \cite{PDG98}). The situation will change once $m_u$ is introduced, i.e., $\tilde{M}_{\rm u}$ takes the same form as $\tilde{M}_{\rm d}$ given in (3.7). In this case the mass matrices result in the following relation \cite{F79}: \begin{equation} \sin \theta_{\rm C} \; = \; \mid R_{\rm u} ~ - ~ R_{\rm d} ~ e^{-{\rm i} \psi} \mid \; , \end{equation} where \begin{eqnarray} R_{\rm u} & = & \sqrt{\frac{m_u}{m_u + m_c}} \, \sqrt{\frac{m_s}{m_d + m_s}} \;\; , \nonumber \\ R_{\rm d} & = & \sqrt{\frac{m_c}{m_u + m_c}} \, \sqrt{\frac{m_d}{m_d + m_s}} \; \; , \end{eqnarray} and $\psi \equiv \arg (\tilde{B}_{\rm u}) - \arg (\tilde{B}_{\rm d})$ denotes the relative phase between the off-diagonal elements $\tilde{B}_{\rm u}$ and $\tilde{B}_{\rm d}$ (in the limit $m_u \rightarrow 0$ this phase can be absorbed through a redifinition of the quark fields). We find that the same structure for the Cabibbo-type mixing matrix has been obtained as in the decoupling limit discussed above. If we set \begin{eqnarray} \tan \tilde{\theta}_{\rm u} & = & \sqrt{\frac{m_u}{m_c}} \; , \nonumber \\ \tan \tilde{\theta}_{\rm d} & = & \sqrt{\frac{m_d}{m_s}} \; , \end{eqnarray} and $\tilde{\varphi} = \psi$ for (3.6), then the result in (3.8) and (3.9) can exactly be reproduced. \begin{figure}[t] \begin{picture}(400,160)(-90,210) \put(80,300){\line(1,0){150}} \put(80,300.5){\line(1,0){150}} \put(150,285.5){\makebox(0,0){$\sin\theta_{\rm C}$}} \put(80,300){\line(1,3){21.5}} \put(80,300.5){\line(1,3){21.5}} \put(80,299.5){\line(1,3){21.5}} \put(71,333){\makebox(0,0){$R_{\rm u}$}} \put(230,300){\line(-2,1){128}} \put(230,300.5){\line(-2,1){128}} \put(178,343.5){\makebox(0,0){$R_{\rm d}$}} \put(107.5,348.5){\makebox(0,0){$\tilde{\varphi}$}} \end{picture} \vspace{-2.6cm} \caption{The light-quark triangle (LT) in the complex plane.} \end{figure} Indeed the relation in (3.6) or (3.8) defines a triangle in the complex plane (see Fig. 1 for illustration), which will be denoted as the ``light-quark triangle''(LT). Taking into account the central values of the Cabibbo angle ($\sin \theta_{\rm C} = |V_{us}| = 0.2205$) and the light quark mass ratios $\left( m_s / m_d = 18.8 \right. $ and $\left. m_c /m_u = 265 \right)$, we can calculate the phase parameter from (3.8) and obtain $\tilde{\varphi} = \psi \approx 79^{\circ}$. If we allow the mass ratios and $\theta_{\rm C}$ to vary in their ranges given above, then $\tilde{\varphi}$ may vary in the range $38^{\circ} - 115^{\circ}$. We find that $\tilde{\varphi}$ has a good chance to be around $90^{\circ}$ (see also Ref. \cite{FX95}). The case $\tilde{\varphi} \approx 90^{\circ}$ (i.e., the LT is rectangular) is of special interest, as we shall see later, since it implies that the area of the unitarity triangle of flavor mixing takes its maximum value for the fixed quark mass ratios -- in this sense, the $CP$ symmetry of weak interactions would be maximally violated. It is worth remarking that the quark mass ratios $m_d/m_s$ and $m_u/m_c$ are essentially independent of the renormalization-group effect from the weak scale to a superhigh scale (or vice versa), so is the Cabibbo angle $\theta_{\rm C}$. As a result the sides and angles of the LT are to a very good degree of accuracy scale-independent. This interesting feature of the light quark sector implies that the prediction for $\tilde{\theta}_{\rm u}$ and $\tilde{\theta}_{\rm d}$ from quark mass matrices at any high scale (e.g., $\mu \sim 10^{16}$ GeV) can directly be confronted with the low-scale experimental data. The two symmetry limits discussed above are both not far from the reality, in which the strong hierarchy of quark masses ($m_u\ll m_c \ll m_t$ and $m_d\ll m_s\ll m_b$) has been observed. They will serve as a guide in the subsequent discussions about generic quark mass matrices and their consequences on flavor mixing. \section{Analysis of generic mass matrices} \setcounter{equation}{0} Now we return to the case of three quark families. In the standard model or its extensions which have no flavor-changing right-handed currents, one can always adopt a basis of flavor space in which both the up- and down-type quark mass matrices are hermitian and have vanishing (1,3) and (3,1) elements, as shown in (2.1). Such a basis is of special interest in case of a strong mass hierarchy (as realized by nature), since no explicit mixing between the very massive $t$ (or $b$) quark and the very light $u$ (or $d$) quark is introduced. The mixing can then be regarded as of the ``nearest neighbour'' type \cite{F78}. Thus without loss of generality one may discuss the model-independent properties of flavor mixing and $CP$ violation on the basis of the mass matrices (2.1), i.e., \begin{eqnarray} M_{\rm u} & = & \left ( \matrix{ E_{\rm u} & D_{\rm u} & {\bf 0} \cr D^*_{\rm u} & C_{\rm u} & B_{\rm u} \cr {\bf 0} & B^*_{\rm u} & A_{\rm u} \cr} \right ) \; , \nonumber \\ M_{\rm d} & = & \left ( \matrix{ E_{\rm d} & D_{\rm d} & {\bf 0} \cr D^*_{\rm d} & C_{\rm d} & B_{\rm d} \cr {\bf 0} & B^*_{\rm d} & A_{\rm d} \cr} \right ) \; . \end{eqnarray} The phases of $D_{\rm u,d}$ and $B_{\rm u,d}$ elements are denoted as $\phi_{D_{\rm u,d}}$ and $\phi_{B_{\rm u,d}}$, respectively. The phase differences \begin{eqnarray} \phi_1 & = & \phi_{D_{\rm u}} - \phi_{D_{\rm d}} \; , \nonumber \\ \phi_2 & = & \phi_{B_{\rm u}} - \phi_{B_{\rm d}} \; \end{eqnarray} are the source of $CP$ violation in weak interactions of quarks. It is clear that $M_{\rm u}$ and $M_{\rm d}$ consist totally of twelve parameters. If the hermiticity is not imposed on the arbitrary up and down mass matrices in the standard model, then they can be taken as the full ``nearest-neighbor'' mixing form with texture zeros in the (1,1), (2,2), (1,3) and (3,1) positions \cite{Branco}: \begin{equation} {\cal M}_{\rm q} \; =\; \left (\matrix{ {\bf 0} & {\cal X}_{\rm q} & {\bf 0} \cr {\cal X}'_{\rm q} & {\bf 0} & {\cal Y}_{\rm q} \cr {\bf 0} & {\cal Y}'_{\rm q} & {\cal Z}_{\rm q} \cr } \right ) \; , \end{equation} where $\arg ({\cal X}'_{\rm q}) = \arg ({\cal X}^*_{\rm q})$ and $\arg ({\cal Y}'_{\rm q}) = \arg ({\cal Y}^*_{\rm q})$ (for q = u and d). In this special basis the light quarks are assumed to acquire masses through an interaction with their nearest neighbors. It is straightforward to find that the non-hermitian mass matrices ${\cal M}_{\rm u,d}$ have the same number of free parameters as the hermitian mass matrices $M_{\rm u,d}$, therefore one could be transformed to the other \footnote{If one imposes the hermiticity on (4.3) or the nearest-neighbor mixing on (4.1), then the resultant mass matrices take the particularly simple form which was first proposed and discussed by one of the authors about twenty years ago \cite{F77}.}. In our point of view the hermitian basis (4.1) is more natural and will be adopted in the subsequent discussions. \subsection{Conditions for $CP$ violation} We first discuss the necessary and sufficient conditions for $CP$ violation in the standard electroweak model and clarify some ambiguity associated with this problem in the literature. As the flavor mixing matrix $V$ is obtained from the diagonalization of the mass matrices $M_{\rm u}$ and $M_{\rm d}$, there must be some kind of relation between the parameters of $V$ and $M_{\rm u,d}$. The conditions for $CP$ violation can be counted either at the level of quark mass matrices or at the level of the flavor mixing matrix. One must distinguish between these two different levels. At the level of quark mass matrices it is obvious that $CP$ symmetry will be violated, if and only if there is at least one nontrivial phase difference between $M_{\rm u}$ and $M_{\rm d}$. In other words, ${\rm Im} (M_{{\rm u}ij} M^*_{{\rm d}ij}) \neq 0$ (for $i,j=1,2,3$ and $i<j$) is the necessary and sufficient condition for $CP$ violation in the standard model. If one defines a commutator for $M_{\rm u}$ and $M_{\rm d}$, $[M_{\rm u}, M_{\rm d} ] \equiv {\rm i} ~ {\cal C}$, then it is easy to find \begin{equation} \frac{{\cal C}_{ii}}{2} \; =\; {\rm Im} \left ( M_{{\rm u}ij} M^*_{{\rm d} ij} \right ) ~ + ~ {\rm Im} \left (M_{{\rm u}ik} M^*_{{\rm d} ik} \right ) \; , \end{equation} for $i,j,k=1,2,3$ but $i\neq j\neq k$. Clearly ${\cal C}_{ii} \neq 0$, if $CP$ symmetry is not conserved. Note that $CP$ symmetry would be conserved, if two quarks with the same charge were degenerate in mass eigenvalues. This is well known, but we shall give a proof here. We assume the $i$ and $j$ quarks in the up sector to be degenerate, then they would not be distinguished from each other by any quantum number. Hence any linear combination of the mass eigenstates $|i\rangle$ and $|j\rangle$, e.g., $|i'\rangle$ or $|j'\rangle$ in the form \begin{equation} \left ( \matrix{ i' \cr j' \cr } \right ) \; =\; \left ( \matrix{ \cos\vartheta ~ e^{+{\rm i} \xi} & +\sin \vartheta ~ e^{+{\rm i} \zeta} \cr - \sin \vartheta ~ e^{-{\rm i} \zeta} & \cos \vartheta ~ e^{-{\rm i} \xi} \cr } \right ) \left ( \matrix{ i \cr j \cr} \right ) \; , \end{equation} remains a mass eigenstate. Without loss of any physical content, the elements of $M_{\rm u}$ in the $i$-th and $j$-th lines and rows can be rearranged by three arbitrary (real) parameters $\vartheta$, $\xi$ and $\zeta$. This, together with other known freedoms, allows one to remove all possible phase differences between $M_{\rm u}$ and $M_{\rm d}$, i.e., ${\rm Im} \left ( M_{{\rm u} ij} M^*_{{\rm d} ij} \right ) = 0$ (for $i\neq j$) appears. A similar proof is valid for the down quark sector. We then conclude that a non-degeneracy between the quarks with the same charge is a necessary (but not sufficient) condition for $CP$ violation in the standard model. This condition can be more explicitly written, in terms of the determinant of $\cal C$, as follows \cite{J89}: \begin{equation} {\rm Det} ~ {\cal C} \; =\; -2 {\cal J} \prod_{i < j} \left (\lambda_i - \lambda_j \right ) \prod_{\alpha < \beta} \left (\lambda_\alpha - \lambda_\beta \right ) \; , \end{equation} where $\cal J$ can be found in (2.6), $\lambda_i$ and $\lambda_\alpha$ denote the quark mass eigenvalues, and the subscripts $(i,j)$ and $(\alpha, \beta)$ run over $(u,c,t)$ and $(d,s,b)$ respectively. However, it should be noted that the parameter $\cal J$ itself {\it does} depend on the product of two mass-eigenvalue differences $(\lambda_i - \lambda_j)$ and $(\lambda_\alpha - \lambda_\beta)$, as we shall prove in the next subsection. In this sense $\cal J$ and ${\rm Det} ~ {\cal C}$ contain the same information about $CP$ violation; i.e., the latter is not more fundamental than the former, contrary to popular belief. Now we discuss the condition for $CP$ violation at the level of the flavor mixing matrix. Of course $CP$ symmetry is violated, if $V$ contains a nontrivial complex phase which cannot be removed through the redefinition of quark-field phases. The most appropriate measure of $CP$ violation (due to the unitarity of $V$) is the rephasing-invariant parameter $\cal J$, whose relation with three mixing angles and the $CP$-violating phase has been given in (2.6). Obviously $\cal J$ vanishes if $\varphi =0$ or $\pi$. Note that for $\theta_{\rm u} =0$ or $\pi/2$ the phase parameter $\varphi$ can be removed from $V$. Therefore the resultant flavor mixing matrix is a real $3\times 3$ matrix described by only two rotation angles ($\theta_{\rm d}$ and $\theta$). A similar situation will appear if $\theta_{\rm d} =0$, $\pi/2$ or $\theta =0$, $\pi/2$. The necessary and sufficient condition for $CP$ violation in the standard model is then ${\cal J} \neq 0$ or $\varphi \neq 0$, $\pi$. Since $s_{\rm u} =0$ or $c_{\rm u} =0$ will definitely (though indirectly) lead to $\varphi =0$ or $\pi$, it is unnecessary to count the condition $\theta_{\rm u} \neq 0$ or $\pi/2$ together with $\varphi \neq 0$ or $\pi$. So is the situation for $\theta_{\rm d}$ and $\theta$. In reality quark masses of each sector have been found to perform a clear hierarchy, and all elements of the flavor mixing matrix are nonvanishing \cite{PDG98}. Therefore the realistic condition for $CP$ violation is only associated with the existence of one nontrivial phase parameter in $V$, which in turn requires (at least) one nontrivial phase difference between $M_{\rm u}$ and $M_{\rm d}$. \subsection{Exact analytical result for ${\cal J}$} Let us derive the exact analytical relation between the $CP$-violating parameter ${\cal J}$ and the quark mass-eigenvalue differences. Without loss of generality, we just adopt the basis of flavor space which accommodates the hermitian quark mass matrices $M_{\rm u}$ and $M_{\rm d}$ in (4.1). A basis-independent proof can similarly be carried out for two arbitrary mass matrices $M'_{\rm u}$ and $M'_{\rm d}$, if one starts from the hermitian products $H_{\rm u} \equiv M'_{\rm u} {M'_{\rm u}}^{\dagger}$ and $H_{\rm d} \equiv M'_{\rm d} {M'_{\rm d}}^{\dagger}$ and arranges them to be of the same form as $M_{\rm u}$ and $M_{\rm d}$. This can always be done by appropriately adjusting the fields of right-handed quarks, which are iso-singlets in the standard model. For convenience we decompose $M_{\rm q}$ into $M_{\rm q} = P_{\rm q}^{\dagger} \overline{M}_{\rm q} P_{\rm q}$, where \begin{equation} \overline{M}_{\rm q} \; =\; \left ( \matrix{ E_{\rm q} & |D_{\rm q}| & {\bf 0} \cr |D_{\rm q}| & C_{\rm q} & |B_{\rm q}| \cr {\bf 0} & |B_{\rm q}| & A_{\rm q} \cr} \right ) \; \end{equation} is a real symmetric matrix, and $P_{\rm q} = {\rm Diag} \{ 1, e^{{\rm i}\phi_{D_{\rm q}}}, e^{{\rm i} (\phi_{B_{\rm q}} + \phi_{D_{\rm q}})} \}$ is a diagonal phase matrix. In the following we shall neglect the subscript ``q'', only if there is no necessity to distinguish between the up and down quark sectors. $\overline{M}$ can be diagonalized by use of the orthogonal transformation $O^{\rm T} \overline{M} O = {\rm Diag} \{ \lambda_1, \lambda_2, \lambda_3 \}$, where $\lambda_i$ (for $i=1,2,3$) are quark mass eigenvalues and may be either positive or negative. As a result, we have \begin{eqnarray} \sum_{i=1}^3 \lambda_i & = & A + C + E \; , \nonumber \\ \prod_{i=1}^3 \lambda_i & = & ACE - A |D|^2 - E |B|^2 \; , \nonumber \\ \sum_{i=1}^3 \lambda^2_i & = & A^2 + 2|B|^2 + C^2 + 2|D|^2 + E^2 \; . \end{eqnarray} It is a simple exercise to solve the nine matrix elements of $O$ in terms of the parameters of quark mass matrices. Explicitly, three diagonal elements of $O$ read \footnote{Here and hereafter, the off-diagonal elements $B$ and $D$ are both taken to be nonvanishing. The relevant calculations will somehow be simplified if one of them vanishes.} : \begin{eqnarray} O_{11} & = & \left [ 1 + \left ( \frac{\lambda_1 - E}{|D|} \right )^2 + \left ( \frac{|B|}{|D|} \cdot \frac{\lambda_1 - E}{\lambda_1 - A} \right )^2 \right ]^{-1/2} \; , \nonumber \\ O_{22} & = & \left [ 1 + \left ( \frac{|D|}{\lambda_2 - E} \right )^2 + \left ( \frac{|B|}{\lambda_2 - A} \right )^2 \right ]^{-1/2} \; , \nonumber \\ O_{33} & = & \left [ 1 + \left ( \frac{\lambda_3 - A}{|B|} \right )^2 + \left ( \frac{|D|}{|B|} \cdot \frac{\lambda_3 - A}{\lambda_3 - E} \right )^2 \right ]^{-1/2} \; ; \end{eqnarray} and then six off-diagonal elements of $O$ can be obtained from the relations \begin{eqnarray} O_{2i} & = & \frac{\lambda_i - E}{|D|} O_{1i} \; , \nonumber \\ O_{3i} & = & \frac{|B|}{\lambda_i - A}O_{2i} \; . \end{eqnarray} The flavor mixing matrix turns out to be $V \equiv O^{\rm T} (P_{\rm u} P_{\rm d}^{\dagger}) O_{\rm d}$. More specifically, we have \begin{equation} V_{i\alpha} \; =\; O^{\rm u}_{1i} O^{\rm d}_{1\alpha} ~ + ~ O^{\rm u}_{2i} O^{\rm d}_{2\alpha} e^{{\rm i}\phi_1} ~ + ~ O^{\rm u}_{3i} O^{\rm d}_{3\alpha} e^{{\rm i}(\phi_1 + \phi_2)} \; , \end{equation} where the Latin subscript $i$ and the Greek subscript $\alpha$ run over $(u,c,t)$ and $(d,s,b)$ respectively, and the phase differences $\phi_{1,2}$ have been defined in (4.2). The $CP$-violating parameter $\cal J$ can be calculated from the common imaginary part of nine rephasing invariants of $V$, i.e., ${\cal J} = | {\rm Im} (V_{i\alpha} V_{j\beta} V^*_{i\beta} V^*_{j\alpha}) |$ for $i\neq j$ and $\alpha \neq \beta$ \cite{J85}. With the help of (4.10) and (4.11), one may express $\cal J$ in terms of the parameters of quark mass matrices. After a lengthy calculation, we arrive at the following exact and rephasing-invariant result: \begin{equation} {\cal J} \; = \; \left (\lambda_i - \lambda_j \right ) \left ( \lambda_\alpha - \lambda_\beta \right ) ~ f^{ij}_{\alpha\beta} \; , \end{equation} where \begin{eqnarray} f^{ij}_{\alpha\beta} & = & \frac{\left ( O^{\rm u}_{1i} O^{\rm u}_{1j} O^{\rm d}_{1\alpha} O^{\rm d}_{1\beta} \right )^2}{|D_{\rm u} D_{\rm d}|} ~ \left [ T_1 \sin \phi_1 ~ + ~ T_2 \sin\phi_2 ~ + ~ T_3 \sin (\phi_1 + \phi_2) \right . \nonumber \\ & & \left . + ~ T_4 \sin (\phi_1 - \phi_2) ~ + ~ T_5 \sin (2\phi_1 + \phi_2) ~ + ~ T_6 \sin (\phi_1 + 2\phi_2) \right ] \; . \end{eqnarray} The expressions of $T_i$ (for $i=1,2,\cdot\cdot\cdot ,6$) are listed in Appendix A. One can see that $\cal J$ depends definitely on the mass-eigenvalue differences $(\lambda_i - \lambda_j)$ of the up sector and $(\lambda_\alpha - \lambda_\beta)$ of the down sector. Since the subscripts $(i,j)$ and $(\alpha, \beta)$ run over the corresponding quarks $(u,c,t)$ and $(d,s,b)$, $\cal J$ would vanish if any two quarks with the same charge were degenerate in mass eigenvalues. Therefore $\cal J$ carries the same information about $CP$ violation as ${\rm Det} ~\cal C$ in (4.6). Two remarks are in order. (a) Note that a phase combination in the form of $\sin (2\phi_1)$, $\sin (2\phi_2)$ or $\sin 2(\phi_1 + \phi_2)$ has no contribution to $\cal J$. The reason is simply that in $\cal J$ the terms associated with $e^{+{\rm i} 2\phi_1}$ and $e^{-{\rm i} 2\phi_1}$ have the same magnitude and cancel each other. So it the case for the terms associated with $e^{\pm {\rm i} 2\phi_2}$ and $e^{\pm {\rm i} 2 (\phi_1 + \phi_2)}$. Once the hierarchy of $M_{\rm u}$ and $M_{\rm d}$ is taken into account, the magnitude of $\cal J$ is expected to be dominated by the term proportional to $\sin\phi_1$ (see the next subsection). (b) The dependence of $\cal J$ on the product of two mass-eigenvalue differences $(\lambda_i - \lambda_j)$ and $(\lambda_\alpha - \lambda_\beta)$ is indeed a basis-independent result, although we have obtained it in a specific basis of flavor space for the quark mass matrices. The basis-independent calculation of $\cal J$ is straightforward, as we have mentioned above, if one starts from $H_{\rm u} = M'_{\rm u} {M'_{\rm u}}^{\dagger}$ and $H_{\rm d} = M'_{\rm d} {M'_{\rm d}}^{\dagger}$ for arbitrary $M'_{\rm u}$ and $M'_{\rm d}$. In this case it is easy to find \begin{equation} {\cal J} \; = \; \left (\lambda^2_i - \lambda^2_j \right ) \left ( \lambda^2_\alpha - \lambda^2_\beta \right ) ~ F^{ij}_{\alpha\beta}\; , \end{equation} where $F^{ij}_{\alpha\beta}$ can be read off from $f^{ij}_{\alpha\beta}$ through the replacements of matrix elements from $M_{\rm u,d}$ to $H_{\rm u,d}$. Of course the results in (4.12) and (4.14) essentially have the same physical meaning. In reality it is known that quark masses show a strong hierarchy in either sector. Therefore $\cal J$ vanishes if and only if both $\phi_1 =0$ (or $\pi$) and $\phi_2 =0$ (or $\pi$) hold. The necessary and sufficient condition for $CP$ violation in the standard model is trivially $\varphi \neq 0$ or $\pi$. \subsection{Flavor mixing angles and the $CP$-violating phase} Now let us take the hierarchy of quark masses ($|\lambda_1| \ll |\lambda_2| \ll |\lambda_3|$) into account for the hermitian mass matrices $M_{\rm u}$ and $M_{\rm d}$ in (4.1). This implies $|A_{\rm q}| \gg |B_{\rm q}|, |C_{\rm q}| \gg |D_{\rm q}|, |E_{\rm q}|$ for both sectors. Our purpose is to calculate the mixing angles ($\theta_{\rm u}$, $\theta_{\rm d}$ and $\theta$) and the $CP$-violating phase ($\varphi$) in an analytically exact way. Certainly the orthogonal matrix $O$ used to diagonalize $\overline{M}$ in (4.7) can further be written as a product of three matrices $R_{12}$, $R_{23}$ and $R_{31}$, which describe simple rotations in the 1--2, 2--3 and 3--1 planes respectively: \begin{eqnarray} R_{12}(\omega) & = & \left ( \matrix{ c_{\omega} & s_{\omega} & 0 \cr - s_{\omega} & c_{\omega} & 0 \cr 0 & 0 & 1 \cr} \right ) \; , \nonumber \\ \nonumber \\ R_{23}(\sigma) & = & \left ( \matrix{ 1 & 0 & 0 \cr 0 & c_{\sigma} & s_{\sigma} \cr 0 & - s_{\sigma} & c_{\sigma} \cr} \right ) \; , \nonumber \\ \nonumber \\ R_{31}(\tau) & = & \left ( \matrix{ c_{\tau} & 0 & s_{\tau} \cr 0 & 1 & 0 \cr - s_{\tau} & 0 & c_{\tau} \cr} \right ) \; , \end{eqnarray} where $s_{\omega} \equiv \sin \omega$, $c_{\omega} \equiv \cos \omega$, etc. Taking $O= R_{13}R_{23}R_{12}$ for example, we arrive at \begin{eqnarray} \tan\tau & = & \frac{|D|}{|B|} \cdot \frac{(E-\lambda_1) + (C-\lambda_2)}{\lambda_3 - E} \; , \nonumber \\ \tan\sigma & = & \frac{|D|}{|B|} \cdot \frac{(E-\lambda_1) + (C-\lambda_2)}{\sqrt{|D|^2 + |B|^2 \tan^2\tau}} \; , \nonumber \\ \tan\omega & = & \frac{|D|}{\lambda_2 - E} \cdot \frac{c^{~}_{\sigma}}{c^{~}_{\tau}} ~ + ~ s^{~}_{\sigma} \tan\tau \; . \end{eqnarray} In view of the hierarchy of quark mass matrices, we find that the magnitude of $\tan\tau$ is highly suppressed, leading to an excellent approximation $\tau \approx 0^{\circ}$. Thus the matrix $O$ is dominantly described by only two rotation angles, $\omega$ and $\sigma$. This is naturally expected, since in lowest order the diagonalization of the mass matrices is provided by a rotation in the 2--3 plane and a rotation in the 1--2 plane. Due to the vanishing of the (1,3) and (3,1) matrix elements, a rotation in the 3--1 plane is essentially unnecessary. This approximation has been used to derive an interesting parametrization of the flavor mixing matrix \cite{F79,Froggatt79,Hall}, whose form is quite similar to that given in (2.2). Note, however, that the exact parametrization (2.2) is indeed independent of the above approximation, because the contribution from rotation matrices $R^{\rm u}_{13}$ (up) and $R^{\rm d}_{13}$ (down) to the flavor mixing matrix can always be absorbed by redefining its three overall mixing angles. Since the concrete calculation of those mixing angles from $R^{\rm u,d}_{12}$, $R^{\rm u,d}_{23}$ and $R^{\rm u,d}_{31}$ is rather complicated and less instructive (see, e.g., Ref. \cite{Rasin}), we shall subsequently follow a different and more straightforward procedure towards the same goal. We make use of the expression of $V$ given in (4.11). The parametrization of $V$ in terms of three mixing angles ($\theta_{\rm u}$, $\theta_{\rm d}$, $\theta$) and one $CP$-violating phase ($\varphi$) has been shown in (2.2). To link these four parameters with the parameters of quark mass matrices in a concise way, we first define four dimensionless quantities: \begin{eqnarray} X_{\rm u} & \equiv & \left | \frac{|D_{\rm u}|}{\lambda^{\rm u}_1 - E_{\rm u}} \cdot \frac{|D_{\rm d}| \left (\lambda^{\rm d}_3 - A_{\rm d} \right )} {|B_{\rm d}| \left (\lambda^{\rm d}_3 - E_{\rm d} \right )} ~ + ~ \frac{\lambda^{\rm d}_3 - A_{\rm d}}{|B_{\rm d}|} e^{{\rm i}\phi_1} ~ + ~ \frac{|B_{\rm u}|}{\lambda^{\rm u}_1 - A_{\rm u}} e^{{\rm i}(\phi_1 + \phi_2)} \right | \; , \nonumber \\ Y_{\rm u} & \equiv & \left | \frac{|D_{\rm u}|}{\lambda^{\rm u}_2 - E_{\rm u}} \cdot \frac{|D_{\rm d}| \left (\lambda^{\rm d}_3 - A_{\rm d} \right )} {|B_{\rm d}| \left (\lambda^{\rm d}_3 - E_{\rm d} \right )} ~ + ~ \frac{\lambda^{\rm d}_3 - A_{\rm d}}{|B_{\rm d}|} e^{{\rm i}\phi_1} ~ + ~ \frac{|B_{\rm u}|}{\lambda^{\rm u}_2 - A_{\rm u}} e^{{\rm i}(\phi_1 + \phi_2)} \right | \; ; \end{eqnarray} and $(X_{\rm d}, Y_{\rm d})$ can directly be obtained from $(X_{\rm u}, Y_{\rm u})$ through the subscript exchange ${\rm u} \leftrightarrow {\rm d}$ in (4.17). After a lengthy but straightforward calculation, we arrive at \begin{eqnarray} \tan\theta_{\rm u} & = & \frac{O^{\rm u}_{21}}{O^{\rm u}_{22}} \cdot \frac{X_{\rm u}}{Y_{\rm u}} \; , \nonumber \\ \tan\theta_{\rm d} & = & \frac{O^{\rm d}_{21}}{O^{\rm d}_{22}} \cdot \frac{X_{\rm d}}{Y_{\rm d}} \; , \end{eqnarray} and \begin{eqnarray} \sin\theta & = & \left [ (O^{\rm u}_{21} )^2 X^2_{\rm u} ~ + ~ (O^{\rm u}_{22} )^2 Y^2_{\rm u} \right ]^{1/2} O^{\rm d}_{33} \; , \nonumber \\ & = & \left [ (O^{\rm d}_{21} )^2 X^2_{\rm d} ~ + ~ (O^{\rm d}_{22} )^2 Y^2_{\rm d} \right ]^{1/2} O^{\rm u}_{33} \; , \end{eqnarray} where $O_{21}$, $O_{22}$ and $O_{33}$ for up and down sectors have been given in (4.9) and (4.10). Also an indirect relation between $\varphi$ and $\phi_{1,2}$ can be obtained as follows: \begin{equation} \cos \varphi \; =\; \frac{s^2_{\rm u} c^2_{\rm d} c^2 + c^2_{\rm u} s^2_{\rm d} - |V_{us}|^2}{2 s_{\rm u} c_{\rm u} s_{\rm d} c_{\rm d} c} \; , \end{equation} where \begin{equation} |V_{us}| \; =\; O^{\rm u}_{11} O^{\rm d}_{22} \left | \frac{|D_{\rm d}|}{\lambda^{\rm d}_2 - E_{\rm d}} + \frac{\lambda^{\rm u}_1 - E_{\rm u}}{|D_{\rm u}|} e^{{\rm i} \phi_1} \left ( 1 + \frac{|B_{\rm u}|}{\lambda^{\rm u}_1 - A_{\rm u}} \cdot \frac{|B_{\rm d}|}{\lambda^{\rm d}_2 - A_{\rm d}} e^{{\rm i} \phi_2} \right ) \right | \; . \end{equation} If the hierarchies of the matrix elements and mass eigenvalues of $M_{\rm u, d}$ are taken into account, one can see that the effect of $\phi_2$ on $|V_{us}|$ is strongly suppressed and thus negligible. Fitting $|V_{us}|$ with current data should essentially determine the magnitude of $\phi_1$. Note also that the terms associated with $\phi_1$ and $\phi_2$ may primarily be cancelled in the ratios $X_{\rm u}/Y_{\rm u}$ and $X_{\rm d}/Y_{\rm d}$ due to the hierarchical structures of $M_{\rm u}$ and $M_{\rm d}$, hence the dependence of $\theta_{\rm u}$ and $\theta_{\rm d}$ on $\phi_{1,2}$ could be negligible in the leading order approximation. Although the mixing angle $\theta$ may be sensitive to $\phi_1$ and $\phi_2$ (or one of them), its smallness indicated by current data makes the factor $\cos\theta$ in the denominator of $\cos\varphi$ completely negligible. As a result, (4.20) and (4.21) imply that the $CP$-violating phase $\varphi$ depends dominantly on $\phi_1$ through $|V_{us}|$, unless the magnitude of $\phi_1$ is very small. Without fine-tuning, we find that a delicate numerical analysis does support the argument made here, i.e., $\phi_2$ plays a negligible role for $CP$ violation in $V$, because of the hierarchy of quark masses. The observed $CP$ violation is linked primarily to the phases in the (1,2) and (2,1) elements of the quark mass matrices. \section{A realistic texture of mass matrices} \setcounter{equation}{0} In order to get definite predictions for the flavor mixing angles and $CP$ violation, we proceed to specify the general hermitian mass matrices in (2.1) or (4.1) by taking $E_{\rm q} =0$: \begin{equation} M_{\rm q} \; = \; \left ( \matrix{ {\bf 0} & D_{\rm q} & {\bf 0} \cr D^*_{\rm q} & C_{\rm q} & B_{\rm q} \cr {\bf 0} & B^*_{\rm q} & A_{\rm q} \cr } \right ) \; . \end{equation} In case of two quark families, this is just the form taken for $\tilde{M}_{\rm d}$ in (3.7). As remarked above, the texture zeros in (1,3) and (3,1) positions can always be arranged. Thus the physical constraint is as follows: in the flavor basis in which (1,3) and (3,1) elements of $M_{\rm u,d}$ vanish, the (1,1) element of $M_{\rm u,d}$ vanishes as well. This can strictly be true only at a particular energy scale. The vanishing of the (1,1) element can be viewed as a result of an underlying flavor symmetry, which may either be discrete or continuous. In the literature a number of such possibilities have been discussed (see, e.g., Refs. \cite{F79} -- \cite{Zero4}). Here we shall not discuss further details in this respect, but concentrate on the phenomenological consequences of such a texture pattern. It is particularly interesting that some predictions of this ansatz for the mixing angles and the unitarity triangle are approximately independent of the renormalization-group effects, therefore a specification of the energy scale at which the texture of $M_{\rm u,d}$ holds is unnecessary for our purpose. We believe that $M_{\rm q}$ given in (5.1) is a realistic candidate for the quark mass matrices of a (yet unknown) fundamental theory responsible for fermion mass generation and $CP$ violation, and we shall make some further speculations about this point at the end of this paper. \subsection{Flavor mixing angles} We take $C_{\rm q} \neq 0$ and define $|B_{\rm q}|/C_{\rm q} \equiv r^{~}_{\rm q}$ for each quark sector \footnote{Note that the special condition $C_{\rm q}/A_{\rm q} = |D_{\rm q}/B_{\rm q}|^2$ has been imposed on $M_{\rm q}$ in a recent paper \cite{Froggatt}. It leads to vanishing flavor mixing among all three quark families in the chiral limit of $u$- and $d$-quark masses, and in turn requires a kind of correlation between the flavor mixing angles $\theta_{\rm u}$, $\theta_{\rm d}$ and $\theta$. This unusual feature is apparently in conflict with our arguments made in section 3 (see also Ref. \cite{F87}). Beyond the chiral symmetry limit the aforementioned condition is equivalent to taking $|r_{\rm u}| \approx \sqrt{m_um_t}/m_c$ and $|r_{\rm d}| \approx \sqrt{m_dm_b}/m_s$ in our case. Clearly both $r_{\rm u}$ and $r_{\rm d}$ are of $O(1)$ in magnitude, consistent with the common expectation.}. The magnitude of $r^{~}_{\rm q}$ is expected to be of $O(1)$. The parameters $A_{\rm q}$, $|B_{\rm q}|$, $C_{\rm q}$ and $|D_{\rm q}|$ in (5.1) can be expressed in terms of the quark mass eigenvalues and $r^{~}_{\rm q}$. Applying such results to the general formulas listed in (4.17) -- (4.19), we get three mixing angles of $V$ as follows: \begin{eqnarray} \tan\theta_{\rm u} & = & \sqrt{\frac{m_u}{m_c}} ~ \left ( 1 + \Delta_{\rm u} \right ) \; , \nonumber \\ \tan\theta_{\rm d} & = & \sqrt{\frac{m_d}{m_s}} ~ \left ( 1 + \Delta_{\rm d} \right ) \; , \nonumber \\ \sin\theta & = & \left | r_{\rm d} \frac{m_s}{m_b} \left (1 - \delta_{\rm d} \right ) ~ - ~ r_{\rm u} \frac{m_c}{m_t} \left ( 1 - \delta_{\rm u} \right ) e^{{\rm i}\phi_2} \right | \; , \end{eqnarray} where the next-to-leading order corrections read \begin{eqnarray} \Delta_{\rm u} & = & \sqrt{\frac{m_c m_d}{m_u m_s}} ~ \frac{m_s}{m_b} ~ \left | {\rm Re} \left [ e^{{\rm i}\phi_1} ~ - ~ \frac{r_{\rm u}}{r_{\rm d}} \cdot \frac{m_c m_b}{m_t m_s} e^{{\rm i}(\phi_1 + \phi_2)} \right ]^{-1} \right | \; , \nonumber \\ \Delta_{\rm d} & = & \sqrt{\frac{m_u m_s}{m_c m_d}} ~ \frac{m_c}{m_t} ~ \left | {\rm Re} \left [ e^{{\rm i}\phi_1} ~ - ~ \frac{r_{\rm d}}{r_{\rm u}} \cdot \frac{m_t m_s}{m_c m_b} e^{{\rm i}(\phi_1 + \phi_2)} \right ]^{-1} \right | \; ; \end{eqnarray} and \begin{eqnarray} \delta_{\rm u} & = & \frac{m_u}{m_c} ~ + \left ( 1 + r^2_{\rm u} \right ) \frac{m_c}{m_t} \; , \nonumber \\ \delta_{\rm d} & = & \frac{m_d}{m_s} ~ + \left ( 1 + r^2_{\rm d} \right ) \frac{m_s}{m_b} \; . \end{eqnarray} Clearly the result for $\hat{\delta}_{\rm u,d}$ in (3.4) can be reproduced from $\delta_{\rm u,d}$ in (5.4), if one takes the chiral limit $m_u \rightarrow 0$, $m_d \rightarrow 0$. From (5.2) we also observe that the phase $\phi_2$ is only associated with the small quantity $m_c/m_t$ in $\sin\theta$. To get the relationship between $\varphi$ and $\phi_1$ or $\phi_2$, we first calculate $|V_{us}|$ from the quark mass matrices by use of (4.21). It turns out that \begin{equation} |V_{us}| \; = \; \left (1 -\frac{1}{2} \frac{m_u}{m_c} - \frac{1}{2} \frac{m_d}{m_s} \right ) \left | \sqrt{\frac{m_d}{m_s}} ~ - ~ \sqrt{\frac{m_u}{m_c}} ~ e^{{\rm i}\phi_1} \right | \; \end{equation} in the next-to-leading order approximation. Note that this result can also be achieved from (3.8) and (3.9), which were obtained in the heavy quark limit. Confronting (5.5) with current data on $|V_{us}|$ leads to the result $\phi_1 \sim 90^{\circ}$, as we have discussed before. Therefore $\cos\phi_1$ is expected to be a small quantity. Then we use (4.20) together with (5.2) and (5.5) to calculate $\cos\varphi$. In the same order approximation, we arrive at \begin{equation} \cos\varphi \; =\; \sqrt{\frac{m_u m_s}{m_c m_d}} ~ \Delta_{\rm u} ~ + ~ \sqrt{\frac{m_c m_d}{m_u m_s}} ~ \Delta_{\rm d} ~ + ~ \left (1 - \Delta_{\rm u} - \Delta_{\rm d} \right ) \cos\phi_1 \; . \end{equation} The contribution of $\phi_2$ to $\varphi$ is substantially suppressed at this level of accuracy. For simplicity, we proceed by taking $r_{\rm u} = r_{\rm d} \equiv r$, which holds in some models with natural flavor symmetries \cite{Democracy}. Then $\sin\theta$ becomes proportional to a universal parameter $|r|$. In view of the fact $m_s/m_b \sim O(10) ~ m_c/m_t$, we find that the result in (5.3) can be simplified as \begin{eqnarray} \Delta_{\rm u} & = & \sqrt{\frac{m_c m_d}{m_u m_s}} ~ \frac{m_s}{m_b} ~ \cos\phi_1 \; , \nonumber \\ \Delta_{\rm d} & = & 0 \; . \end{eqnarray} Also the relation between $\varphi$ and $\phi_1$ in (5.6) is simplified to \begin{equation} \cos\varphi \; =\; \left (1 + \frac{m_s}{m_b} \right ) \cos\phi_1 \; . \end{equation} As $m_s/m_b \sim 4\%$, it becomes apparent that $\varphi \approx \phi_1$ is a good approximation. Note that $\phi_1 = \varphi$ holds exactly in the heavy quark limit, in which $\varphi$ has been denoted as $\tilde{\varphi}$ (see (3.5) as well as Fig. 1). The equality $\phi_1 = \tilde{\varphi}$ follows, i.e., both stand for the phase difference between the mass matrix elements $D_{\rm u}$ and $D_{\rm d}$. Following (4.12) we evaluate the dependence of the $CP$-violating measurable $\cal J$ on $\phi_1$, $\phi_2$ and their various combinations. The results for six coefficients of $f^{ij}_{\alpha\beta}$ are listed in Appendix B. We confirm that the magnitude of $\cal J$ is dominated by the $\sin\phi_1$ term and receives one-order smaller corrections from the $\sin (\phi_1 \pm \phi_2)$ terms. As a result, \begin{equation} {\cal J} \; \approx \; |r|^2 \sqrt{\frac{m_u}{m_c}} \sqrt{\frac{m_d}{m_s}} \left (\frac{m_s}{m_b}\right )^2 \sin\phi_1 \; \end{equation} holds to a good degree of accuracy. Clearly ${\cal J} \sim O(10^{-5}) \times \sin\phi_1$ with $\sin\phi_1 \sim 1$ is favored by current data. The result of $\cal J$ in (5.9) might give the impression that $CP$ violation is absent if either $m_u$ or $m_d$ vanishes. This is not exactly true, however. If we set $m_u=0$, $\cal J$ is not zero, but it becomes dependent on $\sin \phi_2$ with a factor which is about two orders of magnitude smaller (i.e., of order $10^{-7}$): \begin{equation} {\cal J} \; \approx \; |r|^2 ~ \frac{m_c}{m_t} \cdot \frac{m_d}{m_s} \left (\frac{m_s}{m_b} \right )^2 \sin\phi_2 \; . \end{equation} Certainly this possibility is already ruled out by experimental data. Note also that the model predicts \begin{eqnarray} \tan\theta_{\rm u} \; =\; \left | \frac{V_{ub}}{V_{cb}} \right | & = & \sqrt{\frac{m_u}{m_c}} ~ \left (1 + \Delta_{\rm u} \right ) \; , \nonumber \\ \tan\theta_{\rm d} \; =\; \left | \frac{V_{td}}{V_{ts}} \right | & = & \sqrt{\frac{m_d}{m_s}} ~ \left (1 + \Delta_{\rm d} \right ) \; , \end{eqnarray} a result obtained first by one of us from a more specific pattern of quark mass matrices \cite{F78}. In $B$-meson physics, $|V_{ub}/V_{cb}|$ can be determined from the ratio of the decay rate of $B\rightarrow (\pi, \rho) l \nu^{~}_l$ to that of $B\rightarrow D^* l\nu^{~}_l$; and $|V_{td}/V_{ts}|$ can be extracted from the ratio of the rate of $B^0_d$-$\bar{B}^0_d$ mixing to that of $B^0_s$-$\bar{B}^0_s$ mixing. \subsection{The unitarity triangle} \begin{figure}[t] \begin{picture}(400,160)(130,210) \put(300,300){\line(1,0){150}} \put(300,300.5){\line(1,0){150}} \put(370,285.5){\makebox(0,0){$|V_{cd}|$}} \put(300,300){\line(1,3){21.5}} \put(300,300.5){\line(1,3){21.5}} \put(300,299.5){\line(1,3){21.5}} \put(292,333){\makebox(0,0){$S_{\rm u}$}} \put(450,300){\line(-2,1){128}} \put(450,300.5){\line(-2,1){128}} \put(395,343.5){\makebox(0,0){$S_{\rm d}$}} \put(315,310){\makebox(0,0){$\gamma$}} \put(408,309){\makebox(0,0){$\beta$}} \put(328,350){\makebox(0,0){$\alpha$}} \end{picture} \vspace{-2.6cm} \caption{The rescaled unitarity triangle (UT) in the complex plane.} \end{figure} We are now in a position to calculate the unitarity triangle (UT) of quark flavor mixing defined in (2.8), whose three inner angles are denoted as $\alpha$, $\beta$ and $\gamma$ in (2.9). Note that three sides of the unitarity triangle can be rescaled by $V_{cb}^*$ (see Fig. 2 for illustration). The resultant triangle reads \begin{equation} |V_{cd}| \; =\; \left | S_{\rm d} ~ -~ S_{\rm u} ~ e^{-{\rm i}\alpha} \right | \; , \end{equation} where $S_{\rm u} = |V_{ub}V_{ud}/V_{cb}|$ and $S_{\rm d} = |V_{tb}V_{td}/V_{cb}|$. After some calculations $S_{\rm u}$, $S_{\rm d}$ and $\alpha$ are obtained from the above quark mass texture in the next-to-leading order approximation: \begin{eqnarray} S_{\rm u} & = & \sqrt{\frac{m_u}{m_c}} \left ( 1 - \frac{1}{2} \frac{m_u}{m_c} - \frac{1}{2} \frac{m_d}{m_s} + \sqrt{\frac{m_c m_d}{m_u m_s}} ~ \frac{m_s}{m_b} ~ \cos\phi_1 + \sqrt{\frac{m_u m_d}{m_c m_s}} ~ \cos\phi_1 \right ) \; , \nonumber \\ S_{\rm d} & = & \sqrt{\frac{m_d}{m_s}} \left ( 1 + \frac{1}{2} \frac{m_u}{m_c} - \frac{1}{2} \frac{m_d}{m_s} \right ) \; ; \end{eqnarray} and \begin{equation} \sin\alpha \; = \; \left ( 1 - \sqrt{\frac{m_u m_d}{m_c m_s}} ~ \cos\phi_1 \right ) \sin\phi_1 \; . \end{equation} A comparison of the rescaled UT in Fig. 2 with the LT in Fig. 1, which is obtained in the heavy quark limit, is interesting. We find \begin{eqnarray} \frac{S_{\rm u} - R_{\rm u}}{R_{\rm u}} & = & \left ( 1 + \frac{m_c m_s}{m_u m_b} \right ) \sqrt{\frac{m_u m_d}{m_c m_s}} ~ \cos\tilde{\varphi} \; , \nonumber \\ \frac{S_{\rm d} - R_{\rm d}}{R_{\rm d}} & = & \frac{m_u}{m_c} \; , \nonumber \\ \frac{\sin\alpha - \sin\tilde{\varphi}}{\sin\tilde{\varphi}} & = & - \sqrt{\frac{m_u m_d}{m_c m_s}} ~ \cos\tilde{\varphi} \; , \end{eqnarray} which are of order $15\% \cos\tilde{\varphi}$, $0.4\%$ and $1.4\% \cos\tilde{\varphi}$, respectively. Obviously $R_{\rm d} \approx S_{\rm d}$ is an excellent approximation, and $\alpha \approx \tilde{\varphi} \approx \varphi$ is a good approximation. As $\varphi$ (or $\tilde{\varphi}$) is expected to be close to $90^{\circ}$, $R_{\rm u} \approx S_{\rm u}$ should also be accurate enough in the next-to-leading order estimation. Therefore the light-quark triangle is essentially {\it congruent with} the rescaled unitarity triangle! This result has two straightforward implications: first, $CP$ violation is an effect arising primarily from the light quark sector; second, the $CP$-violating observables ($\alpha$, $\beta$, $\gamma$) can be predicted in terms of the light quark masses and the phase difference between up and down mass matrices \cite{FX95}. If we use the value of $|V_{cd}|$, which is expected to equal $|V_{us}|$ within the $0.1\%$ error bar \cite{Xing96}, then all three angles of the unitarity triangle can be calculated in terms of $m_u/m_c$, $m_d/m_s$ and $|V_{cd}|$ to a good degree of accuracy. The three angles of the UT ($\alpha$, $\beta$ and $\gamma$) will be well determined at the $B$-meson factories, e.g., from the $CP$ asymmetries in $B_d\rightarrow \pi^+\pi^-$, $B_d\rightarrow J/\psi K_S$ and $B^{\pm}_u\rightarrow (D^0, \bar{D}^0) + K^{(*)\pm}$ decays \cite{BB}. The characteristic measurable quantities are $\sin (2\alpha)$, $\sin (2\beta)$ and $\sin^2\gamma$, respectively. For the purpose of illustration, we typically take $|V_{us}| = |V_{cd}| =0.22$, $m_u/m_c =0.0056$, $m_d/m_s = 0.045$ and $m_s/m_b = 0.033$ to calculate these three $CP$-violating parameters from the LT and from the rescaled UT separately. Both approaches lead to $\alpha \approx 90^{\circ}$, $\beta \approx 20^{\circ}$ and $\gamma \approx 70^{\circ}$, which are in good agreement with the results obtained from the standard analysis of current data on $|V_{ub}/V_{cb}|$, $\epsilon^{~}_K$, $B^0_d$-$\bar{B}^0_d$ mixing and $B^0_s$-$\bar{B}^0_s$ mixing \cite{PRS98}. Note that among three $CP$-violating observables only $\sin (2\beta)$ is remarkably sensitive to the value of $m_u/m_c$, which involves quite large uncertainty (e.g., $\sin (2\beta)$ may change from $0.4$ to $0.8$ if $m_u/m_c$ varies in the range $0.002 - 0.01$). For this reason we emphasize again that the numbers given above can only serve as an illustration. A more reliable determination of the quark mass values is crucial, in order to test the ans$\rm\ddot{a}$tze of quark mass matrices in a numerically decisive way \footnote{A similar remark based on more delicate numerical analysis has also be made in Ref. \cite{Barbieri}.}. It is also worth mentioning that the result $\tan\theta_{\rm d} = \sqrt{m_d/m_s}$ is particularly interesting for the mixing rates of $B^0_d$-$\bar{B}^0_d$ and $B^0_s$-$\bar{B}^0_s$ systems, measured by $x_{\rm d}$ and $x_{\rm s}$ respectively \cite{PDG98}. The ratio $x_{\rm s} /x_{\rm d}$ amounts to $|V_{ts}/V_{td}|^2 = \tan^{-2} \theta_{\rm d}$ multiplied by a factor $\chi_{\rm su(3)} = 1.45 \pm 0.13$, which reflects the $\rm SU(3)_{\rm flavor}$ symmetry breaking effects \cite{Marti}. As $x_{\rm d} = 0.723 \pm 0.032$ has been well determined \cite{PDG98}, the prediction for the value of $x_{\rm s}$ is \begin{equation} x_{\rm s} \; =\; x_{\rm d} ~ \chi_{\rm su(3)} ~ \frac{m_s}{m_d} \; =\; 19.8 \pm 3.5 \; , \end{equation} where $m_s/m_d = 18.9 \pm 0.8$, obtained from the chiral perturbation theory \cite{Leut}, has been used. This result is certainly consistent with the present experimental bound on $x_{\rm s}$, i.e., $x_{\rm s} > 14.0$ at the $95\%$ confidence level \cite{PDG98}. A measurement of $x_{\rm s} \sim 20$ may be realized at the forthcoming HERA-$B$ and LHC-$B$ experiments. \subsection{Comparison with the Ramond-Roberts-Ross patterns} The quark mass matrices $M_{\rm u}$ and $M_{\rm d}$ given in (5.1) have parallel structures with four texture zeros (here a pair of off-diagonal texture zeros are counted as one zero due to the hermiticity of $M_{\rm u}$ and $M_{\rm d}$). Giving up the parallelism between the structures of $M_{\rm u}$ and $M_{\rm d}$, Ramond, Roberts and Ross (RRR) have found that there exist five phenomenologically allowed patterns of quark mass matrices -- each of them has five texture zeros \cite{RRR}, as listed in Table 1. The RRR patterns I, II or IV can be formally regarded as a special case of our four-texture-zero pattern (5.1), with $B_{\rm u} =0$, $C_{\rm u} =0$ or $B_{\rm d} =0$, respectively. Note that $M_{\rm u}$ of the RRR pattern III or V has nonvanishing (1,3) and (3,1) elements, therefore these two patterns are essentially different from the mass matrices assumed in (4.1) or (5.1). As a comparison, here we make some brief comments on consequences of the RRR patterns on the flavor mixing angles $\theta$, $\theta_{\rm u}$ and $\theta_{\rm d}$. \scriptsize \begin{table}[t] \caption{Five RRR patterns of quark mass matrices.} \vspace{0.2cm} \begin{center} \begin{tabular}{ccccccc}\hline\hline Pattern & I & II & III & IV & V \\ \hline \\ $M_{\rm u}$ & $\left ( \matrix{ 0 & D_{\rm u} & 0 \cr D^*_{\rm u} & C_{\rm u} & 0 \cr 0 & 0 & A_{\rm u} \cr } \right )$ & $\left ( \matrix{ 0 & D_{\rm u} & 0 \cr D^*_{\rm u} & 0 & B_{\rm u} \cr 0 & B^*_{\rm u} & A_{\rm u} \cr } \right )$ & $\left ( \matrix{ 0 & 0 & F_{\rm u} \cr 0 & C_{\rm u} & 0 \cr F^*_{\rm u} & 0 & A_{\rm u} \cr } \right )$ & $\left ( \matrix{ 0 & D_{\rm u} & 0 \cr D^*_{\rm u} & C_{\rm u} & B_{\rm u} \cr 0 & B^*_{\rm u} & A_{\rm u} \cr } \right )$ & $\left ( \matrix{ 0 & 0 & F_{\rm u} \cr 0 & C_{\rm u} & B_{\rm u} \cr F^*_{\rm u} & B_{\rm u}^* & A_{\rm u} \cr } \right )$ \\ \\ $M_{\rm d}$ & $\left ( \matrix{ 0 & D_{\rm d} & 0 \cr D^*_{\rm d} & C_{\rm d} & B_{\rm d} \cr 0 & B^*_{\rm d} & A_{\rm d} \cr } \right )$ & $\left ( \matrix{ 0 & D_{\rm d} & 0 \cr D^*_{\rm d} & C_{\rm d} & B_{\rm d} \cr 0 & B^*_{\rm d} & A_{\rm d} \cr } \right )$ & $\left ( \matrix{ 0 & D_{\rm d} & 0 \cr D^*_{\rm d} & C_{\rm d} & B_{\rm d} \cr 0 & B^*_{\rm d} & A_{\rm d} \cr } \right )$ & $\left ( \matrix{ 0 & D_{\rm d} & 0 \cr D_{\rm d}^* & C_{\rm d} & 0 \cr 0 & 0 & A_{\rm d} \cr } \right )$ & $\left ( \matrix{ 0 & D_{\rm d} & 0 \cr D^*_{\rm d} & C_{\rm d} & 0 \cr 0 & 0 & A_{\rm d} \cr } \right )$ \\ \\ \hline \hline \end{tabular} \end{center} \end{table} \normalsize (a) For the RRR pattern I, the magnitude of $\sin\theta$ is governed by the (2,3) and (3,2) elements of $M_{\rm d}$. Therfore $|B_{\rm d}| \sim |C_{\rm d}|$ is expected, to result in $\sin\theta \sim m_s/m_b$. In the leading order approximation, $\tan\theta_{\rm u} = \sqrt{m_u/m_c}$ and $\tan\theta_{\rm d} = \sqrt{m_d/m_s}$ hold. The next-to-leading order corrections to these two quantities are almost indistinguishable from those obtained in (5.3) and (5.7) for our four-texture-zero ansatz. (b) The mass matrix $M_{\rm u}$ of the RRR pattern II takes the well-known form suggested originally by one of us in Refs. \cite{F79,F77}. To reproduce the experimental value of $\sin\theta$, the possibility $|B_{\rm d}| \gg |C_{\rm d}|$ has been abandoned and the condition $|B_{\rm d}| \sim |C_{\rm d}|$ is required. However, significant cancellation between the term proportional to $\sqrt{m_c/m_t}$ (from $M_{\rm u}$) \cite{F79} and that proportional to $m_s/m_b$ (from $M_{\rm d}$) in $\sin\theta$ may take place, if the phase difference between $B_{\rm u}$ and $B_{\rm d}$ is vanishing or very small. The leading order results $\tan\theta_{\rm u} = \sqrt{m_u/m_c}$ and $\tan\theta_{\rm d} = \sqrt{m_d/m_s}$ can still be obtained here, but their next-to-leading order corrections may deviate somehow from those obtained in the above subsections. (c) From the RRR pattern IV one can arrive at $\sin\theta \sim \sqrt{m_c/m_t}$ with the necessary condition $|C_{\rm u}| \ll |B_{\rm u}|$, since the (2,3) and (3,2) elements of $M_{\rm d}$ vanish. The results for $\tan\theta_{\rm u}$ and $\tan\theta_{\rm d}$ are similar to those obtained from the RRR pattern I. (d) The nonvanishing (1,3) and (3,1) elements of $M_{\rm u}$ in the RRR pattern III make its prediction for the mixing angles $\theta_{\rm u}$ and $\theta_{\rm d}$ quite different from all patterns discussed above. Analytically one can find $\tan\theta_{\rm u} \sim (m_b/m_s) \sqrt{m_u/m_t}$ , while $\tan\theta_{\rm d}$ is a complicated combination of the terms $\sqrt{m_d/m_s}$ and $(m_b/m_s) \sqrt{m_u/m_t}$ with a relative phase. In addition, $\sin\theta \sim m_s/m_b$ holds under the condition $|B_{\rm d}| \sim |C_{\rm d}|$, similar to the RRR pattern I. (e) For the RRR pattern V, the necessary condition $|B_{\rm u}| \gg |C_{\rm u}|$ is required in order to reproduce $\sin\theta \sim \sqrt{m_c/m_t}$ . Here again the nonvanishing (1,3) and (3,1) elements of $M_{\rm u}$ result in very complicated expressions for $\tan\theta_{\rm u}$ and $\tan\theta_{\rm d}$ (even more complicated than those in the RRR pattern III \cite{KX98}). For reasons of naturalness and simplicity, we argue that the RRR patterns III and V are unlikely to be good candidates for the quark mass matrices in an underlying theory of fermion mass generation. \section{Discussions and conclusion} \setcounter{equation}{0} We have studied the phenomena of quark flavor mixing and $CP$ violation in the context of generic hermitian mass matrices. The necessary and sufficient conditions for $CP$ violation in the standard model have been clarified at both the level of quark mass matrices and that of the flavor mixing matrix. Our particular observation is that $CP$ violation is primarily linked to a phase difference of about $90^{\circ}$ in the light quark sector, and this property becomes most apparent in the new parametrization (2.2). To be more specific, we have analyzed a realistic pattern of quark mass matrices with four texture zeros and given predictions for the flavor mixing and $CP$-violating parameters. The approximate congruency between the light-quark triangle (LT) and the rescaled unitarity triangle (UT), which provides an intuitive and scale-independent connection of $CP$-violating observables to quark mass ratios, is particularly worth mentioning. Let us make some further comments on the quark mass matrix (5.1), its phenomenological hints and its theoretical prospects. Naively one might not expect any prediction from the four-texture-zero mass matrices in (5.1), since they totally consist of ten free parameters (two of them are the phase differences between $M_{\rm u}$ and $M_{\rm d}$). This is not true, however, as we have seen. We find that two predictions, $\tan\theta_{\rm u} \approx \sqrt{m_u/m_c}$ and $\tan\theta_{\rm d} \approx \sqrt{m_d/m_s}$ , can be obtained in the leading order approximation. In some cases the latter may even hold in the next-to-leading order approximation, as shown in (5.2) and (5.7). Note again that these two relations, as a consequence of the hierarchy and texture zeros of our quark mass matrices, are essentially independent of the renormalization-group effects. This interesting scale-independent feature can also be seen from the LT and the rescaled UT as well as their inner angles $(\alpha, \beta, \gamma)$. It remains to be seen whether the interesting possibility $\varphi \approx \phi_1 \approx 90^{\circ}$, indicated by current data of quark masses and flavor mixing, could arise from an underlying flavor symmetry or a dynamical symmetry breaking scheme. Some speculations about this problem have been made (see, e.g., Refs. \cite{FX95,Weyers} and Refs. \cite{Democracy,New}). However, no final conclusion has been reached thus far. It is remarkable, nevertheless, that we have at least observed a useful relation between the area of the UT (${\cal A}_{\rm UT}$) and that of the LT (${\cal A}_{\rm LT}$) to a good degree of accuracy: \begin{equation} {\cal A}_{\rm UT} \; \approx \; |V_{cb}|^2 {\cal A}_{\rm LT} \; \approx \; \sin^2\theta ~ {\cal A}_{\rm LT} \; . \end{equation} Since ${\cal A}_{\rm UT} = {\cal J}/2$ measures the magnitude of $CP$ violation in the standard model, we conclude that $CP$ violation is primarily linked to the light quark sector. This is a natural consequence of the strong hierarchy between the heavy and light quark masses, which is on the other hand responsible for the smallness of ${\cal J}$ or ${\cal A}_{\rm UT}$. Is it possible to derive the quark mass matrix (5.1) in some theoretical frameworks? To answer this question we first specify the hierarchical structure of $M_{\rm q}$ in terms of the mixing angle $\theta_{\rm q}$ (for q = d or s). Adopting the radiant unit for the mixing angles (i.e., $\theta_{\rm u} \approx 0.085$, $\theta_{\rm d} \approx 0.204$ and $\theta \approx 0.040$), we have \begin{eqnarray} \frac{m_u}{m_c} & \sim & \frac{m_c}{m_t} \; \sim \; \theta^2_{\rm u} \; , \nonumber \\ \frac{m_d}{m_s} & \sim & \frac{m_s}{m_b} \; \sim \; \theta^2_{\rm d} \; . \end{eqnarray} Then the mass matrices $M_{\rm u}$ and $M_{\rm d}$, which have the mass scales $m_t$ and $m_b$ respectively, take the following {\it parallel} hierarchies: \begin{eqnarray} M_{\rm u} & \sim & m_t \left ( \matrix{ {\bf 0} & \theta^3_{\rm u} & {\bf 0} \cr \theta^3_{\rm u} & \theta^2_{\rm u} & \theta^2_{\rm u} \cr {\bf 0} & \theta^2_{\rm u} & {\bf 1} \cr} \right ) \; \; , \nonumber \\ \nonumber \\ M_{\rm d} & \sim & m_b \left ( \matrix{ {\bf 0} & \theta^3_{\rm d} & {\bf 0} \cr \theta^3_{\rm d} & \theta^2_{\rm d} & \theta^2_{\rm d} \cr {\bf 0} & \theta^2_{\rm d} & {\bf 1} \cr} \right ) \; \; , \end{eqnarray} where the relevant complex phases have been neglected. Clearly all three flavor mixing angles can properly be reproduced from (6.3), once one takes $\theta \approx \theta^2_{\rm d} \gg \theta^2_{\rm u}$ into account. The $CP$-violating phase $\varphi$ in $V$ comes essentially from the phase difference between the $\theta^3_{\rm u}$ and $\theta^3_{\rm d}$ terms. Of course $\theta_{\rm u}$ and $\theta_{\rm d}$, which are more fundamental than the Cabibbo angle $\theta_{\rm C}$ in our point of view, denote perturbative corrections to the rank-one limits of $M_{\rm u}$ and $M_{\rm d}$ respectively. They are responsible for the generation of light quark masses as well as the flavor mixing. They might also be responsible for $CP$ violation in a specific theoretical framework (e.g., the pure real $\theta_{\rm u}$ and the pure imaginary $\theta_{\rm d}$ might lead to a phase difference of about $90^{\circ}$ between $M_{\rm u}$ and $M_{\rm d}$, which is just the source of $CP$ violation favored by current data). The small parameter $\theta_{\rm q}$ could get its physical meaning in the Yukawa coupling of an underlying superstring theory: $\theta_{\rm q} = \langle \Theta_{\rm q} \rangle /\Omega_{\rm q}$, where $\langle \Theta_{\rm q} \rangle $ denotes the vacuum expectation value of the singlet field $\Theta_{\rm q}$, and $\Omega_{\rm q}$ represents the unification (or string) mass scale which governs higher dimension operators (see, e.g., Refs. \cite{Froggatt79,KX98,Ross94}). The quark mass matrices of the form (6.3) could then be obtained by introducing an extra (horizontal) U(1) gauge symmetry or assigning the matter fields appropriately. A detailed study of possible dynamical models responsible for the quark mass matrices (5.1) or (6.3) is certainly desirable but beyond the scope of this work. However, we believe that the texture zeros and parallel hierarchies of up and down quark mass matrices do imply specific symmetries, perhaps at a superhigh scale, and have instructive consequences on flavor mixing and $CP$-violating phenomena. The new parametrization of the flavor mixing matrix that we advocated is particularly useful in studying the quark mass generation, flavor mixing and $CP$ violation. \newpage \begin{flushleft} {\Large\bf Appendices} \end{flushleft}
\section{Introduction\label{sec:introduction}} \nopagebreak Many spiral galaxies display boxy or peanut-shaped bulges when viewed edge-on. Unfortunately, statistical studies of the incidence of these objects have not used objective criteria to quantify the boxiness of the bulges, and they are therefore hard to compare and yield moderately different results. Nevertheless, it seems clear that at least 20-30\% of edge-on spirals possess a boxy/peanut-shaped (hereafter B/PS) bulge (see Jarvis \markcite{j86}1986; Shaw \markcite{s87}1987; de Souza \& dos Anjos \markcite{dd87}1987). Spiral galaxies with a B/PS bulge are therefore a significant class of objects. The fact that the boxy/peanut shape is seen only in edge-on spirals indicates that the shape is related to the vertical distribution of light. Compared to the usual $R^{1/4}$ light distribution of spheroids, B/PS bulges have excess light above the plane at large galactocentric radii (see Shaw \markcite{s93}1993). Furthermore, the three-dimensional (hereafter 3D) light distribution of B/PS bulges must be even more extreme than their projected surface brightness (see, e.g., Binney \& Petrou \markcite{bp85}1985 for axisymmetric models). B/PS bulges also appear to rotate cylindrically, i.e.\ their rotation is independent of the height above the plane (e.g.\ Kormendy \& Illingworth \markcite{ki82}1982; Rowley \markcite{r86}1986). Several models have been proposed to explain the structure of B/PS bulges (e.g.\ Combes \& Sanders \markcite{cs81}1981; May, van Albada, \& Norman \markcite{mvn85}1985; Binney \& Petrou \markcite{bp85}1985). However, despite their prevalence and interesting structural and dynamical properties, B/PS bulges remain poorly studied observationally, probably because the edge-on projection makes the interpretation of observational data difficult. In this paper, we present new kinematical data for a large sample of spiral galaxies with a B/PS bulge. Our main goals are to determine their 3D structure and to identify their likely formation mechanism. We use our data to show that B/PS bulges are simply thick bars viewed edge-on. In \S~\ref{sec:formation}, we discuss the two main scenarios proposed for the formation of B/PS bulges -- accretion of satellite galaxies and buckling of a bar. We describe ways of discriminating between the two scenarios using kinematical data in \S~\ref{sec:diagnostics}. The observations are presented in \S~\ref{sec:observations} and the results discussed at length in \S~\ref{sec:discussion}. We conclude in \S~\ref{sec:conclusions}. \section{Formation Mechanisms\label{sec:formation}} \nopagebreak \subsection{Accretion\label{sec:accretion}} \nopagebreak Accretion of external material such as satellite galaxies was the early favoured mechanism to explain the formation of B/PS bulges in spiral galaxies. Binney \& Petrou (\markcite{bp85}1985) and Rowley (\markcite{r86}1986, \markcite{r88}1988) showed that it is possible to construct axisymmetric cylindrically rotating B/PS bulges from relatively simple distribution functions. Binney \& Petrou adopted a distribution function with a third integral of motion, in addition to the energy $E$ and angular momentum along the symmetry axis $L_z$. This integral favoured orbits reaching a particular height above the plane. Rowley adopted a two-integral distribution function, with a truncation depending on both $E$ and $L_z$. Both authors argued that the required distribution functions can form naturally through the accretion of material onto a host spiral galaxy. Binney \& Petrou (\markcite{bp85}1985) additionally argued that, for the accreted material to form a B/PS bulge, the velocity dispersion of the satellite must be much lower than its orbital speed, and the decay timescale must be much longer that the orbital time. This ensures that the accreted material stays clustered in phase space. Accretion scenarios face several problems. At one extreme, one can consider the accretion of several small satellite galaxies. However, only a narrow range of orbital energy and angular momentum can lead to the formation of a B/PS bulge, so it is improbable that many satellite galaxies would all share these properties. Remaining satellites should then be present but they are not observed (Shaw \markcite{s87}1987). The accretion of a single large companion is also ruled out by the large velocity dispersion and small decay timescale involved (Binney \& Petrou \markcite{bp85}1985). Similarly, the merger of two spiral galaxies of similar sizes (or of a spiral and a small elliptical) seems an unlikely route to form B/PS bulges. This would require a fairly precise alignment of the spins and orbital angular momenta of the two galaxies. We recall that about a third of all spiral galaxy bulges should be formed this way. From the arguments presented above, it seems that the accretion of a small number of moderate-sized satellites is the only accretion scenario which may lead to the formation of B/PS bulges. In favour of accretion is the fact that the best examples of B/PS bulges are found in small groups (e.g.\ NGC~128, ESO~597-~G~036). In addition, the possibly related X-shaped galaxies can form through the accretion of a satellite galaxy (e.g.\ Whitmore \& Bell \markcite{wb88}1988; Hernquist \& Quinn \markcite{hq89}1989). On the other hand, no evidence of accretion (arcs, shells, filaments, etc.) were detected by Shaw (\markcite{s93}1993) near spirals with a B/PS bulge, which argues against any kind of accretion. Furthermore, spiral galaxies with a B/PS bulge are not found preferentially in clusters (Shaw \markcite{s87}1987). We are thus led to the conclusion that, while accretion of external material may play a role in the formation of some B/PS bulges, it is unlikely to be the primary formation mechanism. \subsection{Bar-Buckling\label{sec:buckling}} \nopagebreak Bars can form naturally in $N$-body simulations of rotationally supported stellar disks (e.g.\ Sellwood \markcite{s81}1981; Athanassoula \& Sellwood \markcite{as86}1986), due to global bisymmetric instabilities (e.g.\ Kalnajs \markcite{k71}1971, \markcite{k77}1977). Based on 3D $N$-body simulations, Combes \& Sanders (\markcite{cs81}1981) were the first to suggest that B/PS bulges may arise from the thickening of bars in the disks of spiral galaxies. Their results were confirmed and their suggestion supported by later works (see, e.g., Combes et al.\ \markcite{cdfp90}1990; Raha et al.\ \markcite{rsjk91}1991). In short, the simulations show that, soon after a bar develops, it buckles and settles with an increased thickness and vertical velocity dispersion, appearing boxy-shaped when seen end-on and peanut-shaped when seen side-on. The B/PS bulges so formed are cylindrically rotating, as required. Toomre (\markcite{t66}1966) first considered the buckling of disks, in highly idealised models, and found that, if the vertical velocity dispersion in a disk is less than about a third of the velocity dispersion in the plane, the disk will be unstable to buckling modes (fire-hose or buckling instability; see also Fridman \& Polyachenko \markcite{fp84}1984; Araki \markcite{a85}1985). Bar formation in a disk makes the orbits within the bar more eccentric without affecting much their perpendicular motions, thereby providing a natural mechanism for the bar to buckle. Resonances between the rotation of the bar and the vertical oscillations of the stars can also make the disk vertically unstable (see Pfenniger \markcite{p84}1984; Combes et al.\ \markcite{cdfp90}1990; Pfenniger \& Friedli \markcite{pf91}1991). Irrespective of exactly how a bar buckled, the final shape of the bar is probably due to orbits trapped around the 2:2:1 periodic orbit family (see, e.g., Pfenniger \& Friedli \markcite{pf91}1991). Bar-buckling is the currently favoured mechanism for the formation of B/PS bulges. In particular, it provides a natural way to form B/PS bulges in isolated spiral galaxies, which accretion scenarios are unable to do. A number of facts also suggest a connection between (thick) bars and B/PS bulges, although they do not support bar-{\em buckling} directly. In particular, the fraction of edge-on spirals possessing a B/PS bulge is similar to the fraction of strongly barred spirals among face-on systems ($\approx30\%$; see, e.g., Sellwood \& Wilkinson \markcite{sw93}1993; Shaw \markcite{s87}1987). In a few cases, a bar can also be directly associated with a B/PS bulge from morphological arguments. NGC~4442 is such an example (Bettoni \& Galletta \markcite{bg94}1994). The main observational problem faced by the bar-buckling mechanism is that B/PS bulges seem to be shorter (relative to the disk diameter) than real bars or the strong bars formed in $N$-body simulations. However, this might simply be due to projection (a given surface brightness level being reached at a smaller radius in a more face-on disk) and no proper statistics have yet been compiled. The long term secular evolution of bars is also poorly understood. For example, it is known that bars can transfer angular momentum to a spheroidal component very efficiently (e.g.\ Sellwood \markcite{s80}1980; Weinberg \markcite{w85}1985). On the other hand, Debattista \& Sellwood (\markcite{ds98}1998) showed that, while a bar can be slowed down, it continues to grow and the boxy/peanut shape is preserved. It is less clear what happens if a bar is strongly perturbed. However, Norman, Sellwood, \& Hasan (\markcite{nsh96}1996) showed that, if a bar is destroyed through the growth of a central mass concentration (e.g.\ Hasan \& Norman \markcite{hn90}1990; Friedli \& Benz \markcite{fb93}1993), the boxy/peanut shape is also destroyed. The merits of the bar-buckling mechanism significantly outweigh these potential problems, and the bar-buckling scenario to form B/PS bulges remains largely unchallenged at the moment, despite very little observational support. The main aim of this paper is to test this mechanism, by looking for bars in a large sample of spiral galaxies with a B/PS bulge. Although this stops short of a direct verification of buckling, it does test directly for a possible relationship between bars and B/PS bulges. We will come back to this distinction in \S~\ref{sec:discussion}. Although it is not part of our program, the Milky Way is a primary example of such a galaxy. The Galactic bulge is boxy-shaped (Weiland et al.\ \markcite{wetal94}1994) and it is now well established that it is bar-like (e.g.\ Binney et al.\ \markcite{bgsbu91}1991; Paczy\'{n}ski et al.\ \markcite{psuskkmk94}1994; see Kuijken \markcite{k96}1996 for a brief review of the subject). We note that ``hybrid'' scenarios have also been proposed to explain the formation of B/PS bulges, and have gathered recent support from statistical work on the environment of B/PS bulges by L\"{u}tticke \& Dettmar (\markcite{ld98}1998). An interaction or merger can excite (or accelerate) the development of a bar in a disk which is stable (or quasi-stable) against bar formation (e.g.\ Noguchi \markcite{n87}1987; Gerin, Combes, \& Athanassoula \markcite{gca90}1990; Miwa \& Noguchi \markcite{mn98}1998). Bars formed this way are then free to evolve to a boxy/peanut shape in the manner described above (see, e.g., Mihos et al.\ \markcite{mwhdb95}1995), and the bulges thus formed owe as much to interactions as they do to the bar-buckling instability. However, the {\em shape} of the bulges is due to the buckling of the bars and interaction is merely a way to start the bar formation process. In that sense, hybrid scenarios for the formation of B/PS bulges are really bar-buckling scenarios, and the possible accretion of material is not directly related to the issue of the bulges' shape. \section{Observational Diagnostics\label{sec:diagnostics}} \nopagebreak Our main goal with the observations presented in this paper is to look for the presence of a bar in the disk of spiral galaxies possessing a B/PS bulge. There is no straightforward photometric way to identify a bar in an edge-on spiral. The presence of a plateau in the major-axis light profile of the disk has often been invoked as an indicator of a bar (e.g.\ de Carvalho \& da Costa \markcite{dd87}1987; Hamabe \& Wakamatsu \markcite{hw89}1989). However, axisymmetric or quasi-axisymmetric features (e.g.\ a lens) can be mistaken for a bar and end-on bars may remain undetected. Kuijken \& Merrifield (\markcite{km95}1995; see also Merrifield \markcite{mk96}1996) were the first to demonstrate that bars could be identified kinematically in external edge-on spiral galaxies. They showed that periodic orbits in a barred galaxy model produce characteristic double-peaked line-of-sight velocity distributions when viewed edge-on. This gives their modeled spectra a spectacular ``figure-of-eight'' (or X-shaped) appearance, which they were able to observe in the long-slit spectra of the B/PS systems NGC~5746 and NGC~5965 (see Fig.~\ref{fig:main} for examples). Their approach is analogous to that used in the Galaxy with longitude-velocity diagrams (e.g.\ Peters \markcite{p75}1975; Mulder \& Liem \markcite{ml86}1986; Binney et al.\ \markcite{bgsbu91}1991). Bureau \& Athanassoula (\markcite{ba99}1999, hereafter BA99) refined the dynamical theory of Kuijken \& Merrifield (\markcite{km95}1995). They used periodic orbit families in a barred galaxy model as building blocks to model the structure and kinematics of real galaxies. They showed that the global structure of a position-velocity diagram\footnote{Position-velocity diagrams (PVDs) show the projected density of material in a system as a function of line-of-sight velocity and projected position.} (hereafter PVD) taken along the major axis of an edge-on system is a reliable bar diagnostic, particularly the presence of gaps between the signatures of the different families of periodic orbits. Athanassoula \& Bureau (\markcite{ab99}1999, hereafter AB99) produced similar bar diagnostics using hydrodynamical simulations of the gaseous component alone. They showed that, when $x_2$ orbits are present (corresponding to the existence of an inner Lindblad resonance (hereafter ILR)), the presence of gaps in a PVD, between the signature of the $x_2$-like flow and that of the outer parts of the disk, reliably indicates the presence of a bar. If no $x_2$ orbits are present, one must rely on indirect evidence to argue for the presence of a bar (see, e.g., Contopoulos \& Grosb\o l \markcite{cg89}1989 for a review of periodic orbits in barred spirals). The gaps are a direct consequence of the shocks which develop in relatively strong bars. These shocks drive an inflow of gas toward the center of the galaxies and deplete the outer (or entire) bar regions (see, e.g., Athanassoula \markcite{a92}1992). The simulations of \markcite{ab99}AB99 can be directly compared with the emission line spectra presented here, and will form the basis of our argument. The models of \markcite{ba99}BA99 and \markcite{ab99}AB99 can also be used to determine the viewing angle with respect to a bar, as the signature present in the PVDs changes with the orientation of the line-of-sight. In particular, when a bar is seen end-on, the $x_1$ orbits (and $x_1$-like flow, both elongated parallel to the bar) reach very high radial velocities, while the $x_2$ orbits (and $x_2$-like flow, both elongated perpendicular to the bar) show only relatively low velocities. The opposite is true when a bar is seen side-on. In addition, the presence or absence of $x_2$ orbits can somewhat constrain the mass distribution and bar pattern speed of an observed galaxy. We have not developed specific observational criteria to identify past or current accretion of material in the observed galaxies. As discussed in \S~\ref{sec:accretion}, accretion will occur through interactions or merger events. We will take as the signature of such events, and of possible accretion, the presence of irregularities in the observed PVDs, in particular strong asymmetries about the center of an object (see Fig.~\ref{fig:main} for examples). \section{Observations\label{sec:observations}} \nopagebreak \subsection{The Sample\label{sec:sample}} \nopagebreak Our sample of galaxies consists of 30 edge-on spirals selected from the catalogues of Jarvis (\markcite{j86}1986), Shaw (\markcite{s87}1987), and de Souza \& dos Anjos (\markcite{dd87}1987) (spiral galaxies with a B/PS bulge), and from the catalogue of Karachentsev, Karachentseva, \& Parnovsky (\markcite{kkp93}1993) (spirals with extreme axial ratios; see also Karachentsev et al.\ \markcite{kkpk93}1993). In order to have enough spatial resolution in the long-slit spectroscopy, but still be able to image the galaxies relatively quickly with a small-field near-infrared (hereafter NIR) camera, we have selected objects with bulges larger than 0\farcm6 in diameter and disks smaller than about 7\farcm0 (at the 25~B~mag~arcsec$^{-2}$ level). NIR imaging is important to refine the classification of the bulges and to subsequently study the vertical structure of the identified bars. All objects are accessible from the south ($\delta\leq15\arcdeg$). Three-quarters (23/30) of the galaxies either have probable companions or are part of a group or cluster. A few of these probably are chance alignments, so it is fair to say that we are not biased either against or for galaxies in a dense environment. We should therefore be able to estimate reliably the importance of accretion in the formation of B/PS bulges. Of the sample galaxies described above, 80\% (24/30) have a boxy or peanut-shaped bulge, while 20\% (6/30) have a more spheroidal bulge morphology and constitute a ``control'' sample. Of the former group, it turned out that 17 galaxies have emission lines extending far enough in the disk to apply the diagnostics developed by \markcite{ba99}BA99 and \markcite{ab99}AB99 with the ionised gas; all galaxies in the control group fulfill this condition. In this paper, we will thus concentrate on a main sample of 17 edge-on spiral galaxies with a B/PS bulge and a comparison sample of 6 edge-on spiral galaxies with more spheroidal bulges. The galaxies in each sample are listed in Tables~\ref{ta:main} and \ref{ta:control} respectively, along with information on their properties and environment. The galaxies with no or confined emission are listed in Table~\ref{ta:undetected}. For those, stellar kinematics must be used to search for the presence of bar. We note that the galaxy type listed in Tables~\ref{ta:main}--\ref{ta:undetected} is not precise to more than one or two morphological type, because of the difficulty of classifying edge-on spirals. The bulge-to-disk ratio is effectively the only criterion left to classify the objects. Other than the catalogue of Karachentsev et al.\ (\markcite{kkp93}1993), we are not aware of any general catalogue of edge-on spiral galaxies. This makes it difficult to build a large and varied control sample including edge-on spiral galaxies with large bulges (the Karachentsev et al.\ \markcite{kkp93}1993 catalogue is restricted to galaxies with major to minor axis ratio $a/b\geq7$). Such a catalogue would be very useful, and could probably be constructed from an initial list of candidates taken from a catalogue such as the RC3 (de Vaucouleurs et al.\ \markcite{ddcbpf91}1991), which would then be inspected on survey material. \subsection{Observations and Data Reduction\label{sec:data}} \nopagebreak Our spectroscopic data were acquired between December 1995 and May 1997 (total of 39~nights) using the Double Beam Spectrograph on the 2.3~m telescope at Siding Spring Observatory. A $1752\times532$~pixels SITE ST-D06A thinned CCD was used. The observations discussed in this paper were obtained with the red arm of the spectrograph centered on the H$\alpha$ $\lambda6563$ emission line. All galaxies were observed using a $1\farcs8\times400\arcsec$ slit aligned with the major axis. For objects with a strong dust lane, the slit was positioned just above it. The spectral resolution is about 1.1~\AA\ FWHM (0.55~\AA~pixel$^{-1}$) and the spatial scale is 0\farcs9~pixel$^{-1}$. These data can be directly compared with the gas dynamical models of \markcite{ab99}AB99. When no emission line was detected in an object, the red arm of the spectrograph was moved to the \ion{Ca}{2} absorption line triplet. The blue arm was always centered on the Mg~$b$ absorption feature. These data will form the core of a future paper discussing stellar kinematics in the sample galaxies (including the galaxies in Table~\ref{ta:undetected}). Total exposure times on both arms ranged from 12000 to 21000~s on each object. The spectra were reduced using the standard procedure within IRAF. The data were first bias-subtracted, using both vertical and horizontal overscan regions, and then using bias frames. If necessary, the data were also dark-subtracted. The spectra were then flatfielded using flattened continuum lamp exposures, and wavelength-calibrated using bracketing arc lamp exposures for each image. The data were then rebinned to a logarithmic wavelength (linear velocity) scale corresponding to 25~km~s$^{-1}$~pixel$^{-1}$. The spectra were then corrected for vignetting along the slit using sky exposures. All exposures of a given object were then registered and offset along the spatial axis, corrected to a heliocentric rest frame, and combined. The resulting spectra were then sky-subtracted using source-free regions of the slit on each side of the objects. The sky subtraction was less than perfect in some cases, mainly because of difficulties in obtaining a uniform focus of the spectrograph along the entire length of the slit. This was particularly troublesome for objects like IC~2531 and NGC~5746 which have sizes comparable to that of the slit (see, e.g., Fig.~\ref{fig:lineratios}a). In order to isolate the emission lines, the continuum emission of the objects was then subtracted using a low-order fit to the data in the spectral direction. The resulting spectra constitute the basis of our discussion in the next section. We note that in regions with bright continuum emission, like the center of some galaxies, the continuum subtraction can leave high shot noise in the data, which should not be confused with line emission in the grayscale plots of Figure~\ref{fig:main}--\ref{fig:lineratios}. This is the case for example in the spectra of ESO~240-~G~011, NGC~1032, and NGC~4703. The effect is perhaps best seen when a bright star is subtracted, such as in the spectra of NGC~2788A or NGC~1032 (see Fig.~\ref{fig:main}). We have not extracted rotation curves from our data. This is because the entire two-dimensional spectrum, or PVD, is required to identify the signature of a bar in an edge-on spiral galaxy (see \markcite{ba99}BA99 and \markcite{ab99}AB99). Evidence of interaction and of possible accretion of material is also more easily seen in the PVDs. We present the [\ion{N}{2}] $\lambda6584$ emission line rather than H$\alpha$ because it is not affected by underlying stellar absorption. \subsection{Results\label{sec:results}} \nopagebreak \placefigure{fig:main} \placefigure{fig:control} We present the emission line spectrum for the sample galaxies which have extended emission only. The PVDs of the galaxies in the main sample and the control sample are shown in Figure~\ref{fig:main} and Figure~\ref{fig:control}, respectively. In order to illustrate the range of galaxy type and bulge morphology in the sample, and to allow connections to be made between bulge morphologies and kinematical features in the disks, each PVD is accompanied by a registered image of the corresponding galaxy (from the Digitized Sky Survey) on the same spatial scale. We discuss here the trends observed across the data set. The most important trend, and the main result of this paper, is that most galaxies in the main sample show a clear bar signature in their PVD (as described in \S~\ref{sec:diagnostics}). Of the 17 galaxies in the main sample of spirals with a B/PS bulge and extended emission lines, we conclude that 14 are barred. In these objects, the PVD clearly shows a strong and steep inner component, associated with an $x_2$-like flow, and a slowly-rising almost solid-body component, associated with the outer disk, and joining the flat section of the rotation curve in the outer parts. The two components are separated by a gap, caused by the absence (or low density) of gaseous material with $x_1$-like kinematics in the outer bar regions\footnote{The PVDs of NGC~128 and IC~2531 do not display a bar signature, but Emsellem \& Arsenault (\markcite{ea97}1997) and Bureau \& Freeman (\markcite{bf97}1997) showed, using other data, that each galaxy harbours a bar.}. The best examples of this type of bar signature are seen in the PVD of earlier-type objects, like NGC~2788A, NGC~5746, and IC~5096. However, the signature is still clearly visible in the PVD of galaxies as late as ESO~240-~G~011. In the main sample, only one galaxy, NGC~4469, may be axisymmetric, with no evidence of either a bar or interaction (although it is not possible to rule out an interaction which would have occured a long time ago, leaving no observable trace). Two galaxies, NGC~3390 and ESO~597-~G~036, have a disturbed strongly asymmetric PVD, which we ascribe to a recent interaction (obvious in the case of ESO~597-~G~036). These interactions may have led to the accretion of material. The results for the entire main sample are summarised in Table~\ref{ta:main}. Another significant result of this study is that no galaxy in the control sample shows evidence for a bar. Four of the six galaxies appear to be axisymmetric, without evidence for either a bar or interaction, and two galaxies (NGC~5084 and NGC~7123) have a disturbed PVD, indicating they underwent an interaction recently and possibly accreted material. These results are tabulated in Table~\ref{ta:control}. \section{Discussion\label{sec:discussion}} \nopagebreak \subsection{The Structure of Boxy/Peanut-Shaped Bulges\label{sec:structure}} \nopagebreak The only previous study of this kind was that of Kuijken \& Merrifield (\markcite{km95}1995), who proposed the method and considered two galaxies; Bureau \& Freeman (\markcite{bf97}1997) presented preliminary results of the current work. This is thus the first systematic observational study of the relationship between bars and B/PS bulges. In summary, our main result is that, based on new kinematical data, 14 of the 17 galaxies with a B/PS bulge in our sample are barred, and the remaining 3 galaxies show evidence of interaction or may be axisymmetric. None of the 6 galaxies without a B/PS bulge in our sample shows any indication of a bar. This means that most B/PS bulges are due to the presence of a thick bar viewed edge-on, and only a few may be due to the accretion of external material. In addition, the more spheroidal bulges (i.e.\ non-B/PS) do seem axisymmetric. It appears then that most B/PS bulges are edge-on bars, and that most bars are B/PS when viewed edge-on. However, the small size of the control sample prevents us from making a stronger statement about this converse. There is also a continuum of bar strengths in nature and we would expect to have intermediate cases. The galaxies NGC~3957 and NGC~4703 may represent such objects: one could argue that their PVDs display weak bar signatures, and indeed their bulges are the most flattened in the control sample. If bars were unrelated to the structure of bulges, we would have expected only about 5 galaxies in the main sample to be strongly barred, and about 2 galaxies in the control sample (about 30\% of spirals are strongly barred, Sellwood \& Wilkinson \markcite{sw93}1993). Clearly, our results are incompatible with these expectations. Recent results by Merrifield \& Kuijken \markcite{mk99}(1999) also support our conclusions. Based on a smaller sample of northern edge-on spirals, they show clearly that as bulges become more B/PS, the complexity and strength of the bar signature in the PVD also increase. Our association of bars and B/PS bulges supports the bar-buckling mechanism for the formation of the latter. However, we do not test directly for buckling, but rather for the presence of a barred potential in the plane of the disks. Because bars form rapidly and buckle soon after, on a timescale of only a few dynamical times (see, e.g., Combes et al.\ \markcite{cdfp90}1990), it is unlikely that any galaxy in this nearby sample would have been caught in the act. Thus, other mechanisms which could lead to thick bars cannot be excluded. In addition, we have no way of determining how the bars themselves formed, or even whether they formed spontaneously in isolation or through interaction with nearby galaxies or companions. Therefore, the possibility of hybrid scenarios for the formation of B/PS bulges, where a bar is formed because of an interaction and subsequently thickens due to the buckling instability, remains (see \S~\ref{sec:buckling}). We have looked mainly at objects with large bulges; only two of the galaxies studied are late-type spirals (ESO~240-~G~011 and IC~5176). This is a selection effect due to the difficulty to identify very small B/PS bulges. It would therefore be interesting to search for bars in very late edge-on spiral galaxies, and verify whether thin bars do exist: the bar in ESO~240-~G~011 is not very thick, but it is thicker (isophotally) than the disk. \ion{H}{1} synthesis imaging is probably the best way to achieve this goal, as these objects are often dusty and \ion{H}{1}-rich. Such bars may even provide a novel way to constrain the total (luminous and dark) mass distribution of spirals, in a manner analogous to the use of warps or flaring, as buckling is sensitive to the presence of a dark halo (e.g.\ Combes \& Sanders \markcite{cs81}1981). We will report on \ion{H}{1} synthesis observations of a few objects in our sample in a later paper. Our observations revive the old issue of the exact nature of bulges. In face-on systems, one can often clearly identify a bar and a more nearly axisymmetric component usually referred to as the bulge. However, Kormendy (\markcite{k93}1993) has argued that, at least in some examples, these apparent bulges may just be structures in the disk. In edge-on spirals, we are not aware of any galaxies displaying two separate vertically extended components. This raises the question of whether the bars and bulges of face-on systems are really two distinct structural and dynamical components, despite the fact that they can be separated photometrically. Our data on edge-on galaxies tightly link the presence of a bar with the presence of a B/PS bulge, which suggests that bars and B/PS bulges are very closely related. This view is supported by theoretical and modelling work on barred spiral galaxies (e.g.\ Pfenniger \markcite{p84}1984; Pfenniger \& Friedli \markcite{pf91}1991), as well as by some photometric studies (e.g.\ Ohta \markcite{o96}1996). However, more work is required to settle the issue. Kinematical data covering whole bulges would be particularly useful. \markcite{ba99}BA99 and \markcite{ab99}AB99 proposed diagnostics, again based on the structure of the observed PVD, to determine the viewing angle with respect to a bar in an edge-on disk (see \S~\ref{sec:diagnostics}). When the bar is seen close to side-on, the maximum line-of-sight velocity reached by the $x_2$-like flow is similar to or larger than the flat portion of the rotation curve, and the component of the PVD associated with that flow is very steep. When the bar is seen close to end-on, the $x_2$-like flow only reaches low velocities and extends to relatively large projected distances. These diagnostics are ideally suited to test the main prediction of $N$-body models, that bars are peanut-shaped when seen side-on and boxy-shaped when seen end-on (see, e.g., Combes \& Sanders \markcite{cs81}1981; Raha et al.\ \markcite{rsjk91}1991). Of the 12 barred galaxies in the main sample for which it is possible to apply these criteria (we exclude NGC~128 and IC~2531), two-thirds (8/12) seem to be consistent with the above prediction of $N$-body simulations. For example, in the galaxy with a peanut-shaped bulge IC~4937, it is clear that the steep inner component associated with the $x_2$-like flow extends to higher velocities than the outer parts of the disk (see Fig.~\ref{fig:main}). This situation is reversed in NGC~1886, which has a boxy-shaped bulge. However, caution is required when interpreting this result. Firstly, the present classification of the shape of the bulges is affected by both dust and the low dynamic range of the material used (Digitized Sky Survey). To remedy this problem, we have acquired $K$-band images of all the sample galaxies and will report on these observations in a future paper. Secondly, no clear prediction has been made from $N$-body simulations about the viewing angle at which the transition from a boxy to a peanut-shaped bulge occurs. In that regard, it would be useful to apply quantitative measurements of the boxiness and ``peanutness'' of the bulges to both simulation results and observational data (see, e.g., Bender \& M\"{o}llenhoff \markcite{bm87}1987 and Athanassoula et al.\ \markcite{amwpplb90}1990 for two possible methods). Thirdly, because the $x_2$-like flow is located near the center of the galaxies, the velocities it reaches depend somewhat on the central concentration of the objects (which affects the circular velocity in the central regions). This obviously varies significantly amongst the galaxies in our sample. Therefore, while our observations may support the prediction of $N$-body models concerning the orientation of the bar in B/PS bulges, we believe that it is premature to make a detailed quantitative comparison of the data with the models. For a more detailed comparison with theory, data of higher signal-to-noise and higher spatial resolution than the average PVD presented here would be very desirable. However, it would be worthwhile to model individually the best PVDs obtained in the present study (e.g.\ NGC~5746, IC~5096, and a few others). This would very likely lead to tight constraints on the mass distribution and bar properties of the galaxies, including the orientation and pattern speed of the bars (see \markcite{ba99}BA99; \markcite{ab99}AB99). The $K$-band images could also be used to constrain the mass distributions. Comparing the data with the kinematics (or simply the rotation curve) predicted from an axisymmetric deprojection would provide an easy test of the shape of the bulges. On a related subject, the significant thickness of bars suggests that their 3D structure should be taken into account when deriving the potential of more face-on systems from NIR images. In that regard, we should stress that the bar diagnostics we used rely on the presence of an $x_2$-like flow in the center of the galaxies, and thus on the existence of ILRs (or, at least, one ILR; see \markcite{ba99}BA99; \markcite{ab99}AB99). A priori, barred disks or B/PS bulges need not have ILRs, but at least 13 of the 17 galaxies in the main sample do (we additionally exclude NGC~128 here). Our data therefore strongly support the view that barred spiral galaxies generally have ILRs (see also Athanassoula \markcite{a91}1991, \markcite{a92}1992). \subsection{Dust and Emission Line Ratios in Boxy/Peanut-Shaped Bulges\label{sec:elratios}} \nopagebreak Because many galaxies in our sample have a prominent dust lane, it is important to consider the effects dust may have on our observations. Its principal consequence in edge-on systems is to limit the depth to which the line-of-sight penetrates the disk. To bypass this problem, we selected many galaxies to be slightly inclined, so it was possible in those objects to position the slit just above the dust lane and have a line-of-sight that still goes through most of the disk, as required for a comparison with the models of \markcite{ba99}BA99 and \markcite{ab99}AB99. In the few cases with a strong dust lane and a perfectly edge-on disk, we tried again to offset the slit slightly. Unfortunately, with the observational set-up available at the telescope, it was difficult to position the slit with great precision. The objects where we suspected that dust could affect our observations are indicated in Tables~\ref{ta:main}--\ref{ta:undetected}. A large dust optical depth produces an almost featureless PVD, as one sees only an outer annulus of material in the disk, and the rotation curve appears slowly-rising and solid-body (see, e.g., Bosma et al.\ \markcite{bbfa92}1992). The only objects showing obvious signs of extinction are IC~2531\footnote{This is confirmed by the \ion{H}{1} radio synthesis data of Bureau \& Freeman (\markcite{bf97}1997), which reveal a complex PVD with a bar signature.}, NGC~4703, and possibly ESO~443-~G~042. Because we see a lot of structure in most PVDs, including the PVDs of objects with a significant dust lane, we do not believe that our results are significantly affected by dust. We do detect a clear bar signature in most galaxies in the main sample. This statement is strengthened by the fact that all the PVDs showing a bar signature are close to symmetric. \markcite{ab99}AB99 showed that the bar signature would be strongly asymmetric in a very dusty disk, and this is not observed. Similarly, it is improbable that irregular dust distributions would lead to such well-ordered and symmetric PVDs (very large and localized dust ``patches'' would be required to create the important gaps observed in many objects). \placefigure{fig:lineratios} An unexpected but interesting prospect raised by our observations concerns emission line ratios. For many of the barred galaxies in the main sample, the emission line ratios in the central regions are different from those expected of typical \ion{H}{2} regions. For 9 barred galaxies out of 14, mostly those with the strongest bar signatures, the [\ion{N}{2}] $\lambda6584$ to H$\alpha$ $\lambda6563$ ratio in the bulge region is greater than unity. In particular, in a few objects, the steep inner component of the PVD, associated with the $x_2$-like flow, is much stronger in [\ion{N}{2}] than it is in H$\alpha$, while the slowly rising component, associated with the outer disk, has a [\ion{N}{2}]/H$\alpha$ ratio typical of \ion{H}{2} regions. In fact, the inner component can be almost absent in H$\alpha$. We illustrate these behaviours in Figure~\ref{fig:lineratios}, which shows the PVDs of the galaxies NGC~5746 and IC~5096 in the H$\alpha$ and [\ion{N}{2}] $\lambda6548,6584$ lines. It is possible that the H$\alpha$ emission line is weakened by the underlying stellar absorption. This suggestion is supported by the fact that the spectra of many of the galaxies in the main sample show strong Balmer absorption lines. However, the H$\alpha$ absorption would need to be very large to account for the extreme [\ion{N}{2}]/H$\alpha$ ratios observed in some objects (e.g.\ IC~5096). The strong Balmer lines observed in many objects are nevertheless interesting in themselves, and indicate that a significant intermediate age ($\sim1$~Gyr) stellar population is present in the central regions of the disk of many of the galaxies. It would be interesting to investigate if these past bursts of star formation can be related to the presence of the bars. The high emission line ratios are interesting for two reasons. Firstly, high [\ion{N}{2}]/H$\alpha$ ratios are commonly believed to be produced by shocks (see, e.g., Binette, Dopita, \& Tuohy \markcite{bdt85}1985; Dopita \& Sutherland \markcite{ds96}1996). This is consistent with the view that B/PS bulges are barred spirals viewed edge-on. The steep inner components of the PVDs, which display high [\ion{N}{2}]/H$\alpha$ ratios, are associated with an $x_2$-like flow and the nuclear spirals observed in many barred spiral galaxies (\markcite{ab99}AB99). Athanassoula (\markcite{a92}1992) showed convincingly that these are the locus of shocks. Secondly, if one were to derive H$\alpha$ and [\ion{N}{2}] rotation curves from the data, by taking the upper envelope of the PVDs (the standard method), the H$\alpha$ and [\ion{N}{2}] rotation curves would significantly differ for many objects. The [\ion{N}{2}] lines would yield rapidly rising rotation curves flattening out at small radii, while the H$\alpha$ line would yield slowly rising rotation curves flattening out at relatively large radii. Mass models derived from such data would thus yield qualitatively different results, and our understanding of galactic dynamics and structure gained from this type of work could be seriously erroneous (at least for highly inclined spirals). Of course, now that these galaxies are known to be barred, their rotation curves should not be used directly for mass modelling, as they are not a good representation of the circular velocity. Data such as those presented in Figure~\ref{fig:lineratios} also open up the possibility of determining the ionisation conditions and abundance of the gas in different regions of the galaxies in a single observation. Because the deprojected location of each component of the PVDs is known (see \markcite{ab99}AB99), this provides an efficient way to study the effects of bars on the interstellar medium of galaxies on various scales. \section{Conclusions\label{sec:conclusions}} \nopagebreak In this paper, we discussed the various mechanisms proposed for the formation of the boxy and peanut-shaped bulges observed in some edge-on spiral galaxies. We argued that accretion scenarios were unlikely to account for most boxy/peanut-shaped (B/PS) bulges, but that bar-buckling scenarios, discovered through $N$-body simulations, had this potential. Using recently developed kinematical bar diagnostics, we searched for bars in a large sample of edge-on spiral galaxies with a B/PS bulge. Of the 17 galaxies where the diagnostics could be applied using emission lines, 14 galaxies were shown to be barred, 2 galaxies were significantly disturbed, and 1 galaxy seemed to be axisymmetric. In a control sample of 6 galaxies with spheroidal bulges, none appeared to be barred. Our study supports the bar-buckling mechanism for the formation of B/PS bulges. Our results imply that most B/PS bulges are due to the presence of a thick bar that we are viewing edge-on, while only a few may be due to the accretion of external material. Furthermore, spheroidal bulges do appear to be axisymmetric. This suggests that all bars are B/PS. Our observations also seem to support the main prediction of $N$-body simulations, that bars appear boxy-shaped when seen end-on and peanut-shaped when seen side-on. However, this issue should be revisited in a more quantitative manner in the future. With our data, we have no way of determining whether the bars leading to B/PS bulges have formed in isolation or through interactions and mergers. The association of B/PS bulges and bars is entirely consistent with the properties of the bulge of the Milky Way, which is known to be both boxy and bar-like. We considered the effects of dust on our observations, but concluded that it does not affect our results significantly. We have also shown that emission line ratios correlate with kinematical structures in many barred galaxies. This make possible a direct study of the large scale effects of bars on the interstellar medium in disks. Our study opens up the possibility to study observationally the vertical structure of bars. This was not possible before and represents an interesting spin-off from the use of bar diagnostics in edge-on spiral galaxies. To this end, we have obtained $K$-band images of all our sample galaxies, and will report on this work in a future paper. \acknowledgments We thank the staff of Mount Stromlo and Siding Spring Observatories for their assistance during and after the observations. We also thank A.\ Kalnajs, E.\ Athanassoula, A.\ Bosma, and L.\ Sparke for useful discussions. M.\ B.\ acknowledges the support of an Australian DEET Overseas Postgraduate Research Scholarship and a Canadian NSERC Postgraduate Scholarship during the conduct of this research. The Digitized Sky Surveys were produced at the Space Telescope Science Institute under U.S. Government grant NAG W-2166. The images of these surveys are based on photographic data obtained using the Oschin Schmidt Telescope on Palomar Mountain and the UK Schmidt Telescope. The plates were processed into the present compressed digital form with the permission of these institutions. The 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.
\section{Introduction} Many strong interaction--processes involve meson--baryon coupling constants as the main ingredient. The determination of these fundamental quantities requires information about the physics at large distance. In other words, for a reliable determination of these parameters we need some nonperturbative approach. Among all nonperturbative approaches, QCD sum rules \cite{R1} is one of the most powerful method in studying the properties of hadrons. This method is based on the short distance OPE of vacuum--vacuum correlation function in terms of condensates. For the processes involving light mesons $\pi,~K$ or $\rho$, there is an alternative method to the traditional QCD sum rules, namely, light cone QCD sum rules \cite{R2}. In this approach the expansion of the vacuum--meson correlator is performed near the light cone in terms of the meson wave functions. The meson wave functions are defined by the matrix elements of non--local composite operators sandwiched between the meson and vacuum states and classified by their twists, rather than dimensions of the operators, as is the case in the traditional sum rules. Many applications of light--cone QCD sum rules can be found in \cite{R3}--\cite{R11} and references therein. In this work we use light cone QCD sum rules approach for determination of the coupling constants of the pion to the lowest states of the baryon octet $\Sigma$ and $\Lambda$, $g_{\pi\Sigma\Lambda}$ and $g_{\pi\Sigma\Sigma}$. Note that these coupling constants were investigated in framework of the QCD sum rules based on pion--to--vacuum matrix element in the leading order of the pion momentum $q$ for the structure $\not\!q \gamma_5$ in \cite{R12}, where $q$ is the pion momentum. The results of this work are currently under debate in literature (see discussions in \cite{R13} and \cite{R14}). Moreover in \cite{R15} and \cite{R16} it was pointed out that there is coupling scheme dependence for the structures $\gamma_5$, $\not\!q \gamma_5$, i.e., dependence on the pseudoscalar or pseudovector forms of the effective interaction Lagrangian of pion with hadrons have been used, while the structure $\sigma_{\mu\nu} \gamma_5$ is shown to be independent of any coupling schemes. For this reason, in present work we choose the structure $\sigma_{\mu\nu} \gamma_5 p^\mu q^\nu$, where $p$ and $q$ are the $\Lambda~(\Sigma)$ and the pion momenta, respectively. It should be noted that the sum rules for the $\sigma_{\mu\nu} \gamma_5 p^\mu q^\nu$ structure was derived in \cite{R17} in investigation of $g_{\rho\omega\pi}$ coupling constant. The paper is organized as follows. In Section 2 we derive sum rules for the pion--baryon coupling constants $g_{\pi\Sigma\Lambda}$ and $g_{\pi\Sigma\Sigma}$ for the structure $\sigma_{\mu\nu} \gamma_5 p^\mu q^\nu$. Section 3 is devoted to the numerical analysis of the sum rules for $g_{\pi\Sigma\Lambda}$ and $g_{\pi\Sigma\Sigma}$ and discussion. \section{Formulation of the pion--baryon sum rule for the $g_{\pi\Sigma\Lambda}$ and $g_{\pi\Sigma\Sigma}$} According to the main philosophy of the QCD sum rules, a quantitative estimation for $g_{\pi\Sigma\Lambda}$ and $g_{\pi\Sigma\Sigma}$ couplings can be obtained by matching the representations of a suitable correlator in terms of hadronic (physical part) and quark--gluon language (theoretical part). For this purpose we consider the following two--point correlator function with pion \begin{eqnarray} \Pi(p,q) = \int d^4x \, e^{ipx} \left< \pi(q) \left| \mbox{\rm T} \left[\eta_Y (x) \bar \eta_{\Sigma^+} (0) \right] \right| 0 \right>~, \end{eqnarray} where $p$ and $\eta_Y$ are the four--momentum of the hyperon (in our case $\Lambda^0$ or $\Sigma^0$) and its interpolation current, respectively, $\eta_{\Sigma^+}$ is the interpolating current of $\Sigma^+$ and $q$ is the pion four--momentum. The interpolating currents for $\Lambda^0$, $\Sigma^0$ and $\Sigma^+$ are \cite{R18} \begin{eqnarray} \eta_{Y^0} &=& \alpha \, \epsilon_{abc} \left[ \left( u_a^T {\cal C} \gamma_\mu s_b \right) \gamma_5 \gamma^\mu d_c \mp \left( d_a^T {\cal C} \gamma_\mu s_b \right) \gamma_5 \gamma^\mu u_c \right]~,\nonumber \\ \eta_{\Sigma^+}&=& \epsilon_{abc} \left( u_a^T {\cal C} \gamma_\mu u_b \right) \gamma_5 \gamma^\mu s_c~, \end{eqnarray} where $s,~u$ and $d$ are strange, up and down quark fields, the upper(lower) sign corresponds to $\Lambda^0(\Sigma^0)$ and $\alpha=\sqrt{2/3}$ for $\Lambda^0$ and $\sqrt{2}$ for $\Sigma^0$, respectively, $a,~b,~c$ are the color indices, ${\cal C}$ is the charge conjugation operator. Saturating correlator (1) with the $Y~(=\Lambda^0$ or $\Sigma^0$) and $\Sigma^+$ states in the phenomenological part, we get \begin{eqnarray} \Pi= \frac{\left< \pi(q) Y \vert \Sigma^+ \right> \, \left< \Sigma^+(p+q) \vert \bar \eta_{\Sigma^+} \vert 0 \right> \, \left< 0\vert \bar \eta_Y \vert Y(p) \right>} {(p^2-m_Y^2) \left[(p+q)^2 - m_{\Sigma^+}^2\right]} + \mbox{\rm high. reson.} \end{eqnarray} The matrix elements in Eq. (3) are defined in the following way \begin{eqnarray} \left< 0 \left| \eta_Y(x) \right| Y(p) \right> &=& \lambda_Y u(p)~, \nonumber \\ \left< \Sigma^+(p+q) \left| \bar \eta_{\Sigma^+} \right| 0 \right> &=& \lambda_\Sigma \bar u(p+q)~,\nonumber \\ \left< \pi(q) Y(p)\vert \Sigma^+(p+q) \right> &=& - g_{Y\Sigma^+\pi^-} \, \bar u(p) \gamma_5 u(p+q)~. \end{eqnarray} Substituting Eq. (4) in Eq. (3), and choosing the structure $i\, \sigma_{\alpha\beta} p^\alpha q^\beta \gamma_5$ for the physical part of Eq. (1), we get \begin{eqnarray} \Pi^{phys}= - \frac{g_{\pi^-Y\Sigma^+} \lambda_Y \lambda_{\Sigma^+}} {(p^2-m_Y^2) \left[(p+q)^2 - m_\Sigma^2\right]} + \mbox{\rm higher resonances}~. \end{eqnarray} Let us now consider the theoretical part of the correlator (1). From this correlator we have (we present only the terms which give contributions to the above--mentioned Lorentz structure) \begin{eqnarray} \Pi &=& - \alpha \int d^4x\,e^{ipx} \Big\{ - \gamma_5 \gamma_\mu \gamma_5 \gamma_\varphi \gamma_\rho\, {\cal C}{\cal S}^T {\cal C}^{-1} \gamma_\mu {\cal S}^s \gamma_\rho \gamma_5 \left< \pi \left| \bar u \gamma_5 \gamma_\varphi d \right| 0 \right> \nonumber \\ &&+\frac{1}{2} \gamma_5 \gamma_\mu \sigma_{\alpha\beta} \gamma_\rho\, {\cal C}{\cal S}^T {\cal C}^{-1} \gamma_\mu {\cal S}^s \gamma_\rho \gamma_5 \left< \pi \left| \bar u \sigma_{\alpha\beta} d \right| 0 \right> \nonumber \\ && \pm \Big[ - \gamma_5 \gamma_\mu {\cal S} \gamma_\rho \,{\cal C} (\gamma_5 \gamma_\varphi)^T {\cal C}^{-1} \gamma_\mu {\cal S}^s \gamma_\rho \gamma_5 \left< \pi \left| \bar u \gamma_5 \gamma_\varphi d \right| 0 \right> \nonumber \\ &&+ \frac{1}{2} \gamma_5 \gamma_\mu {\cal S} \gamma_\rho \,{\cal C} \sigma_{\alpha\beta}^T {\cal C}^{-1} \gamma_\mu {\cal S}^c \gamma_\rho \gamma_5 \left< \pi \left| \bar u \sigma_{\alpha\beta}d \right| 0\right>\Big] \Big\} ~, \end{eqnarray} where upper(lower) sign corresponds to $\Lambda(\Sigma)$ case and $\alpha=\sqrt{2/3}$ for $\Lambda$ while $\alpha=\sqrt{2}$ for $\Sigma$. Here ${\cal S}$ and ${\cal S}^s$ are the propagators containing both perturbative and nonperturbative contributions, respectively. Here we present the explicit form of $i {\cal S}^s(x)$ \begin{eqnarray} \lefteqn{ i {\cal S}^s(x) = \frac{i}{2 \pi^2} \frac{\not\!x}{x^4} - \frac{m_s}{4\pi^2} \frac{1}{x^2} - \frac{1}{12} \langle \bar s s \rangle \left( 1 - \frac{i m_s}{4} \not\!x \right)} \nonumber \\ && - \frac{1}{192} m_0^2 \langle \bar s s \rangle \left( 1 - \frac{i m_s}{6} \not\!x \right) - i g_s \frac{1}{16 \pi^2} \int_0^1 du \left\{ \frac{\not\!x}{x^2} \sigma_{\alpha\beta} G^{\alpha\beta}(ux) - 4i \frac{x_\alpha}{x^2}G^{\alpha\beta} \gamma_\beta \right\}~, \end{eqnarray} where $m_s$ is the mass of the strange quark and $G_{\alpha\beta}$ is the gluon field strength tensor. The form of ${\cal S}$ can be obtained from Eq. (7) by making the replacements $\langle \bar s s \rangle \rightarrow \langle \bar q q \rangle$ and $m_s \rightarrow 0$. From Eq. (6) we observe that, in calculation of the correlator function in QCD, the matrix element of the nonlocal operators between the vacuum and pion states are needed. These matrix elements define the two particle pion wave functions and up to twist four they can be written as \cite{R6,R7} \begin{eqnarray} \left< \pi(q) \left| \bar d \gamma_\mu \gamma_5 u \right| 0 \right> &=& - i f_\pi q_\mu \int_0^1 du \, e^{iuqx} \big[ \varphi_\pi(u) + x^2 g_1(u) \big] \nonumber \\ &&+f_\pi \left( x_\mu - \frac{x^2 q_\mu}{qx} \right) \int_0^1 du\, e^{iuqx} g_2(u)~, \nonumber \\ \nonumber \\ \left< \pi(q) \left| \bar d (x) \sigma_{\alpha\beta}\gamma_5 u(0) \right| 0 \right> &=& i \left( q_\alpha x_\beta - q_\beta x_\alpha \right) \frac{f_\pi m_\pi^2}{6(m_u+m_d)} \int_0^1 du \,e^{iuqx} \varphi_\sigma(u)~. \end{eqnarray} Using Eqs. (6), (7) and (8) we get the following result for theoretical part (for the structure $i \sigma_{\alpha\beta} x_\alpha q^\beta \gamma_5$) \begin{eqnarray} \lefteqn{ \Pi^{theor} =} \nonumber \\ && - \alpha f_\pi \int_0^1 dx \,e^{ipx} \Bigg\{ \Bigg[ (1 \pm 1) \frac{m_s}{2 \pi^4 x^6} + (-\lambda \pm \sigma) \left( \frac{1}{6 \pi^2 x^4} + \frac{m_0^2}{96 \pi^2 x^2} \right) \Bigg] \int_0^1 du\, e^{iuqx} \varphi_\pi(u) \nonumber \\ &&+\Bigg[ (1 \mp 1) \frac{\mu_\pi}{6 \pi^4 x^6} + (1 \pm 1) \frac{\mu_\pi \left< g^2 G^2 \right>}{2304 \pi^4 x^2}+ (1 \mp 1) \frac{m_s \mu_\pi \left< \bar s s \right>}{144 \pi^2 x^2} \pm \frac{m_s \mu_\pi \sigma}{72 \pi^2 x^2} \Bigg] \int_0^1 du\, e^{iuqx} \varphi_\sigma(u) \nonumber \\ && +\Bigg[ \frac{1}{6 \pi^2 x^2} (-\lambda \pm \sigma) + (1 \pm 1) \frac{m_s}{2 \pi^4 x^4} \Bigg] \int_0^1 du\, e^{iuqx} \left[ g_1(u) + G_2 (u) \right] \Bigg\}~, \end{eqnarray} where \begin{eqnarray} G_2(u) &=& - \int_0^u g_2(v) dv~, \nonumber \\ \lambda &=& \langle \bar q q \rangle - \langle \bar s s \rangle~, \nonumber \\ \sigma &=& \langle \bar q q \rangle + \langle \bar s s \rangle~. \nonumber \end{eqnarray} Our next task is to perform integration over $x$ and perform double Borel transformation in Eq. (9) with respect to the variables $p^2$ and $(p+q)^2$, in order to get an answer for the theoretical part of the sum rules. As an example, let us demonstrate on one of the terms in Eq. (9) how integration over $x$ and double Borel transformation can be carried. Consider the following term \begin{eqnarray} \int du \varphi_\pi(u) \int d^4 x \frac{e^{i(p+qu)x}}{x^2} i \sigma_{\alpha\beta} x_\alpha q_\beta~. \nonumber \end{eqnarray} In performing the $x$ integration, we will make use of the formula \cite{R19} \begin{eqnarray} \int \frac{d^4 x}{(x^2)^n} e^{ipx} = \frac{i (-1)^n 2^{4-2n} \pi^2}{\Gamma(n-1) \Gamma(n)} (p^2)^{n-2} ln(-p^2) + {\cal P}_{n-2}~,~~~(n \ge 2 )~, \nonumber \end{eqnarray} where ${\cal P}_{n-2}$ is a polynomial of power $n-2$. However this polynomial is inessential, since it vanishes after double Borel transformation. Hence, disregarding this polynomial we have \begin{eqnarray} \int du \varphi_\pi(u) \int d^4 x \frac{e^{i(p+qu)x}}{x^2} i \sigma_{\alpha\beta} x_\alpha q_\beta &=& \int du \varphi_\pi(u) \left( i \frac{\partial}{\partial p_\alpha} \frac{\partial}{\partial p_\rho} \frac{\partial}{\partial p_\rho} \right) \int d^4 x \frac{e^{i(p+qu)x}}{x^4} i \sigma_{\alpha\beta} q_\beta \nonumber \\ &=& 8 \int du \varphi_\pi(u)\frac{P_\alpha}{P^4} i \sigma_{\alpha\beta} q_\beta ~, \nonumber \end{eqnarray} where $P=p + qu$. The last step in this calculation is performing double Borel transformation over the variables $p^2$ and $(p+q)^2$ to the expression \begin{eqnarray} \int du \varphi_\pi(u) \frac{1}{\left[ \left( p+qu \right)^2 \right]^2}~. \nonumber \end{eqnarray} Rewriting $(p+qu)^2 = p^2 \bar u + u (p+q)^2 $ (here the pion mass is neglected) and using the exponential representation for the denominator, \begin{eqnarray} \frac{1}{\left[ p^2 \bar u + u (p+q)^2 \right]^2} = \int_0^\infty d \alpha \alpha e^{-\alpha \left[ p^2 \bar u + u (p+q)^2 \right]}~, \nonumber \end{eqnarray} we have \begin{eqnarray} \int du \varphi_\pi(u) \frac{1}{\left[ \left( p+qu \right)^2 \right]^2} = \int du \varphi_\pi(u) \int d \alpha \alpha e^{-\alpha \left[ p^2 \bar u + u (p+q)^2 \right]}~. \nonumber \end{eqnarray} The double Borel transformation over the variable $p^2$ and $(p+q)^2$ is done with the help of the following general formula \begin{eqnarray} {\cal B}_{p^2}^{M_1^2} {\cal B}_{(p+q)^2}^{M_2^2} \frac{\Gamma(n)}{\left[ - \bar u p^2 - (p+q)^2 u \right]^n} = (M^2)^{2-n} \delta (u-u_0)~,\nonumber \end{eqnarray} where \begin{eqnarray} M^2 = \frac{M_1^2 M_2^2}{M_1^2+M_2^2}~~~~~\mbox{\rm and} ~~~~~ u_0 = \frac{M_1^2}{M_1^2+M_2^2}~,\nonumber \end{eqnarray} and $M_1^2$ and $M_2^2$ are the Borel parameters. Using this expression and performing the integrations over the variables $\alpha$ and $u$, we finally get \begin{eqnarray} \int du \varphi_\pi(u) \int d^4 x \frac{e^{i(p+qu)x}}{x^2} i \sigma_{\alpha\beta} x_\alpha q_\beta = 8 i \sigma_{\alpha\beta} p_\alpha q_\beta \varphi_\pi(u_0)~. \nonumber \end{eqnarray} All other terms can be calculated similarly and for the theoretical part it follows from Eq. (9) that \begin{eqnarray} \lefteqn{ \Pi^{theor} =} \nonumber \\ &&- \alpha f_\pi \Bigg\{ \varphi_\pi(u_0) \Bigg[ (1 \pm 1) \frac{m_s}{8 \pi^2} M^4 f_1(s_0/M^2) - \frac{1}{3} ( -\lambda \pm \sigma ) M^2 f_0(s_0/M^2) +\frac{m_0^2}{12}( -\lambda\pm\sigma ) \Bigg] \nonumber \\ && + \mu_\pi \varphi_\sigma(u_0) \Bigg[ (1 \mp 1) \frac{1}{24 \pi^2} M^4 f_1(s_0/M^2) + (1 \pm 1) \frac{8}{2304} \frac{\left< g^2 G^2 \right>}{\pi^2} + (1 \mp 1) \frac{1}{18} m_s \langle \bar s s \rangle \pm \frac{1}{9} m_s \sigma \Bigg] \nonumber \\ &&+ \left[ g_1(u_0) + G_2(u_0) \right] \Bigg[ -(1 \pm 1) \frac{1}{\pi^2} m_s M^2 f_0(s_0/M^2) + \frac{4}{3} \left( -\lambda \pm \sigma \right) \Bigg] \Bigg\}~, \end{eqnarray} where the function \begin{eqnarray} f_n(s_0/M^2)=1-e^{-s_0/M^2}\sum_{k=0}^n \frac{(s_0/M^2)^k}{k!}~, \nonumber \end{eqnarray} is the factor used to subtract the continuum, which is modeled by the dispersion integral in the region $s_1,~s_2 \ge s_0$, $s_0$ being the continuum threshold (obviously the continuum thresholds for the $\Lambda$ and $\Sigma$ channels are different). Since masses of $\Lambda$ and $\Sigma$ are very close to each other, we can choose $M_1^2$ and $M_2^2$ to be equal to each other, i.e., $M_1^2 = M_2^2 =2 M^2$, from which it follows that $u_0=1/2$. Performing double Borel transformation over the variables $p^2$ and $(p+q)^2$ in the physical part (5) and then equating the the obtained result to Eq. (10), we get the sum rules for $g_{\pi\Lambda\Sigma}$ and $g_{\pi\Sigma\Sigma}$ coupling constants \begin{eqnarray} g_{Y\Sigma^+\pi^-} \lambda_Y \lambda_{\Sigma^+} = e^{m^2/M^2} \Pi^{theor}~, \end{eqnarray} where $m\approx m_\Lambda \approx m_\Sigma$. From Eq. (11) it follows that in determining the strong coupling constants $g_{\pi\Lambda\Sigma}$ and $g_{\pi\Sigma\Sigma}$ the experimentally undetermined residues $\lambda_\Lambda$ and $\lambda_\Sigma$ need to be eliminated from sum rules. The residues $\lambda_\Lambda$ and $\lambda_\Sigma$ are determined from corresponding mass sum rules for the $\lambda$ and $\Sigma$ hyperons \cite{R18,R20} as follows \begin{eqnarray} \left| \lambda_\Lambda \right|^2 e^{-m_\Lambda^2/M^2} 32 \pi^4 &=& M^6 f_2(s_0^\Lambda/M^2) + \frac{2}{3} a m_s (1 - 3 \gamma) M^2 f_0(s_0^\Lambda/M^2)\nonumber \\ &&+ b M^2 f_0(s_0^\Lambda/M^2) + \frac{4}{9} a^2 (3+4 \gamma)~,\\ \left| \lambda_\Sigma \right|^2 e^{-m_\Sigma^2/M^2} 32 \pi^4 &=& M^6 f_2(s_0^\Sigma/M^2) - 2 a m_s (1+\gamma)M^2 f_0(s_0^\Sigma/M^2) \nonumber \\ &&+b M^2 f_0(s_0^\Sigma/M^2) + \frac{4}{3} a^2~, \end{eqnarray} where \begin{eqnarray} a &=& - 2 \pi^2 \langle \bar q q \rangle ~, \nonumber \\ b &=& \frac{\alpha_s \langle G^2 \rangle }{\pi} \simeq 0.012~GeV^4~, \nonumber \\ \gamma &=& \frac{\langle \bar s s \rangle}{\langle \bar q q \rangle} - 1 \simeq - 0.2~, \nonumber \end{eqnarray} and the functions $f_0(x),~f_1(x)$ are presented just after Eq. (10). The ratio of the Eqs. (11), (12) and (13) gives \begin{eqnarray} g_{\pi\Lambda\Sigma} &=& e^{m^2/M^2} \frac{\Pi^{theor}} {\lambda_\Lambda \lambda_\Sigma}~, \\ g_{\pi\Sigma\Sigma} &=& e^{m^2/M^2} \frac{\Pi^{theor}} {\lambda_\Sigma^2}~. \end{eqnarray} The main reason why we consider the above--mentioned ratio rather than the individual sum rules themselves (i.e., first determine $\lambda_\Lambda$ and $\lambda_\Sigma$ independently from Eqs. (12) and (13) and substitute their obtained values in Eq. (11)) is that the sum rules obtained from these ratios are more stable as is similar to the baryon mass sum rules case. In addition to that the uncertainties coming from various parameters such as quark condensate, $m_0^2$, continuum threshold $s_0$ and Borel parameter, are reduced. \section{Numerical analysis} Now we are ready to perform the numerical analysis. The main nonperturbative input parameters in the sum rules (11) are the pion wave functions. In our calculations we have used the set of wave functions proposed in \cite{R6}. The explicit expressions of the wave functions are \begin{eqnarray} \varphi_\pi(u,\mu) &=& 6 u \bar u \left[ 1 + a_2(\mu) C_2^{3/2} (2 u -1 ) + a_4(\mu) C_4^{3/2} (2 u -1 )\right]~, \nonumber \\ \varphi_\sigma(u,\mu) &=& 6 u \bar u \left[ 1 + C_2 \frac{3}{2} \left[5(u-\bar u)^2 -1 \right] + C_4 \frac{15}{8} \left[21(u-\bar u)^4 - 14 (u-\bar u)^2 + 1\right]\right]~, \nonumber \\ g_1(u,\mu) &=& \frac{5}{2} \delta^2(\mu)\bar u^2 u^2 + \frac{1}{2} \varepsilon(\mu) \delta^2(\mu) \Bigg[ u \bar u (2 + 13 u \bar u) \nonumber \\ && + 10 u^3 \ln u \left( 2 - 3u + \frac{6}{5} u^2 \right) + 10 \bar u^3 \ln \bar u \left( 2 - 3\bar u + \frac{6}{5}\bar u^2 \right)\Bigg]~, \nonumber \\ G_2(u,\mu) &=& \frac{5}{3} \delta^2(\mu) \bar u^2 u^2~, \end{eqnarray} where $\bar u = 1-u$, $C_2^{3/2}$ and $C_4^{3/2}$ are the Gegenbauer polynomials defined as \begin{eqnarray} C_2^{3/2} (2 u -1 ) &=& \frac{3}{2} \left[ 5(2 u-1)^2 +1 \right]~,\nonumber \\ C_4^{3/2} (2 u -1 ) &=& \frac{15}{8} \left[ 21 (2 u -1 )^4 - 14 (2 u -1 )^2 +1 \right] ~, \end{eqnarray} and $a_2(\mu=0.5~GeV)=2/3$, $a_4(\mu=0.5~GeV)=0.43$. The parameters $\delta^2(\mu=1~GeV)=0.2~GeV^2$ \cite{R21} and $\epsilon(\mu=1~GeV)=0.5$ \cite{R6}. Furthermore $f_\pi=0.132~GeV$, $\mu_\pi(\mu=1~GeV)=1.65$, $\langle \bar s s \rangle = 0.8 \langle \bar q q \rangle$ and $\langle \bar q q \rangle\vert_{\mu=1~GeV}=-(0.243)^3~GeV^3,~s_0 = s_0^\Lambda \simeq s_0^\Sigma = (3.0 \pm 0.2)~GeV^2$. Moreover all further calculations are performed at $u=u_0=1/2$. Having fixed the input parameters, one must find the range of values of $M^2$ over which the sum rule is reliable. The lowest possible value of $M^2$ is determined by the requirement that the terms proportional to the highest inverse power of the Borel parameters stay reasonable small. The upper bound of $M^2$ is determined by demanding that the continuum contribution is not too large. The interval of $M^2$ which satisfies both conditions is $1 ~GeV^2 < M^2 < 2.5~GeV^2$. The analysis of the sum rules (14) and (15) shows that the best stability is achieved in the region of $M^2$, $1.4~GeV^2 < M^2 < 1.8~GeV^2$. This leads to the following result for the coupling constants \begin{eqnarray} g_{\pi\Lambda\Sigma} &=& 5.3 \pm 1.8 \nonumber \\ g_{\pi\Sigma\Sigma} &=& 12.5 \pm 4.5~, \end{eqnarray} in which the errors coming from the quark condensate (which varies in the region $-(0.24~GeV)^3$ and $-(0.26~GeV)^3$), $m_0^2$ parameter (which varies in the region $0.6~GeV^2$ and $1.4~GeV^2$), variation of the Borel parameter and change of the continuum threshold $s_0$ are all taken into consideration. Our calculation shows that the main error comes from uncertainties of the quark condensate. The central values of the coupling constants are obtained at $m_0^2=0.8~GeV^2$, $\langle \bar q q \rangle = - (0.243~GeV)^3 $ and $s_0=3.0~GeV^2$. At this point we would like to stress that the above--mentioned strong coupling constant have been analyzed using the individual sum rules themselves by first determining $\lambda_\Lambda$ and $\lambda_\Sigma$ independently from Eqs. (12) and (13) and substituting their obtained values in Eq. (11). The results predicted by this approach are close to the ones presented in Eq. (18), however our calculations show that the ratio sum rules prediction is more stable and reliable. Here it should be noted that since the phase of the coupling constants can not be predicted by the sum rules, we take into consideration the magnitudes of these coupling constants to be able to compare them with the predictions of other approaches. Let us now compare our results on the coupling constants $g_{\pi\Lambda\Sigma}$ and $g_{\pi\Sigma\Sigma}$ with SU(3) symmetry prediction. As is known, SU(3) symmetry predicts \begin{eqnarray} G_{\pi\Lambda\Sigma} &=& \frac{2}{\sqrt{3}} (1-\alpha)\, G_{NN\pi} ~,\nonumber \\ G_{\pi\Sigma\Sigma} &=& 2 \alpha \,G_{NN\pi} ~, \end{eqnarray} where $\alpha = F/(F+D)$ (see for example \cite{R22}). Exact SU(3) symmetric analysis of pion--baryon coupling gives $F/D \simeq 0.58$ \cite{R22} (exact SU(6) symmetry predicts $F/D=2/3$). It follows from Eqs. (19) that \begin{eqnarray} R = \frac{G_{\pi\Sigma\Sigma}}{G_{\pi\Lambda\Sigma}} = \frac{\sqrt{3}\alpha}{1-\alpha} \approx 1~, \end{eqnarray} while our analysis yields $R\simeq 2$. It follows from a comparison of these results that SU(3) symmetry is broken significantly. The pion--baryon couplings predicted in \cite{R12} are not presented here since we have already noted that the results of this work is currently under debate in literature. In conclusion we have calculated the strong coupling constants of pion with $\Lambda$ and $\Sigma$ hyperons and found out that our results differ significantly from that of the SU(3) symmetry prediction. \newpage
\section{Introduction} Let $ \cal D $ be the Lie algebra of regular differential operators on ${ \Bbb C } \setminus \{0\}$, and ${ \widehat {\cal D} }= { \cal D } + { \Bbb C } C$ be the central extension of ${ \cal D }$. In the representation theory of the Lie algebra $ \widehat {\cal D} $ the most important are the irreducible quasi-finite highest weight modules. These modules were classified by V. Kac and A. Radul in \cite{KR1}. In \cite{FKRW} was shown that the language of vertex algebras is very useful in $ \widehat {\cal D} $--module theory. In particular, on the irreducible vacuum $ \widehat {\cal D} $--module $L(0,c, \widehat {\cal D} )$ exists a natural structure of a simple vertex algebra (cf. \cite{FKRW}, \cite{K}). This vertex algebra is usually denoted by $W_{1+\infty,c}$. The results from \cite{KR1} give that the representation theory of the vertex algebra $W_{1+\infty,c}$ is nontrivial only in the case $c \in { \Bbb Z }$. When $N \in { \Bbb N }$, the irreducible modules for $W_{1+\infty,N}$ were classified in \cite{FKRW}. The structure of the vertex algebra $W_{1+\infty,-N}$ is much complicated. The representation theory of $W_{1+\infty,-N}$ was began by Kac and Radul in \cite{KR2}. They realized $W_{1+\infty,-N}$ as a vertex subalgebra of the vertex algebra $V_N$ constructed from Weyl algebra $W_N$ (we recall this result in the Section \ref{reckr}). They also constructed a large class of irreducible $W_{1+\infty,-N}$--modules. The next step in this direction was made by Wang in \cite{W1}, \cite{W2}. He considered the case $c=-1$, and proved that $W_{1+\infty,-1}$ is isomorphic to tensor product $W_{3,-2} \otimes H_0$, where $W_{3,-2}$ is a vertex algebra associated with $W_3$--algebra with the central charge $-2$ and $H_0$ is a Heisenberg vertex algebra. He also classified all irreducible modules for $W_{3,-2}$ and $W_{1+\infty,-1}$. The representations obtained in \cite{W2} weren't identified as a highest weight $ \widehat {\cal D} $--modules. In the present paper we will construct $2 N$--dimensional family of irreducible $W_{1+\infty,-N}$--modules. Our family includes the modules constructed in \cite{KR2}, and also coincides with the modules from \cite{W2} in the case $c=-1$. Let us explain our result in more details. We consider lattice vertex superalgebra $V_L$ (cf. \ref{stwist}), and for suitably chosen lattice $L$ we show that $V_N$ and $W_{1+\infty,-N}$ are vertex subalgebras of $V_L$ (cf. Section \ref{wsec}). This fact is in physical literature known as a bosonization of $\beta \gamma$ system (see \cite{FMS}, \cite{W1}). The embedding of $W_{1+\infty,-N}$ into vertex superalgebra $V_L$ will imply that $W_{1+\infty,-N}$ can be realized as a vertex subalgebra of the Heisenberg vertex algebra $M(1)$ with $2 N$--generators. We explicitly identify the generators of $W_{1+\infty,-N}$ in terms of Schur polynomials. Considering $M(1)$--modules $M(1,\lambda)$ we obtain $W_{1+\infty,-N}$--modules $V(\lambda,-N)$ as a irreducible subquotients of $M(1,\lambda)$ (see Section \ref{wsec}). Considering $V(\lambda,-N)$ as a $ \widehat {\cal D} $--module, we identify its highest weights. As a result we get that all modules $V(\lambda,-N)$ are quasi-finite $ \widehat {\cal D} $--modules. \section{Vertex superalgebras} In this section we recall the definition of vertex (super)algebra and a few basic formulas. Fore more details about vertex (super)algebras its representations, and representation theory of certain examples of vertex (super)algebras see \cite{B}, \cite{DL}, \cite{FLM}, \cite{K}, \cite{Li}, \cite{A}. Let $V$ be a vector space. A {\it field} is a series of the form $$ a (z) = \sum_{n \in { \Bbb Z } } a_n z^{- n - 1}, $$ where $a_n \in \mbox{End} V$ are such that for each $v \in V$ one has: $a_n v = 0 \quad \mbox{for} \quad n \gg 0$. Here $z$ is a formal indeterminate. For a ${ \Bbb Z } _2$-graded vector space $W=W^{even}+W^{odd}$ we write $|u|\in { \Bbb Z }_2$, a degree of $u$, only for homogeneous elements: $|u|=0$ for an even element $u\in W^0$ and $|u|=1$ for an odd element $u\in W^1$. For any two $ \Bbb Z _2$-homogeneous elements $u$ and $v$ we define $\epsilon _{u,v}=(-1)^{|u||v|}\in { \Bbb Z }$. \begin{definition} A {\it vertex superalgebra} is a quadruple $(V,{ \bf 1 },D,Y)$, where $V=V^{even}\oplus V^{odd}$ is a $\Bbb{Z}_{2}$-graded vector space, $D$ is a $\Bbb{Z}_{2}$-endomorphism of $V$, ${ \bf 1 }$ is a specified vector called the {\it vacuum} of $V$, and $Y$ is a linear map \begin{eqnarray} Y(\cdot,z):& &V\rightarrow ({\rm End}V)[[z,z^{-1}]];\nonumber\\ & &a\mapsto Y(a,z)=\sum_{n\in \Bbb{Z}}a_{n}z^{-n-1}\;\;(\mbox{where } a_{n}\in {\rm End}V) \nonumber \end{eqnarray} such that % \begin{eqnarray} (V1)& &\mbox{For any }a,b\in V, a_{n}b=0\;\;\;\mbox{ for }n \mbox{ sufficiently large;}\nonumber \\ (V2)& &[D,Y(a,z)]=Y(D(a),z)=\partial_z Y(a,z)\;\;\mbox{ for any }a\in V;\nonumber \\ (V3)& &Y({ \bf 1 },z)=Id_{V}\;\;\;\mbox{(the identity operator of $V$)};\nonumber \\ (V4)& &Y(a,z){ \bf 1 }\in ({\rm End}V)[[z]] \mbox{ and }\lim_{z \rightarrow 0}Y(a,z){ \bf 1 }=a\;\;\mbox{ for any }a\in V;\nonumber \\ (V5)& &\mbox{For }\Bbb{Z}_{2}\mbox{ -homogeneous }a,b\in V, \mbox{ the following {\it Jacobi identity} holds:} \nonumber \end{eqnarray} \begin{eqnarray} & &\;\;\;z_{0}^{-1}\delta\left(\frac{z_{1}-z_{2}}{z_{0}}\right)Y(a,z_{1})Y(b,z_{2}) -\varepsilon_{a,b}z_{0}^{-1}\delta\left(\frac{z_{2}-z_{1}}{-z_{0}}\right) Y(b,z_{2})Y(a,z_{1})\nonumber \\ & &=z_{2}^{-1}\delta\left(\frac{z_{1}-z_{0}}{z_{2}}\right)Y(Y(a,z_{0})b,z_{2}).\nonumber \end{eqnarray} \end{definition} In the case $V= V^{even}$, i.e. when all vectors are even we say that $V$ is vertex algebra. From the Jacobi identity follows % \begin{eqnarray} \label{comut} \left[ a_{m}, b_{n} \right] = \sum^\infty_{j = 0} { m \choose j} (a_{j} b)_{m + n - j}, \quad m, n \in { \Bbb Z }, \end{eqnarray} \begin{eqnarray} \label{5.2} Y (a_{- 1} b, z) = : Y(a, z) Y(b, z):, \end{eqnarray} % where the normally ordered product is defined, as usual, by % $$ : Y (a, z) Y (b, z): = Y (a, z)_- Y(b, z) + Y (b, z) Y (a, z)_+ $$ % and % $$ Y(a, z)_+ = \sum_{j \in { \Bbb Z }_+} a_{j} z^{-j - 1}, \quad Y(a, z)_- = Y(a, z) - Y (a, z)_+ $$ % are the {\it annihilation\/} and the {\it creation\/} parts of $Y (a, z)$. \section{ The vertex algebra $W_{1 + \infty, - N}$ } \label{reckr} In this section we recall some of the results from \cite{KR1}, \cite{KR2}, \cite{FKRW}. Let $ \cal D $ be the Lie algebra of complex regular differential operators on ${ \Bbb C }^\times$ with the usual bracket, in an indeterminate $t$. The elements % $$ J^l (k) = - t^{l + k} (\partial_t)^l \quad (k \in { \Bbb Z }, l \in { \Bbb Z }_+) $$ % form a basis of ${ \cal D }$. The Lie algebra ${ \cal D }$ has the following 2-cocycle with values in ${ \Bbb C }$ : $$ \Psi (f (t) \partial_t^m, g (t) \partial_t^n) = \frac{m! n!}{(m + n + 1)!} \mbox{Res}_{t = 0} f^{(n + 1)} (t) g^{(m)} (t) dt, $$ % where $f^{(m)} (t) = \partial^m_t f (t)$. We denote by ${\widehat { \cal D } } = { \cal D } \oplus { \Bbb C } C$, where $C$ is the central element, the corresponding central extension of the Lie algebra ${ \cal D }$. Another important basis of ${ \cal D }$ is % $$ L^l (k) = - t^k D^l \quad (k \in { \Bbb Z }, l \in { \Bbb Z }_+) $$ % where $D = t \partial_t$. These two bases are related by the formula \cite{KR1}: % \begin{eqnarray} \label{veza} J^l (k) = - t^k D (D - 1) \cdots (D - l + 1). \end{eqnarray} % Given a sequence of complex numbers $\lambda = (\lambda_j)_{j \in { \Bbb Z }_+}$ and a complex number $c$ there exists a unique irreducible ${\widehat { \cal D } }$-module $L (\lambda, c ; {\widehat \cal D })$ which admits a non-zero vector $v_\lambda\/$ such that: % $$ L^j (k) v_\lambda = 0 \quad \mbox{for} \quad k > 0,\; L^j (0) v_\lambda = \lambda_j v_\lambda, \; C = c I. $$ % This is called a highest weight module over $\widehat { \cal D }$ with highest weight $\lambda\/$ and central charge $c$. The module $L (\lambda, c; \widehat \cal D )$ is called {\it quasifinite\/} if all eigenspaces of ${D}$ are finite-dimensional (note that $D\/$ is diagonalizable). It was proved in (\cite{KR1} Theorem 4.2) that $L (\lambda, c; \hat D)$ is a quasi-finite module if and only if the generating series $$ \Delta_\lambda (x) = \sum^\infty_{n = 0} \frac{x^n}{n!} \lambda_n $$ has the form $$ \Delta_\lambda (x) = \frac{\phi (x)}{e^x - 1}, $$ where $$ \phi (x) + c = \sum_i p_i (x) e^{r_i x} \quad \mbox{(a finite sum)}, $$ % $p_i (x)$ are non-zero polynomials in $x\/$ such that $\sum_i p_i (0) = c$ and $r_i\/$ are distinct complex numbers. The numbers $r_i\/$ are called the {\it exponents\/} of this module and the polynomials $p_i (x)$ are called their {\it multiplicities}. Recall now that the ${ \widehat {\cal D} } $-module $L(0, c; {\hat { \cal D } })$ has a canonical structure of a vertex algebra with the vacuum vector ${ \bf 1 } = v_0$ and generated by the fields $J^k (z) = \sum_{m \in { \Bbb Z } } J^k (m) z^{-m - k - 1}$ \cite{FKRW}. In \cite{KR2}, the vertex algebra $L(0,-N, \cal D )$ was realized by using Weyl algebra ${ W_N}$ and the corresponding vertex algebra $V_N$. The Weyl algebra ${ W_N}$ is an associative algebra over ${ \Bbb C }$ generated by $a_i (m), a^{*}_i (m)$ $ (i=1,\dots, N; m \in { \Bbb Z })$ and $C$ with the following defining relations $$ [a_i (m), a_j (n)]= [a^{*}_i (m) , a^{*}_j (n) ] = 0, \quad [a^{*}_i (m), a_j (n) ] = \delta_{i,j} \delta_{m,-n} C $$ for all $i,j \in \{ 1,\dots, N\}$, $n,m \in { \Bbb Z }$, and $C$ is central element. The vacuum ${ W_N}$--module $V_N$ is generated by one vector ${ \bf 1 }$, and the following relations $$ a_i (n) { \bf 1 } = 0, \quad n\ge 0; \quad a^{*}_i (n) { \bf 1 }= 0, \quad n> 0, \quad C { \bf 1 }= { \bf 1 }.$$ Define the following fields acting on $V_N$. % \begin{eqnarray} \label{polja} a_i (z ) = \sum_{n \in { \Bbb Z } } a(n) z^{-n-1}, \quad a^{*} _i (z) = \sum_{n \in { \Bbb Z } } a^{*} (n) z^{-n}. \end{eqnarray} Then there is a unique extensions of the fields (\ref{polja}) such that $V_N$ becomes a vertex algebra (see \cite{KR2}). Denote by $W_{1 + \infty, -N}$ a vertex subalgebra of $V_N$ generated by the fields $$ J^k (z) = - \sum^N _{i = 1} : a_i (z) \partial{^k} _{z} a^{*} _i (z): . $$ % \begin{proposition}(\cite{KR2}, Proposition 6.3) We have an isomorphism of vertex algebras: % $$ L (0, -N; {\hat { \cal D }}) \simeq W_{1 + \infty, -N} $$ % under which the fields (denoted by the same symbol) $J^k (z)$ correspond to each other. \end{proposition} \section{Lattice and Heisenberg vertex algebras } \label{stwist} Let $L$ be a lattice. Set ${ \frak h}={ \Bbb C }\otimes_{ \Bbb Z }L$ and extend the ${ \Bbb Z }$-form $ \langle \cdot, \cdot \rangle $ on $L$ to ${ \frak h}$. Let $\hat{{ \frak h}}={ \Bbb C }[t,t^{-1}]\otimes { \frak h} \oplus { \Bbb C }c$ be the affinization of ${ \frak h}.$ We also use the notation $h(n)=t^{n}\otimes h$ for $h\in { \frak h}, n\in { \Bbb Z }$. Set $ \hat{{ \frak h}}^{+}=t{ \Bbb C }[t]\otimes { \frak h};\;\;\hat{{ \frak h}}^{-}=t^{-1}{ \Bbb C }[t^{-1}]\otimes { \frak h}. $ Then $\hat{{ \frak h}}^{+}$ and $\hat{{ \frak h}}^{-}$ are abelian subalgebras of $\hat{{ \frak h}}$. Let $U(\hat{{ \frak h}}^{-})=S(\hat{{ \frak h}}^{-})$ be the universal enveloping algebra of $\hat{{ \frak h}}^{-}$. Let ${\lambda} \in { \frak h}$. Consider the induced $\hat{{ \frak h}}$-module \begin{eqnarray*} M(1,{\lambda})=U(\hat{{ \frak h}})\otimes _{U({ \Bbb C }[t]\otimes { \frak h}\oplus { \Bbb C }c)}{ \Bbb C }_{\lambda}\simeq S(\hat{{ \frak h}}^{-})\;\;\mbox{(linearly)},\end{eqnarray*} where $t{ \Bbb C }[t]\otimes { \frak h}$ acts trivially on ${ \Bbb C }$, ${ \frak h}$ acting as $\langle h, {\lambda} \rangle$ for $h \in { \frak h}$ and $c$ acts on ${ \Bbb C }$ as multiplication by 1. We shall write $M(1)$ for $M(1,0)$. For $h\in { \frak h}$ and $n \in { \Bbb Z }$ write $h(n) = t^{n} \otimes h$. Set $ h(z)=\sum _{n\in { \Bbb Z }}h(n)z^{-n-1}. $ Then $M(1)$ is vertex algebra which is generated by the fields $h(z)$, $h \in { \frak h}$, and $M(1,{\lambda})$, for $\lambda \in { \frak h}$, are irreducible modules for $M(1)$. Let $\hat{L}$ be the canonical central extension of $L$ by the cyclic group $\< \pm 1\>$: \begin{eqnarray}\label{2.7} 1\;\rightarrow \< \pm 1\>\;\rightarrow \hat{L}\;\bar{\rightarrow} L\;\rightarrow 1 \end{eqnarray} with the commutator map $c(\alpha,\beta)=(-1)^{\< \alpha,\beta\>}$ for $\alpha,\beta \in L$. Let $e: L\to \hat L$ be a section such that $e_0=1$ and $\epsilon: L\times L\to \<\pm 1\>$ be the corresponding 2-cocycle. Then $\epsilon(\alpha ,\beta )\epsilon(\beta ,\alpha )=(-1)^{\<\alpha ,\beta \>},$ \begin{equation}\label{2c} \epsilon (\alpha ,\beta )\epsilon (\alpha +\beta ,\gamma)=\epsilon (\beta ,\gamma)\epsilon (\alpha ,\beta +\gamma) \end{equation} and $e_{\alpha }e_{\beta }= \epsilon (\alpha ,\beta )e_{\alpha +\beta }$ for $\alpha ,\beta ,\gamma\in L.$ Form the induced $\hat{L}$-module \begin{eqnarray*} { \Bbb C }\{L\}={ \Bbb C }[\hat{L}]\otimes _{\< \pm 1\>}{ \Bbb C }\simeq { \Bbb C }[L]\;\;\mbox{(linearly)},\end{eqnarray*} where ${ \Bbb C }[\cdot]$ denotes the group algebra and $-1$ acts on ${ \Bbb C }$ as multiplication by $-1$. For $a\in \hat{L}$, write $\iota (a)$ for $a\otimes 1$ in ${ \Bbb C }\{L\}$. Then the action of $\hat{L}$ on ${ \Bbb C } \{L\}$ is given by: $a\cdot \iota (b)=\iota (ab)$ and $(-1)\cdot \iota (b)=-\iota (b)$ for $a,b\in \hat{L}$. Furthermore we define an action of ${ \frak h}$ on ${ \Bbb C }\{L\}$ by: $h\cdot \iota (a)=\< h,\bar{a}\> \iota (a)$ for $h\in { \frak h},a\in \hat{L}$. Define $z^{h}\cdot \iota (a)=z^{\< h,\bar{a}\> }\iota (a)$. The untwisted space associated with $L$ is defined to be \begin{eqnarray*} V_{L}={ \Bbb C }\{L\}\otimes _{{ \Bbb C }}M(1)\simeq { \Bbb C }[L]\otimes S(\hat{ \frak h} ^{-})\;\;\mbox{(linearly)}. \end{eqnarray*} Then $\hat{L},\hat{{ \frak h}},z^{h}\;(h\in { \frak h})$ act naturally on $V_{L}$ by acting on either ${ \Bbb C }\{L\}$ or $M(1)$ as indicated above. Define ${ \bf 1 }= \iota ( e_0) \in V_L$. We use a normal ordering procedure, indicated by open colons, which signify that in the enclosed expression, all creation operators $h(n)$ $(n<0)$,$a\in \hat{L}$ are to be placed to the left of all annihilation operators $h(n),z^{h}\;(h\in { \frak h},n\ge 0)$. For $a \in \hat{L}$, set \begin{eqnarray*} Y(\iota (a),z)= : e^{\int (\bar{a}(z)-\bar{a}(0)z^{-1})}az^{\bar{a}}:. \end{eqnarray*} Let $a\in \hat{L};\;h_{1},\cdots,h_{k}\in { \frak h};n_{1},\cdots,n_{k}\in { \Bbb Z }\;(n_{i}> 0)$. Set \begin{eqnarray*} v= \iota (a)\otimes h_{1}(-n_{1})\cdots h_{k}(-n_{k})\in V_{L}.\end{eqnarray*} Define vertex operator $Y(v,z)$ with \begin{eqnarray} \label{defvertex} :\left({1\over (n_{1}-1)!}({d\over dz})^{n_{1}-1}h_{1}(z)\right)\cdots \left({1\over (n_{k}-1)!}({d\over dz})^{n_{k}-1}h_{k}(z)\right)Y(\iota (a),z): . \end{eqnarray} This gives us a well-defined linear map \begin{eqnarray*} Y(\cdot,z):& &V_{L}\rightarrow (\mbox{End}V_{L})[[z,z^{-1}]]\nonumber\\ & &v\mapsto Y(v,z)=\sum _{n\in { \Bbb Z }}v_{n}z^{-n-1},\;(v_{n}\in {\rm End}V_{L}). \end{eqnarray*} Let $\{\; h_{i}\;|\;i=1,\cdots,d\}$ be an orthonormal basis of ${ \frak h}$ and set \begin{eqnarray*} \omega ={1\over 2}\sum _{i=1}^{d} h_{i}(-1) h_{i}(-1)\in V_{L}. \end{eqnarray*} Then $Y(\omega,z)=\sum_{n\in { \Bbb Z }}L_n z^{-n-2}$ gives rise to a representation of the Virasoro algebra on $V_{L}$ with the central charged $d$ and \begin{eqnarray} \label{vir.rel} & &L_0\left(\iota(a)\otimes h_{1}(-n_{1})\cdots h_{n}(-n_{k})\right)\nonumber \\ &=&\left({1\over 2}\< \bar{a},\bar{a}\>+n_{1}+\cdots+n_{k}\right) \left(\iota(a)\otimes h_{1}(-n_{1})\cdots h_{k}(-n_{k})\right). \end{eqnarray} The following theorem was proved in \cite{DL} and \cite{K}. \begin{theorem} $(V_L, { \bf 1 }, L_{-1}, Y)$ is vertex superalgebra. \end{theorem} Vertex algebra $M(1)$ can be treated as a subalgebra of $V_L$. Define the Schur polynomials $p_{r}(x_{1},x_{2},\cdots)$ $(r\in { \Bbb Z }_{+})$ in variables $x_{1},x_{2},\cdots$ by the following equation: \begin{eqnarray}\label{eschurd} \exp \left(\sum_{n= 1}^{\infty}\frac{x_{n}}{n}y^{n}\right) =\sum_{r=0}^{\infty}p_{r}(x_1,x_2,\cdots)y^{r}. \end{eqnarray} For any monomial $x_{1}^{n_{1}}x_{2}^{n_{2}}\cdots x_{r}^{n_{r}}$ we have an element $h(-1)^{n_{1}}h(-2)^{n_{2}}\cdots h(-r)^{n_{r}}{ \bf 1 }$ in both $M(1)$ and $V_{L}$ for $h\in{ \frak h}.$ Then for any polynomial $f(x_{1},x_{2}, \cdots)$, \\ $f(h(-1), h(-2),\cdots){ \bf 1 }$ is a well-defined element in $M(1)$ and $V_{L}$. In particular, $p_{r}(h(-1),h(-2),\cdots){ \bf 1 }$ for $r\in { \Bbb Z }_{+}$ are elements of $M(1)$ and $V_{L}$. Suppose $a,b\in \hat{L}$ such that $\bar{a}=\alpha,\bar{b}=\beta$. Then \begin{eqnarray}\label{erelation} Y(\iota(a),z)\iota(b)&=&z^{\<\alpha,\beta\>}\exp\left(\sum_{n=1}^{\infty} \frac{\alpha(-n)}{n}z^{n}\right)\iota(ab)\nonumber\\ &=&\sum_{r=0}^{\infty}p_{r}(\alpha(-1),\alpha(-2),\cdots)\iota(ab) z^{r+\<\alpha,\beta\>}. \end{eqnarray} Thus \begin{eqnarray}\label{eab1} \iota(a)_{i}\iota(b)=0\;\;\;\mbox{ for }i\ge -\<\alpha,\beta\>. \end{eqnarray} Especially, if $\<\alpha,\beta\>\ge 0$, we have $\iota(a)_{i}\iota(b)=0$ for all $i\in { \Bbb Z }_{+}$, and if $\<\alpha,\beta\>=-n<0$, we get \begin{eqnarray}\label{eab} \iota(a)_{i-1}\iota(b)=p_{n-i}(\alpha(-1),\alpha(-2),\cdots)\iota(ab) \;\;\;\mbox{ for }i\in { \Bbb Z }_{+}. \end{eqnarray} We will need one structural result on Heisenberg vertex algebras. Element $L_0$ of the Virasoro algebra defines a ${ \Bbb Z }_+$--graduation on vertex algebra $M(1)= \oplus_{n \in { \Bbb Z }_+ } M(1)_n$. Let $v_{\lambda}$ be the highest weight vector in the $M(1)$--module $M(1, \lambda)$. The following lemma can be proved by using standard calculation in the Heisenberg vertex algebras. \begin{lemma} \label{standard} Let $h \in { \frak h}$, and $n_1, \dots, n_r \in { \Bbb N }$. Let $k= n_1 + \cdots + n_r$. Let $u= h (-n_1) \cdots h (-n_r) { \bf 1 }$, and $ Y( u, z) = \sum_{n \in { \Bbb Z } } u_{n} z^{-n-1}$. Then $u \in M(1)_k$, and we have \begin{eqnarray} (1) && u_n v_{\lambda} = 0 \quad\mbox{for} \ n> k-1, \nonumber \\ (2) && u_{k-1} v_{\lambda} = (-1) ^{n_1 + \cdots + n_r +r}( \langle \lambda, h \rangle ) ^{r} v_{\lambda}. \nonumber \end{eqnarray} \end{lemma} \begin{proposition} \label{schur1} Let $h \in { \frak h}$, and $r\in { \Bbb Z }_+$. Let $u = p_{r}(h(-1),h(-2),\cdots){ \bf 1 } $. Set $ Y( u,z) = \sum _{n \in { \Bbb Z } } u_n z^{-n-1}$. Then $u \in M_r$, and we have \begin{eqnarray} (1) && u_n v_{\lambda} = 0 \quad \mbox{for} \ n>r-1, \nonumber \\ (2)&& u_{r-1} v_{\lambda} = { \langle \lambda, h \rangle \choose r} v_{\lambda}. \end{eqnarray} \end{proposition} {\em Proof.} Since $u \in M(1)_r$, we have that $u_n v_{\lambda} = 0$ for $n > r$. Set $x=\langle \lambda , h \rangle$. Using Lemma \ref{standard} one can easily see that $$ u_{r-1} v_{l}= p_r (x, -x, x,-x,\cdots ) v_{\lambda}.$$ Since $$ \exp \left( \sum_{n=1} ^{\infty} \frac{(-1) ^{n-1} x}{n} y^{n} \right) = ( 1+y) ^{x} = \sum_{r \ge 0} {x \choose r} y^{r},$$ we have that $ p_r (x, -x, x,-x,\cdots ) = {x \choose r}$, and we conclude that $u_{r-1} v_{\lambda}= {x \choose r}$. \qed \begin{remark} Proposition \ref{schur1} can be also proved by using Zhu's algebra theory (see \cite{DLM}, Section 3.) \end{remark} \section{Representations of the vertex algebra $W_{1+\infty, -N}$ } \label{wsec} In this section we will construct $2N$--dimensional family of irreducible modules for the vertex algebra $W_{1+\infty, -N}$. We will prove that $W_{1+\infty, -N}$ can be realized as a vertex subalgebra of Heisenberg vertex algebra $M(1)$. First we will consider the following lattice: \begin{eqnarray} && L=\bigoplus_{i= 1} ^{N} \left( { \Bbb Z } {\alpha }_i + { \Bbb Z } {\beta } _i \right), \nonumber \quad \mbox{where}\\ && \langle {\alpha }_i , {\alpha }_j \rangle = \delta_{i,j}, \quad \langle {\beta }_i , {\beta }_j \rangle = - \delta_{i,j}, \quad \langle {\alpha }_i, {\beta }_j \rangle = \langle {\beta }_j , {\alpha }_i \rangle = 0, \nonumber \end{eqnarray} for every $i,j \in \{1,\cdots,N\}$. Let ${ \frak h}$, $M(1)$ and $V_L$ be defined as in the Section \ref{stwist}. Then ${ \frak h}$ is a Heisenberg algebra, $\mbox{dim} { \frak h} = 2 N$, $M(1)$ is a Heisenberg vertex algebra, and $V_L$ is a lattice vertex superalgebra. For every $ i\in \{1, \cdots, N\}$, let $a^{i}, b^{i} \in {\hat L}$ such that $${\overline{a^{i} } } = \alpha _i + \beta _i, \quad {\overline{b^{i} }}=- (\alpha _i + \beta _i).$$ Then we define $e^{i}, f^{i} \in V_L$ with $ e^{i} = \iota( a^{i} ), f^{i} = \iota( b^{i})$. Set $Y( e^{i}, z) = \sum_{n \in { \Bbb Z } } e^{i} _n z^{-n-1}$, and $Y( f^{i}, z) = \sum_{n \in { \Bbb Z } } f^{i} _n z^{-n-1}$. Recall the definition of Schur polynomial $p_l ( h(-1),h(-2), \dots)$ ($h\in { \frak h}$ ) from Section \ref{stwist}. Set $p_l (h) := p_l ( h(-1),h(-2), \dots) \in S({\hat{ \frak h}} ^{}-)$. Let $\gamma_i = \alpha _i + \beta _i$. Then $ \langle \gamma_i, \gamma_i \rangle =0$. Now, relations (\ref{eab1}) and (\ref{eab}) imply the following lemma. \begin{lemma} \label{fms1} For all $i,j \in \{1, \cdots, N\}$, $k,n \in { \Bbb Z }$ we have : \begin{eqnarray} (1) && e^{i} _n e^{j} = f^{i} _n f^{j}= 0, \quad \mbox{for} \ \ n \ge 0, \nonumber \\ (2) && e^{i} _{-1} f^{j} = \delta_{i,j} { \bf 1 }, \quad f^{i} _{-1} e^{j}= \delta_{i,j} { \bf 1 }, \nonumber \\ (3) && e^{i} _{-l-1} f^{i} = p_l( \gamma_i) { \bf 1 }, \quad f^{i} _{-l-1} e^{i}= p_l (-\gamma_i) { \bf 1 }, \nonumber \\ (4)&& [ \alpha _i (k), e^{j} _n ]= \delta_{i,j} e^{j} _{n+k}, \quad [ \alpha _i (k), f^{j} _n ]= -\delta_{i,j} f^{j} _{n+k}. \end{eqnarray} \end{lemma} Define \begin{eqnarray} &&A^{i} ( z) = \sum_{n \in { \Bbb Z } } A^{i} (n) z^{-n-1}=Y(e^{i}\otimes \alpha_i (-1), z ) = \ :\alpha_i (z) Y(e^{i} , z) :, \nonumber \\ && {{\bar A} }^{i} (z) = \sum_{n \in \Bbb Z } {{\bar A} } ^{i} (n) z^{-n}= Y(f^{i} ,z) . \nonumber \end{eqnarray} \begin{lemma} \label{fms2} We have % \begin{eqnarray} && [A^{i} (n), A ^{j} (m)]= [{{\bar A} } ^{i} (n), {{\bar A} } ^{j} (m)]= 0, \nonumber \\ && [{{\bar A} } ^{i} (n), A^{j} (m) ] = \delta_{i,j} \delta_{m+m,0}, \nonumber \end{eqnarray} i.e. the components of the fields $A^{i}(z), {{\bar A} } ^{i} (z)$, $1\le i \le N$, span an associative algebra isomorphic to the Weyl algebra $W_N$ with $C=1$. \end{lemma} {\em Proof.} Using (\ref{5.2}) we have that $$ A^{i}(m) = \sum_{ n < 0} \alpha _i (n) e^{i }_{ m-n-1} + \sum_{n \ge 0} e^{i}_{m-n-1} \alpha _i (n) $$ for $1\le i \le N$. Using Lemma \ref{fms1} we get \begin{eqnarray} && A^{i} (0) f^{j}= -\delta_{i,j} f^{j}, \quad A^{i} (m) f^{j} = 0 \quad \mbox{for} \ m > 0, \nonumber \\ && A^{i} (m) A^{j} (-1){ \bf 1 } = 0, \quad f^{i} _m f^{j} = 0, \quad \mbox{for} \ m\ge 0. \nonumber \end{eqnarray} Now, the statment of the lemma follows from the commutator formulae (\ref{comut}). \qed From Lemma \ref{fms2} follows the following result. \begin{proposition} \label{wizom}The subalgebra of the vertex superalgebra $V_L$ generated by the fields $A^i (z)$, ${{\bar A} } ^{i} (z) $ $i=1, \dots, N$, is isomorphic to the vertex algebra $V_N$. (Under this isomorphism the fields $a_i (z) $ corresponds to $A^{i} (z)$, and $a^{*} _i (z)$ to $\bar{A} ^{i}(z)$.) \end{proposition} {\em Proof.} From the Lemma \ref{fms2} follows that $V_L$ is a module for the Weyl algebra ${ W_N}$. The subalgebra of $V_L$ generated by the fields $A^i (z)$ and ${{\bar A} } ^{i} (z) $ is exactly the ${ W_N}$ submodule ${ W_N} . { \bf 1 }$ generated by ${ \bf 1 }$. Then we have $$ A^{i} (m) { \bf 1 } = {{\bar A} } ^{i} (n) { \bf 1 } = 0, \quad \mbox{for} \ m\ge0, \ n > 0.$$ Then the fact that $V_N$ is an irreducible ${ W_N}$--module implies that $V_N \cong { W_N}. { \bf 1 } $. \qed \begin{remark} In the physical literature the vertex algebra $V_N$ is known as $\beta \gamma$ system, and the lattice construction of $\beta \gamma$ system is known as Friedan-Martinec-Shenker bosonization (cf. \cite{FMS}, \cite{W1}). \end{remark} With the respect to Proposition \ref{wizom} we can identify vertex algebra $V_N$ with the subalgebra of $V_L$ generated by the fields $A^{i}(z)$, $\bar{A} ^{i}(z)$. Since $W_{1 +\infty,-N}$ is a vertex subalgebra of $V_N$ generated by the fields $$ J^{k} (z) = - \sum_{ i=1} ^{N} : A ^{i} (z)\partial_z ^{k} {{\bar A} } ^{i} (z):$$ we have that $W_{1 +\infty,-N}$ is also a vertex subalgebra of $V_L$. For every $i \in \{1, \cdots, N\}$, let $U^{i}_{k}= A^{i} (-1)f^{i} _{-k-1} { \bf 1 } \in V_L$, and $U_{k}= \sum_{i=1} ^{N} U^{i} _{k}$ . We have $$ J^{k} (z) = -k! \sum_{i=1} ^{N } Y ( A^{i} (-1) {{\bar A} }^{i} (-k) { \bf 1 },z ) =-k! Y(U_k,z).$$ \begin{lemma} \label{ul1} We have \begin{eqnarray} &&U^{i} _{k} = A^{i}(-1) {{\bar A} }^{i} (-k) { \bf 1 }= \alpha _i (-1) p_k (-\gamma_i) { \bf 1 } + p_{k+1} (-\gamma_i) { \bf 1 }. \nonumber \end{eqnarray} In particular, $U^{i}_{k} \in M(1)$. \end{lemma} {\em Proof.} Since $[A^{i} (-1), {{\bar A} }^{i} (-k)] = 0$, we have that $$ U^{i} _k = {{\bar A} }^{i} (-k)A^{i}(-1) { \bf 1 } = f^{i} _{-k-1} \alpha _i (-1) e^{i}= \alpha _i (-1) f^{i}_{-k-1} e^{i} + f^{i} _{-k-2} e^{i}.$$ Then Lemma \ref{fms1} implies that $$ U^{i} _{k} = \alpha _i (-1) p_k (-\gamma_i) { \bf 1 } + p_{k+1} (-\gamma_i) { \bf 1 }, $$ and lemma holds. \qed \begin{theorem} Vertex algebra $W_{1+ \infty, -N}$ is a subalgebra of the vertex algebra $M(1)$. \end{theorem} {\em Proof.} From Lemma \ref{ul1} follows that $U_{k}= \sum_{i=1} ^{N} U^{i} _{k} \in M(1)$, for every $k \in { \Bbb Z }_+$. Since $J^{k} (z) = - k! Y(U_{k},z)$, $k \in { \Bbb Z }_+$, generate $W_{1+ \infty, -N}$, we have that $W_{1+ \infty, -N}$ is a subalgebra of the vertex algebra $M(1)$. \qed \begin{lemma} \label{hw} For every $\lambda \in { \frak h}$, let $v_{\lambda}$ be the highest weight vector in $M(1,\lambda)$. Then we have $ J ^{k}(n) v_{\lambda} = 0$ for $ n > 0$, and $$ J ^{k} (0) v_ {\lambda} = - k! \sum_{i=1} ^{N} \left( {-\langle \lambda , \gamma_i \rangle \choose {k+1} } + \langle \lambda, \alpha _i \rangle {-\langle \lambda, \gamma_i \rangle \choose k} \right) v_{\lambda} . $$ \end{lemma} {\em Proof.} Set $Y( U^{i} _k, z) = \sum_{n \in { \Bbb Z } } U^{i} _k (n) z^{-n-k-1}$. Proposition \ref{schur1}, and Lemma \ref{ul1} implies that \begin{eqnarray} \label{ul2}&& U^{i} _k (n) v_{\lambda} = 0 \quad \mbox{for} \ n >0, \nonumber \\ && U^{i} _k (0) v_{\lambda} = \langle \lambda, \alpha _i \rangle {-\langle \lambda , \gamma_i \rangle \choose k} + {-\langle \lambda, \gamma_i \rangle \choose k+1}. \nonumber \end{eqnarray} Since $ J^{k} (n) = - k!\sum_{i=1} ^{N} U^{i} _k (n)$, we have that the lemma holds. \qed Since $W_{1+ \infty, -N}$ is a subalgebra of VOA $M(1)$, then for every $\lambda \in { \frak h}$, $M(1, \lambda)$ is a $W_{1+ \infty, -N}$--module. Let $V(\lambda,-N)$ be the irreducible $W_{1+ \infty, -N}$--subquotient of $M(1, \lambda)$ generated by the vector $v_{\lambda}$. \begin{theorem} \label{main} For every $\lambda \in { \frak h} $ $ {V} ({\lambda},-N)$ is the irreducible module for the vertex algebra $W_{1 +\infty, - N}$. As a $ \widehat {\cal D} $--module, $V(\lambda,-N)$ is a irreducible quasi-finite highest weight module, and the corresponding generating series is \begin{eqnarray} \label{gener} \Delta_{\lambda} (x) =-\sum_{i=1} ^{N} \left(\frac{e^{ s_i x} -1 }{e^{x} -1} + t_i e^{s_i x} \right) , \end{eqnarray} where \begin{eqnarray} \label{defl}s_i = -\langle \lambda, \alpha _i + \beta _i \rangle, \quad t_i = \langle \lambda, \alpha _i \rangle, \quad i=1,\dots, N. \end{eqnarray} \end{theorem} {\em Proof.} Lemma \ref{hw} implies that ${V}({\lambda},-N)$ is a highest weight $ \widehat {\cal D} $--module. It remains to prove that ${ V}({\lambda},-N)$ is a quasi-finite $ \widehat {\cal D} $--module. In order to prove this we have to identify the generating series $\Delta_{\lambda} (x)$. Using the relation (\ref{veza}), it is straightforward to see the following formulae. \begin{eqnarray} \Delta_{\lambda} (x) = \sum_{k=0} ^{\infty} \frac{x^{k}}{k!} L^{k} (0) v_{\lambda} = \sum_{k=0} ^{\infty} \frac{(e^{x} -1)^{k}}{k!} J^{k} (0) v_{\lambda}. \nonumber \end{eqnarray} Let $s_i, t_i$ be defined with (\ref{defl}). Then Lemma \ref{hw} implies \begin{eqnarray} \Delta_{\lambda} (x) = &&-\sum_{i=1} ^{N}\sum_{k=0} ^{\infty} \left( {s_i \choose {k+1} } + t_i {s_i \choose k} \right) (e^{x} -1)^{k} \nonumber \\ = && -\sum_{i=1} ^{N} \left(\frac{e^{s_i x} -1 }{e^{x} -1} + t_i e^{s_i x} \right) \nonumber \\ =&& \frac{\Phi (x) }{e^{x} -1}, \nonumber \end{eqnarray} where $$ \Phi(x) -N = \sum_{i=1} ^{N} \left( (t_i -1 ) e^{s_i x} -t_i e^{(s_i +1) x}\right).$$ This implies that ${V}(\lambda,-N)$ is a quasi-finite $ \widehat {\cal D} $--module. \qed \begin{remark} Theorem \ref{main} gives the existence of $2 N$--dimensional family of irreducible $W_{1+\infty,-N}$--modules. If we take in (\ref{gener}) $t_i=0$ for every $i =1, \dots, N$, we get exactly $W_{1+\infty,-N}$--modules constructed in \cite{KR2}. \end{remark} We have the following conjecture. \begin{conjecture} \label{slutnja} The set $V(\lambda,-N)$, $\lambda \in { \frak h}$, lists all the irreducible modules for the vertex algebra $W_{1+\infty,-N}$. \end{conjecture} \begin{remark} In Section \ref{sec-1} we will see that the Conjecture \ref{slutnja} is true for $c=-1$. \end{remark} \section{ The case of $c=-1$ } \label{sec-1} In this section we will compare our results with the results from \cite{W1}, \cite{W2}. In \cite{W1}, Wang proved that the vertex algebra $W_{1+\infty,-1}$ is isomorphic to the tensor product $W_{3,-2} \otimes H_0$, where $W_{3,-2}$ is a simple vertex algebra associated to $W_3$--algebra with $c=-2$, and $H_0$ is a Heisenberg vertex algebra. Moreover, in \cite{W2} Wang classified all the irreducible modules for $W_{3,-2}$ and $W_{1+\infty,-1}$. The methods used in \cite{W1}, \cite{W2} didn't imply the identification of $W_{1+\infty,-1}$--modules as a highest weight $ \widehat {\cal D} $--modules (see Section 5 in \cite{W2}). Our approach gives an explicit identification of two-dimensional family $W_{1+\infty,-1}$--modules in terms of highest weights. Let $N=1$, and set $\alpha = \alpha _1$, $\beta = \beta _1$. For $\lambda \in { \frak h}$ set $\lambda_{\alpha } = \langle \lambda, \alpha \rangle$, $\lambda_{\beta } = \langle \lambda , \beta \rangle$. Let $M_{\alpha }(1, \lambda_{\beta })$ (resp. $M_{\beta } (1, \lambda_{\beta })$ ) be the submodules of $M(1,\lambda)$ generated by the highest weight vector $v_{\lambda}$ and $\alpha (n)$ (resp. $\beta (n)$). Set $M_{\alpha } (1) = M_{\alpha }(1,0)$, $M_{\beta } (1) = M_{\beta }(1,0)$. Then \begin{eqnarray} \label{m1ab} M(1) = M_{\alpha } (1)\otimes M_{\beta } (1), \quad M(1,\lambda) = M_{\alpha } (1, \lambda_{\alpha } )\otimes M_{\beta } (1, \lambda _{\beta }). \end{eqnarray} As in \cite{W2} we define \begin{eqnarray} && T(z) = \frac{1}{2} \left( :\alpha (z) ^{2} : + \partial\alpha (z) \right), \nonumber \\ && W(z) = \frac{1}{12} \left( 4 : \alpha (z) ^{3} : + 6 : \alpha (z) \partial\alpha (z) : + \partial^{2} \alpha (z) \right). \nonumber \end{eqnarray} \begin{theorem} \label{wang} \cite{W1}, \cite{W2} \item[(1)] The fields $T(z)$, $W(z)$ span a subalgebra of $M_{\alpha }(1)$ isomorphic to $W_{3,-2}$, and $W_{1+\infty,-1}\cong W_{3,-2} \otimes M_{\beta } (1)$. \item[(2)] Let ${\cal V}_r$ be the irreducible subquotient of $W_{3,-2}$--module $M_{\alpha }(1,r)$. Then ${\cal V}_r$, $r \in { \Bbb C }$, gives all the irreducible $W_{3,-2}$--modules. \end{theorem} Recall the definition of $W_{1+\infty,-1}$--modules $V(\lambda,-1)$ from Section \ref{wsec}. Then we have the following consequence of Theorem \ref{main} and Theorem \ref{wang}. \begin{corollary} We have \item[(1)] $ {V}(\lambda,-1) \ = {\cal V}_{\lambda_{\alpha } } \otimes M_{\beta } (1, \lambda_{\beta } )$ \ \ for every $\lambda \in { \frak h}$. \item[(2)]The set $V(\lambda,-1)$, $\lambda \in { \frak h}$, gives all irreducible $W_{1+\infty,-1}$--modules. \item[(3)] $$\Delta_{\lambda} (x) =-\frac{e ^{-(\lambda_ {\alpha } + \lambda_{\beta }) x }-1}{e^{x}-1}- \lambda_{\alpha } e^{- (\lambda_ {\alpha } + \lambda_{\beta }) x}.$$ \end{corollary} {\em Proof.} (1) follows from the definition of ${\cal V}_{\alpha }$ and (\ref{m1ab}). Then theorem \ref{wang} implies that ${\cal V}_r \otimes M_{\beta } (1,s)$, $r,s \in { \Bbb C }$, are all irreducible $W_{1+\infty,-1}$--modules. Since $ {V}(\lambda,-1) \ = {\cal V}_{\lambda_{\alpha } } \otimes M_{\beta } (1, \lambda_{\beta } )$, we see that $V(\lambda,-1)$, $\lambda \in \frak h$ gives all irreducible $W_{1+\infty,-1}$--modules, and we get (2). The statement (3) follows from the Theorem \ref{main}. \qed