content
stringlengths 1
15.9M
|
|---|
\section{Introduction}
Recently a significant attention is paid to the studying of soliton equations under external perturbations. For KdV equation, in particular, it was shown, that simple sine driven force together with dissipation can lead to a chaotic dynamics, detected by Fourier spectrum, positive Liapunov exponents and phase plane analysis \cite{paper1,paper2}. In this context the reduction of a perturbed KdV equation in the form of stationary travelling waves was considered. The justification of the application of low-dimensional chaos theory to such an infinite-dimensional system as KdV equation is in a numerical investigation, which proves that information dimension, calculated by Kaplan-Yorke formula, and correlation dimension within Grassberger-Procaccia method, are both between 2 and 3 for steady waves \cite{paper3}. On the other hand, perturbed KdV equation can be considered directly whithin numerical approach, or by perturbation technique \cite{paper4}. Besides this series of work it is worth to mention here the cases, when KdV equation is influenced by dissipation and instability, wihtout explicit spatial or temporal dependence \cite{in-proc1}. In such a form this equation describes current-driven ion-acoustic instability in a collision-dominated plasma. For the strongly dissipative case overall evolution demonstrates irregular behaviour, therefore leading to nonstationary and irregular soliton interactions.
In this work we add hamiltonian deterministic perturbation to the KdV equation and study physical consequences of this. The reduction in the form of travelling waves is made in order to turn the system into a second-order ordinary differential equation. The existence of solutions to such an equation was studied in \cite{paper5}, while the construction of periodic orbits can be found in \cite{paper2} for nonresonant and primary resonant cases, and in \cite{paper1} for secondary resonances. We'll, however, perform the analysis of chaotic properties in this situation. It is known, that for any hamiltonian perturbation, which makes the system near-integrable, a stochastic layer around a separatrix appears. This layer is bounded by unbroken KAM-surfaces, and whithin it the system evolves with mixing. For dissipative perturbations, however, these KAM-surfaces are broken and special methods like Melnikov one should be used to determine the conditions for chaos to appear. So, we calculate the width of a stochastic layer, which contains long-period waves and solitons. It is done on the basis of Chirikov criterion for the overlap of resonances, which is widely used for investigation of chaotic properties in hamiltonian systems ( references on original papers in this field can be found in \cite{book1}). Results we obtain are as follows: solitons and nonlinear waves prove to be chaotic in the meaning that in a small distance from the peak of solitons and long-period waves there must be a region of chaotic dynamics, where these waves acquire small irregular deviations. This conclusion is confirmed in experiments with ion-acoustic waves in plasma, where a weak splash before soliton peak was registered \cite{in-proc2}.
The outline of this paper is as follows. In next Section KdV
equation is reduced to ODE and the unperturbed solutions of the latter are
considered. In Section 3 we introduce canonically-conjugated
variables action - angle and obtain general expressions for the
criterion of stochasticity. After that in Section 4 this
criterion is applied to KdV waves directly. Section 5 contains
the conclusions and summary.
\section{Stationary waves}
Let's consider the perturbed KdV equation, taken in the following
form:
\begin{equation}\label{EQ1}
U_t+U\cdot U_x +\beta U_{x x x }=V(U,U_t,U_x,x-vt),
\end{equation}
where $x $ and $t$ denote, respectively, a one-dimensional space
coordinate and time; $U(x ,t)$ is supposed to be differentiable
with respect to $x $ and $t$ sufficient number of times; $V(U,U_t,U_x ,x-vt)$ is a small external perturbation. As far as we restrict ourselves on the case of hamiltonian perturbation, it means that within the class of functions $V$ we'll consider only those, which will not contain derivatives after integration over $(x-vt)\equiv\xi$. It's easy to check that the general form of such perturbations can be represented as:
\begin{equation}
V(U,U_t,U_x,\xi)=f_{1}(U,\xi)\cdot U_t(x,t)+f_{2}(U,\xi)\cdot U_x(x,t)+f_{3}(U,\xi),
\end{equation}
with the condition $\left[ f_2(U,\xi)-c\cdot f_1(U,\xi)\right]_{\xi}=f_3(U,\xi)_U$.
Searching for a solution of equation (\ref{EQ1}) in the
form of a nonlinear stationary wave $U(x ,t)=f(\xi)$, we
obtain the following ODE after one integration over $\xi$ ( prime means differentiation with respect to $\xi $ ):
\begin{equation}\label{EQ2}
\beta f^{^{^{\prime \prime}}}=vf-\frac{f^2}{2}+F(f,f^{^{\prime }},\xi ),
\end{equation}
where $F$ denotes the primitive of the function $V$ after stationary wave substitution. Integration constant is absent here because it can be turned into zero by an appropriate
choice of variables. For convenience let's rewrite equation (\ref{EQ2}) as the following dynamical system:
\begin{equation}\label{EQ3}
\beta \ddot x=vx-\frac{x^2}{2}+F(x,\dot x,t).
\end{equation}
Equation (\ref{EQ3}) without perturbation ( $F(x,\dot x,t) =0$ ) is
analogous to the equation of motion:
\begin{equation}\label{EQ4}
\beta \ddot x=vx-\frac{x^2}{2}
\end{equation}
of a point with mass $\beta$ in a potential field
\begin{equation}\label{EQ5}
U(x)=\frac{x^3}6-\frac{vx^2}2.
\end{equation}
\begin{figure}
\epsfig{width=4cm,file=pp2plot.ps}\hspace{2cm}
\epsfig{width=4cm,file=thick42.ps}
\caption{(Left) Phase plane of the unperturbed system (\ref{EQ4}). (Right) Plot of the potential energy (\ref{EQ5}). Here $v=1$ and $\beta =1$.}
\end{figure}
Full mechanical energy of such a nonlinear oscillator is equal to
\begin{equation}\label{EQ6}
H=\frac{p^2}{2\beta}-\frac{vx^2}2+\frac{x^3}6,
\end{equation}
where $p=\beta \dot x$ is the momentum of the point.
Phase plane of this oscillator together with the plot of the potential energy is presented on the Fig. 1 .
If $H=-\frac{2}{3}v^3$, what corresponds to the bottom of the
potential pit, then the phase curve shrinks to a point. In the case
$-\frac{2}{3}v^3<H<0$ we have closed phase curves representing
cnoidal waves of the form:
\begin{equation}\label{EQ7}
x^{m}(t)=a+b\cdot cn^2(kt\mid m),
\end{equation}
where $cn(z\mid m)$ denotes the Jacobi elliptic cosine function of modulus $m$.
Substitution of (\ref{EQ7}) in (\ref{EQ4}) gives:
\begin{equation}\label{EQ9}
\left\{
\begin{array}{l}
x^{m}(t)=v+v\left[ (1-2m)+3m\cdot cn^2(kt\mid m)\right]/{\sqrt{m^2-m+1}}
\\ T^m=4K(m)/k
\\ k=\frac{1}{2}\sqrt{v/\beta}(m^2-m+1)^{-\frac{1}{4}}
\end{array}
\right.
\end{equation}
Here $K(m)$ is the complete elliptic integral of the first kind, $T^m$ is the period of the corresponding solution, and $m$ is the modulus of elliptic functions. Under $H=0$ $(m=1)$, what corresponds to a separatrix on a phase plane, solution (\ref{EQ9}) yields the homoclinic orbit:
\begin{equation}
x(t)=\frac{3v}{\cosh ^2 \left( \frac{t}{\Delta} \right)},
\end{equation}
where $\Delta =2\sqrt{\frac{\beta }v}$ \cite{book2}.
\section{A criterion for chaotic motion}
Let's transform variables $\left( x,p\right)$ to the canonical pair action -
angle ($I$ - action, $\theta$ - angle) using standard formulae:
\begin{equation}\label{EQ10}
I=\frac 1{2\pi }\oint p(x,H)dx=I(H),\theta =\frac{\partial S(x,I)}{\partial I%
},S(x,I)=\int p(x,H)dx,
\end{equation}
where $S(x,I)$ is the reduced action. The Hamiltonian of a perturbed
motion we can write in the form:
\begin{equation}\label{EQ11}
H(I,\theta ,t)=H_0(I)+V(I,\theta ,t),
\end{equation}
where $H_0(I)$ is the Hamiltonian of the nonperturbed system from
(\ref{EQ6}), and $V(I,\theta ,t)$ corresponds to the perturbation term in equation (\ref{EQ3}).
Assuming perturbation $V$ to be periodical on time with frequency
$\nu $, and taking into consideration the fact that nonperturbed
motion is integrable, we may expand the perturbation into the following
Fourier series:
\begin{equation}\label{EQ12}
\left\{
\begin{array}{l}
V(I,\theta ,t)= \frac {1}{2} \sum\limits_{k,l}V_{kl}(I)\exp
[i(k\theta -l\nu t],
\\ V_{kl}=V_{-k,-l}^{*}.
\end{array}
\right.
\end{equation}
The perturbed equations of motion in variables $(I,\theta )$ take now the form:
\begin{equation}\label{EQ13}
\left\{
\begin{array}{l}
{\dot I=-\frac{1}{2} i \sum\limits_{k,l}kV_{kl}(I)\exp [i(k\theta
-l\nu t]} \\
{\dot \theta =\omega (I)+\frac{1}{2} i \sum\limits_{k,l}}\frac{dV_{kl}}
{dI}\exp [i(k\theta -l\nu t)]
\end{array}
\right.
\end{equation}
where $\omega (I)=\frac{dH_0}{dI}$ is the frequency of the nonperturbed motion.
Let's consider the motion in the vicinity of one particular resonance:
\begin{equation}
m\omega(I_{mn})-n\nu=0.
\end{equation}
The width of this resonance on frequency is $\Omega=\sqrt{4 V_0| \omega ^{^{\prime }}|}$, where $V_0\equiv |V_{mn}(I_{mn})$ and $\omega ^{^{\prime }}\equiv d\omega(I_0)/dI$.
The distance between adjacent resonances is equal to
\begin{equation}\label{EQ17}
\Delta \omega _{\alpha \beta }=\left| \omega (I_{m\pm \alpha ,n\pm \beta
})-\omega (I_{m,n})\right|,
\end{equation}
where $\alpha =0,1$; $\beta =0,1$.
As a condition of chaotic motion, which means an appearance of
mixing, we'll use the Chirikov criterion for the overlap
of resonances \cite{book3}:
\begin{equation}\label{EQ30}
K=\left( \frac \Omega {\Delta \omega }\right) ^2\geq 1.
\end{equation}
where $\Delta \omega $ is the minimal possible value of $\Delta \omega_{\alpha\beta}$ from (\ref{EQ17}). For the cases $(\alpha,\beta)=(1,0)$, $(0,1)$ and $(1,1)$ calculations similar to \cite{book3} give respectively the following conditions for the border of stochasticity:
\begin{equation}\label{EQ31}
\left\{
\begin{array}{l}
K_{10}=4 V_0| \omega ^{^{\prime }}| m^2 / \omega ^2
\geq 1,\mbox{ }m\gg 1,n\geq 1\\
K_{01}=4 V_0| \omega
^{^{\prime }}| n^2 / \omega ^2 \geq 1,\mbox{ }n\gg 1,m\geq 1\\
K_{11}=4 V_0| \omega ^{^{\prime }}| m^2n^2 / \omega
^2(m-n)^2 \geq 1,\mbox{ }m,n\gg 1,\mbox{ }\left| m-{}n\right| \sim
1
\end{array}
\right.
\end{equation}
\section{Chaos of Korteweg-de Vries waves}
To apply these results to the system, governed by KdV equation,
one must first evaluate the
exact expressions for $\omega $ and $\omega ^{^{\prime }}$, which
appear in the conditions (\ref{EQ31}). The frequency $\omega $ is equal to
\begin{equation}
\omega =\frac{2\pi }T=\frac{\pi k}{2K(m)}.
\end{equation}
For convenience let's rewrite it as:
\begin{equation}\label{EQ36}
\omega =\frac{\pi}{K(z)}\sqrt{\frac{c \sin{\frac{\phi}{3}}}{2\beta\sqrt{3}}},
\end{equation}
where we introduced $\varphi$ by the correlation $\cos{\frac{\varphi}{3}}=1+3\frac{H}{c^3}$, and $z\equiv \frac{1}{2}-\frac{\sqrt{3}}{2}\cot{\frac{\varphi}{3}}$.
To evaluate $\omega ^{^{\prime }}$, one should represent it as
\begin{equation}
\omega ^{^{\prime }}=\frac{d\omega }{dH}\frac{dH}{dI}=\omega \frac{d\omega}{d\varphi}\left( \frac{dH}{d\varphi}\right) ^{-1}.
\end{equation}
Tedious calculations give finally
\[
\omega ^{^{\prime }}=\frac{\pi ^2\omega}{16c^{5/2}
\sin \varphi K^2(z)}\sqrt{\frac{3}{2\beta \sin \frac \varphi 3}}\left[ -4\cos \frac \varphi 3 F\left( 1.5;0.5;1;z \right)
+\right.
\]
\begin{equation}\label{EQ37}
\left. +\left( \sqrt{3}\sin \frac \varphi 3+\cos \frac \varphi
3\right)F\left( 1.5;1.5;2;z\right) \right].
\end{equation}
Substituting (\ref{EQ36}) and (\ref{EQ37}) in (\ref{EQ31}) one makes sure that with approach to
separatrix $K\rightarrow $ $\infty $ under arbitrary small
$V_0 $. This means that for any external perturbation,
however small it is, Chirikov criterion is executed starting with
some energies, and this leads to the formation of a corresponding
stochastic layer. The transition from regular to chaotic motioncan be found
approximately from the condition $K\approx 1$. This border is defined on energy by the inequality:
\begin{equation}\label{EQ38}
H_{\min }\leq H\leq 0,H_{\min }=\frac{c^3}3(\cos \varphi _{\min
}-1).
\end{equation}
Here the value of $\varphi _{\min }$ is found from:
\begin{equation}
\frac{4v^3\sin {\varphi_{\min }} \tan (\varphi_{\min
}/3)K(z_{\min })}{-4{}F\left(1.5; 0.5;1;z_{\min }\right) +\left(
\sqrt{3}\tan ( \varphi_{\min }/3) +1\right) {}F\left(
1.5;1.5;2;z_{\min }\right) }=\pi V_0\zeta\left(m,n\right),
\end{equation}
in which $z_{\min }=\frac 12-\frac{\sqrt{3}}2\cot
\frac{\varphi_{\min }}{3}$ and $\zeta(m,n)=m^2$, $n^2$,
$\frac{m^2n^2}{(m-n)^2}$ for three cases in (\ref{EQ31}).
Certainly, there is only half-width of stochastic layer, but the phase curves, laying out separatrix don't constitute an interest, being physically meaningless
(they are unbounded on infinity).
\section{Summary and conclusions}
We have calculated the width of the stochastic layer around a separatrix, corresponding to a soliton solution of Korteweg-de Vries equation under hamiltonian perturbations. As far as the motion in this stochastic layer is chaotic ( in the meaning of mixing), so nonlinear wave solutions, which correspond to the phase curves within this stochastic layer will also posess certain chaotic properties. Let's now consider the question about the spatial region, where this chaotic behaviour can be registered.
The time length, after which stochasticity in a nonlinear oscillator can be found is defined by $\tau _c\ll t$, where $\tau _c$ is a time of the decay of correlations. In the work \cite{book3} the following estimation of
$\tau _c$ for a nonlinear oscillator is obtained:
\begin{equation}
\tau _c\sim \frac 1{\ln K},
\end{equation}
where $K$ is a coefficient for the overlap of resonances, obtained in the previous section. As far as in our consideration time $t$ for a
nonlinear oscillator corresponds to the expression ($x-ct$) for
nonlinear waves, so the region, where chaotic regimes for these waves can be detected, is:
\begin{equation}
\frac 1{\ln K}\ll \left( x-ct\right).
\end{equation}
Therefore, we can conclude that this region represents wave formation, which
outstrips nonlinear wave or soliton, and propagates in the same direction
with the same velocity. Under $t=0$ we obtain:
\begin{equation}
\frac 1{\ln K}\ll x .
\end{equation}
As far as with approach to a separatrix $K\rightarrow $ $\infty $,
so for solitons the minimal distance on which chaotic behaviour should appear $x _c\rightarrow 0$, and this effect
will be realized in a close vicinity of a soliton peak. The
chaos, which we are searching for, will be manifested in
the irregular small deflection from a smooth initial soliton profile.
So, as a main result of this work, it can be inferred that long-period nonlinear waves and especially solitons of Korteweg-de Vries equation obtain chaotic properties in the presence of periodic hamiltonian external perturbations.
Exactly this result was observed in an experiment with ion-acoustic
and Langmuir waves in a nonmagnetized plasma \cite{in-proc2}. There was
registered a faint splash directly before the soliton. From the
concepts developed in this paper this phenomenon can be explained
in the following way. Together with soliton wave generator also produces a group of waves of small amplitude and almost zero frequencies. These waves are produced constantly with generator
working, and therefore can be considered as a small deterministic
periodic external perturbation. Stochastic destruction of soliton
due to the influence of these waves, as it was described above,
must realize itself just before the soliton peak, exactly what was
registered in the experiment.
\section* {\bf Acknowledgements}
The author would like to thank Prof. M. Bestehorn for helpful discussions. He also acknowledges the referee for suggestions, which helped to make the paper clearer.
|
\section{Introduction}
The causal set hypothesis asserts that spacetime, ultimately, is
discrete and that its underlying structure is that of a locally
finite, partially ordered set (a causal set). The approach to quantum
gravity based on this hypothesis has experienced considerable progress
in its kinematic aspects. For example, one possesses natural
extensions of the concepts of proper time and spacetime dimensionality
to causal sets, and these take us a significant way toward an answer
to the question, ``When does a causal set resemble a Lorentzian
manifold?''. The dynamics of causal sets (the ``equations of
motion''), however, has not been very developed to date. One of the
primary difficulties in formulating a dynamics for causal sets is the
sparseness of the fundamental mathematical structure. When all one
has to work with is a discrete set and a partial order, even the
notion of what we should mean by a dynamics is not obvious.
Traditionally, one prescribes a dynamical law by specifying a
Hamiltonian to be the generator of the time evolution. This practice
presupposes the existence of a continuous time variable, which we do
not have in the case of causal sets. Thus, one must conceive of
dynamics in a more general sense. In this paper, evolution will be
envisaged as a process of stochastic growth to be described in terms
of the probabilities (in the classical case, or more generally the
quantum measures in the quantum case \cite{qmeasure}) of forming
designated causal sets. That is, the dynamical law will be a rule
which assigns probabilities to suitable classes of causal sets (a
causal set being the ``history'' of the theory in the sense of
``sum-over-histories''). One can then use this rule -- technically a
probability measure -- to ask physically meaningful questions of the
theory. For example one could ask ``What is the probability that the
universe possesses the diamond poset as a partial stem?''. (The term
`stem' is defined below.)
Why are we interested in a classical dynamics for causal sets, when our
ultimate aim is a quantum theory of gravity? One obvious reason is that
the classical case, being much simpler, can help us to get used to a
relatively unfamiliar type of dynamical formulation, bringing out the
pertinent physical issues and guiding us toward physically suitable
conditions to place on the theory. Is there, for example, an
appropriate form of causality that we can impose? Should we attempt to
express the theory directly in terms of gauge invariant (labeling
independent) quantities, or should we follow precedent by enforcing
gauge invariance only at the end? Some of these issues are well
illustrated with the theories we construct herein.
One of the best reasons to be interested in a classical dynamics for
causal sets is that quantum gravity must possess general relativity as a
classical limit. Thus general relativity should be described as some
type of effective classical dynamics for causal sets, and one may hope
that the relevant dynamical law will be among the family delineated
herein. (One can't be certain this will occur, because general
relativity, as a continuum theory, seems most likely to arise as an
effective theory for coarse-grained causal sets, rather than directly as
a limit of the microscopic discrete theory, and there is no guarantee
that this effective theory will have the same form as the underlying
exact one.)
A question commonly asked of the causal set program is ``How could
nongravitational matter arise from only a partial order?''. One
obvious answer is that matter can emerge as a higher level construct
via the Kaluza-Klein mechanism \cite{JFKK}, but this possibility has
nothing to do with causal sets as such. The theory developed herein
suggests a different mechanism. It is possible to rewrite the theory
in such a way that the dynamics appears to arise from a kind of
``effective action'' for a field of Ising spins living on the
relations of the causal set. A form of ``Ising matter'' is thus
implicit in what would seem at first sight to be a purely
``source-free'' theory.
In subsequent sections of this paper we: describe our notation and
terminology (some new language is required for the detailed derivation
of our causal set dynamics); introduce and briefly discuss the
transitive percolation model; present the physical requirements of
Bell causality and discrete general covariance that we will impose;
derive the (generically)
most general theory satisfying these requirements
(including solving the inequalities which express
that all probabilities must fall between 0 and 1); single out a few
simple choices of the free parameters and exhibit some properties of
the resulting ``cosmologies''; exhibit a pair of state-models for the
dynamics that illustrate how not only geometry, but other matter can
arise implicitly from order.
\subsection{Notation/Terminology/Sequential growth}
First we establish our terminology and notation. For a fuller
introduction to causal sets,
see \cite{causets0,causets1,causets2,causets4,bom87}.
(For recent examples of other discrete models incorporating a causal
ordering see \cite{fot,amb1,amb2}.)
A {\it causal set} (or ``causet'') is a locally finite, partially
ordered set (or ``poset''). We represent the order-relation by
`$\prec$' and use the {\it irreflexive convention} that an element
does not precede itself.
Let $C$ be a poset. The {\it past} of an element $x\in{C}$ is the
subset $\mathrm{past}(x)=\{y\in{C}\,|\,y\prec{x}\}$. The past of a subset of
$C$ is the union of the pasts of its elements. An element of $C$ is
{\it maximal} iff it is to the past of no other element. A {\it
chain} is a linearly ordered subset of $C$ (a subset, every two
elements of which are related by $\prec$); an {\it antichain} is a
totally unordered subset (a subset, no two elements of which are
related by $\prec$). A \emph{partial stem} of $C$ is a finite subset
which contains its own past. (A full stem is a partial stem such that
every element of its complement lies to the future of one of its
maximal elements.) An \emph{automorphism} of $C$ is a one-to-one map
of $C$ onto itself that preserves $\prec$.
A \emph{link} of a poset is an irreducible relation, that is, one not
implied by other relations via transitivity.\footnote%
{Links are often called ``covering relations'' in the mathematical
literature.}
A {\it path} in a poset is an increasing sequence of elements, each
related to the next by a link.
A poset can be represented graphically by a Hasse diagram, which is
constructed as follows. Draw a dot to represent each element of the
poset. Draw a line connecting any two elements $x\prec{}y$ related by a
link, such that the preceding element $x$ is drawn below the following
element $y$.
The dynamics which we will derive can be regarded as a process of
``cosmological accretion'' or ``growth''. At each step of this process
an element of the causal set comes into being as the ``offspring'' of a
definite set of the existing elements -- the elements that form its
past. The phenomenological passage of time is taken to be a
manifestation of this continuing growth of the causet. Thus, we do not
think of the process as happening ``in time'' but rather as
``constituting time'', which means in a practical sense that there
is no meaningful order of birth of the elements other than that
implied by the relation $\prec$.
In order to define the dynamics, however, we will treat the births as if
they happened in a definite order with respect to some fictitious
``external time''. In this way, we introduce an element of ``gauge''
into the description of the growth process which we will have to
compensate by imposing appropriate conditions of ``gauge invariance''.
This fictitious order of birth can be represented as a
{\it natural labeling} of the elements, that is, a labeling by integers
$0,1,2\cdots{}$ which are compatible with the causal order $\prec$ in
the sense that $x\prec{y}\Rightarrow{label}(x)<label(y)$.\footnote%
{A natural labeling of an order $P$ is equivalent to what is called a
``linear extension of $P$'' in the mathematical literature.}
The relevant notion of gauge invariance (which we will call ``discrete
general covariance'') is then captured by the statement that the
labels carry no physical meaning. We discuss this more extensively
later on.\footnote%
{The continuum analog of a natural labeling might be a coordinate system
in which $x^0$ is everywhere timelike (and this in turn is almost the
same thing as a foliation by spacelike slices). One could also
consider arbitrary labelings, which would be analogous to arbitrary
coordinate systems. In that case, there would be a well-defined gauge
{\it group} --- the group of permutations of the causet elements ---
and labeling invariance would signify invariance under this group, in
closer analogy with diffeomorphism invariance and ordinary gauge
invariance. However, we have not found a useful way to do this, and in
this paper only natural labelings will ever be considered.}
It is helpful to visualize the growth of the causal set in terms of
paths in a poset ${\cal P}$ of finite causal sets. (Thus viewed, the
growth process will be a sort of Markov process taking place in
${\cal P}$.)
Each finite causet
(or rather each isomorphism equivalence class of finite causets)
is one element of this poset.
If a causet can be formed by accreting a single element to a second causet,
then the former (the ``child'') follows the latter (the ``parent'')
in ${\cal P}$ and the relation between them is a link.
Drawing ${\cal P}$ as a Hasse diagram of Hasse diagrams,
we get figure \ref{poscau}.
(Of course this is only a portion of the infinite diagram;
it includes all the causal sets of fewer than five elements
and 8 of the 63 five element causets.
\begin{figure}[htbp]
\center
\scalebox{.61}{\includegraphics{poscau.eps}}
\caption{The poset of finite causets}
\label{poscau}
\end{figure}
The ``decorations'' on some of the transitions in figure \ref{poscau} are
for later use.)
Any natural labeling of a causet $C\in{\cal P}$ determines uniquely a
path in ${\cal P}$ beginning at the empty causet and ending at $C$.
Conversely, any choice of upward path through this diagram determines
a naturally labeled causet
(or rather a set of them, since inequivalent labelings can sometimes
give rise to the same path in ${\cal P}$.)\footnote%
{We could restore uniqueness by ``resolving'' each link $C_1\prec{C_2}$
of ${\cal P}$ into the set of distinct embeddings $i:C_1\to{C_2}$ that
it represents. Here, two embeddings count as distinct iff no
automorphism of the child relates them (cf. the discussion of the
Markov sum rule below).}
We want the physics to be independent of labeling, so different paths in
${\cal P}$ leading to the same causet should be regarded as representing
the same (partial) universe, the distinction between them being ``pure
gauge''.
The causal sets which can be formed by adjoining a single
maximal element to a
given causet will be called collectively a \emph{family}. The causet
from which they come is their \emph{parent}, and they are
\emph{siblings} of each other. Each one is a \emph{child} of the
parent. The child formed by adjoining an element which is to the future
of every element of the parent will be called the \emph{timid child}.
The child formed by adjoining an element
which is spacelike to every other element
will be called the \emph{gregarious child}.
Each parent-child relationship in ${\cal P}$ describes a `transition'
$C\to{}C'$, from one causal set to another induced by the birth of a
new element. The past of the new element (a subset of $C$) will be
referred to as the \emph{precursor set} of the transition (or
sometimes just the ``precursor of the transition''). Normally, this
precursor set is uniquely determined up to automorphism of the parent
by the (isomorphism equivalence class of the) child, but (rather
remarkably) this is not always the case. The symbol $\mathcal{C}_n$ will
denote the set of causets with $n$ elements, and the set of all
transitions from $\mathcal{C}_n$ to $\mathcal{C}_{n+1}$ will be called \emph{stage $n$}.
As just remarked, each parent-child transition corresponds to a
choice of partial stem in the parent (the precursor of the transition).
Since there is a one-to-one correspondence between partial stems and
antichains, a choice of child also corresponds to a choice of (possibly
empty) antichain in the parent, the antichain in question being the set
of maximal elements of the past of the new element. Note also that the
new element will be \emph{linked} to each element of this antichain.
\subsection{Some examples}
To help clarify the terminology introduced in the previous section, we
give some examples.
The 20 element causet of figure \ref{20elts}
was generated by the stochastic dynamics described herein,
with the choice of parameters given by
equation (\ref{lifelike}) below.
\begin{figure}[htbp]
\center
\scalebox{.9}{\includegraphics{pasts.eps}}
\caption{An example of a (`typical'?) 20 element causal set}
\label{20elts}
\end{figure}
In the copy of this causet on the left, the past of element $a$ is
highlighted. Notice that since we use the irreflexive convention for
the order, $a$ is not included in its own past. In the the copy on the
right, a partial stem of the causet is highlighted.
Figure \ref{family} shows \No and its children.
\begin{figure}[htbp]
\center
\scalebox{.64}{\includegraphics{family.eps}}
\caption{A family}
\label{family}
\end{figure}
The timid child is $C_b$ and the gregarious child is $C_c$. The
precursor set leading to the transition to $C_d$ is shown in the
ellipse. An example of an automorphism of $C_a$ is the map
$a\leftrightarrow c, b\leftrightarrow d$ (the other elements remaining
unchanged).
\section{Transitive Percolation}
\label{plain_perc}
In a sum over histories formulation of causal set theory,
one might expect sums like
\begin{equation}
\sum_\mathcal{C} A(\mathcal{C},\{q\})
\label{soh}
\end{equation}
to be involved, where $A$ is a complex amplitude for the causal set
$\mathcal{C}$, possibly depending on a set of parameters $\{q\}$. Kleitman and
Rothschild have shown that the number of posets of cardinality $n$ grows
faster than exponentially in $n$ and that asymptotically, almost every
poset has a certain, almost trivial, ``generic'' form.
(See \cite{brightwell}.)
Such a ``generic poset'' consists of three ``tiers'',
with $n/2$ elements in the middle tier
and $n/4$ elements in the top and bottom tiers.
For this reason, one might think that a sum
like (\ref{soh}) would be dominated by causets which in no way resemble
a spacetime, leading to a sort of ``entropy catastrophe''.
Nevertheless, it is not hard to forestall this catastrophe, and in fact
the most naive choice of stochastic dynamics already does so. (Maybe this
is not so different from the situation in ordinary quantum
mechanics, where the smooth paths, which form a set of measure zero in
the space of all paths, are the ones which dominate the sum over histories
in the classical limit.)
The dynamics in question, which we will call ``transitive percolation'',
is perhaps the most obvious model of a randomly growing causet. It is
an especially simple instance of a sequential growth dynamics, in which
each new element forges a causal bond independently with each existing
element with probability $p$, where $p\in[0,1]$ is a fixed parameter of
the model. (Any causal relation implied by transitivity must then be
added in as well.)
{}From a more static perspective, one can also describe transitive
percolation by the following algorithm for generating a random poset:
\begin{enumerate}
\item Start with $n$ elements labeled $0,1,2,\cdots,n-1$ \ ($n=\infty$
is not excluded.)
\item With a fixed probability $p$, introduce a relation between every
pair of points labeled $i$
and $j$, where $i<j$.
\item Form the transitive closure of these relations (e.g. if
$2\prec5$ and $5\prec8$ then enforce that $2\prec8$.)
\end{enumerate}
Expressed in this manner, the model appears as a species of one
dimensional directed percolation; hence the name we have given it
(following D.~Meyer).
{}From a physical point of view, transitive percolation has some appealing
features, both as a model for a relatively small region of spacetime
and as a cosmological model for spacetime as a whole. For
$p\sim{1/n}$, there is a percolation transition, where the causet goes
qualitatively from a large number of small disconnected universes for
$p<p_{\mathrm{crit}}$ to a single connected universe for
$p>p_{\mathrm{crit}}$. Moreover, computer simulations suggest
strongly that the model possesses a continuum limit and exhibits
scaling behavior in that limit with $p$ scaling roughly like
$c\log{n}/n$ \cite{cont_lim,scaling}. The ``cosmology'' of transitive
percolation is also suggestive --- the universe cycles endlessly
through phases of expansion, stasis, and contraction (via fluctuation)
back down to a single element \cite{posts}.
\begin{figure}[htbp]
\center
\scalebox{.6}{\includegraphics{tran_perc.eps}}
\caption{Transitive percolation cosmology}
\label{trans_perc_cos}
\end{figure}
{}From all this, it is clear that the causets generated by transitive
percolation do not at all resemble the 3-tier, generic causets of
Kleitman and Rothschild, but rather they have the potential to reproduce
a spacetime or a part of one.
Nevertheless, the dynamics of transitive percolation is
not viable as a theory of quantum gravity. One obvious reason is that
it is stochastic only in the purely classical sense, lacking quantum
interference. Another reason is that the future of any element of the
causet is completely independent of anything ``spacelike related'' to
that element. Therefore, the only spacetimes which a causal set
generated by transitive percolation could hope to resemble would be
homogeneous, such as the Minkowski or de Sitter spacetimes; but neither
of these possibilities is compatible with the periodic re-collapses
alluded to earlier. At best, therefore, one could hope to reproduce a
small portion of such a homogeneous spacetime.
On the other hand, in computer simulations of transitive percolation
\cite{AChR}, two independent (and coarse-graining invariant) dimension
estimators have tended to agree with each other, one such estimator
being that of \cite{meyer} and the other being a simple ``midpoint
scaling dimension''.
(Some other indicators of manifold-like behavior
have tended to do much more poorly, but those are not invariant under
coarse graining, whereas one would in any case expect to observe
manifold like behavior only for a sufficiently coarse grained causal
set.) In the pure percolation model, however, these dimension
indicators vary with time (i.e. with $n$) and one must rescale $p$ if
one wishes to hold the spacetime dimension constant. One may ask,
then, if the model can be generalized by having $p$ vary with $n$ in
an appropriate sense. We will see in the next section that something
rather like this is in fact possible.
The transitive percolation model, incidentally, has attracted the
interest of both mathematicians and physicists for reasons having
nothing to do with quantum gravity. By physicists, it has been studied
as a problem in the statistical mechanical field of percolation, as we
have already alluded to. By mathematicians, it has been studied
extensively as a branch of random graph theory (a poset being the same
thing as a transitive acyclic directed graph).
Some references on transitive percolation (viewed from whatever angle)
are
\cite{brightwell,posts,bb,schulman,AChR,cont_lim,scaling}.
\section{Physical requirements on the dynamics}
\label{requirements}
As discussed in the previous section, one can think of transitive
percolation as a sort of ``birth process'', but as such, it is only
one special case drawn from a much larger universe of possibilities.
As preparation for describing these more general possible dynamical
rules, let us consider the growth-sequence of a causal set universe.
First element `0' appears (say with probability one, since the universe
exists). Then element `1' appears, either related to `0' or not. Then
element `2' appears, either related to `0' or `1', or both, or neither.
Of course if $1\succ0$ and $2\succ1$ then $2\succ0$ by transitivity.
Then element `3' appears with some consistent set of ancestors,
and so on and so forth.
Because of transitivity, each new element ends up with a partial stem of
the previous causet as its precursor set.
The result of this process, obviously, is a naturally labeled causet (finite
if we stop at some finite stage, or infinite if we do not) whose labels
record the order of succession of the individual births.
For illustration, consider the path in figure \ref{poscau}
delineated by the heavy arrows. Along this path,
element `0' appears initially, then
element `1' appears to the future of element `0', then
element `2' appears to the future of element `0',
but not to the future of `1', then
element `3' appears unrelated to any existing element, then
element `4' appears to the future of
elements `0', `1' (say, or `2', it doesn't matter) and `3', then
element `5' appears (not shown in the diagram), etc.
Let us emphasize once more that the labels 0, 1, 2, etc. are not
supposed to be physically significant. Rather, the ``external time''
that they record is just a way to conceptualize the process, and any two
birth sequences related to each other by a permutation of their labels
are to be regarded as physically identical.
So far, we have been describing the kinematics of sequential growth.
In order to define a dynamics for it,
we may give,
for each $n$-element causet $C$,
the {\it transition probability}
from it to each of its possible children.
Equivalently, we give a transition probability
for each partial stem within $C$. We wish to construct a general
theory for these transition probabilities by subjecting them to
certain natural conditions. In other words, we want to construct the
most general (classically stochastic) ``sequential growth dynamics''
for causal sets.\footnote
{By choosing to specify our stochastic process in terms of transition
probabilities, we have assumed in effect that the process is Markovian.
Although this might seem to entail a loss of generality, the loss is
only apparent, because the condition of discrete general covariance
introduced below would have forced the Markov assumption on us,
even if we had not already adopted it.}
In stating the following conditions, we will employ
the terminology introduced in the Introduction.
\subsubsection{The condition of internal temporality}
By this imposing sounding phrase, we mean simply that each element is
born either to the future of, or unrelated to, all existing elements;
that is, no element can arise to the past of an existing element.
We have already assumed this tacitly in describing what we mean by a
sequential growth dynamics. An equivalent formulation is that the
labeling induced by the order of birth must be {\it natural}, as
defined above. The logic behind the requirement of internal
temporality is that all physical time is that of the intrinsic order
defining the causal set itself. For an element to be born to the past
of another would be contradictory: it would mean that an event
occurred ``before'' another which intrinsically preceded it.
\subsubsection{The condition of discrete general covariance}
As we have been emphasizing, the ``external time'' in which the causal
set grows (equivalently the induced labeling of the resulting poset)
is not meant to carry any physical information. We interpret this in
the present context as being the condition that the net probability of
forming any particular $n$-element causet $C$ is independent of the
order of birth we attribute to its elements. Thus, if
$\gamma$ is any path through the poset ${\cal P}$ of finite causal sets
that originates at the empty causet and terminates at $C$, then the
product of the transition probabilities along the links of $\gamma$
must be the same as for any other path arriving at $C$. (So general
covariance in this setting is a type of path independence).
We should recall here, however, that, as observed earlier, a link in
${\cal P}$ can sometimes represent more than one possible transition.
Thus our statement of path-independence, to be technically correct,
should say that the answer is the same no matter which transition
(partial stem) we select to represent the link.
Obviously, this immediately entails that all
such representatives share the same transition probability.
We might with justice have required here conditions that are
apparently much stronger, including the condition that {\it any} two
paths through ${\cal P}$ with the same initial and final endpoints have
the same product of transition probabilities. However, it is easy to
see that this already follows from the condition stated.\footnote
{If $\gamma$ does not start with the empty causet $C_0$, but at $C_s$,
we can extend it to start at $C_0$ by choosing any fixed path from $C_0$
to $C_s$. Then different paths from $C_s$ to $C_e$ correspond to
different paths between $C_0$ and $C_e$, and the equality of net
probabilities for the latter implies the same thing for the former.}
We therefore do not make it part of our definition of discrete general
covariance, although we will be using it crucially.
\label{physical}
Finally, it is well to remark here that just because the ``arrival
probability at $C$'' is independent of path/labeling, that does not
necessarily mean that it carries an invariant meaning. On the
contrary a statement like ``when the causet had 8 elements it was a
chain'' is itself meaningless before a certain birth order is chosen.
This, also, is an aspect of the gauge problem, but not one that
functions as a constraint on the transition probabilities that define
our dynamics. Rather it limits the physically meaningful {\it
questions} that we can ask of the dynamics. Technically, we expect
that our dynamics (like any stochastic process) can be interpreted as
a probability measure on a certain $\sigma$-algebra, and the
requirement of general covariance will then serve to select the
subalgebra of sets whose measures have direct physical meaning.
\subsubsection{The Bell causality condition}
The condition of ``internal temporality'' may be viewed as a very weak type
of causality condition. The further causality condition we introduce now
is quite strong, being similar to that from which one derives Bell's
inequalities. We believe that such a condition is appropriate for a
classical theory, and we expect that some analog will be valid in the
quantum case as well. (On the other hand, we would have to abandon Bell
causality if our aim were to reproduce quantum effects from a classical
stochastic dynamics, as is sometimes advocated in the context of ``hidden
variable theories''. Given the inherent non-locality of causal sets,
there is no logical reason why such an attempt would have to fail.)
The physical idea behind our condition is that events occurring in some
part of a causal set $C$ should be influenced only by the portion of $C$
lying to their past. In this way, the order relation constituting $C$
will be causal in the dynamical sense, and not only in name. In terms of
our sequential growth dynamics, we make this precise as the requirement
that the ratio of the transition probabilities leading to two possible
children of a given causet depend only on the triad consisting of the
two corresponding precursor sets and their union.
Thus, let $C\rightarrow{}C_1$ designate a transition from $C\in\mathcal{C}_n$ to
$C_1\in\mathcal{C}_{n+1}$, and similarly for $C\rightarrow{}C_2$. Then, the Bell
causality condition can be expressed as the equality of two
ratios\footnote%
{In writing (\ref{BCr}), we assume for simplicity that both numerators
and both denominators are nonzero, this being the only case we will
have occasion to treat in the present paper.}:
\begin{equation}
\frac {prob(C \rightarrow C_1)} {prob(C \rightarrow C_2)}
=
\frac {prob(B \rightarrow B_1)} {prob(B \rightarrow B_2)}
\label{BCr}
\end{equation}
where
$B\in\mathcal{C}_m$, ${m}\le{n}$, is the union of
the precursor set of $C\rightarrow C_1$ with
the precursor set of $C\rightarrow C_2$,
$B_1\in\mathcal{C}_{m+1}$ is $B$ with an element added in the same manner as in
the transition $C\rightarrow{C_1}$, and
$B_2\in\mathcal{C}_{m+1}$ is $B$ with an element added in the same manner as in
the transition $C\rightarrow C_2$.\footnote%
{Recall that the precursor set of the transition $C\to{}C_1$ is the
subposet of $C$ that lies to the past of the new element that forms
$C_1$.}
(Notice that if the union of the precursor sets is the
entire parent causet, then the Bell causality condition reduces to a
trivial identity.)
To clarify the relationships among the causets involved, it may help to
characterize
the latter in yet another way.
Let $e_1$ be the element born in the transition $C\to{}C_1$ and
let $e_2$ be the element born in the transition $C\to{}C_2$.
Then
$C_i=C\cup\{e_i\}$ ($i=1,2$),
and we have
$B=(\mathrm{past}\,e_1)\cup(\mathrm{past}\,e_2)$
and
$B_i=B\cup\{e_i\}$ ($i=1,2$).
By its definition, Bell causality relates ratios of transition
probabilities belonging to one ``stage'' of the growth process to ratios
of transition probabilities belonging to previous stages. For
illustration,
consider the case depicted in figure \ref{bc_ex}.
\begin{figure}[htbp]
\center
\scalebox{.6}{\includegraphics{bc_example1.eps}}
\caption{Illustrating Bell causality}
\label{bc_ex}
\end{figure}
The precursor $P_1$ of the transition $C\rightarrow{}C_1$ contains
only the earliest (minimum) element of $C$, shown in the figure as a
pattern-filled dot.
The precursor $P_2$ of
$C\rightarrow{}C_2$ contains as well the next earliest element, shown
as a (different pattern)-filled dot.
The union of the two
precursors is thus $B={P_1}\cup{P_2}=P_2$. The elements of $C$
depicted as
open dots belong to neither precursor. Such elements will be called
\emph{spectators}. Bell causality says that the spectators can be
deleted without affecting relative probabilities. Thus the ratio of
the transition probabilities of figure \ref{bc_ex} is equal to that of
figure \ref{bc_ex2}.
\begin{figure}[htbp]
\center
\scalebox{.6}{\includegraphics{bc_example2.eps}}
\caption{Illustrating Bell causality - spectators do not affect relative
probability}
\label{bc_ex2}
\end{figure}
\subsubsection{The Markov sum rule}
As with any Markov process, we must require that the sum of the full
set of transition probabilities issuing from a given causet be unity.
However, the set we have to sum over depends in a subtle manner on the
extent to which we regard causal set elements as ``distinguishable''.
Heretofore we have identified distinct transitions with distinct
precursor sets of the parent. In doing so, we have in effect been
treating causet elements as distinguishable (by not identifying with
each other,
precursor sets related by automorphisms of the parent),
and this is what we shall continue to do. Indeed, this is the
counting of children used implicitly by transitive percolation, so we
keep it here for consistency. With respect to the diagram of figure
\ref{poscau}, this method of counting has the effect of introducing
coefficients into the sum rule, equal to the number of partial stems
of the parent which could be the precursor set of the transition. For
the transitions depicted there, these coefficients (when not one) are
shown next to the corresponding arrow.\footnote%
{One might describe the result of setting these coefficients to unity as
the case of ``indistinguishable causet elements''. It appears that
in this case
a dynamics with a richer structure obtains: instead of the transition
probability depending only on the size of the precursor set and the
number of its maximal elements, it is sensitive to more details of the
precursor set's structure.}
We remark here that these sum-rule coefficients admit an alternative
description in terms of embeddings of the parent into the child (as a
partial stem). Instead of saying ``the number of partial stems of the
parent which could be the past of the new element'', we could say ``the
number of order preserving injective maps from the parent onto partial
stems of the child, divided by the number of automorphisms of the
child''. (The proof of this equivalence will appear in \cite{rideout}.)
\section{The general form of the transition probabilities}
We seek to derive a general prescription which gives, consistent with
our requirements, the transition probability from an element of
$\mathcal{C}_n$ to an element of $\mathcal{C}_{n+1}$. To avoid having to deal with
special cases, we will assume throughout that no transition
probability vanishes. Thus the solution we find may be termed
``generic'', but not absolutely general.
In this connection, we want to point out that one probably does not
obtain every
possible solution of our conditions by taking limits of the generic
solution, and the special theories which result from taking certain
transition probabilities to vanish must be treated separately.\footnote%
{Indeed, the requirement of Bell causality itself must be given
an unambiguous interpretation when some of the transition probabilities
involved are zero.}
One such special theory is the \emph{originary percolation} model,
which is the same as the transitive percolation model, but with the
added restriction that each element except the original one must have
at least one ancestor among the previous elements. The net effect is
that the growing causal set is required to have an ``origin'' (=
unique minimum element) at all stages. (Generalizations are also
possible in which a more complex full stem of the causet is enforced.)
The poset of originary causets can be transformed into the poset of
all causets (exactly) by removing the origin from every originary
causet. The transition probabilities for originary percolation are
just those of ordinary transitive percolation with an added factor of
$(1-q^n)$ in the denominator at stage $n$.
\subsection{Counting the free parameters}
A theory of the sort we are seeking provides a probability for each
transition, so without further restriction, it would contain a free
parameter for every possible antichain of every possible (finite)
causet. We will see, however, that the requirements described above in
Section \ref{requirements} drastically limit this freedom.
\begin{lemma}
There is at most one free parameter per family.
\label{1param_family}
\end{lemma}
\noindent
\textbf{Proof:}
Consider a parent and its children.
Every such child, except
the timid child, participates in a Bell causality equation with the
gregarious child.
(See the proof of Lemma \ref{bc_consis} in the Appendix.)
Hence (since Bell causality equates ratios), all
these transitions are determined up to an overall factor.
This leaves two free parameters for the family. The Markov sum
rule gives another equation, which exhausts itself in determining
the probability of the timid child. Hence precisely one
free parameter per family remains after Bell Causality
and the sum rule are imposed. $\Box$
\begin{lemma}
The probability to add a completely disconnected element (the
``gregarious child transition'') depends only on the cardinality of the
parent causal set.
\label{q's}
\end{lemma}
\noindent \textbf{Proof:}
\begin{figure}[htbp]
\center
\scalebox{.9}{\includegraphics{GC.eps}}
\caption{Equality of ``gregarious child'' transitions}
\label{GC}
\end{figure}
Consider an arbitrary causet $A$, with a maximal element $e$, as
indicated in figure \ref{GC}. Adjoining a disconnected
element to $A$ produces the causet $B$. Then, removing $e$ from $B$
leads to the causet $C$, which can be looked upon as the gregarious
child of the causet $D=A\backslash\{e\}$. Adding another disconnected
element to $C$ leads to a causet $E$ with (at least) two completely
disconnected elements. Now, by general covariance,
\begin{displaymath}
ax = bw
\end{displaymath}
and by Bell causality,
\begin{displaymath}
\frac{y}{w} = \frac{b}{a}
\end{displaymath}
(the disconnected element in $C$ acts as the spectator here). Thus
\begin{displaymath}
ax = bw = ay \Longrightarrow x=y
\end{displaymath}
(Recall that we have assumed that no transition probability vanishes.)
Repeating our deductions with $C$ in the place of $A$ in the above
argument (and a new maximal element $f$ in the place of $e$), we see
that $y=z$, where $z$ is the probability for the transition from $F$
to $G$ as shown. Continuing in this way until we reach the antichain
$A_n$ shows finally that $x=q_n$, where we define $q_n$ as the
transition probability from the $n$-antichain to the
($n+1$)-antichain. Since our starting causet $A$ was not chosen
specially, this completes the proof. $\; \; \Box$
If our causal sets are regarded as entire universes, then a
gregarious child transition corresponds to the spawning of a new,
completely disconnected universe (which is not to say that this new
universe will not connect up with the existing universe in the
future). Lemma \ref{q's} proves that the probability for this to
occur does not depend on the internal structure of the existing
universe, but only on its size, which seems eminently reasonable.
In the sequel, we will call this probability $q_n$.
With Lemmas 1 and 2, we have reduced the number of free parameters
(since every family has a gregarious child) to 1 per stage, or what is
the same thing, to one per causal set element. In the next sections we
will see that no further reduction is possible based on our stated
conditions. Thus, the transition probabilities $q_n$ can be identified
as the free parameters or ``coupling constants'' of the theory. They
are, however, restricted further by inequalities that we will derive
below.
\subsection{The general transition probability in closed form}
Given the $q_n$, the remaining transition probabilities (for the
non-gregarious children) are determined by Bell causality and the sum
rule, as we have seen. Here we
derive an expression in closed form for an arbitrary transition
probability in terms of causet invariants and the parameters $q_n$.
\subsubsection{Mathematical form of transition probabilities}
We use the following notation:
\smallskip
\begin{tabular}{|c|c|}
\hline
$\alpha_n$ & an arbitrary transition probability from $\mathcal{C}_n$ to $\mathcal{C}_{n+1}$\\
$\beta_n$ & a transition whose precursor set is not the entire parent
(`non-timid' transition)\\
$\gamma_n$ & a transition whose precursor set \emph{is} the entire parent
(`timid' transition)\\
\hline
\end{tabular}\\
\smallskip
\noindent
Notice that the subscript $n$ here refers only to the number of elements
of the parent causet; it does not exhibit which particular transition
of stage $n$ is intended. A more complete notation might provide $\alpha$,
$\beta$ and $\gamma$ with further indices to specify both the parent
causet and the precursor set within the parent.
We also set $q_0\ideq{}1$ by convention.
\begin{lemma}
Each transition probability $\alpha_n$ of stage $n$ has the form
\begin{equation}
q_n \sum_{i=0}^n {\lambda_i \frac{1}{q_i}}
\label{FN}
\end{equation}
where the $\lambda_i$ are integers depending on the individual transition
in question.
\label{integer_form}
\end{lemma}
\noindent \textbf{Proof:}
This is easily seen to be true for stages 0 and 1.
Assume it is true for stage $n-1$.
Consider a non-timid transition probability $\beta_n$ of stage $n$. Bell
causality gives
\begin{displaymath}
\frac{\beta_n}{q_n} = \frac{\alpha_{n-1}}{q_{n-1}}
\end{displaymath}
where $\alpha_{n-1}$ is an appropriate transition probability from stage $n-1$.
So by induction
\begin{equation}
\beta_n = \alpha_{n-1} \frac{q_n}{q_{n-1}}
= \sum_{i=0}^{n-1} \lambda_i \frac{q_{n-1}}{q_i} \frac{q_n}{q_{n-1}}
= \sum_{i=0}^{n-1} \lambda_i \frac{q_n}{q_i} .
\label{BT}
\end{equation}
For a timid transition probability $\gamma_n$, we use the Markov sum rule:
\begin{equation}
\label{TT}
\gamma_n = 1 - \sum_j \beta_{nj}
\end{equation}
where $j$ labels the possible non-timid transitions (i.e. the set of proper
partial stems of the parent).\footnote%
{Of course, more than one stem will in general correspond to the same
link in ${\cal P}$. If we redefined $j$ to run over links in ${\cal P}$, then
(\ref{TT}) would read
$\gamma_n=1-\sum_j\chi_j\beta_{nj}$, where $\chi_j$ is the
``multiplicity'' of the $j$th link.}
But then, substituting (\ref{BT}) yields immediately
$$
\gamma_n
= 1 - \sum\limits_j \sum\limits_{i=0}^{n-1} {\lambda_{ji}\over q_i} q_n
= 1 - \sum\limits_{i=0}^{n-1} {\sum_j \lambda_{ji} \over q_i} q_n \,,
$$
which we clearly can put into the form (\ref{FN}) by taking
$\lambda_i=-\sum_j\lambda_{ji}$ for $i<n$ and
$\lambda_n=1$. $\Box$
\subsubsection{Another look at transitive percolation}
The transitive percolation model we described earlier is consistent
with the four conditions of Section 3. To see this, consider an
arbitrary causal set $C_n$ of size $n$. The transition probability
$\alpha_n$ from $C_n$ to a specified causet $C_{n+1}$ of size $n+1$ is
given by \begin{equation}
\label{pp}
\alpha_n = p^m (1-p)^{n-\varpi}
\end{equation}
where $m$ is the number of maximal elements in the precursor set and
$\varpi$ is the size of the entire precursor set.
(This becomes clear if one recalls how the precursor set of a newborn
element is generated in transitive percolation: first a set of ancestors
is selected at random, and then the ancestors implied by transitivity
are added. From this, it follows immediately that a given stem
$S\subseteq{}C_n$ results from the procedure iff (i) every maximal
element of $S$ is selected in the first step, and (ii) no element of
$C_n\less{}S$ is selected in the first step.)
In particular, we see that the
``gregarious transition'' will occur with probability $q_n=q^n$, where
$q=1-p$.
Now consider our four conditions. Internal temporality was built in
from the outset, as we know. Discrete general covariance is seen to
hold upon writing the net probability of a given $C_n$ explicitly in
terms of causet invariants (writing it in ``manifestly covariant
form'') as
$$
P(C_n) = W p^L q^{{n\choose 2}-R}
$$
where $L$ is the number of links in $C_n$, $R$ the number of relations,
and $W$ the number of (natural) labelings of $C_n$.
To see that transitive percolation obeys Bell causality, consider an
arbitrary parent causet. The transition probability to a given child
is
exhibited
in eq. (\ref{pp}). Consider two different children, one with
$(m,\varpi)$=($m_1$,$\varpi_1$) and the other with
$(m,\varpi)$=($m_2$,$\varpi_2$).
Bell causality requires that the ratio of their
transition probabilities be the same as if the parent were
reduced to the
union of the precursor sets of the two transitions, i.e.
it requires
\begin{displaymath}
\frac{p^{m_1} q^{n-\varpi_1}}{p^{m_2} q^{n-\varpi_2}} =
\frac{p^{m_1} q^{n'-\varpi_1}}{p^{m_2} q^{n'-\varpi_2}}
\end{displaymath}
where $n'$ is the cardinality of the union of the precursor sets of the two
transitions.
Thus, Bell causality is satisfied by inspection.
Finally, the Markov sum rule is essentially trivial. At each stage of
the growth process, a preliminary choice of ancestors is made by a
well-defined probabilistic procedure, and each such choice is mapped
uniquely onto a choice of partial stem. Thus the induced probabilities
of the partial stems sum automatically to unity.
\subsubsection{The general transition probability}
In the previous section we have shown that transitive percolation
produces transition probabilities
(\ref{pp})
consistent with all our conditions.
By equating the right hand side of (\ref{pp}) to the general form
(\ref{FN}) of Lemma \ref{integer_form}, we can solve for the
$\lambda_i$ and thus obtain the general solution of our conditions:
\begin{displaymath}
\alpha_n = \sum_{i=0}^n{\lambda_i \frac{1}{q_i}} \, q_n
= p^m (1-p)^{n-\varpi}
= (1-q)^m q^{n-\varpi}
\end{displaymath}
Expanding the factor $(1-q)^m$, and using the fact that $q_n=q^n$ for
transitive percolation, we get
\begin{displaymath}
\lambda_i = (-)^{\varpi-i} {m \choose \varpi-i} .
\end{displaymath}
So an arbitrary transition probability in the general dynamics is,
according to (\ref{FN})
\begin{displaymath}
\alpha_n = \sum_{i=0}^n (-)^{\varpi-i} {m \choose \varpi-i} \frac{q_n}{q_i} .
\end{displaymath}
Noting that the binomial coefficients are zero for $\varpi-i \notin
\{0..m\}$, and rearranging the indices,
we obtain
\begin{equation}
\label{gen_trans_prob}
\fbox{
$\displaystyle \alpha_n =
\sum_{k=0}^m (-)^k {m \choose k} \frac{q_n}{q_{\varpi-k}}$ .}
\end{equation}
This form for the transition probability exhibits its causal nature
particularly clearly: except for the overall normalization factor $q_n$,
$\alpha_n$ depends only on invariants of the associated precursor set.
\subsection{Inequalities}
Since the $\alpha_n$ are classical probabilities, each must lie between
0 and 1, and this in turn restricts the possible values of the $q_n$.
Here we show that it suffices to impose only one inequality per stage;
all the others (two per child) then follow. More precisely, what we
show is that, if $q_n>0$ for all $n$, and if $\alpha_n\ge{0}$ for the
``timid'' transition from the $n$-antichain, then all the
$\alpha_n$ lie in $[0,1]$. This we establish in the following two
``Claims''.
\\[3mm]
\noindent \textbf{Claim} \,
\emph{In order that all the transition probabilities $\alpha_n$ fall between 0
and 1, it suffices that each timid transition probability be $\ge 0$.}
\\[3mm]
\noindent \textbf{Proof:}
As described in the proofs of lemmas 1 and 5,
each non-timid transition (of stage $n$) is given (via Bell causality) by
\begin{displaymath}
\alpha_n = \alpha_m \frac{q_n}{q_m}
\end{displaymath}
where $m$ is some natural number less than $n$. The $q$'s are
positive. So if the probabilities of the previous stages are
positive, then the non-timid probabilities of stage $n$ are also positive.
It follows by induction that all but the timid transition probabilities
are positive (since $\alpha_0=q_0=1$ obviously is). But for the timid
transition of each family, we have
\begin{equation}
\gamma_n = 1 - \sum_i \beta_i
\label{GF}
\end{equation}
where each $\beta_i$ is positive. If any of the $\beta_i$ is greater
than one, $\gamma_n$ will obviously be negative. Also (\ref{GF})
plainly cannot be greater than one. Consequently, if we
require that $\gamma_n$ be positive, then all transition probabilities
in the family will be in $[0,1]$. $\Box$
In a timid transition, the entire parent is the precursor set, so $\varpi=n$.
The inequalities constraining each probability
of a given family
to be in $[0,1]$ therefore
reduce to the sole condition
\begin{equation}
\sum_{k=0}^m (-)^k {m \choose k} \frac{1}{q_{n-k}} \geq 0 \,.
\label{tnm}
\end{equation}
\vspace{2mm}
\noindent \textbf{Claim} \,
\emph{The most restrictive inequality of stage $n$ is the one arising from the
$n$-antichain, i.e. the one for which $m=n$. All other inequalities
of stage $n$ follow from this inequality and the inequalities for
smaller $n$.}
\\[2mm]
\noindent \textbf{Proof:}
Assume that we have, for $m=n$,
\begin{displaymath}
\sum_{k=0}^n (-)^k {n \choose k} \frac{1}{q_{n-k}} \geq 0 .
\end{displaymath}
Add to this the inequality from stage $n-1$,
\begin{displaymath}
\sum_{k=0}^{n-1} (-)^k {n-1 \choose k} \frac{1}{q_{n-k-1}} =
\sum_{k=0}^{n} (-)^{k-1} {n-1 \choose k-1} \frac{1}{q_{n-k}} \geq 0
\end{displaymath}
to get
\begin{displaymath}
\sum_{k=0}^{n-1} (-)^k {n-1 \choose k} \frac{1}{q_{n-k}} \geq 0 .
\end{displaymath}
This is the inequality of stage $n$ for $m=n-1$.
(We have used
the identity ${n \choose k} = {n-1 \choose k} + {n-1 \choose k-1}$.)
Adding
to it
the inequality of stage $n-1$ with $m=n-2$ yields the
inequality of stage $n$ for $m=n-2$.
Repeating this process will give all the
inequalities of stage $n$. $\; \; \Box$
It is helpful to introduce the quantities
\begin{equation}
\fbox{ $ \displaystyle
t_n = \sum_{k=0}^n (-)^{n-k} {n \choose k} \frac{1}{q_k} $}
\label{TD}
\end{equation}
Obviously, we have $t_0=1$ (since $q_0=1$), and we have seen that the
full set of inequalities restricting the $q_n$ will be satisfied iff
$t_n\ge0$ for all $n$. (Recall we are assuming $q_n>0, \, \forall n$.)
Moreover, given the $t_n$, we can recover the $q_n$ by inverting
(\ref{TD}):
\begin{lemma}
\begin{equation}
\label{q_of_t}
\fbox{ $ \displaystyle
\frac{1}{q_n} = \sum_{k=0}^n {n \choose k} t_k$}
\end{equation}
\end{lemma}
\noindent \textbf{Proof:}
This follows immediately from the identity
$$
\sum\limits_{k=0}^n {n \choose k} (-)^{n-k} {k \choose m} = \delta^n_m
\qquad\qquad\qquad \Box
$$
Thus, the $t_n$ may be treated as free parameters (subject only to
$t_n\ge{}0$ and $t_0=1$), and the $q_n$ can then be derived from
(\ref{q_of_t}).
If this is done,
the remaining transition probabilities $\alpha_n$ can be
re-expressed more simply in terms of the $t_n$ by inserting
(\ref{q_of_t}) into (\ref{gen_trans_prob}) to get
\begin{displaymath}
\frac{\alpha_n}{q_n}
= \sum_{l} t_l \sum_{k} (-)^k {m \choose k} {\varpi-k \choose l}
= \sum_{l} t_l {\varpi-m \choose \varpi-l}
\end{displaymath}
whence
\begin{equation}
\fbox{
$ \displaystyle
\alpha_n = \frac
{\sum_{l=m}^{\varpi} {\varpi-m \choose \varpi-l} t_l}
{\sum_{j=0}^n {n \choose j} t_j}
$}
\label{alpha_of_t}
\end{equation}
Here, we have used an identity for binomial coefficients
that can be found on page 63 of \cite{Feller}.
In this way, we arrive at the general solution of our inequalities.
(Actually, we go slightly beyond our ``genericity'' assumption that
$\alpha_n\not=0$ if we allow some of the $t_n$ to vanish; but no harm is
done thereby.)
Let us conclude this section by noting that (\ref{q_of_t}) implies
\begin{equation}
q_0 \equiv 1 \geq q_1 \geq q_2 \geq q_3 \geq \cdots
\label{qin}
\end{equation}
If we think of the $q_n$ as the basic parameters or ``coupling
constants'' of our sequential growth dynamics, then it is as if the
universe had a free choice of one parameter at each stage of the
process. We thus get an ``evolving dynamical law'', but the evolution
is not absolutely free, since the allowable values of $q_n$ at every
stage are limited by the choices already made. On the other hand, if
we think of the $t_n$ as the basic parameters, then the free choice is
unencumbered at each stage. However, unlike the $q_n$, the $t_n$
cannot be identified with any dynamical transition probability.
Rather, they can be realized as ratios of two such probabilities,
namely as the ratio $x_n/q_n$, where $x_n$ is the transition
probability from an antichain of $n$ elements to the timid child of
that antichain. (Thus, if we suppose that the evolving causet at the
beginning of stage $n$ is an antichain, then $t_n$ is the probability
that the next element will be born to the future of {\it every}
element, divided by the probability that the next element will be born
to the future of {\it no} element.)
\subsection{Proof that this dynamics obeys the physical requirements}
\label{obeys}
To complete our derivation, we must show that the sequential growth
dynamics given by (\ref{gen_trans_prob}) or (\ref{alpha_of_t}) obeys the
four conditions set out in section \ref{requirements}.
\subsubsection{Internal temporality}
This condition is built into our definition of the growth process.
\subsubsection{Discrete general covariance}
We have to show that the product of the transition probabilities
$\alpha_n$ associated with a labeling of a fixed finite causet $C$ is
independent of the labeling. But this follows immediately from
(\ref{gen_trans_prob}) [or (\ref{alpha_of_t})]
once we notice that what remains after the overall product
$$
\prod\limits_{j=0}^{|C|-1}q_j
$$
is factored out, is a product over all elements $x\in{C}$ of poset
invariants depending only on the structure of $\mathrm{past}(x)$.
\subsubsection{Bell causality}
Bell causality states that the ratio of the transition probabilities
for two siblings depends only on the union of their precursors.
Looking at (\ref{gen_trans_prob}), consider the ratio of two
such probabilities
$\alpha_{n1}$ and $\alpha_{n2}$. The $q_n$ factors will cancel,
leading to an expression which depends only on $\varpi_1$, $\varpi_2$, $m_1$,
and $m_2$. Since these are all determined by the structure of the
precursor sets, Bell causality is satisfied.
\subsubsection{Markov sum rule}
The sum rule states that the sum of all transition probabilities
$\alpha_n$ from a given parent $C$ (of cardinality $|C|=n$) is unity.
Since a child can
be identified with a partial stem of the parent,
we can write this condition, in view of (\ref{alpha_of_t}), as
\begin{equation}
\sum_S \sum_l t_l {|S|-m(S) \choose l-m(S)} = \sum_j t_j {n \choose j}
\label{Msr}
\end{equation}
where $S$ ranges over the partial stems of $C$.
This must hold for any $t_l$, since they may be chosen
freely. Reordering the sums and equating like terms yields
\begin{equation}
\forall l, \ \sum_S {|S|-m(S) \choose l-m(S)} = {n \choose l} \,,
\label{identities}
\end{equation}
an infinite set of identities which must hold if the sum rule
is to be satisfied by our dynamics.
The simplest way to see that (\ref{identities}) is true is to resort to
transitive percolation,
for which $t_l=t^l$, where $t=p/q=p/(1-p)$.
In that case we know that the sum rule is satisfied, so by inspection of
(\ref{Msr}), we see that the identity (\ref{identities}) must be true.
A more intuitive proof is illustrated well by
the case of $l=3$. Group the terms on the left side
according to the number of maximal elements:
\def\clt#1#2{#1\,|\,#2}
\begin{displaymath}
\begin{array}{ccccccccr}
\sum\limits_{\clt{S}{m(S)=0}} {|S|-0 \choose 3-0} & + &
\sum\limits_{\clt{S}{m(S)=1}} {|S|-1 \choose 3-1} & + &
\sum\limits_{\clt{S}{m(S)=2}} {|S|-2 \choose 3-2} & + &
\sum\limits_{\clt{S}{m(S)=3}} {|S|-3 \choose 3-3} & = & {n \choose 3}\\
0 & + &
\sum\limits_{\clt{S}{m(S)=1}} {|S|-1 \choose 2} & + &
\sum\limits_{\clt{S}{m(S)=2}} (|S|-2) & + &
\sum\limits_{\clt{S}{m(S)=3}} 1 & = & {n \choose 3}
\end{array}
\end{displaymath}
The first term is zero because the only partial stem with zero maximal
elements is empty (i.e. $|S|=0$).
The second term is a sum over all partial stems with one maximal
element. This is equivalent to a sum over elements, with the element's
inclusive past forming the partial stem. The summand chooses every
possible pair of elements to the past of the maximal element. Thus the
second term overall counts the 3-element subcausets of $C$ with a
single maximal element. There are two possibilities
here, the three-chain \chain3 and the ``lambda'' \wedge.
The third term sums over partial stems with two maximal elements,
which is equivalent to summing over 2 element antichains,
the inclusive past of the antichain being the partial stem.
The summand then counts the
number of elements to the past of the two maximal ones. Thus the third
term overall
counts the number of three element subcausets with
precisely two maximal elements. Again there are two possibilities, the
``V'' \V, and the ``L'', \Lcauset.
Finally, the fourth term is a sum over partial stems with three maximal
elements, and this can be interpreted as a sum over all three element
antichains \3antichain.
As this example illustrates, then,
the left hand side of (\ref{identities}) counts the number of $l$ element
subcausets of $C$, placing them into ``bins'' according to the number
of maximal elements of the subcauset.
Adding together the bin sizes yields the total number of $l$ element
subsets of $C$, which of course equals ${n \choose l}$.
\subsection{Sample cosmologies}
\label{cosmologies}
The physical consequences of differing choices of the $t_n$ remain to
be explored. To get an initial feel for this question, we list
some simple examples.
(Recall our convention that $t_0=1$, or equivalently, $q_0=1$, where
$q_0$ is the probability that the universe is born at all.\footnote%
{So, is the answer to the old question why something exists rather than
nothing, simply that it is notationally more convenient for it to be so?})
\begin{itemize}
\item{``Dust universe''}
\begin{displaymath}
t_0=1, \ t_i = 0, \; i \geq 1
\end{displaymath}
This universe is simply an antichain, since, according to (\ref{q_of_t}),
$q_n=1$ for all $n$.
\item{``Forest universe''}
\begin{displaymath}
t_0 = t_1 = 1; \; t_i = 0, \; i \geq 2
\end{displaymath}
This yields a universe consisting wholly of trees,
since (see the next example) $t_2=t_3=t_4=\cdots=0$ implies that no
element of the causet can have more than one past link.
The particular choice of $t_1=1$ has in addition the remarkable property
that, as follows easily from (\ref{alpha_of_t}), every allowed transition
of stage $n$ has the same probability $1/(n+1)$.
\item{Case of limited number of past links}
\begin{displaymath}
t_i=0, \; i > n_0
\end{displaymath}
Referring to expression
(\ref{alpha_of_t})
one sees at once that $\alpha_n$ vanishes if $m>n_0$. Hence, no element
can be born with more than $n_0$ past links or ``parents''. This means
in particular that any realistic choice of parameters will have $t_n>0$
for all $n$, since an element of a causal set
faithfully embeddable in Minkowski space
would have an infinite number of past links.
\item{Transitive percolation}
\begin{displaymath}
t_n = t^n
\end{displaymath}
We have seen that for
transitive percolation,
$q_n=q^n$, where $q=1-p$. Using the
binomial theorem, it is easy to learn from
(\ref{q_of_t})
or (\ref{TD}) that this choice of $q_n$ corresponds to $t_n=t^n$ with
$t=p/q$. Clearly, $t$ runs from 0 to $\infty$ as $p$ runs from 0 to 1.
\item{A more lifelike choice?}
\begin{equation}
t_n = \frac{1}{n!}
\label{lifelike}
\end{equation}
We have seen that transitive percolation with constant $p$ yields
causets which could reproduce --- at best --- only limited portions of
Minkowski space. To do any better, one would have to scale $p$ so that
it decreased with increasing $n$
\cite{cont_lim,scaling,AChR}.
This suggests that $t_n$ should fall off faster than in any percolation
model, hence (by the last example) faster than exponentially in $n$.
Obviously, there are many possibilities of this sort
(e.g. $t_n\sim{}e^{-\alpha{}n^2}$), but one of the simplest is
$t_n\sim{c}/n!$ \ This would be our candidate of the moment for a
physically most realistic choice of parameters.
\end{itemize}
\section{The stochastic growth process as such}
\label{class_measure}
We have seen that, associated with every {\it labeled} causet
$\tilde{C}$ of size $N$, is a net ``probability of formation''
$P(\tilde{C})$ which is the product of the transition probabilities
$\alpha_i$ of the individual births described by the labeling:
$$
P(\tilde{C}) = \prod\limits_{i=0}^{N-1} \alpha_i
$$
where $\alpha_i=\alpha(i,\varpi_i,m_i)$ is given by
(\ref{gen_trans_prob}) or (\ref{alpha_of_t}).
We have also
seen that $P$ is in fact independent of the labeling and may be written
as $P(C)$ where $C$ is the unlabeled causet corresponding to
$\tilde{C}$. To bring this out more clearly, let us define
\begin{equation}
\label{lambda-d}
\lambda(\varpi,m) = \sum\limits_{k=m}^\varpi {\varpi-m\choose \varpi-k} t_k
\end{equation}
Then $q_i=\lambda(i,0)^{-1}$ and we have
$$
\alpha_i(i,\varpi_i,m_i) = {\lambda(\varpi_i,m_i)\over\lambda(i,0)} \,,
$$
whence
$ P(\tilde{C}) = \prod\limits_{i=0}^{N-1} \lambda(\varpi_i,m_i) /
\prod\limits_{j=0}^{N-1} \lambda(j,0)$,
or
expressed more intrinsically,
\begin{equation}
\label{PC0}
P(C) = {
\prod\limits_{x\in C} \lambda(\varpi(x),m(x))
\over
\prod\limits_{j=0}^{|C|-1} \lambda(j,0) } \,,
\end{equation}
where $\varpi(x)=|\mathrm{past}{x}|$ and $m(x)=|{\rm maximal}(\mathrm{past} x)|$.
This expression, as far as it goes, is manifestly ``causal'' and
``covariant'' in the senses explained above. As also explained above,
however, it has no direct physical meaning.
Here we briefly discuss
some probabilities which {\it do} have
a fully
covariant meaning and show how,
in simple cases, they are related to $N\to\infty$ limits of
probabilities like (\ref{PC0}).
\def{\rm Prob}{{\rm Prob}}
First, let us notice that the net probability of arriving at a
particular $C\in{\cal P}$ is not $P(C)$ but
$$
{\rm Prob}_N(C) = W(C) \, P(C)
$$
where $N=|C|$ and $W(C)$ is the number of inequivalent\footnote%
{Two labelings of $C$ are equivalent iff related by an automorphism of $C$.}
labelings of $C$, or in other words, the total number of paths through
${\cal P}$ that arrive at $C$, each link being taken with its proper
multiplicity.
Now as a rudimentary example of a truly covariant question, let us take
``Does the two-chain ever occur as a partial stem of $C$?''.
The answer to this question will be a probability, $P$, which
it is natural to identify as
$$
P = \lim_{N\to\infty} {\rm Prob}_N (X_N) \,,
$$
where $X_N$ is the event that ``at stage $N$'', $C$ possesses a partial
stem which is a two-chain.
In this connection,
we conjecture that the
questions of the form
``Does $P$ occur as a partial stem of $C$?'' furnish a
physically complete set, when $P$ ranges over all (isomorphism
equivalence classes of) finite causets.
\section{Two Ising-like state-models}
\label{Ising}
In this section, we present two Ising-like state-models from which
$P(C)$ of equation (\ref{PC0}) can be obtained. In the main we just
indicate the results, leaving the details to appear elsewhere
\cite{rideout}.
The two models come from taking (\ref{gen_trans_prob}) or,
respectively, (\ref{alpha_of_t}) as the starting point. In each case,
the idea is to interpret the binomial coefficients which occur in
these formulas as describing a sum over subsets of relations of $C$.
If we work with (\ref{gen_trans_prob}) these will be subsets of the
set of {\it links} of $C$; if we work with (\ref{alpha_of_t}) they
will be subsets of the set of relations of $C$ that are {\it not}
links.
Let us take first equation (\ref{gen_trans_prob}).
Reinterpreting the binomial coefficients in the manner indicated, and
proceeding as in the derivation of (\ref{PC0}), we arrive at an
expression for $P(C)$ in terms of a sum over
$\hbox{${\rm Z \kern 0.3ex \llap{Z}}$}_2$-valued ``spins'' $\sigma$ living on the relations of
$C$. In summing over configurations, however, the spins $\sigma$ on
the non-link relations are set permanently to 1; only those on the
links vary. With $\sigma=1$ interpreted as ``presence'' and
$\sigma=0$ as ``absence'', the contribution of a particular spin
configuration $\sigma$ is an overall sign times the product of one
``vertex factor'' for each $x\in{C}$. The vertex factor is
$q_r^{-1}$, where $r$ is the number of present relations having $x$ as
future endpoint, and the sign is $(-)^a$, where $a$ is the number of
absent relations. (In addition, there is a constant overall factor
in $P(C)$ of $\prod_{j=0}^{|C|-1}q_j$.)
In the second state model,
we begin with (\ref{alpha_of_t}),
or better (\ref{PC0}) itself,
and proceed similarly.
The result is again a sum over spins $\sigma$
residing on the relations,
this time
with all the terms being positive
(as is required of physical Boltzmann weights).
In this second model,
the spins on the links are set permanently to 1
while those on the non-links vary.
The ``vertex factor'' coming from $x\in{C}$
now is $t_r$,
where $r$ is again
the number of relations
present and ``pointing to $x$''.
These two models (and especially the second) show that our sequential
growth dynamics can be viewed as a form of ``induced gravity''
obtained by summing over (``integrating out'') the values of our
underlying spin variables $\sigma$. This underlying ``matter'' theory
may or may not be physically reasonable (Does it obey its own version
of Bell causality, for example? Is it local in an appropriate sense?),
but at a minimum, it serves to
illustrate how a theory of non-gravitational matter can be hidden
within a theory that one might think to be limited to gravity
alone.\footnote%
{In this connection, it bears remembering that Ising matter can
produce fermionic as well as bosonic fields, at least in certain
circumstances. \cite{id89,ple97}}
\footnote%
{References \cite{kaz} and \cite{staud}
(for which we thank an anonymous referee)
describe
a similar example of ``hidden'' matter fields
in the context of 2-dimensional random surfaces
(Euclidean signature quantum gravity)
and the associated matrix models
in the continuum limit.
Unfortunately, the matter fields used
(Ising spins or ``hard dimers'')
were unphysical in the sense that the partition function
was a sum of Boltzmann weights
which were not in general real and positive.
This is much like our first state model described above.
To the extent that the analogy between
these two, rather different, situations
holds good,
our results here suggest that there might be,
in addition to the matter fields employed in \cite{staud},
another set of fields
with physical choices of the coupling constants,
which could reproduce the same effective dynamics for
the random surface.}
\section{Further Work}
Our dynamics can be simulated;
for $t_n=1/n!$ it takes a minute or so to generate a 64 element causet
on a DEC Alpha 600 workstation.
Analytic results, so far, are available only for the special case of
transitive percolation.
An important question, of course, is whether some choice
of the $t_n$ can reproduce general relativity, or at least reproduce a
Lorentzian manifold for some range of $t$'s and of $n=|C|$.
Similarly, one can ask whether our ``Ising matter'' gives rise to an
interesting effective field theory and what relation it has with the
local scalar matter on a background causal set studied in
\cite{daughton,salgado}?
Another set of questions concerns the possibility of a more ``manifestly
covariant'' formulation of our sequential growth dynamics -- or of more
general forms of causal set dynamics. Can Bell causality be formulated
in a gauge invariant manner, without reference to a choice of birth
sequence? Is our conjecture correct that all meaningful assertions are
logical combinations of assertions about the occurrence of partial
stems (``past sets'')?
(Such questions seem likely to arise with special
urgency in any attempt to generalize our dynamics to the quantum case.)
Also, there are the special cases we left unstudied. There exist
originary analogs of all of our dynamics, for example. Are there other
special, non-generic cases of interest?
We might continue multiplying questions, but let's finish with the
question of how to discover a quantum generalization of our dynamics.
Since our theory is formulated as a type of Markov process, and since a
Markov process mathematically is a probability measure on a suitable
sample space, the natural quantum generalization would seem to be a
{\it quantum} measure\cite{qmeasure} (or equivalent ``decoherence
functional'') on the same sample space.
The question then would be whether
one could find appropriate quantum analogs of Bell causality and general
covariance formulated in terms of such a quantum measure.
If so, we could hope that, just as in the classical case treated herein,
these two principles would lead us to a relatively unique quantum
causet dynamics,\footnote
{See \cite{criscuolo} for a promising first step toward such a dynamics.}
or rather to a family of them among which a potential
quantum theory of gravity would be recognizable.
It is a pleasure to thank Avner Ash for a stimulating discussion
at a critical stage of our work.
The research reported here was supported in part by NSF grant
PHY-9600620 and by a grant from the Office of Research and Computing of
Syracuse University.
\vspace{9mm}
\noindent
\textbf{\Large Appendix: Consistency of the conditions}
\vspace{5mm}
\label{consistency}
\noindent
Our analysis of the conditions of Bell causality et al. unfolded
in the form of several lemmas. Here we present some similar lemmas
which strictly speaking are not needed in the present context, but which
further elucidate the relationships among our conditions. We expect
these lemmas can be useful in any attempt to formulate generalizations
of our scheme, in particular quantal generalizations.
\begin{lemma}
The Bell causality equations are mutually consistent.
\label{bc_consis}
\end{lemma}
\begin{figure}[htbp]
\center
\scalebox{.9}{\includegraphics{BCconsis.eps}}
\caption{two families related by Bell causality}
\label{BCconsis}
\end{figure}
\noindent \textbf{Proof:}
The top of figure \ref{BCconsis} shows three children of an arbitrary
causal set $C_n$. The shaded ellipses represent portions of
$C_n$. The small square indicates the new element whose birth
transforms $C_n$ into a causal set $C_{n+1}$ of the next stage. The
smaller ellipse ``stacked on top of'' the larger ellipse represents a
subcauset of $C_n$ which does not intersect the precursor set of any
of the transitions being considered (i.e. none of its elements lie to
the past of any of the new elements). This small ellipse thus consists
entirely of ``spectators'' to the transitions under consideration. The
bottom part of figure \ref{BCconsis} shows the corresponding parent and
children when these spectators are removed.
Notice that one of the three children is the gregarious child. We
will show that the Bell causality equations between
this child and each of the others imply all remaining Bell causality
equations within this family. Since no Bell causality equation
reaches outside a single family (and since, within a family, the Bell
causality equations that involve the gregarious child obviously always
possess a solution --- in fact they determine all ratios of transition
probabilities except for that to the timid child), this will prove the
lemma.
In the figure $t_1$ and $t_2$ represent a general pair of
transitions related by a Bell causality equation, namely
\begin{equation}
\frac{t_1}{t_2} = \frac{s_1}{s_2} \,.
\label{BCE}
\end{equation}
But, as illustrated,
each of these is also related by a Bell causality equation to the
gregarious child, to wit:
\begin{equation}
\frac{t_1}{t_3} = \frac{s_1}{s_3} \quad \mbox{and} \quad
\frac{t_2}{t_3} = \frac{s_2}{s_3}
\label{BCE2}
\end{equation}
Since (\ref{BCE}) follows immediately from (\ref{BCE2}),
no inconsistencies can arise at stage $n$, and
the lemma follows by induction on $n$. $\Box$
\begin{lemma}
Given Bell causality and the further
consequences of general covariance that are
embodied in Lemma \ref{q's}, all the remaining general covariance
equations reduce to identities, i.e. they place no further restriction
on the parameters of the theory.
\end{lemma}
\noindent \textbf{Proof:}
\begin{figure}[htbp]
\center
\scalebox{.9}{\includegraphics{consis.eps}}
\caption{\label{consis} consistency of remaining general covariance conditions}
\end{figure}
Discrete general covariance states that the probability of forming a
causet is independent of the order in which the elements arise,
i.e. it is independent of the corresponding path through the poset of
finite causets.
Now, general covariance relations always can be taken
to come from `diamonds' in the poset of causets, for the following
reason.
As illustrated in figure \ref{consis}, any two parents $A$, $C$ of a
causet $B$ will have a common parent $D$ (a ``grandparent'' of $B$)
obtained by removing two suitable elements from $B$. (By assumption $B$
must contain an element whose removal yields $A$ and another element
whose removal yields $C$. Then remove these two elements. For example,
consider the case where \wedgeo is the grandchild and it has the parent
\3antichain (by removing the maximal element of the wedge) and the parent
\wedge (by removing the disconnected element). To find the
grandparent \twoach
remove both maximal elements from \wedgeo.)
Now, still referring to figure \ref{consis},
let $|D|=n$ and suppose inductively that all the general covariance
relations are satisfied up through stage $n$. A new condition arising
at stage $n+1$ says that some path arriving at $B$ via $x$ has the
same probability as some other path arriving via $z$. But, by our
inductive assumption, each of these paths can be modified to go
through $D$ without affecting its probability. Thus, the equality of
our two path probabilities reduces simply to $ax=bz$.
Now by Bell causality and lemma \ref{q's},
\begin{displaymath}
\frac{x}{q_n} = \frac{b}{q_{n-1}} \,,
\end{displaymath}
whence
\begin{displaymath}
ax = ab \frac{q_n}{q_{n-1}} \,.
\end{displaymath}
But by symmetry, we also have
\begin{displaymath}
bz = ba \frac{q_n}{q_{n-1}} \,;
\end{displaymath}
therefore $ax=bz$, as required. $\Box$
|
\section{Angular structure of energy losses of hard jet in dense QCD-matter}
Hard jet production is considered to be an efficient probe for formation of
quark-gluon plasma (QGP)~\cite{qm97} in future experiments on
heavy ion collisions at LHC~\cite{cms94,alice}. High $p_T$ parton pair (dijet) from
a single hard scattering is produced at the initial stage of the collision process
(typically, at $\mathrel{\mathpalette\fun <} 0.01$ fm/c). It then propagates through the QGP formed due to
mini-jet production at larger time scales ($\sim 0.1$ fm/c), and interacts strongly
with the comoving constituents in the medium.
We know two possible mechanisms of energy losses of a hard partonic jet evolving
through the dense matter : $(1)$ radiative losses due to gluon "bremsstrahlung" induced
by multiple scattering~\cite{gyul94,baier,zakharov} and $(2)$ collisional losses
due to the elastic rescatterings of high $p_T$ partons off the medium
constituents~\cite{mrow91,lokhtin1}. Although the radiative energy losses of a
high energy parton can dominate over the collisional losses by up to an order of
magnitude~\cite{gyul94}, the angular distribution of the losses is essentially
different for two mechanisms. Indeed, the coherent Landau-Pomeranchuk-Migdal radiation
induces a strong dependence of the jet energy on the jet cone size
$\theta_0$~\cite{baier,lokhtin2}. With increasing of hard parton energy the maximum of
the angular distribution of bremsstrahlung gluons shifts towards the parent parton
direction. This means that measuring the jet energy as a sum of the energies of final
hadrons moving inside an angular cone with a given finite size $\theta_0$ will allow
the bulk of the gluon radiation to belong to the jet. Therefore, the medium-induced
radiation will, in the first place, soften particle energy distributions inside the
jet, increase the multiplicity of secondary particles, but will not affect the total
jet energy. On the other hand, the collisional energy losses turns out to be
practically independent on $\theta_0$ and emerges outside the narrow jet cone: the bulk
of "thermal" particles knocked out of the dense matter by elastic scatterings fly away
in almost transverse direction relative to the jet axis.
The total energy loss experienced by a hard parton due to multiple
scattering in matter is the result of averaging over the dijet production vertex
($R$, $\varphi$), the momentum transfer $t$ in a single rescattering
and space-time evolution of the medium:
\begin{equation}
\Delta E_{tot} =
\int\limits_0^{2\pi}\frac{d\varphi}{2\pi}\int\limits_0^{R_A}dR\cdot P_A(R)
\int\limits_{\displaystyle\tau_0}^{\displaystyle
\tau_L}d\tau \left( \frac{dE}{dx}^{rad}(\tau) + \sum_{b}\sigma_{ab}(\tau)\cdot
\rho_b(\tau)\cdot \nu(\tau) \right) .
\end{equation}
Here $\tau_0$ and $\tau_L = \sqrt{R_A^2-R^2\sin^2{\varphi}} - R\cos{\varphi}$ are the
proper time of the QGP formation and the time of jet escaping from the plasma;
$P_A(R)$ is the distribution of the distance $R$ from the axis of nuclei collision $z$
to the dijet production vertex;
$n_b \propto T^3$ is the density of plasma constituents of type $b$ at temperature $T$;
$\sigma_{ab}$ is the integral cross section of scattering of a jet parton $a$ off the
comoving constituent $b$.
$\frac{dE}{dx}^{rad}(\tau)$ is the
radiative energy losses per unit length; $\nu(\tau) = <Q^2 / 2m_0>$ is the thermal
average collisional energy loss of the jet parton with energy $E$ due to elastic
single scattering off a constituent of the medium with energy $m_0 \sim 3T$. The
scatterings can be treated as independent, if the mean free path of hard parton is
larger than the Debye screening radius, $\lambda \equiv \sum_{b} \sigma_{ab} n_b \gg
\mu_D^{-1}$.
We have suggested a simple generalization of BDPMS result~\cite{baier} for
$dE / dx^{rad}$ to calculate
the gluon energy deposited outside a given cone $\theta_0$, which
is based on the relation between the gluon radiation angle $\theta$ and energy
$\omega$ which holds only in {\it average}~\cite{lokhtin2}:
\begin{equation} \label{angen}
\bar{\theta}=\theta(\omega)
\simeq \theta_M\cdot \left(\frac{E_{LPM}}{\omega}\right)^\frac34,
\end{equation}
where $E_{LPM}=\mu_D^2\lambda_g$ is the minimal radiated gluon energy in the
coherent LPM regime and $\theta_M \>=\> (\mu_D\lambda_g)^{-1}\>$ is the characteristic
angle depending on the local properties of the medium.
Note that the problem of a rigorous description of the differential angular
(transverse momentum) distribution of induced radiation is complicated
by intrinsically quantum-mechanical nature of the phenomenon: large
formation times of the radiation does not allow the direction of the
emitter to be precisely defined~\cite{baier}.
\parindent 0mm
\vspace{6mm}
\epsfig{file=figtet.eps,width=70 mm}
\begin{figure}[htbp]
\end{figure}
\vspace{-15mm}
\begin{small}
Figure 1: ~The average radiative ~(solid curve)
and collisional (dashed curve) energy losses of
quark-initiated ~jet ~as ~a ~function ~of ~the jet
cone size $\theta_0$
\end{small}
\vspace{-83mm}
\hspace{78mm}
\begin{minipage}{81mm}
Figure 1 represents the average radiative (coherent medium-dependent part) and
collisional energy losses of a quark-initiated jet with initial energy $E = 100$ GeV
as a function of the jet cone size $\theta_{0}$. We have used scaling Bjorken's
solution~\cite{bjorken} for temperature and density of gluon-dominated plasma
$T(\tau) \tau^{1/3} = T_0 \tau_0^{1/3},~~ \rho(\tau) \tau = \rho_0 \tau_0$ and initial
conditions, predicted to be achieved in central $Pb-Pb$ collisions at LHC
energies~\cite{eskola94}: $\tau_0 \simeq 0.1$ fm/c, $T_0 \simeq 1$ GeV.
We can see the weak $\theta_0$-dependence of collisional losses,
at least $90 \%$ of scattered ``thermal'' particles
flow outside a rather wide cone $\theta_0 \sim 10^0 - 20^0$.
The radiative losses are almost independent of the initial jet energy and
decrease rapidly with increasing the angular size of the jet at $\theta_0 \mathrel{\mathpalette\fun >} 5^0$.
\end{minipage}
\bigskip
\section{Jet quenching: $jet + jet$, $\gamma + jet$ and $Z + jet$ production}
In a search for experimental evidences in favour of the medium-induced
energy losses a significant dijet quenching (a suppression of high $p_T$ jet
pairs)~\cite{gyul90} was proposed as a possible signal of dense matter formation in
ultra-relativistic nuclear collisions. Note that the dijet rate in $AA$ relative to $pp$
collisions can be studied
by introducing a reference process, unaffected by energy losses and with a rate
proportional to the number of nucleon-nucleon collisions, such as Drell-Yan dimuons or
Z$(\to \mu^+\mu^-)$ production:
$
R^{dijet}_{AA} / R^{dijet}_{pp} = \left( \sigma_{AA}^{dijet} /
\sigma_{pp}^{dijet} \right) / \left(\sigma_{AA}^{DY~(Z)} / \sigma_{pp}^{DY~(Z)}
\right).
$
We have studied the capability of the CMS detector~\cite{cms94} at future LHC collider
to observe the medium-induced energy losses of quarks and gluon detecting
hadronic jets in heavy ion collisions~\cite{note99_016}. The Compact Muon Solenoid
(CMS) is the general purpose detector designed to run at the LHC and optimized mainly
for the search of the Higgs boson in $pp$ collisions. However, a good muon system and
electromagnetic and hadron calorimeters with fine granularity gives the possibility to
cover important "hard probes" aspects of the heavy ion physics.
At LHC ions will be accelerated at $\sqrt{s} = 7 \times (2Z/A)$ TeV per
nucleon pair. In the case of $Pb$ nuclei $\sqrt{s} = 5.5 A$
TeV and the expected average luminosity $L \approx 1.0 \times
10^{27}$~cm$^{-2}$s$^{-1}$. The inelastic interaction cross-section for $Pb-Pb$
collisions is about 8~b, which leads to an event rate of 8~kHz.
The jet recognition efficiency and expected production rates was studied assessing the
CMS calorimeters response for the barrel part of CMS calorimeters, which covers the
pseudorapidity region of $\mid \eta \mid < 1.5$. Using the selection criterion on jet
shape allows getting the maximum efficiency of "true" hard jets
recognition as well as the maximum suppression of "false" jets background at jet
energy $E_T \sim 50-100$ GeV~\cite{note99_016}. In order to test the sensitivity of
the final hadronic jets to the energy losses, the three different
scenarios for jet quenching due to collisional energy losses of a jet
partons were studied~\cite{lokhtin1}: $(i)$ no jet quenching,
$(ii)$ jet quenching in a perfect QGP (the average collisional losses
of a hard gluon $<\Delta E_{g}> \simeq 9$ GeV, $<\Delta E_{q}> = 4/9
\cdot <\Delta E_{g}>$), $(iii)$ jet quenching in a maximally viscous QGP,
resulting in $<\Delta E_{g}> \simeq 18$ GeV.
\vspace{6mm}
\hspace{3mm}
\epsfig{file=figqu1.eps,height=70mm}
\begin{figure}[hbtp]
\end{figure}
\vspace{-4mm}
\begin{small}
Figure 2: ~The probability ~$P_{dijet}$ ~of dijet production ~with
transverse energy $E_T^{jet1,2} > E_T$ in central $Pb-Pb$ collisions
for different quenching scenarios: ~"true" hard ~(histograms)
and ~"false" (point) ~dijets, ~$dN^{\pm}/dy (y=0) = 8000$. ~Solid
curve shows ~the scaled result ~for dijet spectrum ~at parton
level as calculated with PYTHIA.
\end{small}
\vspace{-106mm}
\hspace{98mm}
\begin{minipage}{61mm}
Figure 2 represents the probability of dijet yield as a function of
$E_T$ in simulated central $Pb-Pb$ collisions for CMS.
The significant suppression of hard dijet yield due to energy losses (up to factor
$\sim 7$, or somewhat higher if the radiative energy losses mechanism is included) can
be expected. The quenching factor is almost independent of the jet energy if the
losses do not depend (or depend weakly) on the energy of an initial hard parton.
The expected statistics for dijet production will be large
enough to make the study the dijet rates as a function of impact parameter of the
collision and the transverse energy of jets. The suppression of dijet rates (jet
quenching) due to energy losses of hard partons expected to be much more stronger at
very central collisions in comparison with the peripheral one's.
\end{minipage}
\vspace{10mm}
Other possible signatures that could directly observe the energy losses involve
tagging the hard jet opposite a particle that does not interact strongly
($q + g \rightarrow q + \gamma$~\cite{wang96} and $q + g \rightarrow q +
Z(\to \mu^+\mu^-)$~\cite{kvat95} channels). The jet energy losses
should result in the non-symmetric shape of the distribution of differences in
transverse momentum between the Z-boson ($\gamma$) and jet. The estimated statistics is
rather low for Z$(\to \mu^+\mu^-)$ + jet channel. On the other hand, using
$\gamma + jet$ production is complicated due to large background from $jet+jet$
production when one of the jet in an event is misidentified as a photon (the leading
$\pi ^{0}$). However the shape of the distribution of differences in transverse energy
between the $\gamma$ and jet is well sensitive to the jet quenching effect.
It seems possible to extract the background $\gamma + jet$ events from the experimental
spectra using the background shape from Monte-Carlo simulation and (or) from $pp$ data.
\section{Conclusions}
For small angular jet cone sizes, $\theta_0\mathrel{\mathpalette\fun <} 5^0$,
the radiative energy loss is shown to dominate over the collisional
energy loss due to final state elastic rescattering of the hard
projectile on thermal particles in the medium. Due to coherent
effects, the radiative energy loss decreases with increasing the
angular size the jet. It becomes comparable with the collisional energy
loss for $\theta_0 \mathrel{\mathpalette\fun >} 5^0-10^0$. Relative
contribution of collisional losses would likely become significant for jets with finite
cone size propagating through the hot plasma under LHC conditions
Monte-Carlo study shows that CMS detector at future LHC collider is well suited for
the investigation of high transverse energy jets. Dijet production,
$Z(\to \mu^+\mu^-) + jet$ and $\gamma + jet$ channels are important for
extracting information about the properties of super-dense matter to be created in
heavy ion collisions at LHC.
\section*{Acknowledgments}
I would like to thank A.Nikitenko, L.I.Sarycheva, A.M.Snigirev and
I.N.Vardanian for collaboration on parts of this work. Discussions with M.Bedjidian,
D.Denegri, Yu.L.Dokshitzer O.Drapier, O.L.Kodolova, R.Kvatadze and D.Schiff are
gratefully acknowledged.
|
\section{Introduction}\label{intro}
Extremal principles
are fundamental in our interpretation of phenomena in
nature. One of the best known examples is the second law of
thermodynamics \cite{Keizer,Info,Pri},
governing most physical and chemical systems
and stating the continuous increase of entropy (``disorder'')
in closed systems. Most systems in our natural environment, however, are
open, which is true for driven physical systems, but even more
for biological, economic, and social systems. As a consequence,
these systems are usually characterized by self-organized structures
\cite{Info,Pri,Haken,evol,Ebeling,gamedyn,quantsoc,frank,eshel,biol1a,biol1b,%
biol1c,biol2,biol3,biol4,biol5,biol6,Drasdo,ants1,ants2,mandel,Bouchaud,solomon,stanley,peinke,%
zhang,Stauf,We91,Hub,Hel93,Galam,pre,granular,solid,trail},
which calls for principles that apply
on time scales shorter or comparable to the life spans of these systems.
For example, it is known that in growth and aggregation processes
it is usually the most unstable mode that determines the finally evolving
structure. Hence, there exists an extremum principle of fastest propagation,
which is applicable to so apparently different phenomena like
crystall growth on the one hand \cite{jacob1} and pattern formation in
bacterial colonies on the other hand \cite{jacob2}.
\par
Recent simulations point to the possible existence of
additional optimality principles in certain kinds of
driven multi-particle or multi-agent systems. As examples we mention
\begin{enumerate}
\item
lane formation in pedestrian crowds \cite{pre} (see Fig.~\ref{fig2}),
which appears to be similar to the size segregation in
sheared granular media \cite{granular},
\item
the self-organization of coherent motion in a mixture of cars and trucks
\cite{solid} (see Fig.~\ref{sol}), and
\item the evolution of trail systems \cite{trail} (see Fig.~\ref{tr}).
\end{enumerate}
In these systems, the respective interacting entities (pedestrians,
driver-vehicle units, or
particles) have to coordinate each other in order to reach a
system state which is ``favourable'' to them.
It was conjectured \cite{trail,solid} that the resulting system states are
optimal in some sense, but
there are many open questions to be addressed:
\begin{enumerate}
\item Is there really a quantity which is optimized by the
self-organizing system in the course of time?
\item If yes, which quantity is it? Is there a systematic way to
derive it?
\item What are the conditions for the existence of such a quantity?
\item Is there any systematic connection between self-organization
and optimization?
\item If optimal self-organized systems exist at all, are they
exceptional or quite common?
\end{enumerate}
We will address these questions in the following sections. A particular
problem will be, how to formulate the theoretical approach to the problem
general enough to grasp the variety of different systems mentioned
above. We will manage this by applying concepts related to those from
thermodynamics
\cite{Keizer,Info,Pri,Ebeling} as well as game theoretical ideas
\cite{Ebeling,gamedyn,quantsoc,frank,game1,game2,game3,gd,Nowak,%
life,may,Bernardo}.
\section{An Example: Lane Formation}\label{laneform}
To illustrate the non-trivial aspects of optimal self-organization,
let us consider the dynamics of pedestrian crowds. We begin with a system of
oppositely moving pedestrians in a corridor, for which lane formation has been
observed in empirical studies \cite{book}. Readers who prefer
an example from physics may instead imagine a long vertical
column of a viscous fluid with
light (rising) and heavy (sinking) particles of equal size
(where we will assume that the absolute density difference of the
particles with regard to the fluid is the same). In fact, we encourage
the reader to do this new experiment.
\par
By $\vec{x}_\alpha(t)$ we denote the position of pedestrian or
particle $\alpha$
at time $t$, by $\vec{v}_\alpha(t) = d\vec{x}_\alpha(t)/dt$
its velocity, by $v_0$ its equilibrium speed in the absence
of (interparticle) interactions,
and by $\vec{e}_\alpha$ its ``desired'' or ``preferred'' direction of motion.
Then, the equation mimicking pedestrian or describing particle motion
reads
\begin{equation}
\vec{v}_\alpha(t) = v_0 \vec{e}_\alpha + \sum_{\beta (\ne \alpha)}
\vec{f}_{\alpha\beta}\big(d_{\alpha\beta}(t)\big)
\end{equation}
(in the overdamped limit).
$\vec{f}_{\alpha\beta}$ represents repulsive interactions between
pedestrians or particles $\alpha$ and $\beta$, which were assumed to decrease
monotonically with their distance $d_{\alpha\beta}(t)=
\|\vec{x}_\alpha(t) - \vec{x}_\beta(t)\|$. We will not specify the exact form
of $\vec{f}_{\alpha\beta}$, since it turns out to be
quite irrelevant for the kind of phenomena
we want to describe, here. The forces can be even chosen in a
velocity-dependent way \cite{book,freezing}.
In cases of two opposite desired directions of motion, simulations of the
above model reproduce the
formation of lanes of uniform walking directions observed for pedestrians
(see Fig.~\ref{fig2}),
while there are no stable self-organized states in cases of four different
desired directions of motion (e.g., at intersections).\footnote{The reader
is invited to do his own simulation experiments with our Java Applets for
lane formation and intersections available on this www-page:\\
{\tt http://www.theo2.physik.uni-stuttgart.de/helbing.html}}
It is clear that lane formation will maximize
the average velocity in the respective desired walking direction
and, therefore, the quantity
\begin{equation}
E(t) = \frac{\langle\langle \vec{v}_\alpha \cdot \vec{e}_\alpha
\rangle_\alpha \rangle_t}{v_0} \le 1 \, ,
\end{equation}
which is a measure of the ``efficiency'' or ``success'' of motion.
(Here, $\langle\langle . \rangle_\alpha \rangle_t$ denotes the
average over the pedestrians and over time.) Moreover, optimization of
efficiency immediately implies that the system minimizes the quantity
\begin{equation}
\bigg\langle\!\bigg\langle - \sum_{\beta (\ne \alpha)}
\vec{f}_{\alpha\beta} \cdot \vec{e}_\alpha
\bigg\rangle_{\!\!\alpha}\bigg\rangle_{\!\!t}
= v_0 - \langle\langle \vec{v}_\alpha \cdot \vec{e}_\alpha
\rangle_{\alpha} \rangle_{t} = v_0 (1 - E) \, ,
\end{equation}
i.e., the average interaction intensity
opposite to the respective desired direction of motion.
Hence, even without the use of difficult mathematics we could show that
{\em the system minimizes the interaction intensity}
of the pedestrians (or particles), if it shows segregation into lanes
of uniform directions of motion.
Note, however, that lane formation is not a trivial effect,
but eventually arises only due to the smaller relative velocity
and interaction rate that pedestrians with the same walking
direction have (see Section~\ref{maceq}). In more detail,
the mechanism of lane formation can be understood as follows: Pedestrians
moving in a mixed crowd or moving against the stream will have frequent
and strong interactions. In each interaction, the encountering
pedestrians move a little aside in order to pass each other. This
sidewards movement tends to separate oppositely moving pedestrians.
Moreover, once the pedestrians move in uniform lanes, they will
have very rare and weak interactions. Hence, the tendency to break up
existing lanes is negligible. Furthermore, the most stable configuration
corresponds to a state with a {\em minimal interaction rate.} Therefore, lane
formation and minimal interaction rate are two sides of the same medal.
Nevertheless, lane formation does not occur in all driven repulsive systems.
There are certain conditions for it, which we will work out
later on (see Section~\ref{orgi}).
\section{The Macroscopic Equation}\label{maceq}
To give an analytical description of the macroscopic dynamics of the
considered system, we will set up continuum equations
for the pedestrian densities. By indexes $a$ and $b$,
we will distinguish different (sub-)populations
defined by the different desired walking directions.
For the mathematical description of lane formation, it is
sufficient to focus on the one-dimensional dynamics perpendicular to the
desired walking directions.
(Imagine a projection of pedestrian dynamics on
a cross section of the walkway).
The distribution of the $N_a$ pedestrians of population
$a$ over the locations ${x}$ of this one-dimensional
space will be represented by the densities $\rho_a({x},t)\ge 0$.
\par
Assuming conservation of the number
\begin{equation}
N_a = \int\limits_0^I {d}x \, \rho_a({x},t)
\label{conserv}
\end{equation}
of pedestrians in each population $a$, where $I$ denotes the spatial
extension of the system, we obtain the so-called
continuity equations\cite{Keizer}
\begin{equation}
\frac{\partial \rho_a({x},t)}{\partial t}
+ \frac{\partial}{\partial x} \Big[ \rho_a({x},t) {V}_{\!a}({x},t)
\Big] = 0 \, .
\label{cont}
\end{equation}
Here, ${V}_{\!a}({x},t)$ is the average velocity of pedestrians of
population $a$ {\em perpendicular to their desired walking direction.}
In the following, we will give a rough estimate of this velocity:
It will be proportional to the frequency $\nu_a$ of interactions that
a pedestrian of population $a$ encounters with other pedestrians.
Also, it will be proportional to the average amount $\Delta x$ that
a pedestrian moves aside when evading another pedestrian. Finally,
it will be proportional to the difference of the probabilities
$p_+$ and $p_-$ to move in positive or negative $x$-direction, respectively.
In summary, we have the relation
\begin{equation}
V_{\!a}({x},t) = c \, \nu_a \, \Delta x \, (p_+ - p_-) \, .
\label{eins}
\end{equation}
With a prefactor $c\ne 1$ like $c(x,t)=[1-\sum_a \rho_a(x,t)/\rho_{\rm max}]$,
one can take into account that the motion is slowed down in crowded areas.
It also limits the local density to the maximum density $\rho_{\rm max}$.
\par
The interaction rate of pedestrians belonging to (sub-)population $a$ with
others is
\begin{equation}
\nu_a = C_{aa} \rho_a + C_{ab} \rho_b
\quad \mbox{with} \quad b\ne a \, ,
\label{zwei}
\end{equation}
where $C_{ba} > C_{aa}$ because of the higher
average relative velocity
between oppositely moving pedestrians. (Although this inequality is
enough to know for the following discussion, we mention that
$C_{aa} = (D/I) \chi \sqrt{\pi \theta_a}$ and
$C_{ab} = C_{ba} \approx (D/I) \chi (v_a + v_b)$ \cite{complex}.
Herein, $D$ is the so-called ``total cross section'', which
corresponds to the effective diameter of a pedestrian. The factor
$\chi$ reflects the increase of the interaction rate with
growing density due to the finite space requirements of the pedestrians.
An approximate formula is
$\chi = 1/[1-\sum_a \rho_a /\rho_{\rm max}]^\kappa$,
where $\kappa \ge 1$ is a suitable constant and
$\rho_{\rm max}$ is the maximum pedestrian density. Furthermore,
$\theta_a$ is the velocity variance of pedestrians belonging to
(sub-)population $a$. Finally, $v_a > \sqrt{\theta_a}$
and $v_b > \sqrt{\theta_b}$ represent the average velocities of
subpopulations $a$ and $b$ {\em in their desired walking directions.})
\par
We will assume that the probability of moving by $\Delta x$ in
positive (or negative)
$x$-direction, when evading a pedestrian, is inversely proportional
to the interaction rate at position $x_+=x+\Delta x$
(or $x_-=x-\Delta x$, respectively):
\begin{equation}
p_+ = \frac{1/\nu_a(x_+)}{1/\nu_a(x_+) + 1/\nu_a(x_-)} \, , \qquad
p_- = \frac{1/\nu_a(x_-)}{1/\nu_a(x_+) + 1/\nu_a(x_-)} \, .
\end{equation}
A first order Taylor expansion of the nominator and the denominator
gives the following approximate relation for the difference of these
probabilities:
\begin{equation}
(p_+ - p_-) \approx - \frac{\Delta x}{\nu_a(x)} \,
\frac{\partial \nu_a(x)}{\partial x}
\label{linapp}
\end{equation}
(which, strictly speaking, is restricted to cases of small gradients
$\partial \nu_a/\partial x$ as
in the linear regime around the homogeneous solution).
Hence, with (\ref{eins}) and (\ref{zwei}), we finally obtain the following
formula for the average velocity of motion perpendicular to the
desired walking direction:
\begin{equation}
V_{\!a}({x},t) \approx - c \, (\Delta x)^2 \bigg( C_{aa}
\frac{\partial \rho_a}{\partial x} + C_{ab} \frac{\partial \rho_b}{\partial
x} \bigg) \, .
\end{equation}
Defining $S_{ab} = -(\Delta x)^2 C_{ab}$ and generalizing to an arbitrary
number $A$ of (sub-) populations gives
\begin{equation}
V_{\!a}({x},t) \approx c \, \sum_{b=1}^A S_{ab}
\frac{\partial \rho_b}{\partial x} \, .
\end{equation}
We may rewrite this in terms of a gradient
\begin{equation}
{V}_{\!a}({x},t) = c \,
\frac{\partial S_a({x},t)}{\partial x}
\label{gradient}
\end{equation}
of the linear density-dependent function
\begin{equation}
S_a({x},t) = S_a^0 + \sum_b S_{ab} \, \rho_b({x},t) \, ,
\label{success}
\end{equation}
where the constants $S_a^0$ do not matter at all. Formula (\ref{success})
can be interpreted as a linear approximation of a more general function
$S_a(x,t)$ of the densities.
Notice that, for higher-dimensional spaces, relation
(\ref{gradient}) becomes a {\em potential condition}. Later on, we will see
that this is one of the conditions which must be fulfilled for optimal
self-organization. Since it is not satisfied for pedestrian crowds
with {\em four} different desired walking directions (at intersections),
we can now understand why, in this case, there exist no optimal
self-organized patterns of motion, which are stable. Nevertheless, collective
patterns of motion like ``rotary traffic'' can form
temporarily \cite{book}.
\section{Self-Optimization} \label{optimi}
In the following, we will prove that, under certain conditions, the function
\begin{equation}
S(t) = \sum_a \int {d} x \, \rho_a({x},t) S_a({x},t)
\label{define}
\end{equation}
is a Lyapunov function which monotonically
increases in the course of time.
Notice that $S(t)$ can be viewed as being analogous
to a thermodynamic non-equilibrium potential \cite{Graham},
allowing the determination of the characteristic quantities $S_{ab}$
by functional derivatives. For example, if $S_{ba} = S_{ab}$, we have
\begin{equation}
S_{ab} = \frac{1}{2I} \, \frac{\delta^2 S}{\delta \rho_a \delta \rho_b}
\, .
\end{equation}
By deriving (\ref{define}) with respect to $t$,
using (\ref{success}), and properly interchanging indices $a$ and $b$,
one can eventually obtain
\begin{equation}
\frac{d S(t)}{d t}
= \sum_{a} \int {d} x \, \frac{\partial \rho_a({x},t)}{\partial t}
\sum_b (S_{ab} + S_{ba}) \rho_b({x},t) \, .
\label{zwi}
\end{equation}
If $S_{ab}$ is antisymmetric (i.e. $S_{ba} = -S_{ab}$), $S$ is
obviously an invariant of motion. However, in the following we will focus on
the case $S_{ba} = S_{ab}$ of symmetric interactions, which applies
to our pedestrian, granular, and trail formation examples.
Inserting (\ref{cont}) and (\ref{gradient}) into (\ref{zwi}), and
applying (\ref{success}), we get
\begin{equation}
\frac{d S(t)}{dt} = -2 \sum_a \int {d} x \,
\big[ S_a({x},t) - S_a^0 \big] \,
\frac{\partial}{\partial x} \bigg[ \rho_a({x},t)
\,c\, \frac{\partial S_a({x},t)}{\partial x} \bigg] \, .
\end{equation}
Making use of partial integration
(for spatially periodic systems), we finally arrive at
\begin{equation}
\frac{dS(t)}{dt} =
2 \sum_a \int {d} x \; c \, \rho_a({x},t) \left[
\frac{\partial S_a({x},t)}{\partial x} \right]^2 \ge 0 \, .
\label{endresult}
\end{equation}
This result
establishes {\em self-optimization} for symmetical interactions
and can be easily transferred to discrete spaces
(see Section~\ref{orgi} and Fig.~\ref{fig3}) and to higher-dimensional spaces.
In case of slightly asymmetric interactions,
small non-linear contributions to (\ref{success}), or small diffusion,
relation (\ref{endresult}) will still be a good approximation, i.e. the
system will behave close to optimal. This is exemplified by heterogeneous
freeway traffic \cite{solid}. Later on, we will see that more or less symmetric
interactions are very natural for the kind of self-organizing systems we are
considering, here (see Section~\ref{results}).
\par
Notice that (\ref{endresult}) looks similar to dissipation functions in
thermodynamics \cite{Keizer,Rayleigh,Onsager,deGroot,Graham}.
If we interpret the function $dS(t)/dt$
as a measure of dissipation per unit time in the system,
equation~(\ref{endresult}) immediately implies that {\em the system approaches
a stationary state of minimal dissipation,} since $S(t)$ is bounded for any
finite system (which means $dS(t)/dt = 0$ in the limit of large
times $t$).
This may be viewed as a generalization of the related Onsager principle
of minimal dissipation of entropy \cite{Onsager,deGroot},
dating back to Lord Rayleigh \cite{Rayleigh}. In
contrast to the non-linear, far-from equilibrium systems considered above,
the Onsager principle applies to linearly treatable, close-to-equilibrium
systems only, which usually tend to approach a homogeneous state.
\par
According to (\ref{endresult}), the stationary solution
$\rho_a^{\rm st}({x})$ is characterized by
\begin{equation}
c \, \rho_a^{\rm st}({x}) = 0 \qquad \mbox{or} \qquad
\frac{\partial S_a^{\rm st}({x})}{\partial x} = \sum_b S_{ab}
\frac{\partial \rho_b^{\rm st}({x})}{\partial x} = {0}
\label{statio}
\end{equation}
for all $a$, which is fulfilled by homogeneous or by
step-wise constant solutions. Hence, {\em the stationary solution can
be non-homogeneous, but nevertheless satisfies
minimal dissipation,} which is quite interesting.
We would not have recognized this, if we would have derived
the relations (\ref{statio}) for the stationary solution directly from
the continuity equation (\ref{cont})
with (\ref{gradient}) and (\ref{success}).
\section{The Relation with Game Theory} \label{gameth}
Physicists have recently gained a considerable interest in applications
of methods from statistical physics and nonlinear dynamics to
biological \cite{eshel,biol1a,biol1b,biol1c,biol2,biol3,biol4,biol5,biol6,%
Drasdo,ants1,ants2},
economic \cite{mandel,Bouchaud,solomon,stanley,peinke,zhang,Stauf,We91},
and social \cite{We91,Hub,Hel93,Galam,Ebeling,gamedyn,quantsoc,frank,gd}
systems. Hence, it is worth stressing that the above model can be applied to
such kinds of systems, if we give a more general interpretation to it.
First of all, instead of particles or pedestrians, we may have other kinds
of entities, which we again need to subdivide into uniform (sub-)populations,
according to their behaviors. Second, the space may be rather abstract than
real, for example, a behavioral space or an opinion spectrum
\cite{quantsoc,Hel93}.
The same applies to motion, which may correspond to a change of behavior or
opinion. (In the case of trail formation, a point $\vec{x}$ in space
even corresponds to a {\em path} connecting two places, namely
a pedestrian source with one of their destinations.)
Having again a look at the relation (\ref{gradient}),
it makes sense to interpret
the function $S_a({x},t)$ as the ``(expected) success''
per unit time for an entity of population $a$ at location ${x}$,
since it is plausible that the entities move into the direction of the
greatest increase of success (which is the direction of the gradient).
Furthermore, one can give a more concrete meaning to the coefficients $S_{ab}$:
If an entity of kind $a$ interacts with
entities of kind $b$ at a rate $\nu_{ab}$ and the associated
outcome of the interaction can be quantified by some ``payoff'' $P_{ab}$, we
have the relation $S_{ab} = \nu_{ab}P_{ab}$.
Positive payoffs $P_{ab}$ belong to attractive or, more general, {\em profitable}
interactions between populations $a$ and $b$, while negative ones
correspond to repulsive or {\em competitive} interactions. We think, it is quite
surprising that, based on competitive interactions, there can be
self-organized and even optimal system states at all
(see Section~\ref{results}).
Notice that, despite of the mentioned relations with game theory,
our model differs from the conventional
game dynamical equations \cite{Ebeling,gamedyn,quantsoc}
in several respects:
\begin{enumerate}
\item We have a topology (like in the {\em game of life} \cite{life,may}),
but define abstract games for interactive
motion in space with the possibility of local agglomeration
at a fixed number of entities in each population.
\item The payoff does not depend on the variable that the individual entities
can change (i.e. the spatial coordinate $x$).
\item Individuals can only improve their success by redistributing
themselves in space.
\item The increase of success is not proportional to the
difference with respect to the global average of success,
but to the local change of success in a population.
\end{enumerate}
Moreover, below we will establish a new connection between self-optimization
and self-organization (see Section~\ref{results}),
which we consider to be important.
\section{Self-Organization} \label{orgi}
Our generalized model allows to describe
all kinds of different combinations between attractive or
profitable and repulsive or competitive interactions within and among the
different populations. It is, therefore, desireable to know the exact
conditions under which the corresponding
system forms a self-organized, i.e. a non-homogeneous
state. In order to derive these, we will carry out a linear stability analysis
around the homogeneous stationary solution $\rho_a^{\rm hom}= N_a/I$,
where $I$ again denotes the spatial extension of the system.
For simplicity, we will restrict
ourselves to the case of two (sub-)populations $a, b \in \{1,2\}$.
We start with the continuity equation
\begin{equation}
\frac{\partial \rho_a}{\partial t}
+ \frac{\partial}{\partial x} \bigg[ \rho_a c \,
\frac{\partial S_a}{\partial x} \bigg] = c \, D_a
\frac{\partial^2 \rho_a}{\partial x^2} \, ,
\label{contdiff}
\end{equation}
where we have introduced an additional
diffusion term with diffusion coefficient $c\,D_a$
on the right-hand side, since we will discuss the effect of fluctuations
later on. Next, we write down the corresponding linearized partial
differential equations:
\begin{eqnarray}
\frac{\partial \rho_a}{\partial t}
&=& - c\,\rho_a \frac{\partial^2 S_a}{\partial x^2}
+ c \, D_a \frac{\partial^2 \rho_a}{\partial x^2} \nonumber \\
&=& - \sum_{b=1}^2 c\, \rho_a S_{ab} \frac{\partial^2 \rho_b}{\partial x^2}
+ c\,D_a \frac{\partial^2 \rho_a}{\partial x^2} \, .
\end{eqnarray}
Inserting into these equations the {\em ansatz}
\begin{equation}
\rho_a(x,t) = \rho_a^{\rm hom} + \tilde{\rho}_a \,
\mbox{e}^{{\rm i}kx + \lambda t} \, ,
\end{equation}
where $k$ has the meaning of a wave number,
leads to the following linear eigenvalue problem with
eigenvalue $\lambda$:
\begin{equation}
\lambda \left(
\begin{array}{c}
\tilde{\rho}_1 \\
\tilde{\rho}_2
\end{array} \right) =
\left( \begin{array}{cc}
A_{11} & A_{12} \\
A_{21} & A_{22}
\end{array} \right)
\left(
\begin{array}{c}
\tilde{\rho}_1 \\
\tilde{\rho}_2
\end{array} \right) \, .
\end{equation}
Herein, we have $A_{11} = c\, k^2 (\rho_1^{\rm hom} S_{11} - D_1)$,
$A_{12} = c\, k^2 \rho_1^{\rm hom} S_{12}$,
$A_{21} = c\, k^2 \rho_2^{\rm hom} S_{21}$, and
$A_{22} = c\, k^2 (\rho_2^{\rm hom} S_{22} - D_2)$.
The linear system of equations can be solved for the two eigenvalues
\begin{equation}
\lambda_{1/2} = \frac{A_{11}+A_{22}}{2} \pm \frac{\sqrt{(A_{11}+A_{22})^2
- 4 (A_{11}A_{22} - A_{12}A_{21}) }}{2} \, .
\end{equation}
In order for the homogeneous solution to be stable, the real values of
both eigenvalues need to be negative. This requires
\begin{equation}
(A_{11}+A_{22}) < 0 \quad \mbox{and} \quad
(A_{11}A_{22} - A_{12}A_{21}) > 0\, .
\label{relations}
\end{equation}
Notice that, in the case of an unstable eigenvalue, it is the mode with
the largest wave number (i.e. with the shortest wave length) that grows
fastest. This somewhat unrealistic behavior is a consequence of
simplifications made, namely the linear approximation underlying
relation (\ref{linapp}).
In reality, the spatial extension of the entities will
introduce a natural cutoff for the wave lengths. One may consider this in
equation (\ref{contdiff}) by additional spatial derivatives of higher
(e.g. fourth) order. However, one can easily
circumvent these problems by setting up a discrete version of the model
(where the spatial discretization should be chosen in accordance with
$\Delta x$).
The results of Figures~\ref{fig3} and \ref{fig4},
for example, were obtained with the following discrete analogue
of the model defined by Eqs. (\ref{cont}), (\ref{gradient}),
and (\ref{success}).
We assumed a periodic lattice
with $I$ lattice sites $x\in\{1,\dots,I\}$ and two
populations $a\in\{1,2\}$ with a total of $N=N_1+ N_2 \gg I$ entities.
(In the figures we used $I=40$ and $N_1 = N_2 = 200$.)
At time $t=0$,
our simulations started with a random initial distribution of the entities.
Then, we repeatedly applied the following update steps to determine the
distribution of entities at time $t+1$:
\begin{enumerate}
\item Calculate the successes
\begin{equation}
S_a(x,t) = S_a^0 + \sum_b S_{ab} \, \frac{n_x^b(t)}{I} \, ,
\end{equation}
where $n_x^b(t) = \rho_b(x,t) I$ represents the
number of entities of population $b$ at site $x$.
\item For each entity $\alpha$,
determine a random number $\eta_\alpha$ that is uniformly
distributed in the interval $[0,S_{\rm max}]$ with a large constant
$S_{\rm max}$ ($S_{\rm max}=20$ in the figures).
\item Move entity $\alpha$ belonging to population $a$ from site $x$ to site
$x+1$, if
\begin{equation}
c(x+1,t)[S_a(x+1,t) - S_a(x-1,t)] > \eta_\alpha(t) \, ,
\label{label1}
\end{equation}
but to site $x-1$, if
\begin{equation}
c(x-1,t)[S_a(x-1,t) - S_a(x+1,t)] > \eta_\alpha(t) \, .
\label{label2}
\end{equation}
(Figure~\ref{fig3} and \ref{fig4} are for
$c(x,t)=1$.) In cases, where we assumed errors in the estimation of the
expected success, we replaced $S_a(x,t)$ by $S_a(x,t) + \xi_\alpha(t)$,
where $\xi_\alpha(t)$ is determined according to some
probablility distribution.
\end{enumerate}
We applied a random sequential update rule,
which is most reasonable \cite{Bernardo}.
However, a parallel update, which defines a simple cellular automaton
\cite{Wolfram,Stauffer}, yields
qualitatively the same results (even nicer looking ones, because it does not
have the fluctuations caused by a random update).
\section{Results and Discussion} \label{results}
Let us first discuss the case $D_a \approx 0$ of negligible diffusion.
For this case, the relations (\ref{relations}) imply
that the homogeneous solution of the model
(cf. Figure~\ref{fig3}a) is unstable if
\begin{equation}
\rho_1^{\rm hom} S_{11} + \rho_2^{\rm hom} S_{22} > 0
\label{cond1}
\end{equation}
or
\begin{equation}
S_{12} S_{21} > S_{11} S_{22} \, .
\label{cond2}
\end{equation}
In other words, if one of the conditions (\ref{cond1}) or (\ref{cond2})
is fulfilled, the stable stationary solution is a self-organized,
non-homogeneous state which, according to (\ref{statio}),
corresponds to complete segregation (or aggregation).
If the populations interact
in a symmetric way, the underlying self-organization process is related to
self-optimization (see (\ref{endresult})).
Condition (\ref{cond1}) is satiesfied by systems in which the interactions
within the (sub-) populations are attractive or profitable. Such systems show
always some form of agglomeration (see Figs.~\ref{fig3}c and d).
In the following, we focus on the more common and much more interesting cases
where (\ref{cond1}) is not valid. This will allow us to answer the question,
why self-organizing systems of the considered type tend to reach an
optimal state. The solution is: ``because they tend to be symmetric!''.
This can be seen as follows: Introducing
\begin{equation}
\overline{S} = \frac{S_{12} + S_{21}}{2} \quad \mbox{and} \quad
\Delta S = \frac{S_{12}-S_{21}}{2} \, ,
\end{equation}
we have
\begin{equation}
S_{12} = \overline{S} + \Delta S
\quad \mbox{and} \quad
S_{21} = \overline{S} - \Delta S \, ,
\end{equation}
and condition (\ref{cond2}) becomes
\begin{equation}
(\overline{S}+\Delta S)(\overline{S}-\Delta S)
= \overline{S}^{\,2} - (\Delta S)^2 > S_{11}S_{22} \, .
\end{equation}
Since the interaction strengths
$|S_{11}|$, $|S_{22}|$, $|S_{12}|$, $|S_{21}|$, and
$|\overline{S}|$ will normally have the same order of magnitude, this
condition for self-organization can only be fulfilled for small $|\Delta S|$,
i.e.
\begin{equation}
S_{12} \approx S_{21} \, .
\end{equation}
Hence, there is a tight connection between self-organization and
self-optimization
in the considered kinds of systems: {\em If there is
self-organization,
it is likely to come with optimality,} at least approximately.
For this reason,
one may speak of a ``self-organized system with optimality'' or more compact of
``self-organized optimality''
(although there is no immediate relation
with ``self-organized criticality''\cite{SOC}). However, the more
precise term is probably ``optimal
self-organi\-za\-tion''.\footnote{This implies that there are also
``non-optimal'' forms of self-organization like in systems for which
no Lyapunov function exists. A typical example are systems with
oscillating or chaotic states.}
A good example is uni-directional multi-lane traffic
of cars and trucks \cite{solid}, which develops a self-organized,
coherent state only in a small density range, which is also
characterized by minimal interactions
(minimal lane-changing and interaction rates). From the above, we conclude
that, only in this density range, the interactions of cars and trucks
become sufficiently symmetric to give rise to self-organization and
self-optimization. This conclusion is quite reasonable, since, according to
the assumed traffic model, the interactions of
cars and trucks are more symmetric when their average velocities
are similar.
\par
In our pedestrian example, we find optimal self-organization,
since the symmetry condition is satisfied exactly,
and condition (\ref{cond2}) is also
fulfilled (see Eq. (\ref{zwei}) and below).
Note, however, that there are many other examples of segregation in
the natural and social sciences \cite{granular,seg1,seg2,seg3,We91}.
We point out that the spatial regions occupied by one population need
not be connected (cf. Figures~\ref{fig3}b to \ref{fig3}d), and that
the corresponding configuration may correspond to
a {\em relative} optimum as in Figures~\ref{fig3}c and \ref{fig3}d.
If $S_{aa}< 0$ for all $a$, the distributions $\rho_a^{\rm st}({x})$
tend to be flat, as in the case of
lane formation by repulsive pedestrian interactions (Figure~\ref{fig3}b).
Instead, we have agglomeration (local clustering),
if $S_{aa} >0$ for all $a$ (Figures~\ref{fig3}c and \ref{fig3}d).
Figure \ref{fig3}c describes the
segregation of populations with repulsive
interactions (e.g. ``ghetto formation'').
The case of Figure~\ref{fig3}d allows to understand the conjectured
optimality of trail systems which, based on attractive interactions,
result by a bundling of
trails ending at different destinations \cite{trail}.
\par
Finally, we discuss the influence of fluctuations.
Including the effect of diffusion,
the conditions for self-organization will become
\begin{equation}
\rho_a^{\rm hom} S_{aa} + \rho_b^{\rm hom} S_{bb} > D_a + D_b
\label{cond3}
\end{equation}
or
\begin{equation}
S_{ab} S_{ba}\rho_a^{\rm hom} \rho_b^{\rm hom}
> (S_{aa} \rho_a^{\rm hom} - D_a)(S_{bb} \rho_b^{\rm hom} - D_b) \, .
\label{cond4}
\end{equation}
Hence, {\em large\/} diffusion coefficients will produce homogeneous
equilibrium states, and the principle (\ref{endresult})
of self-optimization is not valid anymore
(see Figure~\ref{fig3}e).\footnote{For example,
lane formation in streams of oppositely moving
particles can be suppressed by sufficiently strong fluctuations,
giving rise to ``frozen'' (blocked) states \cite{freezing}.}
However, {\em small diffusion can further self-organization.} Not only
will the system be able to escape relative optima and
eventually reach the global optimum (cf. Figures~\ref{fig3}f and
\ref{fig4}), although the
interactions are short-ranged (see Eq. (\ref{gradient})).
According to (\ref{cond4}), small diffusion can also reduce the
stability of the homogeneous
stationary solution, which is quite surprising.
\section{Summary and Outlook} \label{outlook}
Our investigations were motivated by the question why many self-organized
systems seem to optimize certain macrosopic quantities, for which we have
given a number of realistic examples. In order to explain this, we have
derived macroscopic equations for lane formation in systems
of oppositely moving driven particles or pedestrians. We could show that,
in cases of repulsive interactions, the system tends to minimize
the interaction rate and the intensity of interactions.
The applicability of the
model, however, could be extended to cases of attractive interactions and,
using game theoretical ideas, also to competitive or profitable
interactions in many kinds of non-physical systems like biological,
economic, or social ones.
After having shown that many driven systems can be represented as
a game between interacting (sub-)populations, we have constructed a
functional for such systems, which is related to
thermodynamic non-equilibrium potentials and can
be interpreted as overall (expected) success. In cases of symmetric
interactions among the populations,
this function increases monotonically in the course of time,
meaning that the overall success of these systems is optimized.
In other words, as individual entities are trying to maximize
their {\em own} success, these systems tend to reach a state with
the highest {\em global} success, which is not trivial at all.
Since the form of the increasing
function reminds of a generalized thermodynamic dissipation
function, one can also say that the system approaches a
state of minimal dissipation (which may be considered as a generalized
Onsager principle). This principle of ``minimal waste of energy''
may be particularly interesting for biology, where it can be conjectured
that organisms use energy very efficient. For example, it is known that
pedestrians tend to move at the speed which is least energy consuming
\cite{Weidmann}. We think it is worth pointing out that
there are quite a number of living systems (for which we
have given some realistic examples) to which existing
methods and notions from thermodynamics
can be successfully applied, if they are generalized in a suitable way.
It would be surely interesting to look for other systems, for which similar
results or principles can be found.
Here, we obtained that
the precondition for self-organization with self-optimization
is symmetric interactions.
However, we could give arguments indicating that {\em the considered systems,
if they self-organize at all, are also (more or less) symmetric and, hence,
behave (close to) optimal.}
Therefore, the phenomenon of ``self-organized optimality'' or
``optimal self-organization'', as we call it, is expected
to be quite common in nature. Already for
two symmetrically interacting populations, one can classify more than
ten different
situations (dependent on whether $S_{11}$, $S_{22}$, and $S_{12}=S_{21}$
are smaller or greater than zero,
and whether conditions (\ref{cond1}) or (\ref{cond2})
are fulfilled or not).
This includes quite surprising cases
as for repulsive interactions within each population
and stronger attractive interactions between them
(e.g. $S_{11}=S_{22}=-1$, $S_{12}=S_{21}=2$), which leads to agglomeration
analogous to Figure~\ref{fig3}d rather than homogeneously distributed,
mixed populations as in Figure~\ref{fig3}a. Apart from the results displayed
in Figure~\ref{fig3}, there are also cases where one population agglomerates,
but the other one is distributed homogeneously (if we choose $S_{11}$ different
from $S_{22}$). In systems with diffusion, the
variety of different cases covered by the above approach is even greater.
In particular, we have observed that, while large noise generally destroys
self-organized solutions, {\em small noise can further self-organization}
in the considered systems, which is surprising.
Finally, we point out that our results are relevant for practical applications.
For example, when optimizing multi-agent systems (like
the coordination of vehicle or air traffic,
or the usage of CPU time in computer networks),
it is desireable to apply control strategies that
are insensitive to system failures (like the temporary breakdown of a control
center). Hence, it would be favourable if the system would
optimize its state by means of the interactions
in the system. For this, one needs to implement a suitable
type of interactions (namely, symmetric ones), which can be reached by
technical means (intelligent communication devices determining the
proper actions of the interacting entities).
Notice that this kind of multi-agent optimization is
decentralized and, therefore, much more robust than classical,
centralized control approaches.
{\em Acknowledgments:}
D.H. is grateful to the DFG for financial support
by a Heisenberg scholarship.
This work was in part supported by OTKA F019299 and FKFP 0203/1997.
The authors also want to thank Ill\'{e}s J. Farkas
for producing Figure~\ref{fig2} and
Robert Axelrod, Eshel Ben-Jacob, and Martin Treiber for valuable comments.
\clearpage
|
\section{Introduction}
There are a number of observational evidences to believe that a large fraction of the
matter in the Universe exists in the form of non-baryonic particle dark
matter. Supersymmetric neutralino is one of the most plausible
candidates for such exotic particle dark matter.
Various experimental efforts are being made aiming at detection of low energy nuclear
recoils caused by the elastic scatterings of
the neutralinos off nuclei\cite{dark matter search}.
Conventional detectors like
semiconductor detectors or scintillators generally have a quenching
factor less than unity. Here the quenching factor is defined as
the ratio of the energy detection efficiency for a nuclear recoil to
that for an electron. On the other hand, because the bolometer
is sensitive to the whole energy deposited in the absorber the
quenching factor of the bolometer should be unity in
principle. Actually a quenching factor close to unity has been
measured by Milan group\cite{Alessandrello}.
We have been developing bolometers with
lithium fluoride absorbers\cite{Tokyo}. Fluorine is
considered to have a large cross section for elastic scattering of
the axially-coupled neutralino off the nucleus compared with other
nuclei\cite{Ellis}. Recently we have
successfully constructed the bolometer array with a total mass of
168\,g and installed it in the Nokogiriyama underground cell with
a depth of 15\,m w. e.
In this paper we report on the first results from the experiment
performed in the Nokogiriyama underground cell using the bolometer array.
\section{Experimental Set-up and Measurement}
The bolometer array used in this work contains eight 21\,g LiF
bolometers. The schematic drawing of the bolometer array
is shown in Fig.\,\ref{fig:multi}. The neutron transmutation doped
(NTD) germanium thermistors with the similar
temperature dependence of the resistance\cite{NTD} are attached to the
crystals. The thermistor senses a small temperature rise of the
absorber crystal caused by the neutralino-nucleus scattering. Each
crystal is placed on four copper posts and
thermally insulated by the Kapton sheets. Moderate thermal anchoring of the
crystal to the copper holder with a temperature of 10\,mK is realized
by a oxygen free copper (OFC) ribbon. The lithium fluoride crystals are checked by
a low-background Ge spectrometer prior to the construction of the
bolometer. The concentration of radioactive contaminations is less
than 0.2\,ppb for U, 1\,ppb for Th, and 2\,ppm for K. The bolometer
array is mounted on a mixing chamber of a dilution refrigerator which
is mostly made of low-radioactivity materials radio-assayed in advance by low-background Ge spectrometer.
Each thermistor is biased through a 100\,M$\Omega$ load
resistor. The voltage change across the thermistor is fed into the
eight channel source follower circuit placed at the 4\,K
stage which include a low noise junction field effect transistor (J-FET),
Hitachi 2SK163. Since the J-FET does not work at this low
temperature, it is connected to a printed circuit board with thin stainless
steel tubes and manganin wires to be thermally isolated from the
circuit board with a temperature of 4\,K and the temperature of the
FET is maintained above 100\,K by the heat produced by itself.
The signal from the source follower circuit is in turn
amplified by an eight channel voltage amplifier placed just above
the refrigerator.
The output of the voltage amplifier is fed into a double
pole low-pass filter with a cut-off frequency of 226\,Hz and in turn into the 16-bit waveform digitizer to record the pulse shape of
the signal for off-line analysis.
The passive radiation shielding consists of
10\,cm-thick oxygen free high conductivity copper layer, 15\,cm-thick
lead layer, 1\,g cm$^{-2}$-thick boric acid layer and 20\,cm-thick
polyethylene layer. The latter two layers act as a neutron shield. In
order to avoid muon-induced background we employ a veto system
which consists of 2\,cm-thick plastic scintillators.
The constructed detector system is installed in the Nokogiriyama
underground cell which is located about 100\,km south from Tokyo and
relatively easy to access. The depth of an overburden of sand is inferred to be
about 15\,m w.e. In this work six
bolometers of the bolometer array are
used and energy spectrum are measured for about ten days. Two
bolometers have some problems in the
cooling procedure.
Since the detector is enclosed in a cryogenic vacuum can during the
measurements it is impossible to place the gamma-ray source close to
the detector for energy calibration. The energy calibration during the
measurements is, therefore, performed by 662\,keV gamma-rays from a
$^{137}$Cs source and 1333\,keV and 1173\,keV gamma-rays from a
$^{60}$Co source placed outside a helium dewar of the dilution refrigerator
and inside the radiation shieldings.
Furthermore, the sharp peak at 4.78\,MeV due to the neutron capture
reaction of $^6$Li observed in the background spectrum is also used for
pthe energy calibration. Fig.\,\ref{fig:calibration} shows one
of the obtained energy calibration plots. Linearities of the six
bolometers up to 5\,MeV are recognized. It must be noted
that linearity down to 60\,keV gamma-ray is confirmed prior to this
measurement using gamma-ray from $^{241}$Am source set inside the cryostat.
\section{Energy Spectra and Dark Matter Limits}
Fig.\,\ref{fig:spectrum} shows the energy spectra obtained by
the six bolometers during ten days. The
bump in the low energy region is considered to be due to microphonics
caused by a helium liquefier which recondenses evaporated
helium gas from the dewar. While the similar spectra are obtained for
four bolometers (D3, D5, D6, and D8), the spectra for the other two
bolometers (D1 and D4) are affected
by microphonics below 30 to 40\,keV because of their low detector gains.
Comparing the measured energy spectrum with the expected recoil spectrum, the
exclusion limits for the cross section for elastic neutralino
scattering off the nucleus can be extracted. The calculation is
performed in the same manner used in Ref.\,\cite{smith}. The theoretical recoil
spectrum is calculated
assuming a Maxwellian dark matter velocity distribution with rms
velocity of 230\,km/s, and then folded with the measured energy
resolution and the nuclear form factor.
We also assume the local halo density of the neutralino to be
0.3\,GeV/cm$^3$. The spin factors calculated assuming an odd group
model as a nuclear shell model are 0.75 for $^{19}$F and 0.417 for
$^{7}$Li\cite{Ellis}. Since the detector responses for the six
bolometers are not the same, the upper limit of the cross
section is evaluated independently from the
spectrum of each detector. For a given neutralino mass the lowest
value of the cross section is taken as a combined limit from the
results of the six detectors.
The calculated exclusion limits in case of the
spin-dependent interaction are given in Fig.\,\ref{fig:limit}. For comparison the exclusion limits derived from the data
in the other experiments at deep underground
sites\cite{EDELWEISS,BPRS,DAMA,smith,OSAKA} and the scatter plots
predicted in the minimal supersymmetric theories are also shown.
Although the other experiments except for Osaka experiment are
performed at deep underground laboratories, our experiment gives comparable limits for the light neutralino. This
owes to the large cross section for the spin-dependent interaction of
$^{19}$F and the low energy threshold of the bolometer.
The sensitivity for neutralinos with a mass below 5\,GeV is
improved by this work.
\section{Prospects}
Compton scattered gamma-rays from the aperture of the shielding and
gamma-rays produced through the interaction of cosmic ray muon within the
shielding materials are considered major background sources.
In order to reduce the muon-correlated background, the veto
efficiency must be improved. The present incompleteness of the veto is
due to the penetrations for vacuum tubes and the tube of the helium liquefier.
Increasing of the coverage of the plastic scintillator will improve
the veto efficiency up to 98\%. Against the Compton scattered gamma-ray
background, internal lead shielding with a thickness of 20\,mm
surrounding the lithium fluoride
bolometer array will be installed. The shielding is made of over 200 year old low-activity lead with a concentration of $^{210}$Pb of
less than 0.05\,pCi/g. The Compton scattered
gamma-rays can be reduced by two orders of magnitude by this internal
shielding. Since lead fluorescence X-rays are produced mainly by the
muon interaction, their contribution can be ignored if muons are sufficiently vetoed. If
these improvements are realized, the sensitivity of
this experiment will be improved by more than an order of magnitude
even at this shallow depth.
The detector system will be installed
in a underground facility with a sufficient depth where cosmic muon
induced background is expected to be negligible. The long-term
measurements in a deep underground site will bring the sensitivity to
the spin-dependent interaction below the level predicted by
the supersymmetric theory.
\section*{Acknowledgments}
We would like to thank Prof. Komura for providing us with the
low-radioactivity old
lead. This research is supported by the Grant-in-Aid for COE Research by the
Japanese Ministry of Education, Science, Sports and Culture. W.O. are
grateful to Special Postdoctoral Researchers Program for support of
this research.
|
\section{Proof of~$\W^{(1)}=I$}
\apxlabel{woneident}
\ifmypprint
{\small
\begin{verbatim}
$Id: apx-woneident.tex,v 2.1 1998/11/13 08:38:25 jeanluc Exp $
\end{verbatim}
}
\fi
Out goal is to demonstrate that through a series of lower-triangular
coordinate transformations we can make~$\W^{(1)}$ (which has an~$n$-fold
degenerate eigenvalue equal to unity) equal to the identity matrix, while
preserving the lower-triangular nilpotent form
of~\hbox{$\W^{(2)},\dots,\W^{(n)}$}.
We first show that we can always make a series of coordinate transformations
to make~\hbox{${\W_\lambda}^{11} = {\delta_\lambda}^1$}. First note that if
the coordinate transformation~$\M$ is of the form~\hbox{$\M = I + L$},
where~$I$ is the identity and~$L$ is lower-triangular nilpotent,
then~\hbox{${\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}^{(1)}=\M^{-1}\,\W^{(1)}\,\M$} still has eigenvalue~$1$, and
for~$\mu>1$ the~\hbox{${\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}^{(\mu)}=\M^{-1}\,\W^{(\mu)}\,\M$} are still
nilpotent.
For~$\lambda>1$ we have
\begin{equation}
{\overline\W_\lambda}^{11} = {{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\lambda}{}^{11}
+ {{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\lambda}{}^{1\nu}\,{L_\nu}^1
= {{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\lambda}{}^{11}
+ \sum_{\nu=2}^{\lambda-1}{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\lambda}{}^{1\nu}\,{L_\nu}^1
+ {L_\lambda}^1,
\eqlabel{apxcoordtrans}
\end{equation}
where we used~\hbox{${{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\lambda}{}^{1\lambda}=1$}. Owing to the triangular
structure of the set of equations~\eqref{apxcoordtrans} we can always solve
for the~\hbox{${L_\lambda}^1$} to make~${\overline\W_\lambda}^{11}$ vanish.
This proves the first part.
We now show by induction that if~\hbox{${\W_\lambda}^{11} =
{\delta_\lambda}^1$}, as proved above, then~$\W^{(1)}$ is the identity
matrix. For~\hbox{$\lambda = 1$} the result is trivial. Assume
that~\hbox{${\W_\mu}^{1\nu} = {\delta_\mu}^\nu$}, for~\hbox{$\mu <
\lambda$}. Setting two of the free indices to one, Eq.~\eqref{Wjacob} can be
written
\[
\begin{split}
{\W_\lambda}^{\mu 1}\,{\W_\mu}^{1\sigma}
&= {\W_\lambda}^{\mu\sigma}\,{\W_\mu}^{11}\\
&= {\W_\lambda}^{\mu\sigma}\,{\delta_\mu}^{1}
= {\W_\lambda}^{1\sigma}\,.
\end{split}
\]
Since~$\W^{(1)}$ is lower-triangular the index~$\mu$ runs from~$2$
to~$\lambda$ (since we are assuming~\hbox{$\lambda > 1$}):
\[
\sum_{\mu = 2}^{\lambda}
{\W_\lambda}^{\mu 1}\,{\W_\mu}^{1\sigma}
= {\W_\lambda}^{1\sigma}\,,
\]
and this can be rewritten, for~$\sigma<\lambda$,
\[
\sum_{\mu = 2}^{\lambda-1}
{\W_\lambda}^{\mu 1}\,{\W_\mu}^{1\sigma}
= 0\,.
\]
Finally, we use the inductive hypothesis
\[
\sum_{\mu = 2}^{\lambda-1}
{\W_\lambda}^{\mu 1}\,{\delta_\mu}^{\sigma}
= {\W_\lambda}^{\sigma 1}= 0\,,
\]
which is valid for~$\sigma < \lambda$. Hence,~\hbox{${\W_\lambda}^{\sigma 1}
= {\delta_\lambda}^{\sigma}$} and we have proved the result.
(${\W_\lambda}^{\lambda 1}$ must be equal to one since it lies on the diagonal
and we have already assumed degeneracy of eigenvalues.)
\section{Antireduction}
\input{sec-conclusion}
\begin{ack}
The authors thank Tom Yudichak for his comments and suggestions. This work
was supported by the U.S. Department of Energy under contract
No.~DE-FG03-96ER-54346. J-LT also acknowledges support from the Fonds pour
la Formation de Chercheurs et l'Aide \`a la Recherche du Canada.
\end{ack}
\section{Casimir Invariants for Low-order Extensions}
\seclabel{caslowdim}
\ifmypprint
{\small
\begin{verbatim}
$Id: sec-casilowext.tex,v 2.1 1998/11/13 08:37:39 jeanluc Exp $
\end{verbatim}
}
\fi
Using the techniques developed so far, we now find the Casimir invariants for
the low-order extensions classified in \secref{lowdimext}. We first find the
Casimir invariants for the solvable extensions, since these are also
invariants for the semidirect sum case. Then, we obtain the extra Casimir
invariants for the semidirect case, when they exist.
\subsection{Solvable Extensions}
Now we look for the Casimirs of solvable extensions.
As mentioned in \secref{localcas}, the Casimirs associated with null
eigenvectors (the only kind of eigenvector for solvable extensions) are
actually conserved locally. We shall still write them in the
form~$\Casi=f(\fv^n)$, where~$\Casi$ is as in
\eqref{formcas}, so they have the correct form as invariants for the
semidirect case of \secref{lowsemicasi} (for which they are no longer locally
conserved).
\subsubsection{n=1}
Since the bracket is Abelian, any function~$\Casi=\Casi(\fv^1)$ is a Casimir.
\subsubsection{n=2}
For the Abelian case we have~$\Casi=\Casi(\fv^1,\fv^2)$. The only other case
is the Casimir of the Leibniz extension,
\[
\Casi(\fv^1,\fv^2) = \fv^1\afi(\fv^2) + \afii(\fv^2).
\]
\subsubsection{n=3}
\seclabel{Casn3}
As shown in \secref{neqthree}, there are four cases. \caseref{n4-00} is the
Abelian case, for which any function~$\Casi=\Casi(\fv^1,\fv^2,\fv^3)$ is a
Casimir. \caseref{n4-01} is essentially the solvable part of the CRMHD
bracket, which we treated in~\secref{CRMHDCas}. \caseref{n4-10} is a direct
sum of the Leibniz extension for~$n=2$, which has the bracket
\[
\lpb(\alpha_1,\alpha_2)\com(\beta_1,\beta_2)\rpb
= (0,\lpb\alpha_1\com\beta_1\rpb),
\]
with the Abelian algebra~$\lpb\alpha_3\com\beta_3\rpb=0$. Hence, the Casimir
invariant is the same as for the~$n=2$ Leibniz extension with the
extra~$\fv^3$ dependence of the arbitrary function (see \secref{Casdirprod}).
Finally, \caseref{n4-11} is the Leibniz Casimir. These results are summarized
in \tabref{Casimirn3}.
Cases~\ref{case:n4-00} and~\ref{case:n4-10} are trivial extensions, that is,
the cocycle appended to the $n=2$ case vanishes. The procedure of then adding
$\fv^n$ dependence to the arbitrary function works in general.
\begin{table}
\caption{Casimir invariants for solvable extensions of order~$n=3$.}
\tablabel{Casimirn3}
\vskip 1em
\begin{center}
\begin{tabular}{ll} \hline
Case & Invariant \\[6pt] \hline
1 & $\Casi(\fv^1,\fv^2,\fv^3)$
\\[9pt]
2 & $\fv^1\afi(\fv^3) + \fv^2\afii(\fv^3) + \afiii(\fv^3)$
\\[9pt]
3 & $\fv^1\afi(\fv^2) + \afii(\fv^2,\fv^3)$
\\[9pt]
4 & $\fv^1\afi(\fv^3) + \frac{1}{2}(\fv^2)^2\afi'(\fv^3)
+ \fv^2\afii(\fv^3) + \afiii(\fv^3)$
\\[9pt] \hline
\end{tabular}
\end{center}
\end{table}
\subsubsection{n=4}
\seclabel{Casn4}
As shown in \secref{neqfour}, there are nine cases to consider. We
shall proceed out of order, to group together similar Casimir invariants.
Cases~\ref{case:n4-00}a,~\ref{case:n4-01},~\ref{case:n4-10}a,
and~\ref{case:n4-11}a are trivial extensions, and as mentioned in
\secref{Casn3} they involve only addition of $\fv^4$ dependence to
their~$n=3$ equivalents. \caseref{n4-10}b is a direct sum of two~$n=2$
Leibniz extensions, so the Casimirs add.
\caseref{n4-10}c is the semidirect sum of the $n=2$ Leibniz extension with an
Abelian algebra defined by $\lpb(\alpha_3,\alpha_4)\com(\beta_3,\beta_4)\rpb =
(0,0)$, with action given by
\[
\rho_{(\alpha_1,\alpha_2)}(\beta_3,\beta_4)
= (0,\lpb\alpha_1\com\beta_3\rpb).
\]
The Casimir invariants for this extension were derived in \secref{singWnex}.
\caseref{n4-10}d has a nonsingular~$\Wn$, so the techniques of
\secref{nsingWn} can be applied directly.
Finally, \caseref{n4-11}b is the~$n=4$ Leibniz extension, the Casimir
invariants of which were derived in \secref{Casleib}. The invariants are all
summarized in \tabref{Casimirn4}.
\begin{table}
\caption{Casimir invariants for solvable extensions of order~$n=4$.}
\tablabel{Casimirn4}
\vskip 1em
\begin{center}
\begin{tabular}{ll} \hline
Case & Invariant \\[6pt] \hline
1a & $\Casi(\fv^1,\fv^2,\fv^3,\fv^4)$
\\[9pt]
1b & $\fv^1\afi(\fv^4) + \fv^2\afii(\fv^4) + \fv^3\afiii(\fv^4)
+ \afiv(\fv^4)$
\\[9pt]
2 & $\fv^1\afi(\fv^3) + \fv^2\afii(\fv^3) + \afiii(\fv^3,\fv^4)$
\\[9pt]
3a & $\fv^1\afi(\fv^2) + \afii(\fv^2,\fv^3,\fv^4)$
\\[9pt]
3b & $\fv^1\afi(\fv^2) + \fv^3\afii(\fv^4) + \afiii(\fv^2,\fv^4)$
\\[9pt]
3c & $\fv^1\afi(\fv^4) + \fv^2\fv^3\afi'(\fv^4) + \fv^3\afii(\fv^4)
+ \afiii(\fv^2,\fv^4)$
\\[9pt]
3d & $\fv^1\afi(\fv^4) + \frac{1}{2}(\fv^2)^2\afi'(\fv^4) +
\fv^3\afii(\fv^4) + \fv^2\afiii(\fv^4) + \afiv(\fv^4)$
\\[9pt]
4a & $\fv^1\afi(\fv^3) + \frac{1}{2}(\fv^2)^2\afi'(\fv^3) +
\fv^2\afii(\fv^3) + \afiii(\fv^3,\fv^4)$
\\[9pt]
4b & $\fv^1\afi(\fv^4) + \fv^2\fv^3\afi'(\fv^4)
+ \frac{1}{3!}(\fv^3)^3\afi''(\fv^4)$\\ &
$\mbox{} + \fv^2\afii(\fv^4) + \frac{1}{2}(\fv^3)^2\afii'(\fv^4)
+ \fv^3\afiii(\fv^4) + \afiv(\fv^4)$
\\[9pt] \hline
\end{tabular}
\end{center}
\end{table}
\subsection{Semidirect Extensions}
\seclabel{lowsemicasi}
Now that we have derived the Casimir invariants for solvable extensions, we
look at extensions involving the semidirect sum of an algebra with these
solvable extensions. We label the new variable (the one which acts on the
solvable part) by~$\fv^0$. In \secref{nsingWn} we showed that the Casimirs of
the solvable part were also Casimirs of the full extension. We also concluded
that a necessary condition for obtaining a new Casimir (other than the linear
case~$\Casi(\fv^0) = \fv^0$) from the semidirect sum was that~$\det \W_{(n)}
\ne 0$. We go through the solvable cases and determine the Casimirs associated
with the semidirect extension, if any exist.
\subsubsection{n=1}
There is only one solvable extension, so upon appending a semidirect part we
have
\[
\W_{(0)} = \l(\begin{array}{cc}
\makebox[1.3em]{1} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r),\ \ \ \
\W_{(1)} = \l(\begin{array}{cc}
\makebox[1.3em]{0} & \makebox[1.3em]{1} \\
\makebox[1.3em]{1} & \makebox[1.3em]{0}
\end{array}\r).
\]
Since~$\det \W_{(1)}\ne 0$, we expect another Casimir. In fact this extension
is of the semidirect Leibniz type and has the same Casimir form as the~$n=2$
solvable Leibniz (\secref{Casleib}) extension. Thus, the new Casimir is
just~$\fv^0\afsd(\fv^1)$.
\subsubsection{n=2}
Of the two possible extensions only the Leibniz one
satisfies~$\det \W_{(2)}\ne 0$. The Casimir is thus
\[
\mathcal{C}_{\rm sd} = \fv^0\afsd(\fv^2)+\frac{1}{2}(\fv^1)^2\afsd'(\fv^2).
\]
\subsubsection{n=3}
Cases~\ref{case:n4-01} and~\ref{case:n4-11} have a nonsingular~$\W_{(3)}$.
The Casimir for \caseref{n4-01} is
\[
\mathcal{C}_{\rm sd} = \fv^0\afsd(\fv^3) + \fv^1\fv^2\afsd'(\fv^3),
\]
and for \caseref{n4-11} it is of the Leibniz form
\[
\mathcal{C}_{\rm sd} = \fv^0\afsd(\fv^3) + \fv^1\fv^2\afsd'(\fv^3)
+\frac{1}{3!}(\fv^2)^3\afsd''(\fv^3).
\]
\subsubsection{n=4}
Cases \ref{case:n4-00}b, \ref{case:n4-10}d, and \ref{case:n4-11}b have a
nonsingular~$\W_{(4)}$. The Casimirs are shown in \tabref{Casimirsemn5}.
\begin{table}
\caption{Casimir invariants for semidirect extensions of order~$n=5$. These
extensions also possess the corresponding Casimir invariants in
\tabref{Casimirn4}.}
\tablabel{Casimirsemn5}
\vskip 1em
\begin{center}
\begin{tabular}{ll} \hline
Case & Invariant \\[6pt] \hline
1b & $\fv^0\afsd(\fv^4) + \l(\fv^1\fv^3
+ \frac{1}{2}(\fv^2)^2\r)\afsd'(\fv^4)$
\\[9pt]
3d & $\fv^0\afsd(\fv^4) + \l(\fv^1\fv^2
+ \frac{1}{2}(\fv^3)^2\r)\afsd'(\fv^4)
+ \frac{1}{3!}(\fv^2)^3\afsd''(\fv^4)$
\\[9pt]
4b & $\fv^0\afsd(\fv^4) + \l(\fv^1\fv^3
+ \frac{1}{2}(\fv^2)^2\r)\afsd'(\fv^4)
+ \frac{1}{2}\fv^2(\fv^3)^2\afsd''(\fv^4)
+ \frac{1}{4!}(\fv^3)^4\afsd'''(\fv^4)$
\\[9pt] \hline
\end{tabular}
\end{center}
\end{table}
\section{Casimir Invariants for Extensions}
\seclabel{casinv}
\ifmypprint
{\small
\begin{verbatim}
$Id: sec-casimirext.tex,v 2.1 1998/11/13 08:37:28 jeanluc Exp $
\end{verbatim}
}
\fi
In this section we will use the bracket extensions of
\secref{classext} to make Lie--Poisson brackets, following the
prescription of \secref{LiePoisson}. In \secref{cascond} we write down the
general form of the Casimir condition (the condition under which a functional
is a Casimir invariant) for a general class of inner brackets. Then in
\secref{Casdirprod} we see how the Casimirs separate for a direct sum
of algebras, the case discussed in \secref{directprod}. \secref{localcas}
discusses the particular properties of Casimirs of solvable extensions. In
\secref{cassoln} we give a general solution to the Casimir problem and
introduce the concept of \emph{coextension}. Finally, in \secref{Casex} we
work out the Casimir invariants for some specific examples, including CRMHD
and the Leibniz extension.
\subsection{Casimir Condition}
\seclabel{cascond}
A generalized Casimir invariant (or Casimir for short) is a
function~$\Cas:\LieA^* \rightarrow \reals$ for which
\[
\lPB F \com\, \Cas \rPB \equiv 0,
\]
for all~$F:\LieA^* \rightarrow \reals$. Using~\eqref{LPB}
and~\eqref{cobracket}, we can write this as
\[
\lang \,\fv \com\,\lpb\frac{\delta F}{\delta \fv} \com
\frac{\delta \Cas}{\delta \fv}\rpb\,\rang
= -\lang \lpb\frac{\delta \Cas}{\delta \fv}\com\,\fv\rpb^\dagger \com\,
\frac{\delta F}{\delta \fv}\,\rang.
\]
Since this vanishes for all~$F$ we conclude
\begin{equation}
\lpb\frac{\delta \Cas}{\delta \fv}\com\,\fv\rpb^\dagger = 0.
\eqlabel{cascond}
\end{equation}
To figure out the coadjoint bracket corresponding to~\eqref{extbrack}, we
write
\[
\lang \,\fv \com\,\lpb\alpha \com \beta\rpb\,\rang =
\lang \,\fv^\lambda \com\,{\W_\lambda}^{\mu\nu}
{\lpb\alpha_\mu \com \beta_\nu\rpb}\,\rang,
\]
which after using the coadjoint bracket of~$\LieA$ becomes
\[
\lang \lpb\beta\com\fv\rpb^\dagger\com \alpha\,\rang =
\lang {\W_\lambda}^{\mu\nu}\lpb\beta_\nu\com\fv^\lambda\rpb^\dagger
\com \alpha_\mu\,\rang
\]
so that
\[
\lpb\beta\com\fv\rpb^{\dagger\,\nu} = {\W_\lambda}^{\mu\nu}
\lpb\beta_\mu\com\fv^\lambda\rpb^\dagger.
\]
We can now write the Casimir condition~\eqref{cascond} for the bracket by
extension as
\begin{equation}
{\W_\lambda}^{\mu\nu} \lpb\frac{\delta \Cas}{\delta\fv^\mu}
\com\fv^\lambda\rpb^\dagger = 0,\ \ \ \ \nu=0,\dots,n.
\eqlabel{cascond2}
\end{equation}
We now specialize the bracket to the case of most interested to us, where the
inner bracket is of canonical form~\eqref{canibrak}. As we saw in
\secref{LiePoisson}, this is the bracket for 2-D fluid flows. The
construction we give here has a finite-dimensional analogue, where one uses
the Cartan--Killing form to map vectors to covectors, but we will not pursue
this here (see Thiffeault~\cite{Thiffeault1998diss}). Further, we assume that
the form of the Casimir invariants is
\begin{equation}
\Cas[\fv] = \int_\fdomain \Casi(\fv(\xv))\d^2x,
\eqlabel{formcas}
\end{equation}
and thus, since~$\Casi$ does not contain derivatives of~$\fv$, functional
derivatives of~$\Cas$ can be written as ordinary partial derivatives
of~$\Casi$. \inthesis{I took a stab at this in RNB III p. 136 but found no
cases that weren't exact derivatives. Also on p. 138 I show that at least for
2--D Euler we cannot have a dependence on the spatial coordinates.} We can
then rewrite~\eqref{cascond2} as
\begin{equation}
{\W_\lambda}^{\mu\nu}
\frac{\partial^2\Casi}{\partial\fv^\mu\partial\fv^\sigma}
\lpb\fv^\sigma\com\fv^\lambda\rpb
= 0,\ \ \ \ \nu=0,\dots,n.
\eqlabel{cascond5}
\end{equation}
In the canonical case where the inner bracket is like~\eqref{canibrak}
the~$\lpb\fv^\sigma\com\fv^\lambda\rpb$ are independent and antisymmetric
in~$\lambda$ and~$\sigma$.\inthesis{Does this hold for other brackets, for
instance finite-dim semisimple?} Thus a necessary and sufficient condition for
the Casimir condition to be satisfied is
\begin{equation}
{\W_\lambda}^{\mu\nu}
\frac{\partial^2\Casi}{\partial\fv^\mu\partial\fv^\sigma}
= {\W_\sigma}^{\mu\nu}
\frac{\partial^2\Casi}{\partial\fv^\mu\partial\fv^\lambda}\ ,
\eqlabel{cascond3}
\end{equation}
for~$\lambda,\sigma,\nu=0,\dots,n$. Sometimes we shall abbreviate this as
\begin{equation}
{\W_\lambda}^{\mu\nu} \Casi_{\dcom\mu\sigma}
= {\W_\sigma}^{\mu\nu} \Casi_{\dcom\mu\lambda}\ ,
\eqlabel{cascond4}
\end{equation}
that is, any subscript~$\mu$ on~$\Casi$ following a comma indicates
differentiation with respect to~$\fv^\mu$. Equation \eqref{cascond4} is
trivially satisfied when~$\Casi$ is a linear function of the~$\fv$'s. That
solution usually follows from special cases of more general solutions, and we
shall only mention it in \secref{singWn} where it is the only solution.
An important result is immediate from \eqref{cascond4} for a semidirect
extension. Whenever the extension is semidirect we shall label the
variables~$\fv^0,\fv^1,\dots,\fv^n$, because the subset~$\fv^1,\dots,\fv^n$
then forms a solvable subalgebra (see \secref{semisimple} for terminology).
For a semidirect extension,~$\W^{(0)}$ is the identity matrix, and
thus~\eqref{cascond4} gives
\begin{eqnarray*}
{\delta_\lambda}^{\mu} \Casi_{\dcom\mu\sigma}
&=& {\delta_\sigma}^{\mu} \Casi_{\dcom\mu\lambda}\ ,\nonumber\\
\Casi_{\dcom\lambda\sigma} &=& \Casi_{\dcom\sigma\lambda}\ ,
\end{eqnarray*}
which is satisfied because we can interchange the order of differentiation.
Hence,~$\nu=0$ does not lead to any conditions on the Casimir. However, the
variables~$\mu,\lambda,\sigma$ still take values from~$0$ to~$n$ in
\eqref{cascond4}.
\subsection{Direct Sum}
\seclabel{Casdirprod}
For the direct sum we found in \secref{directprod} that if we look at the
3-tensor~$\W$ as a cube, then it ``blocks out'' into smaller cubes, or
subblocks, along its main diagonal, each subblock representing a subalgebra.
We denote each subblock of~${\W_\lambda}^{\mu\nu}$
by~${{\W_i}_{\lambda}}^{\mu\nu}$,
\hbox{$i=1,\dots,r$}, where~$r$ is the number of subblocks. We can
rewrite~\eqref{LPB} as
\begin{eqnarray*}
\lPB A\com B\rPB &=& \sum_{i=1}^r \lang\fv_i^{\lambda}\com
{{\W_i}_\lambda}^{\mu\nu}\,
\lpb {\frac{\fd A}{\fd\fv_i^{\mu}}}\com
{\frac{\fd B}{\fd\fv_i^{\nu}}}\rpb\rang \nonumber \\
&\mathrel{=\!\!\raisebox{.069ex}{:}}& \sum_{i=1}^r {\lPB A\com B\rPB}_i\, ,
\end{eqnarray*}
where~$i$ labels the different subblocks and the greek indices run over the
size of the~$i$th subblock. Each of the subbrackets~\hbox{${\lPB \com
\rPB}_i$} depends on different fields. In particular, if the
functional~$\Cas$ is a Casimir, then, for any functional~$F$
\[
\lPB F\com \Cas\rPB = \sum_{i=1}^r {\lPB F\com \Cas\rPB}_i = 0
\ \ \ \Longrightarrow
\ \ \ {\lPB F\com \Cas\rPB}_i = 0,\ \ i=1,\dots,r\, .
\]
The solution for this is
\[
\Cas[\fv] = \Cas_1[\fv_1] + \cdots
+ \Cas_r[\fv_r]\, ,\ \ \
{\rm where}\ {\lPB F\com \Cas_i\rPB}_i =0,\ i=1,\dots,r\, ,
\]
that is, the Casimir is just the sum of the Casimir for each subbracket.
Hence, the question of finding the Casimirs can be treated separately for each
component of the direct sum. We thus assume we are working on a single
degenerate subblock, as we did for the classification in
\secref{classext}, and henceforth we drop the subscript~$i$.
There is a complication when a single (degenerate) subblock has more that one
simultaneous eigenvector. By this we mean~$k$ vectors~$u^{(a)}$,
$a=1,\dots,k$, such that
\[
{\W_\lambda}^{\mu(\nu)}\,u^{(a)}_\mu = \ev^{(\nu)}\,\,u^{(a)}_\lambda.
\]
Note that lower-triangular matrices always have at least the simultaneous
eigenvector \hbox{$u_\mu={\delta_\mu}^n$}. Let~$\eta^{(a)} \mathrel{\raisebox{.069ex}{:}\!\!=}
u^{(a)}_\rho\xi^\rho$, and consider a
form~$\Casi(\eta^{(1)},\dots,\eta^{(k)})$ for the Casimir. Then
\begin{eqnarray*}
{\W_\lambda}^{\mu(\nu)}
\frac{\partial^2\Casi}{\partial\fv^\mu\partial\fv^\sigma}
&=& {\W_\lambda}^{\mu(\nu)} \sum_{a,b=1}^k
u^{(a)}_\mu u^{(b)}_\sigma
\frac{\partial^2\Casi}{\partial\eta^{(a)}\partial\eta^{(b)}}\,,
\nonumber\\
&=& \ev^{(\nu)} \sum_{a,b=1}^k
u^{(a)}_\lambda u^{(b)}_\sigma
\frac{\partial^2\Casi}{\partial\eta^{(a)}\partial\eta^{(b)}}.
\end{eqnarray*}
Because the eigenvalue~$\ev^{(\nu)}$ does not depend on~$a$ (the block was
assumed to have degenerate eigenvalues), the above expression is symmetric
in~$\lambda$ and~$\sigma$. Hence, the Casimir condition~\eqref{cascond3}
is satisfied.
The reason this is introduced here is that if a degenerate block splits into a
direct sum, then it will have several simultaneous eigenvectors. The Casimir
invariants~$\Casi^{(a)}(\eta^{(a)})$ and~$\Casi^{(b)}(\eta^{(b)})$
corresponding to each eigenvector, instead of adding
as~$\Casi^{(a)}(\eta^{(a)}) + \Casi^{(b)}(\eta^{(b)})$, will combine into one
function,~$\Casi{(\eta^{(a)},\eta^{(b)})}$, a more general functional
dependence. However, these situations with more than one eigenvector are not
limited to direct sums. For instance, they occur in semidirect sums. In
\secref{caslowdim} we will see examples of both cases.
\subsection{Local Casimirs for Solvable Extensions}
\seclabel{localcas}
In the solvable case, when all the~$\W^{(\mu)}$'s are lower-triangular with
vanishing eigenvalues, a special situation occurs. If we consider the Casimir
condition \eqref{cascond5}, we notice that derivatives with respect to~$\fv^n$
do not occur at all, since~$\W^{(n)}=0$. Hence the functional
\[
\Cas[\fv] = \int_\fdomain\,\fv^n(\xv')\,\delta(\xv-\xv')\d^2x'
= \fv^n(\xv)
\]
is conserved. The variable~$\fv^n(\xv)$ is \emph{locally} conserved. It
cannot have any dynamics associated with it. This holds true for any other
simultaneous null eigenvectors the extension happens to have, but for the
solvable case~$\fv^n$ is always such a vector (provided the matrices have been
put in lower-triangular form, of course).
Hence there are at most~$n-1$ dynamical variables in an order~$n$ solvable
extension. An interesting special case occurs when the only
nonvanishing~$\W_{(\mu)}$ is for~$\mu=n$. Then the Lie--Poisson bracket is
\[
\lPB F\com G\rPB = \sum_{\mu,\nu=1}^{n-1}{\W_n}^{\mu\nu}
\int_\fdomain\,\fv^n(\xv)
\,\lpb \frac{\fd F}{\fd \fv^\mu(\xv)}\com
\frac{\fd G}{\fd \fv^\nu(\xv)}\rpb\d^2x,
\]
where~$\fv^n(\xv)$ is some function of our choosing. This bracket is not what
we would normally call Lie--Poisson because~$\fv^n(\xv)$ is not dynamical.
It gives equations of motion of the form
\[
\dot\fv^\nu = {\W_n}^{\nu\mu}\,\lpb \frac{\fd H}{\fd \fv^\mu}
\com \fv^n \rpb,
\]
which can be used to model, for example, advection of scalars in a specified
flow given by~$\fv^n(\xv)$. This bracket occurs naturally when a Lie--Poisson
bracket is linearized~\cite{Morrison1998,MarsdenRatiu}.
\subsection{Solution of the Casimir Problem}
\seclabel{cassoln}
We now proceed to find the solution to~\eqref{cascond5}. We assume that all
the~$\W^{(\mu)}$, $\mu=0,\dots,n$, are in lower-triangular form, and that the
matrix~$\W^{(0)}$ is the identity matrix. Although this is the semidirect
form of the extension, we will see that we can also recover the Casimir
invariants of the solvable part. We assume~$\nu>0$ in~\eqref{cascond5},
since~$\nu=0$ does not lead to a condition on the Casimir (\secref{cascond}).
Therefore~${\W_{\lambda}}^{n\nu}=0$. Thus, we separate the Casimir condition
into a part involving indices ranging from~$0,\dots,n-1$ and a part that
involves only~$n$. The condition
\[
\sum_{\mu,\sigma,\lambda=0}^{n} {\W_\lambda}^{\mu\nu} \Casi_{\dcom\mu\sigma}
\lpb\fv^\lambda\com\fv^\sigma\rpb = 0, \ \ \ \nu > 0,
\]
becomes
\[
\sum_{\lambda=0}^{n} \l\lgroup
\sum_{\mu,\sigma=0}^{n-1}{\W_\lambda}^{\mu\nu} \Casi_{\dcom\mu\sigma}
\lpb\fv^\lambda\com\fv^\sigma\rpb
+ \sum_{\mu=0}^{n-1}{\W_\lambda}^{\mu\nu} \Casi_{\dcom\mu n}
\lpb\fv^\lambda\com\fv^n\rpb
\r\rgroup = 0,
\]
where we have used~${\W_{\lambda}}^{n\nu}=0$ to limit the sum on~$\mu$.
Separating the sum on~$\lambda$ gives
\begin{eqnarray*}
\sum_{\lambda=0}^{n-1} \l\lgroup
\sum_{\mu,\sigma=0}^{n-1}{\W_\lambda}^{\mu\nu} \Casi_{\dcom\mu\sigma}
\lpb\fv^\lambda\com\fv^\sigma\rpb
+ \sum_{\mu=0}^{n-1}{\W_\lambda}^{\mu\nu} \Casi_{\dcom\mu n}
\lpb\fv^\lambda\com\fv^n\rpb
\r\rgroup\nonumber\\
+ \sum_{\mu,\sigma=0}^{n-1}{\W_n}^{\mu\nu} \Casi_{\dcom\mu\sigma}
\lpb\fv^n\com\fv^\sigma\rpb
+ \sum_{\mu=0}^{n-1}{\W_n}^{\mu\nu} \Casi_{\dcom\mu n}
\lpb\fv^n\com\fv^n\rpb
= 0.
\end{eqnarray*}
The last sum vanishes because~$\lpb\fv^n\com\fv^n\rpb=0$. Now we separate the
condition into semisimple and solvable parts,
\begin{eqnarray*}
\sum_{\mu=1}^{n-1} \Biggl\lgroup
\sum_{\lambda,\sigma=0}^{n-1}{\W_\lambda}^{\mu\nu}
\Casi_{\dcom\mu\sigma}
\lpb\fv^\lambda\com\fv^\sigma\rpb
- \sum_{\sigma=0}^{n-1}{\W_\sigma}^{\mu\nu} \Casi_{\dcom\mu n}
\lpb\fv^n\com\fv^\sigma\rpb
\nonumber\\
\mbox{} +
\sum_{\sigma=0}^{n-1}{\W_n}^{\mu\nu} \Casi_{\dcom\mu\sigma}
\lpb\fv^n\com\fv^\sigma\rpb
\Biggr\rgroup
+ \sum_{\lambda,\sigma=0}^{n-1}{\W_\lambda}^{0\nu}
\Casi_{\dcom 0\sigma}
\lpb\fv^\lambda\com\fv^\sigma\rpb\nonumber\\
- \sum_{\sigma=0}^{n-1}{\W_\sigma}^{0\nu} \Casi_{\dcom 0 n}
\lpb\fv^n\com\fv^\sigma\rpb
+ \sum_{\sigma=0}^{n-1}{\W_n}^{0\nu} \Casi_{\dcom 0\sigma}
\lpb\fv^n\com\fv^\sigma\rpb
= 0.
\end{eqnarray*}
Using~${\W_\sigma}^{0\nu} = {\delta_\sigma}^\nu$, we can separate the
conditions into a part for~$\nu=n$ and one for~\hbox{$0<\nu<n$}. For~$\nu=n$,
the only term that survives is the last sum
\[
\sum_{\sigma=0}^{n-1} \Casi_{\dcom 0\sigma}
\lpb\fv^n\com\fv^\sigma\rpb = 0.
\]
Since the commutators are independent, we have the conditions,
\begin{equation}
\Casi_{\dcom 0\sigma} = 0, \ \ \ \sigma=0,\dots,n-1.
\eqlabel{C0seq0}
\end{equation}
and for~\hbox{$0<\nu<n$},
\begin{eqnarray*}
\sum_{\mu=1}^{n-1} \Biggl\lgroup
\sum_{\lambda,\sigma=1}^{n-1}{\W_\lambda}^{\mu\nu}
\Casi_{\dcom\mu\sigma}
\lpb\fv^\lambda\com\fv^\sigma\rpb
- \sum_{\sigma=1}^{n-1}{\W_\sigma}^{\mu\nu} \Casi_{\dcom\mu n}
\lpb\fv^n\com\fv^\sigma\rpb
\nonumber\\
\mbox{} +
\sum_{\sigma=1}^{n-1}{\W_n}^{\mu\nu} \Casi_{\dcom\mu\sigma}
\lpb\fv^n\com\fv^\sigma\rpb
\Biggr\rgroup
- \Casi_{\dcom 0 n} \lpb\fv^n\com\fv^\nu\rpb
= 0,
\end{eqnarray*}
where we have used~\eqref{C0seq0}. Using independence of the inner
brackets gives
\begin{eqnarray}
{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\lambda}^{\mu\nu}
\Casi_{\dcom\mu\sigma} &=&
{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\mu\nu}
\Casi_{\dcom\mu\lambda},
\eqlabel{cascondsubext} \\
{\Wn}^{\nu\mu} \Casi_{\dcom\mu\sigma} &=&
{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\mu} \Casi_{\dcom\mu n}
+ {\delta^\nu}_\sigma\, \Casi_{\dcom 0 n},
\eqlabel{Axeqb}
\end{eqnarray}
for~$0< \sigma,\lambda,\nu,\mu < n$. From now on in this section repeated
indices are summed, and all greek indices run from~$1$ to~$n-1$ unless
otherwise noted. We have written a tilde over the~$\W$'s to stress the fact
that the indices run from~$1$ to~$n-1$, so that the~${\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}$ represent a solvable
order~$(n-1)$ subextension of~$\W$. This subextension does not
include~$\W_{(n)}$. We have also made the definition
\begin{equation}
\Wn^{\mu\nu} \mathrel{\raisebox{.069ex}{:}\!\!=} {\W_n}^{\mu\nu}.
\eqlabel{Wndef}
\end{equation}
Equation~\eqref{cascondsubext} is a Casimir condition: it says that~$\Casi$ is
also a Casimir of~${\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}$. We now proceed to solve~\eqref{Axeqb} for the case
where~$\Wn$ is nonsingular. In \secref{singWn} we will solve the
singular~$\Wn$ case. We will see that in both cases \eqref{cascondsubext}
follows from \eqref{Axeqb}.
\subsubsection{Nonsingular $\Wn$}
\seclabel{nsingWn}
The simplest case occurs when~$\Wn$ has an inverse, which we will
call~$\Wni_{\mu\nu}$. Then~\eqref{Axeqb} has the solution
\begin{equation}
\Casi_{\dcom\tau\sigma} =
\coW^\mu_{\tau\sigma}\, \Casi_{\dcom\mu n}
+ \Wni_{\tau\sigma}\, \Casi_{\dcom 0 n}\, ,
\eqlabel{nonsingsol}
\end{equation}
where
\begin{equation}
\coW^\mu_{\tau\sigma} \mathrel{\raisebox{.069ex}{:}\!\!=} \Wni_{\tau\nu}\,{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\mu}.
\eqlabel{coextdef}
\end{equation}
It is easily verified that~$\coW^\mu_{\tau\sigma} = \coW^\mu_{\sigma\tau}$,
as required by the symmetry of the left-hand side of~\eqref{nonsingsol}.
In~\eqref{nonsingsol}, it is clear that the~$n$th variable is ``special'';
this suggests that we try the following form for the Casimir:
\begin{equation}
\Casi(\fv^0,\fv^1,\dots,\fv^n) =
\sum_{i \ge 0}
\ag^{(i)}(\fv^0,\fv^1,\dots,\fv^{n-1})\,\af_{i}(\fv^n),
\eqlabel{Casform}
\end{equation}
where~$\af$ is arbitrary and~$\af_i$ is the $i$th derivative of~$\af$ with
respect to its argument. One immediate advantage of this form is that
\eqref{cascondsubext} follows from \eqref{Axeqb}. Indeed, taking a derivative
of \eqref{Axeqb} with respect to~$\fv^\lambda$, inserting~\eqref{Casform}, and
equating derivatives of~$\af$ leads to
\[
{\Wn}^{\nu\mu}\, \ag^{(i)}_{\dcom\mu\sigma\lambda} =
{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\mu}\, \ag^{(i+1)}_{\dcom\mu\lambda},
\]
where we have used \eqref{C0seq0}. Since the left-hand side is symmetric
in~$\lambda$ and~$\sigma$ then so is the right-hand side, and
\eqref{cascondsubext} is satisfied.
Now, inserting the form of the Casimir~\eqref{Casform} into the
solution~\eqref{nonsingsol}, we can equate derivatives of~$\af$ to obtain,
for~\hbox{$\tau,\sigma=1,\dots,n-1$},
\begin{eqnarray}
\ag^{(0)}_{\dcom\tau\sigma} &=& 0, \ \ \ \ \ \ \ \
\tau,\sigma=1,\dots,n-1;\\ \ag^{(i)}_{\dcom\tau\sigma} &=&
\coW^\mu_{\tau\sigma}\, \ag^{(i-1)}_{\dcom\mu} +
\Wni_{\tau\sigma}\,\ag^{(i-1)}_{\dcom 0}, \ \ \ i \ge 1.
\eqlabel{gcondi}
\end{eqnarray}
The first condition, together with~\eqref{C0seq0}, says that~$\ag^{(0)}$
is linear in~$\fv^0,\dots\fv^{n-1}$. There are no other conditions
on~$\ag^{(0)}$, so we can obtain~$n$ independent solutions by choosing
\begin{equation}
\ag^{(0)\nu} = \fv^\nu, \ \ \ \nu=0,\dots,n-1.
\eqlabel{nsingsolzero}
\end{equation}
The equation for~$\ag^{(1)\nu}$ is
\begin{equation}
\ag^{(1)\nu}_{\dcom\tau\sigma} = \l\{\begin{array}{l}
\Wni_{\tau\sigma},\ \ \ \nu = 0; \\
\coW^\nu_{\tau\sigma},\ \ \ \nu = 1,\dots,n-1.
\eqlabel{nsingsolone}
\end{array}\r.
\end{equation}
Thus~$\ag^{(1)\nu}$ is a quadratic polynomial (the arbitrary linear part does
not yield an independent Casimir, so we set it to zero). Note
that~$\ag^{(1)\nu}$ does not depend on~$\fv^0$
since~$\tau,\sigma=1,\dots,n-1$. Hence, for~$i>1$ we can drop
the~$\ag^{(i-1)}_{\dcom 0}$ term in~\eqref{gcondi}. Taking derivatives
of~\eqref{gcondi}, we obtain
\begin{equation}
\ag^{(i)\nu}_{\dcom\tau_1\tau_2\dots\tau_{(i+1)}} =
\coW^{\mu_1}_{\tau_1\tau_2}\,
\coW^{\mu_2}_{\mu_1\tau_3}\cdots
\coW^{\mu_{(i-1)}}_{\mu_{(i-2)}\tau_{i}}
\,\ag^{(1)\nu}_{\dcom\mu_{(i-1)}\tau_{(i+1)}}.
\eqlabel{nonsingsolni}
\end{equation}
We know the series will terminate because the~${\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}^{(\mu)}$, and hence
the~$\coW_{(\mu)}$, are nilpotent. The solution to \eqref{nonsingsolni} is
\begin{equation}
\ag^{(i)\nu} = \frac{1}{(i+1)!}\,\,
\agc^{(i)\nu}_{\tau_1\tau_2\dots\tau_{(i+1)}}\,
\fv^{\tau_1}\fv^{\tau_2}\cdots\fv^{\tau_{(i+1)}}\,,
\ \ \ \ i > 1,
\eqlabel{Cascoeff}
\end{equation}
where the constants~$\agc$ are defined by
\begin{equation}
\agc^{(i)\nu}_{\tau_1\tau_2\dots\tau_{(i+1)}} \mathrel{\raisebox{.069ex}{:}\!\!=}
\coW^{\mu_1}_{\tau_1\tau_2}\,
\coW^{\mu_2}_{\mu_1\tau_3}\cdots
\coW^{\mu_{(i-1)}}_{\mu_{(i-2)}\tau_{i}}
\,\ag^{(1)\nu}_{\dcom\mu_{(i-1)}\tau_{(i+1)}}.
\eqlabel{agcdef}
\end{equation}
In summary, the~$\ag^{(i)}$'s of~\eqref{Casform} are
given by~\eqref{nsingsolzero},~\eqref{nsingsolone}, and~\eqref{Cascoeff}.
Because the left-hand side of~\eqref{nonsingsolni} is symmetric in all its
indices, we require
\begin{equation}
\coW^{\mu}_{\tau\sigma}\,\coW^{\nu}_{\mu\lambda} =
\coW^{\mu}_{\tau\lambda}\,\coW^{\nu}_{\mu\sigma}, \qquad i>1.
\eqlabel{coextcond}
\end{equation}
This is automatically satisfied for the nonsingular~$\Wn$
case~\cite{Thiffeault1998diss}. Comparing this to~\eqref{Wjacob}, we see that
the~$\coW$'s satisfy all the properties of an extension, except with the dual
indices. Thus we call the~$\coW$'s the \emph{coextension} of~${\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}$ with
respect to~$\Wn$. Essentially~$\Wn$ serves the role of a metric that allows
us to raise and lower indices.
For a solvable extension we simply restrict~$\nu > 0$ and the above treatment
still holds. We conclude that the Casimirs of the solvable part of a
semidirect extension are Casimirs of the full extension. We have also shown,
for the case of nonsingular~$\Wn$, that the number of independent Casimirs is
equal to the order of the extension.
\subsubsection{Singular $\Wn$}
\seclabel{singWn}
In general,~$\Wn$ is singular and thus has no inverse. However, it always has
a (symmetric and unique) pseudoinverse~$\Wni_{\mu\nu}$ such that
\begin{eqnarray}
\Wni_{\mu\sigma}\,\Wn^{\sigma\tau}\,\Wni_{\tau\nu}
&=& \Wni_{\mu\nu},\eqlabel{pseudoinv1}\\
\Wn^{\mu\sigma}\,\Wni_{\sigma\tau}\,\Wn^{\tau\nu}
&=& \Wn^{\mu\nu}.
\eqlabel{pseudoinv2}
\end{eqnarray}
The pseudoinverse is also known as the strong generalized inverse or the
Moore--Penrose inverse~\cite{Osta}. It follows from~\eqref{pseudoinv1}
and~\eqref{pseudoinv2} that the matrix operator
\[
{\Proj^\nu}_\tau \mathrel{\raisebox{.069ex}{:}\!\!=} \Wn^{\nu\kappa}\,\Wni_{\kappa\tau}
\]
projects onto the range of~$\Wn$. The system~\eqref{Axeqb} only has a
solution if the following solvability condition is satisfied:
\begin{equation}
{\Proj^\nu}_\tau\,
\l({{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\tau\mu} \Casi_{\dcom\mu n}
+ {\delta^\tau}_\sigma\, \Casi_{\dcom 0 n}\r)
= {{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\mu} \Casi_{\dcom\mu n}
+ {\delta^\nu}_\sigma\, \Casi_{\dcom 0 n};
\eqlabel{solvcondz}
\end{equation}
that is, the right-hand side of~\eqref{Axeqb} must live in the range of~$\Wn$.
If~\hbox{$\Casi_{\dcom 0 n}\ne 0$}, the quantity~${{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\mu}\,
\Casi_{\dcom\mu n} + {\delta^\nu}_\sigma\, \Casi_{\dcom 0 n}$ has rank equal
to~$n$, because the quantity~${{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\mu}\,\Casi_{\dcom\mu n}$ is
lower-triangular (it is a linear combination of lower-triangular matrices).
Hence the projection operator must also have rank~$n$. But then this implies
that~$\Wn$ has rank~$n$ and so is nonsingular, which contradicts the
hypothesis of this section. Hence,~\hbox{$\Casi_{\dcom 0 n} = 0$} for the
singular~$\Wn$ case, which together with \eqref{C0seq0} means that a Casimir
that depends on~$\fv^0$ can only be of the form~$\Casi = \af(\fv^0)$.
However, since~$\fv^0$ is not an eigenvector of the~$\W^{(\mu)}$'s, the only
possibility is~$\Casi = \fv^0$, the trivial linear case mentioned in
\secref{cascond}.
The solvability condition \eqref{solvcondz} can thus be rewritten as
\begin{equation}
\l({\Proj^\nu}_\tau\,{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\tau\mu}
- {{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\mu}\r) \Casi_{\dcom\mu n} = 0.
\eqlabel{solvcond}
\end{equation}
An obvious choice would be to
require~\hbox{${\Proj^\nu}_\tau\,{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\tau\mu} =
{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\mu}$}, but this is too strong. We will derive a weaker
requirement shortly.
By an argument similar to that of~\secref{nsingWn}, we now assume~$\Casi$ is
of the form
\begin{equation}
\Casi(\fv^1,\dots,\fv^n) =
\sum_{i \ge 0} \ag^{(i)}(\fv^1,\dots,\fv^{n-1})\,\af_{i}(\fv^n),
\eqlabel{singCas}
\end{equation}
where again~$\af_i$ is the $i$th derivative of~$f$ with respect to its
argument. As in \secref{nsingWn}, we only need to show \eqref{Axeqb}, and
\eqref{cascondsubext} will follow. The number of independent solutions of
\eqref{Axeqb} is equal of the rank of~$\Wn$. The choice
\begin{equation}
\ag^{(0)\nu} = {\Proj^\nu}_{\rho}\,\fv^\rho, \ \ \ \nu=1,\dots,n-1,
\eqlabel{singsolzero}
\end{equation}
provides the right number of solutions because the rank of~$\Proj$ is equal to
the rank of~$\Wn$. It also properly specializes to
\eqref{nsingsolzero} when~$\Wn$ is nonsingular, for then~${\Proj^\nu}_\rho =
{\delta^{\,\nu}}_\rho$.
The solvability condition~\eqref{solvcond} with this form for the Casimir
becomes
\begin{equation}
\l({\Proj^\nu}_\tau\,{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\tau\mu}
- {{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\mu}\r) \ag^{(i)\nu}_{\dcom\mu} = 0,\ \ \
i \ge 0.
\eqlabel{solvcondg}
\end{equation}
For~$i=0$ the condition can be shown to simplify to
\[
{\Proj^\nu}_\tau\,{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\tau\mu}
= {{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\tau}\,{\Proj^\mu}_\tau,
\]
or to the equivalent matrix form
\begin{equation}
\Proj\,{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_{(\sigma)} = {\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_{(\sigma)}\,\Proj,
\eqlabel{solvcond0}
\end{equation}
since~$\Proj$ is symmetric~\cite{Osta}.
Equation \eqref{Axeqb} becomes
\begin{eqnarray*}
{\Wn}^{\kappa\mu} \ag^{(0)\nu}_{\dcom\mu\sigma} &=& 0,\\
{\Wn}^{\kappa\mu} \ag^{(i)\nu}_{\dcom\mu\sigma} &=&
{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\kappa\mu} \ag^{(i-1)\nu}_{\dcom\mu},\ \ \ i > 0.
\end{eqnarray*}
If \eqref{solvcond} is satisfied, we know this has a solution given by
\[
\ag^{(i)\nu}_{\dcom\lambda\sigma} =
\Wni_{\lambda\rho}\,{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\rho\mu}\,
\ag^{(i-1)\nu}_{\dcom\mu} + \l({\delta_\lambda}^\mu
- \Wni_{\lambda\rho}\,\Wn^{\rho\mu}\r)
\ae^{(i-1)\nu}_{\mu\sigma} ,\ \ \ i > 0,
\]
where~$\ae$ is arbitrary, and~\hbox{$({\delta_\lambda}^\mu -
\Wni_{\lambda\rho}\,\Wn^{\rho\mu})$} projects onto the null space of~$\Wn$.
The left-hand side is symmetric in~$\lambda$ and~$\sigma$, but not the
right-hand side. We can symmetrize the right-hand side by an appropriate
choice of the null eigenvector,
\[
\ae^{(i)\nu}_{\lambda\sigma} \mathrel{\raisebox{.069ex}{:}\!\!=}
\Wni_{\sigma\rho}\,{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\lambda}^{\rho\mu}\,
\ag^{(i)\nu}_{\dcom\mu},\ \ \ i \ge 0,
\]
in which case
\[
\ag^{(i)\nu}_{\dcom\lambda\sigma} =
\coW^\mu_{\lambda\sigma}\,
\ag^{(i-1)\nu}_{\dcom\mu},\ \ \ i > 0,
\]
where
\begin{equation}
\coW^\nu_{\lambda\sigma} \mathrel{\raisebox{.069ex}{:}\!\!=}
\Wni_{\sigma\rho}\,{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\lambda}^{\rho\nu}
+ \Wni_{\lambda\rho}\,{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\rho\nu}
- \Wni_{\lambda\rho}\,\Wni_{\sigma\kappa}\,
\Wn^{\rho\mu}\,{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\mu}^{\kappa\nu}\,,
\eqlabel{coextension}
\end{equation}
which is symmetric in~$\lambda$ and~$\sigma$. Equation~\eqref{coextension}
also reduces to \eqref{coextdef} when~$\Wn$ is nonsingular, for then the null
eigenvector vanishes. The full solution is thus given in the same manner as
\eqref{nonsingsolni} by
\begin{equation}
\ag^{(i)\nu} = \frac{1}{(i+1)!}\,\,
\agc^{(i)\nu}_{\tau_1\tau_2\dots\tau_{(i+1)}}\,
\fv^{\tau_1}\fv^{\tau_2}\cdots\fv^{\tau_{(i+1)}}\,,
\ \ \ \ i > 0,
\eqlabel{Cascoeffsing}
\end{equation}
where the constants~$\agc$ are defined by
\begin{equation}
\agc^{(i)\nu}_{\tau_1\tau_2\dots\tau_{(i+1)}} \mathrel{\raisebox{.069ex}{:}\!\!=}
\coW^{\mu_1}_{\tau_1\tau_2}\,
\coW^{\mu_2}_{\mu_1\tau_3}\cdots
\coW^{\mu_{(i-1)}}_{\mu_{(i-2)}\tau_{i}}
\,\coW^{\mu_{i}}_{\mu_{(i-1)}\tau_{(i+1)}}
\,{\Proj^{\,\nu}}_{\mu_{i}}\,,
\eqlabel{agcdefsing}
\end{equation}
and~$\ag^{(0)}$ is given by \eqref{singsolzero}.
The~$\coW$'s must still satisfy the coextension condition \eqref{coextcond}.
Unlike the nonsingular case this condition does not follow directly and is an
extra requirement in addition to the solvability condition \eqref{solvcondg}.
Note that only the~$i=0$ case, Eq. \eqref{solvcond0}, needs to be satisfied,
for then \eqref{solvcondg} follows. Both these conditions are
coordinate-dependent, and this is a drawback. Nevertheless, we have found in
obtaining the Casimir invariants for the low-order brackets that if these
conditions are not satisfied, then the extension is a direct sum and the
Casimirs can be found by the method of \secref{Casdirprod}. However, this has
not been proven rigorously.
\subsection{Examples}
\seclabel{Casex}
We now illustrate the methods developed for finding Casimirs with a few
examples. First we treat our prototypical case of CRMHD, and give a physical
interpretation of invariants. Then, we derive the Casimir invariants for
Leibniz extensions of arbitrary order. Finally, we give an example involving a
singular~$\Wn$.
\subsubsection{Compressible Reduced MHD}
\seclabel{CRMHDCas}
The~$\W$ tensors representing the bracket for CRMHD (see \secref{CRMHD}) were
given in \secref{matCRMHD}. We have~$n=3$, so from \eqref{Wndef} we get
\[
\Wn = \l(\begin{array}{cc} 0 & -{\beta_\mathrm{i}} \\ -{\beta_\mathrm{i}} & 0
\end{array}\r),\ \ \
\Wn^{-1} = \l(\begin{array}{cc} 0 & -{\beta_\mathrm{i}}^{-1} \\
-{\beta_\mathrm{i}}^{-1} & 0 \end{array}\r).
\]
In this case, the coextension is trivial: all three matrices~$\coW^{(\nu)}$
defined by \eqref{coextdef} vanish. Using \eqref{Casform} and
\eqref{nsingsolzero}, with~$\nu=1$ and~$2$, the Casimirs for the solvable part are
\[
\Casi^1 = \fv^1\,\afii(\fv^3) = \pvel\,\afii(\magf),\ \ \ \Casi^2 =
\fv^2\,\afiii(\fv^3) = \pres\,\afiii(\magf),
\]
and the Casimir associated with the eigenvector~$\fv^3$ is
\[
\Casi^3 = \afiv(\fv^3) = \afiv(\magf).
\]
Since~$\Wn$ is nonsingular we also get another Casimir from the semidirect
sum part,
\[
\Casi^0 = \fv^0\,\afi(\fv^3)
- \frac{1}{{\beta_\mathrm{i}}}\,\fv^1\,\fv^2\,\afi'(\fv^3)
= \vort\,\afi(\magf)
- \frac{1}{{\beta_\mathrm{i}}}\,\pres\,\pvel\,\afi'(\magf).
\]
\inthesis{Have larger section on this.}
The physical interpretation of the invariant~$\Casi^3$ is given in
Morrison~\cite{Morrison1987} and Thiffeault and
Morrison~\cite{Thiffeault1998}. This invariant implies the preservation of
contours of~$\magf$, so that the value~$\magf_0$ on a contour labels that
contour for all times. This is a consequence of the lack of dissipation and
the divergence-free nature of the velocity. Substituting~\hbox{$\Casi^3(\magf)
= \magf^k$} we also see that all the moments of the magnetic flux are
conserved. By choosing~\hbox{$\Casi^3(\magf) = \heavyside(\magf(\xv) -
\omega_0)$}, a heavyside function, and inserting into \eqref{formcas}, it
follows that the area inside of any $\magf$-contour is conserved.
To understand the Casimirs~$\Casi^1$ and~$\Casi^2$, we also let
$\afii(\magf)=\heavyside(\magf-\magf_0)$ in~$\Casi^1$. In this case we have
\[
\Cas^1[\pvel\,;\magf] = \int_{\fdomain}\pvel\,\afii(\magf)\d^2x
= \int_{\magfcont_0}\,\pvel(\xv)\d^2x,
\]
where~$\magfcont_0$ represents the (not necessarily connected) region of
$\fdomain$ enclosed by the contour $\magf=\magf_0$ and~$\pd\magfcont_0$ is
its boundary. By the interpretation we gave of~$\Casi^3$, the contour
$\pd\magfcont_0$ moves with the fluid. So the total value of~$\pvel$ inside of
a~$\magf$-contour is conserved by the flow. The same is true of the
pressure~$\pres$. (See Thiffeault and Morrison~\cite{Thiffeault1998} for an
interpretation of these invariants in terms of relabeling symmetries, and a
comparison with the rigid body.)
The total pressure and parallel velocity inside of any $\magf$-contour are
preserved. To understand $\Casi^4$, we use the fact
that~$\vort=\lapl\elecp$ and integrate by parts to obtain
\[
\Cas^4[\vort,\pvel,\pres,\magf] = -\int_\fdomain
\l(\grad\elecp\cdot\grad\magf
+ \frac{\pvel\,p}{{\beta_\mathrm{i}}}\r)\afi'(\magf)\d^2x.
\]
The quantity in parentheses is thus invariant inside of any $\magf$-contour.
It can be shown that this is a remnant of the conservation by the full MHD
model of the cross helicity,
\[
V = \int_\fdomain {\mathbf{v}}\cdot{\mathbf{B}}\d^2x\,,
\]
at second order in the inverse aspect ratio, while the conservation
of~$\Cas^1[\pvel\,;\magf]$ is a consequence of preservation of this quantity
at first order. Here ${\mathbf{B}}$ is the magnetic field. The
quantities~$\Cas^3[\magf]$ and~$\Cas^2[\pres\,;\magf]$ they are, respectively,
the first and second order remnants of the preservation of helicity,
\[
W = \int_\fdomain {\mathbf{A}}\cdot{\mathbf{B}}\d^2x,
\]
where ${\mathbf{A}}$ is the magnetic vector potential.
\subsubsection{Leibniz Extension}
\seclabel{Casleib}
We first treat the nilpotent case. The Leibniz extension of \secref{Leibniz}
can be characterized by
\begin{equation}
{\W_\lambda}^{\mu\nu} = {\delta_\lambda}^{\mu+\nu}\,,
\ \ \ \mu,\nu, \lambda = 1,\dots,n,
\tag{\ref{eq:sLeib}}
\end{equation}
where the tensor~$\delta$ is the ordinary Kronecker delta. Upon restricting
the indices to run from~$1$ to~$n-1$ (the tilde notation of \secref{cassoln}),
we have
\[
\Wn^{\mu\nu} = {{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_n}^{\mu\nu} = {\delta_n}^{\mu+\nu}\,,
\ \ \ \mu,\nu = 1,\dots,n-1.
\]
The matrix~$\Wn$ is nonsingular with inverse equal to
itself:~\hbox{$\Wni_{\mu\nu} =
\delta_{\mu+\nu}^{\,\,n}$}. The coextension of~${\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}$ is thus
\[
\coW^\mu_{\tau\sigma} = \sum_{\nu=1}^{n-1}\Wni_{\tau\nu}\,
{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_\sigma}^{\nu\mu}
= \sum_{\nu=1}^{n-1}\delta^n_{\tau+\nu}\,
{\delta_\sigma}^{\nu+\mu}
= \delta^{\mu+n}_{\tau+\sigma}\,.
\]
Equation \eqref{agcdef} becomes
\begin{eqnarray*}
\agc^{(i)\nu}_{\tau_1\tau_2\dots\tau_{(i+1)}} &=&
\coW^{\mu_1}_{\tau_1\tau_2}\,
\coW^{\mu_2}_{\mu_1\tau_3}\cdots
\coW^{\mu_{(i-1)}}_{\mu_{(i-2)}\tau_{i}}\,
\coW^{\nu}_{\mu_{(i-1)}\tau_{(i+1)}}\nonumber\\
&=& \delta^{\mu_1+n}_{\tau_1+\tau_2}\,
\delta^{\mu_2+n}_{\mu_1+\tau_3}\cdots
\delta^{\mu_{(i-1)+n}}_{\mu_{(i-2)}+\tau_{i}}\,
\delta^{\nu+n}_{\mu_{(i-1)}+\tau_{(i+1)}}.\nonumber\\
&=& \delta^{\nu+in}_{\tau_1+\tau_2+\cdots+\tau_{(i+1)}}\,,
\ \ \ \nu = 1,\dots,n-1,
\end{eqnarray*}
which, as required, is symmetric under interchage of the~$\tau_i$. Using
\eqref{Casform}, \eqref{nsingsolzero}, \eqref{nsingsolone}, and
\eqref{Cascoeff} we obtain the~$n-1$ Casimir invariants
\begin{equation}
\Casi^\nu(\fv^1,\dots,\fv^n) =
\sum_{i \ge 0}
\frac{1}{(i+1)!}\,\,
{\delta^{\nu+in}_{\tau_1+\tau_2+\cdots+\tau_{(i+1)}}}\,
\fv^{\tau_1}\cdots\fv^{\tau_{(i+1)}}\,
\af^\nu_{i}(\fv^n),
\eqlabel{CasiLeib}
\end{equation}
for~$\nu=1,\dots,n-1$. The superscript~$\nu$ on~$\af$ indicates that the
arbitrary function is different for each Casimir, and recall the subscript~$i$
denotes the~$i$th derivative with respect to~$\fv^n$. The~$n$th invariant is
simply~$\Casi^\nu(\fv^n) = \af^n(\fv^n)$, corresponding to the null
eigenvector in the system. Thus there are~$n$ independent Casimirs, as stated
in \secref{nsingWn}.
For the Leibniz semidirect sum case, since~$\Wn$ is nonsingular, there will be
an extra Casimir given by \eqref{CasiLeib} with~$\nu=0$, and the~$\tau_i$ sums
run from~$0$ to~$n-1$. This is the same form as the~$\nu=1$ Casimir of the
order~$(n+1)$ nilpotent extension.
For the $i$th term in \eqref{CasiLeib}, the maximal value of any~$\tau_j$ is
achived when all but one (say, $\tau_1$) of the~$\tau_j$ are equal to~$n-1$,
their maximum value. In this case we have
\[
\tau_1+\tau_2+\cdots+\tau_{i+1} = \tau_1 + i(n-1) = \nu + i n,
\]
so that~$\tau_1 = i+\nu$. Hence, the~$i$th term depends only
on~\hbox{$\l(\fv^{\nu+i},\dots,\fv^n\r)$}, and the~$\nu$th Casimir depends
on~\hbox{$\l(\fv^\nu,\dots,\fv^n\r)$}. Also,
\[
\max{\l(\tau_1+\cdots+\tau_{i+1}\r)} = (i+1)(n-1) = \nu + i n,
\]
which leads to~\hbox{$\max i = n - \nu - 1$}. Thus the sum \eqref{CasiLeib}
terminates, as claimed in \secref{nsingWn}. We rewrite \eqref{CasiLeib} in
the more complete form
\[
\Casi^\nu(\fv^\nu,\dots,\fv^n) =
\sum_{k = 1}^{n-\nu}
\frac{1}{k!}\,\,
{\delta^{\nu+(k-1)n}_{\tau_1+\tau_2+\cdots+\tau_{k}}}\,
\fv^{\tau_1}\cdots\fv^{\tau_{k}}\,
\af^\nu_{k-1}(\fv^n),
\]
for~$\nu=0,\dots,n$. \tabref{CasimirLeibn5} gives the~$\nu=1$ Casimirs
up to order~$n=5$.
\begin{table}
\caption{Casimir invariants for Leibniz extensions up to order~$n=5$
($\nu=1$). The primes denote derivatives.}
\tablabel{CasimirLeibn5}
\vskip 1em
\begin{center}
\begin{tabular}{ll} \hline
$n$ & Invariant \\[6pt] \hline
1 & $\af(\fv^1)$
\\[9pt]
2 & $\fv^1\af(\fv^2)$
\\[9pt]
3 & $\fv^1\af(\fv^3) + \frac{1}{2}{(\fv^2)^2}\af'(\fv^3)$
\\[9pt]
4 & $\fv^1\af(\fv^4) + \fv^2\fv^3\af'(\fv^4)
+ \frac{1}{3!}(\fv^3)^3\af''(\fv^4)$
\\[9pt]
5 & $\fv^1\af(\fv^5) + \l(\fv^2\fv^4
+ \frac{1}{2}(\fv^3)^2\r)\af'(\fv^5)
+ \frac{1}{2}\fv^3(\fv^4)^2\af''(\fv^5)
+ \frac{1}{4!}(\fv^4)^4\af'''(\fv^5)$
\\[9pt] \hline
\end{tabular}
\end{center}
\end{table}
\subsubsection{Singular~$\Wn$}
\seclabel{singWnex}
Now consider the~$n=4$ extension from \secref{neqfour}, \caseref{n4-10}c. We
have
\[
{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_{(2)} = \l({\begin{array}{ccc}
\makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}}\r),\ \ \ \
\Wn = \l({\begin{array}{ccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}}\r),
\]
with~${\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_{(1)} = {\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_{(3)} = 0$. The pseudoinverse of~$\Wn$
is~\hbox{$\Wn^{-1}=\Wn$} and the projection operator is
\[
{\Proj^\nu}_\tau \mathrel{\raisebox{.069ex}{:}\!\!=} \Wn^{\nu\kappa}\,\Wni_{\kappa\tau} =
\l({\begin{array}{ccc}
\makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1}
\end{array}}\r).
\]
The solvability condition \eqref{solvcond0} is obviously satisfied. We build
the coextension given by \eqref{coextension}, which in matrix form is
\[
\coW^{(\nu)} = {\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}^{(\nu)}\,\Wn^{-1} + ({\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}^{(\nu)}\,\Wn^{-1})^T
- \Wn^{-1}\,\Wn\,{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}^{(\nu)}\,\Wn^{-1},
\]
to obtain
\[
\coW^{(1)} = \l({\begin{array}{ccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} \\
\makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0}\end{array}}\r),\ \ \
\coW^{(2)} = \coW^{(3)} = 0.
\]
These are symmetric and obviously satisfy \eqref{coextcond}, so we have a good
coextension. Using \eqref{singCas}, \eqref{singsolzero},
\eqref{Cascoeffsing}, and \eqref{agcdefsing} we can write, for~$\nu=1$ and~$3$,
\begin{eqnarray*}
\Casi^1 &=& \fv^1\afi(\fv^4) + \fv^2\,\fv^3\afi'(\fv^4),\nonumber\\
\Casi^3 &=& \fv^3\afii(\fv^4).
\end{eqnarray*}
This extension has two null eigenvectors, so from \secref{Casdirprod} we also
have the Casimir~$\afiii(\fv^2,\fv^4)$. The functions~$\afi$, $\afii$, and
$\afiii$ are arbitrary, and the prime denotes differentiation with respect to
argument.
\section{Classification of Extensions of a Lie Algebra}
\seclabel{classext}
\ifmypprint
{\small
\begin{verbatim}
$Id: sec-classext.tex,v 2.1 1998/11/13 08:37:16 jeanluc Exp $
\end{verbatim}
}
\fi
In this section we return to the main problem introduced in
\secref{theproblem}: the classification of algebra extensions built by
forming~$n$-tuples of elements of a single Lie algebra~$\LieA$. The elements
of this Lie algebra~$\LieAx$ are written as~\hbox{$\alpha \mathrel{\raisebox{.069ex}{:}\!\!=}
\l(\alpha_1,\dots,\alpha_n\r)$},~\hbox{$\alpha_i \in \LieA$}, with a
bracket defined by
\begin{equation}
{\lpb\alpha\com\beta\rpb}_\lambda = {\W_\lambda}^{\mu\nu}\,
\lpb\alpha_\mu\com\beta_\nu\rpb,
\tag{\ref{eq:extbrack}}
\end{equation}
where~${\W_\lambda}^{\mu\nu}$ are constants. We will call~$n$ the \emph{order}
of the extension. Recall (see \secref{theproblem}) the~$\W$'s are symmetric in
their upper indices,
\begin{equation}
{\W_\lambda}^{\mu\nu} = {\W_\lambda}^{\nu\mu}\,,
\tag{\ref{eq:Wcommute}}
\end{equation}
and commute,
\begin{equation}
\W^{(\nu)}\,\W^{(\sigma)} = \W^{(\sigma)}\,\W^{(\nu)},
\tag{\ref{eq:Wcommute}}
\end{equation}
where the~\hbox{$n\times n$} matrices~$\W^{(\nu)}$ are defined
by~${{[\W^{(\nu)}]}_\lambda}^{\mu} := {\W_\lambda}^{\nu\mu}$. Since
the~$\W$'s are 3-tensors we can also represent their elements by matrices
obtained by fixing the lower index,
\begin{equation}
\W_{(\lambda)}\ :\ {\l[\W_{(\lambda)}\r]}^{\mu\nu} :=
{\W_{\lambda}}^{\mu\nu},
\eqlabel{lowindex}
\end{equation}
which are symmetric but do not commute. Either collection of
matrices,~\eqref{Wupdef} or~\eqref{lowindex}, completely describes the Lie
bracket, and which one we use will be understood by whether the parenthesized
index is up or down.
What do we mean by a classification? A classification is achieved if we
obtain a set of normal forms for the extensions that are independent, that is,
not related by linear transformations. We use linear transformations because
they preserve the Lie--Poisson structure---they amount to transformations of
the~$\W$ tensor. We thus begin by assuming the most general~$\W$ possible.
We first show in \secref{directprod} how an extension can be broken down into
a direct sum of degenerate subblocks (degenerate in the sense that the
eigenvalues have multiplicity greater than unity). The classification scheme
is thus reduced to the study of a single degenerate subblock. In
\secref{classcoho} we couch our particular extension problem in terms of the
Lie algebra cohomology language of \secref{extension} and apply the techniques
therein. The limitations of this cohomology approach are investigated in
\secref{furthertrans}, and we look at other coordinate transformations
that do not necessarily preserve the extension structure of the algebra, as
expressed in diagram~\eqref{equivext2}. In \secref{Leibniz} we introduce a
particular type of extension, called the Leibniz extension, which is in a sense
the ``maximal'' extension. Finally, in \secref{lowdimext} we give an explicit
classification of solvable extensions up to order four.
\subsection{Direct Sum Structure}
\seclabel{directprod}
A set of commuting matrices can be put into simultaneous block-diagonal form
by a coordinate transformation, each block corresponding to a degenerate
eigenvalue~\cite{Suprunenko}. Let us denote the change of basis by a
matrix~${\M_{\beta}}^{\bar\alpha}$, with
inverse~${\l(M^{-1}\r)_{\bar\alpha}}^{\beta}$, such that the
matrix~\hbox{${{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}}^{(\nu)}$}, whose components are given by
\[
{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_{\bar\beta}}{}^{\bar\alpha\nu} =
{(M^{-1})_{\bar\beta}}^{\lambda} \,
{\W_{\lambda}}^{\mu\nu} \, {\M_{\mu}}^{\bar\alpha}\ ,
\]
is in block-diagonal form for all~$\nu$~\cite{Suprunenko}.
However,~${\W_\lambda}^{\mu\nu}$ is a 3-tensor and so the third index is
also subject to the coordinate change:
\[
{{\smash{\mbox{$\overline\W$}}\!\mskip.8\thinmuskip}_{\bar\beta}}^{\bar\alpha\bar\gamma} =
{{\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}_{\bar\beta}}\,{}^{\bar\alpha\nu}
{\M_{\nu}}^{\bar\gamma}\, .
\]
This last step adds linear combinations of the~${\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}^{(\nu)}$'s together, so
the~${\smash{\mbox{$\widetilde\W$}}\!\mskip.8\thinmuskip}^{(\nu)}$'s and the~${\smash{\mbox{$\overline\W$}}\!\mskip.8\thinmuskip}^{(\bar\gamma)}$'s have the same
block-diagonal structure. Note that the~${{\smash{\mbox{$\overline\W$}}\!\mskip.8\thinmuskip}}$ tensors are still symmetric
in their upper indices, since this property is preserved by a change of basis.
So from now on we just assume that we are working in a basis where
the~$\W^{(\nu)}$'s are block-diagonal and symmetric in their upper indices;
this symmetry means that if we look at a~$\W$ as a cube, then in the
block-diagonal basis it consists of smaller cubes along the main
diagonal. This is the 3-tensor equivalent of a block-diagonal matrix.
Block-diagonalization is the first step in the classification: each block
of~$\W$ is associated with an ideal (hence, a subalgebra) in the
full~$n$-tuple algebra~$\LieA$. \inthesis{Show this.} Hence, by the definition
of \secref{semidext} the algebra~$\LieA$ is a direct sum of the
subalgebras associated with each block. Each of these subalgebras can be
studied independently, so from now on we assume that we have~$n$ commuting
matrices, each with~$n$-fold degenerate eigenvalues. The eigenvalues can,
however, be different for each matrix. \inthesis{Could say something about
reducibility here.}
Such a set of commuting matrices can be put into lower-triangular form by a
coordinate change, and again the transformation of the third index preserves
this structure (though it changes the eigenvalue of each matrix). The
eigenvalue of each matrix lies on the diagonal; we denote the eigenvalue
of~$\W^{(\mu)}$ by~$\ev^{(\mu)}$. The matrix~${\W_{(1)}}$, which as prescribed
by~\eqref{lowindex} consists of the first row of the lower-triangular
matrices~$\W^{(\mu)}$, is given by
\[
{\W_{(1)}} = \l(\begin{array}{ccccc}
\ev^{(1)} & 0 & 0 & \cdots & 0 \\
\ev^{(2)} & 0 & 0 & \cdots & 0 \\
\vdots & & & & \vdots \\
\ev^{(n)} & 0 & 0 & \cdots & 0
\end{array}\r).
\]
Evidently, the symmetry of~${\W_{(1)}}$ requires
\[
\ev^{(\nu)} = \evone\,{\delta_1}^\nu\, ;
\]
that is, all the matrices~$\W^{(\mu)}$ are nilpotent (their eigenvalues
vanish) except for~$\W^{(1)}$ when~\hbox{$\evone\ne 0$}. If this first
eigenvalue is nonzero then it can be scaled to~$\evone=1$ by the coordinate
transformation~${\M_{\nu}}^{\bar\alpha} =
\evone^{-1}~{\delta_\nu}^{\bar\alpha}$. We will use the
symbol~$\zerorone$ to mean a variable that can take the value 0 or 1.
\subsection{Connection to Cohomology}
\seclabel{classcoho}
We now bring together the abstract notions of \secref{cohoext} with
the~$n$-tuple extensions of \secref{theproblem}. It is shown in
\secref{prelimsplit} that we need only classify the case
of~\hbox{$\zerorone=0$}. This case will be seen to correspond to solvable
extensions, which we classify in \secref{solvext}.
\subsubsection{Preliminary Splitting}
\seclabel{prelimsplit}
Assume we are in the basis described at the end of \secref{directprod} and,
for now, suppose~$\evone = 1$. The set of elements of the form~$\beta =
\l(0,\beta_2,\dots,\beta_n\r)$ is a nilpotent ideal in~$\LieAx$ that we
denote by~$\LieAxb$ ($\LieAxb$ is thus a solvable subalgebra~\cite{Jacobson}).
Hence, we can construct the algebra~$\LieA =
\LieAx/\LieAxb$, so that~$\LieAx$ is an extension of~$\LieA$ by~$\LieAxb$.
If~$\LieA$ is semisimple, then~$\LieAxb$ is the radical of~$\LieAx$ (the
maximal solvable ideal). It is easy to see that the elements of~$\LieA$ are
of the form~$\alpha = \l(\alpha_1,0,\dots,0\r)$. We will now see
that~$\LieAx$ splits; that is, there exist coordinates in which~$\LieAx$ is
manifestly the semidirect sum of~$\LieA$ and the (in general non-Abelian)
algebra~$\LieAxb$.
In~\apxref{woneident} we give a lower-triangular coordinate transformation
that makes~$\W^{(1)}=I$, the identity matrix. Assuming we have effected this
transformation, the mappings~$i$,~$\pi$, and~$\tau$ of \secref{extension} are
given by
\begin{alignat*}{3}
i & :\LieAxb\ &\longrightarrow \LieAx,\ \ \
&i(\alpha_2,\dots,\alpha_n) = (0,\alpha_2,\dots,\alpha_n)
\nonumber\\
\pi & :\LieAx\ &\longrightarrow \LieA,\ \ \
&\pi(\alpha_1,\alpha_2,\dots,\alpha_n)
= \alpha_1,\\
\tau & :\LieA\ &\longrightarrow \LieAx,\ \ \
&\tau(\alpha_1)
= (\alpha_1,0,\dots,0),\nonumber
\end{alignat*}
and the cocycle of Eq.~\eqref{cocycle} is
\begin{eqnarray*}
i\,\omega(\alpha,\beta) &=&
{\lpb\tau\,\alpha\com\tau\,\beta\rpb}_\LieAx
- \tau\,{\lpb\alpha\com\beta\rpb}_\LieA\nonumber\\
&=& {\lpb(\alpha_1,0,\dots,0)\com(\beta_1,0,\dots,0)\rpb}_\LieAx
- (\lpb\alpha_1\com\beta_1\rpb,0,\dots,0)\nonumber\\
&=& 0.
\end{eqnarray*}
\inthesis{Extra steps?} Since~\hbox{$\omega\equiv 0$}, the extension is a
semidirect sum (see \secref{semidext}). The coordinate transformation that
made~$\W^{(1)}=I$ removed a coboundary, making the above cocycle vanish
identically. For the case where~$\LieA$ is finite-dimensional and semisimple,
we have an explicit demonstration of the Levi decomposition theorem: any
finite-dimensional\footnote{The inner bracket can be infinite dimensional, but
the order of the extension is finite.} Lie algebra~$\LieAx$ (of characteristic
zero) with radical~$\LieAxb$ is the semidirect sum of a semisimple Lie
algebra~$\LieA$ and~$\LieAxb$~\cite{Jacobson}.
\subsubsection{Solvable Extensions}
\seclabel{solvext}
Above we assumed the eigenvalue~$\evone$ of the first matrix was unity;
however, if this eigenvalue vanishes, then we have a solvable algebra
of~$n$-tuples to begin with. Since~$n$ is arbitrary we can study these two
solvable cases together.
Thus, we now suppose~$\LieAx$ is a solvable Lie algebra of~$n$-tuples (we
reuse the symbols~$\LieAx$,~$\LieA$, and~$\LieAxb$ to parallel the notation of
\secref{cohoalgebra}), where all of the the~$\W^{(\mu)}$'s are
lower-triangular with zeros along the diagonal. Note
that~$\W^{(n)}=0$, so the set of elements of the form~$\alpha =
(0,\dots,0,\alpha_n)$ forms an Abelian subalgebra of~$\LieAx$. In fact, this
subalgebra is an ideal. Now assume~$\LieAx$ contains an Abelian ideal of
order~\hbox{$n-m$} (the order of this ideal is at least~$1$), which we denote
by~$\LieAxb$. The elements of~$\LieAxb$ can always be cast in the form
\[
\alpha = (0,\dots,0,\alpha_{m+1},\dots,\alpha_n)
\]
via a coordinate transformation that preserves the lower-triangular,
nilpotent form of the~${\W}^{(\mu)}$ \comment{Proved in Lie--Poisson II ring
notepad}.
We also denote by~$\LieA$ the algebra of~$m$-tuples with the bracket
\[
{{\lpb(\alpha_1,\dots,\alpha_{m})
\com(\beta_1,\dots,\beta_{m})\rpb}_\LieA}_\lambda
= \sum_{\mu,\nu = 1}^{m} {\W_\lambda}^{\mu\nu}\,
\lpb\alpha_\mu\com\beta_\nu\rpb\,,\ \ \lambda=1,\dots,m.
\]
It is trivial to show that~$\LieA = \LieAx/\LieAxb$, so that~$\LieAx$ is an
extension of~$\LieA$ by~$\LieAxb$. Since~$\LieAxb$ is Abelian we can use the
formalism of \secref{cohoalgebra} (the other case we used above was
for~$\LieAxb$ non-Abelian but where the extension was semidirect). The
injection and projection maps are given by
\begin{alignat*}{3}
i& : \LieAxb\ &\longrightarrow \LieAx,\ \ \
&i(\alpha_{m+1},\dots,\alpha_n)
= (0,\dots,0,\alpha_{m+1},\dots,\alpha_n),\nonumber\\
\pi& : \LieAx\ &\longrightarrow \LieA,\ \ \
&\pi(\alpha_1,\alpha_2,\dots,\alpha_n)
= (\alpha_1,\dots,\alpha_{m}),\\
\tau & :\LieA\ &\longrightarrow \LieAx,\ \ \
&\tau(\alpha_1,\dots,\alpha_{m})
= (\alpha_1,\dots,\alpha_{m},0,\dots,0).\nonumber
\end{alignat*}
From the definition of the action, Eq.~\eqref{rhodef}, we have
for~\hbox{$\alpha \in \LieA$} and~\hbox{$\eta \in \LieAxb$},
\begin{eqnarray}
i\,\rho_\alpha\,\eta &=& {\lpb \tau\,\alpha\com i\,\eta\rpb}_\LieAx\nonumber\\
&=& {\lpb (\alpha_1,\dots,\alpha_{m},0,\dots,0)
\com (0,\dots,0,\eta_{m+1},\dots,\eta_n)\rpb}_\LieAx
\nonumber\\
&=& \sum_{\mu=1}^{m}\,\sum_{\nu=m+1}^{n-1}(0,\dots,0,
\W_{m+2}^{\,\,\mu\nu}{\lpb\alpha_\mu\com\eta_\nu\rpb}
,\dots,
{\W_n}^{\mu\nu}{\lpb\alpha_\mu\com\eta_\nu\rpb}).
\eqlabel{solvaction}
\end{eqnarray}
In addition to the action, the solvable extension is also characterized by the
cocycle defined in Eq.~\eqref{cocycle},
\begin{eqnarray}
\!\!\!\!\!\!\!\!\!
i\,\omega(\alpha,\beta) &=&
{\lpb\tau\,\alpha\com\tau\,\beta\rpb}_\LieAx
- \tau\,{\lpb\alpha\com\beta\rpb}_\LieA\nonumber\\
&=& {\lpb(\alpha_1,\dots,\alpha_{m},0,\dots,0)
\com(\beta_1,\dots,\beta_{m},0,\dots,0)\rpb}_\LieAx
\nonumber\\
&&\mbox{} - \tau\,{\lpb(\alpha_1,\dots,\alpha_{m})
\com(\beta_1,\dots,\beta_{m})\rpb}_\LieA\nonumber\\
&=& \sum_{\mu,\nu=1}^{m}(0,\dots,0,
\W_{m+1}^{\!\!\mu\nu}{\lpb\alpha_\mu\com\beta_\nu\rpb}
,\dots,{\W_n}^{\mu\nu}{\lpb\alpha_\mu\com\beta_\nu\rpb}).
\eqlabel{solvcocycle}
\end{eqnarray}
We can illustrate which parts of the~$\W$'s contribute to the action and which
to the cocycle by writing
\begin{equation}
\W_{(\lambda)} = \l(\begin{array}{c|c}
{\bf \ww}_\lambda\ & \ {\bf r}_\lambda \\ \hline
{\bf r}_\lambda^T\ & {\bf 0}
\end{array}\r),\ \ \lambda=m+1,\dots,n,
\eqlabel{Wform}
\end{equation}
where the~${\bf\ww}_\lambda$'s are~\hbox{$m\times m$} symmetric matrices that
determine the cocycle~$\omega$ and the~${\bf r}_\lambda$'s are~\hbox{$m\times
(n-m)$} matrices that determine the action~$\rho$. The \hbox{$(n-m)\times(n-m)$}
zero matrix on the bottom right of the~$\W_{(\lambda)}$'s is a consequence
of~$\LieAxb$ being Abelian.
The algebra~$\LieA$ is completely characterized by the~$\W_{(\lambda)}$,
$\lambda = 1,\dots,m$. Hence we can look for the maximal Abelian ideal
of~$\LieA$ and repeat the procedure we used for the full~$\LieAx$. It is
straightforward to show that although coordinate transformations of~$\LieA$
might change the cocycle~$\omega$ and the action~$\rho$, they will not alter
the \emph{form} of~\eqref{Wform} \comment{Proved in Lie--Poisson II ring
notepad}.
Recall that in \secref{cohoalgebra} we defined 2-coboundaries as 2-cocycles
obtained from 1-cochains by the coboundary operator,~$s$. The 2-coboundaries
turned out to be removable obstructions to a semidirect sum structure. Here
the coboundaries are associated with the parts of the~$\W_{(\lambda)}$ that
can be removed by (a restricted class of) coordinate transformations, as shown
below.
Let us explore the connection between 1-cochains and coboundaries in the
present context. Since a 1-cochain is just a linear mapping from~$\LieA$
to~$\LieAxb$, for~\hbox{$\alpha = (\alpha_1,\dots,\alpha_{m}) \in \LieA$} we
can write this as
\begin{equation}
\omega^{(1)}_\mu(\alpha) =
-\sum_{\lambda=1}^{m}{k_\mu}^\lambda\, \alpha_\lambda\,,
\ \ \mu=m+1,\dots,n,
\eqlabel{thecobound}
\end{equation}
where the~${k_\mu}^\lambda$ are arbitrary constants. To find the form of a
2-coboundary we act on the 1-cochain~\eqref{thecobound} with the coboundary
operator; using~\eqref{2cobound} and~\eqref{solvaction} we obtain
\begin{eqnarray}
\omega^{\rm cob}_\lambda(\alpha,\beta)
&=& (s\,\omega^{(1)})(\alpha,\beta),
\nonumber\\
&=& \rho_\alpha\omega^{(1)}(\beta) + \rho_\beta\omega^{(1)}(\alpha)
- \omega^{(1)}({\lpb\alpha\com\beta\rpb}_\LieA),
\nonumber\\
&=& \sum_{\mu=1}^{m}\,\sum_{\nu=m+1}^{n}\,{\W_\lambda}^{\mu\nu}
\lpb\alpha_\mu\com\omega^{(1)}_\nu(\beta)\rpb\eqlabel{cob1}\\
&&\mbox{} - \sum_{\mu=1}^{m}\,\sum_{\nu=m+1}^{n}\,{\W_\lambda}^{\mu\nu}
\lpb\beta_\mu\com\omega^{(1)}_\nu(\alpha)\rpb
+ \sum_{\mu,\nu,\sigma=1}^{m}{k_\lambda}^\sigma\,
{\W_\sigma}^{\mu\nu}
\lpb\alpha_\mu\com\beta_\nu\rpb.\nonumber
\end{eqnarray}
After inserting~\eqref{thecobound} into~\eqref{cob1} and relabeling, we obtain
the general form of a 2-coboundary
\[
\omega^{\rm cob}_\lambda(\alpha,\beta) = \sum_{\mu,\nu=1}^{m}\,
{\Wcob_\lambda}^{\mu\nu}
\lpb\alpha_\mu\com\beta_\nu\rpb,
\ \ \ \lambda=m+1,\dots,n,
\]
where
\begin{equation}
{\Wcob_\lambda}^{\mu\nu} \mathrel{\raisebox{.069ex}{:}\!\!=}
\sum_{\tau=1}^{m}\,
{k_\lambda}^\tau\,{\W_\tau}^{\mu\nu}
- \sum_{\sigma=m+1}^{n}\,
\l(
{k_\sigma}^\mu\,{\W_\lambda}^{\nu\sigma}
+ {k_\sigma}^\nu\,{\W_\lambda}^{\mu\sigma}
\r).
\eqlabel{cob}
\end{equation}
To see how coboundaries are removed, consider the lower-triangular coordinate
transformation
\[
\l[{\M_\sigma}^{\bar \tau}\r] = \l(\begin{array}{c|c}
{\bf I}\ \ & \,\ {\bf 0} \\ \hline
{\bf k}\ \ & \ c\,{\bf I}
\end{array}\r),
\]
where~$\sigma$ labels rows. This transformation subtracts~$\Wcob_{(\lambda)}$
from~$\W_{(\lambda)}$ for \hbox{$\lambda>m$} and leaves the first~$m$ of the
$\W_{(\lambda)}$'s unchanged\comment{Proved in Lie--Poisson II ring notepad}.
In other words, if~${{\smash{\mbox{$\overline\W$}}\!\mskip.8\thinmuskip}}$ is the transformed~$\W$,
\begin{equation}
{{\smash{\mbox{$\overline\W$}}\!\mskip.8\thinmuskip}_{(\lambda)}} = \l\{\begin{array}{ll}
\W_{(\lambda)}\,,\ \ &\lambda = 1,\dots,m;\\
\l(\begin{array}{c|c}
c^{-1}\,({\bf w}_\lambda - {\bf \Wcob}_\lambda)\
& \ {\bf r}_\lambda \\ \hline
{\bf r}_\lambda^T\ & {\bf 0}
\end{array}\r),\ \ &\lambda=m+1,\dots,n.
\end{array}\r.
\eqlabel{newW}
\end{equation}
We have also included in this transformation an arbitrary scale factor~$c$.
Since by~\eqref{solvcocycle} the block in the upper-left characterizes the
cocycle, we see that the transformed cocycle is the cocycle characterized
by~${\bf\ww}_\lambda$ minus the coboundary characterized
by~${\bf\Wcob}_\lambda$.
The special case we will encounter most often is when the maximal Abelian
ideal of~$\LieAx$ simply consists of elements of the
form~\hbox{$(0,\dots,0,\alpha_n)$}. For this case~$m=n-1$, and the action
vanishes since~\hbox{${\W_n}^{\mu n}=0$} (the extension is central). The
cocycle~$\omega$ is entirely determined by~$\W_{(n)}$. The form of the
coboundary is reduced to
\begin{equation}
{\Wcob_n}^{\mu\nu} =
\sum_{\tau=1}^{n-1}\,
{k_n}^\tau\,{\W_\tau}^{\mu\nu},
\eqlabel{cobnoaction}
\end{equation}
that is, a linear combinations of the first~$(n-1)$ matrices. Thus it is easy
to see at a glance which parts of the cocycle characterized $\W_{(n)}$ can be
removed by lower-triangular coordinate transformations.
\subsection{Further Coordinate Transformations}
\seclabel{furthertrans}
In the previous section we restricted ourselves to lower-triangular coordinate
transformations, which in general preserve the lower-triangular structure of
the~$\W^{(\mu)}$. But when the matrices are relatively sparse, there exist
non-lower-triangular coordinate transformations that nonetheless preserve the
lower-triangular structure. As alluded to in \secref{abelianext}, these
transformations are outside the scope of cohomology theory, which is
restricted to transformations that preserve the exact form of the action and
the algebras~$\LieA$ and~$\LieAxb$, as shown by~\eqref{newW}. In other words,
cohomology theory classifies extensions \emph{given} $\LieA$, $\LieAxb$,
and~$\rho$. We need not obey this restriction. We can allow
non-lower-triangular coordinate transformations as long as they preserve the
lower-triangular structure of the~$\W^{(\mu)}$'s.
We now discuss a particular class of such transformations that will be useful
in \secref{lowdimext}. Consider the case where both the algebra of
$(n-1)$-tuples~$\LieA$ and that of $1$-tuples~$\LieAxb$ are Abelian. Then the
possible (solvable) extensions, in lower triangular form, are characterized
by~$\W_{(\lambda)}=0$, $\lambda=1,\dots,n-1$, with $\W_{(n)}$ arbitrary
(except for~${\W_n}^{\mu n}=0$). Let us apply a coordinate change of the form
\[
\M = \l(\begin{array}{c|c} {\bf \mm} \ & {\bf 0} \\ \hline
{\bf 0} & \ c\,
\end{array}\r),
\]
where~${\bf \mm}$ is an~$(n-1)\times (n-1)$ nonsingular matrix and~$c$ is
again a nonzero scale factor. Denoting by~${\smash{\mbox{$\overline\W$}}\!\mskip.8\thinmuskip}$ the transformed~$\W$,
we have
\begin{equation}
{{\smash{\mbox{$\overline\W$}}\!\mskip.8\thinmuskip}_{(\lambda)}} = \l\{\begin{array}{ll}
0\,,\ \ &\lambda = 1,\dots,n-1;\\
\l(\begin{array}{c|c}
c^{-1}\,{\bf m}^T\,{\bf \ww}_\lambda\,{\bf m}\
& \ {\bf 0} \\ \hline
{\bf 0}\ & \ \ 0\
\end{array}\r),\ \ &\lambda=n.
\end{array}\r.
\eqlabel{newW2}
\end{equation}
This transformation does not change the lower-triangular form of the
extension, even if~${\bf\mm}$ is not lower-triangular. The manner in
which~${\bf\ww}_{n}$ is transformed by~$\M$ is very similar to that of a
(possibly singular) metric tensor: it can be diagonalized and rescaled such
that all its eigenvalues are~$0$ or~$\pm 1$. We can also change the overall
sign of the eigenvalues using~$c$ (something that cannot be done for a metric
tensor). Hence, we shall order the eigenvalues such that the~$+1$'s come
first, followed by the~$-1$'s, and finally by the~$0$'s. We will show in
\secref{lowdimext} how the negative eigenvalues can be eliminated to
harmonize the notation.
\inthesis{
Another useful transformation simply exchanges the position of two
matrices~$\W_{()}$.
}
\subsection{Appending a Semisimple Part}
\seclabel{semisimple}
In \secref{classcoho} we showed that because of the Levi decomposition theorem
we only needed to classify the solvable part of the extension for a given
degenerate block. Most physical applications have a semisimple part
($\evone=1$); when this is so, we shall label the matrices
by~$\W^{(0)},\W^{(1)},\dots,\W^{(n)}$, where they are now of
size~\hbox{$n+1$} and~$\W^{(0)}$ is the identity.%
\footnote{The term semisimple is not quite precise: if the base algebra is not
semisimple then neither is the extension. However we will use the term to
distinguish the different cases.} Thus the matrices labeled
by~$\W^{(1)},\dots,\W^{(n)}$ will always form a solvable subalgebra. This
explains the labeling in \secreftwo{matlowbetaRMHD}{matCRMHD}.
If the extension has a semisimple part ($\zerorone=1$, or
equivalently~$\W^{(0)}=I$), we shall refer to it as \emph{semidirect}. This
was the case treated in \secref{prelimsplit}. If the extension is not
semidirect, then it is solvable (and contains~$n$ matrices instead of~$n+1$).
Given a solvable algebra of~$n$-tuples we can carry out in some sense the
inverse of the Levi decomposition and append a semisimple part to the
extension. Effectively, this means that the~\hbox{$n\times n$}
matrices~$\W^{(1)},\dots,\W^{(n)}$ are made~\hbox{$n+1 \times n+1$} by adding
a row and column of zeros. Then we simply append the matrix~$\W^{(0)}=I$ to
the extension. In this manner we construct a semisimple extension from a
solvable one. This is useful since we will be classifying solvable
extensions, and afterwards we will want to recover their semidirect
counterpart.
The extension obtained by appending a semisimple part to the completely
Abelian algebra of~$n$-tuples will be called \emph{pure semidirect}. It is
characterized by~$\W^{(0)}=I$, and~${\W_\lambda}^{\mu\nu}=0$ for~$\mu,\nu>0$.
\subsection{Leibniz Extension}
\seclabel{Leibniz}
A particular extension that we shall consider is called the Leibniz
extension~\cite{Parthasarathy1976}. For the solvable case this extension has
the form
\begin{equation}
{\W}^{(1)} \mathrel{=\!\!\raisebox{.069ex}{:}} \Nilb = \l(\begin{array}{ccccc}
\makebox[1.3em]{0} & & & & \\
\makebox[1.3em]{1} & \makebox[1.3em]{0} & & & \\
& \makebox[1.3em]{1} & \makebox[1.3em]{0} & & \\
& & \cdots & \cdots & \\
& & & \makebox[1.3em]{1} & \makebox[1.3em]{0}
\end{array}\r)
\eqlabel{Nilb}
\end{equation}
or~${\W_\lambda}^{\mu\,1} = {\delta_{\lambda-1}}^{\mu}$,~$\lambda>1$; i.e. the
first matrix is an~\hbox{$n \times n$} Jordan block. In this case the other
matrices, in order to commute with~$\W^{(1)}$, must be in striped
lower-triangular form~\cite{Suprunenko}. After using the symmetry of the
upper indices the matrices can be reduced to
\inthesis{Add a few steps here.}
\begin{equation}
\W^{(\nu)} = (\Nilb)^\nu,
\eqlabel{Nilbmu}
\end{equation}
where on the right-hand side the~$\nu$ denotes an exponent, not a superscript.
An equivalent way of characterizing the Leibniz extension is
\begin{equation}
{\W_\lambda}^{\mu\nu} = {\delta_\lambda}^{\mu+\nu}\,,
\ \ \ \mu,\nu, \lambda = 1,\dots,n.
\eqlabel{sLeib}
\end{equation}
The tensor~$\delta$ is an ordinary Kronecker delta. Note that
neither~\eqref{Nilbmu} nor~\eqref{sLeib} are covariant expressions, reflecting
the coordinate-dependent nature of the Leibniz extension.
The Leibniz extension is in some sense a ``maximal'' extension: it is the only
extension that has~\hbox{$\W_{(\lambda)} \ne 0$} for
\emph{all}~$\lambda=2,\dots,n$ (up to coordinate transformations). Its
uniqueness will become clear in \secref{lowdimext}, and is discussed in
Thiffeault~\cite{Thiffeault1998diss}.
To construct the semidirect Leibniz extension, we append~$\W^{(0)}=I$, a
square matrix of size~$n+1$, to the solvable Leibniz extension above, as
described in \secref{semisimple}.
\subsection{Low-order Extensions}
\seclabel{lowdimext}
We now classify the algebra extensions of low order. As demonstrated in
\secref{classcoho} we only need to classify solvable algebras, which
means that~$\W^{(n)}=0$ for all cases. We will do the classification up to
order~$n=4$. For each case we first write down the most general set of
lower-triangular matrices~$\W^{(\nu)}$ (we have already used the fact that a
set of commuting matrices can be lower-triangularized) with the
symmetry~${\W_\lambda}^{\mu\nu}={\W_\lambda}^{\nu\mu}$ built in. Then we look
at what sort of restrictions the commutativity of the matrices places on the
elements. Finally, we eliminate coboundaries for each case by the methods of
\secreftwo{classcoho}{furthertrans}. This requires coordinate
transformations, but we usually will not bother using new symbols and just
assume the transformation was effected.
Note that, due to the lower-triangular structure of the extensions, the
classification found for an~$m$-tuple algebra applies to the first~$m$
elements of an~$n$-tuple algebra,~\hbox{$n>m$}. Thus,~$\W_{(n)}$ is the
cocycle that contains all of the new information not included in the
previous~\hbox{$m=n-1$} classification. These comments will become clearer as
we proceed.
We shall call an order~$n$ extension \emph{trivial} if~\hbox{$\W_{(n)} \equiv
0$}, so that the cocycle appended to the order~$n-1$ extension contributes
nothing to the bracket.
\subsubsection{n=1}
\seclabel{neqone}
This case is Abelian, with the only possible element~${W_1}^{11}=0$.
\subsubsection{n=2}
\seclabel{neqtwo}
The most general lower-triangular form for the matrices is
\[
\W^{(1)} = \l(\begin{array}{cc} \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
{\W_2}^{11} & \makebox[1.3em]{0} \end{array}\r),
\ \ \ \
\W^{(2)} = \l(\begin{array}{cc} \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} \end{array}\r).
\]
If~${\W_2}^{11} \ne 0$, then we can rescale it to unity. Hence we
let~${\W_2}^{11} \mathrel{\raisebox{.069ex}{:}\!\!=} \zerorone_1$, where~$\zerorone_1 = 0$ or $1$. The
case~$\zerorone_1 = 0$ is the Abelian case, while for~$\zerorone=1$ we have
the~$n=2$ Leibniz extension (\secref{Leibniz}). Thus for~$n=2$ there are only
two possible algebras. The cocycle which we have added at this stage is
characterized by~$\zerorone_1$.
\subsubsection{n=3}
\seclabel{neqthree}
Using the result of \secref{neqtwo}, the most general lower-triangular form is
\[
\W^{(1)} = \l(\begin{array}{ccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\zerorone_1 & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
{\W_3}^{11} & {\W_3}^{21} & \makebox[1.3em]{0} \end{array}\r),
\ \ \ \
\W^{(2)} = \l(\begin{array}{ccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
{\W_3}^{21} & {\W_3}^{22} & \makebox[1.3em]{0} \end{array}\r),
\]
and~$\W^{(3)} = 0$. These satisfy the symmetry condition~\eqref{upsym}, and
the requirement that the matrices commute leads to the condition
\[
\zerorone_1\,{\W_3}^{22} = 0.
\]
The symmetric matrix representing the cocycle is
\begin{equation}
W_{(3)} = \l(\begin{array}{ccc}
{\W_3}^{11} & {\W_3}^{21} & \makebox[1.3em]{0} \\
{\W_3}^{21} & {\W_3}^{22} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \end{array}\r).
\eqlabel{Wiii}
\end{equation}
If~$\zerorone_1 = 1$, then~${\W_3}^{22}$ must vanish. Then,
by~\eqref{cobnoaction} we can remove from~$\W_{(3)}$ a multiple of~$\W_{(2)}$,
and therefore we may assume~${\W_3}^{11}$ vanishes. A suitable rescaling
allows us to write~${\W_3}^{21}=\zerorone_2$, where~$\zerorone_2 = 0$ or $1$.
The cocycle for the case~$\zerorone_1=1$ is thus
\[
W_{(3)} = \l(\begin{array}{ccc}
\makebox[1.3em]{0} & \zerorone_2 & \makebox[1.3em]{0} \\
\zerorone_2 & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \end{array}\r).
\]
For~$\zerorone_2=1$ we have the Leibniz extension (\secref{Leibniz}).
If~$\zerorone_1 = 0$, we have the case discussed in \secref{furthertrans}.
For this case we can diagonalize and rescale~$\W_{(3)}$ such that
\[
W_{(3)} = \l(\begin{array}{ccc}
\pmz_1 & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \pmz_2 & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \end{array}\r),
\]
where~$(\pmz_1,\pmz_2)$ can be~$(1,1)$, $(1,0)$, $(0,0)$, or $(1,-1)$. This
last case, as alluded to at the end of \secref{furthertrans}, can be
transformed so that it corresponds to~$\zerorone_1=0$, $\zerorone_2=1$. The
choice~$(1,0)$ can be transformed to the~$\zerorone_1=1$, $\zerorone_2=0$
case.
\comment{with $M = \{\{1,0,0\},\{0,0,1\},\{0,1,0\}\}$} Finally
for~$(\pmz_1,\pmz_2)=(1,1)$ we can use the complex transformation
\[
\fv^1\rightarrow\frac{1}{\sqrt{2}}(\fv^1+\fv^2),\ \ \
\fv^2\rightarrow-\frac{\imi}{\sqrt{2}}(\fv^1-\fv^2),\ \ \
\fv^3\rightarrow\fv^3,
\]
to transform to the~$\zerorone_1=0$, $\zerorone_2=1$ case.
We allow complex transformations in our classification because we are chiefly
interested in finding Casimir invariants for Lie--Poisson brackets. If we
disallowed complex transformations, the final classification would contain a
few more members. The use of complex transformations will be noted as we
proceed.
There are thus four independent extensions for~$n=3$, corresponding to
\[
(\zerorone_1\com\zerorone_2)
\in \l\{(0\com 0)\com(0 \com 1)\com(1 \com 0)\com(1 \com 1)\r\}.
\]
These will be referred to as Cases~$1$--$4$, respectively.
Cases~\ref{case:n4-00} and~\ref{case:n4-10} have~$\zerorone_2=0$, and so are
trivial ($\W_{(3)}=0$). \caseref{n4-01} is the solvable part of the
compressible reduced MHD bracket (\secref{matCRMHD}). \caseref{n4-11} is the
solvable Leibniz extension.
\subsubsection{n=4}
\seclabel{neqfour}
Proceeding as before and using the result of \secreftwo{neqtwo}{neqthree}, we
now know that we need only write
\begin{equation}
\W_{(4)} = \l(\begin{array}{cccc}
{\W_4}^{11} & {\W_4}^{21} & {\W_4}^{31} & \makebox[1.3em]{0} \\
{\W_4}^{21} & {\W_4}^{22} & {\W_4}^{32} & \makebox[1.3em]{0} \\
{\W_4}^{31} & {\W_4}^{32} & {\W_4}^{33} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r).
\eqlabel{Wfour}
\end{equation}
The matrices~$\W_{(1)}$, $\W_{(2)}$, and~$\W_{(3)}$ are given by their~$n=3$
analogues padded with an extra row and column of zeros (owing to the
lower-triangular form of the matrices). The requirement that the
matrices~$\W^{(1)}\dots\W^{(4)}$ commute leads to the conditions%
\begin{equation}
\begin{split}
\zerorone_2\,{\W_4}^{33} &= 0, \\
\zerorone_2\,{\W_4}^{31} &= \zerorone_1\,{\W_4}^{22}, \\
\zerorone_2\,{\W_4}^{32} &= 0, \\
\zerorone_1\,{\W_4}^{32} &= 0.
\eqlabel{commrel4}
\end{split}
\end{equation}
There are four cases to look at, corresponding to the possible values
of~$\zerorone_1$ and~$\zerorone_2$.
\begin{case}
$\zerorone_1=0$, $\zerorone_2=0$
\caselabel{n4-00}
\end{case}
This is the unconstrained case discussed in \secref{furthertrans}, that is,
all the commutation relations \eqref{commrel4} are automatically satisfied.
We can diagonalize to give
\[
\W_{(4)} = \l(\begin{array}{cccc}
\pmz_1' & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \pmz_2' & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \pmz_3' & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r),
\]
where
\[
(\pmz_1',\pmz_2',\pmz_3')
\in \l\{(1,1,1),(1,1,0),(1,0,0),(0,0,0),(1,1,-1),(1,-1,0)\r\},
\]
so there are six distinct cases. The exact form of the transformation is
unimportant, but the~$(1,1,0)$ extension can be mapped to \caseref{n4-01} (the
transformation is complex\comment{First switch matrices 3 and 4, then use the
same complex transformations for $n=3$.}),~$(1,0,0)$ can be mapped to
\caseref{n4-10}a, and~$(1,-1,0)$ can be mapped to
\caseref{n4-01}. Finally the~$(1,1,1)$ extension can be mapped to
the~$(1,1,-1)$ case by a complex transformation. After transforming
that $(1,1,-1)$ case, we are left with
\[
\W_{(4)} = \l(\begin{array}{cccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r),
\l(\begin{array}{cccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
1 & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r)
\]
These will be called Cases~\ref{case:n4-00}a and~\ref{case:n4-00}b.
\begin{case}
$\zerorone_1=0$, $\zerorone_2=1$
\caselabel{n4-01}
\end{case}
The commutation relations~\eqref{commrel4} reduce to ${\W_{4}}^{31} =
{\W_{4}}^{32} = {\W_{4}}^{33} = 0$, and we have
\[
\W_{(4)} = \l(\begin{array}{cccc}
{\W_4}^{11} & {\W_4}^{21} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
{\W_4}^{21} & {\W_4}^{22} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\eqlabel{fff}
\end{array}\r).
\]
We can remove~${\W_4}^{21}$ because it is a coboundary (in this case a
multiple of~${\W_{(3)}}$). We can also rescale appropriately to obtain four
possible extensions:~$\W_{(4)}=0$, and
\[
\W_{(4)} =
\l(\begin{array}{cccc}
1 & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r),
\l(\begin{array}{cccc}
1 & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r),
\l(\begin{array}{cccc}
1 & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & -1 & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r).
\]
Again, the form of the transformation is unimportant, but it turns out that
the first of the above extensions can be mapped to \caseref{n4-10}c, and the
second and third to \caseref{n4-10}b. This last transformation is complex.
Thus there is only one independent possibility, the trivial
extension~$\W_{(4)}=0$.
\begin{case}
$\zerorone_1=1,\zerorone_2=0$
\caselabel{n4-10}
\end{case}
We can remove~~${\W_4}^{11}$ using a coordinate transformation. From the
commutation requirement \eqref{commrel4} we obtain~${\W_4}^{22} =
{\W_4}^{32} = 0$. We are left with~$\W_{(3)}=0$ and
\[
\W_{(4)} = \l(\begin{array}{cccc}
\makebox[1.3em]{0} & {\W_4}^{21} & {\W_4}^{31} & \makebox[1.3em]{0} \\
{\W_4}^{21} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
{\W_4}^{31} & \makebox[1.3em]{0} & {\W_4}^{33} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r).
\]
Using the fact that elements of the form~$(0,\alpha_2,0,\alpha_4)$ are an
Abelian ideal of this bracket, we find
that~${\W_4}^{33}{\W_4}^{31}=0$. \comment{This involves transforming to move
the Abelian ideal to the end with\\ $M =
\{\{1,0,0,0\},\{0,0,1,0\},\{0,1,0,0\},\{0,0,0,1\}\}$,\\ making use of the
cohomology (see NB II p.69), then transforming back.} Using an
upper-triangular transformation we can also make~${\W_4}^{21}{\W_4}^{31}=0$.
After suitable rescalings we find there are five cases. One of these,
\[
\W_{(4)} = \l(\begin{array}{cccc}
\makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
1 & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r),
\]
may be mapped to \caseref{n4-11} (below) with~$\zerorone_3=0$. We are thus
left with four cases: the trivial extensions,~$\W_{(4)}=0$, and
\[
\W_{(4)} = \l(\begin{array}{cccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r),
\l(\begin{array}{cccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
1 & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r),
\l(\begin{array}{cccc}
\makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
1 & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r).
\]
We will refer to these four extensions as Cases~\ref{case:n4-10}a--d,
respectively (Case~\ref{case:n4-10}a is the trivial extension).
\begin{case}
$\zerorone_1=1$, $\zerorone_2=1$
\caselabel{n4-11}
\end{case}
The elements~${\W_4}^{11}$ and~${\W_4}^{21}$ are coboundaries that can be
removed by a coordinate transformation. From \eqref{commrel4} we
have~${\W_4}^{33} = {\W_4}^{32} = 0, {\W_4}^{22} = {\W_4}^{31}
\mathrel{=\!\!\raisebox{.069ex}{:}} \zerorone_3$, so that
\[
\W_{(4)} = \l(\begin{array}{cccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \zerorone_3 & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \zerorone_3 & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\zerorone_3 & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}\r).
\]
For~$\zerorone_3=1$ we have the Leibniz extension. The two cases will be
referred to as \caseref{n4-11}a for~$\zerorone_3=0$
and~\ref{case:n4-11}b for~$\zerorone_3=1$.
\tabref{n=4extensions} summarizes the results. There are are total
of nine independent~$n=4$ extensions, four of which are trivial
($\W_{(4)}=0$). As noted in \secref{Leibniz} only the Leibniz extension,
\caseref{n4-11}b, has nonvanishing~$\W_{(i)}$ for all~\hbox{$1<i\le n$}.
\begin{table}
\caption{Enumeration of the independent extensions up to~$n=4$. We
have~$\W_{(1)}=0$ for all the cases, and we have left out a row and a column
of zeros at the end of each matrix. We have also omitted cases 1--4a, for
which~$\W_{(4)}=0$.}
\tablabel{n=4extensions}
\vskip 1em
\begin{center}
\begin{tabular}{lccc@{}c@{}c} \hline
Case & $\W_{(2)}$ & $\W_{(3)}$ & \multicolumn{3}{c}{$\W_{(4)}$} \\
& & & b & c & d \\[3pt] \hline
1 &
$\l(\makebox[1.3em]{0}\r)$ &
$\l({\begin{array}{cc} \makebox[1.3em]{0} & \makebox[1.3em]{0} \\ \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}}\r)$ &
$\l({\begin{array}{ccc} \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} \\ \makebox[1.3em]{0} & \makebox[1.3em]{1}
& \makebox[1.3em]{0} \\ \makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}}\r)$& &
\rule[-3em]{0cm}{6.5em}
\\
2 &
$\l(\makebox[1.3em]{0}\r)$ &
$\l({\begin{array}{cc} \makebox[1.3em]{0} & \makebox[1.3em]{1} \\ \makebox[1.3em]{1} & \makebox[1.3em]{0}
\end{array}}\r)$
&&&
\rule[-3em]{0cm}{6.5em}
\\
3 &
$\l(\makebox[1.3em]{1}\r)$ &
$\l({\begin{array}{cc} \makebox[1.3em]{0} & \makebox[1.3em]{0} \\ \makebox[1.3em]{0} & \makebox[1.3em]{0}
\end{array}}\r)$ &
$\l({\begin{array}{ccc} \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} \end{array}}\r)$&
$\l({\begin{array}{ccc} \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \end{array}}\r)$&
$\l({\begin{array}{ccc} \makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} \\
\makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} \end{array}}\r)$
\rule[-3em]{0cm}{6.5em}
\\
4 &
$\l(\makebox[1.3em]{1}\r)$ &
$\l({\begin{array}{cc} \makebox[1.3em]{0} & \makebox[1.3em]{1} \\
\makebox[1.3em]{1} & \makebox[1.3em]{0} \end{array}}\r)$ &
$\l({\begin{array}{ccc} \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} \\
\makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} \\
\makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} \end{array}}\r)$&&
\rule[-3em]{0cm}{6.5em}
\\ \hline
\end{tabular}
\end{center}
\end{table}
The surprising fact is that even to order four the normal forms of the
extensions involve no free parameters: all entries in the coefficients of the
bracket are either zero or one. There is no obvious reason this should hold
true if we try to classify extensions of order~$n>4$. It would be interesting
to find out, but the classification scheme used in this paper becomes
prohibitive at such high order. The problem is that some of the
transformations used to relate extensions cannot be systematically derived and
were obtained by educated guessing.
\subsection{Cohomology of Lie Algebras}
\seclabel{cohoalgebra}
\ifmypprint
{\small
\begin{verbatim}
$Id: sec-cohoalgebra.tex,v 2.1 1998/11/13 08:37:04 jeanluc Exp $
\end{verbatim}
}
\fi
We now introduce the abstract formalism of Lie algebra cohomology.
Historically there were two different reasons for the development of this
theory. One, known as the Chevalley--Eilenberg
formulation~\cite{Chevalley1948}, was developed from de Rham cohomology\@. de
Rham cohomology concerns the relationship between exact and closed
differential forms, which is determined by the global properties (topology) of
a differentiable manifold. A Lie group is a differentiable manifold and so has
an associated de Rham cohomology. If invariant differential forms are used in
the computation, one is led to the cohomology of Lie algebras presented in
this section~\cite{Knapp,Azcarraga,CecileII}. The second motivation is the
one that concerns us: we will show in \secref{extension} that the extension
problem---the problem of enumerating extensions of a Lie algebra---can be
related to the cohomology of Lie algebras.
Let~$\LieA$ be a Lie algebra, and let the vector space~$\vs$ over the
field~$\field$ (which we take to be the real numbers later) be a
left~$\LieA$-module,%
\footnote{When~$\vs$ is a right~$\LieA$-module, we have~\hbox{$\rho_{\lpb
\alpha\com\alpha'\rpb} = -\lpb \rho_\alpha \com \rho_{\alpha'} \rpb$}.
The results of this section can be adapted to a right action by changing the
sign every time a commutator appears.} that is, there is an operator~$\rho:
\LieA \times \vs \rightarrow \vs$ such that
\begin{eqnarray}
\rho_\alpha\, (v + v') &=& \rho_\alpha\,v + \rho_\alpha\,v',\nonumber\\
\rho_{\alpha + \alpha'}\, v &=& \rho_\alpha\,v + \rho_{\alpha'}\,v,
\nonumber\\
\rho_{\lpb \alpha\com\alpha'\rpb}v &=&
\lpb \rho_\alpha \com \rho_{\alpha'} \rpb\,v\,,
\eqlabel{rhomo}
\end{eqnarray}
for~$\alpha, \alpha' \in \LieA$ and~$v,v' \in \vs$. The operator~$\rho$ is
known as a left action. A~$\LieA$-module gives a representation of~$\LieA$
on~$\vs$.
An $n$-dimensional $\vs$-valued cochain~$\omega_n$ for~$\LieA$, or
just~$n$-cochain for short, is a skew-symmetric~$n$-linear mapping
\[
\omega_n :\ \stackrel{\longleftarrow n \longrightarrow}
{\LieA \times \LieA \times \dots \times \LieA}\
\longrightarrow \vs.
\]
Cochains are Lie algebra cohomology analogues of differential forms on a
manifold. Addition and scalar multiplication of~$n$-cochains are defined in
the obvious manner by
\begin{eqnarray*}
(\omega_n + \omega_n')(\alpha_1,\dots,\alpha_n)
&\mathrel{\raisebox{.069ex}{:}\!\!=}& \omega_n(\alpha_1,\dots,\alpha_n)
+ \omega_n'(\alpha_1,\dots,\alpha_n),\nonumber\\
(a\,\omega_n)(\alpha_1,\dots,\alpha_n)
&\mathrel{\raisebox{.069ex}{:}\!\!=}& a\,\omega_n(\alpha_1,\dots,\alpha_n),
\end{eqnarray*}
where~\hbox{$\alpha_1,\dots,\alpha_n \in \LieA$} and~\hbox{$a \in \field$}.
The set of all~$n$-cochains thus forms a vector space over the field~$\field$
and is denoted by~$C^n(\LieA,\vs)$. The $0$-cochains are defined to be just
elements of~$\vs$, so that~\hbox{$C^0(\LieA,\vs) = \vs$}.
The coboundary operator is the map between cochains,
\[
s_n : C^n(\LieA,\vs) \longrightarrow C^{n+1}(\LieA,\vs),
\]
defined by
\begin{eqnarray*}
(s_n\,\omega_n)(\alpha_1,\dots,\alpha_{n+1}) &\mathrel{\raisebox{.069ex}{:}\!\!=}&
\sum_{i=1}^{n+1}(-)^{i+1}\rho_{\alpha_i}\omega_n(\alpha_1,\dots,
\hat\alpha_i,\dots,\alpha_{n+1})\nonumber\\ &&\mbox{}
\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!
+ \sum_{\scriptscriptstyle{j,k=1}\atop\scriptscriptstyle{j<k}}^{n+1}
(-)^{j+k}\omega_n(\lpb\alpha_j\com\alpha_k\rpb,\alpha_1,\dots,
\hat\alpha_j,\dots,\hat\alpha_k,\dots,\alpha_{n+1}),
\end{eqnarray*}
where the caret means an argument is omitted. We shall often drop the~$n$
subscript on~$s_n$, deducing it from the dimension of the cochain on which~$s$
acts.
We shall make use mostly of the first few cases
\begin{eqnarray}
(s\,\omega_0)(\alpha_1) &=& \rho_{\alpha_1}\,\omega_0,
\eqlabel{1cobound}\\
(s\,\omega_1)(\alpha_1,\alpha_2) &=&
\rho_{\alpha_1}\,\omega_1(\alpha_2)
- \rho_{\alpha_2}\,\omega_1(\alpha_1)
- \omega_1(\lpb\alpha_1\com\alpha_2\rpb),
\eqlabel{2cobound}\\
(s\,\omega_2)(\alpha_1,\alpha_2,\alpha_3) &=&
\rho_{\alpha_1}\,\omega_2(\alpha_2,\alpha_3)
+ \rho_{\alpha_2}\,\omega_2(\alpha_3,\alpha_1)
+ \rho_{\alpha_3}\,\omega_2(\alpha_1,\alpha_2)\nonumber\\
&&\mbox{}
\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!
- \omega_2(\lpb\alpha_1\com\alpha_2\rpb,\alpha_3)
- \omega_2(\lpb\alpha_2\com\alpha_3\rpb,\alpha_1)
- \omega_2(\lpb\alpha_3\com\alpha_1\rpb,\alpha_2)\,.
\eqlabel{3cobound}
\end{eqnarray}
It is easy to verify that~$s\,\omega_n$ defines an $(n+1)$-cochain, and it is
straightforward (if tedious) to show that~\hbox{$s_{n+1}s_n = s^2=0$}. For
this to be true, The homomorphism property of~$\rho$ is crucial.
An~$n$-cocycle is an element~$\omega_n$ of~$C^n(\LieA,\vs)$ such
that~$s_n\,\omega_n = 0$. An~$n$-coboundary $\omega_{\rm cob}$ is an element
of~$C^n(\LieA,\vs)$ for which there exists an element~$\omega_{n-1}$
of~$C^{n-1}(\LieA,\vs)$ such that~\hbox{$\omega_{\rm cob} = s\omega_{n-1}$}.
Note that all coboundaries are cocycles, but not vice-versa.
Let
\[
Z^{n}_\rho(\LieA,\vs) \mathrel{\raisebox{.069ex}{:}\!\!=} \ker s_n
\]
be the vector subspace of all~$n$-cocycles,~\hbox{$Z^{n}_\rho(\LieA,\vs)
\subset C^n(\LieA,\vs)$}, and let
\[
B^{n}_\rho(\LieA,\vs) \mathrel{\raisebox{.069ex}{:}\!\!=} \range s_{n-1}
\]
be the vector subspace of all~$n$-coboundaries,~\hbox{$B^{n}_\rho(\LieA,\vs)
\subset C^n(\LieA,\vs)$}. The~$n$th cohomology group
of~$\LieA$ with coefficients in~$\vs$ is defined to be the quotient vector
space
\begin{equation}
H^n_\rho(\LieA,\vs) \mathrel{\raisebox{.069ex}{:}\!\!=} Z^{n}_\rho(\LieA,\vs)/B^{n}_\rho(\LieA,\vs).
\end{equation}
Note that for~\hbox{$n > \dim \LieA$}, we have~\hbox{$H^n_\rho(\LieA,\vs) =
Z^{n}_\rho(\LieA,\vs) = B^{n}_\rho(\LieA,\vs) = 0$}.
\section{Discussion}
\seclabel{conclusion}
\ifmypprint
{\small
\begin{verbatim}
$Id: sec-conclusion.tex,v 2.1 1998/11/13 08:37:51 jeanluc Exp $
\end{verbatim}
}
\fi
Using the tools of Lie algebra cohomology, we have classified low-order
extensions. We found that there were only a few normal forms for the
extensions, and that they involved no free parameters. This is not expected
to carry over to higher orders~($n>4$). The classification includes the
Leibniz extension, which is the maximal extension. One of the normal forms is
the bracket appropriate to compressible reduced
MHD~\cite{Hazeltine1987,Hazeltine1985b}.
We then developed techniques for finding the Casimir invariants of
Lie--Poisson brackets formed from Lie algebra extensions. We introduced the
concept of coextension, which allows one to explicitly write down the solution
of the Casimirs. The coextension for the Leibniz extension can be found for
arbitrary order, so that we were able obtain the corresponding Casimirs in
general.
It would be interesting to generalize the classification scheme presented here
to a completely general form of extension
bracket~\cite{Morrison1980a,Nore1997}. Certainly the type of coordinate
transformations allowed would be more limited, and perhaps one cannot go any
futher than cohomology theory allows.
The interpretation of the Casimir invariants can be pushed further, both in a
mathematical and a physical sense. Mathematically, a precise geometrical
relation between cocycles and the form of the Casimirs could be
formulated. The cocycle and Casimirs should yield information about the
holonomy of the system. For this one must study extensions in the framework of
their principal bundle description~\cite{Azcarraga}. Physically we would like
to attach a more precise physical meaning to these conserved quantities. The
invariants associated with simultaneous eigenvectors can be regarded as
constraining the associated field variable to move with the fluid
elements~\cite{Morrison1987}. The field variable can also be interpreted as
partially labeling a fluid element. Some attempt has been made at formulating
the Casimir invariants of brackets in such a
manner~\cite{Kuznetsov1980,Thiffeault1998}, and an interpretation of cocycles
in the context of dynamical accessiblity has been
offered~\cite{Thiffeault1998diss}.
Sufficient conditions for stability can be obtained via the energy-Casimir
method~\cite{Hazeltine1984,Holm1985,Morrison1987}, or the related technique of
dynamical accessibility~\cite{Morrison1998,Arnold1966b}. In both these case,
we can make use of the coextension to derive the stability conditions for
Lie--Poisson bracket extensions and a large class of
Hamiltonians~\cite{Thiffeault1998diss}.
\qcomment{Zeitlin truncations}
\section{Introduction}
\seclabel{intro}
\ifmypprint
{\small
\begin{verbatim}
$Id: sec-introduction.tex,v 2.1 1998/11/13 08:37:45 jeanluc Exp $
\end{verbatim}
}
\fi
This paper deals with the classification of Lie--Poisson brackets obtained
from extensions of Lie algebras. A large class of finite- and
infinite-dimensional dynamical equations admit a Hamiltonian formulation using
noncanonical brackets of the Lie--Poisson type. Finite-dimensional examples
include the Euler equations for the rigid body~\cite{Arnold} and the moment
reduction of the Kida vortex~\cite{Meacham1997} \inthesis{Also, and a
low-order model of atmospheric dynamics cite{Bokhove1996}.}, while
infinite-dimensional examples include the Vlasov
equation~\cite{Morrison1980b,Marsden1982b} and the Euler equation for the
ideal
fluid~\cite{Morrison1980a,Kuznetsov1980,Morrison1982,Olver1982,Marsden1983}.
Lie--Poisson brackets naturally define a Poisson structure (i.e., a symplectic
structure) on the dual of a Lie algebra. For the rigid body, the Lie algebra
is the one associated with the rotation group,~$\SOthree$, while for the Kida
vortex moment reduction the underlying group is~$\SOtwoone$. For the
two-dimensional ideal fluid, the relevant Lie algebra corresponds to the group
of volume-preserving diffeomorphisms on the fluid domain.
We will classify low-order bracket extensions and find their Casimir
invariants. An extension is simply a new Lie bracket, derived from a base
algebra (for example, $\SOthree$), and defined on~$n$-tuples of that algebra.
We are ruling out extensions where the brackets that appear are not of the
same form as that of the base algebra. We are thus omitting some
brackets~\cite{Morrison1980a,Nore1997}, but the brackets we are considering
are amenable to a general classification.
The method of extension yields interesting and physically relevant algebras.
Using this method we can describe finite-dimensional systems of several
variables and infinite-dimensional systems of several fields. For
finite-dimensional systems an example is the two vector model of the heavy
top~\cite{Holmes1983}. For infinite-dimensional systems there are models with
two~\cite{Morrison1984,Benjamin1984,McLachlan1997}, three
\cite{Morrison1984,Hazeltine1985,Kuvshinov1994}, and four~\cite{Hazeltine1987}
fields. \inthesis{Cite gyro-MHD Morrison1984b paper in thesis} Knowing the
bracket allows one to find the Casimir invariants of the
system~\cite{Trofimov,Kuroda1991,Hernandez1998}. These are quantities which
commute with every functional on the Poisson manifold, and thus are conserved
by the dynamics for any Hamiltonian. They are useful for analyzing the
constraints in the system~\cite{Thiffeault1998} and for establishing stability
criteria~\cite{Hazeltine1984,Holm1985,Morrison1986,Morrison1987,Morrison1998}.
The outline of this paper is as follows. In \secref{LiePoisson}, we review the
general theory behind Lie--Poisson brackets. We give examples of physical
systems of Lie--Poisson type, both finite and infinite-dimensional. We
introduce the concept of Lie algebra extensions and derive some of their basic
properties. \secref{cohoext} is devoted to the more abstract treatment of
extensions through the theory of Lie algebra
cohomology~\cite{Chevalley1948,Knapp,Azcarraga}. We define some terminology
and special extensions such as the semidirect sum and the Leibniz
extension. In \secref{classext}, we use the cohomology techniques to treat the
specific type of extension with which we are concerned, brackets
over~$n$-tuples. We give an explicit classification of low-order
extensions. By classifying we mean reducing---through coordinate changes---all
possible brackets to independent normal forms. We find that the normal forms
are relatively few and involve no free parameters---at least for low-order
extensions. In \secref{casinv}, we turn to the problem of finding the Casimir
invariants of the brackets, those functionals that commute with every other
functional in the algebra. We derive some general techniques for doing so that
apply to extensions of any order. Some explicit examples are derived,
including the Casimir invariants of a particular model of magnetohydrodynamics
(MHD). These are also given a physical interpretation. A formula for the
invariants of Leibniz extenions of any order is also derived. Then in
\secref{caslowdim} we use the classification of \secref{classext} to derive
the Casimir invariants for low-order extensions. Finally in
\secref{conclusion} we offer some concluding remarks and discuss future
directions.
\section{Extension of a Lie Algebra}
\seclabel{cohoext}
\ifmypprint
{\small
\begin{verbatim}
$Id: sec-lieaextension.tex,v 2.1 1998/11/13 08:36:57 jeanluc Exp $
\end{verbatim}
}
\fi
In this section we review the theory of Lie algebra cohomology and its
application to extensions. This is useful for shedding light on the methods
used in \secref{classext} for classifying extensions. However, the
mathematical details presented in this section can be skipped without
seriously compromising the flavor of the classification scheme of
\secref{classext}.
\input{sec-cohoalgebra}
\subsection{Application of Cohomology to Extensions}
\seclabel{extension}
In \secref{theproblem} we gave a definition of extension that is specific to
our problem. We will now define extensions in a more abstract manner. We
then show how the cohomology of Lie algebras of \secref{cohoalgebra} is
related to the problem of classifying extensions. In \secref{classext} we
will return to the more concrete concept of extension, of the form given in
\secref{theproblem}.
Let~$f_i: \LieA_i \rightarrow
\LieA_{i+1}$ be a collection of Lie algebra homomorphisms,
\inthesis{define homomorphism.}
\[
\xymatrix@M=4pt{
\dots \ar[r] & {\LieA_i} \ar[r]^-{f_i} &
{\LieA_{i+1}} \ar[r]^-{f_{i+1}} & {\LieA_{i+2}} \ar[r] & \dots\
} .
\]
The sequence~$f_i$ is called an exact sequence of Lie algebra homomorphisms if
\[
\range f_i = \ker f_{i+1}\,.
\]
Let~$\LieA$, $\LieAx$, and~$\LieAxb$ be Lie algebras. The algebra~$\LieAx$ is
said to be an~\emph{extension} of~$\LieA$ by~$\LieAxb$ if there is a short
exact sequence of Lie algebra homomorphisms
\begin{equation}\xymatrix@M=4pt{
0 \ar[r] & {\LieAxb} \ar[r]^{i} & {\LieAx} \ar[r]^{\pi}
& {\LieA} \ar[r] & 0
}.
\eqlabel{extdef}
\end{equation}
The homomorphism~$i$ is an insertion (injection), and~$\pi$ is a projection
(surjection). We shall distinguish brackets in the different algebras by
appropriate subscripts. We also define~\hbox{$\tau:\LieA\rightarrow\LieAx$}
to be a linear mapping such that~\hbox{$\pi\circ\tau = 1_{|\LieA}$} (the
identity mapping in~$\LieA$). Note that~$\tau$ is not unique, since the
kernel of~$\pi$ is not trivial. Let~$\beta \in
\LieAx$,~$\eta \in \LieAxb$; then
\[
\pi{\lpb\beta\com i\,\eta\rpb}_\LieAx
= {\lpb\pi\,\beta\com\pi\,i\,\eta\rpb}_\LieA
= 0,
\]
using the homomorphism property of~$\pi$ and~\hbox{$\pi\circ i=0$}, a
consequence of the exactness of the sequence. Thus~\hbox{${\lpb\beta\com
i\,\eta\rpb}_\LieAx \in
\ker \pi = \range i$}, and~$i\,\LieAxb$ is an ideal in~$\LieAx$
since~\hbox{$\lpb\beta\com i\eta\rpb \in i \LieAxb$}.
Hence, we can form the quotient algebra~$\LieAx/\LieAxb$, with equivalence
classes denoted by~\hbox{$\beta + \LieAxb$}. By exactness~\hbox{$\pi(\beta +
\LieAxb) = \pi\,\beta$}, so~$\LieA$ is isomorphic to~$\LieAx/\LieAxb$ and we
write~\hbox{$\LieA = \LieAx/\LieAxb$}.
Though~$i\,\LieAxb$ is a subalgebra of~$\LieAx$,~$\tau\,\LieA$ is not
necessarily a subalgebra of~$\LieAx$, for in general
\[
{\lpb\tau\,\alpha\com\tau\,\beta\rpb}_\LieAx
\ne \tau\,{\lpb\alpha\com\beta\rpb}_\LieA,
\]
for~\hbox{$\alpha,\beta \in \LieA$}; that is,~$\tau$ is not necessarily a
homomorphism. The classification problem essentially resides in the
determination of how much~$\tau$ differs from a homomorphism. The cohomology
machinery of \secref{cohoalgebra} is the key to quantifying this difference,
and we proceed to show this.
To this end, we use the algebra~$\LieAxb$ as the vector space~$\vs$ of
\secref{cohoalgebra}, so that~$\LieAxb$ will be a left $\LieA$-module. We
define the left action as
\begin{equation}
\rho_\alpha\,\eta \mathrel{\raisebox{.069ex}{:}\!\!=}
i^{-1}{\lpb \tau\,\alpha\com i\,\eta\rpb}_\LieAx
\eqlabel{rhodef}
\end{equation}
for~\hbox{$\alpha \in \LieA$} and~\hbox{$\eta \in \LieAxb$}. For~$\LieAxb$ to
be a left~$\LieA$-module, we need~$\rho$ to be a homomorphism, i.e.,~$\rho$
must satisfy~\eqref{rhomo}. Therefore consider
\begin{eqnarray*}
{\lpb\rho_\alpha\com\rho_\beta\rpb}\,\eta &=&
(\rho_\alpha\rho_\beta - \rho_\beta\rho_\alpha)\,\eta
\nonumber\\
&=& \rho_\alpha\,i^{-1}{\lpb\tau\,\beta\com i\,\eta\rpb}_\LieAx
- \rho_\beta\,i^{-1}{\lpb\tau\,\alpha\com i\,\eta\rpb}_\LieAx
\nonumber\\
&=& i^{-1}{\lpb\tau\,\alpha\com
{\lpb\tau\,\beta\com i\,\eta\rpb}_\LieAx\rpb}_\LieAx
- i^{-1}{\lpb\tau\,\beta\com
{\lpb\tau\,\alpha\com i\,\eta\rpb}_\LieAx\rpb}_\LieAx,
\end{eqnarray*}
which upon using the Jacobi identity in~$\LieAx$ becomes
\begin{eqnarray}
{\lpb\rho_\alpha\com\rho_\beta\rpb}\,\eta &=&
i^{-1}{\lpb{\lpb\tau\,\alpha\com \tau\,\beta\rpb}_\LieAx\com
i\,\eta\rpb}_\LieAx\nonumber\\
&=& i^{-1}{\lpb\tau\,{\lpb\alpha\com \beta\rpb}_\LieA\com
i\,\eta\rpb}_\LieAx
+ i^{-1}{\lpb\l(
{\lpb\tau\,\alpha\com \tau\,\beta\rpb}_\LieAx
- \tau\,{\lpb\alpha\com \beta\rpb}_\LieA\r)\com
i\,\eta\rpb}_\LieAx\nonumber\\
&=& \rho_{{\lpb\alpha\com\beta\rpb}_\LieA}\,\eta
+ i^{-1}{\lpb\l(
{\lpb\tau\,\alpha\com \tau\,\beta\rpb}_\LieAx
- \tau\,{\lpb\alpha\com \beta\rpb}_\LieA\r)\com
i\,\eta\rpb}_\LieAx.
\eqlabel{mismatch}
\end{eqnarray}
By applying~$\pi$ on the expression in parentheses of the last term
of~\eqref{mismatch}, we see that it vanishes and so is in~$\ker \pi$, and by
exactness it is also in~$i\,\LieAxb$. Thus the $\LieAx$ commutator above
involves two elements of~$i\,\LieAxb$. We define~\hbox{$\omega:
\LieA \times \LieA \rightarrow \LieAxb$} by
\begin{equation}
\omega(\alpha,\beta) \mathrel{\raisebox{.069ex}{:}\!\!=} i^{-1}\l(
{\lpb\tau\,\alpha\com\tau\,\beta\rpb}_\LieAx
- \tau\,{\lpb\alpha\com\beta\rpb}_\LieA\r).
\eqlabel{cocycle}
\end{equation}
The mapping~$i^{-1}$ is well defined on~$i\,\LieAxb$. Equation
\eqref{mismatch} becomes
\begin{equation}
{\lpb\rho_\alpha\com\rho_\beta\rpb}\,\eta =
\rho_{{\lpb\alpha\com\beta\rpb}_\LieA}\,\eta + {\lpb
\omega(\alpha,\beta) \com
\eta\rpb}_\LieAxb.
\eqlabel{rhohomo}
\end{equation}
Therefore,~$\rho$ satisfies the homomorphism property if either of the
following is true:
\begin{enumerate}
\item[(i)] $\LieAxb$ is Abelian,
\item[(ii)] $\tau$ is a homomorphism
\end{enumerate}
Condition~(i) implies~$\lpb\com\rpb_\LieAxb=0$, while condition~(ii)
means
\[
{\lpb\tau\,\alpha\com \tau\,\beta\rpb}_\LieAx =
\tau\,{\lpb\alpha\com\beta\rpb}_\LieA,
\]
which implies~$\omega\equiv 0$. If either of these conditions is
satisfied,~$\LieAxb$ with the action~$\rho$ is a left~$\LieA$-module. We
treat these two cases separately in \secreftwo{abelianext}{semidext},
respectively.
\subsection{Extension by an Abelian Lie Algebra}
\seclabel{abelianext}
In this section we assume that the homomorphism condition~(i) at the end of
\secref{extension} is met. Therefore~$\LieAxb$ is a left~$\LieA$-module, and
we can define~$\LieAxb$-valued cochains on~$\LieA$. In particular,~$\omega$
defined by~\eqref{cocycle} is a 2-cochain,~\hbox{$\omega
\in C^2(\LieA,\LieAxb)$}, that measures the ``failure''
of~$\tau$ to be a homomorphism. We now show, moreover, that~$\omega$ is a
2-cocycle,~\hbox{$\omega \in Z^2_\rho(\LieA,\LieAxb)$}. By
using~\eqref{3cobound},
\begin{eqnarray*}
\!\!\!\!\!\!\!
(s\,\omega)(\alpha,\beta,\gamma) &=&
\rho_{\alpha}\,\omega(\beta,\gamma)
+ \rho_{\beta}\,\omega(\gamma,\alpha)
+ \rho_{\gamma}\,\omega(\alpha,\beta)\nonumber\\
&&\mbox{}- \omega({\lpb\alpha\com\beta\rpb}_\LieA,\gamma)
- \omega({\lpb\beta\com\gamma\rpb}_\LieA,\alpha)
- \omega({\lpb\gamma\com\alpha\rpb}_\LieA,\beta)\,,
\nonumber\\
&=& i^{-1}\l({\lpb\tau\,\alpha\com
{\lpb\tau\,\beta\com\tau\,\gamma\rpb}_\LieAx\rpb}_\LieAx
+ \mathrm{cyc.\ perm.}\r)\nonumber\\
&&\mbox{} + i^{-1}\tau\l({\lpb{\lpb
\alpha\com\beta\rpb}_\LieA\com\gamma\rpb}_\LieA
+ \mathrm{cyc.\ perm.}\r) = 0.
\end{eqnarray*}
\inthesis{Include all commented out steps.}
The first parenthesis vanishes by the Jacobi identity in~$\LieAx$, the second
by the Jacobi identity in~$\LieA$, and the other terms were canceled in
pairs. Hence,~$\omega$ is a 2-cocycle.
Two extensions~$\LieAx$ and~$\LieAx'$ are equivalent if there exists a Lie
algebra isomorphism~$\sigma$ such that the diagram
\begin{equation}
\xymatrix@M=4pt{
& & {\LieAx} \ar[dr]^{\pi} \ar[dd]^{\sigma} & & \\
0 \ar[r] & {\LieAxb} \ar[ur]^i \ar[dr]_{i'} & & {\LieA} \ar[r] & 0 \\
& & {\LieAx'} \ar[ur]_{\pi'} & &
}
\eqlabel{equivext}
\end{equation}
is commutative, that is if~\hbox{$\sigma\circ i=i'$ and~$\pi =
\pi'\circ\sigma$}.
There will be an injection~$\tau$ associated with~$\pi$ and a~$\tau'$
associated with~$\pi'$, such that~\hbox{$\pi\circ\tau = 1_{|\LieA} =
\pi'\circ\tau'$.} The linear map~\hbox{$\nu = \sigma^{-1}\tau' - \tau$} must
be from~$\LieA$ to~$i\,\LieAxb$, so~\hbox{$i^{-1}\nu \in C^1(\LieA,\LieAxb)$}.
Consider~$\rho$ and~$\rho'$ respectively defined using~$\tau,i$ and~$\tau',i'$
by~\eqref{rhodef}. Then
\begin{eqnarray}
(\rho_\alpha - {\rho'}_\alpha)\,\eta
&=& i^{-1}{\lpb\tau\,\alpha\com i\,\eta\rpb}_\LieAx -
{i'}^{-1}{\lpb\tau'\,\alpha\com i'\,\eta\rpb}_{\LieAx'}\ ,
\nonumber\\
&=& i^{-1}{\lpb\tau\,\alpha\com i\,\eta\rpb}_\LieAx -
i^{-1}{\lpb(\nu + \tau)\,\alpha\com
i\,\eta\rpb}_\LieAx\ ,\nonumber\\
&=& -i^{-1}{\lpb\nu\,\alpha\com i\,\eta\rpb}_\LieAx = 0,
\eqlabel{samerho}
\end{eqnarray}
\inthesis{Include all commented out steps.}
since~$\LieAxb$ is Abelian. Hence $\tau$ and~$\tau'$ define the same~$\rho$.
Now consider the 2-cocycles~$\omega$ and~$\omega'$ defined from~$\tau$
and~$\tau'$ by~\eqref{cocycle}. We have
\begin{eqnarray*}
\!\!\!\!\!\!\!\!\!
\omega'(\alpha,\beta) - \omega(\alpha,\beta)
&=& {i'}^{-1}
\l({\lpb\tau'\,\alpha\com\tau'\,\beta\rpb}_{\LieAx'}
- \tau'\,{\lpb\alpha\com\beta\rpb}_\LieA\r)\nonumber\\
&&\mbox{} - i^{-1}\l({\lpb\tau\,\alpha\com\tau\,\beta\rpb}_\LieAx
- \tau\,{\lpb\alpha\com\beta\rpb}_\LieA\r),\nonumber\\
&=& i^{-1}\l({\lpb\tau\,\alpha\com\nu\,\beta\rpb}_\LieAx
+ {\lpb\nu\,\alpha\com\tau\,\beta\rpb}_\LieAx
- \nu\,{\lpb\alpha\com\beta\rpb}_\LieA\r),\nonumber\\
&=& \rho_\alpha\,(i^{-1}\nu\,\beta) - \rho_\beta\,(i^{-1}\nu\,\alpha)
- i^{-1}\nu\,{\lpb\alpha\com\beta\rpb}_\LieA.
\end{eqnarray*}
\inthesis{Include all commented out steps.}
Comparing this with~\eqref{2cobound}, we see that
\begin{equation}
\omega' - \omega = s\,(i^{-1}\nu),
\eqlabel{cobdiff}
\end{equation}
so~$\omega$ and~$\omega'$ differ by a coboundary. Hence they represent the
same element in~$H^2_\rho(\LieA,\LieAxb)$. Equivalent extensions uniquely
define an element of the second cohomology group~$H^2_\rho(\LieA,\LieAxb)$.
Note that this is true in particular for~$\LieAx=\LieAx'$,~$\sigma=1$, so that
the element of~$H^2_\rho(\LieA,\LieAxb)$ is independent of the choice
of~$\tau$.
We are now ready to write down explicitly the bracket in~$\LieAx$. We can
represent an element~\hbox{$\alpha \in \LieAx$} as a two-tuple:~\hbox{$\alpha
= (\alpha_1, \alpha_2)$} where~\hbox{$\alpha_1 \in \LieA$} and~\hbox{$\alpha_2
\in \LieAxb$} (\hbox{$\LieAx = \LieA \oplus \LieAxb$} as a vector space). The
injection~$i$ is then~\hbox{$i\,\alpha_2 = (0,\alpha_2)$}, the
projection~$\pi$ is~\hbox{$\pi\,(\alpha_1,\alpha_2) = \alpha_1$}, and since
the extension is independent of the choice of~$\tau$ we
take~\hbox{$\tau\,\alpha_1 = (\alpha_1,0)$}. By linearity
\begin{eqnarray*}
{\lpb\alpha,\beta\rpb}_\LieAx &=&
{\lpb(\alpha_1,0),(\beta_1,0)\rpb}_\LieAx
+ {\lpb(0,\alpha_2),(0,\beta_2)\rpb}_\LieAx\nonumber\\
&&\mbox{} + {\lpb(\alpha_1,0),(0,\beta_2)\rpb}_\LieAx
+ {\lpb(0,\alpha_2),(\beta_1,0)\rpb}_\LieAx.
\end{eqnarray*}
We know that~${\lpb(0,\alpha_2),(0,\beta_2)\rpb}_\LieAx = 0$ since~$\LieAxb$
is Abelian. By definition of the cocycle~$\omega$, Eq.~\eqref{cocycle}, we
have
\begin{eqnarray*}
{\lpb(\alpha_1,0),(\beta_1,0)\rpb}_\LieAx &=&
{\lpb\tau\,\alpha_1\com\tau\,\beta_1\rpb}_\LieAx \nonumber\\
&=& i\,\omega(\alpha_1,\beta_1) +
\tau\,{\lpb\alpha_1\com\beta_1\rpb}_\LieA\nonumber\\
&=& ({\lpb\alpha_1\com\beta_1\rpb}_\LieA\,,\,\omega(\alpha_1,\beta_1)).
\end{eqnarray*}
Finally, by the definition of~$\rho$, Eq.~\eqref{rhodef},
\[
{\lpb(\alpha_1,0),(0,\beta_2)\rpb}_\LieAx
= {\lpb\tau\,\alpha_1,i\,\beta_2\rpb}_\LieAx
= \rho_{\alpha_1}\,\beta_2,
\]
and similarly for~${\lpb(0,\alpha_2),(\beta_1,0)\rpb}_\LieAx$, with opposite
sign. So the bracket is
\begin{equation}
{\lpb\alpha,\beta\rpb}_\LieAx =
\Bigl({\lpb\alpha_1\com\beta_1\rpb}_\LieA\, , \,
\rho_{\alpha_1}\,\beta_2 - \rho_{\beta_1}\,\alpha_2
+ \omega(\alpha_1,\beta_1)\Bigr).
\eqlabel{abelianbracket}
\end{equation}
As a check we work out the Jacobi identity in~$\LieAx$:
\begin{eqnarray*}
\!\!\!\!\!\!\!\!
{\lpb\alpha\com{\lpb\beta\com\gamma\rpb}_\LieAx\rpb}_\LieAx &=&
\l({\lpb\alpha_1\com{\lpb\beta\com\gamma\rpb}_1\rpb}_\LieA
\,,\, \rho_{\alpha_1}\,{\lpb\beta\com\gamma\rpb}_2
- \rho_{{\lpb\beta\com\gamma\rpb}_1}\,\alpha_2
+ \omega(\alpha_1,{\lpb\beta\com\gamma\rpb}_1)\r)\nonumber\\
&=& \Bigl({\lpb\alpha_1\com{\lpb\beta_1
\com\gamma_1\rpb}_\LieA\rpb}_\LieA
\,,\, \rho_{\alpha_1}(\rho_{\beta_1}\,\gamma_2
- \rho_{\gamma_1}\,\beta_2 + \omega(\beta_1,\gamma_1))
\nonumber\\
&&\mbox{}\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
- \rho_{{\lpb\beta_1\com\gamma_1\rpb}_\LieA}\,\alpha_2
+ \omega(\alpha_1,{\lpb\beta_1\com\gamma_1\rpb}_\LieA)\Bigr).
\end{eqnarray*}
Upon adding permutations, the first component will vanish by the Jacobi
identity in~$\LieA$. We are left with
\begin{eqnarray*}
{\lpb\alpha\com{\lpb\beta\com\gamma\rpb}_\LieAx\rpb}_\LieAx
+ \mathrm{cyc.\ perm.} &=& \Bigl(0\,,\,
\l(\rho_{\alpha_1}\rho_{\beta_1} - \rho_{\beta_1}\rho_{\alpha_1}
- \rho_{{\lpb\alpha_1\com\beta_1\rpb}_\LieA}\r)
\gamma_2\nonumber\\
&&\mbox{} \!\!\!\!\!\!\!\!\!\!\!\!\!\!
+ \rho_{\alpha_1}\,\omega(\beta_1,\gamma_1)
- \omega({\lpb\alpha_1\com\beta_1\rpb}_\LieA,\gamma_1)\Bigr)
+ \mathrm{cyc.\ perm.},
\end{eqnarray*}
which vanishes by the the homomorphism property of~$\rho$ and the fact
that~$\omega$ is a 2-cocycle, Eq.~\eqref{3cobound}.
Equation~\eqref{abelianbracket} is the most general form of the Lie bracket
for extension by an Abelian Lie algebra. It turns out that the theory of
extension by a non-Abelian algebra can be reduced to the study of extension by
the center of~$\LieAxb$, which is Abelian~\cite{Azcarraga}. We will not need
this fact here, as the only extensions by non-Abelian algebras we will deal
with are of the simpler type of \secref{semidext}.
We have thus shown that equivalent extensions are enumerated by the second
cohomology group~$H^2_\rho(\LieA,\LieAxb)$. The coordinate
transformation~$\sigma$ used in~\eqref{equivext} to define equivalence of
extensions preserves the form of~$\LieA$ and~$\LieAxb$ as subsets of~$\LieAx$.
However, we have the freedom to choose coordinate transformations which do
transform these subsets. All we require is that the isomorphism~$\sigma$
between~$\LieAx$ and~$\LieAx'$ be a Lie algebra isomorphism. We can represent
this by the diagram
\begin{equation}\xymatrix@M=4pt{
0 \ar[r] & {\LieAxb} \ar[r]^{i} & {\LieAx} \ar[r]^{\pi}
\ar[d]^\sigma
& {\LieA} \ar[r] & 0 \\
0 \ar[r] & {\LieAxb'} \ar[r]^{i} & {\LieAx'} \ar[r]^{\pi}
& {\LieA'} \ar[r] & 0. \\
}
\eqlabel{equivext2}
\end{equation}
The primed and the unprimed extensions are not equivalent, but they are
isomorphic~\cite{Weiss}. Cohomology for us is not the whole story, since we
are interested in isomorphic extensions, but it will guide our classification
scheme. We discuss this point further in \secref{furthertrans}.
\subsection{Semidirect and Direct Extensions}
\seclabel{semidext}
Assume now that~$\omega$ defined by~\eqref{cocycle} is a coboundary.
By~\eqref{cobdiff} there exists an equivalent extension with~\hbox{$\omega
\equiv 0$}. For that equivalent extension,~$\tau$ is a homomorphism and
condition~(ii) at the end of \secref{extension} is satisfied. Thus the
sequence
\[\xymatrix@M=4pt{
{\LieAx} & {\LieA} \ar[l]_{\tau} & 0 \ar[l]
}
\]
is an exact sequence of Lie algebra homomorphisms, as well as the sequence
given by~\eqref{extdef}. We then say that the extension is a semidirect
extension (or a semidirect sum of algebras) by analogy with the group
case. More generally, we say that~$\LieAx$ splits if it is isomorphic to a
semidirect sum, which corresponds to~$\omega$ being a coboundary, not
necessarily zero. If~$\LieAxb$ is not Abelian, then~\eqref{samerho} is not
satisfied and two equivalent extensions (or two different choices of~$\tau$)
do not necessarily lead to the same~$\rho$.
Representing elements of~$\LieAx$ as 2-tuples, as in \secref{abelianext}, we
can derive the bracket in~$\LieAx$ for a semidirect sum,
\inthesis{Explicitely show the differences, that is, vanishing of cocycle and
nonabelianiticity.}
\begin{equation}
{\lpb\alpha,\beta\rpb}_\LieAx =
\Bigl({\lpb\alpha_1\com\beta_1\rpb}_\LieA\, , \,
\rho_{\alpha_1}\,\beta_2 - \rho_{\beta_1}\,\alpha_2
+ {\lpb\alpha_2\com\beta_2\rpb}_\LieAxb\Bigr),
\eqlabel{sdbracket}
\end{equation}
where we have not assumed~$\LieAxb$ Abelian. Verifying Jacobi
for~\eqref{sdbracket} we find the~$\rho$ must also satisfy
\[
\rho_{\alpha_1}\,{\lpb\beta_2\com\gamma_2\rpb}_\LieAxb
= {\lpb\rho_{\alpha_1}\,\beta_2\com\gamma_2\rpb}_\LieAxb
+ {\lpb\beta_2\com\rho_{\alpha_1}\,\gamma_2\rpb}_\LieAxb\, ,
\]
which is trivially satisfied if~$\LieAxb$ is Abelian, but in general this
condition states that~$\rho_\alpha$ is a derivation on~$\LieAxb$.
\inthesis{Relation to~$H^1_\rho$.}
Now consider the case where~$i^{-1}$ is a homomorphism and~\hbox{$\ker i^{-1}
= \range \tau$}. Then the sequence
\[\xymatrix@M=4pt{
0 \ar@_{->}@<-.3ex>[r]
& {\LieAxb} \ar@_{->}@<-.3ex>[l] \ar@_{->}@<-.3ex>[r]_{i}
& {\LieAx} \ar@_{->}@<-.3ex>[l]_{i^{-1}} \ar@_{->}@<-.3ex>[r]_{\pi}
& {\LieA} \ar@_{->}@<-.3ex>[l]_{\tau} \ar@_{->}@<-.3ex>[r]
& 0 \ar@_{->}@<-.3ex>[l]
}
\]
is exact in both directions and, hence, both~$i$ and~\hbox{$\pi=\tau^{-1}$}
are bijections. The action of~$\LieA$ on~$\LieAxb$ is
\[
\rho_{\alpha}\,\eta = i^{-1}{\lpb\tau\,\alpha\com i\eta\rpb}_\LieAx
= {\lpb i^{-1}\tau\,\alpha\com \eta\rpb}_\LieAxb = 0
\]
since by exactness~\hbox{$i^{-1}\circ\tau = 0$}. This is called a direct
sum. Note that in this case the role of~$\LieA$ and~$\LieAxb$ is
interchangeable and they are both ideals in~$\LieAx$. The bracket in~$\LieAx$
is easily obtained from~\eqref{sdbracket} by letting~\hbox{$\rho=0$},
\begin{equation}
{\lpb\alpha,\beta\rpb}_\LieAx =
\Bigl({\lpb\alpha_1\com\beta_1\rpb}_\LieA\com
{\lpb\alpha_2\com\beta_2\rpb}_\LieAxb\Bigr).
\eqlabel{dbracket}
\end{equation}
Semidirect and direct extensions play an important role in physics. A simple
example of a semidirect extension structure is when $\LieA$ is the Lie
algebra~$\sothree$ associated with the rotation group $\SOthree$ and $\LieAxb$
is $\reals^3$. Their semidirect sum is the algebra of the six parameter
Euclidean group of rotations and translations. That algebra can be used in a
Lie--Poisson bracket to describe the dynamics of the heavy top (see for
example~\cite{Holmes1983,Marsden1984}). We have already discussed the
semidirect sum in \secref{lowbetaRMHD}. The bracket \eqref{RMHDbrak} is a
semidirect sum, with~$\LieA$ the algebra of the group of volume-preserving
diffeomorphisms and~$\LieAxb$ the Abelian Lie algebra of functions
on~$\reals^2$. The action is just the adjoint action~\hbox{$\rho_\alpha\,v
\mathrel{\raisebox{.069ex}{:}\!\!=} \lpb\alpha\com v\rpb$} obtained by identifying~$\LieA$ and~$\LieAxb$.
A Lie--Poisson bracket built from a direct extension is just a sum of the
separate brackets. The interaction between the variables can only come from
the Hamiltonian or from constitutive equations. For example in the baroclinic
instability model of two superimposed two fluid layers with different
potential vorticities the two layers are coupled through the potential
vorticity relation~\cite{McLachlan1997}.
\section{Lie--Poisson Brackets}
\seclabel{LiePoisson}
\ifmypprint
{\small
\begin{verbatim}
$Id: sec-lpbracket.tex,v 2.1 1998/11/13 08:30:06 jeanluc Exp $
\end{verbatim}
}
\fi
Lie--Poisson brackets define a natural Poisson structure on duals of Lie
algebras. Physically, they often arise in the \emph{reduction} of a system.
For our purposes, a reduction is a mapping of the dynamical variables of a
system to a smaller set of variables, such that the transformed Hamiltonian
and bracket depend only on the smaller set of variables. (For a more detailed
mathematical treatment, see for
example~%
\cite{Marsden1974,AbrahamMarsden,Marsden1982,Guillemin,MarsdenRatiu}\@.)
\inthesis{For a brief overview mention cite{Audin}'s book.} The
simplest example of a reduction is the case in which a cyclic variable is
eliminated, but more generally a reduction exists as a consequence of an
underlying symmetry of the system. For instance, the Lie--Poisson bracket for
the rigid body is obtained from a reduction of the canonical Euler angle
description using the rotational symmetry of the system~\cite{Holmes1983}.
The Euler equation for the two-dimensional ideal fluid is obtained from a
reduction of the Lagrangian description of the fluid, which has a relabeling
symmetry~\cite{Morrison1998,Newcomb1967,Bretherton1970,Padhye1996a}.
Here we shall take a more abstract viewpoint: we do not assume that the
Lie--Poisson bracket is obtained from a reduction, though it is always
possible to do so by the method of Clebsch variables~\cite{Morrison1998}.
Rather we proceed directly from a given Lie algebra to build a Lie--Poisson
bracket. The choice of algebra can be guided by the symmetries of the system.
After deriving the basic theory behind Lie--Poisson brackets in
\secref{lpbasic}, we will show some explicit examples in \secref{lpexample}.
We then describe general Lie algebra extensions in \secref{theproblem}.
\subsection{Lie--Poisson Brackets on Duals of Lie Algebras}
\seclabel{lpbasic}
We begin by taking the Lie algebra~$\LieA$ associated with some Lie group.
The Lie group might be chosen to reflect the symmetries of a physical
system. There will be a Lie bracket~\hbox{$\lpb \com \rpb : \LieA
\times \LieA \rightarrow \LieA$} associated with~$\LieA$. Consider the
dual~$\LieA^*$ of~$\LieA$ with respect to the
pairing\index{pairing}~\hbox{$\lang\ \com\
\rang: \LieA^*
\times \LieA \rightarrow \reals$}. Then for real-valued functionals~$F$
and~$G$, that is,~\hbox{$F,G:\LieA^*\rightarrow\reals$}, and~\hbox{$\fv \in
\LieA^*$}, we can define
\begin{equation}
{\lPB F\com G \rPB}_\pm(\fv) = \pm\lang\fv\com {\lpb \frac{\fd
F}{\fd\fv}\com\frac{\fd G}{\fd\fv} \rpb}\rang . \eqlabel{LPB}
\end{equation}
The sign choice comes from whether we are considering right invariant ($+$) or
left invariant ($-$) functions on the cotangent bundle of the Lie
group~\cite{MarsdenRatiu,Marsden1983}, but for our purposes we simply choose
the sign as needed. The functional derivative~$\fd F/\fd\fv$ is defined by
\begin{equation}
\fd F[\,\fv;\fd\fv\,] \mathrel{\raisebox{.069ex}{:}\!\!=}
{\l.\frac{d}{d\epsilon}F[\fv + \epsilon\,
\fd\fv]\r|}_{\epsilon=0}
\mathrel{=\!\!\raisebox{.069ex}{:}} \lang\fd\fv\com\frac{\fd F}{\fd \fv}\rang .
\eqlabel{funcder}
\end{equation}
We shall refer to the bracket~$\lpb\com\rpb$ as the inner bracket and to the
bracket~$\lPB\com\rPB$ as the Lie--Poisson bracket. The dual~$\LieA^*$
together with the Lie--Poisson bracket is a Poisson manifold; that is, the
bracket~$\lPB\com\rPB$ is a Lie algebra structure on real-valued functionals
that is a derivation in each of its arguments. For finite-dimensional groups,
Eq.~\eqref{LPB} was first written down by Lie~\cite{Lie} and was rediscovered
by Berezin~\cite{Berezin1967}; it is also closely related to work of
Arnold~\cite{Arnold1966a}, Kirillov~\cite{Kirillov1962},
Kostant~\cite{Kostant1966}, and Souriau~\cite{Souriau}.
The bracket in~$\LieA$ is the same as the adjoint action of~$\LieA$ on itself:
\hbox{$\lpb\alpha\com\beta\rpb = \ad_\alpha\,\beta$}, where~$\alpha$, $\beta
\in \LieA$. From this we define the coadjoint action~$\ad_\alpha^\dagger$
of~$\LieA$ on~$\LieA^*$ by
\comment{Following the convention of Arnold, p.321}
\begin{equation}
\lang \ad_\alpha^\dagger\,\fv\com \beta\,\rang \mathrel{\raisebox{.069ex}{:}\!\!=}
\lang \,\fv \com\,\ad_\alpha\, \beta\,\rang ,
\eqlabel{coadj}
\end{equation}
where~$\fv \in \LieA^*$. We also define the coadjoint
bracket~\hbox{$\lpb\com\rpb^{\dag}:\LieA
\times \LieA^* \rightarrow \LieA^*$}
by \hbox{$\lpb\alpha\com\fv\,\rpb^{\dag} \mathrel{\raisebox{.069ex}{:}\!\!=} \ad_\alpha^\dagger\,\fv$},
so that
\begin{equation}
\lang \lpb\alpha\com\fv\rpb^\dagger\com \beta\,\rang \mathrel{\raisebox{.069ex}{:}\!\!=}
\lang \,\fv \com\,\lpb\alpha \com \beta\rpb\,\rang ;
\eqlabel{cobracket}
\end{equation}
the bracket~${\lpb\com\rpb}^\dagger$ satisfies the identity
\[
\lang \lpb\alpha\com\fv\rpb^\dagger\com \beta\,\rang =
-\lang \lpb\beta\com\fv\rpb^\dagger\com \alpha\,\rang.
\]
Since the inner bracket is Lie, it satisfies the Jacobi identity, and
consequently the form given by~\eqref{LPB} for the Lie--Poisson bracket will
automatically satisfy the Jacobi identity~\cite[p.~614]{AMR}.
\inthesis{Prove this in thesis.}
Given a Hamiltonian~\hbox{$H:\LieA^*\rightarrow\reals$}, the equation of
motion for~$\fv \in \LieA^*$ is
\begin{eqnarray}
\dot\fv &=& \lPB\fv\com H\rPB
= \pm\lang\fv\com\lpb{\Delta}\com
\frac{\fd H}{\fd \fv}\rpb\rang
\nonumber\\
&=& \mp\lang\lpb \frac{\fd H}{\fd \fv}\com\,\fv\rpb^\dagger
\com\,\Delta\rang
= \mp\lpb \frac{\fd H}{\fd \fv}\com\,\fv\rpb^\dagger,
\eqlabel{motion}
\end{eqnarray}
where~$\Delta$ is a Kronecker or Dirac delta, or a combination of both for an
infinite-dimensional system of several fields.
\subsection{Examples of Lie--Poisson Systems}
\seclabel{lpexample}
We will say that a physical systems can be described by a given Lie--Poisson
bracket and Hamiltonian if its equations of motion can be written as
\eqref{motion}; the system is then said to be Hamiltonian of the Lie--Poisson
type. We give four examples: the first is finite-dimensional (the free rigid
body, \secref{rigidbody}) and the second infinite-dimensional (Euler's
equation for the ideal fluid, \secref{twodfluid}). The third and fourth
examples are also infinite-dimensional and serve to introduce the concept of
extension. They are low--beta reduced magnetohydrodynamics (MHD) in
\secref{lowbetaRMHD} and compressible reduced MHD in \secref{CRMHD}. These
last two examples are meant to illustrate the physical relevance of Lie
algebra extensions.
\subsubsection{The Free Rigid Body}
\seclabel{rigidbody}
The classic example of a Lie--Poisson bracket is obtained by taking
for~$\LieA$ the Lie algebra of the rotation group~$\SOthree$. If the~$\hat
{\bf e}_{(i)}$ denote a basis of~$\LieA = \sothree$, the Lie bracket is given
by
\[
\lpb \hat {\bf e}_{(i)}\com \hat {\bf e}_{(j)}\rpb
= c_{ij}^k\, \hat {\bf e}_{(k)}\,,
\]
where the~$c_{ij}^k = \varepsilon_{ijk}$ are the structure constants of the
algebra, in this case the totally antisymmetric symbol. Using as a pairing
the usual contraction between upper and lower indices, with~\eqref{LPB} we are
led to the Lie--Poisson bracket
\[
\lPB f\com g\rPB = -c_{ij}^k\, \ell_k\,\frac{\pd f}{\pd \ell_i}
\,\frac{\pd g}{\pd \ell_j}\,,
\]
where the three-vector~$\ell$ is in~$\LieA^*$, and we have chosen the minus
sign in \eqref{LPB}. The coadjoint bracket is obtained using~\eqref{coadj},
\[
{\lpb \beta\com \ell\rpb}_i^\dagger = -c_{ij}^k\,\beta^j\,\ell_k.
\]
If we use this coadjoint bracket and insert the Hamiltonian
\[
H = \half {(I^{-1})}^{ij}\,\ell_i\,\ell_j
\]
in~\eqref{motion} we obtain
\[
\dot \ell_m = \lPB \ell_m\com H\rPB
= c_{mj}^k\, {(I^{-1})}^{jp}\,\ell_k\,\ell_p\,.
\]
Notice how the moment of inertia tensor~$I$ plays the role of a metric---it
allows us to build a quadratic form (the Hamiltonian) from two elements
of~$\LieA^*$. If we take~$I = {\rm diag}(I_1,I_2,I_3)$, we recover Euler's
equations for the motion of the free rigid body
\[
\dot \ell_1 = \l(\frac{1}{I_2} - \frac{1}{I_3}\r)\,\ell_2\,\ell_3,
\]
and cyclic permutations of 1,2,3. The~$\ell_i$ are the angular momenta about
the axes and the~$I_i$ are the principal moments of inertia. This result is
naturally appealing because we expect the rigid body equations to be invariant
under the rotation group, hence the choice of~$\SOthree$ for~$G$.
\subsubsection{The Two-dimensional Ideal Fluid}
\seclabel{twodfluid}
Consider now an ideal fluid with the flow taking place over a two-dimensional
domain~$\fdomain$. Let~$\LieA$ be the infinite-dimensional Lie algebra
associated with the Lie group of volume-preserving diffeomorphisms
of~$\fdomain$. In two spatial dimensions this is the same as the group of
canonical transformations on~$\fdomain$. The bracket in~$\LieA$ is the
canonical bracket
\begin{equation}
\lpb a \com b \rpb = \frac{\pd a}{\pd x}\,\frac{\pd b}{\pd y}
- \frac{\pd b}{\pd x}\,\frac{\pd a}{\pd y}.
\eqlabel{canibrak}
\end{equation}
We formally identify~$\LieA$ and~$\LieA^*$ and use as the
pairing~$\lang\com\rang$ the usual integral over the fluid domain,
\[
\lang F\com G \rang = \int_\fdomain F(\xv)\,G(\xv)\d^2x,
\]
where~$\xv \mathrel{\raisebox{.069ex}{:}\!\!=} (x,y)$. For infinite-dimensional spaces, there are
functional analytic issues about whether we can make this identification, and
take~$\LieA^{**}=\LieA$. We will assume here that these relationships hold
formally. See Marsden and Weinstein~\cite{Marsden1982} for references on this
subject and Audin~\cite{Audin} for a treatment of the identification
of~$\LieA$ and~$\LieA^*$.
Assuming appropriate boundary conditions for simplicity, we
get~\hbox{${\lpb\com\rpb}^\dagger = -\lpb\com\rpb$} from
\eqref{cobracket}. (Otherwise the coadjoint bracket would involve extra
boundary terms.) Take the vorticity~$\vort$ as the field variable~$\fv$ and
write for the Hamiltonian
\[
H[\vort] = -\half\lang \vort\com \invlapl\,\vort\rang ,
\]
where
\[
(\invlapl\,\vort)(\xv)
\mathrel{\raisebox{.069ex}{:}\!\!=} \int_\fdomain K(\xv|\xv')\,\vort(\xv')\d^2x',
\]
and~$K$ is Green's function for the Laplacian. The Green's function plays the
role of a metric since it maps an element of~$\LieA^*$ (the vorticity~$\vort$)
into an element of~$\LieA$ to be used in the right slot of the pairing. This
relationship is only weak: the mapping~$K$ is not surjective, and thus the
metric cannot formally inverted (it is called
\emph{weakly nondegenerate}). When we have identified~$\LieA$ and~$\LieA^*$ we
shall often drop the comma in the pairing and write
\[
H[\vort] = -\half\lang \vort\,\streamf\rang =
\half\lang|\grad\streamf|^2\rang,
\]
where~$\vort=\lapl\streamf$ defines the streamfunction~$\streamf$. We work
out the evolution equation for~$\vort$ explicitly:
\begin{eqnarray*}
\dot\vort(\xv) &=& \lPB\vort\com H\rPB
= \int_\fdomain \vort(\xv')
\lpb\frac{\fd\vort(\xv)}{\fd\vort(\xv')}\com
\frac{\fd H}{\fd\vort(\xv')}\rpb\d^2x'\nonumber\\
&=& \int_\fdomain\vort(\xv')
\lpb\delta(\xv-\xv')\com {-\streamf(\xv')}\rpb\d^2x'
\nonumber\\
&=& \int_\fdomain\delta(\xv-\xv')
\lpb\vort(\xv')\com\streamf(\xv')\rpb\d^2x'
\nonumber\\
&=& \lpb\vort(\xv)\com\streamf(\xv)\rpb\, .
\end{eqnarray*}
This is Euler's equation for a two-dimensional ideal fluid. We could also
have written this result down directly from \eqref{motion}
using~\hbox{${\lpb\com\rpb}^\dagger = -{\lpb\com\rpb}$}.
\subsubsection{Low--beta Reduced MHD}
\seclabel{lowbetaRMHD}
This example will illustrate the concept of a Lie algebra extension, the main
topic of this paper. Essentially, the idea is to use an algebra of~$n$-tuples,
which we call an extension, to describe a physical system with more than one
dynamical variable. As in \secref{twodfluid} we consider a flow taking place
over a two-dimensional domain~$\fdomain$. The Lie algebra~$\LieA$ is again
taken to be that of volume preserving diffeomorphisms on~$\fdomain$, but now
we consider also the vector space~$\vs$ of real-valued functions on~$\fdomain$
(an Abelian Lie algebra under addition). The \emph{semidirect sum} of~$\LieA$
and~$\vs$ is a new Lie algebra whose elements are two-tuples~$(\alpha,v)$ with
a bracket defined by
\begin{equation}
\lpb(\alpha,v)\com(\beta,w)\rpb \mathrel{\raisebox{.069ex}{:}\!\!=}
\l(\lpb\alpha\com\beta\rpb\com
\lpb\alpha\com w\rpb - \lpb\beta\com v\rpb
\r),
\eqlabel{RMHDbrak}
\end{equation}
where~ $\alpha$ and~$\beta\in\LieA$, $v$ and~$w\in\vs$. This is a Lie
algebra, so we can use the prescription of \secref{lpbasic} to
build a Lie--Poisson bracket,
\[
{\lPB F\com G \rPB} = \int_\fdomain
\l\lgroup
\vort{\lpb \frac{\fd F}{\fd\vort}\com\frac{\fd G}{\fd\vort}
\rpb}
+ \magf \l({\lpb \frac{\fd F}{\fd\vort}\com\frac{\fd G}{\fd \magf}
\rpb}
- {\lpb \frac{\fd G}{\fd\vort}\com\frac{\fd F}{\fd \magf}
\rpb}\r) \r\rgroup\d^2x.
\]
Let $\vort=\lapl\elecp$, where $\elecp$ is the electric potential, $\magf$
is the magnetic flux, and $\ecurrent=\lapl\magf$ is the current. (We use
the same symbol for the electric field as for the streamfunction in
\secref{twodfluid} since they play a similar role.) The pairing used is a dot
product of the vectors followed by an integral over the fluid domain (again
identifying~$\LieA$ and~$\LieA^*$ as in \secref{twodfluid}). The Hamiltonian
\[
H[\vort;\psi] = \frac{1}{2}\int_\fdomain\,\l\lgroup
|\grad\elecp|^2+|\grad\magf|^2\r\rgroup\d^2x
\]
with the above bracket leads to the equations of motion
\begin{eqnarray*}
\dot\vort &=& \l[\vort,\elecp\r] + \l[\magf,\ecurrent\r] \ ,\nonumber\\
\dot\magf &=& \l[\magf,\elecp\r] \, .
\end{eqnarray*}
This is a model for low-beta reduced
MHD~\cite{Morrison1984,Strauss1977,Zeitlin1992}. It is obtained by an
expansion in the inverse aspect ratio~$\epsilon$ of a tokamak, with~$\epsilon$
small. This is called low beta since the plasma beta (the ratio of plasma
pressure to magnetic pressure) is of order~$\epsilon^2$. With a strong
toroidal magnetic field, the dynamics are then approximately two-dimensional.
Benjamin~\cite{Benjamin1984} used a system with a similar Lie--Poisson
structure, but for waves in a density-stratified fluid. Semidirect sum
structures are ubiquitous in advective systems: one variable (in this
example,~$\elecp$) ``drags'' the others along~\cite{Thiffeault1998}.
\subsubsection{Compressible Reduced MHD}
\seclabel{CRMHD}
In general there are other, more general ways to extend Lie algebras besides
the semidirect sum. The model derived by Hazeltine \emph{et
al.}~\cite{Hazeltine1987,Hazeltine1985b} for two-dimensional compressible
reduced MHD (CRMHD) is an example. This model has four fields, and as for the
system in \secref{lowbetaRMHD} it is also obtained from an expansion in the
inverse aspect ratio of a tokamak. It includes compressibility and finite ion
Larmor radius effects. The Hamiltonian is
\begin{equation}
H[\vort,\pvel,\pres,\magf] = \frac{1}{2}\int_\fdomain\l\lgroup
|\grad\elecp|^2 + \pvel^2
+ \frac{(\pres-2{\beta_\mathrm{i}}\,x)^2}{{\beta_\mathrm{i}}}
+ |\grad\magf|^2\r\rgroup \d^2x,
\eqlabel{CRMHDHam}
\end{equation}
where $\pvel$ is the parallel ion velocity, $\pres$ is the pressure, and
${\beta_\mathrm{i}}$ is a parameter that measures compressibility. The other variables
are as in \secref{lowbetaRMHD}. The coordinate~$x$ points outward from the
center of the tokamak in the horizontal plane and~$y$ is the vertical
coordinate. The motion is made two-dimensional by the strong toroidal
magnetic field. The bracket we will use is
\begin{eqnarray}
\lPB F\com G\rPB &=&
\int_\fdomain\l\lgroup\vort\lpb\frac{\fd F}{\fd \vort}
\com\frac{\fd G}{\fd \vort}\rpb\r.
+ \pvel\l(\lpb\frac{\fd F}{\fd \vort}
\com\frac{\fd G}{\fd \pvel}\rpb
+ \lpb\frac{\fd F}{\fd \pvel}
\com\frac{\fd G}{\fd \vort}\rpb\r)\nonumber\\
&&\mbox{}\!\!\!\!\!\!\!\!\!\! + \pres\l(\lpb\frac{\fd F}{\fd \vort}
\com\frac{\fd G}{\fd \pres}\rpb
+ \lpb\frac{\fd F}{\fd \pres}
\com\frac{\fd G}{\fd \vort}\rpb\r)
+ \magf\l(\lpb\frac{\fd F}{\fd \vort}
\com\frac{\fd G}{\fd \magf}\rpb
+ \lpb\frac{\fd F}{\fd \magf}
\com\frac{\fd G}{\fd \vort}\rpb\r)\nonumber\\
&&\mbox{}\!\!\!\!\!\!\!\!\!\! - \l.{\beta_\mathrm{i}}\,\magf\l(
\lpb\frac{\fd F}{\fd \pres}
\com\frac{\fd G}{\fd \pvel}\rpb
+ \lpb\frac{\fd F}{\fd \pvel}
\com\frac{\fd G}{\fd \pres}\rpb\r)\r\rgroup\d^2x.
\eqlabel{CRMHDbracket}
\end{eqnarray}
Together this bracket and the Hamiltonian \eqref{CRMHDHam} lead to the
equations
\begin{eqnarray*}
\dot\vort &=& \lpb \vort\com\elecp\rpb + \lpb\magf\com \ecurrent\rpb
+ 2\lpb \pres\com x\rpb\\
\dot\pvel &=& \lpb\pvel\com\elecp\rpb + \lpb\magf\com\pres \rpb
+ 2{\beta_\mathrm{i}}\lpb x \com \magf \rpb\\
\dot\pres &=& \lpb\pres\com\elecp\rpb
+ {\beta_\mathrm{i}}\lpb\magf\com\pvel\rpb\\
\dot\magf &=& \lpb \magf\com\elecp\rpb ,
\end{eqnarray*}
which reduce to the example of \secref{lowbetaRMHD} in the
limit~\hbox{$\pvel=\pres={\beta_\mathrm{i}}=0$} (when compressibility effects are
unimportant).
It is far from clear that the Jacobi identity for~\eqref{CRMHDbracket} is
satisfied. A direct verification is straightforward (if tedious), but we shall
see in \secref{theproblem} that there is an easier way.
\subsection{General Algebra Extensions}
\seclabel{theproblem}
We wish to generalize the types of bracket used in
\secreftwo{lowbetaRMHD}{CRMHD}. We build an algebra extension by forming an~$n$-tuple of elements of a single Lie algebra~$\LieA$,
\begin{equation}
\alpha \mathrel{\raisebox{.069ex}{:}\!\!=} \l(\alpha_1,\dots,\alpha_n\r), \eqlabel{algtuple}
\end{equation}
where~\hbox{$\alpha_i \in \LieA$}. The most general bracket on this~$n$-tuple
space obtained from a linear combination of the one in~$\LieA$ has components
\begin{equation}
{\lpb\alpha\com\beta\rpb}_\lambda = \sum_{\mu,\nu=1}^n
{\W_\lambda}^{\mu\nu}\,
\lpb\alpha_\mu\com\beta_\nu\rpb\,,
\ \ \ \lambda=1,\dots,n,
\eqlabel{extbrack}
\end{equation}
where the~${\W_\lambda}^{\mu\nu}$ are constants. (From now on we will assume
that repeated indices are summed unless otherwise noted.) Since the bracket
in~$\LieA$ is antisymmetric the~$\W$'s must be symmetric in their upper
indices,
\begin{equation}
{\W_\lambda}^{\mu\nu} = {\W_\lambda}^{\nu\mu}\,.
\eqlabel{upsym}
\end{equation}
This bracket must also satisfy the Jacobi identity
\[
{\lpb\alpha\com\lpb\beta\com\gamma\rpb\rpb}_\lambda +
{\lpb\beta\com\lpb\gamma\com\alpha\rpb\rpb}_\lambda +
{\lpb\gamma\com\lpb\alpha\com\beta\rpb\rpb}_\lambda = 0, \ \
\lambda=1,\dots,n.
\]
The first term can be written
\[
{\lpb\alpha\com\lpb\beta\com\gamma\rpb\rpb}_\lambda =
{\W_\lambda}^{\sigma\tau}\,{\W_\sigma}^{\mu\nu}\,
{\lpb\alpha_\tau\com\lpb\beta_\mu\com\gamma_\nu\rpb\rpb},
\]
which when added to the other two gives
\[
{\W_\lambda}^{\sigma\tau}\,{\W_\sigma}^{\mu\nu}\,\l(
{\lpb\alpha_\tau\com\lpb\beta_\mu\com\gamma_\nu\rpb\rpb}
+ {\lpb\beta_\tau\com\lpb\gamma_\mu\com\alpha_\nu\rpb\rpb}
+ {\lpb\gamma_\tau\com\lpb\alpha_\mu\com\beta_\nu\rpb\rpb}
\r) = 0.
\]
We cannot yet make use of the Jacobi identity in~$\LieA$: the subscripts
of~$\alpha$, $\beta$, and~$\gamma$ are different in each term so they
represent different elements of~$\LieA$. We first relabel the sums and then
make use of the Jacobi identity in~$\LieA$ to obtain
\begin{eqnarray*}
&&\l({\W_\lambda}^{\sigma\tau}\,{\W_\sigma}^{\mu\nu}
- {\W_\lambda}^{\sigma\nu}\,{\W_\sigma}^{\tau\mu}\r)\,
{\lpb\alpha_\tau\com\lpb\beta_\mu\com\gamma_\nu\rpb\rpb}\nonumber\\
&&\mbox{} + \l({\W_\lambda}^{\sigma\mu}\,{\W_\sigma}^{\nu\tau}
- {\W_\lambda}^{\sigma\nu}\,{\W_\sigma}^{\tau\mu}\r)\,
{\lpb\beta_\mu\com\lpb\gamma_\nu\com\alpha_\tau\rpb\rpb} = 0\,.
\end{eqnarray*}
This identity is satisfied if and only if
\begin{equation}
{\W_\lambda}^{\sigma\tau}\,{\W_\sigma}^{\mu\nu}
= {\W_\lambda}^{\sigma\nu}\,{\W_\sigma}^{\tau\mu}\,,
\eqlabel{Wjacob}
\end{equation}
which together with~\eqref{upsym} implies that the
quantity~${\W_\lambda}^{\sigma\tau}\,{\W_\sigma}^{\mu\nu}$ is symmetric in all
three free upper indices. If we write the~$\W$'s as~$n$
matrices~${\W}^{(\nu)}$ with rows labeled by~$\lambda$ and columns by~$\mu$,
\begin{equation}
{{\l[{\W}^{(\nu)}\r]}_\lambda}^\mu \mathrel{\raisebox{.069ex}{:}\!\!=} {\W_\lambda}^{\mu\nu},
\eqlabel{Wupdef}
\end{equation}
then~\eqref{Wjacob} says that those matrices pairwise commute:
\begin{equation}
\W^{(\nu)}\,\W^{(\sigma)} = \W^{(\sigma)}\,\W^{(\nu)}.
\eqlabel{Wcommute}
\end{equation}
Equations~\eqref{upsym} and~\eqref{Wcommute} form a necessary and sufficient
condition: a set of~$n$ commuting matrices of size~$n\times n$ satisfying the
symmetry given by~\eqref{upsym} can be used to make a good Lie algebra
bracket. From this Lie bracket we can build a Lie--Poisson bracket using the
prescription of~\eqref{LPB} to obtain
\[
{\lPB F\com G \rPB}_\pm(\fv) = \pm\sum_{\lambda,\mu,\nu=1}^n
{\W_\lambda}^{\mu\nu}\lang\fv^\lambda\com
{\lpb \frac{\fd F}{\fd\fv^\mu}\com\frac{\fd G}{\fd\fv^\nu}
\rpb}\rang .
\]
We now return to the two extension examples of \secreftwo{lowbetaRMHD}{CRMHD}
and examine them in light of the general extension concept introduced here.
\subsubsection{Low-beta Reduced MHD}
\seclabel{matlowbetaRMHD}
For this example we have~$(\fv^0,\fv^1)=(\vort,\magf)$, with
\[
\W^{(0)} = \begin{pmatrix} \makebox[1.3em]{1} & \makebox[1.3em]{0} \\
\makebox[1.3em]{0} & \makebox[1.3em]{1}
\end{pmatrix},\ \ \ \
\W^{(1)} = \begin{pmatrix} \makebox[1.3em]{0} & \makebox[1.3em]{0} \\
\makebox[1.3em]{1} & \makebox[1.3em]{0}
\end{pmatrix}.
\]
The reason why we start labeling at~$0$ will become clearer in
\secref{semisimple}. The two~$\W^{(\mu)}$ must commute since~$\W^{(0)}=I$, the
identity. The tensor~$\W$ also satisfies the symmetry
property~\eqref{upsym}. Hence, the bracket is a good Lie algebra bracket.
\subsubsection{Compressible Reduced MHD}
\seclabel{matCRMHD}
We have~$n=4$ and take~$(\fv^0,\fv^1,\fv^2,\fv^3)=(\vort,\pvel,\pres,\magf)$,
so the tensor~$\W$ is given by
\begin{equation}
\begin{array}{rclrcl}
\W^{(0)} &=& \l(\begin{array}{cccc}
\makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1} & \makebox[1.3em]{0}\\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{1}\\
\end{array}\r), & \ \
\W^{(1)} &=& \l(\begin{array}{cccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & -{\beta_\mathrm{i}} & \makebox[1.3em]{0}\\
\end{array}\r),\\
\\
\W^{(2)} &=& \l(\begin{array}{cccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{0} & -{\beta_\mathrm{i}} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\end{array}\r), & \ \
\W^{(3)} &=& \l(\begin{array}{cccc}
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\makebox[1.3em]{1} & \makebox[1.3em]{0} & \makebox[1.3em]{0} & \makebox[1.3em]{0}\\
\end{array}\r).
\end{array}
\eqlabel{matCRMHD}
\end{equation}
It is easy to verify that these matrices commute and that the tensor~$\W$
satisfies the symmetry property, so that the Lie--Poisson bracket given
by~\eqref{CRMHDbracket} satisfies the Jacobi identity. (See
\secref{semisimple} for an explanation of why the labeling is chosen to begin
at zero.)
|
\section{Table Captions}
\begin{enumerate}
\item{\bf Table 1.1} We list the bifurcations from the synchronised
fixed point $x=0$. The type of bifurcation involved,the region in parameter space
where the bifurcation conditions are satisfied, the manner in which the eigenvalue
crosses unit circle and the power law exponent $P(l)\approx l^{\zeta_{i}}$
are listed.
The range of the exponent
$\zeta_{1}$ is $[1.9-2.2]$, $\zeta_{2}$ lies in the range $[1.3-1.35]$ and
$\zeta_{3} $ in the range $[0.61-0.72]$.
We also identify the nature of the intermittency, whether spatial or spatio-temporal(ST).
In {\bf Table 1.2} the same quantities are given for the other fixed point.
\item{\bf Table 2} This table lists the values of the spatial laminar exponent
observed in experiments (values as quoted in Ref. \cite{Colovas}) and compares their values with the laminar exponents of
our CML model. It is clear that the exponents $\zeta_1$ and $\zeta_3$ are directly observed in fluid experiments on convection in an annulus and in the roll coating system. As explained in the text, the exponent $\zeta_2$ serves as a lower bound on the ex
ponent observed in convection in a channel and in the Taylor Dean system. The
experimentally observed exponent actually coincides with the
CML exponent $\zeta_G$ which occurs when the structure associated with the exponent $\zeta_2$ undergoes a further bifurcation.
\end{enumerate}
\end{document}
---------------------------------------------------------------------------------
Neelima M. Gupte
Department of Physics
Indian Institute of Technology, Madras
Chennai 600 036, INDIA.
Phone: +91-44-2351365 Ext. 8679 /4458679 (office) +91-44-2454794 (Home)
Fax: +91-44-2350509
E-mail: <EMAIL>
<EMAIL>
URL: http://www.imsc.ernet.in/~gupte
http://hsb.iitm.ernet.in/~gupte
---------------------------------------------------------------------------------
|
\section{Why quantizing a noncommutative geometry}
The idea that the geometric structure of physical spacetime could
be noncommutative exists in different versions. In some of
versions, the noncommutativity of geometry is viewed as a direct
effect of quantum mechanics, which disappears in the limit in
which we consider processes involving actions much larger than
the Planck constant \cite{noncaltri}. In the noncommutative
geometry approach of Connes et.\,al.\,(NCG)
\cite{dirac,stmodel,sptr,thomas}, on the other hand,
noncommutativity is introduced as a feature of spacetime which
exists independently from quantum mechanics. For instance, in
the noncommutative version of the standard model \cite{stmodel},
the theory is defined over a noncommutative spacetime,
and is {\it then\/} quantized along conventional perturbative
lines. More ambitiously, the spectral triple formulation
\cite{sptr} includes the gravitational field as well. For the
gravitational field, however, conventional perturbative
quantization methods fail \cite{qg}. The problem of going from
the noncommutative, but non-quantum-mechanical, spectral triple
dynamics to the full quantum dynamics is thus open.
In a quantum theory that includes gravity, the geometric
structure of spacetime is to be treated quantum mechanically.
Therefore the noncommutative geometry of spacetime must be
reinterpreted in quantum terms. Thus, in a theory of the
physical world based on NCG and including quantum mechanics, the
geometry of spacetime should be represented by a {\it
quantization\/} of a noncommutative geometry.
There should therefore be two distinct sources of
noncommutativity in the theory: the noncommutativity of the
elements of the algebra describing spacetime {\it and},
separately, the noncommutativity of the quantum mechanical
variables. In this note, we address the problem of understanding
what a quantization of a noncommutative geometry might be, and
what is the relation between geometric noncommutativity and
quantum noncommutativity. We study this problem using a simple
model derived from \cite{thomas} with the aim of developing
structures and notions which, hopefully, could guide us in
addressing the same issue in a full model including general
relativity.
In particular, we consider the spectral triple approach given in
\cite{sptr}. Within this approach, a dynamical model is given by
the spectral triple $(A,H,D)$, where $H$ is a Hilbert space and
$A$ is a $C^{*}$ algebra represented on $H$, which are fixed
once and for all; while $D$ is a Dirac operator (in the sense of
\cite{dirac}) in $H$, which codes the value of the dynamical
fields, and in particular of the spacetime metric, that is, the
gravitational field. Thus $D$ is the dynamical variable of the
model and represents a classical configuration of the theory.
The dynamics is then given by an action $S[D]$. To quantize the
theory, we must find the Hilbert space $K$ of its quantum states.
A state in $K$ will represent a quantum state of the
noncommutative geometry: roughly, a probabilistic quantum
superpositions of (noncommutative) geometries. Such a state will
assign not a number, but rather a probability distribution, to
the observable distance $d(p,p')$ between any two points.
Observable quantities will be represented by operators on $K$.
We construct the Hilbert space $K$ and the dynamical operators
for our simple model. From the quantum theory we obtain a
concrete result: the physical distance between (certain) two
points of the model, which in the classical theory can be an
arbitrary nonnegative number $d$, turns out to be quantized as
\begin{equation}
d = {\frac{ L_{P}}{\sqrt{2n+1}}}, \ \ \ \ n=0,1,2,3\ldots.
\label{qd}
\end{equation}
where $L_{P}$ (the ``Planck length'') is the length determined by
$\hbar$ and the coupling constant of the theory. Furthermore, we
show that the quantum theory can be compactly represented in
terms of a novel triple $(A,{\cal H},\hat D)$, where $\cal H$ is
the tensor product of $H$ with the space of the quantum
geometries $K$.
There exist several attempts to explore the relation between
noncommutative geometry and quantum theory by studying quantum
fields defined over a noncommutative geometry
\cite{noncaltri,ncgqm}. These attempts should not be confused
with the present work. Here we are not concerned with the effect
of a noncommutative structure of spacetime over quantum fields:
we are concerned with the quantum mechanical properties of the
noncommutative geometry itself.
\section{Preliminaries}
\subsection{Mechanics without preferred time}
We begin by recalling a few simple points about classical and
quantum mechanics. This will fix notation and provide an
appropriate conceptual framework. We need to choose a language
sufficiently general to deal with theories, such as general
relativity (GR), in which time evolution enters in a non-simple
manner.
A dynamical theory is defined by a set of equations, the
equations of motion, for a set of variables, the dynamical
variables. By dynamical variable we mean here the full
``history'', or ``motion'', of the physical system. We denote
the space of these variables as $\cal C$, or extended
configuration space. For instance, the dynamics of an oscillator
is given by the equation of motion $d^{2}x(t)/dt^{2}=-\omega
x(t)$, where the variable is the {\em function\/} $x: t \mapsto
x(t)$. Thus, in this example the extended configuration space
$\cal C$ is $C_{\infty}(R)$. When time evolution is standard, as
in this example, $\cal C$ is the space of the maps from $R$ (the
time) into a configuration space. However, we are interested
also in systems in which $\cal C$ does not have this form. We
consider Lagrangian systems, in which the equations of motion can
be expressed as the vanishing of the first variation of a
function $S[x]$ on $\cal C$. $S[x]$ is the action functional.
The solutions of the equations of motion form a subspace of $\cal
C$. This subspace is the phase space of the system, and we
denote it $\Gamma$. In the oscillator, the solutions of the
equations of motion are $x(t)=A \sin(\omega t+\phi)$. Thus,
$\Gamma$ is a two-dimensional subspace of $\cal C$ coordinatized
by $A$ and $\phi$.\footnote{
In the example of the oscillator, and anytime time evolution is
standard, we can represent $\Gamma$ as a space of initial
data. In fact, each set of initial data determines a solution
and viceversa. To do this,
we have to fix a time, say $t=t_{0}$. Define then $q=x(t_{0})$
and $p=dx(t)/dt|_{t=t_{0}}$. The points in $\Gamma$ can then be
coordinatized by $(p,q)$ instead of $(A,\phi)$. The relation
between the two set of coordinates on $\Gamma$ is immediately
obtained from the definitions as $q=A \sin(\omega t_{0}+\phi)$
and $p=\omega A \cos(\omega t_{0}+\phi)$. The $(p,q)$
coordinatization of $\Gamma$ is the one commonly introduced in
textbooks. The definition of the phase space as the space of the
solutions of the equations of motion (known since
Lagrange) has the advantages of being more covariant (this
becomes clear in field theory), of not requiring the fixing of a
preferred time $t_{0}$ and of being extendible to the dynamical
systems without standard time evolution that concern us here.}
A point $s$ in $\Gamma$ represents a physically realizable (that
is, compatible with the classical equations of motion), or
``onshell'', history of the system. This is a classical state of
the system. Here, ``state'' is used in the same (atemporal)
sense as ``Heisenberg state'' in quantum theory. An observable
quantity $f$ corresponds to a function $f: s\in\Gamma \mapsto
f(s)\in R$, where $f(s)$ is the predicted value of $f$ in the
state $s$.\footnote{For instance, the position $x_{t}$ of the
oscillator at time $t$ is the function $x_{t}(A,\phi)= A
\sin(\omega t+\phi)$ on $\Gamma$.} The phase space carries a
symplectic structure. In the example, $\Omega = dp\wedge dq =
\omega A\, dA\wedge d\phi$. This defines Poisson brackets
between observables. If the system has a standard time
structure, the symplectic structure can be determined by standard
methods. The definition of $\Omega$ in more general cases can be
problematic. The Poisson bracket structures of the observables
is the starting point for their quantization.
\subsection{The spectral formulation of general relativity}
The model we introduce in the next section is meant to mimic some
of the features of the spectral formulation of GR
\cite{sptr,landi}. We briefly recall this formulation. Consider
euclidean GR over a fixed compact\footnote{Euclidean GR over a
compact manifold is a theory which is likely to admit only a
finite dimensional space of solutions, and whose relation with
physical GR is questionable. However, the theory is, by itself,
relevant as a model of diffeomorphism invariant field theory of
the geometry. More importantly, the experience with conventional
euclidean quantum field theory suggests that the quantization of
the euclidean theory with compact time might be of great
relevance for understanding the true quantum theory even if the
corresponding classical theories have extremely different
properties.} four dimensional manifold $M$. Let ${\cal G}_{M}$
be the space of the riemanian metrics $g:M\times M \to R^{+}$
over $M$. In the standard formulation of the theory, ${\cal
G}_{M}$ is taken as the extended configuration space
\begin{equation}
{\cal C} \equiv {\cal G}_{M}.
\label{spacegr}
\end{equation}
This is the space of the (four-dimensional) gravitational fields.
The action is then chosen to be the the well known
Einstein-Hilbert action
\begin{equation}
S[g]=\int d^{4}x\ \sqrt{g}R
\label{actiongr1}
\end{equation}
The phase space $\Gamma$ is the space of the Einstein spaces,
namely the gravitational fields that solve the Einstein
equations.
The theory can be formulated also as follows. Fix the 4d compact
(spin) manifold $M$. Consider the spectral triples $(A,H,D)$.
Here $A=C_{\infty}(M)$; the Hilbert space $H$ is the (Hilbert
completion of the) space of the half-densitized Dirac spinor
fields on $M$. Notice that $H$ is defined also in the absence of
a Riemanian structure\footnote{I thank Alain Connes for
clarifying this point.}. $H$ carries a representation $\pi$
of $A$, where $\pi(a)$ is the operator that multiplies the spinor
by the function $a\in A$. $A$ and $H$ are fixed structures. We
then consider the space ${\cal D}_{(A,H)}$ of all the Dirac
operators $D$, in the sense of Connes \cite{dirac}, namely the
set of the operators $D$ on $H$ such that $(A,H,D)$ is a spectral
triple. We take the space of the Dirac operators as the extended
configuration space
\begin{equation}
{\cal C} \equiv {\cal D}_{(A,H)}.
\label{space}
\end{equation}
Thus, a Dirac operator represents here a history of the
gravitational field. The action is chosen to be
\begin{equation}
S[D] = Tr\ [f(D)]
\label{actiongr}
\end{equation}
where $f(\cdot)$ is a suitable simple function given in
\cite{sptr,landi}. The dynamical system defined by (\ref{space})
and (\ref{actiongr}) is physically equivalent to the one defined
by (\ref{spacegr}) and (\ref{actiongr1}) (see \cite{sptr,landi}
for qualifications). The reason is that given a metric
$g\in{\cal G}_{M}$, there exists a corresponding Dirac operator
$D(g)$ on $H$, defined with standard techniques. Viceversa,
given an operator $D$ in ${\cal D}_{(A,H)}$ there is a Riemanian
metric $g$ on $M$ defined by $g=d$, where $d$ is given by the NCG
distance formula
\begin{equation}
d(p,p') = sup_{\{a\in A, |[D,\pi(a)]|
{\scriptscriptstyle \le} 1\}}\, |p(a)-p'(a)|,
\ \ p,p'\in M,
\label{distance}
\end{equation}
which is such that $D=D(g)$. In (\ref{distance}), $p(a)=a(p)$
and $p'(a)=a(p')$ are the Gel'fand states (the points of the
space $M$ correspond to states over the algebra $C_{\infty}(M)$).
Thus, there is a one-to-one correspondence between the space
${\cal D}_{(A,H)}$ of the Dirac operators and the space ${\cal
G}_{M}$ of the Riemanian metrics, and we can identify the two
spaces. Finally, it is shown in \cite{dirac} that $S[D(g)]\sim S[g]$
(again, see \cite{sptr,landi} for qualifications).
The key lesson we learn from this is that the space of the Dirac
operators ${\cal D}_{(A,H)}$ can be viewed as the space $\cal C$
of the extended configurations of the gravitational field.
Accordingly, the phase space $\Gamma$ of the theory is the
subspace of ${\cal D}_{(A,H)}$ where the first variation of the
action functional $S[D]$ vanishes. In a (nonperturbative)
quantization of GR, one chooses a coordinatization $f,f', \ldots
$ of the phase space $\Gamma$ of the theory and then searches for
a representation of the Poisson algebra of the observables $f,f',
\ldots $ in terms of self-adjoint operators on a Hilbert space
$K$. If we want to explore non-perturbatively the quantum
mechanics of a NCG model, we have thus to quantize a phase space
$\Gamma$ formed by the Dirac operator obeying appropriate
equations of motion. In other words, we have to coordinatize
this space, and represent the coordinates as operators on the
quantum state space. Below, we complete this procedure for a
simple, finite dimensional, model.
\section{Definition of the model}
We choose a finite dimensional spectral triple $(A,H,D)$ (see
\cite{thomas}). Let $A$ be the algebra $M_{2}\oplus C$, where
$M_{2}$ is the algebra of complex $2\times 2$ matrices and $C$ is
the complex plane. We write $a=({\mathrm{A}},\alpha)\in A =
M_{2}\oplus C$. Let $H$ be the Hilbert space $M_{3}$, the linear
space of complex $3\times 3$ matrices $\Psi$ with the scalar
product $(\Psi,\Phi)=Tr[\Psi^{\dagger}\Phi]$, and let $A$ act on
$H$ in the representation $\pi$ given by
\begin{equation}
\pi(a) \Psi = \Psi_{a}\Psi
\end{equation}
where
\begin{equation}
\Psi_{a} =
\pmatrix{{\mathrm{A}} & 0 \cr 0 & \alpha}.
\label{f}
\end{equation}
A Dirac operator on $H$ \cite{dirac,thomas}, has the form (see
\cite{thomas})
\begin{equation}
D\Psi= {{\mathrm{D}}}\Psi + \Psi {{\mathrm{D}}}^{\dagger}
\end{equation}
where
\begin{equation}
{{\mathrm{D}}}=\pmatrix{0 & m \cr \overline m & 0} =
\pmatrix{0 & 0 & m_{1} \cr 0 & 0 & m_{2} \cr \overline m_{1} &
\overline m_{2} & 0 }.
\label{dirac}
\end{equation}
$m_{1}$ and $m_{2}$ are complex numbers. As mentioned
above, we view $A$ and $H$ as fixed structures, while $D$ is the
dynamical variable of the model. Since the Dirac operator
(\ref{dirac}) is determined by the two complex numbers $m_{i},
i=1,2$, the space ${\cal D}_{(M_{2}\oplus C,C^{3})}$ of the Dirac
operators compatible with the given $A$ and $H$, namely the space
of the dynamical variables, is isomorphic to $C^{2}$ and
coordinatized by $m_{i}$. Thus the extended configurations space
is
\begin{equation}
{\cal C} = {\cal D}_{(M_{2}\oplus C,\,M_{3})} \sim C^{2}.
\end{equation}
This is the analog, in our model, of the space of the
configurations of the gravitational field in GR
We complete the definition of the spectral triple (see
\cite{dirac}) by defining
\begin{eqnarray}
J\Psi &\equiv& \Psi^{\dagger};
\nonumber \\
\Psi_{\gamma} &\equiv& \pmatrix{1&0\cr 0&-1},
\nonumber \\
\gamma \Psi &\equiv& \Psi_{\gamma}\Psi,
\nonumber \\
\chi\Psi &\equiv& \gamma J\gamma J \Psi =
\Psi_{\gamma}\Psi\Psi_{\gamma}.
\end{eqnarray}
The two involutions $J$ (``charge conjugation'') and $\chi$
(``parity'') satisfy the properties needed to complete the
definition of the spectral triple.
The dynamics is determined by choosing an action $S[D]$, namely a
function on the space of the dynamical variables $\cal C$. Here
we disregard the ``spectral principle'' \cite{sptr} which
requires the action to depend on the spectrum of $D$ only: we
leave the extension of the present ideas to genuinely spectral
invariant actions to further developments. The simplest
possibility is to have a ``free'' theory with a quadratic action.
Thus we search an action of the form
\begin{equation}
S[D] = \ \frac{1}{2}\ Tr[\,{\mathrm{D}}\,\tilde M\,{\mathrm{D}}\,],
\label{action}
\end{equation}
where $\tilde M$ is a $3\times 3$ matrix that determines the
equations of motion. It is easy to see that, because of the
special form (\ref{dirac}) of ${\mathrm{D}}$, the action $S[D]$ in
(\ref{action}) can be rewritten as
\begin{equation}
S[D] = \ \overline m_{i} M_{ij} m_{j},
\label{action2}
\end{equation}
where $M$ is a $2\times 2$ matrix. We want the action to be
real, and thus $M$ must be hermitian. We want the theory to have
a nontrivial space of solutions, and thus $M$ must have vanishing
determinant. These requirements fix $M$ up to a single complex
parameter $\alpha$ and an overall normalization $G$ that does not
affect the field equations but is needed to set the right
physical dimensions
\begin{equation}
M = \frac{1}{G} \pmatrix{|\alpha|^{2} & \overline\alpha \cr
\alpha & 1}.
\label{M1}
\end{equation}
(We write $G$ in the denominator, following the GR use.)
For simplicity, we further choose $\alpha$ to be a pure phase
$\alpha=e^{i\phi}$, although this is not really needed for what
follows. Thus
\begin{equation}
M = \frac{1}{G} \pmatrix{ 1 & e^{-i\phi} \cr e^{i\phi} & 1}.
\label{M}
\end{equation}
This completes the definition of the model. Let us now
analyze its content.
By extremizing the action with respect to $m_{i}$ and $\bar
m_{i}$ we obtain the equations of motion
\begin{equation}
m_{2}= e^{i\phi} \ m_{1},
\label{eqofm}
\end{equation}
the ``Einstein equations'' of the model. These select, out of
all the Dirac operators, a physical phase space $\Gamma$ of
physical ones (in GR: the Dirac operators corresponding to
Einstein metrics). We say that $D$ is ``on shell'' if it
satisfies the equations of motion.
The phase space $\Gamma$ is coordinatized just by $m_{1}$.
Therefore $\Gamma$ is isomorphic to the complex plane $C$. From
now on we will write $ m \equiv m_{1}$ for simplicity. The
complex plane has a natural symplectic structure.
\begin{equation}
\Omega = \frac{i}{G} \ dm \wedge d\overline m =
\frac{2}{G}\ d(\Re(m))\wedge d(\Im(m)).
\label{omega}
\end{equation}
($G$ adjusts dimensions.) We take this as the physical
symplectic structure of the system\footnote{I thank Abhay
Ashtekar for this suggestion}. Thus the basic Poisson brackets
are
\begin{equation}
\{ m , \overline m \} = iG
\label{pb}
\end{equation}
A straightforward computation shows that the symplectic form
$\Omega$ can be written directly in terms of the Dirac operator.
We define by $\Omega_{ex} \equiv d\theta$, where
\begin{equation}
\theta \equiv \frac{i}{12G}\ {\rm Tr}[\gamma D dD] = \frac{i}{4G}
\ Tr[\Psi_\gamma\mathrm{D}d\mathrm{D}].
\end{equation}
(The first trace is the trace of the operators on $H$,
the second is the trace of the $3\times 3$ matrices.)
Then $\Omega$ is the restriction on shell of $\Omega_{ex}$.
The physical interpretation of $D$ is that it determines the
metric structure, and therefore the ``gravitational field'', over
the space of the states over $A$. A state $p$ over $A$ is
determined by a vector $\Psi_{p}$ in $H$
\begin{equation}
p(a) = \langle \Psi_{p} |\pi(a)| \Psi_{p} \rangle.
\end{equation}
In particular, consider the two vectors
$\Psi_{p}=\psi_{p}\otimes\psi_{p}$ and
$\Psi_{p'}=\psi_{p'}\otimes\psi_{p'}$, where $\psi_{p}=(0,0,1)$
and $\psi_{p'}=\frac{1}{\sqrt{2}}(e^{i\phi},1,0)$. They
define the states $p$ and $p'$ over $A$. Their distance is given
by (\ref{distance}) and can be explicitly computed using
Eq.\,(2.175) of Sec\,2.4.1 in \cite{thomas}, obtaining
\begin{equation}
d(p,p')=\frac{1}{\sqrt{|m_{1}|^{2}+|m_{2}|^{2}}}
=\frac{1}{\sqrt{2}|m|}.
\label{d}
\end{equation}
\section{Quantization}
We want to promote $m$ and $\overline m$ to operators
$\hat m$ and $\hat{\overline m}$ on a Hilbert space $K$ in such a
way that their algebra represents ($i\hbar$ times) the Poisson
algebra (\ref{pb}) and that the operators $\Re\hat m=1/2(\hat
m+\hat{\overline m})$ and ${\Im}\hat m =i/2(\hat m-\hat{\overline
m})$, which correspond to real quantities, be self-adjoint. The
solution is well known. The complex structure of $\Gamma$
provides us with a preferred polarization: we choose a
representation in which the wave functions depend on $m$ and are
independent from $\overline m$. Vectors in $K$ are thus analytic
functions $\psi(m)$ and
\begin{eqnarray}
\hat m\ \psi(m) &=& m\ \psi(m), \nonumber \\
\hat{\overline m}\ \psi(m) &=& \hbar G\ \frac{d}{dm}\psi(m).
\label{mbarm}
\end{eqnarray}
The scalar product that implements the desired reality conditions is
\begin{equation}
(\psi,\psi') = \frac{1}{\pi\hbar G} \int_{C}
dm \ e^{-\frac{|m|^{2}}{\hbar G}}\
\overline{\psi(m)} \, \psi'(m)
\label{sp}
\end{equation}
(The notation is $dm=d\Re(m) d\Im(m)$.) A convenient orthonormal
basis is given by the monomials
\begin{equation}
\psi_{n}(m)=\frac{m^{n}}{\sqrt{n!}(\hbar G)^{n/2}}= \langle m | n \rangle,
\label{basis}
\end{equation}
where we have introduced the Dirac notation $| n \rangle$ for the
basis elements. In this basis $\hat m$ and $ \hat{\overline m}$,
\begin{eqnarray}
\hat m\ | n \rangle & = & \sqrt{\hbar G}\ \sqrt{n+1}\ | n+1 \rangle,
\nonumber \\
\hat{\overline m}\ | n \rangle & =
& \sqrt{\hbar G}\ \sqrt{n}\ | n-1 \rangle,
\label{creann}
\end{eqnarray}
are immediately recognized as the creation and annihilation
operators. In fact, equations (\ref{mbarm},\ref{sp},\ref{basis})
form the standard definition of the harmonic oscillator quantum
mechanics in the Bargmann representation. We write the vectors
in $K$ as
\begin{equation}
| \psi \rangle = \sum_{n} \ \psi_{n}\, |n\rangle,
\end{equation}
namely we represent $K$ as the $l_{2}$ space of sequences $\psi_{n}$.
The operator corresponding to the $m_{2}$ variable is related to
$\hat m_{1}=\hat m$ by the (Heisenberg) equations of motion
\begin{equation}
\hat m_{2}= e^{i\phi}\ \hat m_{1}.
\label{hei}
\end{equation}
Consider now the distance $d$ between the two points $p$ and $p'$.
In the classical theory, this is given by (\ref{d}). The
corresponding operator $\hat d$ in the quantum theory is obtained
by replacing the classical quantities $m_{i},{\overline m_{i}}$
with their quantum counterparts $\hat m_{i}, \hat{\overline
m_{i}}$. Using (\ref{hei}), we obtain
\begin{equation}
\hat d = \frac{1}{\sqrt{2N}},
\end{equation}
where $N$ is the operator corresponding to the classical quantity
$m\overline m$. We encounter here an ordering ambiguity. We compute
easily from (\ref{creann})
\begin{equation}
N | n \rangle = \hbar G \ (n+c)\ | n \rangle.
\end{equation}
where we have $c=0$ if we order $N$ as $N=\hat m\hat{\overline
m}$; we have $c=1$ with the inverse ordering $N=\hat{\overline
m}\hat m$; and we have the well known ``harmonic oscillator
vacuum energy'' $c=1/2$ with the ``natural'' symmetric ordering,
which we assume from now on.
Thus we obtain the result that in the quantum theory the distance
$d(p,p')$ is quantized, with discrete eigenvalues
\begin{equation}
d(p,p') = \frac{1}{\sqrt{(2n+1)\hbar G}}
= \frac{L_{P}}{\sqrt{2n+1}},
\label{main}
\end{equation}
where
\begin{equation}
L_{P}=\frac{1}{\sqrt{\hbar G}}.
\label{LP}
\end{equation}
In the classical limit in which $\hbar$ is small and the quantum
number $n$ is large, the eigenvalues of the distance become more
and more dense and approximate the classical continuum.
\section{Quantum noncommutative geometry}
So far we have introduced two Hilbert spaces. The first is the
Hilbert space of the spectral triple we started from. This is
$H=M_{3}$, with vectors $\Psi$. We write their components also
as $\psi^{ab}, a,b=1,2,3$. $H$ is the space of the ``fermions''
over the spectral triple. The second is the Hilbert space of the
quantum theory $K=l_{2}$, with vectors $\psi_{n},
n=0,1,2,\ldots$. $K$ is the space of the quantum geometries.
That is, a vector in $K$ can be seen as a quantum linear
superposition of different classical geometries, or a quantum
superposition of Dirac operators. Each such states assigns not a
fixed value, but rather a probability distribution over the
possible values of the geometry defined by $D$. Accordingly, in
the quantum theory the Dirac operator (an operator on $H$) is
promoted to an operator on $K$, or, more precisely to an object
which is at the same time an operator on $H$ and on $K$:
\begin{equation}
\hat D = \pmatrix{0 & 0 & \hat m_{1} \cr
0 & 0 & \hat m_{2} \cr
\hat{\overline m_{1}} & \hat{\overline m_{2}}& 0} .
\label{hD}
\end{equation}
It is thus natural to consider the tensor product of $H$ and $K$.
We denote this tensor product as ${\cal H}=H\otimes K$. The
vectors in $\cal H$ can be written as $\psi_{n}^{ab}$. In this
basis, the matrix elements of the quantized Dirac operator $\hat
D$ can be computed directly from (\ref{creann}) and (\ref{hD}).
They are
\begin{eqnarray}
\hat D{}_n^{ab}{}^{m}_{cd} &=& \sqrt{\hbar G}\ \delta_{d}^{b} \
\left[\sqrt{n}\ (\delta^{a}_{1}+\delta^{a}_{2})
\ \delta_{c}^{3}\ \delta_{n}^{m-1}\right. \nonumber \\ &&
\left. + \sqrt{n+1}\ \delta^{a}_{3}
\ \delta_{n}^{m+1}\ (\delta_{c}^{1}+\delta_{c}^{2})\right] .
\label{hatD}
\end{eqnarray}
Notice that this is a {\em single\/} operator, not a variable
anymore (as the field operator in a quantum field theory is a
single operator, not a variable as the classical field).
The algebra $A$, as well as $J$ and $\chi$, are still represented
in $\cal H$ (they act trivially on $K$). Thus the quantum theory
defines a novel triple $ (A, \ {\cal H},\ \hat D) $ with respect
to which the set of the initial spectral triples $(A,H,D)$ (for
all possible $D$'s), can be seen as a classical limit. The
representation of $(A, J, \chi)$ in $\cal H$ is highly reducible.
Each state $\psi$ in $K$, namely each quantum geometry, defines a
3d subspace of $\cal H$ carrying a irreducible faithful
representation $\pi_{\psi}$. We call $D_{\psi}$ the operator
with matrix elements
\begin{equation}
(D_{\psi})_{cd}^{ab} = \overline\psi^{n} \hat
D{}_n^{ab}{}^{m}_{cd}
\psi_{m}
\end{equation}
acting on this subspace. Then the expectation value of the
distance between any two states $p$ and $p'$ over $A$, in the quantum
geometry determined by $\psi$ is
\begin{equation}
d_{\psi}(p,p') =
sup_{\{a\in A, |[D_{\psi},\pi_{\psi}(a)]|
{\scriptscriptstyle \le} 1\}}\ |p(a)-p'(a)|,
\end{equation}
where, notice, the Dirac operator and the representation of the
algebra are restricted to the irreducible component. Conversely,
each irreducible representation of $(A, J, \chi)$ in $\cal
H$ determines a state in $K$. It is thus tempting to identify
the quantum states of the geometry with these irreducible
representations. We leave a more detailed analysis of the
structure of the triple $(A,{\cal H}, \hat D)$ for further
developments.
\section{Comments and conclusions}
{\em Spectral invariance.} We have disregarded gauge invariance.
The action we have chosen is not a spectral invariant: it is not
a function solely of the eigenvalues of the Dirac operators, as
is the GR spectral action (\ref{actiongr}). (The model we have
chosen does not have enough degrees of freedom for accommodating
gauges.) Spectral invariance is crucial in GR, as, in
particular, it incorporates diffeomorphism invariance. The most
interesting extension to the technique considered here should
therefore, in our opinion, consists in the incorporation of
spectral invariance. In the case of a spectral invariant action,
the eigenvalues $\lambda_{n}$ of the Dirac operators are natural
real {\em gauge invariant\/} coordinates on $\cal C$. We expect
that in the quantum theory they be represented as self-adjoint
operators $\hat\lambda_{n}$. For GR, some information on the
Poisson algebra of the $\lambda_{n}$'s was obtained in
\cite{landi}.
{\em Zero point distance.} We may attach an intuitive physical
interpretation to the ``zero point distance'' given by
(\ref{main}) with $n=0$. In the classical theory the two points
$p$ and $p'$ can be at infinite distance (when $m=0$). In non
commutative geometry, this can be viewed as a ``classical''
limit. In the quantum theory, on the other hand, there exist a
maximal distance between $p$ and $p'$: the two ``sheets'' of the
universe cannot separate in the quantum theory\footnote{I thank
Thomas Sch\"ucker for pointing this out.}.
{\em Dimensions.} There are two natural possibilities of
assigning physical dimensions to the Dirac operator in NCG. The
traditional choice is to assign to the Dirac operator $D$ the
dimension of a mass. The second possibility is to assign it the
dimension of an inverse length. If $D$ is a mass, $G$ has
dimensions $M/L$. If $D$ is an inverse length, $G$ has
dimensions $1/ML^{3}$. If $D$ has the dimension of a mass, we
must insert the Planck constant (or another constant with the
same dimensions) in the formula (\ref{distance}) for the
distance, in order to adjust the dimensions. For instance we can
write
\begin{equation}
d(p,q) = sup_{\{a\in A, |[D,\pi(a)]|
{\scriptscriptstyle \le} \hbar\}}\ |p(a)-q(a)|.
\label{distance3}
\end{equation}
This is not unreasonable, since in the commutative case the
inequality is meant to fix the derivative of $a$ to be less than
one, but the (commutative) Dirac operator with dimensions of a
mass is proportional to $\hbar \frac{\partial}{\partial x}$. On
the other hand, it is a bit disturbing to invoke the Planck
constant in a theory before quantizing it. Equation
(\ref{distance3}) yields, instead of (\ref{main})
\begin{equation}
d(p,p')
= \frac{\hbar}{\sqrt{(2n+1) \hbar G }}
= \frac{L_{P}}{\sqrt{2n+1}},
\label{main3}
\end{equation}
where (\ref{LP}) is replaced by $ L_{P}=\sqrt{\frac{\hbar}{G}}$.
If, instead, $D$ is an inverse length, all the equations in the
text are dimensionally correct as they are. The two ways of
assigning dimension are obviously equivalent. We find the second
one ($D$ is an inverse length), which does not require us to
insert the Planck constant in the definition of distance, more
natural from the perspective of a gravitational theory.
{\em Functional quantization.} An alternative approach to the
canonical quantization method we have employed here is Feynman's
sum over histories approach. To quantize our model a' la
Feynman, we have to select a measure on the space of the
histories, namely on ${\cal D}=C^{2}$ and integrate the
exponential of the action with a source term added. This defines
the generating functional, from which the Green functions of the
theory can be computed by derivation. In a spectral invariant
theory, this should yield an integration over the Dirac
eigenvalues $\lambda_{n}$:
\begin{equation}
Z = \int [d\lambda_{n}]\ e^{-i/\hbar S[D]}
\end{equation}
Details on the possibility and the difficulties of this approach
will be given elsewhere.
\vskip.5cm
In conclusion, we have studied the quantum theory of an
elementary dynamical noncommutative geometry. We can derive a
few lessons from this exercise. First, a key object is the space
${\cal D}_{(A,H)}$ of the Dirac operators compatible with given
$A$ and $H$. This is the kinematical arena of a dynamical
noncommutative geometry. A second key object is the subspace
$\Gamma$ of ${\cal D}_{(A,H)}$ determined by the dynamical
equations. An important problem is the determination of the
symplectic structure of $\Gamma$. If a time evolution is
defined, general techniques are available. These extend to the
case in which time evolution is a gauge as in GR \cite{landi}. In the model
studied here, $\Gamma$ came equipped with a natural symplectic
structure. In the quantum theory, the matrix elements of $D$ are
operators on a Hilbert space $K$. The two Hilbert spaces $K$ and
$H$ combine naturally into the Hilbert space $\cal H$, on which a
quantum Dirac operator $\hat D$ is defined (see (\ref{hatD})).
The quantum states of the geometry are given by irreducible
representations of the algebra in $\cal H$. The distance between
two physical points (states over the algebra) which can take any
nonnegative real value in the classical noncommutative geometry
is quantized in the quantum theory, with a spectrum given in
(\ref{qd}).
The step from the simple model considered to a full theory
including GR is obviously enormous. The results described form
just an exploration of the structures involved in quantizing a
non-commutative-geometry. Hopefully, these structures can be
relevant in the quantization of more complete noncommutative
models as well.
\vskip.5cm
\centerline{------------------------------------}
\vskip.5cm
Great thanks to Abhay Ashtekar, for a very critical suggestion.
To Thomas Krajewski, whom I have pestered with innumerable
questions. To Thomas Sch\"ucker and Bruno Iochum for carefully
reading and criticizing the manuscript. To Daniel Testard,
Daniel Kastler, and everybody in the NCG group of the CPT for
help, for listening, and for the wonderful welcome I have
received in Marseille. This work has been partially supported by
NSF Grant PHY-95-15506.
|
\section{Introduction}
Compton scattering off the nucleon is determined by 6 independent
helicity amplitudes $A_i$ ($i$ = 1,...,6),
which are functions of two variables, e.g.
the Lorentz invariant variables $\nu$ (related to the $lab$ energy of
the incident photon) and $t$ (related to the momentum transfer to the
target). In the limit $\nu\rightarrow 0$, the general structure of
these amplitudes is governed by low energy theorems (LET) based on
Lorentz and gauge invariance. These theorems require that (I) the
leading term in the expansion in $\nu$ is determined by the global
properties of the nucleon, i.e. its charge, mass and anomalous
magnetic moment, and (II) the internal structure shows up only at
relative order $\nu^2$ and can be parametrized in terms of
polarizabilities. In this way there appear 6 polarizabilities for the
nucleon, the familiar electric and magnetic (scalar) polarizabilities
$\alpha$ and $\beta$ respectively, and 4 spin (vector) polarizabilities
$\gamma_1$ to $\gamma_4$. These polarizabilities describe the response
of the system to an external quasistatic electromagnetic field, and as
such they are fundamental structure constants of the composite system.
In particular, these polarizabilities allow one to make contact with
classical physics phenomena, the dielectric constant and the magnetic
permeability of macroscopic media as well as the Faraday effect in the
case of the spin polarizabilities.
\newline
\indent
As a consequence of LET, the differential cross section for
$\nu\rightarrow 0$ is given by the (model independent) Thomson term.
In a low-energy expansion, the electric and magnetic polarizabilities
then appear as interference between the Thomson term and the
subleading terms, i.e. as contribution of $O(\nu^2)$ in the
differential cross section, and $\alpha$ and $\beta$ can in principle
be separated by studying the angular distributions. However, it has
never been possible to isolate this term and thus to determine the
polarizabilities in a model independent way. The obvious reason is
that, for sufficiently small energies, say $\nu\leq 40$~MeV, the
structure effects are extremely small and hence the statistical errors
for the polarizabilities large. At larger energies, however, the
higher terms of the expansion, $O(\nu^4)$, become increasingly important.
Therefore, a reliable theoretical estimate of these higher terms is of
utmost importance. Moreover, at that order also the spin-dependent
polarizabilities come into the game, which has the further consequence
that a full determination of the 6 polarizabilities will require an
experimental program with polarized photons and polarized nucleons.
\newline
\indent
With the advent of high duty-factor electron accelerators and laser
backscattering techniques, new precision data have been obtained in
the 90's and more experiments are expected in the near future. In 1991
the Illinois group~\cite{Federspiel} measured differential cross
sections with tagged photons at low energy. As was to be expected, the
small counting rate and the low sensitivity to structure effects
allowed for a reduced statistical precision only. Shortly after,
Zieger et al.~\cite{Zieger} determined the cross section for photon
scattering at $\theta=180^{\circ}$ by detecting the recoil proton in a
magnetic spectrometer. In a series of experiments, the
Illinois-Saskatoon group then studied angular and energy distributions
over a wider range. Hallin et al.~\cite{Hallin} investigated the
region from pion production threshold to the $\Delta$(1232) resonance
with a high duty-factor bremsstrahlung beam. Though the statistical
and systematical errors were small, the range of energy was clearly
outside of the low-energy expansion. The presently most accurate
values for the proton polarizabilities were derived from the work of MacGibbon
et al.~\cite{McGibbon} whose experiments were performed with tagged
photons at 70~MeV$\leq \nu \leq 100$~MeV and untagged ones at the
higher energies, and analyzed in collaboration with L'vov~\cite{lvov97}
by means of dispersion relations (in the following denoted by DR) at
constant~$t$. The final results were
\begin{eqnarray}
\alpha \;&=&\; \left(12.1 \,\pm\,0.8 \,\pm\, 0.5 \right)\,
\times\,10^{-4}\,{\mathrm fm}^3 \;, \nonumber\\
\beta \;&=&\; \left(2.1 \,\mp\,0.8 \,\mp\,0.5 \right)\,
\times\,10^{-4}\,{\mathrm fm}^3 \;.
\label{eq1.1}
\end{eqnarray}
\newline
\indent
The physics of the $\Delta$(1232) and higher resonances has been the
objective of further recent investigations with tagged photons at
Mainz~\cite{Molinari,Peise} and with laser-backscattered photons at
Brookhaven~\cite{Tonnison}. The measured differential cross sections
and polarization asymmetries helped to discard a long-standing
problem, a unitarity violation reported in earlier experiments, and
provided useful cross-checks for the magnetic dipole and electric
quadrupole excitation of the $\Delta$ resonance. Such data were also
used to give a first prediction for the so-called backward spin
polarizability of the proton, i.e. the particular combination
$\gamma_{\pi}=\gamma_1+\gamma_2+2\gamma_4$ entering the Compton
spin-flip amplitude at $\theta=180^{\circ}$~\cite{Tonnison},
\begin{equation}
\gamma_{\pi} \;=\; -\,\left[ 27.1 \,\pm\, 2.2 ({\mathrm stat + syst})\,
{+2.8 \atop -2.4} ({\mathrm model})\right] \,
\times\,10^{-4}\,{\mathrm fm}^4 \;.
\label{eq1.2}
\end{equation}
\newline
\indent
In 1991 Bernard et al.~\cite{Bernard} evaluated the one-loop
contributions to the polarizabilities in the framework of relativistic
chiral perturbation theory (ChPT), with the result
$\alpha=10 \cdot \beta=12.1$, (here and in the following,
the scalar polarizabilities are given in units of $10^{-4}$~fm$^3$).
In order to have a systematic
chiral power counting, the calculation was then repeated in heavy
baryon ChPT~\cite{BKKM92} to $O(p^3)$, the expansion parameter being
an external momentum or the quark mass. The result explained the small
value of the magnetic polarizability, which had been difficult to
obtain in quark model calculations. A further calculation to
$O(p^4)$ resulted in the values $\alpha=10.5\pm 2.0$ and $\beta=3.5\pm
3.6$, the errors being due to 4 counter terms entering to that order,
which were estimated by resonance saturation~\cite{BKSM}. One of these
counter terms describes the paramagnetic contribution of the
$\Delta$(1232), which is partly cancelled by large diamagnetic
contributions of pion-nucleon loops.
\newline
\indent
In view of the importance of the $\Delta$ resonance,
Hemmert, Holstein and Kambor~\cite{Holstein} proposed to include the $\Delta$
as a dynamical degree of freedom. This added a further expansion
parameter, the difference of the $\Delta$ and nucleon masses
(``$\epsilon$ expansion''). A calculation to $O(\epsilon^3)$ yielded
the results~\cite{Hemmert} (see also Ref.\cite{HHKK})
\begin{eqnarray}
\alpha & = & 12.2 + 0 + 4.2 = 16.4 \nonumber\ ,\\
\beta & = & 1.2 + 7.2 + 0.7 = 9.1\ ,
\label{eq1.3}
\end{eqnarray}
where the 3 terms on the $rhs$ are the contributions of pion-nucleon
loops (identical to the predictions of the $O(p^3)$ calculation),
$\Delta$ pole terms, and pion-$\Delta$ loops. These predictions are
clearly at variance with the data, in particular $\alpha+\beta=25.5$
is nearly twice the rather precise value determined from DR (see
below). In an optimistic view the problem is to be cured by an
$O(\epsilon^4)$ calculation, which is likely to produce a large
diamagnetism as observed by Ref.~\cite{BKSM}. On the other hand, a
pessimist might doubt that the expansion converges sufficiently well
before the higher orders have introduced a host of unknown counter
terms.
\newline
\indent
The spin polarizabilities have been calculated in both
relativistic one-loop ChPT~\cite{BKKM92,BKKM} and heavy baryon
ChPT~\cite{HHKK}. In the latter approach the predictions are
\begin{eqnarray}
\gamma_0 & = & 4.6-2.4-0.2+0=+2.0\ ,\nonumber \\
\gamma_{\pi} & = & 4.6+2.4-0.2-43.5=-36.7\ ,
\label{eq1.4}
\end{eqnarray}
where the spin polarizabilities are given here and in all of the following
in units of 10$^{-4}$ fm$^4$.
The 4 separate contributions on the $rhs$ of Eq.~(\ref{eq1.4}) refer to
N$\pi$-loops, $\Delta$-poles, $\Delta\pi$-loops, and the triangle
anomaly, in that order. It is obvious that the anomaly or $\pi^0$-pole
contribution is by far the most important one, and that it would
require surprisingly large higher order contributions $O(\epsilon^4)$
to increase $\gamma_{\pi}$ to the value of Ref.~\cite{Tonnison}.
Similar conclusions were reached in the framework of DR. Using the
framework of DR at $t$ = const of Ref.\cite{lvov97},
Ref.~\cite{DKH} obtained a value of $\gamma_{\pi}=-34.3$,
while L'vov and Nathan~\cite{LN} worked in the framework of backward
DR and predicted $\gamma_{\pi}=-39.5\pm2.4$. In the latter approach
the dispersion integral is drawn along a line $t(\nu)$ corresponding
to backward Compton scattering, i.e. on the lower boundary of the
physical $s$-channel region, which is then complemented by a path into
the physical $t$-channel region.
\newline
\indent
As we have stated before, the most quantitative analysis of the
experimental data has been provided by dispersion relations. In this
way it has been possible to reconstruct the forward non spin-flip
amplitude directly from the total photoabsorption cross section by
Baldin's sum rule~\cite{Baldin}, which yields a precise value
for the sum of the scalar polarizabilities
\begin{eqnarray}
\alpha+\beta & = & 14.2 \,\pm\, 0.5 \;\;\;
({\rm{Ref.}}\cite{Damashek}) \nonumber\\
& = & 13.69 \,\pm\, 0.14 \;\;\; ({\rm{Ref.}}\cite{BGM})\ .
\label{eq1.5}
\end{eqnarray}
Similarly, the forward spin-flip amplitude can be evaluated by an
integral over the difference of the absorption cross sections in
states with helicity 3/2 and 1/2,
\begin{eqnarray}
\gamma_0 = \gamma_1-\gamma_2-2\gamma_4 & = & -1.34\; \;
({\rm{Ref.}}\cite{Sandorfi}) \nonumber \\
& = & -0.6\; \; ({\rm{Ref.}}\cite{DKH})\ .
\label{eq1.6}
\end{eqnarray}
While these predictions rely on pion photoproduction multipoles, the
helicity cross sections have now been directly determined by
scattering photons with circular polarizations on polarized
protons~\cite{Ahrens}.
\newline
\indent
In the case of forward Compton scattering the momentum transfer and,
hence, the Mandelstam variable $t$ vanishes. In that sense the above
sum rules of Eqs. (\ref{eq1.5}, \ref{eq1.6})
are derived from DR at $t=0$. At finite angles, however, the
analysis requires DR at $t={\rm{const.}}\leq0$, in a range of values
between 0 (forward scattering) and the largest negative value of $t$
($t = t_{{\rm{max}}}$), determined by the
largest scattering angle at the highest photon energy. As mentioned
above, the most quantitative and detailed such analysis has been
performed by L'vov and collaborators~\cite{lvov97,lvov98} in the framework of
unsubtracted DR at $t=$ const. Unfortunately, not all of the
dispersion integrals converge, as can be inferred from Regge theory.
The reason for the divergence of the integrals is essentially given by
fixed poles in the $t$ channel, notably the exchange of the neutral
pion and of a somewhat fictitious $\sigma$ meson with a mass of about
600~MeV and a large width, which models the two-pion continuum with
the quantum numbers $I=J=0$. In a more formal view, the dispersion
integral is performed along the real axis in the range
$-\nu_{max}\leq\nu\leq+\nu_{max}$, with $\nu_{max}\approx$ 1.5~GeV, and then
closed by a semi-circle with radius $\nu_{max}$ in the upper half of the
complex $\nu$-plane. The contribution of the semi-circle is then
identified with the asymptotic contribution described by $t$-channel
poles. This introduces unknown vertex functions and the mass
of the ``$\sigma$ meson'', which have to be determined from the
experiment. Moreover, the analysis depends appreciably on the choice
of $\nu_{max}$, and there are substantial contributions of intermediate
states beyond the relatively well-known pion-nucleon continuum. These
higher states include multipion, $\eta$- and $\rho$-meson production,
$\Delta \pi$-loops and nonresonant s-wave background. The physics
behind these effects is certainly worthwhile studying, and there can
be no doubt that within the next years we shall learn more about them
by detailed coincidence studies of multipion and heavier meson
production, but also directly from a careful analysis of Compton
scattering at the higher energies~\cite{lvov97}. However, the quest for
the polarizabilities as fundamental structure constants should not be
burdened by too many open questions and phenomenological models.
\newline
\indent
In view of the problems of unsubtracted DR, we propose to analyse
Compton scattering in the framework of subtracted DR at constant $t$,
with the eventual goal to determine the 6
polarizabilities with the least possible model dependence. We choose
to subtract the 6 Compton amplitudes $A_i(\nu, t)$ at the unphysical
value $\nu=0$, i.e. write subtracted DR for $A_i(\nu, t)-A_i(0, t)$ at
constant $t$. As we shall show in the following, these subtracted DR
converge nicely and are quite well saturated already at $\nu\approx
400-550$~MeV, i.e. essentially by one-pion production. Clearly the
price to pay are 6 new functions $A_i(0, t)$, which have to be
determined by another set of dispersion relations, at $\nu$=const=0 and
by use of information obtained from the $t$-channel reaction
$\gamma\gamma\rightarrow$ anything.
\newline
\indent
In order to reduce the dependence on the higher intermediate states in
the $t$-channel, we subtract again, i.e. write DR for $A_i(0,
t)-A_i(0, 0)$, the subtraction constants $A_i(0, 0)$ being linear
combinations of the 6 polarizabilities. Since 4 of these subtraction
constants can be calculated from unsubtracted DR at $t$=const, only 2
parameters have to be fixed by a fit to low energy Compton
scattering, the combinations $\alpha-\beta$ and
$\gamma_{\pi}=\gamma_1+\gamma_2+2\gamma_4$ describing the backward
non spin-flip and spin-flip amplitudes, respectively.
\newline
\indent
In a somewhat similar approach, Holstein and Nathan~\cite{HN} combined
$s$- and $t$-channel information to predict the backward scalar
polarizability $\alpha-\beta$. Using unsubtracted backward DR they
obtained, from the integration along the lower boundary of the
$s-$channel region, the result $(\alpha-\beta)^s=-6\pm3$, and from the
$t$-channel region a contribution of $(\alpha-\beta)^t\approx9$. The
sum of these two contributions, $\alpha-\beta\approx3\pm3$, is at
variance with the presently accepted experimental (global average) value,
$\alpha-\beta=10.0\pm1.5\pm0.9$ of Ref.~\cite{McGibbon}. The
difficulty to predict this observable is due to the bad
convergence of the integrals in both the $s-$ and the $t$-channel
regions. As may be seen from Fig.~3 of Ref.~\cite{HN}, the
$t$-channel integral obtains quite sizeable contributions from
$10 \, m_{\pi}^2<t<40 \, m_{\pi}^2$, in which region the integrand changes its
sign. Independent of the numerical analysis, the authors find an
extremely interesting relationship connecting the polarizabilities of
pions and nucleons. This connection comes via the $t$-channel integral
for the nucleon and the low-energy expansion of the $s$-wave
$\gamma\gamma\rightarrow\pi\pi$ amplitude, and results in
$\delta\alpha=-\delta\beta=0.5\alpha^{\pi}\approx 1.4$, which is the
contribution of the pion polarizability $\alpha^{\pi}$ to the
nucleon's electric and magnetic polarizabilities. In concluding this
discussion we point out the difference of our present approach and the
calculation of Ref.~\cite{HN}. We do not intend to predict
$\alpha-\beta$. Instead we want to develop a dispersion description of
Compton scattering allowing for a derivation of $\alpha-\beta$ with a
minimum of model dependence. For this purpose we use a scheme of
subtracted DR whose subtraction constants are linear
combinations of polarizabilities, which have to be determined by a fit
to the Compton data.
\newline
\indent
In section~\ref{disp} we shall give a general introduction to subtracted DR.
This technique is then applied to the cases of DR at $t$=const
($s$-channel dispersion integral) and DR at $\nu=0$ ($t$-channel
dispersion integral) in sections~\ref{schannel}
and \ref{tchannel}, respectively. Our results
are compared to the existing low-energy Compton data in
section~\ref{results}, and our conclusions are drawn in
section~\ref{conclusions}.
\section{Fixed-$t$ subtracted dispersion relations}
\label{disp}
Assuming invariance under parity, charge conjugation and time reversal
symmetry, the general amplitude for Compton scattering can be
expressed in terms of six independent structure functions $A_i(\nu,
t)$, $i=1,...,6$. These structure functions depend on two Lorentz
invariant variables, e.g. $\nu$ and $t$ as defined in the following.
Denoting the momenta of the initial state photon and proton by $q$ and
$p$ respectively, and with corresponding final state momenta
$q'$ and $p'$, the familiar Mandelstam variables are
\begin{equation}
s=(q+p)^2\ ,\ \ t=(q-q')^2\ ,\ \ u=(q-p')^2\ .
\label{eq2.1}
\end{equation}
These variables fulfill the constraint
\begin{equation}
s+t+u=2M^2\ .
\label{eq2.2}
\end{equation}
The variable $\nu$ is defined by,
\begin{equation}
\nu=\frac{s-u}{4M}=E_{\gamma}+\frac{t}{4M}\ ,
\label{eq2.3}
\end{equation}
where $E_{\gamma}$ is the photon energy in the $lab$ frame and $M$ the
nucleon mass. The Mandelstam plane is shown in Fig.~1, and the
boundaries of the physical and spectral regions are discussed in
Appendix~A.
\newline
\indent
The invariant amplitudes $A_i$ are free of kinematical singularities
and constraints, and because of the crossing symmetry they satisfy the
relation $A_i(\nu, t)=A_i(-\nu, t)$. Assuming further analyticity and
an appropriate high-energy behavior, the amplitudes $A_i$ fulfill
unsubtracted dispersion relations at fixed $t$,
\begin{equation}
{\mathrm Re} A_i(\nu, t) \;=\; A_i^B(\nu, t) \;+\;
{2 \over \pi} \; {\mathcal P} \int_{\nu_{thr}}^{+ \infty} d\nu' \;
{{\nu' \; {\mathrm Im}_s A_i(\nu',t)} \over {\nu'^2 - \nu^2}}\;,
\label{eq:unsub}
\end{equation}
where $A_i^B$ are the Born (nucleon pole) contributions, ${\mathrm
Im}_s A_i$ the discontinuities across the $s$-channel cuts of the
Compton process and $\nu_{thr} = m_\pi + (m_\pi^2 + t/2)/(2 M)$.
However, such unsubtracted dispersion relations
require that at high energies ($\nu \rightarrow \infty$) the
amplitudes ${\mathrm Im}_s A_i(\nu,t)$ drop fast enough so that the
integral of Eq.~(\ref{eq:unsub}) is convergent and the contribution from
the semi-circle at infinity can be neglected. For real Compton
scattering, Regge theory predicts the following high-energy behavior
for $\nu \rightarrow \infty$ and fixed $t$~\cite{lvov97}:
\begin{eqnarray}
&&A_{1,2} \sim \nu^{\alpha(t)} \;,\nonumber\\
&&A_{3,5,6} \sim \nu^{\alpha(t) - 2} \;,\hspace{1cm}
A_{4} \sim \nu^{\alpha(t) - 3} \;,
\end{eqnarray}
where $\alpha(t) \lesssim 1$ is the Regge trajectory. In particular we
note that the Regge trajectory with the highest intercept, i.e.
$\alpha(0) \approx 1.08$, corresponds to soft pomeron exchange. Due to
this high energy behavior, the unsubtracted dispersion integral of
Eq.~(\ref{eq:unsub}) diverges for the amplitudes $A_1$ and $A_2$. In order
to obtain useful results for these two amplitudes, L'vov et
al.~\cite{lvov97} proposed to close the contour of the integral in
Eq.~(\ref{eq:unsub}) by a semi-circle of finite radius $\nu_{max}$
(instead of the usually assumed infinite radius!) in the complex
plane, i.e. the real parts of $A_1$ and $A_2$ are calculated from the
decomposition
\begin{equation}
{\mathrm Re} A_i(\nu, t) \;=\; A_i^B(\nu, t) \;+\;
A_i^{int}(\nu, t) \;+\; A_i^{as}(\nu, t) \;,
\label{eq:aintas}
\end{equation}
with $A_i^{int}$ the $s$-channel integral from pion
threshold $\nu_{thr}$ to a finite upper limit $\nu_{max}$,
\begin{equation}
A_i^{int}(\nu, t) \;=\;
{2 \over \pi} \; {\mathcal P} \int_{\nu_{thr}}^{\nu_{max}} d\nu' \;
{{\nu' \; {\mathrm Im}_s A_i(\nu',t)} \over {\nu'^2 - \nu^2}}\;,
\label{eq13}
\end{equation}
and an `asymptotic contribution' $A_i^{as}$ representing the
contribution along the finite semi-circle of radius $\nu_{max}$ in the
complex plane. In the actual calculations, the $s$-channel integral is
typically evaluated up to a maximum photon energy $E_\gamma =
\nu_{max}(t) - t/(4 M) \approx 1.5$~GeV, for which the imaginary parts
of the amplitudes can be expressed through unitarity by the meson
photoproduction amplitudes (mainly 1$\pi$ and 2$\pi$ photoproduction)
taken from experiment. All contributions from higher energies are
then absorbed in the asymptotic term, which is replaced by a finite
number of energy independent poles in the $t$-channel. In particular
the asymptotic part of $A_1$ is parametrized in Ref.\cite{lvov97}
by the exchange of a scalar particle in the $t$-channel, i.e. an effective
``$\sigma$ meson'',
\begin{equation}
A_1^{as}(\nu, t) \approx A_1^{\sigma}(t) \;=\;
{{F_{\sigma \gamma \gamma} \; g_{\sigma NN}}
\over {t - m_\sigma^2}} \;,
\end{equation}
where $m_\sigma$ is the ``$\sigma$ meson'' mass, and
$g_{\sigma NN}$ and $F_{\sigma \gamma \gamma}$ are the couplings of the
``$\sigma$ meson'' to the nucleons and photons respectively.
The asymptotic part of $A_2$ is parametrized by the $\pi^0$
$t$-channel pole.
\newline
\indent
This procedure is relatively save for $A_2$ because of the dominance
of the $\pi^0$ pole or triangle anomaly, which is well established
both experimentally and on general grounds as Wess-Zumino-Witten term.
However, it introduces a considerable model-dependence in the case of
$A_1$. Though ``$\sigma$ mesons'' have been repeatedly reported in the
past, their properties were never clearly established. Therefore, this
particle should be interpreted as a parametrization of the $I=J=0$
part of the two-pion spectrum, which shows up differently in different
experiments and hence has been reported with varying masses and
widths.
\newline
\indent
It is therefore the aim of our present contribution to avoid the
convergence problem of unsubtracted DR and the
phenomenology necessary to determine the asymptotic contribution. The
alternative we shall pursue in the following is to consider
DR at fixed $t$ that are once subtracted at $\nu=0$,
\begin{equation}
{\mathrm Re} A_i(\nu, t) \;=\; A_i^B(\nu, t) \;+\;
\left[ A_i(0, t) - A_i^B(0, t) \right]
\;+\;{2 \over \pi} \;\nu^2\; {\mathcal P} \int_{\nu_{thr}}^{+ \infty} d\nu' \;
{{\; {\mathrm Im}_s A_i(\nu',t)} \over {\nu' \; (\nu'^2 - \nu^2)}}\;.
\label{eq:sub}
\end{equation}
These subtracted DR should converge for all six invariant amplitudes
due to the two additional powers of $\nu'$ in the denominator, and they are
essentially saturated by the $\pi N$ intermediate states as will be
shown in section \ref{schannel}. In other
words, the lesser known contributions of two and more pions
as well as higher continua are small and may be treated reliably
by simple models.
\newline
\indent
The price to pay for this alternative is the appearance of the
subtraction functions $A_i(\nu=0, t)$, which have to be determined at
some small (negative) value of $t$. We do this by setting up a
once-subtracted DR, this time in the variable $t$,
\begin{eqnarray}
A_i(0, t) \;-\; A_i^B(0, t) &=&
\left[ A_i(0, 0) \;-\; A_i^B(0, 0) \right]
\;+\;
\left[ A_i^{t-pole}(0, t) \;-\; A_i^{t-pole}(0, 0) \right] \nonumber\\
&+&\;{t \over \pi} \; \int_{(2 m_\pi)^2}^{+ \infty} dt' \;
{{{\mathrm Im}_t A_i(0,t')} \over {t' \; (t' - t)}}
\;+\;{t \over \pi} \; \int_{- \infty}^{a} dt' \;
{{{\mathrm Im}_t A_i(0,t')} \over {t' \; (t' - t)}} \;,
\label{eq:subt}
\end{eqnarray}
where $A_i^{t-pole}(0, t)$ represents the contribution of poles in the
$t$-channel, in particular of the $\pi^0$ pole in the case of $A_2$,
which is given by
\begin{equation}
A_2^{\pi^0}(0, t) \;=\; {{F_{\pi^0 \gamma \gamma} \; g_{\pi NN}}
\over {t - m_\pi^2}} \;.
\label{eq:piopole}
\end{equation}
The coupling $F_{\pi^0 \gamma \gamma}$ is determined through the $\pi^0
\rightarrow \gamma \gamma$ decay as
\begin{equation}
\Gamma\left( \pi^0 \rightarrow \gamma \gamma\right) \;=\;
{1 \over {64 \, \pi} } \, m_{\pi^0}^3 \, F_{\pi^0 \gamma \gamma}^2 \;.
\end{equation}
Using $\Gamma\left( \pi^0 \rightarrow \gamma \gamma\right)$ = 7.74 eV
\cite{PDG98}, one obtains
$F_{\pi^0 \gamma \gamma}$ = -0.0252 GeV$^{-1}$,
where the sign is in accordance with the $\pi^0 \gamma \gamma$
coupling in the chiral limit, given by the Wess-Zumino-Witten
effective chiral Lagrangian.
The $\pi NN$ coupling is taken from Ref.~\cite{arndtpin98} : $g_{\pi
NN}^2/4 \pi$ = 13.72. This yields then for the product of the couplings in
Eq.~(\ref{eq:piopole}) : $F_{\pi^0 \gamma \gamma} \, g_{\pi NN}
\approx$ -0.331 GeV$^{-1}$.
\newline
\indent
The imaginary part in the integral from $4 m_\pi^2 \rightarrow \infty$
in Eq.~(\ref{eq:subt})
is saturated by the possible intermediate states for the $t$-channel
process (see Fig.~\ref{fig:tunit}),
which lead to cuts along the positive $t$-axis.
For values of $t$ below the $K \bar K$
threshold, the $t$-channel discontinuity is dominated by $\pi \pi$
intermediate states. The second integral in Eq.~(\ref{eq:subt})
extends from $-\infty$ to $a$, where
$a \,=\, -4\, (m_{\pi}^2 + 2 M m_\pi) \approx - 1.1$ GeV$^2$ is the
boundary of the $su$ spectral region for $\nu = 0$ (see Appendix
\ref{app:spectralcompton} for a detailed discussion). As we are
interested in evaluating Eq.~(\ref{eq:subt}) for small (negative)
values of $t$ ($|t|\ll |a|$), the integral from $- \infty$ to
$a$ will be highly suppressed by the denominator of the subtracted DR,
and can therefore be neglected. Consequently, we shall saturate the
subtracted dispersion integrals of Eq.~(\ref{eq:subt}) by the
contribution of $\pi\pi$ intermediate states, which turns out to be a
good approximation for small $t$. We will show the convergence of
the $t$-channel dispersion integral in section \ref{tchannel}
and thus verify the quality of the approximation.
\newline
\indent
The $t$-dependence of the subtraction functions $A_i(0, t)$ is now
determined, and only the subtraction constants $A_i(0, 0)$ remain to
be fixed. We note that the quantities
\begin{equation}
a_i = A_i(0, 0) \;-\; A_i^B(0, 0)
\label{eq.a_i}
\end{equation}
are directly related to the polarizabilities, which can then be
obtained from a fit to the Compton scattering data.
For the spin-independent (scalar) polarizabilities $\alpha$ and
$\beta$, one finds the two combinations
\begin{eqnarray}
\label{eq:alphaplusbeta}
\alpha + \beta \;&=&\; - {1 \over {2 \pi}} \; (a_3 \;+\; a_6)\;, \\
\alpha - \beta \;&=&\; - {1 \over {2 \pi}} \; a_1\;,
\label{eq:nospinpol0}
\end{eqnarray}
which can be determined from forward and backward scattering
respectively. Furthermore, the forward combination $\alpha + \beta$ is
related to the total absorption spectrum through Baldin's sum rule
\cite{Baldin},
\begin{eqnarray}
(\alpha + \beta)_N \;&=&\
{1 \over {2 \pi^2}} \; \int_{\nu_{thr}}^{\infty} d\nu' \;
{{\sigma(\gamma N \rightarrow X)} \over {\nu'^2}}\;.
\label{eq22}
\end{eqnarray}
The 4 spin dependent polarizabilities $\gamma_1$ to
$\gamma_4$ of Ragusa~\cite{ragusa} are defined by
\begin{eqnarray}
\gamma_0 \;&\equiv&\; \gamma_1 - \gamma_2 - 2 \gamma_4 \;=\;
{1 \over {2 \pi} M} \; a_4\;, \label{eq23} \\
\gamma_{13} \;&\equiv&\; \gamma_1 + 2 \gamma_3 \;=\;
-\;{1 \over {4 \pi} M} \; (a_5 \;+\; a_6)\;, \label{eq24} \\
\gamma_{14} \;&\equiv&\; \gamma_1 - 2 \gamma_4 \;=\;
{1 \over {4 \pi} M} \; (2 \, a_4 \;+\; a_5 \;-\; a_6)\;, \label{eq25} \\
\gamma_\pi \;&\equiv&\; \gamma_1 + \gamma_2 + 2 \gamma_4 \;=\;
-\;{1 \over {2 \pi} M} \; (a_2 \;+\; a_5)\;,
\label{eq:spinpol0}
\end{eqnarray}
where $\gamma_0$ and $\gamma_\pi$ are the spin (vector)
polarizabilities in the forward and backward directions respectively.
Since the $\pi^0$ pole contributes to $A_2$ only, the combinations
$\gamma_0$, $\gamma_{13}$ and $\gamma_{14}$ of Eqs.~(\ref{eq23})-(\ref{eq25})
are independent of this pole term~\cite{DKH}, and only the backward
spin polarizability $\gamma_\pi$ is affected by the anomaly.
\newline
\indent
Although all 6 subtraction constants $a_1$ to $ a_6$ of
Eq.~(\ref{eq.a_i}) could be used as fit parameters, we shall restrict
the fit to the parameters $a_1$ and $a_2$, or equivalently to
$\alpha - \beta$ and $\gamma_\pi$. The subtraction constants $a_4,
a_5$ and $a_6$ will be calculated through an unsubtracted sum rule,
as derived from Eq.~(\ref{eq:unsub}),
\begin{equation}
a_{4, 5, 6} \;=\; {2 \over \pi} \; \int_{\nu_{thr}}^{+ \infty} d\nu' \;
{{ {\mathrm Im}_s A_{4, 5, 6}(\nu',t = 0)} \over {\nu'}}\;.
\label{eq:a4a5a6}
\end{equation}
The remaining subtraction constant $a_3$,
which is related to $\alpha + \beta$ through Eq.~(\ref{eq:alphaplusbeta}),
will be fixed through Baldin's sum rule, Eq.~(\ref{eq22}), using the
value obtained in Ref.~\cite{BGM} : $\alpha + \beta = 13.69$.
\section{s-channel dispersion integral}
\label{schannel}
In this section we describe the calculation of the $s$-channel
contributions, which enter in the once-subtracted
dispersion integral of Eq.~(\ref{eq:sub}) and in the calculation of
subtraction constants $a_4,\, a_5$ and $a_6$ through Eq.~(\ref{eq:a4a5a6}).
The imaginary part of the Compton amplitude due to the $s$-channel
cuts is determined from the scattering amplitudes of
photoproduction on the nucleon by the unitarity relation
\begin{equation}
\label{s-unit}
2\,\mbox{Im}_s\,T_{fi}=
\sum_X (2\pi)^4 \delta^4(P_X-P_i)T^{\dagger}_{X f }\,T_{X i} \;,
\end{equation}
where the sum runs over all possible states that can be formed in the
photon-nucleon reaction. Due to the energy denominator
$1/\nu'(\nu'^2-\nu^2)$ in the subtracted dispersion integrals,
the most important contribution is from the $\pi
N$ intermediate states, while mechanisms involving more pions or
heavier mesons in the intermediate states
are largely suppressed. In our calculation, we evaluate the $\pi N$
contribution using the multipole amplitudes from the analysis of
Hanstein, Drechsel and Tiator~\cite{HDT} at energies $E_\gamma\le
500$~MeV and at the higher energies we take as input the SAID
multipoles (SP98K solution)~\cite{said}. The expansion of
$\mbox{Im}_s A_i$ into this set of multipoles is truncated at a
maximum angular momentum $j_{max}=l\pm 1/2=7/2,$ with the exception of
the lower energy range ($E_\gamma\le 400$~MeV) where we use $j_{{\rm
max}}=3/2$. The higher partial waves with $j\ge j_{max}+1$ are
evaluated analytically in the one-pion exchange (OPE) approximation.
The relevant formulas to implement the calculation are reported in
Appendix B and C of Ref.\cite{lvov97}.
\newline
\indent
The multipion intermediate states are approximated by the inelastic
decay channels of the $\pi N$ resonances. In the spirit of
Ref.~\cite{lvov97} and the more recent work of Ref.~\cite{kamalov}, we
assume that this inelastic contribution follows the helicity structure
of the one-pion photoproduction amplitudes. In this approximation, we
first calculate the resonant part of the pion photoproduction multipoles
using the Breit-Wigner parametrization of Ref.~\cite{arndt}, which is then
scaled by a suitable factor to include the inelastic decays of the
resonances. The resulting contribution to $\mbox{Im}_s\, A_i$ is
\begin{eqnarray}
\left[\mbox{Im}_s\, A_i\right]^{(N^*\rightarrow \pi\pi N,\eta N,...)}=R
\left[\mbox{Im}_s\, A_i\right]^{(N^*\rightarrow \pi N)},
\end{eqnarray}
with the ratio $R$ given by
\begin{eqnarray}
\label{tp-scale}
R=\frac{1-B_\pi}{B_\pi}\frac{\bar\Gamma_{{\rm inel}}(W)}{\bar\Gamma_\pi(W)}.
\end{eqnarray}
In Eq.~(\ref{tp-scale}), $B_\pi$ is the single-pion branching ratio of the
resonance $N^*$ and $\bar\Gamma_\pi(W)$ the energy-dependent pionic
width~\cite{arndt}, while the inelastic width $\bar\Gamma_{{\rm
inel}}(W)$ of the decays $N^*\rightarrow(\pi\pi N,\,\eta
N,\,\pi\pi\pi N,..)$ is parametrized as in Ref.\cite{lvov97} in order
to provide the correct threshold behavior for the resonant two pion
contribution.
\newline
\indent
The $\pi N$ channel consistently reproduces the measured
photoabsorption cross section in the energy range $E_\gamma\le
500$~MeV, while at the higher energies nonresonant mechanisms should
be included in addition to the resonant mechanism
to fully describe the multipion channels. In
Ref.\cite{lvov97}, the non-resonant contribution
to the two-pion photoproduction channel was approximately taken into
account by calculating the OPE diagram
of the $\gamma N\rightarrow \pi\Delta$ reaction.
The difference between the data and the model for two-pion photoproduction
consisting of resonant mechanisms plus the OPE
diagram for the nonresonant mechanism, was then fitted in
Ref.\cite{lvov97} and attributed to a phenomenological,
non-resonant $\gamma N\rightarrow \pi\Delta$ s-wave correction term.
\newline
\indent
A more detailed description of the $\pi \pi N$ channel is clearly
worthwile to be undertaken, especially in view of the new
two-pion photoproduction data (both unpolarized and polarized) that will be
available from MAMI and JLab (CLAS) in the near future.
However, for the extraction of the polarizabilities, the strategy
followed in this paper is to minimize sensitivity and hence model
uncertainty to these higher channels.
\newline
\indent
We show in Fig.~\ref{fig:a1a2conv}
that in the subtracted DR, the
sensitivity to the multipion channels is indeed very small.
For the unsubtracted DR, on the other hand,
the influence of the multipion channels
amounts to about 30 \% of the amplitude $A_2$.
We furthermore note from Fig.~\ref{fig:a1a2conv} that the subtracted DR
are essentially saturated at $\nu~\approx$ 0.4 GeV and
that they only receive a negligible contribution from multipion
channels. The importance of the multipion channels is even weaker in
the case of the amplitudes $A_3$ to $A_6$.
\newline
\indent
In Table I and II, we show our predictions for the
dispersion integral of the spin polarizabilities of the proton and
neutron, respectively. We list the separate contribution of the $\pi
N$ channel, HDT($1\pi$), and the total result, HDT, which includes
the inelastic resonance channels. The last column shows the values of
the dispersion calculation of Ref.\cite{lvov98}, which is based on the
one-pion multipoles of the SAID-SP97K solution and the model for
double-pion production mentioned above. The small differences between
the one-pion multipoles of SAID-SP97K and SAID-SP98K at the higher
energies are practically negligible for the spin polarizabilities,
while the results are very sensitive to the differences between the
HDT and SAID analyses. As discussed in Ref.~\cite{DK},
this fact is mainly due to a different behaviour of the $E_{0+}$
partial wave near threshold, giving rise to substantial effects in the
case of the forward spin polarizability. While the one-pion
contribution from SAID-SP98K is $\gamma^{p}_0=-1.26$ and
$\gamma^{n}_0=-0.03 $, we obtain $\gamma^{p}_0=-0.75$ and
$\gamma^{n}_0=-0.06$ with the HDT multipoles for $E_\gamma\le 500$~MeV.
\section{t-channel dispersion integral}
\label{tchannel}
We next evaluate the $t$-channel dispersion integral in
Eq.~(\ref{eq:subt}) from $4 m_\pi^2$ to $\infty$.
The kinematics of the $t$-channel reaction $\gamma \gamma \rightarrow
N \bar N$ is shown in Fig.~\ref{fig:tkinematics}.
The subtracted dispersion integral is essentially saturated by
the imaginary part of the $t$-channel amplitude
$\gamma \gamma \rightarrow N \bar N$ due to
$\pi\pi$ intermediate states.
To calculate this contribution, we
have to construct the amplitudes $\gamma \gamma \rightarrow \pi \pi$
and $\pi \pi \rightarrow N \bar N$.
\newline
\indent
We start with the isospin and helicity structure
of the $\gamma \gamma \rightarrow \pi \pi$ amplitude, denoted by $F$.
Because of the Bose symmetry of the $\gamma\gamma$
state, only the even isospin values $I$ = 0 and 2 are possible. We
can express the charged ($\gamma \gamma \rightarrow \pi^+ \pi^-$)
and neutral ($\gamma \gamma \rightarrow \pi^0 \pi^0$) amplitudes
in terms of those with good isospin by
\begin{eqnarray}
{F^{(\pi^+ \pi^-)}}\;=\;
\sqrt{2 \over 3}\;{F^{I=0}}\;+\;\sqrt{1 \over 3}\;{F^{I=2}}
{\mathrm \;\;\;(charged\;pions) \;,}\nonumber\\
{F^{(\pi^0 \pi^0)}}\;=\;
-\sqrt{1 \over 3}\;{F^{I=0}}\;+\;\sqrt{2 \over 3}\;{F^{I=2}}
{\mathrm \;\;\;(neutral\;pions) \;.}
\label{eq:isogagapipi}
\end{eqnarray}
The reaction $\gamma \gamma \rightarrow \pi \pi$
has two independent helicity amplitudes
${F}_{\Lambda_\gamma}(t, \theta_{\pi \pi})$,
where $\Lambda_\gamma \equiv
\lambda'_\gamma - \lambda_\gamma$,
being the difference of the final photon helicity ($\lambda'_\gamma$)
and the initial photon helicity ($\lambda_\gamma$),
takes on the values 0 or 2,
depending upon whether
the photons have the same ($\Lambda_\gamma$ = 0) or
opposite ($\Lambda_\gamma$ = 2) helicities.
The $\gamma \gamma \rightarrow \pi \pi$
helicity amplitudes depend upon the c.m. energy squared $t$, and the
pion c.m. scattering angle $\theta_{\pi \pi}$.
In terms of the helicity amplitudes $F_{\Lambda_\gamma}$, the
$\gamma \gamma \rightarrow \pi \pi$ differential c.m. cross section is
given by
\begin{eqnarray}
{\left(\frac{d\sigma}{d\cos\theta_{\pi \pi}}\right)_{\rm c.m.}}\;=\;
\frac{\beta}{64 \, \pi \,t} \;
\left\{|{{F}_{\Lambda_\gamma=0}}(t,\theta_{\pi \pi})|^2\;+\;
|{{F}_{\Lambda_\gamma=2}}(t,\theta_{\pi \pi})|^2\right\} \;,
\label{eq:crossgagapipi}
\end{eqnarray}
with $\beta = \sqrt{1 - 4 m_\pi^2/t}$ the pion velocity.
In Appendix \ref{app:tchannel}, we give the partial wave expansion of
the $\gamma \gamma \rightarrow \pi \pi$ helicity amplitudes
$F^I_{\Lambda_\gamma}(t, \theta_{\pi \pi})$ for a state of isospin $I$,
and thus define the partial wave amplitudes
$F^I_{J \,\Lambda_\gamma}(t)$
(see Eqs.~(\ref{eq:partialgagapipi}) and (\ref{eq:deffgagapipi})),
where $J$ can only take on even values.
\newline
\indent
To construct the helicity amplitudes $F_{\Lambda_\gamma}$ for the
process $\gamma\gamma\,\rightarrow\,\pi\pi$,
we first evaluate the Born graphs as shown in Fig.~\ref{fig:born}.
These graphs only contribute to the charged channel
$\gamma\gamma\,\rightarrow\,\pi^+\pi^-$. The Born
contributions to the helicity amplitudes
${F}^{(\pi^+ \pi^-)}_{\Lambda_\gamma}$ are denoted as
${B}_{\Lambda_\gamma}$ and given by
\begin{eqnarray}
{B_{\Lambda_\gamma = 0}}(t,\theta_{\pi \pi})
&&\;=\;\left( 2 e^2\right)\,\frac{1\,-\,\beta^2}
{1\,-\,{\beta^2}{\cos^2}\theta_{\pi \pi}} \;,\nonumber\\
{B_{\Lambda_\gamma = 2}}(t,\theta_{\pi \pi})
&&\;=\;\left( 2 e^2\right)\,\frac{{\beta^2}{\sin^2}\theta_{\pi \pi}}
{1\,-\,{\beta^2}{\cos^2}\theta_{\pi \pi}} \;.
\label{eq:gagapipibornhel}
\end{eqnarray}
The partial wave expansion of the Born terms
$B_{J \, \Lambda_\gamma}(t)$ is discussed in
Appendix \ref{app:tchannel} (Eq.~(\ref{eq:gagapipiborn})).
As the Born amplitudes are only non-zero for the charged pion channel,
the two isospin amplitudes of Eq.(\ref{eq:isogagapipi}) are related by
\begin{equation}
{B^{I=0}_{J\Lambda_\gamma}}\;=\;\sqrt{2 \over 3}\;{B_{J\Lambda_\gamma}}\,,
\hspace{2cm}
{B^{I=2}_{J\Lambda_\gamma}}\;=\;\sqrt{1 \over 3}\;{B_{J\Lambda_\gamma}}\,.
\end{equation}
\newline
\indent
We now construct the unitarized amplitudes $F^I_{J \Lambda_\gamma}(t)$,
starting from the Born amplitudes $B^I_{J \Lambda_\gamma}(t)$ and
following the method outlined in Refs.~\cite{Morgan88,Pennington95}. We first
note that the low energy theorem requires for each partial wave that
\begin{equation}
\frac{F^{I}_{J\Lambda_\gamma}}{B^{I}_{J\Lambda_\gamma}}\,\rightarrow\,1\,,\;
{\mathrm as}\;\;t\,\rightarrow\,0\,.
\end{equation}
Next, the invariant amplitude for the process
$\gamma\gamma\,\rightarrow\,\pi\pi$ is assumed to have Mandelstam
analyticity. Each partial wave then has a right-hand cut from
$t\,=\,4m^2_\pi$ to $+\infty$ and a left-hand cut from $t\,=\,-\infty$
to $0$. Though the Born amplitude is real for all values of $t$, its
partial waves are complex below $t=0$. The
partial waves of the full amplitude have no other sources of
complexity in this region, and so we can write a
DR for the difference of the full and the Born amplitudes,
\begin{eqnarray}
\frac{F^{I}_{J\Lambda_\gamma}(t)\,-\,B^{I}_{J\Lambda_\gamma}(t)}
{t(t\,-\,4{m^2_\pi})^{J\over2}}\;=\;\frac{1}{\pi}\;
{\int^\infty_{4m^2_\pi}}\;dt'\;
\frac{{\mathrm Im}F^{I}_{J\Lambda_\gamma}(t')}
{t'(t'\,-\,4{m^2_\pi})^{J\over2}(t'-t)}\,,
\label{eq:drgagapipi}
\end{eqnarray}
with an additional factor of $({t(t\,-\,4{m^2_\pi})^{J\over2}})^{-1}$
providing the right asymptotics for the convergence of the integral.
The next step is to evaluate the imaginary part of
the amplitude in Eq.~(\ref{eq:drgagapipi}).
To do this, we exploit the unitarity condition
\begin{eqnarray}
{\mathrm Im}\,{{F}^I_{J \Lambda_\gamma}}
(\gamma\gamma\,\rightarrow\,\pi\pi)\;=\;
{\sum_n}\;{\rho_n}\;
{{F}^{I \ast}_{J \Lambda_\gamma}}(\gamma\gamma\,\rightarrow\,n{)}\;
{{\cal{I}}^I_{J}}(n\,\rightarrow\,\pi\pi)\,,
\label{eq.Im}
\end{eqnarray}
where $\rho_n$ are the appropriate kinematical and isospin factors for
the intermediate channels $n$, and ${\cal I}(n \rightarrow \pi\pi)$ is
a hadronic amplitude.
Below the next inelastic threshold, it follows from unitarity that the phase
${\phi^{I\;(\gamma\gamma\,\rightarrow\,\pi\pi)}_{J}}$ of
each partial wave $F^I_{J \, \Lambda_\gamma}$
is equal to the phase $\delta^{I\,J}_{\pi\pi}$
of the corresponding $\pi\pi\,\rightarrow\,\pi\pi$ partial wave,
\begin{eqnarray}
{\mathrm Im}\,{{F}^I_{J \Lambda_\gamma}}(\gamma\gamma\,\rightarrow\,\pi\pi)
&&\;=\;
{\rho_{\pi\pi}}\;
{{F}^{I \ast}_{J \Lambda_\gamma}}(\gamma\gamma\,\rightarrow\,\pi\pi{)}\;
{{\cal{I}}^I_{J}}(\pi\pi\,\rightarrow\,\pi\pi)\nonumber\\
&&\Downarrow\nonumber\\
{\phi^{I\;(\gamma\gamma\,\rightarrow\,\pi\pi)}_{J}}(t)&&\;=\;
{\delta^{IJ}_{\pi\pi}}(t)\,.
\end{eqnarray}
This fact can be incorporated into the so-called Omn\`{e}s function,
which is constructed to have the phase of the $\pi\pi$ scattering
amplitude above $\pi\pi$ threshold, and to be real otherwise,
\begin{equation}
{\Omega^I_J}(t)\;=\;\exp{\left[
\frac{t}{\pi}\;{\int_{4m_\pi^2}^{\infty}}\,dt'\,
\frac{{\delta^{IJ}_{\pi\pi}}(t')}{t'(t'\,-\,t\,-\,i\varepsilon)}\right]}
\;.
\end{equation}
The function ${{F}^I_{J \Lambda_\gamma}}{(\Omega^I_J{)^{-1}}}(t)$ is
by construction real above $\pi\pi$ threshold,
but complex below threshold due to
the complexity of the Born partial waves ${B^{I}_{J\Lambda_\gamma}}$.
Hence we can write a dispersion relation for
$\left[{{F}^I_{J \Lambda_\gamma}}\;-\;{B^I_{J \Lambda_\gamma}}\right]
{(\Omega^I_J{)^{-1}}}(t) / t (t - 4 m_\pi^2)^{J/2}$,
\begin{eqnarray}
&&{{F}^I_{J \Lambda_\gamma}}(t)\;=\;\nonumber\\
&&{\Omega^I_J}(t)\;\left\{
{B^I_{J \Lambda_\gamma}}(t)\,{\mathrm Re}\left[(\Omega^I_J{)^{-1}}(t)\right]
\;-\;
\frac{t(t\,-\,4m^2_{\pi})^{J/2}}{\pi}\,{\int^{\infty}_{4m^2_{\pi}}}\,dt'\,
\frac{{B^I_{J \Lambda_\gamma}}(t')
\,{\mathrm Im}\left[{(\Omega^I_J{)^{-1}}}(t')\right]}
{t'(t'\,-\,4m^2_{\pi})^{J/2}(t'\,-\,t)}\right\} \;.
\label{eq:gagapipidisprel}
\end{eqnarray}
For $t > 4 m_\pi^2$, this integral is understood to be a principal value
integral, which we implement by subtracting the integrand at $t' = t$.
In this way we obtain a regular integral, which
can be performed without numerical problems,
\begin{eqnarray}
&&{{F}^I_{J \Lambda_\gamma}}(t)\;=\;
{\Omega^I_J}(t)\;\left\{
{B^I_{J \Lambda_\gamma}}(t)\,\left(
{\mathrm Re}\!\left[(\Omega^I_J{)^{-1}}(t)\right]\;+\;
{\mathrm Im}\!\left[(\Omega^I_J{)^{-1}}(t)\right]\,
{1\over\pi}\,\ln(\frac{t}{4m^2_{\pi}}\,-1\,)
\right)\right.\nonumber\\
&&\left.\;-\;
\frac{t(t\,-\,4m^2_{\pi})^{J/2}}{\pi}\,{\int^{\infty}_{4m^2_{\pi}}}
\,\frac{dt'}{t'(t'-t)} \;
\left(\frac{{B^I_{J \Lambda_\gamma}}(t')
\,{\mathrm Im}\!\left[{(\Omega^I_J{)^{-1}}}(t')\right]}
{(t'\,-\,4m^2_{\pi})^{J/2}}\,-\,
\frac{{B^I_{J \Lambda_\gamma}}(t)
\,{\mathrm Im}\!\left[{(\Omega^I_J{)^{-1}}}(t)\right]}
{(t\,-\,4m^2_{\pi})^{J/2}} \right)
\right\} \;.
\end{eqnarray}
In our formalism,
the s($J=0$)- and d($J=2$)-waves are unitarized. For the s- and d-wave
$\pi \pi$ phaseshifts, we use the solutions that were determined in
Ref.~\cite{Frog77}.
For the higher partial waves, the corresponding $\pi \pi$ phaseshifts
are rather small and are not known with good precision.
Therefore, we will approximate all higher partial waves $(J \geq 4)$
by their Born contribution.
The full amplitudes for the charged and neutral channels
can then be cast into the forms
\begin{eqnarray}
&&{{F}^{(\pi^+ \pi^-)}_{\Lambda_\gamma}}(t,\theta_{\pi \pi})
\,=\,{{B}_{\Lambda_\gamma}}(t,\theta_{\pi \pi}) \nonumber\\
&&\hspace{1.2cm}+\;
\sum_{J=0,2}\!\sqrt{2J+1}\sqrt{\frac{(J-\Lambda_\gamma)!}
{(J+\Lambda_\gamma)!}}\left[
\sqrt{2\over 3}{{F}^{I=0}_{J \Lambda_\gamma}}(t)\,+\,
\sqrt{1\over 3}{{F}^{I=2}_{J \Lambda_\gamma}}(t)\,-\,
{{B}_{J \Lambda_\gamma}}(t)\right]
P_J^{\Lambda_\gamma} (\cos \theta_{\pi \pi}) \;,\\
&&{{F}^{(\pi^0 \pi^0)}_{\Lambda_\gamma}}(t,\theta_{\pi \pi})\,=\,
\sum_{J=0,2}\sqrt{2J+1}\sqrt{\frac{(J-\Lambda_\gamma)!}
{(J+\Lambda_\gamma)!}}\left[
-\sqrt{1\over 3}{{F}^{I=0}_{J \Lambda_\gamma}}(t)\,+\,
\sqrt{2\over 3}{{F}^{I=2}_{J \Lambda_\gamma}}(t)\right]
P_J^{\Lambda_\gamma} (\cos \theta_{\pi \pi}) \;.
\end{eqnarray}
\newline
\indent
The two-pion intermediate contribution holds to good precision up to
$K \bar K$ threshold ($\approx$~1~GeV$^2$), because the four-pion
intermediate state couples only weakly and gives only small
inelasticities in the $\pi \pi$ phaseshifts.
\newline
\indent
In Figs.~\ref{fig:gagapipitot} and \ref{fig:diffpippim}, we show our
results for the total and differential
$\gamma\gamma\rightarrow\pi^+ \pi^-$ cross sections
and a comparison to the existing data.
In the threshold region, the charged pion cross sections are clearly
dominated by the Born graphs of Fig.~\ref{fig:born} because of the
vicinity of the pion pole in the $t$-channel of the
$\gamma\gamma\rightarrow\pi^+ \pi^-$ process.
However, the results for the
unitarized calculation show that s-wave rescattering is not
negligible but leads to a considerable enhancement at energies just above
threshold. Besides the low energy structure, driven by the Born
terms, the $\gamma\gamma\rightarrow\pi \pi$ process has a prominent
resonance structure at higher energies corresponding to
excitation of the isoscalar $f_2$(1270) resonance, with mass $m_{f_2}$
= 1275 MeV and width $\Gamma_{f_2}$ = 185.5 MeV \cite{PDG98}.
The $f_2$ resonance shows up in the partial wave $F_{J=2 \,
\Lambda_\gamma = 2}$ as outlined in Appendix~\ref{app:f2}.
Therefore, the most efficient way to unitarize
this particular partial wave is to make a Breit-Wigner ansatz for
the $f_2$ excitation, which is described in Appendix~\ref{app:f2}
where we also give some details of the formalism for a spin-2
particle. The Breit-Wigner ansatz for the $f_2$ contribution to the
partial wave $F_{J=2 \, \Lambda_\gamma = 2}$ depends upon the
couplings $f_2 \pi \pi$ and $f_2 \gamma \gamma$.
The coupling $f_2 \pi \pi$ is known from the decay
$f_2 \rightarrow \pi \pi$ and is taken from Ref.~\cite{PDG98}. The
coupling $f_2 \gamma \gamma$ is then fitted to the
$\gamma\gamma\rightarrow\pi\pi$ cross section at the $f_2$ resonance
position, and is consistent with the value quoted in Ref.~\cite{PDG98}.
The resulting amplitude, consisting of unitarized s-wave,
$f_2$ excitation and Born terms for all other partial waves (with $J
\geq 4$) is seen from Figs.~\ref{fig:gagapipitot} and
\ref{fig:diffpippim} to give a rather good description of the
$\gamma\gamma\rightarrow\pi^+ \pi^-$ data up to $W_{\pi \pi} \simeq$
1.8 GeV. Only in the region $W_{\pi \pi} \approx$ 0.7 - 0.8 GeV, does
our description slightly overestimate the data.
\newline
\indent
Having constructed the $\gamma\gamma\rightarrow\pi\pi$ amplitudes, we
next need the $\pi\pi \rightarrow N \bar N$ amplitudes
in order to estimate the contribution of the $\pi \pi$ states to the
$t$-channel dispersion integral for Compton scattering.
As we only kept s- and d-waves for $\gamma \gamma \rightarrow \pi
\pi$, we will only need the s- and d-waves ($J$ =
0, 2) for $\pi \pi \rightarrow N \bar N$.
For every partial wave $J$, there are two independent
$\pi \pi \rightarrow N \bar N$ helicity amplitudes
$f_{\pm}^J(t)$, depending on
whether the nucleon and anti-nucleon have the same ($f_{+}^J(t)$) or
opposite ($f_{-}^J(t)$) helicities. We refer the reader to
Appendix \ref{app:tchannel} (Eqs.~(\ref{eq:partialpipinnbar}) and
(\ref{eq:frazerfulco})) for details.
In this work, we take the s- and d-waves
from the work of H\"ohler and collaborators~\cite{Hoehler83},
in which the lowest $\pi\pi \rightarrow N \bar N$ partial wave
amplitudes were constructed from a partial wave solution of
pion-nucleon scattering, by use of the $\pi \pi$ phaseshifts of
Ref.~\cite{Frog77}, which we also used to construct the
$\gamma \gamma \rightarrow \pi \pi$ amplitudes.
In Ref.~\cite{Hoehler83}, the
$\pi \pi \rightarrow N \bar N$ amplitudes are given
for $t$ values up to $t \approx 40 \cdot m_\pi^2 \approx$
0.78~GeV$^2$, which will serve well for our purpose since the subtracted
$t$-channel dispersion integral will have converged much below this
value as shown in the following.
\newline
\indent
Finally, we can now combine
the $\gamma \gamma \rightarrow \pi \pi$ and $\pi \pi \rightarrow N
\bar N$ amplitudes to construct the
discontinuities of the Compton amplitudes across the $t$-channel cut.
In Appendix \ref{app:tchannel}, we show in detail how the Compton
invariant amplitudes $A_1,...,A_6$ are expressed by the
$t$-channel ($\gamma \gamma \rightarrow N \bar N$) helicity
amplitudes. Through unitarity we then express the imaginary parts of
these $t$-channel ($\gamma \gamma \rightarrow N \bar N$) helicity
amplitudes in terms of the $\gamma \gamma \rightarrow \pi \pi$ and
$\pi \pi \rightarrow N \bar N$ amplitudes. We
finally express the discontinuities Im$_t A_i$ of the invariant
amplitudes $A_i$ ($i$ = 1,...,6) in terms of the corresponding
$\gamma \gamma \rightarrow \pi \pi$ and $\pi \pi \rightarrow N
\bar N$ partial wave amplitudes (see Eq.~(\ref{eq:ima2pi})).
As we restrict ourselves to s- and d-wave intermediate states
in the actual calculations, we give here the expressions at
$\nu = 0$, including s- and d-waves only,
that are needed for the subtracted $t$-channel dispersion
integral of Eq.~(\ref{eq:subt})~,
\begin{eqnarray}
{\mathrm Im}_t A_1 (\nu = 0, t)^{2 \pi} &=& - \;
\sqrt{{t/4 - m_\pi^2} \over t} \;
{1 \over {t \, (M^2 - t/4)}}\;
F_{0\,\Lambda_\gamma = 0}(t) \; f^{0*}_+(t) \; \nonumber\\
&& - \; \left( {{t/4 - m_\pi^2} \over t} \right)^{3/2} \;
{\sqrt{5} \over 2}\;F_{2\,\Lambda_\gamma = 0}(t) \; f^{2*}_+(t) \;, \nonumber\\
{\mathrm Im}_t A_2 (\nu = 0, t)^{2 \pi} &=& 0 \;, \nonumber\\
{\mathrm Im}_t A_3 (\nu = 0, t)^{2 \pi} &=& -
\left( {{t/4 - m_\pi^2} \over t} \right)^{3/2} \,
{{M^2} \over {(M^2 - t/4)}} \,
{\sqrt{5} \over 2} \; F_{2\,\Lambda_\gamma = 2}(t) \,
\left\{ \sqrt{{3 \over 2}} \, f^{2*}_+(t)
\,-\, M \, f^{2*}_-(t) \right\}\, , \nonumber\\
{\mathrm Im}_t A_4 (\nu = 0, t)^{2 \pi} &=& 0 \;, \nonumber\\
{\mathrm Im}_t A_5 (\nu = 0, t)^{2 \pi} &=& -
\left( {{t/4 - m_\pi^2} \over t} \right)^{3/2} \; M \;
\sqrt{{{15} \over 2}} \;
F_{2\,\Lambda_\gamma = 0}(t) \; f^{2*}_-(t) \;, \nonumber\\
{\mathrm Im}_t A_6 (\nu = 0, t)^{2 \pi} &=& -
\left( {{t/4 - m_\pi^2} \over t} \right)^{3/2} \; M \;
{\sqrt{5} \over 2} \; F_{2\,\Lambda_\gamma = 2}(t) \; f^{2*}_-(t) \;.
\label{eq:ima2pisd}
\end{eqnarray}
The reader should note that the s-wave $\pi \pi$
intermediate state only contributes
to the amplitude $A_1$. It is the $t$-dependence of
this $I = J = 0$ $\pi \pi$ state in the
$t$-channel that is approximated
in Ref.\cite{lvov97} and parametrized by a ``sigma'' pole.
The d-wave $\pi \pi$ intermediate state gives rise to imaginary parts
for the amplitudes $A_1, A_3, A_5$ and $A_6$. The amplitude $A_2$ (at
$\nu = 0$) corresponds to the $t$-channel exchange of an object with
the quantum
numbers of the pion (e.g. $\pi^0$ pole in Eq.~(\ref{eq:piopole})).
Therefore two-pion intermediate states do not have the quantum numbers
to contribute to the amplitude $A_2$.
The imaginary part of $A_4$ receives only contributions from
$\pi \pi$ intermediate states with $J \geq 4$ (see Eq.~(\ref{eq:ima2pi})) and
therefore is zero in our description, as we keep only s- and d-waves.
\newline
\indent
In Fig.~\ref{fig:tchannelconv} we show the convergence of the
$t$-channel integral from $4 m_\pi^2$ to $\infty$
in the subtracted DR of Eq.~(\ref{eq:subt}). We do so by calculating the
dispersion integral as function of the upper integration limit $t_{\rm
upper}$ and by showing the ratio with the integral for
$t_{\rm upper}$ = 0.78~GeV$^2$. The latter value corresponds to the highest
$t$ value for which the $\pi \pi \rightarrow N \bar N$ amplitudes are
given in Ref.~\cite{Hoehler83}. One clearly sees from
Fig.~\ref{fig:tchannelconv} that the unsubtracted $t$-channel DR shows
only a slow convergence, whereas the subtracted $t$-channel DR has
already reached its final value, within the percent level,
at a $t$ value as low as 0.4~GeV$^2$.
\section{Results and discussion}
\label{results}
In this section we shall present our results for Compton
scattering off the nucleon in the dispersion
formalism presented above.
The real and imaginary parts of the six Compton amplitudes are
displayed in Fig.~\ref{fig:a1a6reim}. Note that for the real part, we
only show the subtracted $s$-channel integral of
Eq.~(\ref{eq:sub}). As can be seen from Fig.~\ref{fig:a1a6reim}, these
amplitudes show strong oscillations due to interference effects
between different pion photoproduction multipoles, in particular for
threshold pion production by $E_{0+}$ and $\Delta$-excitation by
$M_{1+}$.
In Figs.~\ref{fig:threshold_gpi} and~\ref{fig:threshold_amb} we show our
predictions in the subtracted DR formalism and compare them with
the available Compton data on the proton
below pion threshold. These data were used in
Ref.~\cite{McGibbon} to determine the scalar
polarizabilities $\alpha$ and $\beta$ through a global fit,
with the results given in Eq.~(\ref{eq1.1}). In the analysis of
Ref.~\cite{McGibbon}, the unsubtracted DR formalism was used and the
asymptotic contributions (Eq.~(\ref{eq:aintas})) to the invariant
Compton amplitudes $A_1$ and $A_2$ were parametrized. In particular,
$A^{as}_2$ was described by the $\pi^0$ pole, which yields the value
$\gamma_\pi \simeq - 45$. The free parameter entering in $A^{as}_1$
was related to $\alpha - \beta$, for which the
fit obtained the value $\alpha - \beta \simeq$ 10.
Keeping $\alpha - \beta$ fixed at that value, we demonstrate in
Fig.~\ref{fig:threshold_gpi}
that the sensitivity to $\gamma_\pi$ is not at all negligible,
especially at the backward angles and the higher energies.
Although we do not intend to give a best fit at
the present stage, the subtracted DR formalism allows one
to directly use the values $\alpha - \beta$ and
$\gamma_\pi$ as fit parameters, as is obvious from Eqs.~(\ref{eq:sub},
\ref{eq:subt}). We
investigate this further in Fig.~\ref{fig:threshold_amb}, where we show
our results for different $\alpha - \beta$ and
for a fixed value of $\gamma_\pi = -37$,
which is consistent with the heavy baryon ChPT
prediction \cite{HHKK} and close to the value obtained in Ref.~\cite{LN}
in a backward DR formalism. For that value of $\gamma_\pi$,
a better description of the data (in particular at the backward angle)
seems to be possible by using a smaller value for $\alpha - \beta$
than determined in Ref.~\cite{McGibbon}. For a more reliable
extraction of the polarizabilities,
more accurate data over the whole angular and energy
range are necessary. Recently, Compton data were taken at MAMI
over a wide angular range below pion threshold \cite{Olmos}. It will
be interesting to perform a best fit for $\alpha - \beta$ and
$\gamma_\pi$ with such an extended data base.
As one moves to energies above pion threshold, the Compton cross section rises
rapidly because of the unitarity coupling to the much stronger
pion photoproduction channel. Therefore this higher energy region is
usually considered less `pure' to extract polarizabilities because the
procedure would require a rather precise knowledge of pion
photoproduction. With the quite accurate pion photoproduction data
on the proton that have become available in
recent years, the energy region above pion threshold for the Compton channel
could however serve as a valuable complement to determine
the polarizabilities, provided
one can minimize the model uncertainties in the dispersion formalism.
In this work, we use the most recent information on the pion
photoproduction channel by taking the HDT \cite{HDT}
multipoles at energies $E_\gamma \leq$ 500 MeV and
the SAID-SP98K solution \cite{said} at higher energies.
In addition, as previously shown in Fig.~\ref{fig:a1a2conv},
the subtracted dispersion relations are practically
saturated by the one-pion channel for photon energies through
the $\Delta$ region, which minimizes the uncertainty due to
the modeling of the two-pion photoproduction channels.
In Figs.~\ref{fig:compton_delta_gpi} and~\ref{fig:compton_delta_amb}
we display the sensitivity of the Compton cross sections to
$\gamma_\pi$ and $\alpha - \beta$ in the lower part of the
$\Delta$ region in comparison with the available data.
As can be seen from Fig.~\ref{fig:compton_delta_gpi}, these
data are quite sensitive to the backward spin
polarizability $\gamma_\pi$. This sensitivity was exploited in
Ref.~\cite{Tonnison} within the context of an unsubtracted DR formalism,
and the value $\gamma_\pi \simeq -27$ was extracted from the LEGS 97
data, which are shown at the higher energies in
Fig.~\ref{fig:compton_delta_gpi}.
Our results for the subtracted DR are obtained in
Fig.~\ref{fig:compton_delta_gpi} by variation of $\gamma_\pi$ at
fixed $\alpha - \beta$ = 10, and by variation of $\alpha - \beta$ at
fixed $\gamma_\pi$ = -37 in Fig.~\ref{fig:compton_delta_amb}.
For $\gamma_\pi$ we show the results for values between $\gamma_\pi = -27$
and $\gamma_\pi = -37$. One sees that the lower energy data
($E_\gamma$ = 149 MeV and 182 MeV) can be easily described by the larger
values of $\gamma_\pi$ if $\alpha - \beta$ decreases to some value below 10.
On the other hand, the higher energy data ($E_\gamma$ = 230 MeV and 287
MeV) seem to favor a smaller value of $\gamma_\pi$, and so far we
confirm the conclusion reached in Ref.~\cite{Tonnison}.
However, we have to point out that, at
these higher energies, the data around 90$^{\rm o}$
cannot be described in our subtracted DR formalism for reasonable
values of $\alpha - \beta$ and $\gamma_\pi$.
This is also seen in Fig.~\ref{fig:fixed_th} at two fixed angles, now
as function of the energy throughout the $\Delta$ resonance region,
for the MAMI data at 75$^{\rm o}$ \cite{Peise} and 90$^{\rm o}$
\cite{Molinari}.
It is again obvious at these angles that the sensitivity to
$\gamma_\pi$ is quite small. Therefore the physics of these data
is basically driven by pion photoproduction. With the multipoles
used here, the 75$^{\rm o}$ data are well described, but
our prediction falls below the 90$^{\rm o}$ data
on the left shoulder of the $\Delta$ resonance.
In the same energy region there exist also both differential cross
section and photon asymmetry data from LEGS
\cite{Blanpied97} by use of the laser backscattering technique.
In Fig.~\ref{fig:asymm} we compare our predictions with these data.
One finds that at both energies ($E_\gamma$ = 265 MeV and 323 MeV) our
subtracted DR formalism provides a good description
of the asymmetries which display only little sensitivity on $\gamma_\pi$,
but underestimates the absolute values of the cross sections. In
particular close to the resonance position at $E_\gamma$ = 323 MeV,
the subtracted DR formalism does not allow us to find any reasonable
combination of $\gamma_\pi$ and $\alpha - \beta$ to describe these
data. Therefore, within the present subtracted DR formalism,
the actual data situation at these higher energies
does not seem to be conclusive to reliably decide on a value of
$\gamma_\pi$. Since the uncertainties due to two-pion and heavier
meson photoproduction are less than 1 \% in our subtracted DR
formalism, the only possibility to describe the $E_\gamma$ = 323 MeV
LEGS data would be an increase of the HDT $M_{1+}$ multipole by about
2.5 \% (see the dotted lines in Fig.~\ref{fig:asymm}). Indeed such a
fit was obtained by Tonnison et al. \cite{Tonnison} by use of the
LEGS pion photoproduction multipole set of Ref.\cite{Blanpied97}
for photon energies between 200 and 350 MeV and the SAID-SM95 multipole
solution \cite{said} outside this interval. However the more recent
SAID-SP98K solution is in very close agreement with the HDT multipoles
in the $\Delta$ region and hence the predictions also fall below the
data at 323 MeV in Fig.~\ref{fig:asymm}.
Before coming to any conclusions, we like to point out that new Compton data
on the proton in and above the $\Delta$-resonance region
and over a wide angular range
have been measured recently at MAMI and reported preliminary \cite{LARA}.
These new data will be most valuable to check if a systematic and
consistent trend becomes visible between the data sets at the lower
energies and in the $\Delta$ region.
Finally, in Fig.~\ref{fig:doublepol} we show that double polarization
observables will be ultimately necessary in order to reliably extract
the polarizabilities $\alpha - \beta$ and $\gamma_\pi$.
In particular, an experiment with a circularly polarized photon
and a polarized proton target displays quite some sensitivity on the
backward spin polarizability $\gamma_\pi$, especially at the somewhat
higher energies of $E_\gamma \approx$~230 MeV.
Such a measurement is more selective to $\gamma_\pi$ due to a lesser
sensitivity to $\alpha - \beta$ (see Fig.~\ref{fig:doublepol}). It
will be indeed a prerequisite to disentangle the scalar and vector
polarizabilities of the nucleon.
\section{Conclusions}
\label{conclusions}
In this work we have presented a formalism of fixed-$t$ subtracted
dispersion relations for Compton scattering off the nucleon at
energies $E_\gamma \leq$ 500 MeV. Due to the subtraction, the
$s$-channel dispersion integrals converge very fast and are
practically saturated by the $\pi N$ intermediate states.
Because of the use of subtracted DR we have minimized the
uncertainty from multi-pion and heavier meson intermediate states.
Hence this formalism provides a direct cross-check between Compton
scattering and one-pion photoproduction.
We have described this dominant
$\gamma N \rightarrow \pi N \rightarrow \gamma N$
contribution by using the recent pion photoproduction multipoles of HDT.
\newline
\indent
To calculate the functional dependence of the subtraction
functions on the momentum transfer $t$, we
have included experimental information on the $t$-channel process
through $\pi \pi$ intermediate states as
$\gamma \gamma \rightarrow \pi \pi \rightarrow N \bar N$. We have constructed a
unitarized amplitude for the $\gamma \gamma \rightarrow \pi \pi$
subprocess and found a good description of available data. In this
way, we have largely avoided the uncertainties in Compton scattering
associated with the two-pion continuum in the $t$-channel, which was
often modeled through the exchange of a somewhat fictitious $\sigma$-meson.
As the polarizabilities directly
enter as subtraction constants in the subtracted DR formalism,
it can be used to extract the nucleon
polarizabilities from data with a minimum of model dependence.
We have demonstrated the sensitivity to the polarizabilities
$\alpha - \beta$ and $\gamma_\pi$ of existing Compton data on the
proton both below pion threshold as well as in the $\Delta$ resonance region.
The effects of the polarizabilities $\alpha - \beta$ and $\gamma_\pi$
on the Compton cross sections are strongly correlated.
Hence these polarizabilities can only be determined
simultaneously from unpolarized observables even below pion
threshold. When comparing the
subtracted DR formalism with the existing data in the $\Delta$
resonance region, we have found that the actual data situation at
these higher energies does not seem to be conclusive to reliably
decide on a value for $\gamma_\pi$. However new Compton data on the
proton both below pion threshold and in the $\Delta$ region are
actually under analysis and will extend considerably the experimental data
base to fit the proton polarizabilities.
We have also argued that double polarization observables
will ultimately be necessary to extract reliably $\alpha - \beta$ and
$\gamma_\pi$. In particular we have shown that experiments of
polarized photons on polarized protons show rather large sensitivity
to $\gamma_\pi$ at energies around $E_\gamma \approx$ 230 MeV and at
backward angles. Therefore such polarization experiments hold the promise to
disentangle scalar and vector polarizabilities of the nucleon and
to quantify the nucleon spin response in an external electromagnetic field.
\section*{acknowledgments}
The authors are grateful to G. Krein, A. L'vov
and members of the A2-Collaboration, in particular J. Ahrens and
V. Olmos de L\'eon, for useful discussions.
This work was supported by the Deutsche Forschungsgemeinschaft
(SFB 443) and in part by the Marie Curie TMR program (contract
ERBFMBICT972758).
\newpage
\begin{appendix}
\section{The Mandelstam plane -- physical and spectral regions for
Compton scattering}
\label{app:spectralcompton}
The kinematics of Compton scattering, $\gamma(q)N(p)\rightarrow
\gamma(q')N(p')$, can be described in terms of the familiar Mandelstam
variables,
\begin{eqnarray}
\label{eqA1}
s=(q+p)^2\ ,\ \ t=(q-q')^2\ ,\ \ u=(q-p')^2\ ,
\end{eqnarray}
with the constraint
\begin{eqnarray}
\label{eqA2}
s+t+u=2M^2\ .
\end{eqnarray}
Furthermore we introduce the coordinate $\nu$ perpendicular to $t$,
\begin{eqnarray}
\label{eqA3}
\nu=\frac{s-u}{4M}=E_\gamma+\frac{t}{4M}\ .
\end{eqnarray}
In these equations, $E_\gamma$ is the photon energy in the $lab$ frame
and $M$ the nucleon mass.
The boundaries of the physical regions in the $s$, $u$ and $t$
channels are determined by the zeros of the Kibble function $\Phi$,
\begin{eqnarray}
\label{eqA4}
\Phi(s,t,u)=t(us-M^4)=0\ .
\end{eqnarray}
The 3 physical regions are shown by the horizontally hatched areas in
Fig.~1. The vertically hatched areas are the regions of non-vanishing
double spectral functions. These spectral regions are those regions
in the Mandelstam plane where two of the three variables
$s, t$ and $u$ take on values that
correspond with a physical (i.e. on-shell) intermediate state.
The boundaries of these regions follow from
unitarity. As discussed in Ref.~\cite{Hoehler83}, it is sufficient to
consider two-particle intermediate states in all channels. Since these
boundaries depend only on the masses, they are the same for all 6
amplitudes $A_i$. In the Mandelstam diagram of Fig.~1 they are
symmetric to the line $\nu=0$ due to crossing symmetry. For the
spectral function $\rho_{su}$ we obtain the boundary
\begin{eqnarray}
\label{eqA5}
b_I(u,s)=b_I(s,u)=[s-(M+m_\pi)^2]
[u-(M+m_\pi)^2]
-(m_\pi^2 + 2 \,M \,m_\pi)^2=0\ ,
\end{eqnarray}
and for the spectral function $\rho_{st}$ we find
\begin{eqnarray}
\label{eqA6}
b_{II}(s,t)=
(t-4 m_\pi^2)[s-(M+m_\pi)^2]
[s-(M-m_\pi)^2]-8m_\pi^4(s+M^2-m_\pi^2/2)=0
\end{eqnarray}
The boundary of the spectral function $\rho_{ut}$ follows from
crossing symmetry. We also note that these boundaries are obtained for
the isovector photon, which couples to a $\pi^+\pi^-$ pair. The
corresponding boundaries for the isoscalar photon are inside the
boundaries of Eqs.~(\ref{eqA5}) and (\ref{eqA6}), because it couples to
3 pions.
\section{{\it t}-channel helicity amplitudes for Compton scattering}
\label{app:tchannel}
The $t$-channel helicity amplitudes for Compton scattering
can be expressed in the orthogonal basis of Prange \cite{Prange58}
in the following form:
\begin{eqnarray}
T_{\lambda_N \lambda_{\bar{N}},\,\lambda'_{\gamma}\lambda_{\gamma}}^t\,
(\nu, t)\;&=&\;(-1)^{\frac{1}{2}-\lambda_{\bar{N}}}\;
{\varepsilon'}_{\mu}(q',{\lambda'}_{\gamma}) \;
{\varepsilon}_{\nu}(q,\lambda_{\gamma}) \nonumber\\
&&\times \;{\bar u}(\vec p~', \lambda_N)\, \left\{
\,- \, {{\tilde{P'}^{\mu}\,\tilde{P'}^{\nu}} \over {\tilde{P'}^{2}}} \;
\left(\, T_{1} \,+\, \tilde{K}\!\!\!\!/ \,T_{2}\, \right) \right.\nonumber\\
&&\hspace{2.5cm}- \, {{\tilde{N}^{\mu}\,\tilde{N}^{\nu}}\over \tilde{N}^{2}}\,
\left(\, T_{3} \,+\, \tilde{K}\!\!\!\!/ \,T_{4}\,\right) \nonumber\\
&&\hspace{2.5cm} +\,i\,
{{\tilde{P'}^{\mu}\tilde{N}^{\nu}\,-\,\tilde{P'}^{\nu}\tilde{N}^{\mu}}\over
{\tilde{P'}^{2}\,\tilde{K}^{2}}} \,\gamma_{5}\,T_{5} \nonumber\\
&&\hspace{2.5cm} \left. +\,i\,
{{\tilde{P'}^{\mu} \tilde{N}^{\nu}\,+\,\tilde{P'}^{\nu} \tilde{N}^{\mu}} \over
{\tilde{P'}^{2}\,\tilde{K}^{2}} }\,\gamma_{5}\,\tilde{K}\!\!\!\!/ \,T_{6}
\right\}\;v(- \vec p, \lambda_{\bar N}) \; ,
\label{eq:thelcov}
\end{eqnarray}
where
\begin{eqnarray}
\tilde{P'}^{\mu} &=& \tilde{P}^{\mu}\,-\,\tilde{K}^{\mu}\,
{{\tilde{P} \cdot \tilde{K}} \over {\tilde{K}^2}} \;,
\hspace{0.5cm} \tilde{P} = {1 \over 2}\,(-\,p \,+\, p')\;,
\hspace{0.5cm} \tilde{K} = {1 \over 2}\,(q \,-\, q'),\nonumber\\
\tilde{N}^{\mu}&=&{\epsilon}^{\mu \alpha \beta \gamma}\,
\tilde{P'}_{\alpha}\,\tilde{Q}_{\beta}\,\tilde{K}_{\gamma} \;, \hspace{.5cm}
\tilde{Q} = {1 \over 2}\,(-\,p \,-\, p')\,=\,{1 \over 2}\,(-\,q' \,-\, q)\, ,
\end{eqnarray}
and using the convention $\epsilon_{0 1 2 3} = + 1$.
\newline
\indent
In the c.m. system of the $t$-channel process $\gamma \gamma
\rightarrow N \bar N$ (see Fig.~\ref{fig:tkinematics} for the
kinematics), we choose the photon momentum $\vec q_t$
(helicity $\lambda'_{\gamma}$) to point in the $z$-direction and the
nucleon momentum $\vec p~' = \vec p_t$ in the $xz$ plane at an angle
$\theta_t$ with respect to the $z$-axis (the anti-nucleon momentum is
then given by $- \vec p = - \vec p_t$). In this frame, the helicity
amplitudes of Eq.(\ref{eq:thelcov}) can be cast into the form
\begin{eqnarray}
T_{\lambda_N \lambda_{\bar{N}},\,\lambda'_{\gamma}\lambda_{\gamma}}^t\,
(\nu, t)
\;=\;(-1)^{\frac{1}{2}-\lambda_{\bar{N}}}\,\bar{u}(\vec{p_t},\lambda_N)\;
&&\left\{ \,-\,\frac{1}{2}\,{\lambda'}_{\gamma} \lambda_{\gamma}
\,\left(T_1 \,+\,|\vec{q_t}|\,\gamma^3 \, T_2\right) \right.\nonumber\\
&&\hspace{.4cm}-\;{1 \over 2}\;
\left( T_3 \,+\,|\vec{q_t}|\,\gamma^3 \,T_4\right) \nonumber\\
&&\hspace{0.4cm}-\,
\frac{1}{2}\;({\lambda'}_{\gamma} + \lambda_{\gamma})
\;\gamma_5 \,T_5 \nonumber\\
&&\left.\hspace{.4cm}-\,\frac{1}{2}\;({\lambda'}_{\gamma} - \lambda_{\gamma})
\,\gamma_5 \,|\vec{q_t}|\,\gamma^3\,T_6 \;\right\}\;
v(-\vec{p_t},\lambda_{\bar{N}}).
\end{eqnarray}
Under parity transformation, these amplitudes behave as
\begin{eqnarray}
T_{\lambda_N\lambda_{\bar{N}},\,\lambda'_{\gamma}\lambda_{\gamma}}^t\,(\nu, t)
\;=\; (-1)^{\Lambda_{N} - \Lambda_{\gamma}}\;
T_{-\lambda_N-\lambda_{\bar{N}},\,-\lambda'_{\gamma}-\lambda_{\gamma}}^t\,
(\nu, t) \;,
\end{eqnarray}
with the helicity differences $\Lambda_\gamma$ and $\Lambda_N$ given
by $\Lambda_\gamma = \lambda'_\gamma - \lambda_\gamma$
(with $\Lambda_\gamma$ = 0 or 2) and $\Lambda_N
= \lambda_N - \lambda_{\bar N}$ (with $\Lambda_N$ = 0 or 1) respectively.
However, the invariant amplitudes $T_i (i = 1,...,6)$ of Prange have
the disadvantage to behave differently under $s \leftrightarrow u$
crossing. While $T_1$, $T_3$, $T_5$ and $T_6$ are even functions of
$\nu$, $T_2$ and $T_4$ are odd functions (note that $\nu
\rightarrow -\nu$ is equivalent to $s \leftrightarrow u$). Therefore,
L'vov used a new set of invariant amplitudes $A_i (i =
1,...,6)$, which are all even functions of $\nu$ and at the same
time free of kinematical singularities~\cite{lvov97}
\begin{eqnarray}
A_1 &=& {1 \over t}\; \left[ T_1\;+\;T_3\;+\;\nu\,(T_2\;+\;T_4)\right]
\;, \nonumber\\
A_2 &=& {1 \over t}\; \left[ 2\,T_5\;+\;\nu\,(T_2\;+\;T_4)\right] \;,
\nonumber\\
A_3 &=& {M^2 \over {M^4 - su}} \;\left[ T_1\;-\;T_3\;-\;
{t \over {4 \nu}}\,(T_2\;-\;T_4)\right] \;,
\nonumber\\
A_4 &=& {M^2 \over {M^4 - su}} \;\left[ 2\,M\,T_6\;-\;
{t \over {4 \nu}}\,(T_2\;-\;T_4)\right] \;,
\nonumber\\
A_5 &=& {1 \over {4\nu}}\; \left[ T_2\;+\;T_4 \right] \;,\nonumber\\
A_6 &=& {1 \over {4\nu}}\; \left[ T_2\;-\;T_4 \right] \;.
\label{eq:aifuncti}
\end{eqnarray}
In terms of the $t$-channel helicity amplitudes $T_{\lambda_N
\lambda_{\bar N}, \, \lambda'_{\gamma} \lambda_\gamma}^t(\nu, t)$ of
Eq.~(\ref{eq:thelcov}) the invariant amplitudes $A_i (\nu, t)$ ($i$ =
1,...,6) of Eq.~(\ref{eq:aifuncti}) read
\begin{eqnarray}
A_1 &=& {{1} \over{t \, \sqrt{t - 4 M^2}}}
\left\{ \; \left[ T_{{1 \over 2}\,{1 \over 2},1\,1}^t \;+\;
T_{{1 \over 2}\,{1 \over 2},-1\,-1}^t \right]
\;-\; {{2 \, \nu \, \sqrt{t}} \over {\sqrt{su - M^4}}}\;
T_{{1 \over 2}\,-{1 \over 2},1\,1}^t \right\} \;, \nonumber \\
A_2 &=& {{1} \over{t \, \sqrt{t}}}
\left\{ \;- \left[ T_{{1 \over 2}\,{1 \over 2},1\,1}^t \;-\;
T_{{1 \over 2}\,{1 \over 2},-1\,-1}^t \right]
\;-\; {{2 \, \nu \, \sqrt{t - 4 M^2}} \over {\sqrt{su - M^4}}}\;
T_{{1 \over 2}\,-{1 \over 2},1\,1}^t \right\} \;, \nonumber \\
A_3 &=& {{M^2} \over {su - M^4}} \; {{1} \over{\sqrt{t - 4 M^2}}}
\left\{ \;2 \; T_{{1 \over 2}\,{1 \over 2},1\,-1}^t
\;+\; {{\sqrt{su - M^4}} \over {\nu \,\sqrt{t}}}
\;\left[ T_{{1 \over 2}\,-{1 \over 2},1\,-1}^t \;+\;
T_{{1 \over 2}\,-{1 \over 2},-1\,1}^t \right] \right\} \;, \nonumber\\
A_4 &=& {{M^2} \over {su - M^4}} \; {{1} \over{\sqrt{su - M^4}}}
\left\{ \;M \left[- \; T_{{1 \over 2}\,-{1 \over 2},1\,-1}^t
\;+\;T_{{1 \over 2}\,-{1 \over 2},-1\,1}^t \right] \right. \nonumber\\
&&\hspace{3.5cm}\left.
\;+\; {{\sqrt{t} \, \sqrt{t - 4 M^2}} \over {4 \, \nu}}
\;\left[ T_{{1 \over 2}\,-{1 \over 2},1\,-1}^t \;+\;
T_{{1 \over 2}\,-{1 \over 2},-1\,1}^t \right] \right\} \;, \nonumber\\
A_5 &=& {{\sqrt{t - 4 M^2}} \over {4 \, \nu \, \sqrt{t} \, \sqrt{su - M^4}}}
\left\{ \;-2 \; T_{{1 \over 2}\,-{1 \over 2},1\,1}^t \right\} \;, \nonumber\\
A_6 &=& {{\sqrt{t - 4 M^2}} \over {4 \, \nu \, \sqrt{t} \, \sqrt{su - M^4}}}
\left\{ \; \left[ T_{{1 \over 2}\,-{1 \over 2},1\,-1}^t \;+\;
T_{{1 \over 2}\,-{1 \over 2},-1\,1}^t \right] \right\} \;.
\label{eq:thelampl}
\end{eqnarray}
In the subtracted DR of Eq.~(\ref{eq:subt}), the
$t$-channel integral runs along the line $\nu$ = 0. Therefore, we have
to determine the imaginary parts Im$_t A_i (\nu = 0, t)$ of the
invariant amplitudes of Eq.~(\ref{eq:thelampl}). We start by
decomposing of the $t$-channel helicity amplitudes for $\gamma \gamma
\rightarrow N \bar N$ into a partial wave series,
\begin{equation}
T_{\lambda_N \lambda_{\bar N}, \, \lambda'_{\gamma} \lambda_\gamma}^t
(\nu, t) \;=\; \sum_J {{2 J + 1} \over {2}} \;
T_{\lambda_N \lambda_{\bar N}, \, \lambda'_{\gamma}
\lambda_\gamma}^{J}
(t)\; d^J_{\Lambda_N \Lambda_\gamma} (\theta_t) \;,
\label{eq:pwgagannbar}
\end{equation}
where $d^J_{\Lambda_N \Lambda_\gamma}$ are Wigner $d$-functions and
$\theta_t$ is the scattering angle in the $t$-channel, which is
related to the invariants $\nu$ and $t$ by
\begin{equation}
\cos \theta_t = {{4 \, M \, \nu} \over {\sqrt{t} \, \sqrt{t - 4 M^2}}}\;.
\label{eq:costht}
\end{equation}
It is obvious from this equation that $\nu = 0$
corresponds to $90^{\rm o}$ scattering for the $t$-channel process.
As explained in Section \ref{tchannel}, we calculate
the imaginary parts of the $t$-channel helicity amplitudes
$T_{\lambda_N \lambda_{\bar N}, \, \lambda'_{\gamma} \lambda_\gamma}^t
(\nu, t)$ through the unitarity equation by inserting $\pi \pi$
intermediate states, which should give the dominant
contribution below $K \bar K$ threshold,
\begin{equation}
2\,{\mathrm Im} T_{\gamma \gamma \, \rightarrow \, N \bar N} \;=\;
{1 \over { {(4 \pi)}^2 } }
\;{ | {\vec {p}}_{{\pi}} | \over {\sqrt {t}} }
\; \int {d \Omega}_{{\pi}}\,
\left [ \, T_{\gamma \gamma \, \rightarrow \, \pi \pi}\, \right ]
\; \cdot \;
\left [ \, T_{\pi \pi \, \rightarrow \, N \bar N}\, \right ] ^ { \ast }.
\end{equation}
Combining the partial wave expansion for $\gamma \gamma \rightarrow \pi
\pi$~,
\begin{equation}
T^{\gamma \gamma \, \rightarrow \, \pi \pi}
_{{\Lambda}_{\gamma}} (t, \theta_{\pi \pi}) \;=\;
\sum_{J even} \, {{2J+1} \over 2} \;
T^{J \;(\gamma \gamma \, \rightarrow \, \pi \pi)}
_{{\Lambda}_{\gamma}}(t) \, \cdot \,
\sqrt {(J-\Lambda_{\gamma})! \over {(J+\Lambda_{\gamma})!}} \, \cdot \,
P_J^{\Lambda_{\gamma}}(\cos \theta_{\pi \pi}),
\label{eq:partialgagapipi}
\end{equation}
and the partial wave expansion for $\pi \pi \rightarrow N \bar N$,
\begin{equation}
T^{\pi \pi \, \rightarrow \, N \bar N}
_{\Lambda_N} (t, \Theta) \;=\;
\sum_{J} \, {{2J+1} \over 2} \;
T^{J \; (\pi \pi \, \rightarrow \, N \bar N)}
_{\Lambda_N}(t) \, \cdot \,
\sqrt {(J-\Lambda_N)! \over {(J+\Lambda_N)!}} \, \cdot \,
P_J^{\Lambda_N}(\cos \Theta) \;.
\label{eq:partialpipinnbar}
\end{equation}
We can now construct the imaginary parts of the
Compton $t$-channel partial waves,
\begin{equation}
2\,{\mathrm Im} T^{J \;(\gamma \gamma \, \rightarrow \, N \bar N)}
_{\lambda_N \lambda_{\bar N}, \, \lambda'_{\gamma} \lambda_\gamma}(t)\;=\;
{1 \over {(8 \pi)} }
\; {p_{\pi} \over \sqrt{t}}
\left [ \, T^{J \;(\gamma \gamma \, \rightarrow \, \pi \pi)}
_{\Lambda_\gamma}(t)\, \right ]
\left [ \, T^{J \;(\pi \pi \, \rightarrow \, N \bar N)}
_{\Lambda_N }(t)\, \right ] ^ { \ast }.
\label{eq:partialtunit}
\end{equation}
The partial wave amplitudes $T^{J \;(\pi \pi \, \rightarrow \, N
\bar N)} _{\Lambda_N}$ of
Eq.~(\ref{eq:partialpipinnbar}) are related to the
amplitudes $ f^J_{\pm}(t)$ of Frazer and Fulco
\cite{Frazer60} by the relations
\begin{eqnarray}
T^{J \;(\pi \pi \, \rightarrow \, N \bar N)}_{\Lambda_N = 0}(t)\;
&&= \frac{16 \pi}{p_N}\,(p_N \; p_\pi)^J \, \cdot \,
f^J_+ (t)\ ,\nonumber\\
T^{J \;(\pi \pi \, \rightarrow \, N \bar N)}_{\Lambda_N = 1}(t)\;
&&= 8 \pi \,\frac{\sqrt {t}}{p_N}\,(p_N \; p_\pi)^J \, \cdot \,
f^J_- (t)\ ,
\label{eq:frazerfulco}
\end{eqnarray}
with $p_N$ and $p_{\pi}$ the c.m. momenta of nucleon and pion
respectively,
\begin{equation}
p_N = \sqrt{t/4 - M^2} \;, \hspace{2cm} p_\pi = \sqrt{t/4 - m_\pi^2} \;.
\end{equation}
For the reaction $\gamma \gamma \, \rightarrow \, \pi \pi$, we will
use the partial wave amplitudes
$F_{J\,\Lambda_\gamma}(t)$, which are
related to those of Eq.~(\ref{eq:partialgagapipi}) by
\begin{equation}
T^{J \;(\gamma \gamma \, \rightarrow \, \pi \pi)} _{\Lambda_\gamma}(t)
\;=\; \frac{2}{\sqrt{2J+1}} \cdot F_{J\,\Lambda_\gamma}(t) \;.
\label{eq:deffgagapipi}
\end{equation}
Denoting the Born partial wave amplitudes for
$\gamma \gamma \;\rightarrow \; {\pi}^+ {\pi}^-$ by
$B_{J \, \Lambda_\gamma}(t)$,
the lowest Born partial waves ($s$ and $d$ waves) are
\begin{eqnarray}
{B_{00}}(t)&&\;=\;2e^2\;{{1-{\beta}^2}\over {2 \beta}}\;
\ln{\left({{1+\beta}\over {1-\beta}}\right)}, \nonumber\\
{B_{20}}(t)&&\;=\;2e^2\;{\sqrt{5}\over 4}\;
{{1-{\beta}^2}\over{\beta^2}}\;\left\{ \,{{3-{\beta}^2}\over \beta}\,
\ln{\left({{1+\beta}\over {{1-\beta}}}\right)} \;-\;6\; \right\}, \nonumber\\
{B_{22}}(t)&&\;=\;2e^2\;{\sqrt{15}\over {4\sqrt{2}}}\;
\left\{ \,{(1-{\beta}^2)^2 \over {\beta^3}}\,
\ln{\left({{1+\beta}\over {1-\beta}}\right)} \;+\;{10 \over 3}\;
-\;{2\over {\beta}^2}\; \right\} \;,
\label{eq:gagapipiborn}
\end{eqnarray}
with $\beta$ = $ p_{\pi} / \left(\sqrt{t}/2 \right)$ the pion
velocity.
Inserting the partial-wave expansion of Eq.~(\ref{eq:pwgagannbar}) into
Eq.~(\ref{eq:thelampl}), we can finally express the $2 \pi$ $t$-channel
contributions Im$_t A_i (\nu = 0, t)^{2 \pi}$ by the partial wave
amplitudes for the reactions $\gamma \gamma \, \rightarrow \, \pi \pi$
and $\pi \pi \, \rightarrow \, N \bar N$,
\begin{eqnarray}
&&{\mathrm Im}_t A_1 (\nu = 0, t)^{2 \pi} = {{p_\pi} \over {\sqrt{t}}} \,
{1 \over {t \, p_N^2}}\,
\sum_{J = 0,2,4,...} \,(p_\pi \,p_N)^J \,\sqrt{2 J + 1} \,
F_{J\,\Lambda_\gamma = 0}(t) \, f^{J*}_+(t) \,
\left[ (-1)^{J/2} \, {{(J - 1)!!} \over {J!!}}\right] \, , \nonumber\\
&&{\mathrm Im}_t A_2 (\nu = 0, t)^{2 \pi} = 0 \;, \nonumber\\
&&{\mathrm Im}_t A_3 (\nu = 0, t)^{2 \pi} = \nonumber\\
&& {{p_\pi} \over {\sqrt{t}}} \;
{{M^2} \over {t \, p_N^4}}\,
\sum_{J = 2,4,...} \,(p_\pi \,p_N)^J \;\sqrt{2 J + 1} \;\;
F_{J\,\Lambda_\gamma = 2}(t) \;
\left[\, {(-1)^{(J - 2)/2}\;{\sqrt{(J + 1)J \over {(J - 1)(J + 2)}}}}
\; {{(J - 1)!!} \over {J!!}} \right] \nonumber\\
&&\hspace{3.5cm} \times \left\{ \;f^{J*}_+(t)
\;-\; \;f^{J*}_-(t) \; {M}
\left[{ {(J + 2)(J - 1) \,-\,2} \over {\sqrt{J(J + 1)}}}
\right] \right\} \;,\nonumber\\
&&{\mathrm Im}_t A_4 (\nu = 0, t)^{2 \pi} = \nonumber\\
&&- \,{{p_\pi} \over {\sqrt{t}}} \; {{M^3} \over {t \, p_N^4}}\,
\sum_{J = 4,...} \,(p_\pi \,p_N)^J \,\sqrt{2 J + 1} \,
F_{J\,\Lambda_\gamma = 2}(t) \, f^{J*}_-(t) \,
{ {(-1)^{(J - 2)/2} \;2\; (J - 2)(J + 3)} \over \sqrt{(J + 2)(J - 1)}}
\, {{(J - 1)!!} \over {J!!}} \, ,\nonumber\\
&&{\mathrm Im}_t A_5 (\nu = 0, t)^{2 \pi} = \nonumber\\
&&-\,{{p_\pi} \over {\sqrt{t}}} \; {M \over {t \, p_N^2}}\;
\sum_{J = 2,4,...} \,(p_\pi \,p_N)^J \;\sqrt{2 J + 1} \;\;
F_{J\,\Lambda_\gamma = 0}(t) \; f^{J*}_-(t) \;
\left[ {{(-1)^{(J - 2)/2}} \over \sqrt{J(J + 1)} } \;
{{(J + 1)!!} \over {(J - 2)!!}}\right] \;, \nonumber\\
&&{\mathrm Im}_t A_6 (\nu = 0, t)^{2 \pi} = \nonumber\\
&&-\,{{p_\pi} \over {\sqrt{t}}} \; {M \over {t \, p_N^2}}\;
\sum_{J = 2,4,...} \,(p_\pi \,p_N)^J \;\sqrt{2 J + 1} \;\;
F_{J\,\Lambda_\gamma = 2}(t) \; f^{J*}_-(t) \;\nonumber\\
&&\hspace{3.5cm} \times
\left[\, { {(-1)^{(J - 2)/2} \;\; \left[(J + 2)(J - 1) \,-\, 2\right]}
\over {\sqrt{(J + 2)(J - 1)}}}
\; {{(J - 1)!!} \over {J!!}} \right] \;.
\label{eq:ima2pi}
\end{eqnarray}
We note that the s-wave $(J=0)$ component of the $2\pi$ intermediate
states contributes only to $A_1$. The amplitude $A_2$,
corresponding to the exchange of pseudoscalar mesons (dominantly
$\pi^0$) in the $t$-channel, gets no contribution from $2\pi$ states,
because the $2 \pi$ system cannot couple to the nucleon through a
pseudoscalar operator. Furthermore, it is found that only waves with
$J\ge4$ contribute to the amplitude $A_4$. In our calculations we
saturate the $t$-channel dispersion integral with s(J = 0)- and
d($J = 2$)-waves, for which the expressions of Eq.~(\ref{eq:ima2pi})
reduce to those given in Eq.~(\ref{eq:ima2pisd}).
\section{${\it F}_2\,$-meson contribution to the
$\gamma\gamma\,\rightarrow\,\pi\pi$ process}
\label{app:f2}
A particle with mass $m$ and spin-2 is described in terms of a
symmetric and traceless field tensor $\Phi^{\mu\nu}$ satisfying the
Klein-Gordon equation. Furthermore, as for a massive spin-1 field,
the `Lorentz gauge' condition requires that the four-divergence with
respect to one of the four-vector indices vanishes,
\begin{eqnarray}
\left(\,\Box \;+\; {m^2}\, \right)\,{\Phi^{\mu\nu}}&&\;=\;0 \;,\nonumber\\
{\Phi^{\mu\nu}}&&\;=\;{\Phi^{\nu\mu}} \;,\nonumber\\
{{\Phi^{\mu}}_{\mu}}&&\;=\;0 \;, \nonumber\\
{\partial_{\mu}}\,{\Phi^{\mu\nu}}&&\;=\;
{\partial_{\nu}}\,{\Phi^{\mu\nu}}\;=\;0\;.
\end{eqnarray}
Therefore, the tensor $\Phi^{\mu \nu}$ has only five independent
components.
A state of spin-2 is characterized by its polarization tensor
${\varepsilon^{\mu\nu}}\,(p, \Lambda),\;{\mathrm where}\; \Lambda
\;=\;\{2,1,0,-1,-2\}$ defines the polarization,
\begin{eqnarray}
{\varepsilon^{\mu\nu}}\,(p, \Lambda)&&\;=\;
{\varepsilon^{\nu\mu}}\,(p, \Lambda) \;,\nonumber\\
{{\varepsilon^{\mu}}_{\mu}}\,(p, \Lambda)&&\;=\;0 \;,\nonumber\\
{p_{\mu}}\,{\varepsilon^{\mu\nu}}\,(p, \Lambda)&&\;=\;
{p_{\nu}}\,{\varepsilon^{\mu\nu}}\,(p, \Lambda)\;=\;0\;.
\end{eqnarray}
The polarization tensor
${\varepsilon^{\mu\nu}}\,(p, \Lambda)$ can be constructed from
(massive) spin-1 polarization vectors by
\begin{equation}
{\varepsilon^{\mu\nu}}\,(p, \Lambda)\;=
\;{\sum_{\lambda=-1,0,1}}\;{\sum_{\lambda'=-1,0,1}}\;
\langle\;1\,\lambda,\;1\,\lambda'\;|\;2\,\Lambda\;\rangle\;
{\epsilon^{\mu}}\,(p,\lambda)\;{\epsilon^{\nu}}\,(p,\lambda') \;.
\label{eq:spin2pol}
\end{equation}
If one chooses the $z$-axis along the momentum of the particle, the three
polarization vectors of a massive spin-1 particle are
\begin{eqnarray}
&&{\epsilon^{\mu}}\,(p,\lambda\,=\,+1)\;=\;
(\;0\,,\,-\frac{1}{\sqrt{2}}\,,\,-\frac{i}{\sqrt{2}}\,,\,0\;) \;,\nonumber\\
&&{\epsilon^{\mu}}\,(p,\lambda\,=\,-1)\;=\;
(\;0\,,\,+\frac{1}{\sqrt{2}}\,,\,-\frac{i}{\sqrt{2}}\,,\,0\;) \;,\nonumber\\
&&{\epsilon^{\mu}}\,(p,\lambda\,=\,0)\;=\;
(\;|\vec{p}|,\;0,\;0,\;{p^0}\;)\;/\;m \;,
\end{eqnarray}
where $p^{\mu}$ = $(p^0, 0, 0, |\vec{p}|)$ and $p^2 = m^2$.
In the case of the spin-2 polarization tensor constructed
as in Eq.~(\ref{eq:spin2pol}), the following normalization
and completeness conditions hold :
\begin{eqnarray}
{\varepsilon^{\mu\nu}}\,(p, \Lambda)\;\cdot\;
{\varepsilon^{\ast}_{\mu\nu}}\,(p, \Lambda')\;&=&\;
\delta_{\Lambda\Lambda'} \;,\nonumber\\
{\sum_{\Lambda}}\;{\varepsilon^{\mu\nu}}\,(p, \Lambda)\;
{\big[\,{\varepsilon^{\alpha\beta}}\,(p, \Lambda)\,\big]^{\ast}}
\;&=&\;\frac{1}{2}\;\left(\;
{K^{\mu\alpha}}\,{K^{\nu\beta}}\;+\;
{K^{\nu\alpha}}\,{K^{\mu\beta}}\;-\;
\frac{2}{3}\,{K^{\mu\nu}}\,{K^{\alpha\beta}}\;\right) \;,\nonumber\\
{\mathrm with}\;\;{K^{\mu\nu}}\;=\;
-\,{g^{\mu\nu}}\;+\;\frac{{p^{\mu}}{p^{\nu}}}{m^2} \;.
\end{eqnarray}
Finally, the propagator of the spin-2 particle with total width
$\Gamma$ takes the form
\begin{equation}
i \;{\Delta^{\mu\nu\alpha\beta}}\,(p)\;=\;
{{{ i \; \sum_{\Lambda}}\;\big[\,{\varepsilon^{\mu\nu}}\,(p, \Lambda)
\,\big]^{\ast}\;{{\varepsilon^{\alpha\beta}}\,(p, \Lambda)}}\over
{{p^2}\;-\;{m^2}\;+\;i\,m\,{\Gamma}}}\ .
\end{equation}
We will now apply this spin-2 formalism to describe the $s$-channel
exchange of the $f_2$ meson in the process $\gamma \gamma \rightarrow
\pi \pi$.
\newline
\indent
The coupling of the (isoscalar) $f_2$(1270) meson to a
pion pair (with momenta $p_\pi$, $p'_\pi$ and cartesian isospin
indices $a, b$) is described by the amplitude
\begin{equation}
{\cal{M}}\left({f_2}\,\rightarrow\,\pi \pi\right)
\;=\;\frac{g_{{f_2}\pi\pi}}{m_{f_2}} \; \delta_{a b}
\;{p'_\pi}^{\mu}\;p_\pi^{\nu}\;{{\varepsilon}_{\mu\nu}}\,(p, \Lambda)\;,
\end{equation}
where $p$ is the $f_2$ meson momentum and $m_{f_2}$ its mass.
The coupling constant $g_{{f_2}\pi\pi}$ is determined from the
${f_2}\,\rightarrow\,\pi \pi$ decay width~:
\begin{equation}
\Gamma({f_2}\,\rightarrow\,\pi \pi)\;=\;{1 \over 40 \pi}\;{g_{{f_2}\pi\pi}^2}\;
\frac{\left(p_\pi\right)^5}{m_{f_2}^4}\ ,
\label{eq:f2pipiw}
\end{equation}
where $p_\pi = \sqrt{m_{f_2}^2 / 4 - m_\pi^2}$
is the pion momentum in the $f_2$ rest frame.
Using the partial width
$\Gamma({f_2}\,\rightarrow\,\pi \pi)\;=\;0.846\,\Gamma_0$ and
the total $f_2$-width $\Gamma_0\;=\;185$~MeV \cite{PDG98},
Eq.~(\ref{eq:f2pipiw}) yields for the coupling :
$g_{{f_2}\pi\pi} \simeq$~23.64.
\newline
\indent
The Lorentz structure of the vertex
${f_2}\,\rightarrow\,\gamma \gamma$ is given by
\begin{equation}
{\cal{M}}\left({f_2}\,\rightarrow\,\gamma \gamma\right)\;=\;-i \;2\,
e^2 \, \frac{g_{{f_2}\gamma \gamma}}{m_{f_2}}
\;{{\cal{F}}^{\mu\delta}}\left(q, \lambda_\gamma \right)
{{{\cal{F}}_{\delta}}^{\nu}}\left(q', \lambda'_\gamma\right)\;
{{\varepsilon}_{\mu\nu}}\,(p, \Lambda)\;,
\label{eq:vertexf2gaga}
\end{equation}
where ${\cal{F}}^{\alpha \beta}$ is the electromagnetic field tensor :
\begin{equation}
{\cal{F}}^{\alpha \beta} \left(q, \lambda_\gamma \right)\;=\;
q^\alpha \, \varepsilon^{\beta *} \left(q, \lambda_\gamma \right) \;-\;
q^\beta \, \varepsilon^{\alpha *} \left(q, \lambda_\gamma \right) \;.
\end{equation}
Using the vertex of Eq.~(\ref{eq:vertexf2gaga}), the
$f_2 \rightarrow \gamma \gamma$ decay width is calculated as :
\begin{equation}
\Gamma({f_2}\,\rightarrow\,\gamma\gamma)\;=\;
\frac{e^4}{80 \pi}\,{g_{{f_2}\gamma\gamma}^2}\,m_{f_2}.
\label{eq:f2gagaw}
\end{equation}
Using the partial width $\Gamma({f_2}\,\rightarrow\,\gamma\gamma)\,=\,
1.32\cdot{10^{-5}}\;\Gamma_0$ \cite{PDG98}, Eq.~(\ref{eq:f2gagaw})
determines the value of the coupling constant :
$g_{{f_2}\gamma\gamma} \simeq$~0.239.
\newline
\indent
Using these couplings and vertices, we can now calculate the
invariant amplitude for
the process $\gamma\gamma\,\rightarrow\,{f_2}\,\rightarrow\,\pi\pi$ :
\begin{equation}
{\cal{M}}\,(\gamma\gamma\,{\rightarrow^{\!\!\!\!\!\!\!{f_2}}}\,\pi\pi)\;=\;
-i \;2 \, e^2\, \frac{{g_{{f_2}\gamma\gamma}}}{m_{f_2}}\,
{{\cal{F}}^{\mu\delta}}\!(q,\lambda_{\gamma})
{{\cal{F}}_{\delta}}^{\nu}\!(q',\lambda'_{\gamma})\;
{\Delta_{\mu\nu\alpha\beta}}(p,\Lambda)\;
\frac{g_{{f_2}\pi\pi}}{m_{f_2}}\,
{p_\pi^\alpha} \,{{p'_\pi}^{\beta}}\,.
\label{eq:gagapipiampli}
\end{equation}
To determine the $\gamma \gamma \rightarrow \pi \pi$ helicity
amplitudes $F_{\Lambda_\gamma}$ defined in Eq.~(\ref{eq:crossgagapipi}),
we shall evaluate Eq.~(\ref{eq:gagapipiampli}) in the c.m. system.
For the case of equal photon helicities
(${\Lambda_\gamma} \,=\,0$) the $f_2$ does not contribute, i.e.
\begin{equation}
F^{(f_2)}_{\Lambda_\gamma \,=\, 0} \;=\;0 \;.
\end{equation}
For the case of opposite photon helicities (${\Lambda_\gamma}\,=\,2$)
we find after some algebra :
\begin{equation}
F^{(f_2)}_{\Lambda_\gamma \,=\,2} \;=\;
-\frac{e^2}{8}\,
\frac{g_{f_2 \gamma\gamma} \, g_{f_2 \pi\pi}}{m_{f_2}^2}\,
\frac{{t^2}{\beta^2}}{t\,-\,{m_{f_2}^2}\,+\,im_{f_2}{\Gamma_0}}\;
{\sin^2}\theta_{\pi \pi} \;,
\label{eq:f2gagapipilam2}
\end{equation}
where $\theta_{\pi \pi}$ is the pion c.m. angle and $\beta$ the pion
velocity as in Eq.~(\ref{eq:crossgagapipi}).
It is immediately seen from Eq.~(\ref{eq:f2gagapipilam2}) that
the $f_2$ meson contribution to the d-wave is given by :
\begin{equation}
F^{(f_2)}_{J=2 \; \Lambda_\gamma = 2}(t)\;=\;-\sqrt{\frac{2}{15}}\;
\frac{e^2}{4}\, \frac{g_{f_2 \gamma\gamma} \, g_{f_2 \pi\pi}}{m_{f_2}^2}\,
\frac{t^2 \beta^2}{t\,-\,{m_{f_2}^2}\,+\,im_{f_2}{\Gamma_0}} \;.
\end{equation}
\end{appendix}
|
\section{\@startsection {section}{1}{\z@}{-3.5ex plus -1ex minus
-.2ex}{2.3ex plus .2ex}{\large\sc}}
\def\subsection{\@startsection{subsection}{2}{\z@}{-3.25ex plus -1ex minus
-.2ex}{1.5ex plus .2ex}{\normalsize\sc}}
\makeatother
\makeatletter
\@addtoreset{equation}{section}
\renewcommand{\theequation}{\thesection.\arabic{equation}}
\makeatother
\newcommand{\newcommand}{\newcommand}
\newcommand{\renewcommand}{\renewcommand}
\newcommand{\begin{equation}}{\begin{equation}}
\newcommand{\end{equation}}{\end{equation}}
\newcommand{\bea}{\begin{eqnarray}}
\newcommand{\eea}{\end{eqnarray}}
\newcommand{\trac}[2]{{\textstyle\frac{#1}{#2}}}
\newcommand{\ex}[1]{\mbox{e}^{\,\textstyle#1}}
\newcommand{\CC}{\Bbb{C}}
\newcommand{\HH}{\Bbb{H}}
\newcommand{\PP}{\Bbb{P}}
\newcommand{\RR}{\Bbb{R}}
\newcommand{\ZZ}{\Bbb{Z}}
\newcommand{\II}{\Bbb{I}}
\newcommand{\EE}{\Bbb{E}}
\newcommand{\SS}{\Bbb{S}}
\renewcommand{\a}{\alpha}
\renewcommand{\b}{\beta}
\newcommand{\al}{\a^{l}}
\renewcommand{\d}{\delta}
\newcommand{\ga}{\gamma}
\newcommand{\la}{\lambda}
\newcommand{\lal}{\la_{l}}
\newcommand{\f}{\phi}
\newcommand{\fb}{\bar{\phi}}
\newcommand{\p}{\psi}
\def\bar{\p}{\bar{\p}}
\newcommand{\e}{\eta}
\newcommand{\eb}{\bar{\eta}}
\renewcommand{\c}{\chi}
\newcommand{\eps}{\epsilon}
\renewcommand{\t}{\theta}
\newcommand{\tb}{\bar{\theta}}
\newcommand{\om}{\omega}
\renewcommand{\P}{\Psi}
\newcommand{\pl}{\P_{L}}
\newcommand{\pdr}{\P^{\dag}_{R}}
\newcommand{\G}[3]{\Gamma^{#1}_{\;{#2}{#3}}}
\newcommand{\Gd}[3]{\Gamma_{{#1}{#2}{#3}}}
\newcommand{\Ga}{\Gamma}
\newcommand{\sig}{\sigma}
\newcommand{\sk}{\sigma_{k}}
\newcommand{\sa}{\sigma_{a}}
\newcommand{\Bb}{\bar{B}}
\newcommand{\symx}{{\mathhexbox@\msafam@73}}
\newcommand{\Q}{\bar{Q}}
\newcommand{\C}{{\cal A}/{\cal G}}
\newcommand{\A}[1]{{\cal A}^{#1}/{\cal G}^{#1}}
\newcommand{\RC}{{\cal R}_{\C}}
\newcommand{\RM}{{\cal R}_{\M}}
\newcommand{\RX}{{\cal R}_{X}}
\newcommand{\RY}{{\cal R}_{Y}}
\newcommand{\ad}{\mathop{\mbox{ad}}\nolimits}
\newcommand{\tr}{\mathop{\mbox{tr}}\nolimits}
\newcommand{\Tr}{\mathop{\mbox{Tr}}\nolimits}
\newcommand{\Det}{\mathop{\mbox{Det}}\nolimits}
\renewcommand{\det}{\mathop{\mbox{det}}\nolimits}
\newcommand{\rk}{\mathop{\mbox{rk}}\nolimits}
\newcommand{\diag}{\mbox{diag}}
\newcommand{\ra}{\rightarrow}
\newcommand{\Ra}{\Rightarrow}
\newcommand{\LRa}{\Leftrightarrow}
\newcommand{\lra}{\leftrightarrow}
\newcommand{\ot}{\otimes}
\renewcommand{\ss}{\subset}
\newcommand{\nul}{\noindent\underline}
\newcommand{\non}{\nonumber\\}
\renewcommand{\S}{\Sigma}
\newcommand{\tp}{2\pi i}
\newcommand{\del}{\partial}
\newcommand{\dbar}{\bar{\del}}
\newcommand{\dx}{\dot{x}}
\newcommand{\dxl}{\dot{x}^{\la}}
\newcommand{\dxm}{\dot{x}^{\mu}}
\newcommand{\dxn}{\dot{x}^{\nu}}
\newcommand{\ddx}{\ddot{x}}
\newcommand{\ddxm}{\ddot{x}^{\mu}}
\newcommand{\dxi}{\dot{\xi}}
\newcommand{\ddxi}{\ddot{\xi}}
\newcommand{\zb}{\bar{z}}
\newcommand{\na}{\nabla}
\newcommand{\nal}{\nabla_{\la}}
\newcommand{\nam}{\nabla_{\mu}}
\newcommand{\nan}{\nabla_{\nu}}
\newcommand{\gint}{\int \sqrt{g} d^{4}x\;}
\newcommand{\gz}{g^{(0)}}
\newcommand{\gt}{g^{(2)}}
\newcommand{\gf}{g^{(4)}}
\newcommand{\ff}{f^{(4)}}
\renewcommand{\lg}{\frak{g}}
\newcommand{\lv}{\log V_{s}}
\newcommand{\vs}{V_{s}}
\renewcommand{\ln}{\log \N}
\newcommand{\ls}{\ell_{s}}
\newcommand{\N}{{\cal N}}
\newcommand{\M}{{\cal M}}
\newcommand{\F}{{\cal F}}
\newcommand{\E}{{\cal E}}
\renewcommand{\P}{{\cal P}}
\newcommand{\I}{{\cal I}}
\newcommand{\IIt}{$\widetilde{\mbox{II}}$}
\newcommand{\gst}{\widetilde{g_{s}}}
\newcommand{\gsh}{\widehat{g_{s}}}
\newcommand{\lsh}{\widehat{\ls}}
\newcommand{\rllh}{\widehat{R_{11}}}
\newcommand{\lph}{\widehat{\ell_{P}}}
\newcommand{\mnd}{M_{Nd}}
\newcommand{\mat}[4]{\left(\begin{array}{cc}#1\\#3\end{array}\right)}
\newcommand{\r}[1]{{\mathbf{#1}}}
\newcommand{\rb}[1]{{\overline{\mathbf{#1}}}}
\newcommand{\gi}{\gamma_{i}}
\newcommand{\gj}{\gamma_{j}}
\newcommand{\adss}{AdS_{5}\times S^{5}}
\newcommand{\adsx}{AdS_{5}\times X^{5}}
\newcommand{\subs}[1]{\begin{center}%
{{\sc #1}}{\addcontentsline{toc}{subsection}{#1}}%
\end{center}}
\newcommand{\chap}[1]{{\clearpage}%
\begin{center}%
{\noindent\underline{\large\sc #1}}{\addcontentsline{toc}{section}{#1}}%
\end{center}%
{\vspace*{0.3cm}}}
\newcommand{\begin{eqnarray}}{\begin{eqnarray}}
\newcommand{\end{eqnarray}}{\end{eqnarray}}
\newcommand{\rightarrow}{\rightarrow}
\newcommand{\leftrightarrow}{\leftrightarrow}
\begin{document}
\begin{titlepage}
\begin{flushright}
{\tt hep-th/9904179}
\end{flushright}
\vspace*{0.5in}
\begin{center}
{\LARGE{\sc
On Subleading Contributions to the \\[.2in] AdS/CFT Trace Anomaly}}\\
\vskip .3in
{\large\sc Matthias Blau}\footnote{e-mail: <EMAIL>} and
{\sc K.S.\ Narain}\footnote{e-mail: <EMAIL>}\\
\vspace{.2in}
{\it ICTP, Strada Costiera 11, 34014 Trieste, Italy}\\
\vskip .3in
{\large\sc Edi Gava}\footnote{e-mail: <EMAIL>}\\
\vspace{.2in}
{\it INFN, ICTP and SISSA, Trieste, Italy}
\end{center}
\begin{abstract}
\noindent In the context of the AdS/CFT correspondence, we perform a direct
computation in $AdS_5$ supergravity of the trace anomaly of a $d=4$,
${\cal N}=2$ SCFT. We find agreement with the field theory result up
to next to leading order in the $1/N$ expansion. In particular, the
order $N$ gravitational contribution to the anomaly is obtained from a
Riemann tensor squared term in the 7-brane effective action deduced from
heterotic - type I duality. We also discuss, in the AdS/CFT context,
the order $N$ corrections to the trace anomaly in $d=4$, ${\cal N}=4$
SCFTs involving SO or Sp gauge groups.
\end{abstract}
\end{titlepage}
\setcounter{footnote}{0}
\section{Introduction}
Recently, a lot of work has been done on the conjectured \cite{jm1}
AdS/CFT correspondence between string theory or M-theory compactifications
on $AdS_{d+1}$ and $d$-dimensional conformal field theories. In particular,
this conjecture relates \cite{gkp,ew1} correlation functions of local
operators in the conformal field theory to amplitudes in the `bulk'
string theory or M-theory, with the boundary values of the bulk fields
interpreted as sources coupling to the operators of the `boundary'
conformal field theory.
An example of particular interest is the conjectured equivalence
between ${\cal N}=4,d=4$ supersymmetric $SU(N)$ Yang-Mills theory
and type IIB superstring theory on $AdS_{5}\times S^{5}$ (with $N$
units of RR five-form $F^{(5)}$ flux on $S^{5}$). This duality
identifies the complex gauge coupling constant
\begin{equation}
\tau_{YM}=\frac{\theta}{2\pi} + \frac{4\pi i}{g^{2}_{YM}}
\end{equation}
of the SYM theory with the constant expectation value
of the type IIB string coupling
\begin{equation}
\tau_{s} = \langle C^{(0)} + i\ex{-\phi}\rangle \equiv
\frac{\chi}{2\pi} + \frac{i}{g_{s}}\;\;.
\end{equation}
The radius of $S^{5}$ (or the curvature radius of $AdS_{5}$) is
\begin{equation}
L=(4\pi g_{s}N)^{1/4} \ell_{s}\;\;,
\label{L1}
\end{equation}
with $\ell_{s}$ the string length,
$\ell_{s}^{2}=\a'$. In terms of the 't Hooft coupling $\la = g^{2}_{YM}N$,
the dimensionless scale $L^{2}/\a'$ of string theory on $AdS_{5}\times
S^{5}$ is related to the SYM parameters by
\begin{equation}
\la^{1/2} = \frac{L^{2}}{\a'}\;\;.
\end{equation}
Correlation functions of string theory on $\adss$ are given by a double
expansion in $g_{s}$ and $\a'/L^{2}$, which can be written as a double
expansion in terms of $1/N= 4\pi g_{s}(\a'/L^{2})^{2}$ and $\la^{-1/2}$.
For closed oriented strings this is actually an expansion in even powers
of $N$, the string theory tree-level (supergravity) contribution being
of order $N^{2}$.
Correlation functions of the SYM theory, on the other hand, have a
$1/N$ expansion, valid when $N$ is large, $g^{2}_{YM}$ is small, and
$\la$ is kept finite (and small). For $SU(N)$ theories with adjoint fields
only, this is once again an expansion in even powers of $N$, the leading
contributions, of order $N^{2}$, coming from planar diagrams.
According to the (strong form of the) AdS/CFT correspondence, these
two theories should give rise to the same function of $\la$ at each
order in $N$. However, as one has an expansion in terms of $\la$ (weak
't Hooft coupling) and the other in terms of $\la^{-1/2}$ (the $\a'$ or,
better, $\a'/L^{2}$, expansion of string theory), in practice this
comparison is restricted to the rather limited set of quantities which
are $\la$-independent, such as global anomalies.
Leading order $N^{2}$
contributions to the chiral anomalies were checked in e.g.\
\cite{ew1,ca}, and trace anomalies were discussed (at the linearized
level) in \cite{lt} by comparing the bulk supergravity action with the
effective action arising from the coupling of ${\cal N}=4$ SYM to
${\cal N}=4$ conformal supergravity.
The complete leading order `holographic Weyl anomaly' was determined
in general in \cite{hs}. In
particular, it was found there that the leading supergravity
contribution to the trace anomaly involves only the squares of the
Ricci tensor and Ricci
scalar of the boundary metric and not the square of the Riemann tensor
itself. This implies that conformal field theories with a
standard (product space) supergravity dual necessarily have $a=c$
to leading order in $N$, where $a$ and $c$ are the coefficients of the
Euler and Weyl terms in the standard expression
\begin{equation}
\langle T^{\mu}_{\;\mu}\rangle = -a E_{4}-cI_{4}
\end{equation}
for the conformal anomaly. In \cite{NO2} the calculations of
\cite{hs} have been generalized to dilatonic gravity.
The check of subleading corrections in the $1/N^{2}$ expansion
is hampered by the fact that, for closed string theory, these correspond
to string loop corrections with RR background fields which are still
not well understood.\footnote{For a preliminary
discussion of some subleading ${\cal O}(1)$
contributions to anomalies see \cite{daak}.}
However, as pointed out in \cite{apty}, certain subleading $1/N$
corrections (i.e.\ terms of order $N$) in theories with open or
unoriented strings, corresponding to $SO(N)$ or $Sp(N)$ gauge theories,
may be accessible. In \cite{apty} an ${\cal N}=2$ superconformal field
theory arising from D3-branes on a $\ZZ_{2}$ orientifold O7-plane with
D7-branes \cite{fts} was analyzed. In particular, in a rather subtle
analysis it was shown that the order $N$ contribution to the chiral $U(1)$
R-current anomaly, proportional to $(a-c)$, is correctly reproduced in
the dual supergravity theory on $\adsx$, where $X^{5} = S^{5}/\ZZ_{2}$
\cite{sgd}, by bulk Chern-Simons couplings on the D7 and O7 world-volumes.
By supersymmetry, this chiral anomaly is related to the trace anomaly and
therefore indirectly \cite{apty} also confirms the AdS/CFT correspondence
for the conformal anomaly to this order in the large $N$ expansion. The
purpose of this note is to perform a direct calculation of the trace
anomaly along the lines of \cite{hs}. By chasing the Chern-Simons
couplings from type I' theory back to ten dimensions, we see that
they originate from the Green-Schwarz couplings $H^{2}=(dB + \omega_{L}
+\ldots)^{2}$ in the heterotic string. These terms are known to be
related by supersymmetry to CP even $R^{2}$-terms
proportional to the Riemann tensor squared and
$F^{2}$-terms for the gauge group $SO(8)\ss SO(32)$ \cite{ghmr2,eric}.
By using heterotic - type I duality and
T-dualizing to type I', we show that these terms
give rise to order $N$ Riemann tensor and gauge field strength
squared terms in eight dimensions
leading to a subleading order $N$ contribution to the conformal
anomaly upon reduction to $AdS_{5}$.
We then show that the external gauge field contribution
and the crucial coefficient of the
$\mbox{Riem}^{2}$-term of the boundary metric in the conformal anomaly,
proportional to $(a-c)$, are precisely reproduced by the supergravity
calculation.
We also find other terms of order $N$, proportional to the squares
of the Ricci tensor and Ricci scalar. This particular linear combination
differs from that of the field theory result precisely by a term
of order $N$ attributable to an effective five-dimensional cosmological
constant. We have been unable to determine this contribution
because of our ignorance regarding other four-derivative terms in the
type I' theory like $(F^{(5)})^{4}$ and $R(F^{(5)})^{2}$. Conversely,
comparing the supergravity calculation with the known field theory
result gives a concrete (but not in itself particularly interesting)
prediction for the $1/N$ contribution to the effective cosmological
constant in this theory.
Another class of theories with subleading order $N$ corrections to the
trace anomaly are ${\cal N}=4$ SYM theories with orthogonal or symplectic
gauge groups. These can be realized as low-energy theories on D3-branes
at an orientifold O3-plane and a candidate for their supergravity dual is
tpye IIB string theory on an $AdS_{5}\times \RR\PP^{5}$ orientifold. At
first, these theories appear to present a puzzle as there are no
D-branes or O-planes wrapping the $AdS_{5}$ and therefore there can be
no orientifold or open string corrections to the bulk theory. We
will show that both the leading and the subleading order $N$
contributions to
the anomaly are correctly reproduced by the classical Einstein
action by taking into account the (fractional) RR charge of the O3-plane.
In section 2, we review the CFT side of the gravitational and
external gauge field contributions to the conformal anomaly.
In section 3, we deduce the relevant $R^{2}$ and $F^{2}$
terms in the $AdS_{5}$ supergravity action via heterotic - type I
- type I' duality. In section 4, we review the calculation of
the leading ${\cal O}(N^{2})$ contribution to the trace anomaly
following \cite{hs}. We then deduce the ${\cal O}(N)$
contributions by extending the analysis of \cite{hs} to include
the $R^{2}$ and $F^{2}$ terms (section 5). In section 6 we
discuss the ${\cal N}=4$ SYM theories for SO and Sp gauge groups,
and we conclude with a discussion of the `missing'
contributions due to an effective cosmological constant of order $N$.
\section{The Trace Anomaly on the CFT Side}
The field theory of interest is \cite{apty} an ${\cal N}=2$
superconformal field theory with $Sp(N)$ gauge group, and 4 fundamental
and one antisymmetric traceless hypermultiplet. It arises \cite{fts} as the
low-energy theory on the world volume on $N$ D3-branes sitting
inside eight D7-branes at an O7-plane. Among the global symmetries
of the theory there are an $SO(8)$-symmetry (from the D7-branes)
as well as an $SU(2)\times U(1)$ R-symmetry of the ${\cal N}=2$
superconformal algebra. Taking the near-horizon limit of this
configuration one finds \cite{sgd} that the conjectural string
theory dual of this theory is type IIB string theory on $\adsx$
where $X^{5}=S^{5}/\ZZ_{2}$ in which the D7 and O7 fill the
$AdS_{5}$ and are wrapped around an $S^{3}$ which is precisely
the fixed point locus of the $\ZZ_{2}$. Because of the $\ZZ_{2}$
action, the relation between the five-form flux $N$ and the
curvature radius of $AdS_{5}$ is now
\begin{equation}
L = (8\pi g_{s}N)^{1/4}\ell_{s}
\label{L2}
\end{equation}
instead of (\ref{L1}). We will set $\ell_{s}=1$ in the following.
The trace anomaly, when the theory is coupled to an external metric, is
\begin{equation}
\langle T^{\mu}_{\;\mu}\rangle = -a E_{4} - cI_{4}\;\;,
\end{equation}
where, using shorthand notation,
\begin{equation}
\mbox{Riem}^{2}=R_{ijkl}R^{ijkl}\;\;,
\end{equation}
with $R_{ijkl}$ the Riemann curvature tensor of the metric of the
(boundary) space-time etc.,
\bea
E_{4}&=&\frac{1}{16\pi^{2}}\left( \mbox{Riem}^{2} - 4 \mbox{Ric}^{2}
+\mbox{R}^{2}\right)\non
I_{4}&=&-\frac{1}{16\pi^{2}}\left( \mbox{Riem}^{2} - 2 \mbox{Ric}^{2}
+\frac{1}{3}\mbox{R}^{2}\right)
\eea
Thus
\begin{equation}
\langle T^{\mu}_{\;\mu}\rangle = \frac{1}{16\pi^{2}}
[(c-a)\mbox{Riem}^{2} +(4a-2c)\mbox{Ric}^{2} +(\trac{1}{3}c-a)\mbox{R}^{2}]
\;\;.
\end{equation}
and we see that the $\mbox{Riem}^{2}$-term is proportional to $(c-a)$.
The coefficients $a$ and $c$ are determined in terms of the field content
of the theory. In particular, for the vector- and hypermultiplets
of the ${\cal N}=2$ theories one has
\bea
a_{V}=\frac{5}{24}&& c_{V}=\frac{1}{6}\non
a_{H}=\frac{1}{24}&& c_{H}=\frac{1}{12}
\eea
Thus, for one ${\cal N}=4$ multiplet ($n_{V}=n_{H}=1$) one has
$a=c=1/4$ and the trace anomaly of ${\cal N}=4$ $SU(N)$ SYM theory
is
\begin{equation}
\langle T^{\mu}_{\;\mu}\rangle =
\frac{N^{2}-1}{32\pi^{2}}[\mbox{Ric}^{2}-\trac{1}{3}\mbox{R}^{2}]\;\;.
\label{tsun}
\end{equation}
For an ${\cal N}=2$ theory with $n_{V}$ vector multiplets and $n_{H}$
hypermultiplets one has
\begin{equation}
\langle T^{\mu}_{\;\mu}\rangle = \frac{1}{24\times 16\pi^{2}}
[(n_{H}-n_{V})\mbox{Riem}^{2} +12 n_{V}\mbox{Ric}^{2} -
\trac{1}{3}(11n_{V}+n_{H})\mbox{R}^{2}]\;\;.
\end{equation}
In the present case, with
\bea
n_{V}&=&N(2N+1)=2N^{2}+N\non
n_{H}&=& 4\times 2N + N(2N-1) - 1 = 2N^{2} + 7N -1
\eea
one finds
\bea
a_{total}&\equiv& n_{V}a_{V} + n_{H}a_{H} = \frac{1}{24}(12N^{2}+12N -1)\non
c_{total}&\equiv& n_{V}c_{V} + n_{H}c_{H} = \frac{1}{24}(12N^{2}+18N -2)
\eea
and
\begin{equation}
a_{total}-c_{total} = \frac{1}{24}(1-6N)\;\;,
\end{equation}
and the conformal anomaly is
\begin{equation}
\langle T^{\mu}_{\;\mu}\rangle = \frac{1}{24\times 16\pi^{2}}
\left[
(6N-1) \mbox{Riem}^{2}+ (24N^{2}+12N)\mbox{Ric}^{2}
-(8N^{2}+6N-\frac{1}{3}) \mbox{R}^{2}\right]\;\;.
\end{equation}
The leading ${\cal O}(N^{2})$ contribution is
\begin{equation}
\frac{N^{2}}{16\pi^{2}}[\mbox{Ric}^{2}-\trac{1}{3}\mbox{R}^{2}]\;\;.
\label{leading}
\end{equation}
This is exactly twice the ${\cal N}=4$ result and this is in accordance
with the expected \cite{hs,sg} relation between volumes, $\mbox{Vol}(S^{5})$
versus $\mbox{Vol}(X^{5})$, and the leading contribution to the anomaly. On
the supergravity side this term arises \cite{hs} from a regularization
of the (divergent) classical gravity action which is just a volume
integral in this case as the $AdS$ scalar curvature (the Einstein-Hilbert
Lagrangian) is constant for $AdS_{5}$. We will review this calculation
below.
The subleading ${\cal O}(N)$-term is
\begin{equation}
\frac{6N}{24\times 16 \pi^{2}}[\mbox{Riem}^{2}+2 \mbox{Ric}^{2}-\mbox{R}^{2}]
\;\;.
\label{subleading}
\end{equation}
We will show that, modulo undetermined volume terms of the form
(\ref{leading}) (with coefficients of order $N$ rather than $N^{2}$),
this term arises from a Riemann curvature squared term in the bulk gravity
action (with the precise numerical coefficient deduced from that appearing
in the heterotic string through heterotic - type I - type I' duality).
One can also couple the theory to external gauge fields of a flavour
symmetry group $G$.
In general, the contribution of gauge fields to the trace anomaly
has been shown in \cite{cdj} to be proportional to the beta-function
of the corresponding gauge coupling constant. The result obtained
in \cite{cdj} is
\begin{equation}
\langle T^{\mu}_{\;\mu}\rangle_{G}
= \frac{\beta(g)}{2g} F^{a}_{ij}F^{a\,ij}\;\;,
\label{tbeta}
\end{equation}
where $\beta(g)$ has the standard form
\begin{equation}
\frac{\beta(g)}{2g}= -\frac{g^{2}}{32\pi^{2}}\left[\frac{11}{3}c_{2}(G)
-\frac{4}{3}T(R_{f}) - \frac{1}{3}T(R_{s}) \right]+{\cal O}(g^{4})\;\;.
\label{beta}
\end{equation}
Here $R_{f,s}$ are the representations of $G$ on the (Dirac)
fermions and (complex) scalars respectively, and $T(R)$ is the
Dynkin index of the representation $R$,
\begin{equation}
\Tr_{R}t_{a}t_{b} = T(R)\d_{ab}\;\;.
\end{equation}
To apply this result in the present situation we note the following.
First of all, in Euclidean space there is a minus sign on the right
hand side of (\ref{tbeta}), as can be seen by tracing through the
derivation in \cite[section 3]{cdj}. Moreover,
for an external gauge field, the first term on the right hand side of
(\ref{beta}) is of course absent.
In the present case, we can choose $G=SO(8)$, and the only fields
that are charged under $G$ are the $8N$ fundamental hypermultiplets
in the fundamental representation of $SO(8)$. As an ${\cal N}=2$
hypermultiplet in four dimensions consists of one Dirac fermion and
two complex scalars, the contribution of external $SO(8)$ gauge fields
to the trace anomaly is
\begin{equation}
\langle T^{\mu}_{\;\mu}\rangle_{G}
= -\frac{N T(\r{8})}{16\pi^{2}} F^{a}_{ij}F^{a\,ij}\;\;.
\label{TG}
\end{equation}
We have dropped the factor $g^{2}$ because we will be working with
the scaled gauge fields in terms of which the action takes the form
$S=(1/4g^{2})\int F^{2}+\ldots$.
We see that this term is also of order $N$, and we will show that
this contribution to the anomaly is reproduced precisely by an
$F^{2}$-term in the heterotic - type I action or, alternatively,
by the $F^{2}$-term of the Dirac-Born-Infeld D7-brane action
(wrapped on the $S^{3}\subset X^{5}$).
\section{The $R^{2}$ and $F^{2}$ Terms}
In \cite{apty}, the relevant Chern-Simons terms in the $AdS_{5}$ bulk
action arose from terms proportional to
\begin{equation}
\int C^{(4)}\wedge \Tr (\Omega\wedge\Omega)\;\;,
\int C^{(4)}\wedge \Tr (F\wedge F)\;\;,
\end{equation}
in the world volume theory of the D7-branes and O7-planes, where
$\Omega$ is the Riemann curvature two-form, $F$ denotes the
$SO(8)$ gauge field and $C^{(4)}$ is the RR
4-form coupling to the D3-brane. T-dualizing these terms
to type I, the RR 4-form becomes a six-form $C^{(6)}$ coupling to
the type I 5-brane. Writing this interaction as $F^{(7)}\wedge\omega_{L,YM}$,
where $\omega_{L,YM}$ is the Lorentz / Yang-Mills
Chern-Simons term, and dualizing
$F^{(7)}=*dB$, we see that this term arises from the
modification
\begin{equation}
H^{2}= (dB + \la_{L} \omega_{L} -\la_{YM}\omega_{YM})^{2}
\end{equation}
of the $B$-field kinetic term in the type I and heterotic supergravity
actions required by the Green-Schwarz anomaly cancellation mechanism.
Now it is known
\cite{ghmr2,eric} that supersymmetry relates this term to a four-derivative
CP even term $R_{LMNP}R^{LMNP}$ in the ten-dimensional heterotic action
together with the $SO(32)$ Yang-Mills term. The relevant part of the
heterotic action for our purposes is thus
\begin{equation}
S_{h}=\frac{1}{16\pi(8\pi^{6})}
\int d^{10}x\sqrt{G^{h}}\ex{-2\phi^{h}}
(R + \frac{1}{4}(R_{LMNP}R^{LMNP} - F^{a}_{MN}F^{a\,MN}))\;\;.
\end{equation}
We will now first check explicitly that these two terms in the end
give rise to terms of order $N^{2}$ and $N$ in the AdS supergravity
action respectively. We will determine the precise numerical factors
below.
First of all, using the rules of heterotic - type I duality,
\bea
\phi^{h}&=&-\phi^{I}\non
G^{h}_{MN} &=& \ex{-\phi^{I}}G^{I}_{MN}\;\;,
\eea
in the type I theory one obtains
\begin{equation}
\int d^{10}x\sqrt{G^{I}}(\ex{-2\phi^{I}} R +
\frac{1}{4} \ex{-\phi^{I}}
(R_{LMNP}R^{LMNP} - F^{a}_{MN}F^{a\,MN}))\;\;.
\end{equation}
Notice that the Riemann tensor square term comes with $\exp (-\phi)$
rather than with $\exp(-2\phi)$. This indicates that it arises from
disc (D9-branes) and crosscap (orientifold O9-planes) world-sheets
and not from
the sphere. The latter was to be expected since the sphere calculation
is identical to that in IIB where one knows that there is no $R^{2}$-term.
The observation that type I - heterotic duality dictates the
appearance of an $R^{2}$-contribution from the disc diagram in
type I theory was originally made in \cite{at}.
For a constant dilaton, which is all that we are interested in, the
dependence of the action on the type I string coupling constant
$g_{I}$ (defined in general by $g_s=\exp<\phi_s>$,
for $s=h, I, I'$ respectively) is thus
\begin{equation}
S_{I} \sim\int d^{10}x\sqrt{G^{I}}(\frac{1}{g_{I}^{2}}R
+ \frac{1}{4} \frac{1}{g_{I}}(R_{LMNP}R^{LMNP} - F^{a}_{MN}F^{a\,MN}))\;\;.
\end{equation}
Now we T-dualize this on a two-torus of volume $V_{I}$ to type I' theory in
eight dimensions. Since the eight-dimensional Newton constant is invariant,
we have (modulo factors of $2$ and $2\pi$)
\begin{equation}
V_{I}/g_{I}^{2} = V_{I'}/g_{I'}^{2}\sim 1/V_{I}g_{I}^{2}\;\;,
\end{equation}
and therefore
\begin{equation}
g_{I} \sim g_{I'}/V_{I'}
\end{equation}
and
\begin{equation}
V_{I}/g_{I} \sim 1/g_{I'}\;\;.
\end{equation}
Thus the T-dualized eight-dimensional action is
\begin{equation}
S_{I'} \sim\frac{V_{I'}}{g_{I'}^{2}}
\int d^{8}x\sqrt{G^{I'}} R +
\frac{1}{4} \frac{1}{g_{I'}}
\int d^{8}x\sqrt{G^{I'}} (R_{LMNP}R^{LMNP}-F^{a}_{MN}F^{a\,MN})\;\;.
\end{equation}
Since T-duality takes D9-branes to D7-branes and O9-planes to O7-planes,
one sees that in type I' the $R^{2}$- and $F^{2}$-terms come from discs
attached to
the D7-branes and crosscaps corresopnding to O7-planes. This explains why
there is no transverse volume factor $V_{I'}$ in these terms.
Now, to extract the $N$-dependence of these terms we scale the
metric to unit radius, not forgetting to scale $V_{I'}$ as well.
Thus
\bea
V_{I'}&\ra& L^{2}V_{I'} \non
d^{8}x\sqrt{G^{I'}} &\ra& L^{8} d^{8}x\sqrt{G^{I'}} \non
R &\ra& L^{-2}R \non
R_{LMNP}R^{LMNP}&\ra& L^{-4}R_{LMNP}R^{LMNP}\;\;,
\eea
and the action becomes
\begin{equation}
S_{I'} \sim\frac{L^{8}V_{I'}}{g_{I'}^{2}}
\int d^{8}x\sqrt{G^{I'}} R +
\frac{1}{4} \frac{L^{4}}{g_{I'}}
\int d^{8}x\sqrt{G^{I'}} (R_{LMNP}R^{LMNP}-F^{a}_{MN}F^{a\,MN})\;\;.
\end{equation}
Using (\ref{L2}) we see that, as anticipated, the string coupling
constant drops out and the Einstein term is of order $N^{2}$ while
the curvature squared terms in the effective 7-brane action
are of order $N$. For a recent discussion of curvature squared
terms in type II D-brane actions see \cite{bmg}.
The precise numerical factors of the five-dimensional action can now also
be determined. For the Einstein term, plus the cosmological constant,
we have the inverse ten-dimensional Newton constant times the volume
$\mbox{Vol}(X^{5})=\mbox{Vol}(S^{5})/2$ times, as we have seen, $L^{8}$,
giving
\bea
S_{E}&=&\frac{1}{16\pi(8\pi^{6} g_{I'}^{2})}\times \frac{\pi^{3}}{2}
\times (8\pi
g_{I'}N)^{2} \times \int_{AdS_{5}}d^{5}x \sqrt{G} (R-2\Lambda)\non
&=&\frac{N^{2}}{4\pi^{2}} \int_{AdS_{5}}d^{5}x \sqrt{G} (R-2\Lambda)\;\;,
\label{adsr}
\eea
where $R$ now denotes the five-dimensional Ricci scalar.
For the Riemann tensor squared term in five dimensions, and related terms
arising from the dimensional reduction of the internal and mixed
components of this term, the numerical coefficient arises as follows.
There is a factor of 1/4 in the ten-dimensional action. It was related
by supersymmetry to the anomaly cancelling Green-Schwarz term for
the gauge group $SO(32)$. By turning on appropriate Wilson lines in the
type I theory, this gauge group can be reduced to $SO(8)^{4}$. Upon
T-duality, these Wilson lines translate into the positions of the
D7-branes in the type I' theory.
Thus there are clusters of 8 D7-branes, each of the clusters located at
one of the 4 O7 orientifold planes. As we have seen above, the total
$R^{2}$-term comes from discs attached to D7-branes and crosscaps for
the O7-planes, each of the four clusters giving $\trac{1}{4}$ of the
total contribution. As in the near horizon limit three of these clusters
are infinitely far away, only one quarter of this term
will be relevant.
Moreover, because of the presence of the orientifold,
the volume of the two-torus should be taken to be $(2\pi^{2})$ rather than
the usual $(2\pi)^{2}$. Wrapping the D7 branes on the $S^{3}$, the
fixed locus of the $\ZZ_{2}$ action, produces another contribution
$\mbox{Vol}(S^{3})$. This $S^{3}$ has \cite{sgd,apty} the standard volume
$2\pi^{2}$. Finally, there is, as we have seen above, a factor of
$L^{4}$ from the scaling of the metric cancelling the $1/g_{I'}$.
Putting everything together,
we find that the coefficient of the $R^{2}$-term (as well as
that of the other components of this term and other related
4-derivative terms in the action) is
\bea
S_{R^{2}} &=&\frac{1}{16 \pi \times 8\pi^{6}}\times
(8\pi N) \times 2\pi^{2} \times 2\pi^{2} \times \frac{1}{16}
\times
\int_{AdS_{5}}d^{5}x \sqrt{G}R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma} +
\ldots\non
&=& \frac{6N}{24 \times 16 \pi^{2}}
\int_{AdS_{5}}d^{5}x \sqrt{G}R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma} +
\ldots
\label{adsr2}
\eea
Note the striking similarity of this coefficient with the subleading
contribution (\ref{subleading}) to the trace anomaly. Even though we
haven't even begun to calculate the contribution of this term to the
trace anomaly, this certainly suggests that we are on the right track.
The same argument shows that the $\Tr F^{2}$-term for the
$SO(8)\subset SO(8)^{4}\subset SO(32)$ gauge fields in the heterotic
action, reinstating the factor of 4 we divided by before,
gives rise to an order $N$ contribution
\begin{equation}
S_{F^{2}}
= -\frac{N}{16 \pi^{2}}
\int_{AdS_{5}}d^{5}x \sqrt{G} F^{a}_{\mu\nu}F^{a\,\mu\nu} + \ldots
\label{adsf2}
\end{equation}
to the bulk action.
Alternatively \cite{apty}, up to an overall
normalization, the coefficient of this
term could have been deduced from the $SO(8)$ Dirac-Born-Infeld action
of the D7-branes. From this point of view it is of
course obvious that this is an open string disc contribution and hence of
order $N$.
The relative factor of 4 between the gravitational and gauge field
couplings mirrors that found in \cite{apty} for the five-dimensional
Chern-Simons terms arising from the D7/O7 RR Chern-Simons couplings.
Note again the striking similarity of this term with
the contribution (\ref{TG}) of external $SO(8)$ gauge fields to the
trace anomaly. Once we have developed the appropriate machinery below,
it will be straightforward to verify that (\ref{adsf2}) reproduces
exactly the anomaly (\ref{TG}).
\section{Review of the ${\cal O}(N^{2})$ Calculation}
\subs{The Strategy}
Before embarking on the calculation of the ${\cal O}(N)$ contribution
to the trace anomaly, let us quickly review the calculation of the
leading ${\cal O}(N^{2})$ contribution \cite{hs}.
Because the AdS metric has a second order pole at infinity, AdS space
only induces a conformal equivalence class $[\gz_{ij}]$ of metrics
on the boundary. To check for conformal invariance, one chooses a
representative $\gz_{ij}$ acting as a source term for the energy-momentum
tensor of the boundary theory. The AdS/CFT correspondence predicts that
the CFT effective action in the large $N$ supergravity limit is
\begin{equation}
W_{CFT}(g_{0})=S_{grav}(g;g_{0})\;\;,
\end{equation}
where $S_{grav}(g;g_{0})$ denotes the gravitational action evaluated on a
classical configuration which approaches (in the conformal sense) the metric
$\gz_{ij}$ on the boundary. The action is the sum of two terms, the standard
bulk action $S_{E} \sim\int (R-2\Lambda)$, and a boundary term, involving
the trace of the extrinsic curvature of the boundary, required
to ensure the absence of boundary terms in the variational principle.
To solve the classical equation of motion
\begin{equation}
R_{\mu\nu} - \trac{1}{2}g_{\mu\nu}(R-2\Lambda)=0\;\;,
\label{eins}
\end{equation}
with this boundary condition, one can \cite{graham,hs} make the following
ansatz for the metric,
\begin{equation}
G_{\mu\nu}dx^{\mu}dx^{\nu} = \frac{L^{2}}{4}\frac{d\rho^{2}}{\rho^{2}}
+ \frac{1}{\rho} g_{ij}dx^{i}dx^{j}\;\;,
\label{adsg}
\end{equation}
with the boundary sitting at $\rho=0$. We will set $L=1$ in the following
as our scaling arguments use the unit radius metric.
The metric $g_{ij}$ has an expansion as \cite{graham,hs}
\begin{equation}
g_{ij} = g^{(0)}_{ij} + \rho g^{(2)}_{ij} + \rho^{2}g^{(4)}_{ij}
+\rho^{2}\log\rho h^{(4)}_{ij} + \ldots\;\;,
\label{grho}
\end{equation}
with $\gz_{ij}$, as above, the chosen boundary value.
Now, for a solution to the classical equations of motion, both the bulk
and the boundary term are divergent (the former because for an Einstein
manifold the classical Einstein-Hilbert action reduces to a volume
integral, and the latter because the induced metric on the boundary is
singular). Therefore, one needs to regularize this expression (which, in
view of its conjectured relation to the CFT effective action is not
surprising). This can be done by introducing a cutoff $\epsilon$
restricting the range of $\rho$ to $\rho \geq \epsilon$. Note that,
in agreement with general arguments on holography \cite{lsewapjp},
this bulk IR cutoff corresponds to an UV cutoff in the CFT. Then the
regularized CFT effective action $W^{\epsilon}_{CFT}(\gz)$ is invariant
under $\d\gz = \la\gz, \d\epsilon = \la\epsilon$.
$W^{\epsilon}_{CFT}(\gz)$ can be written
as a sum of terms diverging as $\epsilon\ra 0$, $W^{\infty}_{CFT}(\gz)$,
and a finite term $W^{fin}_{CFT}(\gz)$. The former is a sum of terms
which are integrals of local covariant expressions in the boundary
metric $\gz$ and hence they can be removed by local counterterms.
Among these terms there is, for $AdS_{d+1}$ with $d$ even, a logarithmically
divergent term (which, interestingly enough, does not arise from the
logarithmic term in the expansion (\ref{grho}) of the metric).
In the standard way, removal of this term will then
induce a conformal anomaly in the finite part $W^{fin}(\gz)$.
The boundary term never contributes to the conformal anomaly (this is
a consequence of the fact that the logarithmic
term $h^{(4)}_{ij}$ in (\ref{grho}) is known to be traceless
with respect to $\gz$ \cite{graham}) and we will not consider it in the
following.
\subs{Calculation of the ${\cal O}(N^{2})$ Contribution}
In the case at hand, the precise form of the anomaly is determined
as follows. For an Einstein space with $R_{\mu\nu}=-4 g_{\mu\nu}$, the
value of the classical Lagrangian is $L_{c}=-8$.
The volume element is
\begin{equation}
\sqrt{\det G}= \trac{1}{2}\rho^{-3}\sqrt{\det g}\;\;,
\label{detG}
\end{equation}
where the latter can be expanded as
\begin{equation}
\sqrt{\det g} = \sqrt{\det g^{(0)}}(1+\trac{1}{2}\rho \Tr g^{(2)}
+\trac{1}{8} \rho^{2}[(\Tr g^{(2)})^{2}-\Tr ((g^{(2)})^{2})])+\ldots\;\;.
\label{detg}
\end{equation}
Here, $\Tr$ denotes the trace with respect to the metric $g^{(0)}$ and
we have made use of the useful identity \cite{hs}
\begin{equation}
\Tr g^{(4)} = \trac{1}{4}\Tr ((g^{(2)})^{2})\;\;.
\label{hs}
\end{equation}
By iteratively solving the Einstein equations as a power series
in $\rho$, one finds \cite{hs}
\begin{equation}
\gt_{ij} =- \trac{1}{2}(r^{(0)}_{ij}-\trac{1}{6}\gz_{ij}r^{(0)})\;\;.
\label{hs2}
\end{equation}
Here $r^{(0)}_{ij}$ denotes the Ricci tensor of $\gz$ etc. Note that
we are using the opposite sign conventions of \cite{hs}. Our conventions
for the curvature tensor,
\begin{equation}
R^{\la}_{\;\sigma\mu\nu}=
\del_{\mu}\G{\la}{\sigma}{\nu}
-\del_{\nu}\G{\la}{\sigma}{\mu}
+\G{\la}{\mu}{\rho}\G{\rho}{\nu}{\sigma}
-\G{\la}{\nu}{\rho}\G{\rho}{\mu}{\sigma}\;\;,
\end{equation}
and the Ricci tensor,
\begin{equation}
R_{\mu\nu} := R^{\la}_{\;\mu\la\nu} = g^{\la\sigma} R_{\sigma\mu\la\nu}
\;\;,
\end{equation}
are such that the curvature of the sphere is positive.
We will need the square of the trace and the trace of the square of this
term. One has
\bea
\Tr \gt &=& -\trac{1}{6}r^{(0)}\non
(\Tr \gt)^{2} &=& \trac{1}{36}(r^{(0)})^{2}\non
\Tr (\gt)^{2} &=& \trac{1}{4}( r^{(0)}_{ij} r^{(0)ij}-
\trac{2}{9}(r^{(0)})^{2})\;\;.
\eea
In particular, therefore, the order $\rho^{2}$-term in the expansion
(\ref{detg}) is
\begin{equation}
(\Tr\gt)^{2}-\Tr(\gt)^{2} = -\frac{1}{4}
[r^{(0)}_{ij}r^{(0)ij}- \frac{1}{3}(r^{(0)})^{2}]\;\;.
\label{volume}
\end{equation}
As one obtains a $\rho^{-3}$ from $\sqrt{G}$, it is clear that a
logarithmically divergent term will arise only from the term of order
$\rho^{2}$ in (\ref{detg}). In particular, we see that for any
gravitational action including only the Einstein term and a cosmological
constant, the leading contribution to the conformal anomaly will be
proportional to (\ref{volume}). Comparing with the discussion in
section 2, we see that this impies $a=c$ to order $N^{2}$ as (\ref{volume})
does not contain a $\mbox{Riem}^{2}$-term.
Let us now apply this to $\adsx$ and thus to the leading
contribution to the trace anomaly of the ${\cal N}=2$ superconformal
field theory considered in \cite{apty} and above.
Using (\ref{adsr}), and noting that the factor of $1/2$ in (\ref{detG})
is cancelled by a conventional factor of 2 in the definition of the
conformal anomaly, one finds that the ${\cal O}(N^{2})$ conformal anomaly,
i.e.\ the coefficient of the $\log\epsilon$-term, is
\bea
&&\frac{N^{2}}{4\pi^{2}} \times (-8) \times \frac{1}{8}
\times [(\Tr g^{(2)})^{2}-\Tr ((g^{(2)})^{2})]\non
&=& \frac{N^{2}}{16\pi^{2}}
[r^{(0)}_{ij}r^{(0)ij}- \frac{1}{3}(r^{(0)})^{2}]\;\;.
\eea
This is indeed {\em precisely} the leading contribution (\ref{leading})
to the conformal anomaly calculated on the CFT side.
\section{The ${\cal O}(N)$ Contribution}
\subs{The Strategy}
Now the strategy for including the Riemann tensor sqared
term should be clear. We take the Einstein plus Riemann squared action
(\ref{adsr}) plus (\ref{adsr2}) (possibly also with the $F^{2}$-term
(\ref{adsf2}) - we will comment on the inclusion of this term below)
\begin{equation}
S=\frac{N^{2}}{4\pi^{2}} \int d^{5}x \sqrt{G} (R-2\Lambda)
+\frac{6N}{24\times 16 \pi^{2}}
\int d^{5}x \sqrt{G}R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma} + \ldots
\label{news}
\end{equation}
(plus boundary terms),
solve the equations of motion with the
given boundary metric $\gz$, and isolate the log-divergent terms in the
action evaluated on this classical solution. Note that, because of
the presence of the term $R^{\mu\nu\rho\sigma}R_{\mu\nu\rho\sigma}$
in (\ref{news}) this calculation will no longer reduce to just a volume
calculation.
In principle, of course, solving the classical equations of motion
of this higher-derivative gravity action to the required order in
$\rho$ is an unpleasant task. In the present case, however, a
drastic simplification is brought about by the fact that we are only
interested in the contributions of order $N$ to the classical action.
For this, it is sufficient to evaluate the term
$R^{\mu\nu\rho\sigma}R_{\mu\nu\rho\sigma}$ on the classical solution
of the previous section to the original Einstein equation (\ref{eins}).
Indeed, as the second term in (\ref{news}) is $1/N$ down with respect to
the Einstein term, we can make an ansatz for the solution to the
full equations in the form
\begin{equation}
G_{\mu\nu}=G_{\mu\nu}^{(0)} + \frac{1}{N}G^{(1)}_{\mu\nu}\;\;,
\end{equation}
where $G^{(0)}_{\mu\nu}$ is a solution of (\ref{eins}). Plugging this
solution into the Einstein term, i.e.\ the first term of (\ref{news}),
one obtains at order $N^{2}$ the leading contribution to the anomaly
calculated in the previous section. A term of order $N$ that could
potentially arise as the next term in the expansion
is actually zero (because we are expanding about a
classical solution to the Einstein action and the boundary term is
precisely there to cancel any residual boundary terms). The second
term in (\ref{news}) will give a contribution of order $N$ when
evaluated on $G_{\mu\nu}^{(0)}$, and any other contributions
involving $G_{\mu\nu}^{(1)}$ will be of order 1 or lower.
Therefore, to find the order $N$ contributions to the trace anomaly,
we need to
\begin{enumerate}
\item calculate the Riemann curvature tensor of the metric
(\ref{adsg}), with $g_{ij}$ given by (\ref{grho}) and (\ref{hs2}),
as a function of $\rho$, and
\item then determine the order $\rho^{-1}$-terms in the
$\rho$-expansion of
\begin{equation}
\sqrt{\det G}G^{\a\mu}G^{\b\nu}G^{\ga\la}G^{\d\sigma}
R_{\a\b\ga\d}R_{\mu\nu\la\sigma}\;\;.
\end{equation}
\end{enumerate}
Since we know that $\sqrt{\det G} \sim \rho^{-3}\sqrt{\det g}$ (\ref{detG}),
this means that we need to pick up the order $\rho^{2}$-terms from
\begin{equation}
\sqrt{\det g}G^{\a\mu}G^{\b\nu}G^{\ga\la}G^{\d\sigma}
R_{\a\b\ga\d}R_{\mu\nu\la\sigma}\;\;,
\end{equation}
Here the $\rho$-expansions of the curvature, of $\sqrt{\det g}$ and of
the inverse metric have to be considered.\footnote{Similar calculations have
recently also been performed in \cite{NO}, however with the diametrically
opposite motivation of trying to reproduce the leading ${\cal O}(N^{2})$
contribution to the anomaly from a higher derivative action \ldots\ldots.}
If one also includes the $F^{2}$-term (\ref{adsf2}), then in principle
one would of corse have to solve the coupled Einstein-Yang-Mills equations.
But as the $F^{2}$-term is also of order $N$ the same argument as above
shows that the resulting subleading corrections to the metric are again
irrelevant. As regards the equation of motion for $F$ itself, we will see
below that only the boundary value of $F$ contributes so we do not have to
solve these equations either.
\subs{External Gauge Fields}
Let us begin with the external gauge field contribution (\ref{TG}) to
the anomaly as it is by far the simplest contribution to determine
(much simpler, in fact, than even the leading ${\cal O}(N^{2})$ contribution
to the anomaly discussed above).
As the above discussion shows, we need
to pick up the order $\rho^{2}$-terms of
\begin{equation}
\sqrt{\det g} G^{\a\mu}G^{\b\nu}F^{a}_{\a\b}F^{a}_{\mu\nu}\;\;.
\end{equation}
Now the components of the inverse metric are $G^{\rho\rho}=4\rho^{2}$,
$G^{ij}=\rho g^{ij}$, where $g^{ij}$ has the expansion
\begin{equation}
g^{ij}= g^{(0)ij} -\rho g^{(2)ij}
+ \rho^{2}(((g^{(2)})^{2})^{ij}-g^{(4)ij}) + \ldots,
\end{equation}
where indices are raised with $g^{(0)ij}$. As the two inverse metrics
contribute at least a factor of $\rho$ each, the only contribution
to the anomaly arises from
\begin{equation}
\sqrt{\det g^{(0)}} g^{(0)ik}g^{(0)jl} F^{(0)a}_{ij}F^{(0)a}_{kl}\;\;,
\end{equation}
where $F^{(0)}_{ij}$ is the boundary value of the gauge field
\begin{equation}
F_{ij} = F^{(0)}_{ij} + {\cal O}(\rho)\;\;.
\end{equation}
In \cite{apty} the relation betwwen the bulk supergravity and boundary
SCFT $SO(8)$-generators, in the fundamental representation $\r{8}$, was
determined from the AdS/CFT correspondence. Using this result,
one obtains that $T(\r{8})$, appearing in the field theoretic
expression (\ref{TG}), is equal to 1. Therefore,
(\ref{adsf2}) gives precisely the
external gauge field contribution (\ref{TG})
to the trace anomaly.
Alternatively, this term could have been deduced (in the Abelian,
non-interacting case) by following the prescription in \cite{ew1}:
On-shell, the bulk Maxwell action reduces to a boundary term, and
this boundary term can be evaluated in terms of Witten's bulk-to-boundary
Green's functions, extracting the local term (relevant to the anomaly)
in the end.
More directly, one can proceed locally, i.e.\ without using Green's
functions, by solving the
Maxwell equations in a $\rho$-expansion as was done for the Einstein
equations in \cite{hs}. From this vantage point, the logarithmic
divergence arises directly in the boundary term $\sim \int
A_{i}\del_{\rho}A_{i}$ because a term of order $\rho\log\rho$ in the
$\rho$-expansion of $A_{i}$ turns out to be
required to solve the bulk Maxwell
equations (cf.\ the $\rho^{2}\log\rho$-term in the expansion (\ref{grho})
of the metric, required for the same reason).
\subs{The Curvature Tensor}
As three different metrics appear here, $G_{\mu\nu}$, $g_{ij}$ and
$g^{(0)}_{ij}$, we will correspondingly denote their
curvature tensors by $R^{\mu}_{\;\nu\la\rho}$, $r^{i}_{\;jkl}$,
$r^{(0)i}{}_{jkl}$. $\nabla$ will denote the covariant derivative
compatible with $g_{ij}$, $\nabla^{(0)}$ that compatible with $g^{(0)}_{ij}$.
$\rho$-derivatives will be denoted by a prime.
The ubiquitous combination $g_{ij}-\rho g'_{ij}$, which we will abbreviate to
$k_{ij}$ in the following, contains no terms linear in
$\rho$. Up to $\rho^{2}\log\rho$-terms one has
\begin{equation}
k_{ij} \equiv g_{ij}-\rho g'_{ij} = g^{(0)}_{ij}-\rho^{2}(g^{(4)}+h^{(4)})
+ \ldots
\end{equation}
We will sometimes also abbreviate $g^{(4)}+h^{(4)}=f^{(4)}$.
For the curvature tensor one then finds
\bea
R_{ijkl} &=& \rho^{-1}[r_{ijkl}+
\rho^{-1} (k_{il}k_{jk}-k_{ik}k_{jl})]\non
R_{\rho ijk} &=& \trac{1}{2}\rho^{-2}(\nabla_{j}k_{ik}-\nabla_{k}k_{ij})\non
R^{\rho}_{\;i\rho j}&=& -2\rho^{-1}(k_{ij}-\rho k'_{ij})
+\rho^{-1} k^{2}_{ij}\;\;.
\eea
where in the last line the product is taken with respect to the metric
$g_{ij}$.
We will need the $\rho$-expansion of these curvature tensors.
For $R_{ijkl}$, we need to expand $r_{ijkl}$ as well as the other
terms. Symbolically we have
\begin{equation}
r^{i}_{\;jkl} = r^{(0)i}_{\;jkl} + \rho
(\nabla^{(0)}_{k}\d\G{i}{j}{l}-\nabla^{(0)}_{l} \d\G{i}{j}{k}) + \ldots,
\end{equation}
where we do not need to know the precise form of the $\d\Ga$'s.
Using this and the definition of $k_{ij}$ one finds
\bea
R_{ijkl} &=& G_{in}R^{n}_{\;jkl} = \rho^{-1}g_{in}R^{n}_{\;jkl}\non
R_{ijkl} &=& \rho^{-2}[\gz_{il}\gz_{jk}-\gz_{ik}\gz_{jl}]\non
&+& \rho^{-1} r^{(0)}_{ijkl}\non
&+& \rho^{0} [ \gz_{ik}\ff_{jl} +\gz_{jl}\ff_{ik}
-\gz_{il}\ff_{jk} -\gz_{jk}\ff_{il}]\non
&+& \rho^{0} [\nabla^{0}_{k}\d\Ga_{ijl}-\nabla^{(0)}_{l} \d\Ga_{ijk}]
\non
&+& \rho^{0} [\gt_{in}r^{(0)n}_{\;jkl}] + {\cal O}(\rho)
\label{rijkl}
\eea
$R_{\rho ijk}$ is simpler, we just keep the first term (and not even that
one will contribute as we will see),
\bea
R_{\rho ijk} &=&
\rho^{-1}[-\trac{1}{2}(\nabla_{j}\gt_{ik}-\nabla_{k}\gt_{ij})]
+ {\cal O}(1)
\eea
For $R^{\rho}_{\;i\rho j}$, one has
\bea
R^{\rho}_{\;i\rho j}&=& \rho ^{-1}[-\gz_{ij}]\non
&+& \rho^{0} [-\gt_{ij}]\non
&+& \rho^{+1}[-5 \ff_{ij} + (\gt)^{2}_{ij}]+ {\cal
O}(\rho^{2})\;\;,
\label{rij}
\eea
where now, of course, in the last line the product is taken with respect
to $\gz$.
As mentioned above, we need to pick up the order $\rho^{2}$-terms from
\begin{equation}
\sqrt{\det g}G^{\a\mu}G^{\b\nu}G^{\ga\la}G^{\d\sigma}
R_{\a\b\ga\d}R_{\mu\nu\la\sigma}\;\;,
\end{equation}
Let us deal with $R_{\rho ijk}$ first. In that case, the factor
entering the contractions is
\begin{equation}
\sqrt{\det g}G^{\rho\rho}G^{im}G^{jn}G^{kp}\;\;.
\end{equation}
This will contribute at least $\rho^{2} \times \rho^{3} = \rho^{5}$,
but the highest negative power of $\rho$ that can arise from the
square of $R_{\rho ijk}$ is $\rho^{-2}$, giving an overall $\rho^{3}$
and therefore no contribution to the anomaly.
For $R_{ijkl}$ we have
\begin{equation}
\sqrt{\det g}G^{im}G^{jn}G^{kp}G^{lq}\;\;.
\label{volijkl}
\end{equation}
This contributes $\rho^{4}$ and higher powers. But the highest negative
power arising from $R_{ijkl}^{2}$ is $\rho^{-4}$. Hence here terms of
order $\rho^{4}$, $\rho^{5}$ and $\rho^{6}$ in the expansion of
the contraction/volume factor (\ref{volijkl}) are relevant.
At order $\rho^{n}$, $n=4,5,6$ respectively, one has:
\bea
\rho^{4}:&& \sqrt{\gz} g^{(0)im} g^{(0)jn} g^{(0)kp} g^{(0)lq} \non
\rho^{5}:&& \sqrt{\gz} \trac{1}{2}\Tr\gt
g^{(0)im} g^{(0)jn} g^{(0)kp} g^{(0)lq} \non
&-&4 \sqrt{\gz} g^{(2)im} g^{(0)jn} g^{(0)kp} g^{(0)lq} \non
\rho^{6}:&& \sqrt{\gz}\trac{1}{8}[(\Tr\gt)^{2}-\Tr(\gt)^{2}]
g^{(0)im} g^{(0)jn} g^{(0)kp} g^{(0)lq} \non
&-&4 \sqrt{\gz} \trac{1}{2}\Tr\gt g^{(2)im} g^{(0)jn} g^{(0)kp} g^{(0)lq} \non
&+& 4\sqrt{\gz} [((\gt)^{2}-\gf)^{im}] g^{(0)jn} g^{(0)kp} g^{(0)lq} \non
&+& 2\sqrt{\gz} g^{(2)im} g^{(2)jn} g^{(0)kp} g^{(0)lq} \non
&+& 4\sqrt{\gz} g^{(2)im} g^{(0)jn} g^{(2)kp} g^{(0)lq}
\label{ijkl}
\eea
Finally, for $R^{\rho}_{\;i\rho j}$, the structure is
\begin{equation}
\sqrt{\det g}G^{im}G^{jn}\;\;.
\label{volij}
\end{equation}
This will be of order $\rho^{2}$ and higher. On the other hand, the
square of the curvature tensor gives terms of order $\rho^{-2}$ and
higher. Hence in the expansion of the contraction/volume factor
(\ref{volij}),
terms of order $\rho^{2}$, $\rho^{3}$, $\rho^{4}$ are relevant. These
are
\bea
\rho^{2}:&& \sqrt{\gz} g^{(0)im} g^{(0)jn} \non
\rho^{3}:&& \sqrt{\gz} \trac{1}{2}\Tr\gt
g^{(0)im} g^{(0)jn} \non
&-&2 \sqrt{\gz} g^{(2)im} g^{(0)jn} \non
\rho^{4}:&& \sqrt{\gz}\trac{1}{8}[(\Tr\gt)^{2}-\Tr(\gt)^{2}]
g^{(0)im} g^{(0)jn} \non
&-&2 \sqrt{\gz} \trac{1}{2}\Tr\gt g^{(2)im} g^{(0)jn} \non
&+& 2\sqrt{\gz} [((\gt)^{2}-\gf)^{im}] g^{(0)jn} \non
&+& \sqrt{\gz} g^{(2)im} g^{(2)jn}
\label{ij}
\eea
\subs{Contributions from $R_{ijkl}$}
Let us call the five contributions in (\ref{rijkl}) $I$, $II$, $III$,
$IV$ and $V$.
Three terms contribute to the $\rho^{4}$-term of (\ref{ijkl}), namely
$II\times II$, $I\times III$ and $I\times V$.
$I\times IV$ only contributes a total
derivative of a covariant quantity and can therefore be cancelled
by the variation of a local counterterm. Using the tracelessness
of $h^{(4)}$ one sees that $h^{(4)}$
will not contribute either. The other terms give
\bea
\rho^{4}, II\times II&& r^{(0)}_{ijkl} r^{(0)ijkl}\non
\rho^{4}, I \times III&& -6 \Tr (\gt)^{2}\non
\rho^{4}, I \times V&& -4 \gt_{ij} r^{(0)ij}
\eea
where we have used (\ref{hs}). The two terms of order $\rho^{5}$
in (\ref{ijkl}) need to be paired with $I\times II$:
\bea
\rho^{5}, I\times II&& -2 r^{(0)}\Tr\gt +16 g^{(2)ij} r^{(0)}_{ij}
\eea
The terms of order $\rho^{6}$ in (\ref{ijkl}) need to be paired with
$I\times I$. From the first three terms of order $\rho^{6}$ we get
\bea
\rho^{6}, I\times I&& -9 (\Tr\gt)^{2} + 15 \Tr (\gt)^{2}
\eea
The fourth and fifth term give
\bea
\rho^{6}, I\times I&& 4 (\Tr\gt)^{2} -4 \Tr (\gt)^{2}\non
&& 4 (\Tr\gt)^{2} +8 \Tr (\gt)^{2}
\eea
Adding all this up, we find the subtotal from $R_{ijkl}$ to be
\bea
&& r^{(0)}_{ijkl} r^{(0)ijkl}
-2 r^{(0)}\Tr\gt +12 g^{(2)ij} r^{(0)}_{ij}
- (\Tr\gt)^{2} + 13 \Tr (\gt)^{2}\non
&=&
r^{(0)}_{ijkl} r^{(0)ijkl}-\frac{11}{4}r^{(0)}_{ij}r^{(0)ij}
+\frac{7}{12}(r^{(0)})^{2}\;\;.
\eea
\subs{Contributions from $R^{\rho}_{\;i\rho j}$}
We proceed as above. The three terms of (\ref{ij}) we call
$I$, $II$, $III$. Every contribution has to be multiplied by
four, because there are four components of the Riemann tensor
with two $\rho$'s.
\bea
\rho^{2}, II\times II&& 4\Tr(\gt)^{2}\non
\rho^{2}, I \times III&& 2 \Tr (\gt)^{2}\non
\rho^{3}, I \times II&&
4(\Tr\gt)^{2}-16 \Tr (\gt)^{2}
\eea
There are four terms of order $\rho^{4}$ in (\ref{ij}), to be paired
with $I\times I$. These give
\bea
\rho^{4}, I \times I&& 2(\Tr\gt)^{2}-2 \Tr (\gt)^{2}\non
&& 4 \Tr (\gt)^{2}\non
&& 6 \Tr (\gt)^{2}\non
&& -4(\Tr\gt)^{2}
\eea
Adding all these up, one gets
\begin{equation}
2(\Tr\gt)^{2}-2\Tr (\gt)^{2}=- \frac{1}{2}r^{(0)}_{ij} r^{(0)ij}
+ \frac{1}{6}(r^{(0)})^{2}
\;\;.
\end{equation}
\subs{The Total ${\cal O}(N)$ Contribution to the Trace Anomaly}
Adding up all the above contributions, and remebering the prefactor
in (\ref{adsr2}), we find that supergravity
predicts the ${\cal O}(N)$ contribution to the trace anomaly to
be
\begin{equation}
\frac{6N}{24 \times 16 \pi^{2}}\times[
r^{(0)}_{ijkl} r^{(0)ijkl}-\frac{13}{4}r^{(0)}_{ij}r^{(0)ij}
+\frac{3}{4}(r^{(0)})^{2}]\;\;.
\end{equation}
A glance at (\ref{subleading}) shows that this does not yet look
particularly encouraging. However, let us split these terms as
\bea
&& \frac{6N}{24 \times 16 \pi^{2}}\times
[ r^{(0)}_{ijkl} r^{(0)ijkl} +2 r^{(0)}_{ij}r^{(0)ij}
- (r^{(0)})^{2}]\non
&-& \frac{6N}{24 \times 16 \pi^{2}} \times \frac{21}{4}
[r^{(0)}_{ij}r^{(0)ij}- \frac{1}{3}(r^{(0)})^{2}]\;\;.
\eea
We see that the first term reproduces {\em precisely} the subleading
contribution (\ref{subleading}) to the conformal anomaly, in particular
with the crucial term proportional to $(a-c)$. The
second (error) term, on the other hand, is exactly (and this is an
important check on our calculation) of the form of a volume contribution
(\ref{volume}), just like the leading ${\cal O}(N^{2})$-term. We will
say more about the possible origin of this volume term below.
\section{${\cal O}(N)$ Corrections for SO and Sp ${\cal N}=4$ Theories}
In this section we will briefly discuss another class of models which,
at first sight, seems to present a puzzle. Looking back at the ${\cal
N}=4$, $SU(N)$ trace anomaly (\ref{tsun}), we see that there is a
tree-level contribution, determined in \cite{hs}, no term of order
$N$ but an ${\cal O}(1)$ correction that ought to arise from a string
one-loop calculation. For other gauge groups $G$, however, the situation
is different. In general, one has
\begin{equation}
\langle T^{\mu}_{\;\mu}\rangle =
\frac{\mbox{dim}(G)}{32\pi^{2}}[\mbox{Ric}^{2}-\trac{1}{3}\mbox{R}^{2}]\;\;.
\label{tson}
\end{equation}
In particular, for orthogonal (symplectic) gauge groups $SO(N)$ ($Sp(N/2)$
for $N$ even),
$\mbox{dim}(G)$ contains both quadratic and linear terms in $N$,
\bea
\mbox{dim}(SO(2k)) &=& k(2k-1)\non
\mbox{dim}(SO(2k+1)) = \mbox{dim}(Sp(k)) &=& k(2k+1)\;\;,
\label{dimG}
\eea
and we want to understand the origin of these linear terms in the AdS/CFT
correspondence.
${\cal N}=4$ theories with gauge groups $SO(N)$, $Sp(N/2)$
can be realized as the low-energy dynamics of $N$ parallel
D3-branes at an orientifold O3-plane \cite{jp}, i.e.\
with the branes sitting at the singularity of a transverse
$\RR^{6}/\ZZ_{2}$. Here $\ZZ_{2}$ acts as $\vec{x}\ra -\vec{x}$ for
$\vec{x}\in\RR^{6}$. Note that, because of the non-compactness of
the transverse space, the number of D3-branes is not fixed by
RR tadpole cancellation.
This strongly suggests \cite{ew2} that a string
theory dual to these theories is given by type IIB string theory
on an $AdS_{5}\times \RR\PP^{5}$ orientifold.
Clearly, unlike for the ${\cal N}=2$ theory we discussed above, now there are
no branes wrapping the entire $AdS_{5}$.
So where are the terms linear in $N$ going to come
from? The answer is: from the classical Einstein action itself. The
reason for this is that O$p$-planes themeselves are carriers of
RR-charge, and hence the numerical value of $N$ appearing in the classical
D$p$-brane solutions (in the coefficient of the term $r^{p-7}$ in the
corresponding harmonic function, $r$ being the transverse distance
from the brane),
will be shifted in the presence of an orientifold
O$p$-plane. In particular, O3-planes carry fractional RR charge
$\pm\frac{1}{4}$ \cite{jp,mukhi}. With the minus sign, one obtains
$SO(N)$, and with the plus sign, for $N$ even, $Sp(N/2)$
gauge theories. In fact, at least prior to taking
the near-horizon limit, in the coefficient of the leading order
${\cal{O}}((\frac{\sqrt{\alpha'}}{r})^{7-p})$ correction to the classical
D$p$-brane solution, this $N$-independent term arises from the addition
of the crosscap $\RR\PP^{2}$ orientifold contribution (of order $g_s$)
to the disc D-brane diagram (of order $g_s N$).
This should extend to all orders to reproduce the
expected result so that, in the AdS-limit,
the net-effect of the presence of
the O3-plane is to replace $N$ as appearing in (\ref{L2}) by
\begin{equation}
N \ra \frac{N}{2} \pm\frac{1}{4}\;\;,
\end{equation}
where we took also into account that only $N/2$ of the $N$
D3-branes lie on $\RR^{6}/\ZZ_{2}$. Consequently
\begin{equation}
L^{4} = 8\pi g_{s}\left(\frac{N}{2}\pm\frac{1}{4}\right)\;\;.
\end{equation}
Repeating the calculation in (\ref{adsr}) and
section 4 for the leading contribution
to the trace anomaly, we now find
\bea
\langle T^{\mu}_{\;\mu} \rangle &=&
\frac{1}{16\pi \times 8\pi^{6}g_{s}^{2}}
\times \mbox{Vol}(\RR\PP^{5}) \times L^{8} \times
\int_{AdS_{5}}d^{5}x \sqrt{G} (R-2\Lambda)\non
&=&
\frac{1}{16\pi \times 8\pi^{6}}
\times \frac{\pi^{3}}{2} \times 64 \pi^{2} \times (\frac{N}{2}
\pm\frac{1}{4})^{2} \times
\frac{1}{4}[\mbox{Ric}^{2}-\trac{1}{3}\mbox{R}^{2}]\non
&=& \frac{1}{32\pi^{2}}\frac{N(N\pm 1)+\frac{1}{8}}{2}
[\mbox{Ric}^{2}-\trac{1}{3}\mbox{R}^{2}] \;\;.
\eea
Comparing with (\ref{dimG}), we see that to order
$N$ this agrees exactly with the trace anomaly formula
(\ref{tson}) for $G=SO(N)$ and $G=Sp(N/2)$. In these cases
we have therefore been able to reproduce both the leading and
the subleading order
$N$ corrections directly from the classical Einstein
action by taking
into account the fractional RR charge of the O3-plane.
\section{Discussion}
We have shown that supergravity calculations with higher-derivative
actions are capable of reproducing the subleading corrections to
the CFT trace anomaly. In this particular example, on the basis
of the results of \cite{apty} this was to be expected on general grounds
since supersymmetry relates the chiral and trace anomalies. Nevertheless,
we find it quite remarkable that the somewhat messy (even though
straightforward) classical calculations performed above conspire to give
precisely the correct result for the trace anomaly in the end.
As regards the `missing' volume contribution, we have of course
attempted to determine this in a variety of ways but it seems
to us that a definitive answer requires a better understanding
of $\a'$-corrections and supersymmetrization of the type I'
effective action.
As the expansion (\ref{detg}) produces $1/8$ times
(\ref{volume}), we see that what we are missing is an effective
cosmological constant term
\begin{equation}
- \frac{168 \times 6N}{24 \times 16 \pi^{2}}\int_{AdS_{5}}\sqrt{G}d^{5}x
\;\;.
\end{equation}
There are many possible terms that can contribute to this cosmological
constant. For instance, we have so far neglected the contributions of
the internal and mixed components of $R_{LMNP}R^{LMNP}$. These
contributions can in principle be determined either from duality arguments
or by a direct two-point function calculation on the type I' side.
There may also be four-derivative terms of the metric involving
the squares of the Ricci tensor or Ricci scalar (perhaps in the form
of the familiar Gauss-Bonnet combination). Such terms are afflicted
by the usual field redefinition ambiguities. Finally, there may also
be terms involving $(F^{(5)})^{4}$ and mixed terms of the
type $R(F^{(5)})^{2}$. The former correspond to four-point functions
on the type I' side and a direct calculation of these terms, although
possible in principle, is somewhat cumbersome.
One might have hoped to be able to invoke heterotic - type I - type I'
duality once more to fix these terms. For instance, in the heterotic
string it is known \cite{eric} that supersymmetry forces the
CP-even four-derivative terms involving $g_{MN}$ and $B_{MN}$ to appear
as curvature-squared terms of the connection with torsion $H=dB + \ldots$.
As $H$ eventually dualizes to $F^{(5)}$ in the type I' theory, this is
the sort of restrictive structure one might have hoped for. However,
chasing these terms through the dualities is somewhat problematic.
For one, as one is taking a large volume limit on the type I' side,
this corresponds to a small two-torus on the type I side, and thus
it appears that winding mode contributions in Type I need also be
considered. Also, at a purely classical level, in order to T-dualize
the D7/O7/D3 configuration underlying the ${\cal N}=2$ theory
we have been considering to a type I confiiguration of D9/D5 branes
one needs to delocalize it in the transverse directions. But if one
does that, it will no longer have the same near-horizon limit
(T-duality and near-horizon limits do not commute).
Conversely, we have been unable to find a classical type I solution
which gives the desired configuration on the type I' side and which
could have been used to calculate the effective cosmological constant
directly on the type I side.
All this just confirms the general picture that appears to be emerging
from the work done on the AdS/CFT correspondence, namely that
whatever can be checked reliably confirms the conjectured
correspondence, but that even simple (one-loop, anomaly) field theory
calculations are difficult to reproduce on the AdS side. Clearly,
what is required among other things is a better understanding of
string theory with RR backgrounds.
\subsubsection*{Acknowledgements}
We are grateful to George Thompson and Hossein Sarmadi for
useful discussions at various stages of this work. This
work was supported in part by the EC under the TMR contract
ERBFMRX-CT96-0090.
\renewcommand{\Large}{\normalsize}
|
\section{Introduction}
Some time ago monopoles in Einstein-Yang-Mills-Higgs(EYMH) model
\cite{{peter},{ortiz},{dieter}}, for $SU(2)$ gauge group with Higgs field in
adjoint representation, were studied as a generalization of the
't Hooft-Ployakov monopole\cite{thooft} to see the effect
of gravity on it. In particular, it was found that solutions exist up to
some critical value of a dimensionless parameter $\alpha $, characterising the
strength of the gravitational interaction, above which there is no regular
solution. The existance of these solutions were also proved analytically
for the case of infinite Higgs mass\cite{dieter}. Also, non Abelian magnetically
charged black hole solutions were shown to exist in this model for both
finite\cite{lee} as well as infinite\cite{bizon} value of the coupling constant
for Higgs field. The Abelian black holes exists for $r_h \ge \alpha $ and non
Abelian black holes exist in a limited region of the $(\alpha , r_h)$ plane.
Recently Born-Infeld theory \cite{born, infeld} has received wide publicity,
especially in the context of string theory\cite{string}.
Bogomol'nyi-Prasad-Sommerfield (BPS) saturated solutions were obtained in
Abelian Higgs model as well as in $O(3)$ sigma model in $2+1$ dimensions
in presence of Born-Infeld term\cite{tuku}. Different models for domain wall,
vortex and monopole solutions, containing the Born-Infeld Lagrangian
were constructed\cite{gib} in such a way that the self-dual equations are
identical with the corresponding Yang-Mills-Higgs model. Recently non
self-dual monopole solutions were found numerically in non Abelian
Born-Infeld-Higgs theory\cite{schap}.
In this paper we consider the Einstein-Born-Infeld-Higgs(EBIH) model and
study the monopole and black hole solutions. The solutions are qualitatively
similar to those of EYMH model. The black hole configurations have
nonzero non Abelian field strength and hence they are
called non Abelian black holes\cite{pbizon}. In Sec. II we consider the model
and find the equations of motion for static spherically symmetric fields.
In Sec III we find the asymptotic behaviours and discuss the numerical
results. Finally we conclude the results in Sec. IV.
\section{The Model}
We consider the following Einstein-Born-Infeld-Higgs action for $SU(2)$
fields with the Higgs field in the adjoint representation
\begin{eqnarray}
S = \int d^4x \sqrt{-g} \left[L_G + L_{BI} + L_H \right]
\end{eqnarray}
with
\begin{eqnarray}
L_G & = & \frac{1}{16\pi G}{\cal R} , \nonumber \\
L_H & = & -\frac{1}{2} D_{\mu}\phi ^a D^{\mu}\phi ^a
-\frac{e^2g^2}{4}\left(\phi ^a\phi ^a - v^2 \right)^2 \nonumber
\end{eqnarray}
and the non Abelian Born-Infeld Lagrangian\cite{nonab},
\begin{eqnarray}
L_{BI} = \beta ^2 Str\left( 1 - \sqrt{ 1
+ \frac{1}{2\beta ^2}F_{\mu\nu}F^{\mu\nu}
- \frac{1}{8\beta ^4}\left(F_{\mu\nu}\tilde{F}^{\mu\nu}\right)^2}\right) \nonumber
\end{eqnarray}
where
\begin{eqnarray}
D_{\mu}\phi ^a = \partial _{\mu}\phi ^a + e \epsilon ^{abc} A_{\mu}^b\phi ^c , \nonumber
\end{eqnarray}
\begin{eqnarray}
F_{\mu\nu} = F_{\mu\nu}^a t^a
= \left(\partial _{\mu}A_{\nu}^a - \partial_{\nu}A_{\mu}^a
+ e \epsilon ^{abc}A_{\mu}^bA_{\nu}^c\right)t^a \nonumber
\end{eqnarray}
and the symmetric trace is defined as
\begin{eqnarray}
Str\left(t_1,t_2...,t_n\right) = \frac{1}{n!}
\sum tr\left(t_{i_1}t_{i_2}...t_{i_n}\right) \nonumber .
\end{eqnarray}
Here the sum is over all permutations on the product of
the $n$ generators $t_i$. Here we are interested in purely magnetic
configurations, hence we have $F_{\mu\nu}\tilde{F}^{\mu\nu} = 0 $.
Expanding the square root in powers of $\frac{1}{\beta ^2}$ and keeping
up to order $\frac{1}{\beta ^2}$ we have the Born-Infeld Lagrangian
\begin{eqnarray}
L_{BI} = -\frac{1}{4}F_{\mu\nu}^a F^{a \mu\nu}
+\frac{1}{96\beta ^2}\left[\left(F_{\mu\nu}^a F^{a \mu\nu}\right)^2
+2F_{\mu\nu}^a F_{\rho\sigma}^a F^{b \mu\nu}F^{b \rho\sigma}\right]
+ O(\frac{1}{\beta ^4}) . \nonumber
\end{eqnarray}
For static spherical symmetric solutions, the metric can be
parametrized as\cite{dieter, komar}
\begin{eqnarray}
ds^2 = -e ^{2\nu(R)}dt^2 + e ^{2\lambda(R)}dR^2
+ r^2(R)(d\theta ^2 + \sin ^2\theta d\varphi ^2)
\end{eqnarray}
and we consider the following ansatz for the gauge and scalar fields
\begin{eqnarray}
A_{t}^a(R) = 0 = A_{R}^a, A_{\theta}^a = e_{\varphi}^a\frac{W(R) - 1}{e},
A_{\varphi}^a = -e_{\theta}^a\frac{W(R) - 1}{e}\sin\theta ,
\end{eqnarray}
and
\begin{eqnarray}
\phi ^a = e_{R}^a v H(R) .
\end{eqnarray}
Putting the above ansatz in Eq.1, defining $\alpha ^2 = 4\pi Gv^2$
and rescaling $ R \rightarrow R/ev, \beta \rightarrow \beta ev^2 $ and
$ r(R) \rightarrow r(R)/ev $ we get the following expression for
the Lagrangian
\begin{eqnarray}
\int dR e^{\nu +\lambda}\left[\frac{1}{2}\left(1
+ e^{-2\lambda}\left((r')^2 + \nu '(r^2)'\right)\right)
- \alpha ^2 \left(e^{-2\lambda} V_1 - e^{-4\lambda} V_2 + V_3 \right)\right],
\end{eqnarray}
where
\begin{eqnarray}
V_1 = (W')^2 +\frac{1}{2}r^2(H')^2 - (W')^2\frac{(W^2 - 1)^2}{6\beta ^2r^4},
\end{eqnarray}
\begin{eqnarray}
V_2 = \frac{(W')^4}{3\beta ^2r^2}
\end{eqnarray}
and
\begin{eqnarray}
V_3 = \frac{(W^2 - 1)^2}{2r^2} + W^2H^2 + \frac{g^2r^2}{4}(H^2 - 1)^2
- \frac{(W^2 - 1)^4}{8\beta ^2r^6} .
\end{eqnarray}
Here the prime denotes differentiation with respect to $R$.
The dimensionless parameter $\alpha $ can be expressed as the mass ratio
\begin{eqnarray}
\alpha = \sqrt{4\pi}\frac{M_W}{eM_{Pl}}
\end{eqnarray}
with the gauge field mass $M_W = e v $ and the Planck mass
$M_{Pl} = 1/ \sqrt{G} $ . Note that the Higgs mass $M_H = \sqrt{2} g e v $.
In the limit of $\beta \rightarrow \infty $ the above action reduces to
that of the Einstein-Yang-Mills-Higgs model\cite{peter,ortiz}.
For the case of $\alpha = 0$ we must have $\nu (R) = 0 = \lambda (R)$ which
corresponds to the flat space Born-Infeld-Higgs theory\cite{schap}.
We now consider the gauge $r(R) = R $, corresponding to the
Schwarzschild-like coordinates and rename $R = r $.
We define $A = e^{\nu + \lambda}$ and $N = e^{-2\lambda}$.
Varying the matter field Lagrangian with respect to the metric we find
the energy-momentum tensor. Integrating the $tt$ component of the
energy-momentum we get the mass of the monopole equal to $M/evG$ where
\begin{eqnarray}
M = \alpha ^2 \int_{0}^{\infty} dr \left(NV_1 - N^2V_2 + V_3\right)
\end{eqnarray}
Following 't Hooft the electromagnetic $U(1)$ field strength
${\cal F}_{\mu\nu}$ can be defined as
\begin{eqnarray}
{\cal F}_{\mu\nu} = \frac{\phi ^aF_{\mu\nu}^a}{\mid \phi \mid}
- \frac{1}{e\mid\phi\mid ^3}\epsilon^{abc}\phi ^aD_{\mu}\phi ^bD_{\nu}\phi ^c. \nonumber
\end{eqnarray}
Then using the ansatz(3) the magnetic field
\begin{eqnarray} B^i = \frac{1}{2}\epsilon ^{ijk}{\cal F}_{jk} \nonumber \end{eqnarray} is equal to
$ {e_{r}^{i}}/{er^2} $ with a total flux
$4\pi /e $ and unit magnetic charge.
The $tt$ and $rr$ components of Einstein's equations are
\begin{eqnarray}
&&\frac{1}{2}\left(1 - (rN)'\right) = \alpha ^2\left( N V_1 - N^2 V_2 +V_3 \right) \\
&&\frac{A'}{A} = \frac{2\alpha ^2}{r}\left(V_1 - 2NV_2\right).
\end{eqnarray}
The equations for the matter fields are
\begin{eqnarray}
&&\left(ANV_4\right)' = A W \left(
\frac{2}{r^2}(W^2 - 1) + 2 H^2 - \frac{(W^2 - 1)^3}{\beta ^2 r^6}
-\frac{2N(W')^2}{3\beta ^2 r^4} (W^2 - 1) \right) \\
&&(ANr^2H')' = A H \left(2W^2 + g^2r^2(H^2 - 1)\right)
\end{eqnarray}
with
\begin{eqnarray}
V_4 = 2W' - \frac{W'}{3\beta ^2 r^4}(W^2 - 1)^2
- \frac{4N}{3\beta ^2 r^2}(W')^3
\end{eqnarray}
It is easy to see that $A$ can be elliminated from the matter field
equations using Eq.(12). Hence we have to solve three differential equations
Eqs. (11),(13) and (14) for the three fields $N, W$ and $H$.
\section{Solutions}
\subsection{Monopoles}
For finite $g$, demanding the solutions to be regular and the
monopole mass to be finite gives the following behaviour near the origin
\begin{eqnarray}
&&H = a r + O(r^3), \\
&&W = 1 - b r^2 + O(r^4), \\
&&N = 1 - c r^2 + O(r^4),
\end{eqnarray}
where $a$ and $b$ are free parameters and $c$ is given by
\begin{eqnarray}
c = \alpha ^2 \left( a^2 + 4b^2 + \frac{g^2}{6} - \frac{20 b^4}{3\beta ^2} \right). \nonumber
\end{eqnarray}
In general, with these initial conditions $N$ can be zero at some finite
$r$ where the solutions become singular. In order to avoid this singularity
we have to adjust the parameters $a$ and $b$ suitably.
For $r \rightarrow \infty $ we require the solutions to be asymptotically
flat. Hence we impose
\begin{eqnarray}
N = 1 - \frac{2M}{r}
\end{eqnarray}
Then for finite mass configuration we have the following expressions for
the gauge and the Higgs fields
\begin{eqnarray}
&& W = C r^{-M} e^{-r}\left(1 + O(\frac{1}{r})\right) \\
&& H = \left\{ \begin{array}{ll}
1 - B r^{-\sqrt{2}gM - 1} e^{-\sqrt{2}gr}, & for ~~~ 0< g \le \sqrt{2} \\
1 - \frac{C^2}{g^2 - 2} r^{-2M-2} e^{-2r}, & for ~~ g = 0 ~~~ and g > \sqrt{2}.
\end{array}
\right.
\end{eqnarray}
Note that the fields have similar kind of asymptotic behaviour in the
EYMH model\cite{dieter}.
We have solved the equations of motion numerically with the boundary
conditions given by Eqs.(16-21). For $\alpha=0$, $g=0$ and $\beta \rightarrow \infty $
they corresponds to the exact Prasad-Sommerfield solution\cite{prasad}.
For nonzero $\alpha ,g$ and finite $\beta $ the qualitative behaviour of the solutions
are similar to the corresponding solutions of EYMH model. For large $r$ these
solutions converges to their asymptotic values given as in Eqs.(19-21).
For a fixed value of $g$ and $\beta $ we solved the equations increasing the
value of $\alpha $. For small value of $\alpha $ the solutions are very close to
flat space solution. As $\alpha $ is increased the minimum of the metric
function $N$ was found to be decreasing. The solutions cease to exist
for $\alpha $ greater then certain critical value $\alpha _{max}$.
For $g=0$ and $\beta =3$ we find $\alpha _{max} \sim 2 $. The profile for the
fields for different values of $\alpha $ with $g = 0$ and $\beta = 3$ are given
in Figs.1,2 and 3. The profile for the fields for $g=.1,
\alpha =1.0 $ and $\beta = 3 $ are given in Fig. 4. We find numerically
the mass $M = 0.7865 $ of the monopole for $g=.1,\alpha =1.0$ and $\beta = 3$.
\subsection{Black holes}
Apart from the regular monopoles, magnetically charged black holes can
also exist in this model. Black hole arises when the field $ N $ vanishes
for some finite $r = r_h$ . Demanding the solutions to be regular near
horizon $r_h$ we find the following behaviour of the fields
\begin{eqnarray}
&& N(r_h + \rho ) = N_{h}'\rho + O(\rho ^2) ,\\
&& H(r_h + \rho ) = H_h + H_{h}'\rho + O(\rho ^2), \\
&& W(r_h + \rho ) = W_h + W_{h}'\rho + O(\rho ^2)
\end{eqnarray}
with
\begin{eqnarray}
&& N_{h}' = \frac{1}{r_h}\left[ 1 - \alpha ^2\left\{
\frac{(W_{h}^2 - 1)^2}{r_{h}^2} + 2 W_{h}^2 H_{h}^2
+\frac{g^2r_{h}^2}{2}(H_{h}^2 - 1)^2
-\frac{(W_{h}^2 - 1)^4}{4\beta ^2 r_{h}^2} \right\}\right] \\
&& H_{h}' = \frac{H_{h}}{N_{h}'r_{h}^2}
\left\{ 2W_{h}^2 + g^2 r_{h}^2 (H_{h}^2 - 1)\right\} \\
&& W_{h}' = \frac{
\frac{W_h}{r_{h}^2}(W_{h}^2 - 1) + W_h H_{h}^2
- \frac{W_h}{2\beta ^2 r_{h}^6}(W_{h}^2 - 1)^3}
{N_{h}' \left[ 1 - \frac{(W_{h}^2 - 1)^2}{6\beta ^2r_{h}^4}\right]}.
\end{eqnarray}
Here $r_h, W_h(\equiv W(r_h)) $ and $H_h(\equiv W(r_h))$ are arbitrary.
For $r \rightarrow \infty $ the behaviour of the fields is same as the
regular monopole solution as given by Eqs.(19-21).
The black hole has unit magnetic charge with nontrivial gauge field strength.
We found numerical solutions to the non Abelian black hole for
different $r_h$. For a fixed value of $r_h $ we find the solutions for
$r > r_h $ adjusting the parameters $W_h $ and $H_h $. For $r_h $ close to
zero the solutions approach the regular monopole solutions. The profile for the
fields are given in Fig.5. We found the mass of the black hole equals to be
$ 0.6796 $ for $\alpha = 1.0, g = 0 $ and $\beta = 3$.
\section{Conclusion}
In this paper we have investigated the effect of gravity on the
Born-Infeld-Higgs monopole. We found that solutions exist only up to some
critical value $\alpha _{max}$ of the parameter $\alpha $. In the limit
$\beta \rightarrow \infty $ these solutions reduces to those of EYMH monopoles.
We also found numerically magnetically charged non Abelian black hole
solutions in this model. It would be interesting to prove analytically the
existence of these solutions for finite value of the parameters. Recently
dyons and dyonic black holes were found in EYMH model numerically\cite{tell}
and the existence of critical value for $\alpha $ was also proved analytically
\cite{tel}. It may be possible to generalize these solutions to find dyons and
dyonic black holes in EBIH model. We hope to report on this issue in future.
\section{Acknowledgements}
I am indebted to Avinash Khare for many helpful discussions as well as for
a careful manuscript reading.
|
\section{Introduction}
Ask any lay person to imitate computer speech and you will be treated to an
utterance delivered in melodic and rhythmic monotone, possibly accompanied by choppy
articulation and a voice quality that is nasal and strained. In fact, current
synthesized speech is far superior. Yet few would argue that synthetic and natural
speech are indistinguishable. The difference, as popular impression suggests, is the
relative lack of interesting and natural variability in the synthetic version. It
may be traced in part to the lack of a common causal account of pitch, timing,
articulation and voice quality. Intonation and stress are usually linked to the
linguistic and information structure of text. Features such as pause location and
word duration are linked mainly to the speaker's cognitive and expressive
capacities, and pitch range, intensity, voice quality and articulation to her
physiological and affective state.
In this paper, I describe a production model that attributes pitch and timing to the
essential operations of a speaker's working memory -- the storage and retrieval of
information. Simulations with this model produce synthetic speech in three of the
prosodic styles likely to be associated with attentional and memory differences: a
child-like exaggerated prosody for limited recall; a more adult but still expressive
style for mid-range capacities; and a knowledgeable style for maximum recall. The
same model also produces individual differences within each style, owing to its
stochastic storage algorithm. The ability to produce both individual and genre
variations supports its eventual use in prosthetic, entertainment and information
applications, especially in the production of reading materials for the blind and
the use of computer-based autonomous and communicative agents.
\section{A Memory Model for Prosody}
Prosody organizes spoken text into phrases, and highlights its most salient
components with {\em pitch accents}, distinctive pitch contours applied to the
word. Pitch accents are both attentional and propositional. Their very use indicates
salience; their particular form conveys a proposition about the words they mark. For
example, speakers typically use a high pitch accent (denoted as H*) to mark salient
information that they believe to be {\it new} to the addressee. Conversely, when
they believe the addressee is already aware of the information, they will typically
de-accent it\cite{brown83} or, if it is salient, apply a low pitch accent
(L*)\cite{pierrehumbert:hirschberg90}. Re-stated as a commentary on working memory,
the H* accent conveys the speaker's belief that the addressee can not retrieve the
accented information from working memory. De-accenting implicitly conveys the
opposite expectation. The L* accent does so explicitly. This view predicts different
speaking styles as a consequence of the speaker's beliefs about an addressee's
storage and retrieval capacities. For example, it ascribes the exaggerated
intonation that adults use with infants and young children\cite{fernald:simon84}, to
the adults' belief that the child's knowledge and attention are extremely limited;
therefore, he needs clear and explicit prosodic instructions as to how to process
language and interaction.
The model of working memory I use shows how retrieval limits can determine
the information status of an item as either given or new, and therefore,
its corresponding prosody. It was developed and implemented by Thomas
Landauer\cite{landauer75} and models {\em working memory} as a periodic
three dimensional Cartesian space, the {\em focus of attention} via a
moving search and storage {\em pointer} that traverses the space in a slow
random walk, and {\em retrieval ability} via a {\em search radius} that
defines the size of a region whose center is the pointer's current
location. Search for familiar items proceeds outward from the pointer, one
city block per time step, up to the distance specified by the search
radius.
As a consequence of the random walk, incoming stimuli are stored in a
spatial pattern that is locally random but globally coherent. That is,
temporal proximity in the stimuli begets spatial proximity in the model. It
contrasts with stack models of memory that are strictly chronological, and
semantic spaces in which distance is conceptual rather than temporal. Most
importantly, it is a valid computational model of attention and working
memory (AWM, from here on). Landauer used it to reproduce the well-known
learning phenomena of recency and frequency, in which subjects tend to
recall stimuli encountered most recently or most
frequently\cite{landauer75}. It has since been used by
Walker\cite{walker95} to show that resource-bound dialog partners will make
a proposition explicit when it is not retrievable or inferable, despite
having been previously mentioned.
Retrieval in AWM is the process of {\em matching} the current stimulus to
the contents of the region centered around the pointer. The search radius
determines the size of this region and therefore is the main AWM simulation
parameter. If a
match is found within the search region, the stimulus is classified as
given, otherwise, it is new. Figure~\ref{fig--awm-ex} illustrates this with
the simple example of filled and unfilled circles, a 4x4 AWM space, and a
search radius of one. At the center of the search region is the current
stimulus, a filled circle. Because the region contains no other filled
circles, the stimulus is classed as new. Had the stimulus been an
unfilled circle, it would have instead been classed as given because a
match is retrievable within the search radius. Or, alternatively, had the
search radius been two instead of one, a matching filled circle would have
been found, and the stimulus again classed as given.
\begin{figure}
\centerline{\psfig{figure=awm-ex.eps,width=50truemm}}
\caption{Using AWM, stimuli are classed as given if they have counterparts within
the the search radius. New items have no such counterparts because they are either
not in working memory, or are stored outside the radius.}
\label{fig--awm-ex}
\end{figure}
The ability to identify given and new items makes AWM a useful producer of prosody
based on this distinction. Ostensibly, it shows how a speaker's processing affects
her prosody. However, although the working memory belongs to the speaker, its
operation and determinations may reflect the speaker's own retrieval capacities, her
estimate of those of the addressee, or a mixture of both. That is, a speaker can
always adapt her style (prosodic and lexical) to the needs of a less knowledgeable
or capable addressee. A cooperative and communicative speaker will usually do this.
However, she cannot model a retrieval capacity greater than her own -- her own
knowledge and attentional limits always constitute the upper bounds on her
performance.
\section{System design}
The AWM component is embedded in a software implementation, {\sc Loq}, that takes a
text-to-speech approach. As shown Figure~\ref{fig--model-vert}, the input to AWM is
text, the output is speech. Therefore, {\sc Loq} models read rather than spontaneous
speech. Text comprehension is the process of searching for a match. Uttering the
text is a question of mapping the search process and its results to prosodic
features and sending the prosodically annotated text to the synthesizer.
Like many commercial text-to-speech synthesizers, the text structure is
analyzed before prosody is assigned. However, the {\sc Loq} analysis is
richer. It takes advantage of on-line linguistic databases to approximate
the speaker's knowledge of English semantics, pronunciation and usage. The
structural analysis is richer as well, providing both grammatical structure
(subject, verb, object), empty categories (ellipses, for example) and
information about clausal attachment. The main qualitative difference is
that {\sc Loq} interposes a model of limited attention and working memory
between the text analysis and prosodic mapping components.
\begin{figure}
\centerline{\psfig{figure=model-vert.eps,width=40truemm}}
\caption{AWM as a component of the {\sc Loq} system.}
\label{fig--model-vert}
\end{figure}
\subsection{Matching}
For the example in Figure~\ref{fig--awm-ex}, the matching criterion is binary and
simple -- a circle is either filled or unfilled. However, language is many more
times complex, and matches may occur for a variety of features, some of which are
more informative than others. The matching criteria used in {\sc Loq} attempt to
distill from the literature (e.g., \cite{nooteboom:terken82,fay:cutler77}) the most
relevant and prevalent ways that items in memory {\em prime} for the current
stimulus, and by the same token, the ways in which the current stimulus can function
as a {\em retrieval cue}. In other words, they gauge the mutual information between
the current stimulus and previously stored items.
Altogether, {\sc Loq} tests for matches on twenty-four semantic, syntactic,
collocation, grammatical and acoustical features. Each test contributes to the total
match score, which is then compared to a {\em threshold}. If it is below, the search
continues; if above, it stops. As shown in Figure~\ref{fig--matching}, matches on
any criterion express priming, and scores above the threshold constitute a match
sufficient to stop the search even before it reaches the edge of the search region.
Because some tests are more informative than others, a high score can reflect the
positive outcome of many un-informative tests, or of one that is definitive. Thus,
in the current ordering, co-reference ensures a match, while structural parallelism
in and of itself does not.
\begin{figure}
\centerline{\psfig{figure=matching.eps,width=80truemm}}
\caption{{\sc Loq} matching algorithm.}
\label{fig--matching}
\end{figure}
\subsection{Input}
The matching criteria determine the form and kind of information in the text input.
As with commercial synthesizers, this includes part of speech tagging. {\sc Loq}
uses the output of Lingsoft's ENGCG (English Constraint Grammar) software which
provides both tags and phrase structure information. However, reliable automatic
means for identifying other information, such as grammatical clauses, empty
categories, attachment and co-reference do not yet exist. Therefore, this
information was entered by hand.
The {\sc Loq} software turns the parsed and annotated text into a sequence of
tokens that assembles clauses in a bottom up fashion, starting with the
word and followed by the syntactic and grammatical clauses to which it
belongs. This models the reader's assembly of the words into meaningful
syntactic and grammatical groupings.\footnote{Adapting this for a
spontaneous speaker would proceed in reverse, from the concept, to
grammatical roles, syntactic phrases and finally, the words.}
To facilitate the matching process, the text is also augmented with information from
the WordNet database for semantic comparisons, a pronunciation
database for acoustical comparisons and the Thorndike-Lorge and Kucera-Francis for
word frequency counts\footnote{As provided in the Oxford Psycholinguistic
Database.} to scale the match score by the prior probability for the language. The
WordNet synonym indices were assigned by hand. However, all subsequent semantic
comparisons using WordNet are automatic as required by the matching process.
\subsection{Mapping}
I have described how AWM produces the L* accent (or none) for retrievable items, and
H* for new ones. However, there are more than two pitch accents -- Pierrehumbert
{\it et al.}\cite{beckman:pierrehumbert86} identify six\footnote{L*, H*, L+H*, L*+H,
H+L*, H*+L.} -- and more components to prosody. Obtaining them from one model first
requires an adjustment such that given or new status is determined from the effect
of the stimulus on the region as a whole, as follows: The result of any one
comparison affects the ``state'' of the item to which the stimulus is compared.
State is simply defined -- a L annotation records a match most any
criterion,\footnote{Some criteria are parasitic and only contribute to the score in
combination with other criteria.} and a H annotation records a match score of zero.
Thus, the comparison process registers both priming and a true match. Both receive L
annotations, but only a match whose score exceeds the threshold stops the search.
A pitch accent is then derived by comparing the contents of the search radius {\it
before} and {\it after} the matching process. Majority rules apply such that the
annotation with the higher count becomes the defining tone. If both the before and
after configurations are composed mainly of L annotations, the accent form is L+L,
which becomes the L* accent. However, if there is a change, for example, from a L to
H majority, the accent form is L+H. The interpretation of L+L is, roughly, that a
familiar item was expected and provided. Likewise, the interpretation for L+H is
that a familiar item was expected but an unexpected one provided.
To complete the bitonal derivation, {\sc Loq} treats the location of the main tone
as a categorical reflection of the magnitude of the effect of the stimulus. If the
stimulus changes the annotations for the majority of items in the search region, the
second tone is the main tone. Otherwise, it is the first. This schema produces the
six pitch accents identified by Pierrehumbert {\it et al.} More generally, the
annotation schema provides the model with a simple form of feedback -- the results
of prior processing persist and contribute to a bias that affects future processing.
\begin{figure}
\centerline{\psfig{figure=bitonals.eps,width=40truemm}}
\caption{In {\sc Loq}, bitonals occur when the L/H counts differ before and after
the matching process. The main tone of a bitonal is treated as a categorical
indicator of the magnitude of the effect of the stimulus on the context.}
\label{fig--bitonals}
\end{figure}
The pitch accent mapping illustrates the main features of the prosodic mapping in
general. First, all mappings reflect the activity and state within the region
defined by the search radius. Second, they express some aspects of prosody as a
plausible consequence of search and storage. For example, storage and search times
are mapped to word and pause duration. However, others -- for example, the bitonal
derivation-- are, at best, coherent with the operation and purpose of the model and
not contradicted by the current (sparse) data on the relation of cognitive capacity
to the prosody of read speech. In all, the mapping from AWM activity and state
produces intonational categories (pitch accent, phrase accent and boundary tone)
and their prominence, word duration, pause duration and the pitch range of an
intonational phrase.
\section{Results}
Although simulations were run using text from three different genres (fiction, radio
broadcast, rhymed poetry), two and three dimensional AWM spaces and three memory
sizes (small, mid-range and large), most of the prosodic output was correlated with
the search radius. Therefore, the results reported here are for the mid-range
two-dimensional memory (22x22) and for the news report text only (one paragraph, 68
words.) Five simulations were run for each radius.
True to the attentional predictions, Figure~\ref{fig--accents} shows that as the
search radius increases, the mean number of unaccented words increases as well,
while the number of H* accents decreases. Under the current mapping, pitch accent
prominence is a function the distance at which the search stops and the number of
comparisons performed prior to stopping. This produces a decrease in the mean
prominence as the search radius increases (Figure~\ref{fig--pa-prom}). These patterns
contribute to the lively and child-like intonation produced for the smallest radii
(1 and 2), the expressive but more subdued intonation for the mid-range radii (3-8)
and flatter intonation of the higher radii.
\begin{figure}
\centerline{\psfig{figure=un-vs-h-star.eps,width=40truemm}}
\caption{Mean unaccented {\it vs.} H* accented word counts as a function of search
radius.}
\label{fig--accents}
\end{figure}
\begin{figure}
\centerline{\psfig{figure=pa_prom.eps,width=40truemm}}
\caption{Mean pitch accent prominence for all six accent types, as a function of search
radius. }
\label{fig--pa-prom}
\end{figure}
The naturalness of synthetic prosody is difficult to evaluate via in perceptual
tests\cite{ross:ostendorf96}. However, informal comments from listeners revealed
that while the three styles were recognizable and the prosody more natural-sounding
than the commercial default, it was best for shorter sections rather than for the
passages as a whole. A comparison with the natural prosody for the same text (the BU
Corpus radio newscasts) showed that when the simulations agreed on pitch accent
location and type, they tended to disagree on boundary location and type, mostly
because the {\sc Loq} simulations produced many more phrase breaks than the natural
speaker.
\section{Conclusion and Future Directions}
{\sc Loq} is a production model. It produces prosody as the consequence of cognitive
processing as modeled by the AWM component. Its focus on retrieval makes it a
performance model as well, demonstrating that prosody is not determined solely by
the text. It produces three recognizable styles that appear to correlate with
retrieval capacities as defined by the search radius: child-like (for radii of 1
and2), adult expressive (for radii between 3 and 8) and knowledgeable (for radii
higher than 8). This is a step towards producing prosody that is both expressive and
natural and, in addition, specific to the speaker,
Currently, the main problem is that the prosody is not entirely cohesive within one
text. Therefore, one next step is to explore variations on the mapping of AWM
activity and state to prosodic features. More distant work includes extending the
model to incorporate other influences, especially the influence of physiology. This
may be the key to producing more than three styles, and to incorporating both the
dynamics and constraints that will produce consistently natural-sounding speech.
|
\section{Introduction}
Many questions are still open in Astronomy and in Particle Physics: what is the origin of the most energetic Cosmic Rays and of the Gamma Ray Bursts, what is the nature of the Dark Matter, do the neutrinos really oscillate as indicated by the SuperKamiokande result \cite{ref1}, ...
To give an answer to some of these questions, it is important to observe the Universe in a wide energy range and at long distance. To reach this goal an abundantly produced messenger is needed, it must able to bring informations out of the source, through long distances, until the detector. Among all the possibilities, only four particles have attracted the scientist's attention: the photon, the proton, the neutron and the neutrino.
{\bf The photon} is the most natural and the favourite probe of the astronomers. Its observation over more than 18 orders of magnitude in energy has revolutionised our conception of the World. Unfortunately, at high energy ($>$ TeV), it interacts with the less energetic ambient photons (infra-red, Cosmic Microwave Background and radio-wave photons) according to the pair-production mechanism:
$$ \gamma \gamma \rightarrow e^+ e^-$$
This so-called Greisen-Zatsepin-Kuzmin (GZK) cutoff prevents the observation of TeV photons above about 50 Mpc. This mechanism explains probably why only the nearest blazars have been observed in this energy range (the only extra-galactic sources observed so-far, Mkn 421 and Mkn 501, are also the two nearest blazars with z$\approx$0.03). Therefore the deep observation of the Universe at high energy is difficult with the photon.
{\bf The proton} is abundantly produced in the Universe and is therefore a good candidate. Unfortunately, at energies lower than a few EeV, it is deflected by the galactic and the extra-galactic magnetic field and its directional information is lost. At ultra high energy the tracing back of the sources is possible. However fluxes are very low and the proton becomes also sensitive to the GZK cutoff by interacting with the photons of the Cosmic Microwave Background (the mean free path is about z=1 for a 10 EeV proton and z=0.03 for a 100 EeV proton).
{\bf The neutron} is electrically neutral and points back to the source. However its life time is too short. Even at ultra high energy ($>$ EeV), it can not travel more than 10 kpc, which is roughly the radius of our Galaxy.
The last candidate is {\bf the neutrino}. It is probably the most promising: it is stable and neutral, allowing it to point back to its production point. Moreover, unlike the photon, it interacts only weakly with matter and photon fields. This last property leads to two major advantages and one drawback:
\begin{itemize}
\item It can easily escape from very dense sources which trap all the others particles.
\item It can travel over very large distances up to the detector.
\item A very large detector will be necessary to observe it on Earth.
\end{itemize}
Up to now this drawback has considerably limited its observation. The supernova SN1987A has been the first, and still the only, source of extraterrestrial neutrinos other than the sun.
The purpose of the ANTARES collaboration is to build a deep underwater neutrino detector. This one should be able to observe astrophysical sources and to study particle physics topics. After having discussed the scientific motivations and the constraints on the detector design, a status report on the project will be presented.
\section{Scientific Motivations}
\subsection{Sources of neutrinos}
The mechanisms leading to the production of neutrinos in the Universe can be classified into two categories: accelerating and non-accelerating mechanisms.
\subsubsection{Accelerating mechanisms}
Neutrinos can be produced according to the ``Beam Dump'' mechanism where an accelerated proton interacts with a target composed of protons and/or photons:
\begin{eqnarray}
\label{eq:nuprod}
p + \mathrm{target} \, (p,\gamma) \longrightarrow & \pi^{0} \quad , & \pi^{\pm} \quad , \quad (\ldots)\\
&\downarrow \quad \: & \: \downarrow \quad \quad \nonumber\\
&\gamma\gamma\quad \: & \mu^{\pm} \nu_{\mu} \rightarrow e^{\pm}
\nu_{\mu} \nu_{e} \nu_{\mu} \nonumber
\end{eqnarray}
This mechanism leads to the production of energetic neutrinos, $\nu_{\mu}$ and $\nu_e$, via the $\pi^{\pm}$ decays. The $\pi^{0}$ decay produces energetic photons\footnote{The high energy photons produced by this mechanism could be the ones detected by the \v{C}erenkov telescopes.}. According to the source of the accelerated proton and to the target, several neutrinos sources can be distinguished.
{\bf Diffuse neutrinos sources} are produced by the interaction of the Cosmic Rays with a target which can be the Earth atmosphere, the matter in the galactic plane or the photons of the Cosmic Microwave Background (table \ref{tab:diffuse}). Atmospheric neutrinos is the only source which has been observed so-far. The production of galactic and cosmologic neutrinos is guaranteed (the source and the targets exist !). However, even if there is a large uncertainty in the computations, all the predictions lead to very low fluxes and it will be difficult to observe them.
{\bf Point-like neutrinos sources} are obviously more attractive. The observation and the identification of a source will lead to a major progress in the knowledge of the origin of the most energetic Cosmic Rays. Except for the supernova remnants in which protons are accelerated in the expanding shell by the Fermi mechanism, all the sources involve a massive object, a pulsar or more efficiently a black hole. The presence of a very strong magnetic field together with an important plasma flow leads to the production of high energy protons by stochastic acceleration. The target can be very diversified: the surrounding matter for the young supernova remnants, a companion star for the X-ray binary systems, ... (table \ref{tab:pointlike}). A particularly interesting potential source is the AGNs. In their core a massive back hole (typically 10$^8$ M$_\odot$) accelerates protons at ultra high energy. Their interaction with the radiation field and/or the matter surrounding the black hole or in the jets could produce neutrinos. In any case, it must be pointed out that the existence of such point-like sources is still very speculative and is the subject of intense discussions between theorists (see for example \cite{ref2}, \cite{ref3} and \cite{ref4}). In any case, the best way to conclude the debate is to observe or not these neutrinos.
{\bf Gamma ray bursts} have a very peculiar place in this bestiary. They have been observed for many years and are currently observed at a rate of about one event per day. The bursts could be generated by highly relativistic shocks, within a concentrated fireball, which would give rise to the production of neutrinos during a very short time interval \cite{ref5}. The expected neutrinos fluxes are very low (their observation could be possible by making use of correlations with the direction and the timing of the signal seen by a gamma ray detector).
\begin{table}[h]
\begin{center}
\begin{tabular}{*{2}{l}}
\hline
{\bf Diffuse sources} & {\bf Target} \\
\hline
Atmospheric neutrinos & Earth atmosphere \\
\hline
Galactic neutrinos & Matter in the galactic plane \\
\hline
Cosmologic neutrinos & Photons of the CMB \\
\hline
\end{tabular}
\caption{Diffuse neutrinos sources produced by the interaction of the Cosmic Rays with different targets. Only the atmospheric neutrinos have been directly observed so-far. The others diffuse sources are guaranteed, but the fluxes are very low.}
\label{tab:diffuse}
\vspace{9mm}
\begin{tabular}{*{3}{l}}
\hline
{\bf Point-like sources}& {\bf Source of proton} & {\bf Target} \\
\hline
Young supernova remnant& Plerion or expanding shell& Expanding shell \\
\hline
Supernova remnant & Expanding shell & Expanding shell \\
\hline
X-ray binary system & Pulsar or black hole & Accreting matter or \\
& & companion star \\
\hline
& & Matter surrounding \\
AGN & Massive black hole & the black hole or \\
& & ambient radiation field \\
\hline
\end{tabular}
\caption{Possible point-like neutrinos sources as a function of the origin of the accelerated protons and of the target.}
\label{tab:pointlike}
\end{center}
\end{table}
\newpage
\subsubsection{Non-accelerating mechanisms}
Neutrinos could be also produced by the annihilation or the decay of elementary heavy particles predicted by some theoretical models in Particle Physics.
One of the most promising extension of the Standard Model in Particle Physics is the Minimal SuperSymmetric Model (MSSM). In the simplest version of this theory the Lightest Supersymmetric Particle (LSP) is the lightest neutralino which is stable and massive (its mass is constrained experimentally above $\sim$50 GeV and theoretically below $\sim$1 TeV). It is a natural candidate to the dark matter which composes most of the matter of the Universe. The LSPs which have been produced just after the Big Bang, move in the Galaxy halo at few hundreds km/s. They can lose energy by elastic scattering on nuclei of the Earth and the Sun matter. They can be trapped by the gravitational field of the Earth or the Sun and can be concentrated in their centres where they would annihilate\footnote{Neutralinos are Majorama particles: they are their own anti-particle.} and would produce neutrinos.
A more exotic candidate for the neutrinos production is the decay of very massive particles which could have been abundantly produced after the Big Bang. The decay of monopoles or cosmic strings would release a huge energy which could produce high energy neutrinos.
\subsubsection{Neutrinos fluxes}
The figures \ref{fig:fluxagn} and \ref{fig:fluxatm} show the expected neutrinos fluxes from a selection of different sources. Two important remarks can be extracted from these figures:
\begin{itemize}
\item Neutrino fluxes are very low and a large detector will be necessary: about \hbox{1 km$^2$.}
\item If we want to observe astronomical diffuse sources, the detector must be able to observe neutrinos with an energy larger than about 10 TeV. Below this limit, atmospheric neutrinos dominate all the other potential sources.
\end{itemize}
\begin{figure}
\begin{minipage}[t]{.495\linewidth}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig1.eps, width=.95\linewidth}
\caption{Expected diffuse neutrinos fluxes from AGNs models. For the models NMB \cite{ref6} and SDSS \cite{ref7}, the neutrinos are produced in the vicinity of the massive black hole. For the models PRO \cite{ref8}, MRLA and MRLB \cite{ref9}, they are produced in the jets. The irreducible background due to the atmospheric neutrinos ATM \cite{ref10} is also indicated.}
\label{fig:fluxagn}
\end{minipage}
\hfill
\begin{minipage}[t]{.495\linewidth}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig2.eps, width=.95\linewidth}
\caption{Expected diffuse flux from the galactic neutrinos \cite{ref11}. Two models of cosmologic neutrinos using two different normalisations are also displayed (COS-2 and COS-4) \cite{ref11}. The irreducible background due to the atmospheric neutrinos ATM \cite{ref10} is indicated.}
\label{fig:fluxatm}
\end{minipage}
\end{figure}
\newpage
\subsubsection{Neutrino oscillation}
Atmospheric neutrinos are an irreducible background for the study of high energy cosmic neutrinos. However, at lower energy (typically between 5 and 60 GeV), they can be a very interesting signal for the neutrino oscillation study.
The SuperKamiokande experiment has announced recently the observation of a $\nu_\mu$ deficit \cite{ref1} which is interpreted as the evidence for $\nu_\mu$ mixing with either $\nu_\tau$ or a sterile neutrino. The mixing angle has been found to be $sin^2 (2 \theta) > 0.82$ and the most likely value of the square mass difference is $\Delta m^2 = 2.2 \cdot 10^{-3}$\,eV$^2$ with a confidence interval between $5 \cdot 10^{-4}$ and $6 \cdot 10^{-3}$ eV$^2$.
A possible method to cross check this result is to measure the energy spectrum of the atmospheric neutrinos. With the hypothesis of two-neutrino mixing, the oscillation probability is given by:
\begin{equation}
P = \sin^2 (2 \theta) \sin^2 \left( 1.27 \frac{L}{E}
\Delta m^2 \right) \label{proba_oscill}
\end{equation}
where $\theta$ is the mixing angle, $L$ is the distance travelled by the neutrino in km, $E$ is the neutrino energy in GeV and $\Delta m^2$ is the difference of the square of the masses in eV$^2$.
If only the upward going neutrinos are studied, the baseline $L$ is constant (the Earth diameter, $\sim$12740 km). Moreover, if the SuperKamiokande result is correct, the first maximal neutrino flux suppression will be observed at an energy given by:
\[ E = 10 \times \left( \frac{\Delta m^2}{10^{-3}{\mathrm{eV}}^2} \right)
{\mathrm{GeV}} \]
Other values of the series can be calculated by dividing this first value by an odd integer number. Therefore, an important dip in the energy spectrum will be visible for the upward going neutrinos in the energy range of 5 to 60 GeV.
\section{The detection of neutrinos}
\subsection{Principle}
A cheap and efficient method to observe a muonic neutrino is to detect the muon produced from its interaction with the matter surrounding the detector. When passing through water or ice, the muon emits \v{C}erenkov light which is detected by a three-dimensional matrix of photomultiplier tubes. The measurement of the arrival time of the \v{C}erenkov light on the photomutipliers allows to reconstruct the muon direction, the amount of light giving an estimate of its energy.
Moreover the $\nu$N cross-section and the muon range increase with the energy: the probability to detect a muon coming from a neutrino increases with the neutrino energy. The most energetic neutrinos are therefore enhanced, improving the detector performance at high energy.
\subsection{Background}
The observation of astronomical sources is subject to two sources of background: the neutrinos and the muons which are copiously produced by the Cosmic Rays interaction with the Earth atmosphere.
Atmospheric neutrinos are an irreducible source of background for the study of the cosmic neutrinos (how to distinguish a neutrino produced in the Earth atmosphere from one produced inside an AGN ?). However, as it is visible on the figure \ref{fig:fluxagn}, they follow a E$^{-3}$ law and their flux becomes less important than the one of the signal beyond 10-100 TeV.
The protection against the atmospheric muons is a more delicate task and have an important impact on the detector design:
\begin{itemize}
\item A very important shielding is necessary (figure \ref{fig:muon1}). The muons flux can be decreased by more than 4 orders of magnitude by installing the detector at 2400 meters deep.
\item Nevertheless this protection is not sufficient: muons flux dominates atmospheric neutrinos flux by more than 6 orders of magnitude at small zenithal angle at 2400 meters deep (figure \ref{fig:muon2}). Even at 4000 meters deep, it overcomes the atmospheric neutrino flux by several orders of magnitude and is too important. However, muons are easily stopped by the Earth matter and no upward going muons are observed: to suppress this background, upward going particles must be detected.
\end{itemize}
\begin{figure}
\begin{minipage}[t]{.495\linewidth}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig3.eps, width=.95\linewidth}
\caption{Flux of downward going muons as a function of the detector depth in water equivalent.}
\label{fig:muon1}
\end{minipage}
\hfill
\begin{minipage}[t]{.495\linewidth}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig4.eps, width=.95\linewidth}
\caption{Muons flux as a function of the zenith angle ($\Theta_z=0^\circ$ corresponds to downward going muons), at a depth of 2400 m water equivalent.}
\label{fig:muon2}
\end{minipage}
\end{figure}
\subsection{Constraints on the detector}
From the following remarks, the most important constraints on the detector design can be summarised:
\begin{itemize}
\item Signal fluxes are very low: the detector surface must be important, about \hbox{1 km$^2$.} However a 0.1 km$^2$ surface should be sufficient to start the study of the cosmic neutrinos.
\item Due to the background, the detector must:
\begin{itemize}
\item Be well shielded.
\item Observe upward going muons to remove the atmospheric muons.
\item Have a good energy resolution to reconstruct the energy spectra.
\item Have a good angular resolution to isolate properly a point-like source and optimise the signal to noise ratio (as discussed in the section {\it 7.3}, the mean number of background events is directly proportional to the pixel size which in turn depends of the angular precision).
\end{itemize}
\end{itemize}
\section{The ANTARES collaboration}
The ANTARES\footnote{ANTARES stands for Astronomy with a Neutrino Telescope and Abyss environmental RESearch.} collaboration \cite{ref12} aims at building a large undersea high energy neutrino telescope. This collaboration, which started in the Spring of 1996, includes astronomers, astrophysicists, oceanographers and particle physicists from France, the Netherlands, Russia, Spain and United Kingdom. The detector should be sufficiently multi-purpose to be able to observe astrophysical sources and to study particle physics topics.
The construction of a km$^2$ undersea detector is a very complex challenge which must be reached by steps. In this objective the ANTARES collaboration aims at building a 0.1 km$^2$ detector which should be able to observe cosmic neutrinos and to demonstrate the feasibility of such a detector. For that the collaboration has defined a R\&D program which can be divided into three main parts:
\begin{enumerate}
\item Learning of the knowhow to install and to operate strings equipped of photomutipliers and connected to the shore.
\item Measurements of the environmental parameters in deep sea to characterise and identify the sites suitable to the installation of a neutrino detector.
\item Computer simulations to optimise the detector geometry and the analysis algorithms.
\end{enumerate}
The km$^2$ detector site has not been chosen yet. It will be guided by the properties of the site itself but also of the closeness to the shore, the meteorological conditions, the support available onshore and the boat disponibilities. During the R\&D program a convenient site in the Mediterranean sea situated at about 20 miles of the Porquerolles Islands (south of France) and at \hbox{2400 m} deep has been used. The measured properties of this site (discussed hereafter) have shown no major drawback and it will be used for the next step of the project, the \hbox{0.1 km$^2$} detector.
The different points of the R\&D program will be now discussed in more details.
\section{Development of the strings}
\subsection{Detector layout}
The final detector will have a large number of instrumented vertical strings. These ones will be installed on the sea bottom to form a tridimensional matrix of photomultipliers (the distance between the strings will be about 80 meters). The whole detector will be connected to the shore by an electro-optical cable which ensures the data transmission and the power supply.
Each string is about 400 meters long and is equipped with about 100 photomultipliers grouped by floors, with an inter-floor distance between 8 m and 15 m. To facilitate their deployment, the strings are flexible. Therefore an accurate positioning system is necessary to monitor their shape during and after the deployment (deep sea currents modify the shape). Some other instruments are also necessary to monitor the properties of the sea water (water optical properties, monitoring of the glass transparency, ...), to calibrate in time and in energy the detector, ...
The conception of such a detector is not an easy task. To succeed in this operation, the development has been divided in several parts which are pursued in parallel. We will now give more details in the three most important studies: the development of the Optical Modules (``the eyes'' of the detector), the design and the deployment of a string prototype and the tests of connection in deep sea.
\subsection{The Optical Module}
Due to the ambient pressure at 2400 meter deep, photomultipliers must be housed inside pressure-resistant glass spheres (figure \ref{fig:om}). In these so-called Optical Modules the photomultiplier is embedded in silicon gel to ensure a good optical coupling and to maintain it mechanically. To optimise the light detection, hemispherical photomultipliers with a photocathode surface larger than 8 inches and with a very good time resolution are used. A high permittivity alloy cage surrounds it, shielding it against the Earth magnetic field. A DC/DC converter supplying power to the photomultiplier is also included in the sphere. Signal outputs, high voltage monitoring, ... are transmitted through water and pressure-resistant connectors to the outside world.
Digital transmission will be used to transmit data to the shore. The digitisation of the photomultiplier signal will be performed at the Optical Module level. An original design for an Application Specific Integrated Circuit (ASIC) has been developed by the ANTARES collaboration. It samples at a frequency up to 1 GHz and memorises analog pulses in an array of MOS capacitors. Data corresponding to the very frequent single photoelectron pulses are not numerized, only a minimal information (charge and arrival time) is transmitted to the shore.
\begin{figure}
\centering
\epsfig{file=Minisymposium08_SBASA1_fig5,width=9cm}
\caption{Sketch of an Optical Module. A large hemispherical photomultiplier (diameter greater than about 8 inches) is protected from the ambient pressure by a pressure-resistant glass sphere. The outer diameter of this sphere is 17 inches (about 42.5 cm).}
\label{fig:om}
\end{figure}
\subsection{Test of the string prototype}
A first string prototype has been developped to learn the knowhow to build, deploy and operate such a structure. A sketch of this string is shown on the figure \ref{fig:line5}.
Even if this string prototype will be equipped of only 8 photomultipliers, all the elements necessary for the good running of the detector have been installed: slow control, energy calibration, time calibration, positioning (the system composed of hydrophon, compass and tiltmeter allows the monitoring of the position of each photomultipliers with a precision better than 10 cm), ... To decrease the influence of the optical background, the photomultipliers are clustered by two and can be put into coincidence. One should note that due to the development schedule of the digital transmission described previously, a first version of the signals readout and transmission using a simple on the shelf analog electronics is being used. However the digital transmission will be used for the next strings.
To test the mechanical behaviour of this string in deep sea, this one has been successfully deployed in July-August 1998. In the Spring 1999, it will be equipped of 8 photomultipliers, deployed and connected to the shore via an electro-optical cable. Data should be taken for several months.
It must be emphasized that the design of the strings foreseen for the next step of the project will be slightly different. For example, computer simulations have shown that the Optical Modules must be clustered by three to optimize the detector performances (they are clustered by two for this prototype).
\begin{figure}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig6.eps, width=14cm}
\caption{Schematic view of the first string prototype. An Optical Module is composed of a photomultiplier housed inside a pressure resistant glass sphere. The hydrophons, compass and tiltmeters are used to monitor the string shape and to measure very precisely the position of each Optical Module.}
\label{fig:line5}
\end{figure}
\subsection{Tests of connection in deep sea}
As already pointed out, an electro-optical cable will be used to connect the detector to the shore. It would not be reasonable in terms of cost and efficiency to have one cable per string. Therefore, each string will be connected to a junction box which in turn will be connected to the shore via the electro-optical cable.
To connect the strings to the junction box, it is foreseen to use a submarine. This one will be furnished by the IFREMER \cite{ref13} which participates to the ANTARES collaboration. This institute owns in particular two deep sea submarines, the NAUTILE (up to 6000 m deep) and the CYANA (up to 3000 m deep) which can be used for this operation. A first test of deep sea connection has been successfully performed in December 1998 with the NAUTILE. These tests will be pursued in the next years.
\section{Measurements of the environmental parameters}
\subsection{Introduction}
The detector geometry and its performances are directly linked to some environmental parameters:
\begin{itemize}
\item {\bf The optical background}. Due to the $^{40}$K decay and to the life activity in sea water (the bioluminescence), the optical background has important constraints on the trigger logics and the electronics as well as the mechanical layout of the photomultipliers.
\item {\bf The effect of the biofouling and the sedimentation on the glass transparency}. Due to the ambient pressure at 2400 m deep, photomutipliers are housed inside a pressure resistant glass sphere. Biofouling and sedimentation deposition can diminish the glass transparency.
\item {\bf The water optical properties}. The absorption and the scattering length influence the detector design (distance between strings and between Optical Modules on the same string).
\end{itemize}
To measure these parameters, the ANTARES collaboration has built and deployed several autonomous strings allowing long term measurements (these environmental parameters are obviously local and seasonal). Each set-up is mounted on a mooring line which is held vertically by a buoy (see for example the figure \ref{fig:line1}). It is placed at about 100 meters from the sea bottom which might be muddy. All the strings are immersed from ships belonging to the French Insitut National des Sciences de l'Univers (INSU-CNRS).
More than 50 trips with the INSU-CNRS ships have been already performed. Most of the measurements have been done at a depth of approximatively 2400 m, in the current ANTARES site.
\begin{figure}
\centering
\psfig{figure=Minisymposium08_SBASA1_fig7.eps, width=9cm}
\caption{Schematic view of the string used to measure the optical background. The string is maintained vertically by the buoy. When the measurement is achieved, the acoustic releases are activated and the string floats back to the surface where a boat can recover it.}
\label{fig:line1}
\end{figure}
\subsection{Optical background}
To measure the spatial extension of the optical background, the string is equipped of three Optical Modules identical to the ones used for the string prototype (figure \ref{fig:line1}): two Optical Modules are 0.5 to 1.5 m apart, and the other one is 10 to 40 m away. A current-meter is also installed to analyse the correlation between the sea current and optical background.
The measurement consists in recording the counting rate of the photomultipliers as a function of time and at different signal amplitude. By putting two photomultipliers in coincidence, it is possible to measure the spatial extension of the optical background. The figure \ref{fig:k40} shows the typical counting rate measured by a 8 inches photomultiplier. The baseline at about 40 kHz is due to the $^{40}$K decay and to a constant contribution of the bioluminescence. The bursts are due to sudden bioluminescence activities.
From our measurement it appears that:
\begin{itemize}
\item By putting two photomultipliers in coincidence, the optical background can be diminished to a reasonable level, lower than 50 Hz for 8 inches photomultipliers.
\item The dead time due to the bioluminescence bursts is small, between 3 and 4 \%.
\item A clear correlation between this bioluminescence activity and the deep sea current has been observed (figure \ref{fig:current}).
\end{itemize}
\begin{figure}
\begin{minipage}[t]{.495\linewidth}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig8.eps, width=.95\linewidth}
\caption{Counting rate measured by a single 8 inches photomultiplier (upper histogram). The counting rate distribution is shown in the lower histogram. The pulse height threshold has been fixed to 1/3 of the mean amplitude of the single photoelectron signal.}
\label{fig:k40}
\end{minipage}
\hfill
\begin{minipage}[t]{.495\linewidth}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig9.eps, width=.95\linewidth}
\caption{Percentage of time where the counting rate of the photomultiplier exceeds by 15 \% the baseline, as a function of the deep sea current. A clear correlation between the sea current and the bioluminscence activity is visible.}
\label{fig:current}
\end{minipage}
\end{figure}
\newpage
\subsection{Biofouling and sedimentation effect}
To measure the effect of the biofouling and of the sedimentation on the glass transparency, a blue LED source has been placed inside a glass sphere and a set of PIN diodes glued inside a second sphere which was placed one meter apart (figure \ref{fig:pindiodes}). The measurement consists in monitoring the amount of light seen by each PIN diodes as a function of time. To measure a significant effect the set-up must be immersed for several months. Therefore, an acoustic modem has been installed to recover the data from time to time.
The effect on the glass transparency is visible on the figure \ref{fig:sedim}. The influence of the biofouling and of the sedimentation is less important at the equator than at the top of the sphere. If we suppose that the attenuation effect is the same on the two spheres, the attenuation at the equator on one glass sphere doesn't exceed 2.5\%/2=1.25\% in 8 months.
From this measurement, it has been decided that the photomultipliers must be faced down to be insensitive to the biofouling and to the sedimentation.
\begin{figure}
\begin{minipage}[t]{.495\linewidth}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig10.eps, width=.95\linewidth}
\caption{Positioning of the LED source and of the PIN diodes to monitor the effect of the biofouling and of the sedimentation on the glass transparency.}
\label{fig:pindiodes}
\end{minipage}
\hfill
\begin{minipage}[t]{.495\linewidth}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig11.eps, width=.95\linewidth}
\caption{Light transmission monitored over 240 days (about 8 months). The polar angle of each PIN diodes is indicated (0$^\circ$ corresponds to the top of the sphere and 90$^\circ$ to the equator). The intensities have been normalised to the first measurement.}
\label{fig:sedim}
\end{minipage}
\end{figure}
\subsection{Water optical properties}
The water optical properties are very important for the definition of the detector geometry (distance between Optical Modules on the same string and distance between strings) and for its performances. In particular a good scattering length is particularly important to obtain a good angular resolution.
To measure the attenuation, the absorption and the scattering length, two different strings have been conceived:
\begin{itemize}
\item The first string is equipped of a 32 m long rail used to guide a mobile cart. A light source is placed at the top of the rail and a photomultiplier on the cart. With this set-up the attenuation length can be measured by monitoring the amount of light seen by the photomultiplier as a function of the distance to the source (figure \ref{fig:att}).
\item A second flexible string has been developped to measure the absorption and the scattering length. The set-up is mainly composed of an isotropic pulsed light source and a 1 inches photomultiplier equipped of a TDC. The distance between then can be either 44 m or 25 m. The measurement consists in monitoring the arrival time of the photons emitted by the pulsed light source: the proportion of scattered photons which are delayed with respect to the direct photons is linked to the scattering length (figure \ref{fig:scatt}).
\end{itemize}
With these two strings, it has been possible to measure in-situ the water optical properties. An attenuation length of 41 $\pm$ 1$_{stat}$ $\pm$ 1$_{syst}$ meters in the blue (460 nm) has been measured and a scattering length greater than 100 meters. From our measurement, it appears that the water optical properties in deep sea are promising.
\begin{figure}
\begin{minipage}[t]{.495\linewidth}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig12.eps, width=.95\linewidth}
\caption{Attenuation length \hbox{measured by monitoring the amount} of light seen by a photomultiplier ($\Phi_{LED}$) as a function of the distance to the light source ($D$). In air, the amount of light measured follows a standard 1/D$^2$ law and the ratio $D^2$/$\Phi_{LED}$ is constant. In water, the slope is due to the attenuation.}
\label{fig:att}
\end{minipage}
\hfill
\begin{minipage}[t]{.495\linewidth}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig13.eps, width=.95\linewidth}
\caption{Time distribution of the arrival time of the photons (1 TDC bin corresponds to 5.5 ns). The distance between the pulsed light source and the photomultiplier is 25 m for the left curve and 44 m for the right curve. The tail is due to the scattered photons. From the shape of these spectra, it is possible to extract the absorption and the scattering length.}
\label{fig:scatt}
\end{minipage}
\end{figure}
\section{Computer simulations}
Computer simulations are mandatory to optimise the detector geometry and to estimate its performances.
\subsection{Detector geometry}
The muon track is defined by 5 parameters: one point on the track (2 coordinates), two angles and one reference time. To reconstruct these parameters the arrival time of the photons on the photomultipliers is used. Therefore, only 5 hit photomultipliers are necessary to reconstruct a muon track.
The estimation of the muon energy is a more delicate task:
\begin{itemize}
\item In the energy range suited for the neutrino oscillation study (between 5 and \hbox{60 GeV),} the muon track is contained in the detector. For the events interacting inside the fiducial volume of the detector, a good estimation of the muon energy can be performed by measuring its range.
\item At high energy ($>$ TeV), the muon range is greater than the detector size and the previous method can not be used. A possible method is to use the number of detected photons. However the energy loss of an energetic muon is dominated by catastrophic energy loss from Bremsstrahlung and pair production, and only a crude estimation of the energy will be possible.
\end{itemize}
Considerable efforts have been performed to optimise the geometry of a 0.1 km$^2$ detector ( many details can be found in \cite{ref14}). It appears that strings must be absolutely arranged in a way to avoid symmetries and the photomultipliers should be slanted at 45$^\circ$ below the horizon (with such an orientation, the detector response over the lower hemisphere is assured). The distance between strings and between Optical Modules on the same strings is not yet clearly defined. For the study of astronomical objects, the most important point is the surface of the detector due to the expected very low fluxes. In this case, the distance between strings would be about 60 meters and the distance between the Optical Modules about 16 meters. At the opposite, for the study of the neutrino oscillation, a better sampling of the signal is necessary to measure accurately the muon range.
\subsection{Performances of a 0.1 km$^2$ detector}
The results given hereafter have been obtained with a geometry optimised for the study of the astronomical objects. The detector is composed of 1000 photomultipliers arranged over 15 strings, the distance between strings is 80 meters and the distance between Optical Modules 16 meters.
With such a detector, a very good angular resolution has be obtained: half of the events have an angular error better than 0.2$^\circ$ (figure \ref{fig:resolangl}). This angular resolution is dominated by the error on the reconstruction: the angular spread due to the physical process is negligible above few TeV. Moreover, even if this angular precision seems poor according to the astronomical telescopes and radio-telescopes (0.2$^\circ$ is the Moon or the Sun radius !), such a precision is the same as the one obtained with the current \v{C}erenkov telescopes which observe TeV photons.
As expected, the energy resolution of such a detector is poor at high energy \hbox{($>$ TeV):}
\begin{itemize}
\item A precision of a factor 3 can be obtained below 10 TeV,
\item A precision of a factor 2 beyond 10 TeV.
\end{itemize}
This precision is sufficient to reconstruct the energy spectrum and to apply the energy cut necessary to suppress the atmospheric neutrinos background (figure \ref{fig:all_spec_rec}).
Moreover the detection efficiency of the detector increases with the energy (figure \ref{fig:area}). An effective surface of 0.055 km$^2$ is reached at 10 TeV and 0.1 km$^2$ beyond 100 TeV. Such a large surface leads to a measurable neutrinos fluxes as displayed on the table \ref{tab:numofevents}. It is also possible to see that the number of atmospheric neutrinos can be kept at a reasonable level. For the atmospheric muons, only an upper limit has been obtained so-far (less than 1000 events per year) due to limitations on the CPU time. However computer simulations are pursued and this limit will be improved.
\begin{figure}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig14.eps, width=10cm}
\caption{Angles between the neutrino and the reconstructed muon as expected with a 0.1 km$^2$ detector. Half of the events have an angular error better than 0.2$^\circ$. The neutrino spectrum which has been simulated follows a E$^{-2}$ law.}
\label{fig:resolangl}
\end{figure}
\begin{figure}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig15.eps, width=0.495\linewidth}
\epsfig{figure=Minisymposium08_SBASA1_fig16.eps, width=0.495\linewidth}
\caption{Reconstructed energy spectrum of muons issued from neutrinos. A 0.1 km$^2$ detector composed of 1000 photomultipliers has been considered. For the models NMB \cite{ref6} and SDSS \cite{ref7}, the neutrinos are produced in the vicinity of the massive black hole. For the models PRO \cite{ref8} and MRLB \cite{ref9}, they are produced in the jets. The atmospheric neutrinos ATM \cite{ref10} which is an irreducible background is also displayed. The error bars are only statistical.}
\label{fig:all_spec_rec}
\end{figure}
\begin{figure}
\centering
\epsfig{figure=Minisymposium08_SBASA1_fig17.eps, width=9cm}
\caption{Triggered and reconstructed effective surface as a function of the muon energy. The surfaces are averaged over the whole lower hemisphere and the errors bars are only statistical.}
\label{fig:area}
\end{figure}
\begin{table}[h]
\begin{center}
\begin{tabular}{*{3}{l}}
\hline
{\bf AGNs models} & E$_\mu^{\mathrm{rec}} > 10$~TeV & E$_\mu^{\mathrm{rec}} > 100$~TeV\\
\hline
SDSS & 251$\pm$12 & 134$\pm$10 \\
NMB & 217$\pm$9 & 64$\pm$4 \\
PRO & 34$\pm$2 & 21$\pm$2 \\
MRLB & 7.8$\pm$0.4 & 2.6$\pm$0.3 \\
\hline
& & \\
\hline
{\bf Background source} & E$_\mu^{\mathrm{rec}} > 10$~TeV & E$_\mu^{\mathrm{rec}} > 100$~TeV\\
\hline
Atmospheric neutrinos & 68$\pm$13 & 0.8$\pm$0.1 \\
\hline
\end{tabular}
\vspace{3mm}
\caption{Number of detected events in one year with a 0.1 km$^2$ deep sea detector for different AGNs models (diffuse neutrinos source). The cut on energy is directly applied to the reconstructed energy of the muon. In the SDSS \cite{ref7} and NMB \cite{ref6} models the neutrinos are produced in vicinity of the black hole and in the PRO \cite{ref8} and MRLB \cite{ref9} models in the jets. The number of atmospheric neutrinos is also indicated. For the atmospheric muons, only an upper limit have been obtained (less than 1000 events per year) due to the present limitations on the CPU time available for the simulation. The indicated errors are statistical.}
\label{tab:numofevents}
\end{center}
\end{table}
\newpage
\subsection{Astronomical potentialities}
Given the detector latitude (43$^\circ$ North), about 3.5 $\pi$ steradian can be observed thanks to the Earth rotation (such a detector works all day long and observes continuously half of an hemisphere). The declinations below -47$^\circ$ are always observable, part of the day between -47$^\circ$ and +47$^\circ$, never above +47$^\circ$. Such an observable sky is particularly convenient: about 300$^\circ$ of the galactic plane (including the galactic center) is visible most of the sideral day. An other convenient feature is the fact that there will have an overlap between ANTARES and AMANDA \cite{ref15} in the South pole which observes between +90$^\circ$ and 0$^\circ$.
With the expected performances, a 0.1 km$^2$ detector should be able to observe diffuse neutrinos sources. For point-like sources, the situation should be also favourable. Due to the very good angular resolution, the observable sky will be covered by about 200000 pixels\footnote{This number of pixel is computed by supposing an angular resolution of 0.2 degrees. However by using the source position in the sky, a better angular resolution could be achieved for point-like sources. Therefore the sky could be covered by a larger number of pixel and the detection of point-like source easier.}. The mean number of background events in one pixel will not exceed 1.3$\cdot 10^{-2}$ per year and an excess of about 3 neutrinos inside a pixel will be sufficient to observe a point-like source.
\subsection{Study of the Neutrino Oscillation}
The ANTARES collaboration has studied the possibility to cross-check the SuperKamiokande result \cite{ref1} by using the method described in the section {\it 2.1.4}. Results are yet preliminaries and more work is needed. Only the most important features will be given here.
First of all a shorter vertical spacing of the Optical Modules is necessary to have a better sampling of the muon track (typically a distance of 8 meters between Optical Modules is necessary). However it appears that the optimal geometry for neutrino oscillation is not very far from the one optimal for astronomy. Concerning the angular resolution, this one is dominated by the physical process, not really by the track reconstruction due to the low muon energy (between 5 and 60 GeV). For example, by studying vertical events seen by only one string, half of the events have an angular error between the neutrino and the muon better than 1.8$^\circ$. However, if the events seen by two strings are used, we are sensitive to events having a larger energy and the angular resolution reaches 0.4$^\circ$.
Finally the background is mainly due to the external $\nu_\mu$ which are reconstructed as contained. The contribution of the atmospheric muons and of the contained $\nu_e$ is negligible. Even if more work is necessary, the background seems to be handled and a 0.1 km$^2$ detector as foreseen by the ANTARES collaboration should be able to explore the region of SuperKamiokande signal in a few years running.
\section{Conclusion}
Since its beginning in Spring 1996, considerable efforts have been done by the ANTARES collaboration to design a deep underwater neutrino detector. During the R\&D phase of the project, several autonomous set-up have been built and deployed many times to measure the environmental parameters. In parallel a string prototype has been conceived to study the problems of deployment and operating of such a structure. Finally extensive computer simulations have been pursued to optimise the detector geometry and to estimate its performances. It appears that a 0.1 km$^2$ should be able to observe diffuse sources and several point-like sources.
Now, the ANTARES collaboration aims at building a 0.1 km$^2$ detector. Its performances should be sufficient to study diffuse neutrinos sources and to observe the nearest point-like sources. The study of neutrino oscillation could be also feasible. The very first string would be deployed at the end of 2000 and the whole detector completed by the end of 2003.
Such a detector could open a new window on the Universe, in the same way as Galileo's first telescope did some 400 years ago...
\section*{Acknowledgements}
I would like to thank F. Bernard, C. C\^arloganu, F. Hubaut and S. Navas for their help and their support, and L. Moscoso for fruitful discussions.
\section*{References}
|
\section{Introduction}
\setcounter{equation}{0}
During the last decade, the problem of the description of hadrons in dense
and hot nuclear matter received great attention. It is well known that
particles change their properties when they are placed into a medium.
Already in the seventies the reduction of the nucleon masses in nuclei was
implemented into the Walecka model in the framework of effective hadron
field theory \cite{Wal,Chin}. Later on this effect was put on firmer grounds
on the basis of a partial restoration of chiral symmetry and finite-density
QCD sum rules \cite{QCD}. The change of the meson properties is also
discussed in the quantum hadrodynamics \cite{Chin} and finite-density QCD
sum rules \cite{MES}. The investigations in the Nambu-Jona-Lasinio model
provide an evidence for reduction of nucleon and meson masses at finite
density and temperature as well \cite{Fae}.
On the other hand, many-body correlations lead to a dressing of the
particles inside the medium and strongly modify spectral properties of the
mesons \cite{herrmann93,rapp97,peters98}. Thus one expects a significant
reduction of the corresponding life times which, loosely speaking, result in
melting the mesons in the nuclear environment.
The melting of the higher nucleon resonances is established experimentally
from the measurement of the total photoabsorption cross section on heavy
nuclei \cite{Bia}. There is a clear signal for a change of the shape of the $%
\Delta $-resonance in nuclei, related to the Fermi motion and, as noticed in
Ref. \cite{Kon}, to the collision broadening effect discussed first by
Weisskopf \cite{Wei} in connection with a broadening of the atomic spectral
lines in gases. The higher nucleon resonances are not seen in the cross
section due to a strong collision broadening effect.
The purpose of the current investigations is to determine mass shifts and
the broadening of hadronic resonances in nuclear matter. The formulation of
this problem can be traced back to the atomic spectroscopy where the shifts
of atomic energy levels and the broadening of atomic spectral lines in dense
and hot gases is a relatively well studied subject (see {\it e.g.} Ref. \cite
{SOB}).
The search for signatures of modified hadron properties such as reduction
masses according to the Brown-Rho scaling \cite{BR95} are presently pursued
with high experimental efforts. However, most hadronic probes loose
important information on the early and violent phase of the reaction by
strong final state interactions. One of the most promising probes to study
in-medium properties of vector mesons are dilepton pairs $\ell ^{+}\ell ^{-}$
($\ell =e,$ $\mu $) which are produced from decays of $\rho ^0$-, $\omega $%
-, and $\phi $-mesons in heavy-ion collisions. The leptonic probes have the
advantage that they are the nearly undistorted messengers from the
conditions at their creation, unlike {\it e.g.} pions.
The dilepton spectra measured by the CERES and HELIOS-3 Collaborations at
CERN SPS have attracted special interest \cite{ceres,helios}. Compared to
theoretical predictions, both experiments found a significant enhancement of
the low-energy dilepton yield below the $\rho $ and $\omega $ peaks. One way
to explain this low-energy dilepton excess is to assume the scenario of a
significant reduction of the $\rho $-meson mass in a dense medium \cite
{koch92,li95,Cassing95}. On the other hand, a more sophisticated treatment
of the $\rho $-meson spectral function which includes the broadening in
dense matter \cite{rapp97,peters98} seems also to be sufficient to account
for these data \cite{Cassing95,Cassing98}.
An excess of low-energy dileptons occurs already at moderate bombarding
energies. However, the spectra obtained by the DLS Collaboration at the
BEVALAC \cite{DLS} for the incident energies around $1$ $A\cdot $GeV cannot
be reproduced by present transport calculations \cite{ernst}. Even using the
reduction of the $\rho $-meson mass, there remains a discrepancy for the
dilepton yield by a factor of $2$ to $3$ \cite{ernst,BK}. Also the medium
dependence of the spectral function, even in combination with a dropping $%
\rho $ mass, does not provide an explanation for this so called 'DLS puzzle'
\cite{brat98,BC}. It is interesting to note that this fact is independent on
the system size and occurs in light ($d+Ca$) as well as in heavier systems ($%
Ca+Ca$). The dilepton yield in elementary $p+p$ collisions measured by the
DLS Collaboration \cite{Wil} is also underestimated by standard theoretical
descriptions \cite{ernst}. In Ref. \cite{pppd} it was claimed that the
discrepancy between the DLS data and theoretical simulations disappears if
additional background contributions from dilepton decays of higher nucleon
resonances (mainly $N^{*}(1520)$) are taken into account. This demonstrates
that a most precise knowledge of the background is indispensable for the
interpretation of the present and future dilepton data. The HADES experiment
at GSI, Germany, will focus on these topics to a large extent \cite{hades}.
While the dilepton spectra stemming from the mesonic decays are not
distorted by final state interactions, the problem of extracting information
on the in-medium properties of the vector mesons is a specific theoretical
task, due to a large amount of decays contributing to the background. A
precise and rather complete knowledge of the relative weights for existing
decay channels is therefore indispensable in order to draw reliable
conclusions from dilepton spectra.
The study of the dilepton decays is useful also for the search on the
dilepton decay modes of the light unflavored mesons. These decays give a
deeper insight into meson structure, allowing to measure transition form
factors at the time-like (resonance) region. In four-body decays the decay
probabilities are determined by the half-off-shell meson transition form
factors which cannot be measured in other reactions. There are plans to
study the $\eta ^{\prime }$ transition form factors in the space-like region
in reactions of the photo- and electroproduction at CEBAF energies \cite{Dav}%
. Recently, the DLS Collaboration published data on the dilepton production
in the elementary $pp$ and $pd$ collisions \cite{Wil}. These data are
analyzed in Refs. \cite{ernst,pppd}. The plans from the HADES Collaboration
include measurements of the dilepton spectra from proton-proton and
pion-proton collisions \cite{hades}. These experiments are stimulated by the
already mentioned discrepancy between the number of the observed dilepton
events and results of the transport simulations, that indicate a limited
understanding of the mechanism for the dilepton emission in heavy-ion
collisions. In such experiments, there is a possibility for exclusive
measurements of the different mesonic channels.
In the present work, we perform a detailed study of possible mesonic $\ell
^{+}\ell ^{-}$ decays which appear in the SIS energy range, {\it i.e.} at
lab. energies below $2$ GeV/nucleon. At $2$ GeV/nucleon, the $\rho $-meson
is produced slightly above the threshold. Due to statistical fluctuations,
the production of heavier mesons is, however, also possible. We consider
decays of unflavored light mesons with masses below the $\phi (1020)$-meson
mass within the framework of the effective meson theory. The vertex
couplings are determined from the measured strong and radiative decay widths
and, when the experimental data are not available, from $SU(3)$ symmetry.
The transition form factors entering the decay rates are calculated using
the Vector Meson Dominance (VMD) model. In this way, we achieve generally
good agreement with the experimental branching ratios for radiative meson
decays. For the dilepton decay modes the branching ratios are, however, only
known in a few cases.
We consider the vector mesons $\rho $, $\omega $, and $\phi (1020)$ ($=V$),
the pseudo-scalar mesons $\pi $, $\eta $, $\eta ^{\prime }$ ($=P$), and the
scalar mesons $f_0(980)$ and $a_0(980)$ ($=S$). The various decay modes can
systematically be classified as follows: (i) There are, first of all, the
direct decays modes $V\rightarrow \ell ^{+}\ell ^{-}$, which contain the
information on the in-medium vector meson masses. In the next Sect., some
useful relations which simplify calculations of the decay rates to final
states with a dilepton pair are derived. In Sect.3 we make a few remarks on
the direct decay modes. There exits then a large number of processes which
mask the vector meson peaks and which should be treated as a background.
(ii)\ These are Dalitz decays of pseudoscalar mesons $P\rightarrow \gamma
\ell ^{+}\ell ^{-}$ and scalar mesons $S\rightarrow \gamma \ell ^{+}\ell
^{-} $. These decays are discussed in Sect.4. (iii) One has also Dalitz
decays with one meson in the final states $V\rightarrow P\ell ^{+}\ell ^{-}$
and $P\rightarrow V\ell ^{+}\ell ^{-}$, which we discuss in Sect. 5. The
radiative decays of these mesons are well studied both from the theoretical
and experimental points of view. The uncertainties in the estimates for the
dilepton decays are connected with the lack of experimental information on
the transition form factors. Constructing the VMD model transition form
factors, we take special care of the quark counting rules. (iv) Finally,
there exist decays to four-body final states $V\rightarrow PP\ell ^{+}\ell
^{-}$, $P\rightarrow PP\ell ^{+}\ell ^{-}$, and $S\rightarrow PP\ell
^{+}\ell ^{-}$. Sect. 6 is devoted to these decays. We calculate almost all
four-body dilepton modes of the unflavored mesons with masses below the $%
\phi (1020)$. The numerical results for radiative widths of the three-body
decays ($V\rightarrow PP\gamma $, {\it etc.}), which provide a useful test
for the model considered, for the dilepton widths ($V\rightarrow P\ell
^{+}\ell ^{-},$ $V\rightarrow PP\ell ^{+}\ell ^{-}$, {\it etc.}), and for
the dilepton spectra from the unflavored meson decays are presented in Sect.
7.
\section{Relation between the decays ${\sf M}\rightarrow {\sf M}^{\prime
}\gamma ^{*}$ and ${\sf M}\rightarrow {\sf M}^{\prime }\ell ^{+}\ell ^{-}$}
\setcounter{equation}{0}
As mentioned above, we consider decays ${\sf M}\rightarrow {\sf M}^{\prime
}\ell ^{+}\ell ^{-}$ where ${\sf M}$ is a meson, ${\sf M}^{\prime }$ is a
photon, a meson, or two mesons, and $\ell ^{+}\ell ^{-}$ is an
electron-positron or muon-antimuon pair. The results of this Sect. are
valid, however, for arbitrary states ${\sf M}^{\prime }$. The decay ${\sf M}%
\rightarrow {\sf M}^{\prime }\ell ^{+}\ell ^{-}$ proceeds through two steps:
${\sf M}\rightarrow {\sf M}^{\prime }\gamma ^{*}$ and $\gamma
^{*}\rightarrow \ell ^{+}\ell ^{-}$, where $\gamma ^{*}$ is a virtual photon
whose mass, $M$, is equal to the invariant mass of the dilepton pair.
The matrix element of the physical process ${\sf M}\rightarrow {\sf M}%
^{\prime }\gamma $ for a real photon $\gamma $ has the form
\begin{eqnarray}
{\cal M}={\cal M}_{\mu }\varepsilon _{\mu }^{*}(k) \label{II.1}
\end{eqnarray}
where $\varepsilon _{\mu }(k)$ is a photon polarization vector. The matrix
element ${\cal M}_{\mu }$ is defined also at $k^{2}=M^{2}\neq 0$ for virtual
photons $\gamma ^{*}$. As a consequence of the gauge invariance, it is
transverse with respect to the photon momentum
\begin{eqnarray}
{\cal M}_{\mu }k_{\mu }=0. \label{II.2}
\end{eqnarray}
The decay rate ${\sf M}\rightarrow {\sf M}^{\prime }\gamma ^{*}$ can
formally be calculated as
\begin{eqnarray}
d\Gamma ({\sf M}\rightarrow {\sf M}^{\prime }\gamma^{*})=\frac{1}{2\sqrt{s}}%
\sum_{f}\overline{{\cal M}_{\mu }{}{\cal M}_{\nu }{}^{*}}(-g_{\mu \nu })%
\frac{(2\pi )^{4}}{(2\pi )^{3n+3}}d\Phi _{n+1} \label{II.3}
\end{eqnarray}
where $\sqrt{s}\ $is mass of the decaying meson and $n$ is number of
particles in the state ${\sf M}^{\prime }$. The phase space in Eq.(\ref{II.3}%
) is defined in the usual way
\begin{eqnarray}
d\Phi _{k}(\sqrt{s},m_{1},...,m_{k})=\prod_{i=1}^{k}\frac{d{\bf p}_{i}}{%
2E_{i}}\delta ^{4}(P-\sum_{i=1}^{k}p_{i}). \label{II.4}
\end{eqnarray}
Here, $P$ is the four-momentum of the meson ${\sf M}$ , $P^{2}=s,$ and $%
p_{i} $ are momenta of the particles in the final state, including the
virtual photon $\gamma ^{*}$. In Eq.(\ref{II.3}), the summation over the
final states and averaging over the initial states of the decaying meson is
performed. The limit $M^{2}\rightarrow 0$ gives the decay rate of the
physical process ${\sf M}\rightarrow {\sf M}^{\prime }\gamma $.
The ${\sf M}\rightarrow {\sf M}^{\prime }\ell ^{+}\ell ^{-}$ decay rate is
given by
\begin{eqnarray}
d\Gamma ({\sf M}\rightarrow {\sf M}^{\prime }\ell ^{+}\ell ^{-})=\frac{1}{2%
\sqrt{s}}\sum_{f}\overline{{\cal M}_{\mu }{}{\cal M}_{\nu }{}^{*}}j_{\mu
}j_{\nu }{}^{*}\frac{1}{M^{4}}\frac{(2\pi )^{4}}{(2\pi )^{3n+6}}d\Phi _{n+2}
\label{II.5}
\end{eqnarray}
where $j_{\mu }\ $is the lepton current. The term $1/M^{4}$ comes from the
photon propagator, and $d\Phi _{n+2}$ is the phase space of $n$ particles in
the state ${\sf M}^{\prime }$ and of the $\ell ^{+}\ell ^{-}$ pair.
The value $\Gamma ({\sf M}\rightarrow {\sf M}^{\prime }\ell ^{+}\ell ^{-})$
can be related to the decay rates $\Gamma ({\sf M}\rightarrow {\sf M}%
^{\prime }\gamma ^{*})$ and $\Gamma (\gamma ^{*}\rightarrow \ell ^{+}\ell
^{-})$. In the analogy with massive vector particles, the width of a virtual
photon $\gamma ^{*}$ can formally be evaluated as
\begin{eqnarray}
M\Gamma (\gamma ^{*}\rightarrow \ell ^{+}\ell ^{-})=\frac{\alpha }{3}%
(M^{2}+2m_{\ell }^{2})\sqrt{1-\frac{4m_{\ell }^{2}}{M^{2}}} \label{II.6}
\end{eqnarray}
where $m_{\ell }$ is the lepton mass. The expression for the product of two
dilepton currents, summed up over the final states of the $\ell ^{+}\ell
^{-} $ pair, has the form
\begin{eqnarray}
\sum_{f}j_{\mu }j_{\nu }{}^{*}=\frac{16\pi \alpha }{3}(M^{2}+2m_{\ell
}^{2})(-g_{\mu \nu }+\frac{k_{\mu }k_{\nu }}{M^{2}}) \label{II.7}
\end{eqnarray}
where $\alpha $ is the fine-structure constant and $k$ is the total momentum
of the pair. Factorizing the $n$-body invariant phase space,
\begin{eqnarray}
d\Phi _{k}(\sqrt{s},m_{1},...,m_{k})=d\Phi _{k-1}(\sqrt{s}%
,m_{1},...,m_{k-2},M)dM^{2}\Phi _{2}(M,m_{k-1},m_{k}), \label{II.8}
\end{eqnarray}
which can be proved by inserting the unity decomposition
\begin{eqnarray}
1=\int d^{4}qdM^{2}\delta (q^{2}-M^{2})\delta ^{4}(q-p_{k-1}-p_{k})
\label{II.9}
\end{eqnarray}
into Eq.(\ref{II.4}), one obtains from Eqs.(\ref{II.3}), (\ref{II.5}), and (%
\ref{II.6}) with the help of Eqs.(\ref{II.7}) and (\ref{II.8}) the following
expression
\begin{eqnarray}
d\Gamma ({\sf M}\rightarrow {\sf M}^{\prime }\ell ^{+}\ell ^{-})=d\Gamma (%
{\sf M}\rightarrow {\sf M}^{\prime }\gamma ^{*})M\Gamma (\gamma
^{*}\rightarrow \ell ^{+}\ell ^{-})\frac{dM^{2}}{\pi M^{4}}. \label{II.10}
\end{eqnarray}
The factor $dM^{2}/(\pi M^{4})$ has the form of a properly normalized
Breit-Wigner distribution for a zero-mass resonance.
The two-body phase space in Eq.(\ref{II.8}) has the form
\begin{eqnarray}
\Phi _{2}(\sqrt{s},m_{1},m_{2})=\frac{\pi p^{*}(\sqrt{s},m_{1},m_{2})}{\sqrt{%
s}} \label{II.11}
\end{eqnarray}
where
\begin{eqnarray}
p^{*}(\sqrt{s},m_{1},m_{2})=\frac{\sqrt{%
(s-(m_{1}+m_{2})^{2})(s-(m_{1}-m_{2})^{2})}}{2\sqrt{s}} \label{II.12}
\end{eqnarray}
is momentum of the particles $1$ and $2$ in the c. m. frame.
In the following, we will work with the matrix elements of the processes $%
{\sf M}\rightarrow {\sf M}^{\prime }\gamma ^{*}$ and use Eq.(\ref{II.3}) to
derive results for the decay rates ${\sf M}\rightarrow {\sf M}^{\prime
}\gamma $ with real photons and Eq.(\ref{II.10}) to get results for decay
rates ${\sf M}\rightarrow {\sf M}^{\prime }\ell ^{+}\ell ^{-}$ with
dileptons in the final states. Now, we will consider the processes which are
interesting for the study of dilepton spectra in heavy-ion collisions.
\section{Decays of the $\rho $-$,$ $\omega $-$,$ and $\phi $-mesons to $\ell
^{+}\ell ^{-}$ pairs}
\setcounter{equation}{0}
The diagram for the $V\rightarrow \ell ^{+}\ell ^{-}$ decays with $V=\rho ,$
$\omega ,$ and $\phi $ is shown in Fig. \ref{fig1}. In terms of the vector
meson fields, $V_{\mu }$, the electromagnetic current has the form \cite{SAK}
\begin{eqnarray}
j_{\mu }=-e\sum_{V}\frac{m_{V}^{2}}{g_{V}}V_{\mu } \label{III.1}
\end{eqnarray}
where $m_{V}$ are the vector meson masses and $e=-|e|$ is the electron
charge. The $SU(3)$ predictions for the coupling constants, $g_{\rho
}:g_{\omega }:g_{\phi }=1:3:\frac{-3}{\sqrt{2}},$ are in good agreement with
the ratios between the values $g_{\rho }=5.03,$ $g_{\omega }=17.1,$ and $%
g_{\phi }=-12.9$ extracted from the $e^{+}e^{-}$ decay widths of the $\rho $-%
$,$ $\omega $-$,$ and $\phi $-mesons with the use of the well known
expression
\begin{eqnarray}
\Gamma (V\rightarrow \ell ^{+}\ell ^{-})=\frac{8\pi \alpha ^{2}}{3g_{V}^{2}}%
(1+2\frac{m_{\ell }^{2}}{m_{V}^{2}})p^{*}(m_{V},m_{\ell },m_{\ell }),
\label{III.2}
\end{eqnarray}
with $p^{*}$ defined in Eq.(\ref{II.12}).
\section{Meson decays to photons and $\ell ^{+}\ell ^{-}$ pairs}
\setcounter{equation}{0}
In this Sect., we discuss meson decays $P\rightarrow \gamma \ell ^{+}\ell
^{-}$ (Fig. 2) and $S\rightarrow \gamma \ell ^{+}\ell ^{-}$ (Fig. 3) for $%
P=\pi ^{0},\eta ,$ and $\eta ^{\prime }$ and $S=f_{0}(980)$ and $%
a_{0}^{0}(980)$. As we shall see, these decays are the dominant $e^{+}e^{-}$
modes for $\pi ^{0}$-, $\eta $-mesons, for $\eta ^{\prime }$-meson at $M$ $%
\gtrsim 250$ MeV, $f_{0}$-meson at $M$ $\gtrsim 500$ MeV and in some other
cases. The $\mu ^{+}\mu ^{-}$ modes are also discussed.
These decays are related to the experimentally measured two photon decays.
The uncertainties in the estimates originate only from the purely known
transition form factors in the time-like region. The $\eta \gamma \gamma
^{*} $ transition form factor is in reasonable agreement with the one-pole
VMD model predictions \cite{LGL}. The experimental errors in the $\eta
^{\prime } $ transition form factor are large \cite{Bayu}. The one-pole VMD
approximation for the $P\gamma \gamma ^{*}$ transition form factors is in
agreement with the quark counting rules which predict for these form factors
a $\sim 1/t$ asymptotics \cite{QCR}.
The nature of the scalar mesons has been a subject of intensive discussions
for a long time. The 4-quark content of the scalar mesons would imply a $%
\sim 1/t^{2}$ asymptotics for the $S\gamma \gamma ^{*}$ transition form
factors. In order to provide the correct asymptotics for the 4-quark meson
transition form factors, the VMD model should be extended to include
contributions from higher vector meson resonances. The $\omega \pi \gamma $
transition form factor has also $\sim 1/t^{2}$ asymptotic behavior \cite{VZ}%
. This form factor is measured in the time-like region \cite{LGL} and the
data show deviations from the naive one-pole approximation. The inclusion of
higher vector meson resonances improves the agreement and provides the
correct asymptotics.
The results of a recent measurement of the $\phi \rightarrow \gamma f_{0}$
branching ratio are interpreted as evidence for the dominance of a 4-quark
MIT bag component in the $f_{0}$-meson wave function \cite{MNA,VEPP}. We
thus calculate branching ratios $S\rightarrow \gamma \ell ^{+}\ell ^{-}$
assuming the 4-quark nature of the scalar mesons and imposing constraints
from the quark counting rules to the $S\gamma \gamma ^{*}$ transition form
factors.
\subsection{Decay modes $\pi ^{0}\rightarrow \gamma e^{+}e^{-},$ $\eta
\rightarrow \gamma \ell ^{+}\ell ^{-},$ and $\eta ^{\prime }\rightarrow
\gamma \ell ^{+}\ell ^{-}$}
The effective vertex for the $P\rightarrow \gamma \gamma $ decays has the
form
\begin{eqnarray}
\delta {\cal L}_{P\gamma \gamma }=f_{P\gamma \gamma }\epsilon _{\tau \sigma
\mu \nu }\partial _{\sigma }A_{\tau }\partial _{\nu }A_{\mu }P \label{IV.1}
\end{eqnarray}
where $P=\pi ^0,\eta ,$ and $\eta ^{\prime }$ and $A_\mu $ is the photon
field. The matrix element for the decay $P\rightarrow $ $\gamma \gamma ^{*}$
with a virtual photon $\gamma ^{*}$ has the form
\begin{eqnarray}
{\cal M}=-if_{P\gamma \gamma }F_{P\gamma \gamma }(M^{2})\epsilon _{\tau
\sigma \mu \nu }k_{\tau }\varepsilon _{\sigma }^{*}(k)k_{1\mu }\varepsilon
_{\nu }^{*}(k_{1}) \label{IV.2}
\end{eqnarray}
where $k$ is the virtual photon momentum ($k^2=M^2$), $k_1$ is the real
photon momentum ($k_1^2=0$), and $F_{P\gamma \gamma }(t)$ is transition form
factor $P\gamma \gamma ^{*}$. The comparison of the $P\rightarrow \gamma
\ell ^{+}\ell ^{-}$ decay width with the decay width of a physical process $%
P\rightarrow $ $\gamma \gamma $ allows to write \cite{LGL}
\begin{eqnarray}
\frac{d\Gamma (P\rightarrow \gamma \ell ^{+}\ell ^{-})}{\Gamma (P\rightarrow
\gamma \gamma )}=2\left( \frac{p^{*}(\sqrt{s},0,M)}{p^{*}(\sqrt{s},0,0)}%
\right) ^{3}\left| F_{P\gamma \gamma }(M^{2})\right| ^{2}M\Gamma (\gamma
^{*}\rightarrow \ell ^{+}\ell ^{-})\frac{dM^{2}}{\pi M^{4}} \label{IV.3}
\end{eqnarray}
where $\sqrt{s}=\mu _P$ is the pseudoscalar meson mass. The value $M\Gamma
(\gamma ^{*}\rightarrow \ell ^{+}\ell ^{-})$ is given by Eq.(\ref{II.6}).
The factor $2$ in Eq.(\ref{IV.3}) occurs due the identity of photons in the
decay $P\rightarrow \gamma \gamma $. The product of the qubic term and the
absolute square of the form factor gives the ratio between squares of the
matrix element (\ref{IV.2}) at $k^2=M^2$ and $k^2=0$, multiplied by the
ratio between the two-particle phase spaces. The quark counting rules \cite
{QCR} imply that the form factor $F_{P\gamma \gamma }(t)$ behaves as $\sim
1/t$ at $t\rightarrow \infty $. The experimental data are described
reasonably well by the monopole formula
\begin{eqnarray}
F_{P\gamma \gamma }(t)=\frac{\Lambda _{P}^{2}}{\Lambda _{P}^{2}-t}
\label{IV.4}
\end{eqnarray}
with $\Lambda _P=0.75\pm 0.03,$ $0.77\pm 0.04,$ and $0.81\pm 0.04$,
respectively, for the $\pi ^0$-$,$ $\eta $-$,$ and $\eta ^{\prime }$-mesons
\cite{HJB}, which reproduces the correct asymptotics. Such a monopole fit
can naturally be interpreted in terms of the vector meson dominance. The
values of the $\Lambda _P$'s are close to the $\rho $- and $\omega $-meson
masses. In a QCD interpolation formula, Ref. \cite{BL}, the pole masses are
related to the PCAC coupling constants of the pseudoscalar mesons, $\Lambda
_P=2\pi f_P$. This expression gives for the pole masses similar numbers.
In the $\eta ^{\prime }\rightarrow \gamma e^{+}e^{-}$ and $\eta ^{\prime
}\rightarrow \gamma \mu ^{+}\mu ^{-}$ decays, the $\rho ^{0}$ and $\omega $
poles occur in the physical region allowed for the spectrum of the dilepton
pairs. The $SU(3)$ symmetry with the $\eta -\eta ^{\prime }$ mixing angle $%
\theta =\arcsin (-\frac{1}{3})=-19.5^{\circ }$ which is quite close to the
experimental value $\theta ^{\exp }=-15.5\pm 1.3$ (see {\it e.g.} \cite{BES}%
) predicts $f_{\rho \gamma \eta ^{\prime }}=3f_{\omega \gamma \eta ^{\prime
}}=\frac{1}{\sqrt{3}}f_{\omega \gamma \pi }$. The experimental branching
ratios $B^{\exp }(\eta ^{\prime }\rightarrow \gamma \rho ^{0})=(30.2\pm
1.3)\%$ and $B^{\exp }(\eta ^{\prime }\rightarrow \gamma \omega )=(3.02\pm
0.30)\%$ indicate that the ratio between the $\rho ^{0}$- and $\omega $%
-meson couplings with the $\eta ^{\prime }$-meson equals $f_{\rho \gamma
\eta ^{\prime }}^{2}/f_{\omega \gamma \eta ^{\prime }}^{2}=10$, in good
agreement with the $SU(3)$ predictions. In the timelike resonance region,
the transition form factor (\ref{IV.4}) should be modified:
\begin{eqnarray}
F_{\eta ^{\prime }\gamma \gamma }(t)=c_{\rho }\frac{m_{\rho }^{2}}{m_{\rho
}^{2}-t-im_{\rho }\Gamma _{\rho }}+c_{\omega }\frac{m_{\omega }^{2}}{%
m_{\omega }^{2}-t-im_{\omega }\Gamma _{\omega }}+c_{X}\frac{m_{X}^{2}}{%
m_{X}^{2}-t} \label{IV.5}
\end{eqnarray}
(recall that $\Lambda _{\eta ^{\prime }}\approx m_{\rho }\approx m_{\omega }$%
), with the weight coefficients $c_{\rho } \sim f_{\rho \gamma \eta ^{\prime
}}/g_{\rho }=f_{\rho \rho \eta ^{\prime }}/g_{\rho }^{2}$ and $c_{\omega }
\sim f_{\omega \gamma \eta ^{\prime }}/g_{\omega }=f_{\omega \omega \eta
^{\prime }}/g_{\omega }^{2}.$ The $SU(3)$ symmetry predicts $f_{\rho \rho
\eta ^{\prime }}=f_{\omega \omega \eta ^{\prime }}=\frac{1}{2\sqrt{3}}%
f_{\rho \omega \pi }$. The normalization condition looks like $c_{\rho }+$ $%
c_{\omega }=1$. The slope is close to $1/\Lambda _{\eta ^{\prime }}^{2}$.
From the relation $g_{\rho }/g_{\omega }\approx 0\allowbreak .\,3$ one gets $%
c_{\rho }/c_{\omega }\approx 10.5$, the relative sign of the values $c_{\rho
}\ $and $c_{\omega }$ is fixed by $SU(3) $ symmetry. The third term in Eq.(%
\ref{IV.5}) is introduced for reasons explained below. We assume for the
moment $c_{X}=0.$ With $c_{\rho }=0.9$ and $c_{\omega }=0.1,$ we obtain $%
B(\eta ^{\prime }\rightarrow \gamma \mu ^{+}\mu ^{-})=0.90\times 10^{-4}$,
in good agreement with the measured value $B^{\exp }(\eta ^{\prime
}\rightarrow \gamma \mu ^{+}\mu ^{-})=(1.04\pm 0.26)\times 10^{-4}$. Data
for the $\eta ^{\prime }\rightarrow e^{+}e^{-}\gamma $ decay are not
available. The coupling constants we used in the calculations are summarized
in Table 1.
The product of the branching ratios $B^{\exp }(\eta ^{\prime }\rightarrow
\gamma \rho ^0)B^{\exp }(\rho ^0\rightarrow \mu ^{+}\mu ^{-})=1.4\times
10^{-5}\ $is almost one order of magnitude smaller than the value $B^{\exp
}(\eta ^{\prime }\rightarrow \gamma \mu ^{+}\mu ^{-})$. The magnitude of the
direct $\rho ^0$-meson contribution to the $\eta ^{\prime }\rightarrow
\gamma \mu ^{+}\mu ^{-}$ decay can be estimated from Eq.(\ref{IV.3}) with
the use of a narrow-width approximation for the transition form factor,
\begin{eqnarray}
\left| F_{\eta ^{\prime }\gamma \gamma }(t)\right| ^{2}\approx \left|
c_{\rho }\right| ^{2}\frac{\pi m_{\rho }^{3}}{\Gamma _{\rho }}\delta
(t-m_{\rho }^{2}), \label{IV.6}
\end{eqnarray}
to give $B(\eta ^{\prime }\rightarrow \gamma \rho ^0\rightarrow \gamma \mu
^{+}\mu ^{-})\approx 2\times 10^{-5}$. This value is already close to the
product $B^{\exp }(\eta ^{\prime }\rightarrow \gamma \rho ^0)B^{\exp }(\rho
^0\rightarrow \mu ^{+}\mu ^{-})$, but still higher. The relative
contributions of the $\rho ^0$- and $\omega $-mesons to the $\eta ^{\prime
}\rightarrow \gamma e^{+}e^{-}$ and $\eta ^{\prime }\rightarrow \gamma \mu
^{+}\mu ^{-}$ decay rates are inversely proportional to the vector meson
widths, as it follows from Eq.(\ref{IV.6}). Since the $\omega $-meson width
is only $8.5$ MeV, its contribution is strongly enhanced. The direct
contribution of the $\omega $-meson to the transition $\eta ^{\prime
}\rightarrow \gamma \mu ^{+}\mu ^{-}$ equals $B^{\exp }(\eta ^{\prime
}\rightarrow \gamma \omega )B(\omega \rightarrow \mu ^{+}\mu ^{-})\approx
\allowbreak 2\times 10^{-6}\ $where $B(\omega \rightarrow \mu ^{+}\mu
^{-})\approx B^{\exp }(\omega \rightarrow e^{+}e^{-})=(7.15\pm 0.19)\times
10^{-5}$ (the quoted experimental values are all from Ref. \cite{PDG}). The
use of the narrow-width approximation gives $B(\eta ^{\prime }\rightarrow
\gamma \omega \rightarrow \mu ^{+}\mu ^{-}\gamma )\approx \Gamma _\rho
/\Gamma _\omega \left| c_\omega /c_\rho \right| ^2B(\eta ^{\prime
}\rightarrow \gamma \rho ^0\rightarrow \gamma \mu ^{+}\mu ^{-})\approx
4\times 10^{-6}$. This value is also greater than the product $B^{\exp
}(\eta ^{\prime }\rightarrow \gamma \omega )B(\omega \rightarrow \mu ^{+}\mu
^{-})$. It can be interpreted as a noticeable contribution from the excited
vector mesons to the form factor $F_{\eta ^{\prime }\gamma \gamma }(t).$
In any case, the above estimates demonstrate that the direct $\rho ^{0}$-
and $\omega $-meson contributions to the $\eta ^{\prime }\rightarrow \gamma
\mu ^{+}\mu ^{-}$ decay are small. Thus, the value $B(\eta ^{\prime
}\rightarrow \gamma \mu ^{+}\mu ^{-})$ is determined mainly by the
background, i.e. by those values of $M^{2}$ which are not near to the $\rho
^{0}$- and $\omega $-meson poles. Since we are interested in the spectrum of
the dilepton pairs, it is important, however, to fix the relative weights of
the vector meson contributions. A $10-20\%$ decrease of the residues $%
c_{\rho }\ $and $c_{\omega }$ due to the admixture of an excited vector
meson $X$ yields for the measured decays $\eta ^{\prime }\rightarrow \gamma
\mu ^{+}\mu ^{-}$, $\eta ^{\prime }\rightarrow \gamma \rho ^{0},$ and $\eta
^{\prime }\rightarrow \gamma \omega $ a consistent interpretation. With $%
c_{\rho }=0.8,$ $c_{\omega }=0.08,$ and $c_{X}=0.12$ we get $B(\eta ^{\prime
}\rightarrow \gamma \rho ^{0}\rightarrow \gamma \mu ^{+}\mu ^{-})\approx
1.5\times 10^{-5}$ and $B(\eta ^{\prime }\rightarrow \gamma \omega
\rightarrow \gamma \mu ^{+}\mu ^{-})\approx 2\times 10^{-6}$, in good
agreement with the products of the branching ratios $B^{\exp }(\eta ^{\prime
}\rightarrow \gamma \rho ^{0})B^{\exp }(\rho ^{0}\rightarrow \mu ^{+}\mu
^{-})$ and $B^{\exp }(\eta ^{\prime }\rightarrow \gamma \omega )B(\omega
\rightarrow \mu ^{+}\mu ^{-}).$ The mass of the radially excited vector
meson $X$ is not well fixed (see discussion in Sect. 5). We set $m_{X}=1.2$
GeV. Since it is out of the physical region, the width $\Gamma _{X}$ is set
equal to zero.
\subsection{Decay modes $f_{0}(980)\rightarrow $ $\gamma \ell ^{+}\ell ^{-}$
and $a_{0}^{0}(980)\rightarrow $ $\gamma \ell ^{+}\ell ^{-}$}
The isoscalar $f_{0}(980)$-meson and the isotriplet $a_{0}(980)$-meson have
quantum numbers $I^{G}(J^{PC})=0^{+}(0^{++})$ and $1^{-}(0^{++}).$ The $%
f_{0} $- and the neutral $a_{0}$-mesons decay to two photons. The effective
vertex for the $S\rightarrow $ $\gamma \gamma $ decay has the form
\begin{eqnarray}
\delta {\cal L}_{S\gamma \gamma }=f_{S\gamma \gamma }F_{\tau \mu }F_{\tau
\mu }S \label{IV.7}
\end{eqnarray}
where $F_{\tau \mu }=\partial _{\mu }A_{\tau }-\partial _{\tau }A_{\mu }$.
The matrix element for the process $S\rightarrow $ $\gamma \gamma ^{*}$ is
given by
\begin{eqnarray}
{\cal M}=-if_{S\gamma \gamma }F_{S\gamma \gamma }(M^{2})(g_{\tau \sigma
}k_{1\lambda }-g_{\tau \lambda }k_{1\sigma })(g_{\mu \sigma }k_{\lambda
}-g_{\mu \lambda }k_{\sigma })\varepsilon _{\tau }^{*}(k_{1})\varepsilon
_{\mu }^{*}(k). \label{IV.8}
\end{eqnarray}
Here, $k_{1}$ and $k$ are real ($k_{1}^{2}=0$) and virtual ($k^{2}=M^{2}$)
photon momenta, $F_{S\gamma \gamma }(t)$ is transition form factor of the
decay $S\rightarrow \gamma \gamma ^{*}$. The square of the matrix element
summed up over the photon polarizations equals $8sp^{*2}(\sqrt{s},0,M)$,
with $\sqrt{s}=m_{S}$ being the scalar meson mass. The width of the $%
S\rightarrow \gamma e^{+}e^{-}$ decay can be written as follows
\begin{eqnarray}
\frac{d\Gamma (S\rightarrow \gamma \ell ^{+}\ell ^{-})}{\Gamma (S\rightarrow
\gamma \gamma )}=2\frac{p^{*3}(\sqrt{s},0,M)}{p^{*3}(\sqrt{s},0,0)}\left|
F_{S\gamma \gamma }(M^{2})\right| ^{2}M\Gamma (\gamma ^{*}\rightarrow \ell
^{+}\ell ^{-})\frac{dM^{2}}{\pi M^{4}}. \label{IV.9}
\end{eqnarray}
The value of $\Gamma (f_0 \rightarrow\gamma \gamma )$ is given by PDG \cite
{PDG}. In case of the $a_0$-meson, PDG gives the quantity $\Gamma (a_0
\rightarrow\gamma \gamma ) \Gamma (a_0 \rightarrow\pi^0 \eta )/\Gamma_{tot}
(a_0)$. Since the mode $a_0 \rightarrow\pi^0 \eta$ is the dominant one, the
two-photon width can also be estimated.
The transition form factor $F_{S\gamma \gamma }(t)$ depends on the nature of
the scalar meson $S$. The asymptotics of the form factor according to the
quark counting rules \cite{QCR} is $\sim 1/t$ for a $2$-quark model \cite
{NAT} and $\sim 1/t^{2}$ for a 4-quark MIT bag \cite{RLJ} and a $K\bar{K}$
molecular models of the $f_{0}$- and $a_{0}$-mesons \cite{CLO}. As pointed
out by N. Achasov and Ivanchenko \cite{AIva}, the decay $\phi \rightarrow
\pi ^{0}\pi ^{0}\gamma $ can provide an important information on the
structure of the $f_{0}$-meson. In the SND experiment at the VEPP-2M $%
e^{+}e^{-}$ collider the branching ratio $\phi \rightarrow \pi ^{0}\pi
^{0}\gamma $ was measured \cite{MNA,VEPP}. The decay goes mainly through the
$\phi \rightarrow \gamma f_{0}$ decay mode. The results are consistent with
the hypothesis of a 4-quark MIT bag nature of the $f_{0}$-meson. We thus
calculate dilepton branching ratios $S\rightarrow \gamma \ell ^{+}\ell ^{-}$
assuming the 4-quark nature of the scalar mesons.
In the framework of the VMD model, the $S\gamma \gamma ^{*}$ transition form
factor can be reproduced assuming contributions from ground-state and
excited vector mesons with masses $m_{i}$%
\begin{eqnarray}
F_{S\gamma \gamma }(t)=\sum_{i}\frac{c_{i}m_{i}^{2}}{m_{i}^{2}-t}.
\label{IV.10}
\end{eqnarray}
The normalization $F_{S\gamma \gamma }(0)=1$ and the asymptotic condition $%
F_{S\gamma \gamma }(t)\sim 1/t^{2}$ at $t\rightarrow \infty $ give
constraints to the residues $c_{i}$:
\begin{eqnarray}
1 &=&\sum_{i}c_{i}, \label{IV.11} \\
0 &=&\sum_{i}c_{i}m_{i}^{2}. \label{IV.12}
\end{eqnarray}
In case of the form factor $f_{0}\gamma \gamma ^{*}$, the OZI rule implies
that $\omega \omega $, $\rho \rho $ and $\phi \phi $ contributions are
small, while the contributions $\omega \rho $, $\phi \rho $ are forbidden by
isospin conservation. The relative weights of the $\omega $- and $\phi $
-mesons in the form factor are fixed and thus at least three vector mesons
should be considered to fit the asymptotic behavior:
\begin{eqnarray}
F_{f_{0}\gamma \gamma }(t)\sim \frac{m_{\omega }^{2}}{m_{\omega }^{2}-t}+%
\frac{m_{\phi }^{2}}{m_{\phi }^{2}-t}+c_{X}\frac{m_{X}^{2}}{m_{X}^{2}-t}.
\label{IV.13}
\end{eqnarray}
The overall normalization factor can be derived from the condition $%
F_{f_{0}\gamma \gamma }(0)=1$, the value $c_{X}$ is fixed from the
requirement $F_{f_{0}\gamma \gamma }(t)\sim 1/t^{2}$ as $t\rightarrow \infty
$. The third meson mass $m_{X}$ is not well determined. The resulting form
factor is given by
\begin{eqnarray}
F_{f_{0}\gamma \gamma }(t)=\frac{m_{\omega }^{2}m_{\phi }^{2}m_{X}^{2}(1+Ct)%
}{(m_{\omega }^{2}-t)(m_{\phi }^{2}-t)(m_{X}^{2}-t)} \label{IV.14}
\end{eqnarray}
where
\begin{eqnarray}
C=-\frac{m_{\omega }^{2}(m_{X}^{2}-m_{\omega }^{2})+m_{\phi
}^{2}(m_{X}^{2}-m_{\phi }^{2})}{m_{\omega }^{2}m_{\phi
}^{2}(2m_{X}^{2}-m_{\omega }^{2}-m_{\phi }^{2})}. \label{IV.15}
\end{eqnarray}
The transition form factor $a_{0}^{0}\gamma \gamma ^{*}$ has the same
structure with the replacement $\omega \leftrightarrow \rho .$
The virtual photon from the scalar meson decay lies in the physical region
of the $\omega $- and $\rho $-mesons, so the $\omega $- and $\rho $-meson
propagators should be modified by introducing the finite meson widths. We
use $m_{X}=1.2$ GeV. The pole $t=m_{X}^{2}$ is outside of the physical
region, so we set $\Gamma _{X}=0$.
\section{Meson decays to one meson and a $\ell ^{+}\ell ^{-}$ pair}
\setcounter{equation}{0}
The radiative decays $V\rightarrow P\gamma $ and $P\rightarrow V\gamma $ are
well studied experimentally. The dilepton modes are measured only for decays
$\omega \rightarrow \pi ^{0}\ell ^{+}\ell ^{-}$ and $\phi \rightarrow \eta
e^{+}e^{-}$. We use data on the radiative decays to fix coupling constants
entering the $VP\gamma $ vertexes. These values are in good agreement with
the $SU(3)$ symmetry relations. We use the framework of the extended VMD
model and impose constraints to the residues of the transition form factors
from the quark counting rules. The calculation of the dilepton spectra is
then a straightforward task.
\subsection{Decay modes $\omega \rightarrow \pi ^{0}e^{+}e^{-}$, $\rho
\rightarrow \pi e^{+}e^{-},$ and $\phi \rightarrow \pi ^{0}e^{+}e^{-}$}
The effective vertex for the $V\rightarrow P\gamma $ radiative decay with $%
V=\omega $, $\rho ^{0}$, and $\phi $ and $P=\pi ^{0}$, $\eta $, and $\eta
^{\prime }$ has the form
\begin{eqnarray}
{\cal \delta L}_{V\gamma P}=-ef_{V\gamma P}\epsilon _{\tau \sigma \mu \nu
}\partial _{\sigma }V_{\tau }\partial _{\nu }A_{\mu }P. \label{V.1}
\end{eqnarray}
The matrix element for the $V\rightarrow P\gamma ^{*}$ transition can be
written as
\begin{eqnarray}
{\cal M}=-ief_{V\gamma P}F_{V\gamma P}(M^{2})\epsilon _{\tau \sigma \mu \nu
}\epsilon _{\tau }(P)P_{\sigma }\varepsilon _{\mu }^{*}(k)k_{\nu }.
\label{V.2}
\end{eqnarray}
The diagram for this transition is shown in Fig. \ref{fig4}. The vertex form
factor $F_{V\gamma P}(t)$ normalized at $t=0$ to unity depends on the square
of the photon four-momentum, $t=M^{2}$. The width of the decay $V\rightarrow
$ $P\gamma ^{*}$ has the form (see. {\it e.g.} \cite{PJain,Ulf})
\begin{eqnarray}
\Gamma (V\rightarrow P\gamma ^{*})=\frac{\alpha }{3}f_{V\gamma P}^{2}\left|
F_{V\gamma P}(k^{2})\right| ^{2}p^{*3}(\sqrt{s},\mu _{P},M) \label{V.3}
\end{eqnarray}
where $\sqrt{s}=m_{V}$ is the vector meson mass and $\mu _{P}$ is the
pseudoscalar meson mass. The limit $M^{2}=0$ describes the radiative decay $%
V\rightarrow $ $P\gamma $. Using the experimental value of the $\omega
\rightarrow \pi ^{0}\gamma $ width, one gets $f_{\omega \gamma \pi }=2.3\;$%
GeV$^{-1}$. The decay width $V\rightarrow P\ell ^{+}\ell ^{-}$ is connected
to the radiative width $V\rightarrow $ $P\gamma $ \cite{CHL}:
\begin{eqnarray}
\frac{d\Gamma (V\rightarrow P\ell ^{+}\ell ^{-})}{\Gamma (V\rightarrow
P\gamma )}=\left( \frac{p^{*}(\sqrt{s},\mu _{P},M)}{p^{*}(\sqrt{s},\mu
_{P},0)}\right) ^{3}\left| F_{V\gamma P}(M^{2})\right| ^{2}M\Gamma (\gamma
^{*}\rightarrow \ell ^{+}\ell ^{-})\frac{dM^{2}}{\pi M^{4}} \label{V.4}
\end{eqnarray}
where $M\Gamma (\gamma ^{*}\rightarrow \ell ^{+}\ell ^{-})$ is given by Eq.(%
\ref{II.6}). The first factor is the ratio between squares of the matrix
element (\ref{V.2}) at $k^{2}=M^{2}$ and $k^{2}=0$, multiplied by the ratio
of the two-particle phase spaces $\Phi _{2}(\sqrt{s},\mu _{P},M)$ and $\Phi
_{2}(\sqrt{s},\mu _{P},0)$.
The vertex form factor $F_{V\gamma P}(t)$ falls off asymptotically as $%
1/t^{2}$ \cite{VZ}. The additional power $1/t$ as compared to the
asymptotics of the pion form factor occurs because of suppression of the
quark spin-flip amplitude due to conservation of the quark helicity in
interactions of quarks with gluons.
The $\omega \gamma \pi $ transition form factor can be reproduced taking the
contribution of the ground-state $\rho $-meson and at least one excited $%
\rho $-meson into account. The extended VMD model with $m_{\rho }=769$ MeV
and $m_{\rho ^{\prime }}=1450$ MeV provides the correct asymptotics and
yields a better description of the experimental data \cite{ARGUS,DOL} than
the one-pole model with only the ground-state $\rho $-meson. However, it
still underestimates the slope of the form factor at $t=0.$ The experimental
value $F_{\omega \gamma \pi }^{\prime }(0)=2.4\pm 0.2$ GeV$^{-2}$ \cite{LGL}
prefers a value $m_{\rho ^{\prime }}\approx 1.2$ GeV.
The possible existence of vector mesons $I=0,$ $1$ with masses around $1.2$
GeV was under discussion for a long time. A quite strong evidence for
significant contributions to the spectral functions of the nucleon form
factors at values $t$ lower than $1.45$ GeV comes from the fact that the
experimental data for the Sachs form factors are reasonably described by the
dipole formula
\begin{eqnarray}
G_{Ep}(t)\approx G_{Mp}(t)/\mu _{p}\approx G_{Mn}(t)/\mu _{n}\approx \frac{1%
}{(1-t/0.71)^{2}} \label{V.5}
\end{eqnarray}
where $t$ is in units GeV$^{2}.$ This formula has a double pole. The
spectral functions are proportional to
\begin{eqnarray*}
\delta^{\prime}(t - 0.71) \approx \frac{ \delta(t - m^2_{\rho}) - \delta(t -
m^2_X)}{-m^2_{\rho} + m^2_X}.
\end{eqnarray*}
The double pole in Eq.(V.5) indicates an enhancement in the spectral
functions at $m_{X}$ close to $m_{\rho}$, whose nature is not yet clear as
long as the existence of the non-strange vector mesons at $1.2$ GeV is not
established. In any case, it is desirable to shift masses of the first
excited non-strange ''vector mesons'' to lower values. This has been done in
some popular fits of the nucleon form factors (see {\it e.g.} \cite{HPM}).
The more recent fits use, however, higher values of the vector meson masses
\cite{NFF}. In our calculations, the value $m_{X}$ is assumed to be $1.2$
GeV, as extracted form the slope of the $F_{\omega \gamma \pi }(t)$ at $t=0$.
The multiplicative representation of the transition form factor $\omega
\gamma \pi $, which in the zero-width limit is completely equivalent to the
additive representation, is given by
\begin{eqnarray}
F_{\omega \gamma \pi }(t)=\frac{m_{\rho }^{2}m_{X}^{2}}{(m_{\rho
}^{2}-t)(m_{X}^{2}-t)}. \label{V.6}
\end{eqnarray}
The experimental data on the transition form factors are available for the $%
\eta ^{\prime }\gamma \gamma ^{*}$, $\eta \gamma \gamma ^{*}$, and $\omega
\pi \gamma ^{*}$ transitions. The parametrization (IV.5) for the $\eta
^{\prime }$ form factor is in good agreement with the data from Ref. \cite
{LGL}, Fig. 26. For the $\eta $-meson form factor (IV.4), our curve
coincides with the VMD curve from Fig. 24, Ref. \cite{LGL}, since the
parameter $\Lambda $ = 0.77 GeV in Eq.(IV.4) is close to the $\rho $-meson
mass.
The one-pole VMD model is known to be in rather pure agreement with the data
on the $\omega \pi \gamma ^{*}$ transition form factor \cite{Dzh}. The
two-pole extended VMD model, Eq.(V.6), improves the agreement without
introducing new parameters. It fits well the experimental points at $t<0.2$
GeV$^{2}$, but underestimates the last three ones from Ref. \cite{Dzh} below
$0.4$ GeV$^{2}$. By price of introducing a new vector meson, $\rho ^{\prime
\prime }$, and one fitting parameter, the last three points can be described
better. The modification consists in inserting an additional multiplier $%
m_{\rho ^{\prime \prime }}^{2}(1+Ct)/(m_{\rho ^{\prime \prime }}^{2}-t)$
into the right side of Eq.(V.6), with $C$ being a free parameter. This model
again satisfies the quark counting rules. However, it overestimates the form
factors values at $t\approx 0.2$ GeV$^{2}$. For the present calculations, we
stay in the framework of the simplest two-pole VMD model, Eq.(V.6). It is
clear, however, that if one needs to calculate the dilepton production for
the DLS and/or future HADES data, the sensitivity of the results on the $%
\omega \pi \gamma ^{*}$ form factor values must be investigated.
The $\phi $-meson decay width is given by Eq.(\ref{V.4}). In the $\phi $%
-meson decay, the emitted photon has isospin $I=1$, so the transition form
factor has the same form as for the $\omega $-meson. The $\ell ^{+}\ell ^{-}$
pairs appear in the physical region of the decay $\phi \rightarrow \pi
^{0}\rho ^{0}$. In order to include the $\rho $-meson direct contribution,
the form factor (\ref{V.6}) should be modified by introducing the finite $%
\rho $-meson width. In the cases of the $\omega \rightarrow \pi
^{0}e^{+}e^{-}$and $\rho \rightarrow \pi e^{+}e^{-}$ decays, vector mesons
appear in the unphysical regions and their widths are not so important.
The direct contribution to the $\phi \rightarrow \pi ^{0}\rho ^{0}$ decay
can be estimated with the use of the narrow-width approximation for the
transition form factor in analogy with the decays $\eta ^{\prime
}\rightarrow \gamma \rho ^{0}\rightarrow \gamma \mu ^{+}\mu ^{-}$ and $\eta
^{\prime }\rightarrow \gamma \omega \rightarrow \gamma \mu ^{+}\mu ^{-}$.
The form factor is proportional to square of the residue $c_{\rho }$ ({\it cf%
}. Eq.(\ref{IV.6})) which is fixed by Eq.(\ref{V.6}). The results of the
calculations $B(\phi \rightarrow \pi ^{0}\rho ^{0}\rightarrow \pi
^{0}e^{+}e^{-})=2.3\times 10^{-6}$ are in reasonable agreement with the
product $B^{\exp }(\phi \rightarrow \pi ^{0}\rho ^{0})B^{\exp }(\rho
^{0}\rightarrow e^{+}e^{-})=(1.9\pm 0.2)\times 10^{-6}$, and similarly for
the $\mu ^{+}\mu ^{-}$ decay mode.
In case of the $\rho ^{\pm }$- and $\rho ^{0}$-meson decays, the effective
vertex has the form
\begin{eqnarray}
{\cal \delta L}_{\rho \gamma \pi } &=&-ef_{\rho \gamma \pi }\epsilon _{\tau
\sigma \mu \nu }\partial _{\sigma }\rho _{\tau }^{\alpha }\partial _{\nu
}A_{\mu }\pi ^{\alpha } \nonumber \\
&=&-ef_{\rho \gamma \pi }\epsilon _{\tau \sigma \mu \nu }(\partial _{\sigma
}\rho _{\tau }^{0}\partial _{\nu }A_{\mu }\pi ^{0}+\partial _{\sigma }\rho
_{\tau }^{+}\partial _{\nu }A_{\mu }\pi ^{-}+\partial _{\sigma }\rho _{\tau
}^{-}\partial _{\nu }A_{\mu }\pi ^{+}). \label{V.7}
\end{eqnarray}
The vertex (\ref{V.1}) for $V=\rho ^{0}$ and $P=\pi ^{0}$ is a part of the
vertex (\ref{V.7}). Eq.(\ref{V.4}) remains valid for $V=\rho ^{\pm }$, $\rho
^{0}$ and $P=\pi ^{\pm }$, $\pi ^{0}$. The photon emitted in the decay $\rho
\rightarrow \pi \gamma ^{*}$ is in the isoscalar state, respectively, the
transition form factor $F_{\rho \gamma \pi }(t)$ receives contributions from
the $\omega $- and $\phi $-mesons. The effect of the $\phi $-meson is small
due to the OZI rule. The small difference between masses of the $\rho $- and
$\omega $-mesons is beyond the accuracy of the VMD model. Away from the $%
\rho $- and $\omega $-mesons poles $F_{\omega \gamma \pi }(t)\approx F_{\rho
\gamma \pi }(t)$, whereas $F_{\omega \gamma \pi }(t)=F_{\phi \gamma \pi }(t)$
everywhere.%
\subsection{Decay modes $\omega \rightarrow \eta \ell ^{+}\ell ^{-}$, $\rho
^{0}\rightarrow \eta \ell ^{+}\ell ^{-},$ and $\phi \rightarrow \eta \ell
^{+}\ell ^{-}$}
These decays are treated in the same way as the decays discussed above. The
effective vertex for the $V\rightarrow \eta \gamma $ transitions with $%
V=\omega $, $\rho ^{0}$, and $\phi $ has the form of Eq.(\ref{V.1}) where
one should put $P=\eta $. The decay widths are given by Eq.(\ref{V.4}) with $%
\mu _{P}$ being the $\eta $-meson mass. In the simplest version of the VMD
model, the transition form factor $F_{V\gamma \eta }(t)$ has the form of Eq.(%
\ref{V.6}). For the $\omega $- and $\phi $-meson decays, the form factor is
determined by the $\rho $-meson mass, while for the $\rho ^{0}$-meson decay
it is determined by the $\omega $-meson mass.
\subsection{Decay modes $\eta ^{\prime }\rightarrow \omega e^{+}e^{-}$ and $%
\eta ^{\prime }\rightarrow \rho ^{0}e^{+}e^{-}$}
These decays are of the same nature. The effective vertex for the $\eta
^{\prime }\rightarrow V\gamma ^{*}$ transitions with $V=\omega $ and $\rho
^{0}$ has the form of Eq. (\ref{V.1}). The spectrum of the $\ell ^{+}\ell
^{-}$ pair in $P\rightarrow V\ell ^{+}\ell ^{-}$ can be obtained from Eq. (%
\ref{V.4}) with the replacements $P\leftrightarrow V$ and $\mu
_{P}\leftrightarrow m_{V}.$ In our case, $V=\omega $ and $\rho ^{0}$, $%
P=\eta ^{\prime }$, the value $\sqrt{s}=\mu _{P}$ stands for the $\eta
^{\prime }$-mesons mass. The transition form factor $F_{\eta ^{\prime
}\gamma V}(t)\ $has the asymptotics $1/t^{2}$ for $t\rightarrow \infty $. It
is described within the VMD model by the vector meson contributions from
isovector and isoscalar channels, respectively.
\subsection{Decay mode $a_{1}(1260)\rightarrow $ $\pi \ell ^{+}\ell ^{-}$}
The axial vector meson $a_1(1260)$ has quantum numbers $%
I^G(J^{PC})=1^{-}(1^{++})$. Its mass is above the $\phi (1020)$-meson mass,
so we do not expect a noticeable effect from the production and the decay of
such a meson at GSI energies. This resonance becomes, however, important at
higher energies (see \cite{BC}, \cite{LiGale}).
The $a_{1}\rightarrow \pi \gamma ^{*}$ transition is an electric dipole $E1$
transition. The effective vertex has the form
\begin{eqnarray}
\delta {\cal L}_{a_{1}\gamma \pi }=f_{a_{1}\gamma \pi }(\partial _{\nu
}a_{1\mu }^{\alpha }-\partial _{\mu }a_{1\nu }^{\alpha })F_{\mu \nu }\pi
^{\alpha } \label{V.8}
\end{eqnarray}
where $F_{\mu \nu }$ is the electromagnetic tensor. The matrix element for
the process $a_{1}\rightarrow $ $\pi \gamma ^{*}$ can be written as follows
\begin{eqnarray}
{\cal M}=-ief_{a_{1}\gamma \pi }F_{a_{1}\gamma \pi }(M^{2})\epsilon _{\tau
}(P)(g_{\tau \sigma }P_{\lambda }-g_{\tau \lambda }P_{\sigma })(g_{\mu
\sigma }k_{\lambda }-g_{\mu \lambda }k_{\sigma })\varepsilon _{\mu }^{*}(k)
\label{V.9}
\end{eqnarray}
where $P$ is the momentum of the decaying meson, $k$ is the photon momentum,
$F_{a_{1}\gamma \pi }(t)$ is the transition form factor, $F_{a_{1}\gamma \pi
}(0)=1$. The square of the matrix element summed up over photon
polarizations and averaged over polarizations of the $a_{1}$-meson is
proportional to $p^{*2}(\sqrt{s},\mu ,M)+\frac{3}{2}M^{2}$ where $\sqrt{s}%
=m_{a_{1}}$. The decay width $a_{1}\rightarrow \pi \ell ^{+}\ell ^{-}$ takes
the form
\begin{eqnarray}
\frac{d\Gamma (a_{1}\rightarrow \pi \ell ^{+}\ell ^{-})}{\Gamma
(a_{1}\rightarrow \pi \gamma )}=\frac{p^{*2}(\sqrt{s},\mu ,M)+\frac{3}{2}%
M^{2}}{p^{*2}(\sqrt{s},\mu ,0)}\frac{p^{*}(\sqrt{s},\mu ,M)}{p^{*}(\sqrt{s}%
,\mu ,0)} \nonumber \\
\times \left| F_{a_{1}\gamma \pi }(M^{2})\right| ^{2}M\Gamma (\gamma
^{*}\rightarrow \ell ^{+}\ell ^{-})\frac{dM^{2}}{\pi M^{4}}. \label{V.10}
\end{eqnarray}
This expression differs from the analogous expression from Ref. \cite{BC},
Eq. (6.18), by the extra term $\frac{3}{2}M^{2}$ in the ratio for squares of
the matrix elements. Such a term is needed to ensure the correct threshold
behavior of the $s$-wave $a_{1}\rightarrow \pi \gamma ^{*}$ decay.
\section{Meson decays to two mesons and $\ell ^{+}\ell ^{-}$ pair}
\setcounter{equation}{0}
The four-body meson decays proceed either through a two-step mechanism
similar to that discussed a long time ago by Gell-Mann, Sharp and Wagner
\cite{GMSW} in connection to the $\omega \rightarrow \pi ^{+}\pi ^{-}\pi
^{0} $ decay or through a bremsstrahlung mechanism when a virtual photon is
emitted form an external meson line. In the first case, the matrix element
is a smooth function of the photon mass. At small $M\;$such that $%
2m_{e}\lesssim M$, the differential widths behave like $1/M$, according to
Eq.(\ref{II.10}). In the second case, the matrix element is singular at
small $M$, as a result of which the differential width increases faster than
$1/M$.
In this Sect., we derive analytical expressions for the double differential
widths $d^{2}\Gamma /ds_{12}dM$ of the $\eta \rightarrow $ $\pi ^{+}\pi
^{-}\ell ^{+}\ell ^{-},$ $\eta ^{\prime }\rightarrow $ $\pi ^{+}\pi ^{-}\ell
^{+}\ell ^{-},\rho ^{0}\rightarrow $ $\pi ^{+}\pi ^{-}\ell ^{+}\ell ^{-},$ $%
\rho ^{\pm }\rightarrow $ $\pi ^{\pm }\pi ^{0}\ell ^{+}\ell ^{-},$ $%
f_{0}\rightarrow $ $\pi ^{+}\pi ^{-}\ell ^{+}\ell ^{-},$ and $a_{0}^{\pm
}\rightarrow $ $\pi ^{\pm }\eta \ell ^{+}\ell ^{-}\;$decays, with $s_{12}$
being the invariant mass of two outgoing mesons. The widths of the other
decays are calculated numerically.
The experimental data are available for the $\eta \rightarrow $ $\pi ^{+}\pi
^{-}e^{+}e^{-}$ decay only, with large errors \cite{RAG66}. Nevertheless,
the dilepton widths can be predicted using the conventional assumptions on a
two-step decay mechanism and/or a bremsstrahlung decay mechanism.
\subsection{Decay modes $\eta \rightarrow $ $\pi ^{+}\pi ^{-}\ell ^{+}\ell
^{-}$ and $\eta ^{\prime }\rightarrow $ $\pi ^{+}\pi ^{-}\ell ^{+}\ell ^{-}$}
The diagram contributing to the decay $\eta \rightarrow $ $\pi ^{+}\pi
^{-}\gamma ^{*}$ is shown in Fig. \ref{fig5}. The $\rho ^{0}\gamma \eta $
vertex is given by Eq.(\ref{V.1}). The vertex for the $\rho \rightarrow \pi
\pi $ decay has the form
\begin{eqnarray}
{\cal \delta L}_{\rho \pi \pi } &=& \frac{1}{2}f_{\rho \pi \pi }\epsilon
_{\alpha \beta \gamma }\rho _{\mu }^{\alpha }(\pi ^{\beta }\stackrel{%
\leftrightarrow }{\partial }_{\mu }\pi ^{\gamma }) \nonumber \\
&=& f_{\rho \pi \pi }(\rho _{\mu }^{0}\pi ^{+}i\stackrel{\leftrightarrow }{%
\partial }_{\mu }\pi ^{-}+\rho _{\mu }^{+}\pi ^{-}i\stackrel{\leftrightarrow
}{\partial }_{\mu }\pi ^{0}+\rho _{\mu }^{-}\pi ^{0}i\stackrel{%
\leftrightarrow }{\partial }_{\mu }\pi ^{+}) \label{VI.1}
\end{eqnarray}
where $\alpha ,$ $\beta ,$ and $\gamma $ are isotopic indices of the $\rho $%
- and $\pi $-mesons. The coupling constant $f_{\rho \pi \pi }=6.0$ is
determined from the $\rho $-meson width
\begin{eqnarray}
\Gamma (\rho ^{0}\rightarrow \pi ^{+}\pi ^{-})=\frac{1}{6\pi s}f_{\rho \pi
\pi }^{2}p^{*3}(\sqrt{s},\mu ,\mu ) \label{VI.2}
\end{eqnarray}
where $\sqrt{s}=m_{\rho }$ is the $\rho $-meson mass. The value $f_{\rho \pi
\pi }$ is in good agreement with the relation $f_{\rho \pi \pi }/g_{\rho }=1$
which follows from the $\rho ^{0}$-meson dominance in the pion form factor.
In the FFGS model \cite{FF} the $t$-dependence of the pion form factor $%
F_{\pi }(t)$ is attributed to the $\rho $-meson propagator. It means that
the constant $f_{\rho \pi \pi }$ does not depend on the value $t$ when the
two pions are on the mass shell.
The radiative decay $\eta \rightarrow $ $\pi ^{+}\pi ^{-}\gamma $ has been
measured in Ref. \cite{JGL}. The $\eta ^{\prime }\rightarrow $ $\pi ^{+}\pi
^{-}\gamma $ decay is studied experimentally much better (see \cite{HA87}-%
\cite{SIB} and references therein). The last decay is more complicated,
since two pions occur in the physical region of the $\rho $-meson. It was
shown that an attempt to fit the experimental data within the framework of
the cascade model $\eta ^{\prime }\rightarrow $ $\rho ^0\gamma \rightarrow $
$\pi ^{+}\pi ^{-}\gamma $ results in overestimating the events with a $\pi
\pi $ invariant mass below the $\rho $-meson mass. In Ref. \cite{SIB}, a
nonresonant contribution is added to the amplitude and its parameters are
fixed by fitting the experimental data. One of the modifications considered
for the $\rho $-meson propagator is the following one
\begin{eqnarray}
(s_{12}-m_{\rho }^{2}+im_{\rho }\Gamma _{\rho })^{-1}\rightarrow
(s_{12}-m_{\rho }^{2}+im_{\rho }\Gamma _{\rho })^{-1}+C_{\rho } \label{VI.3}
\end{eqnarray}
with a complex value of $C_{\rho }$.
It was proposed that the existence of a non-resonant term is connected to
the chiral box anomaly \cite{CPIC}. The anomaly results to a modification of
the $\rho $-meson propagator similar to that of Eq.(\ref{VI.3}) with $C_\rho
=1/(3m_\rho ^2)$. The positive sign of the term added guarantees a reduction
of the number of events on the left hand side of the $\rho $-meson peak. The
contact term describes a production of two mesons in the relative $p$-wave
where a strong resonant pion-pion interaction due to existence of the $\rho $%
-meson is present. The final state interaction can be taken into account by
dividing the bare vertex by the Jost function determined by the $\pi $-meson
$p$-wave phase shift. In this way, the $\rho $-meson propagator reoccurs,
and so we get from the contact term the same old $\rho $-meson pole term of
the earlier models. The net effect of the box anomaly is a redefinition of
the coupling constant for those models which include only the single diagram
of Fig. \ref{fig5}.
We keep the constant term $C_\rho $ in the form suggested by Ref. \cite{CPIC}
and interpret it purely phenomenologically as a modification of the $\rho $%
-meson propagator needed to describe the distribution of the two-pion events
in the $\eta ^{\prime }\rightarrow $ $\pi ^{+}\pi ^{-}\gamma $ decay. The
contact term $C_\rho $ for the $\eta $-meson is assumed to be of the same
magnitude as for the $\eta ^{\prime }$-meson.
The matrix element of the $\eta \rightarrow $ $\pi ^{+}\pi ^{-}\gamma ^{*}$
decay has the form
\begin{eqnarray}
{\cal M}=-ief_{\rho \pi \pi }f_{\rho \gamma \eta }{\cal M}_{\mu }\varepsilon
_{\mu }^{*}(k) \label{VI.4}
\end{eqnarray}
where
\begin{eqnarray}
{\cal M}_{\mu }=\epsilon _{\tau \sigma \mu \nu }(p_{1}-p_{2})_{\tau
}(p_{1}+p_{2})_{\sigma }k_{\nu }(\frac{1}{s_{12}-m_{\rho }^{2}+im_{\rho
}\Gamma _{\rho }}+C_{\rho }). \label{VI.5}
\end{eqnarray}
Here, $p_{1}$ and $p_{2}$ are the pion momenta, $s_{12}=(p_{1}+p_{2})^{2}$, $%
k$ is the photon momentum, and $m_{\rho }$ and $\Gamma _{\rho }$ are the $%
\rho $-meson mass and width. Energy-momentum conservation implies $%
P=p_{1}+p_{2}+k$. After summation of the squared matrix element over the
photon polarizations and averaging over the direction of the two pion
momenta their in c. m., we get
\begin{eqnarray}
{\cal R} &\equiv &\int \frac{d\Omega _{12}}{4\pi }{\cal M}_{\mu }{\cal M}%
_{\nu }{}^{*}(-g_{\mu \nu }) \nonumber \\
&=&\frac{8}{3}sp^{*2}(\sqrt{s},\sqrt{s_{12}},M)p^{*2}(\sqrt{s_{12}},\mu ,\mu
)\left| \frac{1}{s_{12}-m_{\rho }^{2}+im_{\rho }\Gamma _{\rho }}+C_{\rho
}\right| ^{2} \label{VI.6}
\end{eqnarray}
with $d\Omega _{12}=2\pi \sin \theta d\theta $, and $\theta $ being the
angle between the momenta ${\bf p}_{1}$ and ${\bf k}$ in the c. m. frame of
two pions.
The decay width $\eta \rightarrow $ $\pi ^{+}\pi ^{-}\gamma ^{*}$ becomes
\begin{eqnarray}
\Gamma (\eta \rightarrow \pi ^{+}\pi ^{-}\gamma ^{*}) &=& \frac{\alpha }{%
16\pi ^{2}s}f_{\rho \pi \pi }^{2}f_{\rho \gamma \eta }^{2} \left| F_{\rho
\gamma \eta }(M^{2})\right| ^{2} \nonumber \\
&\times& \int_{4\mu ^{2}}^{(\sqrt{s}-M)^{2}}{\cal R\;}\frac{p^{*}(\sqrt{s},%
\sqrt{s_{12}},M)p^{*}(\sqrt{s_{12}},\mu ,\mu )}{\sqrt{s_{12}}}ds_{12}
\label{VI.7}
\end{eqnarray}
We neglect thereby for the $\rho $-meson off-shellness in the form factor $%
F_{\rho \gamma \eta }(t)$. The value $f_{\rho \gamma \eta }=1.86$ GeV$^{-1}$
($=\sqrt{\frac{2}{3}}f_{\omega \gamma \pi }$ due to $SU(3)$ symmetry)\ is
extracted from the $\rho ^{0}\rightarrow \eta \gamma $ width with the use of
Eq.(\ref{V.3}).
The dilepton spectrum in the decay $\eta \rightarrow \pi ^{+}\pi
^{-}e^{+}e^{-}$ is described by expression
\begin{eqnarray}
d\Gamma (\eta \rightarrow \pi ^{+}\pi ^{-}\ell ^{+}\ell ^{-})=\Gamma (\eta
\rightarrow \pi ^{+}\pi ^{-}\gamma ^{*})M\Gamma (\gamma ^{*}\rightarrow \ell
^{+}\ell ^{-})\frac{dM^{2}}{\pi M^{4}}. \label{VI.8}
\end{eqnarray}
The model for the decay $\eta ^{\prime }\rightarrow $ $\pi ^{+}\pi
^{-}\gamma ^{*}$ is a direct extension of the model for the $\eta $-meson
decay. The coupling constant $f_{\rho \gamma \eta ^{\prime }}=1.3$ GeV$^{-1}$
($=\frac{1}{\sqrt{3}}f_{\omega \gamma \pi }$ due to $SU(3)$ symmetry)\ can
be extracted from the radiative $\eta ^{\prime }$-meson decay width. The
two-step decay $\eta (\eta ^{\prime })\rightarrow \rho ^{\pm }\pi ^{\mp
}\rightarrow \pi ^{+}\pi ^{-}\gamma ^{*}$ is forbidden by $C$-parity
conservation. Constructing the dilepton spectra, one should not take into
account {\it e.g.} channels $\eta ^{\prime }\rightarrow $ $\pi ^{+}\pi
^{-}\gamma ^{*}$ and $\eta ^{\prime }\rightarrow $ $\rho ^{0}\gamma ^{*}$
simultaneously. The second one is already contained in the first one.
\subsection{Decay mode $\rho ^{0}\rightarrow $ $\pi ^{+}\pi ^{-}\ell
^{+}\ell ^{-}$}
The diagrams contributing to the decay $\rho ^0\rightarrow $ $\pi ^{+}\pi
^{-}\gamma ^{*}$ are depicted in Fig. \ref{fig6}. The first two diagrams
describe the photon bremsstrahlung. The third contact diagram is added to
restore the gauge invariance.
The vertex $\rho \pi \pi $ is given by Eq.(\ref{VI.1}). The matrix element
of the process $\rho ^{0}\rightarrow $ $\pi ^{+}\pi ^{-}\gamma ^{*}$ has the
form
\begin{eqnarray}
{\cal M}=ief_{\rho \pi \pi }F_{\pi }(M^{2})\epsilon _{\tau }(P){\cal M}%
_{\tau \mu }\varepsilon _{\mu }^{*}(k) \label{VI.9}
\end{eqnarray}
where
\begin{eqnarray}
{\cal M}_{\tau \mu }=(2Q_{1}-P)_{\tau }\frac{(2Q_{1}-k)_{\mu }}{%
Q_{1}^{2}-\mu ^{2}}+(2Q_{2}-P)_{\tau }\frac{(2Q_{2}-k)_{\mu }}{Q_{2}^{2}-\mu
^{2}}-2g_{\tau \mu }. \label{VI.10}
\end{eqnarray}
Here, $Q_{1}=p_{1}+k$, $Q_{2}=p_{2}+k$, $p_{1}$ and $p_{2}$ are the $\pi
^{+} $ and $\pi ^{-}$ momenta, $\mu $ is the pion mass, and $F_{\pi }(t)$ is
the pion electromagnetic form factor. The tensor part of the matrix element,
${\cal M}_{\tau \mu }$, is transverse with respect to the photon momentum, $%
k,$ and the $\rho $-meson momentum, $P$:
\begin{eqnarray}
{\cal M}_{\tau \mu }k_{\mu } &=&0, \label{VI.11} \\
{\cal M}_{\tau \mu }P_{\tau } &=&0. \label{VI.12}
\end{eqnarray}
The role of the last equation is the elimination of the coupling of the
spurious spin-zero component of the $\rho $-meson wave function from the
physical sector. Gauge invariance requires that the form factor entering the
third diagram be equal to the pion form factor.
The square of the matrix element summed up over the photon polarizations,
averaged over the initial $\rho $-meson polarizations and over directions of
the pion momenta in the c. m. frame of the two pions, has a compact form
\begin{eqnarray}
{\cal R} &\equiv& \int \frac{d\Omega _{12}}{4\pi }{\cal M}_{\tau \mu }{\cal M%
}_{\sigma\nu}^{*}\frac{1}{3}(-g_{\tau \sigma}+\frac{P_{\tau }P_{\sigma}}{s}%
)(-g_{\mu \nu}) \nonumber \\
&=&\frac{2}{3B_{\pi }^{2}} \left(4B_{\pi }^{2}+(M^{2}-4\mu
^{2})(s-4\mu^{2})F(\xi) \right. \nonumber \\
&+&\left.((s-4\mu ^{2})(M^{2}-4\mu ^{2}+2s_{12})+2s_{12}M^{2})L(\xi)\right)
\label{II.13}
\end{eqnarray}
where $s=P^2=m_\rho ^2$ and $s_{12}=(p_1+p_2)^2$. We have also
\begin{eqnarray}
F(\xi ) &=&\frac{1}{1-\xi ^{2}}, \label{VI.14} \\
L(\xi ) &=&\frac{1}{2\xi }\ln (\frac{1+\xi }{1-\xi }) \label{VI.15}
\end{eqnarray}
and
\begin{eqnarray}
\xi &=&\frac{2}{B_{\pi }}\sqrt{\frac{s}{s_{12}}}p^{*}(\sqrt{s_{12}},\mu ,\mu
)p^{*}(\sqrt{s},\sqrt{s_{12}},M), \label{VI.16} \\
B_{\pi } &=&\frac{1}{2}(s+M^{2}-s_{12}). \label{VI.17}
\end{eqnarray}
The decay width $\Gamma (\rho ^0\rightarrow \pi ^{+}\pi ^{-}\gamma ^{*})$ is
given by
\begin{eqnarray}
\Gamma (\rho ^{0}\rightarrow \pi ^{+}\pi ^{-}\gamma ^{*})&=&\frac{\alpha }{%
16\pi ^{2}s}f_{\rho \pi \pi }^{2}\left| F_{\pi }(M^{2})\right| ^{2}
\nonumber \\
&\times&\int_{4\mu ^{2}}^{(\sqrt{s}-M)^{2}}{\cal R\;}\frac{p^{*}(\sqrt{s},%
\sqrt{s_{12}},M)p^{*}(\sqrt{s_{12}},\mu ,\mu )}{\sqrt{s_{12}}}ds_{12}.
\label{VI.18}
\end{eqnarray}
The integral over the $s_{12}$ runs from $4\mu ^2\,$to $(\sqrt{s}-M)^2$. In
the case of real photon emission ($M^2=0$), we reproduce the corresponding
expression from Ref. \cite{PS}. The pion form factor can be taken as
\begin{eqnarray}
F_{\pi }(t)=\frac{m_{\rho }^{2}}{m_{\rho }^{2}-t-im_{\rho }\Gamma _{\rho }}.
\label{VI.19}
\end{eqnarray}
The dilepton spectrum can be obtained with the use of Eq.(\ref{II.10}) as
\begin{eqnarray}
d\Gamma (\rho ^{0}\rightarrow \pi ^{+}\pi ^{-}\ell ^{+}\ell ^{-})=\Gamma
(\rho ^{0}\rightarrow \pi ^{+}\pi ^{-}\gamma ^{*})M\Gamma (\gamma
^{*}\rightarrow \ell ^{+}\ell ^{-})\frac{dM^{2}}{\pi M^{4}}. \label{VI.20}
\end{eqnarray}
The square of the matrix element at small invariant masses $2m_e\lesssim M$
and small momenta of the virtual photon is proportional to $1/\omega ^{*2}$
where $\omega ^{*}=B_\pi /\sqrt{s}$ is the photon energy in the c. m. frame
of the decaying meson. The contribution of the region $s_{12}\lesssim (\sqrt{%
s}-M)^2$ to the integral over the $s_{12}$ is of the order $\log ^3(1/M),$
and so $dB/dM\sim \log ^3(1/M)/M$. This is a general feature of the
bremsstrahlung mechanism. The square of the matrix element in a two-step
decay contains no a large parameter. In such a case, $dB/dM\sim 1/M$.
\subsection{Decay modes $\rho ^{0}\rightarrow $ $\pi ^{0}\pi ^{0}\ell
^{+}\ell ^{-}$, $\omega \rightarrow $ $\pi ^{0}\pi ^{0}\ell ^{+}\ell ^{-}$,
and $\omega \rightarrow $ $\pi ^{+}\pi ^{-}\ell ^{+}\ell ^{-}$}
The diagram contributing to the decay $\rho ^{0}\rightarrow $ $\pi ^{0}\pi
^{0}\gamma ^{*}$ is shown in Fig. \ref{fig7}. The vertex $\rho \omega \pi $
has the form
\begin{eqnarray}
{\cal \delta L}_{\rho \omega \pi } &=&f_{\rho \omega \pi }\epsilon _{\tau
\sigma \mu \nu }\partial _{\sigma }\omega _{\tau }\partial _{\nu }\rho _{\mu
}^{\alpha }\pi ^{\alpha } \nonumber \\
&=&f_{\rho \omega \pi }\epsilon _{\tau \sigma \mu \nu }(\partial _{\sigma
}\omega _{\tau }\partial _{\nu }\rho _{\mu }^{0}\pi ^{0}+\partial _{\sigma
}\omega _{\tau }\partial _{\nu }\rho _{\mu }^{+}\pi ^{-}+\partial _{\sigma
}\omega _{\tau }\partial _{\nu }\rho _{\mu }^{-}\pi ^{+}). \label{VI.21}
\end{eqnarray}
The coupling constant $f_{\rho \omega \pi }\approx 16$ GeV$^{-1}$ was
determined by Gell-Mann, Sharp and Wagner \cite{GMSW} from the $\omega
\rightarrow \pi ^{+}\pi ^{-}\pi ^{0}$ decay assuming a two-step mechanism $%
\omega \rightarrow \rho \pi \rightarrow \pi ^{+}\pi ^{-}\pi ^{0}$.
In Refs. \cite{Ulf} a direct contact term $\omega \rightarrow \pi ^{+}\pi
^{-}\pi ^{0}$ originating from the chiral anomaly is taken into account. The
analysis of the $\omega \rightarrow \pi ^{+}\pi ^{-}\pi ^{0}$ and $\phi
\rightarrow \rho \pi $, $\pi ^{+}\pi ^{-}\pi ^{0}$ decays with inclusion of
the contact vertex gives a value $f_{\rho \omega \pi }\approx 12$ GeV$^{-1}$%
. In the contact vertex, one can always select a pion pair with quantum
numbers of the $\rho $-meson. The final-state interaction between these
pions, as usually, can be taken into account dividing the bare vertex by the
Jost function determined from the isovector $p$-wave $\pi \pi $-scattering
phase shift. In the two-pion channels with the $\rho $-meson quantum
numbers, the ordinary $\rho $-meson propagator therefore occurs, as a result
of which the contact vertex reduces to the ordinary pole diagrams of the
same old two-step mechanism.
The coupling constant $f_{\rho \omega \pi }$ can be extracted from the
decays $\rho \rightarrow \pi \gamma $ and $\omega \rightarrow \pi \gamma .$
The one-pole VMD approximation for the transition form factors gives $%
f_{\rho \omega \pi }=f_{\rho \gamma \pi }g_\omega \approx 12.6$ GeV$^{-1}$
and $f_{\rho \omega \pi }=f_{\omega \gamma \pi }g_\rho \approx 11.7$ GeV$%
^{-1}$, respectively. In the two-pole VMD model which provides the
transition form factors with the correct asymptotics these numbers should be
multiplied by a factor $m_X^2/(m_X^2-m_\rho ^2)$. For $m_X=1.2$ GeV, one
gets, respectively, $f_{\rho \omega \pi }\approx 21$ GeV$^{-1}$ and $f_{\rho
\omega \pi }\approx 20$ GeV$^{-1}$. The QCD sum rules \cite{Lubl} give $%
f_{\rho \omega \pi }\approx 16$ GeV$^{-1}$. In Ref. \cite{ABram}, the value $%
f_{\rho \omega \pi }\approx 14$ GeV$^{-1}$ is used.
We consider contributions to the strong part of the decay amplitudes from
the ground-state vector mesons only. The two-step mechanism of the $\omega
\rightarrow \pi ^{+}\pi ^{-}\pi ^{0}$ decay with a ground-state $\rho $%
-meson in the intermediate state is essentially identical to the mechanism
of the decays discussed in this subsection. In our calculations, we use $%
f_{\rho \omega \pi }=16$ GeV$^{-1}$.
The vertex $\omega \gamma \pi $ is defined by Eq.(\ref{V.1}). The matrix
element of the decay $\rho ^{0}\rightarrow $ $\pi ^{0}\pi ^{0}\gamma ^{*}$
has the form
\begin{eqnarray}
{\cal M}=-ief_{\rho \omega \pi }f_{\omega \gamma \pi }F_{\omega \gamma \pi
}(M^{2})\epsilon _{\tau }(P){\cal M}_{\tau \mu }\varepsilon _{\mu }^{*}(k)
\label{VI.22}
\end{eqnarray}
with
\begin{eqnarray}
{\cal M}_{\tau \mu }=\epsilon _{\tau \sigma \rho \lambda }P_{\sigma
}Q_{1\lambda }\epsilon _{\rho \kappa \mu \nu }Q_{1\kappa }k_{\nu }\frac{1}{%
Q_{1}^{2}-m_{\omega }^{2}}+\epsilon _{\tau \sigma \rho \lambda }P_{\sigma
}Q_{2\lambda }\epsilon _{\rho \kappa \mu \nu }Q_{2\kappa }k_{\nu }\frac{1}{%
Q_{2}^{2}-m_{\omega }^{2}} \label{VI.23}
\end{eqnarray}
where $Q_{1}=p_{1}+k$ and $Q_{2}=p_{2}+k$.
The square of the matrix element
\begin{eqnarray}
{\cal R}\equiv \int \frac{d\Omega _{12}}{4\pi }{\cal M}_{\tau \mu }{\cal M}%
_{\sigma\nu}{}^{*}\frac{1}{3}(-g_{\tau \sigma}+\frac{P_{\tau }P_{\sigma}}{s}%
)(-g_{\mu \nu})\text{,} \label{VI.24}
\end{eqnarray}
is averaged numerically over the directions of the photon momentum ${\bf k}$
in the c. m. frame of the pion $1$ and the photon.
The decay widths can be written as
\begin{eqnarray}
\Gamma (\rho ^{0}\rightarrow \pi ^{0}\pi ^{0}\gamma ^{*})&=&\frac{1}{32\pi
^{2}s}f_{\rho \omega \pi }^{2}f_{\omega \gamma \pi }^{2}\left| F_{\omega
\gamma \pi }(M^{2})\right| ^{2} \nonumber \\
&\times&\int_{(\mu +M)^{2}}^{(\sqrt{s}-\mu )^{2}}{\cal R\;}\frac{p^{*}(\sqrt{%
s},\sqrt{s_{12}},M)p^{*}(\sqrt{s_{12}},\mu ,\mu )}{\sqrt{s_{12}}}ds_{12}.
\label{VI.26}
\end{eqnarray}
The integral over the $s_{12}$ runs from $(\mu +M)^{2}$ to $(\sqrt{s}-\mu
)^{2}$. The additional factor $2$ in the denominator occurs due to identity
of two pions. The dilepton spectrum in the decay $\rho ^{0}\rightarrow \pi
^{0}\pi ^{0}\ell ^{+}\ell ^{-}$ can be constructed with the use of Eq.(\ref
{II.10}):
\begin{eqnarray}
d\Gamma (\rho ^{0}\rightarrow \pi ^{0}\pi ^{0}\ell ^{+}\ell ^{-})=\Gamma
(\rho ^{0}\rightarrow \pi ^{0}\pi ^{0}\gamma ^{*})M\Gamma (\gamma
^{*}\rightarrow \ell ^{+}\ell ^{-})\frac{dM^{2}}{\pi M^{4}}. \label{VI.25}
\end{eqnarray}
The $\omega \rightarrow $ $\pi ^{0}\pi ^{0}\gamma ^{*}$ decay is analogous
to the $\rho ^{0}\rightarrow $ $\pi ^{0}\pi ^{0}\gamma ^{*}$ decay. The
corresponding diagrams are shown in Fig. \ref{fig8}. The matrix element has
the form of Eqs.(\ref{VI.22}) and (\ref{VI.23}) with interchanged $V\gamma P$
coupling constants, $f_{\omega \gamma \pi }$ $\leftrightarrow $ $f_{\rho
\gamma \pi }$, and vector meson masses, $m_{\omega }$ $\leftrightarrow $ $%
m_{\rho }$.
The diagrams contributing to the $\omega \rightarrow $ $\pi ^{+}\pi
^{-}\gamma ^{*}$ decay are shown in Fig. \ref{fig9}. The first two diagrams
(a) and (b) correspond to the common two-step mechanism. The last three
diagrams occur due to the $\omega \rho $ mixing and correspond to the
bremsstrahlung. The diagram (e) restores the gauge invariance of the
amplitude. The bremsstrahlung dominates at small values of the $e^{-}e^{+}$
invariant mass. The vertices $\omega \rho \pi $ and $\rho \gamma \pi $
entering into the matrix element of the $\omega \rightarrow $ $\pi ^{+}\pi
^{-}\gamma ^{*}$ decay have the form of Eqs.(\ref{VI.21}) and (\ref{V.7}).
The first two diagrams and the last three diagrams from Fig. \ref{fig9} are
separately gauge invariant. These two classes of diagrams contrary to the
statement of Ref. \cite{FO} interfere: Two pions in the two-step part of the
amplitude are in an isosinglet state. Due to the VMD, the photon in the
diagrams (c) and (d) is emitted from the $\rho ^0$-meson. It is equivalent
to introducing the pion form factor into the $\pi \pi \gamma $\ vertex. The
outgoing pion and the $\rho $-meson with a common vertex form an isotriplet
which, in turn, forms a new isotriplet with the second outgoing pion. The
isospin wave function of the final state looks like $[\vec \pi \times [\vec %
\pi \times \vec \rho ]]=\vec \pi (\vec \pi \cdot \vec \rho )-\vec \rho (\vec %
\pi \cdot \vec \pi ).$\ The first term corresponds to two neutral pions and
can be dropped, while the second term describes two charged pions in an
isoscalar state. Two pions in an isoscalar state from the diagrams (a) and
(b) and a photon interfere with two pions in an isoscalar state from
diagrams (c) and (d) and a photon. The final states are identical.
The interference pattern depends on the production dynamics. The physical $%
\omega $-meson wave function has the form $|\omega >=|\omega _0>+\varepsilon
|\rho _0>$ where $|\omega _0>\ $and $|\rho _0>$ are pure isospin $I=0$ and $%
1 $ states and $\varepsilon $ is a complex $\omega \rho $ mixing parameter.
The total matrix element is given by ${\cal M}={\cal M}^{[p]}+\varepsilon
{\cal M}^{[b]}.$ The value ${\cal M}^{[p]}$ is the same as in Eq.(VI.22).
The value ${\cal M}^{[b]}$ is given by Eq.(VI.9). The $\omega \rightarrow $ $%
\pi ^{+}\pi ^{-}\gamma ^{*}$ decay width represents a sum of three terms
\begin{eqnarray}
\Gamma (\omega \rightarrow \pi ^{+}\pi ^{-}\gamma ^{*})=\Gamma ^{[p]}(\omega
\rightarrow \pi ^{+}\pi ^{-}\gamma ^{*})+\Gamma ^{[b]}(\omega \rightarrow
\pi ^{+}\pi ^{-}\gamma ^{*})+\Gamma ^{[pb]}(\omega \rightarrow \pi ^{+}\pi
^{-}\gamma ^{*}) \label{VI.27}
\end{eqnarray}
originating from the vector-meson pole diagrams $[p]$, the bremsstrahlung
diagrams $[b]$, and from their interference $[pb]$. It was pointed out in
Ref. \cite{PS} that $\Gamma ^{[p]}(\omega \rightarrow \pi ^{+}\pi ^{-}\gamma
^{*})=2\Gamma (\omega \rightarrow \pi ^0\pi ^0\gamma ^{*}).$ The
bremsstrahlung contribution to the $\omega \rightarrow $ $\pi ^{+}\pi
^{-}\gamma ^{*}$ decay equals $\Gamma ^{[b]}(\omega \rightarrow \pi ^{+}\pi
^{-}\gamma ^{*})=|\varepsilon |^2\Gamma ^{[b]}(\rho ^0\rightarrow \pi
^{+}\pi ^{-}\gamma ^{*})$, with $|\varepsilon |=(\Gamma (\omega \rightarrow
\pi ^{+}\pi ^{-})/\Gamma (\rho ^0\rightarrow \pi ^{+}\pi
^{-}))^{1/2}=3.4\times 10^{-2}$ being the $\rho ^0$-meson admixture, as
determined from the $\omega \rightarrow \pi ^{+}\pi ^{-}$ decay.
The interference term
\begin{eqnarray}
{\cal R}^{[pb]}\equiv 2\int \frac{d\Omega _{12}}{4\pi }{\cal M}_{\tau \mu
}^{[p]}{\cal M}_{\sigma \nu }{}^{[b]*}\frac{1}{3}(-g_{\tau \sigma }+\frac{%
P_{\tau }P_{\sigma }}{s})(-g_{\mu \nu }) \label{VI.28a}
\end{eqnarray}
can be found to be
\begin{eqnarray}
{\cal R}^{[pb]} &=&\frac{8}{3}sp^{*2}(\sqrt{s_{12}},\mu ,\mu )p^{*2}(\sqrt{s}%
,\sqrt{s_{12}},M) \nonumber \\
&&\times \frac{1}{B_{\pi }B_{\rho }(\xi +\zeta _{\rho })}[\xi L(\xi )+\zeta
_{\rho }L(\zeta _{\rho })+(1-L(\xi ))/\xi +(1-L(\zeta _{\rho }))/\zeta
_{\rho }] \label{INT}
\end{eqnarray}
where $B_\rho =B_\pi +\mu ^2-m_\rho ^2$ and $\zeta _\rho =B_\pi \xi /B_\rho
. $ The value $B_\pi $ is defined by Eq.(\ref{VI.17}). The interference
contribution to the width is given by
\begin{eqnarray}
\Gamma ^{[pb]}(\omega &\rightarrow &\pi ^{+}\pi ^{-}\gamma ^{*})=\frac{1}{%
16\pi ^{2}s}Re\{(ief_{\rho \pi \pi }F_{\pi }(M^{2}))^{*}(-ie\varepsilon
f_{\rho \omega \pi }f_{\omega \gamma \pi }F_{\omega \gamma \pi }(M^{2}))\}
\nonumber \\
&\times& \int_{(\mu +M)^{2}}^{(\sqrt{s}-\mu )^{2}}{\cal R}^{[pb]}{\cal \;}%
\frac{p^{*}(\sqrt{s},\sqrt{s_{12}},M)p^{*}(\sqrt{s_{12}},\mu ,\mu )}{\sqrt{%
s_{12}}}ds_{12}.
\end{eqnarray}
The couplings $f_{\rho \omega \pi }\ $and $f_{\rho \pi \pi }$ are determined
up a sign. However, the relations $f_{\omega \gamma \pi }\approx f_{\rho
\omega \pi }/g_\rho $ and $f_{\rho \pi \pi }/g_\rho \approx 1$ allow to fix
the sign of the interference term. The phase $\varphi $ of the mixing
parameter $\varepsilon $ is a sum of phases $\varphi _{prod}$ and $\varphi
_{decay}$ originating, respectively, from production and $\pi \pi $ decay of
the $\omega $-meson. The last one is close to $\pi /2$ (see {\it e.g.} Ref.
\cite{ABK}). We set the value $\varphi _{prod}$ equal to zero. Since the
product $F_\pi (M^2)^{*}F_{\omega \gamma \pi }(M^2)$ is approximately real
below $1$ GeV, the interference can then be neglected.
The dilepton spectra of the $\omega \rightarrow \pi ^{+}\pi ^{-}\ell
^{+}\ell ^{-}$ decays can be obtained as usually from Eq.(II.10).
\subsection{Decay modes $\rho \rightarrow $ $\eta \pi e^{+}e^{-}$and $\omega
\rightarrow $ $\eta \pi ^{0}e^{+}e^{-}$}
These decays also proceed through the two-step mechanism. The diagrams are
shown in Figs. \ref{fig10} and \ref{fig11}. The matrix element of the decay $%
\rho \rightarrow $ $\eta \pi \gamma ^{*}$ is given by
\begin{eqnarray}
{\cal M}=ef_{\rho \omega \pi }f_{\omega \gamma \eta }\epsilon _{\tau }(P)%
{\cal M}_{\tau \mu }\varepsilon _{\mu }^{*}(k) \label{VI.28}
\end{eqnarray}
with
\begin{eqnarray}
{\cal M}_{\tau \mu }=\epsilon _{\tau \sigma \rho \lambda }P_{\sigma
}Q_{1\lambda }\epsilon _{\rho \kappa \mu \nu }Q_{1\kappa }k_{\nu }\frac{1}{%
Q_{1}^{2}-m_{\omega }^{2}}+\epsilon _{\tau \sigma \rho \lambda }P_{\sigma
}Q_{2\lambda }\epsilon _{\rho \kappa \mu \nu }Q_{2\kappa }k_{\nu }\frac{1}{%
Q_{2}^{2}-m_{\rho }^{2}} \label{VI.29}
\end{eqnarray}
where $Q_{1}=p_{1}+k$, $Q_{2}=p_{2}+k$, $p_{1}$ is the pion momentum, and $%
p_{2}$ is the $\eta $-meson momentum. We take the $SU(3)$ symmetry relation $%
f_{\rho \omega \pi }f_{\omega \gamma \eta }=2f_{\rho \rho \eta }f_{\rho
\gamma \pi }$ into account.
The average of the squared matrix element,
\begin{eqnarray}
{\cal R}\equiv \int \frac{d\Omega _{12}}{4\pi }{\cal M}_{\tau \mu }{\cal M}%
_{\sigma \nu}{}^{*}\frac{1}{3}(-g_{\tau \sigma}+\frac{P_{\tau }P_{\sigma}}{s}%
)(-g_{\mu \nu})\text{,} \label{VI.30}
\end{eqnarray}
is calculated numerically. The integration over the directions of the
momenta of the two mesons is performed in the c. m. frame of $\pi $ and $%
\eta $. The decay widths can be written as
\begin{eqnarray}
\Gamma (\rho \rightarrow \eta \pi \gamma ^{*})&=&\frac{\alpha }{16\pi ^{2}s}%
f_{\rho \omega \pi }^{2}f_{\omega \gamma \eta }^{2}\left| F_{\omega \gamma
\eta }(M^{2})\right| ^{2} \nonumber \\
&\times&\int_{(\mu +\mu ^{\prime })^{2}}^{(\sqrt{s}-M)^{2}}{\cal R\;}\frac{%
p^{*}(\sqrt{s},\sqrt{s_{12}},M)\;p^{*}(\sqrt{s_{12}},\mu ,\mu ^{\prime })}{%
\sqrt{s_{12}}}ds_{12}. \label{VI.31}
\end{eqnarray}
where $\mu ^{\prime }$ is the $\eta $-meson mass. In the form factor $%
F_{\omega \gamma \eta }(t)$, the shift of the $\omega $-meson from the mass
shell is neglected. The dilepton spectrum in the decays $\rho \rightarrow
\eta \pi e^{+}e^{-}$ can be constructed with the use of Eq.(\ref{II.10}).
The $\omega \rightarrow $ $\eta \pi ^{0}\gamma ^{*}$ decay is depicted in
Fig. \ref{fig11}. The matrix element has the form of Eq.(\ref{VI.28}), its
tensor part once the $SU(3)$ symmetry relation $f_{\rho \omega \pi }f_{\rho
\gamma \eta }=2f_{\omega \omega \eta }f_{\omega \gamma \pi }$ is taken into
account has the form of Eq.(\ref{VI.29}), and the $\omega \rightarrow $ $%
\eta \pi ^{0}\gamma ^{*}$ decay width is given by Eq.(\ref{VI.30}), with the
apparent substitutions $f_{\rho \gamma \eta }\leftrightarrow f_{\omega
\gamma \eta }$ and $m_{\omega }\leftrightarrow m_{\rho }$. The dilepton
spectrum can then be constructed, as usually, using Eq.(\ref{II.10}).\
\subsection{Decay mode $\rho ^{+}\rightarrow $ $\pi ^{+}\pi ^{0}\ell
^{+}\ell ^{-}$}
This decay is depicted in Fig. \ref{fig12}. The first three diagrams
describe the dilepton bremsstrahlung. The last diagram corresponds to the
two-step mechanism of the dilepton emission. These two sets of diagrams,
each of them is the gauge invariant, interfere and we calculate the
interference without further approximations.
The photon emitted from the $\rho ^{+}$-meson line interacts with the
electromagnetic current
\begin{eqnarray}
j_{\mu }(p^{\prime },p)=\frac{1}{2m_{\rho }}\epsilon _{\sigma
}^{*}(p^{\prime })\left[ -(p^{\prime }+p)_{\mu }g_{\tau \sigma }+\lambda
(q_{\tau }g_{\sigma \mu }-q_{\sigma }g_{\tau \mu })\right] \epsilon _{\tau
}(p) \label{VI.32}
\end{eqnarray}
where $q=p^{\prime }-p$, $\epsilon _{\tau }(p)$ is the $\rho $-meson
polarization vector. The value of $\lambda $ is connected to the $\rho ^{+}$
magnetic moment. Let's fix first the value of $\lambda $.
The magnetic moment of a particle can be calculated from expression
\begin{eqnarray}
{\bf \mu }=\int d{\bf x}\frac{1}{2}[{\bf x}\times {\bf j}]. \label{VI.33}
\end{eqnarray}
The wave function of a vector particle in its rest frame where $p=(m,{\bf 0}%
) $ has the form $\epsilon _{\tau }(p)=(0,{\bf e})$. Substituting this
expression into Eq.(\ref{VI.32}) and Eq.(\ref{VI.32}) in turn into Eq.(\ref
{VI.33}), we obtain
\begin{eqnarray}
{\bf \mu }=\frac{\lambda }{2m_{\rho }}e_{\gamma }^{*}{\bf s}e_{\gamma }.
\label{VI.34}
\end{eqnarray}
where ${\bf s}$ is the spin operator acting on the space-like part of the $%
\rho $-meson wave functions according to the rule $s_{\alpha }e_{\beta
}=-i\epsilon _{\alpha \beta \gamma }e_{\gamma }$ ($\alpha ,\beta ,\gamma
=1,2,3$). In the non-relativistic quark model, the $\rho ^{+}$ magnetic
moment is the difference of the magnetic moments of the $u$- and $d$-quarks:
$\mu _{\rho ^{+}}=\mu _{u}-\mu _{d}$. The proton and neutron magnetic
moments are given by the well known expressions
\begin{eqnarray}
\mu _{p} &=&\frac{4\mu _{u}-\mu _{d}}{3}, \label{VI.35} \\
\mu _{n} &=&\frac{4\mu _{d}-\mu _{n}}{3}, \label{VI.36}
\end{eqnarray}
and thus one can write
\begin{eqnarray}
\frac{\lambda }{2m_{\rho }}=\mu _{u}-\mu _{d}=\frac{3}{5}(\mu _{p}-\mu _{n}).
\label{VI.37}
\end{eqnarray}
One gets $\lambda =\allowbreak 2.\,3$. The additive quark model predictions
are valid with an accuracy of $10-30\%$. The one-gluon and one-pion exchange
currents and the anomalous quark moments connected to the violation of the
OZI rule, are shown to be important in order to explain the deviations of
the naive $SU(3)$ predictions from the experimental data \cite{MIK,Bre}.
The diagram (a) in Fig. 12 contains the pion form factor $F_{\pi }(t)$. A
part of the second diagram (b) connected to the $\rho $-meson convection
current contains a $\rho $-meson form factor $F_{1\rho }(t)$ which decreases
as $\sim 1/t^{2}$ at $t\rightarrow \infty $. The second part of the same
diagram connected to the spin current contains a different the $\rho $-meson
form factor $F_{2\rho }(t)$ which decreases also like $\sim 1/t^{2}$ at $%
t\rightarrow \infty $ \cite{VZ}. It is not clear what kind of the propagator
should enter into the diagram (c). The gauge invariance requires, within the
framework of the considered model, that the two form factors $F_{1\rho }(t)$
and $F_{\pi }(t)$ should be equal. This is, however, a shortcoming of the
model. It can be overcome on the basis of a more complete theory. The second
part of the diagram (b) is gauge invariant. The form factor $F_{2\rho }(t)$
is therefore not subjected to additional constraints. The dilepton invariant
masses in the $\rho ^{+}\rightarrow $ $\pi ^{+}\pi ^{0}\ell ^{+}\ell ^{-}$
decay are not very large, so the distinction between these three form
factors does not exceed a $20\%$ effect. We set the form factor $F_{1\rho
}(t)$ and the form factor entering the diagram (c) both equal to the pion
form factor. The sum of the diagrams (a) - (c) is then gauge invariant. The
form factor $F_{2\rho }(t)$ coincides with the transition form factor $%
F_{\omega\gamma\pi }(t)$ in the framework of the extended VMD model.
The matrix element of the $\rho ^{+}\rightarrow $ $\pi ^{+}\pi ^{0}\gamma
^{*}$ decay is given then by
\begin{eqnarray}
{\cal M}=ief_{\rho \pi \pi }F_{\pi }(M^{2})\epsilon _{\tau }(P){\cal M}%
_{\tau \mu }\varepsilon _{\mu }^{*}(k) \label{VI.38}
\end{eqnarray}
Its tensor part consists of the $5$ pieces:
\begin{eqnarray}
{\cal M}_{\tau \mu }=\sum_{i=1}^{5}{\cal M}_{\tau \mu }^{i} \label{VI.39}
\end{eqnarray}
where
\begin{eqnarray}
{\cal M}_{\tau \mu }^{1} &=&\frac{(p_{1}-p_{2})_{\tau }(2P-k)_{\mu }}{%
(P-k)^{2}-m_{\rho }^{2}}, \label{VI.40} \\
{\cal M}_{\tau \mu }^{2} &=&\lambda \frac{-k\cdot (p_{1}-p_{2})g_{\tau \mu
}+k_{\tau }(p_{1}-p_{2})_{\mu }}{(P-k)^{2}-m_{\rho }^{2}}, \label{VI.41} \\
{\cal M}_{\tau \mu }^{3} &=&\frac{(p_{1}-p_{2}+k)_{\tau }(2p_{1}+k)_{\mu }}{%
(p_{1}+k)^{2}-\mu ^{2}}, \label{VI.42} \\
{\cal M}_{\tau \mu }^{4} &=&-g_{\tau \mu }, \label{VI.43} \\
{\cal M}_{\tau \mu }^{5} &=&\gamma \frac{\epsilon _{\tau \sigma \rho \lambda
}P_{\sigma }(p_{2}+k)_{\lambda }\epsilon _{\rho \kappa \mu \nu
}(p_{2}+k)_{\kappa }k_{\nu }}{(p_{2}+k)^{2}-m_{\omega }^{2}}, \label{VI.44}
\end{eqnarray}
with
\begin{eqnarray}
\gamma =\frac{f_{\rho \omega \pi }f_{\omega \gamma \pi }}{f_{\rho \pi \pi }}%
\frac{F_{\omega \gamma \pi }(M^{2})}{F_{\pi }(M^{2})}\approx \frac{f_{\rho
\omega \pi }^{2}}{f_{\rho \pi \pi }^{2}}\frac{F_{\omega \gamma \pi }(M^{2})}{%
F_{\pi }(M^{2})}. \label{VI.45}
\end{eqnarray}
The coupling constants $f_{\rho \omega \pi }$, $f_{\omega \gamma \pi }$, and
$f_{\rho \pi \pi }\ $ are determined earlier (see Table 1). We used in Eq.(%
\ref{VI.45}) the relations $f_{\omega \gamma \pi }\approx f_{\rho \omega \pi
}/g_{\rho }$ and $f_{\rho \pi \pi }/g_{\rho }\approx 1.$ It is seen that,
while signs of the couplings $f_{\rho \pi \pi }$ and $f_{\rho \omega \pi }$
are a matter of convention, the value $\gamma $ is positive and thus the
sign of the interference term is fixed.
The first two terms, (\ref{VI.40}) and (\ref{VI.41}), originate from the
diagram (b) of Fig. \ref{fig12}. The third term (\ref{VI.42}) originates
from the diagram (a), the fourth term comes from the diagram (c), and the
last term originates from the diagram (d). The second and the last terms and
the sum of the rest are transverse with respect to the photon momentum: $%
{\cal M}_{\tau \mu }^{2}k_{\mu }=0,$ ${\cal M}_{\tau \mu }^{5}k_{\mu }=0,$ $%
\sum_{i=1,3,4}{\cal M}_{\tau \mu }^{i}k_{\mu }=0$. The matrix element, $%
{\cal M}_{\tau \mu },$ is not transverse, however, with respect to the $\rho
$-meson momentum, ${\cal M}_{\tau \mu }P_{\tau }\neq 0$. A complete theory
should yield a matrix element which is transverse with respect to both, the
photon and the $\rho $-meson momenta. This is not the case in our effective
theory. In the following, we work exclusively with the transverse part of
the matrix element which is responsible for decays of the physical states,
and ignore the longitudinal part responsible for a decay of a spurious
spin-zero component of the $\rho $-meson. This is achieved by contraction of
the square of the matrix element with the polarization tensor
\begin{eqnarray}
\Sigma _{\tau \sigma }(P)=-g_{\tau \sigma }+\frac{P_{\tau }P_{\sigma }}{s}
\label{VI.46}
\end{eqnarray}
of the massive vector particle, as we did {\it e.g.} in Eq.(VI.24). The
difference is, however, that the term $P_{\tau }P_{\sigma}/s$ in Eq.(VI.24)
can be dropped before calculations. In the case considered in this
subsection the term $P_{\tau }P_{\sigma}/s$ cannot be dropped.
The square of the matrix element averaged over the initial polarizations of
the $\rho $-meson and summed up over the photon polarizations has the form
\begin{eqnarray}
{\cal R}\equiv \sum_{f}\overline{\left| {\cal M}\right| ^{2}}%
=\sum_{i,j=1}^{5}{\cal R}_{ij} \label{VI.47}
\end{eqnarray}
where
\begin{eqnarray}
{\cal R}_{ij}={\cal M}_{\tau \mu }^{i}{\cal M}_{\sigma \nu }^{j*}\frac{1}{3}%
(-g_{\tau \sigma }+\frac{P_{\tau }P_{\sigma }}{s})(-g_{\mu \nu }).
\label{VI.48}
\end{eqnarray}
Let's now consider separately the different interference terms.
We perform an additional averaging of the terms ${\cal R}_{ij}$ with $%
i,j=1\div 4$ and $i=1 \div 3$ and $j=5$ over directions of the pion momentum
${\bf p}_{1}$ in the c. m. frame of two pions. The remaining terms ${\cal R}%
_{45}$ and ${\cal R}_{55}$ are averaged over the directions of the pion
momentum ${\bf p}_{2}$ in the c. m. frame of the pion $\pi ^{0}$ and the
photon. The decay rate will finally be given as a sum of the two integrals
over the variables $s_{12}=(p_{1}+p_{2})^{2}$ and $s_{13}=(p_{2}+k)^{2}$.
The straightforward calculations give
\begin{eqnarray}
{\cal R}_{11} &=&-\frac{4}{9}\frac{2s_{12}+2s-M^{2}}{(s_{12}-s)^{2}}p^{*2}(%
\sqrt{s_{12}},\mu ,\mu )(3+\frac{p^{*2}(\sqrt{s},\sqrt{s_{12}},M)}{s_{12}}),
\nonumber \\
{\cal R}_{22} &=&\frac{4}{9}\frac{\lambda ^{2}}{(s_{12}-s)^{2}}p^{*2}(\sqrt{%
s_{12}},\mu ,\mu )\left[ p^{*2}(\sqrt{s},\sqrt{s_{12}},M)-3M^{2}+\frac{1}{%
4s_{12}}(s-M^{2}-s_{12})^{2}\right. \nonumber \\
&&\left. +\frac{1}{4ss_{12}}(s+M^{2}-s_{12})(s+s_{12}-M^{2})(s-M^{2}-s_{12})%
\right] , \nonumber \\
{\cal R}_{33} &=&-\frac{1}{3B_{\pi }^{2}}\left[ -\frac{1}{s}B_{\pi
}^{2}(3s+s_{12}-4\mu ^{2})\right. \nonumber \\
&&\left. -(s-4\mu ^{2})(M^{2}-4\mu ^{2})F(\xi )+2B_{\pi }(s+M^{2}-8\mu
^{2})L(\xi )\right] , \nonumber \\
{\cal R}_{44} &=&1, \nonumber \\
{\cal R}_{55} &=&\frac{4}{9}\frac{\gamma ^{2}}{(s_{23}-m_{\omega }^{2})^{2}}%
ss_{23}p^{*2}(\sqrt{s},\sqrt{s_{23}},\mu )p^{*2}(\sqrt{s_{23}},\mu ,M),
\label{VI.49}
\end{eqnarray}
\begin{eqnarray}
{\cal R}_{12}+{\cal R}_{21} &=&\frac{8}{9}\frac{\lambda }{(s_{12}-s)^{2}}%
\frac{s+s_{12}-M^{2}}{s_{12}}p^{*2}(\sqrt{s_{12}},\mu ,\mu )p^{*2}(\sqrt{s},%
\sqrt{s_{12}},M), \nonumber \\
{\cal R}_{13}+{\cal R}_{31} &=&-\frac{2}{3}\frac{1}{(s_{12}-s)B_{\pi }}%
\left[ (-4\mu ^{2}+M^{2}-s)B_{\pi }+\frac{1}{3s}B_{\pi }^{3}\xi ^{2}+\right.
\nonumber \\
&&\left. \frac{1}{2}(s+s_{12}-M^{2})(s+s_{12}+M^{2}-8\mu ^{2})L(\xi )\right]
, \nonumber \\
{\cal R}_{14}+{\cal R}_{41} &=&0, \nonumber \\
{\cal R}_{15}+{\cal R}_{51} &=&\frac{4\gamma }{3}\frac{sp^{*2}(\sqrt{s_{12}}%
,\mu ,\mu )p^{*2}(\sqrt{s},\sqrt{s_{12}},M)}{(s_{12}-s)B_{\omega }}\frac{%
1+(\zeta ^{2}-1)L(\zeta )}{\zeta ^{2}}, \nonumber \\
{\cal R}_{23}+{\cal R}_{32} &=&-\frac{2}{3}\frac{\lambda }{(s_{12}-s)B_{\pi }%
}\left[ \frac{B_{\pi }}{12ss_{12}}((4\mu ^{2}-s_{12})M^{4}\right. \nonumber
\\
&&+(-8\mu ^{2}s-10ss_{12}-4s_{12}^{2}+16\mu ^{2}s_{12})M^{2}+(4\mu
^{2}-s_{12})(s-s_{12})(s+5s_{12})) \nonumber \\
&&\left. +\frac{1}{2}M^{2}(s_{12}+M^{2}+s-8\mu ^{2})L(\xi )\right] ,
\nonumber \\
{\cal R}_{24}+{\cal R}_{42} &=&0, \nonumber \\
{\cal R}_{25}+{\cal R}_{52} &=&\frac{2}{3}\frac{\lambda \gamma }{%
(s_{12}-s)B_{\omega }}\left[ \frac{1}{2}B_{\omega }^{3}(\frac{1}{3}\zeta
^{2}+1-L(\zeta ))+\frac{1}{2}s_{12}B_{\omega }^{2}(-1+L(\zeta ))\right.
\nonumber \\
&&+B_{\omega }(-\frac{1}{2}sp^{*2}(\sqrt{s},\sqrt{s_{12}}%
,M)+(s+M^{2}-s_{12})p^{*2}(\sqrt{s_{12}},\mu ,\mu ))(1-L(\zeta )) \nonumber
\\
&&\left. +sp^{*2}(\sqrt{s},\sqrt{s_{12}},M)p^{*2}(\sqrt{s_{12}},\mu ,\mu )%
\frac{1+(\zeta ^{2}-1)L(\zeta )}{\zeta ^{2}}\right] , \nonumber \\
{\cal R}_{34}+{\cal R}_{43} &=&\frac{2}{3}\frac{1}{B_{\pi }}\left[ (-4\mu
^{2}+s_{12})L(\xi )+B_{\pi }(-1+L(\xi ))\right] , \nonumber \\
{\cal R}_{35}+{\cal R}_{53} &=&\frac{8}{3}\frac{\gamma }{B_{\pi }B_{\omega }}%
\frac{1}{\xi +\zeta }\left[ (-\frac{1}{4}s_{12}+\mu ^{2})(s_{12}M^{2}-\frac{1%
}{4}(s-M^{2}-s_{12})^{2})(\xi L(\xi )+\zeta L(\zeta ))\right. \nonumber \\
&&\left. -\frac{1}{4}s_{12}\xi B_{\pi }{}^{2}(-1+L(\xi )+\frac{\xi }{\zeta }%
(-1+L(\zeta )))\right] , \nonumber \\
{\cal R}_{45}+{\cal R}_{54} &=&0. \label{VI.50}
\end{eqnarray}
where $B_{\omega }=B_{\pi }+\mu ^{2}-m_{\omega }^{2}$ and $\zeta =B_{\pi
}\xi /B_{\omega }$.
The decay width $\rho ^{+}\rightarrow \pi ^{+}\pi ^{0}\gamma ^{*}$ is given
by expression
\begin{eqnarray}
\Gamma (\rho ^{+} &\rightarrow& \pi ^{+}\pi ^{0}\gamma ^{*}) = \frac{\alpha
}{16\pi ^{2}s}f_{\rho \pi \pi }^{2}\left| F_{\pi }(M^{2})\right| ^{2}
\nonumber \\
&&\times \left[ \int_{4\mu ^{2}}^{(\sqrt{s}-M)^{2}}(\sum_{i=1}^{4}%
\sum_{j=1}^{4}+2\sum_{i=1}^3\sum_{j=5}^5) {\cal R}_{ij}{\cal \;}\frac{p^{*}(%
\sqrt{s},\sqrt{s_{12}},M)p^{*}(\sqrt{s_{12}},\mu ,\mu )}{\sqrt{s_{12}}}%
ds_{12}\right. \nonumber \\
&&\left. +\int_{(\mu +M)^{2}}^{(\sqrt{s}-\mu )^{2}}{\cal R}_{55}{\cal \;}
\frac{p^{*}(\sqrt{s},\sqrt{s_{23}},M)p^{*}(\sqrt{s_{23}},\mu ,M)}{\sqrt{%
s_{23}}}ds_{23}\right] . \label{VI.52}
\end{eqnarray}
The dilepton spectrum in the $\rho ^{+}\rightarrow \pi ^{+}\pi ^{0}\ell
^{+}\ell ^{-}$ decay can be calculated from equation
\begin{eqnarray}
d\Gamma (\rho ^{+}\rightarrow \pi ^{+}\pi ^{0}\ell ^{+}\ell ^{-})=\Gamma
(\rho ^{+}\rightarrow \pi ^{+}\pi ^{0}\gamma ^{*})M\Gamma (\gamma
^{*}\rightarrow \ell ^{+}\ell ^{-})\frac{dM^{2}}{\pi M^{4}}. \label{VI.53}
\end{eqnarray}
The decay $\rho ^{-}\rightarrow \pi ^{-}\pi ^{0}\ell ^{+}\ell ^{-}$ is
completely equivalent to the $\rho ^{+}$-meson decay.
\subsection{Decay mode $f_{0}(980)\rightarrow $ $\pi ^{+}\pi ^{-}\ell
^{+}\ell ^{-}$}
The dominant decay mode of the $f_{0}(980)$-meson is $f_{0}\rightarrow $ $%
\pi \pi $. The effective vertex for this decay looks like
\begin{eqnarray}
\delta {\cal L}=g_{f_{0}\pi \pi }f_{0}\pi ^{\alpha }\pi ^{\alpha }.
\label{VI.54}
\end{eqnarray}
The $f_{0}\rightarrow $ $\pi ^{+}\pi ^{-}$ decay width has the form
\begin{eqnarray}
\Gamma (f_{0}\rightarrow \pi ^{+}\pi ^{-})=\frac{1}{8\pi s}g_{f_{0}\pi \pi
}^{2}p^{*}(\sqrt{s},\mu ,\mu ), \label{VI.55}
\end{eqnarray}
with $\sqrt{s}=m_{f_{0}}.$ The experimental width $\Gamma
^{tot}(f_{0}\rightarrow \pi \pi )=\frac{3}{2}\Gamma (f_{0}\rightarrow \pi
^{+}\pi ^{-})=3\Gamma (f_{0}\rightarrow \pi ^{0}\pi ^{0})$ lies within the
interval $50-100$ MeV \cite{PDG}. The coupling constant can be found to be $%
g_{f_0 \pi \pi } = 1.6$ GeV$\pm \; 30 \%$.
The diagrams for the $f_{0}\rightarrow $ $\pi ^{+}\pi ^{-}\gamma ^{*}$ decay
are shown on Fig. \ref{fig13}. The matrix element has the form
\begin{eqnarray}
{\cal M}=-ieg_{f\pi \pi }{\cal M}_{\mu }\varepsilon _{\mu }^{*}(k)
\label{VI.56}
\end{eqnarray}
with
\begin{eqnarray}
{\cal M}_{\mu }=\frac{(2p_{1}+k)_{\mu }}{(p_{1}+k)^{2}-\mu ^{2}}-\frac{%
(2p_{2}+k)_{\mu }}{(p_{2}+k)^{2}-\mu ^{2}}. \label{VI.57}
\end{eqnarray}
The tensor ${\cal M}_{\mu }$ is transverse with respect to the photon
momentum. The square of the matrix element summed up over the photon
polarizations and averaged over the directions of the pion momentum in the
c. m. frame of two pions has the form
\begin{eqnarray}
{\cal R} &\equiv &\int \frac{d\Omega _{12}}{4\pi }{\cal M}_{\mu } {\cal M}%
_{\nu }^{*}(-g_{\mu \nu }) \nonumber \\
&=&\frac{2}{B_{\pi }^{2}}\left[ (M^{2}-4\mu ^{2})F(\xi )+(2s_{12}-4\mu
^{2}-M^{2})L(\xi )\right] , \label{VI.59}
\end{eqnarray}
where $s_{12}=(p_{1}+p_{2})^{2}$. The functions $F(\xi )$ and $L(\xi )$ are
defined in Eqs. (\ref{VI.14}) and (\ref{VI.15}) in terms of the parameter $%
\xi $ which, in turn, is defined in Eq. (\ref{VI.16}). The decay width
becomes
\begin{eqnarray}
\Gamma (f_{0}\rightarrow \pi ^{+}\pi ^{-}\gamma ^{*})&=&\frac{\alpha }{16\pi
^{2}s}g_{f_{0}\pi \pi }^{2}\left| F_{\pi }(M^{2})\right| ^{2} \nonumber \\
&\times& \int_{4\mu ^{2}}^{(\sqrt{s}-M)^{2}}{\cal R\;}\frac{p^{*}(\sqrt{s},%
\sqrt{s_{12}},M)p^{*}(\sqrt{s_{12}},\mu ,\mu )}{\sqrt{s_{12}}}ds_{12}.
\label{VI.60}
\end{eqnarray}
The dilepton spectrum can be obtained with the use of Eq.(\ref{II.10}).
\subsection{Decay mode $a_{0}(980)\rightarrow $ $\pi \eta \ell ^{+}\ell ^{-}$%
}
The dominant decay mode of the $a_{0}(980)$-meson is $a_{0}\rightarrow $ $%
\pi \eta .$ The effective vertex for this decay looks like
\begin{eqnarray}
\delta {\cal L}=g_{a_{0}\pi \eta }a_{0}^{\alpha }\pi ^{\alpha }\eta .
\label{VI.61}
\end{eqnarray}
The $a_{0}\rightarrow $ $\pi \eta $ decay width has the form
\begin{eqnarray}
\Gamma (a_{0}\rightarrow \pi \eta )=\frac{1}{8\pi s}g_{a_{0}\pi \eta
}^{2}p^{*}(\sqrt{s},\mu ,\mu ^{\prime }) \label{VI.62}
\end{eqnarray}
where $\mu $ and $\mu ^{\prime }$ are the pion and the $\eta $-meson masses.
The experimental width $\Gamma (a_{0}\rightarrow \pi \eta )$ lies within the
interval $50-100$ MeV. The value of the coupling constant $g_{a_{0}\pi \eta}$
determined from the width is given in Table 1.
Since the $\eta $-meson is neutral, photons can only be emitted from the $%
a_{0}$- and $\pi $-mesons in an isospin $I_{3}=\pm 1$ state. To the lowest
order in $\alpha$, $\Gamma (a_{0}^{0}\rightarrow \pi ^{0}\eta \gamma )=0.$
Two diagrams for the $a_{0}^{+}\rightarrow $ $\eta \pi ^{+}\gamma ^{*}$
decays are shown on Fig. \ref{fig14}. The matrix element can be written in
the form
\begin{eqnarray}
{\cal M}=-ieg_{a\eta \pi }{\cal M}_{\mu }\varepsilon _{\mu }^{*}(k)
\label{VI.63}
\end{eqnarray}
with
\begin{eqnarray}
{\cal M}_{\mu }=\frac{(2P-k)_{\mu }}{(P-k)^{2}-s}+\frac{(2p_{1}+k)_{\mu }}{%
(p_{1}+k)^{2}-\mu ^{2}} \label{VI.64}
\end{eqnarray}
where $p_{1}$ is the pion momentum, $\sqrt{s}=m_{a_{0}}$ is the $a_{0}$%
-meson mass, and $P^{2}=s$. The tensor part of the matrix element, ${\cal M}%
_{\mu }$, is transverse with respect to the photon momentum. The square of
the matrix element summed up over photon polarizations and averaged over the
directions of the pion momentum in the c. m. frame of the $\pi $- and $\eta $%
-mesons has the form
\begin{eqnarray}
{\cal R} &\equiv &\int \frac{d\Omega _{12}}{4\pi }{\cal M}_{\mu }{\cal M}%
_{\nu }{}^{*}(-g_{\mu \nu }) \nonumber \\
&=&-\frac{1}{(s_{12}-s)^{2}}\left[ 2s+2s_{12}-M^{2}\right] -\frac{1}{B_{\pi
}^{\prime 2}}\left[ (4\mu ^{2}-M^{2})F(\xi ^{\prime })+2B_{\pi }^{\prime
}L(\xi ^{\prime })\right] \nonumber \\
&&-\frac{2}{(s_{12}-s)B_{\pi }^{\prime }}\left[ (s+s_{12}-M^{2}+2(\mu
^{2}-\mu ^{\prime 2}))L(\xi ^{\prime })+B_{\pi }^{\prime }\right]
\label{VI.65}
\end{eqnarray}
where $s_{12}=(p_{1}+p_{2})^{2},$ $p_{2}$ is the $\eta $ momentum, $%
p_{2}^{2}=\mu ^{\prime 2}$, and $F(\xi ^{\prime })$ and $L(\xi ^{\prime })$
are functions of the parameter
\begin{eqnarray}
\xi ^{\prime }=\frac{2}{B_{\pi }^{\prime }}\sqrt{\frac{s}{s_{12}}}p^{*}(%
\sqrt{s_{12}},\mu ,\mu ^{\prime })p^{*}(\sqrt{s},\sqrt{s_{12}},M).
\label{VI.66}
\end{eqnarray}
The value $B_{\pi }^{\prime }$ is given by
\begin{eqnarray}
B_{\pi }^{\prime }=\frac{1}{2}(s+M^{2}-s_{12})+\frac{1}{2s_{12}}(\mu
^{2}-\mu ^{\prime 2})(s-M^{2}-s_{12}). \label{VI.67}
\end{eqnarray}
The $a_{0}^{+}\rightarrow \pi ^{+}\eta \gamma ^{*}$ decay width becomes
\begin{eqnarray}
\Gamma (a_{0}^{+}\rightarrow \pi ^{+}\eta \gamma ^{*})&=&\frac{\alpha }{%
16\pi ^{2}s}g_{a_{0}\pi \eta }^{2}\left| F_{\pi }(M^{2})\right| ^{2}
\nonumber \\
&\times&\int_{(\mu +\mu ^{\prime })^{2}}^{(\sqrt{s}-M)^{2}}{\cal R\;}\frac{%
p^{*}(\sqrt{s},\sqrt{s_{12}},M)p^{*}(\sqrt{s_{12}},\mu ,\mu ^{\prime })}{%
\sqrt{s_{12}}}ds_{12}. \label{VI.68}
\end{eqnarray}
The dilepton spectrum can be calculated using Eq.(\ref{II.10}):
\begin{eqnarray}
d\Gamma (a_{0}^{+}\rightarrow \pi ^{+}\eta e^{+}e^{-})=\Gamma
(a_{0}^{+}\rightarrow \pi ^{+}\eta \gamma ^{*})M\Gamma (\gamma
^{*}\rightarrow e^{+}e^{-})\frac{dM^{2}}{\pi M^{4}} \label{VI.69}
\end{eqnarray}
The possible values of $M$ lie in the interval $2m\leq M\leq \sqrt{s}-\mu
-\mu ^{\prime }$.
\section{Numerical results}
\setcounter{equation}{0}
The dilepton meson decays have essentially the same physical origin as the
radiative meson decays. Eq.(\ref{II.3}) allows a direct calculation of the
radiative widths within the framework used for the calculation of the
dilepton widths. Thus, the radiative decays provide a useful test for the
present model.
\subsection{Radiative meson decays}
The results for the radiative decays with two mesons in the final states are
shown in Table 2. The $\rho ^0\rightarrow \pi ^{+}\pi ^{-}\gamma $, $\omega
\rightarrow \pi ^0\pi ^0\gamma $, and $\eta ^{\prime }\rightarrow \pi
^{+}\pi ^{-}\gamma $ branching ratios are in the reasonable agreement with
experiment. The branching of the $\eta \rightarrow \pi ^{+}\pi ^{-}\gamma $
decay is higher than the experimental value. The $\eta $-meson width is
known with an accuracy of $10\%$. This uncertainty is also present in the
calculated branching. The coupling constant $f_{\rho \gamma \eta }^2$ to
which this branching is proportional is known with an accuracy of $20\%$. It
is, however, not possible to use these uncertainties in order to decrease
the predicted branching, since the branching for the $\eta \rightarrow \pi
^{+}\pi ^{-}e^{+}e^{-}$ decay is lower than the experimental value (see
Table 3). The measurement of the $\eta \rightarrow \pi ^{+}\pi
^{-}e^{+}e^{-} $ reaction has been done quite a long time ago and with
rather pure statistics (see Ref. \cite{RAG66}). It would be desirable to
remeasure this process. The experimental branching for the $\eta ^{\prime
}\rightarrow \pi ^{+}\pi ^{-}\gamma $ decay is taken from Ref. \cite{SIB},
others are from \cite{PDG}. The results for the $\rho ^0\rightarrow \pi
^0\pi ^0\gamma $, $\rho \rightarrow \pi ^0\eta \gamma $, $\omega \rightarrow
\pi ^0\pi ^0\gamma $, and $\omega \rightarrow \pi ^0\eta \gamma $ decays are
in agreement with the calculations by Bramon, Grau and Pancheri \cite{ABram}
where the same model for these decays is used. The branching ratios for
decays including photon bremsstrahlung are calculated for photon energies
above $50$ MeV. The data on the radiative decays with one meson in the final
state were used as input to fix the $V\gamma P$ couplings. These decays are
not listed in Table 2.
The results for the dilepton branching ratios are summarized in Table 3.
\subsection{$\rho $-meson decays}
The branching ratio $B(\rho ^{0}\rightarrow \pi ^{+}\pi ^{-}\gamma )=(9.9\pm
1.6)\times 10^{-3}$ is almost one order of magnitude greater than the
branching ratio $B(\rho ^{0}\rightarrow \pi ^{0}\gamma )=(7.9\pm 2.0)\times
10^{-4}$. Despite the fact that the first one depends on the experimental
cut in the photon energy, one can expect that the decay mode $\rho
^{0}\rightarrow $ $\pi ^{+}\pi ^{-}e^{+}e^{-}$ is important. Indeed, it is
seen from Table 3 that the $\rho ^{0}\rightarrow \pi ^{+}\pi ^{-}e^{+}e^{-}$
branching ratio is more than one order of magnitude greater than the $\rho
\rightarrow \pi e^{+}e^{-}$ branching ratio and $4$ times greater than
branching ratio of the direct $\rho ^{0}\rightarrow e^{+}e^{-}$ decay. The
dominant contribution comes, however, from the region of small invariant
masses of the $e^{+}e^{-}$ pair, which is not seen experimentally. The same
is true for the $\rho ^{\pm }\rightarrow \pi ^{\pm }\pi ^{0}e^{+}e^{-}$
decays. For invariant masses $M>100$ MeV, the number of the $e^{+}e^{-}$
pairs due to the $\rho ^{0}\rightarrow \pi ^{+}\pi ^{-}e^{+}e^{-}$ and $\rho
^{\pm }\rightarrow \pi ^{\pm }\pi ^{0}e^{+}e^{-}$ decays increases by $30\%$
as compared to the direct mode, if one assumes that the $\rho ^{0}$- and $%
\rho ^{\pm }$-mesons are produced with equal probabilities. The differential
branching ratios for the $\rho $-meson decays are shown for the $e^{+}e^{-}$
channels in Fig. \ref{fig15} (a) and for the $\mu ^{+}\mu ^{-}$ channels in
Fig. \ref{fig15} (b). At small invariant masses such that $2m_{e}\lesssim M$%
, the slope equals $d\log B^{\prime}/d\log M\approx -1$ where $%
B^{\prime}=dB/dM$, when there is no bremsstrahlung. Such a behavior appears
because of the $dM/M $ dependence of the differential branching ratio
according to Eq.(\ref{II.10}). The bremsstrahlung makes the ${\sf M}%
\rightarrow {\sf M}^{\prime }\gamma ^{*}$ widths more singular ($\sim \log
^{3}(1/M)$) and thus the slope of the other curves is larger.
In order to get a comparison with the direct channel, we plotted also a
weighted $\rho ^{0}\rightarrow e^{+}e^{-}$ distribution according to
equation
\begin{eqnarray}
\frac{dB}{dM}=\frac{1}{\pi }\frac{2Mm_{\rho }\Gamma_{\rho} (M)}{%
(M^{2}-m_{\rho }^{2})^{2}+m_{\rho }^{2}\Gamma_{\rho}(M)^{2}}B(\rho
\rightarrow \ell ^{+}\ell ^{-}). \label{VII.1}
\end{eqnarray}
The $\rho $-meson width as a function of the value $M$ is assumed to have to
the form
\begin{eqnarray}
\Gamma_{\rho}(M)=\Gamma _{\rho }\frac{p^{*3}(M,\sqrt{s_0}/2 ,\sqrt{s_0}/2 )}{%
p^{*3}(m_{\rho },\sqrt{s_0}/2 ,\sqrt{s_0}/2 )}\frac{m_{\rho
}^{2}+p^{*2}(m_{\rho },\sqrt{s_0}/2 ,\sqrt{s_0}/2 )}{m_{\rho }^{2}+p^{*2}(M,%
\sqrt{s_0}/2 ,\sqrt{s_0}/2 )} \label{VII.2}
\end{eqnarray}
where $\Gamma _{\rho }$ is the total experimental width and $\sqrt{s_0} =
2\mu$ is the two-pion threshold energy. In the narrow-width limit, the
integral from the left side of Eq.(\ref{VII.1}) coincides with $B(\rho
\rightarrow e^{+}e^{-})$.
\subsection{$\omega $-meson decays}
The differential branching for the $\omega $-meson decays are shown in Figs.
\ref{fig16} (a,b). The $\omega $ decays are dominated through the $\omega
\rightarrow \pi ^{0}e^{+}e^{-}$ decay mode. The other modes give only a
small correction to the background. It is interesting that strength of the
channel $\omega \rightarrow \pi ^{0}e^{+}e^{-}$ is one order of magnitude
greater than strength of the direct channel $\omega \rightarrow e^{+}e^{-}$.
The $e^{+}e^{-}$ pairs from these two decays have quite different invariant
masses. The background appears below $500$ MeV. Like in the case of the $%
\rho $-meson, we plot also the weighted $\omega \rightarrow \ell ^{+}\ell
^{-}$ distributions according to Eq.(\ref{VII.1}) with substitutions $%
m_{\rho }\leftrightarrow m_{\omega }$ and $\Gamma _{\rho }\leftrightarrow
\Gamma _{\omega }$ for the threshold energy $\sqrt{s_0} = 3\mu$.
\subsection{$\phi $-meson decays}
The differential branching ratios for the $\phi $-meson decays are shown in
Figs. \ref{fig17} (a,b). We do not consider two-meson finals states. The
dominant ones could be $\phi \rightarrow \pi \pi \gamma ^{*}$ and $\phi
\rightarrow \pi ^0\eta \gamma ^{*}$. The first decay with a real photon in
the $\pi ^0\pi ^0$ mode was observed experimentally \cite{MNA,VEPP}. It has
branching $B^{\exp }(\phi \rightarrow \pi ^0\pi ^0\gamma )=(1.14\pm
0.22)\times 10^{-4}$. This decay dominates through the $f_0\gamma $
mechanism. One can expect that branching $B(\phi \rightarrow \pi ^0\pi
^0e^{+}e^{-})$ is two orders of magnitude smaller than branching of the
radiative decay, and so smaller than all $\phi $-meson branching ratios
listed in Table 3. The $\phi \rightarrow \pi \eta \gamma $ decay was also
measured in the SND experiment \cite{MNA,VEPP} to give $B^{\exp }(\phi
\rightarrow \pi ^0\eta \gamma )=(1.3\pm 0.5)\times 10^{-4}$. The branching $%
B(\phi \rightarrow \pi \eta e^{+}e^{-})$ is also expected to be small. The
measured $\phi \rightarrow \pi \pi \gamma $ and $\phi \rightarrow \pi ^0\eta
\gamma $ widths are significantly greater than predictions of Ref. \cite
{ABram} where formation of the intermediate $f_0\gamma $ and $a_0\gamma $
states is not considered. At present there is, however, yet no experimental
evidence for an $a_0$-meson structure in the $\phi \rightarrow \pi ^0\eta
\gamma $ decay, and thus it is not clear how to interpret the excess in the $%
\phi \rightarrow \pi ^0\eta \gamma $ mode. Consequently, the calculation of
the branching ratios for the dilepton $\phi \rightarrow \pi ^0\eta \ell
^{+}\ell ^{-}$ modes is still unclear.
We do not consider decay modes with kaons in the final states. The invariant
mass of the $e^{+}e^{-}$-pairs in such decays does not exceed $50$ MeV, so
they occur in the region excluded by the experimental cuts $M>50$ MeV. In
Figs. \ref{fig17} (a,b), the $\phi \rightarrow \ell ^{+}\ell ^{-}$
distributions are plotted according to Eq.(\ref{VII.1}) modified for the $%
\phi$-meson. The dominant two-kaon decay channel with $\sqrt{s_0}=2\mu_K$
where $\mu_K$ is the kaon mass and also the three-pion decay channel with $%
\sqrt{s_0}=3\mu$ are taken into account in the $M$-dependent width $%
\Gamma_{\phi}(M)$.
\subsection{$\eta $- and $\eta ^{\prime }$-mesons decays}
The differential branching ratios for the $\eta $- and $\eta ^{\prime }$%
-mesons are shown in Figs. \ref{fig18} (a,b). The $\eta $-meson dominates
through the $\eta \rightarrow \gamma \ell ^{+}\ell ^{-}$ modes. The dominant
$e^{+}e^{-}$ modes for the $\eta ^{\prime }$-meson are $\eta ^{\prime
}\rightarrow \gamma e^{+}e^{-}$ at $M>250$ MeV and $\eta ^{\prime
}\rightarrow \pi ^{+}\pi ^{-}e^{+}e^{-}$ at $M\lesssim 250$ MeV. The $\eta
^{\prime }\rightarrow \gamma \ell ^{+}\ell ^{-}$ modes display clear
structures connected to the $\omega $- and $\rho $-meson contributions to
the $\eta ^{\prime }\gamma \gamma $ transition form factors. The $\eta
\rightarrow \pi ^{0}\pi ^{0}\ell ^{+}\ell ^{-}$ decay is forbidden by $C$%
-parity conservation, similarly for the $\eta ^{\prime }$-meson.
\subsection{$\pi ^{0}$-, $f_{0}$-, and $a_{0}$-mesons decays}
The differential branching ratios for the $\pi ^{0}$-, $f_{0}$-, and $a_{0}$%
-mesons are shown in Figs. \ref{fig19} (a,b). At $M\lesssim 500$ MeV, the $%
f_{0}$-meson decays dominate through the $f_{0}\rightarrow \pi ^{+}\pi
^{-}\ell ^{+}\ell ^{-}$ mode. The four-body $a_{0}$-meson decay also becomes
dominant with decreasing the values of $M$. The dilepton spectra of the $%
f_{0}$- and $a_{0}$-mesons display clear structures connected to the $\omega
$- and $\rho $-meson contributions to the transition form factors.
\section{Conclusion}
The present work is an attempt to approach the DLS puzzle through including
new meson decay channels. We studied decay modes of the light mesons below
the $\phi (1020)$-meson into $e^{+}e^{-}$ and $\mu ^{+}\mu ^{-}$ pairs.
Besides the direct decays and some Dalitz decays, which are well known in
the literature, we presented a systematic investigation of the decay modes
which contribute to the dilepton background and have not yet been taken into
account in previous studies. These are special Dalitz decays as, {\it e.g.} $%
\eta ^{\prime }\rightarrow \gamma \ell ^{+}\ell ^{-}$, $f_0\rightarrow
\gamma \ell ^{+}\ell ^{-}$ and most of the decays to four-body final states.
Although many of these processes are found to give small contributions to
the total dilepton branching ratios, this is not the case for all of them:
We found that in the $\rho ^0$-meson decays, the dominant contribution to
the background below $350$ MeV comes from the $\rho ^0\rightarrow \pi
^{+}\pi ^{-}e^{+}e^{-}$ decay. The decay modes $\rho ^0\rightarrow \pi
^{+}\pi ^{-}e^{+}e^{-}$ and $\rho ^{\pm }\rightarrow \pi ^{\pm }\pi
^0e^{+}e^{-}$ increase the number of the $e^{+}e^{-}$ yield with invariant
mass $M>100$ MeV by about $30\%$ compared to the direct mode $\rho
^0\rightarrow e^{+}e^{-}$ at $M<1$ GeV, if one assumes that production cross
sections for the $\rho ^{\pm }$- and $\rho ^0$-mesons are equal. In case of
the $f_0$-meson, the $f_0\rightarrow \pi ^{+}\pi ^{-}e^{+}e^{-}$ decay is in
the interval $100-400$ MeV one to three orders of magnitude more important
than the next to the dominant $f_0\rightarrow \gamma e^{+}e^{-}$ mode. In
most other cases, the four-body decays give only small contributions and
thus there neglection appears to be justified.
The calculated branching ratios can further be used for experimental
searches for the dilepton meson decays.
The production rates of mesons in $p+p$, $p+A$, and in particular in
heavy-ion ($A+A$) collisions differ by orders of magnitude and depend
sensitively on the available energy of the system. The relative strength of
specific channels for different mesons can hardly be estimated without
complete transport calculations. However, for every meson we can decide
which channels are more important. The relative strength of e.g. $\rho
^{0}\rightarrow \pi ^{0}\ell ^{+}\ell ^{-}$ and $\rho ^{0}\rightarrow \pi
^{+}\pi ^{-}\ell ^{+}\ell ^{-}$ decay modes can be found without knowing the
cross section for the $\rho ^{0}$-meson production.
Since theoretical models for the dileption production have to be adjusted to
$p+p$ and possibly $\pi +p$ reactions before applied to $A+A$ reactions, a
detailed knowledge of all relevant decay channels and background processes
is indispensable. In particular, the lack in the understanding of the excess
of dilepton pairs below the $\rho $-meson peak in the BEVALAC as well as in
the CERN data shows that scenarios with a reduced $\rho $-meson mass can be
treated at present as a plausible hypothesis only. To draw definite physical
conclusions, one needs to eliminate possible trivial explanations like those
connected to existence of the nondirect dilepton decays of light unflavored
mesons.
After submitting this paper, we got to know on papers by Koch \cite{koch93}
and Lichard \cite{lichard95} where a quite complete set of the Dalitz decays
(without scalar mesons) and also a few four-body channels were investigated.
In Ref. \cite{koch93}, four-body channels are discussed within an
approximation suggested by Jarlskog and Pilkuhn \cite{JP67}, which consists
in neglecting the $M$-dependence of the matrix elements for emission of the
virtual photons with mass $M$. We performed straightforward calculations
without approximations and included additional Dalitz and four-body
channels. In Ref. \cite{lichard95}, the decay mode $\rho ^{0}\rightarrow \pi
^{+}\pi ^{-}e^{+}e^{-}$ is analyzed in details. Our expressions (VI.13) and
(VI.18) coincide with the corresponding formulae from the Lichard paper
where, however, the pion form factor was neglected. The pion form factor has
a small impact on the total $e^{+}e^{-}$ decay rate. The dimuon mode, where
higher invariant masses are involved, is enhanced by about 40\%.
\begin{acknowledgments}
The authors are indebted to A. Bramon for useful remarks.
M.I.K. is grateful to the Institute for Theoretical
Physics of University of T\"ubingen for kind hospitality.
The work is supported by GSI (Darmstadt) under the contract T\"UF\"AST.
\end{acknowledgments}
\newpage
|
\section{The effect of environment on extragalactic jets}
\subsection{Complex and distorted structure in extragalactic radio
sources}
The jets and hotspots of radio galaxies and quasars often show
complex structure, including bends, twists, knots and multiple
hotspots. Such structure is seen over a huge range of sizes, from the
enormous wide-angle tail (WAT) sources (which can be megaparsecs in
size), to the compact steep-spectrum (CSS) sources (which are about
10--15~kpc across). Jets can appear to bend by over 90$^{\circ}$ and
remain collimated for several jet radii (Bridle \& Perley 1984),
despite the expectation that the oblique shock causing the bend
should decelerate the jet. Explanations for these complex structures
include motion through some intra-cluster medium (Leahy 1984),
perturbations due to mergers (van Breugel et al.\ 1986, Sakelliou,
Merrifield \& McHardy 1996), variations in the direction of the jet
at its source (Williams \& Gull 1985, Scheuer 1982), and collision
with dense clouds in the ambient medium (Burns 1986). Cloud
collisions are particularly applicable in cases where these bends are
very sharp. Other models would produce more gradual bends.
Theoretical studies of jet-cloud collisions appear to be in conflict
over whether they can explain the observations. In some previous
numerical simulations the jet is decelerated and effectively disrupted
by the cloud, and the cloud is subsequently destroyed
(De~Young~1991). In others the jet appears to remain collimated as it
is re-accelerated in a new direction (Norman~1993). Most recently
Raga~\&~Canto~(1996) find from two-dimensional simulations and
analytical studies that the jet bores through the cloud but reaches a
steady configuration. In this paper we describe the results of a study
aimed at resolving this question by investigating the effects of
various hydrodynamical and geometrical parameters. This investigation
uses three-dimensional, adiabatic simulations. We also follow the
development of the interactions for a longer time than previous
simulations, and estimate the intensity of synchrotron emission from
the jet to enable meaningful comparison with observed radio maps.
Through these studies we hope to be able to determine what types of
complex structure can be explained by jet-cloud collisions. Such studies
could shed light on the alignment between radio and optical axes in high
redshift radio galaxies, the contribution of shocks to the spectra of
extended emission lines regions and the
role of environmental effects in unified schemes
for AGN and their evolution.
\subsection{Examples of distorted structure}
Wide-angle tail (WAT) sources are found in rich clusters and have
distorted, C-shaped structure. Large WATs approaching 1~Mpc across
cannot be bent by the motion of the parent galaxy through the cluster
without assuming unreasonably high speeds through unreasonably dense
intra-cluster gas (Burns 1986). To satisfy the requirements of
momentum balance and to reproduce the sharpness of the bends that are
observed, Burns proposed that the jets may collide with clouds of
higher density in the surrounding medium. A good example is the
western jet of the WAT 1919+479, which emerges from the core, vanishes
after a short distance, and then reappears at a bright hotspot (Burns
et al.\ 1986, Pinckney et al.\ 1994). Beyond this hotspot a tail
stretches out for 800~kpc in a direction about 90$^{\circ}$ to that of
the original jet, broadening and fading as it does so. The bend here
is very sharp. There is one other hotspot not far from the beginning
of the tail. Rotation measure and depolarisation are described as
`patchy', and vary significantly over the tail. X-ray observations
show large scale asymmetry in the intra-cluster medium. The
simulations of Loken et al.\ (1995) show that the necessary gas
velocities can arise in cluster mergers, as can shocks that will bend
the jet. However, these simulations still do not explain the sharpness
of the bend or the persistence of the halo.
Barthel et al.\ (1988) present a large sample of quasars with powers greater
than the Fanaroff-Riley division but distorted structures. Twenty out of a
sample of eighty high-redshift quasars showed bending greater than 20{$^{\circ}$},
instead of the classical double structure that would be expected.
Compact Steep Spectrum (CSS) sources make up 10--15\% of AGN. More
than 15\% have axes between the lobes and the core that
are misaligned by more than 20$^{\circ}$ (Saikia et al. 1995). This
distortion seems to be associated with an asymmetrical ambient medium.
In quasars the most prominent jets are in complex or one-sided structures,
and smaller sources are the more powerful (Muxlow \& Garrington 1991).
Stocke, Burns and Christiansen (1985) present observations of `dogleg'
quasars showing strong changes in direction within the lobes.
There is no evidence that
these sources are found any more often in rich or poor cluster environments,
a conclusion supported by recent work (Rector, Stocke \& Ellingson
1995). Only a few dense clouds are needed in each case to explain the
proportion of bent jets, so it would appear that it is the inhomogeneity of
the environment, not its overall density, that causes the bends.
\subsection{Intergalactic clouds in the neighbourhood of
extragalactic jets}
Evidence for inhomogeneity in the medium surrounding AGN comes from
measurements of depolarization and line emission, and correlations
between these properties/features and the radio structure.
Maps of rotation measure and depolarization of a sample of radio sources
show significant inhomogeneity on a range of scales, some as low as 5kpc,
some larger than 50kpc, out to distances of about 100kpc from the nuclei
(Pedelty et al.\ 1989). Such effects could be caused by density
inhomogeneities in the surrounding material.
The infrared, optical and ultraviolet structures of many high-redshift
radio galaxies are closely aligned with the radio structures (McCarthy
1993, Chambers, Miley \& van Breugel 1987). There are also
several correlations between line emission and other aspects of
distorted jets, for example: brighter line emission occurs on the side
of the nearer radio-lobe (McCarthy, van Breugel \& Kapahi 1991); the
level of blue light and the strength of the alignment effect are
correlated with a mix of radio power and spectral index (Dunlop \&
Peacock 1993). The lobes of these radio sources are often asymmetrical
with the lobe nearer the nucleus tending to be more depolarized than
the more distant lobe (Liu \& Pooley 1991), suggesting that the material
responsible for the depolarization may also present more resistance to the
jet.
Regions of optical line emission are associated with knots and bends in the
jets (Wilson 1993). For example, the radio galaxy 4C 29.30 has a region of
line emission close to a bright knot just before a bend (van Breugel et al.\
1986). The radio galaxy PKS2250-41 shows extended emission aligned
with the radio axes, including a large arc-shaped region of line emission
surrounding the radio lobe (Tadhunter et al.\ 1994).
Extended emission line regions have a ``clear spatial association'' with
regions of depolarized radio emission (Baum \& Heckman 1989).
In many sources (especially at higher redshifts) the spectra from this
extended emission line region can only be explained if a significant
shock component is included (Clark \& Tadhunter 1996, and references
therein). Low polarization of the ultra-violet emission shows that
there is not enough scattered AGN emission to account by itself for
the total flux (Tadhunter 1996). The cloud collision model is now
commonly invoked to explain the properties of extended emission line
regions in AGN, for example the Seyfert galaxy NGC 1068 (Axon 1996);
radio galaxies 3C 254 (Crawford 1996) and 3C368 (Stockton, Ridgway \&
Kellog 1996); and Cen A (Sutherland, Bicknell \& Dopita 1993), as well as
the sources mentioned in the previous paragraph.
The most luminous radio sources are known from observations to reside in
regions of high galaxy density (Yates, Miller \& Peacock 1989, Hill \& Lilly
1990). X-ray observations of powerful radio galaxies and quasars show that many
lie in the centres of rich clusters with dense, rapidly cooling IGM in which
cold clouds can condense (Fabian 1993). Cowie et al. (1983) observed
filamentary line emission in cooling flows that could indicate the presence of
such clouds. Observations suggest that these clouds have temperatures of about
10$^4$ K, densities of about 100 cm$^{-3}$ and sizes of 3--15kpc (Baum 1992).
Some contribution from radio jets seems necessary to re-energize the filaments.
\section{Simulating collisions of jets and clouds}
\subsection{Previous simulations}
Analytical studies show that sharp bends of 90$^{\circ}$ are possible
if the jet is thin or only moderately supersonic (Icke 1991). More
recently Raga \& Canto (1996) have published analytical calculations
and two-dimensional simulations showing bending by clouds. They
conclude that slower jets will be bent more, and clouds will be eroded
as jets bore through them.
The first investigation of the effect of off-axis jet-cloud collisions
was by De Young (1991) using the `beam scheme' (Sanders \& Prendergast
1974), to test the proposal of Burns (1986) that bending in large WATs
is the result of collisions with clouds. De Young monitored the jet
flow using test particles, and observed that the jet was considerably
decelerated by the impact with the cloud. The cloud was destroyed
within a few million years, and the jet returned to its original
direction. He concluded that the interaction does not last long
enough to produce anything like a tail.
However, WAT tails have a wide opening angle and show no strong evidence of
supersonic speeds (such as terminal hotspots). It would appear the jet is
disrupted at the impact point anyway. The only question is whether the
interaction can be maintained long enough to produce tails of the observed
length.
A similar interaction was investigated at higher resolution by Balsara \&
Norman using their RIEMANN code (Norman 1993). They argued from plots of the
velocity field that a De Laval nozzle was formed which re-accelerated the
jet in a new direction after impact.
We aimed to resolve this conflict through the work described in this
paper, and to more thoroughly investigate the effect of
different parameters both on the development of the interaction over
time, and the structures produced, and by estimating the appearance of
the source at radio wavelengths for direct comparison with
observations.
\subsection {Numerical techniques}
The simulations we present in this paper were performed using a
hydrodynamic code based on the Godunov method of Falle (1991) in
three-dimensional Cartesian coordinates. This technique solves the inviscid
Euler equations to second-order accuracy in space and time, with an
adiabatic equation of state. In addition to calculating the usual dynamical
variables, we also calculate a parameter representing the fraction of
density within each cell which was originally jet material. This allows us
to follow the evolution of the jet separately from the ambient medium and
to calculate synthetic radio maps as described in section \ref{synchro}.
Since we can draw a plane of symmetry bisecting the cloud and containing
the jet we have only calculated one half of the region around the
interaction (figure \ref{symfig}). The boundaries of this domain are
treated as free flow except for the symmetry plane and the region where the
jet enters the grid. A free flow condition assumes that values on the outer
surfaces of each cell are exactly equal to those on the inner surfaces. The
symmetry condition is that velocities normal to the surface are reflected.
The jet is produced by using an appropriate boundary condition representing
incoming material in the region where the jet enters. Note we have made no
assumptions about the position of the central engine with respect to the
grid.
The simulations can be rescaled so that they represent structures on parsec
or kiloparsec scales, as long as the gas behaves adiabatically. The
simplest rescaling is to change sizes and times in proportion, preserving
all other variables. For example, if the cell side is 1 kpc, then 1 time
unit = 2$\times 10^6$years. We use this scaling as a reference in
discussing the simulations below. We can apply these simulations to other
cases with different pressures and temperatures. For example, in the case
of the fastest jets, the temperature may be a hundred times higher, or for
the slow jets a hundred times lower, and we still have velocities in the
range accepted for extragalactic jets.
\subsection{Testing the hydrodynamic code}
We used the code to calculate two one-dimensional test problems. The
first was Sod's shock tube problem (Sod 1978). The code was used to
compute plane shocks moving along each of the three axes of the grid
(in both forward and reverse directions, as well as at various
resolutions). It produced results in good agreement with the analytical
solution.
The second test problem was the collision of a one-dimensional shock with a
density discontinuity (Nittman, Falle \& Gaskell 1982). This problem is
clearly appropriate to our investigation as a useful indication of the
fidelity of the code in this case. Once again this was run with shocks and
discontinuities normal to each of the three axes, moving both forwards and in
reverse directions and at various resolutions. Figure \ref{test1} shows an
example of the results (full details are given in the caption to the figure).
The positions and values of the two shocks were accurately reproduced.
As a final test we re-calculated a portion of one of the simulations
(simulation 3 -- see section \ref{sumres}) at double the initial
resolution within a volume one-eighth of the size and compared the
results. When the high-resolution results were smoothed to the lower
resolution they showed no significant differences. Nor did there
appear to be any effects caused by allowing the simulated flow to
leave this grid compared to the same region of the lower resolution
simulation. In figure \ref{hirestest} we show density slices at the
same time for the simulations at both resolutions. We are satisfied
from this that the results of our simulations are not significantly
affected by the resolution or boundary effects.
\subsection{Producing synthetic radio maps}
\label{synchro}
Previous studies of jet-cloud collisions relied on the interpretation of
flow patterns produced by the simulations to reach their conclusions. To
allow a more direct comparison we have developed a simple approximation
for the intensity of synchrotron emission in terms of the hydrodynamic
variables. We use this to produce estimates of the surface brightness
distribution of radio emission. These synthetic radio maps can be
compared with observations. In this section we describe our
prescription, and the assumptions it is based on. We then consider how
changing these assumptions might affect the results, and present some
plots for alternative prescriptions.
The intensity $j$ of synchrotron emission at frequency $\nu$ is given by
$j \propto K B^{1+ \alpha}\nu^{-\alpha}$, where $B$ is the magnetic
field strength, $\alpha$ is the spectral index and $K$ is related to
the number density $N(\gamma)$ of relativistic electrons with Lorentz
factors in the range ($\gamma$,$\gamma+d\gamma$) via $N(\gamma)d\gamma =
K \gamma^{-(2\alpha+1)}d\gamma$. In order to calculate $j$ we need to
express the magnetic field and the coefficent $K$ in terms of the
results of our hydrodynamic simulations.
To determine the magnetic field we assume that the field is
insignificant outside the jet (see, for instance, Smith et al. 1985) and the
magnetic flux is frozen into the jet material. By conservation of magnetic
flux we would expect the field strength to be related to the jet density
via $B \propto \rho_{jet}^{2/3}$.
In practise a complicated flow can amplify an initially disordered
field. For the purposes of this work we have neglected this amplification.
To determine the coefficient $K$ we assume that the energy density of
relativistic particles is a fixed fraction of the internal energy
density of the gas with the only changes in the distribution being due
to adiabatic expansion. This is equivalent to assuming that the
relativistic electrons form a supra-thermal tail throughout the gas,
and the efficiency of the acceleration process is similar
everywhere. Following Wilson and Scheuer (1983) we can relate $K$ to
the gas pressure $p$ via $ K \propto p^{(\alpha/2+3/4)}$. The
adiabatic index varies between 5/3 for non-relativistic gas to 4/3 for
relativistic particles which corresponds to an uncertainty of about
0.3 in our choice of $\alpha$.
We substitute for $K$ and $B$ from the formulae above, and therefore
obtain
\begin{equation}
j\nu^\alpha\,\propto\,p^{\alpha/2+3/4}\rho_{jet}^{2\alpha/3+2/3}\
\end{equation}
With $\alpha$=0.5 we have $j\nu^\alpha \propto p \rho_{jet}$, whereas
with $\alpha = 1.0$ we find $j\nu^\alpha \propto
p^{5/4}\rho_{jet}^{4/3}$.
As an alternative to assuming that magnetic flux is frozen into the jet
and that it is zero elsewhere we could assume that the magnetic energy
density is in equipartition with the internal energy density of the gas
(and, because of our other assumption, with the relativistic particle
energy density). This leads to $B\propto p^{1/2}$ and therefore
\begin{equation}
j\nu^\alpha\,\propto\,p^{\alpha+5/4}\
\end{equation}
It is common to use $\alpha$=0.75 so that $j\nu^\alpha \propto p^2$.
Further, if one still believed that the magnetic field was negligible
outside the jet then we could use this expression with one's favourite
$\alpha$ wherever $\rho_{jet}\neq 0$ and set the intensity equal to zero
where $\rho_{jet}=0$.
We can calculate the synchrotron intensity for each computational cell and,
assuming that the emission is optically thin, integrate along lines of
sight to produce a synthetic radio map at any epoch. In figure
\ref{rad_comp} (a -- d) we show four alternative maps from the same fluid
conditions based on the alternatives described above (see caption). Using
the square of total pressure, the hotspot is large with a shape like a
tadpole, but neither the jet nor the deflected tail are detectable.
However, the evidence for higher magnetic fields in the jet is strong.
The other alternatives produce
structures that are similar to each other, so that the general comparisons
we would like to make are not seriously effected.
\section{The effect of jet and cloud properties on the results of the
interaction}
\subsection{Models and coverage of parameter space}
The parameter space of jet cloud collisions is multi-dimensional. In an
attempt to explore the range of behaviour within this space we have
chosen to vary the values of three of the most significant parameters
over ranges applicable to extragalactic jets: the jet Mach number, its
density contrast with the ambient medium, and the contrast between cloud
and ambient density. Using one of these cases, we have also investigated
the effect of impact angle and relative size of cloud and jet on one of
the more interesting cases. Details are given in table \ref{pstbl}.
We have assumed conditions in the ambient medium consistent with
observations, that is a temperature of 5$\times 10^7$K and a particle
number density of 0.01 cm$^{-3}$. These values are used to normalize
the quantities in the computation so that model values for the ambient
density and pressure in the code are set to 1.0. The jet and cloud are
both taken to be in pressure balance with the ambient medium. The
computational domain is divided into a grid of 50 $\times$ 120
$\times$ 120 cells.
\subsection{Summary of results}
\label{sumres}
By simply changing a few parameters we have produced a range of different
structures. We discuss these in this section. The common features in all
the simulations are: the jet is disrupted and decelerated to some extent on
collision; the cloud is eroded by the jet; high pressures and densities are
produced at the point of impact, giving rise to bright hotspots in the
radio emission. All these structures vary in time, and many might be
associated with features observed in real radio sources at different epochs
in the simulation (Higgins, O'Brien, \& Dunlop, 1995).
We illustrate this discussion by presenting some examples. In each
case we show total density beside a synthetic radio map at two epochs.
Density plots show a slice through the symmetry plane (see figure
\ref{symfig}), rendered in a logarithmic greyscale. The synthetic
radio maps are produced by estimating the intensity of synchrotron
emission in each cell (using the formula $j\nu^\alpha \propto p
\rho_{jet}$, as described in section \ref{synchro}). This is then
summed along the line-of-sight perpendicular to the symmetry
plane. The resulting maps are shown on a logarithmic greyscale with a
range of about one order of magnitude. They have very different peak
values, so the hostpots are not directly comparable, but the dynamic
range between the peaks and faintest features are. The faster jets
have left the grid in all cases before the slower jets have developed
any interesting features, so we show each jet shortly after impact,
and then at a later stage of the interaction. For the fast jet this
is $t=$4 units of computational time after it enters the grid (16
million years on the scaling specified above) and $t=$28 (56 million
years). The slower jets are also shown at $t=$28, which is around the
time of impact, and then at $t=$76 (112 million years).
The light, slow jet (simulation 1, figure \ref{jet1}) is scattered into
a very broad area in all directions perpendicular to the jet. A
mushroom or umbrella shaped structure is formed in the radio
emission. The hotspot is no brighter than it was in the ambient medium
before impact, and is recessed a little within the lobe. It is not
obvious from the radio structure that this is a cloud collision.
When the same jet encounters a denser cloud (simulation 2) the interaction
lasts longer, since the jet erodes the cloud much more slowly. The
denser cloud makes no difference to the hotspot behaviour, which is
not significantly brighter on impact with the cloud. This feature
almost certainly depends on the relative sizes of the jet and cloud: a
much larger cloud would not allow deflection in both directions.
The fast, light jet (simulations 3 and 4, figure \ref{jet4}) produces
a weak secondary hotspot at the head of the deflected jet in the radio
map. This is about a tenth as bright as the spot at the impact
point. It becomes disconnected (on the dynamic range of our radio
plots) and disappears within about fifty thousand years, and the jet
rapidly erodes the cloud (whatever its density). As the jet breaks
through the cloud there are two hotspots within a boot-shaped lobe.
Although the cloud impact is the cause of the bending, the deflection
and secondary hotspot is actually produced as the jet bends inside the
distorted cocoon that has been formed during the interaction.
After a few hundred thousand years it has broken
through the cloud and shot off the grid. This faster jet can be
deflected, and produces much more visible results, but only for a
short time. In this case the hotspot shows a small increase in
brightness (about 10~--~20\%) on impact, and remains about the same
throughout the interaction.
The density of the cloud does not make significant
difference to the effect on the jet
in either of these cases.
The slow, heavy jet (simulations 5 and 6, figure \ref{jet7}) shows a
clear deflection by 90$^{\circ}$. The jet forms a bowl-shaped cavity
in the cloud which deflects the jet with a relatively large radius of
curvature. We do not see any secondary hotspot in the radio map. The
primary hotspot brightens steadily after impact, up to a factor
between two and three. The jet erodes the cloud slowly. On the scale
of a WAT source it takes about $3 \times 10^8$ years to erode the less
dense cloud. There is a hotspot down the jet before the impact point.
When the cloud is denser the interaction does of course last
longer. The deflection angle is nearer to 70$^{\circ}$. After about
6$\times 10^8$ years the deflected arm is about four jet radii long,
and about half as wide again as the incoming jet. The jet has worked
about half way through the cloud. The shape of the deflecting face
formed by the interaction is now deeper, so disrupting the jet
more. However the interaction can certainly continue for some time
yet.
The fast, heavy jet (simulations 7 and 8) pushes deep into the cloud,
however dense it may be. The impact hotspot is orders of magnitude
brighter than any other features. The jet breaks through the cloud
within about 50 million years. The radio map shows an extremely bright
deflection spot and weaker secondary (by 1000 times) in the deflected
material. The primary hotspot brightens by a factor of ten on impact
in this powerful jet. Because this interaction is short-lived, and the
hotspot far outshines any other emission, it would be difficult to
detect any deflection in a source with this powerful a jet.
\subsection{The Effect of Impact Angle}
We used the same fluid parameters as in simulation 4, altering only
the angle at which the jet encountered the cloud. With an impact angle
of 45$^{\circ}$ this produces a distinct secondary at the head about
one jet radius away. However, it is shorter lived ($\sim$ 40
million years) and the angle of deflection decreases quickly as the
jet erodes the cloud. With an impact angle of 30$^{\circ}$ the
deflection is shallower ($\sim 70^{\circ}$) and shorter
lived. Shallower impact angles can produce secondary hotspots since
less momentum is lost by the jet, but the interaction is short lived
on the whole. An impact angle of 45$^{\circ}$ is probably the lower
limit for a deflection of 90$^{\circ}$.
\subsection{An Example of a Jet-Cloud Interaction}
Our sixth simulation ($M$=2, $\eta$=0.2, figure \ref{jet7}) shows all
the major features of WAT jets: bent sharply through 90$^{\circ}$ at
bright hotspots and flaring out into long, wide tails (O'Donoghue,
Eilek \& Owen 1990). Due to the higher cloud density the interaction
is long-lived. It lasts for something of the order of 10$^8$ years,
which allows the formation of a tail not only large enough but of an
age consistent with estimates of travel time to the end of the tails
from synchrotron spectral-aging (O'Donoghue, Owen \& Eilek 1993).
We chose this simulation to examine the dynamics of
the interaction in more detail. In particular we look at the
distribution of material and its velocity in three dimensions, along with
the estimated
radio emission, and how it varies with viewing angle. We also give some
indications of the likely location and shape of line emission.
Figure \ref{jet7_1} shows velocity vectors plotted in three dimensions, with
the initial jet direction emerging from the plane of the diagram. This is at a
later time than shown in figure \ref{jet7} (3$\times 10^8$years). These show
how the material is deflected after impact into a fan of opening angle between
60$^{\circ}$~--~80$^{\circ}$. None of this material is moving slower than 20\%
of the speed of the incoming jet. A spine of slightly faster material can be
discerned along the middle of this fan, representing the remnant of the
jet. The wings are material deflected at higher speed on impact; as the
interaction progresses the impact cavity deepens and material deflected in this
direction at later times has slightly lower speeds that fall below the
threshold of this plot. Clearly on deflection the jet is not only decelerated
but almost all collimation is lost. However, this fan is very flat, allowing
the appearance of a deflected tail when viewed from the side.
The next plots (figure \ref{rad}) show the expected radio emission, viewed from
two orientations. Since the fan is fairly thin this gives a
reasonable impression of a bent jet from the side (a), with the tail
showing many similarities to WATs, including filamentary structure.
This seems to be due to vortex action in the tail. In (b) we
see how this radio structure would appear were it not close to the
plane of the sky. The emission from the deflected fan now surrounds
the whole jet. This emission should not be as bright as the previous
case since that had longer path lengths through the emitting
material. Thus it should be more difficult to identify the
counterparts of these objects at different orientations.
\subsection{Line emission}
The compression and distortion of the cloud would be likely to result in
enhanced line emission, due to photoionization by continuum from a hidden
quasar (if not limited by the flux from the nucleus) or shock heating. The
simulations show elongated density enhancements beside the jet, especially
the transmitted shock driven into the cloud. As a crude indicator of the
expected position and distribution of line emission, we have plotted the
distribution of density squared (emission measure) integrated through the
grid in figure \ref{em}. This is overlayed in the figure with contours of
radio emission. Note the bright, elongated region inside the cloud, forming a
cap around the impact point. Compact regions of line emission are seen in HST
images (for example 4C41.17, van Breugel 1996) which we interpret in this
model as shock fronts moving away from the jet, providing density
enhancements for scattering of or photo-ionization by the AGN continuum, or
shock-ionization. This gives a natural explanation for the radio-optical
alignment effect.
\subsection{The Effect of Several clouds}
In realistic situations there are likely to be many clouds, so we have
simulated the passage of a jet through a medium containing an ensemble of
clouds. Figure \ref{multi} shows a three-dimensional rendering of total density and
a synthetic radio map at the same epoch. As the jet progresses through the
grid it collides with clouds, producing prominent hotspots in the radio
emission. These spots persist as the jet moves past the cloud and encounters
further obstructions. They fade as the clouds are eroded by the passage of
the jet. Meanwhile new hotspots form at new encounters, and deflected jet
material percolates through the ambient medium, producing filamentary and
foamy shock structures. The result is a jet that is made visible by a series
of irregular knots, with a crooked ridgeline, filamentary diffuse bridges
and lobes and multiple hotspots at the head of the jet.
It clearly remains collimated and produces
a bow-shock at its head.
\section{Conditions for the deflection of jets}
Synchrotron emission may not trace the fluid flow in an extragalactic
jet in a simple way. Bends in the radio structure may not represent
bends in the flow. Our synthetic radio maps allow us to make more
general comparisons between our simulations and observations. De Young
and Balsara and Norman based their conclusions solely on the flow
patterns. Our results show how a few changes in parameter values can
alter the results of a collision. Under our assumptions, it is
possible to produce structures reminiscent of those seen in a variety
of radio sources. This also explains the apparent conflict between
previous simulations. The simulations performed by De Young involved a
fast, heavy jet (about Mach 25, with a density contrast of $\eta$=1)
so it is not surprising that it ripped up the clouds. In contrast,
Balsara and Norman's jet was of a moderate speed and lightness (Mach 4
and a density contrast of $\eta$=0.2). As we have shown it is easier
to produce deflection under these conditions. The conclusions clearly
depend on the choice of parameter values.
The impact causes a large increase in radio luminosity of fast jets,
which can still be seen after tens of millions of years. Samples of
ultra-luminous radio sources may contain a large proportion of sources
in which jets have been in collision with clouds within this time in
their past, even if the probability of a collision is fairly small for
any source. The technique of ultra-steep-spectrum selection used to
locate luminous sources at high redshift (Rottgering 1992) could make
this bias even stronger.
The range of structures we have produced suggest that deflection may
be easier to produce or detect in lower power jets propagating in the plane
of the sky. We attempt to model sources in more detail in future
papers, but note that in the case of WATs these do not seem very
demanding requirements: the sizes are already very large, so we would
not expect any other orientation. When such sources are seen at some
other orientation the lobe emission may be too diffuse to
detect. However, our simulations use spherical clouds, which are
clearly an idealised case. Other shapes or clouds with some density
variation may produce a better degree of collimation in the deflected
jet. This might allow deflected structures to be observed from a wider
range of angles. Larger clouds may sustain the deflection of high
power jets long enough to produce a deflected arm.
Steffen et al.\ 1997 have studied the interaction of jets with clouds
in the narrow-line region of Seyfert galaxies through two-dimensional,
non-adiabatic simulations. These suggest that only clouds above a
critical density will radiate after being shocked by the impact of the
jet. Thus the absence of detectable emission associated with bends may
be due to a low cloud density.
We have also tentatively associated our other simulations with bent
structure. We discuss this in detail elsewhere. In particular we find
that different structures may be produced by a single set of
parameters as the interaction progresses (Higgins, O'Brien \& Dunlop
1995, Dunlop 1995). Other sources can also be modelled by these
simulations (with rescaling when appropriate). Our studies show that
shallower impact angles can produce secondary hotspots since less
momentum is lost by the jet, although these are short-lived. Thus
compact sources seem to be best modelled with more oblique impacts
and/or several clouds (see figure \ref{jet4}).
\subsection{Further Development}
In conclusion, given the right sets of conditions collisions between
jets and clouds can reproduce some of the distorted structure seen in
observations, and is suggestive of alignments between radio and
optical axes. A more detailed understanding of the physics involved
could allow us to infer the properties of jets and their environment.
The radio maps presented here assume a fixed spectral index throughout
the source. A more sophisticated prescription in which this could vary
would allow us to investigate the variation of the appearance of a
source with frequency. It should also be possible to include
calculation of a few common lines, and their spatial distribution.
These calculations are non-relativistic. Relativistic simulations of
jets show no gross differences to the structures seen in
non-relativistic simulations. The most significant difference is in
the effect of Doppler beaming, which should not apply in our case
since the lobes are not beamed. Recent calculations suggest that
relativistic jets may lose less kinetic energy through entrainment
of ambient material (Bowman, Komissarov \& Leahy 1996). If this was the
case in our model it might allow better collimation after deflection.
We intend to make relativistic calculations to explore
this possibility.
\newpage
\section*{Acknowledgments}
SWH acknowledges the PPARC for receipt of a studentship, and his wife,
Leah, and parents for additional financial subsidies. Computing was
performed using the Liverpool John Moores University Starlink node. We
would like to thank Dr Sam Falle for useful suggestions, and Dr Huw
Lloyd and Dr John Porter for valuable discussions and the referee for pointing
out important considerations.
\newpage
\section*{References}
\parindent=0pt
Axon, D., 1996, in Clark, N.E., ed., Jet-cloud Interactions in
Extragalactic Nuclei, published on the World Wide Web, \vspace{-18pt}
\begin{verbatim}
http://www.shef.ac.uk/~phys/research/astro/conf/
index.html
\end{verbatim}
\vspace{-8pt}
Barthel, P.D., Miley, G.K., Schilizzi, R.T., Lonsdale, C.J., 1988,
A\&AS, 73, 515
Baum, S.A., 1992, in Fabian, A.C., ed., NATO ASI Ser., vol. 366,
Clusters and Superclusters of Galaxies, Kluwer, Dordrecht, p. 171
Baum, S.A., Heckman, T.M., 1989, ApJ, 336, 681
Bridle, A.H., Perley, R.A., 1984, ARA\&A, 22, 319
Bowman, M., Leahy, J.P., Komissarov, S.S., 1996, MNRAS, 279, 899
Burns, J.O., 1986, Can.J.P., 64, 363
Burns, J.O., O'Dea, C.P., Gregory, S.A., Balonek, T.J.,
1986, ApJ, 307, 73
Chambers, K.C., Miley, G.K., van Breugel, W.J.M., 1990, ApJ, 363, 21
Chambers, K.C., Miley, G.K., van Breugel, W.J.M., 1987,
Nat, 329, 624
Cowie, L.L., Hu, E.M., Jenkins, E.B., York, D.G.,
1983, ApJ, 272, 29
Clark, N.E., Tadhunter, C.N., 1996, in Carilli, C.L., Harris, D.E.,
eds., Cygnus A~--~Study of a Radio Galaxy, CUP, Cambridge, p. 15
Crawford, C.S., Vanderriest, C., 1996, MNRAS, 285, 580
De Young, D.S., 1991, ApJ, 371, 69
Dunlop, J.S., 1995, in Hippelein, H., Meisenheimer, K., eds.,
Galaxies in the Young Universe, Lecture Notes in Physics vol. 463,
Springer-Verlag, Berlin
Dunlop, J.S., Peacock, J.A., 1993, MNRAS, 263, 936
Fabian, A.C., 1994, ARA\&A, 32, 277
Falle, S.A.E.G., 1991, MNRAS, 250, 581
Higgins, S.W., O'Brien, T.J., Dunlop, J.S., 1996, in
Ekers, R., Fanti, C., Padrielli, L.,
Extragalactic Radio Sources, IAU Symp. 175, Kluwer, Dordrecht, 467
Higgins, S.W., O'Brien, T.J., Dunlop, J.S., 1995, in Millar, T.J.,
Raga, A., eds, Shocks in Astrophysics, Kluwer, Dordrecht, p. 311
Hill, G.J., Lilly, S.J., 1990 ApJ, 367, 1
Icke, V., 1991, in Hughes, P.A., ed,
Beams and Jets in Astrophysics, CUP, Cambridge, p. 232
Koide, S., Sakai, J-I., Nishikawa, K-I., Mutel, R. L., 1996, ApJ, 464, 724
Lacy, M., Rawlings, S., 1994, MNRAS, 270, 431
Liu, R., Pooley, G., 1991, MNRAS, 249, 343
Leahy, J.P., 1984, MNRAS, 208, 323
Leahy, J.P., Muxlow, T.W.B., Stephens, P.W. 1989, MNRAS, 239, 401
Loken, C., Roettiger, K., Burns, J.O. Norman, M., 1995, ApJ, 445, 80L
Lonsdale, C.J., Barthel, P.D., 1986, ApJ, 303, 617
McCarthy, P.J., 1993, ARA\&A, 31, 639
McCarthy, P.J., van Breugel, W., Kapahi, V.K., 1991,
ApJ, 371, 478
Meisenheimer, K., Hippelein, H., 1992, A\&A, 264, 455
Muxlow, T.W.B, Garrington, S.T., 1991, in Hughes, P.A., ed,
Beams and Jets in Astrophysics, CUP, Cambridge, p. 51
Nittman, J., Falle, S.A.E.G. and Gaskell, P.H.,
1982, MNRAS, 201, 833
Norman, M.L., 1993, in Burgarella, D., Livio, M., O'Dea, C., eds,
STScI Symp. Ser. Vol. 6, Astrophysical Jets, CUP, Cambridge, p. 211
Owen, F.N., O'Dea, C.P., Keel, W.C., 1990, ApJ, 352, 44
O'Donoghue, A.A., Eilek, J.A., Owen, F.N., 1993, ApJ, 408, 428
O'Donoghue, A.A., Owen, F.N., Eilek, J.A., 1990, ApJS, 72, 75
Pedelty, J. A., Rudnick, L., McCarthy, P.J., Spinrad, H.,
1989, AJ, 97, 647
Pinckney, J., Burns, J.O., Hill, J.M., 1994, AJ, 108, 2031
Raga, A.C., Canto, J., 1996, MNRAS, 280, 567
Rector, T.A., Stocke, J.T., Ellingson, E., 1995, AJ, 110, 1492
Rottgering, H., 1993. {\it PhD Thesis}, University of Leiden
Saikia, D.J., Jeyakumar, S., Wiita, P.J., Sanghera, H.,
Spencer, R.E., 1995, MNRAS, 276, 1215
Sakelliou, I., Merryfield, M.R., McHardy, I.M., 1996, MNRAS, 283, 673
Sanders, R.H., Prendergast, K.H., 1974, ApJ, 188, 489
Scheuer, P.A.G., 1982, in Heeschen, D.S., Wade, C.M., eds,
Extragalactic Radio Sources, IAU Symp. 97, Reidel, Dordrecht, p. 163
Smith, M., Norman, M.L., Winkler, K-H.A., Smarr, L.,
1985, MNRAS, 214, 67
Sod, Gary A., 1978, J.Comp.Phys., 27, 1
Steffen, W.S., G\'{o}mez, J.L., Raga, A.C.\ \& Williams, R.J.R, 1997, ApJ, L73
Stocke, J.T., Burns, J.O., Christiansen, W.A., 1985, ApJ, 299, 799
Stockton, A., Ridgway, S., Kellog, M., 1996, AJ, 112, 902
Sutherland, R.S., Bicknell, G.V., Dopita, M.A., 1993, ApJ, 414, 510
Tadhunter, C.N., Clark, N., Shaw, M.A., Morganti, R., 1994
A\&A, 288, L21
Tadhunter, C.N., 1996, in Carilli, C.L., Harris, D.E.,
Cygnus~A~--~Study~of a Radio Galaxy, CUP, Cambridge, p. 33
van Breugel, W.J.M., Heckman, T.M., George, K., Filippenko,
A.V., 1986, ApJ, 311, 58
van Breugel, W.J.M., 1996, in Ekers, R., Fanti, C., Padrielli,
L., Extragalactic Radio Sources, IAU Symp. 175, Kluwer, Dordrecht, 577
Williams, A.G., Gull, S.F., 1985, Nat, 313, 34
Wilson, Andrew S., 1993, in Burgarella, D., Livio, M., O'Dea, C., eds,
STScI Symp. Ser. Vol. 6, Astrophysical Jets, CUP, Cambridge, p. 122
Wilson, M.J., Scheuer, P.A.G., 1983, MNRAS, 205, 449
Yates, M.G., Miller, L., Peacock, J.A., 1989, MNRAS, 240, 129
\newpage
\section*{Figure Captions}
Figure 1: The geometry of the computational grid. This shows the
symmetry plane containing the jet axis (arrow) and bisecting the cloud
(hemisphere). It
also shows the viewpoint used for the density slices in figures
\ref{hirestest} to \ref{jet7}, the $z$-axis.
Figure 2: Example of the results of a test
using a one-dimensional problem involving the impact of a shock on a
density discontinuity. The solid line represents the analytical
solution, the points represent the results of the simulation. The
parameter values are: `jet' (shock) Mach number 2.0, pressure 1.0 and
density 1.0; `cloud' (discontinuity) pressure 1.0 and density 200.
Figure 3: Logarithmic plots (on the same greyscale) of density in the
symmetry plane for the same jet parameters at the same time run at two
different resolutions. The upper panel was produced from a simulation
run at twice the resolution of the second, then smoothed with a 3-D
gaussian and binned down to the same pixel size.
Figure 4: Alternative plots of synchrotron surface brightness distribution,
using a logarithmic grey scale. Panels {\em a)} and {\em b)} show the effect
of changing $\alpha$ on our prescription. Panel {\em a)}~shows the effect
of $\alpha$ = 0.5, which is the value we use in plots shown elsewhere
in this paper. Panel {\em b)}~shows the effect of $\alpha$ = 1.0: jet and
tail are fainter, so that the wider diffuse emission at the edges of
the tail is not seen. Panel {\em c)}~shows pressure squared only where
there is jet material. The jet and tail are as bright as in the
previous plots, but the diffuse emission around them is not seen due
to the negligible magnetic fields in these outer regions, producing a
narrower tail. Panel {\em d)}~shows the square of total pressure: this
is the effect of allowing the same magnetic field in the ambient
medium and the jet.
Figure \ref{jet1}: Logarithmic greyscale plots of total gas
density (in the symmetry plane)
and integrated radio emission form simulation 1 ($M=2,\eta_j=0.01,\eta_c=50.0$).
In the density plots white represents the highest value. For clarity in the radio
plots black represents the peak brightness. The plots are shown at two epochs from
the simulation: {\it a)} 56 million years after the jet enters the grid, and {\it
b)} 112 million years.
Figure \ref{jet4}: Logarithmic greyscale plots of total gas
density (in the symmetry plane)
and integrated radio emission for simulation 4 ($M=10, \eta=0.01, \eta_c=200.0$).
In the density plots white represents the highest value. For clarity in the radio
plots black represents the peak brightness.
The plots are shown at two epochs from the simulation: {\it a)} 16 million years
after the jet enters the grid, and {\it b)} 56 million years.
Figure \ref{jet7}: Logarithmic greyscale plots of total gas
density (in the symmetry plane)
and integrated radio emission for simulation 6 ($M=2, \eta_j=0.2, \eta_c=200.0$).
In the density plots white represents the highest value. For clarity in the radio
plots black represents the peak brightness. The plots are shown at two epochs
from the simulation: {\it a)} 56 million years after the jet enters the grid, and
{\it b)} 112 million years.
Figure \ref{jet7_1}: Three dimensional velocity vector plots of a
Mach 2 jet
with density contrast $\eta=0.2$, seen from
two orientations. The jet
emerges from the plane of the upper panel. Vectors are plotted if they
have a speed greater than 20\% of the speed of the incoming jet.
Figure \ref{rad}: Radio emission seen from two
orientations:{\em (a)}~perpendicular
to the symmetry plane (as in previous radio plots),
and {\em (b)}~rotated
so that the jet axis is about 30$^{\circ}$ from the line of sight.
Figure \ref{em}: Square of density integrated along the line of sight,
indicating
the expected site of line emission, overlaid with synthesized radio
contours corresponding to \ref{rad}~{\em a)}.
Figure \ref{multi}: A jet propagating through a collection of clouds.
This is a fast, light jet
($M$=10, $\eta$=0.01). All clouds have density contrast of 50. They are
distributed at random positions in the grid, with a fixed volume filling factor of
0.2 and a power law distribution of radius up to a fixed maximum size (1.4 times
the radius of the jet). There is no plane of symmetry in this problem, so we had
to calculate the whole grid, which in this case was 90 $\times$ 90 $\times$ 90.
This represents a volume of 729000 kpc$^3$. The plots show the results about four
million years after the jet entered the grid. {\em a)}~A constant density surface
of the jet inside a diffuse rendering of cloud density. {\em b)}~A
synthetic radio map. The jet has been deflected at least twice, where it has
encountered clouds, but clearly remains collimated.
\newpage
\onecolumn
\begin{table}
\begin{tabular}{lllll}
Simulation & Jet density & Cloud density & Jet mach & Jet \\
number & contrast & contrast & number & speed \\
\hline
1 & 0.01 & 50 & 2 & 0.07c \\
2 & 0.01 & 200 & 2 & 0.07c \\
3 & 0.01 & 50 & 10 & 0.36c \\
4 & 0.01 & 200 & 10 & 0.36c \\
5 & 0.2 & 50 & 2 & 0.02c \\
6 & 0.2 & 200 & 2 & 0.02c \\
7 & 0.2 & 50 & 10 & 0.08c \\
8 & 0.2 & 200 & 10 & 0.08c \\
9 & 0.01 & -- & 2 & 0.07c \\
10 & 0.01 & -- & 2 & 0.02c \\
11 & 0.2 & -- & 10 & 0.36c \\
12 & 0.2 & -- & 10 & 0.08c \\
\end{tabular}
\caption{The values of the parameters for twelve different simulations. The
jet density contrast is the ratio of the jet density to the ambient
density; the cloud density contrast is the ratio of the cloud density
to the ambient density; the jet mach number is the ratio of the jet
speed to the sound speed within the jet. There is no cloud in
simulations 9 -- 12.}
\label{pstbl}
\end{table}
\newpage
\begin{figure}
\vspace {18.4truecm}
\special{psfile=3djet.eps hscale=70 vscale=70 angle=270 voffset=300}
\vspace {0.5truecm}
\caption
}
\label{symfig}
\end{figure}
\begin{figure}
\vspace{20truecm}
\special{psfile=jet_tst_080_v_2.eps hoffset=-10
voffset=140 hscale=35 vscale=35 angle=270}
\special{psfile=jet_tst_080_p_2.eps hoffset=-10
voffset=350 hscale=35 vscale=35 angle=270}
\special{psfile=jet_tst_080_d_2.eps hoffset=-10
voffset=550 hscale=35 vscale=35 angle=270}
\vspace{2cm}
\caption{}
\label{test1}
\end{figure}
\begin{figure}
\vspace{17.5cm}
\special{psfile=jet2050_d-3d-l10.eps
hoffset=-100 voffset=500 hscale=50 vscale=50 angle=270}
\special{psfile=jet2050_d-l10.geps
hoffset=-100 voffset=250 hscale=50 vscale=50 angle=270}
\vspace{0.5cm}
\caption{}
\label{hirestest}
\end{figure}
\begin{figure}
a) \hspace{8.3cm} b)
\vspace{8cm}
\special{psfile=jet7140_rada-l10.eps
hoffset=-30 voffset=240 hscale=45 vscale=45 angle=270}
\special{psfile=jet7140_radb-l10.eps
hoffset=215 voffset=240 hscale=45 vscale=45 angle=270}
c) \hspace{8.3cm} d)
\vspace{8.0cm}
\special{psfile=jet7140_radc-l10.eps
hoffset=-30 voffset=240 hscale=45 vscale=45 angle=270}
\special{psfile=jet7140_radd-l10.eps
hoffset=215 voffset=240 hscale=45 vscale=45 angle=270}
\vspace{0.8cm}
\caption{}
\label{rad_comp}
\end{figure}
\newpage
\begin{figure}
a)
\vspace{10cm}
\special{psfile=jet1280_d-l10.ps
voffset=370 hoffset=254 hscale=50 vscale=50 angle=180}
\special{psfile=jet1280_rad5_l10.eps
hoffset=500 voffset=350 hscale=50 vscale=50 angle=180}
\vspace{0.5cm}
b)
\vspace{10cm}
\special{psfile=jet1760_d-l10.eps
voffset=370 hoffset=254 hscale=50 vscale=50 angle=180}
\special{psfile=jet1760_rad2-l10.eps
hoffset=500 voffset=350 hscale=50 vscale=50 angle=180}
\caption{}
\label{jet1}
\end{figure}
\begin{figure}
a)
\vspace{10.5cm}
\special{psfile=jet4080_d-l10.ps
voffset=380 hoffset=260 hscale=50 vscale=50 angle=180}
\special{psfile=jet4080_rad5-l10.ps
hoffset=500 voffset=360 hscale=50 vscale=50 angle=180}
\vspace{0.5cm}
b)
\vspace{10cm}
\special{psfile=jet4280_d-l10.ps
voffset=380 hoffset=260 hscale=50 vscale=50 angle=180}
\special{psfile=jet4280_rad5-l10.ps
hoffset=500 voffset=360 hscale=50 vscale=50 angle=180}
\caption{}
\label{jet4}
\end{figure}
\newpage
\begin{figure}
a)
\vspace{10cm}
\special{psfile=jet7280_d-l10.ps
voffset=370 hoffset=270 hscale=50 vscale=50 angle=180}
\special{psfile=jet7280_rad2-l10.ps
hoffset=500 voffset=350 hscale=50 vscale=50 angle=180}
\vspace{0.5cm}
b)
\vspace{10cm}
\special{psfile=jet7760_d-l10.eps
voffset=370 hoffset=270 hscale=50 vscale=50 angle=180}
\special{psfile=jet7760_rad2-l10.eps
hoffset=500 voffset=350 hscale=50 vscale=50 angle=180}
\caption{}
\label{jet7}
\end{figure}
\newpage
\begin{figure}
\vspace{23cm}
\special{psfile=jetvex_v2.ps
hoffset=-10 voffset=710 hscale=70 vscale=70 angle=270}
\special{psfile=jetvex_v1.ps
hoffset=-10 voffset=-50 hscale=70 vscale=70 angle=0}
\caption{}
\label{jet7_1}
\end{figure}
\newpage
\begin{figure}
\vspace{1cm}
a)
\vspace{10cm}
\special{psfile=jet7_160_rad1.ps hscale=60 vscale=60 voffset=-10 angle=90
hoffset=500}
b)
\vspace{11cm}
\special{psfile=jet7_160_rad2.ps hscale=60 vscale=60 voffset=-10 angle=90
hoffset=500}
\caption{}
\label{rad}
\end{figure}
\newpage
\begin{figure}
\vspace{12cm}
\special{psfile=rademol.ps hscale=67 vscale=67
angle=90 voffset=-30 hoffset=500}
\caption{}
\label{em}
\end{figure}
\begin{figure}
a)
\vspace{11cm} b)
\vspace{10cm}
\special{psfile=jet.ps hoffset=342 voffset=580 hscale=36 vscale=35 angle=180}
\special{psfile=jetrnd080_rad5.eps hoffset=85 voffset=290 hscale=56 vscale=50 angle=270}
\caption{}
\label{multi}
\end{figure}
\end{document}
|
\section{Introduction}
Magnetohydrodynamic (MHD) turbulence appears in many problems in nuclear fusion
research as well as in astrophysics and geophysics (see, e.g. Refs. 1 and 2
for reviews). Compared to the already complicated hydrodynamic turbulence there
are additional degrees of complexity due to the presence of a magnetic field,
the possibility of dynamos and the action of the magnetic field back on the
flow through the Lorentz force. In addition, the distribution of
energy among the modes is controlled through the interaction of
three cascade processes, related to the three conserved quantities
of the dynamical equations in the inviscid limit: the total energy, the cross
helicity, and for two dimensions the mean square magnetic potential.
There is, however, a regime where some of the more complicated processes are
subdominant or absent and where a rather complete analysis can be performed:
two-dimensional (2-D) magnetohydrodynamics with weak magnetic fields.
This regime can be relevant to astrophysical applications,
such as quasi-two-dimensional turbulent processes in thin accretion
discs.\cite{Kui95}
The situation we consider is that of a 2-D turbulent flow, maintained by some
external force, in which a magnetic field is advected. The linearity of the
equation for the magnetic induction implies that without an external seed field
no magnetic field will ever be generated. Moreover, the absence of dynamo
action in two dimensions \cite{Zel80} implies that without permanent driving the
magnetic field will die out eventually. The action of the magnetic field back
on the flow can be neglected if the Lorentz force is sufficiently weak, i.e., if
the magnetic field is sufficiently small. Because of the absence of dynamos in
2-D the amplitude of the external driving
controls the magnitude of the magnetic field.
We thus have a regime of kinematic evolution of
a magnetic field passively advected
by the turbulent flow which can be realized in any 2-D situation
for sufficiently small driving of the magnetic field.
The dynamics of the magnetic field and thus its correlation functions are
nevertheless nontrivial since a transient growth and the development of fractal
like structures on small scales are possible, as evidenced by numerical
simulations. \cite{Bis90,Cat91} Though the range of applicability of this
kinematic approach is limited, it shows some interesting features which we would
like to describe here.
On a technical level, the restriction to a 2-D situation allows to connect the
evolution of the vector potential to that of a scalar field for which detailed
scaling predictions exist within geometric measure
theory. \cite{Con93,Proc93,EckSch98}
On the assumptions of statistical stationarity, homogeneity, and isotropy for
both the magnetic field and the scalar field, geometric measure theory provides
a connection between the structure functions of the velocity field and the
scalar field. In particular, the scaling behavior can be related and the
dependence on the Prandtl number and on another parameter that characterizes the
strength of the driving and appears in 2-D only can be extracted. Using an
exact relation between the correlation function of the scalar and the magnetic
field we can transfer these results to the magnetohydrodynamic problem and in
particular discuss the dependence on the magnetic Prandtl number. The latter
can range from $Pr_m\sim 10^{-6}$ in the solar convection zone to $Pr_m\sim
10^{7}$ in the interstellar medium. \cite{Zel83} Within the kinematic regime we
are able to go beyond previous applications of geometric measure theory to MHD
turbulence. \cite{Bis93a,Gra95}
The outline of the paper is as follows. In sections II and III the theoretical
background in 2-D magnetohydrodynamics and 2-D passive scalar advection is
summarized. For details on the analysis of the 2-D passive scalar within
geometric measure theory we refer to our previous work. \cite{EckSch98} In
section IV both are combined and the correlation and structure functions of
magnetic fields in the kinematic regime are discussed. In section V we compare
with correlation functions and structure functions found in full numerical
simulations; this highlights features of the interaction between both fields
which have to be included when the magnetic field becomes dynamically relevant.
We conclude with a summary in section VI.
\section{MHD basics}
The equations for a magnetic field ${{\bf B}}({\bf x},t)$ in a conducting
fluid moving with velocity ${\bf u}({\bf x},t)$ are
\end{multicols}
\begin{eqnarray}
\rho\left(\partial_t {\bf u} + ({\bf u}\cdot\nabla){\bf u}\right) &=& -\nabla p +
\frac{1}{\mu_0}({\bf\nabla\times}{\bf B})\times{\bf B}
+\rho\nu\Delta{\bf u} + \rho{\bf f}_u \\
\partial_t {\bf B} + ({\bf u}\cdot\nabla){\bf B} &=& ({\bf B}\cdot\nabla){\bf u} +
\eta \Delta {\bf B} + {\bf f}_B \\
\nabla\cdot {\bf u} &=& \nabla \cdot {\bf B} = 0
\end{eqnarray}
\begin{multicols}{2}
where $\nu$ is the kinematic viscosity, $\eta$ is magnetic diffusivity,
$p$ is the thermal pressure, and
${\bf f}_u$ and ${\bf f}_B$ are the external driving for each field. The quantity $\rho$
is the constant mass density and $\mu_0$ denotes the permeability in a vacuum.
We estimate typical amplitudes for the velocity field and the magnetic field by
the root mean square values $u_{r.m.s.}=\langle u_x^2\rangle^{1/2}$ and
$B_{r.m.s.}=\langle B_x^2\rangle^{1/2}$ where $\langle\cdot\rangle$ denotes
a statistical
average over an isotropic and homogeneous ensemble.
If then $L$ is an external length
scale, the fluid Reynolds number $R=u_{r.m.s.} L/\nu$ and the magnetic Reynolds
number $R_m=u_{r.m.s.} L/\eta$ can be formed. Their ratio defines
the magnetic Prandtl number $Pr_m=\nu/\eta=R_m/R$. The relative size
of magnetic field and velocity field is measured by the
Alfv\'en-Mach number $M$, the ratio
of the typical fluid velocity $u_{r.m.s.}$ to the Alfv\'en velocity
$v_A=B_{r.m.s.}/(\mu_0\rho)^{1/2}$,
\begin{eqnarray}
M&=&\frac{u_{r.m.s.}}{v_A},\nonumber\\
&=& \left(\frac{\mu_0\rho u_{r.m.s.}^2}{B_{r.m.s.}^2}\right)^{1/2},\nonumber\\
&=&\left(\frac{E_K}{E_B}\right)^{1/2}\,.
\end{eqnarray}
By the last line this is also the square root of the ratio between
fluid kinetic energy and magnetic field energy.
With these definitions, the dimensionless form of
the first two equations reads
\end{multicols}
\begin{eqnarray}
\partial_t {\bf u} + ({\bf u}\cdot\nabla){\bf u} &=&
-\nabla p
+M^{-2}({\bf\nabla\times}{\bf B})\times{\bf B}
+R^{-1}\Delta{\bf u} +
{\bf f}_u \\
\partial_t {\bf B} + ({\bf u}\cdot\nabla){\bf B} &=& ({\bf B}\cdot\nabla){\bf u} +
R_m^{-1} \Delta {\bf B} + {\bf f}_B \,.
\end{eqnarray}
\begin{multicols}{2}
In two dimensions, both the velocity field and the magnetic field can be
represented by scalar fields by introducing the vorticity
\begin{equation}
\omega {\bf e}_z={\boldmath\nabla\times}{\bf u}\,,
\end{equation}
and the magnetic flux function
\begin{equation}
{\bf B}={\boldmath\nabla}\psi{\boldmath\times}{\bf e}_z
\,.
\end{equation}
The MHD equations then take on the form
\begin{eqnarray}
\label{nseq_nondim}
\partial_t\omega
+({\bf u}\cdot{\bf\nabla})\omega
&=&
M^{-2}({\bf B}\cdot{\bf\nabla})j
+R^{-1}{\bf\nabla}^2\omega
+f_{\omega}\,,\\
\label{indeq_nondim}
\partial_t\psi
+({\bf u}\cdot{\bf\nabla})\psi
&=&
R_m^{-1}{\bf\nabla}^2\psi
+f_{\psi} \,,
\end{eqnarray}
with $j=-\nabla^2\psi$ the
$z$-component of the current density
and $f_{\psi}$ and $f_\omega$ the scalar driving fields .
The situation we want to consider is one of weak magnetic fields, i.e. very
large $M$, so that the term with the Lorentz force density can be ignored.
This can be estimated to be reasonable\cite{Bis90} if the Mach number exceeds the
magnetic Reynolds number, i.e. if $M>R_m$.
Then the equations for the fluid do not depend on
the magnetic field any more and the statistically stationary state that develops
depends only on the strength of the force $f_{\omega}$ and the length scale
$l_f$ on which it acts. In 2-D turbulence it is the enstrophy that cascades
down to the viscous scales. So if we assume homogeneity and isotropy the
appropriate smallest scale is the enstrophy dissipation length
$\eta_{\omega}=\nu^{1/2}\epsilon_{\omega}^{-1/6}$, defined in terms of the
enstrophy dissipation rate
$\epsilon_{\omega}=\nu\langle|{\boldmath\nabla}\omega|^2\rangle$.
The quantities we want to study are the
second order structure function $D_2^{(B)}(r)$,
\begin{equation}
\label{definition_D2}
D_2^{(B)}(r)=\langle|{\bf B}({\bf x}+{\bf r})
-{\bf B}({\bf x})|^2\rangle \,,
\end{equation}
and the correlation function $C_2^{(B)}(r)$,
\begin{equation}
\label{definition_C2}
C_2^{(B)}(r) = \langle {\bf B}({\bf x}+{\bf r})\cdot
{\bf B}({\bf x})\rangle \,,
\end{equation}
of the magnetic field. $D_2^{(B)}(r)$ and
$C_2^{(B)}(r)$ are related by
\begin{equation}
\label{definition_DC}
D_2^{(B)}(r) = 2 \langle B^2 \rangle -2
C_2^{(B)}(r)\,.
\end{equation}
Continuity of the fields on small scales and decay
of correlations on large scales implies the following
limiting behavior for
small and large distances $r$:
\begin{equation}
\label{db_asympt}
D_2^{(B)}(r)=\left\{\matrix{\epsilon_B r^2/(2\eta)\rightarrow 0
&:\;\;r\rightarrow 0\cr
2\langle B^2 \rangle
&:\;\;r\rightarrow L\;,}\right.
\end{equation}
whereas it is the other way round for the correlation function,
\begin{equation}
C_2^{(B)}(r)=\left\{\matrix{2\langle B^2\rangle
&:\;\;r\rightarrow 0\cr
0
&:\;\; r\rightarrow L\;.}\right.
\end{equation}
Similar quantities can be defined for the velocity field and the magnetic
flux function $\psi$. The quantity
$\epsilon_B=\eta\langle|{\bf\nabla}{\bf B}|^2\rangle$ is the
magnetic energy dissipation rate. Since $\psi$ obeys in the kinematic limit
the equations of a passive scalar, we can use our previous results
for the
scaling of a passive scalar in a turbulent fluid to obtain the scaling regimes
for $\psi$. They can be transfered to the magnetic field case
by use of the relations\cite{GroMer92} (see Appendix A for the derivation)
\begin{eqnarray}
\label{laplace}
D_2^{(B)}(r)=2\frac{\epsilon_{\psi}}{\eta}-\Delta D_2^{(\psi)}(r),
\end{eqnarray}
and consequently with (\ref{definition_DC})
\begin{eqnarray}
\label{laplace1}
C_2^{(B)}(r)=\frac{1}{2}\Delta D_2^{(\psi)}(r).
\end{eqnarray}
The quantity $\epsilon_{\psi}=\eta\langle|{\bf\nabla}\psi|^2\rangle
=\eta\langle B^2\rangle$ is the mean flux
dissipation rate.
\section{2-D passive scalar advection}
\label{frame_geo}
In the limit of large $M$ the magnetic
flux $\psi$ does not act on the flow field ${\bf u}$,
so that (\ref{indeq_nondim}) describes the
passive advection of a scalar field in the velocity field that
follows from the hydrodynamic driving. A connection between the
scaling behavior of the fluid, as contained in the
velocity structure function,
\begin{equation}
D_2(r)=\langle|{\bf
u}({\bf x}+{\bf r})-{\bf u}({\bf x})|^2 \rangle\,,
\end{equation}
and its longitudinal part,
\begin{equation}
D_{\parallel}(r)=
\frac{1}{r^2}\int_0^r \rho \,D_2(\rho) \,\mbox{d}\,\rho\,,
\end{equation}
and the scaling behavior of the second order scalar field structure
function,
\begin{eqnarray}
D_2^{(\psi)}(r)&=&\langle|{\psi}({\bf x}+{\bf r})
-{\psi}({\bf x})|^2\rangle\,, \nonumber\\
&\propto& r^{\zeta_2^{(\psi)}}\,,
\end{eqnarray}
can be found within geometric measure theory.\cite{Con93,Proc93}
One estimates the
fractal dimension $\delta_g^{(1)}$ of level sets $\psi_0=\psi({\bf x})$ of the
scalar field graph $G=({\bf x},\psi({\bf x}))$, and obtains in the absence of
intermittency corrections, i.e. ${\zeta_2^{(\psi)}}=2 {\zeta_1^{(\psi)}}$,
the upper bound
\begin{equation}
\label{ineq1}
2 \delta_g^{(1)} \le 4 - {\zeta_2^{(\psi)}}\,.
\end{equation}
Building on Ref. 7 and extending the results of Ref. 8 we were able
to obtain scale resolved and Prandtl number dependent
upper bounds for $\delta_g^{(1)}$, which turned out to
be rather sharp.\cite{EckSch98}
If we assume that the bounds on $\delta_g^{(1)}$
are indeed equalities, (\ref{ineq1}) gives an
upper bound on $\zeta_2^{(\psi)}$, viz.
\begin{eqnarray}
\label{fracdim}
{\zeta_2^{(\psi)}} = 2-
2\frac{\mbox{d}}{\mbox{d}\,\ln r}
\ln\sqrt{1+\alpha\,Pr_m\,r^2+\sqrt{3} Pr_m r \sqrt{D_{\parallel}}}\,.
\end{eqnarray}
In the following we will evaluate the right hand side and in particular
its dependence on the scaling of the velocity field and take this
as an estimate for the scaling exponent $\zeta_2^{(\psi)}$. As mentioned,
this relies
on the assumption that several inequalites are in fact equalities
and that intermittency effects are absent. A careful analysis of the
derivation shows that the presence of intermittency in the scalar
field modifies several approximations and that the cumulative effect
is difficult to control. It is fairly straightforward, however,
to allow for intermittency effects in the velocity structure function.
The dimensionless parameter
$\alpha=\epsilon_{\psi}/(\epsilon_{\omega}^{1/3}\psi_{r.m.s.}^2)$ measures
the ratio of the scalar driving rate $\epsilon_{\psi}/\psi_{r.m.s.}^2$ to the
flow advection rate $\epsilon_{\omega}^{1/3}$ and is specific to two-dimensional
passive scalar advection. For large values of $\alpha$ the scalar cannot be
advected sufficiently rapidly to the small dissipative scales so that a space
filling distribution with $\zeta_2^{(\psi)}=0$ develops. The distance $r$ is
measured in units of $\eta_{\omega}$ and the longitudinal structure function
$D_{\parallel}(r)$ of the velocity field is measured in units of
$\epsilon_{\omega}^{2/3} \eta_{\omega}^2$. The explicit dependence on the
longitudinal velocity structure function $D_\parallel$ allows to substitute
numerical or experimental findings for the velocity structure function, which
very often differ from theoretical expectations, and to obtain predictions for
the corresponding scalar structure functions.
As in the previous work \cite{EckSch98} we use in particular structure functions
obtained from Fourier transforms of model spectra in $k$-space and evaluate
(\ref{fracdim}) numerically. Recent experiments on forced two-dimensional
turbulence\cite{Par97,Rut98} and a number of direct numerical
simulations\cite{FriSul84,SmiYak93,Ben86} support the existence of a
Kolmogorov-like scaling for the energy spectrum, $E(k)\sim\,k^{-5/3}$ in the
inverse energy inertial subrange $k<k_f$ and a scaling $E(k)\sim\,k^{-\beta}$
with $\beta\ge3$ for the direct enstrophy inertial subrange $k>k_f$.
Recent experiments on 2-D turbulence indicate that the velocity increments
in both cascade regimes show a non-intermittent Gaussian
behavior.\cite{Par97,Tab99}
We do not discuss here the role of coherent, large scale vortices which are
formed by the inverse energy cascade in two-dimensional hydrodynamic turbulence
and which can effect the spectra in the enstrophy subrange. This problem is
still a matter of debate. \cite{Bab87,Bas94} We therefore take for our analysis
the following model spectrum for the amplitudes $\langle|{\bf u}_{\bf
k}|^2\rangle$ of the velocity field in a Fourier representation in a periodic
box of size $L=2\pi$:
\end{multicols}
\begin{eqnarray}
\label{spectrum}
\langle|{\bf u}_{\bf k}|^2\rangle\!\sim\!
\left\{
\begin{array}{r@{\quad:\quad}l}
k_f^{-7/3} k_1^{-11/3} k^3 & \frac{2\pi}{L}\le k\le k_1, \\
k_f^{-7/3} k^{-2/3} & k_1<k\le k_f,\\
k^{-3} & k_f<k\le k_{\omega}=\frac{1}{\eta_{\omega}},\\
k^{-3}\exp\left[-\left(\frac{k-k_{\omega}}{k_{\omega}}\right)^{2}\right]
& k_{\omega}<k\;.
\end{array}
\right.
\label{modspec}
\end{eqnarray}
\begin{multicols}{2}
Note the differences in scaling between the amplitudes
$\langle|{\bf u}_{\bf k}|^2\rangle$ and the
energy spectrum $E(k)$ due to phase space factor, i.e. $E(k)\!\sim\!k^{-\beta-1}$
corresponds to $\langle|{\bf u}_{\bf k}|^2\rangle\!\sim\!k^{-\beta}$.
For a dependence of the
results on $\beta$ the reader is refered to our previous work,
where we showed that
a variation of $\beta$ in a range between 2 and 4 left the scaling of $D_2(r)$
in the enstrophy intertial subrange nearly unchanged.
We therefore take the exponent $\beta=3$ for
$\langle|{\bf u}_{\bf k}|^2\rangle$ [see Eq. (\ref{modspec})]
in all calculations here.
Assuming stationarity, homogeneity, and isotropy
the relation between velocity spectrum scaling and the velocity structure
function $D_2(r)$ is given by the volume average
\begin{eqnarray}
D_2(r)&=&\frac{1}{V}\int_{V}\,|\,{\bf u}({\bf x}+{\bf r})-{\bf u}({\bf x})\,|
^{2}\,\mbox{d}\!V,\nonumber \\
&=&\frac{1}{V}\int_{V}\,|\,\sum_{{\bf k}}{\bf u}_{\bf k}\,
\exp(i{\bf k}\cdot{\bf x})
[\exp(i{\bf k}\cdot{\bf r})-1]\,|^{2}\,\mbox{d}\!V,\nonumber \\
&=&2\,\sum_{{\bf k}}\,\langle|{\bf u}_{\bf k}|^{2}\rangle\,
(1-\cos{({\bf k}\cdot{\bf r})})\, .
\end{eqnarray}
By averaging over all directions (due to isotropy) in ${\bf k}$-space
the cosine gives rise to the Bessel function $\mbox{J}_0(kr)$,
\begin{equation}
\label{mod_struc}
D_2(r)=2\,\sum_k\,\langle|{\bf u}_{\bf k}|^{2}\rangle\,
(1-\mbox{J}_{0}(kr)) \,.
\end{equation}
The model spectrum (\ref{modspec}) is then substituted and the summation in
(\ref{mod_struc}) is evaluated numerically using a finite
set of wave numbers, equidistant on a logarithmic scale.
We checked the independence of our results for
several wavenumber resolutions .
The algebraic scaling of $D_2(r)$
with respect to $r$ is shown in Fig.
\ref{p1} (thick dash-dotted line). For the dissipation scales
$r<\eta_{\omega}$ it follows a $r^2$--dependence in
correspondence with the Taylor expansion. This quadratic scaling with respect
to $r$ continues in the enstrophy inertial subrange for scales
$\eta_{\omega}<r<l_f$.
For distances $l_f<r$ the energy inertial subrange with
a $r^{2/3}$--scaling sets in. Finally, a saturation to a constant value due
to finite size effects is observed.
\section{2-D kinematic MHD turbulence}
\label{Res}
Using the energy spectra (\ref{spectrum}) we can now evaluate the passive scalar
structure and correlation function exponents and, via (\ref{laplace}), the
corresponding behavior for the magnetic fields. The results for a large range
of length scales and magnetic
Prandtl numbers are shown in Figs. \ref{p1} and \ref{p2}.
The structure function is plotted in units of
$\epsilon_{\psi}\epsilon_{\omega}^{-1/3}$, and $\log_{10} D_2^{(\psi)}$ is obtained by
numerical integration of $\zeta_2^{(\psi)}(r)$ over $\log_{10} r$. We find in both
figures the following scaling regimes. For small scales $r<\eta_{\omega}$ the
fields are smooth and one finds $D_2^{(\psi)}(r)=(\epsilon_{\psi}/2\eta) r^2$
and thus $\zeta_2^{(\psi)}=2$. The quadratic scaling continues into the
enstrophy inertial subrange $r\gtrsim\eta_{\omega}$. For sufficiently large
$Pr_m$ the third term under the square root in Eq. (\ref{fracdim}) gives the
dominant contributions and the scaling exponent changes to $\zeta_2^{(\psi)}=0$;
this is the viscous--convective Batchelor regime\cite{Bat59}. The flux $\psi$
is then advected chaotically in a velocity field that is still smooth on these
scales, i.e. $D_{\parallel}\sim r^2$ and the flux contours are stretched and
twisted by the fluid motion and form filamented current sheets. The
$r^{2/3}$--scaling regime for larger $r$ and for both the velocity field and the
magnetic flux function is connected with the inverse cascade of the underlying
fluid turbulence; this is the inertial--convective Obukhov--Corrsin
regime\cite{Obu49}. The final saturation to a constant value for very large
separations is due to the finite system size $L$.
The scaling of the structure function $D_2^{(B)}(r)$ as well as of the
correlation function $C_2^{(B)}(r)$ follow from Eqns. (\ref{laplace}) and
(\ref{laplace1}), respectively. The results for the structure function of the
magnetic field $D_2^{(B)}(r)$ in units of $\epsilon_{\psi}\nu^{-1}$ are shown
in Fig.~\ref{p3} (thick solid lines). Clearly, for scales $r<\eta_{\omega}$ the
structure function shows again the Taylor expansion behavior
$D_2^{(B)}(r)=(\epsilon_B/2\eta)r^2$ as stated in Eq. (\ref{db_asympt}). The
larger the scales the more dominant the constant first term in Eq.
(\ref{laplace}), which is proportional to the magnetic Prandtl number, leading
to a saturation at a constant value $D_2^{(B)}(\infty)=2\epsilon_{\psi}/\eta$
for large distances, i.e. where $D_2^{(B)}\sim r^0$. The crossover between
both scaling regimes is relatively sharp and shifted towards smaller $r$ for
growing values of $Pr_m$. We could not observe an intermediate algebraic
scaling between both ranges.
The spatial correlations in the magnetic field are subdominant in $D_2^{(B)}(r)$
for scales $r\gtrsim\eta_{\omega}$, but show up in the correlation
function $C_2^{(B)}(r)$: an algebraic scaling $D_2^{(\psi)}(r)\sim r^{\alpha}$
corresponds to $C_2^{(B)}(r)\sim r^{\alpha-2}$. The results for the correlation
function, derived using (\ref{laplace1}), are shown in Fig.~\ref{p4} for
different values of $Pr_m$ and $\alpha$. In the logarithmic plot the algebraic
decay of the correlation for all discussed parameter sets is clearly visible in
particular, a $r^{-4/3}$--decay is found in regions where $D_2^{(\psi)}(r)\sim
r^{2/3}$ (see also Fig.~\ref{p2}). The correlations decay very fast when the
magnetic flux is advected passively with the turbulent velocity field.
The results show an anticorrelation of the magnetic fields,
i.e. $C_2^{(B)}< 0$. It is
observed near the crossover from $D_2^{(\psi)}\sim r^2$ to
$D_2^{(\psi)}\sim r^0$, i.e. near the transition from the
viscous regime to the Batchelor plateau.
This follows also analytically within the Batchelor
parametrization \cite{Bat51} of $D_2^{(\psi)}(r)$ by setting
\begin{eqnarray}
D_2^{(\psi)}(r)=\frac{r^2}{1+a_2 r^2}\;,
\end{eqnarray}
which approximates such a crossover very well. Equation
(\ref{laplace1}) then gives
\begin{eqnarray}
\label{lap-bat}
\frac{1}{r}\partial_r(r \partial_r D_2^{(\psi)}(r))
=4 \frac{1-a_2 r^2}{(1+a_2 r^2)^3}\;.
\end{eqnarray}
With $a_2 >0$ and $r\ge 0$ we find negative values of (\ref{lap-bat}) for
$r>(a_2)^{-1/2}$. It can be shown that a crossover of
$D_2^{(\psi)}(r)$ from $r^{\alpha}$ to $r^{\beta}$ with $0\le\beta\le\alpha\le
2$ gives negative values
in (\ref{lap-bat}) and thus anticorrelation of the
magnetic field at all when
in addition the constraint $\beta<\alpha/5$ is satisfied.
This is consistent with our results for extremely small magnetic Prandtl numbers
where no anticorrelation is observed. Here, the quadratic scaling with $r$
changes directly to an $r^{2/3}$--scaling (see Figs.~\ref{p2} and \ref{p4}), and
the intermediate Batchelor plateau is not present.
The anticorrelation in the normalized correlation function is magnified in Fig.
\ref{p5} for values of $Pr_m$ reaching from $10^{-5}$ to $10^3$. The negative
contributions vanish for sufficiently low $Pr_m$ where the Batchelor plateau is
supressed (see the cases $Pr_m=10^{-4},\,10^{-5}$). On the other hand we
observe a nearly constant value $\min\left(C_2^{(B)}\right)\simeq
-0.0375\epsilon_{\psi}/\eta$ for the larger values of the magnetic Prandtl
number.
The origin of this anticorrelation presumably lies in the advection
of $\psi$ in turbulent flow structures
and the formation of strongly filamented flux sheets (and thus
current sheets) in which the magnetic energy is stored. The
minimal sheet width $\delta_{CS}$
can be determined by the balance between the advection
and the diffusion term of (\ref{indeq_nondim}), giving
\begin{eqnarray}
\frac{U_l}{l}\psi\simeq\eta\frac{\psi}{\delta_{CS}^2}\,.
\end{eqnarray}
By taking $U_l/l=\epsilon_{\omega}^{1/3}$ as the typical strain rate we end
up with
\begin{eqnarray}
\label{delta}
\delta_{CS}\simeq \eta_{\omega}Pr_m^{-1/2}\,.
\end{eqnarray}
This $Pr_m$ scaling is also found in the position of the
minimum, so that the anticorrelation is due to layers
with opposite orientations of magnetic fields. The saturation
of the anticorrelation in Fig. \ref{p5} suggests that
the ratio of regions with perfect parallel alingment and
deviations, say due to folds or modulations in thickness,
stays the same, independent of
magnetic Prandtl number.
The relation (\ref{delta}) can also be derived by means of (\ref{fracdim}).
Assuming the subdominance of the $\alpha Pr_m r^2$ term, we can ask when the
last term under the square root in (\ref{fracdim}) will exceed unity. By
requiring a smooth flow in the viscous subrange, i.e. setting
$D_{\parallel}\sim r^2$, we get the same $Pr_m$ dependence as in (\ref{delta}).
\section{Full MHD beyond the kinematic regime}
In the following the results of the kinematic approach are compared with direct
numerical simulations \cite{Bis93} as well as integrations of spectral transfer
equations within closure models\cite{Pou78}, both at $Pr_m=1$ and with a kinetic
energy about three orders of magnitude larger than the magnetic energy. This
provides insights into the feedback of the magnetic field on the flow via the
Lorentz force density. The question is whether the phenomena seen in the
kinematic case, in particular the anticorrelation, will still be present.
Both numerical experiments support the existence of a Kolmogorov-type
scaling for the spectrum of the mean square magnetic potential,
$A(k)\sim k^{-7/3}$ for
$k<k_f$, in the inverse mean square flux cascade range. The Alfv\'{e}n
effect causing an equipartion between kinetic and magnetic energy is
subdominant. The spectral closure suggests $A(k)\sim k^{-7/2}$
for the direct energy cascade range $k>k_f$.
We start with a corresponding model spectrum
for the amplitudes $\langle|\psi_{\bf k}|^2\rangle$ (see Fig. \ref{p6}(a))
\end{multicols}
\begin{eqnarray} \langle|\psi_{\bf k}|^2\rangle\!\sim\!
\left\{
\begin{array}{r@{\quad:\quad}l}
k_f^{-7/6}k_1^{-13/3}k^3 & \frac{2\pi}{L}\le k\le k_1, \\
k_f^{-7/6}k^{-4/3} & k_1<k\le k_f,\\
k^{-5/2} & k_f<k\le k_d, \\
k^{-5/2}\exp\left[-\left(\frac{k-k_d}{k_d}\right)^{2}\right]
& k_d<k\;,
\end{array}
\right.
\label{modspec1}
\end{eqnarray}
\begin{multicols}{2}
where $k_d\sim\eta_d^{-1}$ is the wave number at the viscous cut off.
The structure function is calculated as in section III:
\begin{equation}
\label{mod_struc1}
D_2^{(\psi)}(r)=2\,\sum_k\,\langle|\psi_{\bf k}|^{2}\rangle\,
(1-\mbox{J}_{0}(kr)) \,.
\end{equation}
The resulting structure function is shown in panel (b) of Fig.~\ref{p6}. The
inset gives $\zeta_2^{(\psi)}$ vs. $r$ measured in units of $\eta_d$.
Consequently $D_2^{(\psi)}(r)$ has to be taken now in units of
$\epsilon_{\psi}\epsilon^{-1/2}\nu^{1/2}$. The $r^2$--scaling for $r<\eta_d$ is
followed by a range with $\zeta_2^{(\psi)}\approx \frac{4}{3}$ for $r>\eta_d$,
which seems to be connected with the extended inverse cascade of the mean square
magnetic potential. Again structure function and correlation function are
calculated by means of (\ref{laplace}) and (\ref{laplace1}), respectively.
Consistency was checked by calculating $D_2^{(B)}(r)$ directly via $\langle|{\bf
B}_{\bf k}|^2\rangle\sim k^2 \langle|\psi_{\bf k}|^2\rangle$ using a relation
equivalent to (\ref{mod_struc1}). The function $D_2^{(B)}(r)$ is shown in
Fig.~\ref{p3} as the dash-dotted line. We see that it reaches the constant
saturation state for scales $r$ larger than in the kinematic case. The
differences between the kinematic and full dynamic situation are more pronounced
in the correlation functions [see Fig.~\ref{p6}(c)]. The correlations decay
slower with respect to $r$, following now a $r^{-2/3}$--law.
Besides the different scalings we note that the anticorrelation
has disappeared. In the
full dynamic case the magnetic field itself would prohibit the build-up of
elongated current sheets (flux sheets) due to its feedback via the Lorentz
force density
\begin{eqnarray}
\label{Lorentz}
{\bf f}_L=
-{\bf \nabla}\left(\frac{B^2}{2\mu_0}\right)
+\frac{1}{\mu_0}({\bf B}\cdot{\bf\nabla}){\bf B}\;.
\end{eqnarray}
The current sheets, if formed, become sensitive to several resistive
instabilities such as the spontaneous growth of the tearing
mode,\cite{FKR63,Mat86,Pol89} where the stored magnetic energy can be released into
the plasma flow by means of magnetic reconnection. In this process the Lorentz
force density causes the plasma acceleration due to reconnected magnetic flux
and prevents a further steepening of flux up to the dissipative scales. Thus in
addition to the condition $M>R_m$ the Reynolds number R has to be
sufficiently low to avoid resistive MHD instabilities. It was found that
the onset of the tearing mode instability is determined by the
Hartmann number $Ha=\sqrt{R_m R}=R \sqrt{Pr_m}$ and the width
$\delta_{CS}$.\cite{Dah83,SeSch98}
\section{Summary}
Our findings for the magnetic structure function as well as the corresponding
correlation function in a two-dimensional
MHD system can be summarized as
follows.
(1) For the kinematic regime of two-dimensional MHD turbulence ($M\gg 1$), the
magnetic flux function $\psi$ can be treated as a passive scalar advected in a
turbulent flow. By means of the geometric scaling theory the second order
structure function $D_2^{(\psi)}(r)$ was calculated over a wide range of scales
$r$ and of magnetic Prandtl numbers $Pr_m$. For this case a rather complete
analysis is now possible by means of Eq. (\ref{fracdim}).
(2) The second order magnetic structure function $D_2^{(B)}(r)$ and the
corresponding magnetic correlation function $C_2^{(B)}(r)$ were calculated by
exact analytical relations from $D_2^{(\psi)}(r)$. The larger $Pr_m$ the
smaller the scale $r$ where the structure function reaches the constant
saturation. The correlation function decays fast with $r^{-4/3}$ and shows an
anticorrelation. The latter is connected with the transition to the Batchelor
regime and can be associated with an attempt of the system to concentrate its
magnetic energy in strongly filamented structures. This process is limited by
the finite resistivity $\eta$. The anticorrelation minimum was found to be
independent of the magnetic Prandtl number.
(3) The results for the kinematic approach were compared to results in the
dynamic regime $M<R_m$. The magnetic field correlations decay with $r^{-2/3}$,
i.e. slower than for the kinematic case. This can be explained by the feedback
of the magnetic field on the flow which prohibits the build-up of strongly
filamented, elongated current sheets on small scales. Therefore no
anticorrelation was detected.
While the discussion in this paper was limited to the 2-D case, a weak field
regime without dynamo can also be identified in three dimensions. In the
experiments of Odier {\it et al.}\cite{Od98} a scaling of the magnetic field was observed that
is very close to what one expects for a passive scalar. Thus the stretching of
the magnetic field does not seem to have a pronounced effect on the structure
function. Further investigations of this point seem worthwhile.
\acknowledgments
Fruitful discussions with H. Fuchs and K.-H. R\"adler are
gratefully acknowledged.
|
\newpage\sect{\newpage\sect}
\def\text#1{\mbox{\rm #1\ }}
\def{\rm Tr}{{\rm Tr}}
\def{\it Diag}{{\it Diag}}
\def{\rm i.e.,\/}\ {{\rm i.e.,\/}\ }
\def{\rm etc.\/}\ {{\rm etc.\/}\ }
\def{\rm cf.\/}\ {{\rm cf.\/}\ }
\def\mbox{\it id\,}{\mbox{\it id\,}}
\def\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}{\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}}
\newcommand{\Mlo}[1]{\ensuremath{(M_{#1}(\Lambda^{2}))_0}}
\newcommand{{\mbox{\rm Z}\hskip-5pt \mbox{\rm Z}}}{\mathbb{Z}}
\newcommand{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}{\mathbb{R}}
\newcommand{{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}}{\mathbb{C}}
\title{Hopf stars, twisted Hopf stars and scalar products
on quantum spaces
\vspace{0.8cm}
}
\author{R. Coquereaux${}^1$\thanks{~Email: <EMAIL>}$\;$,
A. O. Garc\'{\i}a${}^1$\thanks{~Email: <EMAIL>}$\;$,
R. Trinchero${}^2$\thanks{~Email: <EMAIL>} \\
\\
${}^1$ {\it Centre de Physique Th\'eorique - CNRS} \\
{\it Campus de Luminy - Case 907} \\
{\it F-13288 Marseille - France} \\
\\
${}^2$ {\it Instituto Balseiro and Centro At\'omico Bariloche} \\
{\it CC 439 - CP 8400 - Bariloche - R\'{\i}o Negro - Argentina} \\
\\
}
\date{}
\begin{document}
\begin{titlepage}
\thispagestyle{empty}
\maketitle
\vfill
\abstract{
The properties of Hopf star operations and twisted Hopf stars operations
on quantum groups are discussed in relation with the theory of
representations (star representations). Invariant Hermitian sesquilinear
forms (scalar products) on modules or module-algebras are then defined
and analyzed. Particular attention is paid to scalar products that can
be associated with the Killing form (when it exists) or with the left
(or right) invariant integrals on the quantum group.
Our results are systematically illustrated in the case of a family
of non semi-simple and finite dimensional quantum groups that are
obtained as Hopf quotients of the quantum enveloping algebra
$U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$, $q$ being an $N$-th root of unity. Many explicit
results concerning the case $N=3$ are given.
We also mention several physical motivations for the present work:
conformal field theory, spin chains, integrable models, generalized
Yang-Mills theory with quantum group action and the search for finite
quantum groups symmetries in particle physics.
}
\vspace{0.8 cm}
\noindent PACS: 02.90.+p, 11.30.-j \\
\noindent MSC: 16W30, 81R50 \\
\noindent Keywords: quantum groups, Hopf algebras, star structures,
scalar products, non commutative geometry.
\vspace{0.5cm}
\noindent Anonymous ftp or gopher: cpt.univ-mrs.fr
\vspace{0.3 cm}
\noindent {\tt math-ph/9904037}\\
\noindent CPT-99/P.3808 \\
\noindent
\vspace*{0.3 cm}
\end{titlepage}
\newpage\sect{Introduction}
The purpose of the present paper is to study the concepts of Hopf star
operations and twisted Hopf star operations in the theory of quantum
groups. This study is motivated by a number of physical considerations
that we shall discuss in this introduction.
First of all, one should remember that the notion of quantum group
(Hopf algebra) does not make use of a star operation ---roughly
speaking, the notion of complex conjugate---; chosing one comes only
at a later stage. Such an operation is an anti-multiplicative and
anti-linear involution which could be quite arbitrary when the
(associative) algebra under consideration is not a Hopf algebra.
However, the existence of a coproduct allows one to distinguish two
particular kinds of star operations. The problem is to relate the star
operations that one can define on the algebra $H$ and on its tensor
square $H\otimes H$, since we have a very special embedding of the
first algebra into the latter one given by the coproduct. If
$\Delta a = a_1\otimes a_2$, it may be that the chosen star is such
that $\Delta (a^*) = a_1^*\otimes a_2^*$ (a Hopf star operation) but
it also could happen that $\Delta (a^*) = a_2^*\otimes a_1^*$ (a
twisted Hopf star operation).
Actually, one could define also ``partially twisted stars'' which in
a sense continuously interpolate between a Hopf and a twisted star
(see \cite{Buffenoir-Roche}), but these involve additional data, an
element $f \in H \otimes H$.
In the case of Lie groups or Lie algebras, star operations are used to
define real forms. However, for Hopf algebras the notion of ``real
form'' is slightly more subtle (we shall say more about it later),
but it is a priori clear that the notions of complex conjugate and of
star representations should be discussed as soon as one wants to endow
a representation space with some sort of scalar product.
A general discussion of star versus twisted star operations seems to be
lacking in the literature: mathematical books on quantum groups (for
instance \cite{Chari-Pressley} or \cite{Klimyk}) only discuss (genuine)
Hopf star operations, the same being true for all research papers
studying $C^*$-algebra aspects of ``matrix quantum groups'' (in the
sense of Woronowicz \cite{Woronowicz}). In the physics literature, most
papers dealing with applications of quantum groups to integrable models,
spin chains, or conformal field theory, usually do not choose any
particular star operation at all on the quantum group of interest. But
sometimes they do, and it turns out that the chosen star is often a
twisted star ---although usually the authors do not acknowledge the
fact that it is so\footnote{
With the notable exception of papers by G. Mack and V. Schomerus
\cite{Mack-Schomerus-1, Mack-Schomerus-2}.
},
and this state of affairs creates some confusion. Quantum groups have
also been discussed in relation with the possibility of $q$-deforming
the Lorenz group, and here again, the two possibilities (twisted
versus non twisted) appear in the physical literature: from one side
we have the papers \cite{Wess-Zumino-Ogievetsky} or
\cite{Podles-Woronowicz}, whereas from the other we have the
articles \cite{Lukierski-et-al}.
Another motivation for our work comes from the possibility, as
advocated by A. Connes in \cite{Connes}, that reduced quantum groups
like $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ at a cubic root of unity could have some
essential role to play in the formulation of fundamental interactions
(Standard Model). This suggestion is based upon the following two
observations: first of all, when $q$ is chosen to be a cubic root of
unity the algebra of ``functions'' $Fun(SL_q(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ is a Hopf-Galois
extension of $Fun(SL(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ ---the algebra of complex valued functions
on the Lorenz group---, the fiber being a finite dimensional quantum
group $\mathcal{F}$ whose Hopf dual is a finite dimensional Hopf algebra
quotient $U^{res}_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ (that we shall call $\mathcal{H}$) of
the quantum enveloping algebra $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$. Next, for this $q$ the
semi-simple part of $\mathcal H$ turns out to be isomorphic with the
algebra $M(3,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})\oplus M(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}) \oplus {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}\,$. It is then tempting to
use the tools of non commutative geometry to build a physical model
that would recover the usual Standard Model ---maybe a generalization
of it---, incorporating some action of an hitherto unnoticed finite
quantum group of symmetries. The existence of a non trivial coproduct
mixing the different components of $\mathcal{H}$ and the nature of the
representations of this non semi-simple Hopf algebra make it quite hard
to recover the usual model of strong and electroweak interactions; this
has not been achieved yet. In any case, it is clearly of interest to
analyze in detail the structure of the representation theory of this
Hopf algebra, and to pay particular attention to the different kinds
of ``reality'' structures that one can find for these representations.
For these reasons, and although we decided to write quite a general
paper, most explicitly discussed examples will involve the case of the
finite dimensional algebra $\mathcal{H} = U^{res}_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ at a
cubic root of unity.
Another motivation for studying the reality structures and the type of
scalar products in star representations of quantum groups comes from
our previous work \cite{CoGaTr-l, CoGaTr-e}. Here a new kind of gauge
fields was obtained: starting from the observation that the reduced
quantum plane (identified with the algebra of $N\times N$ complex
matrices) is a module-algebra for the finite dimensional quantum group
$\mathcal{H}$, when $q^N=1$, we built a differential algebra over it by
taking an appropriate quotient of the Wess-Zumino differential algebra
over the ---infinite dimensional--- quantum plane; generalized
differential forms are then obtained by making the tensor product of
the De Rham complex of forms over an arbitrary space-time manifold times
the previous Wess-Zumino reduced differential complex; generalized gauge
fields (and curvatures, {\rm etc.\/}\ ) are finally constructed by standard non
commutative geometrical techniques. Clearly, the wish to construct a
lagrangian model involving the representations of a quantum group (that
knows how to act on such generalized gauge fields) requires the study of
star (or twisted star) operations on the corresponding modules.
Finally, the last motivation comes from spin chains, integrable models
and conformal theories. The $q$-parameter appearing in many conformal
field theory models and integrable models is a primitive root of unity.
Such values, as a rule {\sl exclude} the choice of a Hopf star operation
leading to a compact quantum group like $SU_q(2)$, for instance. For this
reason star operations used in articles like \cite{Pasquier-Saleur}
---where the role of quantum groups is discussed in the context of spin
chains, like $SU_q(2)$ in the $XXZ$ model--- are not true Hopf star
operations; we shall return to this discussion in
Section~\ref{s:discussion}.
\bigskip
The structure of our paper is the following:
\smallskip
In Section~\ref{s:stars} we gather information on stars operations: Hopf
and twisted Hopf stars, compatible stars on modules and module-algebras,
behaviour under tensor product of representations, {\rm etc.\/}\
In Section~\ref{s:invariant_scalar_products} we discuss scalar products
in representation spaces, its quantum invariance and associated star
representations. As everywhere else in this paper, we first discuss
all the general notions and then exemplify by taking the finite
dimensional quantum group $\mathcal{H} = U^{res}_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ for
$q$ a primitive odd root of unity (most of the time we take $N=3$).
The characteristics of the invariant scalar products on the irreducible
and the projective indecomposable representations of $\mathcal{H}$ are
studied in detail, both in the case where a genuine or a twisted Hopf
star is chosen on $\mathcal{H}$. The same analysis is carried out for
the module-algebra $M(N,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$.
In Section~\ref{s:scalar_products_on_left_rr} we examine more particularly
the (left) regular representation of a Hopf algebra $H$ and exhibit two
distinguished invariant scalar products. The first one is defined in terms
of the Killing form. The other is built using the left (or right) invariant
integral on the algebra $H$. We then analyze in detail these scalar
products for the case of $\mathcal{H}$. As we shall see, it happens that
for many properties Hopf stars behave usually much better than twisted Hopf
stars.
\ref{a:structure_of_H} summarizes what is needed for this paper from
the structure and representation theory of the finite dimensional Hopf
algebras $\mathcal{H} = U^{res}_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ when $q$ is an odd
primitive root of unity, in particular the structure of the projective
indecomposable modules and of the corresponding irreducibles.
\ref{a:Killing_form} recalls a few properties concerning the adjoint
representation of quantum groups, together with the notions of quantum
trace and quantum Killing form.
\ref{a:algebras_related_to_H} gives a few explicit results concerning a
``double cover'' of the finite dimensional Hopf algebra $\mathcal{H}$.
\subsection*{About notations}
$F$ will generically denote a complex Hopf algebra, for example the
algebra of ``functions'' on a quantum group. $H$ will be its dual (also
a Hopf algebra), so that it can be thought of as a non commutative
generalization of the group-algebra of a finite group or as the
non-commutative analogue of the enveloping algebra of a Lie algebra.
As already mentioned, the particular examples where $H$ is chosen to
be one of the finite dimensional quotients of $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ will be
called ${\mathcal H}$. $V$ will denote a representation space for $H$
(and we shall have to specify if it is a left or a right module), and
will therefore also be a (left or right) co-representation space of the
Hopf algebra $F$. Finally, $M$ will denote a module-algebra for $H$
({\rm i.e.,\/}\ a comodule-algebra for $F$).
\newpage\sect{Stars}
\label{s:stars}
\subsection{Hopf stars}
Remember that a star on an algebra is an involutive antilinear
antiautomorphism, {\rm i.e.,\/}\
$$
\begin{array}{l}
\left( x^*\right)^* = x \\
(\lambda \, x)^* = \bar{\lambda} x^* \;, \quad \quad \lambda \in {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}
\\
(xy)^* = y^* x^*
\end{array}
$$
\noindent
Now let the algebra on which $*$ acts be a complex Hopf algebra
$H(m,\Delta,\eta,\epsilon,S)$. In this case one requires the star to
satisfy two extra compatibility conditions \cite{Chari-Pressley} with
the Hopf operations:
\begin{eqnarray}
\Delta \, * &=& *_{_{\otimes}} \, \Delta
\label{normal_Hopf_star} \\
\epsilon \,* &=& *_{_{{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}}}\,\epsilon \ . \nonumber
\end{eqnarray}
\noindent
However the $*$'s on the right hand side are operators on different
spaces and are yet to be defined. $*_{_{{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}}}$ should be a star on
${\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$, and therefore is just complex conjugation. The operation
$*_{_{\otimes}}$ should be an involution on $H \otimes H$, the
standard choice is
$$
*_{_{\otimes}} = * \otimes * \ .
$$
\noindent
A star satisfying (\ref{normal_Hopf_star}) with the standard choice
of $*_{_{\otimes}}$ is called a Hopf star, and in such a case $H$ is
called a Hopf star algebra.
Actually one could also make the choice
$*_{_{\otimes}} = \tau(* \otimes *)$,
where $\tau$ is the tensorial flip (twisting); however, making such a
choice and imposing (\ref{normal_Hopf_star}) amounts to make the
standard choice for $*_{_{\otimes}}$ and rewrite
(\ref{normal_Hopf_star}) as
$$
\Delta \, * = *_{_{\otimes}} \, \Delta^{op} \ ,
$$
where $\Delta^{op} \doteq \tau \circ \Delta$ is the opposite coproduct.
We will call this second type of operation a {\it twisted} Hopf star,
or even a twisted star. In this paper, therefore, we shall always make
the standard choice for $*_{_{\otimes}}$. In this section we will
analyze Hopf star algebras, leaving the study of the twisted case to
Section~\ref{s:twisted_star}.
Remark that there is no need to impose a relation between the star
and the antipode (which is a {\sl linear} antiautomorphism) because
this one arises automatically. In fact, it is easy to see that
\begin{eqnarray}
S \,* \, S \, * &=& \mbox{\it id\,} \ .
\end{eqnarray}
\noindent
This is so because $* \,S^{-1}\, * = (*\,S\,*)^{-1}$ satisfies all
the conditions for the antipode, which is unique. We should therefore
remember that for Hopf star algebras
\begin{eqnarray*}
S \,* &=& * \, S^{-1} \ .
\end{eqnarray*}
Notice that in general $S$ has no reason to be equal to $S^{-1}$
(imposing such a property would exclude all the Drinfeld-Jimbo
deformations!).
Given a Hopf algebra $H$, one can consider its dual\footnote{
The examples that we shall consider in this paper are finite
dimensional (and non semisimple) Hopf algebras, therefore it
will be possible to identify canonically a given Hopf algebra
with its bidual.
}
Hopf algebra $F=H^\star$ with operations such that
\begin{eqnarray}
\langle \Delta f, h \otimes h' \rangle &=& \langle f\,, hh'\rangle
\qquad\qquad \forall \:h,h'\in H \nonumber \\
\langle ff', h \rangle &=& \langle f\otimes f' \,, \Delta h\rangle \\
\langle Sf, h \rangle &=& \langle f\,, Sh\rangle \nonumber \\
\epsilon(f) &=& \langle f\,, 1\rangle \nonumber \\
\langle \mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}_F , h \rangle &=& \epsilon(h) \ , \nonumber
\end{eqnarray}
\noindent
where $\langle \: , \: \rangle: F \otimes H \rightarrow {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$ is the
bilinear evaluation pairing. When $H$ is a Hopf star algebra, one may
also define a dual star on $F$. By dual star we mean a star on $F$
which is also a Hopf star. It is easy to verify that the following
formula defines such an operation:
\begin{equation}
\langle f^{\,*},h\rangle =\overline{\langle f\,,(Sh)^{*}\rangle} \ ,
\qquad \forall \:h\in H \ .
\end{equation}
In what follows $F$ will be thought of as the space of functions on a
quantum group, and its dual $H$ as the quantum group analog of the
corresponding group algebra (or the enveloping algebra).
Remark that another standard accepted terminology for denoting the
star structure of a (untwisted) Hopf star algebra is ``real form on
a Hopf algebra''. This name does not imply, and we do not construct
here, any {\it real} Hopf subalgebra of $H$, in the sense of being
an algebra over the field ${\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$ of real numbers (see
\cite{Buffenoir-Roche} for a discussion of this point). Let $T$
be a {\sl linear} involutive Hopf algebra {\sl anti\/}automorphism
(we call it $T$ for transposition like in \cite{Rosso})
of a complex Hopf star algebra $H$, and consider the subspace
$H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}} \doteq \{h \in H / h^{*} = T(h)\}$. Suppose moreover
that $H = H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}\oplus i H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}$, that $T * = * T$ and
that $H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}$ is invariant by the coproduct $\Delta$ ({\rm i.e.,\/}\
$\Delta H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}\subset H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}} \otimes H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}$), then $H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}$ is
a {\it real} Hopf algebra associated with the star $*$ and the
involution $T$. Notice that $c \doteq T \, *$ is an {\sl anti\/}linear
involutive automorphism\footnote{
In the case of our favorite example $\mathcal{H}$, such an operator $c$
can be defined \cite{Kastler}, on the generators, by setting
$c(X_+) = -q KX_-$, $c(X_-) = -q K^{-1}X+$, $c(K) = K$.
}
and that $H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}$ is the set of elements of
$H$ that are invariant under the conjugation $c$. Notice also that if
$h \in H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}\:$, then $ih$, as defined in $H$, cannot belong to
$H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}$ since $(ih)^{*} = - T(ih)$. When $H$ is ``classical'' (the
enveloping algebra of some complex Lie algebra), such a $H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}$ is
the enveloping algebra of a real Lie algebra. Moreover, in this case
one takes $T = S$ (since $S^{2}=\mbox{\it id\,}$), so in $H_{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$ we have
$x^{*}=T(x) = S(x) = x^{-1}$ for group-like elements and
$x^{*}=T(x) = S(x) = -x$ for primitive elements.
\subsubsection{Selfconjugate representations and compatible stars on
modules}
Suppose now that we are given a star $*_H$ on the Hopf algebra $H$,
and a representation on a vector space $V$. We may have to face
possible situations.
The first possibility is that we may want to define a star $*_V$ on
$V$ and decide to constrain it by imposing some sort of compatibility
with the star $*_H$ on the quantum group. The second possibility is
to suppose that we already start with a star $*_V$ on $V$ (a priori
given);
in such a case\footnote{And thinking now only in the Hopf star case,
as it is the only one where this notion makes sense.} one can define
on the same vector space a new representation called the conjugate
representation. It {\em may} happen that both actions ---the original
one and its conjugate--- are equivalent. In this last case the
representation is therefore called {\it self-conjugated}.
Actually, the compatibility condition (see below) between the stars
in the first scenario is just a particular case of the second option,
as we define $*_V$ to be such that the representation precisely
coincides with its conjugate.
Going back to our first problem,
suppose now that we want to define a star $*_V$ on $V$, which is a
representation space for the quantum group $H$ and a corepresentation
space for its dual $F$ ({\rm i.e.,\/}\ $V$ is a right $F$-comodule). Call the
coaction $\delta_R : V\mapsto V\otimes F$.
For a Hopf star $*_F$ on $F$ it can be checked that the operation
$\delta^\prime \doteq (* \otimes *) \, \delta_R \, * \, : V
\longrightarrow V \otimes F$ is again a right coaction on $V$.
Therefore it is natural to impose $\delta^\prime = \delta_R$ as the
compatibility condition between the stars $*_F$ and $*_V$. With a
slight abuse of notation we can even write
\begin{equation}
\delta_R (z^*) = \left( \delta_R z \right)^* \ , \qquad z \in V \ ,
\label{Hopf_star_on_comodule}
\end{equation}
\noindent
where the conjugation on the right hand side is the natural star
structure on $V \otimes F$. In this case we may say that the star is
covariant.
$V$ being a (right) $F$-comodule, it is also a (left) $H$-module. We
have indeed an action $\triangleright : H \otimes V \mapsto V$ given
by
$$
h \triangleright z =
(id \otimes \langle h, \cdot \rangle) \delta_{R}(z) \ . $$
Pairing equation~(\ref{Hopf_star_on_comodule}) with an element $h \in
H$, and using the duality of real structures we get the equation
\begin{equation}
h \triangleright z^* = \left[ (Sh)^* \triangleright z \right]^* \ ,
\qquad z \in V \ .
\label{Hopf_star_on_module}
\end{equation}
\noindent
Assuming nondegeneracy of the duality pairing both expressions are
completely equivalent, and imply some restrictions on $*_V$ given
$*_F$ or $*_H$.
The action $h \,\triangleright$ of $h$ on $V$ is implemented thanks
to an endomorphism $\rho[h]$ of this vector space, so one may also
write $h \,\triangleright \doteq \rho[h]$. Using this notation,
equation (\ref{Hopf_star_on_module}) can also be written $$
\rho[h](z) = \bar\rho[h](z) \ ,
$$
where $\bar\rho$ denotes the conjugate representation $$
\bar\rho[h](z) = \left[ \rho[(Sh)^*] (z^{*}) \right]^* $$
dual to the above $\delta^\prime$ right coaction\footnote{ Remember
that in the ``classical'' case ({\rm i.e.,\/}\ real forms of Lie algebras and
their enveloping algebras), $(Sh)^{*}=h$ for the Lie algebra
generators, and we recognize the usual equation $\bar\rho = {*} \rho
{*}$ defining the conjugate representation. }.
Therefore the compatibility relation (\ref{Hopf_star_on_module})
between the stars on $H$ and $V$
can also be viewed as a very particular case of equivalency of
representations: $\rho$ and $\bar\rho$ should just coincide. Given
the star operations, a representation $\rho$ is called {\it
selfconjugate} if there exists an invertible operator $U:V\mapsto V$
such that
$$
U^{-1} \rho[h] U = \bar\rho[h] \ .
$$
Up to now we did not assume that the representation space $V$ was
endowed with a scalar product $(\cdot\,,\cdot)$. Therefore, we can
not impose, at this point, that $U$ should be unitary. We can not
assume, either, that the star operation on $V$ is an antiunitary
operator, $(v^{*},w^{*})=(w,v)$. For the same reason too, the
notation $\dag$ (adjoint) was avoided. In any case, a Hopf algebra
is, in particular, an associative algebra, and if it is so happens
that a {\sl real} Hopf algebra $H_{{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}}$ can be defined the usual
classification for representations of real associative algebras on
complex Hilbert spaces will, of course, also hold. We could have
three types of representations, complex, real, and quaternionic; we
refer the reader to standard textbooks (see for instance
\cite{Bourbaki}\cite{Simon}).
\subsubsection{Compatible stars on module-algebras}
Instead of a comodule $V$, we now take a right $F$-comodule algebra
$M$, {\rm i.e.,\/}\ we assume that the right coaction $\delta_{R}$ is an algebra
homomorphism from $M$ to $M\otimes F$,
$$
\delta_{R}(zw)=\delta _{R} z \, \delta _{R} w \ .
$$
The map $\delta^\prime: M \rightarrow M \otimes F$ defined as above
will again be an algebra homomorphism, {\rm i.e.,\/}\
$\delta^\prime(zw)=\delta^\prime z \, \delta^\prime w$.
Thus equation (\ref{Hopf_star_on_comodule}) is still a good
requirement when the comodule $M$ is an algebra and shows that
compatibility of the coaction with a given Hopf star operation needs
only to be verified on the (algebra) generators\footnote{
In the case of a module algebra, the star operation is of course
assumed to be antimultiplicative ($(xy)^{*} = y^{*} x^{*}$).
}.
Obviously the dual equation (\ref{Hopf_star_on_module}) defining
compatibility of Hopf stars on left modules will also have the same
properties. Remember that, being a right $F$-comodule algebra, $M$
supports a left action of the dual $H$ of $F$ and indeed this action
is compatible with the product in $M$ (call
$\Delta h = h_{1}\otimes h_{2}$):
$$
h \triangleright (zw) =
(h_1\triangleright z)(h_2\triangleright w) \ .
$$
\subsubsection{Example of the reduced $SL_{q}(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$ at $q^N = 1$}
\label{s:stars_on_FHM}
\medskip
\noindent
{\bf Hopf stars on ${\mathcal F}$ and ${\mathcal H}$}
\medskip
\noindent
First of all, remember that in the quantum case one has three
possibilities for the star operations on $Fun(SL_q(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ (up to
star-Hopf homomorphisms). Given the conventions chosen in
\cite{CoGaTr-e}, they are given on generators by:
\begin{itemize}
\item
The real form $Fun(SU_q(2))$: \quad $a^* = d$, $b^* = -qc$,
$c^* = -q^{-1} b$ and $d^* = a$. Moreover, $q$ should be real.
\item
The real form $Fun(SU_q(1,1))$: \quad $a^* = d$, $b^* = qc$,
$c^* = q^{-1} b$ and $d^* = a$. Moreover, $q$ should be real.
\item
The real form $Fun(SL_q(2,{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}))$: the conjugation is given by
\begin{eqnarray}
a^* &=& a \nonumber \\
b^* &=& b \label{Hopf_star_on_F} \\
c^* &=& c \nonumber \\
d^* &=& d \ . \nonumber
\end{eqnarray}
Here $q$ can be complex but it should be a phase.
\end{itemize}
\noindent
When $q=\pm i \;$ ---hence $q^{4}=1$--- there are still two other
Hopf star structures that have no classical limit (see
\cite{Chari-Pressley} and references therein). A systematic analysis
of real forms for special linear quantum groups $SL_{q}(n)$ was made
by \cite{Jain-Ogievetsky} and, in the case of $GL_{p,q}(2)$ or
$GL_{\alpha}^J(2)$, by \cite{Ewen-Ogievetsky}.
It is already clear from these results that taking $q$ a root of
unity is incompatible with the $SU_q$ and $SU_{q}(1,1)$ real forms.
The only possibility if we assume $q^N = 1$ is to choose the Hopf
star corresponding to $Fun(SL_q(2,{\mbox{\rm I}\hskip-2pt \mbox{\rm R}}))$. Moreover, in such a case
the star is compatible with the finite dimensional Hopf algebra
quotient $\mathcal{F}$ obtained by factoring this quantum group by
the Hopf ideal defined by \cite{CoGaTr-e}
$a^{N}=d^{N} = \mbox{\rm 1}\hskip-2.8pt \mbox{\rm l} \,,\ b^{N}=c^{N}=0$
(take $N$ odd here, and $q$ a primitive $N$-th root of unity).
The corresponding dual star on the dual Hopf algebra $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$
(see \cite{CoGaTr-e} or \ref{a:structure_of_H} for its structure) is
\begin{eqnarray}
X_{+}^{*} &=& -q^{-1}\,X_{+} \nonumber \\
X_{-}^{*} &=& -q\,X_{-} \label{Hopf_star_on_H} \\
K^{*} &=& K \ . \nonumber
\end{eqnarray}
\noindent
Here one can also factor the quantum enveloping algebra by the Hopf
ideal defined by $K^{N}=\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}\,,\ X_{+}^{N}=0\,,\ X_{-}^{N}=0$ and the
same remarks concerning the fact that the stars passes to the
quotient $\mathcal H$ apply \cite{CoGaTr-e}.
\medskip
\noindent
{\bf Compatible star on the quantum plane ${\mathcal M}$}
\medskip
\noindent
The quantum group $Fun(SL_q(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ coacts on the quantum plane
algebra generated by $x,y$ such that $xy=qyx$. For a root of unity
this algebra can be quotiented by the ideal defined by
$x^N=y^N=\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}$, to obtain a finite dimensional algebra that we call
$\mathcal M$. $\mathcal{M}$ is a right comodule algebra for
$\mathcal{F}$ and the right coaction is given by
$\delta_{R} \left( \begin{array}{cc} x & y \end{array} \right) =
\left( \begin{array}{cc} x & y \end{array} \right) \dot{\otimes}
\left( \begin{array}{cc} a & b \\ c & d \end{array} \right)$.
The reduced quantized universal enveloping algebra $\mathcal{H}$ acts
on this quantum plane (for compatible formulae for actions and
coactions, see for instance \cite{CoGaTr-e}). Up to equivalences (now
$*$-homomorphisms) there is only one conjugation on this quantum
plane compatible with the requirements (\ref{Hopf_star_on_comodule})
or (\ref{Hopf_star_on_module}). It works for both the infinite
dimensional algebra or its reduced (finite) quotients when $q^N = 1$.
It is
\begin{eqnarray}
x^{*}&=&x \cr
y^{*}&=&y \ . \nonumber
\end{eqnarray}
Notice that although the star is the identity on the generators it is
non-trivial on $\mathcal{M}$ since it is an antimultiplicative
operation and, for instance, $(xy)^{*} = q^{-1} xy$.
\subsection{Twisted Hopf stars}
\label{s:twisted_star}
As we mentioned before there is an alternative way of relating the
Hopf and star structures on a Hopf algebra. It reduces to replacing
in (\ref{normal_Hopf_star}) the equation for the coproduct
by\footnote{
It could even be written $\Delta * = *_{_{op}} \Delta$ at the expense
of using a flipped definition of the star on the tensor product:
$*_{_{op}} (f\otimes g) = g^* \otimes f^*$.
}
\begin{eqnarray*}
\Delta * &=& (* \otimes *) \Delta^{op} \ ,
\end{eqnarray*}
\noindent
Given such a twisted star on a Hopf algebra $H$, the dual Hopf
algebra $F=H^\star$ can be also endowed with a dual twisted Hopf
star. One just has to define it by
\begin{equation}
\langle f^* \,,h\rangle =\overline{\langle f\,,h^*\rangle} \ .
\label{twisted_dual_*}
\end{equation}
\noindent
It can be readily verified that this operation is a twisted Hopf star
on $F$. As in the untwisted case, a relation involving the antipode
is automatically fulfilled. Now the antipode and the star commute,
\begin{eqnarray}
S \, * &=& * \, S \ .
\end{eqnarray}
\noindent
This is so because $*S*$ is again an antipode, which is unique.
\subsubsection{Compatible twisted stars on modules}
Let $V$ be again a right $F$-comodule. Given $*_F$ a twisted Hopf
star on $F$ we would now like to use it to restrict the possible
choices for a star $*_V$ on $V$, as it was done with equation
(\ref{Hopf_star_on_comodule}) in the pure Hopf case.
$*_F$ being twisted,
$(* \otimes *)\, \delta_R \, * : V \mapsto V\otimes F$ is no longer
a right coaction, however
$\tau (* \otimes *) \,\delta_R \,* : V \mapsto F\otimes V$ is a
left one. Moreover
$(\mbox{\it id\,} \otimes S) (* \otimes *) \, \delta_R \,* =
(* \otimes *) (\mbox{\it id\,} \otimes S) \, \delta_R \,*$
is again a right coaction. Consequently we may require
\begin{equation}
(\mbox{\it id\,} \otimes S) \delta_R (z^*) = \left( \delta_R z \right)^* \ ,
\qquad z \in V \ ,
\label{twisted_star_on_comodule_space}
\end{equation}
or the following dual expression for the corresponding action of $H$
on the module $V$:
\begin{equation}
h \triangleright z^* = \left[ (Sh)^* \triangleright z \right]^* \ ,
\qquad z \in V \ .
\label{Hopf_star_on_module_space}
\end{equation}
Notice that this condition looks formally like
(\ref{Hopf_star_on_module}).
\subsubsection{Compatible twisted stars on module-algebras}
If we now let $V$ be an $F$-comodule {\em algebra} (we then call it
$M$ rather than $V$), it happens that
(\ref{twisted_star_on_comodule_space}) is not a reasonable condition
anymore, because $(\mbox{\it id\,} \otimes S) \delta_R *$ and $* \delta_R$ have
different homomorphism behaviour. It may also be said that $(\mbox{\it id\,}
\otimes S) (* \otimes *) \delta_R *$ is not an R-coaction {\em on an
algebra} but only a coaction; it doesn't preserve the product on $M$.
As $\tau \, (* \otimes *)\, \delta_R \, *$ is a good homomorphism,
the way out to constrain $*_M$ is to choose some {\em other} left
algebra-coaction $\delta_L$ on $M$ and impose
\begin{equation}
\delta_R (z^*) = \left( \delta_L z \right)^{*_{op}} \ ,
\qquad z \in M \ ,
\label{twisted_star_on_comodule_algebra}
\end{equation}
\noindent
where now the star $*_{_{op}}$ on the right hand side includes the
tensorial flip (on $F \otimes M$ it is given by
$*_{_{op}}(f \otimes z) = z^* \otimes f^* \,, \ z \in M, \ f \in F$).
Remark that for many interesting cases we have both natural left and
right coactions; this is for instance the case for quantum planes.
The dual condition involves the left and right actions of $H$ on $M$
which are dual to $\delta_R$ and $\delta_L$, they are respectively
denoted by $\triangleright$ and $\triangleleft$. It reads:
\begin{equation}
z^* \triangleleft h = \left[ h^* \triangleright z \right]^* \ ,
\qquad h\in H\, , \ z\in M \ .
\label{twisted_star_on_module_algebra}
\end{equation}
\subsubsection{Example of the reduced $SL_{q}(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$ at $q^N = 1$}
\medskip
\noindent
{\bf Twisted Hopf stars on ${\mathcal F}$ and ${\mathcal H}$}
\medskip
\noindent
On both the reduced and unreduced $SL_{q}(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$, the twisted stars
are essentially the following\footnote{
The operation defined on generators by
$a^{*}=d, d^{*}=a, b^{*}=\pm b$ and $c^{*}=\pm c$ ``almost works'',
in the sense that it defines a twisted star in $GL_{q}(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$ but it
is incompatible with the determinant condition defining
$SL_{q}(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$.}
({\rm i.e.,\/}\ up to automorphisms):
\begin{eqnarray}
a^*&=&a \nonumber \\ b^*&=&\pm c \label{twisted_star_on_F} \\
c^*&=&\pm b \nonumber \\ d^*&=&d \nonumber \ .
\end{eqnarray}
\noindent So we have two of them, and the corresponding dual twisted
stars are given by
\begin{eqnarray}
X_{+}^* &=& \pm X_{-} \nonumber \\
X_{-}^* &=& \pm X_{+} \label{twisted_star_on_H} \\
K^* &=& K^{-1} \nonumber \ .
\end{eqnarray}
\noindent
Thus we see that, when $q$ is a root of unity, these twisted stars
allow one to recover
the $SU(2)$ ($+$ sign) and $SU(1,1)$ ($-$ sign) real forms, something
that would be otherwise forbidden with a {\it true} Hopf star
operation.
\medskip
\noindent
{\bf Compatible star on the quantum plane ${\mathcal M}$}
\medskip
\noindent
On the quantum plane there is, again up to equivalence, only one star
structure
compatible in the sense (\ref{twisted_star_on_comodule_algebra}) or
(\ref{twisted_star_on_module_algebra}) with each of the twisted stars
(\ref{twisted_star_on_F}) or (\ref{twisted_star_on_H}). These twisted
stars are respectively given by
\begin{eqnarray}
x^{*} &=& x \label{twisted_star_on_M} \\
y^{*} &=& \pm y \ . \nonumber
\end{eqnarray}
\subsection{Stars and tensor products}
\subsubsection{Tensor product of matrices}
If
$$
m = \pmatrix a & b \cr c & d \endpmatrix \qquad \text{and} \qquad
M = \pmatrix A & B \cr C & D \endpmatrix
$$
are two matrices with {\sl non commutative} entries belonging to a
ring $\mathcal B$, then it is standard to define their tensor product
as
$$
m \otimes M \doteq \pmatrix
aA & aB & bA & bB \cr
aC & aD & bC & bD \cr
cA & cB & dA & dB \cr
cC & cD & dC & dD \endpmatrix \ .
$$
We now define a different tensor product, $\otimes_{op}$, by
$$
M \otimes_{op} m = \pmatrix
Aa & Ba & Ab & Bb \cr
Ca & Da & Cb & Db \cr
Ac & Bc & Ad & Bd \cr
Cc & Dc & Cd & Dd \endpmatrix \ ,
$$
the difference being that now the matrix which determines the coarse
structure of the tensor product is the second one. It is clear and
well known that $m \otimes M \neq M \otimes m$, independently of
whether $\mathcal B$ is commutative or not. However, we see that when
$\mathcal B$ is abelian, $m \otimes M = M \otimes_{op} m$. The
previous calculation tells us how to modify this result when
$\mathcal B$ is not commutative: calling $\mathcal B^{op}$ the same
ring with {\em opposite} multiplication (so that
$A\, ._{op} \, a = a.A$, for example), we obtain
$$
m(\mathcal B) \otimes M(\mathcal B) =
M(\mathcal B^{op}) \otimes_{op} m(\mathcal B^{op})
$$
where the notation
$M(\mathcal B^{op}) \otimes_{op} m(\mathcal B^{op})$ means that we
first take the opposite tensor product of the two matrices and
subsequently we multiply the matrix elements in the opposite order.
Suppose in addition that the ring $\mathcal B$ is endowed with a star
operation~$*$, and call $\dag$ the conjugation of matrices with
$\mathcal B$-entries. In the case of $2\times 2$ matrices, this reads
$$
{\pmatrix a & b \cr c & d \endpmatrix}^\dag \doteq
\pmatrix a^{*} & c^{*} \cr b^{*} & d^{*} \endpmatrix \ .
$$
So defined $\dag$ is antimultiplicative. Moreover, direct calculation
shows that
$$
(m \otimes M)^\dag = M^\dag \otimes_{op} m^\dag \ .
$$
When $\mathcal B$ is commutative, the previous right hand side can be
written simply $m^\dag \otimes M^\dag$.
\subsubsection{Tensor product of representations}
Now let $\mathcal A$ be an algebra. Take $\rho_{1}$ and $\rho_{2}$
two representations of $\mathcal A$ in vector spaces $V_1$ and $V_2$.
Then, once bases are chosen, $\rho_{1}(a)$ and $\rho_{2}(a)$, with
$a\in {\mathcal A}$, are two matrices with commutative entries.
It is clear that $\rho_{1} \otimes \rho_{2}$ is a representation of
the algebra ${\mathcal A}\otimes {\mathcal A}$, indeed, with
$a\otimes b \in {\mathcal A}\otimes {\mathcal A}$, we have
$$
[\rho_{1} \otimes \rho_{2}](a\otimes b) =
\rho_{1}(a) \otimes \rho_{2}(b) \ .
$$
However, this is {\em not} a representation of ${\mathcal A}$, unless
we have a coproduct (algebra homomorphism) from ${\mathcal A}$ to
${\mathcal A}\otimes {\mathcal A}$: using
$$
a \in {\mathcal A} \rightarrow \Delta a \doteq a_{1}\otimes a_{2}
\in {\mathcal A} \otimes {\mathcal A}
$$
one defines $\rho_{1} \otimes \rho_{2}$ as a representation of
${\mathcal A}$ by setting
$$
[\rho_{1} \otimes \rho_{2}][a] \doteq
[\rho_{1} \otimes \rho_{2}](\Delta a) \ .
$$
If $\mathcal A$ is a Hopf algebra, we are in such a situation. This
is what we assume from now on.
Now suppose that $\mathcal A$ has a star operation, and that
$(\rho_{1},\dag)$, $(\rho_{2},\dag)$ are star representations of this
Hopf algebra on modules $V_1$, $V_2$ (each one endowed with a scalar
product for which the adjoint is denoted by $\dag$). So, we have
$$
\rho_{1}(u^{*}) = (\rho_{1} (u))^\dag \qquad \mbox{and} \qquad
\rho_{2}(u^{*}) = (\rho_{2} (u))^\dag \ .
$$
We shall now suppose that the star is, somehow, compatible with the
Hopf structure. We shall discuss the Hopf star and twisted Hopf star
cases.
We first suppose that $*$ is a Hopf star. It then commutes with
$\Delta$, and
\begin{eqnarray*}
[\rho_{1} \otimes \rho_{2}][a^{*}] &=&
[\rho_{1} \otimes \rho_{2}](\Delta a^{*}) =
[\rho_{1} \otimes \rho_{2}](* \Delta a) =
[\rho_{1} \otimes \rho_{2}](a_{1}^{*}\otimes a_{2}^{*}) \cr
{} &=& \rho_{1}(a_{1}^{*}) \otimes \rho_{2}(a_{2}^{*}) =
(\rho_{1}(a_{1}))^\dag \otimes (\rho_{2}(a_{2}))^\dag =
(\rho_{1}(a_{1}) \otimes \rho_{2}(a_{2}))^\dag \cr
{} &=& ([\rho_{1} \otimes \rho_{2}](a_{1}\otimes a_{2}))^\dag =
([\rho_{1} \otimes \rho_{2}](\Delta a))^\dag \cr
{} &=& ([\rho_{1} \otimes \rho_{2}][a])^\dag
\end{eqnarray*}
Therefore, $\rho_{1}\otimes \rho_{2}$ is also a $*$-representation.
We now suppose that $*$ is a twisted Hopf star. It no longer commutes
with $\Delta$ but intertwines it with the opposite coproduct
$\Delta^{op}$. In this case
\begin{eqnarray*}
[\rho_{1} \otimes \rho_{2}][a^{*}] &=&
[\rho_{1} \otimes \rho_{2}](\Delta a^{*}) =
[\rho_{1} \otimes \rho_{2}](* \Delta^{op} a) =
[\rho_{1} \otimes \rho_{2}](a_{2}^{*}\otimes a_{1}^{*}) \cr
{} &=& \rho_{1}(a_{2}^{*}) \otimes \rho_{2}(a_{1}^{*}) =
(\rho_{1}(a_{2}))^\dag \otimes (\rho_{2}(a_{1}))^\dag =
(\rho_{1}(a_{2}) \otimes \rho_{2}(a_{1}))^\dag \cr
{} &=& ([\rho_{1} \otimes \rho_{2}](a_{2}\otimes a_{1}))^\dag =
([\rho_{1} \otimes \rho_{2}](\Delta^{op} a))^\dag \cr
{} &\neq& ([\rho_{1} \otimes \rho_{2}][a])^\dag
\end{eqnarray*}
\noindent
Therefore, $\rho_{1}\otimes \rho_{2}$ is {\sl not} a $*$-representation
for a twisted $*$. However, we have the possibility of defining ``another''
tensor product of representations\footnote{
When $\mathcal A$ is quasitriangular, we recall that the two
coproducts are related by an $R$-matrix as follows:
$\Delta^{op}(a) = R \Delta(a) R^{-1}$.
},
called $\otimes_{op}$, as follows:
$$
[\rho_{1}\otimes_{op} \rho_{2}][a] \doteq
[\rho_{1}\otimes \rho_{2}](\Delta^{op} a) \ .
$$
With this notation at hand, we can write
$$
[\rho_{1}\otimes \rho_{2}][a^{*}] =
([\rho_{1}\otimes_{op} \rho_{2}][a])^\dag \ .
$$
For this reason, ``true'' Hopf stars are usually preferred in mathematics,
as the category of $*$-representations is closed under tensor product.
Another possibility, the one employed in CFT's, is to truncate tensor
products (see Section~\ref{s:discussion}). Star representations are
closed under this truncated tensor product ---for both types of stars.
\subsubsection{Hopf action on vectors with non commutative elements}
We now suppose that $\rho_{1}$ and $\rho_{2}$ are no longer complex
matrices but matrices with elements taken in a star algebra $\mathcal
B$. We still assume that we have a left action, in the sense
$\rho_{i}(ab) = \rho_{i}(a) \rho_{i}(b)$, but this is not a
representation in the usual sense. As before we assume that $\mathcal
A$ is endowed with a star operation and that $(\rho_{i},\dag)$ are
star representations, in the sense $\rho(a^*) = (\rho(a))^\dag$,
where $\dag$ transposes the matrix $\rho(a)$ and takes the conjugate
(in $\mathcal B$) of each element.
If we suppose that $*_\mathcal{A}$ is a Hopf star, then a direct
calculation shows that
$$
[\rho_{1}\otimes \rho_{2}][a^{*}] =
([\rho_{1}^{op} \otimes \rho_{2}^{op} ][a])^\dag \ .
$$
Usually, for $\mathcal{B} = {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$, we have $\rho^{op} = \rho$, but
this is not so in general; the upper index ``op'' in $\rho^{op}(a)$
reminds us that we should use the opposite multiplication of
$\mathcal B$ when making product of matrices such as $\rho^{op}(a)$.
If we take instead a twisted Hopf star, the conclusion is now:
$$
[\rho_{1}\otimes \rho_{2}][a^{*}] =
([\rho_{1}^{op} \otimes_{op} \rho_{2}^{op} ][a])^\dag \ .
$$
\newpage\sect{Invariant scalar products}
\label{s:invariant_scalar_products}
\subsection{Compatibility with Hopf stars}
\label{s:invariant_scalar_product_for_Hopf}
Defining the notion of an invariant scalar product $(\cdot,\cdot)$ on
a representation space $V$ of a quantum group $H$ is not as
straightforward as in the classical case. We want the scalar product
to commute with the action
of the Hopf algebra, in the appropriate sense. However, in order to
get a relation which needs to be checked only on the quantum group
generators, we want this condition to be a (linear) homomorphism in
the $H$ variable. Given that the scalar product is antilinear in one
of its variables, there are two ways of achieving this\footnote{ In
the ``classical case'' (real form of some Lie algebra), both formulae
read $(z,h \triangleright w) + (h \triangleright z, w) = 0$.},
\begin{equation}
\epsilon(h) (z,w) =
((*\,S\, h_1) \triangleright z , h_2 \triangleright w)
\label{invariant_scalar_product}
\end{equation}
or
\begin{equation}
\epsilon(h) (z,w) =
((S* h_1) \triangleright z , h_2 \triangleright w) \ .
\label{invariant_scalar_product_2}
\end{equation}
\noindent
We refer the reader to \cite{CoGaTr-l} for a more detailed discussion.
As the Hopf star doesn't commute with the antipode, since
$S \, * = * \, S^{-1}$, (\ref{invariant_scalar_product}) and
(\ref{invariant_scalar_product_2}) are, in general, two different
conditions.
For the scalar product to be invariant in the sense of equation
(\ref{invariant_scalar_product}), one only needs the quantum group
action to be given by a $*$-{\sl representation}\footnote{ If the
action of $h$ is implemented by a linear operator $\rho[h]$ on $V$,
this condition simply reads $\rho[h^{*}]= (\rho[h])^\dag$ where
$\dag$ is the usual adjoint operator.}:
\begin{equation}
(h \triangleright z , w) = (z , h^* \triangleright w) \ .
\label{star_representation}
\end{equation}
\noindent
Notice that (\ref{star_representation}) implies
(\ref{invariant_scalar_product}) but not conversely. In the same way
the alternative requirement
$(h \triangleright z , w) = (z , S^2(h^*) \triangleright w)$ implies
that condition (\ref{invariant_scalar_product_2}) is satisfied.
However, in our examples, we will choose to work with
$*$-representations, and therefore with invariant scalar products in
the sense (\ref{invariant_scalar_product}).
Assuming a non-degenerate pairing between $H$ and its dual $F$, and
extending the notation $(\cdot,\cdot)$ to the following $F$-valued
sesquilinear map on $V\otimes F$
$$
(v \otimes f,w \otimes g) \doteq (v,w) f^* g \ , \qquad v,w\in V, \
f,g \in F
$$
we may write the previous equations in the dual picture in a very
simple way. The first invariance condition reads
\begin{equation}
(\delta_R \, v, \delta_R \, w) = (v,w) \mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}_F \ , \nonumber
\end{equation}
whereas the $*$-representation requirement
(\ref{star_representation}) reads
\begin{equation}
(v,\delta_R w) = ((\mbox{\it id\,} \otimes S)\delta_R v,w) \ . \nonumber
\end{equation}
Again, this latter requirement implies the former.
Now let $\{ v_i \}$ be a basis of the vector space $V$, and call
$G_{ij} = (v_i,v_j)$ the corresponding metric. Moreover, define the
matrix of $h \in H$ in such a basis by
$h \triangleright v_i \doteq ||h||_{ji} v_j$.
{}From equation (\ref{star_representation}) it is now trivial to get
the matrix identities
\begin{equation}
||h||^\dag G = G ||h^*|| \ ,
\label{star_representation_in_basis}
\end{equation}
where $\dag$ denotes the transposed conjugate matrix. In particular, for
an orthonormal basis this reduces to $||h||^\dag = ||h^*||$.
\subsection{Compatibility with twisted Hopf stars}
The previous discussion
(Section~\ref{s:invariant_scalar_product_for_Hopf}) does not use the
fact that the chosen star should be a ``true'' Hopf star operation;
therefore, the same invariance conditions
(\ref{invariant_scalar_product}) and (\ref{invariant_scalar_product_2})
still apply in the twisted case. However now $S*=*S$, so that both
conditions coincide.
The invariance requirement is still automatically satisfied if the
representation of $H$ under study is a $*$-representation (formula
(\ref{star_representation})). However, now the dual formulas are
slightly different, due to the absence of the antipode in the duality
(\ref{twisted_dual_*}). The scalar product will be called invariant if
\begin{equation}
((\mbox{\it id\,}\otimes S)\delta_R \, v, \delta_R \, w) = (v,w) \mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}_F \ , \nonumber
\end{equation}
and the (co)representation will be a $*$-(co)representation if
\begin{equation}
(v,\delta_R w) = (\delta_R v,w) \ . \nonumber
\end{equation}
\noindent
Selecting a basis of $V$ we can write, exactly as in the untwisted case,
$||h||^\dag G = G ||h^*||$ for any $h \in H$.
\subsection{Quantum metric and quantum symplectic form on
$M(2,\mathcal{F})$}
\medskip
\noindent
{\bf Untwisted case}
\medskip
The $q$-deformed symplectic form in two dimensions (one may call it
the $q$-deformed epsilon tensor) is given by the matrix $$
\Sigma \doteq \left( \begin{array}{cc} 0 & q^{-1/2} \\
-q^{1/2} & 0 \end{array} \right) \ .
$$
\noindent
In fact the $*$-representation condition implies for the true Hopf
star case the equation
$$
T^\dag \, \Sigma \, T = \Sigma \ .
$$
Here
$$
T \doteq \left( \begin{array}{cc} a & b \\ c & d \end{array} \right)
$$
is the multiplicative matrix of generators of the quantum function
group $SL_{q}(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$, and the $\dag$ operation corresponds to
applying $*$ to the elements and transposing the matrix:
$$
T^\dag \doteq \left( \begin{array}{cc} a^* & c^* \\ b^* & d^*
\end{array} \right) \ .
$$
Notice that the above equation is different from
(\ref{star_representation_in_basis}) as there is a duality involved,
there $h \in H, \: ||h||_{ij} \in {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$, whereas here $T_{ij} \in F$.
Using the star (\ref{Hopf_star_on_F}) corresponding to $SL_{q}(2,{\mbox{\rm I}\hskip-2pt \mbox{\rm R}})$
and fixing a global factor by requiring hermiticity of $\Sigma$, we
finally obtain the ``invariant metric'' given above.
\medskip
\noindent
{\bf Twisted case}
\medskip
\noindent
Now, as the duality between the star on a Hopf algebra and its dual
differs from the one in the untwisted case, the $*$-representation
condition implies the relation
$$
(ST)^\dag \, \Sigma \, T = \Sigma \ ,
$$
where $S$ is the antipode. Taking the twisted conjugacy
$a^*=a\,,\ b^*=\pm c\,,\ c^*=\pm b$ and $d^*=d$ we get the metric
$$
\Sigma_\pm \doteq
\left( \begin{array}{cc} 1 & 0 \\ 0 & \pm 1 \end{array} \right) \ ,
$$
as we would expect in a (twisted) $SU(2)$ and $SU(1,1)$ case,
respectively.
\subsection{Invariant scalar products for $\mathcal H$ endowed with a
Hopf star}
\subsubsection{Invariant scalar products on the indecomposable
representations of $\mathcal H$}
This was worked out in Appendix E of \cite{CoGaTr-e}. Here we repeat
the expressions for the matrices of scalar products $G$ just for
completeness and to ease the comparison with the twisted case.
Technically this is done by solving the set of linear equations
(\ref{star_representation_in_basis}) for the coefficients $G_{ij}$
taking $h=X_{\pm}$ and $K$, and imposing hermiticity of $G$. Each entry
in the list below corresponds to an indecomposable representation,
remember that $3_{irr}$ is projective and irreducible, whereas $6_{odd}$
and $6_{eve}$ are projective indecomposable (with corresponding
irreducible representations of dimensions $1$ and $2$, respectively). We
only single out the following salient features (notice that $G$ is
always given up to an overall normalization factor):
\medskip
\begin{itemize}
\item[$\bullet \: \mathbf{3_{irr}}$]
We get
{\small
$$
G = \pmatrix 0& 0& -q^2\cr 0& 1& 0\cr -q& 0& 0
\endpmatrix \qquad \text{and}
\qquad \sigma = (++-) \ .
$$}
\noindent
The index of $G$ (maximal dimension of each of the two maximally
isotropic subspaces) is therefore $1$, and the Witt decomposition
reads $3 = 1 + 1 + 1$.
\medskip
\item[$\bullet \: \mathbf{6_{odd}}$]
With $\beta \in {\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$ we have here
{\small
$$
G = \pmatrix 0& 0& 0& q& 0& 0\cr
0& 0&-q& 0& 0& 0\cr
0&-q^2& 0& 0& 0& 0\cr
q^2& 0& 0& 0& 0& 0\cr
0& 0& 0& 0& \beta& 1\cr
0& 0& 0& 0& 1& 0 \endpmatrix \:
\sim {\it Diag}(1,1,-1,-1,\lambda_+,\lambda_-) \ ,
$$}
\noindent
with $\lambda_+ > 0, \lambda_- < 0$. $G$ is neutral, as its signature
is $\sigma = (+++ - - -)$. The index of $G$ is $3$ and the Witt
decomposition reads $6 = 3 + 3$.
\medskip
\item[$\bullet \: \mathbf{5_{odd}}$]
Taking $\beta,\gamma \in {\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$, $g \in {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$,
{\small
$$
G = \pmatrix 0& 0& iq\gamma& g& 0\cr
0& 0& -q^2 \bar g& iq\beta& 0\cr
-i q^2 \gamma& -qg& 0& 0& 0\cr
\bar g& -iq^2 \beta& 0& 0& 0\cr
0& 0& 0& 0& 0 \endpmatrix \:
\sim {\it Diag}(\lambda_+,-\lambda_+,\lambda_-,-\lambda_-,0)
$$}
\noindent
and $\sigma = (++--0)$.
\medskip
\item[$\bullet \: \mathbf{3_{odd}}$]
Now
{\small
$$
G = \pmatrix 0& iq& 0\cr -iq^2& 0& 0\cr 0& 0& 0 \endpmatrix
\qquad \text{and} \qquad \sigma = (+-0) \ .
$$}
\medskip
\item[$\bullet \: \mathbf{6_{eve}}$]
The metric should be ($\beta \in {\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$)
{\small
$$
G = \pmatrix 0& 0& iq\beta& -iq& 0& 0\cr
0& 0& -iq& 0& 0& 0\cr
-iq^2 \beta& iq^2& 0& 0& 0& 0\cr
iq^2& 0& 0& 0& 0& 0\cr
0& 0& 0& 0& 0& i\cr
0& 0& 0& 0& -i& 0 \endpmatrix \ ,
$$}
\noindent
with a signature $\sigma = (+++---)$ for any $\beta$. As in
the $6_{odd}$ case, $G$ is neutral with an index of $3$, and
the Witt decomposition reads $6 = 3 + 3$.
\medskip
\item[$\bullet \: \mathbf{4_{eve}}$]
Having $\alpha, \beta \in {\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$, and $g \in {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$ we may write
{\small
$$
G = \pmatrix 0& 0& 0& 0\cr
0& 0& 0& 0\cr
0& 0& \alpha& g\cr
0& 0& \bar g& \beta \endpmatrix
$$}
\noindent
Now its signature obviously depends on the parameters.
\medskip
\item[$\bullet \: \mathbf{3_{eve}}$]
Here we have simply $G = {\it Diag}( 0, 0, 1)$, and $\sigma = (+00)$.
\medskip
\item[$\bullet \: \mathbf{2_{eve}}$]
In this case
{\small
$$
G = \pmatrix 0& iq\cr -iq^2& 0
\endpmatrix \sim {\it Diag}(1,-1) \ .
$$}
\end{itemize}
\subsubsection{Invariant scalar products on $\mathcal M$}
\label{s:inv_scalar_product_on_M}
It can be seen that for $N=3, q^3=1$ the reduced quantum plane
$\mathcal M$, a module algebra for $\mathcal H$, is isomorphic as an
algebra to the matrix algebra $M(3,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$, whereas as a vector space
splits into the sum of three {\it unequivalent} indecomposable
representations, namely
$\mathcal M \sim 3_{irr} \oplus 3_{eve} \oplus 3_{odd}$.
Actually, this feature can be generalized for all $N$ odd, $q^{N}=1$.
The corresponding quantum plane (which is now isomorphic with
$M(N,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$) splits into the sum of $N$ {\it unequivalent}
indecomposable representations of $\mathcal H$. One of them is the
irreducible $N_{irr}$, and the others are analogous to the
``intermediate modules'' that appear within each lattice of
submodules associated to the other $N-1$ projective indecomposable
modules of $\mathcal H$. This property was proven in
\cite{Coquereaux-Schieber}.
A word of warning seems here necessary: the algebra $M(N,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$
plays an ubiquitous role here. Indeed, on one hand it is isomorphic
with a simple subalgebra of $\mathcal H$ (see the structure of the
regular representation given in \ref{a:structure_of_H}). As such, its
underlying vector space splits into a sum of $N$ subspaces carrying
equivalent representations (all equivalent to the $N_{irr}$), appearing
in the decomposition of the regular representation in projective
indecomposable modules. In this way $M(N,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$ appears as an algebra
and as a module, but not as a module algebra (considering
$Z,W\in M(N,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}) \subset \mathcal{H}$ and $X\in \mathcal{H}$, in general
$X(ZW) \neq (X_1 Z)(X_2 W)$). On the other hand, $M(N,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$ is also
isomorphic with the reduced quantum plane, and as such it is a module
algebra, but not a subalgebra of $\mathcal H$ anymore. Its decomposition
under the action of $\mathcal H$ is now more subtle, since it reads
$M(N,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}) \sim N_{irr} \oplus
N_{1} \oplus N_{2} \oplus \cdots \oplus N_{N-1}$.
Because of this last result, one could be tempted to think that the
most general scalar product on the reduced quantum plane $\mathcal M$
is simply given by the direct sum of its restrictions to the modules
$3_{irr}$, $3_{eve}$ and $3_{odd}$ that are already known\ldots but
it is not so. Indeed, non diagonal blocks may appear as we can have
non-zero projections amongst vectors of different indecomposable
representations.
Being $\mathcal M$ not only a module but a module algebra, we impose
condition (\ref{star_representation}) for the left action of
$\mathcal M$ on itself given by multiplication as well. This singles
out a unique invariant hermitian form $(\:,\,)$, up to an overall
scaling factor. Its structure was studied in Section~$5.5$ of
\cite{CoGaTr-e} and goes as follows: the only non zero scalar products
are those of the type $(x^r y^s,x^p y^t)$ with $r+p=s+t=2$, and they
are all determined by setting $(xy,xy)=1$. The signature of this metric
is $(5+,4-)$, so its index is $4$ and the Witt decomposition reads
$9=4+4+1$. In the basis
$\{\{ x^2,xy,y^2 \},\{ x,y,x^2 y^2 \},\{ \mbox{\rm 1}\hskip-2.8pt \mbox{\rm l},x^2 y,xy^2 \}\}$
the scalar product can be written as
$$
G = \pmatrix B & 0 & 0 \cr
0 & 0 & B \cr
0 & B & 0 \endpmatrix \ ,
$$
where $B$ is the $3 \times 3$ block
$$
B = \pmatrix 0 & 0 & q^2 \cr
0 & 1 & 0 \cr
q & 0 & 0 \endpmatrix \ .
$$
The restriction of this scalar product to the subspace $3_{irr}$
coincides with what was already obtained before, a form of signature
$(2+,1-)$. The restriction to the subspaces $3_{eve}$ and $3_{odd}$ is
actually totally degenerate, so the conclusion we find for $\mathcal M$
does not contradict what was already obtained for $3_{eve}$ and
$3_{odd}$ (just choose an overall scaling factor equal to $0$ in the
latter cases).
\subsubsection{Invariant scalar products on the regular representation
of $\mathcal H$}
\label{s:scalar_product_on_H}
One should not be tempted to think that the most general hermitian
scalar product on $\mathcal{H}$ $(N=3)$ itself is simply given by
its restrictions to the direct sum
$3 [3_{irr}]\oplus 2 [6_{eve}] \oplus 1 [6_{odd}]$ since we may
very well accept ``off-block'' components. As a matter of fact, the
constraints in this case are rather weak:
for any given star, any hermitian form such
that $(X^{*} Y, Z) = (Y, XZ)$ will work, but such a form is totally
determined by the values of $(\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}, X_{+}^a X_{-}^b K^c)$. Since we
have $N^{3}$ terms, we see at once that the most general invariant
scalar product on $\mathcal H$ will depend on $N^{3}$ parameters
(real ones, due to the hermiticity of the scalar product).
If one really wants to obtain an explicit expression for the possible
metrics $G$'s, in the case $N=3$, the thing to do is to write
explicitly $X_{\pm}$ and $K$ as $27 \times 27$ matrices (this is
numerically easy, once we know how to write these generators in
$M(3,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}) \oplus \Mlo{2|1}$; this was done in \cite{Coquereaux} and
recalled in \cite{CoGaTr-e}) and solve the equations
(\ref{star_representation_in_basis}), $||h||^\dag G = G ||h^*||$,
for the coefficients $G_{ij}$ where $h=X_{\pm}$ and $K$.
One can then check that this set of equations indeed lead to a
solution depending on $27$ parameters.
Because of this pretty big number of free parameters, the signature
can be rather arbitrary. This is a slightly dispointing result since
we are looking for some kind of constraint(s) that more or less fix
the hermitian form. It would also be nice if the structure of this
scalar product could somehow reflect the algebraic structure of
$\mathcal H$ itself. As we shall see later, this goal will be
achieved by the choice of a particular scalar product that we call
the ``hermitian Killing form''. Yet another interesting scalar product
on the regular representation can be defined by using the existence
of left (or right) invariant integrals
(see Section~\ref{s:scalar_products_with_integrals}).
\subsection{Invariant scalar products for $\mathcal H$ with a twisted
star}
\subsubsection{Invariant scalar products on the indecomposable
representations of $\mathcal H$}
\label{s:twisted_metrics_on_H_representations}
As it was done for the true Hopf star in Appendix~E of
\cite{CoGaTr-e}, we here show the most general metric on the vector
space of each of the indecomposable representations of $\mathcal H$
using the twisted stars (\ref{twisted_star_on_H}). Since we have two
possible choices, the $\pm$ signs below correspond respectively
to the $\pm$ possibilities defined in (\ref{twisted_star_on_F}),
(\ref{twisted_star_on_H}), (\ref{twisted_star_on_M}). We restrict
the inner product to be a quantum group invariant one, as defined
in Section~\ref{s:invariant_scalar_product_for_Hopf}. On each
representation space we use the basis obtained from appropriate
restrictions of the natural basis (``elementary basis'') associated
with the regular representation of $\mathcal H$ as given in
\ref{a:structure_of_H}. For each indecomposable representation we give
an explicit expression of the most general covariant metric in this
particular base and we calculate its signature.
\medskip
\begin{itemize}
\item[$\bullet \: \mathbf{3_{irr}}$]
Up to a real global normalization the metric is
$$
G = {\it Diag}(1,\mp 1, 1) \ , \text{with signature}\ \sigma = (++\mp) \ .
$$
\medskip
\item[$\bullet \: \mathbf{6_{odd}}$]
Now we get the metric ($\beta \in {\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$)
{\small
$$
G = \pmatrix 1& 0& 0& 0& 0& 0\cr
0& \pm 1& 0& 0& 0& 0\cr
0& 0& \pm 1& 0& 0& 0\cr
0& 0& 0& 1& 0& 0\cr
0& 0& 0& 0& \beta& 1\cr
0& 0& 0& 0& 1& 0 \endpmatrix \ .
$$}
\noindent
A change of basis tells us that
$G \sim {\it Diag}(1,1,\pm 1,\pm 1,\lambda_+,\lambda_-)$, with $\lambda_+
> 0, \lambda_- < 0$. Thus, the signature is $$
\sigma = (+++\pm \pm -) \ .
$$
\medskip
\item[$\bullet \: \mathbf{5_{odd}}$]
If $\alpha,\beta \in {\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$ and $g \in {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$ we may write the metric as
{\small
$$
G = \pmatrix \alpha& g& 0& 0& 0\cr
\bar g& \beta& 0& 0& 0\cr
0& 0& \pm\alpha& \pm g& 0\cr
0& 0& \pm\bar g& \pm\beta& 0\cr
0& 0& 0& 0& 0 \endpmatrix
$$}
\noindent
Given that
$G \sim {\it Diag}(\lambda_1,\lambda_2,\pm\lambda_1,\pm\lambda_2,0)$, with
arbitrary $\lambda_i \in {\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$, its signature may be anything between
$\sigma = (++++0)$ and $\sigma = (----0)$.
\medskip
\item[$\bullet \: \mathbf{3_{odd}}$]
Here
$$
G = {\it Diag}( 1, \pm 1, 0) \qquad \text{and} \ \sigma = (+\pm 0) \ .
$$
\medskip
\item[$\bullet \: \mathbf{6_{eve}}$]
Up to a normalization the metric can be written ($\beta \in {\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$)
{\small
$$
G = \pmatrix \beta& 1& 0& 0& 0& 0\cr
1& 0& 0& 0& 0& 0\cr
0& 0& \pm\beta& \pm 1& 0& 0\cr
0& 0& \pm 1& 0& 0& 0\cr
0& 0& 0& 0& \mp 1& 0\cr
0& 0& 0& 0& 0& -1 \endpmatrix
$$}
\noindent
The signature is clearly
$$
\sigma = (++\mp - - -) \ .
$$
\medskip
\item[$\bullet \: \mathbf{4_{eve}}$]
In this case the result coincides with the one obtained using the
normal Hopf star, as we get the metric
($\alpha, \beta \in {\mbox{\rm I}\hskip-2pt \mbox{\rm R}}$, $g \in {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$)
{\small
$$
G = \pmatrix 0& 0& 0& 0\cr
0& 0& 0& 0\cr
0& 0& \alpha& g\cr
0& 0& \bar g& \beta \endpmatrix
$$}
\noindent
The non-null block in $G$ is an arbitrary hermitian matrix,
therefore the signature is not fixed.
\medskip
\item[$\bullet \: \mathbf{3_{eve}}$]
As in the untwisted case, here we find simply
$$
G = {\it Diag}( 0, 0, 1) \ .
$$
\medskip
\item[$\bullet \: \mathbf{2_{eve}}$]
This irreducible representation has the metric
$$
G = {\it Diag}( 1, \pm 1) \ .
$$
Notice that a positive definite form is obtained for the twisted Hopf
star of $SU(2)$ type.
\end{itemize}
\subsubsection{Invariant scalar product on $\mathcal M$}
A priori, one could think that the discussion goes along the lines of
Section~\ref{s:inv_scalar_product_on_M} and that nothing much should be
changed. This is almost so, in the sense that invariance implies that
the only {\it possibly non zero} scalar product of type $(\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l},z)$ is
$(\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l},x^{2}y^{2})$. However, we shall show that this quantity vanishes
as well (the proof uses the left action of $\mathcal{H}$ on
$\mathcal{M}$ as discussed in Table~$1$, Section~$4.4$ of
\cite{CoGaTr-e}). Indeed
\begin{eqnarray*}
(\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l},x^{2}y^{2}) &=& (x^{2},y^{2}) = q^{-1}(x^{2},Ky^{2})
= q^{-1}(K^{*}x^{2},y^{2}) \\
{} &=& q^{-1}(K^{-1}x^{2},y^{2}) = q^{-1}(qx^{2},y^{2}) \\
{} &=& q^{2}q^{-1}(x^{2},y^{2}) = q(\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l},x^{2}y^{2})
\end{eqnarray*}
Hence $(\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l},x^{2}y^{2}) = 0$, and we see that the bilinear form obtained
on $\mathcal{M}$ is totally degenerate. This result contrasts
drastically with the one obtained in the untwisted Hopf star case.
\subsubsection{Invariant scalar products on the regular representation
of $\mathcal H$}
We refer to Section~\ref{s:scalar_product_on_H} for the general discussion
and to the next section for a study of very specific scalar products on
this representation space.
\newpage\sect{Scalar products on the left regular representation of a Hopf
algebra}
\label{s:scalar_products_on_left_rr}
\subsection{The hermitianized Killing form}
\label{s:hermitianized_Killing_form}
As is recalled in \ref{a:Killing_form}, in the case of Hopf algebras
there is still a notion of a Killing form, which generalizes this
particular bilinear form found in the case of Lie groups and algebras.
Moreover, it is also invariant under an adequate generalization of the
adjoint action of a group on itself, now a left action of a Hopf algebra
on itself.
This Killing form $(.\,,.)_{u}$ is neither symmetric nor hermitian
(actually we did not use any star in its definition), but,
given an arbitrary star operation on $H$, we now define
a sesquilinear form on $H \times H$ by\footnote{
As for the Killing form, there is always an implicit choice of
representation of $H$, here the left regular one.
}
\begin{equation}
(X,Y) \doteq (X^*,Y)_u = {\rm Tr}_q(X^*Y) \ , \qquad X,Y \in H \ .
\end{equation}
This new form is obviously $H$-invariant ---in the sense of
(\ref{star_representation})--- under the left action of $H$ on itself
given by simple multiplication, as
\begin{equation}
(XY,Z) = {\rm Tr}_q(Y^* X^* Z) = (Y,X^* Z) \ .
\label{multiplication_invariance_of_hermitianized_Killing}
\end{equation}
\noindent
The ``symmetry'' property of the Killing form (\ref{symmetry_of_Killing})
gets traduced now in
$$
(Y,X) = (X^*, S^2(Y^*)) \ .
$$
\noindent
In addition, {\it if} the star operation is a true Hopf one, the
invariance of $(\,,)_u$ under the adjoint action
(\ref{invariance_of_Killing}) implies that
\begin{equation}
\left( ad_{(SZ_1)^*}(X), ad_{Z_2}(Y) \right) =
\left( X,Y \right) \, \epsilon (Z) \ .
\label{adjoint_invariance_of_hermitianized_Killing}
\end{equation}
\noindent
This is so because $[ad_{(SZ)^*}(X)]^* = ad_{Z}(X^*)$ for a Hopf star.
Note that both properties
(\ref{multiplication_invariance_of_hermitianized_Killing}) and
(\ref{adjoint_invariance_of_hermitianized_Killing}) are invariances of
this Killing scalar product in the sense of
(\ref{invariant_scalar_product}), but with respect to different actions.
Actually, we also have for this action a $*$-representation, as it is
true that
\begin{equation}
\left( ad_Z(X), Y \right) = \left( X,ad_{Z^*}(Y) \right) \ .
\nonumber
\label{*-representation_of_hermitianized_Killing}
\end{equation}
Finally, when the star involved in this definition is a Hopf star, the
resulting form is ---or can always be chosen to be--- hermitian; we call
it the ``hermitianized Killing form'' or the ``Killing scalar product''
(we will see later that this is not the case when one uses a twisted star).
Indeed, as we are working with a $*$-representation,
${\rm Tr}[h^*] = \overline{ {\rm Tr}[h]} \,,\ h \in H$. Therefore,
\begin{eqnarray*}
(X,Y) &=& {\rm Tr}(u\,X^* Y) = \overline{{\rm Tr}(Y^* X u^*)} \\
&=& \overline{{\rm Tr}(u^* Y^* X)} \ ,
\end{eqnarray*}
and
\begin{eqnarray}
(X,Y) &=& \overline{(Y,X)}
\end{eqnarray}
if $u^* = u$. Using the notation of \ref{a:Killing_form}, we know that
$S^2(h) = u h u^{-1}$ implies, for a Hopf star,
$S^2(h) = u^* h {(u^*)}^{-1}$. Both equations together tell us that
$u^{-1} u^*$ is a central element which, being a matrix on a
representation space, should be proportional to the identity. Moreover
the proportionality factor must be a phase ($(u^*)^* = u$), and this
may always be absorbed in $u$ to have an hermitian form.
\subsubsection{The Killing scalar product for $\mathcal{H}$. Hopf star case.}
\label{s:Killing_scalar_product_on_H}
We have just defined a particular scalar product based on the Killing
form on the regular representation of a quantum group $H$. We analyze
here the case of the finite Hopf algebra $\mathcal H$, taking $N=3$, and
we choose a Hopf star operation. Then
$$
(X,Y) = {\rm Tr}_q(X^*Y) = {\rm Tr}(K^{-1} X^*Y) \ , \qquad X,Y \in \mathcal{H} \ .
$$
In this case, the structure of the corresponding $27\times 27$ hermitian
matrix $G$ in the PBW-basis is not very transparent and we shall not
give it explicitly, although its signature can be read off easily.
However, the expression of $G$ in what we called the ``elementary basis''
is quite remarkable. Here it goes:
\begin{itemize}
\item Its restriction to the $M(3,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$ block, with basis ordering
$$
\{E11,E12,E13,E21,E22,E23,E31,E32,E33\} \ ,
$$
reads:
$$
3 \pmatrix
0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & q^{-1} \\
0 & 0 & 0 & 0 & 0 & 0 & 0 & -q^{-1} & 0 \\
0 & 0 & 0 & 0 & 0 & 0 & q^{-1} & 0 & 0 \\
0 & 0 & 0 & 0 & 0 & -1 & 0 & 0 & 0 \\
0 & 0 & 0 & 0 & 1 & 0 & 0 & 0 & 0 \\
0 & 0 & 0 & -1 & 0 & 0 & 0 & 0 & 0 \\
0 & 0 & q & 0 & 0 & 0 & 0 & 0 & 0 \\
0 & -q & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\
q & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0
\endpmatrix \ .
$$
\item Its restriction to the subspace $\{A11,A12,A21,A22\}$ of the
$\Mlo{2|1}$ block reads
$$
6 \pmatrix
0 & 0 & 0 & q \\
0 & 0 & -q & 0 \\
0 & -q^{-1} & 0 & 0 \\
q^{-1} & 0 & 0 & 0
\endpmatrix \ .
$$
\item Its restriction to the subspace $\{A33\}$ of the $\Mlo{2|1}$
block reads
$$
6 \pmatrix 1 \endpmatrix \ .
$$
\end{itemize}
\noindent
All other entries vanish. This is in particular so for the scalar
products mixing the three aforementioned subspaces. Also vanish
all scalar products between vectors belonging to the $13$
dimensional radical spanned by the generators
$$
\{B11,B12,B21,B22,P13,Q13,P23,Q23,P31,Q31,P32,Q32\} \ .
$$
In other words, $G$ is completely degenerated\footnote{
The fact that the trace of the adjoint map vanishes on the radical,
a result slightly weaker that the one reported here, was separately
observed by \cite{Kastler}.
}
in the direction of the Jacobson radical of $\mathcal H$, and it does
not mix the different simple components of the semi-simple part
$\overline{\mathcal H} \doteq M(3,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}) \oplus M(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}) \oplus {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$.
Moreover, we see at once that $G$, restricted to $\overline{\mathcal H}$
is diagonal in the (hence orthogonal) basis
\begin{eqnarray*}
& &\{ E11 \pm q^{-1} E33, E12 \pm q^{-1} E32, E13 \pm q^{-1} E31,
E21 \pm q^{-1} E23, E22 \,; \\
& & \qquad A11 \pm q^{-1} A22, A12 \pm q^{-1} A21 \,;\, A33 \} \ ,
\end{eqnarray*}
where it actually reads:
$$
G = 3 \, {\it Diag} \, (\pm 1,\pm 1, \pm 1, \pm 1, 1\,;
\, \pm 2, \pm 2 \,;\, 1) \ .
$$
\noindent
The signature of the restriction of $G$ to $\overline{\mathcal H}$
reads therefore $(8+,6-)$, but it is better to write it (with obvious
notations) as
$$
\left[4 (+1,-1) \oplus (+1) \right] \oplus
\left[ (+1,-1) \oplus (+1,-1) \right] \oplus (+1) \ .
$$
\subsubsection{Incompatibility between a Killing scalar product
and a twisted Hopf star}
Here we can follow a discussion along the same lines of the last part of
Section~\ref{s:hermitianized_Killing_form}, but now starting from
$S^2(h) = u h u^{-1}$ it is easy to deduce that
$S^2(h) = {(u^*)}^{-1} h u^*$. Both formulas together imply that
$u u^*$ is a central element, and this means that we will have
$u^* = cu^{-1} \neq u$ ($c \in H$ central).
Therefore, we can not expect to have an hermitian Killing scalar product
if the star is a twisted one. Having a true (hermitian) scalar product
is incompatible with the invariance of the Killing form.
\subsection{Scalar products related to invariant integrals}
\label{s:scalar_products_with_integrals}
We first gather general facts and definitions about left and right
invariant integrals on a Hopf algebra. We then use these
concepts ---together with a star operation--- to define a
particular hermitian scalar product on finite dimensional Hopf
algebras. All these notions are illustrated with our favorite
example $\mathcal H$.
\subsubsection{Integrals}
A left-invariant integral on a Hopf algebra $H$ over ${\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$ is a
linear map $\int_{L} : H \mapsto {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$ such that
$$
\left( \mbox{\it id\,} \otimes \int_{L}\right)\circ \Delta =
\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}_{H} \int_{L} \ ,
$$
where $\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}_{H}$ is the unit of $H$ and $\mbox{\it id\,}$ the identity map in $H$.
Therefore, for any $h \in H$ we have (as always
$\Delta h = h_1 \otimes h_2$)
\begin{equation}
h_1 \int_{L} h_2 = \mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}_{H} \int_{L} h \ .
\label{L_invariance_of_integral}
\end{equation}
\noindent
A right-invariant integral $\int_R$ is defined in the obvious similar
way.
Since $\int_{L}$ (or $\int_{R}$) is a linear object, it can be
identified with an element $\lambda_{L}$ (resp. $\lambda_{R}$) of the
dual $F$ of $H$. Such an element will therefore satisfy
$$
f \lambda_{L} = \epsilon(f) \lambda_{L}
$$
(or $\lambda_{R} f = \epsilon(f) \lambda_{R}$) for any $f \in F$.
Like for groups, a Hopf algebra $H$ is called {\it unimodular} if one
can find left and right integrals which coincide
($\int \doteq \int_{L} = \int_{R}$). Furthermore, such an integral
is called a Haar measure when it is normalizable and normalized, {\rm i.e.,\/}\
$\int(\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}_{H}) = 1$ (in particular $\int$ should not vanish on the
unit!).
We now go back to the example where $H$ is a reduced quantum
enveloping algebra of type $SL_{q}(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})$ at a root of unity,
$\mathcal H$. It is easy to see that here the left and right
integrals are respectively given (up to an overall constant) by
$$
\int_{L} = (X_{+}^{N-1} X_{-}^{N-1} K)^\star
$$
and
$$
\int_{R} = (X_{+}^{N-1} X_{-}^{N-1} K^{-1})^\star \ .
$$
\noindent
Here a particular vector space basis (PBW)
$\{X_{+}^{a}X_{-}^{b}K^{c}\}$ is chosen in $\mathcal H$ and
$\{(X_{+}^{a}X_{-}^{b}K^{c})^\star\}$ denotes its dual basis.
In terms of elements of $\mathcal F$, the same left and right
invariant integrals on $\mathcal H$ read
\begin{eqnarray*}
\lambda_{L} &=& (1 + a + \ldots + a^{N-1}) b^{N-1} c^{N-1} \\
\lambda_{R} &=& b^{N-1} c^{N-1} (1 + a + \ldots + a^{N-1}) \ .
\end{eqnarray*}
\noindent
These two integrals are not proportional and cannot be made equal;
$\mathcal H$ is therefore not unimodular and no Haar measure can be
defined. The dual $\mathcal F$ of $\mathcal H$ turns out to be
unimodular (see \cite{Dabrowski}), but the corresponding integral is
not a Haar measure because it is not normalizable as it vanishes on
the unit.
Further restricting now our class of examples to the case $N = 3$, it
is interesting to decompose the elements $X_+^2 X_-^2 K$ and
$X_+^2 X_-^2 K^{-1}$ on the elementary basis defined in the
\ref{a:structure_of_H}. They read respectively:
{\small
$$
\pmatrix
\pmatrix
q^{2} & 0 & 0 \cr
0 & 0 & 0 \cr
0 & 0 & 0
\endpmatrix & 0 \cr
0 & \pmatrix
-q \theta^{1}\theta^{2} & 0 & 0 \cr
0 & 0 & 0 \cr
0 & 0 & \theta^{1}\theta^{2}
\endpmatrix
\endpmatrix
$$
}
and
{\small
$$
\pmatrix
\pmatrix
q & 0 & 0 \cr
0 & 0 & 0 \cr
0 & 0 & 0
\endpmatrix & 0 \cr
0 & \pmatrix
-q^{2} \theta^{1}\theta^{2} & 0 & 0 \cr
0 & 0 & 0 \cr
0 & 0 & \theta^{1}\theta^{2}
\endpmatrix
\endpmatrix \ .
$$
}
\noindent On the other hand, using the PBW basis, the invariant
integral\footnote{
Notice that on the group ${\mbox{\rm Z}\hskip-5pt \mbox{\rm Z}}_{3} = \{\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}, K, K^2\}$
the integral is given by $\Sigma = 1 + K + K^{2}$.
}
on $\mathcal F$ can be expressed by duality as the element
$\sigma \doteq X_+^{2}X_-^{2} (1 + K + K^{2}) \in \mathcal{H}$.
\subsubsection{Scalar product on the left regular representation}
Using both a star operation (any) and an integral on $H$, we now
define a kind of Hopf algebra analogue of the familiar scalar product
used to discuss square integrable functions in usual complex
analysis. We take
\begin{equation}
(X,Y)_{L,R} \doteq \int_{L,R} X^{*} Y \ ,
\end{equation}
which is then automatically sesquilinear and invariant. In fact, by
construction this scalar product satisfies the $*$-representation
condition, as
$$
(ZX,Y) = (X,Z^* Y) \ .
$$
Here $H$ acts on itself by left-multiplication, and the invariance is
independent of the star chosen (twisted or not).
Other properties of this scalar product will of course depend upon
the kind of star used in its definition.
\paragraph{The Hopf star case}
\begin{itemize}
\item
To have hermiticity of our scalar product we need only to check that
$$
\int_{L,R} X^* = \overline{\int_{L,R} X} \ ,
$$
as $(Y,X) = \int{(X^*Y)^*}$ and
$\overline{\int{X^*Y}}=\overline{(X,Y)}$.
It is easy to see that the above property is compatible with the
left-invariance of this integral (contrarily to what will happen in
the twisted star case). Therefore one needs to check this explicitly
for each case, knowing that a left (or right) invariant integral on a
Hopf algebra is unique ---if it exists---, up to a scalar multiple.
We checked explicitly this property for the case of $H=\mathcal{H}$.
\item
{}From the invariance property of $\int_{L}$, one trivially gets
$$
\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}_H \, (X,Y) = X_1^* Y_1 \, (X_2,Y_2) \ .
$$
But this may also be interpreted ---as happens with the integral---
as an invariance with respect to the right action of $F$:
\begin{eqnarray*}
(X,Y)\triangleleft f &=& \epsilon(f) \, (X,Y) \\
&=& (X \triangleleft (Sf_1)^*\,,\, Y \triangleleft f_2) \ .
\end{eqnarray*}
This expression is the analogue of (\ref{invariant_scalar_product_2})
for a right action.
Recall that this is an extra invariance of the scalar product, as by
construction it is invariant under the left action of $H$ itself.
\item
In our example of $\mathcal H$, with $N=3$, this hermitian form
expressed in terms of the ``elementary basis'' defined in
\ref{a:structure_of_H} gives a $27 \times 27$ hermitian matrix
$G_{ij}$ that we describe now. Its restriction to the $9$-dimensional
subspace spanned by
$$
\{E_{11},E_{12},E_{13},E_{21},E_{22},E_{23},E_{31},E_{32},E_{33}\}
$$
reads
{\small
$$
\frac{1}{3} \pmatrix
0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 &q^{-1}\cr
0 & 0 & 0 & 0 & 0 & 0 & 0 &-q^{-1}& 0 \cr
0 & 0 & 0 & 0 & 0 & 0 &q^{-1} & 0 & 0 \cr
0 & 0 & 0 & 0 & 0 &-1 & 0 & 0 & 0 \cr
0 & 0 & 0 & 0 &1 & 0 & 0 & 0 & 0 \cr
0 & 0 & 0 &-1 & 0 & 0 & 0 & 0 & 0 \cr
0 & 0 &q & 0 & 0 & 0 & 0 & 0 & 0 \cr
0 &-q & 0 & 0 & 0 & 0 & 0 & 0 & 0 \cr
q & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \cr
\endpmatrix \ .
$$
}
\noindent
Its restriction to the $2(4)+2(1)=10$-dimensional subspace spanned by
$$
\{A_{11},B_{11},A_{12},B_{12},A_{21},B_{21},A_{22},B_{22},A_{33},B_{33}\}
$$
reads
{\small
$$
\frac{1}{3} \pmatrix
0 & 0 & 0 & 0 & 0 & 0 &-q &-q & 0 & 0 \cr
0 & 0 & 0 & 0 & 0 & 0 &-q & 0 & 0 & 0 \cr
0 & 0 & 0 & 0 &q &q & 0 & 0 & 0 & 0 \cr
0 & 0 & 0 & 0 &q & 0 & 0 & 0 & 0 & 0 \cr
0 & 0 &q^{-1} &q^{-1} & 0 & 0 & 0 & 0 & 0 & 0 \cr
0 & 0 &q^{-1} & 0 & 0 & 0 & 0 & 0 & 0 & 0 \cr
q^{-1} &-q^{-1} & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \cr
q^{-1} & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \cr
0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & -1 & 1 \cr
0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 1 & 0 \cr
\endpmatrix \ .
$$
}
\noindent
Finally, its restriction to the $8$-dimensional subspace spanned by
$$
\{P_{13},Q_{13},P_{23},Q_{23},P_{31},Q_{31},P_{32},Q_{32}\}
$$
reads
{\small
$$
\frac{1}{3} \pmatrix
0 & 0 & 0 &q & 0 & 0 & 0 & 0 \cr
0 & 0 &-q & 0 & 0 & 0 & 0 & 0 \cr
0 &-q^{2} & 0 & 0 & 0 & 0 & 0 & 0 \cr
q^{2} & 0 & 0 & 0 & 0 & 0 & 0 & 0 \cr
0 & 0 & 0 & 0 & 0 & 0 & 0 & 1 \cr
0 & 0 & 0 & 0 & 0 & 0 &-1 & 0 \cr
0 & 0 & 0 & 0 & 0 &-1 & 0 & 0 \cr
0 & 0 & 0 & 0 &1 & 0 & 0 & 0 \cr
\endpmatrix \ .
$$
}
All the other scalar products vanish.
First of all, we may notice at once that this hermitian form is not
degenerate (this sharply contrasts with the hermitianized Killing form
which is degenerate along the radical, as we saw previously).
Here, the signature is $(14+,13-)$. The $27$ eigenvalues themselves read:
$$
\frac{1}{3} \left\{ (1)_9, (-1)_8, (\beta)_2 ,
(-\beta^{-1})_2, (-\beta)_3, (\beta^{-1})_3 \right\} \ ,
$$
where $\beta = \frac{1 + \sqrt{5}}{2}$ is the golden number.
It is interesting to notice that, although non degenerate, the
restriction of this form to the $9+4+1=14$-dimensional semi-simple part
of $\mathcal H$ is positive definite (this part, isomorphic with the
matrix algebra $M(3,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;})\oplus M(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}) \oplus {\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$, as recalled in
\ref{a:structure_of_H}, is spanned by $E_{ij}$ and $A_{kl}$).
\end{itemize}
\paragraph{The twisted star case}
\begin{itemize}
\item
It is in general {\it not} hermitian. In fact, if we now write down
(\ref{L_invariance_of_integral}) for $h^*$ and conjugate that
equation, we get
$$
h_2 \, \overline{\int_L h_1^*} = \mbox{\rm 1}\hskip-2.8pt \mbox{\rm l} \, \overline{\int_L h^*} \ .
$$
If we assume that $\int_L h^* = \overline{\int_L h}$, the above
equation would tell us that $\int_L$ should also satisfy the {\it
right} invariance condition, which will not be generally true. For
instance, in the case of $\mathcal H$ we know that a biinvariant
integral does not exist. To obtain an hermitian scalar product we
could then add both integrals, $(X,Y) \doteq (\int_L + \int_R) X^* Y$,
but this one would not have any extra invariance property\ldots
\item
{}From the invariance property of $\int_{L}$ results
$$
\mbox{\rm 1}\hskip-2.8pt \mbox{\rm l}_H \, (X,Y) = X_2^* Y_1 \, (X_1,Y_2) \ ,
$$
which shows a left-right mixed behaviour.
\item
The example of $\mathcal H$, with $N=3$, is not particularly enlighting
since the obtained complex bilinear form is not hermitian but symmetric.
A numerical study of this $27 \times 27$ matrix in the elementary
basis defined in \ref{a:structure_of_H} shows that it is not degenerate
and that it is ``almost'' diagonal, in the sense that the only non
diagonal $G_{ij}$ entries are $G(A_{11},B_{11})$, $G(A_{12},B_{12})$,
$G(A_{21},B_{21})$, $G(A_{22},B_{22})$ and $G(A_{33},B_{33})$ together
with the corresponding symmetric coefficients. We however stress
again the fact that, using the twisted star, the scalar product is
not hermitian.
\end{itemize}
To conclude: the twisted Hopf star case is rather bad in this sense.
\newpage\sect{Discussion}
\label{s:discussion}
As it was mentioned in the Introduction, the parameter $q$ that appears
in many integrable and conformal models is often a primitive root of
unity, and such values are generally incompatible with the choice of a
compact real form on the quantum group (like $SU_q(2)$, for instance).
For this reason the stars on ``compact'' quantum groups that one may
define in the context of spin chains, for example, are twisted. The
discussion is however a bit subtle and we want to make the following
comments:
In the case of a spin chain of type $XXZ$, for instance (see
\cite{Pasquier-Saleur}, for example), one may start with the
{\sl usual} rotation group in three dimensions ---or with its double
cover $SU(2)$--- acting at each point of the chain. Another ingredient
is given by the choice of some (unitary) representation of this group,
for instance the fundamental ($s=1/2$). The Hilbert space of the model
is obtained as the $n$-th tensor product of this representation. The
hamiltonian of the model is given by a sum of interaction terms indexed
by a discrete label, each term being itself built in terms of
(hermitian) Pauli matrices. This hamiltonian is {\sl not}, in general,
invariant with respect to the rotation group since the physical system
is clearly not rotationally invariant. However, in some cases, one
notices that the same total hamiltonian commutes with the generators
of a (complex) quantum group, for instance $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$. We should
stress the fact that generators of $SU(2)$ act on the Hilbert space in
a way that is ``local'' (generators rotate the states independently at
each point of the chain), whereas $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ acts in a non local
way (this point of view was emphasized for instance in
\cite{Bernard-Leclerc}). Notice that hermiticity of the hamiltonian
---a Jones projector--- is clearly a required constraint, however this
property does not take place in a representation space for the quantum
group but in its commutant.
Both $SU(2)$ and $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ enter the discussion of the model and
both have two-dimensional representations, but the two related concepts
should not be confused. For physical reasons, it is clear that the
scalar product used on the Hilbert space of the model should not contain
vectors of negative norm; for this reason it should be a {\it bona-fide}
positive definite scalar product. The same Hilbert space could also be
built in terms of tensor products of the fundamental representation of
the quantum group $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$, for $q$ a root of unity; indeed, two
vector spaces over ${\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}$ of the same dimension are clearly isomorphic, as
vector spaces\ldots Nevertheless, in the usual construction the Hilbert
space of the model acquires its Hilbert structure from the scalar product
chosen on representations of $SU(2)$, not from the one chosen on the
representations of the quantum group. Actually, the authors of the
present paper do not see why such a choice should be performed at all;
they cannot exclude however that it may turn out to be be useful. What
is in any case clear, is that {\sl if} one wants to choose a scalar
product on the fundamental representation of $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ such that
it will induce the same (already given) positive scalar product on the
Hilbert space of the model, one has to suppose that the quantum group
is endowed with a star operation which is a twisted Hopf star of
$SU(2)$ type.
We should mention the papers \cite{Mack-Schomerus-1, Mack-Schomerus-2},
where a general study of quantum symmetries in quantum theory is done,
and where the {\sl choice} of twisted star operations is clearly made
right at the beginning. This was actually nothing else than a choice
(related to a way of defining a covariant adjoint for field operators),
and it was subsequently discovered\footnote{
Unpublished addendum by the same authors. We thank G. Mack for this
information.
}
that this choice was not unique and that it would have been also
perfectly possible to define adjoints for field operators after having
decided to use a ``true'' Hopf star operation.
In conformal theories, primary fields are associated with vectors of
highest weight in a representation of some affine algebra, and it was
observed long ago that the fusion table of such primary fields is
identical to the Clebsh Gordan table describing the tensor products of
irreducible representations of some quantum group ---the same quantum
group also appears, via its $6j$-symbols, in the equations describing
the duality properties of the conformal blocks. At this point, one
should stress that the representations of the quantum group that appear
in the associated fusion table are not to be confused with the
representations of the affine algebra. The two structures, although
related (in a way that is apparently not well understood yet, see
\cite{Alekseev-Faddeev}), are quite distinct and the discussion
involving the nature of the scalar product to be used in a given
representation space for the affine or Virasoro generators should
not be confused with the analysis of the scalar product(s) that one
can define on the modules of the emerging quantum group.
When the parameter $q$ is a root of unity, the representation theory
is quite subtle since indecomposable (but not irreducible)
representations of the quantum group appear. Actually, to obtain a
physically meaningful state space one has to choose a so-called
``truncated tensor product'', by selecting only those representations
for which the $q$-trace vanishes (one can also use the formalism of
quasi-Hopf algebras, see \cite{Mack-Schomerus-2}). It is a fact that
discussions involving quantum groups in conformal field theories usually
consider {\sl infinite dimensional} Hopf algebras (like $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$),
which are not ``good'' quantum groups when $q$ is a root of unity since
they are not quasi-triangular in the usual sense. At the contrary, the
{\sl finite dimensional} Hopf algebras that one can obtain from those
ones through division by an (infinite dimensional) Hopf ideal are not
semi-simple but they {\sl are} quasi-triangular: they possess (finite
dimensional) $R$-matrices. The category of representations of these
Hopf algebras is not a modular category (tensor products of irreducible
representations are not necessarily equivalent to direct sums of
irreducibles), but it is again possible to define truncated scalar
products in a very natural way. We conjecture that discussions involving
simultaneously rational conformal field theories and quantum groups
should be done in terms of such finite dimensional Hopf quotients of
the usual quantum enveloping algebras at roots of unity. A general
study of these topics stays outside the scope of the present paper
but we hope that our contribution concerning stars (twisted or not)
and scalar products, together with selected examples involving finite
dimensional Hopf algebra quotients of $U_q(sl(2,{\mbox{\rm C}\hskip-5.5pt\mbox{l} \;}))$ will be useful
in this respect.
\newpage\sect*{Acknowledgements}
A. G. wishes to thank the CPT for the warm hospitality.
R. T. thanks CONICET, ICTP and the Univ. d'Aix Marseille I for financial
support. Also the kind hospitality extended to him at the CPT is
gratefully acknowledged.
|
\section*{References}
|
\section{Introduction}
According to hierarchical cosmological scenarios,
galaxies form through merging of smaller entities.
At least the dark haloes of dissipationless matter
hierarchically merge, and it is expected that
some of the visible galaxies also interact and
exchange mass, while spiraling in a common halo.
If major mergers lead to the formation of
ellipticals, and leave vestiges such
as shells, ripples and loops around present-day
elliptical galaxies (Schweizer \& Seitzer 1988),
signatures of accretion or merger are less easy to
see in spiral galaxies. Yet, mass and
gas accretion is required in spiral
galaxies for several reasons:
\begin{itemize}
\item Metallicity distribution in the
disk (the G-dwarf problem for instance, requires
gas infall)
\item Spiral formation and maintenance: episodes
of spiral density waves heat the disk,
and accreted fresh gas is required to trigger
new instabilities
\item Renewal of bars and nuclear bars, that
drive mass towards the center, and self-destroy
\item Reforming the thin disk after minor mergers:
galaxies such as the Milky Way re-form a thin disk,
while a thick one has been heated by an interacting event
\end{itemize}
In the following, the main evidences
for mass accretion in spiral galaxies will be reviewed,
including: the presence of thick disks,
counter-rotating components, the ubiquity of
warps, or the existence of polar rings.
\section{Galaxy Interactions and
Thickness of Stellar Disks}
In hierarchical cosmologies, it is easy to estimate
analytically the probability of formation of
a dark halo of mass $M$ at time $t$
from the Press-Shechter theory (1974),
revised by Bond et al (1991): a gaussian distribution
of fluctuations is assumed, and structures are followed
through random walk of linear overdensity with
respect to smoothing scale.
From such analytical formulations of merging histories (e.g.
Lacey \& Cole 1993, 1994), it is possible to relate the dark haloes
merger rate to the parameters of the universe (average density,
cosmological constant). The merging rates for visible galaxies should
follow, although the link is presently not well known
(Carlberg 1991, Toth \& Ostriker 1992).
For the standard CDM model ($\Omega$=1) for instance,
80\% of haloes have accreted
at least 10\% of their mass in the last 5 10$^9$ yrs.
To reduce the merging rate today, the solution
is to consider low $\Omega$ models,
for which freezing of halo formation occurs
for z $<$ 1/$\Omega$. After this epoch,
only very few haloes form, and the merger
rate of visible galaxies inside haloes is expected
also very low. When approximations are taken
for the merger conditions of the galaxies,
such as a threshold in their relative velocity
v $< v_{mg} \sim v_{escape}$
(Carlberg 1990, 91), the merger rate can be written
as a power law with redshift,
$dn(mergers) /dt \propto (1+z)^m $,
with the power-law $m$ increasing with $\Omega$
and $\Lambda$ (typically as
$m \propto \Omega^{0.42} (1-\Lambda)^{-0.11}$).
Observations
support a large value of the exponent $m$.
Statistics of close galaxy pairs from faint-galaxy redshift surveys
have shown that the merging rate increases as
$(1+z)^{m}$ with $m=4\pm1.5$ (e.g. Yee \& Ellingson 1995). Lavery
et al (1996) claim that collisional ring galaxies
(Cartwheel-type) are also rapidly evolving,
with $m=4-5$, although statistics are still insufficient.
Many other surveys, including IRAS faint sources,
or quasars, have also revealed a high power-law (Carlberg 1991,
Carlberg et al 1994).
\bigskip
The fragility of disks with respect to interactions
can be used to constrain the merging rate.
During an interaction, stellar disks can
thicken or even be destroyed (e.g. Gunn 1987).
Through the disk thickness of the Milky Way,
Toth \& Ostriker (1992) constrain the frequency of merging and the
value of the cosmological parameters: from analytical and local
estimations of the heating rate, they claim that
the Milky Way disk has accreted less than 4\% of its mass within the
last 5 Gyrs. But these local calculations are only
rough approximations. The first
numerical simulations of the phenomenon of disk
thickening through interactions (Quinn et al 1993, Walker et al 1996)
appear to confirm the analytical results however: they show
that the stellar disk thickening can be large and sudden.
Recently, Huang \& Carlberg (1997) and Velazquez \& White (1999)
reconsider the problem, through numerical simulations,
and find on the contrary that the
heating of disks have been overestimated. In particular,
prograde satellites heat the disks, while
retrograde ones produce only a coherent tilt.
If the halo is rigid, the thickening of the disk is
increased by a factor 1.5 to 2: massive live bulges
can therefore help to keep disks thin, in absorbing
part of the heating.
Also, there are many parameters
to explore in simulations: the most important could be
the compactness of the interacting companion. If
the perturber has a compact core, the heating effect is important,
while a more diffuse companion is destroyed by tidal
shear before damaging the primary disk.
\bigskip
It should be however remarked that gas hydrodynamics and star
formation processes can also alter significantly the
processes, since the thin disk can be reformed
continuously through gas infall.
It is interesting to check on presently interacting galaxies
whether the heating or thickening of disks is measurable.
In normal galaxies, the ratio of radial scale-length $h$
to scale-height $z_0$ is about constant and equal to 5
(Bottema 1993); the ratio only goes up for dwarf galaxies.
Now, in a sample of edge-on interacting galaxies this ratio
was found to be 1.5 to to 2 times lower than normal (Reshetnikov \&
Combes 1997, Schwarzkopf \& Dettmar 1999), as
shown in fig \ref{fig1}. This is surprising,
when taking into account that the visible "interacting phase"
is only transient, and on a Gyr time-scale, interacting
galaxies will return to the "normal phase", with again
a high $h/z_0$ ratio, or thin disk.
\begin{figure}
\psfig{figure=dyn99_f1.ps,width=17cm,bbllx=3cm,bblly=0mm,bburx=12cm,bbury=25cm,angle=-90}
\caption{Distribution of scalelength to scale height ratios
for (a) normal galaxies and (b) interacting galaxies (from Reshetnikov \& Combes, 1997).}
\label{fig1}
\end{figure}
A possible interpretation of this result is that the interacting
galaxies are warped, since the latter is difficult to distinguish
from a thickening at any viewing angle; yet the damping of the warp
will thicken the disk in any case.
If the thickening is in fact transient, this indicates
that the present disk galaxies come from merging of smaller units,
that they acquire mass continuously: through gas accretion
and subsequent star formation,
disks recover their small thickness after galaxy interactions,
or in other words,
the disk of present day spirals has been assembled at low
redshift (Mo et al 1998).
\section {Counter-Rotating Components}
The phenomenon of gas disks counter-rotating with the
stellar component is now well known in ellipticals,
where the ionised gas disks (or dust lanes) are settled
in principal planes (e.g. Bertola et al 1990).
Ellipticals were also first discovered with kinematically decoupled
stellar cores, which are expected in merger remnants, such as NGC 7252
(e.g. Barnes \& Hernquist 1992).
Counter-rotation has also been observed in many spirals during
this last decade, although
it is more difficult than in ellipticals,
since the secondary CR component is not dominating
and the primary component is strongly rotating.
All possibilities have been observed, either two stellar disks
counter-rotating with respect to each-other, or the gas counter to the stars,
or even gas versus gas, but not at the same radii in the galaxies
(see the reviews of Galletta 1996, Bertola \& Corsini 1998).
There is presently about 60 systems of counter-rotation
recorded in the literature.
In general the counter-rotating component is not dominant, but
there is a very special case, NGC 4550, where
two almost identical CR stellar disks are observed
(Rubin et al 1992). This case is a puzzle, since
the second disk cannot have formed through subsequent
accretion of gas: the two stellar disks have the same age.
The only solution is through a merger of pre-existing
spiral galaxies. If major mergers usually give
an elliptical galaxy as a remnant, this is not the case
when they have aligned direction of their angular momentum.
In these rare orientation cases, it is possible to
merge two spiral galaxies in one, and reproduce
the case of NGC 4550, when the momenta are opposite
(Pfenniger 1998, Puerari \& Pfenniger 1999).
Can one consider other explanations than mergers
for CR components? It is possible to artificially
simulate counter-rotation in a certain
region of a galaxy, through perpendicular streaming
motions due to a bar potential, for instance;
but when the 2D velocity field is
obtained, confusion is not possible. There are also
self-consistent models of barred galaxies including retrograde
orbits (Wozniak \& Pfenniger 1997), but the origin
of the retrograde stars is still gas accretion.
A slow bar destruction can be a rare case
where stars in box-orbits in a barred galaxy are
scattered equally in two CR families of tube orbits,
resulting in two opposite streams when the bar has
disappeared (Evans \& Collett 1994). But the process
of bar destruction is in any case related to
galaxy interactions and gas accretion.
\subsection{Stability}
How long such counter-rotating systems can live?
Does this phenomenon favor gas fueling to the
nucleus?
There exists a two-stream instability in flat disks,
similar to that in CR plasmas (Lovelace et al 97).
If there exists a mode in a given disk, the
energy of the modes in the two
streams are of opposite signs: the negative E mode can
grow by feeding energy in the positive E mode,
which produces the instability.
There exist also many bending instabilities
(Sellwood \& Merritt 1994).
If there is only a small fraction of CR stars, these
haveon the contrary a stabilising influence with
respect to bar formation ($m=2$); in a certain sense, they
are equivalent to a system with more velocity dispersion
(Kalnajs 1977).
But in the case of comparable quantities of CR stars,
a one-arm instability is triggered. This is confirmed
through N-body simulations:
a quasi-stationary one-arm structure forms,
and lives for 1--5 periods
(Comins et al 1997), first leading, than trailing, and disappears.
Fig \ref{fig2} shows such a simulation, where the
common $m=1$ pattern is leading for the main direct component,
and trailing for the secondary retrograde one.
\begin{figure}
\psfig{figure=dyn99_f2.ps,width=17cm,bbllx=45mm,bblly=75mm,bburx=11cm,bbury=20cm,angle=-90}
\caption{N-body simulation of two coexisting disks of stars, showing
the $m=1$ instability, a leading arm with respect to the direct stars (left)
and trailing for the retrograde stars (right), here plotted separately.
The instability occurs only if the CR stars represent at least 25\%
of the total.}
\label{fig2}
\end{figure}
\subsection{ Counter-rotating gas}
Accretion of CR gas in a lenticular galaxy
deprived of gas initially is a way to form
two stellar counter-rotating disks, after
star formation has occured.
Thakar \& Ryden (1996) have shown that
both episodic or continuous gas infall are
able to form a stable CR disk, without de-stabilising
the pre-existing disk significantly. The conditions
are that gas must be extended in phase space and not clumpy,
which would heat too much the primary disk.
For example, the merger of a gas-rich dense dwarf will have
a too large heating effect, unless the mass ratio is
quite small, and in such cases, only a small CR disk
is produced. However, the final thickness of
the disk depends drastically on the gas code used,
thicker for sticky particles, and much thinner with SPH
(Thakar \& Ryden 1998), as well as the settling time-scales.
When gas is present in the initial disk,
the presence of two CR streams of gas in the same plane will
be very transient: strong shocks will
produce heating and rapid dissipation will drive
the gas quickly to the center (Kuznetsov et al 1999).
This could be a very efficient way to fuel active
nuclei. However, the gas could also infall
in an inclined plane, or at different radii
than those of the pre-existing gas, which can explain
the observations of two counter-rotating gas systems.
Polar rings (objects similar to the prototype NGC 4650A)
are such cases, where gas settles in a stable plane almost
perpendicular to the primary galaxy.
Polar-ring galaxies are quite rare in the nearby universe:
Whitmore et al (1990) find that about 0.5\% of all nearby
lenticular galaxies are observed with a polar ring. But since
there are projection effects and different selection biases
that prevent to see them all, they estimate to about 5\% the
actual present frequency of PRGs.
An estimation of their frequency as a function of redshift
will be a precious tool to quantify the merging rate evolution.
\section{Warps as clues of matter accretion}
The majority of spirals are warped in their neutral hydrogen (HI)
component (e.g. Sancisi 1976, Bosma 1981, Briggs 1990).
This is a long-standing puzzle, since if the gas is considered
as test-particles in the halo potential, it should
differentially precess, and with a time-scale much shorter
than the Hubble time the disk should end up with a
corrugated shape and thicken.
Many theories have been proposed to solve the problem.
Normal modes of the disk have been ruled out, since they are
quickly damped (Hunter \& Toomre 1969), but
normal modes of the disk in the potential if
a mis-aligned halo have been
a possibility for a while (Toomre 1983, Sparke \& Casertano 1988),
until it was realized that they are
quickly damped through dynamical friction (Nelson
\& Tremaine 1995).
The triggering of warps by tidal interaction
with companions has been ruled out in the past, since
the best examples of warped galaxies appeared isolated.
However, this could be changing now that smallest companions
can be found, or vestiges of a past merger. This is the case
of the warp-prototype NGC 5907, where a conspicuous tidal loop
has been observed by Shang et al (1998). It is obvious that
this galaxy has accreted a small system in the recent
past, and it has also a dwarf companion nearby
(see Fig \ref{fig3}).
Regular and symmetric warps are those that have already
relaxed for a while, and this could explain the apparent lack
of correlation with companions.
Finally, the proposition that gas infall could
maintain warps around galaxies is easily justified
in the framework of hierachical cosmologies (cf Ostriker \& Binney
1989; Binney 1992). Gas infalls with slewed angular momentum
with respect to the main disk. This accretion will
re-align the whole system along a tilted axis.
The transient state is the warped state. This hypothesis
has been recently supported through numerical
simulations by Jiang \& Binney (1999).
They show that the inner halo and disc tilts as one unit.
The halo tilts first in the outer parts, and the
tilt propagates then inwards; the disk is entrained
and aligns with the halo, it plays the role of a tracer
of its orientation. The time-scale of this phenomenon is
about 1 Gyr to re-align by 7$^\circ$.
\begin{figure}
\psfig{figure=dyn99_f3.ps,width=9cm,bbllx=1cm,bblly=5cm,bburx=20cm,bbury=24cm,angle=0}
\caption{NGC 5907 image obtained by Shang et al. (1998) in the visible
(6600\AA). All foreground stars and the dust lane of the edge-on
galaxy have been blanked. The inset diagram shows the position
of the ring/loop (from Shang et al. 1998).}
\label{fig3}
\end{figure}
\section {Conclusions}
In summary, there are many evidences that even spiral
galaxies have experienced a large number of
galaxy interactions in the past, and that their
formation proceeds also through hierachical merging
and accretion: presence of thick and thin disks,
growing number of observed counter-rotating disks,
frequency of polar-ring galaxies, ubiquity of HI warps.
Both present explanations of warps, either through
tidal interactions trigger, or maintenance through gas infall,
are compatible with this paradigm.
\begin{moriondbib}
\bibitem{} Barnes J., Hernquist L., 1992, {\it Ann. Rev. Astron. Astrophys.}
{\bf 30}, 705
\bibitem{}Bertola F. Bettoni D., Buson L.M., Zeilinger W.W., 1990, in
{\it Dynamics and Interaction of Galaxies}, ed. R. Wielen, Springer, p. 249
\bibitem{}Bertola F. \& E. Corsini, 1998 in {\it Galaxy Interactions}
IAU Symp. 186, Kyoto 1997, Kluwer
\bibitem{} Binney J., 1992, {\it Ann. Rev. Astron. Astrophys.}
{\bf 30}, 51
\bibitem{}Bond J.R., Kaiser N., Cole S., Efstathiou G., 1991,
\apj {379}{440}
\bibitem{}Bottema R., 1993, \aa {275}{16}
\bibitem{}Bosma A., 1981, \aj {86}{1825}
\bibitem{}Briggs F., 1990, \apj {352}{15}
\bibitem{}Carlberg R.G. 1990, \apj {359}{L1}
\bibitem{}Carlberg R.G. 1991, \apj {375}{429}
\bibitem{}Carlberg R.G., Pritchet C.J., Infante L. 1994 \apj {435}{540}
\bibitem{}Comins N.F., Lovelace R.V.E., Zeltwanger T., Shorey P., 1997,
\ apj {484}{L33}
\bibitem{}Evans N.W., Collett J.L., 1994, \apj {420}{L67}
\bibitem{}Galletta G., 1996, in {\it Barred Galaxies}, ed R. Buta, D.A. Crocker
\& B.G. Elmegreen, ASP Conf. Series. {\bf 91}, p. 429
\bibitem{}Gunn J.E., 1987, in {\it Nearby Normal Galaxies} ed. S.M. Faber,
Springer New York, p. 459
\bibitem{}Huang S. \& Carlberg R.G. 1997, \apj {480}{503}
\bibitem{}Hunter C., Toomre A., 1969 \apj {155}{547}
\bibitem{}Jiang I-G., Binney J., 1999, \mnras {303}{L7}
\bibitem{}Kalnajs A., 1977, \apj {212}{637}
\bibitem{}Kuznetsov O.A., Prokhorov M.E., Sazhin M.V., Chechetkin V.M.,
1999, {\it Astrophys. J.} preprint (astro-ph/9810429)
\bibitem{}Lacey C., Cole S. 1993, \mnras {262}{627}
\bibitem{}Lacey C., Cole S. 1994, \mnras {271}{676}
\bibitem{}Lavery R.J., Seitzer P., Suntzeff N.B., Walker A.R.,
Da Costa G.S. 1996, \apj {467}{L1}
\bibitem{}Lovelace R.V.E., Jore K.P., Haynes M.P., 1997, \apj {475}{83}
\bibitem{}Mo H.J., Mao S., White S.D.M., 1998, \mnras {295}{319}
\bibitem{}Nelson R.W., Tremaine S., 1995, \mnras {275}{897}
\bibitem{}Ostriker E.C., Binney J.J., 1989, \mnras {237}{785}
\bibitem{}Pfenniger D., 1998, in {\it Galaxy Interactions} IAU Symp. 186, Kyoto
1997, Kluwer
\bibitem{}Press W.H., Schechter P. 1974, \apj {193}{437}
\bibitem{}Puerari I., Pfenniger D., 1999, in {\it The Evolution of Galaxies
on Cosmological Time-Scales}, Tenerife, Spain, Nov 30-Dec 5, 1998, PASP Series
ed T. Mahoney \& J.E. Beckman
\bibitem{}Quinn P.J., Hernquist L., Fullagar D.P., 1993, \apj {403}{74}
\bibitem{}Reshetnikov V., Combes F.: 1997, \aa {324}{80}
\bibitem{}Rubin V.C., Graham J.A., Kenney J.D.P., 1992, \apj {394}{L9}
\bibitem{}Sancisi R., 1976, \aa {53}{159}
\bibitem{}Schwarzkopf U., Dettmar R-J., 1999, in {\it Galaxy Evolution:
Connecting the Distant Universe with the Local Fossil Record}, ed. M.
Spite, Kluwer
\bibitem{}Schweizer F., Seitzer P. 1988, \apj {328}{88}
\bibitem{}Sellwood J.A., Merrit D., 1994, \apj {425}{530}
\bibitem{}Shang Z., Brinks E., Zheng Z. et al., 1998, \apj {504}{L23}
\bibitem{}Sparke L., Casertano S., 1988, \mnras {234}{873}
\bibitem{}Thakar A.R., Ryden B.S., 1996 \apj {461}{55}
\bibitem{}Thakar A.R., Ryden B.S., 1998, \apj {506}{93}
\bibitem{}Toomre A. 1983, in {\it Internal Kinematics and Dynamics of
Galaxies}, IAU Symp. 100, ed. E. Athanassoula, p. 177 (Reidel)
\bibitem{}Toth G., Ostriker J.P. 1992 \apj {389}{5}
\bibitem{}Velazquez H., White S.D.M., 1999, \mnras {304}{254}
\bibitem{}Whitmore B.C., Lucas R.A., McElroy D.B. et al: 1990, \aj {100}{1489}
\bibitem{}Yee H.K.C., Ellingson E. 1995 \apj {445}{37}
\bibitem{}Walker I.R., Mihos J.C., Hernquist L., 1996, \apj {460}{121}
\bibitem{}Wozniak H., Pfenniger D., 1997 \aa {317}{14}
\end{moriondbib}
\vfill
\end{document}
|
\section{Introduction}
The soft X-ray band below 2~keV has been covered by all recent X-ray
imaging satellites. The energy band is rich in spectral features
(absorption, emission lines, excess soft X-ray emission and so on),
which are useful diagnostic tools for understanding the nature of an
X-ray source of interest. It is therefore important that the spectra
are clearly understood and that data from different satellites are
calibrated reasonably well with each other so that results are
independent of the instrument used.
It has been noticed that there are sometimes significant discrepancies
between spectral measurements taken with ROSAT and ASCA (e.g., Yaqoob
et al 1994; Allen \& Fabian 1997; Ptak 1997). Typically, for AGN,
steeper spectral slopes are obtained from ROSAT PSPC measurements than
from ASCA. The steep ROSAT spectral slopes have sometimes been
interpreted as due to excess soft X-ray emission, when compared with
measurements from hard X-ray experiments (e.g., EXOSAT ME, Ginga LAC,
and ASCA). However, similar discrepancies have been found in quasar
spectra between the PSPC and the Einstein Observatory IPC (Fiore et al
1994; Laor et al 1994; Ciliegi \& Maccacaro 1996). As the bandpass of
the two detectors largely overlaps, calibration errors have been
suspected (Fiore et al 1994).
We aim in this paper to address the present status of the discrepancy
between the ASCA SIS and ROSAT PSPC, using data from the bright
Seyfert galaxy NGC5548 for which there was simultaneous coverage with
both satellites. Although the soft X-ray spectrum of NGC5548 is not
ideal for calibration because of complexity due to a warm absorber and
excess soft X-ray emission, the high spectral resolution of the ASCA
SIS is capable of resolving these features. Since the observation was
carried out before any significant degradation of the SIS detector had
occurred, the SIS spectrum has good spectral resolution. There are
several other simultaneous ASCA/ROSAT observations of AGN, but strong
absorption or more complex soft X-ray spectra of these targets (e.g.,
IRAS18325--5926, Iwasawa et al 1996) make cross-calibration difficult.
C. Otani (priv. comm.) finds a similar discrepancy from the
simultaneous ASCA/ROSAT observations of MR2251--178 to that reported
here. NGC5548 is stable and bright in the soft X-ray band where both
the ASCA SIS and ROSAT PSPC are sensitive; it has also been well
studied (Nandra et al 1993; Done et al 1995; Reynolds 1997; George et
al 1998). Preliminary results from efforts of cross-calibration
between ASCA and other satellites, e.g., BeppoSAX and RXTE, are now
available and the consistency between them is discussed.
In Section 2, we give detailed information on the data reduction. In
Section 3, we describe the integrated ASCA spectrum from the whole
observation. The ROSAT PSPC data are presented in Section 4. The
simultaneous data are described in Section 5, where we also examine
the consistency of the two datasets. The results and their implication
are discussed in Section 6.
\section{Observations and data reduction}
\begin{table*}
\begin{center}
\caption{FTOOLS and calibration files used for the ASCA SIS data
reduction and analysis. $^{\ast}$Charge Transfer Inefficiency (CTI) is caused
by radiation damage of the CCDs and has been increasing as a function
of time.}
\begin{tabular}{lll}
Task & FTOOLS & Calibration files \\[5pt]
PI-channel conversion & SISPI (version 1.1) & sisph2pi$\_$110397.fits
(CTI$^{\ast}$ table) \\
Response matrix & SISRMG (version 1.10) & \\
Detector efficiency & ASCAARF (version 2.72) & ${\rm xrt\_ea\_v2\_0.fits}$
(XRT effective area) \\
& (Effective area fudge ON, & ${\rm xrt\_psf\_v2\_0.fits}$ (XRT point spread function) \\
& Arf filter ON) & \\
\end{tabular}
\end{center}
\end{table*}
The ASCA observation started at 1993 July 27, 15:36 (UT) and ended at
1993 July 28, 08:24.
The Solid state Imaging Spectrometer (SIS; S0 and S1) was operated
using 4 CCD chips switching between Faint and Bright modes with
High and Medium telemetry rates, respectively.
No level discriminater was applied.
The ASCA data reduction was performed using FTOOLS (version 4.0)
and standard calibration files
provided by the ASCA Guest Observer Facility (GOF) at NASA/Goddard Space
Flight Center.
We used only the SIS data here.
The Faint mode data were converted to the format of Bright mode so that
all the data could be added together.
This means that the correction for the `Echo' and `Dark Frame Error' (DFE)
(Otani \& Dotani 1995) cannot be applied.
Since the observation was carried out only five
month after the launch of the satellite, those effects on the
data are small and taken into account in the detector response
matrices generated by SISRMG (version 1.1).
Another effect known to degrade the SIS data is Residual Darkframe
Distribution
(RDD, Dotani et al 1998), which is dark current remained after the DFE
correction due to non-uniform dark level between CCD pixels,
although it should be small for this early ASCA observation as
the accumulation of the radiation damage is the main cause.
We have verified that the Faint mode data, taken during this observation,
with DFE and Echo corrections applied
give entirely consistent results with the same data in Bright mode,
apart from the normalization.
The Bright mode data give $\sim 5$ per cent smaller normalization than
the Faint mode data, which can be attributed to the RDD effect.
The RDD effect is largest when all 4 CCD chips are operating,
as in the present observation.
It causes a reduction of efficiency which is almost
independent of energy. The reduction of efficiency
by $\sim 5$ per cent is in the range expected from the RDD effect.
The SIS results shown below are not corrected for this
efficiency reduction.
Observed events with grades of 0, 2, 3 and 4
were selected and hot/flickering pixels on the CCD chips were removed.
The data selection criteria are, 1) source elevation higher than 5$^{\circ}$
and 25$^{\circ}$ above the night and bright Earth rims, respectively;
2) cut-off rigidity greater than 4 GeV c$^{-1}$; and
3) X-ray telescope pointing fluctuation is less than 30 arcsec.
Data taken within 120 seconds of the spacecraft passing through
the transition from bright to night Earth, and the
South Atlantic Anomaly, are discarded.
The total good exposure time for each SIS detector was 30 ks.
The details of the calibration files we used are given in Table 1.
The source data were collected from a circular region with a radius of
4 arcmin (see the details in Table 2)
while the background data were taken from a source-free region
in the same field of view on the detector.
The background counts in the 0.5--10 keV band are only 1 per cent
of the source counts in both SIS spectra.
The effective area as a function
of energy, appropriate for the spectra obtained, was computed with
ASCAARF (version 2.72), assuming a point source lying at the centre of the
photon collecting region (`point'=yes, `simple'=yes in ASCAARF, see also
Table 1).
Note that the source extraction regions spread over the four CCD chips
in each detector, although most of the photons are collected on the
main chip (S0 chip-1, S1 chip-3). Therefore some photons fell into the
interchip gaps. The effective area of the ASCA X-ray telescope (XRT)
is believed to be reasonably accurate for a point source when
integrated over the image, but the azimuthal dependence of the point
spread function (PSF) may not be modelled accurately, as its shape is
like a `maltese cross'. We have made an experiment to check this
issue, using various source extraction-regions with and without
interchip gaps in them. The effective area of each spectrum was
computed with ASCAARF. A comparison of the fluxes obtained from those
spectra shows differences between of 2--3 per cent. Although ASCAARF
does not correct for photons falling in the interchip gaps, the
resulting error in the flux estimate is insignificant. A larger error
could result when the source centroid lies closer to these gaps.
During the ASCA observation, the ROSAT PSPC also observed the galaxy
for a short period, as shown in Fig. 1.
The ROSAT coverage was from 1993 July 28, 00:53 to 09:16.
Total exposure time of the PSPC data is 4.3 ks of which about 3 ks was
covered also by ASCA.
We used the ROSAT data with the standard SASS processing.
The PSPC spectrum of NGC5548 was reduced using XSELECT.
The source photons were collected from a circular region with a radius of
1.9 arcmin while the background data are taken from an anulus of 4.0--5.8
arcmin radii centred on the source.
The response matrix for the PSPC data (`pspcb$\_$gain2$\_$256.rsp'),
which is appropriate for
a point source located at the centre of the field of view, is taken from
the ROSAT calibration database.
We have tested the data corrected for temporal and spatial
gain variations in the PSPC
by PCPICOR (which also fixes a bug in the SASS processed data)
and found no significant difference from the original
data for which results are shown here.
\begin{table}
\begin{center}
\caption{The source extraction regions for the ASCA SIS (S0 and S1).
They are circular regions of $\approx 4$ arcmin radius
centred on each peak of the X-ray images.
The centre and radius are shown in unit of pixel in the detector
coordinates (DETX, DETY).}
\begin{tabular}{ccc}
Detector & Centroid & Radius \\[5pt]
S0 & (619,\thinspace 555) & 147 \\
S1 & (615,\thinspace 599) & 147 \\
\end{tabular}
\end{center}
\end{table}
\begin{figure}
\centerline{\psfig{figure=fig1.ps,width=0.45\textwidth,angle=270}}
\caption{The 0.4--10 keV light curve of NGC5548 observed with the
ASCA SIS0. The epoch of the light curve is 1993 July 27, 15:36 (UT).
Each bin has an exposure of 128 s. The ROSAT PSPC simultaneously observed
the periods (3.34--3.42)$\times 10^4$ s and (3.84--4.02)$\times 10^4$ s,
which are indicated in the light curve.}
\end{figure}
\section{The ASCA data}
\begin{figure}
\centerline{\psfig{figure=fig2.ps,width=0.45\textwidth,angle=270}}
\caption{The soft X-ray part of the integrated ASCA SIS spectrum of NGC5548.
The data from the S0
and S1 are plotted with the best-fit model of a power-law plus three
absorption edges (see Table 3).}
\end{figure}
The ASCA 0.4--10 keV light curve shows moderate flux variation
across the observing run (Fig. 1). We first describe the spectral properties
of NGC5548 in the ASCA band (see also Reynolds 1997; George et al 1998),
using the integrated whole ASCA SIS dataset and then
discuss the data simultaneously observed with both satellites.
Spectral analysis was performed using XSPEC (version 10.0) and the
data from the two SIS detectors are fitted jointly.
A simple power-law fit to the whole band data leaves significant residuals
(see Fig. 1 in Reynolds 1997; note that the upturn of the residual below
1 keV, e.g., in the EXOSAT ME/LE spectrum, has sometimes been interpreted
as a soft excess, but it turns out to be an effect of a warm absorber).
NGC5548 has been known to show spectral
features typical of Seyfert 1 galaxies (e.g., Mushotzky, Done \&
Pounds 1993) such as a warm absorber (Nandra et al 1991),
iron K line and spectral hardening at high energies due to reflection
from optically thick, cold matter (Nandra \& Pounds 1994).
These spectral features modify most of the primary power-law except for
a narrow energy range around 3 keV. We first investigate
the soft X-ray spectrum, which is not affected by reflection.
We fit the 0.4--4 keV data to investigate spectral features due to the
warm absorber (Fig. 2).
At least three absorption edges are significantly detected (Table 3)
when a simple power-law is employed for the continuum. The photon-index of the
power-law is $\Gamma = 1.934\pm 0.011$ and no excess absorption
above the Galactic value (\nH\ $= 1.7\times 10^{20}$cm$^{-2}$,
Dickey \& Lockman 1990) is required. Apart from
a weak line-like feature at 0.55 keV, most likely due to an
under-subtracted atmospheric oxygen line, the quality of the fit is
excellent ($\chi^2 = 260.44$ for 238 degrees of freedom, see Fig. 2).
The three absorption edges are identified with OVII, OVIII and NeIX.
There is no obvious signature of extra soft X-ray emission.
The presence of a weak soft excess has been reported by Reynolds
(1997) using the same data. However, the model he used for the fit
consists of only the two oxygen edges and a simple power-law for the
0.6--10 keV data despite the presence of the 1.2 keV edge and spectral
hardening above 4 keV due to reflection presented here. Although a
blackbody-type soft excess is indeed significantly detected when only
two edges are considered in the model for the 0.4--4 keV data, the
inclusion of the 1.2 keV edge instead provides a better fit to the
data.
\begin{figure}
\centerline{\psfig{figure=fig3.ps,width=0.45\textwidth,angle=270}}
\caption{Confidence contours between photon-index and reflection strength
when the integrated ASCA 0.4--10 keV data are fitted with the model given
in Table 3.
The 68, 90 and 99 per cent confidence levels for two interesting parameters
are plotted.}
\end{figure}
An extrapolation of the best-fit model shows that
the continuum slope becomes flatter above $\sim$4 keV towards 10 keV
(see Fig. 4).
The iron K line is clearly detected (see also Mushotzky et al 1995).
A simple power-law fit to the 4--10 keV band excluding the Fe K band
(5.5--7 keV) gives a photon-index $\Gamma = 1.82^{+0.10}_{-0.09}$.
An absorption edge feature around 8 keV, observed in the Ginga spectra
(Nandra \& Pounds 1994), is
barely detected at $8.1\pm 0.4$ keV ($\tau =0.15^{+0.18}_{-0.13}$).
A likely origin for this is K-shell absorption of partially ionized iron
in the warm absorber. When the iron K edge is included in the fit,
a slightly flatter power-law slope, $\Gamma = 1.76\pm 0.12$, is inferred.
The spectral hardening above
10 keV has been observed clearly in the Ginga spectra (Nandra \& Pounds 1994).
X-ray reflection from cold matter can account for this hard tail
(e.g., George \& Fabian 1991).
\begin{figure}
\centerline{\psfig{figure=fig4.ps,width=0.48\textwidth,angle=270}}
\caption{(a) Spectral features observed in the integrated
ASCA SIS spectrum. Deviations
of the data from the power-law ($\Gamma = 1.934$) modified by the
Galactic absorption are plotted, removing the warm absorption, reflection
and iron line features from the best fit model which gives residuals
shown in lower panel (b). Note that there are alternative models to describe
the data (see text).}
\end{figure}
\begin{table*}
\begin{center}
\caption{Spectral fits to the ASCA data. The warm absorption is described
with three absorption edges. The threshold energies for absorption
edge and line energies are given in the source rest frame. The power-law
reflection model, pexrav (Magdziarz \& Zdziarski 1995), in XSPEC is used with
the Galactic absorption.
$^{\ast}A$ is the normalization of power-law in units of ph\thinspace keV$^{-1}$\thinspace s$^{-1}$\thinspace cm$^{-2}$\
at 1 keV. $^{\dag}$The value for the cosine of inclination of
the reflecting slab, $i$, is fixed at 0.95 (or $i\sim 20^{\circ}$). Strength
of reflection, $I_{\rm Refl}$, is unity when the reflection matter covers
half of the sky. The Galactic absorption, \nH\ $=1.7\times 10^{20}$cm$^{-2}$, is
assumed as no excess absorption is detected.}
\begin{tabular}{cccc}
\multicolumn{4}{c}{Absorption edges}\\[5pt]
& $E_{\rm th}$ & $\tau$ & ID\\
& keV & & \\[5pt]
(1) & $0.73\pm 0.02$ & $0.30\pm 0.05$ & OVII \\
(2) & $0.86\pm 0.02$ & $0.22\pm 0.06$ & OVIII \\
(3) & $1.24\pm 0.04$ & $0.056\pm 0.021$ & NeIX \\[10pt]
\end{tabular}
\begin{tabular}{ccccc}
\multicolumn{5}{c}{Fe K line (double-gaussian)}\\[5pt]
& $E_{\rm line}$ & $\sigma$ & $I$ & $EW$ \\
& keV & eV & $10^{-5}$ph\thinspace s$^{-1}$\thinspace cm$^{-2}$ & eV \\[5pt]
(1) & $5.94\pm 0.10$ & $75^{+125}_{-75}$ & $2.22\pm 1.41$ & $41\pm 26$ \\
(2) & $6.44\pm 0.04$ & $32^{+80}_{-32}$ & $4.27\pm 1.45$ & $91\pm 31$ \\[10pt]
\end{tabular}
\begin{tabular}{ccccc}
\multicolumn{5}{c}{Power-law reflection model}\\[5pt]
$\Gamma$ & $A^{\ast}$ & cos\thinspace $i^{\dag}$ & $I_{\rm Refl}$ & \nH \\
& $10^{-2}$ & & & $10^{20}$cm$^{-2}$ \\[5pt]
$1.934\pm 0.014$ & 1.87 & 0.95 & $1.39\pm 0.47$ & 1.7 \\
\end{tabular}
\end{center}
\end{table*}
The use of the power-law reflection model ({\tt pexrav},
Magdziarz \& Zdziarski 1995)
instead of a simple power-law provides a good fit to the 0.4--10 keV
data if the iron K line is also modelled with a double-gaussian.
Among the parameters of the reflection model, the
abundance of iron and the other
elements in the reflecting matter is assumed to be 1 solar.
As the inclination angle, $i$, of the reflecting slab (or the accretion disk)
cannot be well constrained (when the strength of reflection is assumed to be
1.0, the upper limit is 80$^{\circ}$), cos $i = 0.95$ (or
$i\simeq 20^{\circ}$) is assumed. The best-fit values for photon index and
reflection strength are $\Gamma = 1.93\pm 0.02$ and
$I_{\rm refl}=1.4\pm 0.5$, and
a confidence contour plot for the two parameters is shown in Fig. 3.
The spectral features which have modified the primary power-law are
demonstrated in Fig. 4a, and the model consisting of
a power-law plus warm absorption, reflection and
an iron K line gives a good fit to the data ($\chi^2 = 453.09$ for
475 degrees of freedom, see Fig. 4b).
The data are thus consistent with the reflection from a (nearly)
face-on accretion disk
illuminated by a power-law ($\Gamma\simeq 1.93$) source above it.
The observed fluxes are $2.7\times 10^{-11}$erg~cm$^{-2}$~s$^{-1}$\ in the 0.5--2
keV band and $4.4\times 10^{-11}$erg~cm$^{-2}$~s$^{-1}$\ in the 2--10 keV band.
The 0.5--2 keV flux corrected for the warm absorption is $3.1\times
10^{-11}$erg~cm$^{-2}$~s$^{-1}$. Note that the fluxes could be higher by $\sim
20$ per cent than the SIS values given above on account of the
comparison with GIS results.
The response matrix of the GIS (version 4.0) has a large systematic
error in the low energy range. The corresponding error in
\nH\ measurement is about $\Delta N_{\rm H}\simeq 1\times 10^{21}$cm$^{-2}$\
(private com., T. Dotani and the GIS team), implying that GIS spectra
below 1.5 keV are not as reliable as SIS spectra. The spectral
parameters obtained from the 2--10 keV GIS data are consistent with
the SIS results within errors, but with $\sim 20$ per cent larger
normalization than the SIS. Although about 5 per cent can be explained
by the RDD effect (see Section 2), the remaining $\sim 15$ per cent
may be an error in the flux estimated in the SIS. Telemetry
saturation, which is most severe in 4CCD observations during the PV
phase, might account for part of the error. The GIS has been flux
calibrated with the Crab Nebula, the classic standard X-ray source
used for flight calibration of many previous medium-hard X-ray
instruments. The source is too bright for the SIS to observe because
of pulse pile-up. For X-ray sources with a non Crab-like
spectrum, $\sim$5--10 per cent difference in flux between the GIS and
SIS are often seen, according to the calibration status report
published by the ASCA GOF.
The strength of reflection may be too strong. An acceptable fit to the data
can be obtained by alternative models involving
no strong reflection below.
1) A single power-law
modified by warm absorption with deep edges in the 2--3 keV ranges which could
be attributed to K-edges of ionized Si, S and Ar, in addition to the
three edges considered above. The photon index would be $\Gamma\simeq 1.90$.
This model is far inconsistent with a one-zone
solar abundance warm absorber. 2) A double power-law discussed in George
et al (1998).
This model makes a curving continuum which steepens at lower energies.
There are two solutions for the combination of power-laws
[PL$_i(\Gamma_i, n_i$), where $\Gamma$ and $n$ are photon-index
and normalization at 1 keV of a power-law and $i = 1, 2$]
which provide similar
quality of fit when the three absorption edges are included:
a) $\Gamma_1\simeq 2.3$, $\Gamma_2\simeq 1.8$, $n_1/n_2\sim 0.4$;
and b) $\Gamma_1\simeq 2.0, \Gamma_2\simeq 1.6, n_1/n_2\sim 3.6$.
\begin{figure}
\centerline{\psfig{figure=fig5.ps,width=0.44\textwidth,angle=270}}
\caption{The iron K line profile of NGC5548 obtained from the integrated
ASCA SIS data. The two
SIS detectors are summed together. The energy scale is corrected for
the galaxy redshift ($z=0.017$).}
\end{figure}
The shape of the iron K line was reported to be broad ($\sigma\sim 0.5$ keV)
by Mushotzky et al (1995). With the present analysis, the line profile
appears to be split
into two components: a major one at an energy of 6.4 keV and a weaker
one at 5.9 keV (Fig. 5 and Table 3, also see Otani 1995; Nandra et al 1997).
The shape is difficult to explain with the ordinary diskline models
(e.g., Fabian et al 1989; Laor 1991). It may be the sum of a diskline and
a narrow line at 6.4 keV.
The total equivalent width of the line is $\sim 130$ eV.
The photon-index of the primary power-law source is in a good agreement
with that derived from the simple power-law plus three absorption edge
model for the 0.4--4 keV data (Table 3).
This justifies the use of the ASCA data below 4 keV in investigating
the soft X-ray spectrum, obtained from the simultaneous observation
presented below, without being affected by the reflection component.
It should however be noted that the primary power-law above 4 keV can
be flatter than this. The reflection model fit to the 2--18 keV Ginga
data derived $\Gamma =1.81\pm 0.02$ and $I_{\rm Refl}=0.7\pm 0.4$
(Nandra \& Pounds 1994) with use of the Lightman \& White (1988) type
reflection model ({\tt plrefl} in XSPEC).
This can also describe the 4--10 keV ASCA data
when an iron K edge is included at around 8 keV.
\section{The ROSAT data}
\begin{table*}
\begin{center}
\caption{Spectral fits to the ROSAT PSPC data. Spectral models are
(1) a simple power-law; and (2) a power-law modified by Galactic
absorption (\nH\ $= 1.7\times 10^{20}$cm$^{-2}$) and an absorption edge.
The threshold energy of the absorption edge is measured in the
galaxy rest frame.}
\begin{tabular}{ccccccc}
Model & $\Gamma$ & $A$ & \nH & $E_{\rm th}$ & $\tau$ & $\chi^2$/dof \\
&& $10^{-2}$ & $10^{20}$cm$^{-2}$ & keV & & \\[5pt]
(1) & $2.37\pm 0.04$ & 1.74 & $1.6\pm 0.09$ & --- & --- & 191.52/176 \\
(2) & $2.35\pm 0.02$ & 1.87 & 1.7 & $0.80\pm 0.06$ & $0.23\pm 0.10$ & 177.34/175 \\
\end{tabular}
\end{center}
\end{table*}
\begin{figure}
\centerline{\psfig{figure=fig6.ps,width=0.46\textwidth,angle=270}}
\caption{The ROSAT PSPC spectrum of NGC5548 fitted with a
power-law plus an absorption edge model (Model (2) in Table 4).}
\end{figure}
We now present the spectral analysis of the total (4 ks) ROSAT PSPC data.
A simple power-law fit to the 0.14--2 keV data
gives photon index $\Gamma = 2.38\pm 0.04$
and absorption column density of $(1.65\pm 0.09)\times 10^{20}$cm$^{-2}$,
which is consistent with the Galactic value.
The inclusion of an absorption edge
improves the quality of the fit significantly (Table 4).
It is clear that the PSPC spectrum gives a steeper photon index
($\Gamma = 2.35\pm 0.02$) than
the ASCA SIS spectrum.
The edge energy $\sim 0.8$ keV is an intermediate value of those of the
two oxygen edges (OVII and OVIII, see Table 3).
Little improvement ($\Delta\chi^2 = 0.02$)
is seen when a further edge is added.
The PSPC data and the best-fit model (2) are shown in Fig. 6.
There is no signature of excess soft emission.
The observed fluxes are $6.4\times 10^{-11}$erg~cm$^{-2}$~s$^{-1}$\ in the 0.1--2 keV
band and $3.7\times 10^{-11}$erg~cm$^{-2}$~s$^{-1}$\ in the 0.5--2 keV band,
estimated from the best-fit power-law plus edge model.
The 0.5--2 keV flux corrected for the absorption edge is
$3.9\times 10^{-11}$erg~cm$^{-2}$~s$^{-1}$.
NGC5548 is known to show soft X-ray flares seen below 0.4 keV
(Done et al. 1995). The flux change of the soft X-ray component is
not correlated with the higher energy continuum (Done et al 1995;
Kaastra et al 1998). Here we investigate whether the source is
undergoing a similar soft flare. This is important when excess soft
X-ray emission is disscussed because this
component appears out of the ASCA bandpass.
A monitoring campaign of NGC5548 with ROSAT PSPC was carried out between
1992 December and January 1993, 6--7 months before the present observation
(Done et al 1995).
During the campaign, a large flare over 8 days was observed with strong
spectral softening.
Apart from the flare, the rest of the data show a
similar spectral shape, judging from the fairly constant count rate ratio
in the energy bands of 1.0--2.5 keV and 0.1--0.4 keV.
We thus compared our PSPC data with the pre-flare data
(the dataset {\tt 01b} in Done et al 1995),
which Done et al (1995) used as a template spectrum
and probably a representative one in normal states.
The simple power-law fit to the present data (Table 4) gives very similar
parameters to those for the {\tt 01b} dataset ($\Gamma = 2.36\pm 0.01$ and
$A = 1.69\times 10^{-2}$ph\thinspace keV$^{-1}$\thinspace s$^{-1}$\thinspace cm$^{-2}$, Done et al 1995);
the difference in normalization is only $\sim$2 per cent.
The value of the hardness ratio (0.32) for the present observation is
also similar to that for the {\tt 01b} dataset.
Therefore we conclude that the X-ray source was in its normal state when the
simultaneous observations were carried out.
\section{The ASCA/ROSAT simultaneous data}
\begin{table*}
\begin{center}
\caption{Spectral fit to the ASCA data for the simultaneous coverage.
The 0.4--4 keV SIS data are used. The Galactic value \nH\ =
$1.7\times 10^{20}$cm$^{-2}$\ is assumed for absorption of the power-law.
Two absorption edges are significantly detected. They are attributed to
partially ionized oxygen (OVII and OVIII) in the warm absorber. The
threshold energy of the absorption edges is
measured in the galaxy rest frame.}
\begin{tabular}{ccccccc}
$\Gamma$ & $A$ & $E_{\rm th1}$ & $\tau_1$ & $E_{\rm th2}$ & $\tau_2$ &
$\chi^2$/dof \\
& $10^{-2}$ & keV & & keV & & \\[5pt]
$1.954\pm 0.038$ & 1.83 & $0.74^{+0.05}_{-0.07}$ & $0.30^{+0.15}_{-0.16}$ &
$0.87^{+0.05}_{-0.04}$ & $0.27^{+0.15}_{-0.16}$ & 174.83/182 \\
\end{tabular}
\end{center}
\end{table*}
\begin{figure}
\centerline{\psfig{figure=fig7.ps,width=0.47\textwidth,angle=270}}
\caption{Ratio of the data of the ROSAT PSPC (filled squares)
and the ASCA SIS (crosses) to the best-fit model for the ASCA 0.4--4
keV spectrum are shown. The datasets are obtained from the
simultaneous observation. The SIS normalization has been adjusted to
agree with that of the GIS. A discrepancy in the spectrum between the
two instruments is clearly seen, even in the energy band covered by
both detectors (e.g., 0.4--2 keV band).}
\end{figure}
The period over which both ASCA and ROSAT observed NGC5548 were simultaneous
for about 3 ks.
This is a large fraction of the total PSPC observation, during which
no flux change is observed. The spectral data for the 3 ks is
virtually identical to the total 4 ks spectrum.
On the other hand, the ASCA data of the simultaneous observation is
a small fraction ($\sim 1/10$) of the total exposure.
The ASCA 0.5--2 keV flux during this period is
$3.0\times 10^{-11}$erg~cm$^{-2}$~s$^{-1}$,
$\sim$ 10 per cent higher than the mean value for the whole observation.
The spectrum can be different from the averaged one if not dramatically.
We therefore analyze the ASCA spectrum and then compare it with
the PSPC spectrum to illuminate any discrepancy in spectral measurement
between the two satellites.
Since the spectrum below 4 keV is not affected by reflection, as
discussed in Section 3, a simple power-law modified by
the Galactic absorption and a warm absorber was fitted to the
ASCA SIS 0.4--4 keV data. Results of the spectral
fit are given in Table 5.
The best-fit value of photon-index is slightly larger than that for
the whole dataset, but consistent within uncertainties.
Two absorption edges are significantly detected and are identified
with OVII and OVIII.
In comparing the ROSAT PSPC and the ASCA SIS, we plot the ratio
of the data from both detectors to the best-fit model to the
ASCA 0.4--4 keV spectrum given in Table 5 (Fig. 7).
The SIS normalization has been adjusted to match
that of the GIS, being in favour of the absolute flux calibration in the GIS
(see Section 3).
The significantly better spectral resolution of the ASCA SIS compared to
the PSPC should provide a better description of the observed spectrum
of the source. The deviation of the PSPC data from the ASCA data is clear.
If both detectors are well calibrated with each other,
the PSPC data would lie around the ratio of unity, at least
in the common 0.5--2 keV band.
A steep slope of the PSPC spectrum is however evident in the energy band.
The inclusion of any extra soft X-ray component below 0.4 keV (e.g., a
steep power-law or blackbody emission)
in addition to the ASCA best-fit model fails to provide an acceptable
fit to the PSPC data ($\chi_{\nu}^2>4$).
This rules out the possibility that the excess seen in Fig. 7 can be due to the
energy response of the PSPC when there is a strong soft excess below
the ASCA energy range. Even if the double power-law
(see Section 3) is used for the continuum model, the diecrepancy in ratio
in Fig. 7 is reduced no more than 15 per cent.
The 0.5--2 keV fluxes for the simultaneous data
from the two instruments are
$3.7\times 10^{-11}$erg~cm$^{-2}$~s$^{-1}$\ from the PSPC and
$3.0\times 10^{-11}$erg~cm$^{-2}$~s$^{-1}$\ from the SIS.
As mentioned in Section 3, the ASCA SIS flux
could be
$\sim 20$ per cent larger than the quoted value, and
thus becomes consistent with the ROSAT flux if the
GIS normalization is trusted.
\section{DISCUSSION}
We now examine possible reasons for the discrepancy between ROSAT and
ASCA in spectral measurements. Steep ROSAT spectra have been noticed
in several analysis on AGN spectra by comparing Einstein Observatory
IPC (Fiore et al 1994; Laor et al 1994; Ciliege \& Maccacaro 1996),
EXOSAT ME (Shartel et al 1997), or BeppoSAX LECS (Mineo et al 1999)
measurements. As mentioned in the Introduction, a similar discrepancy
to the present result was reported from the ASCA/ROSAT
simultaneous observation of the quasar MR2251--178 (C. Otani, priv.
comm.), suggesting that the result presented in this paper is typical.
Since the ROSAT bandpass (0.1--2.4 keV) extends to lower energies than
the IPC ($\sim$0.5--4 keV), the steep spectra from ROSAT observations
have been interpreted as soft excess emission. This is a problem which
is particular to AGN spectra (however, it should be noted that the
PSPC data of quasars investigated by Laor et al (1994) are consistent
with a single power-law and no soft excess component below 0.5 keV is
required). For Seyfert galaxies like NGC5548, reflection also makes
spectral slopes flat in the ASCA band ($>4$ keV). These effects from
the difference in bandpass between satellites could result in
different measurements of spectral slope.
We found a difference of photon-index ($\Delta\Gamma\simeq 0.4$)
measured with the ROSAT PSPC
and the ASCA SIS during the simultaneous observation of NGC5548.
The ASCA data below 4 keV are not affected
by reflection (see Section 3) in this object.
Although we are not able to assess exactly how much the soft excess component
addes to the power-law continuum below 0.4 keV, we see no evidence for
a soft X-ray flare during the observation (see Section 4).
The ROSAT PSPC data are fully consistent with a simple power-law when
the warm absorption is taken into account (Fig. 5) and no extra soft
X-ray component is required.
As Fig. 7 demonstrates that the PSPC data are significantly
steeper than the ASCA data in the 0.5--2 keV band, we can rule out
the possibility of bandpass effects for the different spectral
slope measurements. Therefore errors in calibration between the
two satellites are the likely explanation.
NGC5548 was observed with the ROSAT PSPC several times. The first
observation was carried out during the ROSAT PV phase (1990 July
16--21) with a different detector (PSPC-C), which ceased operating on
1991 Jan 25; the results were published by Nandra et al (1993).
The observed flux suggests that the source during the observation was
in its low state with no soft X-ray flare taking place, but a weak
excess soft X-ray component was found below 0.5 keV (Nandra et al
1993). A simple power-law fit to the spectrum gives $\Gamma = 1.98\pm
0.04$, which is slightly steeper than the ASCA value, but similar.
However, when NGC5548 was observed with the PSPC-C once again during
the ROSAT All Sky Survey (Shartel et al 1997) at a flux level similar
to the present observation and {\tt 01b}, a steep photon-index, $\Gamma =
2.4\pm 0.1$, was found (Shartel et al 1997). After the change of
detector (PSPC-B), the ROSAT spectra show similarly steeper slopes,
$\Gamma\approx 2.37$ for the normal state spectrum (Done et al 1995;
this work). The flat power-law slope reported in Nandra et al (1993)
is exceptional possibly because the source was in its low state,
as other Seyfert 1 galaxies often show an anti-correlation between
X-ray spectral slope and flux (e.g., Mushotzky, Done, \& Pounds 1994).
The calibration of the ASCA SIS for the soft X-ray band has been
a controversial issue.
Photon-index and column density measurements of 3C58, a Galactic
synchrotron nebula, with Einstein Observatory IPC, EXOSAT ME/LE, Ginga LAC
and ASCA agree within uncertainties, although the scatter in \nH\ values
derived from different CCD chips of the SIS imply a systematic error
(Dotani et al 1996).
General agreement on the systematic error in column density measurements
indicates that the SIS
over-estimates \nH\ value by 2$\times 10^{20}$cm$^{-2}$.
The other possible sources which could contribute to the systematic
error of the SIS
calibration are a) DFE and Echo corrections in the response matrices
for the Bright-mode data used here; b) the RDD
effect; and c) uncertainty in the SIS gain.
The comparison with the Faint mode data has verified that the effects
of the DFE and Echo are taken into account in the response matrix
with enough accuracy. The RDD effect appears to reduce the efficiency
by $\sim 5$ per cent, but it is unlikely to affect the
spectral form, since the effect is almost energy independent.
The SIS gain is calibrated within 0.5 per cent accuracy (Dotani
et al 1996).
None of the possible systematic errors in the SIS data
considered above is, therefore,
sufficient to explain the difference between the results from the
two satellites in the observation of NGC5548 reported here.
Simultaneous observations of the quasar 3C273 with ASCA, RXTE and
BeppoSAX have been recently carried out.
The three satellites agree on the measurement of spectral slope
($\Gamma\sim 1.62$ from ASCA SIS and GIS, RXTE/PCA, BeppoSAX/MECS).
The BeppoSAX/LECS, which covers the energy range of
0.1--4 keV, found an edge-like feature at 0.5 keV and a steep soft
X-ray component below 0.4 keV (Grandi et al 1997).
The SIS and the LECS are in a good agreement (Orr et al 1998).
Relative differences in \nH\ measured by the two detectors is
reported, by Yaqoob et al (1997) using the latest calibration, as
$\Delta$\nH $<1.1\times 10^{20}$cm$^{-2}$\ for the S0 and
$\Delta$\nH $<1.9\times 10^{20}$cm$^{-2}$\ for the S1.
During ASCA AO-5, NGC5548 was observed with ASCA and EUVE simultaneously.
The extrapolation of the ASCA spectrum matches the highest energy end
of the EUVE/SW measurement (Kaastra et al 1999; Kunieda et al 1998).
As the EUVE data cover the soft portion of the ROSAT bandpass,
this may provide further support for the ASCA result.
There are a number of discontinuities in the effective area curves of
the instruments originating from the materials used in the X-ray
telescope and detectors. These can cause a problem with the energy
calibration. In the ASCA response, uncertainties due to an oxygen edge
at $\sim$0.5 keV in the SIS detector and a gold edge at 2.2 keV in the
XRT (this residual is approximately modelled with ASCAARF) have been
noticed. A very deep carbon edge is present in the PSPC response,
which removes virtually all the efficiency in the 0.3--0.5 keV band.
Misplacement of an edge would introduce an excess or deficit of
effective area. This problem is more relevant to the PSPC than the
ASCA SIS, given the poor spectral resolution.
\section{Summary}
In summary, (1) the ROSAT PSPC yields a photon-index steeper than the ASCA
SIS by $\Delta\Gamma\simeq 0.4$ for the simultaneous observation of NGC5548.
(2) the steep spectral slope obtained from the PSPC cannot be accounted for by
any soft excess emission below the ASCA bandpass (below $\sim $0.5 keV).
(3) results on the spectral form
from ASCA SIS data on other objects agree with
that from other instruments; Einstein IPC, EXOSAT ME/LE, Gigna LAC,
BeppoSAX LECS and MECS, and EUVE SW.
(4) the measured fluxes seem to be consistent between ASCA and ROSAT, when
the error of the SIS flux measurement is taken into account.
The discrepancy between ASCA and ROSAT also appears to affect
measurements of {\it excess} absorption in clusters of galaxies.
Allen \& Fabian (1997) reported that the excess absorption
column densities of the cooling flow cluster Abell 2199
obtained from the ASCA SIS and Einstein Observatory SSS agree well with
each other, but the ROSAT PSPC gives a significantly lower excess column
density.
The determination of excess column densities depends on the detailed
shape of the spectrum over the 0.5--1 keV range.
It is not clear, if Fig. 7 is a guide, that ROSAT PSPC results will
be correct in this respect.
If, however, the ROSAT PSPC energy calibration is correct, and the
source of the discrepancy lies in its response amplitude, then
energy-dependent results such as total column density estimates
of all classes of source
and the temperatures of clusters where lines (e.g., the Fe-L complex)
are important will be unaffected by the problem discussed here.
Our results indicate that caution should be applied when (over)
interpreting differences between ROSAT and ASCA spectra. Given the
fact that the spectra in Fig. 7 depends on the exact spectral form of a
source, it is not advisable to try to correct other PSPC spectra using
this Figure, since it may lead to a false result. The predicted
spectral response curves of both satellites need reconsideration
(although an improved calibration of the PSPC has been attempted (S.
Snowden, priv. comm.), the calibration files are not publicly
available).
\section*{acknowledgements}
We thank all the members of the ASCA PV team and the ASCA Guest
Observer Facility at NASA's Goddard Space Flight Center. Tadayasu
Dotani is especially thanked for valueable comments on the SIS
calibration.
Tahir Yaqoob and the referee are also thanked for their
helpful comments on the paper.
We acknwledge the Royal Society (ACF),
PPARC (KI) and National Research Council of USA (KN) for support.
|
\section{{\bf Introduction}}
In studying the motion of astronomical objects, astrophysicists utilize the
study of frequency shift of spectral lines.
A redshift of spectral lines is the phenomenon of displacement of the lines
towards longer wavelengths. Although for a lot of astronomical objects
such as stars, redshifts and blueshifts are easily understood in terms of
relative motion, the origin of redshift of quasars has been one of the
most controversial topics for the last few decades. Interpretation of redshift
as Doppler shift has been broadly accepted, although this could not solve the
observed anomalies in the
quasar redshifts and some related problems(Wolf,1998 ).
Recent results in Statistical Optics ( Wolf,1986 ) has made it clear that the shift of
the frequency of the spectral lines can be explained without considering
the relative motion of the observer with respect to the source.
Following this idea, dynamic multiple scattering theory(Datta et al. 1998a,b ) has been
developed to account for the shift as well as the
broadening of the spectral line. It
is shown that when light passes through a turbulent(or inhomogeneous ) medium,
due to multiple scattering effects the shift and the width of the line can be calculated.
Here, a sufficient condition for redshift has been derived and when applicable
the shift is shown to be larger than broadening. The width of the
spectral line can be calculated after multiple scatterings and a relation
can be derived between the width and the shift $z$ which applies to
any value of z (Roy et al.1999).
Using the above condition it is shown that there exists a lower bound of
the source frequency below which no spectra is analyzable in the sense that
the broadening will be larger than the shift of the spectral line and the
identity of the line becomes confused.
This critical source frequency depends very much on the actual nature of the medium
through which the light is propagating. We call this the screen effect
due to the medium.
This new screen effect may play an important role in astronomical domain.
This can also be verified in laboratory experiments.
We first discuss briefly the theory of multiple scattering
in section 2. In section 3 we shall introduce the critical source frequency and its
implications while in section 4 we shall discuss a special type of
screening induced by critical source frequency. Finally, a general discussion
will be made on the possible applications of this type of screening.
\vskip10pt
\noindent
\section{{\bf Dynamic Multiple Scattering Theory}}
First, we briefly state the main results of Wolf's scattering mechanism. Let us
consider a polychromatic electromagnetic field of light of central frequency $\omega_0$
and
width $\delta_0$, incident on the scatterer. The incident spectrum is assumed to be of the
form
\begin{equation}\displaystyle{
S^{(i)}(\omega)= A_0e^{\left[-\frac{1}{2\delta_0^2}(\omega-\omega_0)^2\right]}
}
\end{equation}
The spectrum of the scattered field is given by (Wolf
and James,1996 )
\begin{equation}\displaystyle{
S^{(\infty)}(r \vec{u'} ,\omega')=A\omega'^4\int_{-\infty}^{\infty}
K(\omega,\omega',\vec{u}, \vec{u'})S^{(i)}(\omega)d\omega}
\end{equation}
which is valid within the first order Born approximation (Born and Wolf ,1997 ). Here $K(\omega, \omega')$ is the so called scattering kernel and it
plays the most important role in this mechanism. Instead of studying $\cal
K(\omega,\omega')$ in detail, we consider a particular case for the
correlation function $G(\vec{R},T;\omega) $ of the generalized dielectric
susceptibility $\eta(\vec{r},t;\omega)$ of the medium which is characterized by an
anisotropic Gaussian function
\begin{equation}\displaystyle{
\begin{array}{lcl}
G(\vec{R},T;\omega) & = &
<\eta^*(\vec{r}+\vec{R},t+T;\omega)\eta(\vec{r},t;\omega)>\\ \\
& = &G_0exp\left[-\frac{1}{2}\left(\frac{X^2}{\sigma_x^2}
+\frac{Y^2}{\sigma_y^2}
+\frac{Z^2}{\sigma_z^2}+\frac{c^2T^2}{\sigma_\tau^2}\right)\right]
\end{array}
}
\end{equation}
\vskip5pt
\noindent
Here $~G_0~$ is a positive constant, $~\vec{R}~=~(X,~Y,~Z)~$, and
$~\sigma_x~,~\sigma_y~,~\sigma_z~,~\sigma_\tau~$ are correlation
lengths. The anisotropy is indicated by the unequal
correlation lengths in different spatial as well as temporal directions. $\cal
K(\omega,\omega')$ can be obtained from the four dimensional Fourier Transform
of the correlation function $G(\vec{R},T;\omega) $. In this case $\cal
K(\omega,\omega')$ can be shown to be of the form
\begin{equation}\displaystyle{
{\cal{K}} (\omega,\omega')=B exp\left[-{\frac{1}{2}} \left(\alpha'\omega'^2-
2\beta\omega\omega'+\alpha\omega^2 \right) \right]
}
\end{equation}
where
\begin{equation}\displaystyle{
\left.
\begin{array}{lcl}
\alpha &=& {\frac{\sigma_x^2}{c^2}}u_x^2+{\frac{\sigma_y^2}{c^2}}u_y^2+
{\frac{\sigma_z^2}{c^2}}u_z^2+{\frac{\sigma_\tau^2}{c^2}}\\ \\
\alpha' &=& {\frac{\sigma_x^2}{c^2}}u_x'^2+{\frac{\sigma_y^2}{c^2}}u_y'^2+
{\frac{\sigma_z^2}{c^2}}u_z'^2+{\frac{\sigma_\tau^2}{c^2}}\\ \\
{\rm and} \ \ \beta &=& {\frac{\sigma_x^2}{c^2}}u_xu_x'+{\frac{\sigma_y^2}{c^2}}u_yu_y'+
{\frac{\sigma_z^2}{c^2}}u_zu_z'+{\frac{\sigma_\tau^2}{c^2}}
\end{array}
\right\}
}
\end{equation}
Here $\hat{u}=(u_x,u_y,u_z)$ and $\hat{u'}=(u_x',u_y',u_z')$ are the unit vectors in the
directions of the incident and scattered fields respectively.
Substituting (1) and (4) in (2), we finally get
\begin{equation}\displaystyle{ S^{(\infty)}(\omega')=
A'e^{\left[-\frac{1}{2\delta_0'^2}(\omega'-\bar{\omega}_0 )^2\right]} }
\end{equation}
where
\begin{equation}\displaystyle{
\left.
\begin{array}{lcl}
\bar{\omega}_0 &=& \frac{|\beta|\omega_0}{\alpha'+\delta_0^2
(\alpha\alpha'-\beta^2)}\\ \\
\delta_0'^2 &=& \frac{\alpha\delta_0^2+1}{\alpha'+\delta_0^2
(\alpha\alpha'-\beta^2)}\\ \\
{\rm and} \ \ A' &=& \sqrt{\frac{\pi}{2(\alpha\delta_0^2+1)}}ABA_0\omega_0'^4\delta_0
exp\left[\frac{|\beta|\omega_0\bar{\omega}_0-\alpha\omega_0^2}
{2(\alpha\delta_0^2+1)}\right]
\end{array}\right\}}
\end{equation}
Though $A'$ depends on $\omega'$, it was approximated by
James and Wolf ( James and Wolf, 1990)
to be a constant so that $S^{(\infty)}(\omega')$ can be considered to be Gaussian.
The relative frequency shift is defined as
\begin{equation}\displaystyle{
z=\frac{\omega_0-\bar{\omega}_0}{ \bar{\omega}_0}
}
\end{equation}
where $\omega_0$ and $\bar{\omega}_0$ denote the unshifted and shifted central frequencies
respectively. We say that the spectrum is redshifted or blueshifted
according to whether $~z~>~0~ ~
or~~z~<~0~$ , respectively. Here
\begin{equation}
\displaystyle{
z=\frac{\alpha'+\delta_0^2(\alpha\alpha'-\beta^2)}{|\beta|}-1
}
\end{equation}
\vskip5pt
\noindent
It is important to note that this $z$-number does not depend on the incident frequency,
$\omega_0$. This is a very important aspect if the mechanism is
to apply in the
astronomical domain. Expression (9) implies that the spectrum can be shifted to the blue
or to the red, according to the sign of the term $\alpha'+\delta_0^2(\alpha\alpha'-\beta^2)~>~
|\beta|~$. To obtain the no-blueshift condition, we use Schwarz's Inequality
which implies that $\alpha\alpha'-\beta^2~\geq~0~$. Thus, we can take
$$\displaystyle{
\alpha'~>~|\beta|
}$$
as the sufficient condition to have only redshift by this mechanism.
Let's now assume that the light in its journey encounters many such scatterers. What we
observe at the end is the light scattered many times, with an effect as that stated
above in every individual process.
Let there be N scatterers between the source and the
observer and $z_n$ denote the relative frequency shift after the $n^{th}$ scattering of the
incident light from the $(n-1)^{th}$ scatterer, with $\omega_n$ and $\omega_{n-1}$ being the
central frequencies of the incident spectra at the $n^{th}$ and $(n-1)^{th}$ scatterers. Then by
definition,
$$\displaystyle{
z_n =\frac{ \omega_{n-1}-\omega_n}{\omega_n}, ~~~~~~n~=~1,~2,~.~.~.~.,~N
}$$
or,
$$\displaystyle{
\frac{\omega_{n-1}}{\omega_n}=1+z_n,~~~~~~n~=~1,~2,~.~.~.~.,~N
}$$
Taking the product over $ n$ from $n ~=~ 1$ to $n~ =~ N$, we get,
$$ \displaystyle{
\frac{\omega_0}{\omega_N}=(1+z_1)(1+z_2)~.~.~.~.~.~(1+z_N)
}$$
The left hand side of the above equation is nothing but the ratio of the
source frequency and the final or observed frequency $z_f$. Hence,
\begin{equation}\displaystyle{
1 + z_f = (1+z_1)(1+z_2)~.~.~.~.~.~(1+z_N)
}
\end{equation}
\vskip5pt
\noindent
Since the $z$-number due to such effect does not depend upon the central
frequency of the incident spectrum, each $z_i$ depends on $\delta_{i-1}$ only,
not $\omega_{i-1}$ [here $\omega_j$ and $\delta_j$ denote the central frequency
and the width of the incident spectrum at $(j+1)^{th }$ scatterer]. To find the
exact dependence we first calculate the broadening of the spectrum after N number of scatterings.
Here, we are considering the multiple scattering on the assumption that the
scatterers are mutually incoherent. In this case the cross
terms in the spectrum of the scattered field are zero. In other words we are not
including higher order scatterings. Moreover, we are also
considering small angle scatterings so that there will be small deviation from
the actual path of light. As a result the superimposed spectra from different
scatterings will result in a signal in the forward direction only. This is similar to
the single scattering result of James and Wolf.
\vskip8pt
\subsection{Effect of Multiple Scatterings on the Spectral Line Width}
\vskip5pt
From the second equation in (7), we can easily write,
\begin{equation}\displaystyle{
\left.
\begin{array}{lcl}
\delta_{n+1}^2 & = &
\frac{\alpha\delta_n^2+1}{\alpha'+(\alpha\alpha'-\beta^2)\delta_n^2} \\ \\
& = & \left(\frac{\alpha\delta_n^2+1}{\alpha'}\right)\left[1+\delta_n^2\left(
\frac{\alpha\alpha'-\beta^2}{\alpha'}\right)\right]^{-1}
\end{array}
\right\}}
\end{equation}
From (13), we can also write
\begin{equation}
\displaystyle{
\omega_{n+1} =
\frac{\omega_n|\beta|}{\alpha'+(\alpha\alpha'-\beta^2)\delta_n^2}
}
\end{equation}
Then from (11) \& (12), we can write
\begin{equation}
\displaystyle{
\left.
\begin{array}{rcl}
z_{n+1}&=&\frac{\omega_n-\omega_{n+1}}{\omega_{n+1}}\\ \\
&=&\frac{\alpha'+(\alpha\alpha'-\beta^2)\delta_n^2}{|\beta|}-1\\ \\
&=&\frac{\alpha'}{|\beta|}\left\{1+\left(\frac{\alpha\alpha'-\beta^2}{\alpha'}
\right)\delta_n^2\right\}-1
\end{array}
\right\}}
\end{equation}
\noindent
Let's assume that the redshift per scattering process is very small, {\it i.e.,}
$$0~<~\epsilon ~ = ~ z_{n+1}~~<<~1 $$ for all $n$.
\vskip5pt
\noindent
Then,
$$\displaystyle{
\begin{array}{lrcl}
& 1+\epsilon & = & \frac{\alpha'}{|\beta|}\left\{1+\left(\frac{\alpha\alpha'
-\beta^2}{\alpha'}\right)\delta_n^2\right\}\\ \\
{\rm or,}& (1+\epsilon)\frac{|\beta|}{\alpha'} & = & 1+\left(\frac{\alpha\alpha'-
\beta^2}{\alpha'}\right)\delta_n^2
\end{array}
}$$
In order to satisfy this condition and in order to have a redshift
only ( or positive z),
we see that the first factor $\frac{\alpha'}{|\beta|}$ in the right term cannot be much
larger than 1, and, more important,
\begin{equation}
\left(\frac{\alpha\alpha'-\beta^2}{\alpha'}\right)\delta_n^2~~<<~~1
\end{equation}
In that case, from (11), after neglecting higher order terms, the expression for
$~~\delta_{n+1}^2~~$ can be well approximated as:
$$\displaystyle{
\delta_{n+1}^2 \approx
\left(\frac{\alpha\delta_n^2+1}{\alpha'}\right)\left[1-\delta_n^2\left(
\frac{\alpha\alpha'-\beta^2}{\alpha'}\right)\right]
}$$
which, after carrying out a simplification, gives a very important recurrence relation:
\begin{equation}\displaystyle{
\delta_{n+1}^2 = \frac{1}{\alpha'}+\frac{\beta^2}{\alpha'^2}\delta_n^2.
}
\end{equation}
Therefore,
$$\displaystyle{
\begin{array}{lcl}
\delta_{n+1}^2 & = & \frac{1}{\alpha'}+\frac{\beta^2}{\alpha'^2}\delta_n^2 \\
\\
& = & \frac{1}{\alpha'}+\frac{\beta^2}{\alpha'^2}[ \frac{1}{\alpha'}
+\frac{\beta^2}{\alpha'^2}\delta_{n-1}^2] \\ \\
& = & \left(\frac{\beta^2}{\alpha'^2}\right)^2\delta_{n-1}^2
+\frac{1}{\alpha'}\left(1+\frac{\beta^2}{\alpha'^2}\right) \\ \\
& .~. & .~.~.~.~.~.~.~.~.~.~.~.~.~.~ \\ \\
& = & \left(\frac{\beta^2}{\alpha'^2}\right)^{n+1}\delta_0^2
+\frac{1}{\alpha'}\left(1+\frac{\beta^2}{\alpha'^2}+~.~.~.~.~
\frac{\beta^{2n}}{\alpha'^{2n}}\right).
\end{array}
}$$
Thus
\begin{equation}\displaystyle{
\delta_{N+1}^2 =
\left(\frac{\beta^2}{\alpha'^2}\right)^{N+1}\delta_0^2
+\frac{1}{\alpha'}\left(1+\frac{\beta^2}{\alpha'^2}+~.~.~.~.~
\frac{\beta^{2N}}{\alpha'^{2N}}\right).
}
\end{equation}
As the number of scattering increases, the width of the spectrum obviously
increases and the most important topic to be considered is whether this width
is below some tolerance limit or not, from the observational point of view.
There may be several measures of this tolerance limit. One of them is the {\it
Sharpness Ratio}, defined as
$$ Q=\frac{\omega_f}{\delta_f}$$
where $~\omega_f ~$ \& $~ \delta_f ~$ are the mean frequency \& the width of
the observed spectrum.
After $N$ number of scatterings, this sharpness ratio, say $Q_N$, is given by
the following recurrence relation :
$$Q_{N+1}=Q_N \sqrt{\frac{\alpha'}{\alpha'+(\alpha\alpha'-\beta^2)\delta_N^2}-
\frac{1}{\alpha\delta_N^2+1}}.$$
It is easy to verify that the expression under the square root lies between 0 \&
1. Therefore, $Q_{N+1}~~<~~Q_N$, and the line is broadened as the scattering
proceedss on .
Under the sufficient
condition of redshift [i.e., $|\bt|~<~\al'$ ](Datta et al.,1998c) it
was shown that in the observed spectrum
\begin{equation}
\Dl\om_{n+1} \gg \dl_n
\end{equation}
\noindent
if the following condition holds:
\begin{equation}
\frac{\dl_n\om_0(\al\al'-\bt^2)}{\al'+(\al\al'-\bt^2)\dl_n^2} \gg 1
\end{equation}
where $\om_0$ is the source frequency.
The relation (18) signifies that the shift is more prominent than the effective
broadening so that the spectral lines are observable and can be analyzed. If,
on the other hand ,the
broadening is higher than the shift of the spectral line, it will be impossible
to detect the shift from the blurred spectrum. Hence we can take relation
(18) to be one of the conditions necessary for the observed spectrum to be
analyzable. For large $N$ (i.e., $N \rightarrow \infty $ ), the series in the second
term of right hand side of (16) converges to a finite sum and we get
$$ \displaystyle{
\delta_{N+1}^2 =
\left(\frac{\beta^2}{\alpha'^2}\right)^{N+1}\delta_0^2
+\frac{\alpha'}{\alpha'^2 - \beta^2}.}$$
If $\delta_0$ is considered as arising out of Doppler broadening only, we can
estimate $\delta_{\rm Dop} \sim 10^9$ for $T = 10^4$ K. On the other hand, for an anisotropic
medium, we can take $\sigma_x = \sigma_y = 3.42 \times 10^{-1}, \ \sigma_z =
8.73 \times 10^{-1}, \ \alpha' = 8.68 \times 10^{-30}, \ \alpha = 8.536 \times
10^{-30} {\rm and} \beta = 8.607 \times 10^{-30}$
for $\theta = 15^0$(James and Wolf,1990). Then the
second term of the above expression will be much larger than the first
term, and effectively, Doppler broadening can be neglected in comparison to
that due to multiple scattering effect.
\vskip5pt
\noindent
Now if we consider the other condition, $ i.e.,~~\al'~<~|\beta|$, the series in
(16) will be a divergent one and $\dl_{N+1}^2$ will be finitely large for large
but finite $N$. However, if the condition (17) is to be satisfied, then the
shift in
frequency will be larger than the width of the spectral lines. In that case
the
condition $\al'~<~|\beta|$ indicates that blueshift may also be observed but
the width of the spectral lines can be large enough depending on how large the
number
of collisions is. So in general, the blueshifted lines should be of larger widths
than the redshifted lines and may not be as easily observable.
\vskip10pt
\section{\bf Critical Source Frequency}
\vskip5pt
\noindent
Rearranging equation (18) we get,
\beq
\left(\dl_n-\frac{\om_0}{2}\right)^2 \ll
\frac{\om_0^2}{4}-\frac{\al'}{\al\al'-\bt^2}
\eeq
Since the left side is non-negative, the right side must be positive. Moreover,
since the mean frequency of any source is always positive, {\it i.e.,}
$\om_0 \geq 0$, we must have
\beq
\om_0 \gg \sqrt{\frac{4\al'}{\al\al'-\bt^2}}.
\eeq
We take the right side of this inequality to be the {\bf \it critical source
frequency } ${\bf \om_c}$ which is defined here as.
\beq
\om_c = \sqrt{\frac{4\al'}{\al\al'-\bt^2}}.
\eeq
Thus for a particular medium between the source and the observer, the critical
source frequency is the lower limit of the frequency of any
source whose spectrum can be clearly analyzed. In other words, the shift of any
spectral line from a source with frequency less than the critical source frequency for that
particular medium cannot be detected due to its high broadening.
\vskip5pt
\noindent
We can now classify the spectra of the different sources, from which light
comes
to us after passing through a scattering medium characterized by the parameters
$\al$, $\al'$,
$\bt$. If we allow only small angle scattering in order to get prominent
spectra, according to the Wolf mechanism, they will either be blueshifted or
redshifted. The redshift of spectral lines may or may not be detected according
to whether
the condition (20) does or does not hold. In this way those sources whose spectra
are redshifted, are classified in two cases, {\it viz.,} ${\bf \om_0~~>~~\om_c}$ and
{\bf $\om_0~~\leq~~\om_c$}. In the first case, the shifts of the spectral lines can
be easily detected due to condition (17). But in the later case, the spectra
will suffer from the resultant blurring.
\vskip10pt
\subsection{\bf Critical Source Frequency and Anisotropy}
\vskip5pt
\noindent
The anisotropy discussed in this paper is statistical in nature. We have considered
both temporal as well as spatial correlations to induce such spectral changes.
The spatial anisotropy has been characterized by unequal correlation lengths along three
mutually perpendicular directions. We have used the term {\it strong anisotropy}
in a particular direction ( among the above three) to specify that the correlation
length along that direction is very large compared to that of the other two directions.
We now find the critical source frequency in both the isotropic and anisotropic cases.
{\it Spatially Isotropic Case}
If the spatial correlation lengths are equal in every direction , i.e. ${ \sigma_x
= \sigma_y = \sigma_z =\sigma }$ then the parameters ${\alpha, \alpha\prime , \beta}$
reduce to
\beq
\alpha = \alpha^\prime = \frac{{\sigma}^2}{c^2} + \frac{{\sigma_\tau}^2}{c^2}
; \ \ \ \beta = \frac{{\sigma}^2}{c^2} \cos\theta +
\frac{{\sigma_\tau}^2}{c^2}
\eeq
Therefore from (21) we get
\beq
\omega_c = \frac{2c}{\sigma} \sqrt {\frac{\sigma^2 + {\sigma_\tau}^2}
{{(1 - \cos\theta)}{{\sigma^2(1 + \cos\theta)+ 2 {\sigma_\tau}^2}}}}
\eeq
Therefore,
$$\displaystyle{
\begin{array}{lcl}
\omega_c & = & \frac{2c}{\sigma} cosec \theta ; \ \ {\rm if} \ \ \sigma >>
\sigma_\tau \nonumber \\
\omega_c & = & \frac{2c}{\sigma}\sqrt{\frac{1}{2(1-\cos\theta)}} \ \
{\rm if}\ \sigma << \sigma_\tau
\end{array}}
$$
In both cases $\sigma_c$ varies inversely with $\sigma$. This is
equivalent to taking $ \sigma_\tau = 0$ i.e. for white noise.
{\it Spatially Anisotropic Case }
Let there be a strong anisotropy along, say, the x-axis. Then
$\sigma_x >> \sigma_y , \sigma_z$. Consequently,
\beq
\alpha = \frac{\sigma^2}{c^2} + \frac{\sigma_{\tau}^2}{c^2}; \ \
\alpha^\prime = \frac{\sigma^2}{c^2} {u_x^\prime}^2 + \frac{{\sigma_\tau}
^2}{c^2}; \ \
\beta = \frac{\sigma^2}{c^2} u_x {u_x^\prime} + \frac{{\sigma_\tau}^2}{c^2}
\eeq
We can define
$$k = \frac{\sigma_x}{\sigma_\tau}$$ as a ratio of the correlation lengths along the spatial and temporal
directions. We now consider the variation of $\omega_c$ with $k$ in the following two case :
1. $\sigma_\tau$ is fixed ; $k$ varies with $\sigma_x$
2. $\sigma_x$ is fixed ; $k$ varies inversely with $\sigma_\tau$.\\
The nature of the variation of $\omega_c$ against $k$ in the above two
cases has been studied by Datta et al.(Datta et al.1998a,b).
We can easily find the critical source frequency for
$ k = 1$ as
\beq
\omega_c = \frac{2c \sqrt{{u_x ^\prime}^2 + 1}}{\sigma
{|u_x - u_x^\prime |}}
\eeq
From (24) and (25) we can say that the stronger the anisotropy, the lower the
critical source frequency and hence the greater the scope of more and more spectra
to be anlyzable. This induces a special type of screening that we are now going to
discuss.
\section{\bf Screening Effect Induced by the Critical Source Frequency}
Collisional mechanism of light beams give some sort of screening which may
or may not show frequency shift of the spectral lines. Here the scatterer is the
medium with its dielectric susceptibility randomly fluctuating both spatially
and temporally. The critical source frequency induces a special type of screening
which not only changes the colour of the incident lines but also broadens their
widths. Specifically speaking, when an incident line with its peak frequency
$\omega_0$ encounters the scattering medium whose critical source frequency
is $\omega_c$ , then
\begin{enumerate}
\item frequency shift occurs
\item the spectral width increases and this increased width will be less than or
greater than the frequency shift according as $\omega_0$ is greater or less than
$\omega_c$.
\end{enumerate}
Let a spectral line with wavelength greater than the critical wavelength of the
medium propagate and hence be scattered. The scattered line, according to the screening effect, will be of width greater than
the shift of that line. Therefore, the observer will find the broadening of the
line more prominent than its shift and hence it will be difficult to analyze the line.
On the other hand, if the wavelength of the line is less than the critical
wavelength, its shift will dominate the broadening.
In the case of strong spatial anisotropy ( without loss of generality, we may assume
it along the x-axis) it can be shown that the maximum value of this critical wavelength for a
particular medium ( i.e. for a particular set of the spatial correlation
lengths) is given by the following relation:
\beq
\lambda_c = \pi \sqrt{\sigma_x^2 + \sigma_{\tau}^2}
\eeq
Thus this type of screening involves only two parameters - the spatial and temporal
correlation lengths. If we take a particular medium and may know its spatial
and temporal correlation lengths , we can easily find that particular wavelength
from (26) which enables us to get the nature of the screening. To be more specific, we can find
out the wavelength zone beyond which the anlyzability of the spectral lines will
be difficult.
\vskip10pt
\noindent
\section{\bf Possible Implications for Quasar Redshifts}
\vskip5pt
The screening effect as discussed above might play a significant
role in observations of quasars as well as in the laboratory.
If we consider the spectra originating from distant quasars,
it might be possible to explain the large redshift and broadening
by this type of screening. We may try to visualize
this at least in a simple fashion for the redshifts in the case of
NGC 4319 Galaxy and Markarian 205 Quasar pair as follows.
\vskip5pt
\noindent
\subsection{\it Screening Effect and Galaxy-Quasar Pair}
\vskip5pt
\noindent
The above mentioned screening effect gives in this case the simple explanation
for the redshift controversy in a galaxy-quasar pair ( Sulentic, 1989)
[ NGC 4319 is the Galaxy and
Markarian 205 the quasar].
They appear to be connected in very deep photographs but the quasar's redshift
is found to be much higher ( $z$-number is almost 12 times ) than that of the
galaxy. This empirical evidence is normally taken as one of the strongest
arguments in favour of non-cosmological redshifts. Weedman ( Weedman,1986 )
made a critical survey of this situation using statistical arguments for
comparisons of positions in the sky for quasars and galaxies. It has been
claimed ( Weedman, 1986) that if one uses this kind of statistical test
for several samples, none shows evidence for association of high
redshift quasars with low redshift galaxies. This impacts on hypotheses of both
non-cosmological redshifts and gravitational lensing. Here, we want to
emphasize that this kind of association of high redshift quasars with low
redshift galaxies is a possibility in our theory of shift of the spectral lines
by multiple scatterings. We can try to visualise the situation in the following
way :
We know very little little about the atmospheres surrounding those distant galaxies
and quasars. We nevertheless, assume that they are very different. Therefore the medium
through which we observe the quasar is much different from that for the
galaxy. Two sets of parameters $(~~\al,~~~\al',~~~\bt~~)$ characterising the
media thus give rise to two different critical source frequencies, say $\om_G$
( for galaxy ) and $\om_Q$ ( for quasar ). Now if the lines emitted from the
galaxy have frequencies less than $\om_G$, then the Wolf-shift cannot be
observed. On the other hand, if those lines emitted from the quasar have
greater frequencies than $\om_Q$, prominent Wolf-shifts will be observed. The
Doppler shift ( or the shift as interpreted in the expanding Universe picture ) might be
present in both cases. Thus the net shift observed in the case of the quasar
will be higher than in the case of the galaxy. Let the $z$-number of the
galaxy be ${(\rm say z_1=)} a$ and the $z$-number of the quasar due to the Wolf effect be ${(\rm say z_2=)} b$. Again the resultant $z$ number can be written
as $ 1 + z = (1+Z_1)(1+z_2)$. Then
$$ \begin{array}{lrcl} &12a&=&a+b+ab\\
\Rightarrow&b&=&\frac{11a}{1+a}
\end{array}$$
This indicates that the distance of the quasar is just the same as that of the
galaxy, but the shift in the first case is much higher than that of the later.
We can envisage this phenomena as some sort of {\it screening} operation for the
Wolf Effect. In other words, new type of screening arises due to the nature of
the medium . This will be studied in details in future work.
Lastly, we should mention that the critical source frequency relating
to the screening effect which plays a crucial role in explaining both redhsift and
spectral width can be tested in {\it laboratory experiments}.
\vskip20pt
\noindent
{\bf Acknowledgements:} One of the authors ( S.D.) greatly thanks World
Laboratory, Laussane for financial support during this work and Prof. B. K.
Datta ( Director, World Laboratory, Calcutta Branch ) for encouragement.
The author(S.R.) is indebted to Prof. Jack Sulentic,University of Alabama,
for valuable suggestions and comments. The authors like to thank the referee
for valuable suggestions.
\newpage
{\bf REFERENCES}
\begin{enumerate}
\item Born, M.\& Wolf,E. 1997, Principle of Optics, 6th edition,
Pergamon, Oxford.
\item Datta S, Roy S, Roy M \& Moles M, 1998a,, Int. Jour. of Theo. Phys.,
{\bf 37},N4, 1313.
\item Datta S, Roy S, Roy M \& Moles M, 1998b, Int.Jour.Theort.Phys.,
{\bf 37},N5, 1469.
\item Datta S, Roy S, Roy M and Moles M, 1998c, Phys.Rev.A,{\bf 58},720
\item James D.F.V and Wolf E, 1990, Phys. Lett. A, {\bf 146}, 167.
\item Roy S., Kafatos M. and Datta S. 1999, Broadening
of Spectral Lines Due to Dynamic Multiple Scattering and The
Tully-Fisher Realation , Physical Review A ,{\bf 60 },273 .
\item Sulentic J.W.1989, ApJ.{\bf 345}, 54.
\item Weedman D.W.,1986, Quasar Astronomy, Cambridge University
Press.
\item Wolf Emil ,1998, The Redshift Controversy and Correlation-Induced Spectral Changes" Eds.Pratesi R. and Ronchi L.
Spectral changes, Eds.Pratesi R. and Ronchi I.
in Waves, Information and Foundations of Physics
{\bf 60}, p.41 .
\item Wolf E.,1986,Phys.Rev.Lett.{\bf 56}, 1370.
\item Wolf E. and James D.F.V.,1996 ,Rep. Progr. Phys.{\bf 59}, 771.
\end{enumerate}
\end{document}
|
\section{Introduction}
High redshift ($\rm z > 1.8$) QSOs, especially radio-quiet objects, are
rare in the ROSAT All-Sky Survey (RASS, Voges {\it et al. }, 1996).
A comparison of the radio-loud and radio-quiet QSOs detected in the
RASS indicated that the X-ray detection rate of radio-quiet QSOs drops
much faster towards higher redshift than that of radio-loud QSOs
(Yuan {\it et al. } 1998). The catalogue of Yuan {\it et al. } (1998) (see Table 1 in
that paper) contains 18 radio-quiet QSOs with $\rm z > 2$
detected in the RASS. In the northern sky with galactic
latitude $\rm b>35^o$, only 20 QSOs with $z>1.3$ and four QSOs with $z>2$
have been known in the literature so far.
At high redshifts the rest frame energy range is shifted to
higher energies, e.g., for $\rm z = 2.5$ the ROSAT/PSPC
(sensitive
between 0.1 and 2.4\,keV) receives photons emitted between 0.35 and
8.4\,keV. Only upper limits exist for radio-quiet QSOs above 100\,keV
whereas
several radio-loud objects with blazar properties have been detected
above 100 keV (e.g. von Montigny {\it et al. } 1995).
From several high redshift QSOs low energy cut-offs have been
found,
which are interpreted as
due to the strong low energy absorption probably caused by
the material near the QSO itself (Elvis {\it et al. } 1998). Higher resolution
spectroscopy with future X-ray telescopes, e.g., AXAF, XMM,
and
Astro-E,
will be able to quantify the contribution of metals to this proposed strong
absorption.
Low energy cut-offs have mainly been discovered in radio-loud
objects, only some tentative
evidence for low energy cutoffs are known in several
radio-quiet QSOs. However, the statistics
above $\rm z >2.5$ is poor; only
three radio-quiet objects
are included in a previous study (Fiore {\it et al. } 1998).
In addition, high redshift QSOs found in the RASS belong to the
most luminous members of
this class, making the
upper end of the luminosity function. It
is known from the magnification bias (Borgeest {\it et al. } 1991)
that such objects have high probability to be gravitationally
magnified. This is particularly true for radio-quiet objects, which seem to
be the more common but intrinsically
X-ray faint population. Among the radio-quiet $\rm z > 2$ QSOs in the RASS,
two are known gravitationally lensed objects.
HE1104--1805 ($\rm z = 2.4$) is a
double QSO detected in Hamburg (Wisotzki {\it et al. } 1993). The
other object is RX\,J0911.4+0551 ($\rm z = 2.8$, Bade {\it et al. } 1997),
a gravitationally lensed system with at least four QSO images.
The full sky coverage and the sensitivity of the RASS make it the
most appropriate
for finding these rare objects.
Radio-loud QSOs have
been selected mainly during the identifications of bright radio sources.
Projects which may be able to select
radio-quiet QSOs down to $\rm B = 18\,mag$ on large sky areas are under
way
(e.g. SDSS, see Kent 1994; Hamburg/ESO Survey, see Wisotzki {\it et al. } 1996);
but up to now results are only available for comparatively small and
patchy areas of the sky.
Therefore studies using correlations between existing AGN catalogues and
the
RASS to determine the relation between radio bright and quiet QSOs in
the X-rays are probably affected by selection effects.
In order to enlarge the current sample of X-ray luminous radio-quiet
high redshift QSOs,
we start this survey project to select this kind of QSOs in the
RASS
using optical observations. Our study will also verify whether some
selection
effects are serious in the above mentioned correlation
studies.
\section{High redshift QSO candidates selection}
We used the
digitized objective prism spectra from the Hamburg Quasar Survey
(HQS, Hagen {\it et al. } 1995) for the selection of QSOs with high redshifts.
Currently the Hamburg identification project of RASS sources (Bade
{\it et al. } 1998) covers about 10,200\,deg$^{2}$ of the northern high galactic
latitude sky ($\rm |b| > 20^o$). This survey uses photo plates which
are sensitive between 5400\,\AA\ and the
atmospheric limit at $\rm \sim 3400\AA$. Thus the
Ly-$\alpha$ line
can be detected between $\rm 1.8 < z < 3.2$.
If the CIV\,1549
emission line is discernible in the objective prism spectrum, too, the
reliability of the classification is very high ($>90\%$). If only one
strong
emission line is visible, other strong emission lines can be source of
the line
(e.g.. CIII 1909, MgII 2798, [OIII]5007) and smaller redshifts are
possible. Follow-up observations of the QSO candidates are thus necessary.
The classification of AGN candidates in the Hamburg identification
project
of RASS sources is based mainly on their blue continuum which can be
well
described by a power law. For the selection of QSOs with higher
redshifts we
inspected the objective prism spectra by eye, looking for strong
emission lines.
If the magnitude of the AGN candidate is near the plate limit, it is
difficult
to distinguish emission lines from noise.
Another problem for number counts is the varying detection limit
of the objective prism plates. Taking all effects into
account, $\rm B = 18$ is a good approximation of the detection limit
for strong emission lines.
Following these procedures we finally selected 22 QSO candidates
in the northern sky with galactic latitude $\rm b > 35^o$.
In the 1st to 7th columns of Table 1, we list their ROSAT names,
coordinates of the
X-ray sources and their optical counterparts, offsets between the X-ray
and optical
positions, X-ray fluxes, B-band magnitudes and 1.4GHz radio fluxes
from NVSS (Condon {\it et al. } 1998).
The X-ray fluxes between 0.1 and 2.4\,keV are
calculated by assuming a power law with photon index $-1.8$
and only galactic neutral hydrogen absorption.
The follow-up observations were split into two observing
campaigns in the spring of 1998
when the area of 300 HQS plates covering 6500$\,\mbox{deg}^{2}$
was observable.
Because varying\\
weather conditions prevented us from observing the
high redshift QSO candidates uniformly, the surveyed area is only a rough
estimate.
\section{Spectroscopic observations of high redshift QSO candidates}
We have performed the spectroscopic observations on the 22
selected high redshift QSO candidates
using the 2.16m telescope at Xinglong station of Beijing
Astronomical Observatory.
The exposure time varied from 10 to 70 minutes for each object depending
on weather conditions
and object brightness.
The spectra of these high redshift QSO candidates were obtained after the
standard sky subtraction and relative
flux calibration using the MIDAS
software package developed at the European Southern Observatory (Banse {\it et al. }
1983).
The observation dates, exposure times, types and redshifts of these
high redshift candidates are given in the 8th to 11th columns of Table 1. The
12th column gives the rest frame luminosity
$\rm L_{X}$ between $\rm 0.1 - 2.4\,keV$.
For the
determination of
the luminosity
$\rm H_{0} = 50\,km\,s^{-1}\,Mpc^{-1}$ and $\rm q_{0} = 0$
are assumed.
\tabcolsep 1.32mm
\begin{table*}
\caption[]{ Spectroscopic observation results of high redshift QSO
candidates
\label{basistabel}}
\begin{tabular}{lccrrlrcrlcl}\\ \hline
\multicolumn{1}{c} {ROSAT names} & X-ray coordinates &Opt. coordinates&
Offset & $f_x^{*}$
&B & $f_{1.4G}$ & Date & Exp. & Type & z & L$_x^{**}$ \\
& R.A. (2000.0)~~ Decl.&R.A. (2000.0)~~ Decl.&$('')$ & & &(mJy) & &(s)&
&
& \\
\hline
RX J0923.2+4602 &09 23 12.6 ~~46 02 42& 09 23 12.7 ~~46 02 42 & 2
& 4.9&18.4 & 11.7& 97/12/07 & 4200 & QSO & 0.729 & 1.87 \\
RX J0959.8+0049 & 09 59 48.9 ~~00 49 27 &09 59 46.9 ~~00 49 16 &33 &3.5
& 19 & &97/12/07 & 3600 & QSO & 2.243 & 26.9 \\
RX J0959.8+5942 & 09 59 49.3 ~~59 42 53& 09 59 48.5 ~~59 42 50 &8 &4.8&
17.5 & &97/12/07 & 1800 & Seyfert & 0.168 & 0.07 \\
RX J1059.8+0909 & 10 59 51.5 ~~09 09 15& 10 59 51.0 ~~09 09 05 &13& 4.0&
17.1 & &98/03/05 & 3000 & QSO & 1.683 & 13.6 \\
RX J1112.4+1101 & 11 12 25.5 ~~11 01 03& 11 12 25.4 ~~11 01 03 &2& 5.0&
18 & &98/05/24 & 900 & QSO & 0.636 & 1.37 \\
RX J1144.9+5434 &11 44 55.8 ~~54 34 58& 11 44 54.9 ~~54 34 51 &11 &12.7
& 18.4 & &98/03/05 & 1800 & QSO & 0.437 & 1.44 \\
RX J1208.3+5240 & 12 08 22.3 ~~52 40 40& 12 08 22.3 ~~52 40 12 &29
&13.0& 17.1 & 41.1& 98/03/05 & 1800 & QSO & 0.435 & 1.46 \\
RX J1243.8+0828 & 12 43 52..3 ~~08 28 25& 12 43 52.5 ~~08 28 26 &3 &
3.3& 17.6 & &98/05/24 & 900 & QSO & 0.384 & 0.28 \\
RX J1259.8+3423$^a$ &12 59 48.7 ~~34 23 25& 12 59 48.9 ~~34 23 19 &4 & 6.0&
16.9 & 11.8&98/05/24 & 600 & QSO & 1.376 & 11.7 \\
RX J1353.0+2947 & 13 53 00.0 ~~29 47 42& 13 52 59.1 ~~29 47 39 &14&
23.3& 18 & &98/05/24 & 1100 & QSO & 0.436 & 2.63 \\
RX J1404.1+0937 & 14 04 10.4 ~~09 37 45& 14 04 10.7 ~~09 37 45 &3
& 9.1& 17.3 & 20.1&98/05/24 & 600 & QSO & 0.437 & 1.03 \\
RX J1410.2+0811 & 14 10 13.6 ~~08 11 29 & 14 10 13.7 ~~08 11 28 &1 &5.2 &
19 & &98/05/24 & 800 & Seyfert & 0.088 & 0.02 \\
RX J1425.0+2749 &14 25 03.0 ~~27 49 21& 14 25 02.6 ~~27 49 10 &12 & 2.6&
18.8 & & 98/03/05 & 1800 & QSO & 2.346 & 22.8 \\
RX J1428.9+2710 & 14 28 54.0 ~~27 10 44& 14 28 53.0 ~~27 10 39 &17
& 4.7 & 17.8 & 7.2 &98/05/24 & 900 & QSO & 0.443 & 0.55 \\
RX J1441.2+3450 & 14 41 17.3 ~~34 50 53& 14 41 17.6 ~~34 50 52 &2 &
20.0& 16.8 & &98/05/24 & 800 & QSO & 0.352 & 1.34 \\
RX J1451.9+7214 &14 51 53.6 ~~72 14 41 & 14 51 54.1 ~~72 14 44 &6 &4.1 &
18 & &98/03/05 & 600 & QSO & 0.749 & 1.67 \\
RX J1514.3+4244 &15 14 20.4 ~~42 44 39& 15 14 20.4 ~~42 44 44 &5 &
15.9& 18.0 & 6.6 &98/05/24 & 600 & Seyfert & 0.152 & 0.18 \\
RX J1541.2+7126 & 15 41 16.3 ~~71 26 06 & 15 41 15.2 ~~71 25 58 &9 &3.9 &
18 & &98/03/05 & 900 & QSO & 1.418 & 8.26 \\
RX J1548.3+6949 & 15 48 18.7 ~~69 49 30& 15 48 16.7 ~~69 49 33 &11
& 8.3& 18.2 & 33.6& 98/05/24 & 900 & QSO & 0.375 & 0.66 \\
RX J1604.6+5714 &16 04 38.6 ~~57 14 37& 16 04 37.3 ~~57 14 36 &11
&7.3& 17.3 & 496.8& 98/05/24 & 800 & QSO & 0.725 & 2.75 \\
RX J1616.9+3621& 16 16 55.3 ~~36 21 33& 16 16 55.5 ~~36 21 34 &2
&0.7 & 16.9 & 330.9 &98/03/05 & 1200 & QSO & 2.259 & 5.50 \\
RX J1701.4+3511 &17 01 25.3 ~~35 11 50& 17 01 24.6 ~~35 11 56 &19 & 4.7 &
18 & &98/03/05 & 900 & QSO & 2.115 & 30.5 \\
\hline
\end{tabular}
\\ \vskip 0cm
\noindent Notes: $^{*)}$ $f_x$ is in unit of
$10^{-13}\,erg\,cm^{-2}\,s^{-1}$;
$^{**)}$ L$_x$ is the luminosity between $\rm 0.1 - 2.4\,keV$ in units
of
$\rm 10^{45}\,erg\,s^{-1}$;
$^{a)}$ A radio detected object but is formally radio quiet\\
\end{table*}
\vskip -.2cm
\begin{figure*}
\vskip -3.5cm
\hskip -1.5cm
\rule{0.4pt}{4cm
\psfig{file=8350f1.ps,width=21.0cm,height=24.7cm,angle=0}
\vskip -11cm
\caption{The $8'\times 8'$ POSS-I images of six new radio-quiet, X-ray
luminous quasars with redshift
larger than 1.3. The arrow points to the new quasar in each image.}
\end{figure*}
\begin{figure*}
\vskip -2.5cm
\hskip -1.8cm
\rule{0.4pt}{4cm
\psfig{file=8350f2.ps,width=23cm,height=19.2cm,angle=0}
\vskip -4.0cm
\caption{Spectra of 6 new high redshift radio-quiet, X-ray luminous
quasars. Three most significant emission lines in each spectrum are
labeled.}
\end{figure*}
Table 1 shows that
seven objects are QSOs with redshift larger
than 1, and four of them have redshift larger than 2.
12 candidates are
QSOs with redshifts between 0.3 and 1.
Three candidates are low redshift Seyfert galaxies.
If the strongest emission line in the objective prism
spectrum is $\rm
Ly{\,\alpha}$, redshift of the object is larger than 2.
If the strongest line is CIV or MgII in the prism spectrum, the redshift
is larger than 1 or between 0.3 and 1, respectively. However,
in several cases the emission
line in the prism spectrum is produced by a strong forbidden
emission line of Seyfert galaxies, such as [OII] or [NeV].
This is the reason why we also identified three Seyfert
galaxies.
We checked the radio fluxes of our high redshift candidates with three
main radio catalogues,
namely GB 5GHz catalogue (Gregory {\it et al. } 1996), NVSS 1.4GHz catalogue
(Condon {\it et al. } 1998)
and FIRST 1.4GHz catalogue (White {\it et al. } 1997). According to the
definition of radio-loud AGNs, namely $\rm R>10$
with $R$ being the ratio of 5GHz radio flux and the B-band optical flux
(Kellermann {\it et al. } 1989), we found that eight objects
in our sample are radio-loud (see Table 1).
Among them six objects are weak radio sources and only two new quasars have
strong radio emission. RX J1616.9+3621 with $\rm z=2.259$, was found to
be positionally coincident with
a radio-loud object FIRST J161655.5+362134.
Another new
QSO, RX J1604.6+5714 with $\rm z=0.725$, is coincident with 87GB
160335.3+572233.
One weak radio source (RX J1259.8+3423), which does not match the
definition of radio-loud AGNs,
and other 13 sources with no previous radio detection are most probably
radio-quiet objects.
Table 1 shows that five new QSOs have $\rm L_x (0.1 - 2.4\,keV)$\\
$\rm >10^{46}\,erg\,s^{-1}$,
and all of them are radio-quiet QSOs with $\rm z>1.3$. Another radio-quiet
QSO with $\rm z>1.3$, RX J1541.2+7126, also has $\rm L_x
=8.26 \times 10^{45}\,erg\,s^{-1}$. Therefore, these six new radio-quiet
QSOs are really X-ray luminous objects.
In Figure 1 and 2, we give the finding charts and spectra of these six new
QSOs.
The finding charts
were extracted from the Palomar (POSS-I) Digitized Sky-Survey .
Table 1 lists 18 sources having offsets less than $15''$
between
X-ray and optical positions.
They are all the most nearby optical sources to
the X-ray positions, so that we suggest that they are
the most likely true identifications of the X-ray sources. Four
candidates have X-ray to optical offsets larger than $15''$, which
are not
uncommon for ROSAT detection at the
detection limit. Here we
discuss briefly each of these cases.
For {\it RX J0959.8+0049}, the optical counterpart we selected is $33''$
away from
the X-ray position and is identified as a quasar with $\rm z=2.243$. Three
other optical sources (fainter than $O=19.5$) are found to be
closer to the X-ray
position, but all of them are red objects with $\rm O-E>2$.
Although such colors are compatible with a M-type
dwarf, the
optical
weakness makes this
identification unplausible. Therefore,
our identification
of this X-ray source with a $\rm z=2.243$ QSO is highly likely, though a
future high resolution X-ray
image of this source is required to confirm this identification.
For {\it RX J1208.3+5240}, it is identified with a QSO with
$z=0.345$, which is $29''$ away
from the X-ray position. Another nearby optical source is fainter
than our candidate and is $38''$ far away from the X-ray source. Thus we
regard our identification is most likely correct.
For {\it RX J1428.9+2710}, its X-ray position is $17''$ away from
a QSO
with $z=0.443$.
We note that another optical
source, is only $8''$ away from the
X-ray position. This source, with $O=19.2$ and $O-E=1.1$, is fainter
than our candidate QSO and is also point-like. Again future
high quality X-ray observations are needed to confirm our identification.
Finally for {\it RX J1701.4+3511}, the optical source we selected is $19''$
away
from the X-ray position and was found to be a new QSO
with $z=2.115$. Other four nearby optical sources, with angular distances
from
$25''$ to $39''$ to the X-ray position, are faint and red objects with
$O-E>2$.
Therefore, we believe that our identification is reliable.
\section{Discussion}
\begin{figure}
\vskip -0.2cm
\hspace*{0.4cm}
\psfig{file= 8350f3.ps,width=5.2cm,height=8.5cm,angle=270}
\caption{The slit spectrum of WEE 83. The emission lines of
$H_{\gamma}$, $H_{\beta}$ and [OIII] are evident and the
left wing of $H_{\alpha}$ is clearly shown,}
\end{figure}
\begin{figure*}
\vskip -1.5cm
\hspace*{-0.5cm}
\psfig{file= 8350f4.ps,width=18.5cm,height=10cm,angle=0}
\vskip -1.9cm
\caption{The histogram of redshift distributions of previously known
QSOs detected by the RASS in the sky area with
$\rm b > 35^o$, and that of the newly discovered AGNs in our sample. The
redshift
bin is 0.1. The solid and dotted
lines correspond to the previously known AGNs with 5GHz radio fluxes
larger than 20mJy and those with 5GHz radio
fluxes less than 20mJy, respectively. The shaded areas present
the
newly discovered radio-quiet AGNs in our sample.}
\end{figure*}
Among the new radio-quiet QSOs discovered by us, six have redshifts
larger than 1.3 and three of them have
redshifts larger than 2. These identifications substantially
increase the number of such objects detected in the RASS.
In the sky area of high galactic latitudes ($\rm
b > 35^o$), we have made a correlation between the catalogue
of identifications of RASS sources from the HQS Schmidt plates
and the catalogue of QSOs and AGNs (V\'eron-Cetty \& V\'eron 1996).
This area contains only 18 previously known
QSOs with 5GHz radio fluxes less than 20mJy
and redshifts larger than 1.3, and only four of them have redshifts larger
than 2. These four QSOs are RX J09190+3502, KUV 12461+2710, PG
1247+268 and WEE 83. The QSO identified with RX J09190+3502
was
discovered at Hamburg (Bade {\it et al. }
1995), but a confident identification was not made previously because an AGN
with low redshift was detected closer to the X-ray position. WEE 83 was
selected by Weedman
(1985) on objective prism plates, because it showed a strong emission line,
which was regarded as $\rm Ly{\,\alpha}$. In Figure 3
we show the slit
spectrum of WEE 83 which we took in February 1998. Clearly we can see at
least four
emission lines and the
average redshift of this object is 0.311 rather than 2.04.
Therefore, actually only two X-ray luminous and
radio-quiet QSOs with $\rm z>2$ in the sky
area of $\rm b > 35^o$ were known previously. In Figure 4 we show the
redshift distributions
of the previously known radio-loud and radio-quiet AGNs in the area with
$\rm b > 35^o$, and the redshift
distribution of the new radio-quiet AGNs in our sample. It is clear that
the radio-quiet QSOs are more numerous than
the radio-loud QSOs at lower redshift ($\rm z<0.7$);
most of the QSOs with higher redshift have high radio fluxes, and
radio-quiet QSOs with higher redshifts are relatively rare.
The new QSOs discovered by us represent a substantial contribution to
the number of radio-quiet, X-ray luminous QSOs with $\rm z>1.6$.
A histogram of the ratio between the 1.4 GHz radio flux density
and the 1 keV X-ray flux density for the high redshift QSOs in our survey
area is given in
Figure 5. It includes 22 radio-loud QSOs and 9 radio-quiet QSOs with
$\rm z>1.6$
(an upper limit of 2.5mJy was adopted
for the sources with no radio detection in the NVSS catalogue).
It demonstrates again that the radio-loud objects with high redshift are
more numerous than the radio-quiet objects in the RASS,
however the amount of radio-quiet high redshift QSOs was obviously
underestimated in earlier studies.
This histogram shows clearly that the radio-loud QSOs and radio-quiet
QSOs are well separated and there
is a strong distinction between these two populations
of X-ray loud QSOs; it is clear from Figure 5 that all radio-loud high
redshift QSOs have $f_{1.4GHz}/f_{1keV}$
larger than $10^6$ while all radio-quiet QSOs have
$f_{1.4GHz}/f_{1keV}$ less than $10^6$. They seem to support the suggestion
that radio-loud QSOs and radio-quiet QSOs may have
different physical properties.
This has already been shown and discussed in earlier papers.
For example, the differences of the X-ray color and the X-ray spectra
of radio-quiet QSOs and radio-loud QSOs at high redshift have been found
by Bechtold {\it et al. } (1994), and these differences may imply that the
central engine
and the environment of these two kinds of QSOs are probably different.
It has also been found that the X-ray spectra of the radio loud high
redshift QSOs are significantly
harder compared to the radio quiet population (Schartel {\it et al. } 1996;
Brinkmann,
Yuan \& Siebert 1997).
It is thus possible that an additional spectral component is necessary in
order to explain the observations.
Such a component may be contributed by the beamed radiation
from a highly relativistic plasma
jet
aligned nearly to our direction (Wilkes \& Elvis 1987; Kollgaard 1994).
\begin{figure}
\vskip -0.3cm
\hspace{-0.3cm}
\psfig{file= 8350f5.ps,width=9cm,height=6cm,angle=0}
\vskip -0.5cm
\caption{The histogram of distributions of $Log(f_{1.4GHz}/f_{1keV})$
for QSOs with $\rm z > 1.6$ detected by the RASS in the sky area with $\rm b
> 35^o$. The solid and
dotted lines refer to the radio-loud QSOs and radio-quiet QSOs respectively.
The shaded area shows the contribution of the new QSOs in our sample. }
\end{figure}
For high redshift QSOs, the spectral region of the soft X-ray
excess
found with ROSAT is shifted
partially out of the observable window and
with the spectral resolution of ROSAT it is difficult to
distinguish
the soft excess and the power law component. However, a strong
soft X-ray excess
with an extension to higher energies is probably responsible for
the high
X-ray fluxes of the
radio-quiet high redshift QSOs.
The origin of the soft X-ray excess is extensively
discussed in the literature, but no firm conclusion can be drawn currently
(Piro, Matt \&
Ricci 1997). Many authors have proposed
the inner regions of the accretion disk as emission region (e.g. Arnaud
{\it et al. } 1985; Saxton {\it et al. } 1993; Brunner {\it et al. } 1997).
Spectroscopic ASCA observations have shown that
narrow-line
Seyfert 1's and low redshift radio-quiet QSOs have generally steeper
intrinsic hard
X-ray continua (Brandt,
Mathur \& Elvis 1997; Reeves {\it et al. } 1997).
But whether the X-ray luminous, radio-quiet high redshift QSOs
have similar X-ray spectral properties as narrow line
Seyfert 1's (e.g. Boller, Brandt \& Fink 1996)
and low redshift radio-quiet QSOs is still unclear (see Vignali
{\it et al. } 1998).
We expect that the future broad
band spectroscopic observations
on these
QSOs using
XMM and AXAF will
lead to significant progress in understanding
the X-ray emission mechanism
of them.
\begin{acknowledgements}
We thank Weimin Yuan for reading the manuscript critically and
the
referee, Wolfgang Brinkmann, for helpful
comments. X.-B. Wu acknowledges the
support from the Director Funds of Beijing Astronomical Observatory.
This work is
based on the observations made by the 2.16m telescope at Xinglong
station of Beijing Astronomical Observatory.
\end{acknowledgements}
|
\section{Introduction}
One of the most important invariants in algebraic geometry deals with
the computation of cohomologies of sheaves.
The crucial role in this computation is played by some vanishing criteria for
higher-dimensional cohomologies of certain sheaves.
This vanishing then enables one to reduce the calculation of the whole
cohomology group $H^q({\mathbf P},{\mathcal F})$ to the vector space of global sections
${\Gamma}({\mathbf P},{\mathcal F})=H^0({\mathbf P},{\mathcal F})$. For example, the Bott vanishing theorem \cite{Bott}
and the Serre duality theorem help to compute the dimension of
$H^q({\mathbf P}^n,\Omega_{{\mathbf P}^n}^p(k))$ on a complex projective space
${\mathbf P}^n={\mathbf P}^n({\mathbf C})$, where
$\Omega_{{\mathbf P}^n}^p(k):=\Omega_{{\mathbf P}^n}^p\otimes_{{\mathcal O}_{{\mathbf P}^n}} {\mathcal O}_{{\mathbf P}^n}(k)$.
For simplicity, we denote the dimension of the $q$-th cohomology group of the
coherent sheaf ${\mathcal F}$ on a complete projective variety ${\mathbf P}$ by
$h^q({\mathbf P},{\mathcal F}) := \dim_{{\mathbf C}} H^q({\mathbf P},{\mathcal F})$.
\begin{theo}[Bott formula for ${\mathbf P}^n$ {\rm \cite{Bott, Okonek}}]
Let ${\mathbf P}^n$ be a complex projective space.
\begin{itemize}
\item[{\rm 1)}] If $k = 0$, then
\begin{eqnarray*}
h^q({\mathbf P}^n,\Omega_{{\mathbf P}^n}^p)=
\left\{
\begin{array}{ll}
\displaystyle
1,& p=q,\\
\\
0,&otherwise.
\end{array}
\right.
\end{eqnarray*}
\item[{\rm 2)}] If $k \ne 0$, then
\begin{eqnarray*}
h^q({\mathbf P}^n,\Omega_{{\mathbf P}^n}^p(k))=
\left\{
\begin{array}{ll}
\displaystyle
\binom{n+k-p}{k}\binom{k-1}{p},&q=0, \:k>p,\\
\\
\displaystyle
\binom{-k+p}{-k}\binom{-k-1}{n-p},&q=n,\: k<p-n,\\
\\
0,&otherwise.
\end{array}
\right.
\end{eqnarray*}
\end{itemize}
\label{Original.Bott}
\end{theo}
This theorem was also proved by I-C.~Huang \cite{Huang}, who constructed an
explicit basis for the cohomology groups.
Theorem 2.3.2 in \cite{Dolgachev} gives a
similar result for weighted projective spaces.
The first aim of this paper is to give a generalization of the Bott formula
for a complete projective toric variety.
Let $D$ be a Cartier divisor on a toric variety ${\mathbf P}$, and
$\Omega_{\mathbf P}^p(D)$ be the sheaf of $p$-differential
forms of Zariski on ${\mathbf P}$.
Since we have an analogy of the Bott vanishing
theorem, it is possible to compute the dimension of
$H^q({\mathbf P},\Omega_{\mathbf P}^p(D))$ when $D$ is ample.
The answer will be given in terms of combinatorial sums over all faces
of the {\it support polytope} $\Delta$ for ${\mathcal O}_{{\mathbf P}}(D)$, playing the
role of the number $k$.
The case when ${\mathcal O}_{{\mathbf P}}(D)\simeq{\mathcal O}_{{\mathbf P}}$ corresponding to $k=0$ was
extended to a toric variety in \cite{Danilov-Khovanskii} and
\cite{Oda.book}.
Also, note that for $p=0$, we get the well-known result on the cohomologies
$H^q({\mathbf P},{\mathcal O}_{{\mathbf P}}(D))$ of the ample invertible sheaf ${\mathcal O}_{{\mathbf P}}(D)$,
which can be found in any introduction to the theory of toric varieties:
\[
\dim_{\mathbf C} H^q({\mathbf P},{\mathcal O}_{{\mathbf P}}(D))=
\left\{
\begin{array}{ll}
\# (\Delta\cap M),&q=0,\\
\\
0 ,&q>0,
\end{array}
\right.
\]
where $\# (\Delta\cap M)$ is the number of integer points in the polytope
$\Delta$ in a lattice $M$.
Let $N$ be a free ${\mathbf Z}$-module of rank $n$ and $M := {\rm Hom}_{\mathbf Z}(N,{\mathbf Z})$ its
dual. Denote by ${\Sigma}$ a complete {\it fan} of convex polyhedral cones in
$N_{\mathbf R} := N\otimes {\mathbf R}$. We associate a so-called {\it toric variety}
${\mathbf P} = {\mathbf P}({\Sigma})$ with ${\Sigma}$, i.e. a normal complex variety containing a torus
$T_N := {\rm Hom}_{\mathbf Z}(M,{\mathbf C}^*)$ as a dense open set with an algebraic action
of $T_N$ on ${\mathbf P}$. For precise definitions and basic references about toric
varieties see \cite{Danilov}, \cite{Oda.book} and \cite{Fulton.toric}.
In section \ref{section.cohomologies} we will give a brief
review of another, but equivalent, approach to toric varieties which
has been introduced and studied in \cite{Audin}, \cite{Cox} and
\cite{Batyrev-Cox}. Throughout this paper we assume that ${\mathbf P}$ is complete
and simplicial. Here is the precise definition of the sheaf of $p$-differential
forms of Zariski, which is the main object of our study.
\begin{dfn}
Denote {\it the sheaf of $p$-differential forms of Zariski
on ${\mathbf P}$} by $\Omega_{{\mathbf P}}^p$.
These forms are defined by $\Omega_{{\mathbf P}}^p:=i_{*}\Omega_U^p$, where
$i:\,U\hookrightarrow {\mathbf P}$ is the inclusion of the nonsingular locus
$U$ of ${\mathbf P}$.
Also, denote for any invertible sheaf ${\mathcal O}_{{\mathbf P}}(D)$ on ${\mathbf P}$
\[
\Omega_{{\mathbf P}}^p(D):=\Omega_{{\mathbf P}}^p\otimes {\mathcal O}_{{\mathbf P}}(D).
\]
\end{dfn}
The paper is organized as follows:
{\it In sections 2 and 3}, we prove a generalization of the original
Bott formula stated above.
The generalization will be given in two forms:
in theorem \ref{Bott.formula.I} and in theorem
\ref{Bott.formula.II}. The first version of the Bott formula is proved
using the Ishida-Oda complex, while the second version is
proved using the technique of Danilov and Khovanski\^i.
Both versions are formulated in terms of sums over all faces $\Gamma$ of the
support polytope $\Delta$.
{\it In section 4}, we compare our results with the original Bott formula.
{\it In section 5}, we introduce the notion of the $p$-th Hilbert-Ehrhart
polynomial $L_p(k)$ coinciding with the usual Ehrhart polynomial $L(k)$
when $p=0$. Our idea is to compare the Ehrhart polynomial with a
certain Hilbert polynomial corresponding to the sheaf $\Omega_{{\mathbf P}}^p(D)$.
Then, the Serre duality provides a generalization of the well-known
{\it reciprocity law}.
{\it In section 6}, we compare the two versions of the Bott formula for
toric varieties and obtain non-trivial identities between integer points
in simple integer polytopes. We give a simple combinatorial proof
of the identities and the generalized reciprocity law.
{\it In section 7}, we apply the Bott formula to the computation of
cohomologies of quasi-smooth hypersurfaces on a complete simplicial
toric variety.
{\it In section 8}, we compute the dimension of the space of
global sections of weight components of the sheaf
$\Omega_{{\mathbf P}}^p({\rm log}(-K))\otimes {\mathcal L}$.
{\it In section 9}, we obtain a result similar to the Bott formula
on the projective bundle ${\mathbf P}({\mathcal E})$ associated with the
sheaf ${\mathcal E} = {\mathcal L}_0\oplus \ldots \oplus {\mathcal L}_s$, where ${\mathcal L}_i$ is an ample
invertible sheaf on ${\mathbf P}$. Namely, we compute the dimension of the
global sections of the sheaf $\Omega_{Y/{\mathbf P}}^p(k)$ of relative differential
forms on ${\mathbf P}({\mathcal E})$.
{\it Acknowledgments.} The author would like to thank Prof. A.K.~Tsikh for his
wise advices and very useful discussions during the preparation of this paper.
I am very grateful to Prof. A.P.~Yuzhakov, who introduced me to the theory of
combinatorial sums. I appreciate the hospitality of the Mathematisches
Institut of Universit\"at T\"ubingen, where this paper was finished.
\section{The generalized Bott formula (I)}
Let ${\mathbf P}$ be a complete simplicial $n$-dimensional projective toric variety.
Fix an invertible sheaf ${\mathcal L}={\mathcal O}_{{\mathbf P}}(D)$ corresponding to an ample Cartier
divisor $D$ on ${\mathbf P}$.
It is known from the general theory of toric varieties
(see, e. g., \cite{Oda.book}) that the
cohomology group $H^0({\mathbf P},{\mathcal L})$ admits an eigenspace decomposition
coming from the torus action
\[
H^0 ({\mathbf P},{\mathcal L}) = \bigoplus_{m\in M} H^0 ({\mathbf P},{\mathcal L})_m
\]
and $H^q({\mathbf P},{\mathcal L})$ vanishes for $q>0$.
\begin{dfn} Denote the set of all $k$-dimensional cones of ${\Sigma}$ by
${\Sigma}(k)$ and by $|{\Sigma}(k)|$ the number of $k$-dimensional cones
of ${\Sigma}$, i.e. the cardinality of ${\Sigma}(k)$.
\end{dfn}
\begin{dfn}[\cite{Torus.Embeddings}]
The convex hull $\Delta = \Delta({\mathcal L})$ of all lattice points $m\in M$
in $M_{\mathbf R}:=M\otimes {\mathbf R}$ for which $H^0 (X,{\mathcal L})_m \ne 0$ is called the
{\it support polytope for ${\mathcal L}$}.
Moreover, $\Delta$ is equal to the intersection
\[
\Delta = \bigcap_{{\tau}\in{\Sigma}(n)} (m_{\tau}({\mathcal L})+{\tau}^\vee),
\]
where $m_{\tau}({\mathcal L})$ is the unique lattice point of $M$ such that
\[
\{ m\in M:\, H^0(U_{\tau},{\mathcal L})_m \ne 0 \} =
m_{\tau}({\mathcal L}) + {\tau}^\vee \cap M
\]
for any affine chart $U_{\tau}={\rm Spec}{\mathbf C}[M\cap {\tau}^\vee]$
corresponding to the cone ${\tau}$.
\end{dfn}
\begin{rem} $\Delta$ is a {\it simple} convex polytope of
dimension $n$, i.e. only $n$ edges of $\Delta$ meet at each vertex,
if and only if ${\mathbf P}$ is a simplicial toric variety.
\end{rem}
\begin{dfn}
Each face $\Delta_{\sigma}$ of $\Delta$ corresponding to the cone
${\sigma}\in{\Sigma}$ is defined as
\[
\Delta_{\sigma}:=
\bigcap_{\begin{subarray}{l}
{\tau}\in{\Sigma}(n) \\
{\tau}\succ{\sigma}
\end{subarray}}
(m_{\tau}({\mathcal L}) + {\tau}^\vee \cap {\sigma}^\perp),
\]
where ${\tau}\succ{\sigma}$ means that ${\sigma}$ is a face of ${\tau}$.
\end{dfn}
\begin{dfn} Denote by ${\it l} (\Delta)$ the number of integer points in the
polytope $\Delta$, and let ${\it l} (\Delta_{\sigma})$ be the number of integer points
contained in the face $\Delta_{\sigma}$ of $\Delta$.
\label{Int.Points}
\end{dfn}
\begin{dfn}
We identify the closure $V({\sigma})$ of any torus-invariant orbit
$orb({\sigma})$ in ${\mathbf P}$ corresponding to ${\sigma}\in{\Sigma}$ with
a toric variety with respect to a fan glued from the cones
$({\tau} + (-{\sigma}))/{\mathbf R}{\sigma}$ in $N_{\mathbf R}/{\mathbf R}{\tau}$, ${\tau}\in {\rm Star}_{\sigma}({\Sigma})$,
where ${\mathbf R}{\sigma}:={\sigma}+(-{\sigma})$ is the smallest ${\mathbf R}$-subspace containing ${\sigma}$ of
$N_{\mathbf R}$, while ${\rm Star}_{\sigma}({\Sigma}) := \{{\tau}\in{\Sigma}:\,{\tau}\succ{\sigma}\}$
(see \cite{Oda.book}).
\end{dfn}
\begin{prop}[\cite{Batyrev-Cox}, Proposition 4.10]
If ${\mathcal L}$ is an ample invertible sheaf on ${\mathbf P}$, then one has a one-to-one
correspondence between $(n-k)$-dimensional faces $\Delta_{\sigma}$ of the
polytope $\Delta({\mathcal L})$ and $k$-dimensional cones ${\sigma}\in{\Sigma}$ reversing the
face-relation. Moreover, $\Delta_{\sigma}$ is the support polytope for the
sheaf ${\mathcal O}_{V({\sigma})}\otimes {\mathcal L}$.
\label{support.polytope}
\end{prop}
\begin{dfn}
We denote by ${\mathcal F}_k(\Delta)$ the set of all $k$-dimensional faces of $\Delta$
and by $f_k = |{\mathcal F}_k(\Delta)|$ the number of $k$-dimensional faces of $\Delta$.
\end{dfn}
\begin{rem}
It follows from the proposition above that
${\mathcal F}_j(\Delta) \simeq {\Sigma}(n - j)$ and $f_j = |{\Sigma}(n - j)|$ respectively.
\end{rem}
\begin{dfn}
Define the torus-invariant effective divisor $D({\sigma})$ on each $V({\sigma})$
by
\[
D({\sigma}):=
V({\sigma})\backslash orb({\sigma}).
\]
\end{dfn}
Now recall the construction of the complex which has been introduced and
studied by M.-N.~Ishida and T.~Oda in \cite{Ishida}, \cite{Oda.book} and
\cite{Oda.de.Rham}, known as {\it Ishida's $p$-th complex of
${\mathcal O}_{{\mathbf P}}$-modules}. Ishida's complex plays an important role in the
proof of the first version of the generalized Bott formula.
\begin{prop}[\cite{Oda.book}, Corollary 3.2)]
Let $\Omega_{V({\sigma})}^p({\rm log}\, D({\sigma}))$ be a sheaf of meromorphic $p$-forms with
logarithmic poles along $D({\sigma})$.
There exists an isomorphism of ${\mathcal O}_{V({\sigma})}$-modules
\[
\Omega_{V({\sigma})}^p({\rm log}\, D({\sigma}))=
{\mathcal O}_{V({\sigma})}\otimes_{\mathbf Z}\bigwedge\nolimits^p
(M\cap {\sigma}^\perp) \qquad\mbox{for}\quad 0\le p\le \dim V({\sigma}).
\]
\end{prop}
Now, for each pair of integers $p$ and $q$ we let
\[
{\mathcal K}_{{\mathbf P}}^{p,\,q}:=
\bigoplus_{{\sigma}\in{\Sigma}(q)}
\Omega_{V({\sigma})}^{p-q}({\rm log}\, D({\sigma}))=
\bigoplus_{{\sigma}\in{\Sigma}(q)}
{\mathcal O}_{V({\sigma})}\otimes_{\mathbf Z}\bigwedge\nolimits^{p-q}
(M\cap {\sigma}^\perp) \qquad\mbox{if}\quad 0\le q\le p,
\]
and ${\mathcal K}_{{\mathbf P}}^{p,\,q} = 0$ otherwise.
\begin{dfn}
The coboundary map
$
\delta:
{\mathcal K}_{{\mathbf P}}^{p,\,q}
\rightarrow
{\mathcal K}_{{\mathbf P}}^{p,\,q+1}
$
is defined to be the direct sum $\delta=\oplus_{{\sigma},{\tau}} R_{{\sigma}/{\tau}}$
of the maps
\[
R_{{\sigma}/{\tau}}:\,
\Omega_{V({\sigma})}^{p-q}({\rm log}\, D({\sigma}))
\rightarrow
\Omega_{V({\tau})}^{p-q-1}({\rm log}\, D({\tau})),
\]
i.e.
\[
R_{{\sigma}/{\tau}}:\,
{\mathcal O}_{V({\sigma})}\otimes_{\mathbf Z}\bigwedge\nolimits^{p-q}
(M\cap {\sigma}^\perp)
\rightarrow
{\mathcal O}_{V({\tau})}\otimes_{\mathbf Z}\bigwedge\nolimits^{p-q-1}
(M\cap {\tau}^\perp),
\]
where $R_{{\sigma}/{\tau}}$ is the tensor product of the restriction map
${\mathcal O}_{V({\sigma})}\rightarrow {\mathcal O}_{V({\tau})}$ with the interior product $\delta_{{\sigma}/{\tau}}$
whenever ${\sigma}$ is a face of ${\tau}$, and zero otherwise. The definition of
$\delta_{{\sigma}/{\tau}}$ is the following. The homomorphism $\delta_{{\sigma}/{\tau}}$ is defined
to be zero when ${\sigma}$ is not a face of ${\tau}$. On the other hand, if ${\sigma}$ is
a face of ${\tau}$, then we can determine a primitive element $n\in N$
uniquely modulo $N\cap {\mathbf R}{\sigma}$ so that ${\tau}+(-{\sigma})={\mathbf R}_{\ge 0}n+{\mathbf R}{\sigma}$.
Moreover, $M\cap {\tau}^\perp$ is a ${\mathbf Z}$-submodule of corank one in
$M\cap {\sigma}^\perp$. Each element of
$\bigwedge\nolimits^{p-q}(M\cap {\sigma}^\perp)$ can be written as a finite linear
combination of
\[
m_1\wedge m_2\wedge \cdots\wedge m_{p-q},
\quad m_1\in M\cap {\sigma}^\perp,
\quad m_2,\ldots, m_{p-q}\in M\cap {\tau}^\perp.
\]
Now define
\[
\delta_{{\sigma}/{\tau}}:
\bigwedge\nolimits^{p-q} (M\cap {\sigma}^\perp)\rightarrow
\bigwedge\nolimits^{p-q-1} (M\cap {\tau}^\perp)
\]
by
\[
\delta_{{\sigma}/{\tau}}(m_1\wedge m_2\wedge \cdots\wedge m_{p-q}):=
\<m_1,n\>m_2\wedge \cdots\wedge m_{p-q}.
\]
Actually, $R_{{\sigma}/{\tau}}$ is the {\it Poincar\'e residue map} for the component
$V({\tau})$ of the divisor $D({\sigma})$ on $V({\sigma})$.
\end{dfn}
\begin{theo}[\cite{Oda.book}, Theorem 3.6]
If ${\mathbf P}$ is a simplicial toric variety, then for each $0\le p\le n$
there exists an exact sequence
\begin{equation}
0\rightarrow \Omega_{{\mathbf P}}^p
\rightarrow
{\mathcal K}_{{\mathbf P}}^{p,\,0}
\stackrel{\delta}{\rightarrow}
{\mathcal K}_{{\mathbf P}}^{p,\,1}
\stackrel{\delta}{\rightarrow}
\ldots
\stackrel{\delta}{\rightarrow}
{\mathcal K}_{{\mathbf P}}^{p,\,p}
\rightarrow 0
\label{Ishida's.complex}
\end{equation}
of ${\mathcal O}_{{\mathbf P}}$-modules on ${\mathbf P}$.
\end{theo}
\begin{theo}[The generalized Bott formula (I)]
Let ${\mathbf P}$ be a complete simplicial toric variety,
${\mathcal O}_{{\mathbf P}}(D)$ be an invertible sheaf on ${\mathbf P}$ corresponding to an ample
divisor $D$, and $\Delta$ be the support polytope for ${\mathcal O}_{{\mathbf P}}(D)$. Then
\begin{itemize}
\item[{\rm 1)}]
\[
h^q({\mathbf P},\Omega_{{\mathbf P}}^p) =
\left\{
\begin{array}{ll}
\displaystyle
\sum_{j=0}^p(-1)^{p-j}\binom{n-j}{p-j} f_{n - j}, &q=p,\\
\\
0, &q\ne p.
\end{array}
\right.
\]
\item[{\rm 2)}]
\[
h^q({\mathbf P},\Omega_{{\mathbf P}}^p(D)) =
\left\{
\begin{array}{ll}
\displaystyle
\sum_{j=0}^p(-1)^j\binom{n-j}{p-j}
\sum_{F\in{\mathcal F}_{n - j}(\Delta)}{\it l}(F),&q=0,\\
\\
0, &q>0.
\end{array}
\right.
\]
\end{itemize}
\label{Bott.formula.I}
\end{theo}
\begin{proof} The assertion $1)$ is proved in the Theorem 3.11
of \cite{Oda.book} in the form
\[
h^p({\mathbf P},\Omega_{{\mathbf P}}^p) =
\displaystyle
\sum_{j=0}^p(-1)^{p-j}\binom{n-j}{p-j} |{\Sigma}(j)|,
\]
and $h^q({\mathbf P},\Omega_{{\mathbf P}}^p) = 0$ for $p\ne q$.
Let us prove $2)$. We write ${\mathcal L}$ for ${\mathcal O}_{{\mathbf P}}(D)$,
and $\Omega_{{\mathbf P}}^p\otimes{\mathcal L}$ for $\Omega_{{\mathbf P}}^p(D)$ for convenience.
By Bott vanishing theorem (see \cite{Danilov, Oda.book})
$h^q({\mathbf P},\Omega_{{\mathbf P}}^p\otimes{\mathcal L}) = 0$ for $q > 0$. The sequence
(\ref{Ishida's.complex}) remains exact after shifting by ${\mathcal L}$
\begin{equation}
\label{twisted.complex}
0\rightarrow \Omega_{{\mathbf P}}^p\otimes {\mathcal L}
\rightarrow
{\mathcal K}_{{\mathbf P}}^{p,\,0}\otimes {\mathcal L}
\rightarrow
{\mathcal K}_{{\mathbf P}}^{p,\,1}\otimes {\mathcal L}
\rightarrow
\ldots
\rightarrow
{\mathcal K}_{{\mathbf P}}^{p,\,p}\otimes {\mathcal L}
\rightarrow 0.
\end{equation}
Note, that $M\cap {\sigma}^\perp$ is a free ${\mathbf Z}$-module of rank $n-j$ and
$\bigwedge\nolimits^{p-j}(M\cap {\sigma}^\perp)$ is a free ${\mathbf Z}$-module of
rank $\displaystyle \binom{n-j}{p-j}$ when ${\sigma}\in{\Sigma}(j)$.
This observation shows that
\[
{\mathcal K}_{{\mathbf P}}^{p,\,j}\otimes {\mathcal L}
\simeq
{\mathcal O}_{V({\sigma})} \otimes_{\mathbf Z}\bigwedge\nolimits^{p-j}
(M\cap {\sigma}^\perp)\otimes {\mathcal L}
\simeq
\displaystyle
{\mathcal O}_{V({\sigma})}\otimes {\mathcal L}^{\oplus\binom{n-j}{p-j}}.
\]
Assume for the moment that $h^q({\mathbf P},{\mathcal K}_{{\mathbf P}}^{p,\,j}\otimes {\mathcal L}) = 0$ for
$q > 0$. Then (\ref{twisted.complex}) gives the exact sequence of
global sections
\begin{eqnarray*}
0\rightarrow H^0({\mathbf P},\Omega_{{\mathbf P}}^p\otimes {\mathcal L})
\rightarrow H^0({\mathbf P},{\mathcal K}_{{\mathbf P}}^{p,\,0}\otimes {\mathcal L})
\rightarrow H^0({\mathbf P},{\mathcal K}_{{\mathbf P}}^{p,\,1}\otimes {\mathcal L})
\rightarrow
\ldots\\
\ldots
\rightarrow H^0({\mathbf P},{\mathcal K}_{{\mathbf P}}^{p,\,p}\otimes {\mathcal L})
\rightarrow 0.
\end{eqnarray*}
Since $h^0({\mathbf P},{\mathcal O}_{V({\sigma})}\otimes {\mathcal L})=l(\Delta_{{\sigma}})$
(see Proposition \ref{support.polytope}), we have
\begin{eqnarray*}
h^0({\mathbf P},\Omega_{{\mathbf P}}^p\otimes {\mathcal L})
&=&
\sum_{j=0}^p (-1)^j h^0({\mathbf P},{\mathcal K}_{{\mathbf P}}^{p,\,j}\otimes {\mathcal L})\\
&=&
\sum_{j=0}^p (-1)^j \binom{n-j}{p-j}
\sum_{{\sigma}\in{\Sigma}(j)}
h^0({\mathbf P},{\mathcal O}_{V({\sigma})}\otimes {\mathcal L})\\
&=&
\sum_{j=0}^p(-1)^j\binom{n-j}{p-j}
\sum_{{\sigma}\in{\Sigma}(j)}{\it l}(\Delta_{\sigma})\\
&=&
\sum_{j=0}^p(-1)^j\binom{n-j}{p-j}
\sum_{F\in{\mathcal F}_{n - j}(\Delta)}{\it l}(F).
\end{eqnarray*}
It remains to show that $h^q({\mathbf P},{\mathcal K}_{{\mathbf P}}^{p,\,j}\otimes {\mathcal L}) = 0$ for
$q > 0$. Moreover, it is sufficient to prove the vanishing of
$h^q({\mathbf P},{\mathcal O}_{V({\sigma})}\otimes {\mathcal L}) = h^q({\mathbf P},{\mathcal L}|_{V({\sigma})})$.
As observed in \cite{Danilov}, Lemma 6.8.1, the restriction homomorphism
\[
\Gamma({\mathbf P},{\mathcal L})\rightarrow \Gamma({\mathbf P},{\mathcal L}|_{V({\sigma})})
\]
is surjective. Hence, the sheaf ${\mathcal L}|_{V({\sigma})}$ is generated by its
global sections since ${\mathcal L}$ is ample, and all cohomologies
$h^q({\mathbf P},{\mathcal L}|_{V({\sigma})})$ for $q > 0$ vanish by the Bott vanishing theorem.
\end{proof}
\begin{coro} One has the following formulas
\[
\begin{array}{lcl}
&&\displaystyle
\sum_{p=0}^n h^p({\mathbf P},\Omega_{{\mathbf P}}^p)\, y^p =
\displaystyle
\sum_{j=0}^n f_j (1 - y)^j y^{n - j}=
\displaystyle
\sum_{j=0}^n f_j (y - 1)^j,\\
&&\displaystyle
\sum_{p=0}^n h^0({\mathbf P},\Omega_{{\mathbf P}}^p(D))\, y^p =
\displaystyle
\sum_{j=0}^n \Delta^{(j)}
(- y)^j (y + 1)^{n - j},
\quad \mbox{\it if}\; D\; \mbox{\it is ample},
\end{array}
\]
where $\displaystyle \Delta^{(j)} := \sum_{F\in{\mathcal F}_{n - j}(\Delta)}{\it l}(F)$.
\label{Bott.Corollary.I}
\end{coro}
Notice that the first line of the equalities is contained in
Theorem~3.11 of \cite{Oda.book}, where the second equality follows from the
Serre duality $h^p({\mathbf P},\Omega_{{\mathbf P}}^p)=h^{n-p}({\mathbf P},\Omega_{{\mathbf P}}^{n-p})$.
\section{The generalized Bott formula (II)}
Another approach gives the second version of the generalization of the
Bott formula. Here we present two independent proves of this formula.
Suppose that ${\mathbf P}$ is a complete simplicial $n$-dimensional toric variety
as above.
Assume that ${\mathcal O}_{{\mathbf P}} (D)$ is an ample invertible sheaf on ${\mathbf P}$ determining
the convex polytope $\Delta$.
Because of the one-to-one correspondence between the faces
$\Gamma = \Delta_{\sigma}$ of the polytope $\Delta$ and cones ${\sigma}$ of the fan ${\Sigma}$,
we can associate a "small" toric variety, i.e. the closure of torus-invariant
orbit of ${\mathbf P}$, with any face $\Gamma\subseteq\Delta$.
\begin{dfn}
The toric $k$-dimensional subvariety ${\mathbf P}_{\Gamma}$ of
${\mathbf P}$ corresponding to the $k$-dimen\-sional face $\Gamma\subseteq\Delta$
is the closure of a $n - k$-dimensional orbit $orb(\Gamma)$ in ${\mathbf P}$.
For $\Gamma = \Delta$ we let ${\mathbf P}_{\Delta} = {\mathbf P}$.
Define the effective divisor $D_{\Gamma}$ on each ${\mathbf P}_{\Gamma}$ by
\[
D_{{\Gamma}} := {\mathbf P}_{\Gamma}\backslash orb({\Gamma}).
\]
\end{dfn}
\begin{dfn}
\cite{Danilov-Khovanskii, Danilov2}. Denote by
$\displaystyle\Omega_{({\mathbf P}_{\Gamma},\,D_{\Gamma})}^p$ the sheaf of regular
differential $p$-forms on the toric variety ${\mathbf P}_{\Gamma}$ with zeros along
$D_{\Gamma}$, arising from the exact sequence
\begin{equation}
0\rightarrow
\Omega_{({\mathbf P}_{\Gamma},\,D_{\Gamma})}^p
\rightarrow
\Omega_{{\mathbf P}_{\Gamma}}^p
\stackrel{R_\Gamma}{\rightarrow}
\bigoplus_{\Theta}\Omega_{{\mathbf P}_{\,\Theta}}^p
\rightarrow 0,
\label{restr.hom}
\end{equation}
where $\Theta$ runs over all facets (faces of codimension one) of
$\Gamma$ and $R_{\Gamma}$ is the restriction homomorphism.
\end{dfn}
\begin{dfn}
The {\it Euler-Poincar\'e characteristic}
of a coherent ${\mathcal O}_{{\mathbf P}}$-module ${\mathcal F}$ on a $n$-dimensional complete
variety ${\mathbf P}$ over ${\mathbf C}$ is defined to be
\[
\chi({\mathbf P},{\mathcal F}) :=
\sum_{k = 0}^n (-1)^k \dim_{{\mathbf C}} H^k ({\mathbf P},{\mathcal F})=
\sum_{k = 0}^n (-1)^k h^k ({\mathbf P},{\mathcal F}).
\]
\end{dfn}
\begin{dfn}
Denote by ${\it l}^{*}(\Delta)$ the number of integer points in
the relative interior of the polytope $\Delta$.
Also, let ${\it l}^{*}(\Gamma)$ be the number of integer points
contained in the relative interior of a face $\Gamma$ of $\Delta$.
\end{dfn}
Notice the following key property of the sheaf
$\Omega_{({\mathbf P}_{\Gamma},\,D_{\Gamma})}^p$.
\begin{prop} Let ${\mathbf P}$ be a complete simplicial toric variety and
${\mathbf P}_{\Gamma}$ be a closed subset of ${\mathbf P}$ associated with the face
$\Gamma$ of the support polytope $\Delta$ corresponding to an
invertible sheaf ${\mathcal O}_{{\mathbf P}}(D)$ on ${\mathbf P}$.
Then
\begin{eqnarray*}
\chi({\mathbf P}_{\Gamma},\,\Omega_{({\mathbf P}_{\Gamma},\,D_{\Gamma})}^p
\otimes {\mathcal O}_{{\mathbf P}}(D)) =
\left\{
\begin{array}{ll}
\displaystyle
(-1)^{\dim\Gamma}\binom{\dim\Gamma}{p}, &{\mathcal O}_{{\mathbf P}}(D)\simeq {\mathcal O}_{{\mathbf P}},\\
\\
\displaystyle
\binom{\dim\Gamma}{p} {\it l}^{*}({\Gamma}), &{\mathcal O}_{{\mathbf P}}(D)\not\simeq {\mathcal O}_{{\mathbf P}},
\; D \; \mbox{\it is ample}.\\
\end{array}
\right.
\end{eqnarray*}
\label{Remark.in.DK}
\end{prop}
\begin{proof} The statement follows immediately from the Proposition
2.10 of \cite{Danilov-Khovanskii} by taking the restriction of the sheaf
${\mathcal L}$ to ${\mathbf P}_{\Gamma}$.
\end{proof}
Here is the main result of this section.
\begin{theo}[The generalized Bott formula (II)]
Let ${\mathbf P}$ be a complete simplicial toric variety,
${\mathcal O}_{{\mathbf P}}(D)$ be an invertible sheaf on ${\mathbf P}$ corresponding to an ample
divisor $D$, and $\Delta$ be the support polytope for ${\mathcal O}_{{\mathbf P}}(D)$. Then
\begin{itemize}
\item[{\rm 1)}]
\[
h^q({\mathbf P},\Omega_{{\mathbf P}}^p) =
\left\{
\begin{array}{ll}
\displaystyle
\sum_{s = p}^n(-1)^{s + p}\binom{s}{p} f_s, &q = p,\\
\\
0, &q\ne p.
\end{array}
\right.
\]
\item[{\rm 2)}]
\[
h^q({\mathbf P},\Omega_{{\mathbf P}}^p(D)) =
\left\{
\begin{array}{ll}
\displaystyle
\sum_{s = p}^n\binom{s}{p}
\sum_{\Gamma\in{\mathcal F}_s(\Delta)} {\it l}^*(\Gamma), &q = 0,\\
\\
0, &q > 0.
\end{array}
\right.
\]
\end{itemize}
\label{Bott.formula.II}
\end{theo}
\begin{proof}
The formula
\[
h^p({\mathbf P},\Omega_{{\mathbf P}}^p) =
(-1)^p\sum_{\Gamma\subseteq \Delta}(-1)^{\dim\Gamma}\binom{\dim\Gamma}{p} =
\sum_{s = p}^n(-1)^{s + p}\binom{s}{p} f_s
\]
is the assertion of Corollary 2.5 of \cite{Danilov-Khovanskii}.
We prove this statement independently.
Making use of the exact sequence (\ref{restr.hom}), we get
\[
\chi({\mathbf P},\Omega_{\mathbf P}^p) =
\sum_{\Gamma\subseteq\Delta}
\chi({\mathbf P}_{\Gamma},\Omega_{({\mathbf P}_{\Gamma},\,D_{\Gamma})}^p).
\]
Now the requested formula follows from Proposition \ref{Remark.in.DK}
and the vanishing of $h^q({\mathbf P},\Omega_{{\mathbf P}}^p)$ for $p \ne q$.
Let us prove assertion $2)$.
The short exact sequence (\ref{restr.hom}) remains exact after
tensoring by ${\mathcal L} = {\mathcal O}_{{\mathbf P}}(D)$.
As in the case above, we have
\[
\chi({\mathbf P},\Omega_{\mathbf P}^p\otimes{\mathcal L}) =
\sum_{\Gamma\subseteq\Delta}
\chi({\mathbf P}_{\Gamma},\Omega_{({\mathbf P}_{\Gamma},\,D_{\Gamma})}^p\otimes{\mathcal L}),
\]
and from Proposition \ref{Remark.in.DK} and the Bott vanishing
theorem,
\[
\chi({\mathbf P},\Omega_{\mathbf P}^p\otimes{\mathcal L}) =
h^0({\mathbf P},\Omega_{\mathbf P}^p\otimes{\mathcal L}) =
\sum_{\Gamma\subseteq\Delta}
\binom{\dim\Gamma}{p}{\it l}^*(\Gamma) =
\sum_{s = p}^n\binom{s}{p}
\sum_{\Gamma\in{\mathcal F}_s(\Delta)} {\it l}^*(\Gamma).
\]
We prove the second assertion of the theorem by using the description
of the space of global sections of the sheaf $\Omega_{{\mathbf P}}^p\otimes {\mathcal L}$.
We have a decomposition into a direct sum of $M$-homogeneous components
\[
H^0({\mathbf P},\Omega_{{\mathbf P}}^p\otimes {\mathcal L}) = \bigoplus_{m\in \Delta\cap M}
\bigwedge\nolimits^p V_\Delta (m),
\]
where $V_\Delta(m)$ is the ${\mathbf C}$-subspace in $M_{\mathbf C}:=M\otimes{\mathbf C}$ generated by the
smallest face of $\Delta = \Delta({\mathcal L})$ containing $m$.
Since the support polytope $\Delta$ admits the natural partition
$\Delta = \coprod_{\Gamma} Int(\Gamma)$, where $Int(\Gamma)$ is
the relative interior of the face $\Gamma$, we have
\[
h^0({\mathbf P},\Omega_{{\mathbf P}}^p\otimes {\mathcal L})=
\sum_{\Gamma\subseteq \Delta}
\sum_{m\in Int(\Gamma)\cap M}
\binom{\dim V_{\Delta}(m)}{p}=
\sum_{\Gamma\subseteq \Delta}
\binom{\dim\Gamma}{p} {\it l}^*(\Gamma).
\]
Thus we have given a different proof for the second formula.
\end{proof}
\begin{coro} One has the following equality
\begin{eqnarray*}
\displaystyle
\sum_{p = 0}^n h^0({\mathbf P},\Omega_{{\mathbf P}}^p(D)) y^p =
\displaystyle
\sum_{s = 0}^n \Delta_{(s)}
(y + 1)^k,\quad \mbox{\it if} \; D \; \mbox{\it is ample},
\end{eqnarray*}
where $\Delta_{(s)} := \displaystyle\sum_{\Gamma\in{\mathcal F}_s(\Delta)}{\it l}^*(\Gamma)$.
\label{Bott.Corollary.II}
\end{coro}
\section{Relation with the original Bott formula and some \\
combinatorial identities}
In this section we compare our results of theorems \ref{Bott.formula.I}
and \ref{Bott.formula.II} with the original Bott formula of
Theorem \ref{Original.Bott}.
Recall that ${\mathcal O}_{{\mathbf P}^n}(k)$ is ample if and only if $k>0$, and the support
polytope for ${\mathcal O}_{{\mathbf P}^n}(k)$ is the simplex
\begin{eqnarray*}
\Delta = \Delta(k) =
\{m\in M_{\mathbf R}:\, m_1\ge 0,\ldots,m_n\ge 0,\quad m_1+\ldots+m_n\le k\}.
\end{eqnarray*}
It is easy to see that
\[
f_{n - j} = \binom{n+1}{j},\quad
\Delta^{(j)} = \binom{n+k-j}{k}\binom{n+1}{j},
\]
and
\[
f_s = \binom{n+1}{n-s},\quad
\Delta_{(s)} =
\binom{k-1}{s}\binom{n+1}{n-s},
\]
for all $0\le j\le n$ and $0\le s\le n$.
Comparison of the formulas $1)$ of Theorems \ref{Bott.formula.I} and
\ref{Original.Bott} corresponding to the case $q = p$ gives the identity
$$
\sum_{j=0}^p(-1)^{p-j}\binom{n-j}{p-j}\binom{n+1}{j} = 1.
\eqno{(a')}
$$
Now we compare the formulas $2)$ in these theorems corresponding to the
case $q = 0$:
$$
\sum_{j=0}^p(-1)^j\binom{n-j}{p-j}\binom{n+k-j}{k}\binom{n+1}{j}=
\binom{n+k-p}{k}\binom{k-1}{p}.
\eqno{(b')}
$$
Analoguous comparisons of Theorems \ref{Bott.formula.II} and
\ref{Original.Bott} show that
$$
\sum_{s=p}^n(-1)^{p+s}\binom{s}{p}\binom{n+1}{n-s}=1,
\eqno{(a'')}
$$
and
$$
\sum_{s=p}^n\binom{s}{p}\binom{k-1}{s}\binom{n+1}{n-s}=
\binom{n+k-p}{k}\binom{k-1}{p}.
\eqno{(b'')}
$$
These identities can be proved directly.
Let us verify the simplest identity $(a')$.
To prove $(a')$ it is sufficient to check the functional equation
\[
\sum_{j=0}^n\binom{n+1}{j}t^j(1-t)^{n-j}=\sum_{k=0}^n t^k,
\]
and then to compare the coefficients by the monomial $t^p$.
This equation follows from Corollary \ref{Bott.Corollary.I},
but we can deduce it from the obvious identity
\[
(t+1-t)^{n+1}=1,
\]
or
\[
(1-t)\sum_{j=0}^n\binom{n+1}{j}t^j(1-t)^{n-j}+t^{n+1}=1.
\]
Now we get the required by $\displaystyle (1 - t^{n+1})(1 - t)^{-1}=\sum_{k = 0}^n t^k$.
The identity $(a'')$ is equivalent to $(a')$.
The remaining identities $(b')$ and $(b'')$ have been proved in
\cite{Materov-Yuzhakov}.
Our proof is based on the method of integral representations for
combinatorial sums in the spirit of the work \cite{Egorychev-Yuzhakov}.
\section{The Hilbert-Ehrhart polynomials and
the Serre duality}
Let $\Delta\subset{\mathbf R}^n$ be a non-empty integer polytope.
For a positive integer $k$, let $k\cdot\Delta := \{kx:\,x\in\Delta\}$ denote
the dilated polytope. It was proved by Ehrhart \cite{Ehrhart}, and
in somewhat stronger form by Macdonald \cite{Macdonald},
that there is a polynomial $L(k)$ with the properties
\begin{itemize}
\item[(i)] for any integer $k > 0$, one has
\[
L(k) = {\it l}(k\cdot\Delta);
\]
\item[(ii)] for any integer $k > 0$, one has the following
{\it reciprocity law}
\[
L(-k) = {\it l}^*(k\cdot\Delta).
\]
\end{itemize}
The polynomial $L(k)$ is called the {\it Ehrhart polynomial} for $\Delta$.
In this section we study a generalization of the Ehrhart polynomial
using the techniques of algebraic geometry (cf., \cite{Danilov, Oda.book}).
The basic idea is to compare the Ehrhart polynomial with a certain
Hilbert polynomial.
For example, the Serre duality provides a generalization of the
reciprocity law.
\begin{prop}[\cite{Snapper, Kleiman}]
Let ${\mathcal L}$ be an invertible sheaf
on a complete variety ${\mathbf P}$ and ${\mathcal F}$ be a coherent sheaf of ${\mathcal O}_{{\mathbf P}}$-modules
on ${\mathbf P}$. Then the Euler-Poincar\'e characteristic
$\chi({\mathbf P},{\mathcal F}\otimes {\mathcal L}^{\otimes k})$ is a polynomial in $k$
of total degree $\le n$ which assumes integer values whenever $k$ is
integer.
\end{prop}
\begin{dfn}
Let ${\mathcal L}={\mathcal O}_{{\mathbf P}}(D)$ be an invertible sheaf corresponding to
the ample Cartier divisor $D$ of a complete simplicial toric variety ${\mathbf P}$.
We call the polynomial
\[
L_p(k) := \chi({\mathbf P},\Omega_{{\mathbf P}}^p\otimes {\mathcal L}^{\otimes k})=
\chi({\mathbf P},\Omega_{{\mathbf P}}^p(k D))
\]
the {\it Hilbert polynomial} of the sheaf $\Omega_{{\mathbf P}}^p$ with respect
to ${\mathcal L}$, or {\it $p$-th Hilbert polynomial}.
\end{dfn}
It follows from the Bott formula and Bott vanishing theorem
for toric varieties, that
\begin{eqnarray*}
L_p(k) = h^0({\mathbf P},\Omega_{{\mathbf P}}^p(k D))
&=&
\sum_{j = 0}^p (-1)^j \binom{n-j}{p-j}
\sum_{F \in {\mathcal F}_{n - j}(\Delta)} {\it l}(k\cdot F) \\
&=&
\sum_{s = p}^n \binom{s}{p}
\sum_{\Gamma \in {\mathcal F}_{s}(\Delta)} {\it l}^*(k\cdot \Gamma)
\end{eqnarray*}
for $k > 0$. For example, for $p = 0$ the polynomial $L_0(k)$ coincides
with the usual Ehrhart polynomial of the polytope $\Delta = \Delta({\mathcal L})$,
and with $L_n(k) = {\it l}^*(k\cdot \Delta)$ whenever $k > 0$.
Hence, we can consider $L_p(k)$ as a generalization of the Ehrhart
polynomial and also call it the {\it $p$-th Ehrhart polynomial} for
$\Delta$, or simply the {\it $p$-th Hilber-Ehrhart polynomial}.
Note that the reciprocity law can be written as
\[
L_0(-k) = (-1)^n L_n(k).
\]
Here we prove a more general result.
\begin{theo} The polynomial $L_p(k)$ satisfies the duality property
\[
L_p(-k)=(-1)^n L_{n-p}(k)
\]
for any positive integer $k$ and $0 \le p \le n$.
\label{gen.rec.law}
\end{theo}
\begin{proof}
We need a special form of the Serre-Grothendieck duality
(see \S 3.3 of \cite{Oda.book})
\[
H^q({\mathbf P},\Omega_{{\mathbf P}}^p\otimes{\mathcal F})^* \simeq
H^{n-q}({\mathbf P},\Omega_{{\mathbf P}}^{n - p}\otimes{\mathcal F}^*),
\quad 0 \le q \le n
\]
for any locally free sheaf of the ${\mathcal O}_{{\mathbf P}}$-modules ${\mathcal F}$ with the dual
${\mathcal F}^*$ on a complete simplicial toric variety ${\mathbf P}$.
Take ${\mathcal F} = {\mathcal O}_{{\mathbf P}}(kD)$, $k > 0$.
From the Serre duality we have the isomorphism
\[
H^0({\mathbf P},\Omega_{{\mathbf P}}^p(k D))^* \simeq
H^n({\mathbf P},\Omega_{{\mathbf P}}^{n-p}(- k D)).
\]
Hence, using the vanishing of $h^q({\mathbf P},\Omega_{{\mathbf P}}^p(k D))$ for $q > 0$,
we get the equality
\[
\chi({\mathbf P},\Omega_{{\mathbf P}}^p(k D)) =
(-1)^n\chi({\mathbf P},\Omega_{{\mathbf P}}^{n-p}(-k D)),
\]
which is equivalent to
\[
\chi({\mathbf P},\Omega_{{\mathbf P}}^p(-k D)) =
(-1)^n\chi({\mathbf P},\Omega_{{\mathbf P}}^{n-p}(k D)),
\]
and this completes the proof.
\end{proof}
Now let ${\mathbf P}$ be nonsingular.
Recall, that any variety ${\mathbf P}$ has a Todd homology class $td_{{\mathbf P}}$ of ${\mathbf P}$
in $A_*({\mathbf P})_{{\mathbf Q}}$, see \cite{Fulton}.
Since ${\mathbf P}$ is nonsingular, $td_{{\mathbf P}}=Td_{{\mathbf P}}\cap [{\mathbf P}]$,
where $Td_{{\mathbf P}}$ is the Todd cohomogy class in $A^*({\mathbf P})_{{\mathbf Q}}$
and $[{\mathbf P}]$ is the fundamental class of ${\mathbf P}$.
We state without proof some facts on intersection theory on toric
varieties, where part ${\rm (a)}$ slightly generalizes the results in
\cite{Danilov, Fulton, Oda.book}, and ${\rm (b)}$ obviously follows from the
Hirzebruch-Riemann-Roch theorem \cite{Hirzebruch}.
\begin{prop} Let $L_p(k)$ be the Hilbert-Ehrhart polynomial.
{\rm (a)} The coefficients of the polynomial
$L_p(k) = \sum_{i = 0}^n a_{pi} k^i$
are the intersection numbers
\[
a_{pi} =
\displaystyle\frac{1}{i!}\int_{{\mathbf P}}\,D^i\cdot ch(\Omega_{{\mathbf P}}^p)\cdot Td_{{\mathbf P}},
\]
where $\int_{{\mathbf P}}$ denotes the degree homomorphism
$\int_{{\mathbf P}}:A^0({\mathbf P})_{\mathbf Q}\,\rightarrow {\mathbf Q}$ and $ch(\Omega_{{\mathbf P}}^p)$ is the Chern
character of the sheaf $\Omega_{{\mathbf P}}^p$. For example, the leading coefficient is
\[
\displaystyle
a_{pn} = \binom{n}{p}Vol(\Delta),
\]
where $Vol(\Delta)$ is the normalized volume of the polytope
$\Delta\subset M_{{\mathbf R}}$ such that the volume of the unit cube determined by
the basis of the lattice $M$ is $1$.
{\rm (b)} The generating function for the polynomial $L_p(k)$ is equal to
\[
\displaystyle
L(y;k) := \sum_{p = 0}^n L_p(k) y^p =
\int_{{\mathbf P}} e^{kD(y + 1)}\prod_{j = 1}^d Q(y;D_j),
\]
where
\[
Q(y;x) = \frac{x(y + 1)}{e^{x(y + 1)} - 1} + x,
\]
and $D_j$ are the torus-invariant divisors corresponding to the
one-dimensional generators of the fan ${\Sigma}$.
\end{prop}
\section{Combinatorics of simple polytopes}
Comparison of two versions of the Bott formula gives some
elegant corollaries in combinatorics of simple polytopes.
If we compare the first parts of Theorems \ref{Bott.formula.I} and
\ref{Bott.formula.II}, we get the well-known Dehn-Sommerville equations,
conjectured by Dehn \cite{Dehn} and proved by Sommerville \cite{ Sommerville}.
(See also \cite{Broendsted}.) In algebro-geometrical context these equalities
were proved in \cite{Stanley} for any simple rational polytope, using the
Poincar\'e duality on toric varities, and by Oda \cite{Oda.book}, using
the Serre-Grothendieck duality theorem.
\begin{theo} For any simple lattice $n$-polytope $\Delta$
and $p = 0,1,\ldots,n$ one has the identities
\[
\sum_{j = 0}^p (-1)^j \binom{n-j}{p-j} f_{n-j}
=
\sum_{s = p}^n (-1)^{s}\binom{s}{p} f_s.
\]
\end{theo}
Now compare the second parts of Theorems \ref{Bott.formula.I} and
\ref{Bott.formula.II}. We get non-trivial relations between the integer
points in faces of the simple integer polytope $\Delta$, which we prove
here in a purely combinatorial way.
\begin{theo} For any simple lattice $n$-polytope $\Delta$
and $p = 0,1,\ldots,n$ one has the identities
\label{comb_dual}
\[
\sum_{j = 0}^p (-1)^j \binom{n-j}{p-j}
\sum_{F \in {\mathcal F}_{n - j}(\Delta)} {\it l}(F) =
\sum_{s = p}^n \binom{s}{p}
\sum_{\Gamma \in {\mathcal F}_{s}(\Delta)} {\it l}^*(\Gamma).
\]
\end{theo}
\begin{proof}
We have the chain of equalities
\begin{eqnarray*}
\sum_{s = p}^n \binom{s}{p}
\sum_{\Gamma \in {\mathcal F}_{s}(\Delta)} {\it l}^*(\Gamma)
&=&
\sum_{\Gamma \subseteq \Delta} \binom{\dim\Gamma}{p} {\it l}^*(\Gamma)
\\
&=&
\sum_{\Gamma \subseteq \Delta} \binom{\dim\Gamma}{p}
\sum_{F \subseteq \Gamma} (-1)^{\dim\Gamma - \dim F}{\it l}(F)
\\
&=&
\sum_{F \subseteq \Delta} (-1)^{\dim F} {\it l}(F)
\sum_{\Gamma \supseteq F} (-1)^{dim \Gamma} \binom{\dim\Gamma}{p}
\\
&=&
\sum_{F \subseteq \Delta} (-1)^{n + \dim F} \binom{\dim F}{n - p} {\it l}(F)
\\
&=&
\sum_{j = n - p}^n (-1)^{n + j} \binom{j}{n-p}
\sum_{F \in {\mathcal F}_j(\Delta)} {\it l}(F)
\\
&=&
\sum_{j = 0}^p (-1)^j \binom{n-j}{p-j}
\sum_{F \in {\mathcal F}_{n - j}(\Delta)} {\it l}(F),
\end{eqnarray*}
where we have used the identity
\[
\sum_{\Gamma \supseteq F} (-1)^{dim \Gamma} \binom{\dim\Gamma}{p}
=
(-1)^{n} \binom{\dim F}{n - p},
\]
which is proved in the Appendix.
\end{proof}
\begin{rem}
There are two obvious cases of the identities of Theorem \ref{comb_dual}.
For $p = 0$, we have
\[
{\it l}(\Delta) = \sum_{\Gamma\subseteq\Delta} {\it l}^*(\Gamma).
\]
For $p = n$, we get the inclusion-exclusion formula
\[
{\it l}^*(\Delta) = \sum_{j = 0}^n (-1)^{n - j} \sum_{F\in{\mathcal F}_{j}(\Delta)}{\it l}(F).
\]
\end{rem}
We are in a position to give a purely combinatorial proof of the generalized
reciprocity law of Theorem \ref{gen.rec.law}.
\begin{coro}
For any simple lattice $n$-polytope $\Delta$ and any positive integer
$k$, one has
\[
L_p(-k) = (-1)^n L_{n-p}(k),
\]
for any $p = 0,\ldots,n$.
\end{coro}
\begin{proof} Using the equality of Theorem \ref{comb_dual}, we get
\begin{eqnarray*}
L_p(-k)
&=&
\sum_{j = 0}^p (-1)^j \binom{n-j}{p-j}
\sum_{F \in {\mathcal F}_{n - j}(\Delta)} (-1)^{n - j} {\it l}^*(k\cdot F) \\
&=&
(-1)^n
\sum_{s = p}^n \binom{s}{n - p}
\sum_{F \in {\mathcal F}_{s}(\Delta)} {\it l}^*(k\cdot F) \\
&=&
(-1)^n L_{n-p}(k),
\end{eqnarray*}
as required.
\end{proof}
\begin{exam} Let $\Delta$ be a convex polytope in ${\mathbf R}^3$ with vertices at
$(0,\,0,\,0)$, $(1,\,0,\,0)$, $(0,\,1,\,0)$ and $(1,\,1,\,m)$, where $m$
is a positive integer. Then
\begin{eqnarray*}
\begin{array}{l}
\displaystyle
L_0(k)=\frac{m}{6}k^3 + k^2 + \frac{12-m}{6}k + 1,\\
\\
\displaystyle
L_1(k)=\frac{m}{2}k^3 + k^2 - \frac{m}{2}k - 1,\\
\\
\displaystyle
L_2(k)=\frac{m}{2}k^3 - k^2 - \frac{m}{2}k + 1,\\
\\
\displaystyle
L_3(k)=\frac{m}{6}k^3 - k^2 + \frac{12-m}{6}k - 1.
\end{array}
\end{eqnarray*}
\end{exam}
\section{Weighted components of sheaves with
logarithmic poles}
Recall that if ${\mathcal L}$ is an ample invertible sheaf on a complete simplicial
toric variety ${\mathbf P}$, then
\[
H^q({\mathbf P},\Omega_{\mathbf P}^p\otimes {\mathcal L}) = 0
\]
for all $q>0$ and $p\ge 0$ (see \cite{Danilov}).
This statement is a generalization of the theorems of Bott and Steenbrink
for projective spaces and weighted projective spaces respectively.
Batyrev and Cox proved a more general vanishing theorem.
\begin{dfn}[\cite{Batyrev-Cox}]
Denote by $\Omega_{\mathbf P}^p({\rm log} (-K))$ the sheaves of differential $p$-forms
with logarithmic poles along the anticanonical divisor $-K$ on ${\mathbf P}$.
Let $\mathcal{W}$ be the {\it weight filtration}
\[
\mathcal{W}:\, 0\subset
W_0\Omega_{{\mathbf P}}^p({\rm log} (-K))\subset
W_1\Omega_{{\mathbf P}}^p({\rm log} (-K))\subset\cdots\subset
W_p\Omega_{{\mathbf P}}^p({\rm log} (-K))=\Omega_{{\mathbf P}}^p({\rm log} (-K))
\]
on $\Omega_{{\mathbf P}}^p({\rm log} (-K))$ defined by
$W_k\Omega_{{\mathbf P}}^p({\rm log} (-K)):=
\Omega_{{\mathbf P}}^{p-k}\wedge\Omega_{{\mathbf P}}^k({\rm log} (-K))$.
\end{dfn}
\begin{theo}[\cite{Batyrev-Cox}, Theorem 7.2]
Let ${\mathcal L}$ be an ample invertible sheaf on a complete simplicial
toric variety ${\mathbf P}$. Then for any $p\ge 0$, $k\ge 0$, and $q>0$, one has
\[
H^q({\mathbf P},W_k\Omega_{{\mathbf P}}^p({\rm log} (-K))\otimes {\mathcal L})=0.
\]\label{B.S.D}
\end{theo}
Here we give some additional information about the global sections of
$W_k\Omega_{{\mathbf P}}^p({\rm log} (-K))\otimes {\mathcal L}$. We need the following
result of V.I.~Danilov.
\begin{theo}[\cite{Danilov}, \S 15.7]
For any integer $0\le k\le p$ one has the short exact sequence
\begin{equation}
0\rightarrow W_{k-1}\Omega_{{\mathbf P}}^p({\rm log} (-K))
\rightarrow W_k \Omega_{{\mathbf P}}^p({\rm log} (-K))
\stackrel{\rm Res}{\rightarrow}
\bigoplus_{{\sigma}\in{\Sigma}(k)}\Omega_{V({\sigma})}^{p-k}
\rightarrow 0,
\label{weighted.sequence}
\end{equation}
where $V({\sigma})$ is the Zariski closure of the torus-invariant orbit of ${\mathbf P}$
corresponding to ${\sigma}\in{\Sigma}(k)$ and {\rm Res} is the Poincar\'e residue map.
\end{theo}
It is easy to see that by tensoring by ${\mathcal L}$ the short exact sequence
(\ref{weighted.sequence}) and applying the vanishing Theorem
\ref{B.S.D} we obtain
\begin{eqnarray*}
0\rightarrow H^0({\mathbf P},W_{k-1}\Omega_{{\mathbf P}}^p({\rm log} (-K))\otimes {\mathcal L})
\rightarrow H^0({\mathbf P},W_k \Omega_{{\mathbf P}}^p({\rm log} (-K))\otimes {\mathcal L})
\rightarrow\\
\rightarrow
\bigoplus_{{\sigma}\in{\Sigma}(k)}H^0({\mathbf P},\Omega_{V({\sigma})}^{p-k}\otimes {\mathcal L})
\rightarrow 0.
\end{eqnarray*}
Thus, using an induction, we get
\[
h^0({\mathbf P},W_k \Omega_{{\mathbf P}}^p({\rm log} (-K))\otimes {\mathcal L})=
\sum_{s=0}^k\sum_{{\sigma}\in{\Sigma}(s)}
h^0({\mathbf P},\Omega_{V({\sigma})}^{p-s}\otimes {\mathcal L}).
\]
The generalized Bott formula from Theorem \ref{Bott.formula.I} and Proposition
\ref{support.polytope} implies
\[
h^0({\mathbf P},\Omega_{V({\sigma})}^{p-s}\otimes {\mathcal L})=
\sum_{j=0}^{p-s}(-1)^j\binom{n-s-j}{p-s-j}
\sum_{\begin{subarray}{l}
{\tau}\in{\Sigma}(j) \\
{\tau}\succ{\sigma}
\end{subarray}}
{\it l}(\Delta_{\tau}).
\]
We have proved the following statement.
\begin{prop} Let ${\mathbf P}$ be a complete simplicial toric variety and
${\mathcal L}$ be an ample invertible sheaf on ${\mathbf P}$ with the support polytope
$\Delta$. Then one has the equality
\[
h^0({\mathbf P},W_k \Omega_{{\mathbf P}}^p({\rm log} (-K))\otimes {\mathcal L})=
\displaystyle
\sum_{s=0}^k
\sum_{{\sigma}\in{\Sigma}(s)}
\sum_{j=0}^{p-s}(-1)^j\binom{n-s-j}{p-s-j}
\sum_{\begin{subarray}{l}
{\tau}\in{\Sigma}(j) \\
{\tau}\succ{\sigma}
\end{subarray}}
{\it l}(\Delta_{\tau}).
\]
\end{prop}
\section{Cohomology of quasi-smooth hypersurfaces
\label{section.cohomologies}
}
In this section we reprove the ``combinatorial part'' of the well-known result
of Danilov and Khovanski\^i on the Hodge numbers of quasi-smooth
hypersurfaces in toric varieties.
Our proof is simple and uses the Bott formula for toric varieties.
Any complete simplicial toric variety ${\mathbf P} = {\mathbf P}({\Sigma})$ can be constructed as a
geometric quotient as follows (see \cite{Cox}).
Suppose that $S({\Sigma}) := {\mathbf C}[z_1,\ldots,z_d]$ is the
polynomial ring over ${\mathbf C}$ with the variables $z_i$ corresponding to the
integer generators $v_i$ of the one-dimensional cones of ${\Sigma}$.
For every ${\sigma}\in{\Sigma}$ let
$\hat{z}_{\sigma} := \prod_{v_i\not\in{\sigma}} z_i$, and let
\[
Z({\Sigma}) := \bigcap_{{\sigma}\in{\Sigma}}\{z\in{\mathbf C}^d:\,\hat{z}_{\sigma}=0\}.
\]
The toric variety ${\mathbf P}$ is a geometric quotient of the Zariski open set
$U({\Sigma}) := {\mathbf C}^d\backslash Z({\Sigma})$ by the algebraic group
$G({\Sigma}) := {\rm Hom}_{\mathbf Z}(A_{n-1}({\mathbf P}),\,{\mathbf C}^*)$, where $A_{n-1}({\mathbf P})$ is the Chow
group of ${\mathbf P}$. The action of $G({\Sigma})$ on ${\mathbf C}^d$ induces the grading on
$S({\Sigma})$.
Moreover, if ${\mathcal L}$ is an invertible sheaf on ${\mathbf P}$, then for
$\alpha = [{\mathcal L}]\in A_{n-1}({\mathbf P})$ one has the isomorphism
$H^0({\mathbf P},{\mathcal L})\simeq S_\alpha$.
A polynomial $f$ in the graded part $S_\alpha$ of $S({\Sigma})$ corresponding
to the class $\alpha\in A_{n-1}({\mathbf P})$ is said to be $G$-{\it homogeneous}
of degree $\alpha$. Thus, the global sections of ${\mathcal L}$ determine the Cartier
divisor $D$ on ${\mathbf P}$.
\begin{dfn}[\cite{Batyrev-Cox}]
The hypersurface $D\subset {\mathbf P}$ defined by the $G$-homogeneous polynomial $f$
in $S({\Sigma})$ is said to be {\it quasi-smooth} if ${\bf V}(f)\cap U({\Sigma})$
is either empty or a nonsingular subvariety of codimension one in $U({\Sigma})$.
\end{dfn}
\begin{dfn}
Let $X$ be an algebraic variety over ${\mathbf C}$.
Denote by $h^{p,\,q}(H^k(X,{\mathbf C}))$ the dimension of the $(p,\,q)$-component
of the $k$-th cohomology group.
Let us introduce the numbers
\begin{eqnarray*}
&&e^{p,\,q}(X) := \sum_k (-1)^k h^{p,\,q}(H^k(X,{\mathbf C})),\\
&&e^p (X) := \sum_q e^{p,\,q}(X).
\end{eqnarray*}
For a complete nonsingular variety $X$, we have
\[
e^{p,\,q}(X) = (-1)^{p + q} h^{p,\,q}(X),
\]
where $h^{p,\,q}(X)$ are the (ordinary) Hodge numbers of $X$.
\end{dfn}
\begin{rem}
Danilov and Khovanski\^i use cohomologies with compact supports
instead of usual cohomologies, but in our situation there exists
the Poincar\'e duality
\begin{eqnarray}
h^{p,\,q}(H^k(X,{\mathbf C})) = h^{r-p,\,r-q}(H_c^{2r-k}(X,{\mathbf C})),
\label{Poinc}
\end{eqnarray}
where $r = \dim_{{\mathbf C}}X$, which relates the two treatments.
\end{rem}
Let $D\subset {\mathbf P}$ be a nondegenerate divisor on a complete simplicial
$n$-dimensional toric variety. Then the natural map
\[
H^i({\mathbf P})\rightarrow H^i(D)
\]
is an isomorphism for $i < n - 1$ and is injective for $i = n - 1$.
\begin{dfn}
Define the {\it primitive cohomology} $H_0^{n - 1}(D)$ by
the exact sequence
\[
0 \rightarrow H^{n - 1}({\mathbf P})
\rightarrow H^{n - 1}(D)
\rightarrow H_0^{n - 1}(D)
\rightarrow 0.
\]
\end{dfn}
\begin{rem}
As A.~Mavlyutov pointed out to us (see
\cite{Mavlyutov2}, Remark 5.5), the primitive cohomologies $H_0^{n - 1}(D)$
coincide with the residue part $H_{res}^{n - 1}(D)$ of cohomology defined
as the image of the residue map
$Res:\,H^n({\mathbf P}\backslash D)\rightarrow H^{n - 1}(D)$.
\end{rem}
Recall, that since ${\mathbf P}$ is simplicial and $D$ is nondegenerate,
$H_0^{n - 1}(D)$ has a pure Hodge structure (cf., \cite{Batyrev-Cox}).
\begin{theo}[\cite{Danilov-Khovanskii}]
Let $\Delta$ be a support polytope, corresponding
to an ample nondegenerate hypersurface $D$. Then,
\[
h_0^{p,\,n - 1 - p}(D) =
(-1)^n \sum_{\Gamma\subseteq\Delta}
\varphi_{\dim\Gamma - p}(\Gamma),
\]
where
\[
\varphi_i(\Gamma) :=
\sum_{k = 1}^i (-1)^{i - k} \binom{\dim\Gamma + 1}{i - k} {\it l}^*(k\cdot\Gamma).
\]
\label{D-Kh}
\end{theo}
Denote by $\Omega_{{\mathbf P}}^p({\rm log}\, D)$ the sheaf of $p$-differential forms
on ${\mathbf P}$ with logarithmic poles along $D$ (see \S 15 of \cite{Danilov}).
First, we compute the Euler-Poincar\'e characteristic
of the sheaf $\Omega_{{\mathbf P}}^p({\rm log}\, D)$.
\begin{prop}[\cite{Batyrev-Cox}, Proposition 10.1]
If $D$ is a quasi-smooth hypersurface of a complete toric variety
${\mathbf P}$ defined by zeros of a global section of the ample invertible sheaf
${\mathcal O}_{{\mathbf P}}(D)$ on ${\mathbf P}$, then there is an exact sequence
\begin{eqnarray*}
0\rightarrow \Omega_{{\mathbf P}}^p({\rm log}\, D)
\rightarrow
\Omega_{{\mathbf P}}^p(D) \stackrel{d}{\rightarrow}
\Omega_{{\mathbf P}}^{p+1}(2D)/\Omega_{{\mathbf P}}^{p+1}(D)
\stackrel{d}{\rightarrow}
\ldots\\
\ldots\stackrel{d}{\rightarrow}
\Omega_{{\mathbf P}}^n((n-p+1) D)/\Omega_{{\mathbf P}}^n((n-p) D)
\rightarrow 0.
\end{eqnarray*}
\label{Long.exact.sequence}
\end{prop}
\begin{lem} Let $\Delta$ be the support polytope for ${\mathcal O}_{{\mathbf P}}(D)$
corresponding to the ample divisor $D$. Then the Euler-Poincar\'e
characteristics of the sheaf $\Omega_{{\mathbf P}}^p({\rm log}\, D)$ is equal to
\[
\chi({\mathbf P},\Omega_{{\mathbf P}}^p({\rm log}\, D))=
\sum_{k = 1}^{n - p}(-1)^{k + 1}
\sum_{\Gamma\subseteq \Delta}
\binom{\dim\Gamma + 1}{p + k}{\it l}^*(k\cdot\Gamma).
\]
\end{lem}
\begin{proof}
Indeed, from the long exact sequence of
Proposition \ref{Long.exact.sequence}, we have
\[
\chi({\mathbf P},\Omega_{{\mathbf P}}^p({\rm log}\, D))=
\sum_{k = 1}^{n - p}
(-1)^{k + 1}
[\chi({\mathbf P},\Omega_{{\mathbf P}}^{p + k - 1}(kD)) +
\chi({\mathbf P},\Omega_{{\mathbf P}}^{p + k}(kD))].
\]
Since $D$ is ample,
\[
\chi({\mathbf P},\Omega_{{\mathbf P}}^p({\rm log}\, D))=
\sum_{k = 1}^{n - p}
(-1)^{k + 1}
[h^0({\mathbf P},\Omega_{{\mathbf P}}^{p + k - 1}(kD)) +
h^0({\mathbf P},\Omega_{{\mathbf P}}^{p + k}(kD))],
\]
applying the generalized Bott formula given in Theorem
\ref{Bott.formula.II}, we prove the statement.
\end{proof}
We start with the relation
\[
e^p(D) = e^{p + 1}({\mathbf P}) - e^{p + 1}({\mathbf P}\backslash D),
\]
following from the Gysin exact sequence
\[
\ldots\, \rightarrow \, H^{k-2}(D) \,
\rightarrow \, H^k({\mathbf P}) \,
\rightarrow \, H^k({\mathbf P}\backslash D) \,
\rightarrow \, H^{k - 1}(D) \,
\rightarrow \,
\ldots
\]
of hypercohomologies of the exact sequence of complexes
\[
0\, \rightarrow \, \Omega_{{\mathbf P}}^{\cdot} \,
\rightarrow \, \Omega_{{\mathbf P}}^{\cdot}({\rm log}\, D) \,
\rightarrow \, \Omega_{D}^{\cdot - 1} \,
\rightarrow \, 0.
\]
There are two spectral sequences
\[
'E_1^{p,\,q} = H^q ({\mathbf P},\Omega_{{\mathbf P}}^p)
\Rightarrow H^{p + q}({\mathbf P},{\mathbf C}),
\]
and
\[
''E_1^{p,\,q} = H^q ({\mathbf P},\Omega_{{\mathbf P}}^p({\rm log}\, D))
\Rightarrow H^{p + q}({\mathbf P}\backslash D,{\mathbf C}),
\]
both degenerating at the first term and converging to the Hodge
filtration ${\mathcal F}^\cdot$ on $H^*({\mathbf P},{\mathbf C})$ and $H^*({\mathbf P}\backslash D,{\mathbf C})$
respectively \cite{Danilov}. Therefore,
we have the equalities
\[
e^{p + 1}({\mathbf P}) = (-1)^{p + 1} \chi({\mathbf P},\Omega_{{\mathbf P}}^{p + 1}),
\quad
e^{p + 1}({\mathbf P}\backslash D) =
(-1)^{p + 1} \chi({\mathbf P},\Omega_{{\mathbf P}}^{p + 1}({\rm log}\, D)).
\]
Using the Bott formula, one obtains:
\begin{prop} One has the equality
\[
e^p(D) =
(-1)^{p + 1} \sum_{k \ge 0}
(-1)^k \binom{k}{p + 1} f_k -
\sum_{\Gamma\subseteq \Delta}
(-1)^{\dim \Gamma}
\varphi_{\dim\Gamma - p} (\Gamma).
\]
\end{prop}
By applying the Poincar\'e duality (\ref{Poinc}), we restore all the
Hodge numbers $h_0^{p,\,q}(D)$ and get the statement of Theorem \ref{D-Kh}.
\section{The Bott formula on ${\mathbf P}({\mathcal E})$}
Let ${\mathcal L}_0,\ldots,{\mathcal L}_s$ be ample invertible sheaves on a complete
$n$-dimensional toric variety ${\mathbf P}$. Denote $Y = {\mathbf P}({\mathcal E})$
the projective space bundle associated with the sheaf
${\mathcal E} := {\mathcal L}_0\oplus\cdots\oplus {\mathcal L}_s$, with the invertible sheaf ${\mathcal O}_Y(1)$
and the projection $\pi:\,Y\rightarrow {\mathbf P}$. The sheaf ${\mathcal O}_Y(1)$ is ample since
the ${\mathcal L}_j$ are ample (see \S1 of \cite{Hartshorn}).
\begin{dfn}
Let $\Omega^1_{Y/{\mathbf P}}$ be the sheaf of relative differentials arising from
the short exact sequence
(see \cite{Manin})
\begin{eqnarray}
\label{Omega_Y}
0 \rightarrow \Omega^1_{Y/{\mathbf P}}
\rightarrow \pi^*({\mathcal E})\otimes {\mathcal O}_Y(-1)
\rightarrow {\mathcal O}_Y
\rightarrow 0.
\end{eqnarray}
\end{dfn}
The purpose of this section is to give a combinatorial description of
the global sections of the sheaf of relative $p$-th differentials
$\Omega^p_{Y/{\mathbf P}}(k) :=
\bigwedge^p \Omega^1_{Y/{\mathbf P}}\otimes {\mathcal O}_Y(k)$.
Recall the construction of $Y$ as a toric variety (see \cite{Oda.book}).
Suppose that the support polytope associated with the sheaf ${\mathcal L}_j$ has the form
\[
\Delta_j = \{ m\in M_{{\mathbf R}}: \<m,v_i\>\ge -a_{ij},\, i = 1,\ldots,d\}.
\]
Let $N'\simeq {\mathbf Z}^s$ be a lattice with ${\mathbf Z}$-basis $\{e_1,\ldots,e_s\}$
and $\widetilde{N} := N \oplus N'$. The $1$-dimensional cones of the fan
$\widetilde{\Sigma}$ corresponding to the toric variety $Y$ have the
generators
\begin{eqnarray*}
\widetilde{v}_i &=& v_i + \sum_{j = 1}^s (a_{ij} - a_{i0}) e_j,
\quad i = 1,\ldots,d,\\
\widetilde{e}_0 &=& -e_1 - \ldots -e_s, \\
\widetilde{e}_j &=& e_j, \quad j = 1,\ldots,s.
\end{eqnarray*}
Denote by $\widetilde{\sigma}$ the image of each $\sigma\in\Sigma$
under the map $N_{{\mathbf R}}\rightarrow \widetilde{N}_{{\mathbf R}}$ given by
$v_i\mapsto \widetilde{v}_i$ and let $\sigma'$ be the cones generated
by ${e}_0,\ldots,{e}_{i-1},{e}_{i+1},\ldots,{e}_s$. Then the fan
$\widetilde{\Sigma}$ is generated by the cones $\widetilde{\sigma} + \sigma'$
and their faces.
Let $\eta_1,\ldots,\eta_s$ be the generators of the lattice $M'$
dual to $N'$. Denote the torus invariant divisors on $Y$ corresponding
to $\widetilde{v}_i$ and $\widetilde{e}_j$ by $\widetilde{D}_i$ and
$\widetilde{D}_j'$ respectively. Note that
$\pi^*({\mathcal L}_j) = {\mathcal O}_Y(\sum_{i = 1}^d a_{ij} \widetilde{D}_i)$.
\begin{lem} The polytope
\label{support_O_Y}
\[
\nabla =
\{
\lambda_1 \eta_1 + \ldots + \lambda_s \eta_s +
\lambda_0 m_0 + \ldots + \lambda_s m_s :\, \lambda_j\ge -k_j,\,
\sum_{j = 0}^s \lambda_j = k,\, m_j \in \Delta_j
\}
\]
in $\widetilde{M}_{{\mathbf R}} := M_{{\mathbf R}}'\oplus M_{\mathbf R}$ is the support polytope
associated with the sheaf
${\mathcal O}_Y(k)\otimes
{\mathcal O}_Y(k_0\widetilde{D}_0' + \ldots + k_s \widetilde{D}_s')$.
\end{lem}
\begin{proof} It follows from the Lemma 2.1 in \cite{Mavlyutov} that
\[
{\mathcal O}_Y(1) \simeq {\mathcal O}_Y(\widetilde{D}_0')\otimes\pi^*({\mathcal L}_0) =
{\mathcal O}_Y(\widetilde{D}_0' +
\sum_{i = 1}^d a_{i0} \widetilde{D}_i)
\]
and
\[
{\mathcal O}_Y(k)\otimes
{\mathcal O}_Y(k_0 \widetilde{D}_0' + \ldots + k_s \widetilde{D}_s') =
{\mathcal O}_Y((k + k_0)\widetilde{D}_0' + \sum_{j = 1}^s k_j
\widetilde{D}_j' + \sum_{i = 1}^d k a_{i0}
\widetilde{D}_i).
\]
Each $\widetilde{m}\in\widetilde{M}_{{\mathbf R}}$ can be uniquely written as
$\widetilde{m} = m + \lambda_1\eta_1 + \ldots + \lambda_s\eta_s$, where
$m\in M_{{\mathbf R}}$, $\lambda_j\in{\mathbf R}$. The support polytope $\nabla$ is defined
by the inequalities
\begin{eqnarray*}
&&\<\widetilde{m},\widetilde{v}_i\> \ge -k a_{i0}, \quad i = 1,\ldots,d,\\
&&\<\widetilde{m},\widetilde{e}_0\> \ge -k - k_0, \\
&&\<\widetilde{m},\widetilde{e}_j\> \ge -k_j, \quad j = 1,\ldots,s
\end{eqnarray*}
in $\widetilde{M}_{{\mathbf R}}$. Let $\lambda_0 = k - \sum_{j = 1}^s \lambda_j$.
Then the inequalities above are equivalent to
\[
\{
\<m,v_i\> \ge - \sum_{j = 0}^s \lambda_j a_{ij}, \quad i = 1,\ldots,d, \quad
\lambda_j \ge -k_j, \quad j = 0,\ldots,s,\quad \sum_{j = 0}^s \lambda_j = k
\}.
\]
The same arguments as in \S 3 of \cite{Cattani-Cox-Dickenstein}
show that $m = \sum_{j = 0}^s \lambda_j m_j$, where
$m_j\in\Delta_j$, and $\widetilde{m}\in\nabla$ follows immediately.
\end{proof}
\begin{dfn}
For any $J = \{j_1,\ldots,j_t\}\subset I := \{0,\ldots,s\}$
define the polytope
\[
\Delta_{j_1\ldots j_t} (k) :=
\{
\sum_{i = 1}^s \lambda_i \eta_i +
\sum_{t = 0}^s \lambda_t m_t:\, \lambda_j\ge 1,\, j\in J,\,
\lambda_j\ge 0,\,j\in I\backslash J,\,\sum_{j = 0}^s \lambda_j = k,\,
m_t\in\Delta_t
\}.
\]
\end{dfn}
\begin{theo}For any sufficiently large integer $k > 0$
\begin{eqnarray*}
h^q(Y,\,\Omega_{Y/{\mathbf P}}^p(k))
= \left\{
\begin{array}{ll}
\displaystyle
\sum_{t = 0}^p (-1)^{p - t}
\sum_{0 \le j_1 < \ldots < j_t \le s}
{\it l} (\Delta_{j_1 \ldots j_t}(k)), &q = 0,\\
\\
0, &q > 0.
\end{array}
\right.
\end{eqnarray*}
\end{theo}
\begin{proof} From the $p$-th exterior power of the sequence (\ref{Omega_Y}),
we have
\[
0 \rightarrow \Omega^p_{Y/{\mathbf P}}(k)
\rightarrow \bigwedge\nolimits^p \pi^*({\mathcal E})\otimes {\mathcal O}_Y(k - p)
\rightarrow \Omega^{p - 1}_{Y/{\mathbf P}}(k)
\rightarrow 0.
\]
Since ${\mathcal O}_Y(1)$ is ample, for all $k\gg 0$, we obtain the exact
sequence of global sections
\[
0 \rightarrow H^0(Y,\,\Omega^p_{Y/{\mathbf P}}(k))
\rightarrow
H^0(Y,\,\bigwedge\nolimits^p \pi^*({\mathcal E})\otimes {\mathcal O}_Y(k - p))
\rightarrow
H^0(Y,\,\Omega^{p - 1}_{Y/{\mathbf P}}(k))
\rightarrow 0.
\]
Lemma 2.1 of \cite{Mavlyutov} implies the isomorphisms
$\pi^*({\mathcal L}_{j_m})\otimes{\mathcal O}_Y(-1)\simeq{\mathcal O}_Y(-\widetilde{D}_{j_m}')$
and consequently, the isomorphisms
\begin{eqnarray*}
\bigwedge\nolimits^p \pi^*({\mathcal E})\otimes {\mathcal O}_Y(k - p)
&\simeq&
{\mathcal O}_Y(k)\otimes\bigoplus_{0\le j_1 < \ldots < j_p \le s}
\pi^*({\mathcal L}_{j_1})\otimes\cdots\otimes
\pi^*({\mathcal L}_{j_p})\otimes{\mathcal O}_Y(-p) \\
&\simeq&
{\mathcal O}_Y(k)\otimes\bigoplus_{0\le j_1 < \ldots < j_p \le s}
{\mathcal O}_Y(-\widetilde{D}_{j_1}' - \ldots - \widetilde{D}_{j_p}').
\end{eqnarray*}
Hence
we obtain the relations
\[
h^0(Y,\,\Omega^p_{Y/{\mathbf P}}(k)) = \sum_{0\le j_1 < \ldots < j_p \le s}
{\it l}(\Delta_{j_1\ldots j_p }(k)) - h^0(Y,\,\Omega^{p - 1}_{Y/{\mathbf P}}(k)).
\]
By induction we get the requested formula.
\end{proof}
\section{Appendix}
In the proof of Theorem \ref{comb_dual} we have used the following identity.
\begin{lem} For any face $F$ of a simple $n$-polytope $\Delta$
there are relations
\[
\sum_{\Gamma \supseteq F} (-1)^{dim \Gamma} \binom{\dim\Gamma}{p}
=
(-1)^{n} \binom{\dim F}{n - p}.
\]
\end{lem}
\begin{proof} Since $\Delta$ is a simple, there are precisely
\[
\binom{n-s}{n-j}
\]
$j$-faces of $\Delta$ containing a given $s$-face $F$ of $\Delta$.
Hence, the requested formula is equivalent to the combinatorial identity
\[
\sum_{j = 0}^n (-1)^j \binom{j}{p}\binom{n-s}{n-j} =
(-1)^n \binom{s}{n-p},
\]
or
\[
\sum_{j = 0}^{n-s} (-1)^j \binom{n-j}{p}\binom{n-s}{j} =
\binom{s}{n-p}.
\]
We give a proof of this identity based on the method of integral
representations. Rewrite the sum
\[
S = \sum_{j = 0}^{n - s} (-1)^j \binom{n - j}{p}\binom{n - s}{j}
\]
as
\[
S = \sum_{j = 0}^\infty (-1)^j \frac{1}{(2\pi i)^2}
\iint_{\gamma}
\frac{(1 + z)^{n - j} (1 + w)^{n - s}}
{z^{p + 1} w^{j + 1}}\, dz\, dw,
\]
where the cycle
$\gamma$ is $\{|z| = \varepsilon\}\times \{|w| = \delta\}$.
One can choose the numbers $\varepsilon > 0$ and $\delta > 0$
small enough, say $\varepsilon = 1/2$, $\delta = 3$, so
that the geometric series $\sum_{j = 0}^\infty (-(1+ z)^{-1}w^{-1})^{j}$
converges on the contour of integration $\gamma$, and it is possible
to reverse the order of integration and summation.
Therefore, summing up a geometric sequence,
\begin{eqnarray*}
S &=& \frac{1}{(2\pi i)^2} \iint_{\gamma}
\frac{(1 + z)^n (1 + w)^{n - s}}{z^{p + 1} w}
\sum_{j = 0}^\infty
\left( \frac{-1}{(1 + z)w}\right)^j\, dz\, dw
\\
&=&
\frac{1}{(2\pi i)^2}\iint_{\gamma}
\frac{(1 + z)^n (1 + w)^{n - s}}{z^{p + 1} w
\left(1 +\displaystyle\frac{1}{(1 + z)w} \right)}\, dz\, dw
\\
&=&
\frac{1}{2\pi i}\int_{|z| = 1/2}
\frac{(1 + z)^{n + 1}}{z^{p + 1}}\,dz \cdot
\frac{1}{2\pi i}\int_{|w| = 3}
\frac{(1 + w)^{n - s}}{1 + zw + w}\,dw.
\end{eqnarray*}
By the residue theorem, the second integral is
\begin{eqnarray*}
\frac{1}{2\pi i}\int_{|w| = 3}
\frac{(1 + w)^{n - s}}{1 + zw + w}\,dw
&=&
res_{w = w_0} \frac{(1 + w)^{n - s}}{(1 + z)(w + 1/(1 + z))}\\
&=&
\frac{\left(1 - \displaystyle\frac{1}{1 + z}\right)^{n - s}}{1 + z} =
\frac{z^{n - s}}{(1 + z)^{n - s + 1}},
\end{eqnarray*}
where $w_0 = -1/(1 + z)\in\{|w|=3\}$, since
$|w_0| \le \displaystyle\frac{1}{1 - 1/2} = 2 < 3$ for $|z| = 1/2$.
Finally,
\begin{eqnarray*}
S &=& \frac{1}{2\pi i} \int_{|z| = 1/2}
\frac{(1 + z)^{n + 1}}{z^{p + 1}} \cdot
\frac{z^{n - s}}{(1 + z)^{n - s + 1}}\,dz \\
&=&
\frac{1}{2\pi i} \int_{|z|=1/2}
\frac{(1 + z)^s}{z^{p - n + s +1}} \,dz =
\binom{s}{p -n + s} = \binom{s}{n - p},
\end{eqnarray*}
which concludes the proof.
\end{proof}
|
\section{First-Level heading}
Even though many fake
smoothings of $4$-manifolds are known to exists, we know little about the basic
building blocks of exotic smooth manifolds. This is mainly due to
the fact that we still don't know if basic manifolds like $S^{4}$,
$S^{2}\times S^{2}$, and $S^{1}\times S^{3}$ could admit fake smooth
structures. Unable to show this, we demonstrate fake smooth structures
on manifolds that are in a way ``small deformations" of $ S^{2}\times
B^{2} $ (as in \cite{A1},\cite{A2}).
Fishtail $F$ is the tubular neighborhood of an immersed
$2$-sphere in $S^{4}$ with one self intersection; as an handlebody it
can be described as a $4$-ball with $1 $ and $2$-handles attached as in
first picture of Figure 1. Recalling the $1$-handle notation of
\cite{A0}, $F$ is obtained by removing a tubular neighborhood of the
obvious disc
$ B^{2}\subset S^{2}\times B^{2} $ (which the ``circle with dot" bounds).
Cusp $C$ is a $4$-ball with a 2-handle
attached along a trefoil knot with $0$-framing as shown in the second
picture of Figure 1.
\begin{figure}[htb]
\includegraphics[scale=.5]{1.eps}
\caption{}
\end{figure}
\begin{thm} There are compact smooth manifolds $F^{*} $ and
$Q^{*}$, which are homeomorphic but not diffeomorphic to $F$ and $Q$,
respectively. Also, $ F^{*}$ is obtained by removing a tubular
neighborhood of a properly imbedded $2$-disc
$f: B^{2}\hookrightarrow S^{2}\times B^{2} $ from $S^{2}\times B^{2} $.
\end{thm}
In Figure 1 the circle $f(\partial B^{2})$ corresponds to the
circle with dot. Existence of a fake cusp was
first established jointly with R.Matveyev as application of methods of
\cite{AM}
\section{Seiberg-Witten invariants}
Let $X$ be a closed smooth $4$-manifold and $ Spin_{c}(X) $ be the set of
$ Spin_{c} $ structures on $ X $.
In case $H_{1}(X)$ has no $2$-torsion $ Spin_{c}(X) $ can be identified by
$$ Spin_{c}(X) =\{ a\in H^{2}(X,{\bf
Z})\;|\; a=w_{2}(TX) \; (\mbox{mod } 2) \}$$
Recall Seiberg-Witten invariant
$$SW_{X}:Spin_{c}(X) \to {\bf Z} $$
A classes $a\in H^{2}(X,{\bf Z})$ is called {\it basic} if $SW_{X}(a)\neq 0$.
It is known that there are finitely many basic classes and if $a$ is basic then
so is $-a$, and $$SW_{X}(-a)= (-1)^{\varepsilon}SW_{X}(a)$$
where $ \varepsilon = (e(X)+\sigma(X))\;/4 $, and $e(X), \sigma(X)$ denote Euler characteristic
and signature. If ${\it B}=\{\pm \a_1, \pm \a_2,..,\pm a_n\}$ are
the basic classes, by denoting $ a_{0}=SW_{X}(0), a_{j}=SW_{X}(\a_{j}) $
and $t_{j}=exp(\a_{j})$
Seiberg-Witten invariants can be assembled a single polynomial \cite{FS}
$$ SW_{X}= a_0 + \sum_{j=0}^{n}a_{i}(t_{j} + (-1)^{\varepsilon}t_{j}^{-1})$$
In \cite{FS} Fintushel and Stern introduced a method of modifying a
$4$-manifold by using a knot in $ S^{3}$, which changes its
Seiberg-Witten invariants without changing its homeomorphism type.
Let $X$ be a closed smooth $4$-manifold and $ T^2\subset X $ be an
imbedded $2$-torus with trivial normal bundle. Assume that this torus
lies in in a cusp neighborhood; this means that inside of $X$ the
tubular neighborhood $T^2\times B^2 $ of the $2$-torus (first picture of
Figure 2) is contained in a cusp C (the second picture Figure 2). Call
such a $2$-torus in $X$ a
\textit{c-imbedded torus}
\begin{figure}[htb]
\includegraphics[scale=.5]{2.eps}
\caption{}
\end{figure}
Let $ K \subset S^3 $ be a knot, and $ N=N(K)\approx K\times B^2 $ be
the trivialization of its open tubular neighborhood given by the
$0$-framing. Let
$ \varphi: \partial (T^{2}\times B^{2})\to \partial (K\times B^{2}) \times
S^{1}
$ be any diffeomorphism with $ \varphi (p\times \partial B^2)=K\times p $
where
$ p\in T^{2} $ is a point. Define:
$$X_{K}= (X-T^2\times B^2)
\smile_{\varphi} (S^{3}-N)\times S^1$$
Let $[T]$ be the homology class in $H_{2}(X_{K};{\bf Z})$
induced from $T^{2}\subset X$, and $ t=\mbox{exp}(2[T]) $, and let
$\Delta_{K}(t)$ be the Alexander polynomial of the knot $K$ (as a symmetric
Laurent polynomial), then
\begin{thm}{\cite{FS}}: Let $X$ be a smooth manifold as above, and
$ K\subset S^{3} $ be a knot , then
Seiberg-Witten invariants of $ X_{K} $ can be computed
$$ SW_{X_{K}}=SW_{X} . \;\Delta_{K}(t) $$
\end{thm}
\section{Handlebody of $X_{K}$ }
Here we will give a general algorithm of describing
the handlebody description of $X_{K}$ from the handles of $X$. It could
be beneficial for the reader to compare the steps of this section with
\cite{A3}. Let
$ K\subset S^{3}
$ be a knot, depicted as the left handed trefoil knot in Figure 3, and $N$ be
its open tubular neighborhood. We claim that the linking circles
$\alpha $ and
$ \beta $ are the {\it core circles} of the $1$- handles of the
handlebody of $ S^3-N $ (Heegard diagram).
\begin{figure}[htb]
\includegraphics[scale=.5]{3.eps}
\caption{}
\end{figure}
This can easily
checked by the process described in \cite{R} pp.250, which turns a
surgery description of a $3$-manifold to its Heegard diagram.
Steps in Figures 4 describes this process (i.e. attach canceling pair of
$1$ and $2$ - handles to $ S^{3}-N $ until the complement becomes a solid
handlebody). Write:
\begin{figure}[htb]
\includegraphics[scale=.5]{4.eps}
\caption{}
\end{figure}
$$(S^3 -N) \times S^1= (S^3 -N)\times I_{+}
\smile (S^3 -N)\times I_{-}$$
where $I_{\pm} \approx I =[0,1]$
are closed intervals and the union is taken along the boundaries,
i.e. along $ (S^3 -N)\sqcup (S^3 -N)$.
Up to attaching a $3$-handle, $ (S^3 -N)\times I_{-}$ is
obtained by removing the tubular neighborhood of a properly
imbedded arc (with trefoil knot tied on it) from $B^3$, and
crossing it with $I$ as indicated in Figure 5.
\begin{figure}[htb]
\includegraphics[scale=.6]{5.eps}
\caption{}
\end{figure}
Equivalently, $ (S^3 -N)\times I_{-}$ is obtained by removing the ``usual"
slice disc from
$B^4$, which trefoil knot connected summed with its mirror image $K\# (-K)$
bounds, as indicated in Figure 6. The dot on the knot $\;K\#(-K)\;$
in Figure 6 indicates that the tubular neighborhood of the obvious slice
disc which it bounds is removed from $ B^4 $. We will refer this
as \textit{slice 1-handle}. This notation was discussed in
\cite{AK} and \cite{A3}.
\begin{figure}[htb]
\includegraphics[scale=.5]{6.eps}
\caption{}
\end{figure}
To get $ (S^3 -N)\times S^1 $ we glue the upside down
handlebody of $ (S^3 -N)\times I_{+} $ to $ (S^3 -N)\times
I_{-} $. Clearly, up to attaching $3$-handles, this is achieved by
attaching $(S^3 -N)\times I_{-}$ one $1$-handle and two $2$-handles,
resulting from identification of the corresponding $1$-handles $\a$
and
$\b$ (of the knot complements) in the two boundary components
of $(S^3-N)\times I_{-}$, as indicated in Figure 7.
So Figure 7 gives $( S^3-N)\times S^1$.
\begin{figure}[htb]
\includegraphics[scale=.5]{7.eps}
\caption{}
\end{figure}
At this stage it is
instructive to check that the boundary of the handlebody of Figure
7 is indeed $T^3$. This can be seen by removing the ``dot" from
the ``slice 1-handle" then performing handle moves as indicated by
the arrows in Figure 7. This gives the first picture of Figure 8, then by
sliding one of the $0$-framed handles from the other one we get the
second picture Figure 8. Since the trivial knot with $0$-framing
is canceled by a $3$-handle, we see that Figure 8 is $ T^2\times B^2 $
(with boundary $T^3 $). Hence the reverse operation
Figure 8 $\to$ Figure 7 corresponds the modification $X\to X_{K}$
\begin{figure}[htb]
\includegraphics[scale=.5]{8.eps}
\caption{}
\end{figure}
We can see the operation Figure 8 $\to $ Figure 7 directly as
follows: Start with $ T^2\times B^2 $, by attaching a canceling
pair of $2$ and $3$ handles (i.e. by introducing an unknot with
$0$-framing), and then by sliding this new $2$-handle over the
$2$-handle of $T^2\times B^2$ we get another handle description of
$T^2\times B^2$ in Figure 9. Now by removing the ``dot" on the
$1$-handle (i.e. by surgery turning $1$-handle $S^1\times B^3$ to
$B^2\times S^2$) and performing the handle slides as indicated by the
arrows of Figure 9 (and putting a ``dot" on a resulting $0$-framed knot), we
get Figure 7. The ``circled $1/2 $ notation" on one of the arrows of Figure
9 means that when doing handle slide put one half-twist on the band.
\begin{figure}[htb]
\includegraphics[scale=.5]{9.eps}
\caption{}
\end{figure}
Hence to see exactly how the
operation $X\to X_{K}$ modifies the handles of $X$. We simply apply this
process to an imbedded $ T^2\times B^2 $ inside $X$, and trace the
rest of the other handles of $X$ along with it. For example, the operation
Figure 9 $\to$ Figure 7 preserves the linking circles $\gamma$ and $\d$, as
indicated in the figures. Therefore, if $T^2\times B^2$ lies in a
cusp neighborhood in $Q$ (i.e. if there are $-1$ framed
handles attached to the knots $\gamma$ and $\d$ of Figure 9), Figure 9
becomes Figure 10, and the operation $ X\to X_{K} $ corresponds to the
operation Figure 10 $\to$ Figure 11 (here disregard the loop $\t$ in
Figures 10 and 11, it will be explained in the next paragraph).
\begin{figure}[htb]
\includegraphics[scale=.5]{10.eps}
\caption{}
\end{figure}
\begin{figure}[htb]
\includegraphics[scale=.5]{11.eps}
\caption{}
\end{figure}
Warning: even though the operation Figure 9 $\to$ Figure 7 preserves
the loops $ \gamma $, $ \d $ of the $1$-handles, it does not preserve the
linking circle $\t$ of the $2$-handle of $ T^2\times B^2 $. Hence for
example, $\t$ of the cusp of Figure 10 is sent to the quite
complicated loop of Figure 11 (also denoted by $\t$).
Finally by drawing the slice $1$-handle of Figure 11 as two
$1$-handles and one $2$-handle , we see that
Figure 11 is diffeomorphic to Figure 12 (canceling one of the
$1$-handles of Figure 12 by the ``middle" $2$-handle gives the slice $1$-handle
of Figure 11, as in \cite{A3}).
\begin{figure}[htb]
\includegraphics[scale=.5]{12.eps}
\caption{}
\end{figure}
To sum up: The operation $X\to X_{K}$ changes an imbedded $T^{2}\times
B^{2}$ inside of $ X $ by changing one of its $1$-handles with a ``slice
$1$-handle" determined by the knot $K\# (-K) $, and
introducing 2-handles connecting the ``core circles" of $1$-handles of the
two knot complements (all except one) as in Figure 13.
\begin{figure}[htb]
\includegraphics[scale=.5]{13.eps}
\caption{}
\end{figure}
\section{Proof of the theorem }
Attaching $2$-handles with $-1$ framings to $\gamma $ and $\d $ of the first picture
of Figure 13 gives the cusp C. By Eliashberg's theorem $C$ is compact
Stein manifold (see \cite{AM} for brief review). C can be compactified
to a Kahler manifold $ X $; for example $ C $ sits in a K$3$ surface as a
codimension zero submanifold. By applying Theorem 1.1 to the torus
$ T^{2}\subset C\subset X $, and a knot $ K $ is with nontrivial Alexander
polynomial, we obtain a fake copy $ X_{K} $ of $ X $ (because $ X_{K} $
has different Seiberg-Witten invariant than $X$). Define $C^{*}=
C_{K}
\subset X_{K} $, then
$C^{*}$ can not be diffeomorphic to $ C $, otherwise the identity map
$ id: X - \mbox{int}(C)\to X_{K}-\mbox{int}(C^{*}) $ would extend to a
diffeomorphism $ X\to X_{K} $. Recall that all self-diffeomorphisms of the
boundary of the cusp $ C $ extends to $ C $. Figure 12 is the handlebody of
$C^{*}$ (in case $K$ is the trefoil knot)
\begin{figure}[htb]
\includegraphics[scale=.5]{14.eps}
\caption{}
\end{figure}
Attaching one $2$-handle with $-1$ framings to either one of the circles $\gamma $ or
$\d $ (say to $ \gamma $) of the first picture of Figure 13 gives the fishtail
$F$. We can think of of $ F $ being obtained from $ C $ by ``undoing" one of
its
$2$-handles, i.e. removing a thickened disc $ D $ from $C$ (dual $2$-handle of
$ \d $). The boundary
$ \partial D $ corresponds the small trivially linking circle of the $-1$ framed
circle $ \d $ ; so removing $ D $ from $ C $ corresponds to putting a ``dot"
on this dual circle (and hence canceling the
$2$-handle $\d$ from $C$). By extending $ id: X - \mbox{int}(C)\to
X_{K}-\mbox{int}(C^{*}) $ across $D$ gives a diffeomorphism
$ f: \partial F\to
\partial F_{K} $ which does not extend over the interior (otherwise $ X $
and $ X_{K} $ would be diffeomorphic).
\begin{figure}[htb]
\includegraphics[scale=.5]{15.eps}
\caption{}
\end{figure}
Define $ F^{*} = F_{K} $, so removing
thickened $D$ from $C^{*}$ (Figure 11) gives $F^{*}$ (Figure 16). It is
easy to verify that $F^{*}$ is homotopy equivalent to $F$.
Furthermore we can
verify that $F^{*}$ is obtained from $S^{2}\times B^{2}$ by removing an
imbedded disc
$D$ as follows: By the handle moves of Figure 7, we see that on the
boundary the position of the uknotted ``circle with dot" of Figure 16 is
the same as the ``circle with dot" of Figure 1 (i.e. F) ; also removing the
`circle with dot" from Figure 16 results $ S^{2}\times B^{2} $ (this can be
verified by going from Figure 16 to the handle presentation of Figure 12).
\begin{figure}[htb]
\includegraphics[scale=.5]{16.eps}
\caption{}
\end{figure}
It remains to be check that $F^{*}$ is not diffeomorphic
to $F$; so far we only know a particular diffeomorphism of $f:\partial F\to
\partial F^{*}$ does not extend inside. Unfortunately unlike C, not every self
diffeomorphism of $ \partial F $ extends over $F$. But from the construction
it is easy to see that $f$ extends to a homotopy equivalence $F^{*}\to F$,
hence the following lemma implies that $F^{*}$ can not be diffeomorphic to
$F$.
\begin{lem} If a diffeomorphism $ f:\partial F\to \partial F $ extends to a
self homotopy equivalence it extends to a diffeomorphism $ F\to F $
\end{lem}
\begin{proof} (outlined by R.Gompf): It is known that $\partial F$ is
a
$T^{2}$ bundle over
$S^{1}$ with monodromy $ \left( \begin{array}{cc} 1&1\\ 0&1 \end{array}\right) $
\cite{K}.
By standard $3$-manifold theory $f$ can be isotoped
to a fiber preserving isotopy. By composing with obvious diffeomorphism that
extends, we can assume that the fiber orientation is preserved. Since $f$ has to
commute with monodromy it fixes the ``vanishing cycle" C, corresponding to $
(1,0) $. So $f$ is a composition of Dehn twist along the horizontal torus $C\times
S^{1}$ along $S^{1}$ direction, and Dehn twist along the fiber in $C$ direction
(Dehn twist orthogonal to $C$ is ruled out since it does not extend to homotopy
equivalence $F\to F$), all these diffeomorphisms extend to $F$. \end{proof}.
\section{Exotic knottings of the cusp and the fishtail}
We can describe $S^4$ as a union of two fishtails along the boundary, and
$S^{2}\times S^{2}$ as a union of two cusps along the boundary.
Figure 17 describe these
identifications.
For example, attaching an upside-down copy of $-F$ to $F$ has
an affect of attaching a $2$ and $3$-handles to $F$ as described in the
first part of Figure 17: Attaching the $2$-handle to $F$ gives
$S^{2}\times B^{2}$ which, after attaching a $3$-handle, becomes $S^{4}$.
Hence we have imbeddings of singular $2$-spheres $f_{0}:F\to S^{4}$ and $
g_{0}:C\to S^{2}\times S^{2} $
\begin{figure}[htb]
\includegraphics[scale=.6]{17.eps}
\caption{}
\end{figure}
\begin{thm} There are imbeddings
$ f_{1}:F\to S^{4} $ and $ g_{1}:C\to S^{2}\times
S^{2} $ that are topologically isotopic but not smoothly isotopic to the
imbeddings $f_{0}$ and $g_{0}$
\end{thm}
\begin{proof} By replacing one of the fishtails in
$S^{4}=F\smile_{\partial} -F$ by $F^{*}$ we obtain a homotopy $4$-sphere, which
can be easily checked to be $S^{4} = F^{*}\smile_{\partial} -F $ (Figure 18).
Similarly by replacing one of the cusps in $ S^{2}\times
S^{2}= C\smile_{\partial}-C $ we obtain a homotopy $S^{2}\times S^{2}$ which can be
checked to be the standard $ S^{2}\times S^{2}= C^{*}\smile_{\partial}-C $.
Unlike the previous case, this check is surprisingly difficult (it requires the
proof of Scharlemann's conjecture).
Figure 19 is the handlebody of
$ C^{*}\smile_{\partial}-C $, it is diffeomorphic to Figure 20 (to see
this, in Figure 20 slide one of the small $-1$ circles over one of the $0$-framed
handles, going through the $1$-handle, then slide $-1$ framed $2$-handle over it).
To identify Figure 20 by $ S^{2}\times S^{2}$ we need to first recall the
handlebody picture of $ \Sigma \times S^{1} $, where $\Sigma $ is the Poincare
homology sphere (\cite{A3}): Figure 21. From
\cite{A3} we know that surgering $ \Sigma \times S^{1} $ (along the loop trivially
linking the slice 1-handle) gives $ S^{3}\times S^{1}\# S^{2}\times S^{2} $, and
surgering once more the obvious $S^{1}$ gives $S^{2}\times S^{2}$. Performing
these two surgeries corresponds to introducing pair of $0$ and $-1$ framed two
handles as indicated in Figure 22. By canceling two
$2$ and $3$-handle pairs from Figure 22 gives Figure 20. Via the handle moves of
Figure 7 one can check that, introducing the two little $0$-framed handles to
Figure 20, to obtain Figure 22, has the affect of changing the boundary from $
S^{1}\times S^{2} $ to
$ S^{1}\times S^{2}\# S^{1}\times S^{2} \# S^{1}\times S^{2} $ \end{proof}.
\begin{rem} The Fintushel-Stern operation $X\to X_{K}$
can be generalized by
$$X\to X_{T}=(X-T^2\times B^2) \smile_{\varphi} (S^{3}\times S^1 - N_{T})$$
where $N_{T}$ is an open tubular neighborhood of an imbedded $ T^{2} \subset
S^{3}\times S^{1}$. Surprisingly it turns out that this operation does not always
change the smooth structure of
$X$ (in particular it does not change the Seiberg-Witten invariant of $X$).
Another generalization of this operation is by removing an open tubular
neighborhood of a Klein bottle $ N(F) $ (a twisted $ B^{2} $-bundle over $ F $)
from $ X $, and replacing it with a $ S^{3}-N(K) $ bundle over $ S^{1} $,
where $K\subset S^{3}$ is an invertible knot and $\psi : S^{3}- N(K)\to S^{3}-
N(K)$ is the inversion
$$X\to X_{K}=(X-N(F)) \smile_{\partial} (S^{3}- N(K))\times _{\psi} S^{1}$$
These operations will be studied in \cite{A4}.
\end{rem}
\begin{figure}[htb]
\includegraphics[scale=.44]{18.eps}
\caption{}
\end{figure}
\begin{figure}[htb]
\includegraphics[scale=.45]{19.eps}
\caption{}
\end{figure}
\begin{figure}[htb]
\includegraphics[scale=.45]{20.eps}
\caption{}
\end{figure}
\begin{figure}[ht]
\includegraphics[scale=.5]{21.eps}
\caption{}
\end{figure}
\begin{figure}[ht]
\includegraphics[scale=.5]{22.eps}
\caption{}
\end{figure}
\newpage
|
\section*{Acknowledgements}
The work of A.B., H.E., S.K., W.M., and W.P. was supported by
the ``Fonds zur F\"orderung der wissenschaftlichen Forschung''
of Austria, project no. P10843--PHY and P13139--PHY.
The work of T.K. and Y.Y. was supported in part by the
Grant--in--aid for Scientific Research from the Ministry of Education,
Science, and Culture of Japan, Nos.~08640388 and 10740106, respectively.
Y.Y. was also supported in part by Fuju--kai Foundation.
|
\section{Introduction}
The RGB Bump is an intrinsic feature of the RGB Luminosity Function
(LF) of GGCs. It appears as a peak in the differential LF or as a
change in the slope of the cumulative LF. The presence of the Bump is
due to the fact that during the RGB evolution the H-burning shell
crosses the chemical discontinuity left over by the convective envelope
soon after the first dredge-up phase.
Since its first detection in 47 Tuc (King, Da Costa \& Demarque
1985), it became the crossroad of several theoretical and observational
investigations (Alves \& Sarajedini 1999).
The detection of the RGB Bump was mainly hampered by
the size of the available samples of RGB stars. This is particularly
true for the most metal-poor clusters, where the Bump moves toward
brighter magnitudes and therefore less populated RGB regions.
Only recently was this feature firmly detected in a large set of both
Galactic (FP; Brocato et al. 1996) and extragalactic stellar systems
(Desidera 1999).
The analysis by FP showed that the \vhbb\footnote{The
parameter \vhbb is commonly defined as the difference in magnitude
between the Red Giant Branch (RGB) Bump and the Horizontal Branch
(HB) stars located within the RR Lyrae instability strip (see for
a detailed discussion Fusi Pecci et al. 1990, hereinafter FP, and
Cassisi \& Salaris 1997, hereinafter CS).}
values predicted by theory
were 0.4 mag brighter than the empirical estimates
in a sample of 11 GGCs. This observable, which does not depend on
the distance modulus, on the reddening, or on the calibration of
photometric data, is a key parameter for assessing both the accuracy
and the plausibility of the physical assumptions adopted for
constructing the evolutionary models. As a consequence the discrepancy
found by FP needs to be understood.
In order to explain this mismatch between theory and observations
Alongi et al. (1991) suggested the inclusion of convective
undershooting at the boundary between the convective envelope and the
thin H-burning shell, whereas Straniero, Chieffi \& Salaris (1992)
called attention on the role that the global metallicity could play on
the estimate of this observable. The latter hypothesis was confirmed
by CS who found that standard models agree quite well
with the observed \vhbb values in 8 GGCs for which high resolution
spectroscopic determinations of both [Fe/H] and [$\alpha$/Fe] were
available. The significance of this result was limited by the small
number of clusters taken into account and by the small metallicity
range that they cover.
The main aims of this investigation are to provide new homogeneous
measurements of both the Bump position and the corresponding \vhbb value
in a large sample of GGCs covering a wide metallicity range
($-2.1 \le $ [Fe/H] $ \le -0.2$); and also to compare the new
measurements with the predictions of standard evolutionary models.
At present, the most homogeneous photometric sample comes from HST.
Two of us (GP and MZ) are involved in a program aimed at
collecting F439W and F555W WFPC2 images for all the GGC with
($m-M$)$_B<18.0$ not yet observed with HST. Ten of these
clusters have been observed during Cycle 6 (GO6095) and nine more
during the ongoing Cycle 7 (GP7470). Similar observations for nine
additional clusters are available on the HST archive.
In \S 2 we present this sample of 28 GGCs, and discuss the data reduction
strategy as well as the approach adopted for estimating both the
zero age horizontal branch (ZAHB) and the RGB Bump visual magnitudes.
Section 3 deals with the comparison between theory and observations,
while in \S 4 we briefly summarize the main results.
\section{Data Selection and Reduction}
The center of all the clusters were observed in the HST B (F439W)
and V (F555W) bands with the WFPC2. Both very short (a few seconds)
and long (a few hundreds of seconds) exposures are available for both
filters allowing us to map the evolved regions of the color-magnitude
diagram (CMD). The pre-processing, photometric reduction, and
calibration to a standard B and V system of the data for each cluster
were carried out following the same procedure as described in details
in Piotto et al.\ (1999a) for a subset of images of the GO7470. The
photometry for all the clusters extends from the tip of the RGB down
to about 2 magnitudes below the turnoff, and include a number of stars
ranging from $\sim 5000$ for the loosest cluster to $\sim 30000$ for
the most concentrated ones. A detailed description of the CMDs will be
presented in Piotto et al. (1999b).
The large sample of stars, coupled with the high photometric accuracy,
allows us to identify of the RGB Bump even in the most metal-poor clusters.
Figure 1 shows an example of the Bump identification in NGC5824, a
metal-poor cluster in our database.
The upper left panel shows the RGB differential LF. We constructed a
histogram with a fixed bin size of 0.15 magnitude ({\it dotted line})
as well as the multibin histogram ({\it solid line}) described in
Piotto et al. (1999c). The Bump is marked by a vertical arrow.
As a further check of the accuracy in the Bump identification, the
lower left panel shows the RGB cumulative LF. The Bump can be easily
located since it is marked by a change in the slope. The right panel
shows the CMD of NGC5824 and the arrow marks the position of the Bump
along the RGB.
The determination of the ZAHB magnitude ($V_{\rm ZAHB}$)
is a thorny problem, in particular for those clusters which show blue
HBs. A standard method to determine the ZAHB magnitude for both
intermediate and metal-poor clusters is to adopt the mean magnitude of
the cluster RR Lyrae stars. Unfortunately, we could not adopt this
method directly, since our photometry covers a very short time interval, and
therefore the RR Lyrae were always measured at random pulsation phases
(Piotto et al. 1999a).
In order to overcome this problem we have undertaken a different
approach. For metal-poor and intermediate metallicity clusters we
selected from the literature three clusters which offer accurate
photometry, a large number of RR Lyrae stars, and a well defined blue
HB tail. The template clusters are the following: NGC1851
(Walker 1998) for clusters in the metallicity range
$-1.5<$[Fe/H]$<-1.0$, NGC5272 (M3, Buonanno et al. 1994) for clusters
in the range $-1.7<$[Fe/H]$<-1.5$, and NGC4590 (M68, Walker 1994) for
more metal-poor clusters. The CMDs of the template clusters were
artificially shifted in color and in magnitude in order to match both
the RGB and the blue HB tail of each cluster. The mean RR Lyrae magnitude
of the template cluster was scaled according to the magnitude shift
adopted to overlap the stellar distributions
along the HB. The $V_{\rm ZAHB}$ magnitudes were estimated by
using the relation between the mean RR Lyrae magnitude and the
$V_{\rm ZAHB}$ magnitude suggested by CS. It is worth noting that this
method is totally independent from any zero point difference between the
template photometry and ours.
For the most metal-rich clusters ([Fe/H]$>-1$) with well populated red
HBs but no RR Lyraes, first we estimated the photometric error
$\sigma^2_V=\sigma^2_{(B-V)}/2$, where $\sigma^2_{(B-V)}$ is the
standard deviation of the color distribution of the RGB stars at the
level of the HB. Then we fixed $V_{\rm ZAHB}$ at $3 \sigma_V$ magnitudes
above the lower envelope of HB stellar distribution. However, in order
to provide a consistent determination of this parameter over the whole
GGCs sample, the $V_{\rm ZAHB}$ at the level of RR Lyrae instability strip
was evaluated according to the method suggested by Fullton et al. (1995).
Figure 1 shows the fit of the same metal-poor
cluster NGC5824, (full dots) with the template cluster
M68 (crosses). A vertical and a horizontal shift of 2.83 and 0.08
magnitudes were applied to M68 for matching both the RGB and
the HB. The M68 RR Lyrae stars are plotted as open squares.
Taking into account the uncertainties in the fit, we estimated
an error in $V_{\rm ZAHB}$ of the order of 0.1 mag.
Table 1 summarizes the cluster observables: column (1) gives
the NGC and the Messier numbers; column (2) lists the visual magnitudes
of the RGB Bump and the photometric error;
column (3) gives the mean RR Lyrae visual magnitudes estimated according
to the method previously described;
column (4) lists the ZAHB visual magnitude at the lower envelope of red HB
stars in metal-rich clusters; column (5) gives the cluster metallicities
according to the Carretta \& Gratton (1997, hereinafter CG)
scale\footnote{For those clusters whose metallicity was not provided by
CG, the cluster metallicities collected by Harris (1996) were transformed
into the CG scale by adopting the new calibration provided by Cohen
et al. (1999, hereinafter CGBC). In contrast with CG the new scale applies
also to metal-rich clusters ($-2.12 \le$ [Fe/H] $ \le -0.3$).}.
\section{Comparison Between Theory and Observations}
Figure 2 shows the comparison between theory and observations in the
[M/H]-\vhbb plane. The global metallicities were estimated
by adopting a mean $\alpha$ enhancement equal to 0.3 for clusters with
[Fe/H] $ < -1.0$ and to 0.20 for more metal-rich clusters. The former
value was suggested by Carney (1996, hereinafter C96), while
the latter, due to the paucity of data available in the literature,
is a mean between the estimates collected by C96 and by Salaris \&
Cassisi (1996, hereinafter SC).
The observed \vhbb values have been plotted in the top panel according to
the CG metallicity scale.
The \vhbb error bars have been calculated by quadratically combining
the errors on $V_{\rm ZAHB}$ and on $V_{\rm Bump}$. The global metallicity
error bars are a lower limit of the uncertainties affecting both [Fe/H] and
[$\alpha$/Fe] measurements (see C96 and
Rutledge, Hesser, \& Stetson 1998, hereinafter RHS).
The theoretical predictions plotted in the top panel were estimated by
adopting progenitor masses ranging from $M/M_{\odot}\,$=0.8 to 1.0 and a wide
range of global metallicities ($-2.3 \le$ [M/H]$ \le 0.0$). The initial
helium contents adopted in constructing evolutionary models are the
following: $Y=0.23$ for [M/H]$\le -0.5$, $Y=0.255$ for [M/H]$=-0.25$,
and $Y=0.289$ for [M/H]$=0.0$. Metal-poor \vhbb theoretical estimates
up to [M/H]$\approx -0.5$ already been presented in CS,
whereas more metal-rich ones were specifically computed,
to extend the predictions to clusters more metal-rich than 47 Tuc.
Basic assumptions on the input physics adopted for constructing
evolutionary models were extensively described in CS and
Bono et al. (1997 and references therein) and therefore they are
not discussed here. The reader interested in a detailed
discussion on the dependence of \vhbb on mixing-length,
He content, and element diffusion is referred to CS and to
Cassisi, Degl'Innocenti, \& Salaris (1997, hereinafter CDS).
Bolometric magnitudes were transformed into V magnitudes by
adopting bolometric corrections provided by Castelli, Gratton \&
Kurucz (1997).
In order to account for the $V_{\rm Bump}$ dependence on cluster age
as suggested by CS and more recently by Alves \& Sarajedini (1999)
in a detailed investigation on HST data of 8 SMC clusters, we plotted
the theoretical predictions for three different ages: 12 (short-dash line),
14 (solid line), and 16 (long-dash line) Gyr. In the last few years a
large number of theoretical and observational investigations have been
devoted to the absolute and the relative ages of GGCs as well as to the
errors affecting such parameters. In fact, they depend on the physical
assumptions and on the input physics (Cassisi et al. 1998; Vandenberg,
Stetson \& Bolte 1996) adopted for constructing evolutionary models.
As plausible assumptions we adopted an average cluster age of 14 Gyr
(Vandenberg 1999) and an average uncertainty of $\pm2$ Gyr.
The change in the slope toward higher metal contents shown by theoretical
predictions is due to the fact that at fixed age an increase in the evolving
mass causes a smoother increase in the core-mass luminosity relation,
and in turn in the $V_{\rm Bump}$ magnitude.
In fact, for metallicities ranging from Z=0.001 to Z=0.006 the
$M_V({\rm Bump})$ and the $M_V({\rm ZAHB})$ magnitudes changes
according to the following derivatives:
$\partial{M_V({\rm ZAHB})}/\partial{\rm [M/H]}\approx 0.093$
and $\partial{M_V({\rm Bump})}/\partial{\rm [M/H]}\approx 0.753$,
whereas for $0.006 \le Z \le 0.02$ they change according to
$\approx0.289$ and $\approx1.251$ respectively.
The data plotted in the top panel of Figure 2 show clearly that the
discrepancy of 0.4 magnitudes suggested by FP is completely removed
over the entire metallicity range, and indeed the trend of empirical
data is well reproduced by standard models.
This result is even more compelling if we take into account that the
previous comparisons (FP and CS) were hampered by the small number of
metal-poor clusters ([Fe/H] $< -1.5$) for which reliable estimates of
$V_{\rm Bump}$ were available.
Figure 2 also extends the comparison to [Fe/H]$ = -0.2$; note how the
observed values in the high metallicity range show the flattening predicted
by the theory, though with a distribution that is somehow flatter.
A plausible change in the He content (CS) and/or the inclusion of
element diffusion (CDS) can account for this effect only marginally.
As a consequence, this result suggests that the metal-rich clusters
could be younger than the bulk of our clusters (Salaris \& Weiss 1998;
Rosenberg et al. 1999).
In the cluster sample adopted in this investigation there are only
two clusters -NGC7078 and NGC5694- which are marginally in agreement
with the theoretical expectations. For both clusters the uncertainty
on the location of the Bump is very small (Table 1). In the case of
NGC7078, the adopted mean $V_{\rm RR}$ magnitude is also in very good
agreement with the value obtained by Silberman \& Smith (1995).
Therefore for this cluster the discrepance could be due to an
underestimate of the cluster metallicity and/or of the $\alpha$
enhancement (see e.g. CG). An independent estimate of $V_{RR}$ for
NGC5694 is not available, but any plausible assumption on its
uncertainty can hardly remove such a discrepancy.
In order to account for the uncertainty on the metallicity scale
the middle panel shows the same comparison as the top panel, but
the observed points are plotted according to the Zinn \& West (1984,
hereinafter ZW) metallicity scale. The data plotted in this panel show
that the \vhbb values are systematically shifted toward lower metal
contents when compared with theoretical observables.
At present, both systematic and observational errors affecting
the metallicity ranking of GGCs are still controversial issues (RHS; C96).
This notwithstanding, the CG scale is more robust since it relies on
recent high dispersion spectroscopic measurements and up-to-date
atmosphere models.
Even though current observational uncertainties affects the global
metallicity of individual clusters, data plotted in Figure 2 support
the evidence that the ZW scale underestimates the cluster metallicity
in the range $-1.7<$[Fe/H]$<-1.0$. In order to supply a quantitative
estimate of the difference we performed a fit of the empirical
data ($-2.0<$[Fe/H]$<-0.5$) with predictions at 14 Gyr. The standard
deviation is 0.05 mag for the CGBC scale and 0.17 mag for the ZW scale.
The latter value is almost a factor of two larger than the photometric
uncertainty.
\section{Conclusions}
We have presented new homogeneous measurements of the
\vhbb values for a sample of 28 GGCs observed with HST,
and a detailed comparison with the theoretical models. By relying on
homogeneous theoretical and observational frameworks and on the
metallicity scale suggested by CG and by CGBC we found
that, within current uncertainties, observables predicted by standard
H- and He-burning evolutionary models agree with the empirical
data. This result is further strengthened by the fact that this
comparison was extended from metal-poor to metal-rich clusters
($-2.1\le$ [Fe/H] $\le-0.2$).
New theoretical predictions for metal-rich clusters show a change of
the slope of the \vhbb -[M/H] relation. This behavior is supported
by the tail of metal-rich clusters in our sample and does not depend
on the adopted metallicity scale. Leading physical arguments on the
dependence of \vhbb on input physics support the suggestion that
metal-rich cluster could be younger than the bulk of clusters
in our sample.
By adopting the ZW metallicity scale we found that empirical data at
low and intermediate metallicity are shifted toward lower metallicities
when compared with theory.
At the same time the agreement between theory and observations supports
the use of a \vhbb - metallicity relation for constraining the
cluster metallicity (Desidera 1999). However, we note that such a
relation relies on the assumption that all GGCs are coeval within $\pm1$~Gyr
(Stetson et al. 1999), and that the intrinsic accuracy is of the order
of $0.15$~dex provided that the \vhbb values are measured with an
accuracy of 0.10~mag.
We are deeply indebted to E. Carretta and R. Gratton for providing
us with the extension of the CG metallicity scale to [Fe/H]$=-0.3$
in advance of publication.
|
\section{Introduction}
Galactic chemical evolution is the proportional buildup
of helium and heavy elements or metals, i.e. elements other than hydrogen
and helium, within a galaxy over time as a result of the
continuous manufacture and expulsion of these elements by
resident stars. The topic
concerns itself not only with global or pan-galactic changes but
also with regional ones within spatially-resolved galaxies.
The essence of chemical
evolution can be illustrated by imagining a closed box containing
interstellar gas of primordial composition situated at an
arbitrary location within a galaxy. As portions of the gas collapse, fusion
processes within the stars that are formed convert hydrogen into
heavier elements, and this chemically enriched material is
subsequently expelled into the interstellar medium through
stellar winds, planetary nebula formation, or supernova
eruptions. As this cast-off matter mixes with the surrounding
medium, the composition of the latter changes such that the
abundances of helium and the heavy elements increase relative to hydrogen.
As a second generation of stars forms from this
enriched material, the new stars possess a greater fraction of
heavy elements than their predecessors. Thus, the enrichment cycle
continues until enough material has been locked up in
stellar remnants that the star formation process, which depends
upon the availability of interstellar gas, is finally damped.
A galaxy, then, can be thought of as an ensemble of these boxes.
The real picture is more complicated, of course. The boxes have
no walls, and as such are open to matter exchange with their
surroundings in all directions.
Nevertheless, the simple model does suggest some of the crucial processes
which must be understood if we are to have a comprehensive
understanding. For example, we must know the details of star
formation and evolution, stellar nucleosynthesis and the rate of
heavy element production, the details of stellar death and matter
ejection, and the efficiency with which ejected material is mixed
into the interstellar medium.
Studies of galactic chemical evolution involve an interplay
between 1)~global and/or spatially resolved abundances, sometimes
as a function of time, for one or more galaxies; and 2)~models based
upon a set of input parameters determined by the physics being
tested. Observed abundances provide two basic sorts of
information. First,
ratios of heavy elements relative to hydrogen, such
as O/H or Fe/H,
serve as gauges of how far chemical evolution has progressed in a
system, because they
measure the extent to which hydrogen has been converted to
heavier elements. As such, these ratios are particularly
sensitive to the rate at which gas is cycled through stars, i.e. the
star formation rate, and how that rate may have changed with
time. Second,
ratios of two heavy elements, such as N/O or O/Fe, provide information
about differential elemental production by stars. That is, at
what rate, say, is nitrogen produced relative to oxygen, or
oxygen relative to iron? The answer here is tied to the
production rates of individual elements as a function of stellar
mass weighted by the relative number of stars at each mass (the initial mass function; IMF) as
well as to the history of star formation.
Also, there is an element of time involved in all of
this. For example, abundances measured in a star
reveal enrichment levels at
the time the star formed. In summary, chemical evolution can be
traced indirectly by associating abundance patterns within a galaxy with local conditions, where the latter ultimately depend on time, and directly
by observing abundances in stars of different ages or in galaxies of
different look-back times.
The primary goal of this review is to describe the state of affairs
concerning observed abundance patterns in
galaxies. Because of author
expertise, emphasis is placed on abundance patterns in spiral
disks and elliptical galaxies as derived from emission line
analyses and photometric
indices, respectively. However, for completeness and continuity,
we also describe and compare the complimentary results
provided by stellar abundance work in the Milky Way and nearby
galaxies.
Our elemental scope is confined to those elements between carbon
and iron on the periodic table (6$\le$Z$\le$26),
i.e. those elements which are the
most readily observed and for which there is the most
information.
Discussions of helium and the light elements are better
taken up in the context of Big Bang nucleosynthesis, and for
this the reader is urged to consult Chapter~4 of Pagel (1997)
and references therein for recent discussions of this topic.
Likewise, elements beyond iron have been studied in part by
Edvardsson et al. (1993), Wheeler, Sneden, \& Truran (1989), and
McWilliam (1997).
Numerous reviews of galactic chemical evolution
and abundance patterns are available in the literature.
An excellent, approachable introduction to the subject of chemical
evolution is given in the comprehensive review by Tinsley (1980).
The textbook by Pagel (1997) treats numerous topics
related to galactic chemical evolution
and the synthesis of elements. Additional
material on observations and abundance studies in galaxies can
be found in several recent conference proceedings, in particular
Friedli et al. (1998) and Walsh \& Rosa (1999). Other useful works specifically
treating element synthesis include books by Clayton (1983), Rolfs \&
Rodney (1988) and Cowley (1995), the review by Trimble (1991),
and the conference
proceedings by Edmunds \& Terlevich (1992) and Prantzos,
Vangioni-Flam, \& Cass{\'e} (1993). Finally, QSO absorption line systems are enabling chemical evolution studies to be carried out through the study of abundances as a function of look-back time. While these systems are beyond our scope, interested readers are urged to consult Lauroesch et al. (1996), Lu et al. (1996), and Pettini et al. (1999).
We begin with a discussion of abundances derived from emission lines
in spiral galaxies, including the Milky Way, in {\S}2. In {\S}3 we
turn to stars both inside and outside of the Milky way, while abundances in
elliptical galaxies from photometric integrated light are treated in
{\S}4. A summary is given in {\S}5. Appendices explain techniques used
to derive abundances from emission lines (Appendix A), stellar absorption
lines (Appendix B), and the integrated starlight of composite systems
(Appendix C). Unless otherwise stated, elemental abundances and ratios referred to in this review are by number, not mass.
\section{Abundance Patterns In Spiral Galaxies From Emission-Line Objects}
Sampling a galaxy's interstellar medium directly provides a snapshot of
the {\it current} abundance picture at the location being tested, in contrast to stellar abundances which for the most part are indicative of interstellar abundances at the time that the star formed.
The most straightforward way of determining interstellar
abundances is through the analysis of emission
spectra produced by gas heated by nearby hot stars with
continua rich in photons having wavelengths shortward of
912{\AA}, i.e. the ionization edge of hydrogen. Such stars have
effective temperatures exceeding 30,000K and spectroscopically
belong to the O and early B classes. Object types with these
conditions include H~II regions and planetary nebulae, reviews
of which can be found
in Shields (1990), Vila-Costas \& Edmunds (1992), and Zaritsky,
Kennicutt, \& Huchra (1994) for H~II regions and Peimbert (1990),
Henry (1990), Perinotto (1991), Clegg (1993), and
Habing \& Lamers (1997) for planetary nebulae.
Old supernova remnants in which the stellar ejecta have
completely mixed with the interstellar medium in principle
represent a third type of emission line probe, since in this
case the interstellar gas is heated by the shockwave producing
emission lines.
Often, however, full abundance studies are precluded by limited
spectral coverage even within the optical (W.P. Blair, private communication),
and thus there are far fewer
abundance results available.
Ionized gases of the types just mentioned maintain temperature
equilibrium in most cases by radiating photons at discrete wavelengths
following
recombination or collisional excitation processes involving
ion-electron encounters. Spectra of
these objects can then be analyzed to provide abundance,
temperature, and density information.
Measured from the ground in most cases, the strengths
of the resulting emission lines can be converted to
ionic and elemental abundances of He, C, N, O, Ne, S, and Ar
using techniques described in Appendix~A, which includes a table
listing a number of prominent emission features.
While the resulting
abundances refer to levels in the gas phase only, Savage \&
Sembach (1996) indicate that none of these elements is
expected to be heavily partitioned into the solid phase in the
form of dust. Thus, gas phase abundances should represent total
values reasonably well.
Due to their size and therefore their accessibility in external
galaxies, most of the abundance data from emission-line systems
useful in chemical evolution studies relate to H~II
regions, which because of their association with recent star
formation are located in spiral
disks and irregular galaxies. This section focuses on
abundance patterns in spirals.
It should be noted that abundances discussed are
taken directly from the sources listed; no
attempt has been made to homogenize them by recalculating the
abundances in a consistent way. In general, differences
in techniques and atomic data employed produce ranges in
abundances which are smaller than observational uncertainties in
the line strengths, and we believe that presenting unhomogenized
data still provides a realistic representation of patterns and
an opportunity to see the big picture.
\subsection{Metallicity Gradients In Spiral Disks}
Metallicity is the fraction by mass of all elements heavier than He in a
system and is the primary indicator of chemical
evolution as stars convert H into heavier elements and seed their
environments with the products.
Oxygen is the metallicity
tracer of choice in the interstellar medium. Cosmically, its
relative abundance surpasses all elements but H and He. Its
relatively small depletion (Snow \& Witt 1996; Savage \& Sembach
1996) means it is
present almost entirely in the gas phase.
Hot, ionized gas in the vicinity of hot stars or energetic shock
waves give rise to H~II regions, planetary nebulae, or supernova
remnants whose spectra usually display prominent emission lines
of oxygen. This contrasts sharply with the situation for old
stars, for example, where absorption features of iron (heavily
depleted onto grains in the interstellar medium) are prominent in stellar
spectra due to the presence of optimal temperatures. Thus, iron
is usually employed as a metallicity indicator when old stars are
the probes.
Differences in the appearances of H~II region spectra as a function of
galactocentric distance were first noticed by Aller (1942) in his study
of M33. Thirty years later, Searle (1971) connected similar differences
across disks of several Sc galaxies with systematic changes in heavy element
abundances. Early abundance gradient work in spirals is reviewed by
Pagel \& Edmunds (1981), while the more contemporary picture is
available in Friedli et al. (1998).
\subsubsection{Metallicity Gradients In The Milky Way Galaxy}
The disk
of the Milky Way Galaxy (MWG) is arguably the most active and rapidly evolving
region of our galaxy in the chemical sense. Probing it, though,
is complicated by the presence of dust along all lines of sight
within the disk, preventing radiation, particularly in the
ultraviolet, from readily penetrating it. From our location within
the disk, observing emission line objects is restricted not only
from effects of reddening, but the presence of dust limits the
distance over which we can probe. Despite these restrictions,
large amounts of data are now available for H~II regions and planetary
nebulae within the disk of the MWG. The distance range from the
sun in both the directions of the galactic center and anticenter
has been extended by observations in the infrared, where
extinction is at a minimum.
Table 1 summarizes the data for emission line objects
in the disk of the MWG and compiled here. The columns in
order indicate the type of object and spectral region (optical,
far infrared, or radio) followed by the name of the
first author on the paper for the data source, the total number of and
galactocentric distance range for objects included in each study,
and finally an indication of the number of data points for each
abundance ratio which were available in each study.
Six of the eight studies in Table~1 are based upon H~II regions.
The seminal study by Shaver et al. (1983) explored a range in
galactocentric distance centered on the sun. More recent
studies by Fich \& Silkey (1991), V{\'i}lchez \&
Esteban (1996), and Rudolph et al. (1997)
focused on the anti-center direction, while Simpson
et al. (1995) and Afflerbach et al. (1997) studied objects
toward the center of the Galaxy. Although the optical studies were
all ground-based, the far infrared work in all three cases was
carried out using the Cryogenic Grating Spectrometer aboard the
Kuiper Airborne Observatory to observe emission lines such as
[N~III] 57$\mu$m, [O~III] 52,88 $\mu$m, and [S~III] 19,33 $\mu$m.
The remaining two papers in Table~1 were based on studies of
planetary nebulae (Maciel \& K{\"o}ppen 1994) and
supernova remnants (Fesen, Blair, \& Kirshner 1985). Planetary
nebulae comprise ejected material from evolved intermediate mass
stars, and thus generally have an abundance profile which differs somewhat from
that of the progenitor star at the time of birth due to contamination of the nebula by products of stellar nucleosynthesis. However, the contamination affects primarily
helium, carbon, and nitrogen, and thus abundances of oxygen,
sulfur, and argon in the nebula are expected to be good measures
of the levels of those elements in the nascent progenitor star. In particular,
Type~II planetary nebulae (Peimbert 1978) have progenitors of
small enough mass that oxygen is not expected to have been
altered by CNO processing, yet they are disk objects, based upon
their kinematics. Finally, the study of old supernova remnants by Fesen et
al. measured abundances in disk objects over a radial range
similar to that of Shaver et al.
Figure~1 plots
O/H versus galactocentric distance in kiloparcsecs by author,
where 12+log(O/H) is used to represent oxygen\footnote{H~II region distances in Shaver et al.'s data
have been recomputed using their reported radial velocities and
longitudes, eq.~9.3 in Binney \& Merrifield (1998), and
assumptions that $R_{\sun}=8.5$kpc and the average circular
velocity is 240~km/s across the relevant portion of the disk.}.
Representative
error bars are shown in the lower left panel.
Solid lines show first order least squares
fits for which the fitting parameters are
given in Table~2, with the gradient G expressed
in dex/kpc, the absolute abundance A$_{8.5}$ given as 12+log(O/H) at the
solar circle (8.5~kpc), and c is the correlation coefficient. Dot-dashed lines show the composite fit from Table~2 for reference.
Note that reported upper and lower limits on abundance ratios were
generally not used in our compilation. We also point out that the larger
scatter in the Simpson et al. data is probably due in large part
to inferring O/H from O$^{+2}$/S$^{+2}$ observations plus an
assumption of a constant S/O ratio, while the scatter in the Fesen et
al. results for supernova remnants may be due to their using abundance-line strength diagnostic diagrams taken from shock models in the literature for estimating abundances.
Combining several data sets allows a visual comparison to be
made among them as well as a test of the robustness of the trend
exhibited by a single set. Figure 2 shows
12+log(O/H) versus galactocentric distance in kpc for the nebular
data sets of Shaver et al., Afflerbach et al., Maciel \&
K{\"o}ppen, V{\'i}lchez \& Esteban, Fich \& Silkey,
Rudolph et al., and Fesen et al. We also present the B
star results from Smartt \& Rolleston (1997; filled circles) and Gummersbach et al. (1998; open circles). The
sun's position (Grevesse \& Noels 1993) is indicated with an `x',
while the error bars in the lower left show typical
observational uncertainties for all of the data. A monotonic
decrease in oxygen abundance with galactocentric distance is
clearly present in the Milky Way disk. A simple least squares fit to
all points except the B~stars indicates a slope of
-0.06($\pm$0.01), A$_{8.5}$ of 8.68 ($\pm$0.05), and a
correlation coefficient of -0.63. The data from Afflerbach et al.
and V{\'i}lchez \& Esteban extend the trend of the main body of
data toward the galactic center and anti-center, respectively.
The uncertainty of $\pm$0.2 dex in oxygen abundance is consistent
with observational uncertainty, and thus there is no indication
of real abundance scatter at a constant radial distance, in
accord with findings of Kennicutt \& Garnett (1996) in their
study of M101.
Several additional points are illustrated in Fig.~2. First, notice that
the B star oxygen abundance trend is not noticeably different from the one defined by nebular
data; in fact their gradients are very similar to the nebular results.
This represents a major development, as previous attempts to infer the
disk O/H distribution from B stars (Gehren et al. 1985; Fitzsimmons et
al. 1992; Kilian-Montenbruck et al. 1994; Kaufer et al. 1994)
indicated the absence of a gradient. Smartt \& Rolleston speculate that
sample size was the culprit in obscuring the gradient in most of the
previous studies.
Next, the oxygen abundance distribution implied by planetary nebulae is
indistinguishable from the one from H~II regions. This would seem to
confirm the value of PNe to trace disk metallicity and at the same time
perhaps reduce the concern about diffusion (Wielen et al. 1996), i.e.
that positions of PN progenitors shift radially during their lives, and
thus PN abundances do not represent ISM conditions at their present
galactocentric distances.
An interesting wrinkle in the MWG metallicity
gradient picture is the possibility that the gradient flattens
beyond 10~kpc. Results for the three anti-center studies are presented in a
single panel in Fig.~1. V{\'i}lchez \& Esteban conclude that their data
support the presence of a flattened oxygen gradient in the outer
galaxy, although results from Fich \& Silkey and Rudolph et al.
neither support nor counter this claim. Recently,
Maciel \& Quireza (1999) have expanded and updated their sample
to include PNe with larger galactocentric distances than those
presented here. Like V{\'i}lchez \& Esteban above, they find
evidence for a gradient which flattens beyond 12~kpc. A
flattened gradient is both a controversial and interesting conjecture
and is tied to the
dynamics and mass distributions in the disk (Zaritsky 1992;
Moll{\'a} et al. 1996; Samland, Hensler, \& Theis 1997), and we
briefly return to this point in {\S}2.4 in our general discussion of
spiral abundance gradients.
Finally, we have plotted predictions of four chemical evolution models
of the present-day disk for comparison with the data. The dashed line
shows an analytical result based on the ``simple model'' from Pagel
(1997; eq.~8.14), while detailed numerical model results are shown from
Samland, Hensler, \& Theis (1997; solid line), Ferrini et al. (1994;
dot-dashed line), and K{\"o}ppen (1994; long-dashed line), where
K{\"o}ppen (private communication) employed a quadratic star formation
law but recalculated his model for a radial flow velocity of
0.3km/s\footnote{The unpublished K{\"o}ppen model also assumes a disk age of
15~Gyr along with exponentially decreasing infall both in time (5~Gyr
scale) and galactocentric distance (4 kpc scale)}. All models are
scaled so as to match our composite interstellar oxygen abundance of
8.68 at the solar circle (see Table~2). Note that the Samland et al.
model predicts a gradient flattening outward from around 11~kpc, the
result of (according to them) mass loss of long-living metal-poor
intermediate mass stars and additional infall of low metallicity gas in
equilibrium with metal enrichment from condensation of intercloud
medium.
\subsubsection{Metallicity Gradients In External Galaxies}
Results from numerous surveys of spiral galaxy abundance patterns
show clearly
that most spiral disks possess negative gradients qualitatively
similar to the one in the Milky Way. Large surveys of O/H in
extragalactic H~II regions include those of Mc~Call (1982; 40
galaxies; see also Mc~Call, Rybski, \& Shields 1985), Vila-Costas \& Edmunds (1992; 32 galaxies), and
Zaritsky, Kennicutt, \& Huchra (1994; 39 galaxies). We can add
to those the recent
studies by Ferguson, Gallagher, \& Wyse (1998) and
van~Zee et al. (1998), both of which explored the outer
regions of spirals, where star formation rates are much lower and
the regions are less advanced chemically. Vila-Costas \& Edmunds
reprocess reduced line strengths from the literature
to obtain their abundances, while the other authors use
primarily their own data for their studies. All these studies
are based on optical spectra.
As an example of abundance patterns in two external spirals, in Fig.~3
we present a comparison of results for NGC~628 and M33 from Zaritsky
et al. (1994) with Milky Way data from Shaver et al. (1983),
Afflerbach et al. (1997), and V{\'i}lchez \& Esteban (1996), where
12+log(O/H) is plotted against galactocentric distance. The latter
quantity has been normalized to the respective galaxy's isophotal
radius R$_o$\footnote{The isophotal radius R$_o$ is the radial distance from
the galactic nucleus at which the declining disk surface brightness
reaches 25 mag/arcsec$^2$. Comparisons of data among galaxies are made
most frequently using the isophotal radius, but one could also use the
effective radius, i.e. the radius of an aperture admitting one-half of the
light from the disk, or kiloparsecs. To add to the confusion,
literature sources for (non-nebular) bulge or elliptical galaxy
abundances usually express gradients as
$\Delta$log~$Z$/$\Delta$log~$R$! This last notation gives nonsensical
abundances at the nucleus of a galaxy, but seems to match observed
profiles out to the observational limit, which is usually far short of the isophotal radius for integrated starlight spectroscopy.} to account for size
variations among galaxies. R$_o$ for the Milky Way disk was taken
from de~Vaucouleurs \& Pence (1978). Gradient slopes determined from
least squares fits are given in the figure legend. The two external
galaxies clearly resemble the Milky Way in possessing negative
abundance grandients.
A much larger collection of abundance plots for individual spirals can be
found in Zaritsky et al. (1994).
We have extracted results from that paper
and plotted them in Figure~4a, where characteristic
abundances\footnote{The characteristic abundance is the
abundance at 0.4R$_o$ as determined by a least squares fit to the
data. See Zaritsky et al. (1994).}
(top panels) and gradient slopes in dex/R$_o$ (bottom panels)
are shown as functions of galaxy
morphological type (T type), absolute blue magnitude M$_B$, and
maximum circular velocity V$_c$ in km/s. All three of these
parameters track galaxy mass, where
smaller T type indices, more luminous integrated blue magnitudes,
and larger rotational velocities generally correspond with more
massive spirals. Normal
(SA) and barred (SB) spirals are shown separately using filled and
open symbols, respectively. Abundance parameters for the
Milky Way composite fit from Table~2 are indicated in Fig.~4a with plusses,
where we have adopted T=4, M$_B$=-20.08, and R$_o$=11.5kpc
(de~Vaucouleurs \& Pence 1978), along with V$_c$=220~km/s
(Kochanek 1996).
Two important points are implied by Fig.~4a: (1)~Characteristic
abundances increase with galaxy mass, while gradient slopes are
uncorrelated with this parameter; and (2)~Characteristic
abundances in normal and barred spirals are indistinguishable, but
barred spirals appear to have flatter (less negative) gradients.
Both of these results have been noted previously. Garnett \&
Shields (1987) plotted characteristic O/H values against galaxy
mass for numerous spirals and found a direct correlation between
these two parameters, while Pagel et al. (1979) first suggested
that barred spirals may have flatter gradients, a pattern clearly
borne out in the more extensive work by Martin \& Roy (1994),
who
relate gradient slope to bar strength, a quantity which measures
bar ellipticity. Martin \& Roy find direct relations between the slope of
the oxygen abundance gradient of a barred spiral and the galaxy's
bar strength (ellipticity) and length in the sense that stronger bars are accompanied by flatter gradients. This empirical result is
consistent with radial flow models of chemical evolution in which
the presence of a bar enhances large-scale mixing over the
galaxy's disk, damping radial abundance variations.
Interestingly, if gradient slope in dex/kpc (as opposed to
dex/R$_o$ shown here) is plotted versus M$_B$ (see Garnett 1998)
a correlation appears such that more luminous galaxies have
flatter slopes. The dependence of slope behavior on normalization
is no doubt related to the fact that R$_o$ for luminous galaxies
tends to be longer in kiloparsecs. Since the vertical scatter in
the lower panel of Fig.~4a is comparable to observational
uncertainties, this may imply a universal gradient in dex/R$_o$, which
in turn could be associated with similar timescales for viscous
angular momentum transport and star formation, producing
exponential gradients in surface density and abundances (Lin \& Pringle 1987; Yoshii \& Sommer-Larsen 1989).
We illustrate explicitly the correlation between galaxy mass and
characteristic abundance
in Fig.~4b, where we plot 12+log(O/H) at one effective radius versus the
log of the galaxy mass in solar units for a sample of spiral
galaxies. Abundance data are from Garnett \& Shields (1987),
Skillman et al. (1996), and Henry et al. (1996). Sources for
galaxy masses, which for the most part are inferred from rotation curves, are given in Henry et al. The two points connected by a horizontal
line are for NGC~753 whose mass was determined for H$_o$ values
of 50 and 100 km/s. The least squares fit to the data, shown with a solid line, indicates that $12+log(O/H)=3.79+0.47 \times logM$, where $M$ is in solar masses. Note that this relation ignores the low surface brightness spirals (McGaugh 1994) which appear to have low metallicity but high mass. These objects are discussed briefly in {\S}2.1.3.
Finally, Figure~5 shows the observed relation between 12+log(O/H) and total (disk + bulge) surface density, $\Sigma$, in M$_{\sun}$/pc$^2$ at the corresponding location from
Vila-Costas \& Edmunds (1992; their Fig.~7d), where open and
closed squares represent H~II regions residing in late (Scd-Irr)
and early (Sab-Sc) spirals, respectively. Vila-Costas \& Edmunds assumed that the mass distribution follows the light distribution and employed a constant mass-to-light ratio for each galaxy, the latter determined from a rotation curve (see Vila-Costas \& Edmunds and Mc~Call 1982 for details).
The scatter is
consistent with observational uncertainty, and thus we see a
clear positive correlation between abundance and local surface density
in spirals, with earlier spirals generally possessing higher abundances per unit
surface density.
\subsubsection{Assorted Issues About Galaxy Metallicity}
Other issues concerning abundance
gradients include questions about gradients
perpendicular to the disk as well as azimuthal distributions, abundance patterns in low
surface brightness galaxies, effects of cluster environment on
gradients, the mathematical form of abundance profiles, and results of extragalactic planetary nebula studies. We
treat these topics briefly.
{\it A negative vertical gradient in O/H in the Milky Way} is
suggested by planetary nebula studies. Abundance data compiled by
Kaler (1980) for PNe ranging in height above the disk from less
than 0.4~kpc to greater than 1~kpc show a decrease in O/H with
increasing height above the plane. A comparison of more recent
studies of PNe close to the plane (Perinotto 1991), greater than
300pc above the plane (Cuisinier et al. 1996), and in the halo
(Howard, Henry, \& McCartney 1997) shows averages of 12+log(O/H) for these
three samples of 8.68, 8.52, and 8.02 respectively, qualitatively
consistent with Kaler.
Thorough tests for {\it azimuthal gradients in spiral disks} have
yet to be carried out. One example of apparent O/H asymmetry is
discussed by Kennicutt \& Garnett (1996) in their study of M101.
They find that H~II regions located along a spiral arm southeast
of the major axis have a lower oxygen abundance by 0.2-0.4~dex
compared with H~II regions on the opposite side.
{\it Global metallicities in low surface brightness galaxies} are
generally found to be subsolar by roughly a factor of three, according
to McGaugh (1994), indicating that these galaxies evolve very slowly
and form few stars during a Hubble time. Apparently, they also
lack detectable gradients.
This, despite the fact that
these objects are similar in mass and size to prominent spirals
defining the Hubble sequence.
McGaugh suggests that a galaxy's environment and surface
mass density are more relevant to galaxy evolution than gross size.
{\it Effects of cluster environment} on the chemical evolution of
galaxies have been investigated by Skillman et al. (1996), who
studied oxygen profiles in several Virgo spirals representing a
range in H~I deficiency (taken as a gauge of cluster environmental
interactions). Their results imply that global metal abundances in
disks tend to be higher in stripped galaxies, presumably because
reduced infall of metal-poor H~I gas means less dilution of disk
material. Henry et al. (1996 and references therein) investigated
metallicity and heavy element abundance ratios (N/O, S/O) in three
cluster spiral disks with normal H~I and found no clear signatures
of environmental effects. Thus, cluster environment alone is
apparently not a sufficient condition for altered chemical
evolution.
{\it The mathematical form of abundance profiles in spiral disks}
has been investigated recently by Henry \& Howard (1995), who fit
line strength behavior over the disks of M33, M81, and M101 using
photoionization models. Their best fits for O/H versus
galactocentric distance were produced using exponential profiles,
although power law forms could not be ruled out. However, linear
profiles poorly reproduced the observations. Henry and Howard
also concluded that, despite some observational and theoretical
claims to the contrary (see Moll{\'a} et al. 1996),
it is premature to conclude that
gradient flattening is present in the outer parts of some disks.
{\it Planetary nebulae have been used as probes of interstellar
abundances in a small number of external galaxies.} A recent paper by
Jacoby \& Ciardullo (1999) presents abundances for 12 bulge and three
disk planetaries in M31. Their bulge objects have oxygen abundances
whose average is similar to the value observed in the Large Magellanic
Cloud. Interestingly, the implied bulge metallicity is significantly
below the level expected from observations of [Fe/H]. In another study,
Stasi{\'n}ska, Richer, \& Mc~Call (1998) determine abundances of
oxygen, neon, and nitrogen in planetaries in the bulges of the Milky
Way and M31, M32, and the Magellanic Clouds. These authors find higher
oxygen levels in the Milky Way and M31 bulges than in the Clouds, and
also reconfirm the tight correlation between neon and oxygen discussed
below in {\S}2.2.3.
\subsubsection{Summary Thoughts About Spiral Metallicities}
A detailed synthesis based upon the observations is beyond the
scope of our review. However, the following would seem to
provide a reasonable set of explanations.
There appear to be two fundamental physical parameters for a galaxy which
influence its abundance characteristics. These are total mass and the
distribution of material as a function of galactocentric distance. As
supernovae erupt, their metal-rich ejecta are more likely retained in
systems with greater mass. Thus, the more massive galaxies might be expected to exhibit higher global metallicities, which in fact they
do. Furthermore, observations indicate that metallicities tend to be greater in regions where the
total surface density is higher, perhaps because the star formation
process is sensitive to the local density and so more metals are
produced in locations with high densities. Since matter in spirals tends
to form an exponential disk (Binney \& Merrifield 1998) with surface
density falling off with greater galactocentric distance, we might
then expect metallicity locally in the disk to track this pattern.
\subsection{Heavy Element Abundance Ratios In Spiral Disks}
Ratios of heavy elements, i.e. N/O and C/O,
are expected to reveal in particular the
characteristics of the initial mass
function, stellar
yields, and the history of star formation.
Here we consider five ratios which are accessible through
nebular studies,
N/O, C/O, Ne/O, S/O, and Ar/O. Note that because planetary
nebulae are self-contaminating with nitrogen
and (sometimes) carbon, they do not make good probes of the
interstellar levels for these elements, although in the cases of
O, Ne, S, and Ar they seem to work satisfactorily in that
capacity.
\subsubsection{N/O}
We consider the nitrogen abundance studies for the Milky Way
disk indicated in Table~1 along with
H~II region studies by Kobulnicky
\& Skillman (1996), van~Zee et al. (1998), Thurston, Edmunds, \&
Henry (1996), and Izotov \& Thuan (1999) for external spirals.
Figure 6 shows log(N/O) versus 12+log(O/H) for both the Milky Way
and extragalactic objects. Symbols are explained in the caption.
The most striking feature in Fig.~6 is the apparent threshold
running from the lower left to upper right beginning around
12+log(O/H)=8.25 and breached by only a few objects. Behind this
line the frequency of objects drops off toward lower values of
12+log(O/H) and higher values of log(N/O). A second feature is
the behavior of N/O at values of 12+log(O/H)$<$8, where N/O
appears constant, a trend which seems to be reinforced by the upper
limits provided by the damped Ly$\alpha$ objects of Lu et al. (1996; L) at very low
metallicity. This bi-modal behavior of N/O was pointed out
by Kobulnicky \& Skillman (1996).
Although detailed theoretical interpretations are beyond our scope,
we summarize below the basic ideas of nitrogen production and
attempt to tie them to Fig.~6. Readers interested in additional
detail are urged to refer to Vila-Costas \& Edmunds (1993).
Nitrogen is mainly produced in the six steps of the CN branch of the
CNO bi-cycle within H burning stellar zones, where $^{12}$C serves as
the reaction catalyst (see a textbook like Clayton 1983 or Cowley 1995
for nucleosynthesis review). Three reactions occur to transform
$^{12}$C to $^{14}$N: $^{12}$C(p,$\gamma$)$^{13}$N($\beta ^+
\nu$)$^{13}$C(p,$\gamma$)$^{14}$N, while the next step,
$^{14}$N(p,$\gamma$)O$^{15}$, depletes nitrogen and has a
relatively low cross-section. The final two reactions in the
cycle transform $^{15}$O to $^{12}$C. Since the fourth reaction
runs much slower than the others,
the cycle achieves equilibrium only when $^{14}$N accumulates to high
levels, and so one effect of the CN
cycle is to convert $^{12}$C to $^{14}$N. The real issue in
nitrogen evolution is to
discover the source of the carbon which catalyzes the process.
Since stars produce their own carbon during He burning, nitrogen
originating from it is termed {\it primary} nitrogen. Any nitrogen
produced during supernova explosive nucleosynthesis is also termed
{\it primary} since it is created for the first time during the
explosion. On the other hand, stars beyond the first generation in a
galactic system already contain some carbon inherited from the
interstellar medium out of which they formed. Nitrogen produced from
this carbon is termed {\it secondary} nitrogen.
As a system begins to mature chemically from a state of low
metallicity, nitrogen must come from carbon produced by the star
itself, since at this point no significant level of carbon
exists in the ISM which can be incorporated into new
stars and enter into the CN cycle. So, nitrogen production is
primary and its evolution
proceeds at a rate set only by star formation coupled with the
primary production rate of nitrogen. Since the production of
elements such as oxygen is being influenced by similar factors,
the N/O ratio should remain constant as their abundances rise
together.
But as metallicity rises and stars form out of progressively more metal-rich
environments, the amount of carbon present in the star at birth which can
ultimately enter the CN cycle becomes comparable to the amount produced
internally through He burning, and
thus nitrogen production becomes secondary and coupled to the
metallicity of the star.
At this point, N/O versus O/H assumes a positive slope, since
the relation between N and O is now quadratic (Vila-Costas \&
Edmunds 1993).
Based upon the data and models presented in Fig.~6 and allowing
for the scatter, a reasonable explanation for
the observed trend for N/O is that the flatter behavior seen at
12+log(O/H)$<$8.0 corresponds to the dominance of primary
nitrogen production, while the steeper slope in N/O at higher metallcities is
linked to metallicity-sensitive secondary nitrogen
production. We concur
with Shields, Skillman, \& Kennicutt (1991), who found that the
point at which secondary nitrogen production becomes
important is located at roughly
12+log(O/H)=8.3 or 0.6~dex below solar.
A comparison of nitrogen yields from
intermediate mass stars (1-8M$_{\sun}$) by van den Hoek \&
Groenewegen (1997) with those from massive stars by Nomoto et
al. (1997b) suggests that intermediate mass stars are ultimately the main
contributors to nitrogen production, although early-on massive
stars may play a role, due to the longer evolutionary time
scales for less massive stars, and thus their delay in
depositing nitrogen into the interstellar medium.
Vila-Costas and Edmunds (1993) calculated analytical models for
the evolution of N/O assuming a simple, closed-box regime but
accounting separately for primary and secondary nitrogen yields
along with time delays in intermediate-mass star nucleosynthesis.
The two curves in Fig.~6 represent their results using their Eq.~A5
along with two different values for the ratio of time delay
to the system age. We have adopted values for constants
$a$ and $b$ in their formula of 0.025 and 120, respectively, to
force a better fit to the data presented here. The curve
representing the small delay clearly matches the low metallicity
data better, while the curve for greater delay seems to rise
faster at high metallicity and thus fit the data there better.
Further study of the origin of nitrogen will
require especially more abundances for systems of low metallicity
where 12+log(O/H)$<$7. Studies of damped Lyman-$\alpha$ systems
currently offer great promise in this regard.
\subsubsection{C/O}
Carbon is produced during core and shell helium burning in the
triple alpha process, $3 ^4He \to ^{12}C$.
It is an element whose abundance
has lately become more
measurable in extra-galactic H~II regions,
thanks to the Hubble Space Telescope (HST) and its UV capabilities, since the
strong carbon lines of C~III] and C~IV appear in that spectral
region. Recent studies of extragalactic H~II regions have been
carried out by Garnett et al. (1995; 1997; 1999) and Kobulnicky
\& Skillman (1998), while carbon abundances for M8 and the Orion
Nebula, both within the MWG,
have been measured by Peimbert et al. (1993) and Esteban
et al. (1998), respectively.
Results of these measurements are collected together in Fig.~7,
where log(C/O) is plotted against 12+log(O/H). The point
for Orion is indicated with an `O', M8 with `M', and the sun
with an `S' (Grevesse et al. 1996). The vertical lines connect
points corresponding to carbon abundances determined with two
different reddening laws by Garnett et al. (1998). The filled
circles correspond to stellar data from Gustafsson et al. (1999)
for a sample of F and G stars.
A direct correlation between C/O
and O/H is strongly suggested and has been noted before (c.f.
Garnett et al. 1999),
although the result is weakened somewhat by the two points
for I~Zw~18 around 12+log(O/H)=7.25. Ignoring these two points as well as the ones for the sun and stellar data,
and performing a regression analysis, we find that $log(C/O)=-5.34(\pm 0.68)+0.59(\pm 0.08)[log(O/H)+12]$ (solid line in Fig.~7) with a correlation coefficient of 0.88
when we exclude Garnett et al.'s (1999) data points
corresponding to R$_v$=5 (the connected points with lower C/O),
$log(C/O)=-4.45(\pm 0.60)+0.48(\pm 0.07)[log(O/H)+12]$ (dashed line in Fig.~7) with a correlation coefficient of 0.86 when points for R$_v$=3.1 (the
connected points with higher C/O) are excluded. Assuming that with
additional data the trend becomes more robust, it clearly
implies that carbon production is favored by higher
metallicities.
One promising explanation (Prantzos, Vangioni-Flam, \& Chauveau 1994; Gustafsson et al. 1999) is that mass loss
in massive stars is enhanced by the presence of metals in their
atmospheres which increase the UV cross-section to stellar
radiation. Stellar yield calculations by Maeder (1992) appear to support this
claim. The contributions to carbon by different stellar mass
ranges is discussed by both Prantzos et al. and Gustafsson et al., who conclude that the
massive stars are primarily responsible for carbon production.
It is also clear, however, that stars of mass less than about
5M$_{\odot}$ produce and expel carbon as well (van den Hoek \&
Groenewegen 1997), and thus the
relative significance of massive and intermediate mass stars is
still not understood completely.
\subsubsection{Ne/O, S/O, \& Ar/O}
Neon is produced through carbon burning
($^{12}$C+$^{12}$C$\to$$^{20}$Ne+$^4$He),
while both sulfur and argon originate from
explosive oxygen burning in Type~II supernova events
($^{16}$O+$^{16}$O$\to$$^{28}$Si+$^{4}$He, then $^{28}$Si+$^{4}$He$\to$$^{32}$S;
$^{32}$S+$^{4}$He$\to$$^{36}$Ar). In
addition, substantial amounts of
S and Ar may be manufactured in Type~Ia supernova
events (Nomoto et al. 1997a). Note that here we refer only to the dominant isotopes of the respective elements.
Abundance ratios of Ne/O, S/O, and Ar/O are plotted
logarithmically against 12+log(O/H) in Fig.~8. To the data of Shaver et al.
(1983) and Maciel \& K{\"o}ppen (1994) for the Milky Way we
have added data for Ne/O, S/O, and Ar/O from optical studies of
extragalactic H~II regions
from van~Zee et al. (1998) and Izotov \& Thuan (1999)
along with S/O results from Garnett (1989) for both the MWG and
extragalactic H~II regions. Representative uncertainties are
$\pm$0.20~dex in each of the three ratios and $\pm$0.20~dex in
12+log(O/H). The horizontal lines in each panel represent
the predictions from Nomoto et al. (1997a; dashed lines), Woosley \& Weaver (1995; dot-dashed lines), and Samland (1998; solid lines) for massive star yields
integrated over a Salpeter initial mass function between 10-50~M$_{\sun}$ and corrected to give
ratios by number.
All three panels of Fig.~8 show vertical ranges which are
consistent with uncertainties and thus imply in each case a
constant value for each ratio over the 1.5-2 decades of
oxygen abundance. Logarithmic values for unweighted arithmetic
averages (log average antilog) and standard deviations (not
uncertainties) for Ne/O, S/O, and Ar/O are presented in Table~3,
where the first column identifies the sample by the last name of
the first author followed by three numbers indicating the sample
sizes for Ne/O, S/O, and Ar/O, respectively. Also included are
averages for the total of all samples, solar values (Grevesse et
al. 1996), and ratios
found in the Orion Nebula (Esteban et al. 1998) and the Helix
Nebula (Henry, Kwitter, \& Dufour 1999), a nearby planetary
nebula. Generally, for each abundance ratio
there is very good agreement among the
five samples, considering observational
uncertainties. Notice the smaller dispersion
associated with the Izotov \& Thuan data. This may be
explained by their focus
on metal-poor H~II regions possessing very bright emission lines with
resulting signal-to-noise of 20-40 in the
continuum and
abundances frequently having uncertainties of less than $\pm$0.10~dex
compared with the typical $\pm$0.20~dex uncertainties in other
samples (Izotov, private communication). In
addition, we note the significant disparity between Ar/O for Orion
and the other samples along with the sun and the Helix Nebula.
Due to the limited
spectroscopic range, the argon abundance in
Orion was determined using the weak 5192~{\AA} auroral line of
Ar$^{+2}$, where its strength was observed to be on the order of
10$^{-3}$ times H$\beta$. Use
of stronger near IR lines may bring the argon abundance in Orion
into agreement with other objects (Esteban, private
communication). In addition, the S/O ratio found in the Helix Nebula is an order of magnitude below the average value. This is currently difficult to interpret, although a few planetary nebulae do show sulfur abundances which are this low (see Henry et al. 1999).
The evidence provided by Table~3 and Fig.~8 supports the
contention that abundances of Ne, S, Ar, and O evolve in lockstep, a point made by Henry (1989) in his earlier study of Ne and O in planetary nebulae.
This would be expected if these elements are all either produced by
massive stars
within a narrow mass range or stars of different masses but with an
invariant initial mass function. Under these conditions their
buildup is expected to proceed
in lockstep, and the ratios of Ne/O, S/O, and Ar/O should have
constant values over a range of O/H.
Interesting departures from the
universality of the ratios displayed in Fig.~8
appear in halo planetary nebula studies. The
detailed one by Howard, Henry, \& McCartney (1997), for example,
confirms earlier indications that the object BB-1 has
log(Ne/O) of -0.11, while log(Ne/O) for H4-1 has a value of
-1.82. These deviants might be explained by local abundance
fluctuations caused by recent supernova events whose ejecta,
differing in composition because of mass cut differences in the
explosive event, had not yet mixed in with the surrounding ISM
before the PN progenitor formed out of it.
Finally, notice that predicted ratios from the yields for stars in the
10-50~M$_{\odot}$ mass range, represented by the horizontal lines, generally fall below the observed average, with the offset for Nomoto et al. consistently being the largest. This suggests that the theoretical calculations
overproduce oxygen and further imply that the adopted
rate of the $^{12}$C($\alpha,\gamma$)$^{16}$O reaction in the models is too
high, resulting in a higher conversion rate of $^{12}$C to
$^{16}$O with a boost in the oxygen production relative to
elements such as Ne, S, and Ar. This conclusion agrees at least
qualitatively with comparisons by Nomoto et al. (1997b) of yields
from two 25~M$_{\odot}$ stellar models using significantly different
values of the $^{12}$C($\alpha,\gamma$)$^{16}$O rate. Another
explanation in the case of S/O and Ar/O may be that the predicted yields
do not include contributions
from Type~Ia
supernovae, which produce significant amounts of $^{32}$S and $^{36}$Ar,
according to Nomoto et al.'s (1997a) W7 model. Adding this source
to yields of massive stars would raise the theoretical line.
This subject should be explored in more detail, particularly
since the Ne/O ratio ought to provide a good constraint on the
value of the $^{12}$C($\alpha,\gamma$)$^{16}$O rate.
\section{Abundance Patterns In Galaxies From Stars}
Stars are substantially fainter than H II regions, planetary nebulae,
or supernova remnants in general,
and therefore can only be observed one by one in
the Milky Way except for the brightest giants and supergiants, some of
which can be observed out to about 10 Mpc. Additionally, abundance
measurements in stars come from relatively precise measurements of the
depth of absorption features (rather than emission features),
therefore requiring more photons for results of similar
accuracy. Furthermore, it is usually the more inconspicuous lines from
which the most reliable abundances are derived! However, stars have
one distinct astrophysical advantage: they are long-lived. Study of
stars of different ages can reveal chemical history explicitly, rather
than implicitly through chemical evolution models. In this
section we briefly survey the extant stellar results for the MWG
and external galaxies. A brief description of how abundances are
inferred from stellar spectra is provided in Appendix~B.
\subsection{Milky Way Galaxy}
To summarize four decades of work on Milky Way stellar abundances in a
balanced fashion is clearly beyond the scope of a single paper, so we
attempt to provide an executive summary. Workers now subdivide the
Milky Way into the spheroidal halo ($r>2$ kpc) and bulge ($r <2$ kpc)
and two disk-like components, the thick disk (scale height $\approx 1$
kpc) and the thin disk (scale height $\approx 350$ pc). We now
treat these in order.
The halo has a metal abundance of [Fe/H] $\approx -1.6 \pm
1$\footnote{Here we employ the standard bracket notation often used in
expressing abundances, $[X] \equiv log(X)-log(X)_{\sun}$, where
X represents an elemental abundance or an abundance ratio. We will use
the symbol $Z$ to represent all heavy elements at once. } with no
noticeable abundance gradient. This information comes from studies of
individual subdwarfs (Carney et al. 1990, 1996). Globular clusters
fall into two spatially and kinematically distinct groups; the inner,
metal-rich disk clusters and the outer, metal-poor halo
clusters. Internal to either group there is no clear abundance
gradient (Zinn 1996; Richer et al. 1996). Increasing
evidence suggests that the globular cluster system has a significant
age spread of 3-4 Gyr (e.g. Hesser et al. 1997), which may depend on
radius; older toward the center. There is also a pattern of lighter
species becoming enhanced relative to Fe-peak elements in stars more
metal-poor than about [Fe/H] = $-1$, with O, Mg, Al, Si, Ca, and Ti
overabundant by several tenths of a dex relative to a scaled-solar
mixture in the metal-poor group (Wheeler et al. 1989; Edvardsson et
al. 1993). The standard interpretation is that the metal-poor group
was enriched mainly by Type II supernova nucleosynthesis products and
the metal-rich stars contain a mixture of Type II and Type I products,
where the Type I supernovae are thought to produce mainly Fe-peak
elements.
The Galactic bulge has much foreground dust as well as confusion with
foreground disk stars and thus is a difficult place for observational
work. Photometric studies seem to indicate a negative
abundance gradient but the size of the gradient is not yet
well-quantified (Terndrup 1988; Frogel et al. 1990; Harding
1996). Due to the reddening, spectroscopy seems like a safer way to
proceed. Still, results are ambiguous. Ibata \& Gilmore (1995) find a
near-solar metallicity but no gradient outwards from 0.6~kpc, but
Terndrup et al. (1990) and Rich (1998) find a gradient of about $-0.4$
dex/kpc ( or $\Delta$~log($Z$)/$\Delta$~log~$R \approx -0.6$ )
considering regions somewhat further toward the Galactic center.
The thick disk, massing about 10\% of the thin disk, is separable
from the thin disk and halo primarily through kinematics or age, since
its metallicity overlaps at the high end with the thin disk (Wyse \&
Gilmore 1995) and at the low end with the halo (Nissen \& Schuster
1997). No radial or vertical gradient in [Fe/H] has been discovered in
several large data sets (Gilmore et al. 1995; Bell 1996; Robin et
al. 1996). The high-quality data of Edvardsson et al. (1993) confirms
the lack of a {\em strong} gradient for heavy elements, but finds a
probable relation between ``alpha'' elements Si and Ca relative to Fe
as a function of radius over a 4- to 12-kpc span of about
[$\alpha$/Fe]/$R_m =$ +0.03 dex/kpc, where $R_m$ is an estimate of the
radius at which the stars were born, rather than where they are
presently located. Similarly, [$\alpha$/Fe] has been found to increase
with age. Small amplitude results of this nature should become more
common as stellar abundances become more accurate.
The thin disk can be traced by open clusters, most of which are
younger than about half the age of the disk. Friel \& Janes (1993) and
Thogersen et al. (1993) worked with moderate resolution spectra of
open cluster K giants to obtain a mean [Fe/H] gradient of $-0.097
\pm 0.017 $ dex/kpc between 7 and 15 kpc. Photometry of open cluster stars
has yielded similar results (Panagia \& Tosi 1981; Cameron 1985). An
alternative to a steady gradient has been proposed by Twarog et
al. (1997), who from a sample of 76 open clusters find a sharp
falloff of roughly 0.35 dex at around 10~kpc from the Galactic
center with flat gradients interior and exterior to that radius.
Of similar luminosity to K giants are B main sequence stars. B stars
are youthful in age, so their abundances should match those of H II
regions. Accurate spectral analysis of oxygen lines in (at least) the
hotter B stars is dependent on dropping the assumption of ``local
thermodynamic equilibrium'' between the radiation field and the
gas. Smartt \& Rolleston (1997) and Gummersbach et al. (1998) have undertaken such an analysis,
deriving an [O/H] gradient of $-0.07 \pm 0.01$ dex/kpc, a result
different from most of its predecessors, but in agreement with the
nebular results.
Figure 9 shows a summary of the broad-brush abundance pattern in the
Milky Way. The metal-poor halo weighs about $10^9 M_\odot$
(e.g. Freeman 1996) compared to about $60\times 10^9 M_\odot$ for the
total mass (in stars) in the Galaxy, so the number of symbols on the
plot does not reflect where the mass is; most of the mass resides
in the disk. Two trends are evident from the figure: abundance
increases with time, and the abundance is higher toward the Galactic
center. Coupled with stellar kinematics and age information, these
abundances give a picture of Galaxy formation in which the halo formed
early and without much chemical enrichment. The disk may have started
early as well, but it is still gas rich and is still forming stars
today at near-solar abundance.
\subsection{External Galaxies}
Spectroscopy of individual stars in local group galaxies M31 and M33
has become possible in recent years for supergiants, typically
A-type. Like B main sequence stars, A supergiants suffer from serious
non-LTE effects in the outer photosphere, but lines can be chosen that
form deep in the photosphere and a partial non-LTE analysis can be
attempted for other interesting lines. The resultant accuracy can be
$\pm 0.2$ dex (Venn 1995; 1998). For M31, the [O/H] gradient obtained
from A supergiants is consistent within the errors with that obtained
from nebular studies (McCarthy et al. 1998). For M33, based on four B
supergiants, Monteverde et al. (1997) obtain an [O/H] gradient of
$-0.16 \pm 0.06$ dex/kpc, which is also similar to nebular results.
Most stellar abundance work in external galaxies relies on the colors
of red giant stars from older populations. After spectroscopic
abundance work in globular clusters showed a wide range of
metallicities among clusters, it was obvious from published
color-magnitude diagrams (CMDs) that the red giant branches are redder
at progressively higher metallicities. This finding can be used as an
abundance indicator, especially in the HST era where the tip of the
red giant branch can be seen at distances of $\sim 10$ Mpc. A younger
age population has a somewhat bluer giant branch, but this effect is
fairly minimal, and in some cases negligible when the age is already
known. Crowding of stars excludes near-nuclear regions from CMD
analysis.
The halo of M31 has been examined by Durrell et al. (1994) and Rich et
al. (1996) from HST optical colors, with the conclusion that, like the
Milky Way, no abundance gradient is apparent. Unlike the Milky Way,
the average abundance of the stars is [Fe/H] $\approx -0.6$ (Durrell
et al.) or even higher (Rich et al.). Grillmair et al. (1996) derive
an abundance distribution (number of stars per interval [Fe/H]) for
the outer disk of M31 that is identical within the errors with the
abundance distribution of the solar neighborhood. Elliptical galaxies
NGC 5128 (Soria et al. 1996) and M32 (Grillmair et al. 1996) have also
been studied in this fashion, but only at a single radius so far, so
we await further data before we can draw conclusions about abundance
profiles.
\section{Abundance Patterns In Spheroidal Systems From
Photometric Observations}
\subsection{Metallicity Gradients}
We now provide an overview of the abundance profile picture for
spheroidal systems, especially elliptical galaxies. The
principal techniques for measuring abundances in these systems
involve the use of photometric indices of integrated starlight,
since individual stars cannot be resolved. A brief description
of these techniques is given in Appendix~C. Elliptical galaxies look
superficially like a fairly homogeneous class of galaxies, with muted
star formation and no obvious cold gas, and kinematically supported by
almost randomly oriented orbits. Star formation can be seen in most
ellipticals at some level, as can dust lanes and emission-line gas,
but usually at a level below that of spirals. Ellipticals also exhibit
regularity of colors and absorption feature strengths, with larger
galaxies being redder and having stronger metallic absorption features
than smaller ones. This has long been interpreted as a sign that
the metallicity is higher in larger galaxies (e.g. Faber 1972).
To derive abundance profiles in stellar systems, colors and absorption
feature strengths as a function of galactocentric radius are
interpreted through population models. Some color gradient studies
include Kormendy and Djorgovski (1989), Franx \& Illingworth (1990),
and Peletier et al. (1990). Most studies of optical absorption
features have utilized one particular system of feature indices
developed at Lick Observatory (described in Worthey et al. 1994 and
references therein). The last few years have seen a rapid expansion of
galaxy data available in this system. To measure an absorption feature
in the Lick system, one creates a pseudocontinuum by bracketing the
spectral feature of interest with flanking passbands. Flux in the
flanking bands is measured and a straight line is drawn between the
midpoints of the flanking bands to represent the
(pseudo)continuum. The flux difference between the pseudocontinuum and
the absorption feature is integrated and the result is expressed in
\AA\ of equivalent width (or magnitudes, depending on the specific
index; see Worthey et al. 1994 and Worthey \& Ottaviani 1997 for the
details.) Figure 10 illustrates the idea for a portion of the
spectrum. There are 25 indices defined, 5 definitions measuring Balmer
lines and 20 measuring various metallic absorption blends. The index
system operates at a low resolution ($\sim 8$ \AA\ FWHM) necessitated
by Doppler smearing from the substantial (up to $\sigma = 350$ km/s)
velocity dispersions of large elliptical galaxies, and most of the
indices require corrections when velocity dispersions get large.
While most of these 25 indices follow the
$\Delta$~log(Age)/$\Delta$~log~$Z \approx -{{3}\over{2}}$ constant-index slope
described in Appendix~C, a few
(the Balmer indices) are relatively age sensitive, with
$\Delta$~log(Age)/$\Delta$~log~$Z \approx -{{1\ {\rm to}\
2}\over{2}}$, while others, notably a feature called Fe4668 whose main
contributor is really molecular carbon, are relatively metal sensitive,
with $\Delta$~log(Age)/$\Delta$~log~$Z \approx -5$ (Worthey 1994).
Arrayed against each other, it seems possible to separate the effects
of age and metallicity, in the mean.
To derive an abundance gradient in an elliptical galaxy, one compares
observed colors or line strengths with model predictions, often assuming a
constant age throughout the galaxy. For instance, Franx \& Illingworth (1990) find a mean
color gradient in 17 elliptical galaxies of $\Delta(U-R)/\Delta {\rm
log}\ r = -0.23 \pm 0.03$ mag per decade in radius. Entering the
Worthey (1994) models at age 12 Gyr, one finds that a change of 0.15
dex in $Z$ gives the required $\Delta(U-R)$, so the gradient assuming
constant age is $\Delta {\rm log}\ Z/\Delta {\rm log}\ R=-0.15$ dex
per decade. The same number is reached by considering the $B-R$
gradient. Due to the very steep surface brightness dropoff of
elliptical galaxies, projection effects are small and usually
neglected. The steep dropoff also means that long-slit spectroscopy
usually only reaches to 0.5 to 1.0 $R_e$ (the half-light radius)
although color gradient studies and ultradeep spectroscopy can reach
to several $R_e$.
Color studies and line strength studies generally give a consistent
picture of a gradient of about $\Delta$log$Z$/$\Delta$log$R \approx
-0.2$ dex per decade. There is probably a small correction to this
number, however, due to age effects. Simultaneous mean-age, mean-$Z$
estimates using the Balmer-versus-metal feature technique described
above were derived for the Gonz\'alez (1993) and Mehlert et al. (1998)
samples of galaxies, about 60 early type galaxies in a wide variety of
environments, and the distribution of gradients is shown in
Fig. 11a. There are no trends of gradient strength with luminosity or
velocity dispersion (unlike average $Z$, which is larger in larger
galaxies). The average age gradient is younger toward the center by
0.1 dex/decade (a few Gyrs), and more metal-rich by 0.25
dex/decade. The scatter in the average seems mostly due to
observational error, error in correcting for H$\beta$ emission
fill-in, and variation in abundance ratio mixture, and the
residuals scatter along the $-{{3}\over{2}}$ age-metallicity
slope (Fig. 11b) in the way expected for random errors in input index
values.
This $-0.3$ gradient in dex/decade units corresponds to about $-0.02$
dex/kpc assuming an 8~kpc radius, which is a factor of three more
shallow than the gradient found for the Milky Way disk and other
non-barred spirals (see {\S}2.1). But such a value is well within the
range of theoretical models for galaxy formation. For example,
Larson's (1974) dissipative models predict
$\Delta$log$Z$/$\Delta$log$R = -1$, while various Carlberg (1984)
models range from $-0.5$ to 0.0. Pure stellar merging gives zero
gradient, and in fact tends to erase pre-existing gradients by roughly
20\% per event, or even more via changes in radial structure of the
galaxies (White 1980).
The gradient numbers seem fairly robust and consistent from data set
to data set and from model to model. What about absolute abundances?
These are trickier. The nuclei of large elliptical galaxies have mean
[Z/H] in the range 0.0 to 0.4 dex as inferred from Balmer-versus-metal
feature diagrams. In principle, the mean abundance can be known
much more precisely, but there is an additional stumbling block
beyond just the inaccurate models and the complication of age-metal
degeneracy. The elemental mixture in elliptical galaxies is not
scaled-solar. Abundances derived from lighter-element lines like Mg b
or Na D are much higher than those derived from heavier Fe or Ca
lines, and this is the {\it main cause for uncertainty in the absolute
abundance} (Worthey 1998).
\subsection{Enhanced Light-to-Heavy Element Ratios}
The light element\footnote{In this subsection we make a distinction
between ``light'' and ``heavy'' elements, divided at the fourth row of
the periodic table, so that Ca and Fe are heavy, but Na, Mg, and N are
light. The oft-standard terminology is to speak of ``alpha'' elements,
but ``alpha'' usually includes Ca and excludes N, which makes little
sense for the abundance pattern seen in massive elliptical galaxies.}
enhancement can be seen in the case of [Mg/Fe] by plotting a magnesium
feature index (Mg$_2$) versus an average iron feature ($<{\rm Fe}> =$
the arithmetic average of indices Fe5270 and Fe5335). The age
sensitivities of these indices are about the same ($\approx
{{3}\over{2}}$!), so models of different ages and metallicities should
lie on top of one another. They do, as seen in Figure 12, but at high
Mg strength the galaxies follow another distinct trajectory entirely with some
galaxies trending toward strong Mg$_2$ strength at nearly constant
$<$Fe$>$, and hence, we infer, relatively enhanced Mg abundance.
The models are scaled-solar since they are built from local stars, so
they cannot track altered abundances. Composite populations add
approximately like vectors, so any combination of ages and
metallicities still lands on the same model locus. Different models
built by different authors have a spread of something like $\pm 0.5$
dex at constant index strength, but all models follow almost exactly
the same slope in the Fig. 12 diagrams, so that the inferred
[Mg/Fe] for the high-Mg$_2$ group of elliptical galaxies is in the
range [Mg/Fe] = 0.3 to 0.5 dex.
A crucial thing to notice is that velocity dispersion tracks Mg$_2$
very tightly, so the high Mg$_2$ galaxies are also the largest
galaxies (or, more precisely, the ``dynamically hottest''). The
Mg$_2$-$\sigma$ relation is shown in Figure 13; it is one of the
tighter scaling relations known, much tighter than, say, the
$<$Fe$>$-$\sigma$ diagram, which is almost a scatter plot.
With the existence of the Mg-$\sigma$ relation,
[Mg/Fe] increases with galaxy size, where a velocity dispersion of
$\sigma \approx 200$ km/s seems to mark the beginning of noticeable Mg
enhancement.
Figure 12 shows separate diagrams for spiral bulges, S0
galaxies, and elliptical galaxies. No marked difference between Hubble
types is seen except that already ascribed to velocity
dispersion. That is, spiral bulges hover near the solar ratio area,
but only two bulges have $\sigma > 200$ km/s (and those are on the
high-Mg side of the distribution). Elliptical galaxies possess both the
most extreme velocity dispersions and the most extreme Mg enhancement.
The gradient vectors shown in Figure 12 tend to parallel the
age-metallicity direction
traced by the various models rather than the more horizontal slope
defined by the nuclei.
This would suggest that the Mg enhancement is global
throughout the galaxy rather than concentrated only at the nucleus. Is
this a hint that the enrichment mechanism (presumably supernova)
spreads enriched gas 10 or 20 kpc from its origin, or does it merely
imply effective mixing?
The [Mg/Fe] data suggest a variation in enrichment from Type Ia
(mostly Fe) and Type II (all elements) supernovae passing from small galaxies or
bulges to large ones, in the sense that the large galaxies have more
Mg and hence comparatively more Type II enrichment. Figure 14 shows
some corroborating evidence from Trager et al. (1998) nuclear data in
that Ca appears to track Fe, while Na and N are enhanced in a way
similar to that of Mg; only in the larger galaxies or bulges.
The mechanism for modulating Type~I/Type~II enrichment is not
known. It could be due to a time delay in Type I metal production, or
could be some other connection to velocity dispersion like a variable
IMF or a variable fraction of binary stars. (Worthey, Faber, \&
Gonz\'alez 1992; Weiss, Peletier, \& Matteucci 1995).
\subsection{The G Dwarf Problem: a peaked abundance distribution}
The abundance distribution (number of stars versus [Fe/H]) in the
Milky Way galaxy is more strongly peaked than the simplest closed-box
model (see {\S}1) with constant yield predicts. This is known as the G dwarf
problem (van den Bergh 1962; Pagel 1997, Cowley 1995). It seems almost
certain now that other galaxies share this ``problem'' of having a
relatively peaked abundance distribution. Part of the evidence comes
from integrated light (Bressan et al. 1994; Worthey, Dorman, \& Jones
1996) via three lines of evidence.
First, around 2600 \AA\ there is a paucity of ultraviolet flux which
would otherwise be greater if large numbers of metal-poor main
sequence stars are present. Second, in small compact ellipticals for
which data exist, a high-resolution Ca II index (Rose 1994 system, not Lick
system) detects few A-type horizontal branch stars, where
more of these objects would be expected if a large metal-poor
population exists. Third, if a metal-sensitive index like Fe4668 is
modeled using the broad simple model predictions, the strong line
strengths in large galaxies are very difficult to attain, and require
improbably high yield values. Clinching the integrated light results,
recent color-magnitude diagram studies of individual red giants in the
compact elliptical M32 (Grillmair et al. 1996), large elliptical NGC
5128 (Soria et al. 1996), and the disk of M31 (Grillmair et al. 1996)
all indicate a very peaked abundance distribution similar to or more
peaked than that of the Milky Way.
These empirical findings represent important constraints
on some galaxy formation and chemical evolution
models, and their implications are only starting to be explored (Larson
1998).
\subsection{Assorted Issues}
{\it The Mg$_2$-$\sigma$ relation} (Fig. 13) is tighter than other
population-to-structural correlations such as Mg$_2$-$M_B$, or $<{\rm
Fe}>$-$\sigma$. (Bender, Burstein, Faber 1993) There is some powerful
connection between velocity dispersion and Mg abundance
as traced by the Mg b feature,
the exact nature of
which eludes us at the moment. One possibility, suggested by Faber et
al. (1992), is that cloud-cloud collision velocity modulates the IMF
to favor more massive stars in high-$\sigma$ environments. This
suggestion is in harmony with the Mg/Fe abundance trend. Another
possibility is that larger local escape velocities resist supernova
winds more effectively, holding onto heavy-element contaminants
better. Star formation is finally truncated when supernova winds are
able to blow the remaining gas out of the galaxy. This is the
now-standard picture of chemical evolution in elliptical galaxies
(e.g. Arimoto \& Yoshii 1987, Matteucci \& Tornamb\'e 1987). This
picture can also be made harmonious with the Mg/Fe trend if there is
an additional mechanism for varying Mg/Fe as a function of galaxy
size, but the fact that Fe abundance is virtually independent of
galaxy size causes some trouble.
{\it Discontinuities in kinematic profiles are coincident with
discontinuities in line strength profiles.} Bender and Surma (1993)
discovered in the course of studying peculiar kinematics in elliptical
galaxies that many have counter-rotating cores or other kinematic
discontinuities. In every case that they studied, a discontinuity
appeared in the Mg$_2$ line strength profile at the same location as
the kinematical discontinuity. At the very least, this implies that
whatever formation mechanism produced the distinct core also
influenced the local chemistry. Muted echoes of formation exist still
in both the stellar kinematics and the chemical signature in the
stars.
Study of the {\it globular cluster systems around elliptical galaxies}
yields insight into the formation of halos in general and elliptical
galaxies in particular. Some elliptical galaxies host a very large
number of globular clusters per unit luminosity (e.g. M87) while
others have about as many as are seen in spiral galaxies. Questions
remain about how the globular clusters are created and destroyed to
explain the wide variation in number and whether merging events are
important or not (van den Bergh 1995; Zepf \&
Ashman 1993). Most abundance studies concentrate on the integrated
colors of the clusters (Ostrov et al. 1998; Lee et al. 1998; Ajhar et
al. 1994) because of their faintness, but some spectroscopic studies
are beginning to appear (e.g. Cohen et al. 1998). These efforts show a
variety of interesting results. For example,
some globular cluster systems are metal-poor, some
metal-rich, and some bimodal or multimodal. And when both metal-poor
(blue) and metal-rich (red) populations coexist, the red population
tends to be more centrally concentrated than the blue.
\section{Summary and Suggestions}
We have explored in some detail the abundance patterns in spiral disks and elliptical galaxies as revealed through analyses of gaseous nebulae, stars and integrated photometry of galaxies.
The principal points regarding abundance patterns in spiral disks
are:
\begin{itemize}
\item The metallicity as gauged by O/H in nebulae across the Milky Way disk decreases with galactocentric distance, a finding supported by recent abundance results for disk stars. This negative gradient pattern is seen in most other spiral disks. A similar result is seen when luminous stars are used as abundance probes. Scatter at any particular galactocentric distance is consistent with observational uncertainty.
\item Global metallcity, taken as the abundance of oxygen at a standard galactocentric distance, is positively correlated with galaxy mass.
\item Metallicity at any location in a spiral disk appears to be positively correlated with the local total surface density.
\item Abundance gradients are steeper in normal spirals than in barred ones.
\item A plot of N/O versus O/H in spiral disks indicates that production of nitrogen is dominated by primary processes at low metallicities and secondary processes at high metallicities.
\item C/O is positively correlated with O/H in spiral disks, suggesting that
carbon production is sensitive to metallicity, possibly through metallicity-enhanced mass loss in massive stars.
\item Abundance ratios of Ne/O, S/O, and Ar/O appear to be universally constant across the range in metallicities observed, reflecting the idea that either the initial mass function is universally constant; or the stellar mass range responsible for producing these elements is relatively narrow, and thus these ratios are insensitive to IMF variations.
\item Stellar age and galactocentric distance in the Milky Way show rough correlations with metallicity in the sense that metallicity
decreases with increasing age and galactocentric distance. However, all
Galactic components (halo, bulge, thin disk, thick disk) have large scatter
in abundance, and even the metal-poor halo is now thought to display
age scatter of several Gyr.
\end{itemize}
For elliptical galaxies, the main results are:
\begin{itemize}
\item Abundance gradients
are, on average, about a factor of two to three
more shallow than in non-barred spirals. This is well within the range
expected from various formation pictures, including hierarchical
mergers of smaller galaxies.
\item Nuclear or global metallic feature strengths (or colors) become
stronger (or redder) in larger galaxies. The 1970's conclusion that
larger elliptical galaxies must be more metal-rich is reconfirmed, but
every elemental species does not increase in lockstep.
\item Light elements N, Na, and Mg are enhanced relative to heavy
elements Ca and Fe in the largest elliptical galaxies, implying a
modulation of enrichment, plausibly due to variance of the Type II to
Type Ia supernova ejecta, compared to smaller ellipticals, bulges, and
disks.
\item The mean abundance near the nuclei of large elliptical and S0
galaxies is uncertain, but is in the range [Z/H] = 0.0 to 0.4. Most of
the difference in abundance between small and large galaxies is driven
by the increasing abundance of elements {\it lighter} than those near the
Fe-peak, with [Fe/H] staying roughly constant for elliptical galaxies
of all sizes.
\item The
abundance distribution in elliptical galaxies and, so far, every other
well-studied large galaxy type, is strongly peaked like that of the solar
cylinder, not broad like the simplest closed-box model predicts.
\end{itemize}
There are two over-arching patterns which emerge from the combined
results for spirals and ellipticals. First, {\it there is a positive
correlation between galactic metallicity and mass}. This may be
related to the greater retension of heavy elements ejected by
supernovae by the stronger gravitational potentials of massive
galaxies, or perhaps to the effects of galaxy mass on the star
formation process. It is currently difficult to ascertain whether this
relation is completely continuous across galaxy types; in other words,
if one plotted global metallicity versus mass for a sample of galaxies
containing both spirals and ellipticals would there be an unbroken
straight line, or would the correlation for one type be offset from the
other. The difficulty here is in directly comparing abundances between the two
galaxy types. As we have seen, metallicity in spirals is generally
gauged by observing oxygen in nebulae located in their disks. Yet in
ellipticals, metallicity must be measured from integrated light using
numerous photometric indices which are affected not only by
metallicity but by age. Thus, no seamless technique exists for
determining abundances consistently for spirals and ellipticals, and
thus intercomparisons are problematic. This is made all the more
complicated by the fact that we currently don't know how to represent
the global abundance in a galaxy. Do we take the abundance at the
nucleus, or at one effective radius, or at 0.4 optical radii?
The second pattern which has emerged is that {\it abundance gradients
appear to become flatter as one progresses from normal spirals to
barred spirals to ellipticals}. The difference between normal and
barred spirals is currently explained by enhanced radial gas flows in
the disks of barred spirals. To
extend this model to ellipticals it may be neccessary to invoke other
radial mixing mechanisms, either during primordial formation or during
later merging events.
If the pattern is discontinuous between spirals
and ellipticals this might suggest that different processes operate
to affect the gradients in the two galaxy types. Again, our lack of
ability to intercompare spiral and elliptical abundances prevents
further exploration of this pattern at present.
Understanding the broad picture of galactic chemical evolution will
require us to firm up the links between elliptical and spiral galaxy
abundances. While a common elemental yardstick may not exist because of the
different elements which we observe directly in each galaxy type, it
may be possible to tie the two types together abundance-wise by
observing elements in each which share the same production site
nucleosynthetically speaking. An example might be oxygen and
magnesium. In external spirals oxygen is taken as the metallicity gauge
primarily because of its observability. Magnesium, which, like oxygen,
is primarily produced in massive stars (Nomoto et al. 1997a,b) may be
measurable directly through a calibrated Mg$_2$ index. Then oxygen and
magnesium might be linked by assuming a ``cosmic'' Mg/O ratio
calibrated locally. Also, although the work is not started, it may be
possible to construct an oxygen-sensitive photometric index for
integrated light, perhaps revolving around the 2.3$\mu$m CO feature in
conjunction with the C$_2$-sensitive 4668 feature. As synthetic
spectra and stellar abundances grow more precise, these speculative
suggestions might take place, leading to a much more clear
understanding of chemical enrichment and galaxy formation.
\acknowledgments
This collaboration was inspired by the October, 1997, workshop
``Abundance Profiles: Diagnostic Tools For Galaxy History'' held at
Universit{\'e} Laval, Qu{\'e}bec. We are grateful to the
organizers for giving all of us the opportunity to come together and share
our ideas about what one participant, in an attempt to relabel the abundance
field with a trendier and more attention-grabbing name, referred to as
``bio-resources''. We also thank Bill Blair, C{\'e}sar Esteban, Mike
Fich, George Jacoby, Yuri Izotov, Joachim K{\"o}ppen, Walter Maciel,
D\"orte Mehlert, Anne Sansom, Jan Simpson, and Friedl Thielemann for promptly responding
to inquiries with useful answers and information given generously. And finally, we are especially grateful to our referees, Dave Burstein, Karen Kwitter, and Bernard
Pagel, for promptly and carefully reading the manuscript and making
numerous constructive comments which have improved the paper
tremendously.
|
\section{Rejection of atmospheric muon background}
In the Austral summer 96-97 the construction of the
first generation AMANDA detector was completed. The detector consists
of 300 optical sensors on 10 strings located at depths of 1500 to 2000\,m
in the deep Antarctic ice. The calibration and the performance
characteristics of the AMANDA array are described in reference \cite{b4}.
In this report we present a first analysis of data taken during a period
of 113 days during the first year of operation in 1997. The detector
live time corresponds to about 85 days of data.
\begin{table}[htbp]
\vskip -0 cm
\caption{Rejection of background and efficiency for atmospheric neutrinos
at background rejection levels 1 to 4. The meaning of the cuts is explained
in the text. Two categories of "direct hits" are used: B) [-5,+25]\,nsec,
and C) [-5,+75]\,nsec. The results are given for a Monte-Carlo simulation of
cosmic ray muons (14\,h), for a simulation of atmospheric neutrinos (85\,d),
and for experimental data (85\,d).
}
\label{tab_all_in_one}
\begin{center}
\begin{tabular}{|l|c|c|c|c|c|c|c|}
\hline
& Cut Level & 0 & 1 & 2 & 3 & 4 \\
\hline
Filter & filter & & yes & yes & yes & yes \\
\hline
Quality & Direct B Hits & & &$\geq$ 5&$\geq$ 5 &$\geq$ 6 \\
cuts & Direct C Hits & & & &$\geq$ 10 &$\geq$ 15 \\
& D. Length [m] & & & &$\geq$ 100 &$\geq$ 100 \\
& Edge cut & & & & & yes \\
\hline
Zenith & $ \theta_1$ (line fit) & & & & $\geq 80^\circ$ & $\geq 100^\circ$ \\
angle & $\theta_2$ (full fit) & & $\geq80^\circ$ & $\geq 80^\circ$
& $\geq 100^\circ$ & $\geq 100^\circ$ \\
\hline
Results & MC: atmos. $\mu$ & $3.4\cdot 10^6$ & $2.1\cdot 10^5$ & 853 & 0 & 0 \\
& MC: atmos. $\nu$ & 2000 & 1016 & 272 & 89 & 21.1 \\
& Exp. Data & $4.9\cdot 10^8$ & $4.5\cdot 10^7$ & $3.5\cdot 10^5$ & 452 & 17 \\
\hline
\end{tabular}
\end{center}
\end{table}
Atmospheric muons are recorded at a rate of 70\,Hz. Upward going
atmospheric muon neutrinos
are expected to trigger the AMANDA 10-string detector at a rate of about
$3\cdot 10^{-4}$\,Hz or 25 events per day.
The only parameter for background rejection is the
direction of the reconstructed track, which decides whether a muon was
moving upward or downward.
Upward muon tracks are generated by neutrinos, where downward moving
tracks are totally dominated by penetrating cosmic ray muons
generated in the atmosphere.
About 90\% of the cosmic ray muons are rejected with a
simple filter method based on the correlation of arrival times and depth
of the observed Cherenkov photons.
The remaining events are reconstructed by fitting
the Cherenkov light cone generated by a relativistic particle
to the observed arrival times \cite{wiebusch}.
After the initial reconstruction a set of quality cuts are applied
to suppress a remaining background of muons which were reconstructed
as upward moving.
The most important cut is the number of "direct hits" in an event.
A direct hit is a photon that is detected within a time interval
of [-5,+25] nsec of the fitted Cherenkov cone.
Another important criterion is the "direct length" cut which
requires that the direct hits are distributed over a muon track of at
least 100\,m length.
A combination of two other cuts is the "edge cut" which requires
that the event was not exclusively
concentrated at the top or bottom edge of the detector.
\section{Observation of atmospheric neutrino candidates}
\begin{figure}
\vskip -1.7 cm
\hbox{
\hskip -0.3cm\vbox{
\epsfxsize=2.4in
\epsffile{zenith_overlay_cos.ps}
}
\hskip -0.2 cm \raise 1.9cm \vbox{
\epsfxsize=2.2in
\epsffile{skyplot_lev4.eps}
}
}
\vskip -2.3 cm
\caption{ The reconstructed zenith angles of 113 days of AMANDA 10 string data is shown
for quality level 2 (solid lines), 3 (dashed) and 4 (dotted).
The plot on the right shows the sky plot of all events that pass level 4 quality cuts.}
\label{zenith_overlay_cos}
\end{figure}
\begin{figure}
\vskip -0.5 cm
\hbox{
\hskip -0.2cm \raise 0.3cm
\vbox{
\epsfxsize=2.2in
\epsffile{ArrVsDepth-jan99.eps}
}
\hskip -0.2 cm \raise 0.0cm \vbox{
\epsfxsize=2.4in
\epsffile{AmplVsDist-jan99.eps}
}
}
\vskip -0.5 cm
\caption{ Event 1197960: The recorded arrival time of photons
is plotted versus the depth of the observing sensors (left). The slope of the
dashed line is the result of the reconstructed zenith angle of $155^\circ$.
The observed pulse amplitudes are plotted versus
distance of the track (right). }
\label{ArrTimeVsDepth_1197960}
\vskip -0.3 cm
\end{figure}
We reduce the cosmic ray muon background in four steps, to which we
refer as rejection level 1 to 4. The definitions of the four cut levels
are summarized in table \ref{tab_all_in_one}.
Figure \ref{zenith_overlay_cos} shows the distribution of
the reconstructed zenith angles up to $10^\circ$ above the horizon
for the applied quality cuts from level 2 to 4.
In the same figure the sky coordinates of the remaining 17 events
are shown for quality level 4.
Above the horizon the tail of downgoing muons is visible.
However, where at cut level 2 and 3 a background of fakes is present
below the horizon, a cluster of upgoing tracks appears which is separated
from the downgoing background.
The 17 of $4.9\cdot 10^8$ events which pass the highest
quality cuts are concentrated at larger zenith angles.
The distribution in right ascension is statistically consistent with a
random distribution.
A close inspection of the spatial topology and the
amplitudes of the 17 events shows that one of the 17 events is likely to be a
$\nu_e$ initiated cascade or a bremstrahlung event.
The event characteristics of the remaining 16 events are in agreement with the
expectation for upgoing neutrino induced muons.
A display of a neutrino candidate which extends over a length
of 400\,m through the entire detector is shown by Halzen \cite{halzen}.
The upward moving signature of this event is illustrated
in figure \ref{ArrTimeVsDepth_1197960} where the photon arrival times of this event
are plotted versus the
depth of the sensors. The slope matches the vertical velocity of
a track reconstructed at a zenith angle of $155^\circ$,
which agrees with the result of the full Cherenkov cone fit.
Figure \ref{ArrTimeVsDepth_1197960} also shows the amplitudes as a function of
distance of the reconstructed muon track. The observed photon density
is high for sensors close to the track.
\section{Comparison with Monte-Carlo prediction and conclusion}
\begin{figure}
\begin{center}
\vskip -1.80 cm
\hskip 0cm
\hbox{
\hskip -0.5 cm
\raise 0.0cm
\vbox{
\epsfysize=3.6 in
\epsffile{zenith_cos_lev4-e.ps}
}
\hskip -0.5 cm
\raise 2.0cm
\vbox{
\epsfysize=2.3in
\epsffile{atmos-nu-unfilt-b10-superimp.eps}
}
}
\vskip -2.3 cm
\caption{Left: Zenith angle distribution of neutrino candidates and
of MonteCarlo simulated atmospheric neutrinos.
Right:
The simulated energy spectrum (true neutrino energy) is shown
at trigger level of the 10 string array (solid lines),
at level 3 cuts (dashed lines) and at level 4 (dotted).}
\label{level4_zenith_overlay_cos}
\end{center}
\vskip -0.4 cm
\end{figure}
A full simulation of atmospheric neutrinos has been performed
which predicts that 21 $\nu_\mu$ and $\overline{\nu_\mu} $ events pass
the level 4 cuts.
Figure \ref{level4_zenith_overlay_cos} shows the zenith angle
distribution of all events at level 4 along with the prediction of
the atmospheric neutrino simulation. The energy distribution
of simulated atmospheric neutrinos is shown for cut levels 0, 3, and 4.
The energy and angular characteristics of the atmospheric
neutrino spectrum are taken from Lipari~\cite{Lipari}.
We estimate that the combined error of theoretical prediction and
absolute sensitivity of the detector is 50\% or greater.
The angular distribution of the observed upward moving tracks
agrees well with the expectation from atmospheric neutrinos.
It illustrates the higher sensitivity of the 10 string
array to small nadir angles, reflecting that the detector is 400\,m tall,
but only 120\,m in diameter.
Deployments in the 99-00 Antarctic summer will result in a more
symmetric detector \cite{halzen}.
\section*{Acknowledgments}
\small
This research was supported by the following agencies:
1. U.S. National Science Foundation, Office of Polar Programs;
2. U.S. National Science Foundation, Physics Division;
3. University of Wisconsin Alumni Research Foundation;
4. U.S. Department of Energy;
5. U.S. National Energy Research Scientific
Computing Center (supported by the Office of Energy Research
of the U.S. Department of Energy);
6. Swedish Natural Science Research Council;
7. Swedish Polar Research Secretariat;
8. Knut and Alice Wallenberg Foundation, Sweden;
9. Deutsches Elektronen-Synchrotron (DESY).
\normalsize
\section*{References}
\small
|
\section{Introduction}
Several experimental studies \cite{jan,riguidel1,riguidel2,ristow}
have recently been conducted on the problem of a single ball falling
under gravity on a surface of controlled roughness. These works have
revealed interesting new aspects of granular dynamics that are not yet
fully understood. Three distinct dynamical regimes have been
identified \cite{jan,riguidel1,riguidel2,ristow} as the tilting angle
increases. For small inclinations there is (i) a decelerated regime
where the ball always stops, then comes (ii) an intermediate regime
where the ball reaches a steady state with constant mean velocity, and
for larger inclinations the ball enters (iii) a jumping regime.
Computer simulations
\cite{riguidel1,riguidel2,ristow,batrouni1,batrouni2} have confirmed
these results, particularly those concerning regimes (i) and (ii). A
theoretical model \cite{ancey} has also been proposed in which
steady-state solutions (but no detailed dynamics) can be obtained
analytically. More recently, a one-dimensional map \cite{valance} has
been introduced to study the jumping regime. This map in its simplest
version is linear, and to obtain non-linear behavior one has to vary
spatially the properties of the rough surface \cite{valance}, in which
case the model become inaccessible analytically.
In this Paper we present a model for a single particle moving under
the action of gravity on a rough surface of specified shape. Within
this setting we will give a detailed analytical description of all
possible dynamical regimes. Although the model we study is
simplified, its predictions are in good qualitative agreement with the
experimental findings.
Roughly speaking, our conclusions are as follows. There is (i) a sharp
transition (as the surface inclination increases) from a regime of
bounded velocity to one of accelerated motion. Within the region of
bounded velocity various dynamical regimes are possible. First there
is (ii) a range of inclinations for which the dynamics always has a
unique attractor. For higher inclinations two other phases exist:
(iii) a region where we have co-existing attractors for the dynamics
and (iv) a region where instabilities give rise to chaotic behavior.
For a fixed (sufficiently large) inclination a transition to the
chaotic region will take place as the nature of the collisions between
the particle and the surface becomes highly inelastic. Although our
results are derived here in the context of a simple collision rule, it
can be shown \cite{companion} that they remain valid for a wide class
of restitution laws.
The paper is organized as follows. In Sec.\ II we describe the model
and study in detail its dynamical properties. The main results of this
Section are then summarized in the phase diagram shown in Fig.\
\ref{fig:phase}. In Sec.\ III we carry out a comparison between the
model predictions and the experimental findings. In particular, we
argue that the jumping regime seen in the experiments might correspond
to a true chaotic motion, as predicted by the model. Finally, in Sec.\
IV we collect our main conclusions and present further discussions.
\section{the model}
In our model, which is shown in Fig.\ \ref{fig:1}, the rough surface
is considered to have a simple staircase shape whose steps have height
$a$ and length $b$. For convenience, we choose a system of coordinates
in such a way that the step plateaus are aligned with the $x$ axis and
the direction of the acceleration of gravity {\bf g} makes an angle
$\phi$ with the $y$ axis. A grain is then imagined to be launched on
the top of the `staircase' with a given initial velocity. In what
follows, we will be concerned with the problem of a {\it point}
particle falling down this `staircase' and will thus not take into
account any effect due to the finite size of the grain. Upon reaching
the end of a step plateau, the particle will undergo a ballistic
flight until it collides with another plateau located a certain number
$n$ of steps below the departure step (e.g., $n=3$ in Fig.\ 1).
Accordingly, we will refer to the integer $n$ as the {\it jump number}
associated with this flight.
We will assume, for simplicity, that the momentum loss due to
collisions is determined by two coefficients of restitution $e_t$ and
$e_n$, corresponding to the tangential and normal directions,
respectively. More precisely, if ${\bf v} =(v_x,v_y)$ denote,
respectively, the components of the particle velocity parallel and
perpendicular to the surface before a collision, then we will
take the velocity ${\bf v'} = (v_x',v_y')$ after the collision to be
given by
\begin{mathletters}
\label{eq:C}
\begin{eqnarray}
v'_x &=& e_t v_x , \label{eq:en}\\
v'_y &=& - e_n v_y, \label{eq:et}
\end{eqnarray}
\end{mathletters}
where $0\le e_t <1$ and $0\le e_n<1$.
\begin{figure}
\centerline{\hbox{\vbox{\psfig{figure=fig1.ps,width=3.0true in,height=1.8true in}}}}
\vspace{0.5true cm}
\caption{Model for a single particle moving under gravity on an rough
inclined surface.}
\label{fig:1}
\end{figure}
In the present paper we will for simplicity discuss only the case
$e_n=0$; the advantage being that the model can then be described by a
one-dimensional map. When $e_n > 0$ the dynamics is governed by a
three-dimensional map, the analysis of which is more complicated and
will be left for forthcoming publications \cite{us}.
We now derive the equations governing the dynamics of the model
presented above. Let us denote by $E$ the kinetic energy of the
particle at the moment of departure for a given flight. We write
$E={1\over2}mV^2$, where $m$ is the particle mass and $V$ is the
launching velocity at the start of the flight (see Fig.\
\ref{fig:1}). After this flight the particle will first collide with
a step below, then slide along this step (recall $e_n=0$), and
finally take off again on another flight with initial kinetic energy
$E'$. We suppose that the main energy loss is due to collisions
and so we neglect the energy dissipation as the particle slides along
a step, where it then moves with a constant acceleration $g\sin \phi$.
Using simple arguments of energy conservation together
with the collision conditions (1) and (2), one can write $E'$ in terms of $E$.
The result is
\begin{equation}
E'= {1\over2}m e_t^2 v_x^2 + mg\sin\phi \, (nb-x) ,
\label{eq:2}
\end{equation}
where $n$ is the corresponding jump number for the flight and $x$ is
the $x$-coordinate of the landing point. It takes a simple algebra to
show that at the landing point $(x,y)$ we have the following
identities:
\begin{mathletters}
\label{eq:3}
\begin{eqnarray}
&&x = {g \sin\phi\over 2} T^2 + \sqrt{{2E\over m}} T \label{eq:3a}, \\
&&y = {g \cos\phi\over 2} T^2 = na, \label{eq:3b} \\
&&v_x= g \sin\phi \, T + \sqrt{{2E\over m}}, \\
&&v_y= g \cos\phi \, T,
\end{eqnarray}
\end{mathletters}
where $T$ is the flight time.
It is convenient to introduce a dimensionless energy-like variable:
\begin{equation}
{\cal E}={E\over{m g a \cos\phi}} .
\label{eq:5}
\end{equation}
Eliminating $T$ from (\ref{eq:3}) and inserting the result into (\ref{eq:2}),
we obtain that the dynamics of the model in terms of the variable
${\cal E}$ is given by the following map:
\begin{equation}
{\cal E}' = f({\cal E},n)= n \left[ e_t^2 \, \left(\sqrt{{\cal E}/n}+ t\right)^2 + t
\left(\tau -t - 2\sqrt{{\cal E}/n}\right)\right] .
\label{eq:6}
\end{equation}
where we have for conciseness introduced the notation
\begin{eqnarray}
&&t=\tan \phi, \\
&& \tau=b/a.
\end{eqnarray}
The parameter $\tau$ above can be viewed as a measure of the surface
roughness, with $\tau^{-1}=0$ corresponding to a perfectly smooth
surface. As for the inclination parameter $t$, we need to consider
only the interval $0<t<\tau$ for which non-trivial motion
occurs. (Clearly, for $t<0$ the particle will always come to a rest,
whereas for $t>\tau$ the particle undergoes a free fall without ever
colliding again with the ramp.)
The flight jump number $n$ appearing in Eq.\ (\ref{eq:6}) is
determined from the energy $\cal E$ according to the following
condition: $n$ is equal to the smallest integer such that $nb-x\geq 0$
or, alternatively,
\begin{equation}
n(\tau-t) - 2\sqrt{n {\cal E}}\geq 0 .
\label{E-is-pos}
\end{equation}
This means that ${\cal E}$ falls within the interval $I_n$:
\begin{equation}
{\cal E} \in I_{n}(t) \equiv \left( {1\over4}(n-1) (\tau - t)^2,
{1\over4} n (\tau - t)^2\right].
\label{eq:8}
\end{equation}
Thus the function $f({\cal E},n)$, as defined by Eqs.\ (\ref{eq:6})
and (\ref{eq:8}), exhibits jump discontinuities at energy values
${\cal E}= {1\over4} n \left(\tau-t\right)^2$, but each of its
branches is smooth. This is illustrated in Fig.\
\ref{fig:map}, where we graph the function (\ref{eq:6}) for
$e_t=0.7$, $\tau=3.7$, and several values of the inclination $t$,
For later use, we note here that the average velocity $\overline{V}$
between two consecutive flights is given by
\begin{equation}
\overline{V}= {n L\over{T+ (\sqrt{2E'/m}-e_t v_x)/g\sin\phi}},
\label{eq:Vmean}
\end{equation}
where $L=\sqrt{a^2+b^2}$ and the second term in the denominator
corresponds to the time during which the particle moves on the ramp
(see Fig.\ 1). If we now introduce a dimensionless mean velocity
\begin{equation}
\overline{{\cal V}}={\overline{V}\over\sqrt{a g
\cos\phi}}, \label{eq:Vdimless}
\end{equation}
then Eq.\ (\ref{eq:Vmean}) becomes
\begin{equation}
\overline{\cal V}= {t \sqrt{n (1+\tau^2)/2 }\over{(1-e_t)t +
\sqrt{{\cal E}'/n}-e_t \sqrt{{\cal E}/n}}}. \label{eq:vmean}
\end{equation}
In order to study the dynamical properties of the map above, we must first
investigate the existence of fixed points. If we denote by ${\cal E}_n$ a
fixed point with a jump number $n$, then ${\cal E}_n$ will be the
solution to the equation
\begin{equation}
{\cal E}_n = f({\cal E}_n,n) .\label{eq:fp}
\label{hmap:ex}
\end{equation}
In view of the homogeneity of the function $f({\cal E},n)$ [see Eq.\
(\ref{eq:6})] we write
\begin{equation}
{\cal E}_n = n [z_0(t)]^2, \label{eq:En}
\end{equation}
where the quantity $z_0(t)$ no longer bears any dependence on $n$.
Using Eqs.\ (\ref{eq:6}) and (\ref{eq:En}), Eq.\ (\ref{eq:fp}) becomes
\begin{equation}
(z_0+t)^2 = e_t^2 (t+z_0)^2 + \tau t,
\label{eq114}
\end{equation}
whose positive solution is
\begin{equation}
z_0(t) = -t+\sqrt{{\tau t \over 1-e_t^2}}. \label{eq:z0}
\label{eq115}
\end{equation}
Now a fixed point ${\cal E}_n$, as given in Eqs.\ (\ref{eq:En}) and
(\ref{eq115}), will exist if and only if ${\cal E}_n\in I_n(t)$, where
the interval $I_n(t)$ is defined in (\ref{eq:8}). Thus, as $t$
increases, a fixed point with jump number $n$ will be born when ${\cal
E}_n$ equals the left endpoint of $I_n$. Comparing Eqs.\ (\ref{eq:8}),
(\ref{eq:En}) and (\ref{eq:z0}), we see that this happens at an
inclination $t_n$ such that
\begin{equation}
z_0(t_n)=-t_n+\sqrt{{\tau t_n \over 1-e_t^2}} = {1\over2} \sqrt{1-{1\over n}}
\left({\tau - t_n}\right) .
\label{eq116}
\end{equation}
This equation is quadratic in $\sqrt{t_n}$ and can thus be easily
solved. However, we shall not bother to give the result here and will
simply mention a few important facts that follow from Eq.\
(\ref{eq116}). First, we note that $t_1=0$ so that a fixed point with
jump number $n=1$ is always born at $t=0$. Then, as $t$ increases,
fixed points with successively higher $n$ will appear in an increasing
sequence of inclinations $\{t_n\}_{n=1}^{\infty}$. Finally, we have
that for $t>t_\infty$, where $t_\infty = \lim_{n\to\infty} t_n$, all
fixed points cease to exist. Setting $n=\infty$ in Eq.\ (\ref{eq116})
we obtain for the limit point $t_\infty$:
\begin{equation}
t_\infty = \tau {{1-e_t}\over{1+e_t}}.
\label{eq117}
\end{equation}
The appearance of this sequence of fixed points can perhaps be best
visualized by referring to Fig.\ \ref{fig:map}, where we plot the
function $f({\cal E},n)$ at increasing values of $t$, with $e_t$ and
$\tau$ kept fixed. For small $t$ (lower-most curve in Fig.\
\ref{fig:map}) there is only one intersection with the $45^\circ$
line, corresponding to the fixed point with $n=1$. As $t$ increases
fixed points with successively higher $n$ appear (second curve from
the bottom). At $t=t_\infty$ there are infinitely many such fixed points
(second curve from the top) and after this all of them cease to
exist (uppermost curve).
\begin{figure}
\centerline{\hbox{\vbox{\psfig{figure=fig2.ps,width=3.0true in,height=2.5true in}}}}
\vspace{-0.5true cm}
\caption{One-dimensional map $f({\cal E},n)$ for $e_t=0.7$, $\tau=3.7$
and $t=$ 0.2, 0.5, 0.7, 0.9 (from the bottom up).}
\label{fig:map}
\end{figure}
One can also show that for $t>t_\infty$ we always have $f({\cal
E},n)>{\cal E}$, whereas for $0<t<t_\infty$ there exists an energy
${\cal E}^*$ such that $f({\cal E},n)<{\cal E}$ for ${\cal E}>{\cal
E}^*$ (see, e.g., Fig.\ \ref{fig:map}). We thus conclude that for
$t>t_\infty$ the particle velocity will become unbounded for any
initial condition, whereas for $0<t<t_\infty$ the velocity remains
always bounded. In other words, at the critical inclination
$t=t_\infty$ there is a sharp transition (independent of initial
conditions) from a regime of bounded velocity to accelerated motion.
In the region of bounded velocity, several dynamical regimes are
possible, depending on the stability of the fixed points, as discussed
below.
The stability of a fixed point ${\cal E}_n$ is determined by the
parameter $\lambda= f'({\cal E}_n,n)$, where the prime denotes
derivative with respect to $\cal E$, so that if $|\lambda|<1$
($|\lambda|>1$) the fixed point is stable (unstable) \cite{ott}.
Using Eqs.\ (\ref{eq:6}), (\ref{eq:En}) and (\ref{eq:z0}), we obtain
for the derivative $\lambda$ at the fixed point:
\begin{equation}
\lambda(t) = 1 - {{1-e_t^2} \over 1 -
\sqrt{{(1-e_t^2)t/\tau }}} .
\label{eq118}
\end{equation}
Notice that $\lambda$ does not depend on $n$, thus implying that all
existing fixed points ${\cal E}_n$ (for given values of the model
parameters) have the same stability properties. Moreover, since
$\lambda$ is always smaller than unity, we see that instability can occur
only if $\lambda (t)<-1$. Let us then denote by $t_{\rm inst}$ the
inclination such that $\lambda(t_{\rm inst})=-1$. From
Eq. (\ref{eq118}) we obtain that
\begin{equation}
t_{\rm inst}=\tau {(1+e_t^2)^2 \over 4(1-e_t^2)} .
\label{eq119}
\end{equation}
Thus the fixed points are stable for $t<t_{\rm inst}$ and unstable
for $t>t_{\rm inst}$.
If the fixed points are stable, the dynamics of the map will in
general be attracted to one of the existing fixed points. For example,
in the region of parameters such that $0<t<t_2<t_{\rm inst}$ the
particle will almost always reach a periodic motion where the particle
falls by one step at a time, since in this case only the fixed point
with $n=1$ exists and is stable \cite{note}. On the other hand, for
$t_2<t<t_{\rm inst}$ there are co-existing stable fixed points, in
which case the final state (i.e., the fixed points to which the
dynamics is attracted) will depend on the initial condition. Once the
system has reached a given fixed point ${\cal E}_n$ the particle will
accordingly be moving with a constant mean velocity $\overline{\cal
V}_n$ whose value can be readily obtained by inserting Eqs.\
(\ref{eq:En}) and (\ref{eq:z0}) into Eq.\ (\ref{eq:vmean}):
\begin{equation}
\overline{{\cal V}}_n=\left[{n(1+\tau^2)t\over{2 t_\infty}}\right]^{1/2} .
\label{eq:Vn}
\end{equation}
When the fixed points are unstable ($t_{\rm inst}<t<t_\infty$), the
particle motion becomes very irregular and no stationary (periodic)
regime is ever reached. This is illustrated in Fig.\ \ref{fig:chaos},
where we plot the jump number $n$ as a function of time (iteration
step) for two orbits in the region where the fixed points are
unstable. In this figure we clearly see that the jump number
fluctuates erratically around a mean value. We have computed the
Lyapunov exponent for several values of parameters in the region of
unstable fixed points and have found it to be positive for all cases
studied, thus indicating that the motion is indeed chaotic in this
region.
\begin{figure}
\centerline{\hbox{\vbox{\psfig{figure=fig3.ps,width=3.0truein,height=2.5true in}}}}
\vspace{-0.5true cm}
\caption{The jump number $n$ as a function of time (measured in iteration
steps) in the chaotic regime. Here $e_t=0.35$, $\tau=3.73$, and
$t=1.65$ (lower orbit), 1.74 (upper orbit).}
\label{fig:chaos}
\end{figure}
The different dynamical regimes displayed by the model above can be
conveniently summarized in terms of a ``phase diagram'' in the
parameter space $(e_t,t/\tau)$, as shown in Fig.\ \ref{fig:phase}. In
this figure we plot the curves corresponding to $t_\infty$ (solid
line) and $t_{\rm inst}$ (dashed line) given by Eqs.\ (\ref{eq117})
and (\ref{eq119}), respectively. Also plotted is the curve
representing the inclination $t_2$ (dot-dashed line) at which the
fixed point with $n=2$ first appears. Thus in terms of the
existence/stability of the fixed points the model displays the
following four regions: (i) for $0<t<{\rm min}(t_2,t_{\rm inst})$
there is a unique stable fixed point, namely, that with $n=1$; (ii)
for $t_2<t<{\rm min}(t_{\rm inst},t_\infty)$ there are multiple stable
fixed points (at least those with $n=1$ and $n=2$); (iii) for $t_{\rm
inst}<t<t_\infty$ all existing fixed point are unstable and chaotic
motion is observed; (iv) for $t>t_\infty$ no fixed point exists and
the motion becomes accelerated.
Another interesting feature in Fig.\ \ref{fig:phase} is the fact that
the chaotic regime appears when the collisions are highly inelastic
(i.e., small $e_t$). In particular, for $e_t>\sqrt{2}-1$ (at which point
$t_{\rm inst}$ equals $t_\infty$) the fixed points remain stable over
their entire domain of existence. (The results shown in Fig.\
\ref{fig:phase} are qualitatively different from the behavior seen in
the model studied in Ref.\ \cite{valance}, where chaotic motion
appears as the restitution coefficient increases.)
\begin{figure}
\vspace{-2.0true cm}
\centerline{\hbox{\vbox{\psfig{figure=fig4_eps.ps,width=3.0true in,height=3.5true in}}}}
\vspace{-0.7true cm}
\caption{Phase diagram for the model. The solid line corresponds to
$t_\infty$, the dashed line to $t_{\rm inst}$, and the dot-dashed line
to $t_2$.}
\label{fig:phase}
\end{figure}
\section{Comparison with experiments}
In this section we wish to compare our model with recent experimental
studies of a single ball moving under gravity on a rough inclined
surface. In these experiments, first performed by Jan {\it et al.}
\cite{jan} and later expanded by Ristow {\it et al.} \cite{ristow}, a
rough surface was constructed by gluing steel spheres of radius $r$ on
a L-shaped flume. Another steel sphere of radius $R$ was then launched
with a small initial velocity and its subsequent motion analyzed. As
the surface inclinations increases, the following three regimes are
observed \cite{ristow}: for small inclinations the bead always stops
(regime A), then comes a range of inclinations for which the ball
reaches a steady state with constant mean velocity (regime B), and
beyond this point the ball starts to jump (regime C). In Fig.\
\ref{fig:fit} we show data taken from Ref.\ \cite{ristow} for the ball
mean velocity $\overline{V}$ as a function of $\sin \theta$, where
$\theta$ is the inclination angle with respect to the horizontal
direction. As discussed in Ref.\ \cite{ristow}, the change in trend
observed in the data as $\theta$ increases (for a given value of
$R/r$) marks the beginning of the jumping regime.
The regime B seen in the experiments corresponds in our model to a
stable fixed point with $n=1$, for in this case the particle reaches a
periodic motion where it falls one step at a time (as in the
experiments). In order to compare our model more closely with the
experiments let us first express the mean velocity $\overline{V}_1$
(at the fixed point $n=1$) in terms of the angle $\theta$, where
$\theta=\phi+\pi/2-\alpha$ (see Fig.\ 1). Setting $n=1$ in Eq.\
(\ref{eq:Vn}), returning to dimensionful units via Eq.\
(\ref{eq:Vdimless}), and expressing the final result in terms of
$\theta$, we obtain
\begin{equation}
\overline{V}_1=\left[{L g (1+e_t)\over{2(1-e_t)}}\right]^{1/2}
\sqrt{\sin\theta-\tau^{-1} \cos\theta}. \label{eq:V1}
\end{equation}
(We remark parenthetically that a similar expression can be obtained
heuristically if one introduces an effective sliding friction in
addition to inelastic collisions; see Refs. \cite{jan,ristow}. Our
formula follows however from a pure collision model.)
We have fitted the expression (\ref{eq:V1}) to the experimental data
shown in Fig.\ \ref{fig:fit} --- the corresponding results being
displayed as solid curves in this figure. In our fitting procedure, we
took $L=2r=1$ cm \cite{ristow}, $g=980$ cm/s$^2$, and best-fitted the
parameters $\tau$ and $e_t$ for each data set considering {\it only}
points in regime B. As we see in Fig.\ \ref{fig:fit}, the model
prediction for the dependence of $\overline{V}$ with $\theta$ is in a
good agreement with the experimental data (in regime B).
The jumping regime observed in the experiments, on the other hand,
would correspond in our model to the region of unstable fixed points,
since in this case the particle jumps erratically never reaching a
steady state (see Fig.\ \ref{fig:chaos}). This analogy might then
provide a possible explanation for the change in trend observed in the
experimental data for large inclinations. To see
this, consider the region of small $e_t$ in the phase diagram shown in
Fig.\ \ref{fig:phase}. As the inclination $t$ increases (for a given
$e_t$) the system goes from a region of stable periodic motion (with $n=1$)
to a regime of chaotic jumps, in close resemblance to the experimental
transition from steady-state to the jumping regime.
To probe this analogy further, we illustrate in Fig.\ \ref{fig:vmean}
the behavior predicted by the model for the mean velocity
$\overline{V}$ as a function of $\sin \theta$ in the region of small
$e_t$. In this figure, the solid curve corresponds to the expression
(\ref{eq:V1}) for $\overline{V}_1$, up to the point where the fixed
point goes unstable, and the crosses are computed values of
$\overline{V}$ in the ensuing chaotic regime. Comparing Fig.\
\ref{fig:vmean} with Fig.\ \ref{fig:fit}, we see that the change in
behavior predicted by the model at the onset of instability is in
qualitative agreement with what is observed in the experiments (for
small values of $R/r$) as the ball enters the jumping regime. Of
course, more detailed experiments are necessary to verify whether
chaotic motion does indeed take place in the jumping regime.
\begin{figure}
\centerline{\hbox{\vbox{\psfig{figure=fig5.ps,width=3.0true in,height=2.5true in}}}}
\vspace{-0.5true cm}
\caption{Mean velocity $\overline{V}$ (cm/s) as a function of
$\sin \theta$. Points are experimental data taken from Ref.\ [4] for
$R/r=2$ ($+$), 1.5 ($\ast$), 1 ($\triangle$), 0.8 ($\Diamond$). Solid
curves are theoretical fits [Eq.\ (\ref{eq:V1})], ending near
the last data point considered in the fit. Fitted parameters are
$(e_t,\tau)$ = (0.72, 33.18), (0.64, 21.09), (0.41, 10.17),
(0.27, 7.07), from left to right.}
\label{fig:fit}
\end{figure}
\begin{figure}
\centerline{\hbox{\vbox{\psfig{figure=fig6.ps,width=3.0true in,height=2.5true in}}}}
\vspace{-0.5true cm}
\caption{Same as figure \ref{fig:fit} for our model with $e_t=0.1$ and
$\tau=1$. The solid curve corresponds to Eq.\ (\ref{eq:V1}) whereas the
stars give the computed mean velocity in the chaotic regime.}
\label{fig:vmean}
\end{figure}
\section{Conclusions}
We have studied a simple geometrical model for the gravity-driven
motion of a single particle on a rough inclined line. In our model the
rough line was chosen to have a regular staircase shape and a simple
collision law was adopted. With these simplifications the dynamics is
described by a one-dimensional map that is quite amenable to
analytical treatment. Summarizing our findings, we have seen that our
model displays the following four dynamical regimes:
\begin{enumerate}
\item for $0<t<{\rm min}(t_2,t_{\rm inst})$ there is a unique stable
fixed point.
\item for $t_2<t<{\rm min}(t_{\rm inst},t_\infty)$ the system has
multiple stable fixed points.
\item for $t_{\rm inst}<t<t_\infty$ the fixed points are unstable
and the dynamics is chaotic.
\item for $t>t_\infty$ no fixed point exists and the motion becomes
accelerated.
\end{enumerate}
Here the parameter $t$ measures the surface inclination and the
quantities $t_2$, $t_{\rm inst}$, and $t_\infty$ separating the
different regimes are given in terms of the other two model-parameters, namely,
the restitution coefficient $e_t$ and the roughness parameter
$\tau$. These regimes are indicated in the phase diagram shown in
Fig.\ \ref{fig:phase}. Furthermore, it can be shown
\cite{companion} that the above conclusions, which were derived in the
context of a simple collision rule, remain valid for a wide class of
tangential restitution laws.
Despite its simplicity, our model does provide a theoretical framework
within which the generic behavior seen in experiments on a ball
moving on a rough surface can be qualitatively understood. For
example, the model successfully predicts the existence of several
dynamical regimes that are also observed in the experiments. In
particular, the predicted functional dependence of the mean velocity
with the inclination angle $\theta$ (in the steady-state regime) is in
good agreement with the experiments. Moreover, the model provides a
possible explanation for the change in trend seen in the experimental
data as the ball enters the jumping regime. We have suggested that
this jumping regime might correspond to a chaotic motion, as so
happens in the model. Clearly, more experimental studies are required
to investigate this interesting possibility.
This work was supported in part by FINEP and CNPq.
|
\section{Introduction}
\noindent In this paper we describe an extension of the spin
network states to supergravity. The spin network states play a
fundamental role in non-perturbative quantizations of both gauge
theories \cite{KS,bae1} and gravitational theories
\cite{spain,sn1,sn2}. In the gauge theory context they provide an
orthonormal basis for lattice gauge theories \cite{KS,bae1}. In this
case the spin networks are labeled graphs on the lattice, whose
edges are labeled by the finite irreducible representations of the
gauge group $G$. In quantum gravity diffeomorphism invariance
reduces the degrees of freedom, so that a basis of states
invariant under spatial diffeomorphisms and local frame rotations
are given by the diffeomorphism classes of spin networks
\cite{sn1,sn2}. In this case the group is $SU(2)$, for the chiral
formulation based on the Ashtekar-Sen variables \cite{sen,abhay},
or $SU(2)+SU(2)$ in the relativistic
case\cite{barrettcrane1,hologr}.
Over the last ten years there has been a great deal of progress
in our understanding of the non-perturbative structure of quantum
general relativity, leading to the complete formulation of the
quantum theory\footnote{For recent reviews see
\cite{carlo-review,future}.}. Among the key results
are the discovery that diffeomorphism invariant observables that
measure aspects of the spatial geometry such as areas of surfaces
and volumes of regions are finite, and have discrete, computatable
spectra\cite{spain,sn1,sn2,vol2}.
This has led to a physical understanding of the spin
network states as eigenstates of these geometrical
observables.
During this period there have been a number of papers which extend
the methods used to
supergravity\cite{superstuff,GSU1,GSU2,SUG2,N=4}. These have
included the formulation of ${\cal N}=1,2$\cite{GSU1,GSU2,SUG2},
and ${\cal
N}=4$\cite{N=4} supergravity in terms of chiral, Ashtekar-Sen
like variables, as well as the discovery of exact solutions to the
quantum constraints\cite{GSU1,SUG2}. However, much more remains to
be done in this direction. The non-perturbative quantization of
gravitational theories with extended supersymmetry is largely
unexplored territory, despite the fact that extended supersymmetry
is essential to the success of string theory, which remains the
only successful technique for investigating quantum gravity in the
perturbative regime. Another important open area of investigation
is the properties of $BPS$ states in the non-perturbative regime.
This could be very interesting as it could provide a way to
compare results on black hole entropy gotten by both string
theory\cite{string-bh} and loop quantum gravity\cite{kirill-bh}.
In this paper we take a first step to the study of the
non-perturbative quantization of supersymmetric theories of
gravitation by constructing the spin network states for $N=1$
supergravity. We find a number of new features, which suggest
that this could be a fruitful direction of investigation. The
main result is a diagrammatic method for the construction and
evaluation of spin networks for the supergroup $Osp(1|2)$. As a
first example we construct and partly diagonalize the
supersymmetric extension of the area operator. As expected the
spectrum is discrete, but different from that of quantum general
relativity. This means that experimental probes of geometry at the
Planck scale could in principle distinguish different hypotheses
about the local gauge symmetry. This is highly interesting in
light of recent developments that suggest that astrophysical
probes of Planck scale physics can be developed\cite{giovanni}.
Another possible application of the formalism given here is to
supersymmetric Yang-Mills theory. It will be very interesting to
investigate the extent to which the physics of $N=2$ and $N=4$
super-Yang-Mills theory can be expressed in terms of the spin network
states.
It is straightforward to extend the construction here to $N=2$ and
higher supersymmetry, this will be described elsewhere \cite{yi2,
yi3}. Also, in progress \cite{SUG1} is an examination of the
canonical and boundary structure of $N=1,2$ quantum supergravity,
which extends results on a holographic formulation of quantum
general relativity with finite cosmological constant
\cite{linking,hologr}.
In the next section we review some of the basic results about
$N=1$ supergravity in chiral coordinates, first studied by
Jacobson\cite{superstuff}. In section 3 we present some results
from the representation theory of $Osp(1|2)$ which allow us in
section 4 to construct quantum spin networks. The diagrammatic
method for doing calculations with these states is introduced in
section 5, and the following sections describe examples and
calculations.
Finally, we mention that we do not here provide rigorous proofs for
the assertions made, but we see no reason why a straightforward
extension of the rigorous methods
introduced in \cite{rayner,chrisabhay,gangof5,thomas} to the present
case should not be possible.
\section{Review of Quantum Supergravity}
Supergravity in terms of the new variables maybe was initially
investigated in \cite{superstuff} and extended in \cite{GSU1,GSU2}.
In this paper we will consider mainly $N=1$ supergravity. As
shown first by Jacobson in \cite{superstuff}, this can be
formulated in chiral variables which extend the Ashtekar-Sen
variables of general relativity. In this formulation, the
canonical variables are the left handed $su(2)$ spin connection
$A_a^i$ and its superpartner spin-3/2 field $\psi_a^A$. As shown
in \cite{SUG2} these fit together into a connection field of the
superlie algebra $Osp(1|2)$ (which is referred to in some
references \cite{GSU1,SUG2,GSU3} as $GSU(2)$.)
We thus define the graded connection:
\begin{equation}
\cal{A}\mit_a:=A_a^iJ_i+\psi_a^AQ_A
\end{equation}
where $a$ is the spatial index. If $\widetilde{E^a_i}$ and
$\pi^a_A$ are momenta of $A_a^i$ and $\psi_a^A$ respectively, we
can define the graded momentum as:
\begin{equation}
\cal{E}\mit^a:=\widetilde{E^a_i}J^i+\pi^a_AQ^A
\end{equation}
The constraints that generate local gauge
transformations can then be expressed as usual as,
\begin{equation}
{\cal G}_{i} =
D_a{\widetilde{E^a_i}}+\frac{i}{\sqrt{2}}\pi^a_A\psi_{aB}\tau^{AB}_i=0
\label{gauss}
\end{equation}
The left and right handed supersymmetry transformations are
generated by\cite{superstuff},
\begin{equation}
{\cal L}_{A} =
D_a\pi^a_A-ig{\widetilde{E^a_i}}\tau^B_{iA}\psi_{aB}=0 \label{lss}
\end{equation}
\begin{equation}
{\cal R}^{A} =
\epsilon^{ijk}\widetilde{E^a_i}\widetilde{E^b_j}\sigma^A_{kB}
(-4iD_{[a}\psi^B_{b]} +\sqrt{2} g\epsilon_{abc}\pi^{cB})=0
\label{rss}
\end{equation}
where the cosmological constant is given by $\Lambda=-g^2$. The
diffeomorphism and hamiltonian constraints can be derived by
taking the Poisson Brackets of (4) and (5).
These may be written simply in terms of the fundamental
representation of $Osp(1|2)$, which is $3$ dimensional. The
superlie algebra $Osp(1|2)$ is then generated by five $3\times 3$
matrices $G_I(I=1...5)$, given explicitly in \cite{SUG2}. Using
them we can define \begin{equation} \cal{A}\mit_a^I=(A_a^i,\psi_a^A) \end{equation} \begin{equation}
\cal{E}\mit^a_I=(\widetilde{E^a_i},\pi^a_A) \end{equation} where $I=(i,A)$
labels the five generators of $Osp(1|2)$.
Then the first two constraints can be combined into one $Osp(1|2)$
Gauss constraint: \begin{equation} \cal{D}\mit_a\cal{E}\mit^a_I=0 \label{sgauss}
\end{equation}
while the last one combines with the
Hamiltonian constraint to give:
\begin{equation} \cal{E}\mit^a\cal{E}\mit^b\cal{F}\mit_{ab}-ig^2
\epsilon_{abc}\cal{E}\mit^a\cal{E}\mit^b \cal{E}\mit^c=0 \end{equation} where
$\cal{F}\mit_{ab}$ is the curvature of the super connection
$\cal{A}\mit_a$ : \begin{equation}
\cal{F}\mit_{ab}:=d_a\cal{A}\mit_b+[\cal{A}\mit_a,\cal{A}\mit_b]
\end{equation}
A key feature of this kind of approach to supergravity is that the
supersymmetry gauge invariance has been split into two parts,
which play rather different roles. The left handed supersymmetry
transformations generated by eq. (\ref{lss}) combine with the
$SU(2)$ Gauss's law (\ref{gauss}) to give a local $Osp(1|2)_{L}$
gauge invariance. The theory is then written so that the
associated $Osp(1|2)$ connection is the canonical coordinate. The
right handed part of the supersymmetry, generated by (\ref{rss}),
is a dynamical constraint, being quadratic rather than linear in
the momentum. It joins with the Hamiltonian constraint
to form a left handed supersymmetry multiplet of
dynamical operators.
It is then natural in a chiral formulation of supergravity to
represent the left-handed supersymmetry kinematically, and solve
it completely by expressing the theory completely in terms of
$Osp(1|2)$ invariant states. These are the spin network states we
will present shortly. The remaining, right handed, part of the
supersymmetry is then imposed as a dynamical operator, and has the
same status as the Hamiltonian constraint.
The loop representation for supergravity in the chiral
representation was constructed in \cite{SUG2} in terms of
$Osp(1|2)$ Wilson loops. These are defined in terms of the
supertrace taken in the fundamental $3$ dimensional representation
of $Osp(1|2)$.
\begin{equation}
\cal T\mit[\gamma]=Str\cal P\mit
exp(\oint_{\gamma}ds \cal A\mit_a {\gamma}^a) \equiv Str{ \cal
U\mit}_{\gamma}({\cal A\mit })
\end{equation}
These Wilson loop states are subject to additional relations
arising from intersections of loops. These are solved completely
by the introduction of the spin network basis, which are complete
and orthogonal\cite{sn2}.
We can construct the loop-momentum variables by inserting the
$Osp(1|2)$ invariant momentum $\cal E\mit^a$ into the Wilson loops: \begin{equation}
\cal T\mit^a[\alpha](s)=Str[\cal U\mit_{\alpha}(\cal A\mit)\cal
E\mit^a(\alpha(s)] \end{equation}
It is straightforward to show that the
$\cal T\mit[\gamma]$ and $\cal T\mit^a[\alpha](s)$ form a closed
algebra under Poisson brackets, which we will call the $N=1$
super-loop algebra.
We will also need to describe operators quadratic in the conjugate
momenta, which in the loop representation are formed by inserting
two momenta in the loop trace,
\begin{equation}
\cal T\mit^{ab}[\alpha](s,t)=Str[\cal U\mit_{\alpha}(s,t)\cal
E\mit^a(\alpha(t)) \cal U\mit_{\alpha}(t,s)\cal
E\mit^b(\alpha(s))]
\end{equation}
The higher order loop operators are similarly defined as
\begin{equation}
\cal T\mit^{ab...c}[\alpha](s,t,...v)=Str[\cal U\mit_{\alpha}(s,t)\cal
E\mit^a(\alpha(t)) \cal U\mit_{\alpha}(t,u)\cal
E\mit^b(\alpha(u))...\cal U\mit_{\alpha}(v,s) \cal
E\mit^c(\alpha(s))]
\end{equation}
As discussed in \cite{SUG2}, the supersymmetric extension of the
Chern-Simons state may be formed from the Chern-Simons form of the
superconnection ${\cal A}_{a}$,
\begin{equation}
\Psi_{SCS}(\cal{A}\mit_a)=exp[\frac{i}{2\Lambda}\int
d^3xSTr(\cal{A}\mit\wedge\cal{F}\mit-\frac{1}{3}
\cal{A}\mit\wedge\cal{A}\mit\wedge\cal{A}\mit)]
\end{equation}
This state is an exact solution to all the quantum constraints.
Like the ordinary Chern-Simons state it also has a semiclassical
interpretation as the ground state associated with DeSitter or
Anti-DeSitter spacetime.
\section{Finite Dimensional Irreducible Representation of $Osp(1|2)$.}
Spin networks may be constructed for any Lie or Superlie algebra,
$\cal A$ by extending the original definition \cite{pen1,bae1}. An
$\cal A$-spin network is a labeled graph whose edges are labeled
by the finite dimensional irreducible representations of $\cal A$
and whose nodes are labeled by the associated intertwiners. In
quantum gravity spin networks states are associated with the gauge
group of the connection, which we have seen in the case of $N=1$
supergravity in the chiral representation \cite{superstuff} to be
$Osp(1|2)$. The representation theory of $Osp(1|2)$ has been
studied in detail in \cite{GSU3,GSU4,GSU5,GSU7}, here we give some
of the basic facts that we will need to construct the associated
spin network states.
The superlie algebra of $Osp(1|2)$, is constructed by three
bosonic generators $J_i$ (i=1,2,3)and two fermionic generators
$Q_A(A=0,1)$. The commutation relations are :
\begin{equation}
[J_i,J_j]=i\epsilon_{ijk}J_k
\end{equation}
\begin{equation}
[J_i,Q_A]=1/2(\tau_i)_A^BQ_B
\end{equation}
\begin{equation}
\{Q_A,Q_B\}=1/2\epsilon_{AB}\tau^iJ_i
\end{equation}
where $\tau^i$ are Pauli matrices.
Each irreducible representation of the $Osp(1|2)$ contains two
adjacent $SU(2)$ representations. One is labeled by spin $J$ and
the other by $J-1/2$. $J$ may be taken as the label of the
$Osp(1|2)$ representation, and is related to the eigenvalues of
the quadratic Casmier operator of the supergroup $Osp(1|2)$: \begin{equation}
C=J^iJ_i+\epsilon^{AB}Q_AQ_B \end{equation} by \begin{equation} \hat{C}|J>=J(J+1/2)|J> \end{equation}
For each $J$ the representation is a graded vector space with a basis
labeled by $|J;L;M >$,
where J is an integer or half-integer, and L can be
$J$ or $J-1/2$ and $-L\le M\le L$.
The dimension of the space of the representation with spin
$J$ is $4J+1$.
The usual rules for combination of angular momentum can be
extended directly to these states. The result is a super
Racah-Wigner calculus which gives the results of decompositions of
products of representations of $Osp(1|2)$. The tensor product is
completely reducible and is given by \begin{equation}
j_1\textstyle{\bigotimes} j_2=|j_1-j_2|\textstyle{\bigoplus}
|j_1-j_2+1/2|\textstyle{\bigoplus} ...|j_1+j_2| \label{tp} \end{equation}
Note that this
differs from the familiar $SU(2)$ case in that representations
which differ from
$j_{1}+j_{2}$ by both integers and half integers are included.
The Clebsch-Gordon
coefficients for the expansion of the basis elements are
determined uniquely, from
their values for $SU(2)$.
Next we consider the reduction of the tensor product of three
irreducible representations $(j_1,j_2,j_3)$. As in the $SU(2)$
case we have two different recoupling schemes. One can couple the
representations $(j_1,j_2)$ into $j_{12}$ first and then couple
the result to $j_3$ to give the final representation; or one
couples $(j_2,j_3)$ into $j_{23}$ first and then couples to $j_1$
next. These two representations are related to each other by the
Racah sum rule in terms of super rotation 6-symbols. For
$Osp(1|2)$, the parity independent super-rotation 6-symbols are
defined as[21]:
\begin{equation}
{\left\{\begin{array}{ccc}
j_1&j_2&j_{12}\\j_3&j&j_{23}\end{array}\right\}}^s=
(-1)^{\Phi(\lambda_1,\lambda_2,\lambda_3)}
{\left\{\begin{array}{ccc}
j_1\lambda_1&j_2\lambda_2&j_{12}\lambda_{12}\\j_3\lambda_3&j\lambda&j_{23}
\lambda_{23}
\end{array}\right\}}^s
\end{equation}
Then the Racah sum rule reads:
\begin{equation}
{\left\{\begin{array}{ccc}
j_1&j_2&j_{12}\\j_3&j&j_{23}\end{array}\right\}}^s=
\sum_{j_{13}}(-1)^{\Theta}
{\left\{\begin{array}{ccc}
j_1&j_3&j_{13}\\j_2&j&j_{23}\end{array}\right\}}^s
{\left\{\begin{array}{ccc}
j_2&j_1&j_{12}\\j_3&j&j_{13}\end{array}\right\}}^s
\end{equation}
In a similar way the Biedenharn-Elliott identity can be constructed for
the super 6-j symbols:
\begin{eqnarray}
&& {\left\{\begin{array}{ccc}
j_1&j_2&j_{12}\\j_3&j_{123}&j_{23}\end{array}\right\}}^s
{\left\{\begin{array}{ccc}
j_{23}&j_1&j_{123}\\j_4&j&j_{14}\end{array}\right\}}^s\nonumber\\
&&=\sum_{j_{124}}(-1)^{\theta_{be}} {\left\{\begin{array}{ccc}
j_2&j_1&j_{12}\\j_4&j_{124}&j_{14}\end{array} \right\}}^s
{\left\{\begin{array}{ccc}
j_3&j_{12}&j_{123}\\j_4&j&j_{124}\end{array}\right\}}^s
{\left\{\begin{array}{ccc}
j_{14}&j_2&j_{124}\\j_3&j&j_{23}\end{array}\right\}}^s
\end{eqnarray}
where $\theta_{be}$ and $\Theta$ are the sign factors related to
the super-spins involved. The interesting fact is that except
these two sign factors, the structure of two relations are the
same as the structure for SU(2) rotation algebra. Therefore when
we restrict the sum of three super-spins in all the triangles
$(j_1,j_2,j_{12})$, $(j_1,j_3,j_{13})$, $(j_2,j_3,j_{23})$ to be
integers, then all the expressions go back to the normal racah sum
rule and the Biedenharn-Elliott identity for $SU(2)$.
\section {Spin Network States of $N=1$ Supergravity}
We recall that a spin network state of quantum general relativity,
denoted $|\Gamma >$ consists of an embedding of closed graph
$\Gamma$ into a fixed three manifold $\Sigma$ with edges labeled
by the representation of SU(2)($SU(2)_{q}$) and vertices labeled
by intertwining operators, namely distinct ways to decompose the
incoming representations into a singlet. Here we define the {\it
super} spin networks in the same way only by replacing the $SU(2)$
with superlie algebra $Osp(1|2)$.\footnote{It is also possible to
extend the construction to the ``quantum graded group,'' $Osp(1|2)_{q}$.
We do not carry this out here.} The elements
of the super spin networks are links and
vertices. Notice that in quantum general relativity, the link of
color $n$ corresponds to a parallel propogator of connection $A_a$
along this path in the spin n/2 representation of su(2), here
associated to very link we also label a color $n_i$ which is two
times as the superspin $j_i$ which labels the representation of
$Osp(1|2)$. For every vertex $v_e$, there are incoming links with
color $n^{in}_{ei}$ and outcoming links with color $n^{out}_{ei}$.
so we can label the vertex by the total color
$v_e=\sum_{i}n^{in}_{ei}-\sum_{i}n^{out}_{ei}$ which satisfies
$1/2\le v_e\le k/2$.
Corresponding to every super spin network $(\Gamma^{sg},n_i,v_e)$,
there is a super spin network state $<\Gamma^{sg},n_i,v_e|$ in the
Hilbert space of the supergravity. As an independent basis, the
super spin network can also be expressed as the bras so that a
general state in this representation is given by: \begin{equation}
\Psi[\Gamma^{sg}]:=<\Gamma^{sg}|\Psi> \end{equation}
We now list some basic facts about the super spin networks .
\begin{itemize}
\item{} As in the $SU(2)$ case there is no intertwiner associated to
trivalent nodes because
given
\begin{equation}
|J_1-J_2|\le J_3\le |J_1+J_2|
\label{triangle}
\end{equation}
then the map from the tensor products of two
representations to the reduced representation
is unique.
\item{} The condition that the sum of three colors of the links adjacent
to a trivalent vertex
must sum to an even number does not hold, because both even and odd
spins appear in the sum (\ref{tp}). As the color of the link is
twice the spin,
the edges of a trivalent vertex
can be any integers such that eq. (\ref{triangle}) holds.
\item{} If the valences adjacent to the same
vertex is more than three, intertwiners are needed to label the
different maps from the incoming representations to the singlet
state. As in the $SU(2)$ case the multi-valent vertices can be
decomposed in terms of trivalent vertex connected by internal
edges, as described in \cite{sn2}.
\end{itemize}
Note that this means that there is no simple way to decompose the
$Osp(1|2)$ spin networks completely in terms of ordinary spin
networks because there is no ordinary spin network vertex
corresponding to the superspin network vertices where the sum of
incident colors is odd. However, there is still a very useful
decomposition, which we will give below.
As in the $SU(2)$ case, there is a recoupling theory based on the
Racah sum rule and
Biedenharn-elliott identity in terms of the {\it super}-rotation
6-j symbols. We can
express the recoupling theory by fig.1:
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=14cm,height=4cm]{fig005.eps}
\caption{Recoupling theory for $Osp(1|2)$ 6-j symbols} \end{center}\end{figure}
Where the sum is over labels such that the super 6-j symbols
satisfies the triangle inequalities eq. (\ref{triangle}).
\section{Graphic Representation of the Super Spin Networks}
We can now give a diagrammatic notation for $Osp(1|2)$ spin
networks which is useful for computation. We follow the method
developed in \cite{sn2} and elaborated in \cite{vol2} for quantum
general relativity in which a diagrammatic notion for $SU(2)$ spin
network states was developed by modifying notations used by
Penrose\cite{pen1} and Kauffman and Linns\cite{KL}. The result is
a diagrammatic notation of super spin networks based on the
connection between them and the representation theory of the
supergroup $Osp(1|2)$.
\subsection{Element of the Diagrams}
The basic fact about the $SU(2)$ representation theory on which
the Penrose and Kauffman and Linns notation rests is that all
irreducible representations can be obtained by symmetrizing
products of the fundamental representation. In the case of
$Osp(1|2)$ all irreducible representations can be obtained via a
process of graded symmetrizing, in which there are extra signs for
even and odd parts of the representations. There are in fact two
different fundamental representations for the $Osp(1|2)$, which
are complex conjugates of each other. Let us consider first the
left handed fundamental representation. It is a three dimensional
graded vector space, whose elements may be written \begin{equation}
\xi_\alpha=(\begin{array}{c} \psi_A \\ \phi_{\circ}
\end{array})
\end{equation} where $A=( 0,1)$ denotes the left handed $SU(2)$ spinor
index part and $\alpha= \circ$ denotes the third component. Here
we take the $\xi_{A}= \psi_A$ to be fermionic while the
$\xi_{\circ} = \phi$ is bosonic. The grade of the index,
$g(\alpha)$, is defined to be one for $\circ$ and zero for $A$.
Under the action of $Osp(1|2)$, $\xi_\alpha$ transforms as: \begin{equation}
\xi_{\alpha'}={U_{\alpha'}}^\alpha\xi_{\alpha} \end{equation} where $3\times
3$ matrix $U_{\alpha'}^\alpha$ is an element in the fundamental
representation of $Osp(1|2)$.
The higher irreducible representations are formed by taking graded
symmetric products of this fundamental representation.
For instance the basis states for J=1 span a five dimensional space,
which can be
constructed by symmetrized tensor products of two states in the
fundamental representation, as,
\begin{equation}
\xi_{(\alpha\beta)}:=\frac{1}{2}[\xi_{\alpha}\xi_{\beta}+(-1)^{g(\alpha)g(\beta)}
\xi_{\beta}\xi_{\alpha}]
\end{equation}
We can then read off the components of the basis states of the $J=1$
representation. They consist of a pair of $SU(2)$ representations,
given by,
\begin{equation}
\xi_{(\alpha\beta)}=(\xi_{(AB)},\xi_{(A}\phi_{\circ)})
\end{equation}
The first term is the bosonic component defined as,
\begin{equation}
\xi_{(AB)}=\frac{1}{2}(\psi_A^{(1)}\psi_B^{(2)}+\psi_B^{(1)}\psi_A^{(2)})
\end{equation}
and the second term is the fermionic component of the basis states.
\begin{equation}
\xi_{(A}\phi_{\circ)}=\frac{1}{2}(\psi_A^{(1)}
\phi_\circ^{(2)}+\phi_\circ^{(1)}\psi_A^{(2)})
\end{equation}
The other term $\phi_{[\circ \circ]}$ vanishes due to the antisymmetrization.
Under the action of $Osp(1|2)$, the states transform as: \begin{equation}
\xi_{(\alpha'\beta')}={U_{(\alpha'\beta')}}^{(\alpha\beta)}\xi_{\alpha\beta}
\end{equation} where:
\begin{eqnarray}
{U_{(\alpha'\beta')}}^{(\alpha\beta)}&=&\frac{1}{2}[(-1)^{g(\alpha)
[g(\beta')-g(\beta)]}
{U_{\alpha'}}^{\alpha}{U_{\beta'}}^{\beta}\nonumber\\&&+(-1)^{g(\alpha')
g(\beta')}(-1)^{g(\alpha)[g(\alpha')-g(\beta)]}
{U_{\beta'}}^{\alpha}{U_{\alpha'}}^{\beta}] \end{eqnarray}
If we only consider the unit element of $Osp(1|2)$ in this
representation, then we have \begin{equation}
{\delta_{(\alpha'\beta')}}^{(\alpha\beta)}=\frac{1}{2}[
{\delta_{\alpha'}}^{\alpha}{\delta_{\beta'}}^{\beta}+(-1)^{g(\alpha')g(\beta')}
{\delta_{\beta'}}^{\alpha}\delta_{\alpha'}^{\beta}] \end{equation} This
allows us to generalize the Penrose diagrammatic notation for
$SU(2)$ spin networks. We indicate the elements of a super spin
networks by bold lines, the elements with su(2) indices by thin
lines and third component $\phi=\xi_{\circ}$ by dotted lines. Then
we can denote the $\delta_{\alpha'}^{\alpha}$ and its components
as fig.2.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=4cm]{fig002.eps}
\caption{Unit element in fundamental representation of $Osp(1|2)$}
\end{center}\end{figure}
Then it's straightforward to express (35) as fig.3.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=4cm]{fig010-1.1.eps}
\caption{Construction of unit element in representation with spin one}
\end{center}\end{figure}
Let us consider the component formulation of this expression. When
the indices of delta are $SU(2)$ spinor indices, it's easy to see
that it goes back to the normal spin networks expression. If
one index is fermionic and the other one is bosonic, they
commute with other and we can denote the expression by two vertical lines,
one solid and one dotted. If both indices are bosonic, which may
be denoted by two vertical dotted lines. However this term
vanishes because the graded symmetrization antisymmetrizes them
and there is a single bosonic component. The procedure of the
decomposition of the super element can then be drawn as fig.4.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=4cm]{fig001-1.1.eps}
\caption{Decomposition of the link with color 2 in super spin
networks into normal SU(2) ones} \end{center}\end{figure}
Thus, we have a way to decompose the
diagrams for $Osp(1|2)$ spin networks into combinations of $SU(2)$
spin network diagrams and dotted lines representing the single
bosonic component of the fundamental representation. It is
straightforward to see how this works when applied to any higher
dimensional representation of $Osp(1|2)$, which is gotten by
making a graded symmetrization of $n$ fundamental representations.
The basic property is that all the terms whose corresponding
graphs have two or more dotted lines must vanish also since we
need to antisymmetrize them. As a result, the basis states in any
dimensional representation consists of two components, \begin{equation}
\xi_{(\alpha_1...\alpha_n)}=(\xi_{(A_1...A_n)},\xi_{(A_1...A_{n-1}0)})
, \end{equation} where \begin{equation}
\xi_{(A_1...A_n)}=\psi_{(A_1}\psi_{A_2}...\psi_{A_n)} \end{equation} \begin{equation}
\xi_{(A_1...A_{n-1}0)}=\psi_{(A_1}\psi_{A_2}...\psi_{A_n-1}\phi_{0)}
. \end{equation}
The unit element of the supergroup in this representation can be
expressed as:
\begin{equation}
{\delta_{(\alpha'\beta'...\gamma')}}^{(\alpha\beta...\gamma)}:=
{\delta_{(\alpha'}}^{\alpha}
{\delta_{\beta'}}^{\beta}...{\delta_{\gamma')}}^{\gamma}
\end{equation}
and the corresponding graph can be drawn as fig.5. Thus we see
that we can decompose a super spin network into a sum of diagrams,
each of which is a normal spin network together with dotted lines.
In this decomposition each edge of the superspin network, with
color $n$ becomes two ordinary spin network edges, the first an
$n$ line without a dotted line and the second with an $n-1$ line
with a single dotted line. This is shown in fig.5.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=4cm]{fig003.1.eps}
\caption{Decomposition of the super link of color $n$}
\end{center}\end{figure}
\subsection{Trivalent Vertices}
Next we consider the tri-valent vertex. As there is no restriction
that the incident colors must add up to an even number, as in the
$SU(2)$ case, the simplest trivalent node is the one in which all
three edges have color one. This node can be visualized in two
ways, depending on how the direction of time is read. One fermion
with spin one half meets one boson with spin zero and then changes
into one fermion, or two fermions with spin one half meet together
forming into a boson which is also singlet state. These processes
are expressed by fig.6.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=4cm]{fig007-2.1.eps}
\caption{The three terms in the decomposition of the trivalent
vertex, in the case that all colors are equal to one.}
\end{center}\end{figure}
We next consider the case in which every link has color two. This
can be decomposed into the ordinary $SU(2)$ spin networks as shown
in fig.7.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{fig009.1.eps}
\caption{ Decomposition of trivalent vertex in which every link
has color two.}
\end{center}\end{figure}
In general, if the sum of the three colors is even it can be
decomposed into four terms, each of which contains an ordinary
spin network plus, possible dotted edges. We illustrate it in
fig.8.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{e7.eps}
\caption{ Decomposition of trivalent vertex in which the sum of
colors is even}
\end{center}\end{figure}
\subsection{Simple Closed Diagrams: the Super-$\Theta$ Graph}
We have found that edges and nodes of superspin networks decompose
into sums of terms, each of which consists of an ordinary spin
network, perhaps dressed by dotted lines. As a result any closed
super spin network can be decomposed into a sum of such terms. As
an example, we describe the simplest example of a closed spin
network, which is the $\Theta$ graph. The simplest one is the
diagram in which every link has color one. This super $\Theta$ graph
can be decomposed into a sum of three components, each of which is
an ordinary spin network. This is illustrated in fig.9.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=4cm]{fig008-2.eps}
\caption{Decomposition of the simplest $\Theta$ graph }
\end{center}\end{figure}
Another interesting $\Theta$ diagram is the one in which the colors
of three links are $(n,2,n)$. We will use it later in the
calculation of the area spectrum in quantum supergravity. It can
be decomposed into four components in terms of $SU(2)$ spin
networks as shown in fig.10.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=10cm,height=4cm]{fig008-1.1.eps}
\caption{The general case of the decomposition of $\Theta$ graph }
\end{center}\end{figure}
\section{Evaluation of Super Spin Networks}
In the case of $SU(2)$ spin networks, the edges represent
projection operators, which live in the Temperly Lieb algebra.
These can always be decomposed using the bracket identity [see
fig.11].
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=10cm,height=4cm]{fig016.eps}
\caption{The bracket identity for $SU(2)$ spin networks} \end{center}\end{figure}
As a result, associated with any ordinary spin network there is a number
which is called its evaluation. This was first introduced by
Penrose\cite{pen1}. It is now known to be a special case of the
Kauffman bracket polynomial when the quantum deformation parameter
$A= \pm 1$.
For the supergroup $Osp(1|2)$, the spinor identity does not exist
any more. Therefore there is no bracket identity (although in the
super loop representation some identities analogous to the
Mandelstam identity can be expressed by means of the supertraces
of the holonomies \cite{SUG2}). But we can still evaluate a super
spin networks by first decomposing it into ordinary $SU(2)$ spin
networks, using the rules defined in the previous section, and
then evaluating each component.
The evaluation of a super spin network in fact corresponds to
taking the supertrace of a product of projection operators on the
direct product of a number of fundamental representations. The
fact that it can be expressed in terms of the evaluations of
ordinary spin network is a consequence of the fact that the
supertrace can be decomposed into a sum of traces over the $SU(2)$
representations that make up a representation of $Osp(1|2)$. In
fact, the sign factors necessary to turn a sum of traces into a
supertrace are already built into our formalism by the sign
factors that go into the graded symmetrizations that define the
edges and nodes of the super spin networks. In the example of the
super $\Theta$ graph, as well as in the examples that follow, one
can see how this works explicitly.
As a direct application, we can calculate the super standard
closure of the super tangles, which is defined as the supertrace
of the holonomy of the flat connection in this representation:
\begin{eqnarray}
Str_j(
\delta_{\alpha_1\alpha_2...\alpha_n}^{\alpha'_1\alpha'_2...\alpha'_n})
&=& tr_j(
\delta_{A_1A_2...A_n}^{A'_1A'_2...A'_n})+tr_j(\delta_{A_1A_2...A_{n-1}0}
^{A'_1A'_2...A'_{n-1}0})\nonumber\\
&=&(-1)^{(2j)}(2j+1)+(-1)^{2(j-1/2)}[2(j-1/2)+1]
\nonumber\\&=&(-1)^{2j} \label{Str}
\end{eqnarray}
Here when we take the trace
of the dotted line, we find its value is one.
Let us consider the simple super $\Theta$ graph in which the colors of
links are $(1,2,1)$. After decomposing the graph into the normal
spin networks and taking the trace of them as shown in fig.12, we
find the value of the $\Theta$ graph is one.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{e1.eps}
\caption{The Evaluation of $\Theta (1,2,1)$} \end{center}\end{figure} Also since
this graph is equivalent to the super standard closure with color
two (see the first step in figure 12), we can find the value of
this graph by (40) directly, in which $n$ equals two. If we
consider another example in which the colors are (2,2,2), we have
the answer illustrated in fig.13.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{e2.eps}
\caption{The Evaluation of $\Theta (2,2,2)$.} \end{center}\end{figure}
In the third step the coefficient one fourth appears when we try
to separate the dotted loop from the real-line loops. Since
initially the bosonic index is symmetrized with the fermionic
indices, we have four different ways to connect the ropes to form
loops. But the value of any loop which is formed by connecting one
real rope and one dotted rope must be zero, therefore only one
graph has non-zero value. We illustrate the specific expansion in
fig.14.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{e5.eps}
\caption{Separation of the dotted loop from the $\Theta$ graph}
\end{center}\end{figure}
The most interesting $\Theta$ graph, which has important application
to the calculation of the area spectrum in supergravity, is the
one with colors (n,2,n). From the last section, we see this graph
can be divided into four graphs with respect to the ordinary
su(2) spin networks. In fig.16 the bosonic index is symmetrized
with the fermionic indices. To evaluate all these graphs, we also
need to separate the dotted loop from each $\Theta$ graph. In
other words, we must decouple the dotted line from the
symmetrizer just as we have done for the $\Theta$ graph
(1,2,1)and (2,2,2).
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{e6.eps}
\caption{Separation of the dotted line from the symmetrizer.}
\end{center}\end{figure}
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{e4.eps}
\caption{Evaluation of the $\Theta$ graph with color (n,2,n)}
\end{center}\end{figure}
When doing this one must be careful to obtain the right
coefficients for each term. The key point is that the dotted line has to be
connected to the dotted line and for every vertex the triangle
inequality must hold. In general when the dotted line is separated
from the symmetrizer of color $n$, the factor $1/n$ appears and
there are $n$ terms due to the different permutation of the
dotted line as shown in fig.15. For the second and third term in
figure 16, there is only one possible routing of the
dotted line which has non-zero value
among the $2n$ possibilities, therefore the
coefficients are $1/2n$. For the last term, the dotted line can't
be connected to the link of color two so there are $(n-1)(n-1)$
routings with non-zero value and the coefficient is
$(n-1)^2/n^2$.
Now it's straightforward to evaluate the super
$\Theta$ graph by summing all the ordinary $\Theta$ graphs:
\begin{eqnarray}
\Theta (n,2,n)&=&\frac{(-1)^{n+1}(n+1)(n+2)}{2n}+\frac{1}{2n}(-1)^n(n+1)
\nonumber\\&&
+\frac{1}{2n}(-1)^n(n+1)+\frac{(n-1)^2}{n^2}\frac{(-1)^nn(n+1)}{2(n-1)}
\nonumber\\& =&\frac{(-1)^{(n+1)}(n+1)}{2n} \label{eva}
\end{eqnarray}
It is not difficult to generalize this calculation to the case in which
the super $\Theta$ graph has color$(m,2,n)$. One finds that the coefficients
before every ordinary $\Theta$ graphs respectively are
$(1,1/2n,1/2m,(n-1)(m-1)/mn)$. However, it is more
complicated to find a
general formula for the super $\Theta$ graph with color $(m,n,p)$, in
which the separation of the dotted loop from links labeled by
$m,n$ obviously depends on the third link with color $p$.
\section{The Super-area Operator and Its Spectrum}
A natural question concerning the spin network states of
supergravity is whether we
can construct observables such as the area and the volume
of the space in terms of
their action on super spin network states, as in the case
of general relativity[4].
Here we show that the answer is yes, if the operator is
suitably defined. In this
section we construct the area operator and calculating its
eigenvalues in the context
of super spin network basis.
The gauge invariance of supergravity includes the $Osp(1|2)$
symmetry, hence we must require the observables should be
invariant under its full action. The expression for the area
operator in quantum general relativity, computed in
\cite{spain,sn1,sn2}, is not an observable in supergravity, since
it is {\it not} $Osp(1|2)$ gauge invariant. But it is not
difficult to extend the definition of the area of a surface in
general relativity to an expression which is $Osp(1|2)$ invariant.
Given a spatial surface $\cal S\mit$, which is a two-dimensional
manifold embedded in the spacetime manifold $\cal M\mit$, we define
the supersymmetric
area to be:
\begin{equation}
A[\cal S\mit]=\int_{\cal s\mit}d^2s\sqrt{n_an_b\cal
E\mit^{aI}\cal E\mit^b_I}
\end{equation}
where $n_a$ is the normal vector
of the surface and the ${\cal E}^{aI}$ is the conjugate momentum.
The definition of area operator is closely related to the two-hand
loop operator $\cal T\mit^{ab}$ that we have introduced in section three.
When the loop shrinks to a point, following \cite{vol2} and using
the identity about the supertrace of the $Osp(1|2)$ Lie algebra
we find,
\begin{equation} \cal T\mit^{ab}[\alpha](s,t)=Str[\cal
U\mit_{\alpha}(s,t) \cal E\mit^a(\alpha(t)) \cal
U\mit_{\alpha}(t,s)\cal E\mit^b(\alpha(s))]=2\cal E\mit^{aI}\cal
E\mit^b_I
\end{equation}
As a result, the area of the small surface with
side L, to zeroth order, can be written as,
\begin{equation}
A[\cal
S\mit]=\lim_{L\rightarrow\infty}\sum_{I}\sqrt{A_I^2} \end{equation} where: \begin{equation}
A_I^2=\frac{1}{2}\int_{s_I}d^2\sigma \int_{s_I}d^2\tau
n_a(\sigma)n_b(\tau)\cal T\mit^{ab}
[\alpha_{\sigma\tau}](\sigma,\tau)
\end{equation}
Now we define the $Osp(1|2)$ invariant area operator to be,
\begin{equation}
\hat{A}[{\cal S}]=\lim_{L\rightarrow\infty}\sum_{I}
\sqrt{\hat{A}_I^2}
\end{equation}
where,
\begin{equation}
\hat{A}_I^2=\frac{1}{2}\int_{s_I}d^2\sigma\int_{s_I} d^2\tau
n_a(\sigma)n_b(\tau) \hat{\cal T}^{ab}
[\alpha_{\sigma\tau}](\sigma,\tau)
\end{equation}
Next we want to consider
the action of the area operator on spin network states. In
\cite{sn1,vol2}, the discrete spectrum of the area operator in
spin network states is worked out in different ways. One can
divide the link, the element of the spin networks, into ropes in
loop representation so that the area operator acts on the state as
a second order loop operator which can be expressed in terms of
the elementary grasp operation; or equivalently one can define the
action of the area operator on spin networks as inserting two
trivalent intersections on the link by a new link of color $2$,
then calculate the eigenvalues of the operator by recoupling
theory directly. Here we can define the action of super $\cal T\mit$
variables in terms of the elementary grasp operation. This
allows us to calculate the spectrum of the operator
in both ways.
Let us consider the former method first. The action of the super
operator $\cal T\mit^a$ on the super spin networks can be defined
as fig.(17).
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{fig011.eps}
\caption{Action of $\cal T\mit^a$ with one grasp on the super spin
networks }
\end{center}\end{figure}
Basically as we have done in the previous
sections, we can decompose super spin networks into the ordinary
SU(2) ones and then consider the action of the operator on them
separately. From fig.4, we see the super link of color 2 can be
divided into two components, so the corresponding action of the
super operator can be divided into two parts which can be
illustrated in figure 18. For convenience, let's define these
actions as ``real grasp'' and ``dotted grasp'' respectively.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{fig012.eps}
\caption{The super grasps in the view of ordinary SU(2) spin
networks}
\end{center}\end{figure}
When decomposed in terms of $SU(2)$ spin networks, we find that
there are several distinct grasp operations. The first possibility
is the real grasp to the real line, which is exactly the normal
grasps having appeared in \cite{vol2}. The second one is the
dotted grasp acting on the real line, and the third one the
dotted grasp acting on the dotted line. Note that the real grasp
acting on the dotted line vanishes since the only possible result
is that two real lines combine together and go back.
Now it's straightforward to express the action of super operators
on the link of color n, but we need be careful to determine the
multiplicative factors when using the Leibnitz rule to define the
action of area operator on it. Specially, there is a great
difference between the real grasp and the dotted grasp. Since the
area operator is related to the second order super loop operator,
we can take the $\cal T\mit^{ab}$ as the handle with two grasps in
the super spin network basis. When two grasps act on the link of
color $n$, they can grasp the same rope, or any two different
ropes, so there are $n^2$ possible ways to grasp the link. But
when the real grasps act on the dotted rope, the results of the
action are zero. So the number of ``non-zero'' grasps are $n^2$ and
$(n-1)^2$ to the doublet of the super link respectively. Also
after the two dotted grasps act on the link of color $n$, we need
separate the dotted rope from the solid ropes so that we can apply
the formula with respect to the ordinary $SU(2)$ spin networks. As
we have discussed in the last section, the separation involves the
factor $1/2n$. As a result the coefficients before the graphs
acted by the dotted grasps are $n/2$. Figure 19 and 20 show the
actions of these two kinds of grasps on the link of color n.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{fig013.eps}
\caption{Action of the second order real grasp}
\end{center}\end{figure}
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{fig014.1.eps}
\caption{Action of the second order dotted grasp }
\end{center}\end{figure}
Finally we arrive at the last step of this section, that is to
calculate the spectrum of the area operator. Combining the two
actions of the grasps together, we find the super spin network
states are the eigenstates of the $\hat{A}^2$. It is
straightforward to compute the eigenvalues of $\cal T\mit^{ab}$ in
the super spin network basis (see fig.21) and the result is:
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=10cm,height=5cm]{e8.eps}
\caption{The action of the area operator } \end{center}\end{figure}
\begin{equation} \hat
{A}^2|\Gamma^{sg},n_i,v_e>=
\sum_i\frac{n_i(n_i+1)}{4}l_p^4|\Gamma^{sg},n_i,v_e> \end{equation}
Where $l_p$ is Planck length. As a result we find that the eigenvalues of the area operator are
given by, \begin{equation} \hat{A}|\Gamma^{sg},n_i,v_e>=
\sum_i\sqrt{\frac{n_i(n_i+1)}{4}}l_p^2|\Gamma^{sg},n_i,v_e>
=\sum_i\sqrt{j_i(j_i+\frac{1}{2})}l_p^2|\Gamma^{sg},n_i,v_e> \end{equation}
Here we have applied the identities and the formulas in SU(2) spin
networks. This confirms the expected result that the spectrum is
discrete and is directly related to the eigenvalues of the Casmier
operator of $Osp(1|2)$.
Next we conclude that we can get the same solution to eigenvalues
of the area operator directly by employing the identity
associated with the representation of $Osp(1|2)$, in which the
evaluation of the super projectors eq.(\ref{Str}) and super
$\Theta$ graphs (\ref{eva}) is involved. The procedure is shown in
fig.22.
\begin{figure}[h]\begin{center}
\includegraphics[angle=0,width=8cm,height=5cm]{e3.eps}
\caption{The evaluation of area spectrum by means of the identity
for super spin networks}
\end{center}\end{figure}
Finally, we note that we have computed here only a part of the
spectrum of the superarea operator, which is that concerned with
the intersections of edges of the superspin network with the
surface $S$. As in the $SU(2)$ case there are additional
eigenvalues associated with the possibility that the surface $S$
intersects nodes of the superspin network. These eigenvalues may
not be physically relevant as the probability of such
intersections is zero, but in any case they can be computed
directly.
\section{Conclusions}
In this paper we have taken an important step in the extension of
the results of loop quantum gravity to supergravity and string
theory. We have shown that for ${\cal N}=1$ supergravity in $3+1$
dimensions there is a straightforward extension of the methods
developed in \cite{sn1,sn2,vol2} from quantum general relativity.
The extension to ${\cal N}=2$ is in progress and will be reported
shortly \cite{yi2,SUG1}. There is in fact nothing to prevent the
direct extension to any $\cal N$, what is difficult is only the
question of whether, for ${\cal N} >2 $, all the degrees of
freedom of higher $\cal N$ supergravities are represented by an
extension of the connection representation, or whether additional
degrees of freedom need to be introduced. In this connection we
may note that the extension of the loop representation to
represent the states of $p$-form gauge fields is straightforward,
and has been worked out for $p=2$ \cite{rodolofo-p,lee-p} and
$p=3$ \cite{M11}. In the latter case an extension of the loop
representation that describes a limit of $\cal M$ theory in which
only the $3$-form field of $11$ dimensional supergravity survives
can be discussed, and a large set of exact non-perturbative states
found \cite{M11}.
Finally, the quantum deformation of the supersymmetric spin
network states may be developed along the lines of \cite{KL}, and
applied both to yield the supersymmetric extension of the spin
foam models\cite{barrettcrane1,mike} as well as to the background
independent formulation of $\cal M$ theory described in
\cite{tubes,mtheory}.
\section*{Acknowledgements}
We are grateful to Shyamoli Chaudhuri, Laurent Freidel,
Murat Gunyadin,
Renata Loll, Fotini
Markopoulou,
Adrian Ocneanu and Mike Reisenberger
for conversations and encouragement.
This work was supported by the NSF through grant
PHY95-14240
and a gift from the Jesse Phillips Foundation.
|
\section{Introduction}
The idea of exploiting cosmic antiprotons measurements to probe
unconventional particle physics and astrophysics scenarios is
certainly not new.
The investigation on exotic antiproton sources
was stimulated by early reports (Golden et al. 1979;
Buffington et al. 1981) of unexpectedly large values for the
antiproton to proton ratio in cosmic rays.
A possibility which has been discussed in some details is that a
significant contribution to the antiproton flux may be given by
pair annihilation of relic neutralinos. The lightest neutralino,
plausible the lightest supersymmetric particle in the Minimal
Supersymmetric extension to the Standard Model (MSSM),
is one of the leading dark matter candidates.
Its coupling with ordinary particles has the strength of
weak interaction.
This is exactly what it is needed to provide a relic density of
the right order for a flat Universe.
It implies as well that, if one invokes a population of relic
neutralinos in the halo of galaxies to solve the dark matter problem,
these neutralinos can annihilate in pairs into standard model
particles, which might eventually be detected.
In particular it has been shown that the neutralino-induced
antiproton flux can be
at the level of the background flux from cosmic rays
(see e.g. Silk \& Srednicki (1984); Stecker et al. (1985);
Bottino et al. (1998); Bergstr\"om et al. (1999b)).
The question to address is of course if there are distinctive
features in this exotic component so that it can be singled
out unambiguously.
The ordinary cosmic ray induced antiprotons are generated in
collisions of primary particles with the interstellar medium.
The main source are $pp$ collisions which, for kinematical reasons,
produce a characteristic spectrum peaked at a kinetic energy
of about 2~GeV and rapidly decreasing at low energies.
The component from neutralino annihilations may not fall as fast;
in some cases its maximum is actually in the low energy region.
This is the signature which was proposed more than a decade ago
in connection with early antiproton measurements
(Silk \& Srednicki 1984; Stecker et al. 1985).
The more recent (and probably more accurate) data from the {\sc Bess}
experiment (Matsunaga et al.\ 1998; Orito 1998) indicate
unfortunately that the low energy flux may not be so high as
previously thought. It actually seems to be at a level that
is in very good agreement with a standard prediction for the
secondary flux if one takes into account $pHe$ collisions and
energy losses during antiproton propagation (see
Bergstr\"om et al. 1999b, hereafter Paper I). If future
measurements will confirm {\sc Bess} data, this will lead inevitably to
an impasse in this neutralino detection technique:
if the measured flux is consistent with some estimate of the
background, there is no way in improving the signal to background
ratio.
There are however signatures of a neutralino-induced flux
which are alternative to the one we have just discussed. The strategy
we propose here is to search for an exotic component
in the high energy antiproton spectrum,
above the expected maximum of the background.
This kind of investigation is motivated by the fact that
the secondary antiproton flux at high energies is predicted
with great confidence.
It is remarkable that in this energy region nearly all estimates
in the literature are consistent with each other.
The three main ingredients to compute this part of the antiproton
spectrum, unlike in the low energy range, are actually very well
established: data on scattering of protons on nuclei are sufficiently
aboundant, the spectral index of the primary proton flux is known
with good accuracy and, finally, cosmic ray measurements fix quite
strictly the 0.6 power law scaling of the diffusion coefficient
with rigidity. It follows then that at least the shape, if not the
normalization, of the interstellar antiproton flux at high kinetic
energies is significantly constrained.
Moreover in this regime solar modulation, which introduces a further
factor of uncertainty at low energies, does not play a main role.
It was shown in Paper~I that in the canonical scenario
in which neutralinos are assumed to form a perfectly smooth ``gas''
of dark matter particles extended throughout the halo of the Galaxy,
the induced antiproton flux is not sufficiently enhanced to cause
significant distorsions to the high energy secondary spectrum.
As we will show in the next Section a scenario with a clumpy halo
can be much more favourable.
We have singled out three classes of examples in which
the signal we propose can be unambiguously distinguished from the
secondary component. In the first two the main ingredient is clumpy
neutralino dark matter, in the third we consider also the exotic
possibility that antiprotons have a finite lifetime.
Before describing these three cases, we introduce the formalism
needed to give predictions for antiproton fluxes from clumps of
neutralino dark matter in the Milky Way.
\section{Neutralino-induced antiproton flux in a clumpy halo scenario}
\subsection{General discussion}
There are reasons to question whether the distribution of dark matter
in galactic halos has to be regarded as smooth on all scales.
Several theoretical models predict that dark matter may
aggregate in high density substructures, ``clumps'' of dark matter
(see e.g. Silk \& Szalay (1987); Silk \& Stebbins (1993);
Kolb \& Tkachev (1994)).
From the phenomenological point of view, as the annihilation rate
depends quadratically on the neutralino number density locally in
space, the presence of such clumps in the Galaxy may significantly
enhance the chances of detecting neutralino dark matter.
If we postulate that at least a fraction of the dark matter in the
Milky Way is clustered in high density regions, we can associate
to a given clump of density profile
$\rho_{cl}(\vec{r}_{cl})$ and located at the position $\vec{x}_{cl}$,
the following antiproton source term:
\begin{eqnarray}
Q_{\bar{p}}^{\chi,\,cl}(T,\vec{x}\,) &=&
(\sigma_{\rm ann}v)
\frac{1}{{m_{\chi}}^{2}} \sum_{F}^{}\frac{dN^{F}}{dT}B^{F} \cdot
\nonumber \\
&& \cdot \int d^{\,3}r_{cl}\,\left(\rho_{cl}(\vec{r}_{cl})\right)^{2}
\; \delta^3\left(\vec{x}-\vec{x}_{cl}\right)\;.
\label{eq:sourcl}
\end{eqnarray}
The first terms on the right hand side of this equation are related
to the particle physics dark matter candidate.
In our case, $\sigma_{\rm ann}v$ is the total annihilation rate of
non relativistic neutralinos, while
$m_{\chi}$ is the neutralino mass (which enters in the equation
with a -2 power as the dark matter density divided by $m_{\chi}$
is the neutralino number density). We then sum over the annihilation
final states $F$ which can give antiprotons in a fragmentation or
decay process, namely heavy quarks, gluons, gauge and Higgs bosons.
For each of them, $B^{F}$ and $dN^{F}/dT$ are, respectively,
the branching ratio and the fragmentation function.
The hadronization of all final states has been simulated
with the Lund Monte Carlo program {\sc Pythia} 6.115
(Sj\"ostrand 1994).
$\sigma_{\rm ann}v$ and $B^{F}$ are computed in the framework of the
MSSM as described in Paper~I.
Both the value of the annihilation cross section and the relative
importance of different final states can vary largely in different
regions of the MSSM parameter space.
The phenomenological version of the MSSM we use is described in
Bergstr\"om \& Gondolo (1996).
We just remind here few definitions we need in the discussion in
Section~\ref{sec:high}. The lightest neutralino is the lightest
mass eigenstate obtained from the superposition of the
superpartners $\tilde{B}$ and $\tilde{W}^3$ of the neutral
gauge bosons,
and of the neutral CP-even Higgsinos $\tilde{H}^0_1$ and
$\tilde{H}^0_2$.
The neutralino is usually defined
Higgsino-like if it is mainly a linear combination of
$\tilde{H}^0_1$ and $\tilde{H}^0_2$,
while it is called gaugino-like if it is mainly given by the
superposition
of the $\tilde{B}$ and $\tilde{W}^3$ interaction eigenstates.
The formal classification is done introducing the gaugino
fraction $Z_g$.
The remaining terms in the expression for the antiproton source
function, Eq.~(\ref{eq:sourcl}), are related to the neutralino
density distribution; we have actually assumed that the extension of
clumps is much smaller than galactic scales.
The source is hence treated as pointlike: this is not strictly
necessary but simplifies the discussion which follows.
The propagation of antiprotons in the Galaxy is treated with a
two zone diffusion model. Assuming a cylindrical symmetry for the
Galaxy, a solution to a transport equation of the diffusion type
can be derived introducing a Bessel-Fourier series to factorize out
the radial dependence and then solving a one dimensional differential
equation in the vertical direction.
Details on the model and an analytic
solution in case of a pointlike source are given as well in Paper~I.
As the diffusion equation we apply is linear in the antiproton
number density $N$, we can consider each source separately and then
sum the contributions to compute $N$.
To characterize clumps, it is useful to introduce the dimensionless
parameter
\begin{equation}
\delta =\frac{1}{\rho_{0}}
\frac{\int d^{\,3}r_{cl}\,
\left(\rho_{cl}(\vec{r}_{cl})\right)^{2}}
{\int d^{\,3}r_{cl}\,\rho_{cl}(\vec{r}_{cl})}
\end{equation}
which gives the effective contrast between the dark matter density
inside the clump and the local halo density $\rho_0$. We introduce
this definition in Eq.~(\ref{eq:sourcl}) and then, switching to
cylindrical coordinates, we derive the coefficients of the
Bessel-Fourier series for the source function
(see Eq.~(18) in Paper~I). We find:
\begin{eqnarray}
\lefteqn{Q_s^k(z) = (\sigma_{\rm ann}v)
\frac{1}{{m_{\chi}}^2} \sum_{F}^{}\frac{dN^{F}}{dT}B^{F}
\; \rho_0\,M_{cl}\,\delta\,\frac{1}{\alpha_k\,\pi}\cdot}
\nonumber \\
&&
\cdot \frac{2}{{R_h}^2\,{J_{k+1}}^2(\nu_s^k)}
\,J_k \left(\nu_s^k \frac{r_{cl}}{R_h}\right)
\,\cos(k\,\theta_{cl})\,\delta(z-z_{\rm{cl}})
\label{eq:clumpsou}
\end{eqnarray}
where $M_{cl}$ is the mass of the dark matter source we are
considering.
This formula has to be substituted into Eq.~(21) and
(23) of Paper~I to derive the antiproton number density and flux.
It is useful to factorize out in the expression for the flux the
dependence on the MSSM parameters. We define, in analogy to the
expression for a smooth halo scenario,
\begin{equation}
\Phi_{\bar{p}}^{cl}(r_0,T) \equiv
(\sigma_{\rm ann}v) \sum_{F}^{}\frac{dN^{F}}{dT}B^{F}
\frac{\rho_0\,M_{cl}\,\delta}{{m_{\tilde{\chi}}}^{2}}
\, C_{\rm prop}^{cl}(T,\vec{x}_{cl})
\label{eq:signalcl}
\end{equation}
where the coefficient $C_{\rm prop}^{cl}$, which has the dimension
of a length to the power $-2$ times a solid angle to the power $-1$,
contains the dependence of the flux on the neutralino distribution
locally in space and on the parameters in the propagation model.
There is a one to one correspondence between the
antiproton flux computed summing the contributions from the smooth
halo scenario as done in Paper~I, and that from a single clump,
Eq.~\ref{eq:signalcl}; one has just to
replace the coefficient $C_{\rm prop}$ in Eq.~(40) of Paper~I
with $C_{\rm prop}^{cl} \cdot M_{cl}\,\delta / \rho_0$.
To facilitate the comparison we will plot $C_{\rm prop}^{cl}$ in
units of $10^{24}\,\rm{cm}\,\rm{sr}^{-1}\,\rm{kpc}^{-3}$.
\subsection{Antiproton flux from single clumps}
\begin{figure}[tb]
\resizebox{\hsize}{!}{\includegraphics{cpropcl.eps}}
\caption{$C_{\rm prop}^{cl}$ at T=1~GeV and as a function of the distance
from the observer $L$.}
\label{fig:cprpopcl}
\end{figure}
We first suppose that there are few heavy clumps and study their
individual influence on the result for the antiproton flux. Given
the factorization introduced in Eq.~(\ref{eq:signalcl}), we just
analyse the coefficient $C_{\rm prop}^{cl}$ and its behaviour as
a function of the position of the source in the Galaxy. Intuitively we
expect that sources which are nearby give the largest contributions to
the flux. In case of a propagation model which can be regarded to
be as for an infinitely large diffusion box, we would find that
$C_{\rm prop}^{cl}$, rather than being a function of the three
coordinates $\vec{x}_{cl}$, depends just on the distance from the
observer, the line of sight distance $L$. The boundary conditions
in the propagation model, i.e. free escape at border of the diffusion
region, break strictly speaking this symmetry; their effect is
however not very severe for sources which are not too close to the
border of the diffusion zone (in the vertical direction not closer
than 0.8--1~kpc).
In Fig.~\ref{fig:cprpopcl} we plot $C_{\rm prop}^{cl}$ as a function
of the distance $L$, choosing T=1~GeV and the set of canonical
parameters for propagation model introduced in Paper~I (diffusion zone
height $h_h = 3 \,\rm{kpc}$, constant term in the diffusion coefficient
$D^0 = 6 \cdot 10^{27}\,\rm{cm}^2 \rm{s}^{-1}$ and rigidity scale
in the diffusion coefficient
$R_0 = 3 \,\rm{GV}$). The function we have drawn was computed
fixing $z_{cl}$ and moving away from our position in the Galaxy in
the direction of the galactic
centre.\footnote{Details on how to implement the
numerical formula to compute $C_{\rm prop}^{cl}$ are given in
Ullio (1999).}
As can be seen, we find at large distances a very accurate
exponential scaling, while a sharp cusp appears below a distance of
few kiloparsecs: the intuition that if the source happens to be
close to us it gives a large contribution is confirmed.
To show how accurate the assumption is that $C_{\rm prop}^{cl}$ is
essentially just a function of $L$, we plot in Fig.~\ref{fig:level}
its isolevel curves in the plane $z_{cl}=1$~kpc.
We have chosen to normalize $C_{\rm prop}^{cl}$ to 1 at
1~kpc above our position in the Galaxy.
The isolevel curves are basically circumferences, which are just
slightly deformed by our choice of fixing the antiproton number
density to be zero at the border of the diffusion zone in the radial
direction, i.e. $r=20$~kpc (thick solid line in the figure).
\begin{figure}[tb]
\vspace{1 cm}
\centerline{\resizebox{0.8\hsize}{!}{\includegraphics{level.eps}}}
\vspace{0.75 cm}
\caption{Isolevel curves for $C_{\rm prop}^{cl}$ in the
plane $z_{cl}=1$~kpc. $C_{\rm prop}^{cl}$ has been normalized to~1
at 1~kpc above our position in the Galaxy. $r_0$ is our
galactocentric distance.}
\label{fig:level}
\end{figure}
We will not examine here the dependence of $C_{\rm prop}^{cl}$, and
hence of the flux, on the parameters which define the propagation
model: the results are analogous to what it was derived for
$C_{\rm prop}$ in the smooth halo case. We just mention that
introducing a galactic wind fixes a preferred direction of
propagation and breaks therefore the scaling with distance.
As an example we consider the contribution to the antiproton flux
given by a clump of neutralino dark matter defined by the same
choice of parameters as in Bengtsson et al.~(1990), where clumpiness
was introduced in connection with the neutralino induced
$\gamma$-ray signal with continuum energy spectrum.
We fix hence $M_{cl} \sim 10^{8} M_\odot$ and $\delta \sim 10^3$;
the prefactor $M_{cl} \delta$ is then about
$1.3 \cdot 10^4 \rho_0\,\rm{kpc}^{3}$, where we assumed as local
halo density $\rho_0 = 0.3 \,\rm{GeV}\,\rm{cm}^{-3}$.
Comparing the coefficient $C_{\rm prop}^{cl}$ with the analogous
quantity in a smooth halo scenario (with the same choice
of propagation model parameters
$C_{\rm prop}(T=1\,\rm{GeV}) \sim 10^{25}\,\rm{cm}\,\rm{sr}^{-1}$)
we find that the antiproton flux from this single dark matter clump
can be at the level (or much higher) than the sum of the
contributions from the whole dark matter halo if this source is
within about~4.5~kpc.
We might also consider an opposite approach. It is well established
that a dark mass of at least $2 \cdot 10^{6} M_\odot$ is concentrated
within 0.015~pc at the galactic centre (Eckart \& Genzel 1996),
forming probably a massive black hole, possibly the astrophysical
object which is called Sgr A$^*$. Such accretion of matter might be
associated to a region where the density of neutralino dark matter
is enhanced as well.
In an extreme scenario (Berezinsky et al. 1992) the potential well of
a very steep dark matter halo profile ($\sim 1/r^{1.8}$) is the seed
for the formation of the black hole itself. It is also intriguing
that an excess in the high energy $\gamma$-ray flux from the
galactic centre region, which has been found from the analysis of
EGRET data (Mayer-Hasselwander et al. 1998),
can be explained in terms of neutralino annihilations if an appropriate
enhancement of the neutralino density is present there (Ullio 1999).
Regardless of its possible origin, we can estimate how large the
accretion of neutralinos at the galactic centre should be to give a
measurable primary antiproton flux.
Assuming that the galactocentric distance is 8.5~kpc, we find that the
flux induced by a source at the galactic centre is proportional to
$C_{\rm prop}^{cl} (8.5\,\rm{kpc}) \sim 4.5 \cdot 10^{19}
\,\rm{cm}\,\rm{sr}^{-1}\,\rm{kpc}^{-3}$ (at T=1~GeV ).
If we require for instance that its contribution should be at least
one half of the total flux in a smooth halo scenario, we find that
$M_{cl} \delta$ should be at least of the order of $10^{12} M_\odot$.
\subsection{Collective effects of clumpiness}
There is the possibility that a large fraction of the dark matter
mass is in clumps, in the extreme case all of it.
To avoid violating dynamical constraints,
clumps should be light, with masses probably less than
$M_{cl} \sim 10^{4} - 10^{6} M_\odot$.
If one could deal with a model that gives some accurate prediction
for the masses of clumps and their distribution in the Galaxy, it
would be possible to exploit the approach of the previous paragraph
and estimate the antiproton flux by adding the contributions from
individual sources.
As very little is known about the inherently non linear problem of
generating dark matter clumps, it seems more reasonable to follow a
probabilistic approach.
Let $f$ be the fraction of dark matter in clumps and $N_{cl}$ the total
number of clumps, all roughly of about the same mass and overdensity.
We can define a probability density distribution of the clumps in the
Galaxy which in the limit of large $f$, to fulfill
dynamical constraints, has to follow the mass distribution in the
halo. In a Cartesian coordinate system with origin at the galactic
centre, the probability to find a given clump in the volume
element $d^{\,3}x$ at position $\vec{x}$ is:
\begin{equation}
p_{cl}(\vec{x})\,d^{\,3}x = \frac{1}{M_h}\; \rho(\vec{x})\,
d^{\,3}x\;.
\label{eq:prob}
\end{equation}
We introduced here $M_h$, the total mass of the halo, so that $p_{cl}$
has the correct normalization $\int p_{cl} (\vec{x}) d^{\,3}x = 1$.
The antiproton source function in the volume element $d^{\,3}x$ at the
galactic position $\vec{x}$ is then:
\begin{eqnarray}
\lefteqn{Q_{\bar{p}}^{\chi\,\prime}(T,\vec{x}\,)\,d^{\,3}x =
(\sigma_{\rm ann}v)
\frac{1}{{m_{\chi}}^{2}} \sum_{F}^{}\frac{dN^{F}}{dT}B^{F} \cdot}
\nonumber \\
&&
\cdot N_{cl}\,p_{cl}(\vec{x}) \int d^{\,3}r_{cl}\,
\left(\rho_{cl}(\vec{r}_{cl})\right)^{2} \, d^{\,3}x
\nonumber \\
&&
= f\,\delta\;\frac{\rho_0\;\rho(\vec{x})}{{m_{\chi}}^{2}}
(\sigma_{\rm ann}v) \sum_{F}^{}\frac{dN^{F}}{dT}B^{F} \, d^{\,3}x \;,
\label{eq:srclhalo}
\end{eqnarray}
while the antiproton flux is:
\begin{eqnarray}
\lefteqn{\Phi_{\bar{p}}^{\prime}(r_0,T) =
(\sigma_{\rm ann}v) \sum_{F}^{}\frac{dN^{F}}{dT}B^{F}
f\,\delta\; \frac{{\rho_0}^2}{{m_{\tilde{\chi}}}^{2}}\cdot}
\nonumber \\
&&
\cdot\,\int d^{\,3}x\, \frac{\rho(\vec{x})}{\rho_0}
C_{\rm prop}^{cl}(T,\vec{x})
\nonumber \\
&&
\equiv (\sigma_{\rm ann}v) \sum_{F}^{}\frac{dN^{F}}{dT}B^{F}
f\,\delta\; \frac{{\rho_0}^2}{{m_{\tilde{\chi}}}^{2}}
C_{\rm prop}^{\prime}(T) \;.
\label{eq:fluxcl}
\end{eqnarray}
In the equation above, $C_{\rm prop}^{\prime}(T)$ may actually be
computed in the same way as the corresponding coefficient for
a smooth halo $C_{\rm prop}(T)$, just by replacing in the
former $(\rho(\vec{x})/\rho_0)^2$ with $\rho(\vec{x})/\rho_0$.
The two coefficients have similar behaviours; for our canonical
diffusion parameters and an isothermal sphere as dark matter halo
profile, $C_{\rm prop}^{\prime} \sim 0.75 \cdot C_{\rm prop}$
for most of the kinetic energies of interest in our problem.
The antiproton flux in the many small clump scenario can then be
obtained by scaling the result in the smooth halo case by roughly
$0.75 \cdot f\,\delta$. There is quite a large freedom in the
choice of $f\,\delta$, however one has to worry about not violating
existing experimental bounds (Bergstr\"om et al. 1999a).
In all the results in the next Section we have been careful to check
that the models we propose do not induce an overproduction of both
diffuse $\gamma$-rays and of cosmic positrons.
\section{Signatures in the high energy antiproton spectrum}
\label{sec:high}
We have provided two schemes in which the antiproton flux can be
sensibly enhanced with respect to the results in the smooth halo
scenario. We propose here
three cases in which the signal from neutralino dark matter
annihilations in a clumpy halo can distort the high
energy cosmic ray flux, giving a very clear signature
of the presence of an exotic component.
Two experiments have measured the antiproton flux above a
kinetic energy of few GeV (Golden et al. 1979; Hof et al. 1996)
and their data are in contradiction with each other. A new set of
experiments, the {\sc Caprice} experiment (Boezio et al. 1997)
and the space-based {\sc Ams} (Ahlen et al. 1994)
and {\sc Pamela} (Adriani et al. 1995),
should give much more abundant data in the near future.
We will check that our predictions are consistent with the data in
the low energy regime (below 3~GeV) from the {\sc Bess}~97
(Orito 1998) and {\sc Bess}~95 (Matsunaga et al.\ 1998)
flights, which have given a first hint on the actual shape of the
antiproton spectrum. The prediction for the background for our
canonical set of parameters actually provides a very good fit of
the data (see Paper~I; our background prediction has
been confirmed by Bieber et al. (1999) and is consistent
in the high energy region with most the previous estimates in
the literature).
In the first two cases below we allow a little room
for a neutralino-induced component by lowering the normalization
of the primary proton flux by 1~$\sigma$.
For simplicity we will focus on the many small clumps scenario
and quote in each case the value of the parameter $f\delta$ we are
considering.
The sample of supersymmetric models on which this analysis is based
is the same as in Paper~I. As in the smooth halo case we
restrict to those MSSMs for which the neutralino has a cosmologically
interesting relic density, e.g. $0.025 < \Omega_{\chi}h^{2} < 1$.
To compare with {\sc Bess} data we have to take into account
solar modulation. We relate the flux measured at the top of the
atmosphere to the interstellar flux applying the force-field
approximation by Gleeson \& Axford (1967) with solar modulation
parameter $\phi_F \sim 500$~MV (value suggested in the analysis
of the {\sc Bess} collaboration).
\subsection{Broadening of the maximum in antiproton flux}
\begin{figure}[tb]
\resizebox{\hsize}{!}{\includegraphics{speccl1.eps}}
\caption{Distorsion on the secondary antiproton flux (line $b$)
induced by a signal from a heavy Higgsino-like neutralino.}
\label{fig:speccl1}
\end{figure}
The prediction for the background secondary flux has a well defined
maximum at about 2~GeV. We check if it is possible to introduce a
distorsion in the spectrum by adding a component which widens the
shape of the maximum.
We will further require that this is not just some local distorsion
which might be mimicked by introducing for instance
reacceleration effects (Simon \& Heinbach 1996).
In our extreme case we impose that the antiproton flux from
neutralino annihilations must exceed the estimate for the background
by at least one order of magnitude at very high energies,
say about 50~GeV. To obtain a wider maximum one has to add a
neutralino signal rather sharply peaked at an energy higher than the
maximum in the background; its spectrum should then decrease rapidly
at low energies. The effect we are searching for might be produced
by high mass neutralinos with negligible branching ratio into
$b \bar{b}$ or $t \bar{t}$, which is the case e.g. for a very pure
heavy Higgsino-like neutralino.
In Fig.~\ref{fig:speccl1} we show two examples compared to our
standard background estimate (solid line denoted with the letter $b$).
Model 1 is a 964~GeV Higgsino ($Z_g = 0.0027$) with flux rescaled by
$f\,\delta = 4180$; it is the model in our MSSM sample which gives
the highest possible flux at 50~GeV compatibly with the background
normalization we have chosen (dashed line in the figure) and with
the requirement that added to the former it should give a good fit
of the {\sc Bess} data.
Model 2 is instead the case in this class of examples which is
associated with the lowest rescaling, $f\,\delta = 1180$;
it is at the same time the model with the smallest mass,
an Higgsino-like neutralino ($Z_g = 10^{-6}$) with
$m_{\chi} = 777$~GeV.
A further reduction in the background or a less pronounced
overproduction of antiprotons at high energies, allow lower values
of both $m_{\chi}$ and $f\,\delta$.
\begin{figure}[tb]
\resizebox{\hsize}{!}{\includegraphics{speccl2.eps}}
\caption{Distorsion on the secondary antiproton flux (line $b$) induced
by a neutralino signal with flat spectrum.}
\label{fig:speccl2}
\end{figure}
\subsection{Break in the energy spectrum}
We focus in this case on rather flat neutralino-induced spectra.
If one adds such a component to the background, the peak and the low
energy tail in the antiproton flux are not sensibly affected,
but a severe break may occur in the high energy region.
A flat signal spectrum is generated overlapping a low contribution
from a quark annihilation channel to a gauge boson contribution at
higher energies.
In Fig.~\ref{fig:speccl2} model 3 is the case in our MSSM sample
which produces the flattest spectrum and therefore may induce the most
severe break.
It is a very heavy neutralino, $m_{\chi} = 2730$~GeV, and the
rescaling needed in this case may be uncomfortably high,
$f\,\delta \sim 10^5$ (but still it does not violate any
experimental constraint from other cosmic ray measurements).
The second example, model 4, has a mass of 1400~GeV and
$f\,\delta = 7200$.
Clearly these are extreme cases: in this category we may include
a large fraction of models in our sample, which with a
very mild rescaling give a distorsion of the flux at the highest
energies. The chance of detecting less severe breaks in the energy
spectrum clearly depends on how accurately the antiproton flux
is measured.
\subsection{Finite antiproton lifetime}
\begin{figure}[tb]
\resizebox{\hsize}{!}{\includegraphics{spectau.eps}}
\caption{Estimate of possible fluxes in case of finite antiproton
lifetime $\tau_{\bar{p}}$, as compared with the standard background
case when $\tau_{\bar{p}} = \infty$. The total flux is the sum of
a neutralino induced component (not shown) and the secondary
component (dashed lines) which, as plotted in the 3 examples,
can be much suppressed compared to line $b$.}
\label{fig:spectau}
\end{figure}
In exotic scenarios, the antiproton may have a finite lifetime
$\tau_{\bar{p}}$. In the cosmic ray context this is an interesting
ingredient in case $\tau_{\bar{p}}$ is lower than the characteristic
escape time of antiprotons from the Galaxy. The escape time is
determined by the parameters which define the propagation model;
in our case it is about 10~Myr.
This value is clearly much lower than the bound one can infer in CPT
conserving theories from the experimental limit on the proton lifetime
($\tau_{\bar{p}} = \tau_p > 10^{31}$~yr, see Caso et al. (1998)).
On the other hand it is much larger than the most stringent direct
experimental bound ($\tau_{\bar{p}} > 0.05$~Myr from the
$\bar{p} \rightarrow \mu^- \gamma$ decay mode,
see Hu et al. (1998)).
Geer \& Kennedy (1998) have recently claimed that from cosmic
ray measurement it is possible to infer a limit of
$\tau_{\bar{p}} > 0.8$~Myr (90\% C.L.).
In their analysis the secondary component only is included. Adding a
neutralino signal obviously may change this result.
Actually we show here that in a clumpy scenario, even in case of a
finite antiproton lifetime, we can produce models which give at the
same time a very good fit of the existing {\sc Bess} data and
whose spectral features are peculiar enough to be distinguished
from the standard background once measurements at higher energies
will be available. In Fig.~\ref{fig:spectau} we compare three
such models with the case of standard background and infinite
$\tau_{\bar{p}}$.
Model 5 is a 51~GeV gaugino-like neutralino whose antiproton flux
has been scaled by $f\,\delta = 49$ and for
$\tau_{\bar{p}} = 0.82$~Myr. Also shown in the figure is the
reduction in the background flux induced by considering
$\tau_{\bar{p}} = 0.82$~Myr (dashed line labelled by $5b$).
Model~7 is the heaviest model for which the predicted flux is still in
excellent agreement with existing data (90\% C.L. fit for
$m_{\chi} = 477$~GeV,
$f\,\delta = 1.2 \cdot 10^4$ and $\tau_{\bar{p}} = 2.92$~Myr), while
model~6 is some intermediate case ($m_{\chi} = 188$~GeV,
$f\,\delta = 78$ and $\tau_{\bar{p}} = 1.32$~Myr). The trend is that
for heavier neutralinos there is a larger overproduction of
antiprotons in the high energy range.
Applying the largest possible rescalings consistent with $\gamma$-ray
measurements we find that a 56~GeV neutralino model gives a flux
which is consistent with {\sc Bess} data at 90\% C.L. in case
the antiproton lifetime is as low as $\tau_{\bar{p}} = 0.15$~Myr.
The bound of Geer \& Kennedy (1998) is clearly violated. Notice that
we are comparing with a more aboundant data set than in that
reference ({\sc Bess}~97 data were not included there)
and that we used our standard values for the parameters which define
the diffusion model and solar modulation. If uncertainties were
included the lower bound we would get with this method would
probably be very close to the most stringent direct experimental bound
$\tau_{\bar{p}} > 0.05$~Myr or lower.
\section{Conclusion}
To conclude, we have shown that there is a chance of detecting
neutralino dark matter in upcoming measurements of the cosmic
antiproton flux at high energies. The signatures we propose here
are alternative to the signature of an exotic component at low
kinetic energies, which seems not to be required by present data.
We have also discussed the possibility that antiprotons have a finite
lifetime and shown that the limit on $\tau_{\bar{p}}$
which is possible to set
on the basis of cosmic ray measurements
is comparable to those in direct experiments.
\bigskip
\noindent{\bf Acknowlegements}
\smallskip
\noindent
I am grateful to Lars Bergstr{\"o}m and Joakim Edsj{\"o} for many
useful discussions. I thank Paolo Gondolo for collaboration on the
numerical supersymmetry calculations.
|
\section{Strongly correlated liquid and Wigner Crystal}
Let us consider a flat surface uniformly charged with surface density
$-\sigma$ and covered by
concentration $n = \sigma/ Ze$ of counterions with charge $Ze$.
It is well known that the minimum of
Coulomb energy of counterion repulsion and
their attraction
to the background is provided by a triangular close packed WC of
counterions.
Let us write energy per unit surface area of WC as $E=n \varepsilon(n)$
where $\varepsilon(n)$ is the energy per ion. One can estimate
$\varepsilon(n)$ as the interaction
energy of an ion with its Wigner-Seitz cell of background charge
(a hexagon of the background with charge $-Ze$). This estimate gives
$\varepsilon(n)\sim -Z^{2}e^{2}/Da \sim -Z^{2}e^{2}n^{1/2}/D$.
More accurate expression for $\varepsilon(n)$ is~\cite{mara}
\begin{equation}
\varepsilon(n)= - \alpha n^{1/2}Z^{2}e^{2}D^{-1} = -1.1\Gamma k_BT,
\label{energyion}
\end{equation}
where $\alpha=1.96$. At room temperature,
Eq.~(\ref{energyion}) can be rewritten as
\begin{equation}
\varepsilon(n)\simeq - 1.4~Z^{3/2}(\sigma/e)^{1/2}k_BT ~~,
\label{estimate}
\end{equation}
\noindent where $\sigma/e$ is measured in units of nm$^{-2}$.
At $\sigma = 1.0~e/$nm$^{-2}$, Eq.~(\ref{estimate})
gives $|\varepsilon(n)|
\simeq 7 k_BT$ or $\Gamma = 6.3$ at $Z=3$, and $|\varepsilon(n)|
\simeq 13 k_BT$ or $\Gamma = 12$ at $Z=4$. Thus for
multivalent ions at room temperature we are dealing with the
low temperature regime. However,
it is known \cite{Gann} that due to a very small shear modulus, WC melts
at even lower temperature: $\Gamma \simeq 130$.
Nevertheless, the disappearance of
long range order produces only a small effect on thermodynamic
properties. They are determined by
the short range order which does not change significantly in
the range of our interest
$5 < \Gamma < 15$~\cite{Rouzina96,Bruinsma,Shklov98,Shklov99}.
This can be seen from
numerical calculations~\cite{Totsuji,Lado,Gann} of thermodynamic
properties of classical
two-dimensional SCL of Coulomb particles on the neutralizing
background. In the range $0.5 < \Gamma < 50$, the internal
energy of SCL per counterion,
$\varepsilon(n,T)$, was fitted by
\begin{equation}
\varepsilon(n,T) = k_BT ( - 1.1 \Gamma + 0.58 \Gamma^{1/4} + 0.74),
\label{intenergy}
\end{equation}
with an error less than 2\% ~\cite{Totsuji}. The first term on the right
side of
Eq. (\ref{intenergy})
is identical to Eq.~(\ref{energyion}) and dominates at large $\Gamma$.
All other thermodynamic
functions can be obtained from Eq.~(\ref{intenergy}).
In the next section we show that $\kappa_{el}$ and $L_{el}$ are
proportional
to the
inverse isothermal compressibility of SCL at a given number of ions
$N$
\begin{equation}
\chi^{-1}= n(\partial P/\partial n)_T,
\label{compressdef}
\end{equation}
where
\begin{eqnarray}
P &=& - (\partial F/\partial S)_T = (n\varepsilon(n,T) + n
k_BT)/2\nonumber \\
&=& nk_BT ( - 0.55 \Gamma + 0.27 \Gamma^{1/4} + 0.87)
\label{pressure}
\end{eqnarray}
is the two-dimensional pressure,
$F$ is the free energy of SCL and $S=N/n$ is its area.
Using Eq.~(\ref{pressure}) and relation $\partial\Gamma/\partial n
= \Gamma/2n$, one finds
\begin{equation}
\chi^{-1}= nk_BT ( - 0.83 \Gamma + 0.33 \Gamma^{1/4} + 0.87),
\label{compress}
\end{equation}
where the first term on the right side follows from
Eq.~(\ref{energyion})
and describes WC limit. The last two terms give 33\%
correction to the WC term at $\Gamma = 5$ and only 12\% correction at
$\Gamma = 15$.
So one can use zero temperature, Eq.~(\ref{energyion}),
as first approximation
to calculate $\kappa_{el}$ and $L_{el}$. This is how we obtained
Eq.~(\ref{kappaWC}) and Eq.~(\ref{LWC}).
Eqs.~(\ref{pressure}) and (\ref{compress}) show that, in contrast with
most of
liquids and solids, SCL and WC have $negative$ pressure $P$ and
compressibility $\chi$.
We will see below that anomalous behavior is the reason
for anomalous $negative$ rigidity $\kappa_{el}$ and persistence length
$L_{el}$
and $positive$ Gaussian rigidity $\kappa_{G,el}$. The curious negative
sign
of compressibility of two-dimensional electron SCL and WC was first
predicted in Ref.~\onlinecite{Bello}. Later it was discovered in
magneto-capacitance experiments in MOSFETs
and semiconductor heterojunctions~\cite{Krav,Eisen}.
According to Eq.~(\ref{compress})
$\chi^{-1} = 0$ at $\Gamma = 1.48$, $P=0$ at $\Gamma=2.18$
and they become positive at smaller
$\Gamma$.
As one can see from Eqs.~(\ref{estimate}) and (\ref{Gamma}),
at $\sigma \sim 1.0~e/$nm$^{-2}$
such small values of $\Gamma$ correspond to $Z=1$. Thus
surface layer of monovalent ions do not produce large
negative $\kappa_{el}$ and $L_{el}$ in comparison with multivalent ions.
For them conventional results of Eqs.~(\ref{kappaDH}), (\ref{LDH}),
(\ref{kappaPB}),
and (\ref{LPB}) related with counterions in the long distance tail
of screening atmosphere work better.
We will return to this question in Sec. V where we discuss the role of
these
tails.
\section {Membrane}
We will consider a ``thick" membrane for which one can neglect
the effects of the correlation of SCL on two surfaces of the membrane.
If we approximate SCL by WC, the energy of
such correlations between two surfaces of the membrane decay as
$\exp(-2\pi h/a)$, so the condition of ``thickness",
$h \gg 2\pi a$, is actually easily satisfied
for a strongly charged membrane.
Let us first write the free energy of each surface of the membrane as
\begin{equation}
F=Nf(n,T)
\end{equation}
where $f(n,T)$ is the free energy per ion.
\begin{figure}[h]
\input{fig1}
\caption{Bending of membrane (the curvature has been exaggerated).
For simplicity, the WC case is depicted. a) A thick membrane.
The right WC is compressed while the left WC is stretched.
For thick membranes,
this is the dominant cause of the change in free energy. b)
A very thin membrane. Only one Wigner-Seitz cell is shown.
Due to finite curvature of the surface,
the distance from any point of the Wigner-Seitz cell to the central
ion is shorter than that in the flat configuration.
For thin membranes, this is the dominant cause of free energy change.}
\end{figure}
When a membrane is bent (see Fig. 1a), the surface
charge on the right side is compressed to a new
density $n_{R} > n$, while the surface charge on
the left side is stretched to $n_{L} < n$. Since
the total charge on each surface is conserved,
this change in density leads to a change in the
free energy of each surface:
\begin{equation}
\delta F_{L,R} = N\left(\frac{\partial f}{\partial n}\delta n_{L,R} +
\frac{1}{2}\frac{\partial^{2} f}{\partial n^{2}}
\delta n_{L,R}^{2}\right)
~~,
\label{dfn}
\end{equation}
\noindent in which we kept only terms up to second order
in $\delta n_{L,R}=n_{L,R}-n$.
Using the definitions (\ref{pressure}) and (\ref{compressdef})
for the pressure and the compressibility of 2D systems
\begin{eqnarray}
P&=&-\left(\frac{\partial F}{\partial S}\right)_{N,T}
=-N\left(\frac{\partial f}{\partial S}\right)_{N,T}
=n^{2}\frac{\partial f}{\partial n} ~~,\\
\frac{1}{\chi}&=&n\left(\frac{\partial P}{\partial n}\right)_{T}
= 2n^{2}\frac{\partial f}{\partial n}+
n^{3}\frac{\partial^{2} f}{\partial n^{2}}~~, \end{eqnarray}
\noindent Eq.~(\ref{dfn}) can be rewritten as
\begin{equation}
\delta F_{L,R}=\frac{SP}{n}~\delta n_{L,R} +
\frac{S}{n^2}(\frac{1}{2\chi} - P)~\delta n_{L,R}^2 ~~~. \end{equation}
So, the total change in the free energy of the membrane per unit area is
\begin{eqnarray}
\frac{\delta F}{S}&=&\frac{\delta F_{L}+\delta F_{R}}{S}
=\frac{P}{n}(n_{L}+n_{R}-2n) +
\nonumber \\ & & ~~~~\frac{1}{n^2}(\frac{1}{2\chi}
- P)((n_{L}-n)^{2}+(n_{R}-n)^{2}) ~~.
\label{dfLR}
\end{eqnarray}
In the case of cylindrical geometry,
keeping only terms up to second order in the curvature $R_{c}^{-1}$, we
have
\begin{equation}
n_{L,R}=\frac{R_{c}}{R_{c}\pm h/2} n
\simeq \left(1\mp\frac{h}{2R_{c}}+\frac{h^{2}}{4R_{c}^{2}}\right)n ~~.
\label{nLR}
\end{equation}
Substituting Eq.~(\ref{nLR}) into Eq.~(\ref{dfLR}), we get
\begin{equation}
\frac{\delta F^{\rm cyl}}{S}
=\frac{1}{4\chi}h^{2}R_{c}^{-2} ~~.
\label{fcyl}
\end{equation}
Similarly, in the case of spherical geometry we have
\begin{equation}
n_{L,R}=\left(\frac{R_{c}}{R_{c}\pm h/2}\right)^{2} n \simeq
\left(1\mp\frac{h}{R_{c}}+\frac{3h^{2}}{4R_{c}^{2}}\right)n
\end{equation}
\noindent and
\begin{eqnarray}
\frac{\delta F^{\rm sphere}}{S}
=\left(\frac{1}{\chi}-\frac{P}{2}\right)h^{2}R_{c}^{-2} ~~.
\label{fsphere}
\end{eqnarray}
Comparing Eq.~(\ref{fcyl}) and (\ref{fsphere}) with Eq.~(\ref{cyl})
and (\ref{spher}),
we obtain general expressions for the electrostatic contribution to the
bending rigidity
\begin{equation}
\kappa_{el} = \frac{h^2}{2\chi}\: , ~~~~ \kappa_{G,el}= -\frac{h^ 2
P}{2}
\label{kappael} ~~.
\end{equation}
For example, in the case of low surface charge density,
DH approximation can be used to get~\cite{winter}
\begin{equation}
f(n,T)=2\pi \frac{\sigma^2}{D} n^{-1}r_{s} ~~, \end{equation}
\noindent from which, we can easily get a generalization of
Eq. (\ref{kappaDH}) for a ``thick" membrane ($h \gg r_{s}$)
\begin{equation}
\kappa_{DH}=2\pi\frac{\sigma^{2}}{D}h^{2}r_{s},
~~~~ \kappa_{G,DH}=-\frac{1}{2}\kappa_{DH}
~~.
\label{kappaDHthick}
\end{equation}
In the case of high surface charge density we study in this paper, a SCL
of multivalent counterions resides on each surface of the membrane.
The expressions for the pressure and the compressibility given by
Eqs.~(\ref{pressure}) and (\ref{compress}) can be used to calculate
the bending rigidity:
\begin{eqnarray}
\kappa_{SCL}&=&\frac{nh^{2}}{2}
k_BT(-0.83 \Gamma + 0.33 \Gamma^{1/4} + 0.87) ~~,
\label{kappaSCLthick} \\
\kappa_{G,SCL}&=&-\frac{nh^{2}}{2}
k_BT(-0.55 \Gamma + 0.27 \Gamma^{1/4} + 0.87) ~~.
\label{kappaGSCLthick}
\end{eqnarray}
In the limit of a strongly charged surface ($\Gamma \gg 1$),
the first term in
Eqs.~(\ref{kappaSCLthick}) and (\ref{kappaGSCLthick})
dominates, the free energy
of SCL is close to that of WC.
Using Eq.~(\ref{Gamma}) one arrives at Eq.~(\ref{kappaWC})
for the bending rigidity in the WC limit.
As already stated in Sec. 1, for $\Gamma > 3$,
Eqs.~(\ref{kappaSCLthick}),
(\ref{kappaGSCLthick}) give a negative value for the
bending modulus and a positive value
for the Gaussian bending modulus. In other words, multivalent
counterions make the membrane more flexible.
This conclusion is opposite to the standard results
obtained by mean field theories
(Eqs.~(\ref{kappaDH}), (\ref{kappaPB}), (\ref{kappaDHthick})) where
electrostatic effects are known to enhance the bending rigidity of
membranes ($\kappa_{el} > 0$
and $\kappa_{G,el} < 0$). Obviously, this anomaly is related
to the strong correlation between
multivalent counterions condensed on the surface of the membrane,
which was neglected in mean field theories.
We can also look at Eqs.~(\ref{kappaDHthick})
and (\ref{kappaWC}) from
another interesting perspective:
apart from a numerical factor,
Eq.~(\ref{kappaDHthick}) is identical to Eq.~(\ref{kappaWC}) if
we replace $r_{s}$ by $-a$.
So the WC of counterions has effect on bending properties of the
membrane
as if one replaces the
normal 3D screening length of counterions
gas by a $negative$ screening length
of the order of lattice constant.
Such negative screening length of WC or SCL has been derived
for the first time in
Ref.~\onlinecite{Efros}. It follows from the negative
compressibility predicted in Ref.~\onlinecite{Bello},
and observed in Refs.~\onlinecite{Krav} and \onlinecite{Eisen}.
Until now we have ignored the effects related to
Poisson's ratio $\sigma_{P}$ of the membrane material.
We are talking about the
bending induced increase of the thickness
of the compressed (right) half of the membrane, simultaneous decrease
of the thickness of its stretched (left) half, and
the corresponding shift
of the neutral plane of the membrane
(the plane which by definition does not
experience any compression or stretching)
to the left from the central plane.
These deformations can be found following Ref.~\onlinecite{Landau}
and lead to additional term
$\sigma_{P}h^2/(1 - \sigma_{P})R_{c}^2$ in the right side of
Eq.~(\ref{nLR}). It gives for the bending rigidity
\begin{equation}
\kappa_{el} = \frac{h^2}{2\chi} +
\frac{\sigma_P}{1 - \sigma_P}\frac{Ph^2}{2}.
\label{kappaP}
\end{equation}
So, for example, at $\sigma_P =1/3$, the second term of
Eq.~(\ref{kappaP}) gives a 33\% correction to
Eq.~(\ref{kappaWC}).
According to Eqs.~(\ref{kappael}), (\ref{kappaSCLthick}),
(\ref{kappaGSCLthick}) $\kappa_{el} = 0$ at $h=0$.
This happens because in this limit two SCL
merge into one, whose surface charge density
remains unchanged after bending. Nevertheless,
there is another effect directly related to the
curvature of SCL. It can be explained
by concentrating on one curved Wigner-Seitz cell (see Fig. 1b).
One can see, that due to the curvature,
points of the background come closer to the central
counterion of the cell
in the three-dimensional space where Coulomb interaction operates.
As a result, the energy of SCL goes down.
In the Wigner-Seitz approximation, where
energy per ion of WC is approximated by its interaction with
the Wigner-Seitz cell of the background charge, we obtain
\begin{equation}
\kappa_{WC}^{thin} \simeq -0.006\frac{\sigma^{2}a^{3}}{D} ,
~~~~\kappa_{G,WC}^{thin}=-\frac{2}{3}\kappa_{WC}^{thin} ~~~.
\label{kappaWCthin}
\end{equation}
We see that this effect also gives anomalous signs for
electrostatic contribution to rigidity
in the WC limit, but with a very small numerical coefficient.
Also note that, as in the thick membrane case,
we can obtain Eq.~(\ref{kappaWCthin}) for a thin membrane by
replacing $r_s$ in Eq.~(\ref{kappaDH}) by
a negative screening radius of WC with absolute value of the order $a$.
\section {Cylindrical polyelectrolytes}
In this section, we study bending properties of cylindrical
polyelectrolytes with diameter
$d$ and linear charge density $\eta$ (see Fig. 2).
As in the membrane problem, we will assume that the cylinder
is thick, i.e. its circumference
$\pi d$ is much larger than the average distance $a$ between
counterions on it surface.
The calculation is carried out exactly in the same way as in
the case of thick membrane. The only
difference is that, instead of summing the free energy of two
surfaces of the membrane, we
average over the circumference of the cylinder.
Let us denote by $n_{\phi}$ the local density at an angle
$\phi$ on the circumference
on the cylinder (see Fig. 2a). Before bending
$n_{\phi}=n=\eta/\pi dZe$, after bending it changes to a new value
\begin{eqnarray}
n_{\phi}&=&n\frac{R_{c}}{R_{c}-(d/2)\cos\phi } \nonumber \\
&\simeq&n\left(1+\frac{d\cos\phi}{2R_c}+
\frac{d^2\cos^{2}\phi}{4R_c^2} \right) ~~.
\label{nphi}
\end{eqnarray}
Using Eq.~(\ref{dfLR}) the free energy per unit length of
the polymer can be written as
\begin{eqnarray}
\frac{\delta F}{\cal L}&=&\int_{0}^{2\pi} \frac{d}{2} ~d\phi
\left(\frac{P}{n}(n_{\phi}-n) + \frac{1}{n^2}(\frac{1}{2\chi}-P)
(n_{\phi}-n)^{2}\right) \nonumber \\
&=&\frac{\pi}{2\chi}\left(\frac{d}{2}\right)^{3}R_{c}^{-2} \label{dfphi}
\end{eqnarray}
\noindent where we keep terms
up to second order in the curvature $R_{c}^{-1}$.
\begin{figure}[h]
\begin{center}
\input{fig2}
\caption{Bending of cylindrical polyelectrolytes.
a) A thick cylinder. Rigidity is mostly determined by the change
in density of SCL.
b) A thin cylinder. The curvature effect, is the dominant cause of
change in free energy. }
\end{center}
\end{figure}
Comparing Eq.~(\ref{dfphi}) with Eq.~(\ref{rodF}),
(\ref{persistent}),
one can easily calculate the electrostatic
contribution to the persistence length
\begin{equation}
L_{el}=\frac{\pi}{\chi k_BT}\left(\frac{d}{2}\right)^{3} ~~.
\end{equation}
In the case of highly charged polymer, a SCL of counterions
resides on the polymer surface.
For a thick cylinder, the SCL is locally flat
and we can use the numerical
expression (\ref{compress}) for $\chi^{-1}$ to obtain
\begin{equation}
L_{SCL}=\frac{\pi}{8}nd^{3}
( - 0.83 \Gamma + 0.33 \Gamma^{1/4} + 0.87) ~~.
\label{LSCLthick}
\end{equation}
Again, we see that correlations between counterions on the
surface of a polymer lead
to a negative electric contribution to
persistence length for $\Gamma > 1.5$.
In the WC limit $\Gamma \gg 1$, the first term in
Eq.~(\ref{LSCLthick}) dominates, and
using Eq.~(\ref{Gamma}) one can easily obtain Eq.~(\ref{LWC}).
As in the membrane case, for simplicity, in writing down Eqs. (\ref{nphi}),
we have ignored the effect of
finite value of the Poisson's ratio of the polymer material.
In membranes, this effect result in a gain in energy
due to the shift of the neutral plane toward the convex (stretched) sides.
For a cylinder, there is an additional expansion in the $y$
direction (Fig. 2) which reduces the change in surface charge density,
hence compensates the above gain.
These deformations can be found following Ref. \onlinecite{Landau}
and lead to a correction to Eqs. (\ref{nphi})
\begin{eqnarray}
n_{\phi}&=&n\left(1+\frac{d\cos\phi}{2R_c}(1-\sigma_P)+
\frac{d^2\cos^2\phi}{4R_c^2}(1-\frac{\sigma_P}{2}+\sigma_P^2)
\right. \nonumber \\
&&~~~~~~\left. -
\frac{d^2\sigma_P^2}{8R_c^2}(1-\cos^2\phi)\right)
~~.
\end{eqnarray}
This gives, for the persistence length,
\begin{equation}
L_{el}=\frac{\pi}{k_B T}\left(\frac{d}{2}\right)^3
\left(\frac{1}{\chi}(1-\sigma_P)^2+P(3\sigma_P-\sigma_P^2)\right)
~.
\label{Lp}
\end{equation}
Obviously, due to the expansion in $y$ direction, the correction to
energy is not as strong as in the membrane case.
For example, at $\sigma_P=1/3$,
Eq. (\ref{Lp}) gives only 3\% correction to Eq. (\ref{LWC}).
According to Eqs.~(\ref{LSCLthick}) and (\ref{LWC}),
at $d=0$, $\kappa_{el}$ vanishes. In this limit,
we have to directly include
the curvature effect on one dimensional SCL as shown in Fig. 2b.
As already
mentioned in the previous section, after bending,
points on a Wigner-Seitz cell come closer to the central ion, which
lowers the energy of the system. This effect can be calculated
easily in the WC limit. Let's consider the electron at the origin,
its energy can be written as
\begin{equation}
\varepsilon=\sum_i
\frac{Z^2e^2}{Dr_{i}}-\int_{-L}^{L}ds\frac{Ze\eta}{Ds} ~~,
\label{energy}
\end{equation}
\noindent where $r_i=ia$ and $s$ is the contour distance from our ion
to an lattice point $i$ and the element $ds$ of the background charge.
In the straight rod configuration the space distant is the same as the
contour distance, however after bending they change to
\begin{equation}
r_i^{\prime}\simeq r_i(1-r_i^2/24R_c^2) ~~, ~~~
s^{\prime}\simeq s(1-s^2/24R_c^2) ~~.
\label{newrs}
\end{equation}
Using these new distances to calculate the energy of
the bent rod and subtract Eq. (\ref{energy}) from it,
one can easily calculate the change in energy due to curvature and
the corresponding contribution to persistence length:
\begin{equation}
L_{WC}^{thin}=-\frac{l}{96} ~~,
\end{equation}
which is negative and very small.
For e.g., for $Z=3,~4$, $L_{WC}^{thin}=-0.065$ nm and $-0.116$
nm respectively.
\section{Contributions of the tail of the screening atmosphere}
In previous sections, we calculated the contribution of a SCL of
counterions condensed on the surface
of a membrane or polyelectrolyte to their bending rigidity.
We assumed that charge
density $\sigma$ is totally compensated by
the concentration $n= \sigma/Ze$. Actually, for example, for a membrane,
some concentration, $N(x)$, of counterions is distributed
at a distance $x$ from the surface in the bulk of solution
(we call it the tail of the screening atmosphere).
The standard solution of PB equation
for concentration $N(x)$ at
$N(\infty)=0$ has a form
\begin{equation}
N(x) = {1\over{2 \pi l}}{1\over {(\lambda + x)^2}},
\label{GC}
\end{equation}
where $\lambda = Ze/(2 \pi l \sigma)$ is Gouy-Chapman length.
At $ \Gamma \gg 1$, correlations in SCL provide additional
strong binding for counterions,
which dramatically change the form of $N(x)$~\cite{Shklov99}.
It decays exponentially at $\lambda \ll x \ll l/4$,
and at $x \gg l/4$ it behaves as
\begin{equation}
N(x) = {1\over{2 \pi l}}{1\over {(\Lambda + x)^2}}.
\label{GCnew}
\end{equation}
Here $\Lambda = Ze/(2 \pi l \sigma^*)$ is
an exponentially large length and
$\sigma^*$ is the exponentially small uncompensated
surface charge density at the distance $\sim l/4$.
In any realistic situation when $N(\infty)$ is finite or
a monovalent salt is added to the solution,
Eqs.~(\ref{GC}) and (\ref{GCnew}) should be
truncated at the screening radius $r_s$. Then
the solution of the standard PB equation gives~\cite{lekker}
Eq.~(\ref{kappaPB}) at $r_s \gg \lambda$ or Eq.~(\ref{kappaDH})
at $r_s \ll \lambda$.
In the case of SCL, for realistic values of $r_s$ in the range
$l/4 < r_s \ll \Lambda$, we obtain a contribution of
the tail similar to Eq.~(\ref{kappaDH})
%
\begin{equation}
\kappa_{t}=3\pi \frac{(\sigma^*)^{2}r_{s}^{3}}{D}.
\label{kappaelt}
\end{equation}
At reasonable values of $r_s$,
this expression is much smaller than $\kappa_{WC}$
due to very small values of the ratio $\sigma^*/\sigma$.
Now we switch to a cylindrical polyelectrolyte.
In this case, the solution of the PB equation is known~\cite{Zimm}
to confirm the main features of the
Onsager-Manning~\cite{Manning}
picture of the counterion condensation. This solution depends
on
relation between $|\eta|$ and $\eta_c= Ze/l$. In the case
interesting for us,
$|\eta| \gg \eta_c$, the counterion charge $|\eta| - \eta_c$
is
localized at the cylinder surface, while the charge
$\eta_c$, is spread in the bulk of the
solution.
This means that at large distances the apparent charge
density
of the cylinder, $\eta_a$, equals $- \eta_c$ and does not depend
on
$\eta$. Eq.~(\ref{LPB}) can actually be obtained from
Eq.~(\ref{LDH}) by substituting $\eta_c$ for $\eta$.
It is shown in Ref.~\onlinecite{Shklov99} that
at $\Gamma \gg 1$, the existence of SCL at the surface of the cylinder
leads to substantial corrections to the Onsager-Manning theory.
Due to additional binding of counterions by SCL
$|\eta_a| < |\eta_c|$ and is given by the expression
\begin{equation}
\eta_a = - \eta_c {{\ln [N(0)/N(\infty)]}\over{\ln(4 r_{s} / l)}}
~~,
\label{etaapp}
\end{equation}
where $N(0)$ is exponentially small concentration
at the distance $r \geq l/4$
from the cylinder axis, used in Ref.\onlinecite{Shklov99}
as a boundary condition for PB equation at $x=0$.
Therefore, one can obtain for
the tail contribution, the estimate from the above
using Eq.~(\ref{LPB}). For $Z=3$ and $r_s = 5$ nm this
gives $L_t < 1$ nm. For DNA,
this contribution is much smaller than $L_{SCL} \simeq - 5$ nm.
\section{Conclusion}
We would like to conclude with the discussion of
approximations used in this study.
First, we assumed that the surface charges are
immobile. This is true for rigid polyelectrolytes, such as
double helical DNA or actin, as well as for frozen or
tethered membranes. But if the membrane is fluid, its charged
polar heads can move along the surface. In this case surface charges
can accumulate near a $Z$-valent counterion and screen it. Such
screening creates short dipoles oriented perpendicular to the
surface. Interaction energy between these dipoles is
much weaker than the
correlation energy of SCL. Therefore it produces negligible
contribution
to the membrane rigidity. The mobility of the charged polar heads
eliminates effects of counterion correlation only
in the situation where
the membrane has polar heads of two different charges, for example,
neutral and negative ones. In such a membrane, the
local surface charge density can grow due to the
increase of local concentration
of negative heads. But if all of the closely packed
polar heads are equally charged their motion does not lead to
redistribution of the surface charges. Then our theory is valid again.
Another approximation which we used is that
the surface charge is uniformly smeared. This can not be exactly true
because localized charges are always discrete. Nevertheless our approximation
makes sense if the surface charges are distributed evenly, and their
absolute value is much smaller than the counterion charge. For example,
when the surface charged heads have charge -$e$ and the counterion
charge is $Ze\gg e$, then the repulsion between counterions is
much stronger than their pinning by the surface charges. At
$Z\geq 3$ we seem to be close to this picture.
On the other hand, if the surface charges
were clustered, for example, they form compact triplets, the trivalent
counterion would simply neutralize such cluster, creating a small
dipole. Obviously our theory would over-estimate electrostatic
contribution to the bending rigidity in this case.
All calculations in this paper were done for point like counterions.
Actually counterions have a finite size and one can wonder
how this affects our results. Our results, of course,
make sense only if
the counterion diameter is smaller than the average distance between them
in SCL. For a typical surface charge density, $\sigma = 1.0~e/$nm$^{-2}$,
the average distance
between trivalent ions is 1.7 nm, so that
this condition is easily satisfied. The most important correction to
the energy is related to the fact that
due to ion's finite size, the plane of the center of the counterion
charge can be located at some distance from
the plane of location of the surface
charge. This creates an additional planar capacitor at each surface
and results in a positive contribution to
the bending rigidity similar to Eq.~(\ref{kappaDHthick})
which can compensate our negative
contribution.
On the other hand, if the negative ions stick out of
the surface and the centers of counterions are in the same plane with
centers of negative charge this effect disappears.
In general case, one can look at this problem from another angle.
Let us assume that the bare quantities $\kappa_0$ and $L_0$
are constructively defined as experimental values obtained
in the limit of a high concentration of monovalent
counterions. Let us also assume that the distances of closest approach of
monovalent and $Z$-valent counterions to the surface are
the same. This means that the planar capacitor effect
discussed above is already included
in the bare quantities $\kappa_0$ and $L_0$. Then the replacement
of monovalent counterions by $Z$-valent will always lead to
Eq.~(\ref{kappaWC}) and Eq.~(\ref{LWC}).
In summary, we have shown that condensation of multivalent
counterions on the surface of a charged membrane or polyelectrolyte
happens in the form of a strongly correlated Coulomb liquid, which
closely resembles a Wigner crystal. Anomalous properties of
this liquid lead to the observable decrease of the bending rigidity of
a membrane and polyelectrolyte.
\acknowledgements
We are grateful to V. A. Bloomfield and A. Yu. Grosberg
for valuable discussions.
This work was supported by NSF DMR-9616880 (T. N. and B. S.) and
NIH GM 28093 (I. R.).
|
\section{Introduction}
The lepton flavor violating process of neutrinoless
muon-to-electron ($\mu-e$) conversion in a nucleus, represented by
\begin{equation}
\mu^- + (A,Z) \longrightarrow e^- \,+\,(A,Z)^*\, ,
\label{I.1}
\end{equation}
is an exotic process very sensitive to a plethora of new-physics
extensions of the standard model(SM)
\cite{Marci}-\cite{KLV94}.
In addition, experimentally
it is accessible with incomparable sensitivity. Long time ago Marciano and Sanda
\cite{Marci} has proposed it as one of the best probes to search for lepton
flavor violation beyond the standard model.
Recently, in view of the indications for neutrino oscillations
in super-Kamiokande,
solar neutrino
and LSND data, new hope has revived among the experimentalists of nuclear and
particle physics to detect other signals for physics beyond the
SM.
A prominent probe in this spirit is this exotic process (1).
The fact that the upper limits on the branching ratio of the $\mu^--e^-$
conversion relative to the ordinary muon capture,
\begin{eqnarray}\label{rat}
R_{\mu e^-}=\Gamma(\mu^-\to e^-)/\Gamma(\mu^-\to\nu_\mu),
\end{eqnarray}
offer the lowest constraints compared to any purely leptonic rare process
motivated a new $\mu^--e^-$ conversion experiment, the so called MECO experiment
at Brookhaven \cite{Molzon,Molz99,SSK99},
which got recently scientific approval and is planned to start soon.
The MECO experiment is going to use a new very intense $\mu^-$ beam and
a new detector operating at the Alternating Gradient Synchrotron (AGS). The
basic feature of this experiment is the use of a pulsed $\mu^-$ beam to
significantly reduce the prompt background from $\pi^-$ and $e^-$
contaminations. For technical reasons the MECO target has been chosen to be the
light nucleus $^{27}$Al.
Traditionally the $\mu - e$ conversion process was searched by employing medium heavy
(like $^{48}$Ti and $^{63}$Cu) \cite{Schaaf,Dohmen} or very heavy (like $^{208}$Pb
and $^{197}$Au) \cite{Schaaf,Honec,Ahmad} targets (for a historical review on such
experiments see Ref. \cite{KVF98}).
The best upper limits on $R_{\mu e^-}$ set up to the present have been extracted at
PSI by the SINDRUM II experiments resulting in the values
\begin{eqnarray}\label{Ti}
&&R_{\mu e^-} \leq 6.1\times 10^{-13}
\quad\mbox{ for }^{48}\mbox{Ti target \cite{Schaaf}},\\
\label{Pb}
&&R_{\mu e^-} \leq 4.6\times 10^{-11}
\quad\mbox{ for }^{208}\mbox{Pb target \cite{Dohmen}}.
\end{eqnarray}
(at 90\% confidence level).
The experimental sensitivity of the Brookhaven experiment is expected to be
roughly
\begin{eqnarray}\label{Al}
R_{\mu e^-} \leq 2\times 10^{-17}
\quad\mbox{for }^{27}\mbox{Al target \cite{Molzon,Molz99}}
\end{eqnarray}
i.e. three to four orders of magnitude below the existing experimental limits.
It is well known that the process (1) is a very good example of
the interplay between particle and nuclear physics attracting
significant efforts from both sides.
The underlying nuclear physics of the $\mu-e$ conversion has been
comprehensively studied in Refs.
\cite{KVF98}-\cite{KosmasVergados96}.
>From the particle physics point of view, processes like
$\mu-e$ conversion, is forbidden in the SM
by muon and electron quantum number conservation. Therefore
it has long been recognized as an important probe of the flavor changing
neutral currents and possible physics beyond the SM
\cite{Marci}-\cite{KLV94}.
On the particle physics side there are many mechanisms of
the $\mu-e$ conversion constructed in the literature (see
\cite{KLV94,9701381,Huitu,Kos-Kov}
and references therein). All these mechanisms fall into two
different categories:
photonic and non-photonic. Mechanisms from different categories significantly
differ from the point of view of the nucleon and nuclear structure
calculations. This stems from the fact that they proceed at different
distances and, therefore, involve different details of the structure.
The long-distance photonic mechanisms are mediated
by the photon exchange between the quark and the $\mu-e$-lepton currents.
These mechanisms resort to the lepton-flavor non-diagonal
electromagnetic vertex which is presumably induced by
the non-standard model physics at the loop level.
The contributions to the $\mu-e$ conversion via virtual
photon exchange exist in all models which allow $\mu\rightarrow e \gamma$
decay.
The short-distance non-photonic mechanisms contain heavy
particles in intermediate states and can be realized at
the tree level, at the 1-loop level or at the level of box diagrams.
The non-photonic mechanisms are mediated by various particles in intermediate states
such as $W,Z$-bosons
\cite{Marci}-\cite{KLV94}, Higgs bosons \cite{Shanker,KLV94},
supersymmetric particles (squarks, sleptons, gauginos, higgsinos etc.)
with and without $R$-parity conservation in the vertices.
In the supersymmetric (SUSY) extensions of the SM with conserved R-parity
($R_p$SUSY) $\mu^--e^-$ conversion has been studied in Refs.
\cite{KosmasVergados96,R-cons}. In this case the SUSY contributions
appear only at the loop or box level and, therefore, they are suppressed
by the loop factor. The situation is different in the SUSY models with
R-parity non-conservation ($R_p\hspace{-1em}/\ \ $ SUSY). In this framework there exist
the tree level non-photonic contributions
\cite{9701381,Kos-Kov} and the 1-loop photonic contributions
significantly enhanced by the large logarithms \cite{Huitu}.
The primary purpose of this work is to offer a theoretical background
for the running and planned $\mu-e$ conversion experiments. We consider
all the possible tree level contributions to $\mu-e$ conversion in the
framework of the minimal SUSY model with most
general form of R-parity violation ($R_p\hspace{-1em}/\ \ $ MSSM) including the trilinear
$R_p\hspace{-1em}/\ \ $ couplings and the bilinear $R_p\hspace{-1em}/\ \ $ lepton-Higgs terms.
We also examine some non-SUSY and $R_p$SUSY
mechanisms previously studied in \cite{mu-e-nucl,KosmasVergados96,R-cons}.
We develop a formalism of calculating the $\mu-e$ conversion rate for the
quark level Lagrangian with all these terms.
In our study we pay special attention to the effect of nucleon and nuclear
structure dependence of the $\mu-e\ $ conversion branching ratio $R_{\mu e^-}$.
In particular, we take into account the contribution of
the strange nucleon sea which, as we will see, gives a
contribution comparable to the contribution of the valence
quarks of a nucleon.
Thus, we apply our formalism to the case of nuclei $^{48}Ti$ and $^{208}Pb$
by calculating numerically the muon-nucleus overlap integral and
solving the Dirac equations with modern neural networks techniques and
using the PSI experimental data. A similar application is done for the
$^{27}$Al target by employing the sensitivity of the designed MECO experiment.
Our final goal is to derive on this theoretical basis
the experimental constraints on the $R_{p}\hspace{-1em}/$ Yukawa couplings,
the lepton-Higgs mixing parameters and on the sneutrino VEVs.
Towards this end we use the experimental upper limits on
the branching ratio $R_{\mu e^-}$ given above and derive
the new stringent constraints on the R-parity
violating parameters.
\section{The effective $\mu - e$ conversion Lagrangian }
It is well known that, the lepton flavors $(L_i)$ and the total
baryon $(B)$ number are conserved by the standard model interactions
in all orders of perturbation theory. As mentioned above, this is
an accidental consequence of the SM field content and gauge invariance.
Thus $\mu-e$ conversion is forbidden in the SM.
In contrast to $L_i$ the individual quark flavors $B_i$ are not conserved
in the SM by the charged current interactions due to the presence of
non-trivial Cabibbo-Kobayashi-Maskawa (CKM) mixing matrix.
$L_i$ are conserved since the analogous mixing matrix can be rotated
away by the neutrino fields redefinition.
The latter is possible while the neutrino fields
have no mass term and, therefore, are defined up to an arbitrary unitary
rotation.
As soon as the model is extended by inclusion of the right-handed
neutrinos lepton flavor violation can occur since neutrinos may
acquire a non-trivial mass matrix.
In the supersymmetric extensions of SM $L_i$ and $B$ conservation
laws are in general violated. As a result potentially dangerous
total lepton ($L\hspace{-0.5em}/\ \ $) and baryon ($B\hspace{-0.6em}/\ \ $) number violating processes become open.
One may easily eliminate
$L\hspace{-0.5em}/\ \ $ and $B\hspace{-0.6em}/\ \ $ interactions from a SUSY model by
introducing a discrete symmetry known as R-parity. This is a multiplicative
$Z_{2}$ symmetry defined as $R_{p}=(-1)^{3B+L+2S}$, where $S$ is the spin
quantum number. In this framework neutrinos are massless.
However the flavor violation in the lepton sector can occur
at the 1-loop level via the $L_i$-violation in the slepton sector.
Thus, $\mu^--e^-$ conversion is allowed in the SUSY models with R-parity
conservation.
There is no as yet convincing theoretical motivation
for R-symmetry of the low energy Lagrangian and, therefore, SUSY models
with ($R_p$SUSY) and without ($R_{p}\hspace{-1em}/\ \ $ SUSY) R-parity
conservation are "a priori" both plausible.
Despite the above mentioned problems with $L\hspace{-0.5em}/\ \ $, $B\hspace{-0.6em}/\ \ $ interactions
$R_{p}\hspace{-1em}/\ \ $ SUSY looks rather attractive, since
it may offer a clue to the solution of some long standing problems of
particle physics, such as neutrino mass problem.
In the $R_p\hspace{-1em}/\ \ $ SUSY framework neutrinos acquire Majorana masses at the weak-scale
via mixing with the gauginos and higgsinos as well as via $L\hspace{-0.5em}/\ \ $ loop effects
\cite{neutr}. Furthermore, $R_p\hspace{-1em}/\ \ $ SUSY models admit non-trivial contributions to the
lepton flavor violating processes. During the last few years the $R_p\hspace{-1em}/\ \ $ SUSY
models have been extensively studied in the literature (for a recent review see
Refs. \cite{reviews}).
We analyze possible mechanisms for process (1) existing at the tree level in
the minimal $R_p\hspace{-1em}/\ \ $ SUSY model with a most general form of R-parity violation.
A most general gauge invariant form of
the R-parity violating part of the superpotential
at the level of renormalizable operators reads
\begin{eqnarray}
W_{R_p \hspace{-0.8em}/\;\:} =\lambda _{ijk}L_{i}L_{j}E_{k}^{c} + \bar\lambda _{ijk}^{\prime
}L_{i}Q_{j}D_{k}^{c}+\mu _{j}L_{j}H_{2}+\bar\lambda _{ijk}^{\prime \prime
}U_{i}^{c}D_{j}^{c}D_{k}^{c},
\label{sup1}
\end{eqnarray}
where $L$, $Q$ stand for lepton and quark doublet left-handed superfields
while $E^{c},\ U^{c},\ D^{c}$ for lepton and {\em up}, {\em down} quark
singlet superfields; $H_{1}$ and $H_{2}$ are the Higgs doublet superfields
with a weak hypercharge $Y=-1,\ +1$, respectively. Summation over the
generations is implied. The coupling constants
$\lambda $ ($\bar\lambda ^{\prime\prime }$) are antisymmetric
in the first (last) two indices.
The bar sign in $\bar\lambda', \bar\lambda'' $ denotes
that all the definitions are given in the gauge basis for
the quark fields. Later on we will change to the mass basis
and drop the bars.
Henceforth we set $\bar\lambda^{\prime \prime }=0$ which are
irrelevant for our consideration. This condition ensures the proton
stability and can be guaranteed by special discreet symmetries other
than R-parity.
The R-parity non-conservation brings into the SUSY phenomenology
the lepton number ($L\hspace{-0.5em}/\ \ $) and lepton flavor ($L_i\hspace{-0.8em}/\ \ $) violation originating
from the two different sources.
One is given by the $L\hspace{-0.5em}/\ \ $ trilinear couplings in
the superpotential $W_{R_{p}\hspace{-0.8em}/\;}$ of Eq. (\ref{sup1}).
Another is related to the bilinear terms in $W_{R_{p}\hspace{-0.8em}/\;}$
and in soft SUSY breaking sector.
Presence of these bilinear terms leads to the terms linear in the sneutrino
fields $\tilde{\nu}_{i}$ in the scalar potential. The linear terms drive these
fields to non-zero vacuum expectation values $\langle \tilde{\nu}_{i}\rangle\neq 0$
at the minimum of the scalar potential.
At this ground state the MSSM vertices $\tilde{Z}\nu$ $ \tilde{\nu}$ and
$\tilde{W}e\tilde{\nu}$ produce the gaugino-lepton mixing mass terms
$\tilde{Z}\nu\langle \tilde{\nu}\rangle ,\tilde{W}e\langle \tilde{\nu}\rangle $
(with $\tilde{W},\tilde{Z}$ being wino and zino fields). These terms
taken along with the lepton-higgsino $\mu _{i}L_{i}\tilde{H}_{1}$ mixing
from the superpotential of Eq. (\ref{sup1})
form $7\times 7$ neutral fermion
and $5\times 5$ charged fermion mass matrices.
For the considered case of $\mu-e$ conversion the only charged
fermion mixing is essential.
The charged fermion mass term takes the form
\begin{eqnarray}\label{m_term}
{\cal L}^{(\pm)}_{mass} =
- \Psi_{(-)}^{\prime T} {\cal M}_{\pm} \Psi_{(+)}^{\prime} -\mbox{H.c.}
\end{eqnarray}
in the basis of two component Weyl spinors corresponding to
the weak eigenstate fields
\begin{eqnarray}
\Psi _{(-)}^{\prime T} =(e_{\ L}^{-},\,\,\mu _{\ L}^{-},\,\,
\tau _{\ L}^{-},\,\,-i\lambda _{-},\,\,\tilde{H}_{1}^{-}), \ \ \
\Psi _{(+)}^{\prime T} &=&(e_{\ L}^{+},\,\,\mu _{\ L}^{+},\,\,\tau _{\
L}^{+},\,\,-i\lambda _{+},\,\,\tilde{H}_{2}^{+}).
\end{eqnarray}
Here $\lambda _{\pm}$ are the $SU_{2L}$ gauginos while the higgsinos are
denoted as $\tilde{H}_{1,2}^{\pm }$.
These fields are related to the mass eigenstate fields $\Psi _{(\pm )}$
by the rotation
\begin{equation}
\Psi _{(\pm )i}=\Delta _{ij}^{\pm }\ \Psi _{(\pm )j}^{\prime }, \label{mix}
\end{equation}
The unitary mixing matrices $\Delta ^{\pm }$ diagonalize the chargino-charged
lepton mass matrix as
\begin{eqnarray}\label{diag+}
\left(\Delta^{-}\right)^*{\cal M}_{\pm}\left(\Delta^{+}\right)^{\dagger} =
Diag\{m^{(l)}_i, m_{\chi^{\pm}_k}\},
\end{eqnarray}
where $m^{(l)}_i$ and $m_{\chi^{\pm}_k}$ are the physical charged
lepton and chargino masses. In the present paper we use the notations
of Refs. \cite{Now,FKS97}.
Rotating the MSSM Lagrangian to the mass eigenstate
basis according to Eq. (\ref{mix}) one obtains the new lepton
number and lepton flavor violating interactions in addition to those
which are present in the superpotential Eq. (\ref{sup1}).
Note that the mixing between the charged
leptons $(e_{\ L}^{+},\,\,\mu _{\ L}^{+})$
and the chargino components $(-i\lambda_{+},\,\, \tilde H^+_2)$,
described by the off diagonal blocks of the $\Delta^+$,
is proportional to the small factor $\sim m_{e,\mu}/M_{SUSY}$
\cite{Now,FKS97} and is, therefore, neglected in our analysis.
Let us write down the MSSM terms generating by the rotation (\ref{mix})
the new lepton flavor violating interactions relevant for the $\mu^--e^-$
conversion. In the two component form they can be written as \cite{mssm}
\begin{eqnarray}\label{MSSM-rel}
{\cal L}_{_{MSSM}} = \frac{g_2}{2 \cos\theta_W} Z^{\mu}
\bar{\Psi'}_i A_{ij} \bar\sigma_{\mu} \Psi'_j + i g_2 \lambda^- u_L
\tilde d^*_L,
\end{eqnarray}
where $A_{ij} = (1 -2 \sin^2\theta_W)\delta_{ij} + \delta_{i4}\delta_{4j}$.
Rotating this equation to the mass eigenstate basis
we write down in the four-component Dirac notation
\begin{eqnarray}\label{MSSM-mes}
{\cal L}_{_{MSSM}} = \frac{g_2}{2 \cos\theta_W} a_Z Z^{\mu}
\bar e \gamma_{\mu} P_{_L} \mu + g_2 \zeta_i \cdot
\bar u_k P_{_R} e^c_i \tilde d_{Lk}.
\end{eqnarray}
Here $e_{i} = (e, \mu, \tau)$, $P_{_{L,R}}=(1\mp \gamma_{5})/2$.
The lepton flavor violating parameters in this formula are given by
\begin{eqnarray}\label{param1}
a_Z = \Delta^-_{14} {\Delta^{-}}^*_{24} \approx \xi_{11} \xi^*_{21},
\ \ \ \ \ \ \zeta_i = {\Delta^{-}}^*_{i4} \approx \xi_{i1}.
\end{eqnarray}
The approximate expressions for these parameters were found
by the method of Ref. \cite{Now} expanding the mixing matrix
$\Delta^-$ in the small matrix parameter
\begin{eqnarray}\label{xi-mix}
\xi_{i1}^{*} = \frac{g_{2}(\mu \langle\tilde\nu_i\rangle - \langle H_1\rangle \mu_i)}{\sqrt{2}\
(M_{2}\mu -\sin \!2\beta M_{W}^{2})},
\end{eqnarray}
where $\mu, M_2$ are the ordinary MSSM mass parameters from
the superpotential term $\mu H_1 H_2$ and from the $SU_2$ gaugino
soft mass term $M_2 \tilde W^k \tilde W^k$. The MSSM mixing angle
is defined as $\tan\beta = \langle H_2\rangle/\langle H_1\rangle$.
The other lepton flavor violating interaction contributing to the
$\mu^--e^-$ conversion come from the superpotential (\ref{sup1}).
The leading diagrams describing possible tree
level $R_p\hspace{-1em}/\ \ $ MSSM contributions to the $\mu-e$ conversion are
presented in Fig. 1. The vertex operators encountered in these
diagrams are
\begin{eqnarray}\nonumber
&&{\cal L}_{\mu e^-} =
2\,\lambda _{i21}\,\,\tilde{\nu}_{_{Li}}
\bar{e}\,P_{_L}\,\,\mu +
\lambda _{ijj}^{\prime }\,\tilde{\nu}_{_{Li}} \bar{d}_{_j}\,P_{_L}\,d_{_j} -
\lambda_{ijk}^{\prime *}V_{nj}\left(\tilde{u}_{_{Ln}}^{*}\overline{e_i}
\,P_{_R}\,d_k + \tilde{d}_{_{Rk}}\bar{u}_{_n} \,P_{_R}\,e^{c}_i\right) +\\
&&+ g_{2}\,\zeta_{i}\ V_{nk}\cdot \bar{u}_n P_{_R}e_{_i}^{c}\ \tilde{d}_{_{Lk}}+
\frac{1}{2} \frac{g_2}{\cos\theta_W} a_{Z}
Z^{\mu }\bar{e}\gamma _{\mu }P_{_L}\mu -
\frac{g_2}{\cos\theta_W} Z^{\mu}
\bar q\gamma_{\mu}(\epsilon_L P_{_L} + \epsilon_R P_{_R})q,
\label{q-lev}
\end{eqnarray}
where the first three terms originate from the superpotential
(\ref{sup1}), the fourth and fifth terms correspond to the chargino-charged
lepton mixing terms in Eq. (\ref{MSSM-mes}) and the last one is the ordinary
SM neutral current interaction.
\begin{figure}[h!]
\vspace{-3.cm}
\hspace{-2.58cm}
\mbox{\epsfxsize=19. cm\epsffile{fig1.ps}}
\vspace{-14cm}
\caption{
Leading $R_p\hspace{-1em}/\ \ $~MSSM diagrams contributing to $\mu-e$ conversion at the tree level.
(i) The upper diagrams originate from the trilinear $\lambda$, $\lambda'$
terms in the superpotential Eq. (\ref{sup1}).
(ii) The lower diagrams originate from the chargino-charged lepton mixing
schematically denoted by crosses (X) on the lepton lines.}
\vskip 1cm
\end{figure}
In Eq. (\ref{q-lev}) the SM neutral current parameters
are defined as usual
$$
\epsilon_L = T_3 - \sin^2\theta_W \ Q \ , \qquad
\epsilon_R = - \sin^2\theta_W \ Q
$$
with $T_3$ and $Q$ being the 3rd component of the weak isospin and
the electric charge.
The Lagrangian (\ref{q-lev}) is given in the quark mass eigenstate basis
which is related to the flavor basis $q^0$ through
$$q_{L,R} = V_{L,R}^q \cdot q^0.$$
For convenience we introduced the new couplings
$$\lambda_{ijk}^{\prime} = \bar\lambda_{imn}^{\prime }
\left(V^d_L\right)^*_{jm} \left(V^d_R\right)_{kn}.$$
The CKM matrix
is defined in the standard way as $V = V^u_L V^{d\dagger}_L.$
Integrating out the heavy fields from the diagrams in Fig. 1
and carrying out Fierz reshuffling we obtain the 4-fermion effective
Lagrangian which describes the $\mu-e$ conversion at the quark level
in the first order of perturbation theory. It takes the form
\begin{equation}
{\cal L}_{eff}^{q}\ =\ \frac{G_F}{\sqrt{2}}j_{\mu }\left[ \eta _{L}^{ui}J_{uL(i)}^{\mu }+
\eta_{R}^{ui}J_{uR(i)}^{\mu }+\eta_{L}^{di}J_{dL(i)}^{\mu }+
\eta_{R}^{di}J_{dR(i)}^{\mu}\right] + \frac{G_F}{\sqrt{2}}\left[
\bar\eta_{R}^{di}J_{dR(i)}j_{L} + \bar\eta_{L}^{di}J_{dL(i)}j_{R}\right].
\label{eff-q}
\end{equation}
The index $i$ denotes generation so that $u_i = u,c,t$ and $d_i = d,s,b$.
The coefficients $\eta $ accumulate dependence on the $R_{p}\hspace{-1em}/\ \ $
SUSY parameters in the form
\begin{eqnarray} \nonumber
\eta_{L}^{ui} &=&- \frac{1}{\sqrt{2}} \sum_{l,m,n}\frac{\lambda _{2ln}^{\prime }
\lambda_{1mn}^{\prime *}}{G_F \tilde m_{dR(n)}^{2}}V^*_{il} V_{im} +
\,2(1-\frac{4}{3}\sin^{2}\theta _{W})\, a_Z +
4 \sum_n \, {\zeta_{1} \zeta_{2}^*}
\frac{M_W^2}{\tilde m_{dL(n)}^2}|V_{in}|^2,\\
\label{coeff12}
\eta_{R}^{di} &=& \frac{1}{\sqrt{2}}\sum_{l,m,n}
\frac{\lambda_{2mi}^{\prime }\lambda_{1li}^{\prime *}}
{G_F \tilde m_{{u}L(n)}^{2}} V^*_{nm} V_{nl}
+\frac{4}{3} \sin^{2}\theta _{W} \, a_Z, \\ \nonumber
\eta_{L}^{di} &=& - 2(1-\frac{2}{3} \sin^{2}\theta_{W}) \, a_Z, \ \ \
\eta_{R}^{ui} =
- \frac{8}{3} \sin^{2}\theta_{W} \, a_Z
\\ \nonumber
\bar\eta_{L}^{di} &=& - \sqrt{2} \sum_{n}
\frac{\lambda_{nii}^{\prime }\lambda _{n12}^{*}}{{G_F \tilde m^2_{\nu(n)}}},
\ \ \
\bar\eta_{R}^{di} = - \sqrt{2} \sum_{n}
\frac{\lambda_{nii}^{\prime*}\lambda_{n21}}{G_F \tilde m^2_{{\nu}(n)}}.
\end{eqnarray}
Here $\tilde m_{q(n)}, \tilde m_{\nu(n)}$ are the squark and sneutrino
masses.
In Eq. (\ref{eff-q}) we introduced the color singlet currents
\begin{eqnarray}\label{currents}
J_{q_{L/R}(i)}^{\mu }=\bar{q}_i \gamma^{\mu }P_{_{L/R}}q_{i},\ \ \
J_{d_{L/R}(i)}=\bar{d}_i P_{_{L/R}}d_{i}, \ \ \
j^{\mu }=\bar{e}\gamma^{\mu} \mu, \ \ \
j_{_{L/R}}=\bar{e} P_{_{L/R}}\mu,
\end{eqnarray}
where $q_i = (u_i, d_i)$.
Since in the next sections we report the new results for
the nuclear matrix elements of $^{48}$Ti, $^{27}$Al and $^{208}$Pb
we are going also to update $\mu^-\to e^-$ constraints on
the effective lepton flavor violating parameters corresponding
to certain non-SUSY and $R_p$SUSY mechanisms shown in Fig. 2.
\begin{figure}[h!]
\vspace{-2cm}
\vspace{-1.cm}
\hspace{1.0cm}
\mbox{\epsfxsize=16. cm\epsffile{fig2.ps}}
\vspace{-3cm}
\caption{Photonic and non-photonic mechanisms of the $\mu^-- e^-$ conversion within
some extensions of the standard model: (a-c) the SM with massive neutrinos
and (a-c) R-parity conserving supersymmetric extensions of the SM.
}
\label{fig1}
\end{figure}
These mechanisms were previously studied in Refs. \cite{KosmasVergados96,R-cons}.
Let us shortly summarize these mechanisms for completeness.
The long-distance photonic mechanisms mediated
by the photon exchange between the quark and
the $\mu-e$-lepton currents is realized at the 1-loop level
as the $\nu-W$ loop [Fig. 2(a)] with the massive neutrinos
$\nu_i$ and the loop with the supersymmetric particles such as
the neutralino(chargino)-slepton(sneutrino) [Fig. 2(d)].
In the R-parity violating SUSY models there are also lepton-slepton
and quark-squark loops generated by the superpotential couplings
$\lambda L L E^c$ and $\lambda' L Q D^c$ respectively \cite{Huitu}.
The short-distance non-photonic mechanisms in
Fig. 2 contain heavy particles in intermediate states
and is realized at the 1-loop level [Fig. 2(a,b,d,e)]
or at the level of box diagrams [Fig. 2(c)].
The 1-loop diagrams of the non-photonic mechanisms
include the diagrams similar to those for the photonic
mechanisms but with the Z-boson instead of the photon [Fig. 2(a,d)]
as well as additional Z-boson couplings to the neutrinos and
neutralinos [Fig. 2(b,e)].
The box diagrams are constructed of the W-bosons and massive neutrinos
[Fig. 2(c)] as well as similar boxes with neutralinos and sleptons
or charginos and sneutrinos. The branching ratio formula for these mechanisms
is given in the next section.
\section{Nucleon and nuclear Structure dependence of the $\mu-e$ conversion rates.}
One of the main goals of this paper is the calculation of the $\mu-e$
conversion rate using realistic form factors of the participating nucleus
(A,Z).
This can be achieved by applying the conventional approach based on
the well known non-relativistic impulse approximation, i.e. treating
the nucleus
as a system of free nucleons \cite{Berna}. To follow this approach as a first
step one has to reformulate the $\mu-e$ conversion effective Lagrangian
(\ref{eff-q}) specified at the quark level in terms of the nucleon
degrees of freedom.
The transformation of the quark level effective Lagrangian, ${\cal L}_{eff}^{q}$,
to the effective Lagrangian at the nucleon level, ${\cal L}_{eff}^{N}$,
is usually done by utilizing the on-mass-shell matching condition \cite{FKSS97}
\begin{equation}
\langle \Psi_F|{\cal L}_{eff}^{q}|\Psi_I\rangle \approx \langle \Psi_F|{\cal L}%
_{eff}^{N}|\Psi_I\rangle , \label{match}
\end{equation}
where $|\Psi_I\rangle $ and $\langle \Psi_F|$ are the initial and
final nucleon states. In order to solve this equation we use
various relations for the matrix elements of the quark operators
between the nucleon states
\begin{eqnarray}\label{mat-el1}
\langle N|\bar{q}\ \Gamma_{K}\ q|N\rangle = G_{K}^{(q,N)}
\bar{\Psi}_N\ \Gamma_{K}\ \Psi_N,
\end{eqnarray}
with $q=\{u,d,s\}$, $N=\{p,n\}$ and $K = \{V,A,S,P\}$,
$\Gamma_K = \{\gamma_{\mu}, \gamma_{\mu}\gamma_5, 1, \gamma_5\}$.
Since the maximum momentum transfer ${\bf q}^{2}$ in $\mu -e$ conversion,
i.e. $|{\bf q}| \approx m_\mu/c$ with $m_\mu=105.6 MeV$ the muon mass,
is much smaller than the typical scale of nucleon structure we can safely neglect
the ${\bf q}^{2}$-dependence of the nucleon form factors $G_{K}^{(q,N)}$.
For the same reason we drop the weak magnetism and
the induced pseudoscalar terms proportional to the small
momentum transfer.
The isospin symmetry requires that
\begin{eqnarray}\label{isosym}
G_{K}^{(u,n)}=G_{K}^{(d,p)}\equiv G_{K}^{d}, \ \ \
G_{K}^{(d,n)}=G_{K}^{(u,p)}\equiv G_{K}^{u},\ \ \
G_{K}^{(s,n)}=G_{K}^{(s,p)}\equiv G_{K}^{s},
\end{eqnarray}
with $K=V,A,S,P$.
Conservation of vector current postulates the vector charge to be equal to
the quark number of the nucleon. This allows fixing of the vector nucleon
constants
\begin{eqnarray}\label{cvc}
G_{V}^{u}=2,\ \ \ G_{V}^{d}=1, \ \ \ G_{V}^{s}=0.
\end{eqnarray}
The axial-vector form factors $G_A$ can be extracted from the experimental
data on polarized nucleon structure functions \cite{EMC,SMC} combined with
the data on hyperon semileptonic decays \cite{Hyp}. The result is
\begin{eqnarray}\label{axial}
G_A^{u} \approx 0.78,\ \ \ G_A^{d} \approx -0.47, \ \ \
G_A^{s} \approx -0.19.
\end{eqnarray}
The scalar form factors are extracted from the baryon octet $B$ mass
spectrum $M_B$ in combination with the data on the pion-nucleon sigma term
\begin{eqnarray}\label{sigma}
\sigma_{\pi N} = (1/2)(m_u + m_d)\langle p|\bar u u + \bar d d|p\rangle.
\end{eqnarray}
Towards this end we follow the QCD picture of the baryon masses which
is based on the relation \cite{SVZ,Cheng}
\begin{eqnarray}\label{emt}
\langle B|\theta^{\mu}_{\mu}|B\rangle = M_B \bar B B
\end{eqnarray}
and on the well known representation for the trace of
the energy-momentum tensor \cite{SVZ}
\begin{eqnarray}\label{trace}
\theta^{\mu}_{\mu} = m_u \bar u u + m_d \bar d d
+ m_s \bar s s - (\tilde b\alpha_s/8\pi)G_{\mu\nu}^a G^{\mu\nu}_a,
\end{eqnarray}
where $G_{\mu\nu}^a$ is the gluon field strength, $\alpha_s$ is
the QCD coupling constant and $\tilde b$ is the reduced Gell-Mann-Low
function with the heavy quark contribution subtracted. Using these
relations in combination with $SU(3)$ relations \cite{Cheng} for
the matrix elements $\langle B|\theta^{\mu}_{\mu}|B\rangle$ as well as
the experimental data on $M_B$ and $\sigma_{\pi N}$ we derive
\begin{eqnarray}\label{scalar}
G_S^{u} \approx 5.1, \ \ \ G_S^{d} \approx 4.3, \ \ \ G_S^{s} \approx 2.5.
\end{eqnarray}
The nucleon matrix elements of the pseudoscalar
quark currents can be related to the divergence of the baryon octet
axial vector currents \cite{Cheng}. Utilizing this fact we
find the pseudoscalar form factors
\begin{eqnarray}\label{pseud}
G_P^{u} \approx 103, \ \ \ G_P^{d} \approx 100, \ \ \ G_P^{s} \approx 3.3.
\end{eqnarray}
Note that the strange quarks of the nucleon
sea significantly contribute to the nucleon form factors
$G_A$, $G_P$ and $G_S$. This result dramatically differs from
the na\"{\i}ve quark model and the MIT bag model where $G_{A,S,P}^{s} = 0$.
The contribution of the strange nucleon sea will allow us to extract
additional constraints on the second generation $R_p\hspace{-1em}/\ \ $ parameters.
Now we can solve Eq. (\ref{match}) and write
the effective $\mu-e$ conversion Lagrangian at the nucleon level as
\begin{equation}
{\cal L}_{eff}^{N}= \frac{G_F}{\sqrt{2}}\left[\bar{e}
\gamma_{\mu }(1-\gamma _{5})\mu \cdot J^{\mu }+
\bar{e}\mu \cdot J^{+}+\bar{e}\gamma_{5}\mu \cdot J^{-}\right]. \label{nucl1}
\end{equation}
where we have introduced the nucleon currents
\begin{eqnarray}
J^{\mu } &=&\bar{N}\gamma ^{\mu }\left[ (\alpha _{V}^{(0)}+\alpha
_{V}^{(3)}\tau _{3})+(\alpha _{A}^{(0)}+\alpha _{A}^{(3)}\tau _{3})\gamma
_{5}\right] \,N, \label{nucur} \\
J^{\pm } &=&\bar{N}\,\,\left[ (\alpha _{\pm S}^{(0)}+\alpha _{\pm
S}^{(3)}\tau _{3})+(\alpha _{\pm P}^{(0)}+\alpha _{\pm P}^{(3)}\tau
_{3})\gamma _{5}\right] \,N, \nonumber
\end{eqnarray}
for nucleon isospin doublet $N^T = (p, n)$.
The coefficients in Eq. (\ref{nucur}) are defined as
\begin{eqnarray}
\alpha _{V}^{(0)} &=&\frac{1}{8}(G_{V}^{u}+
G_{V}^{d})(\eta _{R}^{u1}+\eta_{L}^{u1}+\eta _{R}^{d1}+\eta _{L}^{d1}),
\nonumber \\ \nonumber
\alpha _{V}^{(3)} &=& \frac{1}{8}%
(G_{V}^{u}-G_{V}^{d})(\eta_{R}^{u1}+\eta_{L}^{u1}-\eta_{R}^{d1}-
\eta_{L}^{d1}), \nonumber \\ \nonumber
\alpha _{A}^{(0)} &=&\frac{1}{8}(G_{A}^{u}+
G_{A}^{d})(\eta _{R}^{u1}-\eta_{L}^{u1}+\eta_{R}^{d1}-\eta_{L}^{d1}) +
\frac{1}{4}G_A^{s}(\eta _{R}^{d2} - \eta _{L}^{d2}),\\ \nonumber
\alpha _{A}^{(3)} &=&
\frac{1}{8}(G_{A}^{u}-G_{A}^{d})(\eta_{R}^{u1}-\eta _{L}^{u1}-
\eta_{R}^{d1}+\eta_{L}^{d1}),
\label{alp1} \\
\alpha_{\pm S}^{(0)} &=&\frac{1}{16}(G_{S}^{u}+G_{S}^{d})
(\bar\eta_{L}^{d1}\pm \bar\eta_{R}^{d1}) +
\frac{1}{8}G_S^{s}(\bar\eta_{L}^{d2} \pm \bar\eta_{R}^{d2}),\\
\alpha_{\pm S}^{(3)} &=& -\frac{1}{16}
(G_{S}^{u}-G_{S}^{d})(\bar\eta_{L}^{d1}\pm \bar\eta_{R}^{d1}),
\nonumber \\ \nonumber
\alpha_{\pm P}^{(0)} &=&-\frac{1}{16}(G_{P}^{u}+G_{P}^{d})
(\bar\eta_{L}^{d1}\mp \bar\eta_{R}^{d1}) -
\frac{1}{8}G_P^{s}(\bar\eta_{L}^{d2} \mp \bar\eta_{R}^{d2}),
\nonumber \\ \nonumber
\alpha_{\pm P}^{(3)} &=& \frac{1}{16}
(G_{P}^{u}-G_{P}^{d})(\bar\eta_{L}^{d1}\mp \bar\eta_{R}^{d1}),
\ \ \ \nonumber
\end{eqnarray}
Starting from the Lagrangian
(\ref{nucl1}) it is straightforward to deduce the formula
for the total $\mu-e$ conversion rate. In the present paper
we focus on the coherent process, i.e. ground state to ground
state transitions. This is a dominant channel of $\mu-e$ conversion which,
in most of the experimentally interesting nuclei, exhausts more than
$90\%$ of the total $\mu^- \to e^-$ branching ratio
\cite{Chiang,mu-e-nucl}.
To the leading order of the non-relativistic expansion the coherent
$\mu-e$ conversion rate takes the form
\begin{equation}
\Gamma_{\mu e^-}^{coh} = \frac{G^2_F p_e E_e}
{2 \pi } \ {\cal Q} ({\cal M}_p + {\cal M}_n)^2 \, ,
\label{Rme}
\end{equation}
where
\begin{eqnarray}
{\cal Q} \,=\, 2|\alpha_V^{(0)}+\alpha_V^{(3)}\ \phi|^2 +
|\alpha_{+S}^{(0)}+\alpha_{+S}^{(3)}\ \phi|^2 + |\alpha_{-S}^{(0)}+
\alpha_{-S}^{(3)}\ \phi|^2
\nonumber \\
+2\ {\rm Re}\{(\alpha_V^{(0)}+\alpha_V^{(3)}\ \phi)[\alpha_{+S}^{(0)}+\alpha_{-S}^{(0)}+
(\alpha_{+S}^{(3)}+\alpha_{-S}^{(3)})\ \phi] \}\, .
\label{Rme.1}
\end{eqnarray}
The transition matrix elements
${{\cal M}}_{p,n}$
in Eq. (\ref{Rme})
depend on the final nuclear state populated during the $\mu-e$ conversion.
We should stress that, after computing the nuclear matrix elements
${{\cal M}}_{p,n}$ the data provide constraints on the quantity $\cal Q$ of
Eq. (\ref{Rme.1}).
For the ground state to ground state transitions in spherically symmetric
nuclei the following integral representation is valid
\begin{equation}
{\cal M}_{p,n} = 4\pi \int j_0(p_e r) \Phi_\mu (r) \rho_{p,n} (r) r^2 dr
\label{V.1}
\end{equation}
where $j_0(x)$ the zero order spherical Bessel function and
$\rho_{p,n}$ the proton (p), neutron (n) nuclear density normalized to
the atomic number $Z$ and neutron number $N$ of the participating
nucleus, respectively. The space dependent part of the muon
wave function $\Phi_{\mu}$ is a spherically symmetric function which in our
calculations (see Sect. 4) was obtained by solving numerically the Shr\"ondinger
and Dirac equations with the Coulomb potential.
In defining $\cal Q$, Eq. (\ref{Rme.1}), we introduced the ratio
\begin{eqnarray}\label{phi}
\phi = ({\cal M}_p - {\cal M}_n)/({\cal M}_p + {\cal M}_n)
\approx (A-2Z)/A,
\end{eqnarray}
where $A$ and $Z$ are the atomic weight and the total charge of the nucleus.
The quantity ${\cal Q}$ depends weakly on the nuclear parameters through the factor
$\phi$. In fact, the terms depending on $\phi$ are small since $\phi < 1$ (see
Table 1) and $G_S^{u}$, $G_S^{d}$ as well as $G_V^{u}$, $G_V^{d}$ have the same
sign. In practice the nuclear dependence of ${\cal Q}$ can be neglected
and thus, ${\cal Q}$ coincides with the value of $2 \rho$ where $\rho$ is defined below.
It can be considered as a universal effective
$R_p\hspace{-1em}/\ \ $ parameter measuring the $R_p\hspace{-1em}/\ \ $ SUSY contribution to the $\mu-e$ conversion.
It also represents a suitable characteristic which allows
comparison of $\mu-e$ conversion experiments on different targets treating
the corresponding upper bounds on ${\cal Q}$ as their
sensitivities to the $R_p\hspace{-1em}/\ \ $ SUSY signal.
For completeness, in Sect. 4 the limits for some non-SUSY
\cite{KosmasVergados96,R-cons} as well as $R_p$SUSY contributions to the
$\mu^--e^-$ nuclear conversion (see in Fig. 2) are updated.
The corresponding expression for $R_{\mu e^-}$ is written as \cite{KVF98}
\begin{equation}
R_{\mu e^-} = \rho\gamma,
\label{III.2}
\end{equation}
where $\rho$ is nearly independent of nuclear physics~\cite{mu-e-nucl} and
contains the lepton flavor violating parameters corresponding to the
contributions in Fig. 2. Thus, e.g. for photon-exchange mode $\rho$ is given by
\begin{equation}
\rho = (4\pi\alpha)^2\frac{\vert f_{M1}+f_{E0}\vert ^2+
\vert f_{E1}+f_{M0}\vert^2}{(G_Fm^2_{\mu})^2}
\label{rho}
\end{equation}
where the four electromagnetic form factors $f_{E0}$, $f_{E1}$, $f_{M0}$,
$f_{M1}$ are parametrized in a specific elementary model~\cite{mu-e-nucl}.
The factor $\gamma(A,Z)$ in Eq. (\ref{III.2}) accumulates about all the
nuclear structure dependence of the branching ratio $R_{\mu e^-}$.
Assuming that the total rate of the ordinary muon
capture is given by the Goulard-Primakoff function,
$f_{GP}$, the nuclear structure factor $\gamma(A,Z)$ takes the form
\begin{equation}
\gamma(A,Z)\equiv\gamma = \frac{E_e p_e}{m_\mu^2}
\frac{ M^2}{G^2 Z f_{GP}(A,Z)},
\label{III.3}
\end{equation}
where $G^2 \approx 6$. Thus, a non-trivial nuclear
structure dependence of the $\mu^-\to e^-$ conversion
branching ratio $R_{\mu e^-}$ is mainly concentrated
in the nuclear matrix elements $M^2$ \cite{KLV94}.
In the proton-neutron representation one can write down
\begin{equation}
M^2 = [ M_p+ Q \ M_n ]^2
\label{ME2}
\end{equation}
where $Q$ takes the values of Eq. (32) of Ref. [15a] and
$M_{p,n}$ are given by writing the matrix elements of Eq. (\ref{V.1}) in terms
of an effective muon wavefunction as
\begin{equation}
{\cal M}_{p,n} = \langle \Phi_{\mu} \rangle M_{p,n}
\label{factoriz}
\end{equation}
In our present approach the role of $\rho$ in Eq. (\ref{III.2}) is played
by $\cal Q$, since ${\cal Q }= 2 \rho$, and the corresponding $\gamma$
function defined in Eq. (\ref{III.3}) for R-parity violating interactions,
$\gamma_{R_p \hspace{-0.8em}/\;\:}$, is obtained from
Eq. (\ref{III.3}) by putting $Q=1$ in Eq. (\ref{ME2}).
The separation of nuclear physics from the elementary particle parameters
is not complete but we have seen that $\phi$ is quite small. In any case
we present in Table 1 the values of $\phi$ for the various nuclear systems.
\section{Results and Discussion.}
The pure nuclear physics calculations needed for the $\mu-e$ conversion
studies refer to the integrals of Eq. (\ref{V.1}). The results of ${\cal M}_p$
and ${\cal M}_n$ for the currently
interesting nuclei $Al$, $Ti$ and $Pb$ are shown in Table 2. They have been
calculated using proton densities $\rho_{p}$ from the electron scattering
data \cite{Vries} and neutron densities $\rho_{n}$
from the analysis of pionic atom data \cite{Gibbs}.
We employed an analytic form for the muon wave function $\Phi_\mu(r)$ obtained by
solving the Schr\"odinger equation using the Coulomb potential produced
by the charge densities discussed above.
This way the finite size of a nucleus was taken into consideration.
Moreover, we included vacuum polarization corrections as in Ref. \cite{Chiang}.
In solving the Schr\"odinger equation we have used modern neural
networks techniques \cite{Lagaris} which give the wave function $\Phi_\mu (r)$
in the analytic form of a sum over sigmoid functions. Thus, in Eq. (\ref{V.1})
only a simple numerical integration is finally required.
To estimate the influence of the non-relativistic
approximation on the muon wave function $\Phi_\mu({\bf r})$, we have also
determined it by solving the Dirac equation. The results for
${\cal M}_{p,n}$
do not significantly differ from those of the Schr\"odinger picture.
In Table 2 we also show the muon binding energy $\epsilon_b$ (obtained
as output of the Dirac and Schr\"odinger solution) and the experimental values
for the total rate of the ordinary muon capture $\Gamma_{\mu c}$ taken from
Ref. \cite{Suzuki}.
Using the values of ${\cal M}_{p}$, ${\cal M}_{n}$ for a set of nuclei
throughout the periodic table we can estimate the nuclear
structure dependence of the $\mu-e$ conversion branching ratio, i.e. the
function $\gamma (A,Z)$ in Eq. (\ref{III.2}).
The results are quoted in Table 3. For comparison in this table we also
list the results of $\gamma_{R_p \hspace{-0.8em}/\;\:}$ corresponding to the $R_p\hspace{-1em}/\ \ $ SUSY mechanisms
studied in the present paper and calculated as we stated in Sect. 3.
These results can be exploited for setting constraints on
the quantities $\rho$ and $\cal Q$ corresponding to specific models
predicting the $\mu-e$ process.
In Table 4 we quote the upper bounds for the quantities $\rho$ and
$\cal Q$ corresponding to the mechanisms shown in Figs. 1,2.
These bounds were derived from the recent experimental upper bounds on
the branching ratio $R_{\mu e^-}$ for Ti and Pb targets given in
(\ref{Ti}) and (\ref{Pb}) and from the expected experimental sensitivity
(\ref{Al}) of the Brookhaven MECO experiment.
The limits on $\rho$ and ${\cal Q}$ quoted in Table 4 for $^{27}$Al
are improvements by about four orders of magnitude over the existing ones.
We should stress that limits on the quantities $\rho$ of Eq. (\ref{rho})
and ${\cal Q}$ of Eq. (\ref{Rme.1}), are the only constraints imposed by
the $\mu-e$ conversion on the underlying elementary particle physics.
One can extract upper limits on the individual lepton flavor violation
parameters ($R_p\hspace{-1em}/\ \ $ couplings, effective scalar and vector couplings,
neutrino masses etc. \cite{Marci,KLV94,mu-e-nucl,Kos-Kov})
under certain assumptions like the commonly assumed
dominance of only one component of the $\mu-e$ conversion
Lagrangian which is equivalent to constrain one
parameter or product of the parameters at a time.
Using the upper limits for ${\cal Q}$ given in Table 3 we can derive
under the above assumptions the constraints on $\alpha^{(\tau)}_K$
of Eq. (\ref{Rme.1}) and the products of various $R_p\hspace{-1em}/\ \ $ parameters.
Thus, the bounds obtained for the scalar current couplings
$\alpha^{(0)}_{\pm S}$
in the $R$-parity violating Lagrangian for the $^{27}$Al target
\cite{SSK99} are $|\alpha^{(0)}_{\pm S}| < 7 \times 10^{-10}$.
The limit for $\alpha^{(0)}_{\pm S}$ obtained with the data for
the Ti target \cite{Schaaf} is $|\alpha^{(0)}_{\pm S}| < 1.1\times 10^{-7}$,
i.e. more than two orders of magnitude weaker than the limit of $^{27}$Al.
With these limits it is straightforward to derive constraints on
the parameters of the initial Lagrangian (\ref{q-lev}). In Tables 5,6
we list the upper bounds on the products of the trilinear
$R_p\hspace{-1em}/\ \ $ couplings
which we obtained from the experimental limit on $\mu-e$ conversion
in $^{48}$Ti and from the expected experimental sensitivity of MECO
detector using $^{27}$Al as a target material.
The corresponding constraints for $^{208}Pb$ are significantly weaker and
not presented here.
In Tables 5,6 the quantity $B$ denotes a scaling factor defined as
\begin{eqnarray}\label{scale}
B_{Ti} = (R^{exp}_{\mu e}/6.1 \times 10^{-13})^{1/2},\ \ \ \ \
B_{Al} = (R^{exp}_{\mu e}/2.0 \times 10^{-17})^{1/2},
\end{eqnarray}
which can be used for reconstructing the limits on the listed products
of the $R_p\hspace{-1em}/\ \ $ parameters corresponding to the other
experimental upper limits on the branching ratio $R^{exp}_{\mu e}$.
In the 2nd column of Tables 5,6 we present previous limits existing
in the literature and taken from \cite{Huitu,reviews}.
As seen from Tables 5,6 the $\mu-e$-conversion limits on the products
$\lambda'\lambda'$ and $\lambda\lambda'$, except only
$\lambda'_{232}\lambda'_{132}$ and $\lambda'_{233}\lambda'_{133}$,
are significantly more stringent than those previously known in
the literature.
The two products $\lambda'\lambda'$ and $\lambda\lambda'$ are
less stringently constrained by the present experimental data on $^{48}$Ti
within the tree level non-photonic mechanism. Note that the corresponding
previous constraints on these products (2nd column of Table 5) were obtained
from the photonic 1-loop mechanism \cite{Huitu} of $\mu^--e^-$ conversion
which are better than existing in the literature non-$\mu^- \to e^-$
constraints on $\lambda'\lambda'$ and $\lambda\lambda'$.
The products in Table 6 are not constrained by this mechanism \cite{Huitu}.
However it constrains the products $\lambda\lambda$ not constrained by
the tree level $R_p\hspace{-1em}/\ \ $ SUSY mechanism. At present the $\mu^-\to e^-$
constraints on $\lambda\lambda$ within the 1-loop $R_p\hspace{-1em}/\ \ $ SUSY mechanism
are weaker than those derived from the other processes.
As we have mentioned at the beginning, significant improvement on
these $\mu^- \to e^-$ limits is expected from the ongoing experiments
at PSI \cite{Schaaf} and even better from the MECO experiment at Brookhaven
\cite{Molzon}. This would make the $\mu-e$ conversion constraints on
all the products of the $R_p\hspace{-1em}/\ \ $ trilinear coupling attainable in the $\mu-e$
conversion at the tree and at the 1-loop levels better than those from
the other processes in all the cases.
Note that the last four limits for $\lambda'\lambda$ in
Table 6 originate from the contribution of the strange nucleon sea. These
limits are comparable to the other $\mu^- \to e^-$ constraints derived from
the valence quarks contributions.
Finally we put constraints on the products of the bilinear $R_p\hspace{-1em}/\ \ $ parameters
evaluated from the $\mu-e$ conversion in ${}^{48}$Ti as
\begin{eqnarray}
\label{bilin}
\langle\tilde\nu_1\rangle\langle\tilde\nu_2\rangle, \mu_1\mu_2,
\langle\tilde\nu_1\rangle \mu_2, \langle\tilde\nu_2\rangle \mu_1 \
\leq \left(80\mbox{MeV}\right)^2
\left(\frac{\tilde m}{100\mbox{GeV}}\right)^{2} B,
\end{eqnarray}
where $B$ is a scaling factor defined in Eq. (\ref{scale}).
These constraints are weaker than those derived in Ref. \cite{BFK}
from the Super-Kamiokande atmospheric neutrino data.
The limits in Tables 5,6 and in Eq. (\ref{bilin})
were extracted under the following assumptions. We
assumed that all the scalar masses in Eq. (\ref{coeff12})
are equal $\tilde m_{{u}L(n)}\approx \tilde m_{{d}L,R(n)}
\approx \tilde m_{\nu(n)}\approx \tilde m$ and also that
there is no significant compensation between the different
terms contributing to the ratio $R_{\mu e^-}$.
Note that the last assumption is, in practice,
equivalent to the other well known assumptions about the dominance of only
one parameter or product of the parameters at a time.
These assumptions are widely used in the literature for
derivation of constraints on the R-parity violating parameters.
Instead of considering only one specific combination of $\lambda$ and
$\lambda^\prime$ as dominant, one may attempt to consider all
combinations of the R-parity violating couplings. To this end one needs
constraints on the relative magnitudes of the R-parity violating
couplings. This can be accomplished in a fashion analogous to the theory
of textures for the Yukawa couplings. This theory, which expresses the
entries of the down fermion mass matrix in powers of a parameter
$\bar \epsilon = 0.23$, has been successful in describing the charged
fermion mass matrices (for review see \cite{lola:98} and references therein).
This can be extended to the R-parity violation
couplings themselves \cite{U1Rp,Leon-Riz}.
In fact using solution B of Ref. \cite{Leon-Riz}
we obtain
\begin{equation}
\lambda^{\prime}_{112}=\lambda^{\prime}_{121}=\lambda^{\prime}_{212}=
\lambda^{\prime},
\qquad \lambda_{1 2 1} = {\bar \epsilon}^2 \lambda^{\prime}
\label{textures}
\end{equation}
This way, $\mu - e$ conversion can be used to constrain the overall scale of
the R-parity violating interaction using phonomenologically acceptable textures
\cite{Leon-Riz}.
\section{Summary and Conclusions.}
The transition matrix elements of the flavor violating $\mu^-- e^-$
conversion are of notable importance in computing accurately the corresponding
rates for each accessible channel of this exotic process.
Such calculations provide useful nuclear-physics inputs for the expected
new data from the PSI and MECO experiments to put more severe bounds on the
muon-number violating parameters
determining the effective currents
in various models that predict the exotic $\mu^-\to e^-$ process.
We developed a systematic approach which allows one to calculate
the $\mu^--e^-$ conversion rate in terms of the quark level Lagrangian
of any elementary model taking into account the effect of the nucleon
and nuclear structure.
Our conversion rate formula (\ref{Rme}) is valid for the interactions
with the (axial-)vector and (pseudo-)scalar quark and lepton currents,
as shown in Eq. (\ref{nucl1}). In the previous $\mu^- \to e^-$
calculations found
in the literature only the (axial-)vector currents were considered.
In the case of the R-parity violating interactions discussed here
we have investigated all the possible tree level contributions to the
$\mu-e$ conversion in nuclei.
We found new important contribution to $\mu^--e^-$ conversion
originating from the strange quark sea in the nucleon which
is comparable with the usual contribution of the valence $u,d$ quarks.
We introduced the quantities $\rho$ and $\cal Q$ defined
in Eqs. (\ref{rho}) and (\ref{Rme.1}) which can be associated with
theoretical sensitivities of a $\mu^--e^-$ conversion experiment
to the charged lepton flavor violating interactions discussed in the present
paper. These quantities are independent of a target material and, therefore,
might be helpful for comparison of searching potentials
of different $\mu-e$ conversion experiments and for planning future experiments.
>From the existing data on $R_{\mu e^-}$ in $^{48}$Ti and $^{208}$Pb
and the expected sensitivity of the designed MECO experiment \cite{Molzon}
we obtained stringent upper limits on the quantities $\rho$ and $\cal Q$.
Then we extracted the limits on the products of the trilinear $R_p\hspace{-1em}/\ \ $
parameters of the type $\lambda\lambda'$, $\lambda'\lambda'$
which are significantly more severe than those existing in the literature.
Let us conclude with the following important remark.
As was observed in Refs. \cite{Huitu,Kos-Kov}, if the ongoing
experiments at PSI \cite{Dohmen} and Brookhaven \cite{Molzon}
will have reached the quoted sensitivities in the branching ratio
$R_{\mu e^-}$ then the $\mu^- \to e^-$ constraints on all the products of
the $R_p\hspace{-1em}/\ \ $ parameters $\lambda\lambda$, $\lambda'\lambda$,
$\lambda'\lambda'$ measurable in $\mu-e$ conversion
will become more stringent than those from any other processes.
This is especially important since no comparable improvements of
the other experiments probing these couplings is expected in the near future.
\vskip10mm
\noindent
{\bf Acknowledgments}\\
The research described in this publication was made possible
in part by the INTAS grant 93-1648-EXT and
Fondecyt (Chile) under grant 1000717. JDV would like to express
his appreciation to the Humboldt Foundation for their award and
thanks to the Institute of Theoretical Physics at University of
T\"ubingen for hospitality.
S.K. thanks V.A. Bednyakov and F. Simkovic for discussions.
\bigskip
|
\section{\@startsection {section}{1}{\z@}{-3.5ex plus -1ex minus
-.2ex}{2.3ex plus .2ex}{\large\bf}}
\def\subsection{\@startsection{subsection}{2}{\z@}{-3.25ex plus%
-1ex minus -.2ex}{1.5ex plus .2ex}{\sc}}
\advance \voffset by -0.9in
\advance \hoffset by -0.8in
\textheight=9.5in
\textwidth=6.8in
\newcommand{\mathbb{N}}{\mathbb{N}}
\newcommand{\mathbb{Z}}{\mathbb{Z}}
\newcommand{\mathbb{Q}}{\mathbb{Q}}
\newcommand{\mathbb{R}}{\mathbb{R}}
\newcommand{\mathbb{C}}{\mathbb{C}}
\newcommand{{\rm tr}}{{\rm tr}}
\def\begin{eqnarray}{\begin{eqnarray}}
\def\end{eqnarray}{\end{eqnarray}}
\def\halmos{\hbox{\vrule height0.31cm width0.01cm\vbox{\hrule height
0.01cm width0.3cm \vskip0.29cm \hrule height 0.01cm width0.3cm}\vrule
height0.31cm width 0.01cm}}
\def\hhalmos{{\unskip\nobreak\hfil\penalty50
\quad\vadjust{}\nobreak\hfil\halmos
\parfillskip=0pt\finalhyphendemerits=0\par}}
\let\sPP=\smallbreak
\let\mPP=\medbreak
\let\bPP=\bigbreak
\def\par\noindent{\par\noindent}
\def\smallbreak\noindent{\smallbreak\noindent}
\def\medbreak\noindent{\medbreak\noindent}
\def\bigbreak\noindent{\bigbreak\noindent}
\def\alpha{\alpha}
\def\beta{\beta}
\def\gamma{\gamma}
\def\delta{\delta}
\def\varepsilon{\varepsilon}
\def\kappa{\kappa}
\def\Delta{\Delta}
\def{\rm id}{{\rm id}}
\let\lan=\langle
\let\ran=\rangle
\let\ten=\otimes
\let\wb=\overline
\newcommand{\bra}[1]{\langle\,#1|}
\newcommand{\ket}[1]{|#1\,\rangle}
\newcommand{\sket}[1]{#1\,\rangle}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{definitionn}[theorem]{Definition}
\newtheorem{remark}[theorem]{Remark}
\begin{document}
\baselineskip 18pt
\parskip 7pt
\begin{flushright}
ITFA-99-6, UvA-Math-99-08 \\
math.QA/9904029
\end{flushright}
\vspace{0.7cm}
\begin{center}
\baselineskip 24 pt
{\LARGE Fourier transform and the Verlinde formula for} \\
{\LARGE the quantum double of a finite group}
\vspace{1cm}
\baselineskip 18pt
{\Large
T.H. Koornwinder\footnote{e-mail: {\tt <EMAIL>}} }\\
KdV Institute for Mathematics, University of Amsterdam\\
Plantage Muidergracht 24, 1018 TV Amsterdam\\
The Netherlands\\
\vspace{1cm}
{\Large B.J. Schroers \footnote{e-mail: {\tt <EMAIL>}} }\\
Department of Mathematics and Statistics, University of Edinburgh\\
King's Buildings, Mayfield Road, Edinburgh EH9 3JZ\\
United Kingdom\\
\vspace{1cm}
{\Large
J.K. Slingerland\footnote{e-mail: {\tt <EMAIL>}} and
F.A. Bais\footnote{e-mail: {\tt <EMAIL>}} }\\
Institute for Theoretical Physics, University of Amsterdam\\
Valckenierstraat 65, 1018 XE Amsterdam\\
The Netherlands
\vspace{0.5cm}
4 October 1999
\vspace{0.3cm}
\end{center}
\date{\today}
\begin{abstract}
\noindent We define a Fourier transform $S$ for the quantum double
$D(G)$ of a finite group $G$. Acting on characters of $D(G)$, $S$ and the
central ribbon element of $D(G)$ generate a unitary matrix
representation of the group $SL(2,\mathbb{Z})$. The characters form
a ring over the integers under both the algebra multiplication
and its dual, with the latter encoding the fusion rules of $D(G)$.
The Fourier transform relates the two ring structures. We use
this to give a particularly short proof of the Verlinde formula
for the fusion coefficients.
\end{abstract}
\vspace{1cm}
\section{Introduction}
The quantum double $D(G)$ of a
finite group $G$ is a quasi-triangular ribbon Hopf
algebra \cite{qugroups} constructed,
via Drinfeld's double construction \cite{Drinfeld},
out of the Hopf algebra $C(G)$ of ${\mathbb{C}}$-valued functions on $G$.
Such quantum doubles arise in physics in
orbifold conformal field theories \cite{DVVV} and in the classification
of flux-charge composites in the massive phases of (2+1)-dimensional
gauge theories \cite{BDP},\cite{BP}. The mathematical structure of $D(G)$
was clarified in \cite{DPR}.
There and in \cite{DVVV} it was also pointed out that
the set of irreducible representations of $D(G)$ carries a
representation of the group $SL(2,\mathbb{Z})$ by unitary, symmetric
matrices. In particular one has a symmetric and unitary matrix
$S$ and a diagonal, unitary matrix $T$ acting on the
set of irreducible representations which satisfy the modular
relation $(ST)^3=S^2$ and $S^4=1$. Although perhaps surprising from the
point of view of Hopf algebras, the appearance of the
$SL(2,\mathbb{Z})$ action in the representation theory of $D(G)$
is physically motivated by application
of $D(G)$ in orbifold conformal field theories.
In particular, it was already pointed out
in \cite{DVVV} that the matrix $S$ plays the role of the Verlinde
matrix which diagonalises the fusion rules in orbifold conformal field
theory (for a general review
of fusion rules in conformal field theory
we refer the reader to \cite{Fuchs}).
As a result one has a Verlinde formula \cite{erik}
for integer fusion coefficients
in terms of (generally non-integer) matrix elements of the
Verlinde matrix $S$.
A central goal of the present paper is to understand the role
of the group $SL(2,\mathbb{Z})$ in the representation theory of $D(G)$
without reference to conformal field theory.
Our starting point here is the set of characters of irreducible
representations of $D(G)$. The group $SL(2,\mathbb{Z})$ acts on this
set in a geometrically natural way. We identify two generators
$S$ and $T$ of this action (satisfying$(ST)^3=S^2$ and $S^4=1$)
which play a natural role in the theory of
$D(G)$ and its dual $D(G)^\star$.
It was already noted in
\cite{DPR} that the diagonal matrix $T$ is related to
the central ribbon element of $D(G)$. As a vector spaces, both
$D(G)$ and $D(G)^\star$ can be identified with the space $C(G\times G)$
of $\mathbb{C}$-valued functions on $G\times G$, and here we show
that $S$ can be extended to an automorphism of the vector space
$C(G\times G)$.
We prove a convolution theorem
for this extension which shows that it has a natural interpretation
as a Fourier transform.
Returning to the set of characters we show that it
can be given a ring structure
in two dual ways. One, using the algebra multiplication,
is essentially determined by Schur orthogonality relations.
The other, using the dual multiplication, encodes the fusion
rules of $D(G)$. Our Fourier transform relates the two ring
structures, and we use this to give a very short proof of the
Verlinde formula for $D(G)$.
Quantum doubles can also be defined for locally compact
groups $G$ \cite{KM} and we have used a notation which
anticipates the generalisation of the arguments given here
to quantum doubles of locally compact groups. There are
a number of technical problems, however, which we intend
to address in a future publication.
Finally, we observe that the Fourier transform we will
define in this paper is related to the non-abelian Fourier transform
defined by Lusztig in the context of finite
group theory, see \cite{Lusztig} and \cite{charred}, and to the
quantum Fourier transform defined by Lyubashenko in \cite{Ly2}
and discussed further by Lyubashenko and Majid in
\cite{LyMa2} and \cite{LyMa1}.
We will clarify the relationship between these definitions and
ours in the course of the paper.
There are several other places in the literature where
Fourier transforms are discussed in the context of quantum groups.
The focus of the papers \cite{KeMa} and \cite{BKLT} is braided quantum groups
and thus different from ours.
In section 3.4 of \cite{Chryss} a Fourier transform is defined for finite
dimensional Hopf algebras. However, when applied to the quantum double of a
finite group that definition yields something essentially different
from our Fourier transform.
\section{The quantum double of a finite group}
Let $G$ be a finite group, with
invariant measure given by
\begin{eqnarray}
\label{maat}
\int_G f(z)\,dz:=|G|^{-1}\,\sum_{z\in G} f(z).
\end{eqnarray}
We will use delta functions $\delta_x$ on $G$,
normalised so that $\delta_x(y)=0$ if $x\ne y$ and
$\delta_x(x)=|G|$.
The quantum double $D(G)$ of a finite group $G$ was first
studied in detail in \cite{DPR}.
The definitions we are about to give are equivalent to the ones
given there, but
we adopt a different notation. The advantage of our notation is
that it easily generalises to the case where $G$ is a locally
compact Lie group \cite{KM}.
As a linear space we identify the quantum double $D(G)$ of $G$
with $C(G\times G)$. On $D(G)$ we have a non-degenerate pairing
\begin{eqnarray}
\lan f_1,f_2 \ran:=\int_G\!\int_G f_1(x,y)\,f_2(x,y)\,dx\,dy.
\end{eqnarray}
By this pairing we can also identify the dual $D(G)^\star$ of $D(G)$ with
$C(G\times G)$ as a linear space.
On $D(G)$ we have multiplication $\bullet$, identity 1,
comultiplication $\Delta$,
counit $\varepsilon$, antipode $\kappa$ and involution ${}^*$ given by:
\begin{eqnarray}
\label{algebra}
(f_1\bullet f_2)(x,y)&:=&\int_G f_1(x,z)\,f_2(z^{-1}xz,z^{-1}y)\,dz,
\nonumber \\
1(x,y)&:=&\delta_e(y),\nonumber \\
(\Delta f)(x_1,y_1;x_2,y_2)&:=&f(x_1x_2,y_1)\,\delta_{y_1}(y_2).\nonumber \\
\varepsilon(f)&:=&\int_G f(e,y)\,dy,\nonumber \\
(\kappa f)(x,y)&:=&f(y^{-1}x^{-1}y,y^{-1}),\nonumber \\
f^*(x,y)&:=&\wb{f(y^{-1}xy,y^{-1})}.
\end{eqnarray}
By duality we have multiplication $\star$, identity $\iota$, comultiplication
$\Delta^\star$, counit $\varepsilon^\star$, antipode $\kappa^\star$ and involution ${}^\circ$
on $D(G)^\star$:
\begin{eqnarray}
\label{coalgebra}
(f_1\star f_2)(x,y)&:=&\int_G f_1(z,y)\,f_2(z^{-1}x,y)\,dz,\nonumber \\
\iota(x,y)&:=&\delta_e(x),\nonumber \\
(\Delta^\star f)(x_1,y_1;x_2,y_2)&:=&f(x_1,y_1y_2)\,\delta_{x_2}(y_1^{-1}x_1y_1),
\nonumber \\
\varepsilon^\star(f)&:=&\int_G f(x,e)\,dx,\nonumber \\
(\kappa^\star f)(x,y)&:=&f(y^{-1}x^{-1}y,y^{-1}).\nonumber \\
f^\circ(x,y)&:=&\wb{f(x^{-1},y)}.
\end{eqnarray}
Later, we will also refer to the ribbon algebra structure of
$D(G)$. Following the conventions for
ribbon Hopf algebras of sect. 4.2 in \cite{qugroups},
we define the universal
$R$-matrix $R\in D(G)\otimes D(G)$:
\begin{eqnarray}
R(x_1,y_1;x_2,y_2) = \delta_e(y_1)\delta_e(x_1y_2^{-1})
\end{eqnarray}
and the central ribbon element $c \in D(G)$:
\begin{eqnarray}
\label{randc}
c(x,y) = \bullet\circ(\kappa\otimes{\rm id})(R_{21}) = \delta_e(xy),
\end{eqnarray}
where $R_{21}(x_1,y_1;x_2,y_2):=R(x_2,y_2;x_1,y_1)$.
The monodromy element $Q\in D(G)\times D(G)$ is
\begin{eqnarray}
\label{monodromy}
Q(x_1,y_1;x_2,y_2):=\bigl(R_{21}\bullet R\bigr)
(x_1,y_1;x_2,y_2)=\delta_{y_1}(x_2)
\delta_{y_2}(x_2^{-1}x_1x_2).
\end{eqnarray}
Together with $c$, it satisfies the ribbon relation
\begin{eqnarray}
\label{ribbon}
\Delta c=Q^{-1} \bullet \bigl( c\otimes c \bigr).
\end{eqnarray}
In the representation theory of $D(G)$ and $D(G)^\star$
an important role
is played by the Haar functionals $h^\star \colon D(G)^\star \to\mathbb{C}$
and $h \colon D(G) \to \mathbb{C}$, respectively. They are given by
\begin{eqnarray}
\label{haar}
h^\star (f):=\int_G f(e,y)\,dy \quad \mbox{and} \quad
h (f):=\int_G f(x,e)\,dx
\end{eqnarray}
Here we have chosen the normalisation
$h^\star(\iota)=h(1)=|G|$.
Direct computation shows left and right invariance:
\begin{eqnarray}
\label{haarprop}
\bigl((h^\star\ten{\rm id})\circ\Delta^\star \bigr)(f)=h^\star(f)\,\iota=
\bigl(({\rm id}\ten h^\star)\circ\Delta^\star \bigr)(f),
\end{eqnarray}
and similarly for $h$. Furthermore,
centrality, positivity and faithfulness of $h$ and $h^\star$
follow from the formulae
\begin{eqnarray}
\label{inpro}
h(f_1\bullet f_2^*)=h^\star(f_1\star f_2^\circ)=
\int_G\int_G f_1(x,y)\,\wb{f_2(x,y)}\,dx\,dy
\end{eqnarray}
and
\begin{eqnarray}
h^\star(f\star f^\circ)=\int_G\int_G |f(x,y)|^2\,dx\,dy.
\end{eqnarray}
Thus, $C(G\times G)$ has a hermitian inner product
\begin{eqnarray}
\label{inprod}
(f_1,f_2)\mapsto h^\star(f_1\star f_2^\circ)=\lan f_1,\wb{f_2}\ran.
\end{eqnarray}
From the existence of faithful
positive linear functionals on $D(G)$ and $D(G)^\star$
(namely $h$ and $h^\star$) we can conclude that $D(G)$ and
$D(G)^\star$ both have a faithful $*$-representation on a finite dimensional
Hilbert space, so they are $C^*$-algebras. Therefore, the theory of
Woronowicz \cite{woronowicz}
for compact matrix quantum groups holds both for $D(G)$
and $D(G)^\star$. Moreover, this theory simplifies because we are in the
finite dimensional case, see \cite{woronowicz}, Appendix A.2, and
\cite{vanDaele}.
These simplifications are already evident in our explicit results that
$\kappa^2={\rm id}$ and $(\kappa^\star)^2={\rm id}$, and that $h$ and $h^\star$ are central.
Furthermore, in \cite{vanDaele} after the proof of Proposition 2.2,
van Daele gives a
formula in terms of dual bases for the Haar functional in the finite
dimensional $C^*$-algebra case.
For $D(G)^\star$ this can be written as
\begin{eqnarray}
h^\star(f)={\rm const.} \sum_{x,y\in G}(f\star \delta_{x,y})(x,y).
\end{eqnarray}
A simple calculation indeed shows that this agrees with (\ref{haar}).
An analogous formula holds for $h$.
\section{Irreducible representations and their characters}
The irreducible representations of the quantum double of a finite
group were classified in \cite{DPR}. We adopt some of the notation
used there, but for our description of the representations
we follow the approach used in the discussion of the double
of a locally compact group in \cite{KM} and \cite{KBM}.
Thus let $\{C_{A}\}_{A=0,...,p}$ be
the set of conjugacy classes in $G$, with $C_0=\{e\}$. In each
$C_A$ pick an element $g_A$ and write $N_A$ for the centraliser of
$g_A$ in $G$. For later use it is also convenient to pick, for
each $x\in G$,
an element $B_x\in G$ such that
\begin{eqnarray}
\label{bdef}
B_xg_A B_x^{-1} =x \quad \mbox{if} \quad x \in C_A.
\end{eqnarray}
Write $q_A$ for the number of irreducible representations of $N_A$
and let $\{\pi_\alpha\}$ be a complete set
of such representations. The label $\alpha$ is a positive integer
running from 1 (for the trivial representation) to $q_A$.
We denote the carrier spaces by $V_{\alpha}$ and their dimensions
by $d_{\alpha}$. Then
the irreducible representations $\pi^A_{\alpha}$
of $D(G)$ are labelled by pairs
$(A,\alpha)$ of conjugacy classes and centraliser representations.
The carrier space $V^A_{\alpha}$ of $\pi^A_{\alpha}$ is
\begin{eqnarray}
\label{irrep}
V^A_{\alpha}:=\{\phi:G\rightarrow V_{\alpha}|\phi(xn)=\pi_\alpha(n^{-1})\phi(x),
\forall x\in G,\forall n\in N_A\},
\end{eqnarray}
and the action of an element $f\in D(G)$ is
\begin{eqnarray}
\label{action}
\left(\pi^A_{\alpha}(f)\phi\right)(x):=\int_G f(x g_A x^{-1},z)
\, \phi(z^{-1}x)\, dz.
\end{eqnarray}
The set $\{\pi^A_{\alpha}\}$ is a complete set of mutually inequivalent
irreducible matrix $*$-representations of $D(G)$.
We write $d_{A,\alpha}$ for the dimension of $V^A_{\alpha}$ and note that
$d_{A,\alpha} = |C_A|\cdot d_\alpha$.
Then, after choosing an orthonormal basis of $V^A_{\alpha}$,
$\pi^A_\alpha$ can be represented by a matrix $(\pi^A_\alpha)_{ij}$,
$i,j=1,...,d_{A,\alpha}$. The matrix elements $(\pi_{\alpha}^A)_{ij}$
are elements of $D(G)^\star $ and we write
$M_{A,\alpha}$ for the span of the matrix elements $(\pi^A_\alpha)_{ij}$
($i,j=1,\ldots,d_{A,\alpha}$). Then it follows from Woronowicz's general
theory that $D(G)^\star$ is the orthogonal direct sum
of the spaces $M_{A,\alpha}$. Finally we define the
character
\begin{eqnarray}
\label{chardef}
\chi^A_\alpha=\sum_{i=1}^{d_{A,\alpha}}( \pi^A_\alpha)_{ii}.
\end{eqnarray}
Characters play a fundamental role in the following discussion. {}From
\cite{KBM} we have the following formula:
\begin{eqnarray}
\label{char}
\chi^A_{\alpha}(f)= \int_G\int_{N_A}f(zg_Az^{-1}, znz^{-1})
\,\chi_{\alpha}(n)\,dn\,dz.
\end{eqnarray}
Changing integration variables, this can be rewritten as
\begin{eqnarray}
\label{chara}
\chi^A_{\alpha}(f)= \int_G\int_G\, f(v,w) \,
{\bf 1}_A(v)\,\,\delta_e(vwv^{-1}w^{-1})\,\,\chi_\alpha
(B^{-1}_v wB_v) \,dv\,dw,
\end{eqnarray}
where
${\bf 1}_A$ is the characteristic function of the conjugacy class $C_A$,
normalised so that
${\bf 1}_A(v)= 1 $ if $v\in C_A$ and ${\bf 1}_A(v)= 0 $ otherwise.
By definition, characters are elements of $D^\star (G)$. Using the
pairing $\langle\, ,\rangle$ we can therefore identify them with functions
on $G\times G$. To do this explicitly we
insert a delta function for $f$,
\begin{eqnarray}
\label{dprchar}
\chi^A_{\alpha}(x,y):=\chi^A_{\alpha}(\delta_x\otimes\delta_y) =
\delta_e(xyx^{-1}y^{-1})\,{\bf 1}_A(x) \, \chi_{\alpha}(B_x^{-1}yB_x),
\end{eqnarray}
and reproduce the character formula given in \cite{DPR}.
One checks that the characters
enjoy the orthogonality relation
\begin{eqnarray}
\label{ortho}
\langle\chi_{\alpha}^A,\wb{\chi_{\beta}^B}\rangle =
|G|\,\,\delta_{\alpha\beta}\,\delta_{AB}.
\end{eqnarray}
As elements of $D( G)^\star$ characters have the
property that they are cocentral, i.e. they satisfy
$\Delta^\star \chi^A_\alpha = \sigma \circ \Delta^\star \chi^A_\alpha$,
where $\sigma: D(G)^\star
\times D(G)^\star\rightarrow D(G)^\star \times D(G)^\star $ is
the flip operation, $\sigma(\lambda,\mu) = (\mu,\lambda)$.
Using again the identification of $D(G)^\star$
with $C(G\times G)$, the cocentrality of the characters (\ref{dprchar})
means that their support lies in
\begin{eqnarray}
\label{comm}
G_{\rm comm}:=\{(x,y)\in G\times G|xy=yx\}.
\end{eqnarray}
and that they are
invariant under the simultaneous
conjugation of both arguments, in symbols
$\chi^A_\alpha(gxg^{-1},gyg^{-1}) =\chi^A_\alpha(x,y)$ for all $g,x,y\in G$.
These properties are also evident in the explicit expression (\ref{dprchar}).
We write $C(G_{\rm comm})$ for the space of functions in $C(G\times G)$
whose support lies in $G_{\rm comm}$, and $C^{0}(G_{\rm comm})$
for the space of functions in $C(G_{\rm comm})$ which
are invariant under the simultaneous conjugation of both arguments.
It follows from the remark in \cite{woronowicz} after Corollary 5.10 that
the characters in fact span $C^{0}(G_{\rm comm})$. It is instructive
to see this explicitly. The dimension of $C^{0}(G_{\rm comm})$
is equal to the number of $G$-conjugacy
classes in $G_{\rm comm}$. To count these,
introduce an integer label $a$ for $N_A$-conjugacy classes in $N_A$.
Since the number of such conjugacy classes is equal
the number of irreducible representations, $a$ runs from $1$ to $q_A$.
In the $N_A$-conjugacy class labelled by $a$ pick an element $g^a_A$.
Then every $G$-conjugacy class in $ G_{\rm comm}$ contains a unique element
of the form $(g_A,g_A^a)$ for suitable $A$ and $a$.
The number of conjugacy classes is therefore
\begin{eqnarray}
\mbox{dim}\bigr( C^{0}(G_{\rm comm})\bigr) = \sum_{A=0}^p q_A.
\end{eqnarray}
This is also the number of irreducible representations $(A,\alpha)$ of
$D(G)$ and hence the number of characters.
By the orthogonality relation, the characters
are certainly linearly independent and therefore form
an orthogonal basis of $C^{0}(G_{\rm comm})$.
The vector space $C^{0}(G_{\rm comm})$
is closed under both the multiplication $\bullet $
and the dual multiplication
$\star $. This means that the vector space spanned by
the characters can be given two algebra structures which
are dual to each other. Both these algebras are initially
defined over the field $\mathbb{C}$, but in the following section we will show
that the structure constants for both algebras are integers.
Therefore the integer linear combinations
of characters form a ring over $\mathbb{Z}$
under both the multiplications $\bullet$
and $\star$.
\section{Character rings}
Again following the general theory given in \cite{woronowicz},
the matrix elements $(\pi_{\alpha}^A)_{ij}$ form a complete set
of mutually inequivalent irreducible matrix
corepresentations $\pi^A_{\alpha}$ of
$D(G)^\star $. We therefore have
\begin{eqnarray}
\label{costar}
\Delta^\star \bigl((\pi^A_\alpha)_{ij}\bigr)=\sum_k
(\pi^A_\alpha)_{ik}\ten(\pi^A_\alpha)_{kj}.
\end{eqnarray}
The quantum analogues of Schur's orthogonality relations, given
in \cite{woronowicz}, simplify for $D(G)$
because \hbox{$(\kappa^\star )^2={\rm id}$}. We have
\begin{eqnarray}
\lan (\pi^A_\alpha)_{ij}, (\wb{\pi}^B_\beta)_{kl}\ran= h^\star
((\pi^A_\alpha)_{ij} \star
(\pi^B_\beta)_{kl}^\circ)=\delta_{\alpha\beta}\,\delta_{AB}\,
\delta_{ik}\,\delta_{jl}\,
h^\star(\iota)/d_{A,\alpha}.
\end{eqnarray}
Note that due to a standard theorem in
the theory of finite groups, see \cite{Serre} or \cite{groupth},
the ratio
\begin{eqnarray}
\label{ndef}
n^A_\alpha:=h^\star(\iota)/d_{A,\alpha}=|N_A|/d_\alpha
\end{eqnarray}
is an integer. These relations
are sufficient to establish the following theorem.
\begin{theorem}
\label{projthm}
The map $f\mapsto (n^A_\alpha)^{-1}\,
\chi^A_\alpha\bullet f$ is the orthogonal projection of $D(G)^\star$ onto
$M_{A,\alpha}$.
\end{theorem}
{\bf Proof}\quad We have:
$$
\lan\chi^A_\alpha\bullet(\pi^B_\beta)_{ij},\wb{(\pi^C_\gamma)_{kl}}\ran
=
\lan \chi^A_\alpha\ten(\pi^B_\beta)_{ij},\Delta^\star \wb{(\pi^C_\gamma)_{kl}}\ran
=
\sum_r \lan\chi^A_\alpha,\wb{(\pi^C_\gamma)_{kr}}\ran\,
\lan(\pi^B_\beta)_{ij},\wb{(\pi^C_\gamma)_{rl}}\ran
$$
$$
=
\sum_{m,r}\lan(\pi^A_\alpha)_{mm},\wb{(\pi^C_\gamma)_{kr}}\ran\,
\lan(\pi^B_\beta)_{ij},\wb{(\pi^C_\gamma)_{rl}}\ran
=\delta_{AC}\,\delta_{BC}\,
\delta_{\alpha\gamma}\,\delta_{\beta\gamma} \,\delta_{ik}\,\delta_{jl}\, \bigl(n^A_\alpha\bigr)^2.
\eqno \halmos
$$
As an immediate consequence we note:
\begin{lemma}
\label{bul}
The characters of the quantum double $D(G)$ of a finite group $G$
form a ring over $\mathbb{Z}$ under the multiplication
$\bullet$. The multiplication rule is
\begin{eqnarray}
\label{bulalg}
\chi^A_\alpha\bullet \chi^B_\beta = \delta_{AB}\,\delta_{\alpha\beta}\,
n^A_\alpha \,\chi^A_\alpha.
\end{eqnarray}
\end{lemma}
Next consider the algebra structure of the characters under the
dual multiplication $\star $. This is related to the tensor product
decomposition into irreducible representations:
\begin{eqnarray}
\label{fusion}
\pi^A_\alpha \otimes \pi^B_\beta \simeq \bigoplus_{C,\gamma}
N_{\alpha\beta C}^{A B \gamma} \pi^C_\gamma.
\end{eqnarray}
We will refer to
the non-negative integers $N_{\alpha\beta C}^{A B \gamma}$ as fusion
coefficients.
By definition of the dual multiplication we have,
for $\pi,\rho \in D(G)^\star $
and $f\in D(G)$
\begin{eqnarray}
\label{reminder}
\lan\pi\otimes\rho, \Delta (f)\ran = \lan\pi\star \rho, f\ran.
\end{eqnarray}
Thus, upon taking the trace we find that for all $f\in D(G)$
\begin{eqnarray}
{\rm tr}\bigl(\pi^A_\alpha \otimes \pi^B_\beta \bigl(\Delta(f)\bigr)\bigr)=
\sum_{C,\gamma} \,N_{\alpha\beta C}^{A B \gamma}\, {\rm tr}\bigl( \pi^C_\gamma (f)\bigr).
\end{eqnarray}
Using (\ref{reminder}) we deduce the following lemma.
\begin{lemma}
\label{starr}
The characters of the quantum double $D(G)$ of a finite group $G$
form a ring over $\mathbb{Z}$ under the multiplication
$\star$. The multiplication rule is
determined by the fusion coefficients
$N_{\alpha\beta C}^{A B \gamma}$:
\begin{eqnarray}
\label{starmult}
\chi^A_\alpha \star \chi^B_\beta = \sum_{C,\gamma} \,N_{\alpha\beta C}^{A B \gamma}
\,\chi^C_\gamma.
\end{eqnarray}
\end{lemma}
\section{$SL(2,\mathbb{Z})$-action, Fourier transform and the Verlinde formula}
In this section, a central role is played
by a natural action of the group $SL(2,\mathbb{Z})$
of integer unimodular $2\times 2$ matrices on space $C(G_{\rm comm})$.
Let
\begin{eqnarray}
M= \left(
\begin{array}{cc}
a&b\\
c&d
\end{array}\right),
\end{eqnarray}
with $a,b,c,d$ integers such that $ad-bc=1$,
be a generic element of $SL(2,\mathbb{Z})$ and define a right action
on $G_{\rm comm}$ via
\begin{eqnarray}
(x,y) M: = (x^ay^c,x^b y^d).
\end{eqnarray}
This induces an action of $M \in SL(2,\mathbb{Z})$ on elements
$f\in C(G_{\rm comm})$, which we write as
\begin{eqnarray}
(Mf)(x,y):= f(x^ay^c,x^b y^d).
\end{eqnarray}
The generators
\begin{eqnarray}
S=
\left(
\begin{array}{cc}
0&-1\\
1&0
\end{array}\right), ~~~~~
T=
\left(
\begin{array}{cc}
1&1\\
0&1
\end{array}\right)
\end{eqnarray}
satisfy the modular relation
\begin{eqnarray}
\label{modrel}
( ST)^3= S^2
\end{eqnarray}
and $S^4=1$,
and both have a natural interpretation
within the quantum double $D(G)$. To see this, first note that
the actions
\begin{eqnarray}
\label{four}
(Sf)(x,y) =f(y,x^{-1})
\end{eqnarray}
and
\begin{eqnarray}
\label{spinf}
(Tf)(x,y)= f(x,xy)
\end{eqnarray}
also make sense for general $f\in C(G\times G)$.
We keep the notation (\ref{four}) and (\ref{spinf})
even when the arguments $x$ and $y$ do not commute. Note that
$S$ and $T$ are unitary operators on $C(G\times G)$ with the inner
product (\ref{inprod}).
Moreover, one finds that the action of $T$ on $f\in C(G\times G)$ is equal to
the multiplication of $f$ by the central element $c$ (\ref{randc}):
\begin{eqnarray}
\label{spinff}
Tf =c\bullet f.
\end{eqnarray}
Acting on $ C(G\times G)$, $S$ and $T$ no longer satisfy the
modular relation (\ref{modrel}) but we still have $S^4=1$.
This last property and the following convolution
theorem are our reasons for calling $S$ a Fourier transform.
\begin{theorem}
\label{invol}
If $f,g\in C(G\times G)$ and ${\rm supp}(f)\in G_{\rm comm}$ then
\begin{eqnarray}
\label{fourthe}
S(f\star g)=S(f) \bullet S( g)\quad \hbox{and } \quad
S(f\bullet g)=S(g)\star S( f).
\end{eqnarray}
\end{theorem}
\medbreak\noindent
{\bf Proof}\quad If the first factor in a $\bullet$-product has
support in $ G_{\rm comm}$, the formulae simplify, yielding
\begin{eqnarray}
\bigl(S(f)\bullet S(g)\bigr)(x,y)
&=&\int_G f(z,x^{-1}) g(z^{-1}y, x^{-1}) \,dz\nonumber \\
&=& \bigl( S(f\star g)\bigr)(x,y). \nonumber
\end{eqnarray}
Similarly we have
\begin{eqnarray}
\bigl(S(g)\star S(f) \bigr)
(x,y) &=& \int_G g(y,z^{-1}) f(y,x^{-1}z)\,dz \nonumber \\
&=& \int_G f( y, w ) \,g (y,w^{-1} x^{-1})\,dw \nonumber \\
&=& \bigl(S(f\bullet g)\bigr)(x,y),\nonumber
\end{eqnarray}
where we have again used
${\rm supp}(f)\in G_{\rm comm}$ and exploited the
invariance of the measure under
the change of integration variable $w= x^{-1}z$.\hhalmos
Since $S$ leaves the space of functions with support in $G_{\rm comm}$
invariant, we deduce the following
\begin{corollary}
\label{sinv}
If $f,g\in C(G\times G)$ and ${\rm supp}(f)\in G_{\rm comm}$ then
\begin{eqnarray}
S^{-1}(f\bullet g)=S^{-1}(f)\star S^{-1}(g)\quad \hbox{and } \quad
S^{-1}(g\star f)= S^{-1}(f)\bullet S^{-1}(g).
\end{eqnarray}
\end{corollary}
At this point it is instructive to make contact with
related discussions of Fourier transforms in the literature.
By defining a slight variant of the operator $S$ we can establish
a connection with
the non-abelian Fourier transform given by
Lusztig in \cite{Lusztig} and \cite{charred}.
For $f\in C(G\times G)$ put
\begin{eqnarray}
(Uf)(x,y):=f(y,x),\quad
(J_1f)(x,y):=f(x^{-1},y),\quad
(J_2f)(x,y):=f(x,y^{-1}).
\end{eqnarray}
Then $U=J_2S=SJ_1$. The operators $U$, $J_1$, $J_2$ correspond to
$2\times 2$ matrices with integer entries but with determinant $-1$.
Note also that the operators
$U$, $J_1$ and $J_2$, like $S$, leave the
space of functions
with support in $G_{\rm comm}$ invariant and commute with conjugations by
elements of $G$. We can therefore in particular
consider the restriction of $U$
to $C^0(G_{\rm comm})$.
It turns out that the matrix elements of $U$ with respect to the basis of
characters formally coincide with
Lusztig's Fourier kernel:
\begin{eqnarray}
\label{umatrix}
U^{BA}_{\beta\alpha}:&=&|G|^{-1}\lan U\chi_\alpha^A,\wb{\chi_\beta^B}\ran
\nonumber \\
&=&{1\over |N_A|\,|N_B|}\,
\sum_{\textstyle{g\in G\atop g_A\,gg_Bg^{-1}=gg_Bg^{-1}\,g_A}}
\chi_\alpha(gg_Bg^{-1})\,\wb{\chi_\beta(g^{-1}g_Ag)} \nonumber \\
&=&: \{(g_A,\alpha),(g_B,\beta)\}.
\end{eqnarray}
However, Lusztig takes $\{\,,\,\}$ with values in
the field $\wb \mathbb{Q}_l$, i.e., in an algebraic completion of the field
$\mathbb{Q}_l$ of l-adic numbers. He also has a bar operation on $\wb \mathbb{Q}_l$.
Straightforward computations shows that
the following analogues of Theorem (\ref{invol}) hold. For
$f,g\in C(G\times G)$ we have
\begin{eqnarray}
J_1(f)\star J_1(g)=J_1(g\star f),\quad
J_1(f)\bullet J_1(g)=J_1(f\bullet g)\quad
\mbox{and}\quad J_2(f)\star J_2(g)=J_2(f\star g).
\end{eqnarray}
If supp$(f)\in G_{\rm comm}$ one checks that furthermore
\begin{eqnarray}
U(f)\star U(g)=U(f\bullet g)\quad{\rm and}\quad
U(f)\bullet U(g)=U(f\star g).
\end{eqnarray}
Finally, if supp$(f)$ and supp$(g)\in G_{\rm comm}$ then
\begin{eqnarray}
J_2(f)\bullet J_2(g)=J_2(g\bullet f).
\end{eqnarray}
In the abstract setting of tensor categories, a quantum Fourier
transform
was defined by Lyubashenko in \cite{Ly2} and discussed further
in \cite{LyMa2} and \cite{LyMa1}
for finite-dimensional
factorisable ribbon Hopf algebras. Applied to $D(G)$ and in our
notation their formula for the Fourier transform $\tilde S$
of an element $f$ of $D(G)$ reads
\begin{eqnarray}
\label{shahn}
\tilde Sf: = (1\otimes h )\circ \bigl(R^{-1}\bullet (1\otimes f) \bullet
R_{21}^{-1}\bigr).
\end{eqnarray}
An explicit calculation shows that
\begin{eqnarray}
\label{fourti}
\bigl(\tilde S f\bigr)(x,y) = f(xy^{-1}x^{-1}, x).
\end{eqnarray}
and therefore the relation between $\tilde S$ and $S$ can
be expressed via
\begin{eqnarray}
\label{shrel}
\tilde S = \kappa \circ S.
\end{eqnarray}
The fourth power of the map $\tilde S$ is not equal to the identity,
but if, following \cite{LyMa2}, one defines $\tilde T = T^{-1}$ one
finds that the modular relation $ (\tilde S \tilde T)^3 = {\tilde S}^2$
is satisfied.
Convolution formulae similar to ours
can be proven for the map (\ref{shahn}). Again, at least one of the two
elements $f$ and $g$ to be multiplied has to have support
in $ G_{\rm comm}$. Finally, we observe that restricted to $C(G_{\rm comm})$,
the map $\tilde S$ agrees with our $S^{-1}$.
For the rest of this paper we focus our attention on
the space $C^{0}(G_{\rm comm})$.
In particular we consider the restriction of the map
$S$ to $C^{0}(G_{\rm comm})$, and again denote it by $S$.
The characters
form a natural orthogonal basis of the space
$C^{0}(G_{\rm comm})$,
and we
define the matrix $S^{BA}_{\beta\alpha}$ as the
matrix representing the map $S$ on the basis of characters:
\begin{eqnarray}
\label{verlmatrix}
S^{BA}_{\beta\alpha} := |G|^{-1}\,\lan S\chi^A_\alpha, \wb{\chi^B_\beta}\ran.
\end{eqnarray}
Here the normalisation is chosen so that
$S\chi^A_\alpha=\sum_{B,\beta} S^{B A}_{\beta\alpha} \chi^B_\beta$.
The matrix $S^{BA}_{\beta\alpha}$ is unitary because the map $S$ is.
Using the explicit expression for the characters (\ref{dprchar})
one finds the following formula, first given in \cite{DPR}:
\begin{eqnarray}
\label{matform}
S^{BA}_{\beta\alpha}=\int_G\int _G \delta_e(xyx^{-1}y^{-1})\,
\, {\bf 1 }_A(x)\, {\bf 1}_B(y)\,
\,\wb{\chi}_{\alpha}(B_x^{-1}yB_x)\,\,\wb{ \chi}_\beta(B_y^{-1}x B_y)\,\,dx\,dy.
\end{eqnarray}
This expression shows that the matrix $S^{AB}_{\alpha\beta}$ is symmetric,
$S^{AB}_{\alpha\beta}= S^{B A}_{\beta\alpha }$. Since it is also unitary
its inverse is given by its complex conjugate. We can also read off
the useful relation
\begin{eqnarray}
\label{dimform}
S^{ 0A}_{1 \alpha}=\frac{1}{n^A_\alpha}.
\end{eqnarray}
Armed with this notation, we can now use the convolution
theorem to relate the $\bullet$ and $\star $-ring structures
of the characters. The result is the Verlinde formula.
\begin{theorem}[Verlinde Formula] \quad
Acting on characters, the inverse Fourier transform
$S^{-1}$
diagonalises the fusion rules of $D(G)$.
The fusion coefficients can be expressed in terms
of the matrix $S^{AB}_{\alpha\beta}$ :
\begin{eqnarray}
N^{AB\gamma}_{\alpha\beta C}=
\sum_{D,\delta}\frac{ S^{DA}_{\delta\alpha}\,S^{DB}_{\delta\beta}\,
\wb{S}^{CD}_{\gamma\delta}}{S^{0D}_{1\delta}}.
\end{eqnarray}
\end{theorem}
\medbreak\noindent
{\bf Proof} \quad
It follows from the definition of $S$ and from Lemma (\ref{bul}) that
$$
S\chi^A_\alpha\bullet \chi^B_\beta =\frac{ S^{BA}_{\beta\alpha}} {S^{0B}_{1\beta}}\,\,
\chi^B_\beta.
$$
Now apply $S^{-1}$ to both sides and use
the first formula in Corollary (\ref{sinv})
to obtain
$$
\chi^A_\alpha\star S^{-1}\chi^B_\beta =\frac{ S^{AB}_{\alpha\beta}} {S^{0B}_{1\beta}}\,\,
S^{-1}\chi^B_\beta,
$$
yielding the diagonalised fusion rules, with eigenvalues
$ S^{AB}_{\alpha\beta}/ S^{0B}_{1\beta}$. A quick derivation
of the formula for
the fusion coefficients follows again
from the definition of $S$ and from Lemma (\ref{bul}):
$$
S\chi^A_\alpha\bullet S\chi^B_\beta =
\sum_{D,\delta}\frac{ S^{DA}_{\delta\alpha}\, S^{DB}_{\delta\beta}} {S^{0D}_{1\delta}}\,\,
\chi^D_\delta.
$$
Again apply $S^{-1}$ to both sides and use
the first formula in Corollary (\ref{sinv})
to obtain
$$
\chi^A_\alpha\star\chi^B_\beta = \sum_{C\gamma} \,\left(
\sum_{D,\delta}\frac{ S^{DA}_{\delta\alpha}\,S^{DB}_{\delta\beta}\,
\wb{S}^{CD}_{\gamma\delta}}{S^{0D}_{1\delta}}
\right) \,\chi_\gamma^C.
$$
Comparing this expression
with (\ref{starmult}) shows that the expression
in brackets is equal to the fusion coefficient
$N^{AB\gamma}_{\alpha\beta C}$.\hhalmos
\noindent{\bf Remark} \quad
There is an interesting connection with Lusztig's matrix
$U^{AB}_{\alpha \beta}$ (\ref{umatrix})
here. We find that $\wb{U_{\alpha\beta}^{AB}}=U_{\beta\alpha}^{BA}=
(U^{-1})_{\beta\alpha}^{BA}$ and that
$U_{\alpha\beta}^{AB}=S_{\alpha \wb \beta}^{AB}$.
Verlinde's formula can also be expressed in terms of $U$:
\begin{eqnarray}
\chi_\alpha^A\star U\chi_\beta^B={U_{\beta\alpha}^{BA}\over U_{ 1\beta}^{0B}}\,
U\chi_\beta^B
\end{eqnarray}
and
\begin{eqnarray}
N_{\alpha\beta C}^{AB\gamma}=
\sum_{D,\delta}{{U_{\delta\alpha}^{DA}}\,\,{U_{\delta\beta}^{DB}}\,\,
U_{\gamma\delta}^{CD}\over
U_{1\delta}^{0D}}\,.
\end{eqnarray}
\vspace{1.3cm}
The simplicity of our proof of the Verlinde formula
shows that the Fourier transform
$S$ is a very natural tool for proving the Verlinde formula for
$D(G)$. While we have restricted attention to a particular
ribbon Hopf algebra here, we have tried to indicate as far
as possible how our definitions and equations for $D(G)$
can be formulated using only natural operations (such the
Haar measure, the antipode, the universal $R$-matrix
or the central ribbon element)
which exist for a large class of (quasi-triangular ribbon)
Hopf algebras. More generally it is natural to ask
for which class of
(quasi) Hopf algebras a Fourier transform with analogous
properties can be defined. In view of the tight connection
between fusion rules in rational conformal field theory and
tensor decomposition rules in (quasi) Hopf algebras (see e.g.
\cite{FGV}, or \cite{Fuchs} for a review) such a generalised Fourier
transform, if it exists, could be expected to play an important role in
both Hopf algebra theory and conformal field theory.
\vspace{1cm}
\noindent {\bf Acknowledgements}
\noindent BJS was a post doc at the Instituut voor Theoretische Fysica
of the University of Amsterdam
while the research for this paper was carried out.
BJS thanks J\"urgen Fuchs for illuminating discussions about
fusion rules.
|
\section{Introduction}
The nature of the dominant energy source in ultraluminous infrared
galaxies (ULIGs: $L_{\rm ir} > 10^{12}\ L_\odot$)\footnote{$L_{\rm ir}
\equiv L (8-1000 \micron)$, computed from the observed infrared fluxes
in all four {\em IRAS} bands according to the prescription outlined in
Perault (1987)} is the object of a vigorous debate (see, e.g., review
by Sanders \& Mirabel 1996). Results from recent studies at
mid-infrared (e.g., Lutz et al. 1996, 1998; Genzel et al. 1998) and
radio wavelengths (Lonsdale et al. 1998; Smith et al. 1998a,b) suggest
that several ULIGs are powered predominantly by hot stars rather than
by an active galactic nucleus (AGN). However, the great majority of
the infrared galaxies in these samples are of relatively modest
luminosity, with only a few having $L_{\rm ir} > 2 \times 10^{12}\
L_\odot$. This distinction may be important as there is growing
evidence from optical/near-infrared spectroscopy that the frequency of
occurence of AGN among luminous infrared galaxies (LIGs: {\em L}$_{\rm
ir}$ $\ga$ 10$^{11}$ {\em L}$_\odot$) increases with increasing
infrared luminosity. Veilleux et al. (1995; hereafter VKSMS) carried
out a sensitive optical spectroscopic survey of a sample of 200 LIGs
and classified the nuclear spectra of these galaxies using a large
number of optical line-ratio diagnostics corrected for the underlying
stellar absorption features. VKSMS found that 7 of the 21 ULIGs in
their sample showed Seyfert characteristics. In contrast, only 19 of
the 161 objects with lower luminosities ($L_{\rm ir} < 10^{12}\
L_\odot$) were optically classified as Seyferts. More recently, Kim,
Veilleux, \& Sanders (1998; hereafter KVS) extended this type of
analysis to a subset of 45 ULIGs from the {\em IRAS} 1-Jy sample (Kim
1995; Kim \& Sanders 1998). The fraction of Seyfert 1 and 2 galaxies
among LIGs was found to increase dramatically above $L_{\rm ir}
\approx 10^{12.3}\ L_\odot$ with more than 50{\thinspace}\% of the galaxies
with $L_{\rm ir} > 10^{12.3}\ L_\odot$ having Seyfert
characteristics. Near-infrared spectroscopy of a representative subset
of ULIGs from the 1-Jy sample appears to confirm the presence of an
energetically important AGN in many of these Seyfert galaxies
(Veilleux, Sanders, \& Kim 1997, 1999b; hereafter VSK97 and VSK99,
respectively).
It is clearly important to verify the results of KVS using the entire
1-Jy sample. This sample provides a complete list of the brightest
ULIGs with F[60~$\mu$m] $>$ 1 Jy which is not biased toward `warm'
quasar-like objects. The `1-Jy' sample contains 118 objects with $z$ =
0.02 -- 0.27 and log~[L$_{\rm ir}$/L$_\odot$] = 12.00 -- 12.90. The
infrared luminosities of these objects therefore truly overlap with
the bolometric luminosities of optical quasars. The present paper
discusses the optical spectra of 63 of the 73 ULIGs from the 1-Jy
sample that were not observed by KVS, and combine these results with
our previously published optical and near-infrared data to look for
systematic trends with infrared luminosity among galaxies with $L_{\rm
ir} \approx 10^{10.5}-10^{13}\ L_\odot$.
The structure of this paper is as follows. Section 2 discusses the
procedures used to obtain and reduce the new data. The results
derived from these data are described and compared with those of
previous optical spectroscopic studies in \S 3. In \S 4, the
implications of our results on the nature of the dominant energy
source in these objects are discussed in the context of recent
investigations at longer and shorter wavelengths. The main
conclusions are summarized in \S 5. We use $H_{\rm o} =
75$~km~s$^{-1}$~Mpc$^{-1}$ and $q_{\rm o} = 0$ throughout this paper.
\section{Observations and Data Reduction}
The new spectroscopic data were obtained with the Gold Cam
Spectrograph on the Kitt Peak 2.1-meter telescope. Table 1 lists the
dates of the observations, grating used, spectral coverage,
resolution, and seeing during the observations. The exposure times
range from 300 sec to 3,600 sec depending on the $R$ magnitude of each
object (cf.~Kim 1995). In all cases, a slit with a width of 2\arcsec\
was used and the slit was positioned in the N-S direction ($PA =
0^\circ$). Most of the observations were made under photometric
conditions. Those that were not are indicated with a dagger ($\dag$)
in Tables 3 -- 5. The moderate seeing prevented us from extracting
useful spatial information along the spectrograph slit. The results
derived from the KPNO data therefore only refer to the nuclear
regions. For consistency with VKSMS and KVS, the KPNO data were
reduced using the same techniques as those used in these earlier
papers. Once again, aperture-related effects were minimized using a
window for the extraction of the nuclear spectrum that varied
according to the redshift of each object so that it corresponds to a
constant linear scale (total diameter) of 4 kpc for most of the
galaxies, with the exceptions of three distant galaxies with $z$ $>$
0.2 (F00397--1312, F12032+1707, and F23499+2423)\footnote{Object names
that begin with `F' are sources identified in the {\em IRAS} Faint
Source Catalog, Version 2 (FSC: Moshir et al. 1992)} for which an
8-kpc window was used.
The spectra of the flux standard stars were used to remove the
absorption bands near 6870{\thinspace}\AA~and 7620{\thinspace}\AA~produced by
atmospheric O$_2$ (the B- and A-bands, respectively).
\section{Results}
The final calibrated spectra obtained at KPNO are plotted in Figure 1.
Only 10 of the 118 sources in the 1-Jy sample could not be observed
with the KPNO 2.1-meter telescope because of low declination and/or
lack of time. Refer to Figure 1 in KVS to view the calibrated spectra
of the 45 objects observed using the Mauna Kea facilities. Tables 2,
3 and 4 summarize the results of our analysis on the new KPNO data and
combine them with those of KVS and VKSMS. The methods used to derive
the various quantities listed in these tables are described in detail
in Kim et al. (1995) and KVS and are summarized below.
The line fluxes were measured using two methods. First, the standard
plotting package in IRAF (``splot'') was used to measure the flux of
isolated emission lines. To deal with blended lines (e.g., H$\alpha$ +
[N~II] $\lambda\lambda$6548, 6583 and [S~II] $\lambda\lambda$6716,
6731) and emission lines affected by stellar absorption features
(H$\beta$ and H$\alpha$), we used ``specfit'', an interactive IRAF
procedure kindly provided by Gerard A. Kriss. This routine can fit a
wide variety of emission-line, absorption-line, and continuum models
to the observed spectrum. The input parameters for the fit were
determined through ``splot'' in IRAF. We chose to fit the continuum with
a simple first-order polynomial, and the emission and absorption lines
with Gaussian profiles. The actual fitting was done via a chi-square
minimization using a simplex algorithm. The output parameters were
the flux levels and slopes of the underlying continuum emission, the
fluxes, centroids, and widths (FWHM) of the emission lines, and
equivalent widths, centroids, and widths (FWHM) of the absorption
lines.
Uncertainties in the H$\alpha$ fluxes are typically 25 --
50\%. The other emission line fluxes listed in Table 3 are normalized
to the observed H$\alpha$ flux. The uncertainty in these relative flux
ratios is typically 5 -- 10\% . Colons (:) and double colons (::)
indicate values with relative uncertainties of about 25\% and 50\%,
respectively. The adopted H$\beta$ flux, listed in column (5), is the
weighted average of the values obtained from the ``splot'' and ``specfit''
methods, with weights estimated from the relative uncertainties in the
two types of measurements.
The absorption line strengths, line widths, and continuum measurements
are presented in Table 4. Column (2) of Table 4 lists the equivalent
widths of H$\beta$ derived from the fitting method. These measurements
are rather uncertain ($\sim$50\%) because of the generally strong
H$\beta$ emission line affecting the profile of this feature. Columns
(3) and (4) list the equivalent widths of the Mg~Ib $\lambda$5174 and
Na~ID $\lambda$5896 absorption lines, respectively. Note that the
Na~ID feature is blended with the emission line of He I at
$\lambda$5876 in some galaxies. Column (5) of Table 4 lists the
values of the line widths (FWHM) of [O~III] $\lambda$5007. This line
was selected for line width measurements because it is strong in most
ULIGs and free of any nearby emission or absorption lines (in contrast
to H$\alpha$). The line widths listed in Table 4 have been corrected
for the finite instrument resolution of the data ($\sim${\thinspace}8.3\AA)
using the quadrature method. This method assumes that the intrinsic
and instrument profiles are Gaussian and gives corrected widths that
are too large for profiles which are more peaky than Gaussians (e.g.,
the emission-line profiles in optically-selected AGN -- Whittle 1985;
Veilleux 1991). For this reason, the [O~III] $\lambda$5007 line widths
should be treated with caution. Uncertainties on these measurements
are about $\pm$ 100 -- 200 km s$^{-1}$ or more. Corrected line widths
below 100 km s$^{-1}$ are flagged with a colon in Table 4 to emphasize
the large uncertainties on these measurements. Finally, the intensity
levels of the continuum near H$\beta$ and H$\alpha$ are listed in
columns (6) and (7) as C4861 and C6563, respectively.
\subsection{Spectral Classification}
The results from the spectral analysis of the {\em IRAS} 1-Jy sample
are quantitatively similar to those derived by KVS on the subsample of
45 objects. The emission-line ratio measurements used for this
analysis are listed in Table 5 and are plotted in Figure 2. As
described in the previous section, these line ratios were corrected
for the underlying Balmer absorption features. For consistency with
KVS and VKSMS, the same dereddening and spectral classification
techniques were used in the analysis. The line ratios were corrected
for reddening using the values of {\em E(B--V)} determined from the
H$\alpha$/H$\beta$ ratio and the Whitford reddening curve as
parameterized by Miller \& Mathews (1972; see next section). The
boundaries of Veilleux \& Osterbrock (1987) were used to classify each
object as H~II or AGN-like galaxies. These boundaries are based on
large data sets on optically selected galaxies and the predictions of
photoionization models by power-law spectra and by hot stars. A
distinction was made among AGN-like galaxies between the objects of
high ([O~III] $\lambda$5007/H$\beta \ge$ 3) and low ([O~III]
$\lambda$5007/H$\beta \leq$ 3) excitation. The first group represents
the ``classic'' Seyfert 2 galaxies while galaxies in the second group
were classified as LINERs (``low-ionization nuclear emitting
regions''). This LINER definition differs from the original definition
of Heckman (1980) which partly relies on the strength of the [O~II]
$\lambda$3727 line (this lines is outside the wavelength range of our
spectra). However, results from spectroscopic studies of optically
selected objects (e.g., Veilleux \& Osterbrock 1987) suggest that the
two ways of defining LINERs are often (close to 95{\thinspace}\% of the time)
equivalent. Finally, galaxies with Fe II multiplets at
$\lambda\lambda$5100--5560 and very broad ($\Delta V_{\rm FWHM}
\gtrsim 2,000$ km s$^{-1}$) H~I Balmer and He~I $\lambda$5876 emission
lines were classified as Seyfert 1s.
Following VKSMS and KVS, ULIGs with double nuclei were assigned the
more Seyfert-like spectral type of the two nuclei: galaxy pairs with
spectral types H~II -- LINER, H~II -- Seyfert 2, H~II -- Seyfert 1,
LINER -- Seyfert 2, LINER -- Seyfert 1, and Seyfert 1 -- Seyfert 2
were classified as LINER, Seyfert 2, Seyfert 1, Seyfert 2, Seyfert 1,
and Seyfert 1, respectively. These widely separated interacting
systems are more common at lower luminosities (VKSMS; Kim 1995; Surace
1998; Veilleux, Kim, \& Sanders 1999a). Consequently, this
classification method of the double-nucleus systems is conservative in
the sense that it overestimates the fraction of AGN among the lower
luminosity objects and thus cannot explain the trends with infrared
luminosity discussed below.
Out of the 108 ULIGs that make up our spectroscopic sample, 30{\thinspace}\%
(32/108) were found to have spectra characteristic of photoionization
by hot stars (H~II region-like). AGN-like emission lines were
observed in 70{\thinspace}\% (76/108) of the total sample including 9{\thinspace}\%
(10/108) Seyfert 1s, 21{\thinspace}\% (23/108) Seyfert 2s, and 40{\thinspace}\%
(43/108) LINER-like objects. These results were combined with the
measurements obtained by VKSMS for the LIGs from the Bright Galaxy
Survey (BGS; Soifer et al. 1987, 1989) to search for systematic
variations of the spectral types with infrared luminosity. For this
exercise, ULIGs were further divided into two luminosity bins:
$10^{12}\ L_\odot \leq L_{\rm ir} < 10^{12.3}\ L_\odot$ and $L_{\rm
ir} \geq 10^{12.3}\ L_\odot$. A summary of this analysis is given in
Table 6 and plotted in Figure 3.
This figure confirms the tendency noted by VKSMS and KVS for the more
luminous objects to be more Seyfert-like. As noted by KVS, only one
Seyfert 1 (NGC~7469) has $L_{\rm ir} < 10^{12}\ L_\odot$, whereas
48{\thinspace}\% (15/31) of the galaxies with $L_{\rm ir} \geq 10^{12.3}\
L_\odot$ present Seyfert characteristics. The percentage of Seyfert
1s relative to Seyfert 2s increases with infrared luminosity (from
$\sim$ 0\% among LIGs up to $\sim$ 50\% among ULIGs with $L_{\rm ir} >
10^{12.3}\ L_\odot$). This result is not compatible with the
predictions of the standard unification model of Seyfert galaxies
which purports that Seyfert 1s and 2s are basically the same kind of
objects observed from different angles. The far-infrared emission in
these objects is believed to be emitted more or less isotropically and
therefore does not depend on viewing angle (e.g., Mulchaey et
al. 1994). The trend with infrared luminosity in our data can be
explained if the covering factor of the obscuring material (opening
angle of torus?) in Seyfert ULIGs decreases with increasing infrared
luminosity.
LINER-like galaxies constitute a large fraction (30--40{\thinspace}\%) of the
total sample regardless of $L_{\rm ir}$. The LIGs classified as H~II
galaxies generally have low-excitation lines ([O~III]
$\lambda$5007/H$\beta$ $<$ 3; Fig. 2) and no detectable Wolf-Rayet
emission features (e.g., combinations of He~II $\lambda$4686, He~I
$\lambda$5876, C~III $\lambda$5696, N~IV $\lambda$5737, N~III
$\lambda\lambda$4634, 4640, and 4642), confirming previous results
(KVS; VKSMS; Leech et al. 1989; Armus et al. 1989; Allen et al. 1991;
Ashby, Houck, \& Hacking 1992; Ashby et al. 1995) and strongly arguing
{\em against} a very recent burst ($<<$ 10$^7$ yr) of star formation
in these objects (cf.~\S 3.5; Evans \& Dopita 1985; McCall, Rybski, \&
Shields 1985; Allen et al. 1991). Therefore, there is no evidence at
optical wavelengths for the type of extremely young (5 $\times$ 10$^6$
yrs) starbursts purported by Downes \& Solomon (1998) to exist in
Arp~220 and other ULIGs. The implications of this spectral
classification are discussed in \S 4.
\subsection{Reddening}
Following KVS, the reddening in our sample galaxies was estimated
using the emission-line H$\alpha$/H$\beta$ ratios corrected for the
underlying stellar absorption features. An intrinsic
H$\alpha$/H$\beta$ ratio of 2.85 was adopted for H II region-like
galaxies (Case B Balmer recombination decrement for T = 10$^4$ K and
N$_e$ = 10$^4$ cm$^{-3}$) and 3.10 for Seyferts and LINERs (e.g.,
Halpern \& Steiner 1983 and Gaskell \& Ferland 1984). For consistency
with the studies of VKSMS and KVS, the Whitford reddening curve as
parameterized by Miller \& Mathews (1972) was used. The reddenings
derived from this method are listed as color excesses, {\em E(B--V)},
in the third column of Table 5. Since all the objects in the 1-Jy
sample lie at $|b| > 30 ^\circ$, no correction was made for Galactic
reddening. Note that this method assumes that the obscuration is due
to a foreground screen of dust and underestimates by a factor
[exp($\tau$)--1]/$\tau$ the actual amount of extinction if the
line-emitting gas and dust are spatially mixed. Moreover, the adopted
procedure does not take into account possible differences between
Galactic and extragalactic reddening curves (e.g., Calzetti, Kinney,
\& Storchi-Bergmann 1994). However, these differences will be small at
optical wavelengths.
The distribution of {\em E(B--V)} for the ULIGs of our sample is shown
in Figure 4. The results are quantitatively similar to those of
KVS. Not surprisingly, the color excesses measured in the ULIGs of our
sample are considerably larger than those measured in
optically-selected Seyfert and starburst galaxies (Dahari \&
De~Robertis 1988) and in extragalactic H~II regions (Kennicutt, Keel,
\& Blaha 1989). These results confirm the importance of dust in
infrared-selected galaxies. The median {\em E(B--V)} are 0.80, 1.11
and 1.21 for the H~II galaxies (30 objects), LINERs (45) and Seyfert 2
galaxies (22), respectively. Kolmogorov-Smirnov (K-S) tests indicate
that the differences between the various spectral types are not
significant. These color excesses are similar to those obtained by
VKSMS in the lower-luminosity galaxies of the BGS [{\em E(B--V)} =
1.05, 1.24, and 1.07 for H~II galaxies, LINERs, and Seyfert 2
galaxies, respectively]. However, VKSMS found that the color excesses
of the LINERs were significantly larger than those of the H~II and
Seyfert 2 galaxies; this difference is not observed among the 1-Jy ULIGs.
The correlation reported by VKSMS between {\em E(B--V)} and the equivalent
width of the Na{\thinspace}ID absorption feature, {\em EW}(Na{\thinspace}ID), in
H~II galaxies is also present, but at a weaker level,
among our high-luminosity objects. The probability, P[null], that
this correlation is fortuitous is 0.04, 0.01, and 0.001 among the
H~II galaxies, LINERs, and Seyfert 2 galaxies of the 1-Jy sample. An
important fraction of the Na{\thinspace}ID feature is therefore of interstellar
origin. A similar correlation is observed between {\em E(B--V)} and
the observed continuum colors of H~II galaxies (P[null] = 3 $\times$
10$^{-5}$) and LINERs (P[null] = 1 $\times$ 10$^{-4}$), but not among
Seyfert 2 galaxies (P[null] = 0.16). As first reported by VKSMS,
these results suggest that the scatter in the continuum colors of the
Seyfert LIGs is predominantly intrinsic rather than caused by
variations in the amount of reddening from one object to the other
(cf. \S 3.5 for a more detailed discussion of continuum colors).
The {\em IRAS} flux density ratio ${\em f}_{25}/{\em f}_{60}$ is a
well-known indicator of Seyfert activity in infrared galaxies (e.g.,
deGrijp et al 1985; Miley, Neugebauer \& Soifer 1985). Galaxies with
``warm'' {\em IRAS} 25-to-60 $\mu$m colors are generally believed to
be less affected by dust obscuration than the ``cooler'' objects
(e.g., Sanders et al 1988a,b; Surace et al. 1998; VSK97, VSK99). One
should therefore expect to see a positive correlation between the {\em
IRAS} 25-to-60 $\mu$m color and the color excess in our sample;
however, none is detected (Fig. 5). This result suggests that the
color excess derived from the optical emission lines is not always a
good indicator of the total column of material towards the nucleus of
these objects. This problem is more severe in the cooler, dustier
objects because the optical method saturates beyond $\tau_{\rm V} \sim
5$ or, equivalently, {\em E(B--V) $\sim${\thinspace}2}. Extinction
measurements at longer wavelengths are more reliable (e.g., VSK97,
VSK99, Genzel et al. 1998).
\subsection{Line Widths}
Figure 6 shows the distribution of the [O~III] line widths for each
spectral type. The median line widths of the H II galaxies and LINERs
are comparable (200 km s$^{-1}$ and 320 km s$^{-1}$) whereas the
[O~III] line widths of Seyfert 1 and 2 galaxies are larger (1,120 and
610 km s$^{-1}$, respectively). K-S tests confirm these
differences. The probability that the line widths of H~II galaxies and
LINERs are drawn from the same population is 0.70, but it is less than
0.001 when comparing the line widths of H~II and Seyfert galaxies, or
when comparing the LINERs and Seyfert galaxies.
Figure 7 presents the distribution of [O~III] line widths of the
combined sample as a function of infrared luminosity and as a function
of the {\em IRAS} flux density ratio ${\em f}_{25}/{\em f}_{60}$. The
new data strengthen the conclusions of VKSMS and KVS: there is no
obvious correlation between line width and infrared luminosity or {\em
IRAS} color, but nearly all of the objects with line widths
larger than 600 km s$^{-1}$ have $L_{\rm ir} \gtrsim 10^{11}\ L_\odot$
(the only exception is the H~II galaxy NGC~3597 which has log~[$L_{\rm
ir}$/L$_\odot$] = 10.91). Objects with the most extreme profiles
($\Delta V_{\rm FWHM} \ga 1200$ km s$^{-1}$) all have $L_{\rm ir} \ga
10^{ 12}\ L_\odot$, optical Seyfert characteristics, and warm {\em
IRAS} colors (${\em f}_{25}/{\em f}_{60}$ $\ga$ 0.12). These results
suggest that the nuclear activity in the more powerful Seyfert ULIGs
contributes to the line broadening. Non-gravitational processes
associated with AGN-driven outflows have been proposed by VKSMS and
KVS to explain the broad and complex line profiles in the nuclear and
circumnuclear regions of some LIGs.
\subsection{Stellar Absorption Features}
In Figures 8 and 9, the distributions of the equivalent widths of
H$\beta$ and Mg~Ib are presented as a function of spectral type.
These distributions are similar to those obtained by KVS in their
subset of the 1-Jy sample. The conclusions derived by KVS on the
stellar absorption features are therefore generally confirmed by the
analysis of the entire 1-Jy sample. The median {\em EW}(H$\beta$) for
the 1-Jy ULIGs (including those with undetected H$\beta$) are
0.8{\thinspace}\AA, 0.6{\thinspace}\AA, 0.0{\thinspace}\AA\, and 0.0{\thinspace}\AA\ for the H~II
galaxies (30 objects), LINERs (43), Seyfert 2 galaxies (22), and
Seyfert 1 galaxies (9), respectively. K-S tests indicate that the
equivalent widths of Seyfert 1 and 2 galaxies are significantly
smaller than those of LINERs and H~II galaxies (P[null] $\la$ 0.01).
The equivalent widths of the 1-Jy ULIGs taken as a whole are also
significantly smaller than those measured by VKSMS in the BGS LIGs
(P[null] = 4 $\times$ 10$^{-9}$ when comparing both samples, and
P[null] = 0.002 and 0.01 for LINERs and H~II galaxies, respectively.
There are too few Seyfert galaxies among the BGS LIGs to carry out
this analysis). The intermediate-age (10$^8$--10$^9$ yr) population
of stars that is responsible for the H$\beta$ absorption feature is
therefore less prominent in our sample of ULIGs, possibly because the
H$\beta$ feature is diluted by a hot, featureless continuum from
young stars (particularly in H~II galaxies) or from an AGN (in the
Seyfert galaxies).
The strength of the Mg~Ib absorption feature provides additional
constraints on the underlying stellar population in ULIGs. As found
by KVS, the {\em EW}(Mg Ib) for our sample of ULIGs [median values of
0.0{\thinspace}\AA, 0.1{\thinspace}\AA, 0.2{\thinspace}\AA, and 0.0{\thinspace}\AA~for H II
galaxies (30 objects), LINERs (43), Seyfert 2 galaxies (22), and
Seyfert 1 galaxies (9), respectively] are considerably smaller that
those measured in non-active spiral galaxies [{\em EW}(Mg~Ib) =
3.5--5.0{\thinspace}\AA: Keel 1983; Stauffer 1982; Heckman, Balick, \& Crane
1980]. Moreover, {\em contrary} to KVS findings, the {\em EW}(Mg Ib)
for our sample of ULIGs are also significantly smaller than those
measured in the BGS objects of VKSMS (1.12{\thinspace}\AA, 1.49{\thinspace}\AA, and
1.17{\thinspace}\AA~for H~II galaxies, LINERs, and Seyfert 2 galaxies,
respectively). The very weak H$\beta$ and Mg~Ib features in 1-Jy
ULIGs can be explained if the age of the {\em dominant} stellar
population in these objects is less than a few $\times$ 10$^7$ yrs and
$\sim${\thinspace}10{\thinspace}\% of the galaxy mass is taking part in a burst of
star formation (Bica, Alloin, \& Schmidt 1990). Alternatively, an AGN
may account for $\sim${\thinspace}50 -- 75{\thinspace}\% of the optical continuum in
some ULIGs (e.g., Seyfert 1 and 2 galaxies) and weaken the spectral
signatures of the underlying old stellar population. The analysis of
the 1-Jy sample indicates that the young stellar population and/or AGN
is more preeminent in ULIGs than in the lower luminosity objects.
\subsection{Continuum Colors}
The observed continuum colors of the 1-Jy ULIGs are presented in
Figure 10 as a function of spectral type. Seyfert 1s have continuum
colors (median C6563/C4861 of 0.70) which are significant bluer than
all other classes of objects (median C6563/C4861 = 0.98, 1.01, and
1.15 for H~II galaxies, LINERs, and Seyfert 2 galaxies, respectively).
This result is consistent with the presence in Seyfert 1s of an AGN
emitting a strong energetic continuum. Figure 10 also shows that the
optical continuum of ULIGs is on average bluer than that of the BGS
LIGs (P[null] = 2 $\times$ 10$^{-7}$ when considering all
objects). This is especially evident among the LINERs (P[null] = 4
$\times$ 10$^{-5}$) and the H{\thinspace}II galaxies (P[null] = 2 $\times$
10$^{-3}$). These results confirm the presence of a strong blue
featureless continuum from a young stellar population and/or AGN in
the high-luminosity objects (\S 3.4).
The expected continuum color, C6563/C4861, of galaxies where
$\sim${\thinspace}10{\thinspace}\% of the galaxy mass has taken part in a burst of
star formation less than a few 10$^7$ yrs ago is $\sim${\thinspace}0.90
(e.g., Bica, Alloin, \& Schmidt 1990). Yet, the continuum colors
measured in ULIGs generally are redder than this value. This is a
sign that the continuous emission from these objects is probably
affected by dust obscuration. However, if the observed continuum
colors are dereddened using the amount of dust derived from the
emission-line spectrum (\S 3.2), a strong {\em negative} correlation
becomes apparent between the dereddened continuum colors and
dereddened H$\alpha$ luminosities (P[null] = 10$^{-14}$ for the whole
sample; in contrast, P[null] = 0.1 when the {\em observed} continuum
colors are considered). As mentioned in KVS, this result strongly
suggests that this dereddening method overcorrects for reddening in
the continuum (cf. also \S 3.7). The emission-line gas near hot stars
or the AGN are therefore dustier than the relatively cold stellar
population which is producing the optical continuum. This discrepancy
is also observed among optically-selected starbursts (e.g., Calzetti
et al. 1994; Gordon, Calzetti, \& Witt 1997). We have no easy way to
properly deredden the continuum colors of our sample galaxies. The
dereddened continuum colors will not be considered any further in the
present discussion.
\subsection{H$\alpha$ Luminosities and Equivalent Widths}
Here again, many of the conclusions based on the smaller sample of KVS
are confirmed in the complete 1-Jy sample. The distributions of
H$\alpha$ equivalent widths, observed H$\alpha$~luminosities, and
infrared-to-H$\alpha$~luminosity ratios are plotted in Figures 11 --
13 for each spectral type. The large H$\alpha$ luminosities and
equivalent widths and small infrared-to-H$\alpha$~luminosity ratios
among the (4) Seyfert 1 ULIGs [their median {\em EW}(H$\alpha$) in
\AA, log ({\em L}$_{{\rm H}\alpha}$/{\em L}$_\odot$), and log({\em
L}$_{\rm ir}$/{\em L}$_{{\rm H}\alpha}$) are 234, 8.6, and 3.8,
respectively] again suggest the presence of an additional source of
ionization (AGN) which is not visible or present in starburst-powered
galaxies. LINER ULIGs appear to be slightly deficient in H$\alpha$
relative to H~II and Seyfert 2 ULIGs: the median \{{\em EW}(H$\alpha$)
in \AA, log ({\em L}$_{{\rm H}\alpha}$/{\em L}$_\odot$)\} are \{88,
7.8\}, \{50, 7.4\}, and \{84, 7.9\} for H~II galaxies (30 objects),
LINERs (43), and Seyfert 2 galaxies (22), respectively. The observed
infrared-to-H$\alpha$~luminosity ratios in LINER ULIGs are also
somewhat larger than those of Seyfert 2 and H~II ULIGs [median
log({\em L}$_{\rm ir}$/{\em L}$_{{\rm H}\alpha}$) = 4.2, 4.7, and 4.2
for the H~II (29), LINER (38), and Seyfert 2 galaxies (21),
respectively], confirming the H$\alpha$ deficiency in LINERs. This
deficiency persists even after correcting the emission-line fluxes for
reddening using the color excesses derived from the optical Balmer
decrements. The H$\alpha$ equivalent widths of HII and LINER ULIGs are
slightly larger than those found in LIGs of the same type (55 \AA~in
HII galaxies and 29 \AA~in LINERs). This is consistent with the
presence of a more preeminent starburst in these ULIGs (\S 3.4 and \S
3.5).
\subsection{Infrared Spectral Properties}
The infrared spectral properties of the various classes of LIGs were
investigated. We used the definitions of Dahari \& De Robertis (1988)
for the ``{\em IRAS} color indices'' and ``infrared color excess''
[$\alpha_1$ = -- log(${\em f}_{60}/{\em f}_{25}$)/log(60/25),
$\alpha_2$ = -- log(${\em f}_{100}/{\em f}_{60}$)/log(100/60), $IRCE$
= (($\alpha_1$ + 2.48)$^2$ + ($\alpha_2$ + 1.94)$^2$)$^{0.5}$]. The
infrared color excess is a measure of the deviation from colors of
non-active spiral galaxies ($\alpha_1$ = --2.48 and $\alpha_2$ =
--1.94; Sekiguchi 1987).
As described in Kim \& Sanders (1998), the {\em IRAS} 60-to-100 flux
ratios were used to pre-select the targets for the 1-Jy sample.
Consequently, the 1-Jy ULIGs have median $\alpha_2$ and $IRCE$ (--0.41
and 1.63, respectively) which are significantly larger than those of
the BGS LIGs (--0.77 and 1.19, respectively), while the median
$\alpha_1$ for ULIGs and BGS LIGs (--2.31 and --2.46, respectively)
are similar. The differences in {\em IRAS} colors between ULIGs and
normal galaxies are especially striking among Seyfert 1 galaxies
(median $\alpha_1$, $\alpha_2$, and $IRCE$ of --1.70, +0.05, and 2.16,
respectively) and Seyfert 2 galaxies (--2.27, --0.38, and 1.61). The
infrared spectral properties of the LINER ULIGs (--2.49, --0.53, and
1.53) are indistinguishable from those of H{\thinspace}II ULIGs (--2.49,
--0.45, 1.59) and are closer to the parameters of normal spiral
galaxies. Therefore, there is no convincing optical or infrared
evidence for an AGN in LINER ULIGs. VKSMS and KVS argue that hot stars
and/or shocks are the most likely sources of ionization in these
objects (cf. \S 4).
Optically classified H~II galaxies in our sample show correlations
between infrared colors and H$\alpha$ equivalent widths. H~II
galaxies with large H$\alpha$ equivalent widths tend to have small
${\em f}_{12}/{\em f}_{25}$ and large ${\em f}_{25}/{\em f}_{60}$ and
${\em f}_{60}/{\em f}_{100}$ (P[null] = 0.001, 0.002 and 4 $\times$
10$^{-9}$, respectively; Fig. 14). Similar correlations were reported
by Mazzarella, Bothun, \& Boroson (1991) among optically-selected
starburst galaxies. These results suggest that the most active or
youngest star-forming regions produce far-infrared emission with the
warmest dust temperatures (Mazzarella et al. 1991). The fact that
these correlations are considerably weaker among the infrared-selected
Seyfert 2 galaxies (P[null] = 0.11, 0.005 and 0.002, respectively) is
another indication that star formation may not be the only source of
energy in these objects.
As shown in Figure 15, the well-known correlation between the
optical-to-infrared luminosity ratio and the infrared luminosity or
{\em IRAS} flux ratio ${\em f}_{60}/{\em f}_{100}$ among luminous
infrared galaxies (e.g., VKSMS) disappears when the 1-Jy ULIGs are
included in the analysis. In this figure, the optical continuum
luminosity, L(4861), is defined as P(4861) $\times$ 4861 where P(4861)
is the monochromatic power of the continuum at 4861 \AA. No reddening
correction was applied to the continuum luminosity. Note that L(4861)
refers to the nuclear region while L$_{\rm ir}$ and ${\em f}_{60}/{\em
f}_{100}$ are integrated quantities. Integrated continuum luminosities
are not yet available for most of these galaxies. However, recent imaging
studies at infrared and millimeter wavelengths (e.g., Evans 1998;
Egami 1998; Scoville et al. 1998; Downes \& Solomon 1998; Sakamoto et
al. 1999) now strongly suggest that the active regions in ULIGs are
generally centered on the inner kpc of the galactic nuclei, with the
rest of the galaxy contributing very little to the bolometric
luminosity (the situation is different in objects of lower
luminosities). The correlations among lower luminosity objects have
been interpreted in the past as an indication that high global dust
temperature is associated with large infrared dust emission and small
optical continuum extinction. The tendency reported by VKSMS for the
color excess of H~II and LINER LIGs from the BGS sample to decrease at
large log~L(4861)/L$_{ir}$ is consistent with this hypothesis, but
this tendency disappears when the 1-Jy ULIGs are included in the
analysis. It is not clear at present why that is the case. An attempt
to correct L(4861)/L$_{ir}$ for reddening effects using the color
excesses determined from the H$\alpha$/H$\beta$ flux ratios fails to
bring new insight into this issue. Instead, a very strong {\em
positive} correlation between {\em E(B--V)} and L(4861)$_0$/L$_{ir}$
(P[null] = 10$^{-43}$ when considering all the objects) is created,
confirming that the reddening correction derived from the emission
lines severely overestimates the actual extinction of the optical
continuum (\S 3.5).
\section{Discussion: Nature of the Energy Source}
Most of the trends observed by KVS in their subset of 45 galaxies from
the 1-Jy sample are confirmed in the present analysis of the entire
sample. Most relevant to the question, raised in the Introduction, of
the nature of the energy source in ULIGs is the confirmation that
Seyfert 1 or 2 spectral characteristics are present in
$\sim${\thinspace}30{\thinspace}\% of all ULIGs in the 1-Jy sample, and in
$\sim${\thinspace}50{\thinspace}\% of the high-luminosity objects (log[L$_{\rm
ir}$/L$_\odot$] $>$ 12.3). The
weaker H$\beta$ and Mg~Ib stellar absorption features, bluer continuum
colors, larger H$\alpha$ luminosities and equivalent widths, smaller
infrared-to-H$\alpha$ luminosity ratios, and warmer {\em IRAS} ${\em
f}_{25}/{\em f}_{60}$ colors in these Seyferts relative to LINER or
H~II ULIGs are all consistent with the presence of a genuine AGN in
these objects. The absence of correlations between line widths and
line ratios (not shown here) argues against the possibility that
high-velocity shocks (e.g., Binette, Dopita, \& Tuohy 1985; Innes
1992; Dopita \& Sutherland 1995) are the dominant source of ionization
in these Seyfert 2 ULIGs. The very restrictive range of shock
conditions (e.g., V$_{\rm shock}$ = 300 -- 500 km s$^{-1}$) needed to
produce the high-excitation spectrum of Seyfert 2 galaxies from
high-velocity shocks also is inconsistent with the broad range in line
widths observed in these objects (cf. also VKSMS).
Recent near-infrared spectroscopy of 22 optically classified Seyfert 2
galaxies from the 1-Jy sample by VSK97 and VSK99 reveals the presence
of broad recombination lines with FWHM $\ga$ 2,000 km s$^{-1}$ or
strong high-ionization [Si~VI] emission in $\sim${\thinspace}70{\thinspace}\% of
these objects (especially those with warm {\em IRAS} 25-to-60 $\mu$m
colors). The absence of broad features in the profiles of the H$_2$
lines and forbidden [Fe~II]~$\lambda$1.257 suggests that the broad
emission is produced by high-density gas near the nucleus like that in
the BLR of AGN rather than by outflowing lower density material in the
circumnuclear region. The bright [Si~VI] feature detected in these
objects is also believed to be a good indicator of AGN activity since
this feature has never been detected in starburst galaxies despite
extensive searches (e.g., Marconi et al. 1994). The optical and
near-infrared results taken together, therefore suggest that the total
fraction of objects in the 1-Jy sample with signs of bonafide AGN is
at least $\sim${\thinspace}20 -- 25{\thinspace}\%, but reaches $\sim${\thinspace}35 --
50{\thinspace}\% for those objects with log[L$_{\rm ir}$/L$_\odot$] $>$ 12.3.
Nevertheless, the presence of an AGN in ULIGs does not necessarily
imply that AGN activity is the dominant energy source in these
objects. A more detailed look at the AGN in these ULIGs is needed to
answer this question.
A strong linear correlation has long been known to exist between the
continuum (or, equivalently, bolometric) luminosities of broad-line
AGN and their emission-line luminosities (e.g., Yee 1980; Shuder 1981;
Osterbrock 1989). This correlation has often been used to argue that
the broad-line regions in AGN are photoionized by the nuclear
continuum. If this is the case, the broad-line--to--bolometric
luminosity ratio is a measure of the covering factor of the BLR (e.g.,
Osterbrock 1989). This correlation was also used by VSK97 to estimate
the importance of the AGN in powering ULIGs. In ULIGs powered uniquely
by an AGN, we expect the broad-line luminosities to fall along the
correlation for AGN. Any contribution from a starburst will increase
the bolometric luminosity of the ULIG without a corresponding increase
in the broad-line luminosity. Starburst-dominated ULIGs are therefore
expected to fall below the ``pure-AGN'' correlation traced by the
optical quasars in a diagram of $L_{\rm H\beta}$(BLR) plotted as a
function of $L_{\rm bol}$\footnote{Note that the origin of the
far-infrared/submm emission in optical quasars is still a matter of
some debate. Some of it may be produced by a dusty circumnuclear
starburst (e.g., Rowan-Robinson 1995). In the following discussion,
we identify the ``pure-AGN'' $L_{\rm H\beta}$(BLR) -- $L_{\rm bol}$
relation as the one traced by optical quasars. We therefore implicitly
assume that all of the far-infrared/submm emission is powered by the
central AGN (Sanders et al. 1989). The far-infrared/submm emission
generally contributes about 20\% of the total bolometric luminosities
of optical quasars. }.
In Figure 16, the dereddened {\em broad-line} H$\beta$ luminosities of
the Seyfert 1 ULIGs in our sample are plotted as a function of their
bolometric luminosities. Data on optically selected quasars and on
ULIGs with obscured BLRs (i.e. ``buried quasars'' detected at
near-infrared wavelengths by VSK97) are shown for comparison. $L_{\rm
bol}$ for the QSOs was determined using the bolometric correction
factor 11.8 (i.e. $L_{\rm bol} = 11.8 \nu_{\rm B} L_{\rm \nu}(B)$:
Elvis et al. 1994; Sanders \& Mirabel 1996), except for those few
sources that were detected by {\em IRAS}, in which case $L_{\rm bol}$
was taken from Sanders et al. (1989). For the 15 ULIGs $L_{\rm bol}$
was taken to be $1.15 \times L_{\rm ir}$ (Kim \& Sanders 1998). The
H$\beta$ data for the optically selected QSOs are from Yee (1980)
corrected for $H_{\rm o} = 75$~km~s$^{-1}$~Mpc$^{-1}$ and $q_{\rm o} =
0$. The H$\beta$ luminosities tabulated in Yee (1980) include a
contribution from narrow H$\beta$ and are not corrected for
reddening. The flux from narrow H$\beta$ is often difficult to measure
in these broad-line objects, but the high-quality rest-frame optical
spectra that exist on a few of these quasars show that $L_{\rm
H\beta}$(NLR)/$L_{\rm H\beta}$(BLR) is generally much less than
$\sim${\thinspace}30{\thinspace}\% (e.g., Boroson \& Green 1992; Corbin \& Boroson
1996). This correction would lower the quasar data points in Figure
16. The reddening correction to the broad-line luminosities would, on
the other hand, raise these data points. The reddening towards the
BLRs of these {\em optically-selected} objects is likely to be
modest. In the end, no attempt was made to correct for narrow-line
contamination or reddening effects. The best log-linear fit through
the quasar data points, log $L_{H\beta}$(BLR) = 1.05 log $L_{\rm bol}$
-- 3.61, is shown as a solid line in Figure 16. The broad-line
H$\beta$ luminosities of the ULIGs with optical BLRs were dereddened
using the broad-line H$\alpha$/H$\beta$ ratio and assuming an
intrinsic ratio of 3.1 (see MacAlpine 1981 for uncertainties
associated with using this dereddening method for BLRs). The
broad-line H$\beta$ luminosities of the ULIGs with obscured BLRs were
calculated from the measured broad Pa$\alpha$ fluxes assuming Case B
recombination (except for Mrk~463E = F13536+1836 where the broad
Pa$\beta$ flux was used). The reddening correction for these objects
was carried out using the color excesses derived from the infrared
broad-line ratios (cf.~VSK97).
Considering the various sources of uncertainties in the derivations of
the dereddened broad-line luminosities plotted in Figure 16 (overall
uncertainties of order 30{\thinspace}\%) and the intrinsic scatter of the
quasar data points around the ``pure'' quasar relation (of order 0.2
dex in log[L$_{\rm bol}$]), caution should be used when using this
figure to make quantitative statements about the dominant energy
source in these objects. The positions of 6 or 7 of the 10 optical
BLRs (Mrk~231 = F12540+5708, F07599+6508, F13218+0552, F13342+3932,
F15462-0450, F21219-1757 and perhaps also Mrk~1014 = F01572+0009) and
of all 5 obscured BLRs (Mrk~463E, PKS~1345+12, F20460+1925,
F23060+0505, and F23499+2423) fall close (or above) to that of optical
quasars in Figure 16. This result suggests that a substantial fraction
of the luminosity in these objects is powered by a quasar rather than
a starburst. The only broad-line ULIGs which clearly fall below the
``pure-AGN'' correlation in Figure 16 are NGC~7469, F11119+3257, and
F11598-0112. NGC~7469 has long been known to host a Seyfert 1 nucleus
and a powerful circumnuclear starburst (De~Robertis \& Pogge 1986;
Wilson et al. 1986). Recent ISO results (Genzel et al. 1998) indicate
that the starburst is indeed the dominant infrared energy source in
this object. The only other broad-line 1-Jy ULIGs in common with the
ISO sample are Mrk~231, Mrk~463E and F23060+0505. In all three cases,
the results from the ISO survey support our conclusion that the
dominant source of infrared radiation is a quasar rather than a
starburst (Genzel et al. 1998).
It is also instructive to study the mid-infrared results on the
optically classified LINERs of our sample. Five of these objects were
studied by Genzel et al. (1998; UGC 5101, F12112+0305, F14348-1447,
F15250+3609, and Arp 220 = F15327+2340). All of them are considered powered
predominantly by a starburst (Genzel et al. 1998). This result is also
confirmed when a larger sample of ULIGs is considered: if at all
present, the AGN in ULIGs with optical LINER characteristics does not
appear energetically important in the great majority of these objects
(Lutz, Veilleux, \& Genzel 1999).
Another method commonly used to estimate the importance of the AGN in
ULIGs is based on the strength of the hard (2 -- 10 keV) X-ray
emission from these objects. Nakagawa et (1998) have recently
summarized the ASCA results on ULIGs. Among our subset of broad-line
ULIGs, three objects were observed by ASCA: Mrk 231, F20460+1925, and
F07599+6508. The first two objects show clear signs of AGN activity in
the hard X-rays while the last object was not detected by ASCA.
Assuming a X-ray luminosity fraction $L$(2 -- 10 keV)/$L_{\rm FIR}$
$\sim$ 0.1 for typical optical quasars, Nakagawa et al. concluded that
the AGN contribution to the total luminosity is significant in
F20460+1925, in agreement with our optical data, but is small in
Mrk~231 and F07599+6508, in contrast to the optical results (and the
ISO results on Mrk~231). A multiwavelength analysis of the
Palomar-Green Bright Quasars (Sanders et al. 1989) shows, however,
that the X-ray luminosity fraction is typically only about 1{\thinspace}\%
(and can be as small as 0.1{\thinspace}\% in some objects). The AGN contribution
in Mrk~231 would be significant if Mrk 231 happens to lie in the faint
X-ray tail of the optically selected quasars. The same conclusion
applies to F07599+6508 if the actual hard X-ray flux from this object
is close to the upper limit determined by Nakagawa et
al. Consequently, the current X-ray data on the broad-line ULIGs of
our sample are not inconsistent with the presence of energetically
important AGN in many of these broad-line objects.
High-resolution radio observations of several ULIGs have provided
additional information on the nature of the energy source of ULIGs. A
good correlation seems to exist between the core radio power and
bolometric luminosity of optically-selected quasars (Lonsdale, Smith,
Lonsdale 1995). This correlation was used by Lonsdale et al. to argue
that buried quasars are capable of powering the infrared luminosities
in most ULIGs. However, recent VLBI observations by the same group
indicate that a starburst may be able to explain the observed radio
characteristics of at least a few of these objects (e.g., Arp 220 West;
Smith et al. 1998b, Lonsdale et al. 1998). The current VLBI sample
includes only two broad-line galaxies from our sample, Mrk~231 and
NGC~7469. The VLBI data on these two objects are difficult to explain
with a starburst alone, and therefore strongly suggest the presence of
an AGN in both objects (Smith et al 1998a). High-resolution radio
maps of more broad-line ULIGs from our sample will be needed to
properly test our conclusions of the existence of energetically
important AGN in these objects.
\section{Summary}
The results from an optical spectroscopic study of the nuclear regions
of 108 of the 118 ULIGs from the {\em IRAS} 1-Jy sample of Kim \&
Sanders (1998) are reported. These spectra are combined with
previously published optical data on 200 LIGs from the BGS sample
(VKSMS) to examine the spectral properties of LIGs over the range
$L_{\rm ir} \approx 10^{10.5}-10^{13.0}\ L_\odot$. The results from
this analysis confirm many of the trends already seen by KVS in a
subset of 45 of these objects. The fraction of LIGs with Seyfert
characteristics increases with increasing $L_{\rm ir}$. For $L_{\rm
ir} > 10^{12.3}\ L_\odot$, about 48{\thinspace}\% of the ULIGs (15/31
objects) are classified as Seyfert galaxies. The fraction of Seyfert
1s relative to Seyfert 2s increases with infrared luminosity. This
result is not compatible with the predictions of the standard
unification model of Seyfert galaxies unless the covering factor of
the obscuring material (opening angle of torus?) in Seyfert ULIGs
decreases with increasing infrared luminosity.
Many of the optical and infrared spectroscopic properties of the
Seyfert galaxies point to the existence of an AGN which is not present
or visible in LINER or H~II ULIGs. About 30{\thinspace}\% (10/33) of the
Seyfert galaxies in the 1-Jy sample are of type 1, presenting broad
Balmer lines and strong Fe~II emission similar to what is observed in
optically selected quasars. Near-infrared spectroscopy (VSK97, VSK99)
often reveals broad (FWHM $>$ 2,000 km s$^{-1}$) Paschen emission
lines or strong high-ionization [Si~VI] emission in the remaining Seyferts
(especially those with warm {\em IRAS} 25-to-60 $\mu$m
colors). Seyfert ULIGs (especially those of type 1) have weaker
H$\beta$ and Mg~Ib stellar absorption features, bluer continuum
colors, larger H$\alpha$ luminosities and equivalent widths, smaller
infrared-to-H$\alpha$ luminosity ratios, and warmer {\em IRAS} ${\em
f}_{25}/{\em f}_{60}$ colors than LINER or H~II ULIGs. The [O~III]
$\lambda$5007 line widths in the nuclei of the Seyfert galaxies are
also significantly broader on average than those measured in the H~II
and LINER ULIGs.
Comparisons between the emission-line luminosities of the optical or
near-infrared broad-line regions in the Seyfert galaxies of the 1-Jy
sample and the broad-line luminosities of optical quasars suggest that
the AGN in these ULIGs generally is energetically important. Only
three objects (NGC~7469, F11119+3257, and F11598-0112) are clearly
powered predominantly by a starburst. The general agreement between
this optical analysis and the current ISO, X-ray, and VLBI results is
encouraging but only based on a limited number of broad-line
objects. Results from on-going multiwavelength studies of
high-luminosity ULIGs (especially those with $L_{\rm ir} \ga
10^{12.3}\ L_\odot$) will provide more stringent constraints to test
these conclusions.
The weak H$\beta$ and Mg~Ib features in ULIGs optically classified as
H~II galaxies or LINERs suggests the presence of a young ($\la$ few
$\times$ 10$^7$ yrs) stellar population comprising
$\sim${\thinspace}10{\thinspace}\% of the total galaxy mass in these objects. While
this starburst is the likely source of ionization among H~II galaxies
and some LINERs, long-slit information from VKSMS and KVS indicates
that the LINER-like emission may also be produced through shocks
caused by the interaction of starburst-driven outflows with the
ambient material. Comparisons of the optical results with recent
mid-infrared data obtained with ISO suggest that the main source of
energy in these infrared-selected H~II galaxies and LINERs is a
starburst rather than an AGN. The observed continuum colors and
strengths of the stellar absorption features and H$\alpha$ emission
line indicate that the starburst becomes increasingly important with
increasing infrared luminosity in both H~II galaxies and LINERs.
The large H$\alpha$/H$\beta$ ratio and red continuum colors of the
1-Jy ULIGs confirm the importance of dust in all these objects. As was
found among optically selected starbursts, the relatively cool stars
which are producing the continuous optical emission are less reddened
than the emission-line gas in ULIGs. No significant differences are
found between the mean emission-line color excess of ULIGs and that of
{\em IRAS} galaxies of lower infrared luminosity. The color excess in
the nuclei of ULIGs does not depend strongly on the optical spectral
type or {\em IRAS} 25-to-60 $\mu$m color. The most likely explanation
for this surprising result is that the optical method used to derive
the color excess in infrared galaxies underestimates the amount of
dust in the dustier, cooler objects.
\clearpage
\acknowledgments
S. V. and D. B. S. thank the organizers of the 1998 Ringberg meeting
where some of the issues in this paper were discussed. We are also
grateful to the editor, Dr. Greg Bothun, and the anonymous referee for
suggestions which significantly improved this paper. This research was
supported in part by JPL contract no. 961566 to the University of
Hawaii (D. B. S.). S. V. is grateful for partial support of this
research by NASA/LTSA grant NAG 56547 and Hubble fellowship
HF-1039.01-92A awarded by the Space Telescope Science Institute which
is operated by the AURA, Inc. for NASA under contract
No. NAS5--26555. This research has made use of the NASA/IPAC
Extragalactic Database (NED) which is operated by the Jet Propulsion
Laboratory, California Institute of Technology, under contract with
NASA.
\clearpage
|
\section{Introduction}
Studies of effective continuum field theories have resulted in detailed
predictions for the low-energy physics of quantum spin systems in two
dimensions.\cite{csh,neuberger,fisher,hasenfratz} In order to make use of
these predictions for a given model Hamiltonian, the ground state
parameters appearing in the Lagrangian formulation have to be determined.
Spin-wave theory \cite{spinwave1} can in some cases give good estimates,
\cite{spinwave2} but in general some numerical method has to be
employed in order to obtain accurate results. Since the theories can
also predict the finite-size scaling behavior of various physical
quantities, numerical results for a series of lattice sizes can be used
to extract the ground state parameters. Such calculations are also
important for testing theoretical predictions.
With todays computers, Lanczos and related exact diagonalization methods
can be used for square lattices with up to $6 \times 6$ spins.
\cite{schulz,einarsson,hamer}
This relatively small maximum size, and the small number of different lattices
available, can make the finite-size scaling procedures problematic if
sub-leading corrections are significant. It is therefore important to
consider also alternative methods that can reach larger lattice sizes.
Here we discuss quantum Monte Carlo (QMC) results for the $S=1/2$ XY model,
defined by the Hamiltonian
\begin{equation}
H = -J\sum\limits_{\langle i,j\rangle} [S^x_iS^x_j + S^y_iS^y_j],
\label{model}
\end{equation}
where $S^x_i$ and $S^y_i$ are the $x$- and $y$-components of a spin-1/2
operator at site $i$, and $\langle i,j\rangle$ denotes a pair of
nearest-neighbor sites on a square lattice. Numerical studies of this
model have a long history.\cite{pearson,oitmaa,loh,okabe,zhang,harada,hamer}
Exact finite-lattice calculations gave the first indications that the
O(2) symmetry is spontaneously broken at $T=0$ in the thermodynamic limit.
\cite{oitmaa} This was later proved rigorously.\cite{orderproof} QMC
simulations have resulted in a quite precise value of the ordered moment
\cite{zhang} and have also shown \cite{loh} that there is a
Kosterliz-Thouless (KT) transition at a temperature
$T_{\rm KT} \approx 0.343J$.
\cite{harada} For $T \le T_{\rm KT}$ the system acquires quasi-long-range
order, i.e., power law decay of the spin-spin correlation function and
a non-zero spin stiffness constant. In accordance with the Mermin-Wagner
theorem,\cite{mermin} true long-range order develops only at $T=0$.
Our motivation for carrying out calculations for the XY model to even
higher precision is to test predictions of effective Lagrangian theories.
In particular, the chiral perturbation calculations by Hasenfratz and
Niedermayer have resulted in detailed predictions for the finite-size and
finite-temperature corrections of various quantities,\cite{hasenfratz}
in some cases beyond leading order. The finite-size and finite-temperature
scaling behavior of the O(3) symmetric Heisenberg model has been the
topic of numerous studies,\cite{schulz,runge,wiese,troyer,sandvik}
and the agreement with the predictions has been confirmed to high
precision. The predictions for the O(2) symmetric XY model have not yet
been tested exhaustively, however. A recent finite-size scaling
study of exact energies for systems with up to $6 \times 6$ spins has quite
convincingly demonstrated agreement with the leading finite-size
behavior.\cite{hamer} Here we consider also several other physical
observables. This allows us to carry out a number of independent tests
of the consistency of the scaling predictions. With access to larger
system sizes, we can also improve considerably on the accuracy of the
extrapolated ground state parameters and finite-size corrections. Our
data are sufficiently accurate for addressing also sub-leading corrections.
In addition to calculations in the ground state, we have also studied
systems at finite temperature, on lattices sufficiently large to enable
extraction of the leading finite-temperature corrections in the
thermodynamic limit.
We have used a numerically exact finite-temperature QMC method based on
``stochastic series expansion'',\cite{sse} i.e., importance sampling
of diagonal matrix elements of the power series expansion of
${\rm exp}(-\beta H)$, where $\beta=1/T$ is the inverse temperature.
The ground state can be obtained by choosing a sufficiently large
$\beta$ value. This method has previously been used to study the
finite-size scaling behavior and the ground state parameters of the
square-lattice Heisenberg model.\cite{sandvik} Recently, a new way of
sampling the terms of the expansion was proposed (the ``operator-loop''
algorithm),\cite{loop} which enables calculations to even higher
precision. We have used this improved method to study the ground state
of the XY model on lattices with $N=L\times L$ spins with $L$ up to
$16$. We have carried out calculations at finite temperature for
systems with $L$ up to $64$.
In Sec.~II we discuss the physical quantities that we have calculated.
The numerical ground state data and the results for finite-size
corrections and extrapolations to infinite size are presented in
Sec.~III. In Sec.~IV we discuss calculations at $T > 0$. Sec.~V
concludes with a brief summary.
\begin{table}
\begin{tabular}{lddd}
$L$ & ~~~~$-E_0$ & ~~~~$\rho$ & ~~~~~~~~$M_x^2$ \\ \hline
4 & 0.562485(4) & 0.2769(1) & 0.13282(2) \\
6 & 0.552696(4) & 0.2718(1) & 0.11885(4) \\
8 & 0.550436(4) & 0.2705(2) & 0.1126(2) \\
10 & 0.549643(4) & 0.2700(3) & 0.1087(2) \\
12 & 0.549296(4) & 0.2698(4) & 0.1065(3) \\
14 & 0.549118(4) & 0.2695(3) & \\
16 & 0.549020(4) & 0.2699(4) & \\
\end{tabular}
\vskip2mm
\caption{QMC results for the ground state energy, the spin stiffness,
and the squared magnetization per spin. The numbers within parentheses
indicate the statistical errors of the least significant digit of the
results.}
\label{tab1}
\end{table}
\begin{figure}
\centering
\epsfxsize=8cm
\leavevmode
\epsffile{e0.epsi}
\vskip1mm
\caption{Ground state energy vs the inverse cubed of the system
size for $L=8,10,12,14$, and $16$ (points with error bars). The
curve is a fit to Eq.~(\protect{\ref{e0scale}}), including also $L=4$
and $L=6$.}
\label{fige0}
\end{figure}
\section{Calculated quantities}
The QMC algorithm \cite{loop} can be implemented in two different ways;
with the spin quantization either along the $z$ or $x$ axis. The internal
energy, $E=\langle H\rangle /N$,
can be easily calculated in both cases. In the $z$-representation,
the spin stiffness constant $\rho$ can also be evaluated. It is defined
as the second derivative of the internal energy with respect to a twist
$\phi$ under which, for all nearest-neighbor spin pairs in either the
$x$ or $y$ lattice direction, the spin-spin interaction is modified
according to
\begin{eqnarray}
S^x_iS^x_j && + S^y_iS^y_j \to
(S^x_iS^x_j + S^y_iS^y_j)\cos{(\phi)} \nonumber \\
&& - S^x_iS^y_j\sin{(\phi)} + S^y_iS^x_j\sin{(\phi)}.
\end{eqnarray}
In analogy with the superfluid density of a boson system,\cite{pollock}
one can show that\cite{sandvik}
\begin{equation}
\rho = {\partial ^2 E(\phi) \over \partial^2 \phi } =
{\langle W_x^2 \rangle + \langle W_y^2 \rangle \over 2N\beta},
\end{equation}
where $E(\phi)$ is the internal energy per spin and $W_x$ and $W_y$ are
the ``winding numbers'', i.e., the net spin currents across the
periodic boundaries in the $x$ and $y$ directions that characterize
configurations in simulations carried out in the $z$-representation.
We also calculate the spin susceptibility, given by
\begin{equation}
\chi = {\beta\over N^2}
\left\langle \left ( \sum\limits_{i=1}^N S^z_i\right )^2 \right\rangle .
\label{suscept}
\end{equation}
In the $z$-representation,
it is not convenient to calculate spin-spin correlations involving
the $x$ or $y$ spin components. We therefore also use an
algorithm implemented in the $x$ representation, and there calculate
the squared magnetization $M_x^2$;
\begin{equation}
M_x^2 = {1\over N^2} \left\langle \left ( \sum\limits_{i=1}^N S^x_i
\right ) ^2 \right\rangle .
\end{equation}
Since the O(2) symmetry is not broken on a finite lattice, the magnetization
$m$ per spin in the thermodynamic limit is given by the infinite-size
extrapolated value of
\begin{equation}
m=\sqrt{2M_x^2}
\label{mM}.
\end{equation}
We use periodic boundary condition and include each spin pair only once
in Eq.~(\ref{model}). For technical details on the simulation
algorithm we refer to previously published work.\cite{loop,sandvik}
\section{Ground state results}
In order to obtain the ground state, we have carried out simulations
at inverse temperatures as high as $\beta=1024$ for lattices with $L=4-16$.
In the $z$-representation simulations, we have calculated the ground
state energy to within relative statistical errors lower than $10^{-5}$.
The relative accuracy of the stiffness is on the order of $10^{-3}$.
Simulations in the $x$-representation result in statistical errors for
the energy roughly twice as large as in the $z$-representation (for
simulations of comparable length). The squared magnetization is the
least accurate quantity, with statistical errors of roughly $0.3\%$
for the largest system size considered in this case ($L=12$).
In Table \ref{tab1} we list the results used in the finite-size
analysis presented below. The results for the
energy for $L=4$ and $6$ are in perfect agreement with previous
\cite{oitmaa,hamer} exact diagonalizations. The magnetization for
$L=4$ also agrees with the exact diagonalization result. However,
for $L=6$ there is an $\approx 5\sigma$ deviation from the result
presented in Ref.~\onlinecite{hamer}. We do not know the reason
for this disagreement. The QMC method has previously been shown to
give perfect agreement with isotropic Heisenberg results for both
$L=4$ and $L=6$,\cite{sandvik} and therefore a failure for (only) the
XY-model magnetization for $L=6$ would be surprising.
\begin{figure}
\centering
\epsfxsize=8cm
\leavevmode
\epsffile{rho0.epsi}
\vskip1mm
\caption{Spin stiffness vs the inverse cubed of the system size.
The line is a fit to the cubic form (\protect{\ref{rho0scale}}).}
\label{figrho0}
\end{figure}
For a model with O(2) symmetry, chiral perturbation theory predicts
the size dependence of the energy as \cite{hasenfratz}
\begin{equation}
E_0(L) = E_0 + {e_3 \over L^3} + {e_5 \over L^5} + \ldots,
\label{e0scale}
\end{equation}
with no $O(1/L^4)$ term. The leading-order correction is given
by
\begin{equation}
e_3 = -\gamma c/2 ,
\label{e3}
\end{equation}
where the constant $\gamma=1.437745$ and $c$ is the spin-wave velocity.
In order to obtain a good fit to our $L=4-16$ data, the $O(1/L^5)$ term
has to be included. The fit then has a chi-squared value per degree of
freedom of $0.7$. Using instead an $O(1/L^4)$ term gives chi-squared
$\approx 6$. We can therefore conclude that our results
support the prediction \cite{hasenfratz}
that $e_4=0$. In Figure \ref{fige0} we show the
data for $L \ge 8$ along with the best fit. The extrapolation
$L \to \infty$ gives $E_0 = -0.548824(2)$. The finite-size correction
constants are $e_3 = -0.807(2)$ and $e_5 = -1.07(3)$. Using
Eq.~(\ref{e3}) we obtain the spin-wave velocity $c=1.123(2)$. These
results are in agreement with the previous extrapolations using
exact diagonalization data for $L=2,3,4,5,6$,\cite{hamer} but the statistical
errors are considerably smaller. The energy is also in perfect agreement
with a previous Green's function Monte Carlo calculation, which gave
$E_0 = 0.54883(1)$.\cite{zhang}
We are not aware of any predictions for the size dependence of
the spin stiffness of the XY model. For the Heisenberg model, the
leading correction is $O(1/L)$.\cite{einarsson} In contrast, our XY
data can be very well fit with only an $O(1/L^3)$ term;
\begin{equation}
\rho (L) = \rho + {r_3 \over L^3} + \ldots,
\label{rho0scale}
\end{equation}
with the infinite-size value $\rho=0.2696(2)$ and the cubic correction
$r_3 = 0.47(1)$. This fit is shown in Figure \ref{figrho0}.
\begin{figure}
\centering
\epsfxsize=8cm
\leavevmode
\epsffile{ej.epsi}
\vskip1mm
\caption{The difference between the energy in the $j=1$ and $j=0$
magnetization sectors graphed vs $1/L^4$. The open circle is the exact
result for $L=6$. The points with error bars are the QMC results for
$L=8,10$ and $12$. The curve is a fit to Eq.~(\protect{\ref{eje0}}).}
\label{figej}
\end{figure}
The stiffness result
is in perfect agreement with the value obtained using exact diagonalization
and finite-size scaling of the ground state energy
in higher magnetization sectors.\cite{hamer} The energy per spin in the sector
with magnetization $j=\sum_iS^z_i$ should scale as \cite{hasenfratz}
\begin{equation}
E_j - E_0 = {j^2c^2 \over 2\rho L^4} + O(j^4,1/L^6).
\label{eje0}
\end{equation}
We have also carried out some QMC calculations in the $j=1$ sector. Figure
\ref{figej} shows results for $L=8,10$, and $12$, along with the exact
result for $L=6$. Using also the exact $L=4$ result, a fit to
Eq.~(\ref{eje0}), including an $O(1/L^6)$ term, gives $c^2/\rho = 4.70(4)$.
This value agrees with the above separate estimates of
$c$ and $\rho$. The consistency between the results, obtained in two
different ways, clearly gives very strong support to the
Lagrangian theory.
The susceptibility can be obtained from $c$ and $\rho$ using
the standard hydrodynamic relation
\begin{equation}
\chi = \rho / c^2 ,
\end{equation}
which with our values of $c$ and $\rho$ gives $\chi = 0.2138(8)$.
\begin{figure}
\centering
\epsfxsize=8cm
\leavevmode
\epsffile{m.epsi}
\vskip1mm
\caption{Magnetization vs inverse system size and a fit to the
scaling form (\protect{\ref{scalem}}).}
\label{figm}
\end{figure}
For the magnetization, we assume a finite-size scaling
\begin{equation}
M^2_x(L) = M^2_x + {a_1 \over L} + {a_2 \over L^2} + \ldots .
\label{scalem}
\end{equation}
A fit to this form, including the quadratic term, is shown in Figure
\ref{figm} and gives $M_x^2 = 0.0956(6)$ and $a_1 = 0.1212(5)$. According to
Eq.~(\ref{mM}), the magnetization is thus $m=0.437(2)$, in good agreement
with a previous Green's function Monte Carlo result [$m=0.441(5)$].\cite{zhang}
For the Heisenberg model, the linear correction factor is related to ground
state parameters according to $a_1 = \alpha M^2_x /(c\chi)$, where
$\alpha = 0.62075$.\cite{neuberger} The $a_1$ obtained here is instead
consistent to within a few percent with $a_1 = \alpha M^2_x /(2c\chi)$.
As in the scaling of the energy,\cite{hasenfratz} the leading size
correction is hence proportional to the number of gapless modes
in the symmetry-broken system.
\section{Finite-temperature results}
We now discuss calculations aimed at extracting the leading finite-temperature
corrections to the ground state. Previously, extensive simulations were
carried out in order to study the KT transition, and a critical temperature
$T_{\rm KT} \approx 0.343J$ was found.\cite{harada} Here we focus on the
susceptibility and the internal energy density at lower temperatures.
Figure \ref{figxt} shows the susceptibility, defined in Eq.~(\ref{suscept}),
calculated for lattices with $L=32$ and $64$ for temperatures $T/J \ge 0.05$.
Within statistical errors there are no differences between the data for
the two system sizes, and hence the results represent the thermodynamic
limit. Chiral perturbation theory predicts a temperature-independent
susceptibility for the XY model.\cite{hasenfratz} Within error bars,
our data is temperature independent for $T/J \alt 0.15$, with the
constant value estimated at $\chi = 0.2096(2)$. This result is
lower (by about five standard deviations, or $2$\%) than the value
extracted in Sec~III from the ratio $\chi=\rho/c^2$. The spin-wave
velocity was obtained from the
leading finite-size correction to the ground state energy. It is
possible that this correction is affected by the presence of sub-leading
corrections that are not completely taken into account by only including
up to $O(1/L^5)$ terms in Eq.~(\ref{e0scale}). Using the data shown in
Fig.~\ref{figxt} is a more direct way of extracting the $T=0$
susceptibility, and we therefore consider it more reliable. An
improved estimate of $c$ is then $c=\rho/\chi =1.134(2)$.
\begin{figure}
\centering
\epsfxsize=8cm
\leavevmode
\epsffile{xt.epsi}
\vskip1mm
\caption{Spin susceptibility vs temperature for lattices with $L=32$
(solid circles) and $L=64$ (open circles). The dashed line is the
estimated constant low-temperature value. The inset shows results
for $L=32$ over a wider temperature range.}
\label{figxt}
\end{figure}
For a finite lattice, the gap between the $j=0$ ground state and
the finite-magnetization states implies an exponential decay of $\chi$
to zero as $T \to 0$. Chiral perturbation theory predicts that the
form of this decay for $TL^2\rho/c^2 \ll 1$ is given by \cite{hasenfratz}
\begin{equation}
\chi = {1 \over TL^2} {\rm exp}[-c^2/(2\rho TL^2)].
\label{xltprediction}
\end{equation}
In Figure~\ref{figxlt} we show QMC data for $L=4-16$ along with this
prediction, where we have used $c^2/\rho=4.77$, corresponding to our best
estimates of $c$ and $\rho$. The agreement is not perfect, but satisfactory
considering that there are no adjustable parameters and that there
should also be corrections to Eq.~(\ref{xltprediction}). For $L=4$,
we have also calculated $\chi$ using exact diagonalization down to much
lower temperatures. As shown in Figure \ref{figxt4}, there is a very good
agreement with the predicted form over a sizable low-temperature range.
Note, however, that the asymptotic $T \to 0$ decay is always purely
exponential, without the $1/T$ factor in Eq.~(\ref{xltprediction}),
as only the lowest $j=\pm 1$ states contribute.
\begin{figure}
\centering
\epsfxsize=8cm
\leavevmode
\epsffile{xlt.epsi}
\vskip1mm
\caption{The logarithm of the spin susceptibility for different
system sizes vs the logarithm of the temperature (solid circles
connected by dashed lines, $L=4,6,\ldots,16$ from right to left).
The solid curves are obtained from Eq.~(\protect{\ref{xltprediction}})
with $c^2/\rho = 4.77$.}
\label{figxlt}
\end{figure}
Chiral perturbation theory predicts that the low-temperature form of the
internal energy is given by\cite{hasenfratz}
\begin{equation}
E(T) = E(0) + {\zeta (3) \over \pi c^2}T^3 + O(T^5),
\label{etscale}
\end{equation}
where $\zeta (3) = 1.20206$. In Figure \ref{figet} we show results for
$T/J \ge 0.05$ calculated for $L=64$ lattices. We also show a comparison
between $L=16,32$, and $64$, which suggests that $L=64$ gives the
thermodynamic limit within error bars. A fit to the $L=64$ data with
$O(T^3)$ and $O(T^5)$ terms, and including also the $T=0$ energy extracted
in Sec.~III, gives the cubic correction $\zeta (3)/ \pi c^2=0.284(5)$. This
corresponds to $c=1.16(1)$, which is higher than the value $1.134(2)$
extracted above using the estimates of $\rho$ and $\chi$ by about 2.5
error bars. Again, this small deviation may indicate some influence of
higher-order corrections in the energy scaling.
\begin{figure}
\centering
\epsfxsize=8cm
\leavevmode
\epsffile{xt4.epsi}
\vskip1mm
\caption{The logarithm of the exact spin susceptibility for a $4\times 4$
lattice vs the logarithm of the temperature (dashed curve) compared
with the predicted form (\protect{\ref{xltprediction}}) with
$c^2/\rho = 4.77$ (solid curve).}
\label{figxt4}
\end{figure}
\section{Summary}
In summary, we have presented extensive QMC calculations for the
two-dimensional $S=1/2$ XY model. We have carried out finite-size and
finite-temperature scaling of several physical quantities. The results
are consistent with the predictions of effective Lagrangian theory
\cite{hasenfratz} to within $1-2$\%.
The best estimates of the ground state parameters resulting from
our calculations are
\begin{eqnarray}
E_0 & = & 0.548824(2), \nonumber \\
\rho & = & 0.2696(2), \nonumber \\
m & = & 0.437(2), \nonumber \\
\chi & = & 0.2096(2), \nonumber \\
c & = & 1.134(2). \nonumber
\end{eqnarray}
The ground state energy, $E_0$, the spin stiffness, $\rho$, and the
square of the magnetization, $m^2$, were all calculated directly
in the ground state for systems of linear size $L=4-16$ and extrapolated
to infinite size. The susceptibility $\chi$ perpendicular to the XY
spin-plane was calculated at finite temperature and extrapolated to $T=0$.
Only the spin-wave velocity $c$ was obtained by a more indirect procedure,
using the relation $c^2 = \rho/\chi$. The results are in good agreement
with previous exact diagonalization \cite{oitmaa,hamer} and QMC
work,\cite{zhang} but the precision is considerably improved over
other estimates. Our results for both the energy and the magnetization
are in remarkably good agreement with a series expansion calculation.
\cite{hamer2} The magnetization is also in excellent agreement with
second order spin-wave theory.\cite{hamer2} The energy obtained in
spin-wave theory \cite{hamer2} is only $0.2$\% higher than the
numerical result obtained here. It would be interesting to carry
out high-order spin-wave calculations also for the other quantities
discussed here.
\section{Acknowledgments}
We would like to thank Jaan Oitmaa for discussions and comments on the
manuscript. A. W. S. would like to thank the School of Physics at the
University of New South Wales for hospitality and financial support
during a visit. Support from the NSF under Grant No.~DMR-9712765
is also acknowledged.
\begin{figure}
\centering
\epsfxsize=8cm
\leavevmode
\epsffile{et.epsi}
\vskip1mm
\caption{The internal energy calculated for $L=64$ lattices, and a fit
to the form (\protect{\ref{etscale}}). The inset shows data for $L=16$
(squares), $32$ (open circles), and $64$ (solid circles).}
\label{figet}
\end{figure}
|
\section{Introduction} \label{intro}
The paper is organized as follows: section 1 describes the
basic facts of hierarchy of nonlinear equations, Baker-Akhiezer (BA) function
theory for finite--gap integrationn method.
Particular solutions in terms of generalized Novikov, Lam\'e, Hermite type
polynomials are presented.
\section{ Preliminaries }
We review in this section some basic facts about Baker--Akhieser
function which will be used in the sequel. For systematic treatments
and proofs, we refer the reader to \cite{Kr2}, \cite{Cher}.
\begin{eqnarray}
[\partial_{x}-U,\partial_{t}-V^{(N)}]=0, \\
U=U(x,t,\lambda),\qquad V^{(N)}=V^{(N)}(x,t,\lambda),
\label{ZqE}
\end{eqnarray}
where $V^{(N)}=V_{0}\lambda^N+V_{1}\lambda^{(N-1)}+\ldots V_N$.
The nonlinear equations in the form (\ref{ZqE}) are obtain as
compatibility condition of the following linear differential
equations
\begin{eqnarray}
\partial_{x}\Psi=U(x,t,\lambda)\Psi, \qquad
\partial_{t}\Psi=V_{t}^{(N)}(x,t,\lambda)\Psi.
\label{LS}
\end{eqnarray}
where $\Psi$ is a vector-collumn $(\Psi_{1} \ldots \Psi_{n})^{T}$.
The system (\ref{LS}) play an impotran role in the theory of
finite-gap integration method. The function $\Psi$, which is
general solution of the system (\ref{LS}), is meromorphic
function on the Riemann surface and have esential singularities
of prescribed form near fixed points on Riemann surface. This surface
is defined by
\begin{equation}
\mbox{det}(L(x,t,\lambda)-\mu\mbox{I})=0.
\end{equation}
We call finite-gap solutions of (\ref{ZqE}) these for which there
exists the meromorphic in $\lambda$ matrix function $L(x,t,\lambda)$
such that
\begin{eqnarray}
L_{t} = [V^{i}, L], \quad i=1,\ldots ,N \\
L_{x} = [U,L],\quad L=L_{0}\,\lambda^{N} +
L_{1}\,\lambda^{N-1} + \ldots L_{N} .
\end{eqnarray}
\section{Resultant of two differential operators}
Let us consider the following two differential operators
\begin{eqnarray}\label{A.1}
A =\sum_{k=0}^n a_{k}D^{k}, \quad a_{n}\neq 0, \qquad
B =\sum_{l=0}^m b_{l}D^{l}, \quad b_{m}\neq 0, \quad a_{k},
b_{l}\in C^{k(l)}_{I}
\end{eqnarray}
where $D = d/dx$, $C^{k(l)}_{I}$ is the set of $k(l)$-th
differentiable functions on the interval $I$ of real variable $x$.
Right resultant of operators $A$ and $B$ $\mbox{RRes}\,(A,B)$ is called
the determinant of resultant matrix $R$ of degree $(m+n)$.
Analogically left resultant of operators $A$ and $B$ $\mbox{LRes}\,(A,B)$
is called the resultant matrix $R^{*}$ of degree $(m+n)$, i.e.
$\mbox{RRes}\,(A,B) = \det (R)$, $\mbox{LRes}\,(A,B) = \det (R^{*})$.
By definition $\mbox{LRes}\,(A,B)$ = $\mbox{RRes}\,(A^{*},B^{*})$,
where
$A^{*}$ and $B^{*}$ are the conjugated operators. The differential
resultant answers to the question when the operators $A$ and $B$ have
right(left) divisor i.e. when the overdetermined system $Ay=0$, $By=0$
(or $A^{*}y=0$, $B^{*}y=0$) has soluion. By construction to define right
resultant $\mbox{RRes}\,(A,B)$ we act with operators $D^{m-1},\ldots ,D$,
$D^{0}=1$, by left to $A$ and with operators $D^{n-1},\ldots, D$,
$D^{0}=1$. to $B$. In this case the system $Ay=0$, $By=0$ has the
following form \begin{eqnarray}\label{A.2} \sum_{k=0}^{n+s}a_{k,s}y^{(k)}
= 0 , \qquad s = 0 \div (m-1), \qquad \sum^{m+p}_{l=0} b_{l,p}y^{(l)} =
0, \qquad p = 0 \div (n-1), \end{eqnarray} where the coefficients
$a_{k,s}$ and $b_{l,p}$ are computed by: \begin{eqnarray}\label{A.3}
a_{k,s}= \sum_{i=0}^{s} \left({s\atop i}\right) a^{(s-i)}_{k-i},\qquad
b_{l,p} = \sum_{j=0}^{p}\left({p\atop j}\right) b^{(p-j)}_{l-j}.
\end{eqnarray}
The system (\ref{A.2}) is constructed by homogeneous linear algebraic
equations in terms of the following variables
$y^{(n+m-1)},\ldots, y'$, $y$. The matrix of coefficients of this system is the
right resultant of the matrix
$R$, $\det (R)= \mbox{RRes}\,(A,B)$.
\begin{eqnarray}\label{A.4}
\mbox{RRes}\,(A,B) = \left| \begin{array}{cccccc}
a_{n+m-1,m-1} & a_{n+m-2,m-1} & \ldots &\ldots &\ldots & a_{0,m-1}\\
0 & a_{n+m-1,m-2} & \ldots &\ldots &\ldots & a_{0,m-1}\\
\ldots &\ldots &\ldots &\ldots &\ldots &\ldots \\
0 &\ldots &\ldots &a_{n,0} &\ldots &a_{0,0} \\
b_{n+m-1,n-1} & b_{n+m-2,n-1} & \ldots &\ldots &\ldots & b_{0,n-1}\\
0 & b_{n+m-2,n-2} & \ldots &\ldots &\ldots & b_{0,n-1}\\
\ldots &\ldots &\ldots &\ldots &\ldots &\ldots \\
0 &\ldots &\ldots &b_{m,0} &\ldots &b_{0,0} \end{array} \right| .
\end{eqnarray}
If the system $Ay=0$, $ By=0$ is consistent then
$\det(R)=\mbox{RRes}\, (A,B)=0$, in the opposite case the system is
not consistent.
For applications of differential resultant see for example \cite{BerZ}.
The most important of them are the following: the cirterion of existing the
greater right and left divisor of the system of operators; the criteion
of concictency of the system of linear ordinary differential equations with
one unknown function; the criterion of commutation of two linear differential
operators; he criterion of existing polynomial and exponential solutions of
linear differential equtions with variable coefficients; the criterion of
factrorization of operators of degree $n$ into products of operators
of degrees $n-1$ and $1$.
\medskip
{\bf Example.} Let us introduce the following equations,
\begin{eqnarray}
& L1 \equiv Ay = x^{2}y'' + xy' - (x^{2} + 1/4 )y = 0, & \\
& L2 \equiv By = 2xy'' + (3-4x)y' + (2x-3)y = 0 . &
\end{eqnarray}
Using (\ref{A.4}) we obtain
$$
\mbox{RRes}\,(A,B) = \left| \begin{array}{cccc}
x^{2} & 3x & -x^{2}+3/4 & -2x \\
0 & x^2 & x & - x^2 - 1/4 \\
2x & 5x-4 & 2x-7 & 2 \\
0 & 2x & 3 - 4x & 2x -3 \end{array} \right|
$$
\medskip
{\bf Problem 1.} Let us define two LODE:
\begin{eqnarray}
Ay =\sum_{k=0}^n a_{k}D^{k}y =0 , \qquad a_{n}\neq 0, \qquad
By =\sum_{l=0}^m b_{l}D^{l}y =0 , \qquad b_{m}\neq 0.
a_{k},
b_{l}\in C^{k(l)}_{I}
\end{eqnarray}
Find differential resultant of operators $A$ and $B$.
\medskip
{\bf Algorithm 1.} We solve problem 1 by the procedure
DIFRESULT$(a,n,b,m,x)$;
\begin{description}
\item[Input]
\item[$a$] is the array of coefficients of differential operator $A$;
\item[$n$] is the degree of differential operator $A$;
\item[$b$] is the array of coefficients of differential operator $B$;
\item[$m$] is the degree of differential operator $B$;
\item[$x$] is independent variable.
\item[Output:]{}
\end{description}
Using (\ref{A.3}) with given coefficients of differential operator $A$
and $B$ we compute the ellements of the resultant matrix , after that we
we find the determinant. The output is $\mbox{RRes}\,(A,B) =
\det (R)$, where $R$ is the resultant matrix.
\section{The generalized Riccati equation}
{\bf 4.2.1.\, Prime right divisor of operator $L$.}
Let us define the differential operator $L$ of degree $n$:
$$
L =\sum_{k=0}^n a_{k}D^{k}, \qquad a_{n}\neq 0, a_{k}\in C^{k}_{I} .
$$
A necessary and sufficient condition for the following factorization:
$$
L = L_{2}(D - \alpha ),\qquad \mbox{ord}\,L_{2} = n - 1
$$
is given by
$\mbox{RRes}\,(L,D-\alpha ) = 0$.
\begin{eqnarray}\label{AA.4}
\mbox{RRes}\,(L,D-\alpha) = \left| \begin{array}{cccccc}
a_{n} & a_{n-1} & \ldots &a_{2} &a_{1} & a_{0}\\
1 & -\alpha & \ldots & -\left( {n-1 \atop n-3 } \right) \alpha^{(n-3)} &
-\left( {n-1 \atop n-2 } \right) \alpha^{(n-2)} & -\alpha^{(n-1)} \\
\ldots &\ldots & \ldots & \ldots & \ldots & \ldots \\
0 & 0 & \ldots & 1 & -\alpha & -\alpha' \\
0 & 0 & \ldots & 0 & 1 & -\alpha
\end{array} \right| = 0.
\end{eqnarray}
Then we find the following generalized Riccati equation of first kind
$N(\alpha )$ of degree $(n-1)$:
$$
N(\alpha ) = L(\alpha ) + M(\alpha ),$$
where
$$
L(\alpha ) =\sum_{k=0}^n a_{k}\alpha ^{k},
$$
and $L(\alpha )$ is called he pseudo-characteristic Riccati equation of
first kind
$$
M(\alpha ) =\sum_{k=0}^n a_{k}M_{k-1}, \qquad M_{k-1} = (D + \alpha
)^{(k-1)} - \alpha ^{(k-1)}
$$
and $M(\alpha )$ is called the reduced Riccati equation of first kind.
When $n=2$ the generalized Riccati equation of first kind have the form
$$
N(\alpha ) = a_{2}\alpha ' + a_{2}\alpha ^{2} + a_{1}\alpha + a_{0}
$$
{\bf 4.2.2. Prime left divisor of operator $L$.}
Let us define the following differential operator $L$ of degree $n$:
$$
L =\sum_{k=0}^n a_{k}D^{k}, \qquad a_{n}\neq 0 , \quad a_{k}\in C^{k}_{I}
$$
To find factorization of following type,
$$
L = L_{2}(D - \alpha ),\qquad \mbox{ ord} L_{2} = n - 1
$$
the necessary and sufficient condition is $\mbox{RRes}\,(L,D-\alpha ) =
0$, or to find factorization of the following type,
$$
L =(D - \alpha ) L_{1}, \qquad \mbox{ ord }L_{1} = n - 1
$$
\noindent the necessary and sufficient condition is
$\mbox{LRes}\,(L,D-\alpha ) = \mbox{RRes}\,(L^{*},D+\alpha )$, i.e.
$$
\left|
\begin{array}{ccccccc} a^*_n & a^*_{n-1} & \ldots & a^*_2 & a^*_1 &
a^*_0 \\
1 & \alpha & \ldots & \left( {n-1 \atop n-3 } \right) \alpha^{(n-3)} &
\left( {n-1 \atop n-2 } \right) \alpha^{(n-2)} & \alpha^{(n-1)} \\
\ldots &\ldots & \ldots & \ldots & \ldots & \ldots & \ldots \\
0 & 0 & \ldots & 1 & \alpha & \alpha' \\
0 & 0 & \ldots & 0 & 1 & \alpha
\end{array} \right| = 0.
$$
Thus we obtain the generalzed Ricati equation
$N^{*}(\alpha )$ of second kind and degree $(n-1)$:
$$
N^{*}(\alpha ) = L^{*}(\alpha ) + M^{*}(\alpha ),
$$
where
\begin{eqnarray}
L^{*}(\alpha ) \equiv \tau L = \sum_{r=0}^n \sum_{k=r}^n (-1)^{k} \left(
{k \atop r} \right) a^{(k-r)}_{k} \alpha^k = L^{*}_{1}(D+\alpha) .
\end{eqnarray}
Here $\tau$ is the conjugation operator, $L^{*}(\alpha )$ is
pseudo-characteristic Riccati equation of second kind and
\begin{eqnarray}
M^{*}(\alpha ) = \sum_{r=1}^n \sum_{k=r}^n (-1)^{k} \left( {k \atop r}
\right) a^{(k-r)}_{k} M_{r-1}, \qquad
M_{r-1} = (D + \alpha )^{(r-1)} - \alpha ^{(r-1)},
\end{eqnarray}
$M^{*}(\alpha )$ is the reduced Riccati equation of second kind.
When $n=2$ the generalized Riccati equation of second kind have the form,
$$
N(\alpha ) = a_{2}\alpha ' + a_{2}\alpha ^{2} + (-a_{1}+2a_2')
\alpha + a_{0} - a_1' + a_{2}''= 0
$$
As a consequence of problem $1$ we have:
\medskip {\bf Problem 2.} Let us define the following equations,
\begin{eqnarray}
&\displaystyle
L_{1}y =\sum_{k=0}^n a_{k}D^{k}y = 0, \qquad a_{n}\neq 0, \quad a_{k}\in
C^{k}_{I} ,& \\
&\displaystyle
L^{*}_{1}y \equiv \tau L_{1} =\sum_{r=0}^n \sum_{k=r}^n (-1)^{k}
\left( {k \atop r} \right) a^{(k-r)}_{k}y^{k} =0.
\end{eqnarray}
Find the generalized equation of first kind,
$$
N_{1}(\alpha ) =\mbox{RRes}\, ( L_{1}, D - \alpha )
$$
and the generalized Riccati equation of second kind,
$$
N^{*}_{2} (\alpha ) =\mbox{RRes}\,( L^{*}_{1}, D + \alpha ) ).
$$
\medskip {\bf Algorithm 2.} We solve the problem $2$
by the procedure
$RICCATI(a,n,pp,x)$.
\begin{description}
\item[Input]{}
\item[$a$] is the array of coefficients of differential operator $L_{1}$;
\item[$n$] is the degree of differential operator $L_{1}$;
\item[$pp$] is the kind of seeking Riccati equation $(pp=1$ and $pp=2)$;
\item[$x$] is the independent variable.
\item[Output:]
The generalized Riccati equation of first kind $(pp=1)$ and
of second kind $(pp=2)$.
\item[R1:] If $pp=1$ then go to R2 else go to R3;
\item[R2:] $\ll L2 = (D - \alpha )y$;
\begin{description}
\item[$b$] is the array of coefficients of operator $L2$;
\item[$m:=1$];\quad
$N_{1}(\alpha ):=\mbox{DIFRESULT}(a,n,b,m,x)$; \quad return
$N_{1}(\alpha ) \gg $;
\end{description}
\end{description}
\begin{description}
\item[R3:] $\ll L^{*}_{1} = \tau L_{1}$,\quad $\tau$ is operator of
conjugation,
$$
L^{*}_{1} = \sum_{r=0}^n \sum_{k=r}^n (-1)^{k} \left( {k \atop r} \right)
a^{(k-r)}_{k}y^{k},
$$
\end{description}
\begin{description}
\item[$a^{*}$] is the array of coefficients of differential operator
$L^{*}_{1}$, \quad $L^{*}_{2}= (D + \alpha )y$;
\item[$b^{*}$] is the array of the coefficients of differential operator
$L^{*}_{2}$;
\item[$m:=1$];\quad
$N_{2}(\alpha ):=\mbox{DIFRESULT}(a^*,n, b^*,m, x)$;\quad return
$N_{2}(\alpha ) \gg $;
\end{description}
\section{ ODE resolved by algebraic means}
{\bf 4.3.1. Linear differential equations with exponential solutions
of type
$y = \exp (-\alpha x)$,\quad $ \alpha =
\mbox{const}$.}
The necessary and sufficient condition for existence of such solutions is
the pseudo-characteristic Riccati equation
of first kind $L(\alpha )$ and of second kind $L^{*}(\alpha )$
to have solution $\alpha = \mbox{const}$, i.e.
$\mbox{RRes}\,(L,D+\alpha)=0$
or $\mbox{RRes}\,(L^{*},D-\alpha ) = 0$. Then the linear differential
equation
$Ly =\sum_{k=0}^n a_{k}D^{k}y$ have the form
$Ly = L_{2}(D + \alpha )y$, where $L_{2}$ is differential
operator,
$\mbox{ord}\,(L_{2})=n-1,$ and the differential equation have
the exponential solution of following type $y = \exp (-\alpha x)$.
In the case of linear equation of second degree
$$
Ly = a_{2}(x)y'' + a_{1}(x)y' + a_{0}(x)y =0, \qquad a_{k}\neq 0 , \quad
a_{k}\in C^{k}_{I}, \quad k=0\div 2 ,
$$
the problem is to find the coefficients of factrization
$\alpha _{1}$, $\alpha _{2}$ in
$Ly = (D - \alpha _{1})(D - \alpha _{2})y = 0$, where $\alpha_1 =
\mbox{const}$ and $\alpha_2 = \mbox{const}$.
i.e. we consider the classes of equations for which the associated
Riccati equation is of first type or of second type in terms of $\alpha $
have constant solutions.
\medskip {\bf Problem 3.} Given LEDE of degree $n$,
$$
Ly =\sum_{k=0}^n a_{k}D^{k}y =0 ,\qquad a_{n}\neq 0, a_{k}\in C^{k}_{I}.
$$
Find exponential solution $y = \exp^(-\alpha x)$,
where $\alpha = \mbox{const.}$.
\medskip {\bf Algorithm 3.} In our program the problem 3 is solved by the
procedure
\begin{description}
\item[] $LIVDIF(a,d,n,x)$.
\item[Input]
\item[$a$] is the array of coefficents of operator $L$;
\item[$d$] is nonhomogeneous part of equation $Ly$;
\item[$n$] is the degree of differential operator $L$;
\item[$x$] is the independent variable.
\item[Output]
If we find the particular solution of the following form
$y = \exp (-\alpha x)$, $\alpha = \mbox{const}$ and differential
equation is of degree two the coefficients of factorization
$\alpha_{1}$, $\alpha_{2}$ and
the fundamental system of solutions
$y_{1}(x)$, $y_{2}(x)$ are obtained. If the degree of LODE
is $n > 2$, then the degree of LODE is reduced by 1.
If the exponential solution does not exists the message is received.
\item[D1:] Algorithm 2 for finding the generalized Riccati equation
of first type $N_{1}(\alpha )$, \\
$N_{1}(\alpha ):= RICCATI(a,n,1,x)$;
\item[D2:] $L_{1}(\alpha ) =\sum_{k=0}^n a_{k}\alpha ^{k}$;
$L_{1}(\alpha )$ - pseudo-characteristic equation of first kind;
\item[D3:] $M_{1}(\alpha )=N_{1}(\alpha )-L_{1}(\alpha )$;
$M_{1}(\alpha )$ - reduced equation of first kind;
\item[D4:] Subalgorithm for finding constant solution of algebraic
\\ equation $L_{1}(\alpha )$ in terms of $\alpha$. \\
{\bf \qquad Output:} constant solution $\alpha _{0}$, or \\ obtain
message that the solution of this type does not exists,
\item[D5:] If $\alpha _{0}$ exists and is solution of
$M_{1}(\alpha ) = 0$ then go to D11
else go to D6;
\item[D6:] Algorithm 2 for finding the generalized Riccati equation of
second type $N_{2}(\alpha )$.\\
$N_{2}(\alpha ):= \mbox{RICCATI}(a,n,2,x)$;
\item[D7:]
\begin{eqnarray}
L_{2}(\alpha ) = \sum_{r=0}^{n} \sum_{k=r}^{n} (-1)^{k}
\left( {k\atop r}\right) a_{k}^{(k-r)}\alpha^k
\end{eqnarray}
($L_{2}(\alpha)$-pseudo-characteristic equation of second type)
\item[D8:]
\begin{equation}
M_{2}(\alpha)=N_{2}(\alpha) - L_{2}(\alpha)
\end{equation}
($M_{2}(\alpha)$-reduced Riccati equation of second type)
\item[D9:]
Subalgorithm of finding constant solution of algebraic equation
$L_{2}(\alpha)$ in terms of $\alpha$
(Output- $\mbox{const.}$ the solution $\alpha_{0}$, or the message
that such a solution does not exists)
\item[D10:] If $\alpha _{0}$ exists and is solution of
$M_{2}(\alpha ) = 0$ then go to D11
else go to D13;
\item[D11:] If $n=2$ then go to D12 else go to D13;
\item[D12:]
<< $\alpha_{2}=\alpha_{0};$
$\alpha{1}=-(a_{1}+\alpha_{2});$
\begin{equation}
y_{1}=\exp(\int\alpha_{1}dx), \quad
y_{2}=y_{1}\int\exp(\int(\alpha_{2}-\alpha_{1})dx)dx;
\end{equation}
Message "The coefficients of factorization $\alpha_{1}$ and
$\alpha_{2}$ and fundamental solutions $y_{1}$ and $y_{2}$
are obtained " >>;
\item[D13:]
<< $ y_{1}=\exp(\int\alpha_{0}dx)$,
Message "The degree of differential equation is reduced by 1";
New Lode := Sub($y=y_{1}\int zdx, Ly$);
In equation
\begin{equation}
Ly =\sum_{k=0}^n a_{k}D^{k}y =0, \quad a_{n}\neq 0 ,
\quad a_{k}\in C^{k}_{I}
\end{equation}
we make the change of variables $y=y_{1}\int zdx$
where $z$ is new variable)
Return New Lode >>;
\item[D14:]
Message "The solution of exponential type does not
exists";
Subalgorithm for finding of constant solution of algebraic equation
$L(\alpha)$ in terms of $\alpha$.
In our program this subalgorithm is realized by procedure
$\mbox{HAREQ}(L(\alpha),\alpha,x)$;
Output:
$L(\alpha)$ is algebraic equation in term of $\alpha$.
$x$ is independen variable.
Output:
We have $\mbox{const.}$ is the solution $\alpha_{0}$ of equation
$L(\alpha)$, or message, such a solution does not exists
Example:
\begin{equation}
Ly=y'' + (8 + \sin(x)^{2})y' + 8\sin(x)^{2}y =0; \quad (')=d/dx
\end{equation}
In the array $a(n)$, $n=0\div 2$, we write the coefficients of the equation
\begin{equation}
a(2):=1; \quad a(1):=8 + \sin(x)^{2}; \quad a(0):=8\sin(x)^{2}; \quad
n:=2; \, d:=0;
\end{equation}
Using the procedure $\mbox{LIVDIF}(a,d,n,x)$; we seek particular
solution of the following type $y = \exp^(-\alpha x)$,
where $\alpha = \mbox{const.}$, and if the differential equation is of
degree $n=2$ , then we find the coefficents of factorization
$\alpha_{1}$, $\alpha_{2}$ and the fundamental system of solutions
$y_{1}(x)$, $y_{2}(x)$, also the general solution.
Then we have the solution of differential equation of type
$y = \exp^(-8x)$, and the coefficents of factorization are
\begin{equation}
\alpha_{1}=8; \quad \alpha_{2}=\sin(x)^2,
\end{equation}
and due to the fact that the equation is of second degree we may found and
second solution
\begin{equation}
y_{2}=y_{1}\int\exp(\int(\alpha_{2}-\alpha_{1})dx)dx,
\end{equation}
and general solution
\begin{equation}
y=C1y_{1}+C2y_{2}.
\end{equation}
{\bf 4.3.2. LODE with polynomial solutions}
To exists in the set of solutions of the equation
\begin{equation}
Ly =\sum_{k=0}^n a_{k}D^{k}y =0, \quad a_{n}\neq 0 ,
\quad a_{k}\in C^{k}_{I}
\end{equation}
polynomial of degree $p\neq m$ the necessary and sufficent condition
is $\mbox{RRes}(L,D^{m+1})=0$, where
\begin{equation}
L =\sum_{k=0}^n a_{k}D^{k}, \quad D=d/dx ,
\quad D^{m+1}=d^{m+1}/dx,
\end{equation}
i.e. using the differential resultant of operators $L$ and
$D^{m+1}$ we find the degree of existing polynomial solution
of LODE and after that by metod of indefinite coefficents we find
the coefficents of this polynomial and if the degree
of LODE is $n=2$ we may obtain the general solution and when $n>2$ we
reduce by 1 the degree of LODE.
\medskip {\bf Problem 4.} Given LODE of degree $n$:
\begin{equation}
Ly =\sum_{k=0}^n a_{k}D^{k}y =0, \quad a_{n}\neq 0 ,
\quad a_{k}\in C^{k}_{I}
\end{equation}
find the polynomial solution.
\end{description}
\medskip
{\bf Algorithm 4.} To solve the problem 4 we apply the procedure
POLDIF(a,n,x,hs);
\begin{description}
\item[Input]
\item[$a$] is the array of coefficents of LODE $Ly=0$,
\item[$n$] is the degree of differential operator $L$,
\item[$x$] is independent variable,
\item[$hs$] is the maximal degree of the polynomial solution,
\item[Output:]
\end{description}
If there exists the polynomial solution of degree $p\leq hs$, then
in the case of $n=2$ we find the general solution, and when
$n > 2$ the degree is reduced by 1. If there is no solution we have the
message that polynomial soluion does not exists.
\begin{description}
\item[P1:] $\quad i:=1$;
\item[P2:] Algorithm 1 for finding the differential resultant
$\mbox{RRes}\,(L,D^{i})$;
\item[P3:] If $\mbox{RRes}\,(L, D^{i}) = 0$, then go to P6 else go to
P4;
\item[P4:] If $i>hs$ then go to P12 else go to P5;
\item[P5:] $\ll i:=i+1$;\quad go to \quad P2 $\gg $;
\item[P6:] P$:=i-1$;
\item[P7:] Message ``$P$ is the degree of polynomial solution'';
\item[P8:] Subalgorithm for finding the coefficents of the
polynomial solution ( Output- polynomial soluition)
\item[P9:] If $n=2$ then go to P10 else go to P11;
\item[P10:] Find the fundamental system of LODE.\\
$\ll y_{1}=$ plynomial solution;
$$
y_{2}= y_{1}\int \exp \left(- \int a_{1}dx \right) y^{-2}_{1} dx
$$
Return \quad $y_{2},y_{1}\gg $;
\item[P11:] The degree of LODE is reduced by 1.\\
<< $ y_{1}=$ plynomial solution;\\
New Lode $:= \mbox{Sub}( y = y_{1}\int zdx, Ly)$, where
in equation
\begin{equation}
Ly =\sum_{k=0}^n a_{k}D^{k}y =0, \quad a_{n}\neq 0 ,
\quad a_{k}\in C^{k}_{I}
\end{equation}
we make the change of variables $y=y_{1}\int zdx$
where $z$ is the new variable\\
Return New Lode >>;
\item[P12:] Message ``The finded polynomial solution of the
LODE '';
\end{description}
Subalgorithm for finding the coefficents of the polynomial solution
of the LODE.
In our program this subalgorithm is realized by the procedure
$DEPOLDIF(a,n,x,p)$;
\begin{description}
\item[Input:]
\item[$a$] is the array of coefficents of of the differential equation $y=0$
\item[$n$] is the degree of differential operator $L$,
\item[$x$] is independent variable,
\item[$p$] is the degree of polynomila solution of $Ly=0$.
\item[Output:]
The coefficents of the polynomial solution of by the method of indefinite
coefficents.
\end{description}
\medskip
{\bf Example. }
$$
Ly = x^2 (\ln (x)-1)y'' - x y' + y =0, \qquad (') = d/dx
$$
The array $a(n), n=0\div 2,$ contain the coefficents of the given LODE.
$$
a(2):= x^2(\ln (x)-1); \qquad a(1):=-x; \qquad a(0):=1; \quad n:=2;
\quad hs:=3;
$$
By using the procedure $POLDIF(a,n,x,hs)$ we find the fundamental system
of solution
$y1:=x; y2:=-\ln (x)$. Then the general solution is given by:
$y:=C1 x + C2\ln (x)$;
{\bf 4.3.3. The equations in terms of exact differentials.}
Let us define
\begin{equation}
Ly =\sum_{k=0}^n a_{k}D^{k}y =0, \quad a_{n}\neq 0 ,
\quad a_{k}\in C^{k}_{I}, \quad D=d/dx
\end{equation}
This equation is expressed in terms of exact differentials of
there exists the following form
\begin{equation}
Ly =D\sum_{k=0}^{n-1} b_{k}D^{k}y =0, \quad D=d/dx,
\end{equation}
where
\begin{equation}
a_{n}=b_{n-1}(x), \quad a_{0}=b_{0}'(x),
\quad a_{k+1}=( b_{k}+b_{k+1}'(x)),
\qquad (')=d/dx, \quad k=0\div (n-2).
\end{equation}
The necessary and sufficient condition the equation $Ly=0$ to be
in exact differential is,
\begin{equation}
\sum_{k=0}^n (-1)^{k}a_{k}^{k}(x) =0,
\end{equation}
If the equation
\begin{equation}
Ly =\sum_{k=0}^n a_{k}D^{k}y =0,
\end{equation}
is expressed in terms of exact differentials in the case $n=2$ we may
find the general solution of this equation and when $n>2$ we may
find fist integral.
\medskip {\bf Problem 5.} Given LODE of degree $n$:
\begin{equation}
Ly =\sum_{k=0}^n a_{k}D^{k}y = f(x), \quad a_{n}\neq 0 ,
\quad a_{k}\in C^{k}_{I}
\end{equation}
Check is this solution of LODE in exact differentials and if the equation is
of this type find the general solution (when $n=2$) and
first integral (when $n>2$).
{\bf Algorithm 5.} To solve the problem 5 we apply
the procedure
EQDIF(a,n,dl,x);
\begin{description}
\item[Input:]
\item[$a$] is the array of coefficents of LODE
\item[$n$] is the degree of differential operator $L$,
\item[$dl$] is the nonhomogeneous part of given LODE
\item[$x$] is independent variable,
\item[Output:]
\end{description}
Check is given equation in exact differentials and if it is true find
the general solution in the case $n=2$ and first integral
if $n>2$, and if it not the case,
message that the equation is not in exact differentials
\begin{description}
\item[E1:]
\begin{equation}
D:=\sum_{k=0}^n (-1)^{k}a_{k}^{k}(x) =0,
\end{equation}
\item[E2:] If $D=0$ then go to E4 else go to E3;
\item[E3:] Message:
"This equation is not in exact differentials".
\item[E4:]
<< $b(n-1):=a(n)$;
For $k:=(n-2)$ Step (-1) Until 0 Do
\begin{equation}
b(k):=a(k+1)-b(k+1)', \qquad (') = d/dx
\end{equation}
\item[E5:]
If $n=2$ Then Go To E6 Else Go To E7;
\item[E6:]
\begin{equation}
y=\exp(-\int a(1)dx) (c_{2} +
\int(\int dl dx +C_{1})\exp(\int a(1)dx)dx);
\end{equation}
( $C_{1}$, $C_{2}$ - are constant of integration,
$y$ - general solution of LODE)
\item[E7:]
\begin{equation}
L_{2}y =D\sum_{k=0}^{n-1} b_{k}D^{k}y , \quad D=d/dx,
\end{equation}
\end{description}
{\bf Appendix 3: Differential resultant and algebra of commuting diffrential
operators.}
\begin{center}
{\bf Example 1}
\end{center}
\begin{eqnarray} \label{HHii}
H = \frac{1}{2} \left(p_1^2 + p_2^2 \right) +
\frac{1}{2} q_1q_2^2 + q_1^3 .
\end{eqnarray}
We associate to this Hamilonian (\ref{HHii}) the following
linear differential operators
\begin{eqnarray}
&& L=D^2 + q_1, \\
&& M=16 D^5 + 40 q_1 D^3 + 60 p_1 D^2 -
(120 q_1^2 + 25 q_2^2) D \nonumber \\ && -60q_1p_1 - 15 q_2 p_2,
\end{eqnarray}
This example show the abilities of the differntial resultant to
find the first inetgrals and algbraic curve associated to
given linear diferential operators. Using REDUCE 3.4 sintactis we
have
\begin{verbatim}
depend qq1,x; depend qq2,x; depend pp1,x; depend pp2,x;
a(0):=-60*qq1*pp1-15*qq2*pp2-zz;
a(1):=-120*qq1**2-25*qq2**2; a(2):=60*pp1;
a(3):=40*qq1; a(4):=0; a(5):=16;
b(0):=qq1-ll; b(1):=0; b(2):=1;
b(3):=0;
out nik52kdv;
res52:=difresult(a,5,b,2,x)$
off mcd; res52:=num(res52)/den(res52);
write "denumerator:=",den(res52);
p1:=deg(res52,zz); write "p1:=",p1;
p2:=deg(res52,ll); write "p2:=",p2;
for i:=0:p1 do
<<
d1(i):=coeffn(res52,zz,i);
LET df(qq1,x,2)=-3*qq1**2-qq2**2/2;
LET df(qq2,x,2)=-qq1*qq2;
LET df(pp1,x)=-3*qq1**2-qq2**2/2;
LET df(pp2,x)=-qq1*qq2;
LET df(qq2,x,3)=df(-qq1*qq2,x);
LET df(qq2,x)=pp2,df(qq1,x)=pp1;
LET df(qq2,x,5)=df(-qq1*qq2,x,3);
LET df(pp1,x,2)=df(-3*qq1**2-qq2**2/2,x);
LET df(pp2,x,2)=df(-qq1*qq2,x);
LET df(pp2,x,4)=df(-qq1*qq2,x,3);
Determinant of resultant matrix
i-degree of zz:=0
j - degree of ll:=0
d2(0,0):=0
***********************************
i-degree of zz:=0
j - degree of ll:=1
2 2 2 4
d2(0,1):=4*qq1 *qq2 - 8*qq1*pp2 + qq2 + 8*qq2*pp1*pp2
***********************************
i-degree of zz:=0
j - degree of ll:=2
3 2 2 2
d2(0,2):= - 32*qq1 - 16*qq1*qq2 - 16*pp1 - 16*pp2
***********************************
i-degree of zz:=0
j - degree of ll:=3
d2(0,3):=0
***********************************
i-degree of zz:=0
j - degree of ll:=4
d2(0,4):=0
***********************************
i-degree of zz:=0
j - degree of ll:=5
d2(0,5):=-256
***********************************
p3:=0
i-degree of zz:=1
j - degree of ll:=0
d2(1,0):=0
***********************************
p3:=0
i-degree of zz:=2
j - degree of ll:=0
d2(2,0):=1
***********************************
\end{verbatim}
The associated alegebraic curve has the form
(the results are obtaine dby differential resultant
\begin{eqnarray}
z^2 = -256\lambda^5 -32 E\lambda^2 + 8 K\lambda
\end{eqnarray}
where
\begin{eqnarray}
& &E = \frac{1}{2}\left(p_1^2 + p_2^2\right) +
q1^3 + \frac{1}{2} q_1 q_2^2, \\
& &K=q_2 p_1p_2-q_1p_2^2+q_2^4 + 1/2 q_1^2q_2^2.
\end{eqnarray}
\begin{center}
{\bf Example 2} (Fordy example \cite{Fordy}
\end{center}
\begin{verbatim}
a(0):=-zz; a(1):=-5*qq2**2; a(2):=15*pp1;
a(3):=15*qq1; a(4):=0; a(5):=9;
b(0):=-ll; b(1):=qq1; b(2):=0;
b(3):=1;
out nik53sk;
res53:=difresult(a,5,b,3,x)$
off mcd; res53:=num(res53)/den(res53); write "denumerator:=",den(res53);
p1:=deg(res53,zz); write "p1:=",p1;
p2:=deg(res53,ll); write "p2:=",p2;
for i:=0:p1 do
<<
d1(i):=coeffn(res53,zz,i);
LET df(qq1,x,2)=-qq1**2/2-qq2**2/2;
LET df(qq2,x,2)=-qq1*qq2;
LET df(pp1,x)=-qq1**2/2-qq2**2/2;
LET df(pp2,x)=-qq1*qq2;
LET df(qq2,x,3)=df(-qq1*qq2,x);
LET df(qq2,x)=pp2,df(qq1,x)=pp1;
LET df(qq2,x,5)=df(-qq1*qq2,x,3);
LET df(pp1,x,2)=df(-qq1**2/2-qq2**2/2,x);
LET df(pp2,x,2)=df(-qq1*qq2,x);
LET df(pp2,x,4)=df(-qq1*qq2,x,3);
d1(i):=d1(i);
p3:=deg(d1(i),ll); write "p3:=",p3;
for j:=0:p3 do
<< d2(i,j):=coeffn(d1(i),ll,j);
write "i-degree of zz:=",i;
write "j - degree of ll:=",j;
write "d2(",i,",",j,"):=",d2(i,j);
write "***********************************" >> >>;
Determinant of resultant matrix
denumerator:=1
i-degree of zz:=0
j - degree of ll:=0
d2(0,0):=0
***********************************
i-degree of zz:=0
j - degree of ll:=1
4 2 2 4 2
d2(0,1):= - 9/4*qq1 *qq2 - 3/2*qq1 *qq2 - 9*qq1 *qq2*pp1*pp2
6 3 2 2
- 1/4*qq2 - 3*qq2 *pp1*pp2 - 9*pp1 *pp2
***********************************
i-degree of zz:=0
j - degree of ll:=2
d2(0,2):=0
***********************************
i-degree of zz:=0
j - degree of ll:=3
3 2 2 2
d2(0,3):=27*qq1 + 81*qq1*qq2 + 81*pp1 + 81*pp2
***********************************
i-degree of zz:=0
j - degree of ll:=4
d2(0,4):=0
***********************************
i-degree of zz:=0
j - degree of ll:=5
d2(0,5):=-729
***********************************
p3:=0
i-degree of zz:=1
j - degree of ll:=0
d2(1,0):=0
***********************************
p3:=0
i-degree of zz:=2
j - degree of ll:=0
d2(2,0):=0
***********************************
p3:=0
i-degree of zz:=3
j - degree of ll:=0
d2(3,0):=1
***********************************
\end{verbatim}
The associated algebraic curve has the form
(the results are obtained by differntial resultant)
\begin{eqnarray}
z^3 = 729\lambda^5 -162 E\lambda^3 + K^2\lambda,
\end{eqnarray}
where
\begin{eqnarray}
& &E = \frac{1}{2}\left(p_1^2 + p_2^2\right) +
\frac{1}{6}q1^3 + \frac{1}{2} q_1 q_2^2, \\
& &K=3p_1p_2 + \frac{3}{2}q_2 q_1^2 +\frac{1}{2} q_2 q_2^2,
\end{eqnarray}
\begin{verbatim}
print(`A program for calculation differential resultants`):
print(`written by N.A. Kostov, Z.T. Kostova 8 August 1994`):
#find the elements of series
procoff:=proc(s,k,m,b,x)
local i,s1,s2,cr;
s1:=max(0,k-m); s2:=min(s,k);
if k < 0 or k > (m+s) then cr:=0 else
cr:=0; for i from s1 to s2 do
if (s-i)=0 then cr:=cr+binomial(s,i)*b(k-i) else
cr:=cr+binomial(s,i)*diff(b(k-i),x$s-i);
fi;
od;
fi;
RETURN(eval(cr))
end:
difresult:=proc(a,n,b,m,x)
local i1,i2,i2,j2,i3,j3,k,s,l,p;
#****************************
#** Programm for finding **
#** differential resultant **
#****************************
r:=array(1..n+m,1..n+m):
for s from 0 to (m-1) do
for k from 0 to (n+s) do
ak(k,s):=procoff(s,k,n,a,x);
od;
od;
for p from 0 to (n-1) do
for l from 0 to (m+p) do
bk(l,p):=procoff(p,l,m,b,x);
od;
od;
for i1 from 1 to m do
for j1 from 1 to (n+m) do
if i1 > 1 and j1 < i1 then r[i1,j1]:=0 else
r[i1,j1]:=ak(n+m-j1,m-i1);
fi;
od;
od;
for i2 from 1 to n do
for j2 from 1 to (n+m) do
if i2>1 and j2<i2 then r[m+i2,j2]:=0 else
r[m+i2,j2]:=bk(n+m-j2,n-i2);
fi;
od;
od;
print (`"Element of resultant matrix"`);
# for i3 from 1 to (n+m) do
# for j3 from 1 to (n+m) do
# print(`r:=`,r[i3,j3],i3,j3);
# od;
# od;
with(linalg,det);
rres1:=det(r):
RETURN(rres1)
end:
#alias( dq1=dq1(x),q1=q1(x),q2=q2(x),dq2=dq2(x)):
#b(0):=dq1-z;
#b(1):=2*q1; b(2):=0; b(3):=1;
#a(0):=-5*q2*dq2-60*q1*dq1-ll;
#a(1):=-120*q1**2-(35/4)*q2**2; a(2):=45*dq1;
#a(3):=30*q1; a(4):=0; a(5):=9;
#res3:=difresult(a,5,b,3,x);
#alias(UU=uu(x),WW=ww(x)):
#alias(UUx=diff(uu(x),x),UUxx=diff(uu(x),x,x),UUxxx=diff(uu(x),x,x,x)):
#alias(WWx=diff(ww(x),x),WWxx=diff(ww(x),x,x),WWxxx=diff(ww(x),x,x,x)):
#alias(WWxxxx=diff(ww(x),x,x,x,x),WWxxxxx=diff(ww(x),x,x,x,x,x)):
#b(0):=-Z+diff(ww(x),x); b(1):=2*WW; b(2):=0; b(3):=1;
#a(0):=20*WW*diff(ww(x),x)+10*diff(ww(x),x$3)-L;
#a(1):=20*WW**2+35*diff(ww(x),x$2); a(2):=45*diff(ww(x),x);
#a(3):=30*WW; a(4):=0; a(5):=9;
#difresult(a,5,b,3,x);
alias( DQ1=dq1(x),Q1=q1(x),Q2=q2(x),DQ2=dq2(x)):
alias(DQ1x=diff(dq1(x),x),DQ1xx=diff(dq1(x),x,x),DQ1xxx=diff(dq1(x),x,x,x)):
alias(DQ2x=diff(dq2(x),x),DQ2xx=diff(dq2(x),x,x),DQ2xxx=diff(dq2(x),x,x,x)):
alias(DQ1xxxx=diff(dq1(x),x,x,x,x),DQ1xxxxx=diff(dq1(x),x,x,x,x,x)):
alias(DQ2xxxx=diff(dq2(x),x,x,x,x),DQ2xxxxx=diff(dq2(x),x,x,x,x,x)):
alias(Q1x=diff(q1(x),x),Q1xx=diff(q1(x),x,x),Q1xxx=diff(q1(x),x,x,x)):
alias(Q2x=diff(q2(x),x),Q2xx=diff(q2(x),x,x),Q2xxx=diff(dq2(x),x,x,x)):
alias(Q1xxxx=diff(q1(x),x,x,x,x),Q1xxxxx=diff(q1(x),x,x,x,x,x)):
alias(Q2xxxx=diff(q2(x),x,x,x,x),Q2xxxxx=diff(dq2(x),x,x,x,x,x)):
#alias(Q2Nxxx=diff(Q2xx(x),x));
#alias(Q2xx=Q2xx(x));
DQ1:=Q1x:
DQ2:=Q2x:
b(0):=DQ1-Z;
b(1):=2*Q1; b(2):=0; b(3):=1;
a(0):=-5*Q2*DQ2-60*Q1*DQ1-L;
a(1):=-120*Q1**2-(35/4)*Q2**2; a(2):=45*DQ1;
a(3):=30*Q1; a(4):=0; a(5):=9;
ppp:=difresult(a,5,b,3,x):
rrr:=expand(subs(DQ1=Q1x,Q1xx=-4*Q1^2-Q2^2/4,Q2Nxxx=Q2xxx,
Q2xx=-Q1*Q2/2,Q2xxx=-Q1x*Q2/2-Q1*Q2x/2,
Q1xxxxx=-24*Q1x*Q1xx-8*Q1*Q1xxx-Q2x*Q2xx*3/2-Q2*Q2xxx/2,
DQ2=Q2x,Q1xxx=-8*Q1*Q1x-Q2*Q2x/2,ppp)):
Pres:=coeffs(rrr,[L,Z],`tres`):
printdifres:=proc(sss)
local rr;
for i from 1 to nops([tres]) do
rr:=expand(Pres[i]):
print(`number=`);
print(i);
print(`spectral paremeters=`);
print(tres[i]);
print(`element=`);
print(rr);
od;
end:
print(`*******************************************************************`);
print(`spectral parameters`,-tres[2]*162);
EE:=expand(subs(Q2xx=-Q1*Q2/2,Q1xx=-4*Q1^2-Q2^2/4,Pres[2])):
EE:=expand(EE)/(81*2):
print(`energy is equal to res2`);
print(EE);
print(`*******************************************************************`);
print(`spectral parameters`,3*tres[4]/4);
K:=expand(subs(Q2xx=-Q1*Q2/2,Q1xx=-4*Q1^2-Q2^2/4,Pres[4])):
K:=-4*expand(K)/3:
print(`second integral is res4`);
print(K);
print(`*******************************************************************`);
print(`spectral parameters`,tres[7]);
res7:=expand(subs(Q2xx=-Q1*Q2/2,Q1xx=-4*Q1^2-Q2^2/4,
Q1xxx=-8*Q1*Q1x-Q2*Q2x/2,Q2xxx=-Q1x*Q2/2-Q1*Q2x/2,
Pres[7])):
res7:=expand(res7):
print(`element res7=`);
print(res7);
print(`*******************************************************************`);
print(`spectral parameters`,tres[3]);
res3:=expand(subs(Q2xx=-Q1*Q2/2,Q1xx=-4*Q1^2-Q2^2/4,
Q1xxx=-8*Q1*Q1x-Q2*Q2x/2,Q2xxx=-Q1x*Q2/2-Q1*Q2x/2,
Pres[3])):
res3:=expand(res3):
print(`element res3 =`);
print(res3);
print(`*******************************************************************`);
print(`spectral parameters`,tres[1]);
res1:=subs(Q2xx=-Q1*Q2/2,Q1xx=-4*Q1^2-Q2^2/4,
Q1xxx=-8*Q1*Q1x-Q2*Q2x/2,Q2xxx=-Q1x*Q2/2-Q1*Q2x/2,
Pres[1]):
res1:=expand(res1):
print(`element res1=`);
print(res1);
print(`*******************************************************************`);
print(`spectral parameters`,tres[5]);
res5:=subs(Q2xx=-Q1*Q2/2,Q1xx=-4*Q1^2-Q2^2/4,
Q1xxx=-8*Q1*Q1x-Q2*Q2x/2,Q2xxx=-Q1x*Q2/2-Q1*Q2x/2,
Pres[5]):
res5:=expand(res5):
print(`element res5=`);
print(res5);
print(`*******************************************************************`);
print(`*******************************************************************`);
print(`spectral parameters`,tres[8]);
res8:=subs(Q2xx=-Q1*Q2/2,Q1xx=-4*Q1^2-Q2^2/4,
Q1xxx=-8*Q1*Q1x-Q2*Q2x/2,Q2xxx=-Q1x*Q2/2-Q1*Q2x/2,
Pres[8]):
res8:=expand(res8):
print(`element res8=`);
print(res8);
print(`*******************************************************************`);
print(`*******************************************************************`);
print(`spectral parameters`,tres[6]);
res6:=subs(Q2xx=-Q1*Q2/2,Q1xx=-4*Q1^2-Q2^2/4,
Q1xxx=-8*Q1*Q1x-Q2*Q2x/2,Q2xxx=-Q1x*Q2/2-Q1*Q2x/2,
Pres[6]):
res6:=expand(res6):
print(`element res6=`);
print(res6);
print(`*******************************************************************`);
\end{verbatim}
\begin{verbatim}
alias(q[1]=q[1](x),q[2]=q[2](x)); alias(wp=wp(x));
alias(u=u(x));
Eq[1]:=diff(q[1],x$2)+q[2]^2/2+3*q[1]^2+a[0]*q[1]-a[1]/4:
Eq[2]:=q[2]^3*diff(q[2],x$2)+q[2]^4*q[1]+q[2]^4*a[0]/4+a[4]/4:
wp1:=(4*wp^3-g2*wp-g3)^(1/2):
wp2:=eval(diff(wp1,x)):
wp3:=eval(expand(subs(diff(wp,x)=wp1,wp2)));
F:=-6; B:=0; a[0]=0; G:=-144;
for i from 1 to 2 do
SEq[i]:=eval(expand(subs(q[1]=F*wp+B,
q[2]=(G*wp^2+E)^(1/2),Eq[i]))):
SSEq[i]:=expand(subs(diff(wp,x)^2=(4*wp^3-g2*wp-g3),
diff(wp,x$2)=wp3,expand(SEq[i]))):
#print(`SSEq(`,i,`):=`,SSEq[i]);
n[i]:=degree(SSEq[i],wp):
for j from 0 to n[i] do
Cowp[i,j]:=coeff(SSEq[i],wp,j):
#Cowp1[i,j]:=expand(subs(F=0,G=0,Cowp[i,j]));
od:
#FF1[i]:=seq(Cowp[i,i1]=0,i1=0..n[i]);
FF2[i]:=seq(Cowp[i,i3],i3=0..n[i]);
od:
for i1 from 1 to 2 do
for j1 from 0 to n[i1] do
#print(`Cowp(`,i1,`,`,j1,`):=`, Cowp[i1,j1]);
print(`Cowp(`,i1,`,`,j1,`):=`,subs(a[0]=0,E=36*g2, Cowp[i1,j1]));
od;
od;
#for i2 from 1 to 2 do
#for j2 from 0 to n[i2] do
#print(`Cowp1(`,i2,`,`,j2,`):=`,Cowp1[i2,j2]);
#od:
#od:
#Fs:={FF1[1],FF1[2]};
#sols:=solve(Fs,{F,A,B,G,D,E,C1^2,C2^2,gama,beta});
with(grobner);
Fg:=[FF2[1],FF2[2]];
#Fg1:=subs(eta^(-1)=eta1,C1^2=CC1,C2^2=CC2,Fg);
#gsolve(Fg);
#gbasis(Fg1,[A,b,F,G,D,E,CC1,CC2,r,s,sigma,gama,beta,eta,eta1,g2,g3],plex);
#gbasis(Fg1,[A,D,C1^2,C2^2,gama,beta],tdeg);
#gsolve(Fg1,[A,D,C1^2,C2^2,gama,beta]);
#A1:=expand(subs(A=(-6*s-sigma*D)/eta,B=A^2/F-F*g2/4,G=-eta*F/sigma,
# E=D^2/G-G*g2/4,Cowp[1,3]));
#D1:=expand(subs(A=(-6*s-sigma*D)/eta,B=A^2/F-F*g2/4,G=-eta*F/sigma,
# E=D^2/G-G*g2/4,Cowp[2,3]));
#C12:=solve(Cowp1[1,0],C1^2);
#Gama1:=solve(Cowp1[1,1],gama);
#Gama2:=solve(Cowp1[1,2],gama);
#AA:=solve(Cowp1[1,3],A);
#C22:=solve(Cowp1[2,0],C2^2);
#Be1:=solve(Cowp1[2,1],beta);
#Be2:=solve(Cowp1[2,2],beta);
#DD:=solve(Cowp1[2,3],D);
\end{verbatim}
{\bf Acknowledgments.} N.A.K. acknowledges support of contract
F--442/1994 of the Bulgaria Ministry of Science, Education and
Technology.
|
\section{Introduction}
The {\em Leitmotiv} of this survey paper is our belief that in some
way the true geometry of spacetime is actually dictated by
quantum field theories as currently used by particle physicists in
the calculation of radiative corrections.
There are two major ingredients in this use of the theory, the
first is the renormalization technique, with all its combinatorial
intricacies, which is perfectly justified by its concrete physical
roots and the resulting predictive power. The second is the
specific Lagrangian of the theory, the result of a long dialogue
between theory and experiment, which, of course, is essential in
producing meaningful physical results.
This Lagrangian is a unique mixture of several pieces, some very
geometrical and governed by the external symmetry group of the
equi\-va\-lence principle, i.e.~the diffeomorphism group ${\bf
Diff}$, the others governed by the internal group of gauge
transformations ${\bf Gauge}$. The overall symmetry group is the
semi-direct product ${\bf G}={\bf Gauge} \mathop{>\!\!\!\triangleleft}{\bf Diff}$, which
is summarized by the following exact sequence of groups
\[
1\to{\bf Gauge}\to{\bf G}\to {\bf Diff}\to 1.
\]
But there is more group structure than the external and internal
symmetries of this Lagrangian of gravity coupled with matter. Our
goal in this paper is to explain how the understanding of the
group-theoretic principle underlying the working machine of
renormalization \cite{dhopf} should allow one to improve the
understanding of the geometrical nature of particle physics which
was proposed in \cite{alain1}.
The main point in the latter proposal is that the natural symmetry
group ${\bf G}$ of the Lagrangian is isomorphic to the group of
diffeomorphisms ${\bf Diff}(X)$ of a space $X$, provided one
stretches one's geometrical notions to allow slightly
noncommutative spaces. Indeed, the automorphism group ${\bf
Aut}({\cal A})$ of a noncommutative algebra ${\cal A}$ which
replaces the diffeomorphism group of any ordinary manifold has
exactly this very feature of being composed of two pieces, one
internal and one external, which, again, can be given equivalently as
an exact sequence of groups
\[ 1\to{\bf Int}({\cal
A})\to{\bf Aut}({\cal A})\to {\bf Out}({\cal A})\to 1.
\]
In the general framework of NCG the confluence of the two notions
of metric and fundamental class for a manifold led very naturally
to the equality
\begin{equation}
ds=1/D,\label{E1}
\end{equation}
which expresses the line element $ds$ as the inverse of the Dirac
operator $D$, hence under suitable boundary conditions as a
propagator. The significance of $D$ is two-fold. On the one hand,
it defines the metric by the above equation, on the other hand its
homotopy class represents the K-homology fundamental class of the
space under consideration. While this new geometrical framework
was immediately useful in various mathematical examples of
noncommutative spaces, including the noncommutative torus
\cite{alain2,alain3}, it is obvious that it cannot be a
satisfactory answer for spacetime precisely because QFT will
unavoidably dress the free propagator, as Fig.(\ref{dress})
indicates. It nevertheless emphasizes that spacetime itself ought
to be regarded as a derived concept, whose structure in we believe
follows from the properties of QFT if one succeeds describing the
latter in purely combinatorial terms.
\bookfig{dress}{!}{dress}{The dressed line element is obtained
from the transition of the bare to the dressed propagator.}{0.5}
Whereas it was simple in the undressed case to recover the
standard ingredients of intrinsic geometry directly from the Dirac
propagator and the algebra ${\cal A}$, it is much more challenging
in the dressed case, and we shall naturally propose the
Schwin\-ger-Dy\-son equation for the full fermion propagator as the proper
starting point in the general case.
Let us now describe various appearances of Hopf algebras relevant
to Quantum Field Theory. We will start with the Hopf algebra of
renormalization, whose antipode turned out to be the key to a
conceptual understanding of the subtraction procedure. We shall
then describe several occurences of this, or closely related Hopf
algebras, in other mathematical domains, such as foliations
\cite{CM}, Runge-Kutta methods \cite{Br}, iterated integrals
\cite{chen,dchen} and multiple zeta values \cite{sasha}.
We will finally address the Schwinger Dyson equation as a
prototype of a generalized form of an ordinary differential
equation (ODE).
\section{The pertinence of Hopf algebras}
\subsection{Renormalization and the antipode in the algebra of rooted trees}
Renormalization occurs in evaluating physical observable
quantities which in simple terms can be written as formal
functional integrals of the form
\[
\int e^{-{\cal L}(\phi,\partial \phi)}P(\phi,\partial\phi)[d\phi],\;\;\;
{\cal L}={\cal L}_0+{\cal L}_I.
\]
Computing such an integral in perturbative terms leads to a formal
power series, each term $G$ of which is obtained by integrating a
polynomial under a Gaussian $e^{-{\cal L}_0}$. Such Feynman
diagrams $G$ are given by multiple integrals over
spacetime or, upon Fourier transformation, over momentum space,
and are typically divergent in either case.
The renormalization
technique consists in adding counter\-terms ${\cal L}_G$ to the
original Lagrangian ${\cal L}$, one for each diagram $G$, whose
role is to cancel the undesired divergences. In good situations to
which we are allowed to restrict ourselves if we take guidance
from nature, this can be done by local counterterms, polynomials
in fields and their derivatives.
The main calculational complication of this subtraction procedure
occurs for diagrams which possess non-trivial subdivergences,
i.e.~subdiagrams which are already divergent. In that situation
the procedure becomes very involved since it is no longer a simple
subtraction, and this for two obvious reasons: i) the divergences
are no longer given by local terms, and ii) the previous
corrections (those for the subdivergences) have to be taken into
account.
To have an example for the combinatorial burden imposed by these
difficulties consider the problem below of the
renormalization of a two-loop four-point function in massless
scalar $\phi^4$ theory in four dimensions, given by the following
Feynman graph:
\[
\Gamma^{[2]} = \bobfig{g2}.
\]
It contains a divergent subgraph
\[
\Gamma^{[1]}=\bobfig{g1}.
\]
We work in the Euclidean framework and introduce a cut-off
$\lambda$ which we assume to be always much bigger than the square
of any external momentum $p_i$. Analytic expressions for these
Feynman graphs are obtained by utilizing a map $\Gamma_\lambda$
which assigns integrals to them according to the Feynman rules and
employs the cut-off $\lambda$ to the momentum integrations. Then
$\Gamma_\lambda[\Gamma^{[1,2]}]$ are given by
\[
\Gamma_\lambda[\Gamma^{[1]}](p_i)=\int d^4k
\frac{\Theta(\lambda^2-k^2)}{k^2}\frac{1}{(k+p_1+p_2)^2},
\]
and \[ \Gamma_\lambda[\Gamma^{[2]}](p_i)=\int d^4l
\frac{\Theta(\lambda^2-l^2)}{l^2(l+p_1+p_2)^2}
\Gamma_\lambda[\Gamma^{[1]}(p_1,l,p_2,l)].
\]
It is easy to see that $\Gamma_\lambda[\Gamma^{[1]}]$
decomposes into the form $b
\log\lambda$ (where $b$ is a real number)
plus terms which remain finite for
$\lambda\to\infty$, and hence will produce a
divergence which is a non-local function of external momenta
\[
\sim\log\lambda\int d^4l
\frac{\Theta(\lambda^2-l^2)}{l^2(l+p_1+p_2)^2}\sim\log\lambda\;\log(p_1+p_2)^2.
\]
Fortunately, the counterterm ${\cal L}_{\Gamma^{[1]}}$
$\sim\log\lambda$ generated to subtract the divergence in
$\Gamma_\lambda[\Gamma^{[1]}]$ will precisely cancel this
non-local divergence in $\Gamma^{[2]}$.
That these two diseases actually cure each other in general is a
very non-trivial fact that took decades to prove \cite{BPHZ}. For
a mathematician the intricacies of the proof and the lack of any
obvious mathematical structure underlying it make it totally
inaccessible, in spite of the existence of a satisfactory formal
approach to the problem \cite{EG}.
The situation is drastically changed by the understanding of the
mechanism behind the actual computations of radiative corrections
as the antipode in a very specific Hopf algebra, the Hopf algebra
${\cal H}_R$ of decorated rooted trees discovered in
\cite{dhopf,CK,doverl}.
This algebra is the algebra of coordinates in an affine nilpotent
group $G_R$, and all the non-linear aspects of the subtraction
procedure are subsumed by the action of the
antipode, i.e.~the operation $g\to
g^{-1}$ in the group $G_R$.
This is not only very satisfactory from the conceptual point of
view, but it does also allow the automation of
the subtraction procedure
to arbitrary loop order (see \cite{BK} for an example of Feynman
graphs with up to twelve loops) and to describe the more elaborate
notions of renormalization theory, -change of scales,
renormalization group flow and operator product expansions-, from
this group structure \cite{BK,dchen,DelKr}.
In \cite{CK} we characterized this Hopf algebra of
renormalization, ${\cal H}_R$, abstractly as the universal
solution for one-dimensional Hochschild cohomology. The Hopf
algebra ${\cal H}_R$ together with the collecting map
(cf.\cite{CK}) $B_+:{\cal H}_R\to {\cal H}_R$ is the universal
solution of the following Hochschild cohomology equation. We let
${\cal H}$ be a commutative (not necessarily cocommutative) Hopf
algebra together with a linear map $L:{\cal H}\to {\cal H}$. The
Hochschild equation is then $bL=0$, i.e.
\[
\Delta L(a)=L(a)\otimes 1+(id\otimes L)\Delta(a).
\]
We shall see later the intimate relation between this problem and
the generalized form of ODE's.
Renormalization can now be summarized succinctly
by saying that it
maps an element $g\in G_R$ to another element $g_o^{-1}g\in G_R$.
Typically, $g$ is associated with bare diagrams and $g_o$ is a
group element in accordance with a chosen renormalization scheme.
In the ratio $g_o^{-1}g$ the undesired divergences drop out.
Hence, $g_o^{-1}$ delivers the counterterm and $g_o^{-1}g$ the
renormalized Green function. It is precisely this group law which
allowed the description of the change of scales and renormalization
schemes in a comprehensive manner \cite{dchen}. Indeed, if
$g_o^{-1}g$ is one renormalized Green function, and
$g_o^{-1}g^\prime=(g^{-1}_og)(g^{-1}g^\prime)$ another, then we
immediately see the group law underlying the evolution of the
renormalization group.
\subsection{Foliations}
Quite independently, the problem of computation of certain
characteristic clas\-ses coming from differential operators on
foliations led \cite{CM} to a very specific Hopf algebra ${\cal
H}(n)$ associated to a given integer $n$, which is the codimension
of the foliation.
The construction of this Hopf algebra was actually dictated by the
commutation relations of the algebra ${\cal A}$ (corresponding to
the noncommutative leaf space) with the operator $D$
(cf.(\ref{E1})) describing the transverse geometry in the sense of
(\ref{E1}). The structure of this Hopf algebra revealed that it
could be obtained by a general procedure from the pair
of subgroups $G_0,G_1$ of the group ${\bf Diff}({\bf R}^n)$ of
diffeomorphisms of ${\bf R}^n$: One lets $G_0$ be the subgroup of
affine diffeomorphisms, while $G_1$ is the subgroup of those
diffeomorphisms which fix the origin and are tangent to the
identity map at this point. This specific structure of a Hopf
algebra not only led in this case to a solution of the
computational problem, but also provided the proper generalization
of Lie algebra cohomology to the general Hopf framework in the
guise of cyclic cohomology (see the paper {\em Cyclic cohomology
and Hopf algebras} by Connes and Moscovici in this volume).
In \cite{CK} we established a relation between the two Hopf algebras
which led us to extend various results of \cite{CM}
to the Lie algebra ${\cal L}^1$ of rooted trees, in particular we
extended ${\cal L}^1$ by an
additional generator $Z_{-1}$, intimately related to natural
growth of rooted trees.
Let us mention at this point that the extended Lie algebra ${\cal
L}$ has a radical as well as a simple quotient. The former is
nilpotent and infinitely generated, the latter turns out to be
isomorphic to the Lie algebra of ${\bf Diff}({\bf R})$, thus
exhibiting an intrinsic relation between the two groups.
Extension of this relation between ${\cal H}(1)$ and ${\cal H}_R$
leads to the Hopf algebra of decorated planar rooted trees,
explored by R.~Wulkenhaar \cite{W}, whose new feature is that the
decoration of the root provides the information necessary to
define the notion of parallel transport, while the decorations of
other vertices are mere spacetime indices.
\subsection{Numerical Integration Methods}
Our description in \cite{CK} of the simplest realization of the
Hopf algebra ${\cal H}_R$ and its related group, together with the
elaboration of this group structure in \cite{BK,dchen}, allowed
C.~Brouder \cite{Br} to recognize a far reaching relation with the
combinatorics of numerical analysis as worked out by Butcher
several decades ago.
Indeed, Butcher, in his seminal work on the algebraic aspects of
Runge-Kutta methods (and other numerical integration methods)
\cite{Bu} discovered the very same Hopf algebra and group $G_R$.
Let us first remind the reader of the simplest numerical
integration methods in the integration of a given ordinary
differential equation
\begin{equation}
\dot{y}=f(y),\;y(0)=y_0,
\end{equation}
where $t\to y(t)$ is a curve in Euclidean space $E$. The Runge
method, designed to approximate the value $y(h)$ of $y$ at $t=h$
is given by the formula,
\begin{equation}
\hat{y}(h)=y_0+h f(y_0+\frac{h}{2}f(y_0)).
\end{equation}
The virtue of this simple iteration of $f$ is that the Taylor
expansion of $\hat{y}$ for small $h$ agrees up to order two with
the Taylor expansion of the actual solution $y(h)$.
More generally a Runge-Kutta method is an iterative procedure,
governed by two sets of data, traditionally denoted by $a$ and $b$, where
$b_i$, $i=1,\ldots,n$, and $a_{ij}$ are scalars while the implicit
equations are
\begin{eqnarray}
X_i(h) & = & y_0+h\sum a_{ij} f(X_j(h)),\\ \hat{y} & = & y_0+h
\sum b_i f(X_i(s)).
\end{eqnarray}
Their solution is known to be a sum over rooted trees, involving
the very combinatorial quantities ascribed to rooted trees by
the study of QFT, namely tree factorials, CM weights and symmetry
factors \cite{Bu,Br,BK,dchen}.
It has been known since Cayley \cite{Cayley} that rooted
trees are the correct way to label polynomials in higher
derivatives of a smooth map $f:E\to E$. For instance an $n$th
derivative $f^{(n)}$ corresponds to a vertex with $n$ adjacent
branches, as in Fig.(\ref{f1}). \bookfig{deriv}{!}{f1}{The tree
structure of derivatives.}{2}
What Butcher discovered is that Runge-Kutta methods naturally form
a group, whose elements are actually scalar functions of rooted
trees. He gave explicit formulas for the composition of two
methods as well as for the inverse method. He also showed how the
standard solution of a differential equation is obtained from a
particular (continuous) method which he called the Picard method.
There is a nuance between Runge-Kutta methods and the Butcher
group of scalar functions of rooted trees, but Butcher proved that
the Runge-Kutta methods are dense in the latter group.
In \cite{dchen} scalar functions of rooted trees were used to
parametrize and generalize iterated integrals, allowing a
unified description of renormalization schemes. This immediately
allowed C.~Brouder to identify the above group $G_R$ (in the
simplest undecorated case) with the Butcher group.
The supplementary freedom in constructing group elements or
operations on them provided by the Runge-Kutta description matches
the freedom to choose a renormalization scheme or to describe the
change of a chosen scheme. Hence the group product, as well as
the counterterm, -the inverse group element, have immediate and
equally elegant counterparts in the Runge-Kutta language worked
out recently by Brouder \cite{Br}.
Moreover, comparing the data $a,b$ of a Runge-Kutta method with our
characterization of the Hopf algebra ${\cal H}_R$ by the
Hochschild cohomology problem leads us to the following natural
framework for the formulation of a universal differential equation
(which also covers the case of the Schwinger-Dyson equation)
given a non-linear map $f:E\to E$. The only nuance between our
framework and the Butcher framework of an algebra $B$ together
with a linear map $a:B\to B$ (cf.\cite{Bu}) is that we now
assume that the abelian algebra $B$ is in fact a Hopf algebra
while the map $a$ satisfies our Hochschild equation
\[
\Delta a(P)=a(P)\otimes 1+(id\otimes a)\Delta(P).
\]
The simplest example of such data is already given by the Picard
method of \cite{Bu} where the Hopf algebra structure (not provided
in \cite{Bu}) is given by
\[
\Delta x=x\otimes 1+1\otimes x,
\]
\[
a(P)(x)=\int_0^x P(u)du,
\]
cf.\cite{CK}. The difference between such data and a Runge-Kutta
method is that we now have translation invariance available.
Exactly as the Runge-Kutta method was producing an element in the
Butcher group, the above data give a homomorphism from the
underlying group (by the Milnor-Moore theorem) to the Butcher
group, a situation which is dual to our theorem on the
universality of ${\cal H}_R$.
We can now reformulate the ODE in general as an equation for $y\in
B\otimes E$,
\[
y=1\otimes\eta_0+(L\otimes id)f(y),
\]
where $\eta_0\in E$ is the initial datum and $f(y)$ is uniquely
defined by $f(y)(x)=f(y(x))$ for any $x\in spec(B)$.
This is very suggestive: somehow the usual solution curve for
an ODE should not be considered as an ultimate solution and
the universal problem should be thought of as a refinement of the
idea of a scalar time parameter. We regard Butcher's work on the
classification of numerical integration methods as an impressive
example that concrete problem-oriented work can lead to far
reaching conceptual results.
\subsection{Iterated integrals and numbers from primitive
diagrams} The circle of ideas described so far certainly allows us
to come to terms with the combinatorics of the subtraction procedure
so that we can now concentrate only on those diagrams without
subdivergences, i.e.~the decorations at vertices of rooted trees.
The proper definition for the integral in the new calculus used in
Noncommutative Geometry is the Dixmier trace, i.e. the invariant
coefficient of the logarithmic divergence of an operator trace. A
superficial understanding of QFT would lead one to consider it as
far too limited a tool to confront the divergences of QFT. This
misgiving is based on the ignorance of the plain fact that the
divergence of a sub\-di\-ver\-gence-free diagram is a mere overall
logarithmic divergence, and that such diagrams, for which
Fig.(\ref{F2}) gives archetypical examples, appear at all
loop orders (the expert will notice that the decorations which are
to be used are actually Feynman graphs with appropriate
polynomial insertions which are the remainders when we shrink
subgraphs to points). \bookfig{qed}{!}{F2}{Diagrams free of
subdivergences and hence delivering a first order pole in
dimensional regularization. They are typically reflecting the
skeleton expansion of the theory.}{2.3}
In di\-men\-sio\-nal re\-gu\-la\-ri\-za\-tion the
de\-co\-ra\-tions appearing at the roo\-ted trees deliver a
first-order pole in $D-4$. In other words, the Dixmier trace, or
rather its residue guise $Res_-$, is perfectly sufficient to
disentangle and determine a general Feynman graph $G$, making full
use of the decomposition dictated by the Hopf algebra structure,
which gives the joblist \cite{BK} for the practitioner of QFT: the
list of diagrams which correspond to analytic expressions
suffering from merely an overall divergence.
Analyzing the corresponding residues of the decorations and their
number-theoretic properties is a far reaching subject (discussed
in part by M.~Kontsevich in this issue) which by itself reveals
some interesting Hopf algebra structures, based on the algebraic
properties of iterated integrals.
Iterated integrals came to prominence with the work of K.T.~Chen \cite{chen}.
Essentially, they are governed by two rules,
Chen's lemma
\[
F_{r,t}^I=F_{r,s}^I+F_{s,t}^I+\sum_{I=(I^\prime,I^{\prime\prime})}
F_{r,s}^{I^\prime}F_{s,t}^{I^{\prime\prime}}
\]
and the shuffle product
\[
F_{r,s}^{I^\prime}F_{r,s}^{I^{\prime\prime}}=\sum_\sigma
F_{r,s}^{\sigma[I]},
\]
where the sum is over all $(p,q)$ shuffles of the symmetric group
$S_{p+q}$ acting on a string $I$ of, say, $n=p+q$ integers
parametrizing the one-forms which give the iterated integral as
their integral over the $n$-dimensional standard simplex.
If we define $t^I$ to be a tree without sidebranches decorated
with these one-forms in the appropriate order, then each iterated
integral corresponds to a map from such a decorated tree to a real
number, so that Chen's lemma expresses the familiar group law of
the Butcher group in this special case. It is then the shuffle
product which guarantees that iterating one-forms in accordance
with a general rooted tree \cite{dchen} will not produce anything
new: any such integral is a linear combination of the standard
iterated integrals.
Most interestingly, the calculus of full perturbative QFT can be
understood as a calculus of generalized iterated integrals, where
the boundaries $r,s$ are to be replaced by elements of the Butcher
group: scalar functions of (decorated) rooted trees \cite{dchen}.
The group law still holds naturally for such generalized
integrals, but the shuffle product only holds for the leading term
in the asymptotic expansion of bare diagrams \cite{DelKr}.
We hope that an investigation of this situation has far reaching
consequences for the understanding of the number-theoretic content
of Feynman diagrams, whose richness is underwritten by a wealth of
empirical data which provide a plethora of interesting numbers
like multiple zeta values \cite{zagier}, their alternating cousins
\cite{BK15}, and even sums involving non-trivial roots of unity in
numerators \cite{David1}.
It is gratifying then that the regularization of iterated
integrals based on forms $dx/x$, $dx/(1-x)$, providing the
iterated integrals representation of multiple zeta values, is in
full accord with the renormalization picture developed here. The
appearance of such numbers in the solution of the
Knizhnik-Zamolodchikov equation, and the relation to the Drinfel'd
associator upon renormalizing this solution so that it extends to
the unit interval, clearly motivates one to investigate the
differential equations which encapsulate the iterative structure
of Green functions in perturbative QFT, the bare and renormalized
Schwinger-Dyson equations.
Before we close this survey with a couple of remarks concerning
these equations, let us mention yet another application where the
freedom of having tree-indexed scales, hence elements of the
Butcher group, will prove essential.
The formulation of operator product expansion clearly takes place
in a configuration space of the same nature as the Fulton-MacPherson
compactification \cite{FMPh}, known to be stratified by
rooted trees. However, there is a very important nuance
which can already be fully appreciated in the case of two points.
In that case, the Fulton-MacPherson space is simply the blow-up of
the diagonal in the space $X\times X$, whereas the geometric data
which encodes most of the semiclassical deformation aspect in a
diffeomorphism invariant manner are provided by
a smooth groupoid, called the
tangent groupoid in \cite{alain3}.
Essentially, the relation between the latter and the former is the
same as the relation between a linear space and the corresponding
projective space. That the transition to a linear space is a
crucial improvement can be appreciated from the fact that it is
only in the linear space that Fourier transform takes its full
power.
It is thus of great interest to extend the construction of the
tangent groupoid from the two-point case to the full set-up of the
configurations of $n$ points. This clearly involves the freedom
of having scales available for any strata in the compactification,
hence again elements of the Butcher group.
This is in full accord with the
use of tree-indexed parameters in the momentum space description
of the operator product expansion undertaken in \cite{dchen},
which clearly shows the necessity of allowing for the full linear
space to be able to describe the variety of renormalization
prescriptions employed by the practitioners of QFT.
\subsection{The Schwinger-Dyson Equations}
Recalling that equation (\ref{E1}) demands an understanding of the
full dress\-ed line element (cf.~Fig.( \ref{dress})) , we finally
consider Schwinger-Dyson equations. The propagator, the vertex and
the kernel functions provide a system of such Schwinger-Dyson
equations, whose solution is at the heart of any Lagrangian QFT
\cite{BjD}. Though the fermion propagator, hence the line element,
comes to prominence in these equations due to gauge covariance, a
full solution is not yet known for any QFT of interest.
In the simplest form, the Schwinger-Dyson equation is typically
an equation of the form
\[ \Gamma(q)=\gamma+\int d^4k \Gamma(k)P(k)^2K(k,q),
\]
which is the structure of the Schwinger-Dyson
equation for the QED vertex at zero momentum transfer. Here, $P$
is a propagator, hence a geometric series in a self-energy (so
that the resulting equations are highly non-linear) and $K$ is a
QED four-point kernel function, actually one which is a typical
generator (in its undressed form) of decorations upon closure of
two of its legs.
Nevertheless it is not difficult to identify an operator $L_K$
such that the solution is the formal series
\[
\Gamma(q)=\gamma+\sum_i L_K^i(\gamma),
\]
where the operator $L_K$ amounts to the operator $B_+$ combined
with all decorations provided by the kernel $K$. If the above
Schwinger-Dyson equation is for the bare vertex, the one for the
renormalized vertex ${\bf\Gamma}$ is obtained by multiplying it with
the counterterm $Z$ which amounts, in full accordance with the
$g_o^{-1}g$ notation used before, to the equation
\[
{\bf \Gamma}(q)=Z\gamma+\int d^4k {\bf \Gamma}(k)P(k)^2K(k,q),
\]
which neatly summarizes the Hopf algebra structure of
perturbative QFT. From the above structure,
we conclude that Runge-Kutta methods are fully available
for this system.
Let us close this paper by noting an amusing coincidence: If we
restrict the kernel $K$ to the first two terms $K^{[1]},K^{[2]}$
in its skeleton expansion, restrict $P$ to the free propagators
and label each of the two terms $K^{[1]},K^{[2]}$ by noncommuting
variables $k_1$ and $k_2$ say, then we are naturally led to the
equation
\[
\Gamma(q)= \gamma+k_1\left[\int d^4k
\frac{1}{k^4}\Gamma(k)K^{[1]}\right] +k_2\left[\int d^4k
\frac{1}{k^4}\Gamma(k)K^{[2]}\right],
\]
which obviously involves a sum over all words in $k_1,k_2$ in its
solution and is fascinatingly close to the K-Z equation in two
variables considered before. \section*{Acknowledgements}
D.K.~thanks the IHES for extended hospitality and marvelous
working conditions. D.K.~also thanks the DFG for support as a
Heisenberg fellow.
|
\section*{Motivation}
Current physical theory predicts that at small scales the
conventional picture of spacetime as a 4-dimensional differential
manifold breaks down to something more discrete, finitary and
quantum. This inadequacy of the smooth spacetime manifold is on one
hand due to the ideal character of event determinations of a
classical observer, on the other due to the appearance of
singularities.
To deal with the first shortcoming of the manifold model,
we insist that realistic models of measurement should be pragmatic:
we actually perform a finite number of observations and record a
finite number of events. Thus, the conventional infinitude of
events that we adopt to model classical spacetime structure seems
to be a gross generalization of little operational value: we have
no actual experience of a continuous infinity of events and
their infinitesimal differential separation can not be recorded
in the laboratory.
However, due to the success of the classical model of observation
at large scales one expects a connection between the
realistic models and their ideal counterpart. The anticipated
connection could be formulated in terms of a correspondence
principle. That is to say, the structure of ideal observations
arises in some kind of limit of the structures of pragmatic
measurements. The aim of this paper is to provide a physical
account for this correspondence.
The central object of the correspondence will be the {\em classical
event} living in the limit manifold. This is modelled by a point
there and a classical observer records no quantum interference of
events. But in the pragmatic regime we interpret the event as a
{\em pure state of spacetime} and we admit coherent quantum
superposition between events. In this sense
the quantum substratum of pragmatic events 'decoheres' to the classical
point event of the limit manifold and the physical meaning of
the correspondence principle is the usual quantum one due to
Bohr.
We use the finitary substitutes proposed by Sorkin (1991) to model
combinatorial relations between events in realistic measurements.
The incidence algebras due to Rota (1968) are employed to
accomplish the same thing but operationally, that is,
algebraically. There is a duality implicit here that is pregnant to
familiar notions about the duality of quantum dynamics. In our
treatment the Sorkin model is held to represent an evolution of
states much like the Schr\"odinger picture of quantum systems,
while the Rota model recalls the evolution of operators similar to
the Heisenberg picture. Our approach lies with
the latter picture although one is always able to switch
back to the Sorkin scheme (Zapatrin, 1998). In brief, we propose
that posets describe the dynamical evolution of events when the
algebras describe the dynamical evolution of event determinations
(operations).
Our algebraic approach is constructive, that is, we provide a
matrix representation for the algebras employed. The latter possess
preferred elements that represent the pragmatic observations that
in the ideal limit are expected to yield the irreducible elements
of classical observations: the manifold point events. They
constitute abelian subalgebras of the incidence algebras and are
coined {\em stationaries}. Of course, actual pragmatic observations
are expected to effect dynamical transitions between quantum states
of spacetime (stationaries). These are modelled by non-commuting
operations in the algebra of pragmatic events and are called {\em
transients}. We anticipate that stationaries will correspond to
classical point events in the limit manifold (in Section \ref{s4} we
show how sequences of station{\it a}ries become station{\it e}ry
recording classical events), while transients to some kind of tangent
structure at an event.
To deal with the second shortcoming of the manifold model we note
that at the pragmatic level of observations there are no
points, but only algebras. We call this feature {\em
alocality}. Nevertheless, we require the classical correspondence
limit to yield the familiar local structure of spacetime: the point
event and the space tangent to it. Since any point of the limit
manifold can be the host of a singularity of some important
physical field, the alocal quantum pragmatic substratum may prove
to be an effective resolution of spacetime into finite quantum
elements. The quantum substratum is asingular (finite) because
alocal. Thus, from our perspective, locality, that is, the
assumption of a differential continuum for spacetime (Einstein,
1924), is the prime reason for
singularities, so that at the pragmatic level of observation we
abandon it. We suggest a plausible quantum theory of spacetime
structure with a strong operational and finitistic character.
Then, based on the algebraic models of pragmatic observations we
may develop a non-commmutive differential geometry to erect a
quantum theory of gravity on it (Parfionov and Zapatrin, 1995).
The correspondence limit suggested in the present paper may also
be employed in this context to recover the classical algebra of
spacetime observables and the conventional differential geometry
of the spacetime manifold (used to describe general relativity)
from the pragmatic non-commutative quantum substratum.
It should be mentioned that our algebraic approach is rather
flexible in the following sense: alternatively to the novel notion
of alocality in the pragmatic regime we can formulate
the notion of {\em nearest neighbour} connections. The
latter was assumed by Finkelstein (1985)
to be the principal characteristic of the physical causal topology in
the quantum deep so as to localise in some sense a causality relation
between events (Bombelli {\em et al.}, 1987). This causality relation
was modelled by a partial order. Thus, if we physically interpret
Sorkin's finitary substitutes as causal sets (Sorkin, 1995), a recent
result (Zapatrin, 1998) allows us to represent the 'nearest neighbour'
causal connection between events algebraically in the pragmatic
regime thus vindicate Finkelstein's demand for an algebraic
representation of immediate causality (Finkelstein, 1985). There, in
turn, we have the advantage of interpreting this connection
operationally and study its quantum properties. The question we are
confronted with is: what is the physical meaning in the pragmatic
algebraic regime of the Sorkin--Finkelstein local causality? As an
answer we expect a formulation of local causality in operational terms
with a quantum interpretation, something that is missing in Sorkin's
picture which is dual to ours. Affine to this question is the
following one: how our pragmatic event determinations accord with and
be adequate models of the causal structure of the world at small
scales?
Finally, inspired by the Sorkin (1991) approach, we contend that
pragmatic measurements can be subjected to refinements. In passing
to the dual picture we deal with algebras and, in accordance with
the correspondence limit, the ideal ultimate refinement corresponds
to what is known as the algebra of classical observables
(coordinates) and the manifold supporting them.
\section{Finitary preliminaries}\label{s1}
A finitary topological space is defined in (Sorkin, 1991) as a
space with any bounded region in it consisting of a finite number
of points. This seems to be a reasonable model for actual
measurements involving a finite number of events during experiments
of finite spatiotemporal extent.
Any finitary topological space $\mmm$ can be equivalently pictured as
poset. Introduce the relation "$\to$" between points of $\mmm$
\[
p\to q \quad \Leftrightarrow \quad \mbox{ the constant sequence }
\{p,p,\ldots,p,\ldots\} \quad \mbox{ tends to } q
\]
\noindent using the standard definition of convergence: a sequence
$\{p_1,p_2,\ldots \}\to q$ iff for any open set $U$ containing $q$
there is a number $N_U$ such that $p_n\in U$ for any $n\ge N_U$.
The obtained relation "$\to$" is always reflexive ($p\to p$) and
transitive ($p\to q$, $q\to r$ imply $p\to r$). Vice versa, any
quasiordered set $(\mmm, \to)$ acquires a topology defined through the
closure operator on subsets $P\subseteq M$:
\[
{\rm\bf Cl} P =
\{q\,:\, \exists p \in P \quad p\to q \}
\]
For technical reasons (see Section \ref{s2}) we employ the
Alexandrov (1956) construction of {\em nerves} to substitute the
continuous topology. Recall that the nerve ${\cal K}$ of a covering
$\uuu$ of a manifold $\mmm$ is the simplicial complex whose
vertices are the elements of $\uuu$ and whose simplices are formed
according to the following rule. A set of vertices (that is,
elements of the covering) $\{U_0, \ldots , U_k\}$ form a
$k$-simplex of ${\cal K}$ if and only if they have nonempty
intersection:
\[
\{U_0, \ldots , U_k\} \in {\cal K} \: \Leftrightarrow \:
U_0 \cap U_1 \cap \ldots \cap U_k \neq \emptyset
\]
\noindent Any nerve ${\cal K}$ being a simplex can be as well treated
as a poset, denoted also by ${\cal K}$. The points of the poset ${\cal K}$
are the simplices of the complex ${\cal K}$, and the arrows are drawn
according to the rule:
\[
p \to q \: \Leftrightarrow \: p \mbox{ is a face of } q
\]
\medskip
In the nondegenerate cases the posets associated with nerves and
those produced by Sorkin's (1991) `equivalence algorithm' are the
same. We choose nerves because their specific algebraic structure
makes it possible to build the dual algebraic theory.
\section{Incidence algebras}\label{s2}
The notion of incidence algebra of a poset was introduced by Rota
(1968) in a purely combinatorial context. Let $P$ be a poset.
Consider the set of formal symbols $\ketbra{p}{q}$ for all $p, q\in
P$ such that $p\le q$ and its linear span
\be{e38a}
\Omega = {\rm span}\{\ketbra{p}{q}\}_{p\le q}
\end{equation}
\noindent and endow it with the operation of multiplication
\be{38}
\ketbra{p}{q} \cdot \ketbra{r}{s} =
\ket{p} \braket{q}{r} \bra{s} =
\braket{q}{r} \cdot \ketbra{p}{s} =
\left\lbrace \begin{array}{rcl}
\ketbra{p}{s} &,& \mbox{if } q=r \cr
0 && \mbox{otherwise}
\end{array} \right.
\end{equation}
The correctness of this definition of the product, that is, the
existence of $\ketbra{p}{s}$ when $q=r$ is due to the transitivity
of the partial order. The obtained algebra $\Omega$ with the
product (\ref{38}) is called {\em incidence algebra} of the poset
$P$.
The incidence algebra $\Omega$ is obviously associative, but not
commutative in general. Namely, it is commutative if and only if
the poset $\ppp$ contains no arrows.
Let us split $\Omega$ into two subspaces
\[
\Omega = {\cal A} \oplus {\cal R}
\]
\noindent where
\be{e37a}
{\cal A} = {\rm span}\{\ketbra{p}{p}\}_{p\in P}
\end{equation}
\noindent and call
\[
{\cal R} = {\rm span} \{ \ketbra{p}{q} \}_{p<q}
\]
\noindent the {\em module of differentials} of the poset $P$. It is
a fact that ${\cal R}$ is a bimodule over ${\cal A}$.
As we refine the poset, the limit space is intended to be a
manifold. The incidence algebras are dual objects to posets,
therefore their behavior should be similar to that of differential
forms in classical geometry. The algebra ${\cal A}$ is intended to play
the r\^ole of classical coordinates, while ${\cal R}$ should be graded
being an analogue of the module of differential forms.
For this aim we consider only simplicial complexes which are
treated as posets. $p\le q$ means that $p$ is a face of $q$. The
elements of simplicial comlexes are naturally graded. Then
any basic element $\ketbra{p}{q}$ of the incidence algebra $\Omega$
acquires a degree being the difference of the degrees of its
constituents:
\be{e40a}
{\rm deg}\ketbra{p}{q} = \mbox{'the difference of cardinalities of
$p$ and $q$'}
\end{equation}
\noindent splitting $\Omega$ into linear subspaces
\be{40}
\Omega = \Omega^0 \oplus \Omega^1 \oplus \ldots
\end{equation}
\noindent with
\[
\begin{array}{rcl}
\Omega^0 &=& {\rm span}\{\ketbra{p}{p}\} \: = \: {\cal A} \cr
\ldots & \ldots & \ldots \cr
\Omega^n &=& {\rm span}\{\ketbra{p}{q}\}_{\deg\ketbra{p}{q} = n} \cr
\ldots & \ldots & \ldots
\end{array}
\]
\noindent making $\Omega$ graded algebra:
\[
\forall \omega \in \Omega^m, \omega' \in \Omega^n \qquad
\omega\om' \in \Omega^{m+n}
\]
\noindent and therefore making the module of differentials
${\cal R}$ graded ${\cal A}$-bimodule:
\[
{\cal R} = \Omega^1 \oplus \Omega^2 \oplus \ldots
\]
This grading makes the incidence algebras discrete differential
manifolds (Dimakis and M\"uller-Hoissen, 1999) as they posssess both
commutative scalars (the subalgebra ${\cal A}$) and differentials over it
(the module ${\cal R}$). For a more detailed account the reader is referred
to (Breslav and Zapatrin, 1999).
\section{Rota topology and the duality}\label{s2a}
In this section we establish a duality between a certain class of
finitary substitutes and their incidence algebras. We select
this class in such a way that canonical mappings between the
points admit conjugate homomorphisms of incidence algebras
making the correspondence between posets and algebras
functorial.
As it was shown in the previous section, with any poset its
non-commutative incidence algebra can be associated. It was proved by
Stanley (1968) that if two posets have isomorphic incidence algebras
then they are isomorphic. The reverse procedure building a poset
$P(\Omega)$ from an arbitrary finite-dimensional algebra $\Omega$ was
suggested in (Zapatrin, 1998). Let us briefly describe the
construction.
The elements of the poset $P(\Omega)$ are the irreducible representations
of the algebra $\Omega$. Building the partial order on $P(\Omega)$ cosists
of two steps. First, the nearest neighbour connections $p\to q$ are
built according to the following rule: let $p,q$ are two irreducible
representations of $\Omega$, denote by $p^0,q^0$ their kernels:
\[
p^0 = p^{-1}(0) \quad ; \quad q^0 = q^{-1}(0)
\]
\noindent which are ideals in $\Omega$. Then define the nearest
neighbours $p\to q$:
\be{e93}
p\to q \quad \Leftrightarrow \quad
p^0q^0 \neq p^0\cap q^0
\end{equation}
\noindent where the left-hand side $p^0q^0$ is understood as the
product of subsets of $\Omega$. The resulting partial order on the set
$P(\Omega)$ is obtained as the transitive closure of the relation
(\ref{e93}). The topology associated with this partial order is
referred to as {\em Rota topology}.
When the algebra $\Omega$ is commutative, the Rota topology is discrete
(no linked pairs $p\to q$). The obtained topology becomes non-trivial
only when $\Omega$ is non-commutative.
\paragraph{Remark.} When all irreducible representations of $\Omega$ are
one-dimensional we can build two topologies on the set $P(\Omega)$
Gel'fand and Rota ones, and it is interesting to compare them. The
result is the following: the Gel'fand topology is always discrete,
while the Rota topology may be non-trivial.
The possibility of mutual transitions between between finitary
topological spaces and algebras is based on the following theorem
(Zapatrin, 1998):
{\em
If the algebra $\Omega$ is the incidence algebra $\Omega(\ppp)$ of a poset
$\ppp$ then the resulting poset is isomorphic to $\ppp$:
}
\[
\ppp \simeq \ppp(\Omega(\ppp))
\]
As it was mentioned, Stanley (1968) theorem claims that
\[
\Omega(\ppp) \simeq \Omega(\qqq)
\Leftrightarrow
\ppp \simeq \qqq
\]
\noindent and one could expect that a poset homomorphism, that
is, a continuous mapping of appropriate finitary topological
spaces, should give rise to a homomorphism of their incidence
algebras. Alas, this is not the case for general posets $\ldots$
To gather functoriality we have to restrict somehow the class of
posets we are dealing with and the mappings between them. We did
it already in the previous section in order to make the
incidence algebras graded. Namely, we restricted ourselves to
simplicial complexes. To make incidence algebras dual objects,
we, following Alexandrov (1956), restrict the mappings between
simplicial complexes to simplicial mappings only. Recall that a
mapping $\omega: {\cal K}_\alpha \to {\cal K}_{\alpha'}$ between two
simplicial complexes ${\cal K}_\alpha$ and ${\cal K}_{\alpha'}$ is said
to be simplicial if
\begin{itemize}
\item the $\omega$-image of any vertex in ${\cal K}_\alpha$ is a vertex
in ${\cal K}_{\alpha'}$
\be{e77}
\omega({\cal K}^0_\alpha) \subseteq {\cal K}^0_{\alpha'}
\end{equation}
\item $\omega$ is completely defined by its values on the vertices
of ${\cal K}_\alpha$.
\item $\omega$ preserves simplices
\end{itemize}
Under this condition the correspondence between posets and their
incidence algebras becomes functorial. With any $\omega:
{\cal K}_\alpha \to {\cal K}_{\alpha'}$ its adjoint $\omega^*:
\Omega({\cal K}_{\alpha'}) \to \Omega({\cal K}_\alpha)$ is defined in the
following way. Let $\ketbra{P}{Q}$ be a basic element of
$\Omega({\cal K}_{\alpha'})$. Then $Q$ may be represented as a
disjoint sum $Q = P+S$ and we put
\be{e79}
\omega^*(\ketbra{P}{Q}) =
\sum\{ \ketbra{p}{q} \:: \omega(p) = P;\, q=p+s,\, \omega(s)=S \}
\end{equation}
The direct verification shows that the so-defined $\omega^*$ is
always a homomorphism of the incidence algebras. As a result, we
have the duality between simplicial complexes and their
incidence algebras: if $\omega$ is surjective its adjoint $\omega^*$
is injective and vice versa.
\section{Physical interpretation of incidence algebras}\label{s3}
The differential manifold model of spacetime is an ill-founded
assumption and a gross generalisation of what we actually
experience as spacetime. It is essentially based on the
non-operational supposition that we can pack an infinity of
events in an infinitesimal spacetime volume element when, in
fact, we only record a finite number of them during experiments
of finite duration in laboratories of finite size. It is exactly
due to this trait of the manifold model that at small scales our
theories of quantum spacetime structure and dynamics are plagued
by non-renormalizable infinities of the values of many important
physical fields. On the other hand, the requirement that the laws
of nature are local almost mandates the assumption of smoothness
for spacetime and we seem to get back to square one. However, the
success the manifold has enjoyed in picturing the local dynamics of
matter should not mask the unphysicality of its character,
especially at small scales. In particular, quantum theory, when
applied to investigate the structure and dynamics of spacetime in
the small, is simply incompatible with a classical,
non-operational ideal of a continuous infinity of events labelled
by commutative coordinates. Pragmatic measurements of quantum
spacetime are finite and inevitably induce uncontrollable
dynamical perturbations to it. Thus, the requirement for
operationality and finiteness as well as the success that a quantum
theory of matter has had when formulated algebraically motivate us
to formulate an algebraic theory of pragmatic finite measurements
of spacetime at quantum scales.
The local structure of the differential manifold is the point event
and its infinitesimal differential neighbours in the tangent space.
As mentioned above, this classical geometric structure serves us
well in casting dynamical laws as differential equations (classical
locality), but is rather inadequate for picturing actual quantum
spacetime measurements that are finite hence free from infinities
(singularities). Especially in the quantum deep this classical
conception of locality can not survive. We propose to revise it by
substituting the geometrical point events and the space of
directions tangent to them by finitely generated algebras affording
a cogent quantum spacetime interpretation for their structure. In
this sense our scheme is alocal and finitary and more likely to
evade the infinities of the differential manifold. Of course, the
'naturalness' of our substituting quantum alocality for the
classical differential locality will be vindicated if we are able
to recover the limit manifold with its classical observables and
differential structure by some kind of correspondence principle
applied to the alocal algebras of pragmatic measurements. We carry
this out in the section \ref{s4}.
We give the following physical meaning to the elements of $\Omega$ in
(\ref{e37a}):
\begin{enumerate}
\item ${\cal A}$ constitutes the space of stationaries. The latter can
be thought of as elementary acts of determination of the pure
states of quantum spacetime. We interpret them as quantum
spacetime events. The algebraic connective '$+$' between them is
interpreted as coherent superposition between quantum events. The
commutativity of stationaries foreshadows the compatibility of the
determinations of the coordinates of events in the classical
manifold regime (Section \ref{s4}). In the dual (poset) picture the
sationaries correspond to self-incidences $p\to p$.
\item $\Omega^1$ constitutes the space of transients. These can be
thought of as elementary quantum dynamical processes between
stationaries, thus they represent discrete one-step transitions
between quantum spacetime events. Transients do not commute with
each other and this foreshadows the Lie structure of covectors in
the limit space.
\item $\Omega^i$ ($i\ge 2$) constitute the spaces of paths which are
thought of as composites of transients. If we associate with a
transient a quantum of an additive physical
quantity like energy (or its dual time), then the total grade of the
appropriate element of the algebra corresponds to the total energy
associated with it (or to the duration of the whole transition process).
\end{enumerate}
In the Motivation we alluded to the Sorkin poset scheme as being an
analog of the Schr\"odinger picture of quantum dynamics while our
algebraic approach as being the simile of the Heisenberg picture:
this is based on the duality of the two approaches (Section
\ref{s2a}). In an analogous way quantum states are the linear duals
of the operators in the conventional algebraic approach to quantum
mechanics.
Here too any finitary substitute is associated with an incidence
algebra in such a way that the topology of the poset is the same as
that encoded in the algebra. This resembles the fact that the
Schr\"odinger and the Heisenberg pictures encode the same
information about quantum dynamics. Furthermore, the arrows
between point events in the Sorkin scheme can be thought of as the
directed dynamical transitions of spacetime event-states while in
our picture such dynamical connections are between pragmatic
operations. The topology in both schemes is physically interpreted
as dynamical connections between events although our picture being
algebraic naturally affords a quantum interpretation.
\section{Limiting procedure and the correspondence principle
}\label{s4}
When spacetimes are subsituted by finitary topological spaces,
we may consider finer or coarser experiments. That is why we
have to formalize the notion of refined experiment. Within the
Sorkin discretization procedure (Section \ref{s1}) a refinement
means passing to an inscribed covering of the manifold. In this
case any element of the finer covering is contained in an
element of the coarser one. Since we are dealing with nerves and
simplicial mappings between them we have to take care of the
condition (\ref{e77}). Recall that a vertex of the nerve is
associated with an element of the covering. In general it
may happen that a small region of the fine covering can belong
to two elements of a coarser one. So, we have have to require
for any element of the fine covering to keep track of its origin
in order for (\ref{e77}) to hold.
Each step of a limiting procedure, that is, a refined covering,
gives rise to a projection of appropriate complexes: the finer
one is projected to the coarser one. In the dual framework
we have an injection of the smaller algebra associated with a
coarser measurement to the bigger one.
In general, limiting procedures for approximating systems (whatever
they be, posets or algebras) are organised using the notion of
converging nets. Namely, each pragmatic observation is labelled by
an index $\alpha$ and we have the relation of refinement $\succ$ on
observations: $\alpha \succ \alpha'$ means that the observation
$\alpha$ is a refinement of $\alpha'$.
When we are dealing with posets with each pair $\alpha,\alpha'$
such that $\alpha \succ \alpha'$ a canonical projection
$\omega_{\alpha'}^\alpha : {\cal K}_\alpha \to {\cal K}_{\alpha'}$ is defined
such that for any $\alpha \succ \alpha' \succ \alpha''$
\[
\omega_{\alpha''}^\alpha =
\omega_{\alpha''}^{\alpha'}\omega_{\alpha'}^\alpha
\]
\noindent and we introduce the set of threads. A thread is a
collection $\{t_\alpha\}$ of elements $t_\alpha \in {\cal K}_\alpha$
such that
\[
t_{\alpha} = \omega_{\alpha}^{\alpha'} t_{\alpha'}
\]
\noindent whenever $\alpha \succ \alpha'$. Denote by $\tttt$ the
set of all threads.
The next step is to make $\tttt$ a topological space which is done
in a standard way (Alexandrov, 1956): $\tttt$ is a subspace of the
total cartesian product
\[
\tttt_0 = \times_\alpha {\cal K}_\alpha
\]
\noindent while each of ${\cal K}_\alpha$ is a topological space.
Endow $\tttt_0$ with the product Tikhonov topology, then $\tttt$
being a subset of $\tttt_0$ becomes topological space.
Finally we obtain the limit space $X$ as the collection of all
closed threads from $\tttt$. This procedure is described in detail
in (Sorkin, 1991).
\medskip
The scheme for building limit algebras is exposed in (Landi and
Lizzi, 1998). As mentioned above, with any pair $\alpha \succ
\alpha'$ of pragmatic observations we have a canonical injection
\[
\omega^{*\,\alpha'}_{\alpha} : \Omega({\cal K}_{\alpha'}) \to
\Omega({\cal K}_{\alpha})
\]
Moreover, due to the requirement (\ref{e77}) the restriction of
each $\omega^*$ on commutative subalgebras ${\cal A} = \Omega^0 \subseteq
\Omega$ is well defined. Now we first consider the set of all
sequences
\[
{\bf \Omega} = \times_\alpha \Omega({\cal K}_\alpha) =
\{ \{a_\alpha\}\,|\, a_\alpha \in \Omega({\cal K}_\alpha) \}
\]
\noindent and select the set of {\em converging} sequences in the
following way. Note that ${\cal A}$ is an algebra. Introduce a norm
$||\cdot||_\alpha$ in each finite-dimensional algebra
$\Omega({\cal K}_\alpha)$, then a sequence $\{a_\alpha\}$ converges if
and only if for any $\epsilon > 0$ there exists a filter
$\fff_\epsilon$ of indices $\alpha$ such that
\[
\forall \alpha, \alpha'\in \fff_\epsilon
\quad \alpha \succ \alpha' \Rightarrow
||\omega^{*\,\alpha}_{\alpha'}a_\alpha - a_{\alpha'}||_{\alpha'} <
\epsilon
\]
Since any element of the limit algebra is a net we may consider the
coupling between the limit algebra and the limit space which
consists of nets. The result of this coupling is a converging net
of numbers whose limit is thought of as the value of an element of
the limit algebra at a point of the limit space.
The Sorkin scheme recovers the manifold in the limit of
refinements of finitary posets. Our dual picture aspires to the
same in the limit of resolution of pragmatic event
determinations. Since our algebraic scheme affords a quantum
spacetime interpretation, this limit can be thought of as a
correspondence principle linking the finitary quantum spacetime
substrata with the smooth classical spacetime manifold. The
alocal, algebraic quantum spacetime determinations of the substrata
converge to the local geometric spacetime point and its cotangent
space. This is to be contrasted for instance with the Bombelli {\em et
al.} (1987) causal set scenario where the limiting procedure may be
thought of as a `random sprinkling' of events according to some
appropriate distribution so that the `limit spacetime manifold', with
its topological, differential and Lorentz-causal structure, arises as a
statistical average of causal sets, thus it is essentially of {\em
thermodynamic} nature.
On the other hand, our correspondence limit is well-defined in the
{\em quantum} (rather than statistical) sense as the well-known
correspondence principle: the pragmatic quantum stationaries
decohere to the point events of the limit manifold, while the
non-commuting transients to covectors.
\section*{Concluding remarks}
In the present paper we gave quantum spacetime interpretation to
the incidence algebras induced by posets which, in turn,
correspond to finitary topological spaces. Sorkin's limit for
recovering the manifold as a maximal refinement of finitary
posets is cast here as Bohr's correspondence principle. Still,
due to the implausibility of any notion of pre-existing space in
the quantum dynamical deep, we would rather give a more
physical, causal or temporal interpretation to the posets'
partial order (Sorkin, 1995), so that we can link our algebraic scheme
with Bombelli {\em et al.} (1987) causal set approach to quantum
gravity. Our quantum interpretation of the incidence algebras
induced by causal sets is a first step into yet another attempt
at quantizing causality (Finkelstein, 1969). It is one of the
authors' previous result (Zapatrin, 1998) and Finkelstein's
(1985) claim for immediate causal links between
events to represent the physical causal topology that caught our
attention and motivated us to try to link the present work with causal
sets. This project, however, is still at its birth.
\paragraph{Acknowledgments.} The work was carried out within the
program "T\^ete-\`a-t\^ete in St.Petersburg" under the auspices of the
Euler Mathematical Institute. The authors express their
gratitude to the participants of the Friedmann seminar
(St.Petersburg, Russia) for their kind suggestions.
One of the authors (IR) wholeheartedly acknowledges
numerous exchanges on quantum spacetime structures with Professor
Anastasios Mallios. The second author (RRZ) acknowledges the
support from the RFFI research grant and the research grant
"Universities of Russia".
\medskip
\noindent {\bf References}
Aleksandrov, P.S. (1956).
{\it Combinatorial Topology},
Greylock, Rochester, New York
\smallskip
Bombelli L., Lee J., Meyer D., and Sorkin R.D. (1987).
Spacetime as a causal set,
{\it Physical Review Letters},
{\bf 59}, 521
\smallskip
Breslav, R., and R.R.Zapatrin (1999).
Differential structure of Greechie logics,
{\it International Journal of Theoretical Physics},
submitted, eprint quant-ph/9903011
Dimakis, A., and F.M\"uller-Hoissen (1999).
Discrete Riemannian geometry,
{\it Journal of Mathematical Physics},
{\bf 40}, 1518
\smallskip
Einstein, A. (1924).
\"Uber den \"Ather,
{\it Schweizerische naturforschende Gesellschaft Verhanflungen},
{\bf 105}, 85-93 (English translation by Simon Saunders:
'On the Ether' in {\it The Philosophy of Vacuum},
S.Saunders and H.Brown, Eds.,
Oxford University Press (1991),
13--20)
\smallskip
Finkelstein D. (1969).
Space-time code,
{\it Physical Review},
{\bf 184}, 1261
\smallskip
Finkelstein, D. (1985).
Superconducting Causal Nets,
{\it International Journal of Theoretical Physics},
{\bf 27}, 473
\smallskip
Landi, G., and F.Lizzi (1999).
Projective Systems of Noncommutative Lattices as a Pregeometric
Substratum,
in {\it Quantum Groups and Fundamental Physical Applications},
ISI Guccia, Palermo, December 1997, D. Kastler and M. Rosso Eds.,
Nova Science Publishers, Inc.
\smallskip
Parfionov, G.N. and R.R.Zapatrin (1995).
Pointless Spaces in General Relativity,
{\it International Journal of Theoretical Physics},
{\bf 34},
737
\smallskip
Rota, G.-C., (1968).
On The Foundation Of Combinatorial Theory, I. The Theory Of
M\"obius Functions,
{\it Zetschrift f\"ur Wahrscheinlichkeitstheorie,\/}
{\bf 2},
340
\smallskip
Sorkin, R.D. (1991).
Finitary Substitute for Continuous Topology,
{\it International Journal of Theoretical Physics}, {\bf 30}, 7, 923
\smallskip
Sorkin, R.D. (1995).
A specimen of theory construction from quantum gravity,
in {\it The Creation of Ideas in Physics},
Jarret Leplin Ed.,
Kluwer Academic Publishers,
Dordrecht
\smallskip
Stanley, R.P. (1986).
{\it Enumerative Combinatorics},
Wadsworth and Brook, Monterey, California
\smallskip
Zapatrin, R.R. (1998).
Finitary Algebraic Superspace,
{\it International Journal of Theoretical Physics},
{\bf 37},
799
\smallskip
\end{document}
|
\section{Introduction}
The classical Anti-de Sitter (AdS) space can be defined as the
four-dimensional hyperboloid
$$\eta_{ab}x^ax^b=-\frac{1}{a^2}$$
in $R^5$ with Cartesian coordinates $x^a$, where $\eta_{ab}={\rm diag}(-1,~1,~
1,~1,~-1)$. In polar coordinates $y^\mu=(t,~r,~\theta,~\phi)$ the line element
may be written
$$g_{\mu\nu}dy^\mu dy^\nu=-(1+a^2r^2)dt^2+(1+a^2r^2)^{-1}dr^2+r^2(d\theta^2+
\sin^2\theta d\phi^2)~.$$
The symmetry group (isometry
group) of the classical AdS space is $SO(3,2)$, which plays the role of the
Poincar\'{e} group on Minkowski space. In fact, the Poincar\'{e} group can
be obtained from $SO(3,2)$ by a contraction.
The AdS space has rather peculiar features. Its time-like geodesics are finite
and closed. The time translations are a $U(1)$ subgroup of $SO(3,2)$. The
space-like geodesics are unbounded. Furthermore the casual structure is
somewhat complicated.
The AdS group, AdS space and quantum field theory based on it
has been an interested topic for a long time\cite{01}.
In the early 80's, there was great interest in four-dimensional $N$-extended
supergravities for which the global $SO(N)$ is promoted to a gauge
symmetry\cite{02}.
In these theories the underlying supersymmetry algebra is no longer Poincar\'{e}
but rather AdS.
An important ingredient in these developments was that the AdS$\times S^7$
geometry was not fed in by hand but resulted from a spontaneous
compactification, {\it i.e.}, the vacuum state was obtained by finding a stable
solution of the higher-dimensional field equations.
There has recently been a revival of interest in AdS space brought about by
the conjectured duality between physics in the bulk of AdS and a conformal
field theory on the boundary\cite{03}. This is one of the most important
progresses in the so-called non-perturbative superstring theory. Many new
results have been obtained by making use this conjecture\cite{04}\cite{05}.
Here the AdS geometry plays a central role.
It might seem very strange that quantum theories in different space-time
dimensions could be equivalent. This possibility is related to the fact
that the theory in the large dimension is (among other things) a quantum
theory of gravity. For such theories the concept of holography has been
introduced as a generic property. The Maldacena conjecture is an example
of realization of the holography.
According to general relativity, gravity is nothing but space-time geometry.
The poor understanding of physics at very short distances indicates that
the small scale structure of space-time might not be adequately described by
classical continuum geometry.
It is long been suspected that the noncommutivity of space-time
might be a realistic picture of how space-time behaves near
the Planck scale where strong quantum fluctuations of gravity may make points
in space fuzzy. As it is well known, superstring theory is now the only
possible candidate for quantum theory of gravity\cite{06}. The recent efforts to unify
by non-perturbative dualities all the known five perturbative superstring theory is
referred as M theory\cite{07}. Many physicists believe that the M theory is a fundamental
quantum theory in eleven dimensional space-time. The BFSS matrix
model\cite{08} is proposed for the microscopic description of M theory in discrete
light-cone quantization. The basic block in the matrix model is a set of $N$ partons, called $D_0$-branes,
on which strings can end. A novel feature of the M(atrix) theory is that the
nine transverse coordinates of the $D_0$-branes are promoted into $N\times N$
Hermitian matrices. Then the noncommutative geometry becomes a natural way
to deal with spaces whose coordinates are noncommutative.
Because the noncommutative geometry description is a strong candidate of
realistic pictures of space-time behaviors at Planck scale and the Madacena
conjecture a concrete realization of holography --the generic feature of
quantum theory of gravity, it is nature to assume that the proper geometry
for the Madacena conjecture be the noncommutative AdS space.
In this paper, we discuss the quantum AdS space and quantum AdS group by
investigating real forms of quantum orthogonal group and quantum orthogonal
space. Differential calculus on the quantum AdS space are presented. In
particular, we set up relationships between quantum group and quantum
algebra, which may be referred as the quantum counterpart of the classical
exponential. It is the relationship which deduces the quantum AdS algebra naturally
from the conjugation operation on the quantum AdS group. The presentations of
the quantum AdS algebra are foundations for constructing quantum field theory
on the quantum AdS space.
This paper is organized as follows. In Section 2, we present some preliminary
knowledge about quantum group theory and differential calculus on quantum orthogonal
space. A realization of the quantum algebra $U_q(so(5))$ in quantum orthogonal
space is shown in Section 3. In Section 4, kinds of conjugations on quantum
orthogonal group and quantum orthogonal space are constructed to get quantum
AdS group and quantum AdS space both for $\vert q\vert=1$ and $q$ being real
cases. We state the differential calculus on quantum AdS space in Section 5.
Reality of differential calculus is discussed. To deduce quantum AdS algebra
from conjugation on quantum orthogonal space and quantum orthogonal group, we set up explicit
relationships between quantum group $SO_q(5)$ and quantum algebra $U_q(so(5))$
in Section 6. Section 7 is devoted to study of quantum AdS algebra.
\section{Differential calculus on quantum space}
The quantum ${\cal R}$ matrix\cite{09} for the quantum group $SO_q(5)$ is of the form
\begin{equation}
\begin{array}{rcl}
{\cal R}&=&q \displaystyle\sum_{\stackrel{i=-2}{i\not=0}}^2 e^i_i\otimes e^i_i
+ \displaystyle\sum_{\stackrel{\stackrel{i,j=-2}{i\not=j,j'}}
{{\rm or}~i=j=0}}^2
e^i_i\otimes e^j_j
+ \displaystyle q^{-1}\sum_{\stackrel{i=-2}{i\not=0}}^2 e^{i'}_{i'}\otimes e^i_i\\
& &+ \displaystyle \lambda\left[\sum_{\stackrel{i,j=-2}{i>j}}^2 e^i_j\otimes e^j_i
-\sum_{\stackrel{i,j=-2}{i>j}}^2q^{\rho_i-\rho_j}e^i_j\otimes e^{i'}_{j'}\right]~,
\end{array}
\end{equation}
where $\lambda=q-q^{-1}$, $i'=-i$ and $\rho_i=\left(\frac{3}{2},~\frac{1}{2},~0,~-\frac{1}{2},~
-\frac{3}{2}\right)$.
By making use of the ${\cal R}$ matrix, the quantum group $SO_q(5)$ with
entries
$$T=\left(\begin{array}{ccccc}
t_{11}&t_{12}&t_{13}&t_{14}&t_{15}\\
t_{21}&t_{22}&t_{23}&t_{24}&t_{25}\\
t_{31}&t_{32}&t_{33}&t_{34}&t_{35}\\
t_{41}&t_{42}&t_{43}&t_{44}&t_{45}\\
t_{51}&t_{52}&t_{53}&t_{54}&t_{55}\end{array}\right)$$
can be written into the standard form
\begin{equation}
{\cal R}T_1T_2=T_2T_1{\cal R}~,
\end{equation}
where $T_1\equiv T\otimes 1$ and $T_2\equiv 1\otimes T$.
Of course, the $T$ matrix also should satisfy the orthogonal relations
\begin{equation}
T^tCT=C~,~~~~~~TCT^t=C~,
\end{equation}
where $C_{ij}$ is the metric on the quantum orthogonal space,
$$C=\left(\begin{array}{ccccc}
& & & &q^{-3/2}\\
& & &q^{-1/2}& \\
& &1& & \\
&q^{1/2}& & & \\
q^{3/2}& & & &\end{array}\right)~.$$
The Hopf algebra structure of the quantum orthogonal group is
\begin{equation}
\begin{array}{l}
\Delta(T)=T\dot{\otimes} T~,\\
\epsilon(T)=I~,\\
S(T)=CT^tC^{-1}=\left(\begin{array}{ccccc}
t_{55}&q^{-1}t_{45}&q^{-3/2}t_{35}&q^{-2}t_{25}&q^{-3}t_{15}\\
qt_{54}&t_{44}&q^{-1/2}t_{34}&q^{-1}t_{24}&q^{-2}t_{14}\\
q^{3/2}t_{53}&q^{1/2}t_{43}&t_{33}&q^{-1/2}t_{23}&q^{-3/2}t_{13}\\
q^2t_{52}&qt_{42}&q^{1/2}t_{32}&t_{22}&q^{-1}t_{12}\\
q^3t_{51}&q^{2}t_{41}&q^{3/2}t_{31}&qt_{21}&t_{11}
\end{array}\right)~,
\end{array}
\end{equation}
where the operation $\dot{\otimes}$ between two matrices $A$ and $B$ is defined
as$$(A\dot{\otimes}B)_{ij}=A_{ik}\otimes B_{kj}~.$$
To define differential calculus on the non-commutative algebra generated
by the coordinates ${\bf x}(=\{x^i\},~i=-2,~-1,~0,~+1,~+2)$, we write down
the $\hat{\cal R}$ ($\equiv {\cal R}P$, and $P$ a permutation operator $P:~~~A\otimes B
=B\otimes A$) matrix explicitly,
{\small
\begin{equation}
\hat{\cal R}=\left(\begin{array}{ccccccccccccccccccccccccc}
q& & & & & & & & & & & & & & & & & & & & & & & & \\
&\lambda& & & &1& & & & & & & & & & & & & & & & & & & \\
& &\lambda& & & & & & & &1& & & & & & & & & & & & & & \\
& & &\lambda& & & & & & & & & & & &1& & & & & & & & & \\
& & & &\frac{(1-\frac{1}{q^3})}{\lambda^{-1}}& & & &-\frac{\lambda}{q^2}& &
& &-\frac{\lambda}{q^{\frac{3}{2}}}& & & &-\frac{\lambda}{q}& & & &
\frac{1}{q}& & & & \\
&1& & & & & & & & & & & & & & & & & & & & & & & \\
& & & & & &q& & & & & & & & & & & & & & & & & & \\
& & & & & & &\lambda& & & &1& & & & & & & & & & & & & \\
& & & &-\frac{\lambda}{q^2}& & & &\frac{1-\frac{1}{q}}{\lambda^{-1}}&
& & &-\frac{\lambda}{q^{\frac{1}{2}}}& & & &\frac{1}{q}& & & & & & & & \\
& & & & & & & & &\lambda& & & & & & & & & & & &1& & & \\
& &1& & & & & & & & & & & & & & & & & & & & & & \\
& & & & & & &1& & & & & & & & & & & & & & & & & \\
& & & &-\frac{\lambda}{q^{\frac{3}{2}}}& & & &-\frac{\lambda}{q^{\frac{1}{2}}}&
& & &1& & & & & & & & & & & & \\
& & & & & & & & & & & & &\lambda& & & &1& & & & & & & \\
& & & & & & & & & & & & & &\lambda& & & & & & & &1& & \\
& & &1& & & & & & & & & & & & & & & & & & & & & \\
& & & &-\frac{\lambda}{q}& & & &\frac{1}{q}& & & & & & & & & & & & & & & & \\
& & & & & & & & & & & & &1& & & & & & & & & & & \\
& & & & & & & & & & & & & & & & & &q& & & & & & \\
& & & & & & & & & & & & & & & & & & &\lambda& & & &1& \\
& & & &\frac{1}{q}& & & & & & & & & & & & & & & & & & & & \\
& & & & & & & & &1& & & & & & & & & & & & & & & \\
& & & & & & & & & & & & & &1& & & & & & & & & & \\
& & & & & & & & & & & & & & & & & & &1& & & & & \\
& & & & & & & & & & & & & & & & & & & & & & & &q
\end{array}\right)~.\end{equation} }
The $\hat{\cal R}$ matrix is semisimple and has the spectral resolution
\begin{equation}
\begin{array}{l}
\hat{\cal R}=q{\cal P}_S-q^{-1}{\cal P}_A+q^{-4}{\cal P}_1~,\\
\hat{\cal R}^{-1}=q^{-1}{\cal P}_S-q{\cal P}_A+q^{4}{\cal P}_1~,\\
\end{array}
\end{equation}
where ${\cal P}_S$, ${\cal P}_A$ and ${\cal P}_1$ are projection operators
which act on the tensor product ${\bf x}\otimes {\bf x}$ of the fundamental
representation ${\bf x}$, and project into the symmetric, antisymmetric
and singlet irreducible representations with dimension $14$, $10$ and $1$,
respectively. It is convenient to give a concrete representation of the
projector,
$$\begin{array}{l}
{\cal P}_1=\frac{1-q^2}{(1-q^5)(1+q^3)}(C^{-1})^{ij}C_{kl}e^k_i\otimes e^l_j~,\\
{\cal P}_A=\frac{1}{q+q^{-1}}\left(-\hat{\cal R}+q-(q-q^{-4}){\cal P}_1\right)~,\\
{\cal P}_S=\frac{1}{q+q^{-1}}\left(\hat{\cal R}+q^{-1}-(q^{-1}+q^{-4}){\cal P}_1
\right)~.\end{array}$$
Analogous to the commutative case, we define the commutation relations of the
quantum orthogonal space by requiring the vanishing of their antisymmetric products.
Here the quantum antisymmetric products is given by the projector ${\cal P}_A$,
\begin{equation}
{\cal P}_A({\bf x}\otimes {\bf x})=0~.
\end{equation}
In components, the commutation relations among coordinates $x^i$ is
\begin{equation}
\begin{array}{l}
x^ix^j=qx^jx^i~,~~~~~{\rm for}~i<j~ ~{\rm and}~~i\not=-j~,\\
qx^{+2}x^{-2}-q^{-1}x^{-2}x^{+2}=\frac{q^{1/2}-q^{-1/2}}{q-1+q^{-1}}
\frac{1}{a^2}~,\\
qx^{+1}x^{-1}-q^{-1}x^{-1}x^{+1}=(1-q^2)x^{+2}x^{-2}+q\frac{q^{1/2}-
q^{-1/2}}{q-1+q^{-1}}\frac{1}{a^2}~.
\end{array}
\end{equation}
Here we have used the constraint that the quantum length ($L\propto
{\bf x}^tC{\bf x}=1/a^2$)
is invariant under quantum orthogonal group transformations.
The exterior derivative $d$\cite{10} is an operator which gives the mapping from
the coordinates to the differentials
$$d:~~~x^i\longrightarrow dx^i~.$$
It satisfies two properties
$$d^2=0~,~~~~~~~~d(fg)=(df)g+(-1)^ff(dg)~.$$
where $f$ and $g$ are $p$-forms and $(-1)^f$ is $-1(+1)$ if $f$ is odd(even)
element. The exterior derivative $d$ is invariant under the quantum group
transformation and the differential $dx^i$ transforms in the same way as the
coordinate $x^i$ under the quantum group transformation.
The derivative $\partial _i$ can be introduced by
$$d=dx^i\partial_i=C_{ij}dx^i\partial^j~.$$
For the differential calculus on the quantum orthogonal space\cite{11}, the following relations
are satisfied
\begin{equation}\label{diff}
\begin{array}{l}
x^idx^j=q\hat{R}^{ij}_{kl}dx^kx^l~,\\
{\cal P}_S(d{\bf x}\wedge d{\bf x})=0~,~~~~~
{\cal P}_1(d{\bf x}\wedge d{\bf x})=0~,\\
\partial^i x^j=(C^{-1})^{ij}+q(\hat{\cal R}^{-1})^{ij}_{kl}x^k\partial^l~,\\
\left.{\cal P}_A\right.^{ij}_{kl}\partial_j\partial_i=0~,\\
\partial^idx^j=q^{-1}\hat{\cal R}^{ij}_{kl}dx^k\partial^l~,\\
\partial^id=q^{-2}d\partial^i-(q^{-2}-q^3)\frac{1-q^2}{(1-q^5)(1+q^{-3})}
dx^iC_{jk}\partial^j\partial^k~.
\end{array}
\end{equation}
\section{Realization of quantum algebra on quantum space}
An action of an algebra $A$ on a space $V$ is a bilinear map,
$${\cal A}:~~~A\otimes V\longrightarrow V~,~~~~p\otimes x\longrightarrow p
\cdot x~,$$
such that
$$(pq)\cdot x=p\cdot(q\cdot x)~,~~~~~1\cdot x=1~.$$
We call $V$ an $A$-module. In the same sense as comultiplication is the dual
to multiplication, coactions are dual to actions. The coaction of a coalgebra
$B$ on a space $V$ is a linear map
$$\omega_B:~~~V\longrightarrow B\otimes V~,$$
such that
$$(\omega_B\otimes {\rm id})\omega_B=({\rm id}\otimes \omega_B)\omega_B~,
~~~~~
({\rm id}\otimes\epsilon)\omega_B={\rm id}~.$$
The coalgebra is referred as a co-module.
We know that the quantum group co-acts on quantum space and is a co-module.
From the above general discussion, we conclude that, as a dual Hopf algebra
of the quantum group, the quantum algebra acts on the quantum space. Thus,
analogue to the classical case, a realization of quantum algebra on the
quantum space should be exist. And when the deformation parameter
$q\rightarrow 1$, familiar results on Lie algebra should be recovered.
Following these guidances, we present a realization of quantum algebra
$U_q(so(5))$ on the quantum orthogonal space.
For convenient, we introduce the dilatation operator $S_m$ ($m\leq 2$) as
$$\begin{array}{l}
\displaystyle S_m=1+q\lambda E_m+q^{2m+1}\lambda^2L_m\Delta_m~,\\
\displaystyle E_m=\sum_{j=-m}^mx^j\partial_j~,\\
\displaystyle\Delta_m=\sum_{j=1}^mq^{\rho_j}\partial_j\partial_{-j}+\frac{q}{1+q}
\partial_0\partial_0~,\\
\displaystyle L_m=\sum_{i=1}^m q^{\rho_i}x^{-i}x^i+\frac{q}{1+q}x^0x^0~.
\end{array}$$
The dilatation operator satisfies
\begin{equation}
S_m x^k=q^2x^k S_m~,~~~~~~~S_m\partial _k=q^{-2}\partial_kS_m~,~~~~{\rm for}~
\vert k\vert\leq m~.
\end{equation}
Using the notations
$$\begin{array}{l}
y^{-1}=x^{-1}+q^{3/2}\lambda L_1\partial_{+1}~,\\
\delta_{-1}=\partial_{-1}+q^{3/2}\lambda \Delta_1x^{+1}~,\\
y^{-2}=x^{-2}+q^{5/2}\lambda L_2\partial_{+2}~,\\
\delta_{-2}=\partial_{-2}+q^{5/2}\lambda \Delta_2x^{+2}~,
\end{array}$$
we can construct a set of independent basis on the quantum orthogonal space
\begin{equation}
\begin{array}{ll}
{\cal X}^{-2}=S_{2}^{-1/2}\mu_{+2}^{-1/2}y^{-2}~,~~~~~~&{\cal D}_{-2}=q^{-1}
S_2^{-1/2}\mu_{+2}^{-1/2}\delta_{-2}~,\\
{\cal X}^{-1}=\mu_{+2}^{-1/2}S_1^{-1/2}\mu_{+1}^{-1/2}y^{-1}~,~~~~~~&
{\cal D}_{-1}=\mu_{+2}^{-1/2}q^{-1}S_1^{-1/2}\mu_{+1}^{-1/2}\delta_{-1}~,\\
{\cal X}^0=\mu_{+2}^{-1/2}\mu_{+1}^{-1/2}x^0~,~~~~~~&{\cal D}_0=\mu_{+2}^{-1/2}
\mu_{+1}^{-1/2}\partial_0~,\\
{\cal X}^{+1}=\mu_{+2}^{-1/2}x^{+1}~,~~~~~~&{\cal D}_{+1}=\mu_{+2}^{-1/2}
\partial_{+1}~,\\
{\cal X}^2=x^2~,~~~~~~&{\cal D}_2=\partial_2~,
\end{array}
\end{equation}
where $(\mu_{\pm i})^{\pm 1}={\cal D}_{\pm i}{\cal X}^{\pm i}-
{\cal X}^{\pm i}{\cal D}_{\pm i}$ and $\mu_0^{1/2}={\cal D}_0{\cal X}^0-
{\cal X}^0{\cal D}_0$.\\
We note that the $\mu_i$'s satisfy simple commutation relations with the
independent basis ${\cal X}$ and ${\cal D}$,
$$[\mu_i,\mu_j]=0~,~~~~~~
\mu_i{\cal X}^j={\cal X}^j\mu_i\cdot\left\{\begin{array}{l}
q^2~,~~~{\rm for}~i=j\\
1~,~~~{\rm for}~i\not=j
\end{array}\right.~,~~~~~~
\mu_i{\cal D}^j={\cal D}^j\mu_i\cdot\left\{\begin{array}{l}
q^{-2}~,~~~{\rm for}~i=j\\
1~,~~~{\rm for}~i\not=j
\end{array}\right.~.$$
For the independent basis of coordinates and derivatives on the quantum
space\cite{0a}\cite{12}, it is not difficult to show that
\begin{equation}
\begin{array}{l}
{\cal D}_{-2}{\cal X}^{-2}=1+q^{-2}{\cal X}^{-2}{\cal D}_{-2}~,\\
{\cal D}_{-1}{\cal X}^{-1}=1+q^{-2}{\cal X}^{-1}{\cal D}_{-1}~,\\
{\cal D}_{0}{\cal X}^{0}=1+q{\cal X}^{0}{\cal D}_{0}~,\\
{\cal D}_{+1}{\cal X}^{+1}=1+q^{2}{\cal X}^{+1}{\cal D}_{+1}~,\\
{\cal D}_{+2}{\cal X}^{+2}=1+q^{2}{\cal X}^{+2}{\cal D}_{+2}~,\\[2mm]
[{\cal D}_i,{\cal D}_j]=0~,~~~~~[{\cal X}^i,{\cal X}^j]=0~,\\[2mm]
{\cal D}_i{\cal X}^j={\cal X}^j{\cal D}_i~,~~~~~{\rm for}~ i\not=j~.
\end{array}
\end{equation}
We know the Cartan matrix for its rank two Lie algebra $B_2$ is
\begin{equation}
A_{ij}=\left(\begin{array}{cc}
2&-1\\
-2&2\end{array}\right)~,~~~~~~\langle\alpha_i,\alpha_j\rangle
=\left(\begin{array}{cc}
2&-1\\
-1&1\end{array}\right)~.
\end{equation}
So that $d_1=1$, $d_2=1/2$.
The quantum universal enveloping algebra\cite{16} $U_q\left(so(5)\right)$ is
\begin{equation}
\begin{array}{l}
[\tilde{H}_i,\tilde{H}_j]=0~,~~~~~~i,~j=1,~2,\\[1mm]
[\tilde{H}_i,\tilde{X}_j^\pm]=\pm a_{ij}\tilde{X}_j^\pm~,\\[5mm]
\displaystyle [\tilde{X}_i^+,\tilde{X}_j^-]=\delta_{ij}\frac{q_i^{\tilde{H}_i}-
q_i^{-\tilde{H}_i}}
{q_i-q_i^{-1}}~,~~~~~q_i\equiv q^{d_i}~,\\[5mm]
\displaystyle\sum_{m=0}^{1-a_{ij}}\left[\begin{array}{c}
1-a_{ij}\\
m\end{array}\right]_{q_i}
(\tilde{X}_i^\pm)^m \tilde{X}_j^\pm (\tilde{X}_i^\pm)^{1-a_{ij}-m}=0~,~~~~
i\not=0~,
\end{array}
\end{equation}
where $[m]_q=\frac{q^m-q^{-m}}{q-q^{-1}}$, and $\left[\begin{array}{c}
m\\
n\end{array}\right]_q=\frac{[m]_q![m-n]_q!}{[n]_q!}$.
Redefine $H_i=d_i\tilde{H}_i$, $X_i^\pm=\sqrt{[d_i]_q}\tilde{X}_i^\pm$, we have
\begin{equation}
\begin{array}{l}
[H_1,H_2]=0~,\\[1mm]
[H_1,X_1^\pm]=\pm 2X_1^\pm~,~~~~~[H_1,X_2^\pm]=\mp X_2^\pm~,\\[1mm]
[H_2,X_1^\pm]=\mp X_1^\pm~,~~~~~[H_2,X_2^\pm]=\pm X_2^\pm~,\\[2mm]
[X_1^+,X_1^-]=\displaystyle\frac{q^{H_1}-q^{-H_1}}{q-q^{-1}}~,\\[2mm]
[X_2^+,X_2^-]=\displaystyle\frac{q^{H_2}-q^{-H_2}}{q-q^{-1}}~,\\[2mm]
X_2^\pm (X_1^\pm)^2-[2]X_1^\pm X_2^\pm X_1^\pm +(X_1^\pm)^2X_2^\pm=0~,\\
X_1^\pm (X_2^\pm)^3-[3]_{\sqrt{q}}X_2^\pm X_1^\pm (X_2^\pm)^2+[3]_{\sqrt{q}}
(X_2^\pm)^2X_1^\pm X_2^\pm-(X_2^\pm)^3X_1^\pm=0~.
\end{array}
\end{equation}
The Hopf algebra structure of the quantum universal enveloping algebra
$U_q(so(5))$ is of the form
\begin{equation}
\begin{array}{l}
\Delta(H_i)=H_i\otimes 1+1\otimes H_i~,\\
\Delta(X_i^\pm)=X_i^\pm\otimes q^{H_i/2}+q^{-H_i/2}\otimes X_i^\pm~,\\
\epsilon(H_i)=0=\epsilon(X_i^\pm)~,\\
S(H_i)=-H_i~,~~~~~~S(X_i^\pm)=-q^{\mp d_i}X_i^\pm~.
\end{array}
\end{equation}
In terms of the set of independent basis ${\cal X}^i,~{\cal D}_i$ on the
quantum space, we write down a realization
of the quantum universal enveloping algebra $U_q(so(5))$ in fundamental
representation,
\begin{equation}
\begin{array}{l}
q^{H_1}=\left(\frac{\mu_{-2}\mu_{+1}}{\mu_{-1}\mu_{+2}}
\right)^{1/2}~,\\
X^-_1=\sqrt{q^{-1}\lambda}\left(\frac{\mu_{-1}\mu_{+2}}{\mu_{-2}\mu_{+1}}\right)^{1/4}
\mu_{+2}^{-1/2}
\left((\mu_{-2}\mu_{+1})^{1/2}{\cal X}^{-1}
{\cal D}_{-2}-q{\cal X}^{+2}{\cal D}_{+1}\right)~,\\
X^+_1=\sqrt{q^{-1}\lambda}\left(\frac{\mu_{-1}\mu_{+2}}{\mu_{-2}\mu_{+1}}\right)^{1/4}
\mu_{+2}^{-1/2}
\left((\mu_{-2}\mu_{+1})^{1/2}{\cal X}^{-2}
{\cal D}_{-1}-q{\cal X}^{+1}{\cal D}_{+2}\right)~,\\
q^{H_2}=\left(\frac{\mu_{-1}}{\mu_{+1}}\right)^{1/2}~,\\
X^-_2=\sqrt{q^{1/2}\lambda}(\mu_{-1}\mu_{+1})^{-1/4}\left(q^{-3/2}\mu_{-1}^{1/2}
{\cal X}^0{\cal D}_{-1}-{\cal X}^{+1}{\cal D}_0\right)~,\\
X^+_2=\sqrt{q^{1/2}\lambda}(\mu_{-1}\mu_{+1})^{-1/4}\left(q^{1/2}\mu_{-1}^{1/2}
{\cal X}^{-1}{\cal D}_0-{\cal X}^0{\cal D}_{+1}\right)~.
\end{array}
\end{equation}
Therefore, we indeed have a natural action of the quantum algebra $U_q(so(5))$
on the
quantum orthogonal space. It may be referred as the quantum counterpart of
the Lie algebra realized on Euclidean space. In fact, at the case of
$q=1$, this realization recovered a ordinary realization of Lie algebra.
This is helpful for clearly understanding of meanings of quantum group and quantum
algebra.
\section{Quantum AdS group and quantum AdS space}
To study quantum AdS group, real forms and $*$-conjugations of quantum
orthogonal group should be introduced. A $*$-structure on a Hopf algebra $A$
is an algebra anti-automorphism $(\eta ab)^*=\bar{\eta}b^*a^*~(\forall a,~b\in
A,~\forall\eta\in C$), coalgebra automorphism $\Delta\cdot *=(*\otimes *)
\cdot\Delta$, $\epsilon\cdot *=\epsilon$ and involution $*^2=1$.
Following FRT\cite{09}, on quantum orthogonal groups conjugations can be defined.
The first type is trivially as $T^\times =T$.
Compatibility with the Yang-Baxter
equation requires $\overline{\cal R}={\cal R}^{-1}$ and
$\overline{C}C^t=1$. This occurs only for
$\vert q\vert=1$. Then the $CTT$ relations are invariant under the
$*$-conjugation. As we know the quantum orthogonal group
co-acts on the quantum orthogonal space, and may induce an associated
conjugation on the quantum space as well. More precisely a conjugation
on the quantum space is compatible with a conjugation on its quantum symmetry
group if the coaction $\omega(x^a)=T^a_{~b}\otimes x^b$ satisfies
$\omega({\bf x}^\times)=T^\times\otimes {\bf x}^\times\equiv \delta^\times
({\bf x})$. The unique associated
quantum space conjugation here is $(x^a)^\times=x^a$.
Clearly, we can not get the desired quantum AdS space by this kind of
conjugation on quantum orthogonal space.
Further structure should be added
to the conjugation to arrive at the quantum AdS group and quantum AdS space.
For the aim, we introduce another operation on the quantum orthogonal group as
\begin{equation}
T^\dagger={\cal D}T{\cal D}^{-1}~,
\end{equation}
where the matrix ${\cal D}$ is of the form
$${\cal D}=\left(\begin{array}{ccccc}
1& & & & \\
&1& & & \\
& &-1&& \\
& & &1& \\
& & & &1\end{array}\right)~.$$
It is easy to check that the ${\cal D}$ matrix is a special element of the
quantum orthogonal group\cite{13}, {\it i.e.},
$$\begin{array}{l}
{\cal R}_{12}{\cal D}_1{\cal D}_2={\cal D}_2{\cal D}_1{\cal R}_{12}~,\\
{\cal D}^tC{\cal D}=C~,~~~~{\cal D}C{\cal D}^t=C~.
\end{array}$$
Thus the $\dagger$ operation is compatible with the Hopf algebra structure
of the quantum orthogonal group,
$$
\begin{array}{l}
{\cal R}_{12}T^\dagger_1T^\dagger_2=T^\dagger_2T^\dagger_1{\cal R}_{12}~,\\
(T^\dagger)^tCT^\dagger=C~,~~~~~T^\dagger C(T^\dagger)^t=C~,\\
\Delta(T^\dagger)={\cal D}\Delta(T){\cal D}^{-1}=T^\dagger\otimes
T^\dagger~,\\
\epsilon(T^\dagger)=\epsilon(T)~,~~~~~S(T^\dagger)=[S(T)]^\dagger~.
\end{array}$$
We finally obtain the desired AdS quantum group by using the combined operation of the
$\times$ and $\dagger$, i.e., the conjugation $T^*\equiv
T^{\times\dagger}={\cal D}T{\cal D}^{-1}$.
The induced conjugation on the quantum space is ${\bf x}^*\equiv {\bf x}^{
\times\dagger}={\cal D}{\bf x}$.
To show that the conjugation really gives the quantum AdS group and quantum
AdS space, we should find a linear transformation ${\bf x}\longrightarrow
{\bf x}'=M {\bf x},~ T\longrightarrow T'=MTM^{-1}$
such that the new coordinates ${\bf x}'$ and $T'$ are real and the new metric
$C'=(M^{-1})^tCM^{-1}$ diagonal in the $q\longrightarrow 1$ limit,
$C'\vert_{q=1}={\rm diag}(1,-1,-1,-1,1)$.
Here the $M$ matrix is given by
$$M=\frac{1}{\sqrt{2}}\left(\begin{array}{ccccc}
1& & & &1\\
-1& & & &1\\
& &i\sqrt{2}& &\\
&-1& & 1& \\
& 1& & 1& \end{array}\right)~.$$
The second conjugation of the quantum orthogonal group given by FRT\cite{09}
is realized via the
metric, {\it i.e.}, $T^\star=C^tTC^t$. The condition on the quantum
${\cal R}$ matrix is
$\overline{\cal R}={\cal R}$, which happens for $q\in R$. Again the $CTT$
relations are invariant under such a conjugation operation.
The corresponding real form is $SO_q(5;R)$.
The induced conjugation on the quantum space is ${\bf x}^\star=C^t{\bf x}$.
Again, we can not get the desired quantum AdS group and quantum AdS space by
this kind of conjugation along.
To obtain quantum AdS group, we introduce the operation $\ddagger$ on the
quantum orthogonal group $SO_q(5)$ as
\begin{equation}
T^\ddagger={\cal N}T{\cal N}^{-1}~,
\end{equation}
where ${\cal N}$ matrix is of the form
$${\cal N}=\left(\begin{array}{ccccc}
1& & & & \\
&-1& & & \\
& &-1& & \\
& & &-1& \\
& & & &1\end{array}\right)~.$$
The ${\cal N}$ matrix is also a special element of the
quantum orthogonal group, {\it i.e.},
\begin{equation}\label{rnn}
\begin{array}{l}
{\cal R}_{12}{\cal N}_1{\cal N}_2={\cal N}_2{\cal N}_1{\cal R}_{12}~,\\
{\cal N}^tC{\cal N}=C~,~~~~{\cal N}C{\cal N}^t=C~.
\end{array}
\end{equation}
Thus the $\ddagger$ operation is also compatible with the Hopf algebra
structure of quantum group.
We obtain the desired AdS quantum group by using the combined operation of the
$\star$ and $\ddagger$, {\it i.e.}, the conjugation $T^*\equiv
T^{\star\ddagger}={\cal N}C^tTC^t{\cal N}^{-1}$.
The induced conjugation on the quantum space is ${\bf x}^*\equiv
{\bf x}^{\star\ddagger}=C^t {\cal N}{\bf x}$.
Parallel to the case of $\vert q\vert=1$, we prove the combined conjugation really
gives the quantum AdS group and quantum AdS space by trying to
find a linear transformation ${\bf x}\longrightarrow {\bf x}'=N {\bf x},~ T\longrightarrow
T'=NTN^{-1}$
such that the new coordinates ${\bf x}'$ and $T'$ are real and the new metric
$C'=(N^{-1})^tCN^{-1}$ diagonal in the $q\longrightarrow 1$ limit,
$C'\vert_{q=1}={\rm diag}(1,-1,-1,-1,1)$. We find a $N$ matrix which
satisfies all these requirements of the form
$$N=\frac{1}{\sqrt{2}}\left(\begin{array}{ccccc}
1& & & &q^{3/2}\\
& -i& &-iq^{1/2}& \\
& &i\sqrt{2}& &\\
&q^{-1/2}& &-1& \\
iq^{-3/2}& & & &-i \end{array}\right)~.$$
\section{Differential Calculus on the quantum AdS space}
To study differential calculus on the quantum AdS space, we should
introduce conjugate derivatives. Because there are two different conjugations
on the quantum AdS space corresponds to the deformation parameter $q$ being root of unit
or real, respectively. We present differential calculus on the quantum AdS
space for the two cases separately.
In the case of $q$ being root of unit, we note the conjugation on $x^i$,
$\hat{\cal R}$ and $C$ is given by
\begin{equation}
\left.x^i\right.^*={\cal D}_{ij}x^j~,~~~~~\overline{\hat{\cal R}^{uv}_{ij}}=\left.
\hat{\cal R}^{-1}\right.
^{vu}_{ji}~,~~~~~\overline{C^{ij}}=C^{ji}~.
\end{equation}
Because the matrix ${\cal D}$ is a special element of the quantum orthogonal group,
we can prove that
$$\left.\hat{\cal R}^{-1}\right.^{ik}_{ld}
{\cal D}_{kn}{\cal D}_{sl}{\cal D}_{ip}{\cal D}_{td}=
\left.\hat{\cal R}^{-1}\right.^{st}_{pn}~.$$
The $\hat{\cal R}$-matrix has the following symmetry properties
\begin{equation}\label{I}
C^{im}\hat{\cal R}^{jn}_{mk}C_{nl}=\left.\hat{\cal R}^{-1}\right._{kl}^{ij}
=C_{km}\hat{\cal R}^{mi}_{ln}C^{nj}~,
\end{equation}
and
\begin{equation}\label{II}
\hat{\cal R}^{ij}_{kl}=\hat{\cal R}^{kl}_{ij}~.
\end{equation}
Using the above relations as well as the conjugation of the definition of
derivatives,
$\partial_ix^j=\delta^j_i+q\hat{\cal R}^{jk}_{il}x^l\partial_k$, we obtain the
action of the conjugate derivatives on the quantum AdS space,
\begin{equation}
\hat{\partial}_mx^s=\delta^s_m+q\hat{\cal R}^{sl}_{mk}x^k\hat{\partial}_l~,
\end{equation}
where $\hat{\partial}_m\equiv -q^{-1}{\cal D}_{mk}C_{kv}\left.\partial^v\right.^*$.
This show that the action of the conjugate derivatives on the quantum AdS space
is almost the same with that of the derivatives at the case of $\vert q\vert=1$.
In fact, we can represent
the conjugate derivatives $\partial^*$ linearly in terms of the
derivatives $\partial$,
\begin{equation}
\left.\partial_v\right.^*=-qC_{vk}{\cal D}_{km}\partial_m~.
\end{equation}
From a very general consideration\cite{10}, we know that there are two types of
consistent derivatives $\partial_i$ and $\tilde{\partial}_i$ satisfy the
following relations, respectively,
\begin{equation}\label{conjugate}
\begin{array}{l}
\partial_ix^j=\delta^j_i+q\hat{\cal R}^{jk}_{il}x^l\partial_k~,\\
\tilde{\partial}_kx^v=\delta^v_k+q^{-1}\left.\hat{\cal R}^{-1}\right.^{vi}_{kj}
x^j\tilde{\partial}_i~.
\end{array}
\end{equation}
The first one is just what given in Eq.(\ref{diff}). We show that the second
one is related with the conjugate derivatives in the case of $q\in R$.
In the following, we present actions of the conjugate derivatives on the quantum
AdS space at the case of $q\in R$.
To proceed, we first conjugate the first relation in Eq.(\ref{conjugate})
and invert it to find an
expression for $\partial_i^*x^j$ in terms of $x^j\partial_i^*$ by using
$$\left.x^i\right.^*=C_{ji}{\cal N}_{jk}x^k$$
and the symmetry properties of the $\hat{\cal R}$-matrix Eq.(\ref{I}) and (\ref{II}).
The result is:
\begin{equation}
-q^5{\cal N}_{ip}C_{ia}C^{va}\left.\partial^v\right.^*x^s=q^{4}C_{ia}C_{p\delta}
\hat{\cal R}^{p\delta}_{al}{\cal N}_{sl}{\cal N}_{ip}-q^4\left.\hat{\cal R}^{-1}\right.^{ik}_{ld}
{\cal N}_{kn}{\cal N}_{sl}{\cal N}_{ip}{\cal N}_{td}x^nC_{db}C^{mb}
{\cal N}_{dt}\left.\partial^m\right.^*~.
\end{equation}
Making use of Eq.(\ref{rnn}), we know that
$$\left.\hat{\cal R}^{-1}\right.^{ik}_{ld}
{\cal N}_{kn}{\cal N}_{sl}{\cal N}_{ip}{\cal N}_{td}=
\left.\hat{\cal R}^{-1}\right.^{st}_{pn}~.$$
Then, we get
\begin{equation}
\hat{\partial}_px^s=\delta^s_p+q^{-1}\left.\hat{\cal R}^{-1}\right.^{st}_{pn}
x^n\hat{\partial}_t~,
\end{equation}
where
$$\hat{\partial}_p=-q^5{\cal N}_{ip}C_{ia}C^{va}\left.\partial^v\right.^*~.$$
$\hat{\partial}_i$ (or equivalient $\partial_i^*$)
can be expressed algebraically in terms of $x^k$ and $\partial_l$\cite{14},
\begin{equation}
\hat{\partial}_k=q^3S_2^{-1}[\Delta_2,x^k]~.
\end{equation}
To complete the differential calculus on the quantum AdS space, we give the
consistent relations satisfied by $\hat{\partial}$ and $\partial$
\begin{equation}
\begin{array}{l}
\hat{\partial}_i\hat{\partial}_j=q^{-1}\hat{\cal R}^{lk}_{ji}\hat{\partial}_k
\hat{\partial}_l~,\\
\hat{\partial}_i\partial_j=q\hat{\cal R}^{lk}_{ji}\partial_k
\hat{\partial}_l~.
\end{array}
\end{equation}
\section{Quantum group via quantum algebra}
It is well known that, in Lie group theory, there is an exponential
corresponding between group and algebra. Representations of Lie group are
easily deduced from that of Lie algebra. And then quantum field theory
is constructed based on representations of Lie algebra. However, in general,
there is no such a direct transformation from quantum algebra to quantum group,
besides duality between them. Bearing the goal of constructing quantum field
theory on noncommutative geometry, we try to set up an explicit relationship\cite{15}
between quantum orthogonal group and quantum orthogonal algebra.
It is convenient to introduce the operator $L^+$ as
\begin{equation}
L^+=\left(\begin{array}{ccccc}
l^+_{11}&l^+_{12}&l^+_{13}&l^+_{14}&l^+_{15}\\
&l^+_{22}&l^+_{23}&l^+_{24}&l^+_{25}\\
& &l^+_{33}&l^+_{34}&l^+_{35}\\
& & &l^+_{44}&l^+_{45}\\
& & & &l^+_{55}\end{array}\right)~,
\end{equation}
where
\begin{equation}
\begin{array}{l}
l^+_{11}=q^{H_1+H_2}~,\\
l^+_{12}=\lambda q^{-1/2}X_1^+q^{\frac{H_1}{2}+H_2}~,\\
l^+_{13}=\lambda q^{-1/2}\left(X_1^+X_2^+
-q^{-1}X_2^+X_1^+\right)q^{\frac{H_1+H_2}{2}}~,\\
l^+_{14}=\lambda q^{-5/2}\left(-qX_1^+(X_2^+)^2+(q+1)
X_2^+X_1^+X_2^+-(X_2^+)^2X_1^+\right)
q^{H_1/2}~,\\
l^+_{15}=\frac{\lambda^2 q^{-2}}{1+q}\left((X_1^+)^2(X_2^+)^2-
(q+1+q^{-1})X_1^+X_2^+X_1^+X_2^++(q+1+q^{-1})X_1^+(X_2^+)^2X_1^+\right.\\
~~~~~~~~~~~~~~~~\left. -(q+1+q^{-1})X_2^+X_1^+X_2^+X_1^++X_2^+(X_1^+)^2X_2^+
+(X_2^+)^2(X_1^+)^2\right)~,\\
l^+_{22}=q^{H_2}~,\\
l^+_{23}=\lambda q^{-1/2}X_2^+q^{H_2/2}~,\\
l^+_{24}=-\frac{\lambda^2}{q(1+q)}(X_2^+)^2~,\\
l^+_{25}=\lambda q^{-5/2}\left(X_1^+(X_2^+)^2-(1+q)
X_2^+X_1^+X_2^++q(X_2^+)^2X_1^+\right)q^{-H_1/2}~,\\
l^+_{33}=1~,\\
l^+_{34}=-\lambda q^{-1}X_2^+q^{-H_2/2}~,\\
l^+_{35}=\lambda q^{-1}\left(-q^{-1}X_1^+X_2^++
X_2^+X_1^+\right)q^{-\frac{H_1+H_2}{2}}~,\\
l^+_{44}=q^{-H_2}~,\\
l^+_{45}=-\lambda q^{-1/2}X_1^+q^{-(\frac{H_1}{2}+H_2)}~,\\
l^+_{55}=q^{-(H_1+H_2)}~.
\end{array}
\end{equation}
In the same manner, we introduce another operator $L^-$ as
\begin{equation}
L^-=\left(\begin{array}{ccccc}
l^-_{11}& & & & \\
l^-_{21}&l^-_{22}& & & \\
l^-_{31}&l^-_{32}&l^-_{33}& & \\
l^-_{41}&l^-_{42}&l^-_{43}&l^-_{44}& \\
l^-_{51}&l^-_{52}&l^-_{53}&l^-_{54}&l^-_{55}\end{array}\right)~,
\end{equation}
where
\begin{equation}
\begin{array}{l}
l^-_{11}=q^{-(H_1+H_2)}~,\\
l^-_{21}=-\lambda q^{1/2}X_1^-q^{-\left(\frac{H_1}{2}+H_2\right)}~,\\
l^-_{31}=\lambda q^{3/2}\left(X_1^-X_2^-
-q^{-1}X_2^-X_1^-\right)q^{-\frac{H_1+H_2}{2}}~,\\
l^-_{41}=\lambda q^{1/2}\left(qX_1^-(X_2^-)^2-(q+1)
X_2^-X_1^-X_2^-+(X_2^-)^2X_1^-\right)q^{-H_1/2}~,\\
l^-_{51}=\frac{\lambda^2 q^{2}}{1+q}\left((X_1^-)^2(X_2^-)^2-
(q+1+q^{-1})X_1^-X_2^-X_1^-X_2^-+(q+1+q^{-1})X_1^-(X_2^-)^2X_1^-\right.\\
~~~~~~~~~~~~~~~~\left. -(q+1+q^{-1})X_2^-X_1^-X_2^-X_1^-+X_2^-(X_1^-)^2X_2^-
+(X_2^-)^2(X_1^-)^2\right)~,\\
l^-_{22}=q^{-H_2}~,\\
l^-_{32}=-\lambda q^{1/2}X_2^-q^{-H_2/2}~,\\
l^-_{42}=-\frac{\lambda^2 q}{1+q}(X_2^-)^2~,\\
l^-_{52}=\lambda q^{1/2}\left(-X_1^-(X_2^-)^2+(1+q)
X_2^-X_1^-X_2^- -q(X_2^-)^2X_1^-\right)q^{H_1/2}~,\\
l^-_{33}=1~,\\
l^-_{43}=\lambda X_2^-q^{H_2/2}~,\\
l^-_{53}=\lambda q\left(-q^{-1}X_1^-X_2^-+X_2^-X_1^-\right)q^{\frac{H_1+H_2}{2}}~,\\
l^-_{44}=q^{H_2}~,\\
l^-_{54}=\lambda q^{1/2}X_1^-q^{\frac{H_1}{2}+H_2}~,\\
l^-_{55}=q^{H_1+H_2}~.
\end{array}
\end{equation}
By making use of the operators $L^\pm$, we can write the quantum universal
enveloping algebra $U_q(so(5))$ into a compact form
\begin{equation}\label{algebra}
\begin{array}{l}
{\cal R}_{12}L^\pm_1L^\pm_2=L^\pm_2 L^\pm_1{\cal R}_{12}~,\\
{\cal R}_{12}L^-_1L^+_2=L^+_2 L^-_1{\cal R}_{12}~,\\
\Delta(L^\pm)=L^\pm\dot{\otimes}L^\pm~,~~~~~~\epsilon(L^\pm)=1~,\\
S(L^+)=\left(\begin{array}{ccccc}
l^+_{55}&q^{-1}l^+_{45}&q^{-3/2}l^+_{35}&q^{-2}l^+_{25}&
q^{-3}l^+_{15}\\
&l^+_{44}&q^{-1/2}l^+_{34}&q^{-1}l^+_{24}&q^{-2}l^+_{14}\\
& &l^+_{33}&q^{-1/2}l^+_{23}&q^{-3/2}l^+_{13}\\
& & &l^+_{22}&q^{-1}l^+_{21}\\
& & & &l^+_{11}\end{array}\right)~,\\
S(L^-)=\left(\begin{array}{ccccc}
l^-_{55}& & & &\\
ql^-_{54}&l^-_{44}& & &\\
q^{3/2}l^-_{53}&q^{1/2}l^-_{43}&l^-_{33}& &\\
q^2l^-_{52}&ql^-_{42}&q^{1/2}l^-_{32}&l^-_{22}&\\
q^{3}l^-_{51}&q^{2}l^-_{41}&q^{3/2}l^-_{31}&ql^-_{21}&l^-_{11}
\end{array}\right)~.
\end{array}
\end{equation}
As well known, in the classical case, many important properties of Lie group
are conveniently studied using corresponding Lie algebra and vise verse. This
is based on the exponential relation of Lie group and Lie algebra. However,
in the quantum group theory, there is no such a counterpart of exponential
has been found. All we know is the duality between quantum group and quantum
algebra.
We give here an explicit relation between elements of quantum group and
generators
of the quantum universal enveloping algebra,
\begin{equation}
T=L^+\dot{\otimes} L^-~,~~~~~{\rm or}~~~T=L^-\dot{\otimes} L^+~.
\end{equation}
Then elements of the $T$ matrix are expressed as
\begin{equation}
\begin{array}{l}
t_{11}=l_{11}^+\otimes l_{11}^-+l_{12}^+\otimes l_{21}^-+l_{13}^+\otimes l_{31}^-
+l_{14}^+\otimes l_{41}^-+l_{15}^+\otimes l_{51}^-~,\\
t_{12}=l_{12}^+\otimes l_{22}^-+l_{13}^+\otimes l_{32}^-+l_{14}^+\otimes l_{42}^-
+l_{15}^+\otimes l_{52}^-~,\\
t_{13}=l_{13}^+\otimes l_{33}^-+l_{14}^+\otimes l_{43}^-+l_{15}^+\otimes l_{53}^-~,\\
t_{14}=l_{14}^+\otimes l_{44}^-+l_{15}^+\otimes l_{54}^-~,\\
t_{15}=l_{15}^+\otimes l_{55}^-~,\\
t_{21}=l_{22}^+\otimes l_{21}^-+l_{23}^+\otimes l_{31}^-+l_{24}^+\otimes l_{41}^-
+l_{25}^+\otimes l_{51}^-~,\\
t_{22}=l_{22}^+\otimes l_{22}^-+l_{23}^+\otimes l_{32}^-+l_{24}^+\otimes l_{42}^-
+l_{25}^+\otimes l_{52}^-~,\\
t_{23}=l_{23}^+\otimes l_{33}^-+l_{24}^+\otimes l_{43}^-+l_{25}^+\otimes l_{53}^-~,\\
t_{24}=l_{24}^+\otimes l_{44}^-+l_{25}^+\otimes l_{54}^-~,\\
t_{25}=l_{25}^+\otimes l_{55}^-~,\\
t_{31}=l_{33}^+\otimes l_{31}^-+l_{34}^+\otimes l_{41}^-+l_{35}^+\otimes l_{51}^-~,\\
t_{32}=l_{33}^+\otimes l_{32}^-+l_{34}^+\otimes l_{42}^-+l_{35}^+\otimes l_{52}^-~,\\
t_{33}=l_{33}^+\otimes l_{33}^-+l_{34}^+\otimes l_{43}^-+l_{35}^+\otimes l_{53}^-~,\\
t_{34}=l_{34}^+\otimes l_{44}^-+l_{35}^+\otimes l_{54}^-~,\\
t_{35}=l_{35}^+\otimes l_{55}^-~,\\
t_{41}=l_{44}^+\otimes l_{41}^-+l_{45}^+\otimes l_{51}^-~,\\
t_{42}=l_{44}^+\otimes l_{42}^-+l_{45}^+\otimes l_{52}^-~,\\
t_{43}=l_{44}^+\otimes l_{43}^-+l_{45}^+\otimes l_{53}^-~,\\
t_{44}=l_{44}^+\otimes l_{44}^-+l_{45}^+\otimes l_{54}^-~,\\
t_{45}=l_{45}^+\otimes l_{55}^-~,\\
t_{51}=l_{55}^+\otimes l_{51}^-~,\\
t_{52}=l_{55}^+\otimes l_{52}^-~,\\
t_{53}=l_{55}^+\otimes l_{53}^-~,\\
t_{54}=l_{55}^+\otimes l_{54}^-~,\\
t_{55}=l_{55}^+\otimes l_{55}^-~;
\end{array}
\end{equation}
or, equivalently
\begin{equation}
\begin{array}{l}
t_{11}=l_{11}^-\otimes l_{11}^+~,\\
t_{12}=l_{11}^-\otimes l_{12}^+~,\\
t_{13}=l_{11}^-\otimes l_{13}^+~,\\
t_{14}=l_{11}^-\otimes l_{14}^+~,\\
t_{15}=l_{11}^-\otimes l_{15}^+~,\\
t_{21}=l_{21}^-\otimes l_{11}^+~,\\
t_{22}=l_{21}^-\otimes l_{12}^++l_{22}^-\otimes l_{22}^+~,\\
t_{23}=l_{21}^-\otimes l_{13}^++l_{22}^-\otimes l_{23}^+~,\\
t_{24}=l_{21}^-\otimes l_{14}^++l_{22}^-\otimes l_{24}^+~,\\
t_{25}=l_{21}^-\otimes l_{15}^++l_{22}^-\otimes l_{25}^+~,\\
t_{31}=l_{31}^-\otimes l_{11}^+~,\\
t_{32}=l_{31}^-\otimes l_{12}^++l_{32}^-\otimes l_{22}^+~,\\
t_{33}=l_{31}^-\otimes l_{13}^++l_{32}^-\otimes l_{23}^++l_{33}^-\otimes l_{33}^+~,\\
t_{34}=l_{31}^-\otimes l_{14}^++l_{32}^-\otimes l_{24}^++l_{33}^-\otimes l_{34}^+~,\\
t_{35}=l_{31}^-\otimes l_{15}^++l_{32}^-\otimes l_{25}^++l_{33}^-\otimes l_{35}^+~,\\
t_{41}=l_{41}^-\otimes l_{11}^+~,\\
t_{42}=l_{41}^-\otimes l_{12}^++l_{42}^-\otimes l_{22}^+~,\\
t_{43}=l_{41}^-\otimes l_{13}^++l_{42}^-\otimes l_{23}^++l_{43}^-\otimes l_{33}^+~,\\
t_{44}=l_{41}^-\otimes l_{14}^++l_{42}^-\otimes l_{24}^++l_{43}^-\otimes l_{34}^+
+l_{44}^-\otimes l_{44}^+~,\\
t_{45}=l_{41}^-\otimes l_{15}^++l_{42}^-\otimes l_{25}^++l_{43}^-\otimes l_{35}^+
+l_{44}^-\otimes l_{45}^+~,\\
t_{51}=l_{51}^-\otimes l_{11}^+~,\\
t_{52}=l_{51}^-\otimes l_{12}^++l_{52}^-\otimes l_{22}^+~,\\
t_{53}=l_{51}^-\otimes l_{13}^++l_{52}^-\otimes l_{23}^++l_{53}^-\otimes l_{33}^+~,\\
t_{54}=l_{51}^-\otimes l_{14}^++l_{52}^-\otimes l_{24}^++l_{53}^-\otimes l_{34}^+
+l_{54}^-\otimes l_{44}^+~,\\
t_{55}=l_{51}^-\otimes l_{15}^++l_{52}^-\otimes l_{25}^++l_{53}^-\otimes l_{35}^+
+l_{54}^-\otimes l_{45}^++l_{55}^-\otimes l_{55}^+~.
\end{array}
\end{equation}
It is straightforward to verify that the matrix $T$ defined above really
satisfies all relations which a quantum group should be satisfied.
Let $P^\otimes$ be the transposition operator, for arbitrary matrices $A$ and
$B$, which satisfies
$$P^\otimes:~~~A\dot{\otimes}B=B\dot{\otimes}A~.$$
Then the comultiplication $\Delta$ for these tensor operator $T$ can be
introduced as
\begin{equation}
\Delta=({\rm id}\otimes P^\otimes \otimes{\rm id})\Delta_{L^\pm}\dot{\otimes}
\Delta_{L^\mp}~.
\end{equation}
It is easy to check that
$$\begin{array}{rcl}
\Delta (T)&=&({\rm id}\otimes P^\otimes \otimes{\rm id})\Delta_{L^\pm}\dot{\otimes}
\Delta_{L^\mp}(L^\pm\dot{\otimes}L^\mp)\\
&=&({\rm id}\otimes P^\otimes \otimes{\rm id})
(L^\pm\dot{\otimes}L^\pm\dot{\otimes}L^\mp\dot{\otimes}L^\mp)\\
&=&(L^\pm\dot{\otimes}L^\mp)\dot{\otimes}(L^\pm\dot{\otimes}
L^\mp)\\
&=&T\dot{\otimes}T~.\end{array}$$
Define counit operator $\epsilon$ as
$$\epsilon=\epsilon_{L^\pm}\otimes\epsilon_{L^\mp}~.$$
Then we have $\epsilon(T)=1$. Finally, the antipode operator $S$ is of the
form,
$$S=P(S_{L^\mp}\otimes S_{L^\pm})P^\otimes~.$$
By making use of the relations (\ref{algebra}), it is a straightforward calculation to check that
$$\begin{array}{rcl}
S(T)&=&S(L^\pm\dot{\otimes}L^\mp)\\
&=&P(S_{L^\mp}\otimes S_{L^\pm})P^\otimes(L^\pm\dot{\otimes}L^\mp)\\
&=&P(S_{L^\mp}(L^\mp)\otimes S_{L^\pm}(L^\pm))\\
&=&CT^tC^{-1}~.\end{array}$$
At the stage, we can say that a realization of quantum group $SO_q(5)$ is
given in terms of quantum algebra $U_q(so(5))$.
\section{Quantum AdS Algebra}
Quantum field theory is usually constructed based on representations of an algebra. To
get a quantum field theory on noncommutative geometry, we should investigate properties of the quantum
AdS algebra. Of course, first of all, we should give conjugations on quantum
AdS algebra, which is consistent with conjugations on quantum AdS group and quantum
AdS space.
At the case of $\vert q\vert=1$, conjugation on quantum AdS group is given by
\begin{equation}
\begin{array}{rcl}
T^*&=&{\cal D}T{\cal D}^{-1}\\
&=&\left(\begin{array}{ccccc}
t_{11}&t_{12}&-t_{13}&t_{14}&t_{15}\\
t_{21}&t_{22}&-t_{23}&t_{24}&t_{25}\\
-t_{31}&-t_{32}&t_{33}&-t_{34}&-t_{35}\\
t_{41}&t_{42}&-t_{43}&t_{44}&t_{45}\\
t_{51}&t_{52}&-t_{53}&t_{54}&t_{55}\end{array}\right)~.
\end{array}
\end{equation}
From the relationship between quantum group and quantum algebra, {\it i.e.},
$T=L^\pm\dot{\otimes} L^\mp$, we deduce directly the corresponding quantum AdS algebra as
\begin{equation}
\begin{array}{l}
\left.X_1^+\right.^*=X_1^+~,~~~~~
\left.X_1^-\right.^*=X_1^-~,~~~~~
H_1^*=-H_1~,\\
\left.X_2^+\right.^*=-X_2^+~,~~~~~
\left.X_2^-\right.^*=-X_2^-~,~~~~~
H_2^*=-H_2~.
\end{array}
\end{equation}
It is easy to show that this type of conjugation on quantum algebra satisfies
all requirements for a conjugation.
For the case of $q$ being real, we know from the previous section
that the conjugation structure of the quantum AdS group is given by
\begin{equation}\label{mmm}
\begin{array}{rcl}
T^*&=&{\cal N}C^tT(C^{-1})^t{\cal N}^{-1}\\
&=&\left(\begin{array}{ccccc}
t_{55}&-qt_{54}&-q^{3/2}t_{53}&-q^2t_{52}&q^3t_{51}\\
-q^{-1}t_{45}&t_{44}&q^{1/2}t_{43}&qt_{42}&-q^2t_{41}\\
-q^{-3/2}t_{35}&q^{-1/2}t_{34}&t_{33}&q^{1/2}t_{32}&
-q^{3/2}t_{31}\\
-q^{-2}t_{25}&q^{-1}t_{24}&q^{-1/2}t_{23}&t_{22}&
-qt_{21}\\
q^{-3}t_{15}&-q^{-2}t_{14}&-q^{-3/2}t_{13}&-q^{-1}t_{12}&
t_{11}\end{array}\right)~.
\end{array}
\end{equation}
For $T=L^\pm\dot{\otimes}L^\mp$, the proper conjugation operation is given by
\begin{equation}\label{nnn}
T^*=\left.L^\mp\right.^*\dot{\otimes}\left.L^\pm\right.^*~,~~~~~~
\left(\left.L^\pm\right.^*\right)_{ij}\equiv \left(L^\pm_{ij}\right)^*~.
\end{equation}
Comparing of Eq.(\ref{mmm}) and (\ref{nnn}) gives the corresponding quantum
AdS algebra to quantum AdS group and quantum AdS space,
\begin{equation}
\begin{array}{l}
\left.X_1^+\right.^*=-q^2X_1^-~,~~~~~
\left.X_1^-\right.^*=-q^{-2}X_1^+~,~~~~~
H_1^*=H_1~,\\
\left.X_2^+\right.^*=q^{3/2}X_2^-~,~~~~~
\left.X_2^-\right.^*=q^{-3/2}X_2^+~,~~~~~
H_2^*=H_2~.
\end{array}
\end{equation}
Therefore, finally, we obtain the quantum AdS algebra for both the case of
$\vert q\vert=1$ and $q\in R$. The representations of the quantum AdS
algebra\cite{17} should be foundations of constructing quantum field theory on the
quantum AdS space. Further investigating on this direction is still on
progressing.
\bigskip
\bigskip
\centerline{\large\bf Acknowledgments }
I would like to thank Prof. J. Wess for introducing the problem to me and for
enlightening discussions.
I am grateful to B. L. Cerchiai and H. Steinacker for valuable discussions.
The work was supported in part by the National Science Foundation of
China under Grant 19625512.
|
\section{Introduction}
Light-cone gauge in quantum field theory is a fascinating subject. It is a
``simple'' gauge choice in the sense that the emerging propagator for the boson
field has a deceivingly simple structure, when compared to other non-covariant
gauge choices. Yet, as it has soon been realized, it hid subtle complications
when one actually wanted to work with it.
To make things more concrete, let us analyse it in the framework of vector
gauge fields, e.g. the pure Yang-Mills fields, where, after taking the limit of
vanishing gauge parameter, the propagator reads,
\begin{equation}
\label{prop}
D^{ab}_{\mu\nu}(k)=\frac{-i\,\delta^{ab}}{k^2+i\varepsilon}
\left[g_{\mu\nu}-\frac{n_{\mu}k_{\nu}+n_{\nu}k_{\mu}}{k\!\cdot\! n} \right ]
\end{equation}
where $n_{\mu}$ is the arbritary and constant null four-vector which defines
the gauge, $n\!\cdot\!A^a(x)=0;\;n^2=0$. This generates $D$-dimensional
Feynman integrals of the following generic form,
\begin{equation}
\label{struc}
I_{lc}=\int
\frac{d^Dk_i}{A(k_j,\;p_l)}\frac{f(k_j\!\cdot\!n^*,\;p_l\!\cdot\!n^*)}{h(k_j\!\cdot\!n,\;p_l\!\cdot\!n)},
\end{equation}
where $p_{l}$ labels all the external momenta, and $n^{*}_{\mu}$ is a null
four-vector, dual to $n_{\mu}$. We will evaluate, as a pedagogical example, an
integral where,
\begin{equation} \label{exemplo} A(k_j,\;p_l) =
(k^2)^{-i}\left[(k-p)^2\right]^{-j}, \;\; h(k_j\!\cdot\!n,\;p_l\!\cdot\!n) =
(k\cdot n)^{-l},\;\; f(k_j\!\cdot\!n^*,\;p_l\!\cdot\!n^*) = (k\cdot n^*)^m,
\end{equation} with $i,j,l$ negative and $m$ positive or zero.
A conspicuous feature that we need to note
first of all, is that the dual vector $n_{\mu}^{*}$, when it appears at all,
it does so {\em always } and {\em only} in the numerators of the integrands.
And herein comes the first seemingly ``misterious'' facet of light-cone gauge.
How come that from a propagator expression like (\ref{prop}), which contains no
$n^*$ factors, can arise integrals of the form (\ref{struc}), with prominently
seen $n^*$ factors? Again, this is most easily seen in the framework of
definite external vectors $n$ and $n^*$. An alternative way of writing the
generic form of a light-cone integral is
\begin{equation}
\label{struc1}
I_{lc}^{\mu_1\cdots \mu_n}=\int
\frac{d^{D}k_{i}}{A(k_{j},\;p_{l})}\frac{g(k^{\mu_j}_{j},\;p_l^{\mu_l})}{h(k_{j}\!\cdot\! n,\;p_{l}\!\cdot\! n)},
\end{equation}
where the numerator $g(k^{\mu_j}_{j},\;p_l^{\mu_l})$ defines a tensorial
structure in the integral. For a vector, we have $k^{\mu}=(k^+,\;k^-,\;{\bf
k}^{\sc t})$, where $k^+=\lambda(k^0+k^{D-1})$ and $k^-=\lambda(k^0-k^{D-1})$
with $\lambda$ being a normalization factor. If we choose definite $n$ and
$n^*$ such that $n_{\mu}=(1,\;0,\;\cdots,\;1)$, and
$n^*_{\mu}=(1,\;0,\;\cdots,\;-1)$, this allows us to write $k^+\equiv
k\!\cdot\!n$ and $k^-\equiv k\!\cdot\!n^*$. We have therefore traced back the
origin for the numerator factors containing $n^*$. By the way, these terms have
nothing whatsoever to do with some kind of prescription input as it is in
(\ref{presc}) below. There, the $n^*$ factors are added into the structure {\em
by hand} via the {\em ad hoc} prescription.
\section{Prescriptions}
Now, the troublesome factors in the denominator, represented by
$h(k_{j}\!\cdot\!n,\;p_{l}\!\cdot\! n)$, for one-loop, two-point functions are
typically products of the form
\begin{equation}
\frac{1}{h(k_{j}\!\cdot\!n,\;p_{l}\!\cdot\! n)} \sim \frac {1}{(k\cdot n)\:(k-p)\cdot n}.
\end{equation}
In the standard approach, these are dealt with first by partial fractioning
them, a trick sometimes called ``decomposition formula'' (see, for example,
\cite{leib,leib-nyeo,leib-nyeo2})
\begin{equation}
\label{decomp}
\frac {1}{(k\!\cdot\! n)\:(p-k)\!\cdot\! n}\rightarrow \frac{1}{p\!\cdot\!
n}\left[\frac{1}{k\!\cdot\! n}+\frac{1}{(p-k)\!\cdot\! n}\right ]\:,
\:\:\:\:\:\:\:p\!\cdot\! n \neq 0
\end{equation}
It is easy to see that such kind of trick becomes very clumsy to deal with when
one has higher powers of denominator factors. In fact, the authors of
\cite{leib-nyeo2} agree that the application of (\ref{decomp}) complicates
considerably the calculation of such Feynman integrals. For example, consider,
\begin{equation}
\label{decomp1}
\frac {1}{(k\!\cdot\! n)^2\:(p-k)\!\cdot\! n}\rightarrow \frac{1}{(p\!\cdot\!
n)^2}\left[\frac{p\!\cdot\!n}{(k\!\cdot\!n)^2}+\frac{1}{k\!\cdot\! n}+\frac{1}{(p-k)\!\cdot\! n}\right ]\:,
\:\:\:\:\:\:\:p\!\cdot\! n \neq 0
\end{equation}
and so on and so forth. Yet, the standard approach based upon prescriptions,
cannot handle denominator products like these without first decomposing them.
As far as we know only Leibbrandt and Nyeo said explicitly before,
that ``decomposition formulas'' like (\ref{decomp}) and (\ref{decomp1}) are in
fact part of the prescription\cite{leib-nyeo2}. Be it as it is, after the
decomposition is made, {\em then and only then}, one seeks to handle the
isolated denominators calling for prescriptions that would turn them
manageable mathematically.
Early efforts in this direction draw heavily from the use of the Cauchy
principal value (CPV or PV for short),
\begin{equation}
PV \left(\frac{1}{q\!\cdot\! n}\right) \longrightarrow \frac{1}{2} \lim_{\epsilon\to 0} \left(
\frac{1}{q\!\cdot\! n +i\epsilon}+ \frac{1}{q\!\cdot\! n - i\epsilon} \right) .
\end{equation}
Unfortunately, this prescription, although preserves the general structure of
the light-cone integral (\ref{struc}), and is mathematically consistent, does
not provide physically acceptable results in the light-cone
gauge\cite{bass,bass2}.
Later on, independently Mandelstam\cite{mandelstam} and Leibbrandt\cite{leib2}
proposed new prescriptions to treat the gauge-dependent poles,
\begin{eqnarray}
\label{presc}
\frac{1}{q\!\cdot\! n} &\longrightarrow &\frac{1}{q\!\cdot\! n +i\epsilon \;
sgn (q\!\cdot\! n^*)}\nonumber \\
& = & \frac{q\!\cdot\! n^*}{(q\!\cdot \! n) \, (q\!\cdot \! n^*) +i\epsilon}.
\end{eqnarray}
where the former is the modified Mandelstam prescription and the latter is the
Leibbrandt's one. It can be proven that, in fact, they are completely
equivalent to each other, so that sometimes people refer to this as the
Mandelstam-Leibbrandt (ML) prescription. A conspicuous feature of these
prescriptions is that they introduce the factor $(q\cdot n^*)$ in the
denominator of integrals, thus modifying the original structure. This, however,
does not seem to affect the end result, as far as physically meaningful results
are concerned.
Applying the ML prescription for example, in the one-loop two-point function
integrals, double poles still appear but they cancel against each other and one
is left with physical poles only. In other words, those singularities that are
merely gauge artifacts cancel out when one uses this approach.
Experts\cite{bass2} argue, therefore, that adopting the ML prescription makes
the light-cone gauge acceptable, at least perturbatively. Along this line, some
two-loop calculations have been performed explicitly and can be found in the
pertinent literature\cite{twoloop}.
Pimentel {\em et al.} realized that when the physical principle of causality is
correctly taken into account --- either by careful, stepwise watching out for
it to be preserved along the whole computation, or by implementing it directly
onto the propagator, via causal considerations the same picture arises, namely,
unphysical, gauge-dependent poles cancel out leaving solely physical
ones\cite{pimentel,covariantization}. Actually, the procedure corresponds
exactly to taking out the zero-frequency mode from the Fourier series expansion
for the field operators, ensuring that only positive energy quanta propagate into
the future and vice versa. Mathematically this can be acomplished via
\begin{equation}
\frac{1}{(q\!\cdot\! n)^j} = PV\left(\frac{1}{(q\!\cdot\! n)^j}\right) -i\pi
\frac{(-1)^{j-1}}{(j-1)!} \delta^{(j-1)}(q\!\cdot\! n) \;sgn(q^0) ,
\end{equation}
where $PV$ stands for Cauchy principal value. The second term of this expression
is crucial. It is the term that guarantees that the troublesome zero-frequency
mode of the field quanta is subtracted out from the energy spectrum. It is
exactly tantamount to avoiding the pathological mixing of quanta of opposite
energy signs propagating into the future, thus violating causality.
By the very nature in which these prescriptions are introduced, they are
completely powerless to deal with products of light-cone poles without ere
using the partial fractioning trick. Moreover, in addition to the need to use
the ``decomposition formula'', the necessity to resort to prescriptions or even
clever maneuvers in order to extract out the culprit factor from the original
integral is an awkward trail to walk along to reach our objective. Because a
prescription is solely a prescription, a mere device to by-pass a problem,
prescriptions are usually inconvenient and undesirable. Yet, as awkward and
burdensome as it might be, there was no other way we could do things
properly in dealing with perturbative light-cone gauge computations.
\section{Negative dimensional approach}
Not so now, with the advent of NDIM\cite{halliday}. NDIM is a technique wherein
the principle of analytic continuation plays a key role. With it we solve a
``Feynman-like'' polynomial fermionic integral, i.e., a loop integral in
negative $D$-dimensional space with propagators raised to positive powers in
the numerator. Solutions arise as linear combinations of solutions for simple
systems of linear equations and these are then analytically continued to allow
for negative values of exponents (i.e. propagators now become raised to
negative powers, becoming denominator terms in the integrands) and positive
dimension\cite{flying}. This procedure, of course, when applied to the
light-cone case has to consider the general structure (\ref{struc}) above which
is characteristic of this gauge. That is, terms like $f(k_{j}\!\cdot\!
n^{*},\;p_{l}\!\cdot\! n^{*}) \sim (q\!\cdot\! n^*)^a[(q-p)\!\cdot\! n^*]^b$
which can appear in the numerator, evidently must remain there; therefore, no
exponents of such terms are to be analytically continued to allow for negative
values\cite{probing}. Schematically (see for instance second reference in
\cite{flying}),
\begin{equation}
\label{aci}
\int d^{D}k_{i}\;A(k_{j},\;p_{l})\;f(k_{j}\!\cdot\!
n^{*},\;p_{l}\!\cdot\! n^{*})\:h(k_{j}\!\cdot\! n,\;p_{l}\!\cdot\! n)\longrightarrow
^{\!\!\!\!\!\!\!\!\!\!\!\!AC}\int
\frac{d^{D}k_{i}}{A(k_{j},\;p_{l})}\frac {f(k_{j}\!\cdot\!
n^{*},\;p_{l}\!\cdot\! n^{*})}{\:h(k_{j}\!\cdot\!
n,\;p_{l}\!\cdot\! n)}
\end{equation}
where the left-hand side shows the negative dimensional integral to be
performed and the right-hand side displays the generic form (\ref{struc}) for
the light-cone integrals. Note that the factor $f(k_{j}\!\cdot\!
n^{*},\;p_{l}\!\cdot\! n^{*})$ remains in the numerator of the integrands in
both sides. The left-hand side of (\ref{aci}), which is evaluated via NDIM
methodology, is defined from projecting out powers of Gaussian type
$D$-dimensional momentum integrals of propagators\cite{halliday}. It is worth
mentioning here that the usual Schwinger exponentiation of propagators for the
factor $[h(k_{j}\!\cdot\! n,\;p_{l}\!\cdot\! n)]^{-1}$ in positive dimensional
calculation does not work for light-cone integrals \cite{covariantization}.
Now, one could rightfully ask: how can one get the standard ML results for one
and two loop ML dimensionally regulated Feynman-integrals with the procedure of
negative dimensions? Substitute (\ref{exemplo}) in (\ref{struc}), that is,
consider the integral,
\begin{equation}
B(i,j,l,m) = \int d^D\!q\;
(q^2)^i\left[(q-p)^2\right]^j(q\cdot n)^l (q\cdot n^*)^m.
\end{equation}
Let us evaluate it in great detail following the steps thoroughly described in
our previous papers\cite{flying}. Let our starting point be,
\begin{eqnarray}
\label{Gaussian}
G &=& \int
d^D\!q\; \exp{\left[ -\alpha q^2 - \beta (q-p)^2 -\gamma q\cdot n -\theta
q\cdot n^*\right]}\nonumber\\ && = \left(\frac{\pi}{\lambda}\right)^{D/2}
\exp{\left[-\frac{1}{\lambda} \left( \alpha\beta p^2 + \beta\gamma p\cdot n +
\beta\theta p\cdot n^* -\frac{1}{2}\gamma\theta n\cdot
n^*\right)\right]},
\end{eqnarray}
where $\lambda=\alpha+\beta$. Taylor
expanding the exponentials above and solving for the integral $B(i,j,l,m)$ we
obtain,
\begin{eqnarray} B(i,j,l,m) &=& (-\pi)^{D/2} i!j!l!m!(-\sigma-D/2)!
\sum_{\{X,Y\}=0}^\infty
\frac{(p^2)^{X_1}(p^+)^{X_2} (p^-)^{X_3}}{X_1!X_2!X_3!X_4!Y_1!Y_2!}
\left(\frac{-nn^*}{2}\right)^{X_4}\nonumber\\
&&\times \delta_{a,i}\delta_{b,j}\delta_{c,l}
\delta_{d,m}\delta_{e,\sigma} ,\end{eqnarray}
where $\sigma=i+j+l+m+D/2$, $a=X_1+Y_1$, $b=X_1+X_2+X_3+Y_2$, $c=X_2+X_4$,
$d=X_3+X_4$, $e=X_1+X_2+X_3+X_4$. This system of linear algebraic equations
has six possible solutions -- since the number of unknowns is bigger than the
number of equations. In all six cases eliminating the deltas leaves us with
one remaining sum, a hypergeometric series $_3F_2$. Three out of these six
series have as its variable,
$$ z=\left(\frac{p^2 nn^*}{2p^+ p^-}\right),$$
while the other three have as variable $$
w \equiv z^{-1}=\left(\frac{2p^+ p^-}{p^2 nn^*}\right) .$$
In our previous work \cite{probing}, we have considered the simplest one-loop
integral in the light-cone gauge, evaluating it via NDIM method, with one- and
two-degree violation of Lorentz covariance (more specifically, with $n_\mu$
only and with both $n_\mu$ and its dual $n_\mu^{*}$). For the first case we
obtained the usual PV prescription result, whereas in the second case we
obtained the Mandelstam-Leibbrandt prescription result. However, that toy
integral is easily evaluated in the NDIM scheme, since there are no left over
summation indices, that is, the number of linear equations in the system equals
the number of unknowns and the result can be expressed just in terms of product
of gamma functions. The integrals we consider here are more complicated; they
generate several linearly independent and dependent solutions which need to be
sorted out carefully. That is, the space of solutions spanned by base functions
are not all linearly independent ones. This is characteristic of these
particular cases of integrals we are considering here, and this is the reason
we treat them at lenght and in depth.
From our previous works \cite{flying} we know that the Feynman integral will
be represented by a linear combination of linearly independent series.
Moreover, these representations are related through analytic continuation. Let
us consider the representation of $B(i,j,l,m)$ in terms of $z^{-1}$,
\begin{equation} B(i,j,l,m) = B_1 + B_2 ,\end{equation}
where
\begin{eqnarray} B_1 &=& (-\pi)^{D/2}\frac{\Gamma(1+j)\Gamma(1+l)
\Gamma(1+m)(p^2)^i(p^+)^{j+l+D/2} (p^-)^{j+m+D/2}}{\Gamma(1+j+l+D/2)
\Gamma(1+j+m+D/2) \Gamma(1-j-D/2)}\nonumber\\
&&\times\left(\frac{-nn^*}{2}\right)^{-j-D/2} \,
_3F_2(a_1,b_1,c_1;e_1,f_1|w) ,\end{eqnarray}
and
\begin{eqnarray} B_2 &=& (-\pi)^{D/2}\frac{\Gamma(1+i)\Gamma(1+j)
\Gamma(1+m)\Gamma(1-\sigma-D/2) (p^2)^{\sigma-m}
(p^-)^{-l+m}}{\Gamma(1-j-l-D/2) \Gamma(1-i-m-D/2)
\Gamma(1-l+m)\Gamma(1+\sigma-m)} \nonumber\\
&&\times\left(\frac{-nn^*}{2}\right)^{l}
\, _3F_2(a_2,b_2,c_2;e_2,f_2|w) ,\end{eqnarray}
where $a_1=-i$, $b_1=D/2+j$, $c_1=\sigma+D/2$, $e_1=1+j+l+D/2$,
$f_1=1+j+m+D/2$, $a_2=-l$, $b_2=-i-j-l-D/2$, $c_2=i+m+D/2$, $e_2=1-l+m$,
$f_2=1-j-l-D/2$ and $_3F_2$ is the generalized hypergeometric
function\cite{luke}. This is the result in the negative dimension region and
$i,j,l,m$ positive. Now, in order to be physically meaningful it must be
analytically continued to positive dimension and $i,j,l$ negative, while the
exponent $m$ must be kept untouched. To carry out this analytic continuation
one must rewrite the gamma factors as Pochhammer symbols and use one of its
properties \begin{equation}
(a|j) \equiv (a)_j = \frac{\Gamma(a+j)}{\Gamma(a)}, \;\;\;\; (a|-j) =
\frac{(-1)^j}{(1-a|j)}. \end{equation}
Performing these simple operations one gets,
\begin{eqnarray}
B^{AC}(i,j,l,m) &=& \pi^{D/2}
\frac{(-j|-l-D/2)(-l|j+l+D/2)}{(1+m|j+D/2)} (p^2)^i (p^+)^{j+l+D/2}\nonumber\\
&&\times(p^-)^{j+m+D/2}\left(\frac{nn^*}{2}\right)^{-j-D/2}\,
_3F_2(a_1,b_1,c_1;e_1,f_1|w) \nonumber\\
&&+\pi^{D/2}
\frac{(-j|i+j+m+D/2)(-i|i+j+l+D/2)}{(1+m|-l)}\nonumber\\
&&\times (\sigma+D/2| -2\sigma-D/2+m) (p^2)^{\sigma-m}
(p^-)^{-l+m} \left(\frac{nn^*}{2}\right)^{l}\nonumber\\
&&\times \,_3F_2(a_2,b_2,c_2;e_2,f_2|w). \end{eqnarray}
In fact, from the theory of hypergeometric functions\cite{luke}, we know that the hypergeometric differential equation for
$_pF_q$ has up to $p$ linearly independent solutions, so, in writing
down $B(i,j,l,m)$ we would expect it to contain up to three terms. Indeed,
among the three series of funtions $_3F_2(\cdots|w)$ there is a third
term, namely,
\begin{eqnarray} B_3 &=&
(-\pi)^{D/2}\frac{\Gamma(1+i)\Gamma(1+j)
\Gamma(1+l)\Gamma(1-\sigma-D/2)(p^2)^{i+j+m+D/2}
}{\Gamma(1+i+j+m+D/2)\Gamma(1+l-m) \Gamma(1-j-m-D/2)
}\nonumber\\
&& \times \frac{(p^+)^{l-m}}{\Gamma(1-i-l-D/2)}
\left(\frac{-nn^*}{2}\right)^{m}\,_3F_2(a_3,b_3,c_3;e_3,f_3|w) ,
\end{eqnarray}
where $a_3=-i-j-m-D/2$, $b_3=-m$, $c_3=i+l+D/2$, $e_3=1+l-m$,
$f_3=1-j-m-D/2$. However, this solution is not a linearly independent one.
Indeed, from the fact that $m$ is always positive or zero we can
rewrite $B_3$ using eq.(6), section 3.2 of ref.\cite{luke} --- if we choose
$\alpha_1=c_3$, $\alpha_2=a_3$, $n=-e_3$, $\rho_1=f_3$ --- as a linear
combination of $B_1$ and $B_2$.
The integral we are considering here was studied by Lee and Milgram\cite{lee}. Our result
can also be written in terms of the so-called $G-$function or Meijer's
function and agrees with their calculation. Taking the special case where $i=j=l=-1$
and $m=0$ one gets,
\begin{eqnarray}
B(-1,-1,-1,0) &=& -(-\pi)^{D/2}\frac{\Gamma(2-D/2)\Gamma(D/2-1)}{\Gamma(D/2)}
\frac{(p^+)^{D/2-2}(p^-)^{D/2-1}}{p^2}\nonumber\\
&&\times \left(\frac{-2}{nn^*}\right)^{D/2-1}\, _2F_1(1,D-3;D/2|w)
-\pi^{D/2}\Gamma(D/2-1)\nonumber\\
&&\times\frac{\Gamma(D/2-2)\Gamma(3-D/2)}{ \Gamma(D-3)} (p^2)^{D/2-3}p^-
\left(\frac{-2}{nn^*}\right)\nonumber\\
&&\times \, _2F_1(1,D/2-1;2|w) .\end{eqnarray}
Observe that when $D=4-2\epsilon$ there are simple poles in these two terms
but they cancel since, $$ \lim_{\epsilon\to 0}
\Gamma(\epsilon)+\Gamma(-\epsilon) = {\cal O}(1). $$
To extract the finite part one can proceed in the same way as we did in the second
paper of ref.\cite{flying}. The other representation for the integral
$B(i,j,l,m)$ also coincide with that presented in \cite{lee}. In fact, they
are related through analytic continuation (see ref.\cite{luke} for such
formulas) -- as all representations provided by NDIM for Feynman integrals in
covariant\cite{flying} and non-covariant gauges\cite{probing}.
Consider now an integral containing two factors of the form $(k\cdot n)$,
\begin{equation} T_2(i,j,l,m) = \int d^D\! q \, (q-p)^{2i}(q\cdot n)^j
\left[(q-p)\cdot n\right]^l (q\cdot n^*)^m ,\end{equation}
where the exponent $m\geq 0$. The standard approach makes use of decomposition
formulas (\ref{decomp}) and (\ref{decomp1}). On the other hand, NDIM can solve
all integrals of such kind at the same time, giving in positive
dimension\cite{prescriptionless},
\begin{eqnarray}
T_2(i,j,l,m) &=& \pi^{D/2} \left(\frac{2p^+ p^-}{n\cdot n^*}\right)^{D/2+i}
\frac{\Gamma(i+l+D/2) \Gamma(1+m)}{(p^+)^{-j-l}(p^-)^{-m}\Gamma(-i)
\Gamma(-j)}\nonumber\\
&&\times \frac{\Gamma(-i-j-l-D/2)}{\Gamma(1+i+m+D/2)} ,\end{eqnarray} observe
that the Pochhammer symbol which contain $(1+m)$ was not analytically
continued since $m$ must be either positive or zero.
\section{Conclusion}
Finally, in concluding this work we would like to emphasize some important
points concerning the application of NDIM technique to light-cone integrals.
First of all, why one needs to define the original Gaussian integral
(\ref{Gaussian}) with two-degree violation of Lorentz covariance in order to get
``causality'' preserving results? The answer is related to the very definition
of light-cone gauge condition where one has the external four-vector $n_\mu$
such that it is chosen to be light-like, i.e., $n^2=0$. However, the
``light-likeness'' condition $n^2=0$ does not uniquely define the needed
external vector $n_\mu$ to implement the gauge condition. The reason for this
is more easily seen considering a particular case where
$n_\mu=(n_0,\,n_3,\,0,\,0)$, in which case the condition $n^2=0$ gives as
solutions either $n_0=+n_3$, or $n_0=-n_3$ with $n_0>0$. Therefore, the
components of the ``light-like'' vector are not linearly independent; hence the
two possibilities: either $n_\mu\equiv (n_0,\,+n_3,\,0,\,0)$ or $n_\mu^*\equiv
(n_0,\,-n_3,\,0,\,0)$. These peculiar light-cone properties have been shown by
G. Leibbrandt \cite{NP} to be connected to the Newman-Penrose tetrad formalism
in the context of gravitation and cosmology, where a four-dimensional basis is
spanned entirely by null vectors. In his work, Leibbrandt demonstrated that the
two-dimensional vector sub-space $(n_0,\,n_3)$ {\em cannot} be spanned solely
by the vector $n_\mu$; it is not sufficient for this purpose because this
vector possesses linearly dependent components. A thorough discussion,
including not only the sub-space $(n_0,\,n_3)$, but the entire four-dimensional
space, is found in the reference above given.
This is the reason why our NDIM calculation done in \cite{probing} with only
one degree violation of Lorentz covariance failed to be ``causal'', reproducing
the well-known ``causality'' breaking PV result. The cure for this pathological
result could only be achieved by introducing the needed additional $n_\mu^*$
vector, so that the basis for spanning the entire four-dimensional vector space
be unique and well-defined. Therefore our calculations here always included
this needed dual vector $n_\mu^*$, so that our results are well-defined and
unambiguous.
This NDIM technique is therefore the most powerful and beautiful machinery
ever to come to front until now to handle the light-cone Feynman integrals.
All the desirable features are embedded in it and we can list them as follows:
{\em i)} No ``decomposition formula'' like (\ref{decomp}) is needed to
separate gauge dependent poles before the integral is evaluated; {\em ii)} It
preserves the general structure of the light-cone integral, namely, no factors
containing $n^*$ are introduced in the denominators; {\em iii)} No
prescription is needed to handle $(q\cdot n)^{-\alpha}=0$ type singularities
to solve the integral; {\em iv)} There are no parametric integrals to perform;
{\em v)} There is no need to split the dimensionality of space-time to work
out integrations over component sub-spaces like in \cite{leib-nyeo}; {\em vi)}
Results are obtained for arbitrary {\it negative} exponents of propagators
{\em and} gauge-dependent poles, so that special cases -- which
agree\cite{leib,bass,probing,prescriptionless} with the ones calculated using
ML prescription -- are contained in them; {\em vii)} Results are always within
the context of dimensional regularization, i.e., preserving gauge symmetry.
The power of NDIM is therefore readily apparent. If one remembers how integrals
over Grassmann variables are introduced, he/she can recall that the underlying
property employed to define them is translation invariance. It is an outstanding
thing that this sole property in the fermionic integration is able to guarantee
a prescriptionless light-cone and more important than that, a causality
preserving method of direct computation of Feynman integrals.
Of course, like in any gauge, two-loop integrals are more demanding when
compared to the one-loop case, and specially in the light-cone gauge.
Preliminary results we have show that NDIM can also handle them with more
easiness than in the usual standard approaches.
\section*{Acknowledgements}
A.G.M.S. gratefully acknowledges FAPESP (Funda\c c\~ao de
Amparo \`a Pesquisa do Estado de S\~ao Paulo, Brazil) for financial support.
|
\section{Experiments}
Fifty-mg samples of Li$_{1-x}$Ni$_{1+x}$O$_2$ powder, prepared as reported in
Ref. \cite{Delmas-prep}, were placed in Pyrex tubes,
$4$ mm in diameter, filled with Helium gas. The amount of extra Nickel
was previously
determined by Rietveld analysis in the $\bar{R3m}$ crystallographic space group.
The magnetic measurements were carried out using an Oxford Instrument Maglab$2000$
cryostat equipped
with a $7$~T superconducting magnet and a variable temperature insert
allowing temperature scans between $1.8$ and $450$ K. During experiments
at constant field, the magnet was set in the persistent mode.
The measurement system comprises an excitation coil for ac susceptibility, with two inner
compensated coils, used either for ac susceptibility or for dc moment measurements.
The dc moment and ac susceptibility signals are measured respectively with a
Keithley $2001$ voltmeter and with a SR$830$ lock-in-amplifier.
The geometry of the system allows a sensitivity of $10^{-5}$ (dc moment) and $10^{-8}$ emu
(ac susceptibility).
The static properties were first investigated under external
field of $2$ and $5$ T as a function of $x$.
These field values are sufficiently large to suppress the ferromagnetic domains
due to dipole-dipole interactions and to make negligible the effect of magnetic anisotropy
which reaches $2.5$~T in this compound \cite{Barra}. In these
conditions, the system is only subject
to local exchange interactions in which we are interested.
The data were corrected for demagnetizing effects using
a model of spherical thin grains. The correction is only significant below $100$ K
for the largest $x$ values and does not exceed $5\%$ at $3$ K.
In Fig. \ref{mag_lab}, we represent the magnetization curves
obtained under $2$ T. It is clearly seen that
the magnetic response of the Li$_{1-x}$Ni$_{1+x}$O$_2$ family is strongly influenced
by the amount of extra Nickel ions. The important point is that the topology of the
experimental curves changes with $x$. This suggests that, as $x$ changes,
the system undergoes
a transition between distinct macroscopic magnetic behaviors probably related
to Ni clusters formation. Indeed, the lowest ($x<0.14$) and
highest ($x>0.14$) concentrations present respectively superparamagnetic and
ferromagnetic like behavior.
To illustrate and quantify this point we measured the ac susceptibility
of the samples at a
frequency of $10$ kHz, with an excitation field of $3$ Oe. In Fig. \ref{khi_lab}, we
show the data obtained for $x=0.02$, $x=0.14$ and $x=0.20$. The susceptibility curves
display maxima which are characteristic for ferromagnets at temperatures of
$116$K and $168$K
respectively for $x=0.14$ and $x=0.20$. In contrast, for $x=0.02$, the
temperature variation
of $\chi_{ac}$ is rather monotonic. These observations confirm the
superparamagnetic behavior of
compounds with low extra Nickel concentrations, in agreement with previous neutron
experiments \cite{Hira}.
The second point concerns the magnetization of the different samples at
the lowest temperatures.
Assuming that all the magnetic moments in Li$_{1-x}$Ni$_{1+x}$O$_2$ are
parallel, the magnetization
should saturate to:
\begin{equation}
M_s = \left( 1-x \right) N \mu _B g_1 S_1 + 2xN\mu_Bg_2 S_2,
\end{equation}
$N , \mu_B, g$ and $S$ denoting respectively the Avogadro's constant,
Bohr's magneton, Land\'e's factor and spin value.
The first term represents the contribution of Ni$^{3+}$ ions in the
$(111)_{Ni}$ planes while the
second one is related to Ni$^{2+}$ ions in $(111)_{Li}$
and $(111)_{Ni}$ planes. This saturation should be reached when the Zeeman energy exceeds
the interplane antiferromagnetic coupling.
This cannot be reasonably the case for the moderate magnetic fields involved in this work.
In Fig. \ref{mag_satur}, we plot the magnetization measured at $4.3$ K, as a function of $x$,
together with $M_s(x)$ (Eq. 1).
We can see that the maximum saturation is not reached.
This is in good agreement with previous results
of Rougier {\em et al.} \cite{these} who reported the absence of saturation
for $x=0.02$, even at $11$ T. In addition, our results show that the measured magnetization
presents a maximum at $x \simeq 0.14$.
In the light of the experimental results, it is difficult to conclude about the exact
nature of the magnetic order. In the next section, we propose a simple model to investigate
this point further.
\section{Model}
Our model incorporates the different magnetic interactions described in Section I.
The main ones ($J_\perp$ and $J_\perp'<0$ ) are the $180^\circ$ Ni--O--Ni
super-exchange
coupling one Ni ion sitting in a $(111)_{Li}$ plane to its six NNs
sitting in the adjacent $(111)_{Ni}$ planes. It is reasonable to assume that these two coupling
constants are of the same order of magnitude. In the following we suppose
$J_\perp=J_\perp'< 0$.
It is obvious that changing the sign of $J_\perp$ does not modify the magnetization, the
susceptibility and other related quantities, since there
is a compensation between the magnetic moments of the two Ni$^{2+}$ ions.
We also consider the $90^\circ$ Ni--O--Ni intra-plane interaction $J_\parallel$.
We make no {\em a priori} assumption about the sign of $J_\parallel$.
Recent neutron diffraction studies \cite{Pouillerie} on this material family
($x \leq 0.20$) have shown that there is no
Lithium in the Nickel planes.
We verified this point with the help of the lattice gas
structural model introduced by Reimers {\em et al.} \cite{Reimers}.
For a composition $x=0.2$, which is the highest considered here, the quantity
of Li ions in the
$(111)_{Ni}$ planes is not significant (less than $2\%$) and can be reasonably
neglected.
Furthermore, we suppose that the excess Ni ions are randomly distributed
in the $(111)_{Li}$
planes.
Since we assumed that the excess Ni ions are antiferromagnetically coupled
to their six NNs,
three-dimensional magnetic Ni
clusters are formed in the system, as shown in Fig. \ref{Struct} (for the time being,
we neglect intra-plane coupling).
For $x<x_c$ all these clusters have a finite size. At the critical
concentration $x_c$ the
largest cluster percolates and for higher concentrations it grows in size.
This percolation transition strongly affects the
magnetic properties. One may expect different types of behaviors in samples
with $x$ below, at,
or above $x_c$.
Numerical simulation gave $x_c\simeq 0.136$ for the percolation
threshold. The cluster mass distribution $n_s$ was
calculated at $x_c$ : it follows
a power law $n_s \propto s^{-\tau}$ with a critical exponent $\tau = 2.19 \pm 0.01$.
This last
result indicates a regular three-dimensional percolation transition \cite{JM},
as was predicted
in a dedicated
study of percolation in multi-layered structures \cite{Dayan}.
We note that a small amount of Nickel impurities can give rise to large clusters
because each impurity is
coupled to six other Ni ions. This is illustrated in Fig. \ref{Clu2} where
we represent the Ni ions distribution
in a sample with $x=0.02$. For such a low value of $x$, most of the magnetic
clusters contain seven ions,
but larger clusters coupling three successive Ni layers are also observed.
In order to calculate the magnetic properties, we use an Ising model in which
each Ni ion has a spin $\sigma=\pm 1$.
The choice of an Ising-type Hamiltonian is justified by intrinsically high
magnetic anisotropy of Ni clusters in LiNiO$_2$ \cite{Barra}, and {\em a posteriori} by
the good agreement found between the
numerical results and the experimental data. In our model, we neglect the differences
between the Ni ions. Indeed, the $(111)_{Ni}$ planes are mostly composed of N$i^{3+}$ ions
in a low spin state $S=1/2$
while extra Nickels are Ni$^{2+}$ ions with $S=1$ \cite{Azzoni}. Two types of spin can be
easily taken into account in the model, but we checked that it is not a
crucial point, probably because
we consider small concentrations of extra Nickel ions.
The Hamiltonian then simply reads:
\begin{equation}
H = -g\mu_B B \sum_i \sigma_i - J_\perp \sum_{<i,j>} \sigma_i \sigma_j - J_\parallel
\sum_{<i,j>} \sigma_i \sigma_j
\end{equation}
The first term describes the interaction with the magnetic field B.
In the second term, $\sigma_i$ sits on a Ni in a $(111)_{Li}$ plane,
and the $\sigma_j$'s sit on the six NN Ni ions in the adjacent $(111)_{Ni}$ planes.
In the last term, the sum runs
over all NN pairs in the same plane.
For a given spin configuration, the Boltzmann factor is:
\begin{equation}
\exp \left[ {K \left( b\sum_i \sigma_i
+ {\rm sign}(J_\perp) \sum_{<i,j>} \sigma_i \sigma_j + j_\parallel
\sum_{<i,j>} \sigma_i \sigma_j \right) } \right]
\end{equation}
where $b=g\mu_B B/\left|J_\perp\right|$, $K=\left|J_\perp\right|/kT$,
and $j_\parallel=J_\parallel/\left|J_\perp\right|$ are respectively
the dimensionless field, the inverse temperature, and the ratio
of intra- to inter-plane coupling constants. \\
The classical "single spin flip" Monte Carlo algorithm is not well suited
to simulate this system
because of critical slowing down around the transition and at low
temperatures \cite{MC}.
To avoid very long simulation times, we adapted Wolff's algorithm \cite{Wolff}
to our system. This method largely compensates the correlation length
increase when
the temperature is lowered because the system evolves by flipping clusters
of ordered spins.
One Monte Carlo step (MCS) consists in constructing a Wolff cluster and then
trying to flip it.
In our case, the construction of a cluster is made as follows: we randomly
choose a first spin $\sigma_a$.
All the inter-plane bonds (if any) between $\sigma_a$ and the
closest spins $\{\sigma_b\}$
are examined in turn.
Those with $\sigma_a=\sigma_b$ are discarded; for the other bonds,
spin $\sigma_b$ is added to the cluster with an
arbitrary probability $p$. Now set $\{\sigma_a\}$ is replaced
by set $\{\sigma_b\}$ and the procedure
is repeated until an empty set is found.
Choosing $p=1-e^{-2K}$, as in the classical Wolff algorithm,
the expression for the flipping
probability satisfying the detailed balance condition simplifies to
\begin{equation}
\min \left( 1, \exp \left[-2 K\left( b M_c + j_\parallel
\Gamma_c \right)\right] \right)
\end{equation}
with
\begin{equation}
M_c=\sum_{cluster} \sigma_i
\end{equation}
and
\begin{equation}
\Gamma _c = \sum_{<i,j>_{surf}} \sigma_i \sigma_j
\end{equation}
The last sum runs over all the pairs of spins at the cluster surface,
$\sigma_i$ being part of the cluster and $\sigma_j$ at the exterior.
The exponential argument is the energy cost to flip the cluster: the
first term is due to the field, the second to
the intra-plane interactions at the cluster surface.
We considered systems of about $3000$ spins (9 pure $(111)_{Ni}$ planes,
each
containing $18 \times 18$ Ni ions, and $x\%$ excess Ni in the $(111)_{Li}$
planes).
Periodic boundary conditions
were applied in the three directions. To check for finite size effects, we
simulated a bigger system of approximately 30000 spins without significant
changes in the
results. For a concentration $x$, a
spatial configuration was built up by randomly placing Ni excess ions in
the Li planes.
At each temperature, ranging
from $300$ K down to $20$ K,
we first equilibrated the spin system through about half a million MCSs and
collected
thermodynamic quantities
during another million MCSs. The same operation was repeated for several
samples with different spatial configurations.
Quantities such as magnetization $M$ and susceptibility $\chi$ , were finally
obtained by averaging over the different samples.
We first present the magnetization curves for samples with $x=0.02,
0.14$ and $0.2$, in
an external field $B=5$~T. The numerical results are given in Fig. \ref{MagB5},
together with the experimental ones.
The value of the intra-plane interaction $J_\parallel$ is fixed to zero,
while the antiferromagnetic inter-plane
coupling is equal to $-110 $ K. This value, adjusted to obtain the best
agreement between our model
and the experimental data, is reasonably close to the values proposed
in the literature for the $180^\circ$ Ni--O--Ni interaction \cite{NiO}.
This estimate of $J_\perp$ rules out the possible presence
of oxygen holes at these concentrations, since, as discussed above, this
would give
a much larger inter-plane interaction.
The model reasonably reproduces the experimental magnetization curves,
especially in the high temperature region. We can interpret our results in
the following way.
For $x=0.02$ the Ni clusters
do not percolate and are thus "magnetically isolated": they are at the origin of
the superparamagnetic behavior observed
both numerically and experimentally. For an extra Nickel concentration above the
critical value ($x_c\simeq 0.136$), a magnetic
cluster spans the sample. When the temperature is decreased, this magnetic cluster
undergoes a second order transition
to a ferrimagnetic state. Since the external field is high, the transition is
broadened on the $x=0.14$ and $x=0.2$
curves. We also calculated magnetization curves for the same concentrations but
without
external magnetic field. In this case, the sample with $x=0.02$ has no
magnetization,
as expected for a
superparamagnetic system under zero applied magnetic field. On the other hand,
in the samples with $x>x_c$, the
transition is well marked with the appearance of a spontaneous magnetization
at low temperature.
From the $M(T)$ curves, it is possible to extract values for the temperature $T_o$
locating the onset of magnetic ordering.
For $x>x_c$, we first locate the inflexion point $I$. We then draw the line
tangent to the curve at point $I$, and define $T_o$
as the intercept with the $T$ axis. For $x<x_c$, there is no longer an inflexion point,
since the system is superparamagnetic. However, we can still estimate the onset
of magnetic ordering
inside the isolated clusters. To do so, we define $T_o$ as the temperature
at which magnetization reaches
$7$ percent of its saturation value. For $x>x_c$, this definition approximately
gives the same $T_o$ as the inflexion point method. In Fig. \ref{Tcrit},
we represent
the onset of magnetic ordering temperatures, $T_o$ deduced from
the experimental and numerical $M(T)$ curves under an applied field of $2$ T.
The good agreement confirms that our simple model permits an accurate description
of the onset of magnetic ordering, thus providing information for the estimate
of $x$.
Let us now examine more closely the low temperature magnetization.
The disagreement between the experimental and numerical results can be traced
to the different approximations made in the model. At first,
the choice of Ising spins is certainly too crude and it affects more
and more the shape of the calculated magnetization curve as $T$ is decreased. Secondly,
we did not
yet introduce the intra-plane interaction, known to be weaker than
the inter-plane one, and to influence the
magnetization at lower temperatures. Assuming a strong AF inter-plane
coupling $J_\perp$ and a weak intra-plane coupling
$J_\parallel$, we still have to discuss the sign of the intra-plane coupling.
We consider two different magnetic subsystems: the first one is composed
of spins belonging to the ferrimagnetic clusters and the second one contains all the
remaining spins which lie in the $(111)_{Ni}$ planes. We denote $\beta(x)$ the
proportion of spins belonging to the clusters. We calculated
$\beta(x)$ numerically for $x$ ranging between $0$
and $0.2$: the results are well fitted by the polynomial $\beta(x)= 7x -
22x^2 + 40.7x^3$.
Assuming a ferromagnetic coupling, $J_\parallel \geq 0$, at low temperature all
the Ni$^{3+}$ ions
in the $(111)_{Ni}$ planes adopt the field direction while the
extra Ni$^{2+}$ spins
adopt the opposite one. As one can see on Fig. \ref{Struct}(b) the
spin $S=1$ on the extra Ni$^{2+}$ is exactly compensated by one of his
NNs having a spin $S=1$ too.
One can then expect a maximum ferrimagnetic magnetization
\bq
\frac{M_1}{N \mu_B g_1 S_1}=\left(1-x\right).
\eq
On the opposite, an AF coupling, $J_\parallel < 0$, leads to frustration
in the second subsystem. In the case where this frustration is large enough to cancel
the corresponding
magnetization, only the ferrimagnetic clusters contribute to the low
temperature magnetization.
One then obtains
\bq
\frac{M_2}{N \mu_B g_1 S_1}=\left[\left(1+x\right)\beta\left(x\right)-2x\right] .
\eq
In Fig. \ref{mag_satur} we plot the two curves corresponding to $M_1(x)$ and $M_2(x)$
together with the experimental results for $B\!=\!2$ T at $T\!=\!4.3$ K.
Since temperature is too high and magnetic field too low
to ensure perfect saturation of the different subsystems,
neither $M_1$ nor $M_2$ fit the experimental data on the whole range. We can however
qualitatively discuss the two limits. $M_1(x)$ decreases linearly with $x$, in
contradiction
with the increase of magnetization observed at low $x$. Conversely, the $M_2(x)$
curve approximately follows the experimental data points up to a systematic vertical shift.
Since Eq. (8) assumes that the second subsystem has no magnetization,
this is consistent with an AF intra-plane coupling ($J_\parallel < 0$).
We thus performed additional simulations to examine the effect of the
intra-plane interaction $J_\parallel$. Our results
show that, for a ferromagnetic coupling ($J_\parallel \geq 0$), the magnetization
curve changes concavity around $20$ K to reach the expected saturation value $M_1$ as
$T\rightarrow 0$ K (Fig. \ref{In-plane}). For an AF coupling, ($J_\parallel<0$),
the curves at low temperatures
behave much more like the experimental ones: saturation
is not reached because the AF coupling induces a frustrated state in the triangular
$(111)_{Ni}$ planes. The best qualitative agreement between simulations and
experiments is found for
$J_\parallel\simeq -1.5 $ K. Of course, this rough estimate of
$J_\parallel$ is obtained for
a given value of $J_\perp$, determined from the high temperature
behavior of the model with
$J_\parallel=0$. For a better estimate, one would have to take
both $J_\perp$, $J_\perp'$ and $J_\parallel$
as free parameters, simulate several magnetization curves and
fit them to the experimental data.
\section{Summary and discussion}
In summary, we introduced a simple model to investigate the
magnetic properties of non stoichiometric Li$_{1-x}$Ni$_{1+x}$O$_2$
layered compounds. The formation of three-dimensional (3D) Ni clusters
around the extra Nickel ions was simulated by simple percolation of the Ni
ions in the direction perpendicular to the metal layers.
An Ising Hamiltonian was then used to compute the magnetic properties
of the disordered system made of both 3D Ni clusters (magnetic
coupling $J_\perp$) and 2D intra-plane Ni clusters (magnetic
coupling $J_\parallel$). The transition from a superparamagnetic
to a ferrimagnetic behavior, observed experimentally as $x$ increases,
is interpreted in our model as the onset of percolation for the 3D
Ni clusters, at a critical threshold $x_c$. Our numerical estimate
$x_c\simeq 0.136$, agrees nicely with the experimental evidence of
magnetic ordering at $x=0.14$.
The best fit of our model to the experimental
data is found for $J_\parallel=-1.5K$ and $J_\perp=-110K$. The assumption that
$|J_\perp| \simeq |J_\perp'|$ is validated by
the good agreement between the model and the experiments. The order of
magnitude found for $J_\perp$ allows us to confirm magnetic coupling through
the Ni-O-Ni super-exchange interactions.
The possibility of charge compensation on Oxygen
proposed for $x$ close to $1$ \cite{Kuiper}
is thus inadequate for the low $x$ values considered in this paper.
Another important outcome of our
work is the relation found between the ordering temperature $T_o$
and the concentration $x$ of excess Nickel, which offers
a new way to determine $x$ precisely. In fact the accurate
determination of $x$ was overlooked for a long time.
If new techniques \cite{Delmas-prep} permit now to determine
the proportion of extra Ni ions with a reasonable accuracy, one should
certainly be more careful about
older results for stoichiometric LiNiO$_2$, which very likely concern
samples with a non zero effective value of $x$.
Moreover, Bajpai {\em et al.} \cite{Bajpai} have recently reported the coexistence
of ordered and random phases in the same sample, the proportion of each phase
depending on the heat treatment during elaboration. The coexistence of
two distinct phases may very well explain the anomaly observed at 240K
in the ac susceptibility of some samples, a point currently
investigated \cite{Andre_Nico}.
Finally, the antiferromagnetic nature of $J_\parallel$ implies 2D magnetic
frustration because the Ni planes have a trigonal symmetry. The magnetic properties
of the system at very low temperatures ($T<\vert J_\parallel \vert$)
would thus probably be better described in a 2D model with a random distribution
of quenched spins due to the intra-plane coupling. We plan to investigate this
point in the near future. \\ \\
*e-mail: <EMAIL>
|
\section{\bf Introduction}
\hspace{1cm}
The null (tensionless) $p$-branes correspond to usual $p$-branes with their
tension $T_p$ taken to be zero. This relationship between null
$p$-branes and the tensionful ones may be regarded as a
generalization of the massless-massive particles correspondence.
The dynamics of the null branes in curved backgrounds is
interesting also as a generalization of the motion of null
strings in such backgrounds \cite{RZ}, \cite{Kar}, \cite{DabLar},
\cite{PorPap}, \cite{KKP}.
A model for null $p$-branes living in Friedmann-Robertson-Walker
space-time (with ${\it k}=0$) was proposed in \cite{RZh}. The
motion equations were solved and it was shown there that an ideal
fluid of null $p$-branes may be considered as a source of gravity.
Here we investigate the classical evolution of the null branes in
a curved background. In Sec. 2 we give the corresponding
Lagrangian formulation. In Sec. 3 the BRST-BFV approach in its
Hamiltonian version is applied to the considered dynamical system.
In Sec. 4 some exact solutions of the equations of motion and of
the constraints for the null membrane in general stationary,
axially symmetrical, four dimensional, gravity background are
found. The examples of Minkowski, de Sitter, Schwarzschild,
Taub-NUT and Kerr space-times are considered in Sec. 5. Sec. 6 is
devoted to comments and conclusions.
\section{\bf Lagrangian formulation}
\hspace{1cm}
The action for the bosonic null $p$-brane in a D-dimensional curved
space-time with metric tensor $g_{\mu\nu}(x)$ can be written in the
form:
\begin{eqnarray}\label{a}
S=\int d^{p+1}\xi L \hspace{1cm},\hspace{1cm} L=V^JV^K\p_J x^\mu\p_K
x^\nu g_{\mu\nu}(x),
\\ \nl
\p_J=\partial/\partial\xi^J, \hspace{1cm} \xi^J=(\xi^0,\xi^j)=(\tau,\sigma^j),
\\ \nl
J,K=0,1,...,p \hspace{1cm},\hspace{1cm} j,k=1,...,p \hspace{1cm},\hspace{1cm} \mu,\nu=0,1,...,D-1.
\end{eqnarray}
It is an obvious generalization of the flat space-time action
given in \cite{CQG}.
To prove the invariance of the action under infinitesimal
diffeomorphisms on the world volume (reparametrizations),
we first write down the corresponding
transformation law for the (r,s)-type tensor density of weight $a$
\begin{eqnarray}\nl
\delta_{\varepsilon}T^{J_1...J_r}_{K_1...K_s}[a]&=&
L_{\varepsilon}T^{J_1...J_r}_{K_1...K_s}[a]=
\varepsilon^L\partial_L T^{J_1...J_r}_{K_1...K_s}[a]\\
\label{diff}
&+&
T^{J_1...J_r}_{KK_2...K_s}[a]\partial_{K_1}\varepsilon^K+...+
T^{J_1...J_r}_{K_1...K_{s-1}K}[a]\partial_{K_s}\varepsilon^K \\ \nl
&-&
T^{JJ_2...J_r}_{K_1...K_s}[a]\partial_J\varepsilon^{J_1}-...-
T^{J_1...J_{r-1}J}_{K_1...K_s}[a]\partial_J\varepsilon^{J_r} \\ \nl
&+&
aT^{J_1...J_r}_{K_1...K_s}[a]\partial_L\varepsilon^L ,
\end{eqnarray}
where $L_\varepsilon$ is the Lie derivative along the vector field
$\varepsilon$. Using (\ref{diff}), one verifies that if
$x^\mu(\xi)$, $g_{\mu\nu}(\xi)$ are world-volume scalars
($a=0$) and $V^J(\xi)$ is a world-volume (1,0)-type tensor density
of weight $a=1/2$, then $\p_J x^\nu$ is a (0,1)-type tensor,
$\p_J x^\mu \p_K x^\nu g_{\mu \nu}$ is a (0,2)-type tensor and $L$ is
a scalar density of weight $a=1$. Therefore,
\begin{eqnarray}\nl
\delta_{\varepsilon}S=\int d^{p+1}\xi\partial_J\bigl ( \varepsilon^J L\bigr )
\end{eqnarray}
and this variation vanishes under suitable boundary conditions.
The equations of motion following from (\ref{a}) are:
\begin{eqnarray}\nl
\p_J\Bigl (V^J V^K \p_K x^{\lambda}\Bigr ) +
\Gamma^{\lambda}_{\mu\nu} V^J V^K \p_J x^{\mu}\p_K x^{\nu} = 0 ,\\
\nl
V^J \p_J x^{\mu}\p_K x^{\nu} g_{\mu\nu}(x) = 0 ,
\end{eqnarray}
where $\Gamma^{\lambda}_{\mu\nu}$ is the connection
compatible with the metric $g_{\mu\nu}(x)$:
\begin{eqnarray}\nl
\Gamma^{\lambda}_{\mu\nu}=\frac{1}{2}g^{\lambda\rho}\bigl(
\partial_{\mu}g_{\nu\rho}+\partial_{\nu}g_{\mu\rho}-\partial_{\rho}g_{\mu\nu}\bigr) .
\end{eqnarray}
For the transition to Hamiltonian picture it is convenient to
rewrite the Lagrangian density (\ref{a}) in the form
($\p_\tau=\partial/\partial\tau, \p_j=\partial/\partial\sigma^j$):
\begin{eqnarray}\label{L}
L=\frac{1}{4\mu^0} g_{\mu\nu}(x)\bigl (\p_\tau-\mu^j\p_j\bigr )x^\mu
\bigl (\p_\tau-\mu^k\p_k\bigr )x^\nu ,
\end{eqnarray}
where
\begin{eqnarray}\nl
V^J=\bigl(V^0,V^j\bigr)=\Biggl(-\frac{1}{2\sqrt{\mu^0}},
\frac{\mu^j}{2\sqrt{\mu^0}}\Biggr) .
\end{eqnarray}
Now the equation of motion for $x^\nu$ takes the form:
\begin{eqnarray}\label{eqx}
\p_\tau\Bigl [\frac{1}{2\mu^0}\bigl (\p_\tau-\mu^k\p_k\bigr )x^{\lambda}\Bigr ]
-\p_j\Bigl [\frac{\mu^{j}}{2\mu^{0}}\bigl (\p_\tau-\mu^k\p_k\bigr )x^{\lambda}
\Bigr ] \\
\nl
+ \frac{1}{2\mu^0}\Gamma^{\lambda}_{\mu\nu}
\bigl (\p_\tau-\mu^j\p_j\bigr )x^\mu \bigl (\p_\tau-\mu^k\p_k\bigr )x^\nu = 0 .
\end{eqnarray}
The equations of motion for the Lagrange multipliers $\mu^{0}$ and
$\mu^{j}$ which follow from (\ref{L}) give the constraints :
\begin{eqnarray}\label{Tx1}
g_{\mu\nu}(x)\bigl (\p_\tau-\mu^j\p_j\bigr )x^\mu
\bigl (\p_\tau-\mu^k\p_k\bigr )x^\nu = 0 , \\
\nl
g_{\mu\nu}(x)\bigl (\p_\tau-\mu^k\p_k\bigr )x^\mu \p_j x^\nu = 0 .
\end{eqnarray}
In terms of $x^\nu$ and the conjugated momentum $p_\nu$ they read:
\begin{eqnarray}\label{Tpx}
T_0=g^{\mu\nu}(x)p_\mu p_\nu = 0 \hspace{1cm},\hspace{1cm} T_j=p_\nu\p_j x^\nu = 0 .
\end{eqnarray}
\section{\bf Hamiltonian formulation}
\hspace{1cm}
The Hamiltonian which corresponds to the Lagrangian density
(\ref{L}) is a linear combination of the constraints (\ref{Tpx}) :
\begin{eqnarray}\nl
H_0=\int d^p\sigma\bigl (\mu^0 T_0+\mu^j T_j \bigr ) .
\end{eqnarray}
They satisfy the following (equal $\tau$) Poisson bracket algebra
\begin{eqnarray} \nl
\{T_0(\underline \sigma_1),T_0(\underline \sigma_2)\}&=&0,
\\ \label {CA}
\{T_0(\underline \sigma_1),T_{j}(\underline \sigma_2)\}&=& [T_0(\underline \sigma_1) + T_0(\underline
\sigma_2)] \partial_j \delta^p (\underline \sigma_1 - \underline \sigma_2) ,
\\ \nl
\{T_{j}(\underline \sigma_1),T_{k}(\underline \sigma_2)\}&=&
[\delta_{j}^{l}T_{k}(\underline \sigma_1) + \delta_{k}^{l}T_{j}(\underline
\sigma_2)]\partial_l\delta^p(\underline \sigma_1-\underline \sigma_2) ,\\
\nl
\underline\sigma=(\sigma^1,...,\sigma^p) .
\end{eqnarray}
The equalities (\ref{CA}) show that the constraint algebra is the
same for flat and for curved backgrounds. Having in mind the above
algebra, one can construct the corresponding BRST charge $\Omega$
\cite{FF} (*=complex conjugation)
\begin{eqnarray} \label{O}
\Omega = \Omega^{min}+\pi_J \bar {\cal P}^J , \hspace{1cm}
\{\Omega,\Omega\} = 0 , \hspace{1cm} \Omega^* = \Omega .
\end{eqnarray}
$\Omega^{min}$ in (\ref{O}) can be written as \cite{MPL}
\begin{eqnarray}\nl
\Omega^{min}&=&\int d^p\sigma\{T_0\eta^0+T_j\eta^j+
{\cal P}_0 [(\partial_j\eta^j)\eta^0 + (\partial_j\eta^0)\eta^j ] +
{\cal P}_k(\partial_j\eta^k)\eta^j \} ,
\end{eqnarray}
and can be represented also in the form
\begin{eqnarray}\nl
\Omega^{min}=\int d^p\sigma\bigl [\bigl (T_0+
\frac{1}{2}T_0^{gh}\bigr )\eta^0
+\bigl (T_j+\frac{1}{2}T_j^{gh}\bigr )\eta^j \bigr ]
+\int d^p\sigma\partial_j\Bigl (\frac{1}{2}{\cal P}_k\eta^k\eta^j \Bigr ) .
\end{eqnarray}
Here a superscript $gh$ is used for the ghost part of the total gauge
generators
\begin{eqnarray}\nl
\nl
T_J^{tot}=\{\Omega,{\cal P}_J\}=\{\Omega^{min},{\cal P}_J\}=
T_J+T_J^{gh} .
\end{eqnarray}
We recall that the Poisson bracket algebras of $T_J^{tot}$ and
$T_J$ coincide for first rank systems which is the case under
consideration. The manifest expressions for $T_J^{gh}$ are:
\begin{eqnarray}\nl
T_0^{gh}&=&2{\cal P}_0\partial_j\eta^j+\bigl (\partial_j{\cal P}_0\bigr )\eta^j ,
\\ \nl
T_j^{gh}&=&2{\cal P}_0\partial_j\eta^0+\bigl (\partial_j{\cal P}_0\bigr )\eta^0+
{\cal P}_j\partial_k\eta^k+{\cal P}_k\partial_j\eta^k+
\bigl (\partial_k{\cal P}_j\bigr )\eta^k .
\end{eqnarray}
Up to now, we introduced canonically conjugated ghosts
$\bigl (\eta^J,{\cal P}_J\bigr )$,
$\bigl (\bar \eta_J,\bar {\cal P}^J\bigr)$ and momenta $\pi_J$ for
the Lagrange multipliers $\mu^J$ in the Hamiltonian. They have
Poisson brackets and Grassmann parity as follows ($\epsilon_J$ is
the Grassmann parity of the corresponding constraint):
\begin{eqnarray}\nl
\bigl \{\eta^J,{\cal P}_K\bigr \}&=&\delta^J_K ,
\hspace{1cm}
\epsilon (\eta^J)=\epsilon ({\cal P}_J)=\epsilon_J + 1 , \\
\nl
\bigl \{\bar \eta_J,\bar {\cal P}^K \bigr \}&=&-(-1)^{\epsilon_J\epsilon_K}
\delta^K_J ,
\hspace{1cm}
\epsilon (\bar\eta_J)=\epsilon (\bar {\cal P}^J)=\epsilon_J + 1 ,
\\ \nl
\bigl \{\mu^J,\pi_K\bigr \}&=&\delta^J_K ,
\hspace{1cm}
\epsilon (\mu^J)=\epsilon (\pi_J)=\epsilon_J .
\end{eqnarray}
The BRST-invariant Hamiltonian is
\begin{eqnarray}\label{H}
H_{\tilde \chi}=H^{min}+\bigl \{\tilde \chi,\Omega\bigr \}=
\bigl \{\tilde \chi,\Omega\bigr \} ,
\end{eqnarray}
because from $H_{canonical}=0$ it follows $H^{min}=0$. In this
formula $\tilde \chi$ stands for the gauge fixing fermion
$(\tilde \chi^* = -\tilde \chi)$. We use the following representation
for the latter
\begin{eqnarray}\nl
\tilde \chi=\chi^{min}+\bar\eta_J(\chi^J+\frac{1}{2}\rho_{(J)}\pi^J) ,
\hspace{1cm}
\chi^{min}=\mu^J{\cal P}_J ,
\end{eqnarray}
where $\rho_{(J)}$ are scalar parameters and we have separated the
$\pi^J$-dependence from $\chi^J$. If we adopt that $\chi^J$ does
not depend on the ghosts $(\eta^J,{\cal P}_J)$ and
$(\bar\eta_J,\bar {\cal P}^J)$, the Hamiltonian $H_{\tilde\chi}$
from (\ref{H}) takes the form
\begin{eqnarray}
\label{r}
H_{\tilde\chi}&=&H_{\chi}^{min}+{\cal P}_J \bar {\cal P}^J -
\pi_J(\chi^J+\frac{1}{2}\rho_{(J)}\pi^J)+ \\
\nl
&+&\bar\eta_J \bigl \{\chi^J,T_K\bigr \}\eta^K ,
\end{eqnarray}
where
\begin{eqnarray}\nl
H_{\chi}^{min}=\bigl \{\chi^{min},\Omega^{min}\bigr \} .
\end{eqnarray}
One can use the representation (\ref{r}) for $H_{\tilde\chi}$ to obtain the
corresponding BRST invariant Lagrangian density
\begin{eqnarray}\nl
L_{\tilde\chi}=L+L_{GH}+L_{GF} .
\end{eqnarray}
Here $L$ is given in (\ref{L}), $L_{GH}$ stands for the ghost part
and $L_{GF}$ - for the gauge fixing part of the Lagrangian density.
The manifest expressions for $L_{GH}$ and $L_{GF}$ are \cite{MPL}:
\begin{eqnarray}
\nl
L_{GH}=-\partial_\tau\bar{\eta_0}\partial_\tau\eta^0-\partial_\tau\bar{\eta_j}
\partial_\tau\eta^j+\mu^0[2\p_\tau\bar{\eta_0}\p_j\eta^j+
(\p_j\p_\tau\bar{\eta_0})\eta^j]
\\
\nl
+\mu^j[2\p_\tau\bar{\eta_0}\p_j\eta^0+
(\p_j\p_\tau\bar{\eta_0})\eta^0+\p_\tau\bar{\eta_k}\p_j\eta^k+
\p_\tau\bar{\eta_j}\partial_k\eta^k+
(\partial_k\p_\tau\bar{\eta_j})\eta^k]
\\
\nl
+\int d^p\sigma'\{\bar{\eta_0}(\underline{\sigma'})
[\{T_0,\chi^0(\underline{\sigma'})\}\eta^0
+\{T_j,\chi^0(\underline{\sigma'})\}\eta^j]
\\
\nl
+\bar{\eta_j}(\underline{\sigma'})[\{T_0,\chi^j(\underline{\sigma'})\}
\eta^0+\{T_k,\chi^j(\underline{\sigma'})\}\eta^k]\} , \\
\nl
L_{GF}=\frac{1}{2\rho_{(0)}}(\partial_\tau \mu^0-\chi^0)(\partial_\tau
\mu_0-\chi_0)+ \frac{1}{2\rho_{(j)}}(\partial_\tau \mu^j-\chi^j) (\partial_\tau
\mu_j-\chi_j) .
\end{eqnarray}
If one does not intend to pass to the Lagrangian
formalism, one may restrict oneself to
the minimal sector $\bigl (\Omega^{min},\chi^{min},H_\chi^{min}\bigr )$.
In particular, this means that Lagrange multipliers are not considered as
dynamical variables anymore.
With this particular gauge choice, $H_\chi^{min}$
is a linear combination of the total constraints
\begin{eqnarray}\nl
H_\chi^{min}=\int d^p\sigma\Bigl [\Lambda^0 T_0^{tot}(\underline\sigma)+\Lambda^j
T_j^{tot}(\underline\sigma)\Bigr ] ,
\end{eqnarray}
and we can treat here the Lagrange multipliers $\Lambda^0,\Lambda^{j}$
as constants. Of course, this does not fix the gauge completely.
\section{\bf Null membranes in D=4}
\hspace{1cm}
Here we confine ourselves to the case of membranes moving in a
four dimensional, stationary, axially symmetrical, gravity
background of the type
\begin{eqnarray}\label{gm}
ds^2&=&g_{00}(dx^0)^2+g_{11}(dx^1)^2+g_{22}(dx^2)^2+
g_{33}(dx^3)^2+2g_{03}dx^0 dx^3 , \\
\nl
g_{\mu\nu}&=&g_{\mu\nu}(x^1,x^2) .
\end{eqnarray}
We will work in the gauge $\mu^0, \mu^j = constants $, in which the
equations of motion (\ref{eqx}) and constraints (\ref{Tx1})
for the membrane $(j,k=1,2)$ have the form:
\begin{eqnarray}\label{eqxf}
\bigl(\p_\tau-\mu^j\p_j\bigr)\bigl (\p_\tau-\mu^k\p_k\bigr )x^{\lambda}
+ \Gamma^{\lambda}_{\mu\nu}
\bigl (\p_\tau-\mu^j\p_j\bigr )x^\mu \bigl (\p_\tau-\mu^k\p_k\bigr )x^\nu = 0 .
\\ \label{cf1}
g_{\mu\nu}(x)\bigl (\p_\tau-\mu^j\p_j\bigr )x^\mu
\bigl (\p_\tau-\mu^k\p_k\bigr )x^\nu = 0 , \\
\nl
g_{\mu\nu}(x)\bigl (\p_\tau-\mu^k\p_k\bigr )x^\mu \p_j x^\nu = 0 .
\end{eqnarray}
To establish the correspondence with the null geodesics we note
that if we introduce the quantities
\begin{eqnarray}\label{unu}
u^{\nu}(x)=\bigl(\p_\tau-\mu^{j}\p_j\bigr)x^{\nu} ,
\end{eqnarray}
the equations of motion (\ref{eqxf}) can be rewritten as
\begin{eqnarray}\nl
u^{\nu}\bigl(\partial_{\nu}u^{\lambda}+\Gamma^{\lambda}_{\mu\nu}u^{\mu}\bigr)=0 .
\end{eqnarray}
Then it follows from here that $u^2$ do not depend on $x^{\nu}$.
In this notations, the constraints are:
\begin{eqnarray}\nl
g_{\mu\nu}u^{\mu}u^{\nu} = 0 \hspace{1cm},\hspace{1cm}
g_{\mu\nu}u^{\mu}\p_j x^{\nu} = 0 .
\end{eqnarray}
Taking into account the metric (\ref{gm}), one can write the
equations of motion (\ref{eqxf}) and the constraints (\ref{cf1})
in the form:
\begin{eqnarray}\nl
\bigl(\pu-\mu^{j}\pj\bigr) u^0+2\Bigl(\Gamma^0_{01}u^0+\Gamma^0_{13}u^3\Bigr)\bigl(\pu-\mu^{j}\pj\bigr) x^1 +
\\ \nl
+2\Bigl(\Gamma^0_{02}u^0+\Gamma^0_{23}u^3\Bigr)\bigl(\pu-\mu^{j}\pj\bigr) x^2 = 0 ,\\
\nl
\bigl(\pu-\mu^{j}\pj\bigr)\bigl(\pu-\mu^{k}\pk\bigr) x^1+\Gamma^1_{11}\bigl(\pu-\mu^{j}\pj\bigr) x^1\bigl(\pu-\mu^{k}\pk\bigr) x^1+
\\ \nl
+2\Gamma^1_{12}\bigl(\pu-\mu^{j}\pj\bigr) x^1\bigl(\pu-\mu^{k}\pk\bigr) x^2 +\Gamma^1_{22}\bigl(\pu-\mu^{j}\pj\bigr) x^2\bigl(\pu-\mu^{k}\pk\bigr) x^2+
\\ \nl
+\Gamma^1_{00}(u^0)^2+ 2\Gamma^1_{03}u^0 u^3+\Gamma^1_{33}(u^3)^2 = 0 ,
\\ \label{l2}
\bigl(\pu-\mu^{j}\pj\bigr)\bigl(\pu-\mu^{k}\pk\bigr) x^2+\Gamma^2_{11}\bigl(\pu-\mu^{j}\pj\bigr) x^1\bigl(\pu-\mu^{k}\pk\bigr) x^1+
\\ \nl
+2\Gamma^2_{12}\bigl(\pu-\mu^{j}\pj\bigr) x^1\bigl(\pu-\mu^{k}\pk\bigr) x^2+\Gamma^2_{22}\bigl(\pu-\mu^{j}\pj\bigr) x^2\bigl(\pu-\mu^{k}\pk\bigr) x^2+
\\ \nl
+\Gamma^2_{00}(u^0)^2+2\Gamma^2_{03}u^0 u^3+\Gamma^2_{33}(u^3)^2 = 0 ,
\\ \nl
\bigl(\pu-\mu^{j}\pj\bigr) u^3+2\Bigl(\Gamma^3_{01}u^0+\Gamma^3_{13}u^3\Bigr)\bigl(\pu-\mu^{j}\pj\bigr) x^1 +
\\ \nl
+2\Bigl(\Gamma^3_{02}u^0+\Gamma^3_{23}u^3\Bigr)\bigl(\pu-\mu^{j}\pj\bigr) x^2 = 0 ,\\
\nl
g_{11}\bigl(\pu-\mu^{j}\pj\bigr) x^1\bigl(\pu-\mu^{k}\pk\bigr) x^1+g_{22}\bigl(\pu-\mu^{j}\pj\bigr) x^2\bigl(\pu-\mu^{k}\pk\bigr) x^2+
\\ \nl
+g_{00}(u^0)^2+2g_{03}u^0 u^3+g_{33}(u^3)^2 = 0 ,\\
\nl
g_{11}\bigl(\pu-\mu^{k}\pk\bigr) x^1\p_j x^1+g_{22}\bigl(\pu-\mu^{k}\pk\bigr) x^2\p_j x^2 +
\\ \nl
+\bigl(g_{00}\p_j x^0+g_{03}\p_j x^3\bigr)u^0 +
\bigl(g_{03}\p_j x^0+g_{33}\p_j x^3\bigr)u^3 = 0 ,
\end{eqnarray}
where the notation introduced in (\ref{unu}) is used. To simplify
these equations, we make the ansatz
\begin{eqnarray}\nl
x^0(\tau,\underline\sigma)&=&f^0(z^1,z^2)+t(\tau) ,
\\ \label{az}
x^1(\tau,\underline\sigma)&=&r(\tau) \hspace{1cm},\hspace{1cm}
x^2(\tau,\underline\sigma)=\theta(\tau) ,
\\ \nl
x^3(\tau,\underline\sigma)&=&f^3(z^1,z^2)+\varphi(\tau) ,
\\ \nl
z^j&=&\mu^j \tau + \sigma^j ,
\end{eqnarray}
where $f^0, f^3$ are arbitrary functions of their arguments.
After substituting (\ref{az}) in (\ref{l2}), we receive
(the dot is used for differentiation with respect to the affine
parameter $\tau$):
\begin{eqnarray}\label{l01}
\dot u^0+\Bigl[\left(g^{00}\partial_1 g_{00}+g^{03}\partial_1 g_{03}\right)u^0
+\left(g^{00}\partial_1 g_{03}+g^{03}\partial_1 g_{33}\right)u^3\Bigr]\dot r
\\ \nl
+\Bigl[\left(g^{00}\partial_2 g_{00}+g^{03}\partial_2 g_{03}\right)u^0
+\left(g^{00}\partial_2 g_{03}+g^{03}\partial_2
g_{33}\right)u^3\Bigr]\dot\theta = 0 ,
\\ \label{l11}
2g_{11}\ddot r+\partial_1 g_{11}\dot r^2+2\partial_2 g_{11}\dot r\dot\theta
-\partial_1 g_{22}\dot\theta^2
\\ \nl
-\bigl[\partial_1 g_{00}(u^0)^2+2\partial_1 g_{03}u^0 u^3
+\partial_1 g_{33}(u^3)^2\bigr]=0 ,
\\ \label{l21}
2g_{22}\ddot\theta+\partial_2 g_{22}\dot\theta^2+2\partial_1 g_{22}\dot
r\dot\theta-\partial_2 g_{11}\dot r^2
\\ \nl
-\bigl[\partial_2 g_{00}(u^0)^2+2\partial_2 g_{03}u^0 u^3
+\partial_2 g_{33}(u^3)^2\bigr]=0 ,
\\ \label{l31}
\dot u^3+\Bigl[\left(g^{33}\partial_1 g_{03}+g^{03}\partial_1 g_{00}\right)u^0
+\left(g^{33}\partial_1 g_{33}+g^{03}\partial_1 g_{03}\right)u^3\Bigr]\dot r
\\ \nl
+\Bigl[\left(g^{33}\partial_2 g_{03}+g^{03}\partial_2 g_{00}\right)u^0
+\left(g^{33}\partial_2 g_{33}+g^{03}\partial_2
g_{03}\right)u^3\Bigr]\dot\theta = 0 ,
\\ \label{c11}
g_{11}\dot r^2+g_{22}\dot\theta^2+g_{00}(u^0)^2
+2g_{03}u^0 u^3+g_{33}(u^3)^2 = 0 ,
\\ \label{c21}
\bigl(g_{00}\p_j f^0+g_{03}\p_j f^3\bigr)u^0
+\bigl(g_{03}\p_j f^0+g_{33}\p_j f^3\bigr)u^3 = 0 .
\end{eqnarray}
If we choose
\begin{eqnarray}\nl
f^0(z^1,z^2)=f^0(w) \hspace{1cm},\hspace{1cm} f^3(z^1,z^2)=f^3(w) ,
\end{eqnarray}
where $w=w(z^1,z^2)$ is an arbitrary function of $z^1$ and $z^2$,
then the system of equations (\ref{c21}) reduces to the single
equation
\begin{eqnarray}\label{c22}
\Bigl(g_{00}\frac{df^0}{dw}+g_{03}\frac{df^3}{dw}\Bigr)u^0
+\Bigl(g_{03}\frac{df^0}{dw}+g_{33}\frac{df^3}{dw}\Bigr)u^3=0 .
\end{eqnarray}
To be able to separate the variables $u^0, u^3$ in the system of
differential equations (\ref{l01}), (\ref{l31}) with the help of
(\ref{c22}), we impose the following condition on $f^0(w)$ and
$f^3(w)$ \begin{eqnarray}\nl f^0(w) = C^0 f[w(z^1,z^2)] \hspace{1cm},\hspace{1cm} f^3(w) = C^3
f[w(z^1,z^2)] , \end{eqnarray} where $C^0, C^3$ are constants, and $f(w)$ is an
arbitrary function of $w$. Then the solution of (\ref{l01}),
(\ref{l31}) and (\ref{c22}) is \cite{PorPap} ($C_1 = const $): \begin{eqnarray}\nl
u^0(\tau)=-C_1\left(C^0 g_{03}+C^3 g_{33}\right)\exp(-H) ,
\\ \label{s032}
u^3(\tau)=+C_1\left(C^0 g_{00}+C^3 g_{03}\right)\exp(-H) ,
\\ \nl
H = \int\Bigl(g^{00}dg_{00}+2g^{03}dg_{03}+g^{33}dg_{33}\Bigr) .
\end{eqnarray}
The condition for the compatibility of (\ref{s032}) with
(\ref{l11}), (\ref{l21}) and (\ref{c11}) is:
\begin{eqnarray}\nl
u^0(\tau)&=&-C_1\left(C^0 g_{03}+C^3 g_{33}\right)h^{-1}
\\ \nl
&=&-C_1\left(C^3 g^{00}-C^0 g^{03}\right)=\dot t(\tau) ,
\\ \label{su3}
u^3(\tau)&=&+C_1\left(C^0 g_{00}+C^3 g_{03}\right)h^{-1}
\\ \nl
&=&-C_1\left(C^3 g^{03}-C^0 g^{33}\right)=\dot\varphi (\tau) ,
\\ \nl
h&=&g_{00}g_{33}-g_{03}^2 .
\end{eqnarray}
On the other hand, from (\ref{l21}) and (\ref{c11}) one has:
\begin{eqnarray}\label{r.2}
\dot r^2&=&-g^{11}\Bigl[C_1^2\frac{G}{h}+g^{22}\left(g_{22}^2
\dot\theta^2\right)\Bigr]
\\ \nl
&=&g^{11}\Bigl\{C_1^2\bigl[2C^0 C^3 g^{03}-(C^3)^2 g^{00}-(C^0)^2
g^{33}\bigr]-g^{22}\bigl(g_{22}^2\dot\theta^2\bigr)\Bigr\} ,
\\ \label{t.2}
g_{22}^2\dot\theta^2&=&C_2+C_1^2\int\limits^{\theta}d\theta h^{-2}
\Bigl[g_{22}G\frac{\partial h}{\partial\theta}-h\frac{\partial g_{22}G}{\partial\theta}\Bigr],
\\ \nl
G&=&(C^0)^2 g_{00}+2C^0 C^3 g_{03}+(C^3)^2 g_{33}.
\end{eqnarray}
In obtaining (\ref{t.2}), we have used
the gauge freedom in the metric (\ref{gm}),
to impose the condition \cite{Chandra}:
\begin{eqnarray}\nl
\partial_2\Biggl(\frac{g_{22}}{g_{11}}\Biggr) = 0 .
\end{eqnarray}
As a final result we have
\begin{eqnarray}\nl
x^0 &=& C^0 f[w(z^1,z^2)] + t(\tau), \\ \nl
x^1 &=& r(\tau), \\ \nl
x^2 &=& \theta (\tau), \\ \nl
x^3 &=& C^3 f[w(z^1,z^2)] + \varphi (\tau) ,
\end{eqnarray}
where $\dot t(\tau), \dot r(\tau), \dot\theta (\tau),
\dot\varphi (\tau)$ are given by (\ref{su3}), (\ref{r.2}),
and (\ref{t.2}).
In the particular case when $x^2=\theta=\theta_0=const$, one can
integrate to obtain the following exact solution of the equations of
motion and constraints for the null membrane in the gravity
background (\ref{gm}): \begin{eqnarray}\nl x^0 (\tau,\sigma^1,\sigma^2)&=&C^0
f[w(z^1,z^2)] + t_0 \\ \nl &\pm&\int\limits_{r_0}^{r}dr\left(C^3
g^{00}-C^0 g^{03}\right)W^{-1/2} ,
\\ \label{GSC}
x^3 (\tau,\sigma^1,\sigma^2)&=&C^3 f[w(z^1,z^2)] + \varphi_{0}
\\ \nl
&\pm&\int\limits_{r_0}^{r}dr\left(C^3 g^{03}-C^0
g^{33}\right)W^{-1/2} ,
\\ \nl
C_1 (\tau -\tau_{0})&=& \pm\int\limits_{r_0}^{r}drW^{-1/2} ,\\ \nl
W&=&g^{11}\left[2C^0 C^3 g^{03}-\left(C^3\right)^2
g^{00}-\left(C^0\right)^2 g^{33}\right] ,
\\ \nl
t_0, r_0, \varphi_{0}, \tau_{0} &-& constants.
\end{eqnarray}
\section{\bf Examples}
\hspace{1cm} Here we give some examples of solutions of the type
received in the previous section. To begin with, let us start with
the simplest case of {\it Minkowski space-time}. The metric is \begin{eqnarray}\nl
g_{00}=-1,\hspace{1cm} g_{11}=1,\hspace{1cm} g_{22}=r^2,\hspace{1cm} g_{33}=r^2 \sin^2\theta , \end{eqnarray}
and equalities (\ref{su3}), (\ref{r.2}), (\ref{t.2}) take the form:
\begin{eqnarray}\nl \dot t&=&C_1 C^3 ,\\ \nl \dot r^2&=&(C_1 C^3)^2 -
\frac{C_2}{r^2} ,\\ \nl r^4 \dot\theta^2&=&C_2-\frac{(C_1
C^0)^2}{\sin^2\theta} ,
\\ \nl
\dot\varphi&=&\frac{C_1 C^0}{r^2 \sin^2\theta} . \end{eqnarray} When
$\theta=\theta_0=const$, the solution (\ref{GSC}) is: \begin{eqnarray}\nl x^0
(\tau,\sigma^1,\sigma^2)&=&C^0 f[w(z^1,z^2)] + t_0 \mp C^3
\int\limits_{r_0}^{r}\frac{dr} {\bigl [(C^3)^2-(C^0)^2
r^{-2}\sin^{-2}\theta_0\bigr ]^{1/2}} ,
\\ \nl
x^3 (\tau,\sigma^1,\sigma^2)&=&C^3 f[w(z^1,z^2)] + \varphi_{0}
\mp\frac{C^0}{\sin^2\theta_0}\int\limits_{r_0}^{r}\frac{dr} {r^2\bigl
[(C^3)^2-(C^0)^2 r^{-2}\sin^{-2}\theta_0\bigr ]^{1/2}} ,
\\ \nl
C_1(\tau - \tau_{0}) &=& \pm\int\limits_{r_0}^{r}\frac{dr}{\bigl
[(C^3)^2 -(C^0)^2 r^{-2}\sin^{-2}\theta_{0}\bigr ]^{1/2}} . \end{eqnarray}
Our next example is the {\it de Sitter space-time}. We take the
metric in the form \begin{eqnarray}\nl g_{00}=-\left(1-kr^2\right),
g_{11}=\left(1-kr^2\right)^{-1}, g_{22}=r^2, g_{33}=r^2 \sin^2\theta
, \end{eqnarray} where $k$ is the constant curvature. Now we have \begin{eqnarray}\nl \dot
t&=&\frac{C_1 C^3}{1-kr^2} ,\\ \nl \dot r^2&=&(C_1 C^3)^2 + C_2
(k-r^{-2}) ,
\\ \nl
r^4 \dot\theta^2&=&C_2-\frac{(C_1 C^0)^2}{\sin^2\theta} ,
\\ \nl
\dot\varphi&=&\frac{C_1 C^0}{r^2 \sin^2\theta} , \end{eqnarray} and the
corresponding solution (\ref{GSC}) is: \begin{eqnarray}\nl x^0
(\tau,\sigma^1,\sigma^2)&=&C^0 f[w(z^1,z^2)] + t_0
\\ \nl
&\mp& C^3 \int\limits_{r_0}^{r}\frac{dr}{(1-kr^2) \bigl
[(C^3)^2+(C^0)^2 (k-r^{-2})\sin^{-2}\theta_0\bigr ]^{1/2}} ,
\\ \nl
x^3 (\tau,\sigma^1,\sigma^2)&=&C^3 f[w(z^1,z^2)] + \varphi_{0}
\\ \nl
&\mp&\frac{C^0}{\sin^2\theta_0}\int\limits_{r_0}^{r}\frac{dr}
{r^2\bigl[(C^3)^2+(C^0)^2 (k-r^{-2})\sin^{-2}\theta_0\bigr]^{1/2}} ,
\\ \nl
C_1(\tau - \tau_{0}) &=& \pm\int\limits_{r_0}^{r}\frac{dr}{\bigl
[(C^3)^2 +(C^0)^2 (k-r^{-2})\sin^{-2}\theta_0\bigr ]^{1/2}} . \end{eqnarray}
Now let us turn to the case of {\it Schwarzschild space-time}.
The corresponding metric may be written as
\begin{eqnarray}\nl
g_{00}&=&-(1-2Mr^{-1}) \hspace{1cm},\hspace{1cm} g_{11}=(1-2Mr^{-1})^{-1} ,
\\ \nl
g_{22}&=&r^2 \hspace{1cm},\hspace{1cm} g_{33} = r^2 \sin^2\theta , \end{eqnarray} where $M$ is the
Schwarzschild mass. The equalities (\ref{su3}), (\ref{r.2}) and
(\ref{t.2}) read \begin{eqnarray}\nl \dot t&=&\frac{C_1 C^3}{1-2Mr^{-1}} ,\\ \nl
\dot r^2&=&(C_1 C^3)^2 -\frac{C_2}{r^2}(1-2Mr^{-1}) ,
\\ \label{S}
r^4 \dot\theta^2&=&C_2-\frac{(C_1 C^0)^2}{\sin^2\theta} ,
\\ \nl
\dot\varphi&=&\frac{C_1 C^0}{r^2 \sin^2\theta} . \end{eqnarray} When
$\theta=\theta_0=const$, one obtains from (\ref{GSC}) \begin{eqnarray}\nl x^0
(\tau,\sigma^1,\sigma^2)&=&C^0 f[w(z^1,z^2)] + t_0
\\ \nl
&\mp& C^3\int\limits_{r_0}^{r}\frac{dr}{(1-2Mr^{-1}) \bigl [(C^3)^2
-(C^0)^2 r^{-2}(1-2Mr^{-1})\sin^{-2}\theta_0]^{1/2}} ,
\\ \nl
x^3 (\tau,\sigma^1,\sigma^2)&=&C^3 f[w(z^1,z^2)] + \varphi_{0}
\\ \nl
&\mp&\frac{C^0}{\sin^2\theta_0}\int\limits_{r_0}^{r}\frac{dr}
{r^2\bigl[(C^3)^2 -(C^0)^2
r^{-2}(1-2Mr^{-1})\sin^{-2}\theta_0\bigr]^{1/2}} ,
\\ \nl
C_1(\tau - \tau_{0}) &=& \pm\int\limits_{r_0}^{r}\frac{dr}{\bigl
[(C^3)^2-(C^0)^2 r^{-2} (1-2Mr^{-1})\sin^{-2}\theta_0\bigr ]^{1/2}} .
\end{eqnarray}
For the {\it Taub-NUT space-time} we take the metric as \begin{eqnarray}\nl
g_{00}=-\frac{\delta}{R^2}\hspace{1cm},\hspace{1cm} g_{11}=\frac{R^2}{\delta}\hspace{1cm},\hspace{1cm}
g_{22}=R^2 ,\\ \nl
g_{33}=R^2\sin^2\theta-4l^2\frac{\delta\cos^2\theta}{R^2}\hspace{1cm},\hspace{1cm}
g_{03}=-2l\frac{\delta\cos\theta}{R^2} , \\ \nl
\delta(r)=r^2-2Mr-l^2\hspace{1cm},\hspace{1cm} R^2(r)=r^2+l^2 , \end{eqnarray} where $M$ is the mass
and $l$ is the NUT-parameter. Now we have from (\ref{su3}),
(\ref{r.2}) and (\ref{t.2}): \begin{eqnarray}\nl \dot t&=&\frac{C_1}{R^2}\Biggl[
C^3\Biggl(\frac{R^4}{\delta}+4l^2\Biggr)-\frac{2l}{\sin^2\theta}\left(
C^0\cos\theta+2C^3 l\right)\Biggr], \\ \nl R^4\dot
r^2&=&\left(C_1C^3\right)^2\left(R^4+4l^2\delta\right)-C_2\delta, \\
\nl R^4\dot\theta^2&=&C_2-\frac{C_1^2}{\sin^2\theta}\Bigl[\left(
C^0\right)^2+\left(2C^3 l\right)^2+4C^0 C^3 l\cos\theta\Bigr],
\\ \nl \dot
\varphi&=&\frac{C_1}{R^2\sin^2\theta}\left(C^0+2C^3
l\cos\theta\right). \end{eqnarray} In the Taub-NUT metric the solution
(\ref{GSC}) is \begin{eqnarray}\nl x^0 (\tau,\sigma^1,\sigma^2)&=&C^0 f[w(z^1,z^2)] + t_0
\\ \nl
&\pm&\int\limits_{r_0}^{r}dr\Bigl[C^3\left(R^4
\delta^{-1}+4l^2\right)-2l
\sin^{-2}\theta_0\left(C^0\cos\theta_{0}+2C^3
l\right)\Bigr]U^{-1/2}(r) , \\ \nl x^3 (\tau,\sigma^1,\sigma^2)&=&C^3
f[w(z^1,z^2)] + \varphi_{0}
\\ \nl
&\pm&\frac{1}{\sin^2\theta_0}\int\limits_{r_0}^{r}dr\left(C^0+2C^3
l\cos\theta_{0}\right)U^{-1/2}(r) ,\\ \nl C_1(\tau - \tau_{0})&=&
\pm\int\limits_{r_0}^{r}drR^2 U^{-1/2}(r) ,\end{eqnarray} where \begin{eqnarray}\nl
U(r)=\left(C^3\right)^2\left(R^4-4l^2\delta\cot^2\theta_{0}\right)
-\delta\sin^{-2}\theta_{0}\left[\left(C^0\right)^2+4C^0 C^3
l\cos\theta_{0}\right] . \end{eqnarray}
Finally, we consider the {\it Kerr space-time} with metric taken
in the form \begin{eqnarray}\nl g_{00}&=&-\Biggl(1-\frac{2Mr}{\rho^2}\Biggr)\hspace{1cm},\hspace{1cm}
g_{11}=\frac{\rho^2}{\Delta},
\\ \nl
g_{22}&=&\rho^2 \hspace{1cm},\hspace{1cm} g_{33}=\Biggl(r^2+a^2+\frac{2Ma^2
r\sin^2\theta}{\rho^2}\Biggr)\sin^2\theta ,
\\ \nl
g_{03}&=&-\frac{2Mar\sin^2\theta}{\rho^2} , \end{eqnarray} where \begin{eqnarray}\nl
\rho^2=r^2+a^2\cos^2\theta \hspace{1cm},\hspace{1cm} \Delta=r^2-2Mr+a^2 , \end{eqnarray} $M$ is the
mass and $a$ is the angular momentum per unit mass of the Kerr black
hole. With this input in equations (\ref{su3}), (\ref{r.2}) and
(\ref{t.2}), we have: \begin{eqnarray}\nl \dot
t&=&\frac{C_1}{\Delta\rho^2}\Bigl[C^3 (r^2+a^2)^2 -C^3 a^2\Delta
\sin^2\theta - 2C^0 M a r\Bigr] ,
\\ \nl
\rho^4\dot r^2&=&C_1^2\Bigl[(C^3)^2 (r^2+a^2)^2
-4C^0 C^3 Mar+(C^0)^2 a^2\Bigr] - C_2 \Delta ,
\\ \nl
\rho^4\dot\theta^2&=&C_2-C_1^2\Biggl(\frac{(C^0)^2}{\sin^2\theta}
+(C^3)^2 a^2 \sin^2\theta\Biggr)
\\ \nl
\dot\varphi&=&\frac{C_1}{\Delta\rho^2}\Biggl( \frac{C^0
\Delta}{\sin^2\theta}+2C^3 Mar-C^0 a^2\Biggr) . \end{eqnarray} In this case, the
exact solution (\ref{GSC}) takes the form: \begin{eqnarray}\nl
x^0
(\tau,\sigma^1,\sigma^2)&=&C^0 f[w(z^1,z^2)] + t_0
\\ \nl
&\pm&\int\limits_{r_0}^{r}dr\Delta^{-1} \Bigl\{2C^0 Mar-C^3\left[
(r^2+a^2)\rho_{0}^2+2Ma^2 r\sin^2\theta_0\right]\Bigr\}V^{-1/2}(r)
,\\ \label{K} x^3 (\tau,\sigma^1,\sigma^2)&=&C^3 f[w(z^1,z^2)] + \varphi_{0}
\\ \nl
&\pm&\frac{1}{\sin^2\theta_0}\int\limits_{r_0}^{r}dr\Delta^{-1}
\left[C^0\left(2Mr-\rho^2_{0}\right)-2C^3
Mar\sin^2\theta_0\right]V^{-1/2}(r) ,\\ \nl C_1(\tau - \tau_{0})&=&
\pm\int\limits_{r_0}^{r}dr\rho_0^2 V^{-1/2}(r) ,\\ \nl
V(r)&=&\left(C^0\right)^2\left(a^2-\Delta\sin^{-2}\theta_{0}\right)
-4C^0 C^3 Mar\\ \nl &+&\left(C^3\right)^2\Bigl[\left(r^2+a^2\right)^2
-a^2\Delta\sin^2\theta_{0}\Bigr] ,\\ \nl
\rho^2_{0}&=&r^2+a^2\cos^2\theta_{0} . \end{eqnarray}
\section{\bf Comments and conclusions}
\hspace{1cm} In the previous section we restrict ourselves to some
particular cases of the generic solution (\ref{GSC}) and do not pay
attention to the existing possibilities for obtaining solutions in
the case $\theta\neq const$ in the considered examples.
Obviously, the examples given in Sec.5 do not exhaust all
possibilities contained in the metric (\ref{gm}) \cite{KSMH}. On the
other hand, in different particular cases of this type of metric,
there exist more general brane solutions. They will be published
elsewhere. Here we only mention that in the gauge $\mu^{k}=const$,
\begin{eqnarray}\nl x^{\nu}(\tau,\underline\sigma)=x^{\nu}(\mu^{k}\tau+\sigma^k) \end{eqnarray} is an obvious
nontrivial solution of the equations of motion and of the constraints
(\ref{eqx}), (\ref{Tx1}) depending on $D$ arbitrary functions of $p$
variables for the null $p$-brane in arbitrary $D$-dimensional gravity
background.
From the results of the previous sections, it is easy to extract the
corresponding formulas for the null string case simply by putting
$\sigma^1=\sigma, \mu^1=\mu, \sigma^2=\mu^2=0$. For example, our equalities
(\ref{S}) coincide with the ones obtained in \cite{DabLar} for the
null string moving in Schwarzschild space-time after identification:
\begin{eqnarray}\nl E=C_1 C^3 ,\hspace{1cm} L=C_1 C^0 ,\hspace{1cm} L^2+K=C_2 . \end{eqnarray} Moreover, our
solution (\ref{K}) in the case $p=1$, generalizes the solution given
in \cite{PorPap}. The latter corresponds to fixing the arbitrary
function $f(w)$ to a linear one and fixing the gauge to $\mu=0$, i.e.
\begin{eqnarray}\nl f[w(\mu^1\tau+\sigma^1,\mu^2\tau+\sigma^2)]\mapsto
f(\mu\tau+\sigma)=\mu\tau+\sigma ,\hspace{1cm} \mbox{with}\hspace{1cm} \mu=0 . \end{eqnarray}
In this paper we perform some investigation on the classical dynamics
of the null bosonic branes in curved background. In the second
section we give the action, prove its reparametrization invariance
and give the equations of motion and constraints in an arbitrary
gauge. Then we construct the corresponding Hamiltonian and compute
the constraint algebra. Following \cite{FF}, we obtain the manifest
expressions for the classical BRST charge, the total constraints, the
BRST invariant Hamiltonian and the BRST invariant Lagrangian. All
this gives the possibility for quantization of the null $p$-branes in
a curved space-time. In the fourth section we consider the dynamics
of the null membranes ($p=2$) in a four dimensional, stationary,
axially symmetrical, gravity background. Some exact solutions of the
equations of motion and of the constraints are found there. The next
section is devoted to examples of such solutions in Minkowski, de
Sitter, Schwarzschild, Taub-NUT and Kerr space-times.
\vspace*{1cm}
{\bf Acknowledgments}
The author would like to thank D. Mladenov for useful discussions
and B. Dimitrov for careful reading of the manuscript.
\vspace*{.5cm}
|
\section{Introduction}
Starting in 1996 we undertook a long term project, DIRECT (as in
``direct distances''), to obtain the distances to two important
galaxies in the cosmological distance ladder -- M31 and M33 -- using
detached eclipsing binaries (DEBs) and Cepheids. These two nearby
galaxies are stepping stones to most of our current effort to
understand the evolving universe at large scales. First, they are
essential to the calibration of the extragalactic distance scale
(Jacoby et al.~1992; Tonry et al.~1997). Second, they constrain
population synthesis models for early galaxy formation and evolution
and provide the stellar luminosity calibration. There is one simple
requirement for all this---accurate distances.
DEBs have the potential to establish distances to M31 and M33 with an
unprecedented accuracy of better than 5\% and possibly to better than
1\%. These distances are now known to no better than 10-15\%, as there
are discrepancies of $0.2-0.3\;{\rm mag}$ between various distance
indicators (e.g.~Huterer, Sasselov \& Schechter 1995; Holland 1998;
Stanek \& Garnavich 1998). Detached eclipsing binaries (for reviews
see Andersen 1991; Paczy\'nski 1997) offer a single step distance
determination to nearby galaxies and may therefore provide an accurate
zero point calibration---a major step towards very accurate
determination of the Hubble constant, presently an important but
daunting problem for astrophysicists. A DEB system was recently used
by Guinan et al.~(1998) and Udalski et al.~(1998) to obtain an
accurate distance estimate to the Large Magellanic Cloud.
The detached eclipsing binaries have yet to be used (Huterer et
al.~1995; Hilditch 1996) as distance indicators to M31 and M33.
According to Hilditch (1996), there were about 60 eclipsing binaries
of all kinds known in M31 (Gaposchkin 1962; Baade \& Swope 1963, 1965)
and only {\em one} in M33 (Hubble 1929), none of them observed with
CCDs. Only now does the availability of large-format CCD detectors
and inexpensive CPUs make it possible to organize a massive search for
periodic variables, which will produce a handful of good DEB
candidates. These can then be spectroscopically followed-up with the
powerful new 6.5-10 meter telescopes.
The study of Cepheids in M31 and M33 has a venerable history (Hubble
1926, 1929; Gaposchkin 1962; Baade \& Swope 1963, 1965). Freedman \&
Madore (1990) and Freedman, Wilson \& Madore (1991) obtained
multi-band CCD photometry of some of the already known Cepheids, to
build period-luminosity relations in M31 and M33, respectively.
However, both the sparse photometry and the small samples (11 Cepheids
in M33 and 38 Cepheids in M31) do not provide a good basis for
obtaining direct Baade-Wesselink distances (see, e.g., Krockenberger,
Sasselov \& Noyes 1997) to Cepheids---the need for new digital
photometry has been long overdue. Recently, Magnier et al.~(1997)
surveyed large portions of M31, which have previously been ignored,
and found some 130 new Cepheid variable candidates. Their light
curves are, however, rather sparsely sampled and in the $V$-band only.
In Kaluzny et al.~(1998, 1999, hereafter: Papers I and IV)
and Stanek et al.~(1998, 1999, hereafter: Papers II and III),
the first four papers of the series, we presented the catalogs
of variable stars found in four fields in M31, called M31B, M31A,
M31C and M31D. Here we present the catalog
of variables from the field M31F. In Sec.2 we discuss the
selection of the fields in M31 and the observations. In Sec.3 we
describe the data reduction and calibration. In Sec.4 we discuss
briefly the automatic selection we used for finding the variable
stars. In Sec.5 we discuss the classification of the variables. In
Sec.6 we present the catalog of variable stars, followed by a brief
discussion of the results in Sec.7.
\section{Fields selection and observations}
M31 was primarily observed in 1996 with the 1.3~m McGraw-Hill
Telescope at the Michigan-Dartmouth-MIT (MDM) Observatory. We used the
front-illuminated, Loral $2048^2$ CCD ``Wilbur'' (Metzger, Tonry \&
Luppino 1993), which at the $f/7.5$ station of the 1.3~m telescope has
a pixel scale of $0.32\; arcsec\; pixel^{-1}$ and field of view of
roughly $11\;arcmin$. We used Kitt Peak Johnson-Cousins $BVI$ filters.
Data for M31 were also obtained, mostly in 1997, with the 1.2~m
telescope at the F. L. Whipple Observatory (FLWO), where we used
``AndyCam'' (Szentgyorgyi et al.~1999), with a thinned,
back-illuminated, AR coated Loral $2048^2$ pixel CCD. The pixel scale
happens to be essentially the same as at the MDM 1.3~m telescope. We
used standard Johnson-Cousins $BVI$ filters.
Fields in M31 were selected using the MIT photometric survey of M31 by
Magnier et al.~(1992) and Haiman et al.~(1994) (see Paper I, Fig.1).
We selected six $11'\times11'$ fields, M31A--F, four of them (A--D)
concentrated on the rich spiral arm in the northeast part of M31, one
(E) coinciding with the region of M31 searched for microlensing by
Crotts \& Tomaney (1996), and one (F) containing the giant star
formation region known as NGC206 (observed by Baade \& Swope
1965). Fields A--C were observed during September and October 1996
five to eight times per night in the $V$ band, resulting in total of
110--160 $V$ exposures per field. Fields D--F were observed once a
night in the $V$-band. Some exposures in $B$ and $I$ were also
taken. M31 was also observed, in 1996 and 1997, at the FLWO 1.2~m
telescope, whose main target was M33.
In this paper we present the results for the M31F field. We obtained
for this field useful data during 29 nights at the MDM, collecting a
total of $28\times 900\;sec$ exposures in $V$ and $2\times 600\;sec$
exposures in $I$. We also obtained for this field useful data during
22 nights at the FLWO, in 1996 and 1997, collecting a total of
$80\times 900\;sec$ exposures in $V$, $67\times 600\;sec$ exposures in
$I$ and $7\times 1200\;sec$ exposures of $B$.\footnote{The complete
list of exposures for this field and related data files are available
through {\tt anonymous ftp} on {\tt cfa-ftp.harvard.edu}, in {\tt
pub/kstanek/DIRECT} directory. Please retrieve the {\tt README} file
for instructions. Additional information on the DIRECT project is
available through the {\tt WWW} at {\tt
http://cfa-www.harvard.edu/\~\/kstanek/DIRECT/}.}
\section{Data reduction, calibration and astrometry}
The details of the reduction procedure were given in Paper I.
Preliminary processing of the CCD frames was done with the standard
routines in the IRAF-CCDPROC package.\footnote{IRAF is distributed by
the National Optical Astronomy Observatories, which are operated by
the Associations of Universities for Research in Astronomy, Inc.,
under cooperative agreement with the NSF} Stellar profile photometry
was extracted using the {\it Daophot/Allstar} package (Stetson 1987,
1992). We selected a ``template'' frame for each filter using a
single frame of particularly good quality. These template images were
reduced in a standard way (Paper I). Other images were reduced using
{\it Allstar} in the fixed-position-mode using as an input the
transformed object list from the template frames. For each frame the
list of instrumental photometry derived for a given frame was
transformed to the common instrumental system of the appropriate
``template'' image. Photometry obtained for the $B,V$ and $I$ filters
was combined into separate data bases. M31F $I$-band images obtained at the
MDM were reduced using FLWO ``templates''. Two templates were used in the
case of $V$-band images. An MDM template was used to fix the positions of
the stars. The $V$ photometry was transformed to the instrumental
system of an FLWO template.
The photometric $VI$ calibration of the MDM data was discussed in
Paper I. In addition, for the field M31F on the night of 1997 October
9/10 we have obtained independent $BVI$ calibration with the FLWO
1.2~m telescope. There was an offset of $-0.020\;{\rm mag}$ in $V$ and
$0.057\;{\rm mag}$ in $V-I$ between the FLWO and the MDM calibration.
The $V$ offset is well within our estimate of the total $0.05\;mag$
systematic error discussed in Paper I, and the $V-I$ offset falls slightly
above.
We also derived equatorial coordinates for all objects included in the
data bases for the $V$ filter. The transformation from rectangular
coordinates to equatorial coordinates was derived using 79 stars
identified in the USNO-A2.0 catalog. We have also compared these
coordinates with those given by Magnier et al.~(1992) and find a
good agreement, with an average difference of $0.5\; arcsec$
for 66 of the transformation stars found in both catalogs.
\section{Selection of variables}
The procedure for selecting the variables was described in detail in
Paper I, so here we only give a short description, noting changes when
necessary. The reduction procedure described in previous section
produces databases of calibrated $BVI$ magnitudes and their standard
errors. The $V$ database for M31F field contains 7997 stars, with up
to 108 measurements, the $I$ database contains 26540 stars with up to
69 measurements and the $B$ database contains 3382 stars with up to 7
measurements. Figure~\ref{fig:dist} shows the distributions of stars
as a function of mean $\bar{B}$, $\bar{V}$ or $\bar{I}$ magnitude. As
can be seen from the shape of the histograms, our completeness starts
to drop rapidly at about $\bar{B}\sim22$, $\bar{V}\sim22$ and
$\bar{I}\sim20.5$. The primary reason for this difference in the depth
of the photometry between $BV$ and $I$ is the level of the combined
sky and background light, which is about three times higher in the $I$
filter than in the $BV$ filters.
\begin{figure}[t]
\plotfiddle{dist.ps}{8cm}{0}{50}{50}{-160}{-85}
\caption{Distributions in $B$ (dotted line), $V$ (dashed line) and $I$
(continuous line) of stars in the field M31F.}
\label{fig:dist}
\end{figure}
The measurements flagged as ``bad'' (with unusually large {\em
Daophot} errors, compared to other stars) and measurements with errors
exceeding the average error, for a given star, by more than $4\sigma$
are removed. Usually zero to 10 points are removed, leaving the
majority of stars with roughly $N_{good}\sim95-105$ $V$\/
measurements. For further analysis we use only those stars that have
at least $N_{good}>N_{max}/2\;(=54)$ measurements. There are 5838
such stars in the $V$ database of the M31F field.
Our next goal is to select a sample of variable stars from the total
sample defined above. There are many ways to proceed, and we largely
follow the approach of Stetson (1996), also described in Paper~I. In
short, for each star we compute the Stetson's variability index $J_S$
(Paper I, Eq.7), and stars with values exceeding some minimum value
$J_{S,min}$ are considered candidate variables. The definition of
$J_S$ is rooted in the assumption that on each visit to the program
field at least one pair of observations is obtained, and only when
both observations have the residual from the mean of the same sign
does the pair contribute positively to the variability index. The
definition of Stetson's variability index includes the standard errors
of individual observations. If, for some reason, these errors were
over- or underestimated, we would either miss real variables, or
select spurious variables as real ones. Using the procedure described
in Paper I, we scale the {\em Daophot} errors to better represent the
``true'' photometric errors. We then select the candidate variable
stars by computing the value of $J_S$ for the stars in our $V$
database. We used a cutoff of $J_{S,min}=0.75$ and additional cuts
described in Paper I to select 122 candidate variable stars (about 2\% of
the total number of 5838). In Figure~\ref{fig:stetj} we plot the
variability index $J_S$ vs. apparent visual magnitude $\bar{V}$ for 5838
stars with $N_{good}>54$.
\begin{figure}[t]
\plotfiddle{stetj.ps}{8cm}{0}{50}{50}{-160}{-85}
\caption{Variability index $J_S$ vs. mean $\bar{V}$ magnitude for 5838
stars in the field M31F with $N_{good}>54$. Dashed line at $J_S=0.75$
defines the cutoff applied for variability.}
\label{fig:stetj}
\end{figure}
\section{Period determination, classification of variables}
We based our candidate variables selection on the $V$ band data collected
at the MDM and the FLWO telescopes. We also have the $BI$-bands data for
the field, up to 69 $I$-band epochs and up to 7 $B$-band epochs, although
for a variety of reasons some of the candidate variable stars do not have a
$B$ or $I$-band counterpart. We will therefore not use the $BI$ data for
the period determination and broad classification of the variables. We will
however use the $BI$ data for the ``final'' classification of some
variables.
Next we searched for the periodicities for all 122 candidate
variables, using a variant of the Lafler-Kinman (1965) string-length
technique proposed by Stetson (1996). Starting with the minimum period
of $0.25\;days$, successive trial periods are chosen so
\begin{equation}
P_{j+1}^{-1}=P_{j}^{-1}-\frac{0.02}{\Delta t},
\end{equation}
where $\Delta t=t_{N}-t_{1}=398\;days$ is the time span of the series.
The maximum period considered is $150\;days$. For each candidate
variable 10 best trial periods are selected (Paper I) and then used in
our classification scheme.
The variables we are most interested in are Cepheids and eclipsing
binaries (EBs). We therefore searched our sample of variable stars for
these two classes of variables. As mentioned before, for the broad
classification of variables we restricted ourselves to the $V$ band
data. We will, however, present and use the $BI$-bands data, when
available, when discussing some of the individual variable stars.
For EBs, we used the search strategy described in Paper II. Within our
assumption the light curve of an EB is determined by nine parameters:
the period, the zero point of the phase, the eccentricity, the
longitude of periastron, the radii of the two stars relative to the
binary separation, the inclination angle, the fraction of light coming
from the bigger star and the uneclipsed magnitude. A total of six
variables passed all of the criteria. We then went back to the CCD
frames and tried to see by eye if the inferred variability is indeed
there, especially in cases when the light curve is very noisy/chaotic.
We decided to remove two dubious eclipsing binaries, classifying one as a
periodic variable with half the determined period. Its light curve is
presented in Section~6.3. The remaining four EBs with their parameters and
light curves are presented in the Section~6.1.
In the search for Cepheids we followed the approach by Stetson (1996)
of fitting template light curves to the data. We used the
parameterization of Cepheid light curves in the $V$-band as given by
Stetson (1996). There was a total of 64 variables passing all of the
criteria (Paper I and II), but after investigating the CCD frames we
removed 12 dubious ``Cepheids'', which leaves us with 52 probable
Cepheids. Their parameters and light curves are presented in
Section~6.2.
After the selection of four eclipsing binaries, 52 Cepheids and one
periodic variable, we were left with 65 ``other'' variable stars.
After raising the threshold of the variability index to
$J_{S,min}=1.2$ (Paper I) we are left with 17 variables. After
investigating the CCD frames we removed 10 dubious variables from the
sample, which leaves seven variables which we classify as
miscellaneous. Their parameters and light curves are presented in the
Section~6.4.
\section{Catalog of variables}
In this section we present light curves and some discussion of the 64
variable stars discovered by our survey in the field M31F.
\footnote{Complete $V$ and (when available) $BI$ photometry and
$128\times128\;pixel$ ($\sim 40''\times40''$) $V$ finding charts for
all variables are available from the authors via the {\tt anonymous
ftp} from the Harvard-Smithsonian Center for Astrophysics and can be
also accessed through the {\tt World Wide Web}.} The variable stars
are named according to the following convention: letter V for
``variable'', the number of the star in the $V$ database, then the
letter ``D'' for our project, DIRECT, followed by the name of the
field, in this case (M)31F, e.g. V244 D31F. Tables~\ref{table:ecl},
\ref{table:ceph}, \ref{table:per} and \ref{table:misc} list the
variable stars sorted broadly by four categories: eclipsing binaries,
Cepheids, other periodic variables and ``miscellaneous'' variables, in
our case meaning ``variables with no clear periodicity''.
\subsection{Eclipsing binaries}
In Table~\ref{table:ecl} we present the parameters of the four eclipsing
binaries in the M31F field. The light curves of these variables are
shown in Figure~\ref{fig:ecl}, along with the simple eclipsing binary
models discussed in the Papers I and II. The variables are sorted in the
Table~\ref{table:ecl} by the increasing value of the period $P$. For
each eclipsing binary we present its name, J2000.0 coordinates (in
degrees), period $P$, magnitudes $V_{max}, I_{max}$ and $B_{max}$ of
the system outside of the eclipse, and the radii of the binary
components $R_1,\;R_2$ in the units of the orbital separation. We
also give the inclination angle of the binary orbit to the line of
sight $i$ and the eccentricity of the orbit $e$. The reader should
bear in mind that the values of $V_{max},\;I_{max},\;B_{max},\;
R_1,\;R_2,\;i$ and $e$ are derived with a straightforward model of the
eclipsing system, so they should be treated only as reasonable
estimates of the ``true'' value.
One of the eclipsing binaries found, V1835 D31F, is a good DEB candidate.
However, a much better light curve is necessary to accurately establish the
properties of the system.
\begin{figure}[p]
\plotfiddle{M31F_ecl1.ps}{19.5cm}{0}{83}{83}{-260}{-40}
\caption{$BVI$ lightcurves of eclipsing binaries found in the field
M31F. The thin continuous line represents the best fit model for each
star and photometric band. $B$-band lightcurve is shown in the bottom
panel and $I$-band lightcurve (when present) is shown in the top
panel.}
\label{fig:ecl}
\end{figure}
\begin{small}
\tablenum{1}
\begin{planotable}{lrrrrrrrrcrl}
\tablewidth{40pc}
\tablecaption{\sc DIRECT Eclipsing Binaries in M31F}
\tablehead{ \colhead{Name} & \colhead{$\alpha_{J2000.0}$} &
\colhead{$\delta_{J2000.0}$} & \colhead{$P$} & \colhead{} & \colhead{} &
\colhead{} & \colhead{} & \colhead{} & \colhead{$i$} & \colhead{} &
\colhead{} \\
\colhead{(D31F)} & \colhead{$(\deg)$} & \colhead{$(\deg)$}
& \colhead{$(days)$} & \colhead{$V_{max}$} & \colhead{$I_{max}$}
& \colhead{$B_{max}$} & \colhead{$R_1$} & \colhead{$R_2$} &
\colhead{(deg)} & \colhead{$e$} & \colhead{Comments}}
\startdata
V207\ldots & 10.2157 & 40.6399 & 2.5202 & 20.53 & 20.34 & 20.40 & 0.48 & 0.39 & 81 & 0.05 & \nl
V4320\dotfill & 10.1076 & 40.7456 & 4.6694 & 19.46 & \nodata & 19.17 & 0.58 & 0.42 & 70 & 0.00 & \nl
V1835\dotfill & 10.1433 & 40.7184 & 6.3452 & 20.06 & 19.62 & 19.93 & 0.37 & 0.30 & 86 & 0.03 & DEB\nl
V763\dotfill & 10.1807 & 40.7263 & 6.8930 & 19.60 & 19.42 & 19.55 & 0.64 & 0.31 & 63 & 0.02 & W UMa\nl
\enddata
\label{table:ecl}
\end{planotable}
\end{small}
\subsection{Cepheids}
In Table~\ref{table:ceph} we present the parameters of 52 Cepheids in
the M31F field, sorted by the period $P$. For each Cepheid we present
its name, J2000.0 coordinates, period $P$, flux-weighted mean
magnitudes $\langle V\rangle$ and (when available) $\langle I\rangle$
and $\langle B\rangle$, and the $V$-band amplitude of the variation
$A$. In Figure~\ref{fig:ceph} we show the phased $B,V,I$ lightcurves
of our Cepheids. Also shown is the best fit template lightcurve
(Stetson 1996), which was fitted to the $V$ data and then for the $I$
data only the zero-point offset was allowed. For the $B$-band data,
lacking the template lightcurve parameterization (Stetson 1996), we
used the $V$-band template, allowing for different zero-points and
amplitudes. With our limited amounts of $B$-band data this approach
produces mostly satisfactory results, but extending the
template-fitting approach of Stetson (1996) to the $B$-band (and
possibly other popular bands) would be most useful.
Some Cepheids seem to be brighter in the $B$ band than in $V$. This
effect is most likely caused by blending, since these variables are
located in regions densely populated by stars, but could partially be
due to blue binary companions of Cepheids (Evans 1994; Evans \&
Udalski 1994).
\begin{figure}[p]
\plotfiddle{M31F_ceph1.ps}{19.5cm}{0}{83}{83}{-260}{-40}
\caption{$BVI$ lightcurves of Cepheid variables found in the field
M31F. The thin continuous line represents the best fit Cepheid
template for each star and photometric band. $B$ (if present) is
usually the faintest and $I$ (if present) is usually the brightest.}
\label{fig:ceph}
\end{figure}
\addtocounter{figure}{-1}
\begin{figure}[p]
\plotfiddle{M31F_ceph2.ps}{19.5cm}{0}{83}{83}{-260}{-40}
\caption{Continued.}
\end{figure}
\addtocounter{figure}{-1}
\begin{figure}[p]
\plotfiddle{M31F_ceph3.ps}{19.5cm}{0}{83}{83}{-260}{-40}
\caption{Continued.}
\end{figure}
\addtocounter{figure}{-1}
\begin{figure}[p]
\plotfiddle{M31F_ceph4.ps}{19.5cm}{0}{83}{83}{-260}{-40}
\caption{Continued.}
\end{figure}
\addtocounter{figure}{-1}
\begin{figure}[p]
\plotfiddle{M31F_ceph5.ps}{19.5cm}{0}{83}{83}{-260}{-40}
\caption{Continued.}
\end{figure}
\addtocounter{figure}{-1}
\begin{figure}[p]
\plotfiddle{M31F_ceph6.ps}{19.5cm}{0}{83}{83}{-260}{-40}
\caption{Continued.}
\end{figure}
\addtocounter{figure}{-1}
\begin{figure}[p]
\plotfiddle{M31F_ceph7.ps}{19.5cm}{0}{83}{83}{-260}{-40}
\caption{Continued.}
\end{figure}
\addtocounter{figure}{-1}
\begin{figure}[p]
\plotfiddle{M31F_ceph8.ps}{19.5cm}{0}{83}{83}{-260}{-40}
\caption{Continued.}
\end{figure}
\addtocounter{figure}{-1}
\begin{figure}[p]
\plotfiddle{M31F_ceph9.ps}{6.25cm}{0}{83}{83}{-260}{-230}
\caption{Continued.}
\end{figure}
\clearpage
\subsection{Other periodic variables}
For one of the variables preliminarily classified as an eclipsing
binary we decided upon closer examination to classify it as an ``other
periodic variable''. In Table~\ref{table:per} we present the
parameters of this possible periodic variable. In Figure~\ref{fig:per}
we show its phased $BVI$ lightcurves. We present its name, J2000.0
coordinates, period $P$, error-weighted mean magnitudes $\bar{V}$,
$\bar{I}$ and $\bar{B}$. To quantify the amplitude of the variability,
we also give the standard deviations of the measurements in the $BVI$
bands, $\sigma_{V},\sigma_{I}$ and $\sigma_{B}$.
The period of V7438 D31F was taken to be half of the period determined
by fitting a simple eclipsing binary lightcurve, so it should only be
treated as a first approximation of its true value. Inspection of the
$V,V-I$ and $V,B-V$ color-magnitude diagrams (Figure~\ref{fig:cmd})
reveals that the variable lands in the regions occupied by
Cepheids. V7438 D31F has been previously identified by Baade \& Swope
(1965) as a Cepheid with a period of 5.12 days, very close to our
value. In the P-L diagram (Figure~\ref{fig:pl}), however, it is
located above the region occupied by Cepheid variables, indicating
that it is possibly a blend. Another fact which may favor this
explanation is the small amplitude of its variability.
\begin{figure}[tp]
\plotfiddle{M31F_per1.ps}{6.45cm}{0}{83}{83}{-260}{-418}
\caption{$BVI$ lightcurves of the other periodic variable found in the
field M31F. $B$-band data (shown with the open circles) is the
faintest and $I$ is the brightest.}
\label{fig:per}
\end{figure}
\begin{figure}[p]
\plotfiddle{cmd.ps}{13cm}{0}{85}{85}{-260}{-135}
\caption{$V,\;V-I$ (upper panels) and $B,\;B-V$ (lower panels)
color-magnitude diagrams for the variable stars found in the field
M31F. The eclipsing binaries and Cepheids are plotted in the left
panels and the other periodic variables and miscellaneous variables
are plotted in the right panels. The dashed lines correspond to the
$I$ detection limit of $I\sim21\;{\rm mag}$ (upper panels) and the $B$
detection limit of $B\sim22.5.\;{\rm mag}$ (lower panels).
\label{fig:cmd}}
\end{figure}
\begin{figure}[t]
\plotfiddle{pl.ps}{8cm}{0}{50}{50}{-160}{-85}
\caption{Diagram of $\log{P}$ vs. $I$ for the Cepheids (open circles),
RV Tau (dotted circles) variables and the other periodic variable
(filled circle). The sizes of the circles are proportional to the $V$
amplitude of the variability.}
\label{fig:pl}
\end{figure}
\begin{small}
\tablenum{2}
\begin{planotable}{lrrrrrrr}
\tablewidth{35pc}
\tablecaption{DIRECT Cepheids in M31F}
\tablehead{ \colhead{Name} & \colhead{$\alpha_{J2000.0}$} &
\colhead{$\delta_{J2000.0}$} & \colhead{$P$} &
\colhead{} & \colhead{} & \colhead{} & \colhead{}\\
\colhead{(D31F)} & \colhead{(deg)} & \colhead{(deg)}
& \colhead{$(days)$} & \colhead{$\langle V\rangle$} & \colhead{$\langle I\rangle$}
& \colhead{$\langle B\rangle$} & \colhead{$A$} }
\startdata
V3441\ldots & 10.1201 & 40.7667 & 4.678 & 21.28 & 20.52 & 21.18 & 0.35 \nl
V4254\dotfill & 10.1111 & 40.6766 & 5.718 & 21.48 & 20.50 & 21.99 & 0.36 \nl
V7832\dotfill & 9.9963 & 40.7972 & 5.814 & 21.39 & 20.47 & 22.02 & 0.41 \nl
V3732\dotfill & 10.1189 & 40.6764 & 6.070 & 20.98 & 19.96 & 21.05 & 0.37 \nl
V3054\dotfill & 10.1243 & 40.7911 & 6.105 & 21.09 & 20.13 &\nodata& 0.31 \nl
V893\dotfill & 10.1716 & 40.7771 & 6.505 & 21.64 & 20.71 &\nodata& 0.40 \nl
V7441\dotfill & 10.0247 & 40.6533 & 6.514 & 21.43 & 19.88 &\nodata& 0.37 \nl
V3860\dotfill & 10.1154 & 40.7284 & 6.529 & 21.24 & 19.65 & 22.08 & 0.36 \nl
V1599\dotfill & 10.1470 & 40.7687 & 6.640 & 20.94 & 19.86 & 21.78 & 0.28 \nl
V5856\dotfill & 10.0799 & 40.6725 & 6.660 & 21.24 & 20.37 & 21.99 & 0.32 \nl
V5711\dotfill & 10.0828 & 40.6802 & 6.707 & 21.07 & 20.10 & 21.87 & 0.43 \nl
V6623\dotfill & 10.0591 & 40.6814 & 6.999 & 20.86 & 20.04 &\nodata& 0.35 \nl
V5886\dotfill & 10.0788 & 40.6903 & 7.458 & 21.12 & 20.54 & 21.91 & 0.36 \nl
V6406\dotfill & 10.0662 & 40.6569 & 7.563 & 20.95 & 20.24 &\nodata& 0.35 \nl
V3289\dotfill & 10.1232 & 40.7453 & 7.599 & 20.89 & 19.76 & 20.77 & 0.27 \nl
V5893\dotfill & 10.0797 & 40.6513 & 7.655 & 21.00 & 20.02 & 21.70 & 0.34 \nl
V6962\dotfill & 10.0462 & 40.6923 & 7.782 & 21.31 & 20.47 &\nodata& 0.24 \nl
V7741\dotfill & 10.0076 & 40.6885 & 8.099 & 21.03 & 20.17 & 20.81 & 0.24 \nl
V6098\dotfill & 10.0748 & 40.6395 & 8.471 & 20.52 & 19.52 &\nodata& 0.28 \nl
V4855\dotfill & 10.0966 & 40.8035 & 9.064 & 20.70 & 19.70 &\nodata& 0.30 \nl
V5498\dotfill & 10.0883 & 40.6611 & 9.387 & 20.72 & 20.27 & 20.98 & 0.24 \nl
V7074\dotfill & 10.0388 & 40.7778 & 9.478 & 20.78 & 19.70 & 21.51 & 0.30 \nl
V5097\dotfill & 10.0967 & 40.6804 & 9.662 & 20.66 & 19.92 & 21.16 & 0.25 \nl
V6483\dotfill & 10.0637 & 40.6739 & 9.736 & 20.83 & 20.13 & 21.20 & 0.23 \nl
V5994\dotfill & 10.0774 & 40.6532 & 9.886 & 20.67 & 19.53 & 21.45 & 0.22 \nl
V4556\dotfill & 10.1054 & 40.7022 & 9.894 & 20.54 & 19.41 & 21.20 & 0.25 \nl
V6195\dotfill & 10.0722 & 40.6463 & 9.924 & 21.21 & 20.05 & 22.54 & 0.35 \nl
V5178\dotfill & 10.0949 & 40.6807 & 9.932 & 20.30 & 19.30 & 20.18 & 0.22 \nl
V7393\dotfill & 10.0278 & 40.6364 & 9.937 & 20.50 & 19.51 & 21.15 & 0.31 \nl
V3550\dotfill & 10.1188 & 40.7607 & 10.468 & 20.55 & 19.85 & 21.03 & 0.32 \nl
V2320\dotfill & 10.1341 & 40.7656 & 10.868 & 20.27 & 19.54 & 20.60 & 0.34 \nl
V4125\dotfill & 10.1109 & 40.7332 & 11.139 & 20.46 & 19.63 & 21.11 & 0.31 \nl
V1549\dotfill & 10.1488 & 40.7569 & 11.764 & 20.72 & 19.68 & 21.49 & 0.35 \nl
V5442\dotfill & 10.0892 & 40.6766 & 11.902 & 19.65 & 18.58 & 19.45 & 0.16 \nl
V5696\dotfill & 10.0806 & 40.7520 & 12.287 & 20.74 & 19.54 & 21.62 & 0.43 \nl
V5598\dotfill & 10.0855 & 40.6741 & 12.311 & 20.34 & 19.45 & 20.96 & 0.37 \nl
V6267\dotfill & 10.0684 & 40.6995 & 12.324 & 20.38 & 19.30 & 19.27 & 0.17 \nl
V2156\dotfill & 10.1353 & 40.7925 & 12.831 & 20.37 & 19.40 & 20.76 & 0.27 \nl
V3373\dotfill & 10.1199 & 40.7971 & 12.872 & 20.94 & 19.55 & 21.81 & 0.41 \nl
V4955\dotfill & 10.0959 & 40.7808 & 13.034 & 20.70 & 19.67 & 21.45 & 0.41 \nl
V1633\dotfill & 10.1461 & 40.7760 & 13.043 & 21.30 & 19.85 &\nodata& 0.45 \nl
V1619\ldots & 10.1503 & 40.6545 & 13.141 & 20.61 & 19.45 & 21.47 & 0.34 \nl
V4682\dotfill & 10.1046 & 40.6552 & 13.297 & 20.73 & 19.30 & 21.72 & 0.34 \nl
V6640\dotfill & 10.0593 & 40.6558 & 13.761 & 19.67 &\nodata&\nodata& 0.18 \nl
V4861\dotfill & 10.0972 & 40.7849 & 13.991 & 20.47 & 19.31 & 21.22 & 0.41 \nl
V6503\dotfill & 10.0615 & 40.7215 & 15.210 & 20.77 & 19.55 & 21.44 & 0.38 \nl
V4708\dotfill & 10.1039 & 40.6631 & 15.690 & 20.50 & 19.33 & 21.38 & 0.35 \nl
V1893\dotfill & 10.1415 & 40.7417 & 16.756 & 18.79 & 17.47 & 19.54 & 0.18 \nl
V6208\dotfill & 10.0714 & 40.6561 & 17.572 & 19.74 & 18.78 & 20.28 & 0.48 \nl
V821\dotfill & 10.1760 & 40.7640 & 20.547 & 21.35 & 19.61 &\nodata& 0.55 \nl
V602\dotfill & 10.1880 & 40.7392 & 31.416 & 21.05 & 19.12 &\nodata& 0.31 \nl
V2203\dotfill & 10.1369 & 40.7306 & 55.373 & 17.94 & 17.17 & 18.32 & 0.17 \nl
\enddata
\label{table:ceph}
\end{planotable}
\end{small}
\subsection{Miscellaneous variables}
In Table~\ref{table:misc} we present the parameters of seven
miscellaneous variables in the M31F field, sorted by increasing value
of the mean magnitude $\bar{V}$. In Figure~\ref{fig:misc} we show the
unphased $VI$ lightcurves of the miscellaneous variables. For each
variable we present its name, J2000.0 coordinates and mean magnitudes
$\bar{V}, \bar{I}$ and $\bar{B}$. To quantify the amplitude of the
variability, we also give the standard deviations of the measurements
in $BVI$ bands, $\sigma_{V}, \sigma_{I}$ and $\sigma_{B}$. In the
``Comments'' column we give a rather broad sub-classification of the
variability.
All of the variables seem to represent the LP type of variability. A
closer inspection of the color-magnitude diagrams
(Figure~\ref{fig:cmd}) reveals that three variables (V667, V1229 and
V2285 D31F) land in the same area as Cepheids. Based on their
lightcurves it was possible to roughly estimate the periods of the
first two to be around 90 and 100 days, respectively. Using these
periods to place the stars on the P-L diagram (Figure~\ref{fig:pl})
suggests they may be RV Tauri type variables. V1665 and V1724 D31F are
most likely Mira-type variables, based on their location in the
color-magnitude diagrams.
\begin{small}
\tablenum{3}
\begin{planotable}{cccrccccccl}
\tablewidth{40pc}
\tablecaption{DIRECT Other Periodic Variables in M31F}
\tablehead{ \colhead{Name} & \colhead{$\alpha_{J2000.0}$} &
\colhead{$\delta_{J2000.0}$} & \colhead{$P$} &
\colhead{} & \colhead{} & \colhead{} & \colhead{} & \colhead{} & \colhead{} & \colhead{} \\
\colhead{(D31F)} & \colhead{(deg)} & \colhead{(deg)} &
\colhead{$(days)$} & \colhead{$\bar{V}$} &
\colhead{$\bar{I}$} & \colhead{$\bar{B}$} & \colhead{$\sigma_V$} &
\colhead{$\sigma_I$} & \colhead{$\sigma_B$} & \colhead{Comments} }
\startdata
V7438\ldots & 10.0252 & 40.6442 & 5.1 & 20.41 & 19.41 & 21.02 & 0.15 & 0.11 & 0.13 & Cepheid?\nl
\enddata
\label{table:per}
\tablecomments{V7438 was identified by Baade \& Swope (1965) as Cepheid
variable 232 with $P=5.12$ days.}
\end{planotable}
\end{small}
\begin{small}
\tablenum{4}
\begin{planotable}{llllllllll}
\tablewidth{35pc}
\tablecaption{DIRECT Miscellaneous Variables in M31F}
\tablehead{ \colhead{Name} & \colhead{$\alpha_{J2000.0}$} &
\colhead{$\delta_{J2000.0}$} & \colhead{} & \colhead{} &
\colhead{} & \colhead{} & \colhead{} & \colhead{} & \colhead{} \\
\colhead{(D31F)} & \colhead{(deg)} & \colhead{(deg)} &
\colhead{$\bar{V}$} & \colhead{$\bar{I}$} & \colhead{$\bar{B}$} &
\colhead{$\sigma_V$} & \colhead{$\sigma_I$} & \colhead{$\sigma_B$} &
\colhead{Comments} }
\startdata
V244\ldots & 10.2103 & 40.7380 & 18.39 & 16.68 & 19.87 & 0.12 & 0.07 & 0.05 & LP\nl
V1665\dotfill & 10.1460 & 40.7562 & 19.27 & 16.01 & 21.40 & 0.20 & 0.16 & 0.10 & LP\nl
V1724\dotfill & 10.1446 & 40.7499 & 19.63 & 16.34 & 0.00 & 0.24 & 0.13 & 0.00 & LP\nl
V2285\dotfill & 10.1373 & 40.6841 & 19.67 & 18.81 & 19.94 & 0.08 & 0.07 & 0.08 & RV Tau?\nl
V1229\dotfill & 10.1590 & 40.7389 & 20.47 & 19.57 & 21.07 & 0.23 & 0.10 & 0.18 & RV Tau?\nl
V764\dotfill & 10.1775 & 40.8110 & 20.51 & 0.00 & 0.00 & 0.35 & 0.00 & 0.00 & LP\nl
V667\dotfill & 10.1852 & 40.7265 & 21.18 & 19.33 & 0.00 & 0.35 & 0.22 & 0.00 & RV Tau?\nl
\enddata
\label{table:misc}
\end{planotable}
\end{small}
\subsection{Comparison with other catalogs}
The area of the M31F field coincides with two overlapping fields observed
by Baade. The catalogs of variable stars discovered in those fields are
given by Gaposchkin (1962, field II) and Baade \& Swope (1965, field III).
We succeeded in the cross-identification of all but one of the 55 Cepheid
variables found in field III with stars on our template. We have discovered
independently 24 of those Cepheids and found a very good agreement between
the period determinations. We have also confirmed the periods of 23 other
Cepheids, which eluded our detection, in large part due to their faintness
and the strict criteria we have imposed in our process of Cepheid selection
(see Table~\ref{table:crossid} for cross-ids).
Out of the 38 unique Cepheid variables listed in the field II catalog,
located within our M31F field, we have found 14 Cepheids and confirmed the
periods of two. The remaining field II Cepheids have evaded positive
cross-identification with our template stars.
Another overlapping variable star catalog is given by Magnier et
al.~(1997, hereafter Ma97). Out of the three variable stars in Ma97
which are in our M31F field, we cross-identified one, also classifying
it as a Cepheid. The other two did not qualify as variable star
candidates because of low $J_S$ values.
\begin{small}
\tablenum{5}
\begin{planotable}{lrrrrrrr}
\tablewidth{35pc}
\tablecaption{Cross-Identifications of the DIRECT Cepheid Variables in M31F}
\tablehead{ \colhead{Name} & \colhead{$P$} &\colhead{} &
\colhead{$P$} & \colhead{} & \colhead{$P$} & \colhead{} &\colhead{$P$}\\
\colhead{(D31F)} & \colhead{$(days)$} & \colhead{field II} &
\colhead{($days$)} & \colhead{field III} & \colhead{$(days)$} &
\colhead{Other} & \colhead{$(days)$}}
\startdata
V3441\ldots & 4.678 & & & 108 & 5.000 & & \nl
V4254\dotfill & 5.718 & 351 & 5.719 & 153 & 5.721 & & \nl
V3732\dotfill & 6.070 & 352 & 6.071 & 175 & 6.074 & & \nl
V3054\dotfill & 6.105 & & & 92 & 6.101 & & \nl
V7441\dotfill & 6.514 & 230 & 6.508 & & & & \nl
V3860\dotfill & 6.529 & & & 24 & 6.526 & & \nl
V1599\dotfill & 6.640 & & & 145 & 6.637 & & \nl
V5856\dotfill & 6.660 & 328 & 6.660 & 30 & 6.700 & & \nl
V5711\dotfill & 6.707 & 330 & 6.709 & 29 & 6.708 & & \nl
V6623\dotfill & 6.999 & 326 & 7.000 & & & & \nl
V5886\dotfill & 7.458 & 332 & 7.457 & 70 & 7.463 & & \nl
V6406\dotfill & 7.563 & 319 & 7.553 & & & & \nl
V5893\dotfill & 7.655 & 320 & 7.823 & & & & \nl
V6962\dotfill & 7.782 & 225 & 7.780 & & & & \nl
V7741\dotfill & 8.099 & 222 & 8.094 & & & & \nl
V6098\dotfill & 8.471 & 315 & 8.464 & & & & \nl
V4855\dotfill & 9.064 & & & 51 & 9.085 & & \nl
V5097\dotfill & 9.662 & 348 & 9.662 & 68 & 9.678 & & \nl
V6483\dotfill & 9.736 & 325 & 9.483 & & & & \nl
V4556\dotfill & 9.894 & 341 & 9.881 & 60 & 9.870 & & \nl
V6195\dotfill & 9.924 & 318 & 9.921 & & & & \nl
V7393\dotfill & 9.937 & 234 & 9.933 & & & Ma97 4& 9.0 \nl
V3550\dotfill & 10.468 & & & 72 & 10.461 & & \nl
V2320\dotfill & 10.868 & & & 109 & 10.858 & & \nl
V4125\dotfill & 11.139 & & & 20 & 11.147 & & \nl
V1549\dotfill & 11.764 & & & 54 & 11.766 & & \nl
V5696\dotfill & 12.287 & & & 17 & 12.286 & & \nl
V5598\dotfill & 12.311 & 329 & 12.312 & 135 & 12.358 & & \nl
V6267\dotfill & 12.324 & 334 & 12.294 & 133 & 12.284 & & \nl
V2156\dotfill & 12.831 & & & 128 & 12.821 & & \nl
V4955\dotfill & 13.034 & & & 14 & 13.051 & & \nl
V1633\dotfill & 13.043 & & & 107 & 13.021 & & \nl
V1619\dotfill & 13.141 & 415 & 13.125 & 31 & 0.000 & & \nl
V4682\dotfill & 13.297 & 357 & 13.293 & & & & \nl
V4861\dotfill & 13.991 & & & 74 & 13.966 & & \nl
V6503\dotfill & 15.210 & 339 & 15.232 & 150 & 15.216 & & \nl
V4708\dotfill & 15.690 & 355 & 15.699 & 114 & 15.625 & & \nl
V6208\dotfill & 17.572 & & 17.569 & & & H22 & 17.60 \nl
\enddata
\label{table:crossid}
\tablecomments{
field II refers to the catalog published by Gaposchkin (1962); field
III - Baade \& Swope (1965); Ma97 - Magnier et al. (1997); H - Hubble(1929)}
\end{planotable}
\end{small}
\begin{figure}[p]
\plotfiddle{M31F_misc1.ps}{19.5cm}{0}{83}{83}{-260}{-40}
\caption{$VI$ lightcurves of the miscellaneous variables found in the
field M31F. $I$ (if present) is plotted in the two right panels.
$B$-band data is not shown.}
\label{fig:misc}
\end{figure}
\addtocounter{figure}{-1}
\begin{figure}[p]
\plotfiddle{M31F_misc2.ps}{2cm}{0}{83}{83}{-260}{-495}
\caption{Continued.}
\end{figure}
\section{Discussion}
\begin{figure}[t]
\plotfiddle{xy.ps}{8cm}{0}{50}{50}{-160}{-85}
\caption{Location of eclipsing binaries (filled squares) and Cepheids
(open circles) in the field M31F, along with the blue stars
($B-V<0.4$) selected from the photometric survey of M31 by Magnier et
al.~(1992) and Haiman et al.~(1994). The sizes of the circles
representing the Cepheids variables are proportional to the logarithm
of their period. Fields II and III observed by Baade are marked with
dashed lines.
\label{fig:xy}}
\end{figure}
In Figure~\ref{fig:cmd} we show $V,\;V-I$ and $V,\;B-V$
color-magnitude diagrams for the variable stars found in the field
M31F. The eclipsing binaries and Cepheids are plotted in the left
panels and the other periodic variables and miscellaneous variables
are plotted in the right panels. As expected, the eclipsing binaries
occupy the blue upper main sequence of M31 stars. The Cepheid
variables group near $B-V\sim1.0$, with considerable scatter probably
due to the differential reddening across the field. The other periodic
variable is located on the CMD in the part occupied by Cepheids. The
miscellaneous variables are scattered throughout the CMDs and
represent several classes of variability. Two of them are very red
with $V-I>2.0$, and are probably Mira variables.
In Figure~\ref{fig:xy} we plot the location of eclipsing binaries and
Cepheids in the field M31F, along with the blue stars ($B-V<0.4$)
selected from the photometric survey of M31 by Magnier et al.~(1992)
and Haiman et al.~(1994). The sizes of the circles representing the
Cepheids variables are proportional to the logarithm of their
period. As could have been expected, both types of variables group
along the spiral arms, as they represent relatively young populations
of stars. Many Cepheid variables are located in the star-forming
region NGC206. We will explore various properties of our sample of
Cepheids in the future paper (Sasselov et al.~1999, in preparation).
\acknowledgments{We would like to thank the TAC of the F.~L.~Whipple
Observatory (FLWO) and the TAC of the Michigan-Dartmouth-MIT (MDM)
Observatory for the generous amounts of telescope time devoted to this
project. We are grateful to Bohdan Paczy\'nski for motivating us to
undertake this project and always helpful comments and suggestions.
We thank Lucas Macri for taking some of the data described in this
paper, Przemek Wo\'zniak for his FITS-manipulation programs and Eugene
Magnier for the Cepheid catalogs. The staff of the MDM and the FLWO
observatories is thanked for their support during the long observing
runs. KZS was supported by the Harvard-Smithsonian Center for
Astrophysics Fellowship. JK was supported by NSF grant AST-9528096 to
Bohdan Paczy\'nski and by the Polish KBN grant 2P03D011.12.}
|
\section{Introduction}
There exist many systems which exhibit relaxation times long enough to
keep them from reaching equilibrium on experimental time scales. Primary
examples are spin glasses and polymer glasses, but also systems as simple
as the Ising model when prepared in an arbitrary initial state, or phase
separation dynamics in systems with conserved parameter such as Ostwald
ripening in binary alloys. As a consequence, the relaxation depends
for all these system on the waiting time $ t_w $ already spent in the low
temperature phase: the systems age. To understand the aging phenomena
observed in these models one has to investigate their non-equilibrium
dynamics.
In this context the investigation of the non-equilibrium dynamics of
spherical spin models with two-spin interactions is interesting because
they
exhibit nontrivial dynamical behaviour, despite their simplicity, which
makes their
non-equilibrium dynamics exactly solvable. These systems never reach
equilibrium,
hence correlation and response functions depend on the waiting time even
in the
limit of large times \cite{curev,bray,cude}.
Our main aim is to complete for this class of models the analysis of
the spherical SK model presented in \cite{cude} by identifying
{\em all\/} relevant time scales of the problem. We are going to show
that, in
addition to the two time regimes found in \cite{cude}, there exists an
intermediate time scale $ t_p \gg 1 $ satisfying $ t_p/t_w \to 0 $ for
$t_w\to\infty$. It is this intermediate time scale on which the
fluctuation
dissipation theorem is beginning to be violated. Interest in this time
scale stems
from the fact that a thorough understanding of the dynamics at these
intermediate
times is important with respect to the study of the non-equilibrium
dynamics
in models more complicated than those considered in the present paper,
such
as that of the spherical p-spin glass with $ p > 2 $, because it is
the
behaviour at the time scale $ t_p $ which determines the behaviour at the
time scale $ t_w $ in a unique way. It is thus the key ingredient towards
the solution of the so far unsolved problem of selecting a unique
solution
within an infinite family of time reparametrization covariant solutions
on
diverging time scales, as has been demonstrated within a multi-domain
crossover scaling approach for the closely related problem of a slowly
dragged
particle in a random potential \cite{hordr}. The analysis of the simple
spherical spin models is presented here, because their behaviour at the
intermediate time scale can be studied analytically and in instructive
detail.
Moreover we shall see that, despite the similarity of these models to
the more difficult case of the spherical p-spin glass, their dynamics is
not
spin glass dynamics. This has been realized for some time from
considerations
concerning fluctuation dissipation ratios or parametric plots of an
integrated
response versus correlation (see e.g. \cite{curev,cukupe}).
Alternatively,
one may look at the thermoremanent magnetization (another form of
integrated
response) as a quantity sensitive to the complicated phase space
structure,
to distinguish spin glasses from the simpler magnetic systems. While the
thermoremanent magnetization, when plotted against logarithmic time,
exhibits
a waiting time dependent plateau in spin glasses, this plateau is absent
in
the models considered here.
We have organized our material as follows. In Sec. II we introduce the
models
and briefly review the general method for solving their non-equilibrium
dynamics,
as first presented in \cite{cude}. In Sec. III we specialize to the
spherical
SK model and to $d$ dimensional hyper-cubic spherical ferromagnets, which
independently of the dimension $d$ of the latter, exhibit formally the
same
type of long time non-equilibrium dynamics; exponents describing the
decay of
correlation and response for the latter vary, of course, with $d$. Time
scales
are identified and analyzed in Sec. IV, while Sec. 5 contains a
discussion of
our results.
\section{The model}
We consider spherical spin models with two spin interactions consisting of
$ N $ continuous spins $ s_i(t) $, $ i= 1,\ldots,N $, which satisfy for
all
times $ t $ the spherical constraint $ \sum_{i=1}^N {s_i}^2(t) = N $. The
Hamiltonian of the system is given by
\begin{equation}\label{spham}
H = -\frac{1}{2}\sum_{i\neq j}J_{ij}s_i s_j\ .
\end{equation}
The coupling matrix $ J_{ij} $ is supposed to be an arbitrary symmetric
matrix. Denoting the eigenvalues of the matrix $ J $ by $ a_i $,
$ i = 1,\ldots,N $, the system of Langevin equations, which describes the
dynamics of the model, decouples in terms of the projections $ s_{a_i}(t)
$
of the spins $ s_i(t) $ onto the eigenvectors
\begin{eqnarray}
\label{langevinsp}
\partial_t s_{a_i}(t) = (a_i - \mu(t))s_{a_i}(t) + h_{a_i}(t) && +
\xi_{a_i}(t)\ ,
\nonumber\\ && i = 1,\ldots,N\ ,
\end{eqnarray}
where $ h_{a_i}(t) $ is the corresponding component of an external
magnetic
field and $ \xi_{a_i}(t) $ is thermal Gaussian white noise with zero mean
and correlation $ \langle\xi_{a_i} (\tau + t_w)\xi_{a_j} (t_w)\rangle
= 2T\delta_{ij}\delta (\tau) $. The parameter $ \mu(t) $ is the Lagrange
multiplier enforcing the spherical constraint. Henceforth we will use
$ \langle\cdot\rangle $ to represent the average over the thermal noise.
If
it were not for the Lagrange parameter $ \mu(t) $, the dynamics
(\ref{langevinsp}) would just be that of $ N $ independent harmonic
oscillators under the influence of thermal noise. This means that solving
the
non-equilibrium dynamics of these models reduces to determining $ \mu(t)
$.
It was shown in reference \cite{cude} that for a given waiting time $ t_w
$
and given time separation $ \tau\geq 0 $ the autocorrelation
$ q(\tau,t_w) := 1/N\left[\sum_{i=1}^N\langle s_i(\tau + t_w)s_i(t_w)
\rangle\right]_J $ and response function $ r(\tau,t_w) := 1/N
\left[\sum_{i=1}^N \delta \left.\langle s_i(\tau+t_w)\rangle/
\delta h_i(t_w)\right|_{h=0}\right]_J $ of this class of models are in
terms
of
\begin{equation}\label{deflambda}
\Lambda(t) := \exp\left( 2\int_0^t ds\,\mu(s)\right)
\end{equation}
given by
\begin{eqnarray}\label{qsp}
q(\tau,t_w) =&&
\frac{\Lambda(t_w+\tau/2)}{\sqrt{\Lambda(\tau+t_w)\Lambda(t_w)}}\ \Bigg[1
- \\
&&T\int_0^\tau ds\,\frac{\Lambda(t_w+\tau/2-s/2)}{\Lambda(t_w+\tau/2)}\,
\langle\langle\exp(as)\rangle\rangle\Bigg]\nonumber
\end{eqnarray}
and
\begin{equation}\label{rsp}
r(\tau,t_w) = \sqrt{\frac{\Lambda(t_w)}{\Lambda(\tau+t_w)}}\,
\langle\langle\exp(a\tau)\rangle\rangle\ ,
\end{equation}
where we have specialized the expressions in \cite{cude} to the case of
zero
external field and constant temperature $ T $ and have chosen the initial
condition to be $ s_{a_i}(t=0) = 1 $. By $ \left[\cdot\right]_J $ we have
denoted a possible disorder average and by
$ \langle\langle\cdot\rangle\rangle $
we denote the integration $ \int da\,\rho(a)\cdot $ over an eigenvalue
density
$ \rho(a) $ which in the thermodynamic limit $ N\to\infty $ describes the
distribution of eigenvalues of the coupling matrix $ J $. The
quantity $ \Lambda(t) $ itself is determined by
\begin{equation}
\label{lambda}
\Lambda(t) = \langle\langle \exp(2at)\rangle\rangle + 2T\int_0^t ds\,
\Lambda(s)\langle\langle\exp(2a(t-s))\rangle\rangle\ ,
\end{equation}
which together with (\ref{qsp}) immediately implies
$ \Lambda(0) = q(0,t_w) = 1 $. Another dynamical observable we will be
interested in is the thermoremanent magnetization $ m_r(\tau,t_w) $.
Given that
the system is kept in a small magnetic field $ h $ in the time interval
$ [0,t_w] $ , the magnetization measured at time $ \tau + t_w $ is given
by
\begin{equation}\label{defrm}
m_r(\tau,t_w) = h\int_0^{t_w}ds\,r(\tau+s,t_w-s)\ .
\end{equation}
For models such as the $d$ dimensional ferromagnets considered in what
follows,
in which the interaction matrix has a geometrical structure, off-diagonal
correlations of the form $q_{ij}(\tau , t_w) = \langle s_i(\tau+t_w)
s_j(t_w)
\rangle$ are of course also of interest. We have not investigated them in
the present paper, however, as our main interest here is in results which
will have further bearing on spin glass models of the mean field type.
\section{Spherical SK model and spherical ferro\-magnet}
While the expressions given so far are valid for any choice of the
coupling
matrix $ J $ we now want to treat two special cases. Our aim is to solve
the non-equilibrium dynamics of these particular models in the limit of
large
waiting times $ t_w \gg1 $ by explicitly determining $ \Lambda(t) $. As
we are
only interested in the behaviour of the dynamical observables for times
$ \tau,t_w \gg 1 $ it is sufficient to determine the asymptotic behaviour
of
$ \Lambda(t) $ for $ t \gg 1 $. In the following we will discuss the
d-dimensional spherical ferromagnet and the spherical SK model. The
latter
is the special case $ p=2 $ of the disordered spherical p-spin model and
we will present the results found in \cite{cude} in a slightly different
form.
In the case of the spherical ferromagnet we consider a $ d $-dimensional
hyper-cubic lattice with periodic boundary conditions, whose lattice
constant we take to be unity and whose lattice sites with coordinate
vectors
$ \vec{x}_i $ are occupied by spins $ s_i $. The couplings are chosen to
be
ferromagnetic nearest neighbour interactions, whose strength is set to
unity.
The standard diagonalization procedure using Fourier modes \cite{bax}
yields for this choice of the matrix $ J $ in the limit $ N\to\infty $
for
the spectrum of eigenvalues and the eigenvalue density $ \rho^{fm}(a) $
the
result
\begin{eqnarray}\label{dichtefm}
\rho^{fm}(a) = \frac{1}{\pi}\int_0^{\infty}dy\, \cos(ay) &&
\left[J_0(2y)\right]^d\ ,\nonumber\\
&&\ a\in [-2d,2d]\ ,
\end{eqnarray}
where $ J_0(y) $ denotes the Bessel function of zeroth order.
The spherical SK model is defined by choosing the coupling matrix $ J $ to
be a random matrix whose entries $ J_{ij} $ are independent and
identically
distributed Gaussian random variables with zero mean and variance
$ [(J_{ij})^2]_J = 1/N $. A general result of random matrix theory
\cite{meht} states that, for
this choice of $ J $, the eigenvalue density in the thermodynamic limit
$ \rho^{sk} $ is given by the Wigner semi-circle law
\begin{equation}\label{dichtesk}
\rho^{sk}(a) = \frac{1}{2\pi}\sqrt{4 - a^2}\ ,\quad a\in[-2,2]\ .
\end{equation}
Solving the non-equilibrium dynamics of these models means solving
(\ref{lambda}) for the eigenvalue densities (\ref{dichtefm}) and
(\ref{dichtesk}). This is best done using the Laplace transform
$ \tilde{\Lambda}(s) = \int_0^{\infty}dt\,\Lambda(t)\exp(-st) $
to obtain from (\ref{lambda}) for $ t > 0 $ the relation
\begin{equation}\label{laplam}
\tilde{\Lambda}(s) = \frac{\tilde{f}(s)}{1-2T\tilde{f}(s)}\ ,
\end{equation}
where the function $ \tilde{f}(s) := \left\langle\left\langle
1/(s-2a)\right\rangle\right\rangle $ is characteristic of the given
model. In terms of the
function $ f(t) $, which yields $ \tilde{f}(s) $ via Laplace
transformation,
the expressions (\ref{qsp}) and (\ref{rsp}) can be rewritten as
\begin{eqnarray}\label{qfmsk}
q(\tau,t_w) =&&
\frac{\Lambda(t_w+\tau/2)}{\sqrt{\Lambda(\tau+t_w)\Lambda(t_w)}}\ \Bigg[1
\\ -
&&\frac{1}{2}T\int_0^{2\tau} dx\,
\frac{\Lambda(t_w+\tau/2-x/4)}{\Lambda(t_w+\tau/2)}\,f(x/4)\Bigg]\ .
\nonumber
\end{eqnarray}
and
\begin{equation}\label{rfmsk}
r(\tau,t_w) = \sqrt{\frac{\Lambda(t_w)}{\Lambda(\tau+t_w)}}\,f(\tau/2)\ ,
\end{equation}
respectively. Inserting the expressions for the eigenvalue densities
(\ref{dichtefm}) and (\ref{dichtesk}) in the definition of $ \tilde{f}(s)
$
we find that in the case of the spherical ferromagnet the function $ f(t)
$
is given by
\begin{equation}\label{ftfm}
f^{fm}(t) = [I_0(4t)]^d\ ,
\end{equation}
while for the spherical SK model it is calculated to be
\begin{equation}\label{ftsk}
f^{sk}(t) = \frac{I_1(4t)}{2t}\ .
\end{equation}
In these expressions $ I_0(t) $ and $ I_1(t) $ denote the modified Bessel
function of zeroth and first order, respectively.
The critical temperature of the dynamic phase transition is found from
(\ref{laplam}) to be
\begin{equation}\label{tcsp}
T_c = \frac{1}{2\tilde{f}(2a_m)}\ ,
\end{equation}
where $ a_m $ denotes the maximal eigenvalue of the eigenvalue spectrum
$ [-a_m,a_m] $ of the corresponding model. In the special cases
considered
here we have $ a_m^{fm} = 2d $ and $ a_m^{sk} = 2 $. Expression
(\ref{tcsp}) implies in the case of the spherical ferromagnet that we get
a
phase transition in $ d > 2 $ only. For the spherical SK model the
critical
temperature can be calculated explicitly and one finds $ T_c^{sk} = 1 $.
These results are in both models in agreement with the ones obtained from
static calculations of the transition temperature \cite{bk},\cite{ktj}.
In the following we will only be interested in the low temperature phase
with
$ T < T_c $. In this phase the asymptotic behaviour of $ \Lambda(t) $
for large times $ t $ is determined by the behaviour of the Laplace
transform
$ \tilde{\Lambda}(s) $ at the right bound $ s = 2a_m $ of the branch cut.
In the case of the spherical SK-model the inverse Laplace transformation
can
be done exactly and the result found in \cite{cude} reads
\begin{equation}\label{lambdask}
\Lambda^{sk}(t) = \frac{1}{T}\sum_{k=1}^{\infty} kT^k\,\frac{I_k(4t)}{2t}\ .
\end{equation}
For the spherical ferromagnet there is no simple analytic expression, but
we
can find the leading asymptotic behaviour for $ t\gg 1 $ in integer
dimension
$ d > 2 $ by expanding $ \tilde{\Lambda}(s) $ around $ s = 2a_m $ and
performing the inverse Laplace transformation. Performing this
calculation
and comparing the result with the leading order of expression
(\ref{lambdask})
for large times $ t\gg 1 $, we find that in both models the leading
asymptotic
behaviour of $ \Lambda(t) $ for $ t\gg 1 $ can be written as
\begin{equation}\label{asylambda}
\Lambda(t) \simeq \frac{\Lambda_0}{(1-T/T_c)^2}\frac{e^{2a_mt}}{t^{\nu_s}}
\hspace{1cm}t\gg 1\ ,
\end{equation}
where the prefactor $\Lambda_0$ is given by $ \Lambda_0^{fm} =
(8\pi)^{-\nu_s} $ for the spherical ferromagnet and by $ \Lambda_0^{sk} =
(32\pi)^{1-\nu_s} $ for the spherical SK model. The exponent $ \nu_s $
appearing in these expressions is $ \nu_s^{fm} = d/2 $ for the spherical
ferromagnet and $ \nu_s^{sk} = 3/2 $ for the spherical SK model. In the
same way it follows from (\ref{ftfm}) and (\ref{ftsk}) that the
asymptotic
behaviour of $ f(t) $ can in both models be written as
\begin{equation}\label{asyft}
f(t) \simeq \Lambda_0 \frac{e^{2a_mt}}{t^{\nu_s}}\hspace{1cm}t\gg 1\ ,
\end{equation}
with the same factor $ \Lambda_0 $ as in (\ref{asylambda}). These two
formulas
indicate already the close correspondence of the asymptotic behaviour of
autocorrelation and response function in both models which we will study
in more detail in the following section. We will see that we can write
the
expressions for these dynamical observables for both models in a unified
way
using the exponent $ \nu_s $ defined above. This means in particular that
we
will find the same time scales appearing in the limit of large waiting
times
$ t_w $ and large time separations $ \tau $.
\section{Asymptotic dynamics and time scales}
Before entering the discussion of the asymptotic behaviour for large
waiting times $ t_w $, we want to present for later reference the
expressions
for correlation and response function for finite waiting time $ t_w \sim
1 $
and large times $ \tau \gg 1 $. Using (\ref{asylambda}) and (\ref{asyft})
in the
terms containing the time variable $ t $ in (\ref{qfmsk}) and
(\ref{rfmsk})
we obtain for the autocorrelation
\begin{equation}\label{kurztwq}
q(\tau,t_w) \sim f_q(t_w)\tau^{-\nu_s/2}\hspace{1cm}\tau\gg t_w\simeq 1\ ,
\end{equation}
where $ f_q(t_w) \simeq 1 $ for $ t_w \sim 1 $, and for the
response
\begin{equation}\label{kurztwr}
r(\tau,t_w) \simeq f_r(t_w)\tau^{-\nu_s/2}\hspace{1cm} \tau \gg t_w\simeq
1\ ,
\end{equation}
where again $ f_r(t_w) \sim 1 $.
We are, however, mainly interested in the case of large waiting times
$ t_w \gg 1 $ and large time separations $ \tau \gg 1 $. In this case
we can insert the asymptotic expansions (\ref{asylambda}) and
(\ref{asyft})
in expressions (\ref{qfmsk}) and (\ref{rfmsk}) and get for $ \tau,t_w \gg
1 $
\begin{eqnarray}\label{asyq}
q(\tau,t_w) && \simeq
\frac{\left(1+\frac{\tau}{t_w}\right)^{\nu_s/2}}{\left(1
+\frac{\tau}{2t_w}\right)^{\nu_s}}
\times \\ &&
\Bigg[1-\frac{1}{2}T\int_0^{2\tau}dx\,e^{-a_mx/2}\frac{f(x/4)}{\left(1
-\frac{x}{4t_w(1+\tau/2t_w)}\right)^{\nu_s}}\Bigg] \nonumber
\end{eqnarray}
for the leading asymptotic behaviour of the autocorrelation and
\begin{equation}\label{asyr}
r(\tau,t_w) \simeq
b\left(1+\frac{\tau}{t_w}\right)^{\nu_s/2}\,\tau^{-\nu_s}
\end{equation}
for the response.
The prefactor $ b $ is $ b^{fm} = (4\pi)^{-\nu_s} $ in the case of the
ferromagnet and $ b^{sk} = (4\pi)^{1-\nu_s} $ for the SK model. These
equations show that the asymptotic dynamics in the limit $ \tau,t_w \gg 1
$
of the spherical ferromagnet and the spherical SK model possesses the same
{\em formal\/} structure. In particular we find that the scaling
behaviour of the dynamical observables autocorrelation and response of
the
spherical ferromagnet in $ d=3 $ and the spherical SK model is equivalent
(on the level of exponents). This result was independently stated in
\cite{curev}. The formal correspondence of the two models implies in
particular that in both models the same characteristic time scales appear.
Before entering the discussion of the relevant time scales in the problem,
let us simplify expression (\ref{asyq}) further. Expanding the
denominator of
the integrand in this expression in a power series, one can prove that in
the limit of large waiting times $ t_w \gg 1 $ the dominant contribution
to
the integral comes for all times $ \tau \gg 1 $ from the zeroth order
term of
the expansion. Defining
\begin{equation}\label{qeasp}
q_p := 1-\frac{T}{2}\int_0^{\infty}dx\,e^{-a_mx/2}f(x/4) = 1 -
\frac{T}{T_c}\ ,
\end{equation}
where the last equality follows from (\ref{tcsp}), we find
\begin{equation}\label{leadasyq}
q(\tau,t_w) \simeq
\frac{\left(1+\frac{\tau}{t_w}\right)^{\nu_s/2}}{\left(1
+\frac{\tau}{2t_w}\right)^{\nu_s}}\left(q_p + c_0\tau^{1-\nu_s}\right)
\end{equation}
for the leading behaviour of the autocorrelation in the limit $ \tau,t_w
\gg 1 $,
in which $c_0 = b T/(\nu_s -1)$.
Using (\ref{asyr}) and (\ref{leadasyq}) it is now straightforward
to identify the different time scales of the problem. At first sight
we find the two time scales already discussed
in \cite{cude} for the case of the spherical SK model. The first is the
time scale $ t_0 \sim 1 $ of the microscopic relaxation. At the upper end
of this scale we have $ \tau \gg 1 $ but still $\tau/t_w \ll 1 $, such
that we
can neglect all waiting time dependent corrections. On this time scale the
dynamics corresponds to the dynamics in equilibrium, i. e. it is time
translation invariant with autocorrelation $ q(\tau,t_w) =
\tilde{q}_0(\tau_0)$
and response $ r(\tau,t_w) = \tilde{r}_0(\tau_0) $ being functions of
the scaling variable $ \tau_0 := \tau/t_0 $ only, and autocorrelation and
response satisfy the FDT $ -\partial_{\tau} \tilde{q}_0(\tau_0)
= T\tilde{r}_0 (\tau_0) $ of equilibrium dynamics. Therefore we will
refer to
this time scale as the FDT regime. At the upper end $ \tau_0 \gg 1 $ of
this
scale the response is found from (\ref{asyr}) to be
\begin{equation}\label{fdtr}
r(\tau,t_w) = \tilde{r}_0(\tau_0) \simeq \hat b_0\,\tau_0^{-\nu_s}\ ,
\end{equation}
with $\hat b_0 = b t_0^{-\nu_s}$, while (\ref{leadasyq}) implies for the
correlation
\begin{equation}\label{fdtqsp}
q(\tau) = \tilde{q}_0(\tau_0) \simeq q_p + \hat c_0\tau_0^{1-\nu_s}
\end{equation}
with $\hat c_0 = c_0 t_0^{1-\nu_s}$. This corresponds to a power law
decay of
the correlation to a plateau value $ q_p $, which in the case of the
spherical
ferromagnet is just the square of the static spontaneous magnetization
$\langle s_i\rangle^2$ \cite{bk}, while in the case of the spherical SK
model
it is the static Edwards-Anderson parameter $ q_{EA} = \left[\langle s_i
\rangle^2\right]_J $ \cite{cude,ktj}. The exponent of the decay is
\begin{equation}
\nu_0 := 1-\nu_s
\end{equation}
which in the case of the spherical SK model is just the special case $
p=2 $
of the result found in \cite{hcsed} for the equilibrium decay of the
correlation of the spherical p-spin glass. If we speak of a plateau in the
correlation, it is of course understood that this plateau in the
autocorrelation is only visible in a plot against the logarithm of time
$\tau$.
The second obvious time scale is the waiting time itself. For $ \tau \sim
t_w $
correlation and response can be written as functions of the scaling
variable
$ \tau_w := \tau/t_w $ and
one finds asymptotically for $ \tau_w \gg 1 $ power law decays of the
dynamical observables to zero \cite{cude}.
From (\ref{asyr}) it is obvious that these two are the only time scales
which can be identified from the behaviour of the response function.
However,
it turns out that there exists a further nontrivial time scale in the
problem, which can be identified from the autocorrelation function.
Looking at
expression (\ref{leadasyq}) and taking into account the leading waiting
time
dependent correction in the prefactor we arrive at
\begin{equation}\label{compasyq}
q(\tau,t_w) \simeq
\left(1-\frac{\nu_s}{8}\left(\frac{\tau}{t_w}\right)^2\right)
\left(q_p + c_0\tau^{1-\nu_s}\right)\ .
\end{equation}
This expression shows that there exists a waiting time dependent scale
$ t_p(t_w) $, on which the correlation begins to decay away from the
plateau
value $ q_p $. To be more precise, we define this time scale $ t_p $ by
requiring $ q(t_p,t_w) = q_p $, such that $ t_p $ corresponds to the
middle
of the plateau of the correlation function. This means that $ t_p $ is
the
time for which the competing corrections in (\ref{compasyq}) are of the
same
order of magnitude. Hence we find that the time scale $ t_p(t_w) $ scales
as
\begin{equation}\label{tp}
t_p(t_w) \sim {t_w}^{2/(1+\nu_s)}\ll t_w
\end{equation}
with the waiting time $ t_w $. The latter inequality follows as $ \nu_s >
1 $.
The plateau regime corresponding to time scale $ t_p $ is the so far
missing
link between the stationary dynamics within the FDT regime and the
non-stationary dynamics for times of the order of the waiting time $ t_w
$
itself. We will shortly see that it is in particular the time scale on
which
the FDT of equilibrium dynamics is violated.
In terms of the scaling variable $ \tau_p := \tau/t_p $, the correlation
within
the plateau regime $ \tau \sim t_p $ can be expressed in the scaling form
\begin{equation}\label{qanstp}
q(\tau,t_w) = q_p + \hat{q}_p(t_w)\tilde{q}_p(\tau_p)
\end{equation}
with the prefactor $\hat{q}_p(t_w) = t_w^{-2(\nu_s - 1)/(\nu_s + 1)} \sim
t_p^{\nu_0}$ and the scaling function
\begin{equation}
\tilde{q}_p(\tau_p) \simeq c_0 \tau_p^{\nu_0} + c_p{\tau_p}^{\nu_1}
\label{obtpq}
\end{equation}
in which $ c_p = -\frac{\nu_s}{8}\,q_p $ and $ \nu_1 = 2 $ (recall that
$\nu_0 = 1-\nu_s < 0$). This scaling function describes the decay of the
correlation towards $q_p$ ad the lower end of the plateau scale, i.e.
for
$\tau_p \ll 1$, and its subsequent decay away from $ q_p $ at the upper
end
of the plateau scale where $\tau_p \gg 1$.
In order to study the violation of the FDT we introduce a quantity
$n(\tau,t_w)$ which characterizes this violation quantitatively \cite{hordr}
via
\begin{equation}\label{defn}
-\partial_\tau q(\tau,t_w) =: T\,(1+n(\tau,t_w))r(\tau,t_w)\ .
\end{equation}
Note that differentiating with respect to the time separation is
equivalent
to differentiating with respect to the later time $t$. Alternatively one
may
differentiate with respect to the earlier time $t_w$. In the
representation
of the correlation in terms of $t_w$ and time difference $\tau = t - t_w$
this
gives rise to the corresponding definition
\begin{equation}\label{defnw}
\hat\partial_{t_w} q(\tau,t_w) =: T\,(1+n_w(\tau,t_w))r(\tau,t_w)\ ,
\end{equation}
with $\hat\partial_{t_w} = \partial_{t_w} -\partial_\tau$. The latter is
related to the fluctuation dissipation ratios $X(\tau,t_w)$ studied e.g.
in
\cite{curev,cukupe} via $X(\tau,t_w) = 1/(1+n_w(\tau,t_w))$. For
$\tau,t_w
\gg 1$ we get
\begin{eqnarray}
1+n(\tau,t_w) && \simeq \frac{1}{\left(1
+\frac{\tau}{2t_w}\right)^{\nu_s}}
\times \label{noftau}
\\ && \Bigg[1 + \frac{\nu_s q_p}{4bT}\frac{\tau^{1+\nu_s}}{t_w^2} \frac{1}
{\left(1 +\frac{\tau}{t_w}\right)\left(1 +\frac{\tau}{2t_w}\right)} \Bigg]\ ,
\nonumber\\
1+n_w(\tau,t_w) && \simeq \frac{1}{\left(1
+\frac{\tau}{2t_w}\right)^{\nu_s}}
\times \label{nwoftau} \\
&& \Bigg[1+ \frac{\nu_s q_p}{4bT}\frac{\tau^{1+\nu_s}}{t_w^2} \frac{1}
{\left(1 +\frac{\tau}{2t_w}\right)} \Bigg]\ . \nonumber
\end{eqnarray}
Fig. 1 shows these two functions for $T=0.6\,T_c$, and $t_w = 10^{10}$ so that $t_p=10^8$.
\begin{figure}
[ht]
{\centering
\epsfig{file=nt.eps,width=0.500\textwidth}
\par}
\caption{$1+n(\tau,t_w)$ and \ $1+n_w(\tau,t_w)$ \ as functions of $\tau$ for $t_w=10^{10}$. Vertical arrows mark the plateau scale $t_p(t_w)$ and the
waiting time scale $t_w$.}
\vspace{0.2truecm}
\label{figure}
\end{figure}
As we have seen before, the decay towards the plateau satisfies the FDT,
that is, we have in leading order $ n(\tau,t_w) = n_w(\tau_tw) = 0 $ for
$ \tau \ll t_p $. On the intermediate time scale $ t_p =
t_w^{2/(1+\nu_s)}$,
however, we obtain the scaling form
\begin{equation}
n(\tau,t_w) \simeq n_w(\tau,t_w) \simeq\tilde{n}(\tau_p) =
\frac{\nu_s q_p}{4 b T}\,{\tau_p}^{1+\nu_s}\ ,
\end{equation}
which approaches zero at the lower end of the plateau scale (where the FDT
holds) but is non-zero (indicating FDT violation) for all $\tau_p = {\cal
O}(1)$,
and exhibits a power law divergence at the upper end of the $t_p$
scale. This can be traced back to the fact that the behaviour of the
response
function $ r(\tau,t_w) $ does not change on the time scale $ t_p $
whereas
that of the correlation does. It is this divergence which is responsible
for
the fact noted in \cite{curev,cukupe} that parametric representations of
integrated response $\chi(\tau,t_w) = \int_0^\tau d s r(s,t_w +\tau -s)$
versus correlation $q(\tau,t_w)$ saturate at the value $\chi = (1-
q_p)/T$ for
$q(\tau,t_w) \le q_p$ for the models considered in the present paper.
Note
that the nature of this divergence is in the large $t_w$ limit of course
not
detectable in such parametric plots, as it occurs entirely on the plateau
scale, on which $q(\tau,t_w)$ is basically arrested at $q_p$. It may
however
be obtained from the finite $t_w$-corrections to such plots which may be
extracted from (\ref{noftau}) and (\ref{nwoftau}).
Note that (\ref{noftau}) and (\ref{nwoftau}) imply that $n$ and $n_w$
exhibit
{\em different\/} scaling on the $t_w$ scale. With $\tau_w = \tau/t_w$ we
have
\begin{eqnarray}
1+n(\tau,t_w) && \simeq \frac{1}{\left(1 +\tau_w/2\right)^{\nu_s}}
\times \label{nnoftau}\\
&& \Bigg[1+ \frac{\nu_s q_p}{4bT}\,t_w^{\nu_s-1} \tau_w^{1+\nu_s} \frac{1}
{\left(1 +\tau_w\right)\left(1 +\tau_w/2\right)} \Bigg]\ ,\nonumber\\
1+n_w(\tau,t_w) && \simeq \frac{1}{\left(1 +\tau_w/2\right)^{\nu_s}}
\times \label{nnwoftau}\\
&& \Bigg[1+ \frac{\nu_s q_p}{4bT}\,t_w^{\nu_s-1} \tau_w^{1+\nu_s} \frac{1}
{\left(1 +\tau_w/2\right)} \Bigg]\ ,\nonumber
\end{eqnarray}
implying that both are infinite in the $t_w\to \infty$ limit, but show
different behaviour at large $\tau_w$ when $t_w$ is large but finite:
Whereas
$1 + n(\tau,t_w) \simeq \frac{\nu_s q_p}{bT}\, (2 t_w)^{\nu_s-1}\,
\tau_w^{-1}
\to 0$ as $\tau_w \to \infty$, we have $1+ n_w(\tau,t_w) \simeq
\frac{\nu_s
q_p}{bT}\, (2 t_w)^{\nu_s-1}$ in the same limit.
To summarize, the FDT is broken already on the time scale $ t_p $ rather
than only on the scale $ t_w $, the former being much smaller than the
latter when $ t_w $ becomes large, since $ t_p/t_w \to 0 $ as $ t_w \to
\infty $. Moreover, the divergence of $n$ and $n_w$ on the $t_p$ scale
implies that the QFDT solution, which was found in \cite{hcsed} for the
spherical p-spin glass with $ p >2 $ and in \cite {hk} for manifolds in
disordered potentials, does not exist in the case of the spherical SK
model
(and in the ferromagnetic systems). From (\ref{compasyq}) it is also
obvious that for the models considered here the plateau regime $ t_p $ is
the only further time scale in the problem. This, too, is in contrast to
the expectations for spherical p-spin glass with $ p >2 $, which will
be considered in a forthcoming paper \cite{neqps}.
\section{Discussion}
Considering the simplicity of the models discussed in the previous
sections,
the complexity of the dynamical behaviour seems rather astonishing.
However, in the case of spherical ferromagnet the explicit dependence on
the
waiting time of both correlation and response even in the limit
$ \tau\gg t_w\gg 1 $ is a well
known result in the theory of phase ordering kinetics \cite{bray}.
According
to the scaling hypothesis of coarsening dynamics there exists for large
times
$ t =\tau + t_w \gg 1 $ a single length scale $ L(t) $ in the system,
which can
be interpreted as the typical size of a domain at time $ t $. This means
that
for large times $ t\gg t_w \gg 1 $ the two-time-autocorrelation function
of
such a system is a function of the ratio of the two length scales
$ L(t)\gg L(t_w)\gg 1 $ only. The exact solution of the phase ordering
dynamics of the spherical ferromagnet yields for the autocorrelation at
$ T=0 $ in the limit of large times the result \cite{bray}
\begin{equation}\label{phoqasy}
q(t-t_w,t_w) = \left(\frac{4t t_w}{(t +t_w)^2}\right)^{d/4}\ ,
\end{equation}
which is just equation (\ref{asyq}) for $ T=0 $.
It has been noted that aging behaviour in the correlation functions of
coarsening systems has a simple interpretation in terms of domain growth
\cite{bray,curev,cukupe,be+}. This holds in particular for the emergence
of
the plateau scale. After the system has spent the waiting time $ t_w\gg1$
in
the low temperature phase, an arbitrary spin will on average be found in a
domain of size $ L(t_w) $. The autocorrelation of this spin will for short
times decay towards $ q_p $, which is the square of the local
spontaneous
magnetization, due to spin fluctuations within this domain. This decay is
equivalent to a decay within a local equilibrium state and satisfies time
translational invariance and FDT. The further asymptotic decay of the
autocorrelation away from this value $ q_p $ towards zero can only be
produced by a change of the environment of the chosen spin, which means
that a domain wall has to pass by its site. As the size of the original
domain grows as a power of the waiting time, it is very plausible that
the
time spent near the plateau value should also grow as a power of the
waiting time.
Since the growth of the domains and therefore the wandering of the
domain walls is a slow process the asymptotic decay towards zero is also
a
slow power law decay. For the spherical SK model, such arguments are of
course not available, as the model does not possess a geometry.
At the heart of it, the {\em formal\/} equivalence of the asymptotic
dynamics
for $ \tau,t_w \gg 1 $ of the spherical ferromagnet and the spherical SK
model
is due to the fact that both interaction matrices exhibit eigenvalue
densities
$\rho(a)$ whose behaviour at the upper end $a_m$ of the spectrum can be
characterized by a power law
\begin{equation}
\rho(a) \sim (a_m - a)^{\nu_s - 1}\ ,\quad {\rm as} \quad a\to a_m\ ,
\end{equation}
with the exponent $\nu_s$ introduced earlier. It is this feature which
determines the behaviour of correlation and response in these systems at
$\tau,t_w \gg 1$. The origin of the power law may be disorder, as in the
case
of the SK model and the semi-circle law, but it need not, as exemplified
by the
$d$-dimensional ordered systems. Thus aging in the spherical SK model
cannot
be interpreted as spin glass aging as it is observed experimentally
\cite{sveno,aohrm,vinsu} as well as in model calculations \cite{cuku1}
and simulations \cite{rie}. Indeed, it is well known that from a static
point
of view this model does not have the properties of a typical spin glass,
as
it has a replica symmetric solution for all temperatures and does not
possess
many degenerate ground states. The results above imply that the spherical
SK
model is neither a spin glass from a dynamical point of view, despite
the
existence of a plateau in the correlation function as it is observed in
spin
glasses and related systems \cite{hcsed,hordr}.
Obviously the autocorrelation is not a suitable quantity to distinguish
aging
in a spin glass from the simpler case of coarsening dynamics in magnetic
systems, whose nonequlibrium dynamics is determined by domain growth. A
dynamical observable which characterizes a spin glass, however, is given
by
the thermoremanent magnetization defined in (\ref{defrm}). This is due to
the
fact that it is the response function which is most sensitive to the
complex
phase space structure exhibited by spin glasses. To be more precise, the
particular metastable configurations of a spinglass depend strongly on a
magnetic
field. During the waiting time the system is expected to move to
configurations of
increasing stability. On the other hand, a state which is relatively
stable in a
given field might become less stable if the field is slightly changed.
This means
that, after a change of the field at $t_w$, the system has to move to new
states
of increasing stability. The time scale of this process depends on the
degree of
stability reached at $t_w$. This leads to a plateau in the thermoremanent
magnetization similar to the one found in the correlation function. This
will
be derived for the spherical $ p $-spin glass with $ p >2 $ in a
forthcoming paper
\cite{neqps}. A mechanism of this kind is of course absent in a
coarsening system
and as a consequence $ m_r(\tau,t_w) $ decays in the limit of large
waiting times $
t_w \gg 1 $ for all $ \tau\ll t_w $ as
\begin{equation}\label{shortrm}
m_r(\tau,t_w) \sim \tau^{1-\nu_s}.
\end{equation}
To prove this result, let us denote by $ t_1(t_w) $ a lower bound of the
waiting time scale satisfying $ \tau \ll t_1\ll t_w $. Let us
further choose a time $ t_2 $, such that $ t_w -t_2 \sim 1 $. Then we can
split the integration in (\ref{defrm}) as follows
\begin{eqnarray}\label{rmsplit}
m_r(\tau,t_w) \simeq && h\Big(\int_0^{t_1}ds\,r(\tau + s,t_w)\\
&& + \int_{t_1}^{t_2}ds\,r(s,t_w-s)
+\int_{t_2}^{t_w}ds\,r(s,t_w-s)\Big).\nonumber
\end{eqnarray}
Using (\ref{fdtr}) in the first integral we find that this term yields the
leading order contribution given in (\ref{shortrm}) as the contribution
from
the upper bound is negligible in the limit $ t_w \gg 1 $. In the last
integral
in (\ref{rmsplit}) the argument $ t_w-s $ is always of order unity and
with
(\ref{kurztwr}) we find that it scales as $ {t_w}^{-\nu_s/2} $ with the
waiting time, such that it is negligible in the limit of large waiting
times.
Thus we just have to consider the contributions from the middle of the
integration range for the remaining integral in (\ref{rmsplit}). Rewriting
this integral in terms of the scaling variable $ \sigma := s/t_w $
we get using (\ref{asyr}) that this term scales as $ {t_w}^{1-\nu_s} $
which
leads to (\ref{shortrm}) as the dominant contribution. Hence we have
indeed
found that in the type of models considered here the thermoremanent
magnetization does not exhibit a plateau in the limit $ t_w\gg 1 $ nor
does it depend on the waiting time for all times $ t\ll t_w $. For
coarsening systems this is what we expected as this relaxation stems
from spin fluctuations within a certain
domain, which do not know anything about the waiting time. As noted in
\cite{curev,cukupe} an alternative criterion to distinguish aging in
coarsening
systems from spin glass aging is the integrated response $\chi(\tau,t_w)$
mentioned
in Sec. IV. Both, the saturation of $\chi(\tau,t_w)$ and the absence
of a plateau in the thermoremanent magnetization are due to the same
reason,
namely due to the divergence of the function $n(\tau,t_w)$, equivalently
due
to the vanishing of the fluctuation dissipation ratio $X(\tau,t_w)$ on
the plateau scale.
Let us finally stress that the time scale $ t_p $ also appears in the
more
complicated case of the spherical $ p $-spin glass with $ p >2 $, the
spherical SK model being just the simplest of this class of models, and
it is
the behaviour of correlation and response on this time-scale which is
needed
to uniquely fix the dynamics at later times. This will be explicitly
shown in
a forthcoming paper \cite{neqps}.
\acknowledgements
It is a pleasure to thank H. Kinzelbach for numerous illuminating discussions.
|
\section{Introduction}
Gravitational lensing is one of the most powerful and direct methods
for studying the gravitational potentials of massive objects in the
universe. An important type of lensing study is `weak lensing.' It is
the study of mild systematic distortions of background sources as
their light rays are perturbed by gravitational fields on their way to
us. Weak lensing has already provided important results in the study
of galaxy clusters (e.g., Tyson et al.~\cite{tvw}; Bonnet et
al.~\cite{q2345}; Fahlman et al.~\cite{fahlman}; Squires et
al.~\cite{a2163}; Fischer et al.~\cite{q0957}; Clowe et
al.~\cite{clowe}; Hoekstra et al.~\cite{hoekstra}), halos of
individual galaxies (e.g., Brainerd et al.~\cite{brainerd}), and
large-scale structure (Schneider et al.~\cite{schneider98}). As the
techniques are becoming better understood, research is progressing to
the search for weaker and weaker distortions, which would enable the outer
regions of galaxy clusters and galaxy halos, as well as lensing
signals from large-scale structure (e.g., Jain \&
Seljak~\cite{jain97}; Kaiser~\cite{kaiser98}), to be studied.
To be able to detect such very weak signals, it is important to
accurately remove the dominant systematic effect affecting weak
lensing measurements: anisotropy of, and smearing by, the point-spread
function (PSF). In this paper we will first investigate the limits of
the most commonly-used technique for weak lensing analysis, devised by
Kaiser et al.~(\cite{ksb}, henceforth KSB). We will show that after
PSF anisotropy correction, residual effects on the order of 1\% shear
are difficult to avoid with this method, even for moderately elongated
PSF's. Since the ability to detect percent signals is important for a
variety of scientific questions, we have therefore devised a new
method which does not have such residuals, but which nevertheless has
noise properties comparable to those of the KSB method.
There are several other methods for PSF anisotropy correction in the
literature. The Autocorrelation Function method of Van Waerbeke et
al.~(\cite{waerbeke}) is a variant of the KSB method in which not
individual galaxy images, but the autocorrelation function of many of
them, is analyzed. The Bonnet \& Mellier~(\cite{bm}) method uses a
different aperture weighting function from KSB, and treats the PSF
convolution as a shear term. Fisher \& Tyson~(\cite{ft97}) convolve
the image with a kernel constructed to make the PSF rounder again.
A more sophisticated such kernel has recently been presented
by Kaiser~(\cite{k99}).
\section{Limitations of the KSB formalism for PSF anisotropy correction}
We first examine the Kaiser et al.~(\cite{ksb}) method, following our
earlier limited investigation in the context of analysis of Hubble
Space Telescope images (Hoekstra et al.~\cite{hoekstra}, Appendix~D).
\subsection{The KSB method}
The technique of weak (or statistical) lensing involves measuring the
systematic, gravitationally induced, distortion of background images
behind a gravitational lens. In the weak lensing regime small
background images are distorted by a shear $(\gamma_1,\gamma_2)$ and a
convergence $\kappa$, whose combined effect is represented by the
mapping
\begin{eqnarray}
\pmatrix{x\cr y}
&\to
\pmatrix{1-\kappa-\gamma_1 & -\gamma_2 \cr -\gamma_2 & 1-\kappa+\gamma_1}
\pmatrix{x\cr y}\cr
&\equiv
(1-\kappa)
\pmatrix{1-g_1 & -g_2 \cr -g_2 & 1+g_1}
\pmatrix{x\cr y}
\label{eq:dist}
\end{eqnarray}
where $g_i=\gamma_i/(1-\kappa)$. For simplicity, in what follows we
neglect $\kappa$ as it is small in the weak lensing regime, and
pretend we are deriving the true shear $\gamma$ instead of the reduced
shear $g$. Thus, our results on shape measurements are valid, but their
interpretation as a lensing signal may require consideration of the
$(1-\kappa)$ factor. Our analysis makes no assumptions on the
smallness of $g_i$, though.
Kaiser, Squires and Broadhurst (1995, henceforth KSB) describe a
method for recovering $(\gamma_1,\gamma_2)$ from images of distant
galaxies. Essentially, they derive galaxy ellipticities from weighted
second moments of the observed images, and then correct these for the
effects of the weight function and of smearing by the point spread
function (PSF). By averaging over many galaxies, which are assumed to
be intrinsically randomly oriented, the effect of individual galaxy
ellipticities should average out, leaving the systematic lensing
signal. The KSB method has proved to be very effective, especially in
the study of galaxy cluster potentials.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{psfd.3.ps}}
\caption{An example of a PSF with radially varying ellipticity, but
zero overall polarization. (Eq.~\ref{eq:psfone} with $\delta=0.3$).
Contours differ by a factor of $2^{1/2}$. The dotted curves are
circles, shown for comparison.}
\label{fig:modelpsf}
\end{figure}
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{dcoeff.ps}}
\caption{The polarization of the PSF of Figure~\ref{fig:modelpsf} as a
function of the weight function's Gaussian radius $w$. The plot shows
that with compact weight functions such as those that are required to
suppress photon noise, even mildly elongated PSF's may erroneously
yield a polarization of several percent.}
\label{fig:dcoeff}
\end{figure}
KSB define various ``polarizabilities'', which express the ratio
between an input distortion (gravitational shear or PSF anisotropy)
and the measured polarization
\begin{equation}
e=\left({I^w_{xx}-I^w_{yy}\over I^w_{xx}+I^w_{yy}},{2I^w_{xy}\over
I^w_{xx}+I^w_{yy}}\right)
\end{equation}
of an image $f(x,y)$. These polarizations are derived from weighted second
moments
\begin{equation}
\label{eq:wtfn}
I^w_{xx}=\int f(x,y) x^2 W(r) dx dy,\qquad\hbox{etc.,}
\end{equation}
of the image intensities, where $W$ is a weight function which goes
to zero at large radii. The weight function is required as otherwise
the sky noise in the outer parts of the image dominates the measured
moment. The significance of the measurement is optimized by taking the
weight function to be relatively compact, of a size comparable to the
image itself.
Details of the method can be found in KSB, and in Hoekstra et al.\
(1998, henceforth HFKS), where a few small errors in the formulae of
KSB were corrected. For the purposes of the present paper, it is
sufficient to know that in the KSB formalism, the ``smear
polarizability" $P^{\rm sm}$ defines the ratio between the PSF
anisotropy $p=(I_{xx}-I_{yy},2I_{xy})$, constructed from the
unweighted second moments $I_{ij}$ of the (normalized) PSF, and the
resulting change in image polarization $e$. The ``shear
polarizability" $P^{\rm sh}$ is the ratio between the applied shear
$(\gamma_1,\gamma_2)$ and the resulting change in the image
polarization. KSB show how the polarizabilities can be derived from
higher weighted moments of the observed PSF and galaxy images.
\subsection{How accurate is KSB?}
In the context of ground-based cluster weak lensing, the KSB method
works well. Nevertheless, it does involve some approximations. Now
that weaker and weaker signals are of interest, it is therefore
important to understand the limitations of the method. As already
discussed by HFKS, for strongly non-Gaussian PSF's the KSB method does
not completely correct PSF anisotropy. This is particularly true when
analyzing small galaxies in deep HST images, where it turns out that
the choice of weight function in eq.~\ref{eq:wtfn} is important.
A simple PSF model can be used to illustrate why such residuals are,
at some level, unavoidable. Consider the following PSF:
\begin{equation}
\label{eq:psfone}
P(x,y)=G(1+\delta,1)+G(4-\delta,4),
\end{equation}
where $G(a^2,b^2)$ is a unit-area Gaussian of $x$- and $y$-dispersions
$a$ and $b$. $\delta$ is a small parameter. The case $\delta=0.3$ is
plotted in Figure~\ref{fig:modelpsf}.
The PSF of equation \ref{eq:psfone} has exactly zero anisotropy $p$:
the second moments in $x$ and in $y$ are equal. However, the
ellipticity of the PSF varies with radius, which means that the
{\em weighted} second moments are not equal: weighting the central parts
more will enhance the $x$-moment preferentially. In fact, it is easy
to show that the polarization constructed with weighted moments is
$O(\delta)$. The precise result for a Gaussian weight function
$W=\exp(-{1\over2}r^2/w^2)$ is
\begin{equation}
e_1= {\frac{9\,{w^2}\,\left( 7 + 5\,{w^2} + {w^4} \right) \,\delta}
{2\,\left( 1+{w^2}\right) \,
\left( 4 + {w^2} \right) \,
\left( 20 + 16\,{w^2} + 5\,{w^4} \right) }} +
{{{\rm O}(\delta^2)}}.
\label{eq:dcoeff}
\end{equation}
The polarization of the $\delta=0.3$ PSF plotted in
Figure~\ref{fig:modelpsf} is plotted in Figure~\ref{fig:dcoeff}. It
has a value of 0.03 near $w=2$, roughly the radius of maximum
significance which should be used to minimize photon noise in the
polarization measurement.
The KSB polarizabilities are derived assuming that the PSF can be
written as the convolution of a {\em compact} anisotropic part with an
{\em extended} circular part (KSB eq.~A1). This assumption allows the
anisotropy to be characterized in terms of $p$ only. However, our
example shows that this assumption may be too restrictive: it
effectively couples the radial intensity profile of the PSF with its
ellipticity profile. For example, a single Gaussian with constant
ellipticity can be written as such a convolution, but a sum of two
elliptical Gaussian such as the PSF of eq.~\ref{eq:psfone} cannot. The
systematic errors that arise are the result of this.
\section{A new method}
\begin{figure*}
\begin{picture}(250,350)(0,20)
\put(160,330){\framebox(140,30){Measure PSF and ${\overline{g}}$ radii}}
\put(230,330){\vector(0,-1){20}}
\put(160,280){\framebox(140,30){Pick circular basis functions $f_n$}}
\put(230,280){\vector(0,-1){20}}
\put(160,230){\framebox(140,30){Until $\chi^2$ optimized over $\gamma_1,\gamma_2$:}}
\put(230,230){\vector(0,-1){20}}
\put(200,180){\framebox(60,30){Pick $\gamma_1,\gamma_2$}}
\put(230,180){\vector(0,-1){20}}
\put(145,130){\framebox(170,30){Construct sheared basis functions $f_n^\gamma$}}
\put(230,130){\vector(0,-1){20}}
\put(145,080){\framebox(170,30){Construct PSF-smeared $F_n^\gamma$ }}
\put(230,080){\vector(0,-1){20}}
\put(115,030){\framebox(230,30){Optimize $\chi^2=\int dx\, dy\,(\sum_n a_n F_n^\gamma-{\overline{g}})^2$
over $a_n$}}
\put(115,045){\vector(-1,0){40}}
\put(075,045){\vector(0,1){200}}
\put(075,245){\vector(1,0){85}}
\put(390,245){\oval(80,30)}
\put(300,245){\vector(1,0){50}}
\put(350,230){\makebox(80,30){Best-fit $(\gamma_1,\gamma_2)$}}
\end{picture}
\caption{The schematic algorithm used to derive the shear from an
observed mean galaxy and PSF image.}
\label{fig:algo}
\end{figure*}
Here we present a new method, with which the PSF effects can be
corrected for with greater accuracy. The essence of the method is not
to work with the moments of the observed images; instead each image is
fit directly as a PSF-convolved, sheared circular source of unknown
radial profile.
Assume for the moment that we have managed to sum the images of many
galaxies into an `average galaxy' image ${\overline{g}}(x,y)$.
Analysing a stacked galaxy image is similar to the approach discussed
by Lombardi \& Bertin (\cite{lb}), who average image second moments
before corrections are applied. It differs from methods such as KSB or
Bonnet \& Mellier (\cite{bm}) in which galaxies are individually
corrected for PSF effects before they are combined to produce a shear
estimate.
Intrinsically, ${\overline{g}}$ is circular if the galaxies are
randomly oriented, but the image we observe has been distorted first
by gravitational lensing shear, then by the atmospheric seeing, and
finally by the camera optics. The observed ${\overline{g}}$ is
therefore a sheared circular source, convolved with a (known) PSF. We
therefore fit ${\overline{g}}$ directly to such a model, with the
minimum of further assumptions. This approach addresses the apparent
difficulty in the KSB methodology in the case of radially changing
ellipticity profiles: a sheared circular source has constant
ellipticity at all radii, and so after convolution with the PSF only a
subset of ``allowed'' ellipticity profiles remain.
Assuming that the PSF is known, e.g., from analysis of star images in
the field, the model for ${\overline{g}}$ is specified by an unknown
radial brightness profile, and by the shear parameters
$(\gamma_1,\gamma_2)$ that we are interested in. In practice we model
the radial profile as the superposition of several Gaussians of
different fixed widths, and unknown amplitude. We have found that the
following recipe for assigning the basis functions gives good results:
(i) determine the best-fit circular Gaussian radii to the observed PSF
and galaxy images, $r_{\rm PSF}$ and $r_{\rm GAL}$. (ii) Take
$r=(r_{\rm GAL}^2-r_{\rm PSF}^2)^{1/2}$ as an estimate for the
intrinsic radius of ${\overline{g}}$. (iii) Use four components to
describe the radial profile of $\overline{g}$, with
Gaussian radii $(0.5,1,2,4)\times r$.
The algorithm is laid out in Figure~\ref{fig:algo}. We now describe
the results of tests to verify the accuracy of the PSF anisotropy
correction, and to investigate how well it fares in the presence of
noise in the images.
\subsection{Simulations in the absence of noise}
We tested how well PSF anisotropy can be corrected for by considering
the case where there is no gravitational shear, only a range of PSF
shapes of varying anisotropy. An accurate analysis should yield zero
shear after correction for the PSF. On a large number of model images,
described below, we compared the results of the algorithm of
Figure~\ref{fig:algo} with those from the KSB algorithm as described
in HFKS (implying in particular that the same weight function is used
in the derivation of polarizations and polarizabilities of galaxy and
PSF images). The weight function was taken to be the best-fit circular
Gaussian to the post-seeing galaxy image.
The algorithm described in this paper directly yields an estimate for
the shear. In the comparisons, the KSB galaxy polarization after
seeing anisotropy correction was divided by the ``pre-seeing shear
polarizability" $P^\gamma$, for which we use the expression given by
Luppino \& Kaiser (1997).
\subsubsection{Double-Gaussian images and PSF}
In most of our simulations, we modeled the round average galaxy images as
\begin{equation}
\label{eq:gal}
{\overline{g}}=G(r_g)+G(k_g r_g)
\end{equation}
where $G(\sigma)$ is a unit-integral Gaussian of dispersion $\sigma$
(a double-Gaussian PSF was also considered by ???REF???). The
parameter $k_g$ is unity for a Gaussian profile, and is larger for
more radially extended profiles. A reasonable, though admittedly crude,
approximation to an exponential profile is given by setting $k_g=2$,
while $k_g=3$ gives a reasonable approximation to a de vaucouleurs
profile (Figure \ref{fig:kprofiles}).
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{k23profiles.ps}}
\caption{The double-Gaussian profiles used in the modeling in this
paper. Left, the $k=2$ profile is plotted logarithmically to show its
similarity to an exponential profile; right the $k=3$ profile is
plotted logarithmically vs. $r^{1/4}$ to show it is similar to a de
Vaucouleurs model.}
\label{fig:kprofiles}
\end{figure}
The PSF's were modeled in a similar way, but with anisotropy. Writing
now $G(a,b)$ for a Gaussian with $x$- and $y$-dispersions $a$ and $b$,
we have
\begin{equation}
\label{eq:psf}
PSF=
G\left(r_p,(1-\epsilon_1)r_p\right)+G\left(k_pr_p,(1-\epsilon_2)k_pr_p\right).
\end{equation}
Again we introduced a shape parameter $k_p$, but we also included
ellipticities $\epsilon_i$ for the two components. We considered three
kinds of PSF ellipticity profile: we either set
$\epsilon_1=\epsilon_2$ (constant ellipticity with radius), or set one
of the $\epsilon$'s to zero, to give a radial increase or decrease of
the PSF anisotropy. These three possibilities, though by no means
exhaustive, form a representative set of PSF's.
\begin{figure*}
\resizebox{\hsize}{!}{\includegraphics{ksb82.ps}}
\caption{The result of correcting simulated unsheared images for PSF
anisotropy, following the KSB method (solid lines) and the method
presented here (dashed lines). $\delta\gamma_1$ is in each case the
shear that is deduced after the PSF correction, and should be zero for
a perfect analysis. The $k$'s are luminosity profile shape parameters
for galaxy and PSF, and are explained in the text. In each panel the
heavy line represents the case where the galaxy image is intrinsically
of the same radius as the PSF ($r_g=r_p$ in eqs. \ref{eq:gal} and
\ref{eq:psf}), the lighter lines those where the galaxy is 0.5 (upper)
and 1.5 (lower) times this size. The PSF ellipticity is constant with
radius in these simulations. While for the $k_{\rm PSF}=1$ case
(Gaussian PSF) the KSB method leaves a residual which is third-order
in PSF ellipticity, other PSF luminosity profiles give rise to
first-order residuals.
The residuals of the new method in the upper panels disappear if more
radial components are used in the fit for $\overline{g}$, highlighting
that the dominant source of error in this method is the extent to
which the radial profile is modeled correctly.
}
\label{fig:rnda}
\end{figure*}
\begin{figure*}
\resizebox{\hsize}{!}{\includegraphics{ksb83.ps}}
\caption{As figure \ref{fig:rnda}, but only the outer component of the
PSF is elliptical.}
\label{fig:rndb}
\end{figure*}
\begin{figure*}
\resizebox{\hsize}{!}{\includegraphics{ksb84.ps}}
\caption{As figure \ref{fig:rnda}, but only the inner component of the
PSF is elliptical.}
\label{fig:rndc}
\end{figure*}
The advantage of the multiple-Gaussian formulation is that the PSF
convolution can be done analytically. We thus constructed a large
number of PSF-smeared galaxy images, calculated the polarizations and
polarizabilities following HFKS, subtracted the PSF anisotropy correction
\[\delta e=P^{{\rm sm}\, g} (P^{\rm sm\,\star})^{-1}e^\star\]
to the galaxy polarization, and divided the result by Luppino and
Kaiser (1997)'s pre-seeing shear polarizability $P^\gamma$. We then
compared these results with the results of our implementation of the
new fitting algorithm of Figure~\ref{fig:algo}.
The results of the simulations are
presented in figures \ref{fig:rnda}, \ref{fig:rndb} and
\ref{fig:rndc}.
They show that the KSB method can suffer from systematic
residuals around the 0.01 shear level once the PSF ellipticity
exceeds 0.2 or so, whereas this is not so for the new method developed
here. The systematic effects are strongest for small galaxies, for PSF
profiles with long tails, and for radially increasing PSF
ellipticity. The effect is clearly driven by the PSF shape,
not by the galaxy brightness profile.
Notice that in the constant-ellipticity case (Figure~\ref{fig:rnda}),
with a Gaussian PSF ($k_{\rm psf}=1$) the residuals left by the KSB
method are high order in PSF ellipticity, but that for non-Gaussian
PSF's a low-order residual dominates. (We have verified this result
analytically using symbolic mathematics.) This is a consequence of the
fact that only the single elliptical Gaussian PSF can be written as a
convolution of a compact anisotropic function with a round extended
one, as assumed in the KSB derivation. It is clearly important to test
algorithms not only for single-Gaussian PSF's!
\subsubsection{A WFPC-2 PSF}
In order to test whether our results are specific to the
double-Gaussian formulation of the PSF, a test was also performed with
a model PSF for the WFPC-2 camera on the Hubble Space Telescope. The
model was generated with the Tiny TIM software package, provided
on-line at STScI by J.~Krist. An oversampled PSF was calculated for a
position near the corner of CCD\#4, and convolved with a Gaussian
circular galaxy of FWHM 0.25arcsec. This `galaxy' and the PSF
(Figure~\ref{fig:wfpc}) were then binned to a resolution of half a
WFPC-2 pixel to avoid under-resolving the PSF, and analyzed as
above. The results are summarized in table~\ref{tab:wfpcsim}, and
confirm the results obtained from the large number of double-Gaussian
simulations described earlier.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{ksb10.ps}}
\caption{PSF and simulated galaxy image for an observation in the
corner of one of the WFPC2 CCD's. Axis units are 0.05 arcsec.}
\label{fig:wfpc}
\end{figure}
\begin{table}
\caption[Simulation with a WFPC-2 PSF]
{Results of a simulation based on a WFPC-2 PSF calculated
using the TinyTim software. In agreement with earlier results (HFKS),
the KSB technique appears to over-correct for the anisotropic WFPC-2
PSF images slightly.}
\label{tab:wfpcsim}
\begin{tabular}{ll}
Weight function radius: &1.3 WFPC-2 pixels\\
PSF polarization:&$(-0.078,0.024)$\\
Uncorrected galaxy $e$:&$(-0.020,-0.006)$\\
PSF-corrected galaxy $\gamma$:&$(-0.000,-0.019)$ (KSB)\\
&$(-0.000,-0.001)$ (new method)\\
\end{tabular}
\end{table}
\subsection{Noise properties}
\subsubsection{Analytic estimate}
\label{sec:noise-an}
The error on the estimated shear due to photon noise can be estimated as
follows. Let the 1-$\sigma$ error on each pixel of $\overline g$ be
$\sigma$ (for simplicity we take this to be the same on every pixel,
appropriate for background-limited work). Then the fit involves
finding the minimum of
\begin{equation}
\chi^2=\sum_k\left[\overline{g}_k-(P\otimes f({\mathbf
x}\cdot{\mathbf\Gamma}^2\cdot{\mathbf x}))_k\right]^2/\sigma^2
\end{equation}
where $P(x,y)$ is the PSF, $f(r^2)$ is the intrinsic radial profile of
the average galaxy, $\otimes$ denotes convolution, ${\mathbf x}_k$ is
the position of the $k$th pixel, and ${\mathbf\Gamma}$ is the
distortion matrix of equation~\ref{eq:dist}. If the fit parameters
$\gamma_i$ are uncorrelated with the radial profile, their inverse
variances are given by $\frac12\partial^2\chi^2/\partial \gamma_i^2$. For
example, at the best fit
\begin{eqnarray}
\frac12{\partial^2\chi^2\over\partial \gamma_1^2}
=&\displaystyle
\sum_k\left({\partial\over\partial \gamma_1}
(P\otimes f({\mathbf x}\cdot{\mathbf\Gamma}^2\cdot{\mathbf
x}))_k\right)^2/\sigma^2,\cr
=&\displaystyle
\sum_k\left(P\otimes [4f'(r^2)(y^2-x^2)]\right)^2/\sigma^2.
\label{eq:noise-an}
\end{eqnarray}
The right-hand side of equation~\ref{eq:noise-an} can be estimated
assuming that the PSF and observed average galaxy are Gaussians with
dispersions $r_{\rm PSF}$ and $r_{\rm GAL}$ pixels of integral 1 and
$F$, respectively. Then the 1-$\sigma$ error on $\gamma_1$ evaluates to
\begin{eqnarray}
\sigma(\gamma_1)=&
\displaystyle
\left(\frac12{\partial^2\chi^2\over\partial
\gamma_1^2}\right)^{-1/2}
={2\pi^{1/2}r_{\rm GAL}^3\sigma \over (r_{\rm GAL}^2-r_{\rm PSF}^2)F}\cr
=&
\displaystyle
{r_{\rm GAL}^2\over (r_{\rm GAL}^2-r_{\rm PSF}^2)}{\delta F\over F},
\label{eq:noise-an2}
\end{eqnarray}
where we have used the results that the PSF-fitting error on $F$ for a
Gaussian source is $\delta F=2\pi^{1/2}r_{\rm GAL}\sigma$. (The error
on $\gamma_2$ is the same.) We have verified this formula by means of
simulations, similar to those described below. Equation
\ref{eq:noise-an2} shows the expected increase in noise for small
objects, as well as the limit, even for large objects, of
\begin{equation}
\sigma(\gamma_i)\la{\delta F\over F}.
\end{equation}
\subsubsection{Simulations}
We have checked the sensitivity to noise in the images by Monte Carlo
simulation. Many realizations of random Gaussian noise superimposed on
a PSF-smeared, intrinsically round galaxy image were analyzed
with both algorithms, and the distributions of the resulting
$(\gamma_1,\gamma_2)$ estimates
compared. Selected results are shown in Table~\ref{tab:noise}.
Interestingly, the dispersions in the shears derived with both methods
are very similar over a range of galaxy sizes. As we have
already seen, the small bias in the results from KSB is present in the
simulations with non-circular PSF's, but not in the method advocated in
this paper.
A possible way to avoid the systematic residuals of the KSB method is
to increase the radius of the weight function $W$ in
eq.~\ref{eq:wtfn}, since the problems arise from the imperfect way in
which the polarizabilities represent the effect of $W$. However, the
primary function of $W$ is to control the noise in the
images. Doubling the Gaussian radius of $W$ does in fact improve the
anisotropy correction in the mean, but at the cost of almost doubling
the noise on the result (see Table~\ref{tab:noise}).
A by-product of our algorithm is an estimate of the intrinsic radial
profile of $\overline{g}$. In practice, this estimate appears to be
rather sensitive to the noise, especially for small images---not
surprising given that this is effectively a deconvolution, albeit a
constrained one. Nevertheless, it may be possible to use the
information in the best-fit radial profile in several ways. If a
suitable prior for the intrinsic radial profile of the average galaxy
selected can be formulated (e.g., by combining results over a wide
field, or from deeper, higher resolution images), this information
might help to refine the best-fit shear solution
further. Alternatively, the width of $\overline{g}$ might be used to
attempt to derive the lensing convergence $\kappa$ directly, since in
principle it is a direct measure of the magnification of faint
galaxies. This possibility is yet to be explored in detail, but is
likely to be difficult in practice.
\begin{table*}
\caption[Noise Simulations] {Results from representative noise
simulations of the KSB method and the one presented in this paper. In
each case, 100 noise realizations (noise per pixel of 0.001, with
$r_{\rm psf}=2$pixels, $k_{\rm gal}=2$ and total flux 4) were analyzed
with the standard KSB method, with the KSB method using a weight
function double the radius of the best-fit Gaussian, and with the new
method described in this paper. The first six simulations were of
cases without PSF anisotropy, and in the last six the PSF has a
constant axis ratio of 0.7. In all cases, the dispersions of the
standard KSB method and the new one are very similar, but note the
imperfect correction from the KSB method. The simulations with a
larger weight function show that the PSF anisotropy is corrected
better, but at the cost of increased noise.}
\label{tab:noise}
\begin{tabular}{ll@{$\qquad$}ll@{$\qquad$}ll@{$\qquad$}l}
\multicolumn{4}{c}{KSB$\quad$}&
\multicolumn{2}{c}{This paper$\quad$}\\
\multicolumn{2}{c}{Standard $W$$\quad$}&\multicolumn{2}{c}{Wider $W$$\quad$}&
\multicolumn{2}{c}{(4 radial cpts.)$\quad$}\\
Mean & \multicolumn{1}{c}{$\sigma$} &
Mean & \multicolumn{1}{c}{$\sigma$} &
Mean & \multicolumn{1}{c}{$\sigma$} &Comments\\
0.0010 & 0.0084 & 0.0016 & 0.0142 & 0.0010 & 0.0079 &
Round Gaussian PSF, $r_{\rm gal}=0.5r_{\rm PSF}$\\
0.0008 & 0.0067 & 0.0009 & 0.0105 & 0.0008 & 0.0061 &
Round Gaussian PSF, $r_{\rm gal}=r_{\rm PSF}$\\
0.0008 & 0.0072 & 0.0007 & 0.0113 & 0.0009 & 0.0066 &
Round Gaussian PSF, $r_{\rm gal}=1.5r_{\rm PSF}$\\
0.0019 & 0.0149 & 0.0031 & 0.0310 & 0.0023 & 0.0150 &
Round $k=3$ PSF, $r_{\rm gal}=0.5r_{\rm PSF}$\\
0.0012 & 0.0104 & 0.0014 & 0.0206 & 0.0017 & 0.0110 &
Round $k=3$ PSF, $r_{\rm gal}=r_{\rm PSF}$\\
0.0012 & 0.0105 & 0.0012 & 0.0186 & 0.0014 & 0.0107 &
Round $k=3$ PSF, $r_{\rm gal}=1.5r_{\rm PSF}$\\
0.0116 & 0.0056 & 0.0045 & 0.0097 & 0.0011 & 0.0061 &
$\epsilon=0.3$ Gaussian PSF, $r_{\rm gal}=0.5r_{\rm PSF}$\\
0.0041 & 0.0054 & 0.0017 & 0.0087 & 0.0009 & 0.0054 &
$\epsilon=0.3$ Gaussian PSF, $r_{\rm gal}=r_{\rm PSF}$\\
0.0020 & 0.0064 & 0.0009 & 0.0102 & 0.0008 & 0.0060 &
$\epsilon=0.3$ Gaussian PSF, $r_{\rm gal}=1.5r_{\rm PSF}$\\
0.0277 & 0.0101 & 0.0222 & 0.0214 & 0.0023 & 0.0116 &
$\epsilon=0.3$ $k=3$ PSF, $r_{\rm gal}=0.5r_{\rm PSF}$\\
0.0244 & 0.0085 & 0.0112 & 0.0162 & 0.0007 & 0.0091 &
$\epsilon=0.3$ $k=3$ PSF, $r_{\rm gal}=r_{\rm PSF}$\\
0.0191 & 0.0091 & 0.0061 & 0.0157 & 0.0008 & 0.0094 &
$\epsilon=0.3$ $k=3$ PSF, $r_{\rm gal}=1.5r_{\rm PSF}$\\
\end{tabular}
\end{table*}
\subsubsection{The effect of centroiding errors}
The centroid of an image can be determined in different ways, each of
them susceptible to errors due to photon noise. The effect of
centroiding errors on the summed galaxy image will be a convolution
with the distribution of centroid errors. Thus, the PSF needs to be
convolved with this distribution before analysis of $\overline g$, so
that the effect of the centroiding error can be compensated.
\section{Galaxy-by-galaxy application}
The method as described so far involves analysing the average galaxy
$\overline g$. Very accurate shear measurements require $\overline g$
to be the average of a large number of galaxies ($\sim1000$ for a
1-$\sigma$ shear accuracy of 0.01), otherwise intrinsic ellipticity
scatter will dominate. However, constructing such a $\overline g$ is
only possible if the shear and the PSF are constant over a large part
of the image. Often this is not the case.
To cope with this limitation, we have therefore experimented with the
algorithm in `galaxy-by-galaxy' mode, where the algorithm is applied
to individual galaxies and the resulting shear estimates averaged.
Mathematically this approach is not perfect, because it involves
fitting a constant-ellipticity model to individual galaxies even
though this is not necessarily appropriate. Nevertheless it turns out
to work better than might have been expected, and better than existing
methods.
We tested this approach on various model galaxies, of differing axis
ratios. To simulate typical galaxies, we include a round, central
`bulge' component, and an outer `disk' of axis ratio between 0.1 and
1. (Simulations with different bulge axis ratios yielded the same
results.) These were placed at all orientations, smeared with an
elliptical PSF, and analysed with the algorithm described above. The
best-fit $(\gamma_1,\gamma_2)$ values thus derived for each galaxy are
then averaged to give an estimate of the shear.
As may be seen in Figure~\ref{fig:galbygal}, the algorithm performs
very well, essentially correcting all PSF anisotropy signal in the
measured shear. By comparison the slightly biased answer returned by
the KSB algorithm is apparant as before. Residual systematics of the
new method are at the level of a few tenths of a percent.
\begin{figure*}
\resizebox{\hsize}{!}{\includegraphics{ksb9abc.ps}}
\caption{The derived shear values from the algorithm applied to
individual elongated `disk+bulge' galaxy images, after smearing with
an elongated PSF. Each ring represents a galaxy of different shape,
seen at many orientations. Left: the raw $(e_1,e_2)$ polarizations measured
with the standard KSB method, without correction for the PSF. Centre:
the result of applying the KSB prescription for PSF anisotropy and
circularization correction. The small bias seen before remains. Right:
the result of the new algorithm on the same galaxies. In the latter
case, the correct
average shear estimate (zero) is recovered even though individual
galaxies are not correctly described as intrinsically
constant-ellipticity sources. The three rows refer to the three kinds
of PSF ellipticity profile considered in
figures \ref{fig:rnda}--\ref{fig:rndc}: constant with radius (top),
outward-increasing (middle), and outward-decreasing (bottom).}
\label{fig:galbygal}
\end{figure*}
\section{Summary}
In this paper we have studied possible systematic errors arising from
the correction for anisotropic point-spread functions in weak lensing
analyses based on the well-known Kaiser et al.\ (1995) method. While
such effects are small, generally below a few percent in the deduced
gravitational shear components $(\gamma_1,\gamma_2)$, they are at a
level that is important for studies such as galaxy-galaxy
lensing, lensing by large-scale structure or cluster lensing at large
radii. A range of simulations shows that modelling the PSF as
a convolution of a compact anisotropic function with a more extended,
circular function, which underlies the KSB
formulation, is not sufficiently general to describe many PSFs, and
leads to these systematic residuals.
We have presented a new algorithm with which to carry out the
PSF-correction in a single fitting step, and show with simulated
images that the low-level residuals left in the KSB analysis can thus
be avoided. The whole image is used in the fitting, so that not just
the lowest moments are used to characterize the image shapes. We have
also shown that the noise properties of this algorithm compare well
with those of KSB. While mathematically the algorithm requires an
intrinsically circular source, as may be constructed by stacking many
observed galaxy images, in practice nearly unbiased results can also
be obtained when the algorithm is used to correct individual galaxy
shapes for the PSF. This galaxy-by-galaxy application of the algorithm
allows observations with spatially varying PSF and/or shear fields to
be handled.
Weak lensing is a unique technique with which to study gravitational
potentials at large radii in galaxy clusters, galaxy halos and in the
field. The present method holds the promise of allowing a little more
information to be extracted from the large volumes of data that will
be gathered with the coming generation of wide-field imagers.
\begin{acknowledgements}
I would like to thank Peter Schneider, Marijn Franx, Henk Hoekstra and
the referee for critical readings of the manuscript and for suggesting
several improvements.
\end{acknowledgements}
|
\section{Neutrino Astronomy: Multidisciplinary Science}
Using optical sensors buried in the deep clear ice or deployed in deep ocean and lake waters, neutrino astronomers are attacking major problems in astronomy, astrophysics, cosmic ray physics and particle physics by commissioning a first generation of neutrino telescopes\cite{gaisser}. According to estimates covering a wide range of scientific objectives, a neutrino telescope with effective telescope area of 1 kilometer squared is required to address the most fundamental questions\cite{halzen}. Planning is already underway to instrument a cubic volume of ice, 1 kilometer on the side, as a neutrino detector. This infrastructure provides unique opportunities for yet more interdisciplinary science covering the geosciences and biology.
Among the many problems which high energy neutrino telescopes will address are the origin of cosmic rays, the engines which power active galaxies, the nature of gamma ray bursts (GRB), the search for the annihilation products of halo cold dark matter and, possibly, even the structure of the Earth's interior. In burst mode they scan the sky for galactic supernovae and, more speculatively, for the birth of supermassive black holes. Coincident experiments with Earth- and space-based gamma ray observatories, cosmic ray telescopes and gravitational wave detectors such as LIGO can be contemplated. With high-energy neutrino astrophysics we are poised to open a new window into space and back in time to the highest-energy processes in the Universe.
Neutrino telescopes can do particle physics. This is often illustrated by their capability to detect the annihilation into high energy neutrinos of neutralinos, the lightest supersymmetric particle which may constitute the cold dark matter. Also, with cosmological sources such as active galaxies and GRBs we will be observing $\nu_e$ and $\nu_\mu$ neutrinos over a baseline of $10^3$\,Megaparsecs. Above 1\,PeV these are absorbed by charged-current interactions in the Earth before reaching a detector at the opposite surface. In contrast, the Earth never becomes opaque to $\nu_\tau$ since the $\tau$ produced in a charged-current $\nu_\tau$ interaction decays back into $\nu_\tau$ before losing significant energy\cite{saltzberg}. This penetration of tau neutrinos through the Earth above $10^2$\,TeV provides an experimental signature for neutrino oscillations. The appearance of a $\nu_{\tau}$ component in a pure $\nu_{e,\mu}$ beam would be signalled by a flat angular dependence of a source intensity at the highest neutrino energies. Such a flat zenith angle dependence for the farthest sources is a signature for tau neutrino mixing with a sensitivity in $\Delta m^2$ as low as $10^{-17}$\,eV$^2$. With neutrino telescopes we will also search for ultrahigh-energy neutrino signatures from topological defects and magnetic monopoles; for properties of neutrinos such as mass, magnetic moment, and flavor-oscillations; and for clues to entirely new physical phenomena. The potential of neutrino ``telescopes" to do particle physics is evident.
\subsection{Cosmic Particle Accelerators: Gamma Ray Bursts Take Center Stage}
Recently, GRBs may have become the best motivated source for high energy neutrinos~\cite{waxman}. Although neutrino emission may be less copious and less energetic than anticipated in some models of active galaxies, the predicted fluxes can be calculated in a relatively model-independent way. There is increasing observational support for a model where an initial event involving neutron stars, black holes or the collapse of highly magnetized rotating stars, deposits a solar mass of energy into a radius of order 100~km. Such a state is opaque to light. The observed gamma ray display is the result of a
relativistic shock which expands the original fireball by a factor $10^6$ in 1~second. Gamma rays arise by synchrotron radiation by relativistic electrons accelerated in the shock, possibly followed by inverse-Compton scattering.
It has been suggested~\cite{waxmanprime} that the same cataclysmic events produce the highest energy cosmic rays. This association is reinforced by more than the phenomenal
energy and luminosity. Both GRBs and the highest energy cosmic rays are produced in cosmological sources, {\it i.e.}, distributed throughout the Universe. Also, the average rate $\dot E \simeq 4\times10^{44}\rm~Mpc^{-3}~yr^{-1}$ at which energy is injected into the Universe as gamma rays from GRBs is similar to the rate at which energy must be injected in the highest energy cosmic rays in order to produce the observed cosmic ray flux beyond the ``ankle'' in the spectrum at $10^{18}$~eV.
\break
The association of cosmic rays with GRBs obviously requires that kinetic energy in the shock is converted into the acceleration of protons as well as electrons. It is assumed that the efficiency with which kinetic energy is converted to accelerated protons is comparable to that for electrons. The production of high-energy neutrinos is inevitably a feature of the fireball model because the protons will photoproduce pions and, therefore, neutrinos in interactions with the gamma rays in the burst. We have a beam dump configuration where both the beam and target are constrained by observation: the beam by the observed cosmic ray spectrum and the photon target by astronomical measurements of the high energy photon flux.
From an observational point of view, the predicted flux can be summarized in terms of the main ingredients of the model:
\begin{equation}
N_\nu (km^{-2}year^{-1}) \simeq 25 \left[ f_\pi\over 20\% \right] \left[ \dot E\over 4\times10^{44}
{\rm\, Mpc^{-3} \, yr^{-1}} \right] \left[ E_\nu\over 700\rm\ TeV
\right]^{-1} \,,
\end{equation}
{\it i.e.},
we expect 25 events in a km$^2$ detector in one year. Here $f_{\pi}$, estimated to be 20\%, is the efficiency by which proton energy is converted into the production of pions and $\dot E$ is the total injection rate into GRBs averaged over volume and time. The energy of
the neutrinos is fixed by particle physics and is determined by the threshold for photoproduction of pions by the protons on the GRB photons in the shock. Note that GRBs produce a ``burst" spectrum of neutrinos. After folding the falling GRB energy spectrum with the increasing detection efficiency, a burst energy distribution results centered on an average energy of several hundred TeV.
Interestingly, this flux may be observable in operating first-generation detectors. The effective area for the detection of 100~TeV neutrinos can approach 0.1~km$^2$ as a result of the large size of the events. A $\nu_e$ of this energy initiates an electromagnetic shower which produces single photoelectron signals in ice over a radius of 250~m. The effective area for a $\nu_{\mu}$ is larger because the muon has a range of 10~km (water-equivalent) and produces single photoelectrons as far as 200~m from the track by catastrophic energy losses. Because these spectacular events arrive with the GRB time-stamp of order 1 second precision and with an unmistakable high-energy signature, background rejection is greatly simplified. Although, on average, we expect much less than one event per burst, a relatively near burst would produce multiple events in a single second. The considerable simplification of observing high energy neutrinos in the burst mode has inspired proposals for highly simplified dedicated detectors\cite{crawford}.
\break
The importance of making GRB observations cannot be overemphasized\cite{waxman}:
\begin{itemize}
\item The observations are a direct probe of the fireball model of GRBs.
\item They may unveil the source of the highest energy cosmic rays.
\item The zenith angle distribution of the GRB neutrinos may reveal the appearance of $\nu_{\tau}$ in what was a $\nu_e, \nu_{\mu}$ beam at its origin. The appearance experiment with a baseline of thousands of megaparsecs has a sensitivity to oscillations of $\Delta m^2$ as low as $10^{-17}$\,eV$^2$, as previously discussed.
\item The relative timing of photons and neutrinos over cosmological distances will allow unrivaled tests of special relativity.
\item The fact that photons and neutrinos should suffer the same time delay travelling through the gravitational field of our galaxy will lead to better tests of the weak equivalence principle.
\end{itemize}
In response to the evidence that atmospheric neutrinos oscillate\cite{superK}, workers in this field have been investigating the possibility of studying neutrino mass with the atmospheric neutrinos which, up to now, are used for calibration only. These studies may significantly reshape the architecture of some detectors. We will return to this topic later on.
\section{Large Natural Cherenkov Detectors}
The study of GRBs is one more example of a science mission that requires kilometer-scale neutrino detectors. This is not a real surprise. The probability to detect a PeV neutrino is roughly $10^{-3}$. This is easily computed from the requirement that, in order to be detected, the neutrino has to interact within a distance of the detector which is shorter than the range of the muon it produces\cite{gaisser}. At PeV energy the cosmic ray flux is of order 1 per m$^{2}$ per year and the probability to detect a neutrino of this energy is of order 10$^{-3}$. A neutrino flux equal to the cosmic ray flux will therefore yield only a few events per day in a kilometer squared detector. At EeV energy the situation is worse. With a cosmic ray rate of 1 per km$^2$ per year and a detection probability of 0.1, one can only detect several events per year in a kilometer squared detector provided the neutrino flux exceeds the proton flux by 2 orders of magnitude or more. For the neutrino flux generated by cosmic rays interacting with CMBR photons and for sources like active galaxies and topological defects\cite{schramm}, this is indeed the case. All above estimates are however conservative and the rates should be higher because absorption of protons in the source is expected, and the neutrinos escape the source with a flatter energy spectrum than the protons. In summary, at least where cosmic rays are part of the beam dump, their flux and the neutrino cross section and muon range define the size of a neutrino telescope. A telescope with kilometer squared effective area represents a neutrino detector of kilometer cubed volume.
The first generation of neutrino telescopes, launched by the bold decision of the DUMAND collaboration over 25 years ago to construct such an instrument, are designed to reach a large telescope area and detection volume for a neutrino threshold of order 10~GeV. This relatively low threshold permits calibration of the novel instrument on the known flux of atmospheric neutrinos. The architecture is optimized for reconstructing the Cherenkov light front radiated by an up-going, neutrino-induced muon. Only up-going muons made by neutrinos reaching us through the Earth can be successfully detected. The Earth is used as a filter to screen the fatal background of cosmic ray muons. This makes neutrino detection possible over the lower hemisphere of the detector. Up-going muons must be identified in a background of down-going, cosmic ray muons which are more than $10^5$ times more frequent for a depth of 1$\sim$2 kilometers. The method is sketched in Fig.\,1.
\begin{figure}[h]
\centering
\hspace{0in}\epsfxsize=2.25in\epsffile{wf+oms.eps}
\caption{The arrival times of the Cherenkov photons in 6 optical sensors determine the direction of the muon track.}
\end{figure}
The optical requirements of the detector medium are severe. A large absorption length is required because it determines the spacings of the optical sensors and, to a significant extent, the cost of the detector. A long scattering length is needed to preserve the geometry of the Cherenkov pattern. Nature has been kind and offered ice and water as adequate natural Cherencov media. Their optical properties are, in fact, complementary. Water and ice have similar attenuation length, with the role of scattering and absorption reversed; see Table~1. Optics seems, at present, to drive the evolution of ice and water detectors in predictable directions: towards very large telescope area in ice exploiting the large absorption length, and towards lower threshold and good muon track reconstruction in water exploiting the large scattering length.
\begin{table}[t]
\def\arraystretch{1.5}\tabcolsep=1em
\caption{Optical properties of South Pole ice at 1750\,m, Lake Baikal water at 1\,km, and the range of results from measurements in ocean water below 4\,km.}
\smallskip
\centering\leavevmode
\begin{tabular}{lccc}
\hline
& (1700 m)&&\\[-1.5ex]
$\lambda = 385$ nm\,$^*$& AMANDA& BAIKAL& OCEAN\\
\hline
attenuation& $\sim 30$ m\,$^{**}$& \llap{$\sim$}8 m& 25--30 m\,$^{***}$\\
absorption& 100 m~~ & 8 m& ---\\
scattering& 25 m& 150--300 m& ---\\[-2ex]
length&&&\\
\hline
\multicolumn{4}{l}{\llap{$^*$}\,peak PMT efficiency}\\[-1ex]
\multicolumn{4}{l}{\llap{$^{**}$}\,same for bluer wavelengths}\\[-1ex]
\multicolumn{4}{l}{\llap{$^{***}$}\,smaller for bluer wavelengths}
\end{tabular}
\end{table}
\section{Baikal, ANTARES, Nestor and NEMO: Northern Water.}
Whereas the science is compelling, the real challenge is to develop a reliable, expandable and affordable detector technology. With the termination of the pioneering DUMAND experiment, the efforts in water are, at present, spearheaded by the Baikal experiment\cite{domogatsky}. The Baikal Neutrino Telescope is deployed in Lake Baikal, Siberia, 3.6\,km from shore at a depth of 1.1\,km. An umbrella-like frame holds 8 strings, each instrumented with 24 pairs of 37-cm diameter {\it QUASAR} photomultiplier tubes (PMT). Two PMTs in a pair are switched in coincidence in order to suppress background from natural radioactivity and bioluminescence. Operating with 144 optical modules since April 1997, the {\it NT-200} detector has been completed in April 1998 with 192 optical modules (OM). The Baikal detector is well understood and the first atmospheric neutrinos have been identified.
The Baikal site is competitive with deep oceans although the smaller absorption length of Cherenkov light in lake water requires a somewhat denser spacing of the OMs. This does however result in a lower threshold which is a definite advantage, for instance for oscillation measurements and WIMP searches. They have shown that their shallow depth of 1 kilometer does not represent a serious drawback. By far the most significant advantage is the site with a seasonal ice cover which allows reliable and inexpensive deployment and repair of detector elements.
With data taken with 96 OMs only, they have shown that atmospheric muons can be reconstructed with sufficient accuracy to identify atmospheric neutrinos; see Fig.\,2. The neutrino events are isolated from the cosmic ray muon background by imposing a restriction on the chi-square of the Cherenkov fit, and by requiring consistency between the reconstructed trajectory and the spatial locations of the OMs reporting signals. In order to guarantee a minimum lever arm for track fitting, they were forced to reject events with a projection of the most distant channels on the track smaller than 35 meters. This does, of course, result in a higher threshold.
\begin{figure}[h]
\centering\leavevmode
\epsfxsize=2.9in\epsffile{nobel5.eps}
\caption{Angular distribution of muon tracks in the Lake Baikal experiment after the cuts described in the text.}
\end{figure}
In the following years, {\it NT-200} will be operated as a neutrino telescope with an effective area between $10^3 \sim 5\times 10^3$\,m$^2$, depending on energy. Presumably too small to detect neutrinos from extraterrestrial sources, {\it NT-200} will serve as the prototype for a larger telescope. For instance, with 2000 OMs, a threshold of $10 \sim 20$\,GeV and an effective area of $5\times10^4 \sim 10^5$\,m$^2$, an expanded Baikal telescope would fill the gap between present underground detectors and planned high threshold detectors of cubic kilometer size. Its key advantage would be low threshold.
The Baikal experiment represents a proof of concept for deep ocean projects. These have the advantage of larger depth and optically superior water. Their challenge is to find reliable and affordable solutions to a variety of technological challenges for deploying a deep underwater detector. Several groups are confronting the problem, both NESTOR and ANTARES are developing rather different detector concepts in the Mediterranean.
The NESTOR collaboration\cite{resvanis}, as part of a series of ongoing technology tests, is testing the umbrella structure which will hold the OMs. They have already deployed two aluminum ``floors", 34\,m in diameter, to a depth of 2600\,m. Mechanical robustness was demonstrated by towing the structure, submerged below 2000\,m, from shore to the site and back. These test should soon be repeated with fully instrumented floors. The actual detector will consist of a tower of 12 six-legged floors vertically separated by 30\,m. Each floor contains 14 OMs with four times the photocathode area of the commercial 8~inch photomultipliers used by AMANDA and ANTARES.
The detector concept is patterned along the Baikal design. The symmetric up/down orientation of the OMs will result in uniform angular acceptance and the relatively close spacings in a low threshold. NESTOR does have the advantage of a superb site, possibly the best in the Mediterranean. The detector can be deployed below 3.5\,km relatively close to shore. With the attenuation length peaking at 55\,m near 470\,nm the site is optically superior to that of all deep water sites investigated for neutrino astronomy.
The ANTARES collaboration\cite{feinstein} is investigating the suitability of a 2400\,m deep Mediterranean site off Toulon, France. The site is a trade-off between acceptable optical properties of the water and easy access to ocean technology. Their detector concept requires, for instance, ROV's for making underwater connections. First results on water quality are very encouraging with an attenuation length of 40\,m at 467\,nm and a scattering length exceeding 100\,m. Random noise exceeding 50\,khz per OM is eliminated by requiring coincidences between neighboring OMs, as is done in the Lake Baikal design. Unlike other water experiments, they will point all photomultipliers sideways or down in order to avoid the effects of biofouling. The problem is significant at the Toulon site, but only affects the upper pole region of the OM. Relatively weak intensity and long duration bioluminescence results in an acceptable deadtime of the detector. They have demonstrated their capability of deploying and retrieving a string.
With the study of atmospheric neutrino oscillations as a top priority, they plan to deploy in 2001-2003 10 strings instrumented over 400\,m with 100 OMs. After study of the underwater currents they decided that they can space the strings by 100\,m, and possibly by 60\,m. The large photocathode density of the array will allow the study of oscillations in the range $255 < L/E < 2550 \rm~km\,GeV^{-1}$ with neutrinos in the energy range $5 < E_{\nu} < 50$~GeV.
A new R\&D initiative based in Catania, Sicily has been mapping Mediterranean sites, studying mechanical structures and low power electronics. One must hope that with a successful pioneering neutrino detector of $10^{-3}\rm\, km^3$ in Lake Baikal, a forthcoming $10^{-2}\rm\, km^3$ detector near Toulon, the Mediterranean effort will converge on a $10^{-1}\rm\, km^3$ detector at the NESTOR site\cite{spiro}. For neutrino astronomy to become a viable science several of these, or other, projects will have to succeed besides AMANDA. Astronomy, whether in the optical or in any other wave-band, thrives on a diversity of complementary instruments, not on ``a single best instrument". When, for instance, the Soviet government tried out the latter method by creating a national large mirror project, it virtually annihilated the field.
\section{AMANDA: Southern Ice}
Construction of the first-generation AMANDA detector was completed in the austral summer 96--97. It consists of 300 optical modules deployed at a depth of 1500--2000~m; see Fig.\,3. Here the optical module consists of an 8~inch photomultiplier tube and nothing else. It is connected to the surface by a cable which transmits the HV as well as the anode current of a triggered photomultiplier. The instrumented volume and the effective telescope area of this instrument matches those of the ultimate DUMAND Octogon detector which, unfortunately, could not be completed.
\begin{figure}[t]
\centering\leavevmode
\hspace{0in}\epsfxsize=4in%
\epsffile{amanda_eiffel.eps}
\caption{The Antarctic Muon And Neutrino Detector Array (AMANDA).}
\end{figure}
As predicted from transparency measurements performed with strings near 1\,km depth\cite{science}, it was found that ice is bubble-free below 1400\,m. Its optical properties were a surprise, nevertheless the bubble-free ice turned out to be an adequate Cherenkov medium. We explain this next.
The AMANDA detector was antecedently proposed on the premise that inferior
properties of ice as a particle detector with respect to water could be compensated by additional optical modules. The technique was supposed to be a
factor $5 {\sim} 10$ more cost-effective and, therefore, competitive. The
design was based on then current information\cite{dublin}:
\begin{itemize}
\item
the absorption length at 370~nm, the wavelength where photomultipliers are
maximally efficient, had been measured to be 8~m;
\item
the scattering length was unknown;
\item
the AMANDA strategy was to use a large number of closely spaced OM's to overcome the short absorption length. Simulations indicated that muon tracks triggering 6 or more OM's were reconstructed with degree accuracy. Taking data with a simple majority trigger of 6 OM's or more at 100~Hz resulted in an average effective telescope area of $10^4$~m$^2$, somewhat smaller for atmospheric neutrinos and significantly larger for the high energy signals.
\end{itemize}
\noindent
The reality is that:
\begin{itemize}
\item
the absorption length is 100~m or more, depending on depth\cite{science};
\item
the scattering length is ${\sim} 25$~m (preliminary, this number represents an average value which may include the combined effects of deep ice and the refrozen ice disturbed by the hot water drilling);
\item
because of the large absorption length, OM spacings are similar, actually
larger, than those of proposed water detectors. Also, in a trigger 20 OM's report, not 6. Of these more than 6 photons are, on average, ``not scattered\rlap." A ``direct" photon is typically required to arrive within 25\,nsec of the time predicted by the Cherenkov fit. This allows for a small amount of scattering and includes the dispersion of the anode signals over the 2\,km cable. In the end, reconstruction is therefore as before, although additional information can be extracted from scattered photons by minimizing a
likelihood function which matches their measured and expected time delays.
\end{itemize}
The most striking demonstration of the quality of natural ice as a Cherenkov detector medium is the observation of atmospheric neutrino candidates with the partially deployed AMANDA detector which consisted of only eighty 8 inch photomultiplier tubes\cite{B4}. The up-going muons are separated from the down-going cosmic ray background once a sufficient number of direct photons and a minimum track length guarantee adequate reconstruction of the Cherenkov cone. For details, see Ref.~\citenum{B4}. The analysis methods were verified by reconstructing cosmic ray muon tracks registered in coincidence with a surface air shower array.
After completion of the AMANDA detector with 300 OMs, a similar analysis led to a first calibration of the instrument using the atmospheric neutrino beam. The separation of signal and background are shown in Fig.\,4 after requiring, sequentially, 5 direct photons, a minimum 100\,m track length, and 6 direct photons per event. The details are somewhat more complicated; see Ref.~\citenum{albrecht}. A neutrino event is shown in Fig.\,5. By requiring the long muon track the events are gold-plated, but the threshold high, roughly $E_{\nu} \geq 50$~GeV. This type of analysis will allow AMANDA to harvest of order 100 atmospheric neutrinos per year, adequate for calibration.
\begin{figure}[t]
\centering\leavevmode
\epsfxsize=2.7in\epsffile{zenith.eps}
\caption[]{Angular distribution of muon tracks in AMANDA at three levels of quality cuts. Roughly, $N_{\rm direct}\geq 5$, muon track longer than 100~m, and $N_{\rm direct} \geq 6$. Details in Ref.~\citenum{albrecht}.}
\end{figure}
While calibration continues, science started. Ongoing efforts cover a wide range of goals: indirect search for dark matter, active galaxies, gamma ray bursts and magnetic monopoles.
\looseness=-1
While water detectors exploit large scattering length to achieve sub-degree angular resolution, ice detectors can more readily achieve large telescope area because of the long absorption length of blue light. By instrumenting a cube of ice, 1~kilometer on the side, the planned IceCube detector will reach the effective telescope area of 1 kilometer squared which is, according to estimates covering a wide range of scientific objectives, required to address the most fundamental questions\cite{halzen}. A strawman detector with effective area in excess of 1\,km$^2$ consists of 4800\,OM's: 80 strings spaced by $\sim$\,100\,m, each instrumented with 60\,OM's spaced by 15\,m. IceCube will offer great advantages over AMANDA and AMANDA II\cite{albrecht} beyond its larger size: it will have a much higher efficiency to reconstruct tracks, map showers from electron- and tau-neutrinos (events where both the production and decay of a $\tau$ produced by a $\nu_{\tau}$ can be identified) and, most importantly, adequately measure neutrino energy.
\begin{figure}
\centering\leavevmode
\epsfxsize=4.7in\epsffile{1197960.ps}
\caption{Neutrino candidate in AMANDA. The shading (size) of the dots represents time (amplitude) of triggered photomultipliers. The reconstructed muon track moves upward over more than 300~m.}
\end{figure}
\newpage
\noindent{\bf Acknowledgments:} \ This work was supported in part by the University of Wisconsin Research Committee with funds granted by the Wisconsin Alumni Research Foundation, and in part by the U.S.\,Department of Energy under Grant No.\,DE-FG02-95ER40896.
\medskip
\noindent{\bf References}
|
\section{INTRODUCTION}
\label{sec:level1}
This paper summarises a study of the problem of escapes of energetically
unbound orbits in strongly non-integrable two-degree-of-freedom Hamiltonian
systems as an example of phase space transport in complex systems.
This work has led to two significant conclusions: (1) When evolved into the
future,
ensembles of orbits of fixed energy often exhibit a rapid approach towards a
constant escape probability $P_{0}$, the value of which is independent of
the details of the ensemble and exhibits interesting
scaling behaviour.
Moreover, the values of the critical exponents appear to be relatively
insensitive to the choice of Hamiltonian. (2) At later times, the escape
probability decreases in a fashion which, for at least one model system, is
well fit by a power law $P(t){\;}{\propto}{\;}t^{-\mu}$ with
${\mu}{\;}{\approx}{\;}0.35-0.40$. This nontrivial time-dependence is
attributed to the fact that the possibility of escape to infinity is
controlled by cantori, which can trap chaotic orbits near regular
regions for extremely long times.
The first three papers in this series\cite{1,2,3} (hereafter
Papers $1$ - $3$) described a numerical investigation
of the statistical properties of orbit ensembles evolving in nonintegrable
two-degree-of-freedom Hamiltonian systems where it is possible energetically
for trajectories to escape to infinity. Earlier investigations of individual
orbits in these systems had led to two significant conclusions\cite{4,5}:
(1)
Just because escape is possible energetically does not mean that escape
will inevitably occur; and, even if escape does occur, the time required for a
trajectory to cross a Lyapunov curve\cite{Church} and hence escape to
infinity may be very long compared with the natural crossing time. (2) In some
phase space regions,
the time and direction of escape vary smoothly as a function of initial
conditions, but in other cases one discovers instead an apparent near-fractal
dependence on the specific choice of initial data (cf.\cite{Henon}).
This problem of escapes is an example of phase space transport in complex
two-degree-of-freedom Hamiltonian systems, a subject which has been explored
in detail over the past two decades (cf. \cite{Wiggins,Meiss,Reichl,LL} and
references cited therein). The passage of orbits through Lyapunov curves and
their subsequent escape to infinity is the most conspicuous aspect of the
transport, but crucial features of the bulk flow, especially at late times,
appear to be controlled by diffusion through
cantori\cite{Percival,Aubry,Mather,Shirts}, which can trap orbits for very
long times (cf. \cite{Meiss}). The chaotic behaviour
of late escapers indicates that this problem is closely related to chaotic
scattering (cf. \cite{Ott1,Ott2}), where an incident trajectory scatters to
infinity
at a time and in a direction that can exhibit a fractal dependence on the
value of the impact parameter. However, the problem of escapes is also related
to a variety of other physical problems, including, e.g., the dissociation of
molecules (cf. \cite{Noid}) or the evaporation of stars from a cluster (cf.
\cite{Heggie}).
Indeed, the phase space interpretation of the escape problem suggested in
Section IV
is completely consistent with the detailed phase
space description deduced recently for escapes in the planar isoceles
three-body problem.\cite{Zare}
Paper I showed that, for at least one particular Hamiltonian system (the
$H_{1}$ of eq.~[1]), the microscopic chaos exhibited by individual orbits
leads macroscopically to bulk regularities: (1) For sufficiently large
deviations from integrability, localised ensembles of initial conditions
evolve so as to exhibit a rapid approach towards a near-constant escape
probability $P_{0}$, which is independent of the specific choice of initial
conditions and which, if at all, only changes on a significantly longer time
scale. (2) The value of $P_{0}$ scales in terms of an ``order parameter''
${\epsilon}-{\epsilon}_{1}$. (3) For ensembles that probe a phase space region
of specified size $r$, the time $T$ required to converge towards $P_{0}$ also
scales in ${\epsilon}-{\epsilon}_{1}$. (4) For fixed values of ${\epsilon}-
{\epsilon}_{1}$, $T$ also scales in the size $r$ of
the phase space region sampled by the initial ensemble.
Subsequent work, summarised and extended here, has addressed several
questions not considered in Paper 1:
\par\noindent
(1) Do other potentials exhibit similar behaviour and, if so, are the scaling
exponents the same? In other words, {\it could this behaviour be universal}?
\par\noindent
(2) What happens at much later times? Does $P$ remain constant or is there
a different asymptotic behaviour for larger values of $t$?
\par\noindent
(3) Is there any obvious correlation between the time of escape for individual
orbits in the ensemble and the exponential instability of those orbits, as
probed, e.g., by short time Lyapunov exponents
(cf.\cite{Grass,Badii,KAA,Voglis})?
\par\noindent
(4) Can one identify a simple, physically well motivated model to
explain the observed behaviour?
This work is based on a detailed investigation of orbits in three different
Hamiltonians, namely:
$$H_{1}={1\over 2}({\dot x}^{2}+{\dot y}^{2}+x^{2}+y^{2})-{\epsilon}x^{2}y^{2},
\eqno(1) $$
$$H_{2}={1\over 2}({\dot x}^{2}+{\dot y}^{2}+x^{2}+y^{2})-{\epsilon}xy^{2},
\eqno(2) $$
and
$$H_{3}={1\over 2}{\Biggl(}{\dot x}^{2}+{\dot y}^{2}+x^{2}+y^{2}-
{2\over 3}y^{3}{\Biggr)}+{\epsilon}x^{2}y, \eqno(3) $$
where ${\dot x}$ and ${\dot y}$ denote canonical momenta.
In each case, the Hamiltonian is of the form
$$H=H_{0}+{\epsilon}H' ,\eqno(4) $$
with $H_{0}$ integrable and ${\epsilon}H'$ a nonintegrable correction.
${\epsilon}$ is
taken to be non-negative but is {\it not} assumed to be small.
These Hamiltonians exhibit very different symmetries. $H_{1}$ is invariant
under $x\to -x$ and/or $y\to -y$ and has four identical escape channels.
$H_{2}$ is only symmetric with respect to $y\to -y$, and has two channels of
escape. For ${\epsilon}=1$, $H_{3}$ reduces to motion in the H\'enon-Heiles
potential, which manifests a $2{\pi}/3$ rotation symmetry, but for
${\epsilon}>1$ this discrete symmetry is broken.
Representative equipotential surfaces
are exhibited in Fig. 1.
This research has led to four principal conclusions:
\par\noindent
(1) For all three systems, there exists a critical ${\epsilon}_{1}$,
larger than ${\epsilon}_{0}$, the smallest ${\epsilon}$ for which escapes
can occur, which signals a qualitative change in short time behaviour:
Below ${\epsilon}_{1}$ the escape probability $P$ decays towards zero
exponentially, but for larger values of ${\epsilon}$ one observes
instead an initial approach towards a near-constant nonzero $P_{0}$,
the value of which is independent of the detailed choice of initial conditions.
\par\noindent
2) For all three systems, $P_{0}$ scales in ${\epsilon}-{\epsilon}_{1}$, i.e.,
$P_{0}{\;}{\propto}{\;}({\epsilon}-{\epsilon}_{1})^{\alpha}$ with
${\alpha}>0$. For a uniform sampling of a given phase space region of
fixed size $r$, the time $T$ required to converge towards $P_{0}$ also scales,
i.e., $T{\;}{\propto}{\;}({\epsilon}-{\epsilon}_{1})^{-\beta}$, with
${\beta}>0$. For fixed ${\epsilon}$, $T$ also scales in the linear size $r$ of
the phase space region initially sampled, i.e.,
$T{\;}{\propto}{\;}r^{-\delta}$ with ${\delta}>0$. Finally, the data are
at least consistent with the possibility
that the numerical values of ${\alpha}$,
${\beta}$, and ${\delta}$ are the same for all three potentials; and that
${\alpha}-{\beta}-{\delta}=0$.
\par\noindent
(3) At later times, $P$ deviates from $P_{0}$ by exhibiting a slow
decrease towards zero. For at least one Hamiltonian, namely $H_{3}$, this
later time evolution is well fit by a power law $P{\;}{\propto}{\;}t^{-\mu}$,
with a positive constant ${\mu}<1$.
\par\noindent
(4) At least for $H_{3}$, and possibly for $H_{1}$ and $H_{2}$,
orbits that escape early on tend to be more unstable than orbits which only
escape at much later times, in that they have a larger short time
Lyapunov exponent. Computed distributions of short time Lyapunov
exponents and surfaces of section both support the hypothesis that the chaotic
orbits divide, at least approximately, into relatively distinct subclasses,
presumably separated by cantori.
\par\noindent
The last two points suggest strongly that the late time evolution of $P(t)$ is
controlled by cantori, which can trap chaotic orbits near regular islands for
very long times. It is well known (cf. \cite{Meiss}) that cantori typically
constitute the principal impediment for efficient phase space transport in
two-degree-of-freedom systems and that diffusion through cantori is usually
not characterised by a constant escape probability.
Section II describes the observed short time behaviour, summarising results
presented in Papers 1 - 3 and discussing the evidence for universality.
Section III focuses on longer time evolution, using surfaces of section and
short time Lyapunov exponents to provide insights into flows associated
with initially localised orbit ensembles. Section IV suggests a tentative
physical interpretation of the numerical results in terms of flows in a
chaotic phase space partitioned by cantori. Phase space is
presumed to be dominated by ``unconfined'' chaotic orbits which, in the
absence of any cantori or Lyapunov curves, would evolve towards a statistical
equilibrium (cf. \cite{LL}). The constant $P_{0}$ observed at relatively
early times is attributed to the fact that orbits sampling this
near-equilibrium will escape at a near-constant rate. The decaying $P(t)$
later on reflects the fact that the phase space also includes an appreciable
measure of temporarily ``confined'' or ``sticky'' (cf. \cite{Shirts,Contop})
chaotic
orbits which, albeit not trapped within the system forever, can only escape
much later once they have breached one or more cantori to become unconfined.
\section{SHORT TIME BEHAVIOUR}
\label{sec:level1}
\subsection{Description of the experiments}
\label{sec:level2}
The experiments described here entailed a study of orbit ensembles with fixed
energy $h$ evolving in $H_{1}$, $H_{2}$, and $H_{3}$ with variable
${\epsilon}$. Attention focused exclusively on ${\epsilon}>{\epsilon}_{0}(h)$,
the smallest value of ${\epsilon}$ for which escape to infinity is possible
energetically. The values of $h$ and ${\epsilon}_{0}$ for all three cases are
given in Table 1.
Ensembles of initial conditions were generated by uniformly sampling a square
cell of linear dimension $r$ in the $(x,{\dot x})$ plane, setting $y=0$, and
then computing ${\dot y}={\dot y}(x,{\dot x},h)>0$. The cells were chosen to
sample ``interesting'' phase space regions where most of the orbits do {\it
not} escape at very early times. This implied that the time of escape for any
given orbit typically manifested a sensitive dependence on
initial conditions. Most of the computations involved a fiducial cell size
$r=0.05$. However, when exploring the effects of varying cell size, cells as
small as $r=1.0\times 10^{-4}$ were also used.
Each initial condition was integrated into the future and the location of the
orbit on a Poincar\'e section noted at successive consequents, i.e.,
successive crossings of the $x=0$ phase space hyperplane with ${\dot x}<0$.
If after consequent $t-1$ but before consequent $t$ the orbit crossed one of
the Lyapunov curves, the orbit was recorded as having escaped at $t$. The
experiments with $H_{1}$ \cite{1,4,5}
used a time series integrator. However, the experiments involving $H_{2}$ and
$H_{3}$ \cite{2,3} exploited a more efficient Lie integrator truncated at
twelfth order\cite{Dvorak}, which facilitated integrations of significantly
larger
ensembles typically containing $2000\times 2000$ initial conditions or more.
The fundamental object of interest is $P({\epsilon},t)$, the probability that
a randomly chosen orbit is an escaping orbit with escape occuring between
consequents
$t-1$ and $t$. This escape probability, along with an estimated
uncertainty ${\Delta}P({\epsilon},t)$, was defined by the obvious relation
$$P({\epsilon},t){\pm}{\Delta}P({\epsilon},t)=
{N_{esc}{\pm}\sqrt{N_{esc}}\over N_{tot}} ,\eqno(5)$$
where $N_{esc}$ denotes the number of trajectories that escape between $t-1$
and $t$ and $N_{tot}$ the total number present at $t-1$.
\subsection{Results from the experiments}
\label{sec:level2}
At early times, the escape probability can exhibit a complex, highly irregular
behaviour, the details of which depend sensitively on the size and location of
the initial cell. However, at somewhat later times, $P$ tends instead to
exhibit a more systematic behaviour that is seemingly independent of the
initial cell and depends only on the value of ${\epsilon}$. The qualitative
form of this behaviour depends crucially on whether ${\epsilon}$ is above or
below a critical ${\epsilon}_{1}>{\epsilon}_{0}$.
Below ${\epsilon}_{1}$, the escape probability $P({\epsilon},t)$ decays
towards zero in a fashion that is well fit by an exponential. This is,
e.g., illustrated in Figs. 2 a and b, which exhibit $P(t)$ and
${\rm ln}\,P(t)$ for one value of ${\epsilon}<{\epsilon}_{1}$ in $H_{2}$,
namely ${\epsilon}=1.04$. The solid curve superimposes a best fit exponential
$P{\;}{\propto}{\;}\exp (-t/{\tau})$, with ${\tau}=62.0$.
For ${\epsilon}>{\epsilon}_{1}$, $P({\epsilon},t)$ appears instead to
converge towards a nonzero $P_{0}({\epsilon})$, the value of which depends on
${\epsilon}$ but is independent of
the cell of initial
conditions. The evidence is particularly compelling for $H_{1}$ and $H_{3}$,
but somewhat weaker for $H_{2}$ where, especially for small
${\epsilon}-{\epsilon}_{1}$, the convergence is relatively slow and can merge
into the later time evolution described in Section III. Examples of this
behaviour are provided in Figs. 3 a-c, which exhibit $P(t)$ for representative
values ${\epsilon}> {\epsilon}_{1}$ in $H_{1}$, $H_{2}$, and $H_{3}$. (Other
examples are provided in Papers 1 - 3.) The transition at or near
${\epsilon}_{1}$ is quite abrupt and, for ${\epsilon}>{\epsilon}_{1}$,
$P_{0}({\epsilon})$ is a monotonically increasing function
of ${\epsilon}$. Moreover, for all three Hamiltonians one finds that, at least
for relatively small values of the order parameter ${\epsilon}-{\epsilon}_{1}$,
the escape probability $P_{0}({\epsilon})$ exhibits a simple scaling, namely
\cite{1,2,3}
$$P_{0}({\epsilon}){\;}{\propto}{\;}({\epsilon}-{\epsilon}_{1})^{\alpha},
\eqno(6) $$
with a constant ${\alpha}>0$. Illustrations of the goodness of fit of this
scaling relation for the Hamiltonians $H_{1}$ and $H_{3}$ are provided,
respectively, by Figs. 7 in Paper 1 and Figs. 4 in Paper 3, which
exhibit plots of $P$ vs.~${\epsilon}$ and
${\rm ln}\,P$ vs.~${\rm ln}\,({\epsilon}-{\epsilon}_{1})$. The values of
${\epsilon}_{1}$ are again given in Table 1.
That the escape probability approaches a roughly time-independent value can
be interpreted by supposing that the cell of initial conditions has dispersed
to fill certain large regions inside the Lyapunov curves in a nearly uniform
fashion, and that trajectories are escaping near-randomly at a constant rate.
Using operational prescriptions described in Paper 1, one can also estimate
the time $T$ required for an ensemble with ${\epsilon}>{\epsilon}_{1}$ to
approach $P_{0}$. For cells of fixed size $r$, this $T$ appears to be roughly
independent of initial conditions, depending only on ${\epsilon}$. Moreover,
one finds that, for all three Hamiltonians, $T$ also scales in ${\epsilon}$
\cite{1,2,3}, i.e.,
$$T({\epsilon}){\;}{\propto}{\;}({\epsilon}-{\epsilon}_{1})^{-{\beta}},
\eqno(7)$$
with ${\beta}>0$. Illustrations of the goodness of fit of this
scaling relation for the Hamiltonians $H_{1}$ and $H_{3}$ are provided,
respectively, by Figs.~8 in Paper 1 and Figs.~5 in Paper 3, which
exhibit plots of $T$ vs.~${\epsilon}$ and
${\rm ln}\,T$ vs.~${\rm ln}\,({\epsilon}-{\epsilon}_{1})$.
To the extent that a constant $P_{0}$ reflects a population that has dispersed
throughout the regions inside the Lyapunov curves, one might anticipate that
the convergence time $T$ would depend on
the size $r$ of the initial cell, smaller cells approaching
$P_{0}({\epsilon})$ more slowly. This too was confirmed numerically, Indeed,
for all three Hamiltonians one finds that, for a fixed value of ${\epsilon}$,
the convergence time $T(r)$ also scales in $r$ \cite{1,2,3}, i.e.,
$$T(r){\;}{\propto}{\;}r^{-\delta}, \eqno(8)$$
with ${\delta}>0$. This is illustrated in Figs.~9 and 10 of Paper 1
for two different values of ${\epsilon}$ for the Hamiltonian $H_{1}$.
It is obviously important to determine how abrupt the transition at
${\epsilon}{\;}{\approx}{\;}{\epsilon}_{1}$ actually is. However, this is
difficult numerically. Eq. (7) implies that $T({\epsilon})$ diverges for
${\epsilon}\to{\epsilon}_{1}$, but this critical slowing down implies an
intrinsic limitation in one's ability to probe the details near the transition
point.
\subsection{Possible evidence for universality}
\label{sec:level2}
That the scaling relations (6) - (8) hold for all three Hamiltonians
is clearly interesting. Even more striking, however, is the fact that, as is
evident from Table 1, the values of the exponents ${\alpha}$, ${\beta}$, and
${\delta}$ are very similar for all three systems. In each case,
$${\alpha}{\;}{\sim}{\;}0.5, \hskip .2in
{\beta}{\;}{\sim}{\;}0.4, \hskip .2in {\rm and} \hskip .2in
{\delta}{\;}{\sim}{\;}0.1. \eqno(9) $$
The uncertainties in ${\alpha}$ are dominated by uncertainties in the correct
value of ${\epsilon}_{1}$. As discussed in Paper 1, because of the
aforementioned critical slowing down ${\epsilon}_{1}$ is best estimated by
looking at somewhat higher values of ${\epsilon}$ and extrapolating to smaller
${\epsilon}$. The uncertainties in ${\beta}$ are dominated by the
precise prescription used to identify a convergence time. In particular, even
though it is usually easy to determine a lower bound on the convergence time,
the determination of an upper bound proves more difficult. It follows that the
quoted error bars in Table 1 can be asymmetric. As regards the best fit
${\delta}$, there are two principal sources of uncertainty, namely (1) that
the effect is relatively small (so that the fractional error is large) and (2)
that, especially for very small values of $r$, different ensembles can exhibit
significant variability.
Given these uncertainties, one cannot conclude unambiguously that
${\alpha}$, ${\beta}$, and ${\delta}$ are strictly equal
for all three Hamiltonians. However, one {\it can} conclude that
they are all comparable in size for all three systems and that, consistent
with the uncertainties, they may in fact be equal.
It is also true that, to within statistical uncertainties,
$${\alpha}-{\beta}-{\delta}{\;}{\approx}{\;}0. \eqno(10)$$
For $H_{1}$ and $H_{3}$ the evidence for this assertion is relatively strong.
For $H_{2}$ the case is somewhat weaker, largely because the best fit
${\beta}$ is somewhat larger than for the other two systems. However, it
should be noted that the error bars on the ${\beta}$ for $H_{2}$
are especially big. This reflects the fact that, for this system, the short
time behaviour described in this Section merges relatively quickly into the
later time evolution described in Section III, where $P(t)$ begins to decay to
values below $P_{0}$. For $H_{1}$ and $H_{3}$ this subsequent decay only
become significant at somewhat later times, at least for larger values of
${\epsilon}-{\epsilon}_{1}$.
The evidence for universality summarised here is certainly much weaker than
for the universality first identified by Feigenbaum\cite{Feig} for
one-dimensional maps or by Escande and Doveil\cite{Escande} and
MacKay\cite{MacKay} in
their renormalisation group analyses of tori, but it is, nevertheless,
intriguing. Moreover, certain points are seemingly unambiguous. (1) For all
three Hamiltonians, there is clear evidence for an abrupt change in behaviour
at or near some critical value ${\epsilon}_{1}$. (2) For
${\epsilon}>{\epsilon}_{1}$, orbit ensembles evolve towards an escape
probability
$P(t)$ which is (a) largely independent of the choice of initial ensemble
and (b) nearly time-independent, at least for relatively short times. (3)
The convergence time $T$ depends both on cell size $r$ and
${\epsilon}-{\epsilon}_{1}$, larger cells and larger
${\epsilon}-{\epsilon}_{1}$ leading to a more rapid convergence. (4) Because
$T({\epsilon},r)$ increases with decreasing ${\epsilon}-{\epsilon}_{1}$,
pinning down the precise value of ${\epsilon}_{1}$ is quite hard. (5)
The resulting uncertainties in ${\epsilon}_{1}$ do not impact the apparent
fact that $P$ and $T$ scale in ${\epsilon}-{\epsilon}_{1}$. However, they
{\it do} impact estimates of the precise values of ${\alpha}$, ${\beta}$, and
${\delta}$, thus making it difficult to determine whether these exponents are
the same for all three potentials. What is clear is that, for all three
potentials, the values of the exponents are comparable in magnitude.
\section
PHASE SPACE FLOW AT LATER TIMES}
\label{sec:level1}
\subsection{Late time evolution of $P(t)$}
\label{sec:level2}
Section II summarised experiments indicating that, on a relatively short time
scale, the escape probability $P(t)$ evolves towards a near-constant value
$P_{0}({\epsilon})$. However, there is no reason to expect that $P$ will
remain approximately constant if the orbit ensembles are evolved for much
longer times. If, e.g., the initial ensembles contain a few regular orbits
that cannot escape, these will eventually dominate the orbits that remain in
the system and $P(t)$ must decay towards zero exponentially.
Even if the initial ensembles contain no regular orbits that never escape, one
might anticipate more complicated behaviour at late times. If, e.g., some
small subset of the chaotic escape orbits are stuck by cantori near some
regular island for relatively long times, the escape probability should
decrease below the initial near-constant $P_{0}$ once the other chaotic orbits
have almost all escaped.
Such a decrease was first noted for orbit ensembles in $H_{2}$ and
subsequently studied more systematically for ensembles in $H_{3}$ \cite{3}.
The principal conclusion is that, at least for ${\epsilon}>{\epsilon}_{1}$,
when orbit ensembles are evolved for somewhat longer times the probability
$P(t)$ begins to exhibit a slow, monotonic decrease. This is illustrated
in Fig. 4 which, for one ensemble evolved in $H_{3}$, exhibits $P(t)$ for an
ensemble with ${\sim}{\;}1.1 \times 10^{9}$ orbits carefully chosen from a
tiny region of
size $r=1 \times 10^{-5}$ to be dominated by ``slow escapers,'' so that
$P(t)$ can be tracked for a comparatively long interval.
Visually, $P(t)$ decays too slowly, and has the wrong curvature, to be well
fit by an exponential. However,
as is illustrated in Fig. 4b, which plots ${\rm ln}\,P$ as a
function of ${\rm ln}\,t$, the data for $5<t<180$ or so can be fit quite well
by a power law,
$$P(t){\;}{\propto}{\;}t^{-{\mu}}, \eqno(11) $$
with ${\mu}=0.39{\pm}0.02.$
This algebraic decay seems very robust, with ${\mu}$ apparently independent
of the initial cell and, at least within a limited range, the value of
${\epsilon}$. The best fit value for several different values of ${\epsilon}$
(most of which were sampled for shorter times with far fewer orbits) is
${\mu}=0.35{\pm}0.07$.
\subsection{Tools of analysis}
\label{sec:level2}
To ascertain why $P(t)$ changes in time and to better understand the
qualitative character of the flow, orbit ensembles evolved in $H_{3}$ were
also analysed in two other ways.
The first involved computing surfaces of section for an evolving ensemble.
Each orbit in the ensemble was integrated into the future and, provided that
it had not yet escaped, its values of $y$ and ${\dot y}$ were recorded at
successive consequents and sorted to generate sequences of
surfaces of section exhibiting $(y,{\dot y})$ pairs.
Such surfaces of section allow one to determine the extent to which the
ensemble has evolved to cover a large portion of the allowed phase space.
Moreover, they can facilitate the detection of ``zones of avoidance''
associated with regular islands and/or with orbits that have immediately
escaped, as well as phase space regions with excess concentrations of orbits,
corresponding presumably to regions from which escape is especially difficult.
The second involved computing short time Lyapunov exponents (cf.
\cite{Grass,Badii,KAA,Voglis}), which probe the average exponential
instability of
chaotic orbits over finite time intervals. By analogy with ordinary Lyapunov
exponents, a finite time ${\chi}(t)$ can be defined by the obvious prescription
$${\chi}(t)=\lim_{{\delta}Z(0)\to 0}
{1\over t}\ln {|{\delta}Z(t)|\over |{\delta}Z(0)|} \eqno(12)$$
where
$|{\delta}Z|^{2}=({\delta}x)^{2}+({\delta}y)^{2}
+({\delta}{\dot x})^{2}+({\delta}{\dot y})^{2}$
denotes the magnitude of the initial phase space perturbation, defined with
respect to the natural Euclidean norm. These exponents were determined
computationally in the usual way \cite{Bennetin} by introducing a small
initial perturbation ${\delta}x(0)=
1\times 10^{-12}$ in the $x$-direction and evolving simultaneously
both the perturbed and unperturbed orbits, periodically renormalising the
amplitude of the perturbation to assure that $|{\delta}Z(t)|$ remains smaller
than $1\times 10^{-8}$.
At early times, the ${\chi}(t)$ computed in this way will depend strongly on
the initial perturbation. However, if the orbit segment is chaotic,
${\chi}(t)$ will quickly become dominated by the component of the perturbation
in the most unstable direction and become relatively insensitive to the
initial ${\delta}Z(0)$. Note also that, given ${\chi}(t)$ for two different
times, $t_{1}$ and $t_{2}$, one can identify the average exponential
instability for the interval
$t_{1}<t<t_{2}$ as
$$ {\chi}(t_{2}-t_{1})=
{t_{2}{\chi}(t_{2})-t_{1}{\chi}(t_{1})\over t_{2}-t_{1}}.\eqno(13)$$
Short time Lyapunov exponents were used to confirm that, for the values of $h$
and ${\epsilon}$ under consideration, most, if not all, of the computed orbits
in $H_{3}$ are chaotic. Distributions of short time Lyapunov exponents were
also used to show (1) that the chaotic orbits which have not escaped often
appear to divide into distinct populations and (2) that there are correlations
between the magnitude of ${\chi}$ and the time at which the orbit escapes.
\subsection{Surfaces of section}
\label{sec:level2}
Figures 5 a-f exhibit a sequence of sections, generated for one ensemble at
six different consequents, $t=2$, $6$, $10$, $15$, $20$, and $25$. This
ensemble was comprised of $160,000$ orbits evolved with ${\epsilon}=1.13$,
a value only marginally above the critical ${\epsilon}_{1}$. The
cell of initial conditions, with $0<x<0.05$, $0.04<{\dot x}<0.09$, $y=0$,
$h=1/6$, and hence $0.5681<{\dot y}<0.5780$ is located near the top
of the energetically accessible portion of the surface of section. The first
escapes occured at $t=4$, and the largest number of escapes was
at $t=7$. $P(t)$ first settled down towards a smoothly varying form around
$t=10-12$.
Inspection of these, and other intermediate, sections indicates
that the orbits remaining inside the Lyapunov curves tend systematically to
spread over a relatively large fraction of the energetically allowed phase
space. Indeed, the elongated striae associated with the specific
choice of initial conditions, so conspicuous at consequents $t=2$ and $6$,
have been significantly blurred by $t=10$ and have nearly disappeared by
$t=25$.
For $t<6-8$, the form of the sections is strongly time-dependent. However, by
$t=10$ the ensemble has evolved to yield sections characterised by three
seemingly distinct regions which persist to later times, namely: (1) two
large holes accompanied by smaller surrounding whorls, (2) several overdense
regions at positive values of $y$, and (3) a larger region characterised by a
substantially lower density. As time passes, the occupied regions all decrease
in density, but the overdense regions remain overdense.
The two holes and their surrounding whorls are associated with escapes through
Lyapunov curves: any orbit with values of $y$ and ${\dot y}$ in these
regions would already have escaped before intersecting the $x$-axis. For
larger values of ${\epsilon}$ it is apparent that the visible whorls are part
of an elaborate set of structures that penetrate throughout large portions of
the lower density regions, and that this lower density region, which appears
macroscopically to be populated in a near-uniform fashion, is really laced
with tiny zones of avoidance.
Assuming that essentially all the orbits in the ensemble are chaotic, the
overdense regions can be interpreted as reflecting trapping near regular
islands: Even though these islands may be so small as to be almost
unobservable, cantori can significantly impact relatively large portions of
the chaotic phase space\cite{Meiss,LL}. The idea then is that orbits in the
overdense regions
are trapped by cantori and can only escape once they have diffused through the
cantori and can travel unimpeded throughout the remainder of the chaotic sea.
The existence of this three-part structure -- holes, less dense regions, and
more dense regions -- is independent of the choice of initial
conditions. Moreover, the general locations of the holes and the higher
density regions are insensitive to the precise value of ${\epsilon}$. This
latter fact is manifested in Fig. 6, which shows the analogue of Fig. 5 c, now
generated for an ensemble with ${\epsilon}=1.06$. This reinforces
the interpretation that one is seeing the effects of basic phase space
structures, rather than transient streaming motions reflecting the choice of
ensemble.
\subsection{Short time Lyapunov exponents}
\label{sec:level2}
Perhaps the most obvious way to search for correlations between the degree of
exponential instability exhibited by a chaotic orbit and the time at which it
escapes is to compute the mean short time Lyapunov exponent,
${\langle}{\chi}(t_{E}){\rangle}$, for all the orbits in an ensemble that
escape between successive consequents $t_{E}$ and $t_{E}+1$. The results of one
such computation are presented in Fig. 7, which was generated from an
ensemble of $1\times 10^{\,6}$ orbits evolved in $H_{3}$ with
${\epsilon}=1.30$. ${\langle}{\chi}(t_{E}){\rangle}$ clearly exhibits an
initial relatively rapid decrease for $t_{E}<5-10$, followed by a more
extended period during which ${\langle}{\chi}(t_{E}){\rangle}$ decreases more
slowly. (The initial point in Fig. 7 at $t_{E}=5$ {\it is} statistically
significant.)
As described above, interpreting the computed
${\langle}{\chi}(t_{E}){\rangle}$ at very early times as an accurate probe of
the average maximum short time exponent is suspect. Given, however, that the
typical values of ${\chi}$ are greater than or of order unity, the computed
${\langle}{\chi}(t_{E}){\rangle}$ should be relatively reliable for $t_{E}>5$
or so, which means that the initial rapid decrease is most likely real. It is
not completely clear whether ${\langle}{\chi}(t_{E}){\rangle}$ will continue
to decrease at late times, or whether it asymptotes towards a nonzero value.
However, the existence of a continued decrease out to at
least $t=70$ or so is unquestionably significant statistically.
The observed decrease in ${\langle}{\chi}(t_{E}){\rangle}$ can be easily
interpreted by assuming that the chaotic phase space inside the
Lyapunov curves divides into two different regions, namely (1) a region where
orbit segments are less unstable exponentially and from which direct escape to
infinity is difficult, if not impossible, and (2) a region where orbit segments
are more unstable and from which escape to infinity can proceed on a relatively
short time scale. The idea is that orbits which remain inside the
Lyapunov curves for a long time will typically spend most of their time in the
less unstable region before entering the more unstable region and subsequently
escaping to infinity. It follows that, for orbits that only escape at late
times, the short time exponent ${\chi}(t_{E})$, which probes the average
instability for $0<t<t_{E}$, will typically be smaller than for early escapers.
Suppose, oversimplistically, that orbit segments in the less and more unstable
regions can be characterised respectively by unique short time exponents
${\chi}_{L}$ and ${\chi}_{H}$, and that any orbit that enters the high
${\chi}$ region will escape after exactly a time $t_{e}$. It then follows
from eq. (13)
that the total ${\chi}(t_{E})$ for an orbit escaping at time $t_{E}$ satisfies
$${\chi}(t_{E})={1\over t_{E}}{\Bigl[}t_{e}\,{\chi}_{H}
+ (t_{E}-t_{e})\,{\chi}_{L}{\Bigr]}. \eqno(14)$$
Consistent with Fig.~7, this implies that ${\chi}(t_{E})$ decreases
monotonically, but eventually asymptotes towards a nonzero ${\chi}_{L}$. For
$t_{E}>15$ or so the computed ${\langle}{\chi}(t_{E}){\rangle}$ exhibited
in Fig. 7 is in fact well fit by eq. (14) with ${\chi}_{L}=0.92$ and
$t_{e}({\chi}_{H}-{\chi}_{L})=7.80$.
To confirm that the observed decrease in ${\langle}{\chi}(t_{E}){\rangle}$
reflects transitions between relatively distinct orbit
populations, it is also instructive to determine how short time Lyapunov
exponents computed for the same set of orbits change as a function of time.
One way to do this is to consider all the orbits in an initial ensemble that
escape at a given time $T_{E}$ and, given expressions for ${\chi}(t)$ for
times $t<T_{E}$, computed using eq.~(12), extract short time exponents for
different intervals $t_{1}<t<t_{2}$.
Figure 8, generated from the same ensemble as Fig.~7, focuses on all
$832$ orbits that escaped at $t=50$, computing distributions of short time
exponents for the intervals $15<t<20$ (solid curve) and $45<t<50$ (dashed
curve).
Both distributions are
bimodal, seemingly comprised of two different populations with peaks at or
near the same values of ${\chi}$, but it is clear that the relative
height of the two peaks changes significantly in time. For the earlier
interval, the lower ${\chi}$ population dominates whereas for the later
interval the low and high ${\chi}$ populations are of more nearly equal
importance. For $10<t<15$ the mean ${\langle}{\chi}{\rangle}=0.70$; for
$45<t<50$ the mean ${\langle}{\chi}{\rangle}=1.11$. This is consistent with
the interpretation that most of the orbits that escaped at $t=50$ were members
of a low ${\chi}$ population at early times but shifted to the higher ${\chi}$
population shortly before escaping from the system.
\section{PHYSICAL INTERPRETATION}
\label{sec:level1}
\subsection{General considerations}
\label{sec:level2}
This section suggests a simple model for the behaviour observed in Sections II
and III which is based on the assumption that, over finite intervals, orbits
divide at least approximately into three distinct classes, namely (1) regular
orbits, (2) sticky, or temporarily confined,
orbits, and (3) unconfined chaotic orbits, even though the distinction between
confined and unconfined chaos disappears entirely in the $t\to\infty $ limit.
Conventional wisdom would suggest that the presence of stable periodic orbits,
which one expects for generic Hamiltonians, implies that there must exist a
finite measure of regular orbits, even at very high energies $h$ and/or large
values of ${\epsilon}$. However, the existence of such regular regions,
separated from the remaining chaotic orbits by invariant {\it KAM} tori,
suggests in turn that the surrounding chaotic sea should contain cantori which,
albeit not impenetrable barriers, can trap chaotic orbits near the regular
islands for relatively long intervals of time. Even if there is no absolute
distinction between different types of orbits in the chaotic sea, there may
exist short time {\it de facto} distinctions which can have significant
implications for the Hamiltonian flow. (Strictly speaking, in general there
will also exist a finite measure of chaotic orbits {\it inside} the
{\it KAM} tori. However, these can never breach the invariant tori and, as
such, will never be able to escape to enter the surrounding stochastic sea.
For this reason, they may be lumped together with the regular orbits in the
following discussion.)
If a localised ensemble of phase space points, each corresponding to a chaotic
orbit with energy $h$, is evolved in a time-independent potential which, as
for the Hamiltonians (1) - (3) for ${\epsilon}<{\epsilon}_{0}$, has a compact
constant energy hypersurface, one anticipates a coarse-grained evolution
towards an invariant distribution corresponding to a uniform population of the
accessible phase space\cite{LL}. Numerical experiments suggest (cf.
\cite{KPRE}) that, if the chaotic phase space is not significantly impacted by
cantori, so that a single chaotic orbit can easily access the entire region
without having to diffuse through any barriers (cf. \cite{MMP1,MMP2}),
this approach will proceed exponentially in time on a time scale of order the
natural crossing time, $t_{cr}$. If, however, cantori play an important role
in partitioning the chaotic phase space regions over relatively short time
scales, one can instead observe a more complex, seemingly two stage process
\cite{MMP1,MMP2,MABK}. Sets of orbits in different nearly disjoint regions
will
rapidly approach near-invariant distributions, corresponding to near-uniform
populations of the separate regions; but only later, on a significantly longer
time scale, will orbits from different regions ``mix'' to yield an approach
towards a true invariant distribution.
Suppose now that ${\epsilon}>{\epsilon}_{0}$, so that orbits are no longer
bound energetically. It then seems reasonable to interpret the observed
behaviour of orbits in terms of two distinct sorts of ``escape,''
namely (1) unconfined chaotic orbits which pass through Lyapunov curves to
escape to infinity and (2) confined chaotic orbits, originally stuck near the
regular regions, which pass through one or more cantori to become unconfined,
after which they too can escape to infinity. To the extent that the escape
channels -- both the simple gaps breached by Lyapunov curves and the
more complex cantor sets of holes in cantori -- are small, or that one is
considering phase space regions relatively far from the escape channels, it
should be reasonable to visualise what is happening in terms of ensembles that
have evolved towards a near-invariant distribution.
Suppose, in particular, that one selects an initial ensemble where most of the
orbits are unconfined chaotic, but that there are also a significant number of
confined chaotic orbits and, perhaps, a few regular orbits. It is then easy to
explain the qualitative evolution.
On a relatively short time scale, the unconfined chaotic orbits should evolve
towards a near-uniform population of the phase space regions far from the
Lyapunov curves and outside any cantori which significantly impede phase space
transport. (The fact that, in the late time sections of Fig. 5, different
parts of the lower density region seem to have the same relative density at
different times supports this idea.) However, once this near-invariant
distribution has been achieved, escapes to infinity should proceed ``at
random'' at a near-constant rate, so that the unconfined chaotic orbits will
be characterised by a constant escape probability. To the extent that the
total orbit population is dominated by the initially unconfined orbits, and
that appreciable numbers of confined orbits have not yet become unconfined,
the total escape probability $P$ should be approximately constant.
Eventually, however, most of the unconfined chaotic orbits will have escaped,
so that, assuming that unconfined orbits cross the Lyapunov curves
more quickly than confined orbits diffuse through cantori, the total escape
probability is impacted significantly, and ultimately dominated, by the
remaining orbits. If all these orbits were regular and unable to escape, one
would expect $P$ to decay exponentially in time (cf. \cite{1}). Given, however,
that most of the remaining orbits are chaotic orbits originally trapped by
cantori, as seems true for ${\epsilon}>{\epsilon}_{1}$, one expects a slower
decay in $P$ reflecting transitions from confined to unconfined chaos: The
idea here is that confined orbits will only diffuse through cantori to become
unconfined very slowly, but that, once unconfined, they will quickly escape to
infinity. The observed $P{\;}{\propto}{\;}t^{-\mu}$ is thus driven by the
diffusion of orbits through cantori rather than their escape through
Lyapunov curves. That diffusion through cantori need not characterised by a
constant escape probability is in fact well known (cf. \cite{Meiss}).
The same qualitative picture should remain valid even if the initial
orbit ensemble is carefully selected to contain no chaotic orbits
trapped near the regular regions. Even if most of the original unconfined
orbits quickly escape to infinity through one of the Lyapunov curves, a
small fraction of those orbits could become trapped by cantori. However,
once trapped most of these orbits will only leak out at significantly later
times, when the population of unconfined orbits inside the Lyapunov curves
has become significantly reduced.
\subsection{A simple model}
\label{sec:level2}
Consider an ensemble of initial conditions of energy $h$, located inside the
Lyapunov curves, which may be divided into three different classes --
unconfined chaotic, temporarily confined chaotic, and regular -- characterised
by numbers $N_{u}(0){\;}{\gg}{\;}{\hat N}_{c}(0){\;}{\gg}{\;}N_{r}(0)$. Now
implement a probabilistic description, treating the regular orbits as a
separate population that can never escape, but allowing for three sorts
of transitions, namely escapes of unconfined orbits through Lyapunov curves,
trapping of unconfined orbits by cantori, and leakage of confined orbits to
become unconfined orbits inside the Lyapunov curves.
Calculating the total escape rate $R$, the continuum limit of the discrete
probability $P$ computed in the numerical experiments summarised above,
is straightforward if one makes two basic assumptions, each implicit in
the preceding and seemingly consistent with the numerical experiments. (1) The
rate at which unconfined orbits escape to infinity assumes a constant value
${\lambda}$, independent of time. (2) ${\lambda}$ is much larger than the
rates at which unconfined orbits become temporarily confined and temporarily
confined orbits become unconfined.
These assumptions imply that, at early times, the total escape rate is
dominated by the escape of initially unconfined orbits, and that
the details of any early trapping of unconfined orbits are irrelevant. This
means that, when computing the total escape rate, the confined population may
be approximated by an expression of the form
$$N_{c}(t)=N_{c}(0)f(t), \eqno(15)$$
where $N_{c}(0)$ allows for a possible early time trapping of some small
fraction of the originally unconfined orbits and
$f(t)$ is a monotonically decreasing function of time, assumed to
satisfy an initial condition $f(0)=1$. The rate $R_{cu}$ at which temporarily
confined orbits become unconfined is thus
$$R_{cu}(t)=-{1\over f}{df\over dt} \eqno(16). $$
Similarly, $N_{u}(t)$ must satisfy a simple rate equation
$$\hskip -.4in {dN_{u}(t)\over dt}={dN_{u}(t)\over dt}{\Biggl|}_{out}+
{dN_{u}(t)\over dt}{\Biggl|}_{in}$$
$$=-{\lambda}N_{u}(t)-N_{c}(0){df(t)\over dt}. \eqno(17) $$
This latter equation is easily solved to yield
$$N_{u}(t)=N_{u}(0)e^{-{\lambda}t}-N_{c}(0)\,\int_{0}^{t}\,
d{\tau}e^{{\lambda}({\tau}-t)}\;{df({\tau})\over d{\tau}} \;.\eqno(18) $$
However, this implies that the total escape rate
$$R(t){\;}{\equiv}{\;}{-1\over (N_{r}+N_{c}+N_{u})}{\;}{d\over dt}
(N_{r}+N_{c}+N_{u}) $$
$$={{\lambda}{\Bigl[}1-{\nu}_{c}\,\int_{0}^{t}\,d{\tau}
e^{{\lambda}{\tau}}df({\tau})/d{\tau}{\Bigr]}\over
{\Bigl[}1-{\nu}_{c} \,\int_{0}^{t}\,d{\tau}
e^{{\lambda}{\tau}}df({\tau})/d{\tau}+{\nu}_{c}f(t)e^{{\lambda}t}+{\nu}_{r}
e^{{\lambda}t}{\Bigr]}}\;, \eqno(19)$$
\vskip .1in
\par\noindent
where
${\nu}_{c}=N_{c}(0)/N_{u}(0)$ and ${\nu}_{r}=N_{r}/N_{u}(0)$ reflect
original relative abundances.
That the escape rate through the Lyapunov curves in much larger than the
rate at which confined orbits become unconfined means that
${\lambda}{\;}{\gg}{\;}|(1/f)(df/dt)|.$
This implies, however, that the integrals in the preceding equation can be
evaluated perturbatively, expanding $df({\tau})/d{\tau}$ about its value at
time ${\tau}=t$. Recognising that the second term in the denominator is
small compared with the third, one thus concludes that
$$R(t){\;}{\approx}{\;}{{\lambda}-{\nu}_{c}e^{{\lambda}t}df(t)/dt\over
1+{\nu}_{c}e^{{\lambda}t}f(t)+{\nu}_{r}e^{{\lambda}t}}\; . \eqno(20)$$
Granted that ${\nu}_{r}{\;}{\ll}{\;}{\nu}_{c}{\;}{\ll}{\;}1$, $R(t)$ typically
admits three different, asymptotic regimes, namely
\par\noindent
(1) an early time regime where $R{\;}{\approx}{\;}{\lambda}$;
\par\noindent
(2) an intermediate regime where $R \to
-(1/f)df/dt$; and
\par\noindent
(3) a late time
regime where $R \to -({\nu}_{c}/{\nu}_{r})df/dt$.
\par\noindent
At early times, the total escape rate is fixed by the rate at which initially
unconfined orbits cross the Lyapunov curves. Later, once most of these
original unconfined orbits have escaped, $R(t)$ is set by the rate at
which confined chaotic orbits become unconfined. Finally, once most of the
initially confined orbits have escaped, there is a more rapid decrease in $R$
towards zero, reflecting the fact that the now dominant regular population can
never escape.
It is natural to interpret the probability $P{\;}{\propto}{\;}t^{-{\mu}}$
described in Section III as reflecting the intermediate regime. If ${\mu}$
were equal to unity, one would then infer a population
$$N_{c}(t){\;}{\propto}{\;}t^{-q}, \eqno(21)$$
for some constant $q>0$, i.e., an eventual power law decay in the number of
confined orbits. Given, however, that the best fit ${\mu}<1$, one infers
instead a population
$$N_{c}(t){\;}{\propto}{\;}\exp (-qt^{1-\mu}), \eqno(22)$$
which decays faster than an a power law but slower than the exponential
decrease associated with a constant escape rate.
That $P(t)$ is not constant is hardly surprising. Indeed, diffusion through
cantori, interpreted as orbits wending their way through a self-similar
collection of turnstiles,\cite{MMP1,MMP2} would suggest that $P(t)$ decay
in time (cf.
\cite{Chirikov,Karney,Hanson}). Rather, what is interesting is that
${\mu}{\;}{\ne}{\;}1$.
In the context of chaotic scattering, one seems to see a sharp distinction
between hyperbolic scattering (cf. \cite{Ott2,Smil}), where the number of
incident
particles remaining within the scattering region at time $t$ decays
exponentially, i.e., $N(t){\;}{\propto}{\;}\exp (-{\lambda}t)$, and
nonhyperbolic scattering (cf. \cite{Yau}), where the number remaining decays
as a power law, i.e., $N(t){\;}{\propto}{\;}t^{-q}$. The origin of the
intermediate behaviour observed here is, at the present, unclear.
This picture presupposes that
${\nu}_{r}{\;}{\ll}{\;}{\nu}_{c}{\;}{\ll}{\;}1$. Indeed, if these
inequalities fail, the qualitative evolution can change significantly.
Suppose, for instance, that temporarily confined chaotic orbits are
unimportant compared with regular
orbits, so that one need consider only two orbit classes -- regular and
unconfined --, and that, even early on, there are many more regular than
unconfined orbits. It then follows immediately that, already at early times,
$R(t)$ should decay towards zero exponentially:
$$R(t){\;}{\approx}{\;}{{\lambda}\over 1 + (N_{r}/N_{c}){\exp}(-{\lambda}t)}
{\;}\to{\;}{N_{c}\over N_{r}}{\,}{\exp}(-{\lambda}t) .\eqno(23)$$
This is consistent with the observed behaviour for
${\epsilon}_{0}<{\epsilon}<{\epsilon}_{1}$ where, as noted already, $P(t)$
decays towards zero without first asymptoting
towards a near constant nonzero value.
\subsection{Discussion}
\label{sec:level2}
To place the preceding in an appropriate context, three important
points should be noted.
\par\noindent
1. Extracting a near-constant $P_{0}({\epsilon})$ is
necessarily a somewhat imprecise operation. The algorithm described in Paper 1,
or any obvious alternative, depends crucially on the idea that $P(t)$
will evolve towards a form that is largely independent of the initial
conditions on a time scale sufficiently short that appreciable numbers of
sticky chaotic orbits do not escape and the total escape probability is
dominated by the near-constant rate at which unconfined orbits escape. If the
ensemble ``forgets'' its initial conditions sufficiently quickly, it is
possible operationally to identify a reasonable estimate of
$P_{0}({\epsilon})$. Strictly speaking, however,
the ensemble is really evolving towards a characteristic $P({\epsilon},t)$
which manifests a slow, but nontrivial, time-dependence. This leads to an
inherent inaccuracy in the determination of the values of the critical
${\epsilon}_{1}$ and, especially, the exponents ${\alpha}$, ${\beta}$, and
${\delta}$.
\par\noindent
2. The calculations described in Section III were restricted to relatively
short times, $t<200$ or so; and, for this reason, one cannot preclude the
possibility that, on a significantly longer time scale, $P(t)$ could assume
a form very different from what was suggested in Section IV. For example,
one cannot preclude the possibility that, for much later times,
$P(t){\;}{\propto}{\;}t^{-1}$, in agreement with Karney's \cite{Karney}
experiments. Ideally one might like to integrate for much longer times, but
this quickly becomes very expensive computationally: because the overwhelming
majority of the orbits in an initial ensemble escape relatively early on, one
would need to start with an absolutely enormous collection of initial
conditions in order to derive statistically significant conclusions about
behaviour at much later times. Nevertheless, even though one cannot exclude
the possibility of different later time behaviour, it is significant that the
observed scaling $P(t){\;}{\propto}{\;}t^{-\mu}$ for $t<180$ or so appears to
be robust
\par\noindent
It should also be noted that one cannot completely exclude the possibility that
the long time integrations described in Section III are contaminated by
accumulating errors which, e.g., make the system slightly dissipative.
Numerical tests described in Paper I allow one to be confident that the early
time computations ($t<20$ or so) are reliable, but there is less hard evidence
to justify accepting long time integrations at face value. All that can be said
definitively is that, even for the longest time integrations, the relative
energy error for an orbit was never larger than $2\times 10^{-8}$ and usually
orders of magnitude smaller.
\par\noindent
3. This simple three-component model, based on two, and only two, nearly
distinct classes of chaotic orbits, may well be oversimplistic.
Indeed, lumping together every chaotic orbit that is not unconfined into a
single population assumed to have reached a statistical near-equilibrium seems
less justified than assuming that, at least away from the Lyapunov curves, the
unconfined phase space is characterised by a near-invariant distribution.
Furthermore, the
entire analysis assumes an abrupt transition occuring at or near some
critical ${\epsilon}_{1}$. However, one can argue that these limitations are
not completely unreasonable. Detailed examinations of orbit ensembles
on a compact phase space hypersurface indicate that, oftentimes, many of the
qualitative features of a flow can be interpreted by allowing only for two
classes of chaotic orbits, named sticky and unconfined (cf.
\cite{MABK,MMP1,MMP2}). Moreover,
investigations of the effects of increasing deviations from integrability
suggest that, as the control parameter ${\epsilon}$ becomes larger, holes in
cantori can abruptly increase in size, so that what was initially a barrier
that could only be penetrated on a very long time scale ceases to play a
significant role in impeding phase space transport \cite{CVB}.
\acknowledgments
H. E. K. acknowledges useful discussions with Elaine Mahon and Ed Ott
regarding short time Lyapunov exponents and the problem of convergence
towards invariant and near-invariant distributions.
H. E. K. was supported in part by the NSF grant PHY92-03333.
The remaining authors were supported in part by the European Community Human
Capital and Mobility Program ERB4050 PL930312.
Some of the computations in this paper were effected using computer time
made available through the Research Computing Initiative at the Northeast
Regional Data Center (Florida) by arrangement with IBM.
|
\section{Introduction}
The luminosity distribution of ellipticals and bulges of S0 galaxies has been
modelled for long using the $r^{1/4}$ formula (de Vaucouleurs 1948), which is
characterized by just two scale parameters, the effective radius $R_{e}$ and the
effective surface brightness $\mu_{e}$ of the isophote encircling half of
the total light.
More recently \inlinecite{caon93} showed that a generalized de Vaucouleurs
formula, $r^{1/n}$, obtained by adding a free exponent $n$ \cite{sersic}, not
only gives a better fit to the light profiles (which is expected in view of
the larger number of parameters), but it provides also an interesting
correlation between shape ($n$) and total luminosity ($L\sim I_e R^2_e$),
which in turn suggests that the family of elliptical galaxies is not
homologous.
The main criticism moved to this result is that
the observed trend of $n$ with $R_e$ may be due to the presence of
hidden disks embedded in elliptical galaxies: the larger is this disk,
the smaller comes the value of $n$.
Caon et al. (1993) had obtained their result by fitting the main-axes light
profiles using just the $r^{1/n}$ law. In view of the above criticism, we
decided to pinpoint the question by adopting a 2D fitting procedure allowing
the presence of the disk; 2D is needed in order to strongly couple and
exploit the different projection properties of components such as disk and
bulge which are expected to have quite different intrinsic thickness.
A non-linear least square fit minimizing the $\chi^2$ has been applied to the
same volume-limited sample of Caon et al. (1993): it contains $\sim 80$
early-type galaxies belonging to either the Virgo or the Fornax cluster. This
sample is $\sim 80\%$ complete down to $B_T\sim14$ mag, equivalent to $M_B
\sim -17.3$ if we assume a distance modulus $\Delta\mu = 31.3$ for both
clusters.
In the following we outline the adopted methodology and summarize
the results of this work (D'Onofrio et al., in preparation).
\section{The 2D fit}
Our 2D fitting models are based on the superposition of up to two components,
each one characterized by concentric and coaxial elliptical isophotes with
constant flattening. One such component may be thought as the projection of a
spheroid with finite intrinsic thichness, thus mimicking a bulge. For this
component only we allow a moderate degree of boxiness/diskiness. The other
component is thought to be a disk.
{\bf Model 1} is the 2D extension of the one used in their work by Caon et al.
(1993).
It consists of a spheroidal component only whose projected light distribution
follows the generalized de Vaucouleurs law:
\begin{equation}
\mu_{b}(x,y)=\mu_{e}+k \left[\left(\frac{r_{b}}{r_{e}}\right)^{1/n}-1 \right]
\end{equation}
with $k=2.5(0.068n-0.142)$ and
$r_{b}=\left[{x^{c}+\frac{\displaystyle y^{c}}{\displaystyle
(b/a)_b^{c}}}\right]^{1/c}$; $c$ is the parameter accounting for some
boxiness/diskiness ($c>2$ for boxy isophotes and $c<2$ for disky isophotes).
\medskip
{\bf Model 2} is made by the superposition of a bulge-like component with a
$r^{1/4}$ light distribution (same as eq. 1, with $n=4$) to an exponential
disk represented as:
\begin{equation}
\mu_{d}(x,y)=\mu_{0}+1.086\left(\frac{r_{d}}{h}\right)
\end{equation}
with $r_{d}=\left[{x^{2}+\frac{\displaystyle y^{2}}
{\displaystyle (b/a)_d^{2}}}\right]^{1/2}$.
\medskip
{\bf Model 3} is a generalization of model 2; the light distribution of
the bulge follows a $r^{1/n}$ law, but the disk remains exponential.
\bigskip
In summary, the structural parameters involved in these models are: the
effective surface brightness $\mu_{e}$ and the effective radius $R_{e}$ of the
spheroid, the free exponent $n$ when considered, the central surface
brightness $\mu_{0}$ and the scale length of the disk $h$ if included, the
apparent axial ratios $b/a$ for the bulge and the disk components, and
the boxiness/diskiness parameter $c$.
\begin{figure}
\centerline{
\psfig{file=fig1.ps,width=24pc,angle=-90}}
\caption{\small {
Plot of the effective surface brightness $\mu_{e}$ versus the
logarithmic effective radius $\log(R_{e})$ for the bulge component.
The bulges of spiral galaxies studied by de Jong (1996)
are marked by asterisks.
Filled circles and triangles mark the best fit for E and S0 galaxies, open
pentagons and triangles the fair fits, open squares and open squares with a
cross the poor fits.
The solid line represents a constant absolute magnitude
$M_{B}=-19.3$.}}
\end{figure}
\begin{figure}
\centerline{
\psfig{file=fig2.ps,width=24pc,angle=-90}}
\caption{\small {Plot of the effective radius $\log(R_e)$ of the bulge
component versus the exponent $n$ of the generalized de Vaucouleurs law
(symbols as in Fig.1). Open circles represent the data
from Caon et al. 1993.}}
\end{figure}
A number of simulations on artificial galaxies have been performed in order to
test the tool and to unveil the systematic effects that influence the model
parameters; in particular the seeing, which produces strongly biased results
even after proper convolution of the model, and can be taken care only by
masking out the central area of the galaxies. The simulations also provided a
more correct estimate of the errors of the model parameters.
\section{Result of the fit}
The three adopted models do not provide the same accuracy in reproducing the
light distribution of early-type galaxies. By examining the reduced $\chi^2$,
the 2D residuals, and by comparing the geometry of the model (ellipticity and
$a_{4}$ profiles) with the real galaxies, we concluded that the $r^{1/n} +
exp$ model has to be preferred to either the simple $r^{1/n}$ and the $r^{1/4}
+ exp$ models. Model 1 is comparable to model 3 in reproducing dwarf
elliptical galaxies ($M_B>-17$). Model 2 often produces extended luminous
halos not observed in the real objects.
It is evident that the reduced $\chi^2$ by itself does not guarantee the
physical significance of the best-fit solution for models with different
numbers of parameters. We performed also an F-test of the best-fit $\chi^2$
distributions, in order to verify whether the variances are different even
from a statistical point of view. Actually, however, we may try to attach a
physical meaning to the most complex and best-fitting model, that of $r^{1/n}
+ exp$, only after analyzing and falsifying the correlations among the
different parameters.
\begin{figure}
\centerline{
\psfig{file=fig3.ps,width=24pc,angle=-90}}
\caption{\small
{Plot of the central corrected surface brightness $\mu_0^c$ versus the scale
length $h$ of the disks (symbols as in Fig.1). The solid line is the Freeman
value of constant $\mu_0^c(B)=20.75$. The data seem to follow the lines of
constant absolute magnitude (dashed lines).}}
\end{figure}
In general the best fitting model does not work in the inner seeing-dominated
region ($<1''$) and in the outer parts, where an extended luminous `halo' is
often found, but it is good in the intermediate regions for a large range of
$\mu$.
\section{Analysis of the bulge parameters}
We confirm the correlation between the effective radius $R_{e}$ and the
effective surface brightness $\mu_{e}$. The $\mu_{e} - \log(R_{e})$ relation
(Fig. 1) shows a family of large bulges with $R_{e}>3$ kpc and with low surface
brightness $\mu_{e}$, as found by \inlinecite{cap92}. The separation between
{\it ordinary} and {\it bright} galaxies is well visible, even if the gap
between the two families is less clear. The bulges of spirals all belong to
the ordinary family ($R_{e} <3$ kpc).
Fig. 2 shows the $n-\log(R_{e})$ relation.
The open circles represent the old distribution, found by \inlinecite{caon93}
for the same sample of galaxies by fitting the equivalent light profiles.
The dashed line marks the value $n=4$ of the de Vaucouleurs law.
By comparing the old relation with the new one, we observe
that there is not a variation within the errors of the parameters.
\begin{figure}
\centerline{
\psfig{file=fig4.ps,width=24pc,angle=-90}}
\caption{\small
{Logarithmic plot of the scale length $R_e$ of the bulge
versus the the scale length $h$ of the disk components (symbols as
in Fig. 1). The dashed lines mark the seeing dominated regions.}}
\end{figure}
The $n$ parameter spans a wide range of values ($1<n<10$): low values
are typical of exponential bulges (already known to exist in spiral galaxies),
while high values are found in bright and large galaxies.
This result confirms that the different values of $n$ are not due to the
presence of a disk, but that there is a clear trend in the values of $n$ with
the galaxy size (and therefore with the total luminosity) for the
spheroidal components. It follows that the non-homology is likely a
characteristic of hot stellar systems.
\section{Analysis of the disk parameters}
Our best fit solutions tell us that in quite a few ($\sim 25\%$) elliptical
galaxies there is a central exponential component. This is not surprising for
the disky ellipticals, in which an `embedded disk', generally smaller and
brighter than normal disks in S0's and spirals, gives a characteristic disky
shape to the isophotes (\opencite{cap90}, \opencite{scorza95}).
However, we found an inner exponential component (with bright $\mu_0$ and
small $h$) also in the bright, boxy ellipticals.
We verified {\it a posteriori\/} that such small and bright `disks' are a first
order approximation of the inner cores of elliptical galaxies, well described
by the 'Nuker law' (\inlinecite{lauer}, \inlinecite{byun} and
\inlinecite{faber}).
Fig.3 shows the central surface brightness $\mu_{0}^{c}$ (corrected for
inclination assuming a zero-thickness disk), plotted versus the scale length
$\log(h)$ of the detected `disks'. The distribution of values is large: it
goes from small and bright `disks' ($\mu_{0}^{c} < 19$, $h < 0.63$ kpc), many
of them hosted by luminous E galaxies, to less luminous and great disks,
preferably found in S0 and low luminosity ellipticals. The disk parameters of
normal spirals (from \opencite{dejong}) are also included in the diagram for
comparison.
It seems that the whole distribution follows approximately the
lines of constant disk luminosities (dashed lines).
The spiral disks of de Jong's sample are brighter than the early-type disks
of our sample and extend towards fainter $\mu_{0}^{c}$ and longer scale
length $h$. The distribution of $\mu_{0}^{c}$ does not peak around
the Freeman value (solid line)
(\opencite{freeman}, \opencite{kruit}).
In Fig. 4 we plotted the effective radius $\log(R_{e})$ of the bulge component
versus the scale length $\log(h)$. It is apparent there is a correlation
between these parameters, but with two separated trends, one for the
bulge dominated systems (upper part of the diagram), and another for the disk
dominated galaxies (bottom part of the diagram).
It seems also that galaxies with very small bulges and disks ($R_{e} \sim h <
0.6$ kpc) as well as systems with large $R_{e}$ and $h$ ($>2$kpc) do not exist.
We note that the lack of objects with large size cannot be a bias, since our
sample is complete in luminosity.
\section{Conclusions}
By applying a full 2D fit of the light distribution of a volume-limited
sample of early-type galaxies we obtained the following results:
\begin{enumerate}
\item The luminosity distribution of bright E galaxies is well
represented by the superposition of an inner exponential component and
a large outer bulge.
This inner structure is characterized by an hight central surface
brightness and a small scale length.
It is likely that the detected component is the core of the galaxy.
It seems important, however, that this component follows the same photometric
properties of normal disks, although shifted in zero point.
\item The distribution of $\mu_{0}^{c}$ for all detected `disks' is
a function of the scale length $h$. The slope of the relation
is that of constant disk luminosities. The central surface brightness
$\mu_{0}$ spans a wide range of values, in contrast with the
claimed constant value.
\item The effective radii of the bulges and the scale lengths of
the `disks' are correlated: this suggests a possible coupling during
the galaxy formation between the two components, and that the angular momentum
should be an important parameter for the Hubble sequence.
\item The effective radius $R_{e}$ correlates with $\mu_{e}$ and with the
exponent $n$ of the generalized de Vaucouleurs law.
The existence of two families of spheroids is
confirmed as well as the increasing of $n$ for large
luminous bulges. This suggests that elliptical galaxies are not homologous
systems.
\end{enumerate}
|
\section{Introduction}
About ten years ago a concerted effort began to apply modern
first-principles band structure methods to the problems of ferroelectricity,
with a vision that electromechanical properties could perhaps one day be
designed computationally \cite{1}. This effort has been at least
as successful as hoped, and a range of problems that we didn't even realize
existed at the time have been identified and solved. Understanding
electromechanical response has turned out to be a rich problem, both from a
fundamental condensed matter physics perspective, as well as from the
applied physics and even engineering perspectives. There is much expanded
interest in the field now that a new class of materials have been
discovered, single crystal piezoelectrics with ultra-high electromechanical
couplings such as PMN-PT (PbMg$_{\frac{1}{3}}$Nb$_{\frac{2}{3}}$O$_{3}$-PbTiO%
$_{3}$) which will revolutionize fields ranging from sonar and hydrophones
to medical ultrasonic imaging \cite{2}. The current state of the
field, as will be discussed below, is that we now can compute all of the parameters related to
electromechanical coupling, particularly the piezoelectric constants, for
pure, ordered, single crystal phases \cite{3,4}. This can be done with good
accuracy at zero temperature, and
indications are that quite reliable results can also be obtained as
functions of temperature \cite{5}. The problems now being studied
are how to treat disordered solid solutions, and to understand the
underlying physics behind the observed material behavior--why is one
material better than another. In the future, using the computational
techniques developed, and the insights obtained from theory and experiment,
we will be able to predict material properties of as-yet unsynthesized
complex materials. The current state of the field and future prospects are
outlined here. This is not intended as a comprehensive review, but should be
a useful introduction to the field of first-principles studies of
ferroelectrics and piezoelectrics.
The reason to use first-principles methods is that (1) one is not reliant on
parameterized theories or (2) on fitting possible inaccurate experiments,
(3) one has access to the underlying potential surfaces, (4) can clearly see
the origin of observed behavior, and (5) they can be applied to hypothetical
or not yet synthesized materials, or (6) for temperature, pressure or
compositions for which data are not available. By ``first-principles,'' we
mean that experimental data are not used to constrain parameters; rather one
starts from the fundamental interactions among electrons and nuclei. Most of
the first-principles methods that have been applied to ferroelectrics are
based on the density functional theory (DFT) \cite{6} and
some are based on Hartree-Fock theory. The DFT states that the ground state
properties of a system are given by the charge density, and Kohn and Sham
\cite{7} showed how to compute the charge density and energy
self-consistently, using an effective exchange-correlation potential (V$%
_{xc} $) that accounts for the quantum mechanical interactions between
electrons. The local density approximation (LDA) takes the
exchange-correlation potential from the uniform electron gas at the density
for each point in the material. The Generalized Gradient Approximation (GGA)
includes the effects of local gradients in the density \cite{8}.
Given a form for the exchange correlation energy E$_{xc}[\rho]$, one
can find the self-consistent charge density and compute the energy for any
arrangement of nuclei (atoms). From the energies, zero temperature phase
diagrams, phonon frequencies, and elastic constants can be computed. In the
frozen phonon method, for example, one displaces atoms and computes the
change in energy as a function of displacement, from which the potential
surface for an instability, or phonon frequencies, can be derived. In
principle one can also compute finite temperature properties, for example
using Monte Carlo or molecular dynamics methods. In practice,
self-consistent methods are extremely computationally demanding, especially
for large systems, so instead one can find effective potentials or
Hamiltonians fit to the first-principles results to obtain finite
temperature properties.
Properties of ferroelectrics are extremely sensitive to volume
(pressure), which can cause problems since the present
first-principles methods are not perfect, and small errors in volume
(typically several percent, or more in some cases) can result in large
errors in computed ferroelectric properties. The GGA is generally an
improvement over the LDA, but still many computations for
ferroelectrics so far have been done at the experimental lattice.
The GGA gives reasonable $c/a$ at the experimental volume. For
PbTiO$_3$ at the experimental volume (427.27 bohr$^3$), for example,
we found $c/a$ equal to 1.12 (90\% high) for LDA (Hedin-Lundqvist),
1.09 for LDA (Wigner), and 1.073 for GGA (PBE), compared to the
experimental value of 1.063. However, the volume is still too large
in GGA and $c/a$ is thus overestimated. A new form for the exchange
and correlation called the Weighted Density Approximation (WDA)
shows much promise, and gives excellent results for the lattice
constants and energetics
\cite{9}, so there is hope that in the future all properties of
ferroelectrics could be computed truly
\textit{ab initio} without recourse to the experimental volume.
\section{Milestones}
\subsection{Total energy and electronic structure}
At first it was unknown whether first-principles methods within the LDA
would even give ferroelectric ground states where they should. If fact, in
the initial computations for BaTiO$_{3}$ (read ``initial'' as over one
year!) the author's LAPW computations did \emph{not }give a ferroelectric
ground state due to a minor bug that apparently did not effect other
systems! After this problem was fixed, the ferroelectric instability was
found at the experiment volume, though it was extremely sensitive to
pressure. The initial questions then after finding the instability were (1)
what is the underlying cause of the ferroelectric instability and (2) why do
different similar materials behave differently; particularly why does BaTiO$%
_{3}$ show a series of phase transitions from cubic, to tetragonal, to
orthorhombic, and to rhombohedral with decreasing temperature whereas PbTiO$%
_{3}$ has a single phase transition to tetragonal? The potential surface was
initially mapped out using the LAPW method, and the charge density and
electronic structure were analyzed
\cite{10,11,12}.
PbTiO$_{3}$ showed a much deeper well when the tetragonal strain was
included, whereas in BaTiO$_{3}$ the rhombohedral phase had the lowest
energy. Thus the tetragonal strain is responsible for the tetragonal ground
state in PbTiO$_{3}$. Analysis of the charge densities and densities of
states showed that the ferroelectric instability is due to hybridization
between the O 2p states and the Ti 3d states, and if the 3d variational
freedom was removed from the problem, the ferroelectric instability
vanished \cite{10}. In BaTiO$_{3}$, the Ba is quite ionic and
spherical, whereas the Pb in PbTiO$_{3}$ is very not spherical in the
ferroelectric phase, and polarization of the Pb helps stabilize the large
strain and the tetragonal ground state in PbTiO$_{3}$.
Thus the following picture emerged. All Coulomb lattices are unstable with
respect to ferroelectric displacements; the short-range repulsive forces
tend to stabilize crystals with respect to off-center displacements. In
perovskite like BaTiO$_{3}$ and PbTiO$_{3}$, the O 2p states strongly
hybridize with the d$_{0}$ cation, in this case Ti$^{4+}$, reducing the
short-range repulsions thus allowing off-center displacements. Without
strain, or when strain effects are minor, the lowest energy off-center
displacements are along the (111) directions. Thus the ground state is
rhombohedral. As temperature is raised, the off-center displacements
disorder over two directions, giving an average orthorhombic symmetry, then
there is a disordering over four directions giving tetragonal symmetry, and
final a disordered cubic phase at high temperatures. The same features are
found in BaTiO$_{3}$ and KNbO$_{3}$ \cite{13,14}. One
should not overemphasize this order-disorder picture too much, however. In
the ideal order disorder picture, the atoms are in off-center sites
at all temperatures, and there should be a large configurational
entropy change at the ordering phase transitions, which is not
observed. Furthermore, in the ideal disorder picture, the local
potential for displacing a single B-site cation such as Ti or Nb is
unstable, but calculations show that the diagonal, self-force
constant, is positive, and so there would be a restoring force for
displacing a single cation from its ideal position in the perfect
cubic perovskite, and only when groups of ions are moved (for
example in a $\Gamma$-point displacement) is there a multiple well
potential surface \cite{15}.
It is clear that the polarizability of Pb plays a special role in
ferroelectrics. PbTiO$_{3}$ has a
very large strain (6\%) where tetragonal BaTiO$_{3}$ has only a 1\% strain.
The ground state of PbZrO$_{3}$ is antiferroelectric and it has a complex
phase diagram \cite{16}, whereas BaZrO$_{3}$ is cubic \cite{17}. PbZrO$_{3}$ has a
complex structure at low temperatures, and it
was structural total energy relaxations within the LDA that clarified the
crystal structure \cite{16}. Thus first-principles theory has the
power now to find problems with experimental determinations of complex
structures.
Another example of theory finding problems with experimental analysis is
LiNbO$_{3}$. In the LiNbO$_{3}$ structure there are chains of oxygen
octahedra with Li ions shared between two octahedra. Based on neutron
scatter experiments, it was suggested that the Li atoms hopped between two
octahedra at high temperatures in the high temperature phase, and ordered to
occupied one set of octahedra at low temperatures in the polar
phase \cite{18,18b}. Analysis of experimental data suggested that the
underlying potential surface had a triple well structure for Li
displacements, and all models considered the Li motions to the important
physics driving the ferroelectricity. Inbar and Cohen performed LAPW frozen
phonon computations and found a quite different picture \cite{19}.
They found no triple well, and they found that the double well for Li
displacements was quite shallow, and the double well for oxygen displacements
much deeper. The deepest wells were found for coupled displacements of Li
and O. They found that the ferroelectric instabilities in LiNbO$_{3}$ and LiTaO%
$_{3}$ are quite similar to the instabilities in the oxide perovskites, and
that a primary driving force is the d$_{0}$ configuration of the Nb$^{5+}$
and Ta$^{5+}$ ions, and hybridization with the surrounding oxygens that
allows the d$_{0}$ cation to go off-center. This sets up local fields that
drive or order the Li off-centering.
Though static total energy computations are very powerful, and further
examples will be given below, they are limited in the ability to find phonon
or instability eigenvectors or behavior for arbitrary wavevector. Advances
that give the linear response for arbitrary wavevector are discussed next.
\subsection{First-principles lattice dynamics--linear response}
Using conventional methods, first-principles methods with the LDA or GGA
scale as N$^{3}$, where N is the size (say number of atoms) of the system.
This makes it prohibitive to study arbitrary wavevectors. For small
wavevectors a very large supercell would be required. In linear response
computations that problem is circumvented, and one can compute the response
of the system to small perturbations with arbitrary wavevector with
approximately the same cost as the primitive cell computation. Yu and
Krakauer developed a linear response LAPW code and applied it to phonon
dispersion in KNbO$_{3}$ \cite{13,20}. They found the
ferroelectric instability to be dispersive in certain directions, indicating
correlations in displacements in real space. In other words, if the unstable
band were flat, it wouldn't matter how the atoms were displaced in one cell
relative to the next. Analysis of the instability dispersion showed that
there should be chain-like correlations, consistent with streaking observed
in x-ray diffraction \cite{21}. Linear response methods also allow
computation
of Born effective charges (discussed below) and the dielectric constant.
A number of different ferroelectric perovskites have now been studied using
linear response methods. The unstable modes of KNbO$_{3}$ and BaTiO$_{3}$
show very similar dispersion, consistent with the same sequence of phase
transitions for the two. The soft branch has primarily B-ion character. PbTiO%
$_{3}$ shows a much more flat soft branch, indicating that correlations
between cells are less important than in KNbO$_{3}$ or BaTiO$_{3}$. The soft
modes also have appreciable A-ion character (Pb) in PbTiO$_{3}$ compared
with KNbO$_{3}$ or BaTiO$_{3}$. PbZrO$_{3}$ is similar to PbTiO$_{3}$ in
that the soft branch remains unstable for any wavevector, but the zone
boundary R- and M-point instabilities are deeper than the $\Gamma $-point
instability, consistent with the fact that the ground state in PbZrO$_{3}$
is antiferroelectric, and PbTiO$_{3}$ is ferroelectric.
\subsection{Effective Hamiltonians--finite temperature properties}
Although self-consistent methods are very powerful, they generally cannot
provide anything but static, essentially zero temperature, properties, at
least for the complex systems we are discussing. Ferroelectric phase
transitions, however, occur with increasing temperature from one
ferroelectric state to another, or to the high temperature paraelectric
state. Dielectric and piezoelectric properties are strongly dependent on
temperature, especially around the phase transitions. Thus we want
techniques that can give finite temperature behavior. One approach is to use
a Gordon-Kim based model and Molecular Dynamics or Monte Carlo
\cite{22,23}, but these models have not been successful so far
to
compute finite temperature properties for ferroelectrics. An effective
Hamiltonian approach has, however, been very successful \cite{24,25}. One develops
a symmetrized expansion of the free
energy in terms of local mode coordinates, effective charges, and strains.
Only the important local modes are used, and thus the number is degrees of
freedom is reduced from the full atomic Hamiltonian. Local modes are
important if they are related to the soft mode, and the choice of local
modes is key to the success of the method. A rigorous formulation of local
modes as Wannier-like functions has been developed \cite{26}. Though
the effective Hamiltonian method contains all of the physics to give
qualitatively correct results, one might question whether it would be
quantitatively accurate since it neglects the higher frequencies modes, and
does not even try to estimate their average self-consistent effects on the
ferroelectric behavior. One such example is the importance of thermal
expansivity, which is not included in the effective Hamiltonian models so
far. Since ferroelectricity is very sensitive to volume, one would think
that neglect of thermal expansivity might lead to serious errors. This might
be partly responsible for the shifts in predicted transition temperatures
relative to experiment. So far, however, it appears, that the effective
Hamiltonian models are very accurate, and they are the only game in town for
finite temperature properties of ferroelectrics at the present time. Since
the underlying first-principles calculations are also not perfect, it is
surprising and encouraging that such high accuracy can be obtained.
\subsection{Physics of Polarization}
\subsubsection{Polarization in Periodic Boundary Conditions}
Contrary to intuition and many textbook discussions, polarization
cannot be determined directly from the self-consistent charge
density or changes in the charge density alone within periodic boundary
conditions. Rather one must consider changes in polarization as current
flow, and this can be determined from the phases of the wave functions.
King-Smith and Vanderbilt showed how this can be computed using a Berry's
phase approach \cite{27}. A more intuitive approach is the
realization that polarization changes can also be considered as the charge
transport on displacing Wannier functions \cite{28}. Practical
calculations generally use the Berry's phase approach, and the development
of this technique was a key advance in the ability to compute
electromechanical properties. A review of the theory of polarization
is given by Resta \cite{29}.
Born effective charges are defined as the change in polarization with
displacement of a nucleus, $Z_{i\alpha \beta }=dP_{\beta }/du_{i\alpha }$,
for the change in \ $\overrightarrow{P}$ moving ion $i$; note that the
effective charge is a tensor. The effective charges are not necessarily
equal to the nominal ionic charges, and can be quite different. In oxide
ferroelectrics they tend to be greatly enhanced for certain directions, with
values up to several times the nominal values. The
enhancement of the effective charges is due to strong covalent
hybridization \cite{30,31,32}.
Effective charges have been predicted for a number of ferroelectric systems
now. They are accessible to experimental determination, for example by
optical measurements of LO-TO splitting. Very large values were
experimentally obtained for BaTiO$_3$ many years ago \cite{33}, but there
was no theoretical basis for understanding the results for many years.
It would be of great interest for experimentalists to measure these values
for a variety of ferroelectrics and compare with the theoretical predictions.
Piezoelectric constants can also be determined theoretically from changes in
polarization with respect to strain, as discussed below, using the Berry's
phase technique.
\subsubsection{Density polarization theory}
Another area of much discussion is in even more fundamental problems of
polarization under Kohn-Sham theory. It was argued that DFT
did not properly account for electronic polarization, and implied that
previous calculations were incorrect, and that the gap problem and low
dielectric constants in LDA were due to this failure \cite{34}. Mazin and
Cohen
\cite{35} questioned this, and showed that there is no problem
with established methods such as the LDA. The situation has been clarified
\cite{36,37} and it appears that to obtain the
proper polarization in exact Kohn-Sham theory (which no one knows how to do
for real bulk systems of interest, anyway) one would have to consider the
polarization as well as the charge density. Realistic estimates of this
effect suggest it may be very small (3\%, \cite{38}). Regardless of the
theoretical
importance of density polarization in the exact Kohn-Sham theory, it is
questionable if it is solely responsible for the gap problem and LDA
dielectric constants. The gap problem varies considerably from one material
to another, as do dielectric constants. For example, although the LDA\ high
frequency dielectric constant is 20\% too high in BaTiO$_{3}$, it is
very close to experiment in PbTiO$_{3}$ \cite{39}. A study of a set of 11
materials gives an r.m.s. error of 8\% from experiment for
$\epsilon_\infty$ using LDA \cite{40}. It is not at all clear than
there is a uniform cause for these deviations.
\subsection{Piezoelectricity}
The electromechanical response important in
transducer applications is given by the piezoelectric constants, or by the
figure of merit of the electromechanical coupling factors. Piezoelectric
constants can be written in different ways and one can transform from one
set to the other using thermodynamic transformations \cite{41}. The
set most easily computed are the piezoelectric strain coefficients $%
e_{ijk}=(\partial P_{i}/\partial \varepsilon _{jk})_{E=0}$, the change in
polarization with strain at zero applied field, usually reduced to the Voigt
form $\varepsilon _{j}$, $j=1...6$. Since the polarization can now be
computed using the Berry's phase approach, one can compute the polarization
versus strain and obtain the derivatives for the piezoelectric constants
numerically. It is important then to allow the atomic coordinates to
displace as functions of strain, because this is in fact the major
contribution to the piezoelectric response. The piezoelectric constant can
be written as
\begin{eqnarray}
e_{ijk} &=&\left( \frac{\partial P_{i}}{\partial \varepsilon _{jk}}\right)
=\left( \frac{\partial P_{i}}{\partial \varepsilon _{jk}}\right)
_{u}+\sum_{l\alpha }\left( \frac{\partial P_{i}}{\partial u_{l\alpha }}%
\right) _{\varepsilon }\left( \frac{\partial u_{l\alpha }}{\partial
\varepsilon _{jk}}\right) \\
&=&e_{ijk}^{h}+\sum_{l\alpha }\frac{e^{0}a_{i\alpha }}{\Omega }Z_{l,i\alpha
}^{\ast }\left( \frac{\partial u_{l\alpha }}{\partial \varepsilon _{jk}}%
\right) , \notag
\end{eqnarray}
where the first term, $e_{ijk}^{h}$, is the change of polarization with
homogeneous strain, that is no atomic relaxations, and the second is from
the displacements of the ions in response to strain, carrying effective
charges $Z$.
This was first carried through for PbTiO$_{3}$ \cite{3}.
Ferroelectric PbTiO$_{3}$ is tetragonal, and there are three independent
piezoelectric constants which can be found by three strains, tetragonal,
orthorhombic, and a monoclinic strain. Two methods can be used, the direct
method, that is finding the polarization versus strain, relaxing the
internal coordinates at each strain, and using the effective charges as
shown above. Both methods give the same final results, but the two step
procedure also illustrates the relative importance of the two terms.
First one must find the optimum minimum energy structure at the volume of
interest, in our case, the experimental zero pressure volume. Then, for
example for $e_{33}$(using Voigt notation), one varies the $c$-axis length,
and computes the change in polarization, giving $e_{33}^{h}$. One finds the
effective changes by displacing each ion separately, and computing the
change in polarization in each case, and then finds how the atoms relax as a
function of strain. Finally one can compute the polarization at the final
relaxed strained lattice. This gives two computations of $e_{33}$, which we
checked are identical, and give 3.23 C/m$^{2}$. The homogeneous contribution
is -0.88, and the internal relaxations give 4.11, so the total is dominated
by the atomic relaxations. Similarly, we found $e_{15}$=3.15 C/m$^{2}$ and $%
e_{31}$=-0.93 C/m$^{2}$. These are the proper moduli that do not include
homogeneous deformations of the original spontaneous polarization, which are
not observable. As pointed out later in this volume, the straightforward way
to extract the proper piezoelectric response is to compute differences in
Berry's phases, and convert to piezoelectric constants only in the last
step \cite{42}. The computed moduli were very close to some
experimental results, but there is a wide spread between different
experiments even for PbTiO$_{3}$, indicating the importance of obtaining
better values of experimental piezoelectric moduli under well-defined
electrical boundary conditions in order to test theory.
We also computed $e_{33}$ for two ordered structures of PZT50/50, near the
morphotropic phase boundary and obtained values significantly lower than
experimental measurements for PZT ceramics (single crystals were never made
successfully for PZT) extrapolated to low temperatures \cite{4}.
This lead us to conclude either that (1) extrinsic effects are very
important even at low temperatures in PZT and/or (2) that the single crystal
$e_{33}$ does not dominate the piezoelectric response in the poled direction
in PZT.
Using an effective Hamiltonian and Monte Carlo, Garcia and Vanderbilt
studied piezoelectric response of BaTiO$_{3}$ as a function of applied field
and temperature \cite{43}. They obtained good agreement with the
temperature dependence of the piezoelectric constants, and found a field
induced rhombohedral to tetragonal phase transition at very large fields.
Rabe and Cockayne used an effective Hamiltonian for \ PbTiO$_{3}$ and found d%
$_{33}$ as a function of temperature using Monte Carlo, also finding good
agreement with the experimental temperature dependence, which peaks strongly
at the ferroelectric phase transition \cite{5}.
\subsection{Solid solutions}
Most useful piezoelectrics are solid solutions rather than pure
ordered compounds.
This allows their properties to be tuned to meet engineering specifications.
But more than that, many complex solid solution ferroelectrics, for example,
the new single crystal piezoelectrics with giant electromechanical coupling,
like PMN-PT, have much enhanced properties relative to the pure compounds.
Even the endmember PMN is complex, with ions of different valence, Mg$^{2+}$
and Nb$^{5+}$ on the same crystallographic site. There has been some work to
study the ordering energetics in these systems. In PST-PT, for example, the
PIB (Potential Induced Breathing) model was used to study the relative
energetics of ordering, and to find an order-disorder phase diagram basic on
the Cluster Variation Method (CVM) \cite{44}. More recently,
self-consistent methods have been used along with Monte Carlo
\cite{45}. Even simple Madelung sums have been shown to illustrate some
of the
important behavior in these systems \cite{46}. Only limited work so
far
has studied how ordering (short-range and long-range) interacts with
ferroelectricity and piezoelectricity. Computations show that different
ordering schemes can drastically affect piezoelectric response in one
case \cite{47}, but how general this is not yet known.
\subsection{Defects, domain boundaries, and surfaces}
Real crystals are not perfect and infinite, and defects, domain boundaries,
and surfaces play important roles in ferroelectrics. Park and Chadi studied oxygen
vacancies in PbTiO$_3$, which may be important in fatigue. Domains have been
studied using an effective Hamiltonian, and it was found that 180${{}^\circ}$
domain boundaries are very sharp, and even one unit cell away the
structure looks like bulk \cite{48} Large scale LAPW
computations were performed on periodic BaTiO$_{3}$ slabs. The
computations showed that just one unit cell away from the surface, the
charge density is indistinguishable from bulk BaTiO$_{3}$. There are two
types of surfaces with either TiO$_{2}$ or BaO terminations. On the TiO$_{2}$
surface the dangling Ti-bond relaxes back onto the surface, but atomic
relaxations are quite small. A surface state was found in the band
structure. LAPW computations for symmetrically terminated surfaces were
straightforward to interpret, and using results from slabs with two BaO and
two TiO$_{2}$ surfaces, an average unrelaxed surface energy of 920 erg/cm$^2$ was
determined by comparing with the bulk BaTiO$_{3}$ energy \cite{49}. Using
plane waves and pseudopotentials, Padilla and Vanderbilt found an average
relaxed surface energy of 1260 erg/cm$^2$ \cite{50}. Hartree-Fock
calculations using Gaussian basis sets gave 1690 erg/cm$^2$
(unrelaxed) \cite{51}. The differences in basis sets as well as Hartree-Fock
versus LDA may be
responsible for the differences. The effect of periodic slabs versus isolated
slabs, as used in the Hartree-Fock study, amount to only about 3\% in the average
surface energy for symmetric slabs. Clearly these are exploratory studies, and we
cannot confidently state the exact surface energy for BaTiO$_3$ except all agree it
is high, which explains the difficulty of of cleavage in BaTiO$_3$ and the rough
surfaces that form on fracture.
In the LAPW study, periodic asymmetric and polar slabs were also
studied \cite{49}. As discussed there, asymmetric or polar periodic slabs
produce
artificial
potential gradients across the slab which is like an external applied
field. This is an undesirable effect that makes it difficult to interpret
the results. Nevertheless, it was interesting that the polar slab has a
higher energy than the ideal slab, due to the unscreened depolarization
field. The field is large enough to cause the slab to be metallic with
densities of states overlapping from one side of the slab to the other. It
is much less metallic than would be the case in the rigid band picture,
however, with oxygen density of states piling up at the Fermi level. This is
due to the fact that O$^{2-}$ really wants to be closed shell, and not
metallic, even on surfaces.
In the pseudopotential study, symmetric slabs were studied, but a
polarization parallel to the slab was also considered \cite{50}. The
isolated slab Hartree Fock study is the most straightforward to interpret, and they
found results similar to bulk BaTiO$_{3}$. The use of isolated, rather than
periodic, slabs, greatly facilities interpretation of results for asymmetric or
polar slabs \cite{51}.
Padilla and Vanderbilt also studied SrTiO$_3$ surfaces and found a
very similar surface energy of 1360 erg/cm$^2$ compared with 1260 in
BaTiO$_3$. In SrTiO$_3$, they found without relaxation a value 14\%
higher, very close to the estimate of 12\% for the surface
relaxation energy in BaTiO$_3$ of Cohen \cite{49}.
\section{The Future: A Vision}
We see that in the last ten years significant advances have been made in
computing properties and understanding origins of ferroelectric behavior. We
now have the ability to predict ferroelectric instabilities,
electromechanical coupling, phonon dispersion and optical spectroscopy,
elasticity, pressure behavior, order-disorder, and some surface and defect
properties of structurally and chemically relatively simple pure compounds.
Now the challenge is to address problems in more complex solid solutions,
some that are structurally and/or chemically heterogeneous and some with
frequency dependent properties.
\subsection{Relaxor systems}
In relaxor systems the dielectric response has a broad peak as a function of
temperature, rather than a sharp peak in a normal ferroelectric, and a
frequency dependent response. The origin of this behavior is still
controversial, but is most likely due to heterovalent disorder. It will be a
challenge to compute this behavior from first-principles, but models can be
parametrized using first-principles results, as in the effective Hamiltonian
models discussed above, and then molecular dynamics or Monte Carlo
simulations could be used to simulate still rather small disordered systems.
To simulate much larger, mesoscopic, systems, it may be necessary to
parametrize models for interactions between nanoregions, leaving the atomic
domain all together. But this may be possible using only \textit{ab initio}
results.
\subsection{New piezoelectrics}
The discovery of new single crystal piezoelectrics with huge
electromechanical coupling \cite{2,52,53}
are revolutionizing the field of piezoelectric transducers.
Theory did not play a role in discovering these materials, but it may play
an important role in showing how they work and helping design better
materials. The new materials are rhombohedral, with a morphotropic phase
boundary to a tetragonal phase nearby. The large response is d$_{33}$, which
is an elongation along the cubic (001) direction, and parallel to the applied
field. The rhombohedral phase though has polarizations locally about
(111). As the
the field is increased, the numbers of domains with polarizations that
project in the (0,0,-1) direction decrease, and there is a large
piezoelectric strain induced as the local polarizations are slowly rotated
up towards the (001) direction. At a critical field, a field induced phase
transition to true tetragonal symmetry occurs, and the piezoelectric
response changes and becomes more normal. An idea of why the strain is so
large in these materials can be understood by considering PbTiO$_{3}$. PbTiO$%
_{3}$ is tetragonal, with a large 6\% strain at zero pressure and low
temperatures. There is no rhombohedral phase stable in PbTiO$_{3}$ at zero
or positive pressures, but if there was, one would see a small strain in
that phase, and one would see a giant strain as one went from the
rhombohedral to tetragonal phases. One can think of phases like PMN-PT\ as
chemically engineered ``rhombohedral PbTiO$_{3}$.'' It is clear that the
polarization of the Pb$^{2+}$ ion is key to this large strain
behavior \cite{12}, though a thorough understanding of exactly how
the Pb drives the strain has not yet been fully elucidated
\subsection{Materials by design}
The goal of this research program is to lead to the ability to
computationally design useful materials. We are clearly on that path, but it
will be a few years before one can practically hope to routinely design
materials by computer. On the other hand, it is also largely a matter of
having the right good idea. We understood the role of Pb in the strain of
PbTiO$_{3}$ some years ago, and if we smart enough to think of trying to
make ``rhombohedral PbTiO$_{3}$'' we might have found this large strain
effect before experiments. However, theoretical developments were just a few
years behind to realistically do this, because the first first-principles
computation of piezoelectric constants in pure tetragonal PbTiO$_{3}$ were
only done this last year, and it hasn't been that long since the Berry's
phase approach to computing polarization was discovered. As we make further
advances, however, in the next revolution perhaps theory will lead the way
rather than follow.
\begin{ack}
This work was supported by ONR grant N00014-92-J-1019. I would like to thank T.
Egami, H. Fu, H. Krakauer, I. Mazin, K. Rabe, R. Resta, G. Saghi-Szabo, and D.
Vanderbilt for helpful discussions. Computations were performed on the Cray J90-
16/4096 at the Geophysical Laboratory, supported by NSF EAR-9512627, The Keck
Foundation, and the Carnegie Institution of Washington.
\end{ack}
|
\section*{Abstract}
We predict the amount of cometary, interplanetary, and interstellar
cosmic dust that is to be measured by the Com\-et\-ary and Interstellar
Dust Analyzer (CIDA) and the aerogel collector on-board the
Stardust spacecraft during its fly-by of comet P/Wild 2 and during the
interplanetary cruise phase. We give the dust flux on the spacecraft
during the encounter with the comet using both, a radially symmetric
and an axially symmetric coma model. At closest approach, we predict a
total dust flux of $10^{6.0}\;{\rm m}^{-2}\;{\rm s}^{-1}$ for the
radially symmetric case and $10^{6.5}\;{\rm m}^{-2}\;{\rm s}^{-1}$ for
the axially symmetric case. This prediction is based on an observation of
the comet at a heliocentric distance of $1.7\;{\rm AU}$. We reproduce
the measurements of the Giotto and VEGA missions to comet P/Halley
using the same model as for the Stardust predictions.
The planned measurements of {\em interstellar} dust by Stardust
have been triggered by the discovery of interstellar
dust impacts in the data collected by the Ulysses and Galileo dust
detector. Using the Ulysses and Galileo measurements we predict that
$25$ interstellar particles, mainly with masses of about $10^{-12}\
{\rm g}$, will hit the target of the CIDA experiment. The
interstellar side of the aerogel collector will contain $120$
interstellar particles, $40$ of which with sizes greater than $1\ {\rm
\mu m}$. We furthermore investigate the ``contamination'' of the CIDA
and collector measurements by interplanetary particles during the
cruise phase.
\section{Introduction}
The Stardust mission to comet 81P/Wild 2 is dedicated to the in situ
measurement and sample return of cosmic dust. At the comet the
pristine cometary material is investigated which is assumed to be a
good sample of the nebula from which the Solar System has
formed. Pre-solar interstellar grains that survived the formation
process may therefore be present in the samples collected or measured
in situ. On the way to the com\-et the opportunity is taken to
investigate contemporary interstellar dust that traverses the Solar
System and originates from the local interstellar low-density diffuse
cloud which crosses the Sun's way through the Milky Way
\cite{frisch95}. We predict the dust flux on the two main instruments
on-board Stardust, the aerogel dust collector and the Cosmic and
Interstellar Dust Analyzer (CIDA) \cite{brownlee97}. The dust
collector is a plate with an area of $\approx 0.1\;{\rm m}^2$ that is
to be exposed to the dust stream. Particles hitting the collector are
trapped inside the aerogel \cite{albee94}. CIDA is an in-situ impact
plasma dust detector with a time-of-flight mass spectrometer similar
to the PUMA/PIA instruments flown on Giotto \cite{kissel86}.
Stardust launches in February 1999 and flies by Earth in January
2001 for gravity assist. The slow ($6.1\ {\rm km}\ {\rm s}^{-1}$)
fly-by of comet P/Wild 2 occurs on January 1st, 2004 at a heliocentric
distance of $1.86\ {\rm AU}$ when the comet is on the out-bound part of
its orbit. During the fly-by the distance of the
spacecraft to the comet nucleus is $\approx 100\ {\rm km}$, and the
phase angle, i.e. the Sun--comet--spacecraft angle, is
$70^\circ$. After fly-by Stardust returns to Earth in January 2006,
where the sample-return capsule that contains the aerogel collector is
separated from the spacecraft and sent to a direct entry into Earth's
atmosphere. The collection and in situ measurement of interstellar
dust occurs during the cruise phase as shown in figure
\ref{fig_traj}.
\begin{figure}[htb]
\hspace{-5mm}
\epsfbox{sd_traj.ps}
\caption{\label{fig_traj} The trajectory of Stardust (solid line) in
the ecliptic plane (the verenal equinox direction is to the
right). The orbits of P/Wild 2 and Earth are indicated as dashed
lines. Mission events are marked with the time given in days after
launch in brackets. The flow direction of interstellar particles from
the LIC as derived from the Ulysses/ Galileo measurements is shown by
the arrow in the lower left corner. Thick lines indicate the part of
the orbit used to collect interstellar dust with the Aerogel
collector.}
\end{figure}
During the short time of the encounter with the comet cometary dust
particles are collected which are lifted from the comet's nucleus and
accelerated into the coma by the drag of the outflowing gas. The
amount of material in a comet's coma depends on the activity of its
nucleus, which, in turn, depends on the distance from the Sun and on
the properties of the nucleus' surface. To determine P/Wild 2's
activity we use ground based observations. For the properties of the
dust phase, especially the mass distribution of dust particles,
contained in a comet's coma, we rely on data collected with the VEGA
and Giotto spacecraft at comet P/Halley during its visit of the inner
Solar System in 1986. Since P/Halley and P/Wild 2 differ in activity
and size, and since the VEGA and Giotto measurements had a different
geometry than the measurements of Stardust, we have to construct
a model of the coma and use
the measured activity, brightness and mass
distribution as an input to this model. By using the model to
reproduce the spacecraft measurements of the dust fluence at P/Halley,
we can validate the model and the procedure of taking the input
activity and brightness from ground based observations.
The secondary goal of the Stardust mission is the collection and
in-situ measurement of contemporary interstellar dust particles which
enter the Solar System from the local interstellar cloud (LIC). The
same instruments are used for the interstellar dust measurements as
for the cometary part of the mission. During definite parts of
Stardust's cruise, the dust collector will expose its backside to the
interstellar dust stream and on other occasions CIDA will be pointed
into the upstream-direction (see figure \ref{fig_traj}). The
interstellar dust stream was discovered by
Gr\"un et al. (1993) \cite{gruen93} in the data collected by the
dust-detector on-board the
Ulysses spacecraft after fly-by of Jupiter. It was clearly identified
and distinguished from interplanetary dust by its opposite impact
direction, its impact velocity in exceess of the escape velocity at
Jupiter distance, and its constant impact rate at high heliocentric
latitudes. The Ulysses measurements have been confirmed by impacts
from the retrograde direction measured by the identical
dust-detector
on-board the Galileo spacecraft
\cite{baguhl95b}. The total flux of
interstellar particles was determined to be $1.5\cdot 10^{-4}\ {\rm
m}^{-2}\ {\rm s}^{-1}$
\cite{gruen94}. The mass distributions of the Ulysses and Galileo
measurements have been found to be nearly identical and range from
$10^{-15}\ {\rm g}$ to $10^{-8}\ {\rm g}$ with particles of masses of
$10^{-13}\ {\rm g}$ being most abundant
\cite{landgraf98a}. From the
impact direction it was derived that interstellar dust particles come
from the same direction \cite{baguhl95a} as interstellar neutral
Helium which was measured by the Ulysses/GAS experiment
\cite{witte93}. The upstream-direction of the interstellar Helium was
determined to be $\lambda_{\infty,{\rm gas}}=254.7^\circ \pm
1.3^\circ$ (heliocentric longitude) and $\beta_{\infty,{\rm
gas}}=4.6^\circ\pm 0.7$ (heliocentric latitude)
\cite{witte96}. Furthermore the relative velocity of the gas was
given by $v_{\infty,{\rm gas}}=25.4\pm 0.5\ {\rm km}\ {\rm
s}^{-1}$. These parameters relate the interstellar material measured
in the Solar System with the LIC which has been identified in the
Doppler shift measured in the UV by HST-GHRS
\cite{lallement92}. The Doppler-shift indicates a relative motion between
the Sun and the LIC with a velocity of $25.7\pm 0.5\;{\rm km}\;{\rm
s}^{-1}$.
\begin{figure}[htb]
\epsfxsize=.9\hsize
\centering\epsfbox{srtmp.ps}
\caption{\label{srtmp}
Equilibrium surface temperature versus the angle $\vartheta_s$ from
the sub-solar point as given by the solution of equation
(\ref{surt}).}
\end{figure}
In this paper we use the expressions ``flux'' for the number of impacting
particles of a given mass per unit area and time, ``fluence'' for the flux
integrated over time, ``total flux'' for the flux of particles of all
masses, and ``total fluence'' for the total flux integrated over time.
\section{Coma Model\label{commod}}
In the following we describe the coma model we use to determine the
cometary dust flux on the Stardust spacecraft. Because there is no
standard model of the gas and dust environment of a comet nucleus,
yet, we describe the assumptions and calculations we use in
detail. So the reader can see how different assumptions or values of
parameters change the results.
Since the dust is dragged into the coma by outflowing gas, we first
have to calculate the gas density and velocity inside the coma. We
assume that the outflow is symmetric about the Sun-nucleus
axis. Therefore, all spatial distributions depend only on the angle
$\vartheta_s$ between the considered location and the axis of symmetry
and the radial distance $r$. We furthermore assume that
the dust particles do not affect the gas flow, i.e. we use a
test-particle approach.
\subsection{Gas Phase}
To calculate the state of the gas around the nucleus, we need to
know the activity distribution on
the nucleus' surface, which is governed
by the temperature distribution.
The temperature at a location on the day side of the nucleus' surface
is calculated such that there is an equilibrium between solar
illumination, water sublimation, and blackbody reradiation. Therefore,
the surface temperature $T_s(\vartheta_s)$ is the solution of the
following equation (the meaning and values of the
parameters in the next two equations are given in table
\ref{surtab}).
\begin{equation}\label{surt}
\renewcommand{\arraystretch}{1.5}
\begin{array}{l}
\left( 1 - A_B \right) \left( \frac{f_{Sun}}{(r_h/1AU)} \right)^2
\cdot \cos( \vartheta_s ) \\
\hspace{2cm} = \frac{L}{N_A} Z_{\rm HK}(T_s(\vartheta_s))
+ \epsilon \left( \sigma(T_s(\vartheta_s)) \right)^4,
\end{array}
\renewcommand{\arraystretch}{1}
\end{equation}
where $Z_{\rm HK}(T_s)$ is the Hertz--Knudsen sublimation
rate given by:
\begin{eqnarray}\label{eqn_zhk}
Z_{\rm HK}(T_s) & = & \frac{ p(T_s) }{ \sqrt{2\pi m_{\rm H_2O} k T_s} }
\end{eqnarray}
We use the approximation $p(T_s)=Ae^{(-B/T_s)}$ for the vapor pressure
\cite{fanale}.
\begin{table}
\caption{\label{surtab}
Parameters for computation of the surface
temperature.}
\begin{tabular}{ll}
\hline
Solar constant &$f_{\rm Sun}=1370\;{\rm W}\;{\rm m}^{-2}$\\
nucleus surface albedo\footnotemark[1] & $A_B=6\%$\\
Emissivity of comet surface &$\epsilon=1$\\
Boltzmann constant & $k=1.38\cdot 10^{-23}\;{\rm J}\;{\rm K}$\\
Stefan Boltzmann constant & $\sigma=5.67\cdot 10^{-7}\;{\rm W}\;{\rm
K}^{-4}\;{\rm m}^{-2}$\\
Advogadro Number & $N_A=6\cdot 10^{23}{\rm mol}^{-1}$ \\
Latent heat of water ice & $L=36\;{\rm kJ}\;{\rm mol}^{-1}$\\
Vapor pressure parameter &
$A=3.56\cdot 10^{12}\;{\rm N}\;{\rm m}^{-2}$\\
Vapor pressure parameter & $B=6141\;{\rm K}$ \\
\hline
{\footnotesize $^1$ bolometric Bond albedo}
\end{tabular}
\end{table}
On the night side of the nucleus the temperature is
constantly set to $150\;{\rm K}$. This value is also used at the day
side near the terminator if the solution of equation (\ref{surt}) is
smaller than $150\;{\rm K}$. Figure \ref{srtmp} shows the equilibrium
temperature distribution determined from equation (\ref{surt}).
\begin{figure}[htb]
\epsfxsize=.9\hsize\epsfbox{gas_bw.ps}
\caption{\label{gasiso} Solid lines represent lines of constant gas
density (arbitrary units) in the region up to 25 comet radii
($=50\;{\rm km}$) around the nucleus. The Sun direction is upward and
the short radial lines indicate the gas velocity (the scale on the top
can be used to determine the modulus of the velocity). The gas density
above the comet night side (bottom) is much lower than the density
above the day side. However, the gas velocity above the day and night
side is comparable. Apart from a region above the terminator, the gas
flow is almost radial close to the nucleus.}
\end{figure}
Now, we obtain the activity $Q_{\rm HK}$ of the comet by inserting the
temperature distribution into equation (\ref{eqn_zhk}) and integrate
over $\vartheta_s$.
\begin{eqnarray}
Q_{\rm HK} & = & 2\pi r_n^2 \int_{-1}^{1} Z_{\rm HK}(\vartheta_s)
d\cos(\vartheta_s).
\end{eqnarray}
In our model we use the outgasing activity of the nucleus given by
Farnham and Schleicher (1997) \cite{obs1}. On 1997 March 5 they report an
$\rm OH$-activity $Q_{\rm OH}=5.9\cdot 10^{27}\;{\rm s}^{-1}$ and a value of
the albedo--filling factor--radius product of $Af\rho=4.27\;{\rm
m}$. At this time the
comet was at a heliocentric distance of $1.7\;{\rm AU}$ and was
observed at a phase angle of $27^\circ$. Although the observation was
taken on the in-bound part of the comet's orbit, and the Stardust
spacecraft will approach the comet on the out-bound part, we assume
that the observed activity is comparable to the comet's activity
during the encounter. Therefore, we run our coma model for a comet at
$1.7\;{\rm AU}$ with a water activity equal to the observed
$\rm OH$-activity $Q_{\rm H_2O}=5.9\cdot 10^{27}\;{\rm s}^{-1}$. To match the
observed activity, the water sublimation rate is
calculated by scaling the Hertz--Knudsen sublimation rate:
\begin{eqnarray}
Z_{\rm H_2O}(\vartheta_s) & = & \frac{Q_{\rm H_2O}}{Q_{\rm HK}} \cdot Z_{\rm
HK}(T_s(\vartheta_s))
\end{eqnarray}
In addition to the water activity, we assume a $\rm CO$-activity of $10\%$
of the water activity, i.e. $Q_{\rm CO}=5.9\cdot 10^{26}\;{\rm s}^{-1}$,
and a constant distribution above the nucleus' surface
$Z_{\rm CO}(\vartheta_s)=Q_{\rm CO}/(4\pi r_n^2)$. For calculating the
acceleration of the dust particles by the cometary gas, we have to
consider the most abundant gas molecules only,
therefore other constituents than
$\rm H_2O$ and $\rm CO$ are not considered in the model and
the total gas activity is set to $Q_{\rm gas}=Q_{\rm H_2O}+Q_{\rm CO}=6.5\cdot
10^{27}\;{\rm s}^{-1}$. Accordingly, we set the total gas activity
distribution to $Z_s(\vartheta_s) = Z_{\rm H_2O}(\vartheta_s) +
Z_{\rm CO}(\vartheta_s)$. Using $Z_s(\vartheta_s)$, the thermodynamic
state of the gas can be determined by a numerical procedure described
by Crifo et al. (1995) \cite{crifo2} and Knollenberg (1993) \cite{knoll}.
As an input to the calculation of the gas state we use a mean mass of
a gas molecule of $m_{\rm gas} = (m_{\rm H_2O} Q_{\rm H_2O} + m_{\rm
CO} Q_{\rm CO})$ $/
(Q_{\rm H_2O} + Q_{\rm CO})$ and a mean degree of freedom per molecule of
$f_{\rm gas} = (f_{\rm H_2O} Q_{\rm H_2O} + f_{\rm CO} Q_{\rm CO}) /
(Q_{\rm H_2O} +
Q_{\rm CO})$. This simplification introduces some physical inconsistency
into the model, but the results do not depend significantly on these
choices.
Another important parameter for modeling the gas outflow is the initial gas
temperature above the nucleus' surface, because it determines the
velocity reached by the gas, and therefore also has an impact on
the dust acceleration. The initial gas temperature above the nucleus'
surface does not necessarily coincide with the surface temperature, because
the gas is not at rest with respect to the surface. This
effect was taken into account as it is described in
Crifo and Rodionov (1997) \cite{crifo1} (Appendix B).
As shown in figure \ref{gasiso}, the gas density determined by
our thermodynamic model is distributed
highly asymmetric around the
nucleus. On the day side much more gas is available to accelerate dust
from the surface. The flow direction is nearly radial. In the outer
region of the coma, the gas is not dense enough to exert a significant
drag force on the dust particles. We assume that the gas drag is
negligible outside a maximum distance $r_{\rm max}$ of $25$ times the
radius of the nucleus. After leaving this region, the dust particles
move independently from the gas.
\subsection{Dust Phase}
The dust phase of a comet's coma consists of all dust particles lifted
and accelerated from the nucleus by the gas drag. The mass range of
dust particles emitted by a comet covers many orders of magnitude and
the dynamics of these particles depend on their mass. Therefore we
divide the mass range $[10^{-20}\;{\rm kg},10^{0}\;{\rm kg}]$ into
$20$ logarithmic mass decades $[m_{i-1},$ $m_i]$, $i=1,\ldots,20$ which
we call ``dust classes'' in the following. All dust particles which
are released with a mass inside one of the intervals $i$ are modeled
by a dust particle of a representative mass $m_{d,i}$. We assume that
no processes which change the properties of a dust particle,
e.g. fractionation due to mutual collisions, take place after a
particle has left the nucleus' surface. Furthermore, we assume the
dust particles to be spherical with a constant bulk density
$\rho_d=1\;{\rm g}\;{\rm cm}^{-3}$ and a radius
$s_i=(3m_{d,i}/(4\pi\rho_d))^{(1/3)}$. For the determination of the
amount of dust in the coma, we derive a dust-to-gas mass ratio
$\chi_i$ for each dust class. Using the dust to gas ratios $\chi_i$,
we calculate the number of dust particles released per unit area and
time.
\begin{eqnarray}
Z_{d,i}(\vartheta_s) & = & \chi_i \frac{m_{\rm gas}}{m_{d,i}}Z_{\rm
gas}(\vartheta_s)
\end{eqnarray}
We take into account that particles cannot be lifted from the comet
surface above a given mass by setting the dust activity to zero at
locations on the comet surface where the gravitational attraction of
the comet nucleus exceeds the drag force exerted by the gas.
To calculate the gravitational force of the nucleus, we determined the
nucleus' size by using ground-based observations of P/Wild 2 at large
heliocentric distances. Fitzsimmons and Cartright (1995) \cite{obs2}
report a red magnitude of $R=22$ at the heliocentric distance of
$r_h=4.4\;{\rm AU}$, a geocentric distance of $\Delta=4.0\;{\rm AU}$,
and a phase angle of $\alpha=12^\circ$. We assume that P/Wild 2's
nucleus has a spherical shape. The radius of the nucleus $r_n$ can
be calculated by (compare Jewitt (1991) \cite{jew})
\begin{eqnarray}\label{eqrn}
r_n & = & r_h \frac{\Delta} {1\;{\rm AU}} \sqrt{ \frac{10^{-0.4 (R -
R_{\rm Sun})}} {p_{\rm nuc} \cdot j_{\rm nuc}(\alpha)} },
\end{eqnarray}
where $R_{\rm Sun}=-27.22$ is the red magnitude of the Sun \cite{land},
$p_{\rm nuc}=4\%$ is used for the geometric albedo of the nucleus and
$j_{\rm nuc}(\alpha)=10^{-0.4\delta\alpha}$ is the phase function of the nucleus
with $\delta=0.035\;{\rm deg}^{-1}$ \cite{jew}.
We find $r_n=2.3\;{\rm km}$. Because
Fitzsimmons and Cartright (1995) \cite{obs2}
reported a ''near--stellar appearance'', there was already some
contribution of the dust coma to the observed intensity. Therefore,
$r_n=2.3\;{\rm km}$ is an overestimation of the radius of the
nucleus and we use the value $r_n=2\;{\rm km}$ in our model.
\begin{figure}[htb]
\epsfxsize=.8\hsize\epsfbox{dustra09.ps}
\caption{\label{dsttra09}
Trajectories of dust particles with radii of $8.8\;{\rm \mu m}$, which
start at equidistant positions on the surface. The scales in the image
are nucleus radii. The velocity of the particles is
dominated by the radial component, far from the nucleus the tangential
velocity component is negligible. The distance of the trajectories
above the terminator is increased due to the tangential gas flow in
this region. }
\end{figure}
Assuming a bulk density of $\rho_n=0.5\;{\rm g}\;{\rm cm}^{-3}$, the
gravitational acceleration is given by:
\begin{eqnarray}
\vec{a}_{\rm grav} & = & \mu_n
\frac{\vec{r}} {\left|\vec{r} \right|^3}
\hspace{0.5cm} \mbox{with}
\hspace{0.5cm} \mu_n=1118\;{\rm m}^3\;{\rm s}^{-2}.
\end{eqnarray}
The acceleration of a dust particle due to gas drag
can be expressed using the drag coefficient $C_D$:
\begin{eqnarray}
\vec{a}_{\rm drag} & = & \frac{1}{2} m_{\rm gas} n_{\rm gas} \cdot
\frac{\pi s_i^2}{m_{d,i}} C_D \left| \vec{v}_{\rm gas}
- \vec{v}_{\rm dust} \right| \nonumber \\
&& \cdot \left( \vec{v}_{\rm gas} - \vec{v}_{\rm dust} \right)
\end{eqnarray}
The drag coefficient in the free molecular approximation is given by
Probstein (1968) \cite{prob}
\begin{eqnarray}
C_D & = & \frac{2 \sqrt{\pi}} {3s} \sqrt{\frac{T_d}{T_{\rm gas}}} +
\frac{2s^2+1} {s^3 \sqrt{\pi}} e^{(-s^2)} \nonumber \\
&& + \frac{4s^4+4s^2-1} {2s^4}
\mbox{erf}(s),
\end{eqnarray}
where $s=|\vec{v}_{\rm gas} - \vec{v}_{dust}| /\sqrt{2kT_{\rm gas} /
m_{\rm gas}}$. The dust temperature is set to $T_d=228\;{\rm K}$
(compare Divine (1981) \cite{div1}: $T_d=310\;{\rm K} \cdot (r_h/1\;{\rm
AU})^{-0.58}$ and $r_h = 1.7\;{\rm AU}$). At every point on the
nucleus' surface for which the gas drag acceleration exceeds the
gravitational acceleration of the nucleus, the equation of motion of a
dust particle can be solved numerically inside the gas flow (see
figure \ref{dsttra09} and
\ref{dsttra18}). For large particles which can be lifted only from
the comet's day side, there are trajectories for which the radial
velocity is not directed away from the nucleus but fall back into the
nucleus' direction (see figure \ref{dsttra18}). Since this behavior
is only found in a narrow region, we do not consider these particles
in the following.
\begin{figure}[htb]
\epsfxsize=.8\hsize\epsfbox{dustra18.ps}
\caption{\label{dsttra18}
Trajectories of dust particles with a radius of $8.6\;{\rm
mm}$. These big
dust particles can only be lifted in a region around the sub-solar
point. Near the boundary of this region there are trajectories which
start to fall back in nucleus direction. In our model we neglect the
contribution of these dust particles to the dust number density of the
coma.}
\end{figure}
Apart from a narrow region near back-falling particles, the dust
velocity is dominated by the radial velocity component. Since the
contribution of the tangential velocity to the total dust velocity
decreases with increasing distance from the nucleus' surface, the dust
trajectories can be approximated by straight radial lines with
constant radial velocity at large distances. Outside the maximum
distance $r_{\rm max}$ we approximate the dust trajectories by
straight, radial lines and constant velocities $v_{e,i}(\vartheta)$
which is equal to the radial velocity computed for a dust particle at
the distance $r_{\rm max}$. The angle $\vartheta$ denotes the angle of
the position of the dust particle with respect to the Sun direction at
$r_{\rm max}$, and is also the angle of the extrapolated trajectory
with the Sun direction.
The dust number density outside the radius $r_{\rm max}$ can then be
computed using the position $\vartheta_s$ of the trajectory at the
nucleus' surface as a function of the angle $\vartheta$ with respect
to the Sun direction. To calculate the number density $n_{d,i}$ at
positions with $r>r_{\rm max}$ and angle $\vartheta$, we introduce
the activity distribution of the comet outside the acceleration zone:
\begin{eqnarray}
Z_{e,i} & = & r_n^2 Z_s(\vartheta_s(\vartheta)) \frac{d
\cos(\vartheta_s)} {d \cos(\vartheta)}
\end{eqnarray}\label{numb}
For the number density $n_{d,i}$ we get:
\begin{eqnarray}
n_{d,i}(r,\vartheta) & = & \frac{Z_{e,i}(\vartheta)}{r^2 v_{e,i}(\vartheta)}
\end{eqnarray}
The number density inside the acceleration cone can be calculated in
an analogous way.
As a result, we show the contours of constant dust particle density in
figure \ref{dstiso}. The region of high dust density occupies a larger
volume on the day side than on the night side. Due to the drop in
activity at the terminator, the dust particle density right above the
terminator is lower than on the night side. Dust particles from the
day side can propagate freely to the night side which extends the
region of high dust particle density just behind the terminator in a
narrow radial feature. If this is a real feature at a comet and if it
can be observed in reality remains an open question. The dust particle
density opposite to the Sun direction is the result of the night side
activity.
\begin{figure}[htb]
\hspace{3mm}\epsfxsize=\hsize\epsfbox{dst09_bw.ps}
\caption{\label{dstiso}
The lines of constant dust number density of particles with radii of
$8.8\mu m$ in the region within 25 radii of the nucleus. The
radial lines indicate the velocity of the dust particles. Because of
the low gas density, the dust particle velocity is smaller above the
night side. Due to the increased tangential
distance of the dust trajectories, the dust number
density is small above the terminator (see figure \ref{dsttra09}).}
\end{figure}
We want to use the model to determine the dust flux on the Stardust
instruments during the fly-by of the nucleus. We can neglect
the influence of radiation pressure on the particle's trajectories,
because Stardust approaches the
nucleus sufficiently close. Because
the velocity of Stardust relative to nucleus is much larger than the
velocity of the dust particles, the dust impact velocity at the
spacecraft is dominated by the spacecraft's motion. We therefore
calculate the dust flux on the spacecraft by the product of the dust
number density and the spacecraft velocity $v_{\rm SC}$ relative to
the nucleus.
\begin{eqnarray}\label{fsc}
f_{\rm SC} & = & n_{d,i} \cdot v_{\rm SC}
\end{eqnarray}
From the result in equation (\ref{fsc}) we can derive the dust fluence
on the instruments, which is given by the time integral over the flux
and represents the column density of dust particles along the
spacecraft's trajectory. We need the column density of dust particles
also to calculate the brightness measured by an ground-based observer
for a given line of sight.
As we show in Appendix A, the column density (in number of
dust particles per ${\rm m}^2$) for an image plane with phase angle
$\alpha$ can be calculated by
\begin{eqnarray}\label{nproj}
\frac{d n_{d,i}}{d(x',y')} & = & \frac{1} {\rho} \int_{-u}^{u}
d\cos(\vartheta) \; \frac{Z_{e,i}(\vartheta)} {v_{e,i}(\vartheta)}
\nonumber \\
&&\cdot \frac{1} {\sqrt{u^2 - \left(\cos(\vartheta)^2\right)}}
N(\alpha,\eta,\vartheta),
\end{eqnarray}
where $u = u(\alpha,\eta) = \sqrt{1 - (\sin(\eta) \cdot
\sin(\alpha))^2}$. The variables $\rho$ and $\eta$ represent the
position of the line of sight in the projected plane: $\rho$ is the
distance from the (projected) nucleus position and $\eta$ denotes the
direction of the line of sight position with anti-solar direction
projected in the image plane (compare figure \ref{app1}).
$N(\alpha,\eta,\vartheta)$ is equal to the number of intersections of
a line of sight with the cone of constant $\vartheta$ and is equal to
$0$, $1$, or $2$ (see table \ref{naet}).
\begin{table}
\caption{\label{naet} Number of intersections of a line of sight of
angle $\eta$ with a cone of opening angle $\vartheta$ for different cases as
explained in the text. Note that $(N(\alpha, \eta, \vartheta) +
N(\alpha, \pi-\eta, \vartheta))/2 = 1$ for $|\cos(\vartheta)| < u$.}
\begin{tabular}{lll}
\hline
Condition 1 & Condition 2 & $N(\alpha,\eta,\vartheta)$ \\
\hline
\hline
$u < \cos(\vartheta)$ & none & $0$ \\
\hline
$|\cos(\alpha)|<\cos(\vartheta)<u$ &
\begin{tabular}{l}
$\cos(\eta)<0$\\
$\cos(\eta)>0$
\end{tabular}
&
\begin{tabular}{@{\hspace{0cm}}l@{\hspace{0cm}}}
$2$\\
$0$
\end{tabular}
\\
\hline
$\begin{array}{l}
|\cos(\alpha)|>|\cos(\vartheta)>\\
-|\cos(\alpha)| \end{array} $ & none & 1 \\
\hline
$-|\cos(\alpha)|>\cos(\vartheta)>-u$ &
\begin{tabular}{l}
$\cos(\eta)<0$\\
$\cos(\eta)>0$
\end{tabular}
&
\begin{tabular}{@{\hspace{0cm}}l@{\hspace{0cm}}}
$0$\\
$2$
\end{tabular}
\\
\hline
$-u>\cos(\vartheta)$ & none & $0$ \\
\hline
\end{tabular}
\end{table}
For comparison with our axially symmetric model we provide the column
density in the radially symmetric case:
\begin{eqnarray} \label{nprojrsym}
\frac{d n_{d,i}}{d(x',y')} & = & \frac{\pi Z_{e,i}}{\rho v_{e,i}}.
\end{eqnarray}
\subsection{Adjustment of the Dust Activity to the Observed $Af\rho$-value}
We adjust the dust activity of our model so that the observed
$Af\rho_{\rm obs}=4.27\;{\rm m}$ is matched. The $Af\rho$-parameter is related to
the intensity $f_{\rm dust}$ received from dust particles inside a field
of view with radius $\rho$ by A'Hearn et al. (1984) \cite{ahearn}:
\begin{eqnarray}
Af\rho & = & 4\frac{\Delta^2(r_h/1AU)^2} {f_{\rm Sun}} \cdot
\frac{f^*_{\rm dust}} {\rho},
\end{eqnarray}
where $f^*_{\rm Sun}$ is the radiation flux density of the Sun at a
heliocentric distance of $1\;{\rm AU}$ observed in the same band of
the spectrum as the comet, $\Delta$ is the geocentric distance and
$r_h$ the heliocentric distance.
The intensity received due to a single dust particle is given by
\begin{eqnarray}
I_{sng,i} & = & pj(\alpha) \cdot \frac{\pi s_i^2}{\pi \Delta^2} \cdot
\frac{f^*_{\rm Sun}} {(r_h/1AU)^2},
\end{eqnarray}
where $p$ is the dust geometric albedo and $j(\alpha)$ is the
dust phase function of the phase angle $\alpha$.
Using the number of dust particles inside the aperture $N_{d,i}$
(see equation (\ref{naper})), we can determine
the $Af\rho$-parameter of our model by
\begin{eqnarray}
Af\rho_{\rm mod} & = & 4 pj(\alpha) \sum_i s_i^2 \int_{0}^{2\pi}
\int_{-u}^{u} \frac{Z_{e,i}(\vartheta)} {V_{e,i}(\vartheta)}\nonumber
\\
&& \cdot
\frac{1} {\sqrt{u^2 - \cos^2(\vartheta)}} d\cos(\vartheta)\; d\eta.
\end{eqnarray}
Here we assume that all dust particles have a geometric albedo
and a phase function independent of size. For the geometric albedo we assume
a value of $4$\% and take the phase function from Divine (1981) \cite{div1}.
To match the observed $Af\rho_{\rm obs}$ with the coma model, we have to
scale the dust activity of the comet, i.e. the dust to gas ratio of
all dust classes have to be scaled by the factor
$Af\rho_{\rm obs}/Af\rho_{\rm mod}$.
\subsection{Dust Mass Distribution}
The mass distribution of the dust particles is the most crucial
parameter of the model, because a change in the mass distribution can
change the total dust fluence by orders of magnitude. Therefore we take
special care to choose a reasonable mass distribution. In this
section we explain our choice and discuss the resulting fluence in
comparison with the measured fluence in the coma of P/Halley. The
reader might want to check what the effects of a different choice of
the mass distribution are.
The VEGA spacecraft measured the number of particles per unit area as
a function of mass. We use this measurement as the mass distribution
of our model. As an analytical representation of the mass distribution,
we use a fit by Divine and Newburn (1987) \cite{div2}.
\begin{eqnarray}\label{msd}
F(m) & = & F_t \left(\frac{(1+x)^{b-1}} {x^b}\right)^{ac}
\hspace{0.5cm} \\
&& \mbox{with} \hspace{0.5cm}
x = \left(\frac{m} {m_t}\right)^{1/c}. \nonumber
\end{eqnarray}
The parameters which fit the cumulative mass distribution of the VEGA
2 fluence best are $a=0.9$, $b=0.29$, $c=2.16$, and $m_t=1.6\cdot
10^{-13}\;{\rm kg}$ \cite{div2}. Since the mass
distribution is only used to determine the relative but not the total
abundance of particles in different dust classes, $F_t$ can be set to
an arbitrary value. Note that $-a=-0.9$ is the exponent of the
cumulative mass distribution for big dust particles ($m \gg m_t, x
\gg 1$), and $-ac = -1.9$ is the exponent for small particles ($m \ll
m_t, x \ll 1$).
For each dust class $[m_{i-1},m_i],\; i=1,\ldots,20$ considered in the
model, the fluence on the VEGA 2 spacecraft inside one of the mass
bins is proportional to
\begin{eqnarray}
n_{VG2,i} & \propto & F(m_{i-1})-F(m_i)
\end{eqnarray}
The dust fluence on the VEGA 2 spacecraft can be modeled analogous
to the fluence on Stardust. At closest approach, VEGA 2 was also far
inside the envelope of the measured dust particles, therefore, the
total fluence was dominated by direct particles.
Because of the dependency of the integrand in equation (\ref{nproj})
on the dust escape velocity $v_{e,i}$, which, in turn, depends on the
mass of the dust particle, the mass distribution of particles which
hit the VEGA 2 instruments does not coincide with the mass
distribution of particles at the nucleus' surface, i.e. the
distribution of $Z_{d,i}$. Therefore a correction has to be applied to
the VEGA 2 fluence before it can be used as mass distribution at the
nucleus' surface. For this purpose we compute the dust escape
velocities $v_{e,i,{\rm Halley}}$ for an analogous radially symmetric
P/Halley model (The model parameters are
$Q_{\rm H_2O}=6.8\cdot10^{29}\;{\rm s}^{-1}$ (see Krankowsky (1986)
\cite{krank},
$Q_{\rm CO}=0\;{\rm s}^{-1}$, $r_n=5\;{\rm km}$; all other
model parameters are not changed). The fluence of dust class $i$ is
inversely proportional to the dust escape velocity $v_{e,i,Halley}$
for a radially symmetric coma model (see equation (\ref{nprojrsym})).
Therefore, the product $n_{srf,i}=n_{VG2,i} \cdot v_{e,i,{\rm
Halley}}$ is proportional to the mass distribution at the comet's
surface and the dust-to-gas mass ratios of the dust classes can be
computed by
\begin{eqnarray}
\chi_i & = & \chi_{\rm aux} \frac{m_{d,i}n_{srf,i}} {\sum_j
m_{d,j}n_{srf,j}},
\end{eqnarray}
where $\chi_{\rm aux}$ is an auxiliary dust to gas ratio.
$\chi_{\rm aux}$ would have the meaning of the ratio of dust mass to
gas mass released per unit time from the comet's nucleus, if all
dust classes could be lifted from the comet's surface. Since there
are dust classes which can not be lifted from the comet's surface or
can only be lifted inside some region around the sub-solar point,
no physical interpretation can be linked to this number and is
in fact only an auxiliary value needed during the computation.
\section{Model of LIC Dust Traversing the Solar System}
The impact parameters of interstellar particles on the dust collector
and the CIDA target are determined by the impact di\-rec\-tion and
velocity of the particles, and also by the articulation of the
spacecraft during the cruise phase. Since both is non-trivial, we
separate the effects by discussing two different types of particles:
(1) the reference particles as defined by the Stardust mission plan
\cite{stardust97} which move on straight lines through the Solar
System in a direction derived from Ulysses and Galileo data, and (2)
larger particles coming from the same direction, but being deflected
due to solar gravity. For the reference particles the change in impact
parameters is caused only by the spacecraft's trajectory and the
articulation strategies defined in the mission plan for the cruise
phase. Larger particles are mainly interesting for the dust-collector,
because CIDA is unlikely to be hit by such a particle due to its small
target area and the low abundance of these particles. Furthermore,
larger particles can be more easily extracted from the aerogel, this
extraction might be difficult for particles with sizes of about a few
tenth of a micron. In the following we argue why the reference
particles are a good approximation of the most abundant interstellar
particles, which are to be measured with Stardust and why smaller
interstellar particles are not expected to reach the spacecraft.
In general, the dynamics of a charged, massive dust-particle in the
interplanetary environment can be described by
\begin{eqnarray}
\ddot{\vec{r}} & = & -\gamma \left(1 - \beta\right) \frac{M_\odot}
{\left| \vec{r}\right|^3} + \frac{Q}{m}\left(\vec{v}_{\rm swf} \times
\vec{B} \right).\label{eqn_om}
\end{eqnarray}
This equation of motion takes into account gravity, radiation
pressure, and the Lorentz force induced by the solar wind magnetic
field. $\gamma$ is the coupling constant of gravity, $M_\odot$ is the
solar mass and the parameter $\beta$ is defined as the ratio of the
magnitudes of radiation pressure force and gravity. $Q$ is the
equilibrium charge to which a dust-particle will be charged in the
interplanetary radiation and plasma environment, $\vec{v}_{\rm swf}$
is the velocity vector of the particle, measured in the frame of the
radially expanding solar wind, and $\vec{B}$ represents the solar wind
magnetic field. In equation (\ref{eqn_om}) drag forces like
Poynting-Robertson drag and solar wind drag are neglected, because
these drag forces are small and only have a long-term effect on
particles on bound orbits. The dynamical parameters $\beta$ and $Q/m$
depend on the particle size and can be represented as functions of
particle radius $a$, if we assume a spherical shape for the particles
\cite{gustafson94,mukai81}.
\subsection{Reference Particles\label{straightlines}}
Physically, the reference particles are implemented by setting $\beta=1$
and $Q/m=0\ {\rm C}\ {\rm kg}^{-1}$. In this case the solution to equation
(\ref{eqn_om}) are trajectories along straight lines,
i.e. $\dot{\vec{r}}=\vec{v}_{\infty,{\rm dust}}={\rm const.}$. Since
the Ulysses measurements indicate that interstellar dust and gas are
kinematically coupled, we set $\vec{v}_{\infty,{\rm
dust}}=\vec{v}_{\infty,{\rm gas}}$ using the parameters derived for
the gas by Witte et al. (1996) \cite{witte96}
Gustafson (1994) \cite{gustafson94} calculated $\beta$ as a function
of particle radius assuming a spherical shape and a composition
like the astronomical silicates defined by
Draine and Lee (1984) \cite{draine84} (see figure \ref{fig_beta}). Using
these calculations
it was shown \cite{landgraf98b}, that for $57\%$ of all interstellar
particles measured by Ulysses and Galileo $|1-\beta|\leq 0.4$. If the
particles are measured at about $2\ {\rm AU}$, this deviation from the
reference behavior would change the impact angle by less than
$15^\circ$, which is the uncertainty in the determination of the
interstellar dust upstream direction from the Ulysses and Ga\-li\-leo
measurements. The impact velocity will deviate by less than $30\%$ for
these particles. Furthermore, the variation of impact direction and
velocity due to electro-magnetic interaction of interstellar
particles with the solar wind magnetic field ($Q/m \neq 0\ {\rm C}\
{\rm kg}^{-1}$) can be neglected if they have radii larger than $0.2\
{\rm
\mu m}$.
\begin{figure}[htb]
\hspace{-2mm}\epsfbox{bobeta.ps}
\caption{\label{fig_beta}The parameter $\beta$ as a function of
particle radius $a$. Homogeneous, spherical particles with the optical
constants of astronomical silicates have been assumed (see
\protect\cite{gustafson94}).}
\end{figure}
\subsection{Gravitational Particles\label{grav_grains}}
As shown in figure \ref{fig_beta}, $\beta \approx 1$ is not valid for
large particles. Large interstellar particles ($m>10^{-12}\ {\rm g}$)
are less abundant than smaller ($m<10^{-12}\ {\rm g}$) ones
\cite{landgraf96}, so they are more important for the dust collector
than for CIDA due to its larger surface area. If we neglect
electro-magnetic effects, the trajectories are hyperbolae and the
spatial density distribution of these particles can be calculated
analytically (see equation (\ref{nsngsol}) in Appendix A). Figure
\ref{fig_focus} shows this distribution as a grey-scale in the
ecliptic plane. In this simple model the density enhancement of large
particles due to gravitational focusing during the collection phase is
less than a factor of $2$.
\begin{figure}[htb]
\epsfbox{sd_nd.ps}
\caption{ \label{fig_focus} The normalized density distribution of
gravitationally deflected ($\beta=0.1$) interstellar particles in the ecliptic plane
represented by a grey-scale. The normalization is defined to set the
spatial density at infinity equal to $1$. Stardust's trajectory is
indicated as the white solid line (compare to figure 1).}
\end{figure}
The enhancement in spatial density downstream the Sun is caused by
gravitational focusing. Since the down\-stream vector of the
interstellar dust flux has an angle of $\approx 5^\circ$ with respect
to the ecliptic plane, the actual focus does not lie in the ecliptic,
but a peripheral spatial density enhancement is present in the
ecliptic plane. The main effect of the dynamics of large particles
will be a higher impact velocity and a deviation in impact angle with
respect to the reference particles.
Particles with large $\beta$-values ($\beta > 1.4$) have diameters of
$\approx 0.45\ {\rm \mu m}$ which correspond to the maximum
wavelength of the solar spectrum. As we will argue in the next
section, we do not expect these particles to be abundant in the Solar
System during the Stardust mission due to their interaction with the
solar wind magnetic field.
\subsection{Electromagnetic Effects\label{emeffects}}
In section \ref{straightlines} we have argued that electromagnetic
effects can be neglected for the determination of the impact direction
and velocity of interstellar particles. But we have to take into
account these effects when considering the abundance of interstellar
particles in the Solar System since long-term effects might reduce or
enhance their spatial density. It was argued by Levy and Jokipii
(1976) \cite{levy76} that classical (very small) interstellar
particles are removed from the Solar System by Lorentz force. The
Ulysses and Galileo measurements show \cite{landgraf96} a depletion of
small particles (but still one order of magnitude above the detection
threshold). Models of the electromagnetic interaction of the charged
interstellar particles with the solar wind magnetic field
\cite{gruen94,gustafson96a,grogan96} predict a periodic focusing and
defocusing of the particles to the solar equator plane with the
$22$-years solar cycle. A new magnetic cycle starts at solar maximum
and the mean deflection effect is strongest when the magnetic field is
in the most ordered configuration during the solar minimum. The last
solar maximum in 1991 started a defocusing cycle. We have calculated
the total flux of interstellar particles on the Ulysses detector as a
function of time to check if the defocusing causes the total flux to
drop. Figure \ref{fig_ulsflux} shows that the total flux (in the
heliocentric inertial frame) of interstellar particles was constant
after Ulysses' fly-by of Jupiter in February 1992 (when Ulysses left
the ecliptic plane).
\begin{figure}[htb]
\epsfbox{uls_flux.ps}
\caption{\label{fig_ulsflux} Total flux $f$ of interstellar
particles measured by Ulysses as a function of time. The error-bars are
due to the fraction of particles which could not be clearly identified as
interstellar dust.}
\end{figure}
The gap in the data between beginning and mid of 1995 is caused by the
ecliptic crossing in March 1995 when distinction between interstellar
and interplanetary particles was difficult. After mid of 1996 the
measured total flux of interstellar particles drops by a factor of
$3$. We interpret this phenomenon as the effect of electromagnetic
defocusing. Unfortunately this can not be confirmed by Galileo data,
because Galileo is inside the Jovian system since the end of 1995,
therefore, Jupiter dust dominates the data and interstellar dust
impacts can not be identified \cite{krueger98a}. We assume that the
period of reduced total flux will last for half a solar cycle, which
is $11$ years, i.e. from mid of 1996 to mid of 2007. This period of
time covers the duration of the Stardust mission. Following this
estimate, we predict from the Ulysses data that the total dust flux on
the Stardust experiments will be reduced by a factor of $3$ compared
to the Ulysses and Galileo measurements.
For Ulysses and Galileo a total flux of $f_{\rm U/G} = 1.5\cdot
10^{-4}$ ${\rm m}^{-2}\ {\rm s}^{-1}$ has been determined
\cite{baguhl95b}, so we predict $f_{\rm SD}=5\cdot 10^{-5}\ {\rm
m}^{-2}\ {\rm s}^{-1}$.
\section{Results}
\subsection{Cometary Dust Measurements}
In this section we apply the coma model developed in section
\ref{commod} to quantitatively predict the flux and the fluence of
cometary dust particles on the Stardust spacecraft during the P/Wild 2
encounter. The results we obtain for an axially symmetric model are
compared with the results we get if we use a radially symmetric
model. We can compare our results for Stardust at P/Wild 2 with the
fluence measured by the VEGA spacecraft inside the coma of P/Halley by
scaling the predicted fluence with the difference in brightness,
encounter distance, and phase angle.
\paragraph{Fluence as a function of mass}
We approximate the trajectory of Stardust as a straight line passing
the coma with a phase angle of $70^\circ$ with respect to the Sun
direction. Stardust's closest approach to the nucleus is above the
comet's day side. The fluence of dust particles of a given dust class
$i$ during the whole trajectory is equal to the column density along
the spacecraft's trajectory which is given by equation
(\ref{nproj}). As the fluence is inversely proportional to the
distance of closest approach of the spacecraft to the nucleus $\rho$,
our results can easily be scaled to other closest approach
distances. The numerical values of the fluence for a trajectory above
the sub-solar point is given in table \ref{cfltab}.
\paragraph {Flux and Fluence as a Function of Spacecraft Position}
The fluence as a function of time from closest approach is calculated
by the time integral, or, equivalently, by the integral over the path
$s$ along the spacecraft's trajectory, of the dust flux on the
spacecraft as given in equation (\ref{fsc}).
\begin{eqnarray}
F_{SC,i}(t) & = & \int_{-\infty}^{V_{SC}t} f_{SC}\;dt =
\int_{-\infty}^{s(t)}n_{d,i}(\vec r(s))\;ds
\end{eqnarray}
\begin{table}
\caption{\label{cfltab} Fluence of dust particles on the spacecraft
for the radial and axially symmetric model for a closest approach
distance of $\rho=100\;{\rm km}$. For different distances
$\rho^\prime$ of closest approach the fluence can be scaled by
$\rho/\rho^\prime$. In the axially symmetric model we assume that the
spacecraft's trajectory crosses the sub-solar point.}
\begin{tabular}{lllll}
\hline
Dust & particle & particle && \\
class & mass & radius &
\multicolumn{2}{c}{fluence $[{\rm m}^{-2}]$}
\\
$i$ & $m_{d,i}\ [{\rm kg}]$ & $s_{i}\ [{\rm m}]$ & radial-symm. &
axial-symm.\\
\hline
$1$&$
10^{-20}$&$ 1.34\cdot 10^{-8}$&$ 5.86\cdot 10^{+07}$&$
1.11\cdot 10^{+08}$\\ $2$&$ 10^{-19}$&$ 2.88\cdot 10^{-8}$&$
3.86\cdot 10^{+07}$&$ 7.05\cdot 10^{+07}$\\ $3$&$ 10^{-18}$&$
6.20\cdot 10^{-8}$&$ 2.52\cdot 10^{+07}$&$ 4.46\cdot 10^{+07}$\\ $4$&$
10^{-17}$&$ 1.34\cdot 10^{-7}$&$ 1.60\cdot 10^{+07}$&$
2.76\cdot 10^{+07}$\\ $5$&$ 10^{-16}$&$ 2.88\cdot 10^{-7}$&$
9.67\cdot 10^{+06}$&$ 1.64\cdot 10^{+07}$\\ $6$&$ 10^{-15}$&$
6.20\cdot 10^{-7}$&$ 5.39\cdot 10^{+06}$&$ 8.98\cdot 10^{+06}$\\ $7$&$
10^{-14}$&$ 1.34\cdot 10^{-6}$&$ 2.58\cdot 10^{+06}$&$
4.26\cdot 10^{+06}$\\ $8$&$ 10^{-13}$&$ 2.88\cdot 10^{-6}$&$
9.37\cdot 10^{+05}$&$ 1.53\cdot 10^{+06}$\\ $9$&$ 10^{-12}$&$
6.20\cdot 10^{-6}$&$ 2.32\cdot 10^{+05}$&$ 3.75\cdot 10^{+05}$\\ $10$&$
10^{-11}$&$ 1.34\cdot 10^{-5}$&$ 4.11\cdot 10^{+04}$&$
6.62\cdot 10^{+04}$\\ $11$&$ 10^{-10}$&$ 2.88\cdot 10^{-5}$&$
6.02\cdot 10^{+03}$&$ 9.64\cdot 10^{+03}$\\ $12$&$ 10^{-09}$&$
6.20\cdot 10^{-5}$&$ 8.11\cdot 10^{+02}$&$ 1.29\cdot 10^{+03}$\\ $13$&$
10^{-08}$&$ 1.34\cdot 10^{-4}$&$ 1.06\cdot 10^{+02}$&$
1.67\cdot 10^{+02}$\\ $14$&$ 10^{-07}$&$ 2.88\cdot 10^{-4}$&$
1.37\cdot 10^{+01}$&$ 2.14\cdot 10^{+01}$\\ $15$&$ 10^{-06}$&$
6.20\cdot 10^{-4}$&$ 1.78\cdot 10^{+00}$&$ 2.72\cdot 10^{+00}$\\ $16$&$
10^{-05}$&$ 1.34\cdot 10^{-3}$&$ 2.40\cdot 10^{-01}$&$
3.44\cdot 10^{-01}$\\ $17$&$ 10^{-04}$&$ 2.88\cdot 10^{-3}$&$
3.55\cdot 10^{-02}$&$ 4.34\cdot 10^{-02}$\\ $18$&$ 10^{-03}$&$
6.20\cdot 10^{-3}$&$ 6.31\cdot 10^{-03}$&$ 5.37\cdot 10^{-03}$\\ $19$&$
10^{-02}$&$ 1.34\cdot 10^{-2}$&$ 0.00\cdot 10^{+00}$&$
5.71\cdot 10^{-04}$\\ $20$&$ 10^{-01}$&$ 2.88\cdot 10^{-2}$&$
0.00\cdot 10^{+00}$&$ 1.15\cdot 10^{-05}$\\ \hline
\end{tabular}
\end{table}
For a radially symmetric model an analytical expression is given by:
\begin{eqnarray}\label{eqn_fsci_radsymm}
F_{SC,i} & = & \int_{-\infty}^{V_{SC}t} \frac{Z_{e,i}} {V_{e,i}
\rho^2+s^2}\;ds\nonumber \\
& = & \frac{Z_{e,i}} {V_{e,i}\rho} \cdot \left(\xi +
\frac{\pi} {2}\right),
\end{eqnarray}
where $\xi = \arctan(V_{SC}t / \rho)$ is the angle of the spacecraft
position to the point of closest approach. From equation
(\ref{eqn_fsci_radsymm}) we see that $F_{SC,i}$ is a linear function
of $\xi$ in the radially symmetric model.
We show the calculated total dust fluence on Stardust in figure
\ref{uflu}.
\begin{figure}[htb]
\epsfxsize=.5\hsize
\epsfbox{uflu.ps}
\caption{\label{uflu}
Total fluence on Stardust versus the angle $\xi$ from the location of
closest approach for the radially symmetric (dashed line) and axially
symmetric model (solid line). The total fluence calculated by the
axially symmetric model deviates significantly from the result of the
radially symmetric calculation.}
\end{figure}
\begin{figure}[htb]
\epsfxsize=.5\hsize
\epsfbox{uflx.ps}
\caption{\label{uflx}
Total dust flux versus the angle $\xi$ between Stardust's location
on the trajectory and the location of closest approach P/Wild 2. The
total flux of the axis and radially symmetric model are represented by
the solid and dashed lines, respectively. It can be seen that the
total flux predicted by the radially symmetric model is symmetric with
respect to closest approach and reaches its maximum at that
point.}
\end{figure}
In the axially symmetric model the total fluence is approximately a
linear function of $\xi$ for $-40^\circ \leq \xi \leq 40^\circ$ ,
like it would be for the radially symmetric model. The radially
symmetric model predicts a too low dust flux during closest approach,
because it does not take into account the enhanced activity on the day
side. Therefore, the slope of the total fluence is flatter between
$\xi=-60^\circ$ and $\xi=60^\circ$ in the radially symmetric
case. For $\xi > 60^\circ$ Stardust enters the region above the
terminator and consequently the total flux (and therefore the slope of
the total fluence) drops below the value predicted by the radially
symmetric model. We conclude that, for predicting the total dust
fluence on Stardust during closest approach, one must take into
account that Stardust approaches the comet from the day side and
passes the terminator shortly after closest approach. The different
results from the radially and axially symmmetric calculations show
that the total fluence is strongly affected by the activity
distribution on the surface of the nucleus. Unfortunately, we do not
have any information about the activity distribution on P/Wild 2, so
we have to rely on the axially symmetric coma model as a simple yet
complete physical model of the inner coma for which we can constrain
the parameters by measurements.
In figure \ref{uflx} we show the total particle flux versus
$\xi$. In the radially symmetric model the closest approach coincides
with the point of highest total dust flux. Using the axially symmetric model,
we find that the maximum total flux is reached $3\;{\rm s}$ before closest
approach and that the total flux stays nearly constant until after closest
approach. This is because the spacecraft enters into a region of lower
activity after it has passed the sub-solar point but the decrease
of cometary activity is apparently compensated by the still decreasing
distance from the nucleus.
\paragraph{Reproduction of the VEGA and Giotto measurements at P/Halley}
To validate our prediction of the fluence on the Stardust instruments,
we scale our model to the conditions at P/Halley during the fly by of
the Giotto and VEGA spacecraft. We stress that, although the VEGA 2
data were used to derive a mass distribution of our coma model, the
comparison of the fluence is in fact a check of the model. This is
because the VEGA 2 data were only used to derive the relative
abundance of particles in different dust classes, but we derive the
total amount of the dust activity from the value of
$Af\rho$ that has been determined from ground based observations.
The fluence on the Stardust and VEGA spacecraft is proportional to
the value of $Af\rho$ of the coma, inversely proportional to the
distance $\rho$ of closest approach, and the dust phase function
during the observation. Thus, the fluence which was computed for the
Stardust spacecraft can be scaled to the fluence measured on board the
VEGA spacecraft by multiplying the fluence with the following three
ratios:
\begin{equation}
\begin{array}{ccccc}
\multicolumn{5}{l}{\mbox{closest approach distance ratio:}} \\
\frac{\mbox{$\rho_{\mbox{\tiny Stardust}}$}} {\mbox{$\rho_{\mbox{\tiny
VEGA}}$}}
&\approx &\frac{100\;{\rm km}}{8000\;{\rm km}} &=& \frac{1} {80}
\\
\\
\multicolumn{5}{l}{\mbox{$Af\rho$-ratio:}}\\
\frac{\mbox{$Af\rho_{\mbox{\tiny Halley}}$}}
{\mbox{$Af\rho_{\mbox{\tiny Wild 2}}$}}& \approx &
\frac{219\;{\rm m}}{4.27\;{\rm m}} & \approx & 51.3
\\
\\
\multicolumn{5}{l}{\mbox{dust phase function ratio:}}\\
\frac{\mbox{$j(\alpha_{\mbox{\tiny Wild 2}}\approx 30^\circ)$}}
{\mbox{$j \left(\alpha_{\mbox{\tiny Halley}}
\approx 60^\circ\right)$}} &
\approx &\frac{0.04} {0.032}& = & 1.25
\end{array}
\end{equation}
The $Af\rho$ value of P/Halley during the VEGA encounter was taken
from Schleicher et al. (1998) \cite{afrhalley}. This leads to a total
scaling factor
of $51.3\cdot 1.25 / 80\approx 0.8$. Hence the fluence which we
predict for Stardust at P/Wild 2 is compatible with the fluence
measured on the VEGA spacecraft at P/Halley. In figure \ref{whcmp} we
show the data of the fluence measured by VEGA together with the our
prediction of the fluence on Stardust scaled by $0.8$.
\begin{figure}[htb]
\epsfxsize=.5\hsize
\epsfbox{whcmp.ps}
\caption{\label{whcmp}
Fluence of dust particles on the VEGA spacecraft. Filled and outlined
triangles represent VEGA 1 and VEGA 2 data, respectively. The fit to
the VEGA 2 data by Divine and Newburn (1987) \protect\cite{div2} is
shown as the dashed
line and the solid lines represent the scaled fluences as predicted by
the axially symmetric (upper solid curve) and radially symmetric model
(lower solid curve)}
\end{figure}
The good agreement of
the number of dust particles collected on the VEGA spacecraft with
the scaled Stardust fluence validates our procedure to derive the total dust
activity of the comet from the observations and it shows that the
parameter set we use in our model is consistent.
Unfortunately, because of the loss of the Giotto data near closest
approach due to large particle impacts, we can not perform any
quantitative comparison of the Giotto fluence to our model. However,
we can estimate the total mass fluence on Giotto on the basis of the
well known deceleration of $\Delta V=23\;{\rm cm}\;{\rm s}^{-1}$
\cite{eden} of Giotto during the encounter. As for the considered
mass distribution, the total fluence is dominated by large particles
and the momentum enhancement factor is believed to
be rather low for large particles \cite{eden},
the total mass fluence is estimated by assuming inelastic dust particle
impacts on Giotto: With the Giotto mass of $M_{\rm Giotto}=573.7\;{\rm
kg}$ and the velocity of $V_{\rm Giotto}=68.37\;{\rm km}\;{\rm
s}^{-1}$ during encounter, the total mass which hit Giotto can be
estimated to be $m_{\rm Giotto}=M_{\rm Giotto}/$ $V_{\rm Giotto}\cdot
\Delta V\approx 2\;{\rm g}$. Using the area of the
Giotto dust shield of $A=2.64\;{\rm m}^2$, we determine the total mass
fluence on Giotto to be $0.76\;{\rm g}\;{\rm m}^{-2}$. The total mass
fluence on the Stardust at P/Wild 2 can also be scaled to the total
mass fluence on Giotto at P/Halley by taking into account that the
closest approach distance of Giotto was approximatively $600\;{\rm
km}$. We find the values $0.038\;{\rm g}$ and $0.064\;{\rm g}$ for the
radially symmetric and axially symmetric model, respectively. Thus, the
extrapolation of the mass distribution which was fitted by Divine and
Newburn (1987) \cite{div2}
to the fluence on the VEGA 2 spacecraft to larger masses leads to an
underestimation of the total mass fluence by about one order of magnitude
when compared to the total mass fluence derived from the final impact
on Giotto. We discuss this discrepancy and its significance in section
\ref{discussion}.
\subsection{Interstellar Dust Measurements}
We predict the measurements of interstellar dust during the
interplanetary cruise phase by both main instruments on-board
Stardust, CIDA and the aerogel collector. Unlike in the case of the
measurements at P/Wild 2, the geometry of the measurements of both
instruments is different, because they are used at different parts of
the interplanetary trajectory. Therefore, we discuss the predictions
for both instruments sepa\-rate\-ly.
\subparagraph{CIDA}
To comprehend the predictions we make for CIDA one has to understand
the three attitude strategies which have been defined by the mission
plan \cite{stardust97}. Since CIDA has a fixed position on Stardust,
its pointing is determined by the spacecraft attitude. The overall
strategy should be to choose the attitude in a way that a maximum
number of interstellar particles are detected assuming that they all
behave like the reference particles. But the more one optimizes the
attitude, the more complicated the spacecraft operations will become
during the cruise phase. Thus, a trade-off has to be found between
maximum number of detected particles and operation complexity.
We now explain the three different attitude strategies defined in the
mission plan. Figure \ref{fig_cidaconfig} shows a simplified sketch of
Stardust containing the high gain antenna defining the $+z$-direction
, the solar panels, which lie in the $x$-$y$-plane and the CIDA field
of view (FOV) in the $x$-$z$-plane.
\begin{figure}[htb]
\epsfbox{cidaconfig.ps}
\caption{\label{fig_cidaconfig} Configuration of the CIDA field of
view (FOV) with respect to Stardust. The $+z$-direction is defined by
the pointing of the high gain antenna and the fixed solar panels,
which should always be pointed roughly towards the Sun.}
\end{figure}
In this plane the angle of FOV is $52^\circ$. Strategy \#1 fixes the $+x$-axis
to the upstream-direction of reference particles. This means that the
$+z$ axis, i.e. the solar panels and the high gain antenna, are not
always pointing to the Sun. The mission plan restricts this off-pointing to
$45^\circ$ due to power concerns. If more than $45^\circ$ off-pointing
is required to keep the $+x$-axis pointing upstream, the spacecraft
is to be turned to full Sun-pointing again, which means that the
upstream-direction is turned out of the FOV. Strategy \#2 is the most
complex one. If pointing of the $+x$-axis towards the upstream
direction is possible without more than $45^\circ$ off-pointing of the
$+z$-axis with respect to the Sun, strategy \#2 is identical to
strategy \#1. But if this configuration is not possible any more,
strategy \#2 keeps the $45^\circ$ off-pointing as long
as the upstream-direction lies inside the FOV. If even this is
impossible, the spacecraft should return to Sun-pointing again. The
most simple strategy is strategy \#3. It fixes Sun-pointing of the
$+z$-axis all the time, so CIDA will only collect interstellar particles
when the upstream direction of the flux is occasionally in the
FOV. Table \ref{tab_strategies} summarizes the three attitude
strategies.
\renewcommand{\arraystretch}{2}
\begin{table}
\caption{\label{tab_strategies} Attitude strategies for the
Stardust mission during the cruise phase as given by the mission plan.}
\begin{tabular}{ll}
\hline
strategy & description \\
\hline
\# 1 &
\begin{minipage}[t]{6cm}
$+x$-axis points upstream. Maximum Sun off-pointing is $45^\circ$.
\end{minipage}\\
\# 2 &
\begin{minipage}[t]{6cm}
upstream-direction is kept in the FOV. Maximum Sun off-pointing is
$45^\circ$.
\end{minipage}\\
\# 3 &
\begin{minipage}[t]{6cm}
$+z$-axis always points towards the Sun.
\end{minipage}\\
\hline
\end{tabular}
\end{table}
\renewcommand{\arraystretch}{1}
Since the reference particles have a constant velocity of $26\;{\rm
km}\;{\rm s}^{-1}$, the impact velocity of the reference particles as
a function of time is determined by the spacecraft velocity. The
impact velocity, the impact energy, and the quantity $m v_i^{3.5}$,
which is proportional to the impact charge for impact plasma detectors
\cite{gruen95a}, where $m$ is the mass and $v_i$ is the impact velocity
of the particle, are shown in figure
\ref{fig_imp_ref}.
We predict the impact rate $\nu$ of reference particles on the CIDA
target. We can estimate $\nu$ by the given effective sensitive target
area of $A_{\rm CIDA} = 80\ {\rm cm}^2\cdot \cos 40^\circ = 60\;{\rm
cm}^2$
\cite{jochenpers}, the flux $f_{\rm SD}=5\cdot 10^{-5}\;{\rm
m}^{-2}\;{\rm s}^{1}$ (see section
\ref{emeffects}) in the inertial heliocentric frame, and the enhancement due
to the upstream motion of Stardust. We get
\begin{eqnarray}
\nu_{\rm max} & = & A_{\rm CIDA} f_{\rm SD} \left(\frac{v_{\rm
rel,max}} {v_{\rm dust,ecl}} \right) \approx 7.0 \cdot 10^{-7}\ {\rm
s}^{-1},
\end{eqnarray}
where $v_{\rm rel,max}=60\ {\rm km}\ {\rm s}^{-1}$ is the maximum
relative velocity of Stardust to the stream of reference particles (see
above) and $v_{\rm dust,ecl}=26\ {\rm km}\ {\rm s}^{-1}$ is the
velocity of the reference particles in the inertial ecliptic
frame. Figure \ref{fig_rate} shows the predicted impact rate on the
CIDA target as a function of time for all three strategies explained
above.
The total number of reference particles that CIDA will detect is given by
the accumulated impact rate over time. We show the prediction in figure
\ref{fig_fluence_ref}.
\begin{figure*}[htb]
\epsfxsize=\hsize\epsfbox{sd_predict_imp.ps}
\caption{\label{fig_imp_ref} The values of $v$, $E=mv^2/2$, and $m
v^{3.5}$ as a function of time, where $m$ is the mass and $v$ the
velocity of the impacting reference particle. The shaded areas labeled
``Arogel'' indicate the phases of deployment of the Aerogel collector
and ``encounter Phases'' labels the phase of preparations of the comet
encounter. During these phases, CIDA is not taking data.}
\end{figure*}
\begin{figure*}[htb]
\epsfxsize=\hsize\epsfbox{sd_predict_rate.ps}
\caption{\label{fig_rate} Impact rate on the CIDA target as a
function of time. Prediction in panel (a) assumes strategy \#1, in
panel (b) strategy \#2, and in panel (c) strategy \#3. Mission phases
during which CIDA is not allowed to be turned on due to other
activities are indicated by the shaded areas (compare figure
\ref{fig_imp_ref}).}
\end{figure*}
\begin{figure*}[htb]
\epsfxsize=\hsize\epsfbox{sd_predict_fluence.ps}
\caption{\label{fig_fluence_ref} Total fluence on the CIDA target as a
function of time. Prediction in panel (a) assumes strategy \#1, in
panel (b) strategy \#2, and in panel (c) strategy \#3. Mission phases
during which CIDA is not allowed to be turned on due to other
activities are indicated by the shaded areas (compare figure
\ref{fig_imp_ref}).}
\end{figure*}
In summary, we predict that, assuming attitude strategy \#2, CIDA will
detect about $25$ particles which behave very much like reference
particles and hit the target with velocities between $30\ {\rm km}\
{\rm s}^{-1}$ and $60\ {\rm km}\ {\rm s}^{-1}$ and energies between
$18\;{\rm eV}\;{\rm N}^{-1}$ and $4\;{\rm eV}\;{\rm N}^{-1}$ (${\rm
eV}\;{\rm N}^{-1}$ = electronvolt per nucleon). For a summary of the
prediction see table \ref{tab_ip}, this table also contains the number
of interplanetary dust particles as determined from the
five-population model of interplanetary dust by Staubach et al. (1997)
\cite{staubach97} ``contaminating'' the measurement.
\begin{table}[ht]
\caption{\label{tab_ip} Number of interstellar and interplanetary
particles predicted to be measured by Stardust during the cruise. For
the collector we give the numbers for the front (which is exposed to
the comet) and the back side (which is exposed to the interstellar
upstream direction) seperately.}
\begin{tabular}{rr|rr}
\hline
& & \multicolumn{2}{c}{collector}\\
& CIDA\footnotemark[1]
& front & back \\
\hline
interstellar & $25$ & $0$ & $120$\\
interplanetary & $5$ & $25$ & $20$\\
\hline
\multicolumn{4}{l}{\footnotesize $^1$ Attitude strategy \#2, see table
\ref{tab_strategies}}
\end{tabular}
\end{table}
\clearpage
\subparagraph{The Dust Collector}
\begin{figure*}[htb]
\epsfxsize=\hsize\epsfbox{sd_predict_coll_fluence.ps}
\caption{\label{fig_coll_fluence} Total fluence on the aerogel detector
of reference particles (a) and gravitationally deflected particles (b) as a function of
time. Aerogel collection is not possible in the hatched periods.}
\end{figure*}
\begin{figure*}[htb]
\epsfxsize=\hsize\epsfbox{sd_predict_dev_angle.ps}
\caption{\label{fig_dev_angle} Deviation of the impact angle of
gravitationally deflected particles from the impact angle of reference particles, which
should be close to $0^\circ$ as a function of time. Aerogel collection
is not possible in the hatched regions.}
\end{figure*}
\begin{figure*}[htb]
\epsfxsize=\hsize\epsfbox{sd_predict_imp_grav.ps}
\caption{\label{fig_imp_grav} Values of $v$, $E=mv^2/2$, and $m
v^{3.5}$ as a function of time, where $m$ is the mass and $v$ the
velocity of the impacting gravitationally deflected particle (compare figure
\ref{fig_imp_ref}).}
\end{figure*}
For the dust collector larger particles are more interesting, because
their trajectories can more easily be measured and they can be
extracted from the aerogel. Since the collector is turned to maintain
a pointing to the upstream direction of the reference particles, we
expect a spread in impact direction due to dynamical effects. For
larger particles the dynamics should be dominated by gravity and we
assume $\beta=0.1$ (see section \ref{grav_grains}). The
Mie-calculations
\cite{gustafson94} give a radius of $1\ {\rm \mu m}$ for these particles,
which translates to a mass of $1\cdot 10^{-11}\ {\rm g}$ assuming a
density of $2.5\ {\rm g}\ {\rm cm}^{-3}$. In the Ulysses and Galileo
data particles heavier than $10^{-11}\ {\rm g}$ make up $\approx 10\%$ of
the total number \cite{landgraf96}, but they are not depleted due to
electro-magnetic effects (see section \ref{emeffects}). So we assume a
flux of $1.5\cdot 10^{-5}\ {\rm m}^{-2}\ {\rm s}^{-1}$ of these large
particles. We show the total fluence of interstellar particles on the
collector for both, reference and gravitationally deflected
particles, in figure \ref{fig_coll_fluence}.
Due to particle dynamics, tracks of larger particles should show a direction
distribution. We show the deviation of impact angle as a function of
time in figure \ref{fig_dev_angle}.
For the collector it is furthermore interesting how fast
gravitationally deflected
particles impact into the aerogel. Since the particles are accelerated
towards the Sun we expect their impact velocities to be higher than
the impact velocities of the reference particles. In figure
\ref{fig_imp_grav} we show the same plot as in figure
\ref{fig_imp_ref} for gravitationally deflected particles.
To summarize our prediction for the aerogel collector, we give the
total number particles collected during the $290$ days of exposure time
\cite{stardust97} to be $120$. $80$ of these particles are small, like
reference particles. The remaining $40$ are large particles, the impact
tracks of which have angular deviations between $10^\circ$ and
$30^\circ$. For the large particles we expect impact velocities on the
aerogel between $20\ {\rm km}\ {\rm s}^{-1}$ and $40\ {\rm km}\ {\rm
s}^{-1}$.
\clearpage
\section{Discussion\label{discussion}}
We think that the Stardust mission to comet P/Wild 2 will enhance our
understanding of the environment and the properties of comets by a
large amount. According to our prediction, Stardust is able to fulfill
its primary goal, the in situ measurement and collection of cometary
dust particles. Concerning the secondary goal, the in situ detection
and collection of interstellar dust particles from the local
interstellar cloud, we find that the total number of
particles we expect to be measured by Stardust is quite small
considering the seven year duration of the mission. But having
material from beyond the Solar System in a laboratory on the ground
would be a big achievement.
We have determined the fluence on the Stardust spacecraft using
a coma model. In our model the dust activity was determined using
observations of P/Wild 2 at $1.7\;{\rm AU}$ heliocentric distance. Our
model can reproduce the measurements of the VEGA spacecraft at
P/Halley when scaling the model to the geometry of this measurement. The
total mass fluence on Giotto, deduced from the fact that the spacecraft
was hit by a very large particle, is underestimated by our model by
one order of magnitude. Of course, a one-particle hit has not a very
high statistic significance and we can hypothesize that this one hit was
a low probability event. The possibility that the mass distribution
used in our model has a deficit of large dust particles, as suggested
by the DIDSY data \cite{fulle}, exists, but
simply adding large particles to the mass distribution would either
contradict ground based observations of the total brightness of P/Wild
2 during its in-bound part of its orbit at $1.7\;{\rm AU}$, or the
dust fluence measured by VEGA at P/Halley. The real
mass distribution of dust particles at the nucleus is still unknown
and the reader may choose a different mass distribution and put it into
our calculation to calculate the expected flux on Stardust.
Stardust approaches the comet from the sunlit side and therefore more
dust particle impacts are expected before than after
closest approach. Thus, a radially symmetric model is not
sufficient to calculate the dust fluence on CIDA and the aerogel
collector. As a result of the axially symmetric model, radial features
on the night side of the nucleus appear (see figure
\ref{dstiso}). Unfortunately, we can not prove or disprove the
existence of these features by the Stardust measurements, because
Stardust is passing by the day side of P/Wild 2. The prove or disprove
of the existence of such features remains a task for future ground
based observations and space missions.
On the basis of the Galileo and Ulys\-ses measurements we predict that
during the interplanetary cruise phase $25$ mainly small
($m\approx 10^{-12}\ {\rm g}$) particles will hit the CIDA target. As the
Ulysses data indicate, Stardust's timing with the solar magnetic cycle
is unfortunate (see section \ref{emeffects}). To confirm that the
decrease in the total flux of interstellar particles measured by
Ulysses is due to their interaction with the solar wind magnetic
field, modeling of this interaction is under way
\cite{landgraf98b}. Even if the total number of interstellar
particles detected by CIDA could be low, the scientific significance
of the data, in any case, is high because the composition of LIC dust
is to be directly measured for the first time by CIDA. The
reliability of this measurement depends very much on the ability to
identify interstellar impacts clearly. Interplanetary particles are a
possible source of ``contamination'' for CIDA's measurements. Table
\ref{tab_ip} shows that 5 interplanetary dust particles with masses
greater than $10^{-14}\;{\rm g}$ are expected to hit the CIDA target
during the cruise phase. For the determination of this number we have
used the five-population-model of interplanetary dust by
Staubach et al. (1997) \cite{staubach97}. So, interstellar particles
should dominate the
impacts on the CIDA target during cruise, but one has to make sure that an
individual detection is really from an interstellar particle.
We expect the aerogel collector to contain about $120$ interstellar
particles inside its interstellar collection layer. Of these, $40$
should be large particles (size $>1\ {\rm \mu m}$) that are probably
the only ones which are extractable from the aerogel. Due to their
high impact velocities, these large particles might be able to
penetrate very deep into the aerogel and are possibly altered by the
entry process. The distribution of impact angles can be determined by
measuring the geometry of the entry-tracks. This will give valuable
information on the dynamics of interstellar particles in the solar
system. The highest amount of information is of course contained in the
particles themselves when put into the laboratory and analyzed for
their chemical and mineralogical properties. Like CIDA, the collector
might also contain a contamination by interplanetary particles on both
sides, the interstellar and the cometary dust collection layer. As
shown in table \ref{tab_ip} we predict $25$ interplanetary particles
with masses greater than $10^{-13}\;{\rm g}$ to hit the cometary side
of the collector and $20$ to hit the interstellar side.
If the interplanetary particles can be identified as such, it might be
interesting to compare them with the IDPs collected by stratospheric
aircraft \cite{brownlee85,love94}.
|
\section{Discussion}
For completeness and for comparison we also present the numbers for
the non factorizable contributions of the lighter quarks and the
Z-contribution of the b-quark. The result for these
contributions are taken from \cite{light1}.
To slightly improve them, we performed a Pad\'{e} approximation for
the expansions in $x=1$ for $\Delta^Z$ and $x=m_Z^2/m_W^2$
for $\Delta^W$ (however this gives only minor changes of order
of few percents in formulae (\ref{DeltaZlight}) and (\ref{DeltaWlight}) ).
For $\Delta^Z$ we have
\begin{eqnarray}
\label{DeltaZlight}
\Delta^Z = \left\{
\begin{array}{ll}
-0.489 & \mbox{for u,c} \\
-0.796 & \mbox{for d,s,b}
\end{array}
\right.
\end{eqnarray}
Taking into account (\ref{DeltaWb}), we have for $\Delta^W$
\begin{eqnarray}
\label{DeltaWlight}
\Delta^W = \left\{
\begin{array}{ll}
-3.652 & \mbox{for u,c} \\
-3.745 & \mbox{for d,s} \\
+18.811 & \mbox{for b}
\end{array}
\right.
\end{eqnarray}
The above numbers demonstrate again that due to the heavy
top quark the $\Delta_b^W$ contribution is larger by almost an
order of magnitude than all the other contributions. The smallness of
the subleading terms in $1/m_t^2$ compared is
surprising, however. This is a highly nontrivial result. It is, e.g.,
in contrast to the results found in Ref. \cite{Degrassi}, where an
$O({\alpha}^2 m_t^2/m_W^2)$ calculation of $\Delta r$ was performed
and it was found that this correction is of the same order as the leading
$O({\alpha}^2 m_t^4/m_W^4)$ correction.
A detailed comparison of our results with those of Ref. \cite{Zbb1}
is given in \cite{Barselona}.
|
\section{Introduction}
It is well known that all spin-glass systems have complicated free energy
structures with many metastable states separated by large barriers \cite
{Fischer93}. The calculation of heights of these barriers is a difficult
task even for the long-range spin-glass models because most of the methods
developed in the spin glass theory do not give such information. For
example, the application of the standard replica method \cite{Fischer93}
allows to find the equilibrium free energy which does not contain any
information about the heights of the barriers. The standard dynamical
approach \cite{Dotsenko90,kucu} gives the properties of different
metastable states and their history dependence and might contain, in
principle, the information about the transition rates between them but in
the limit of long-range interaction the probability of transitions becomes
exponentially low in the number of spins, $N$, and, thus, such processes are
neglected by the mean field derivation of the equations for the correlation
and response functions on which this aproach is usually based. The
modification of the replica method that allows to estimate the energy
barriers between metastable states was suggested in Ref.\cite{cgp}. The main
idea of this modification is to study the free energy of the state
constrained to have a certain overlap with the given state. The main
drawback of it is that it is not clear whether the state corresponding to
the energy maximum found by this method is dynamically accessible and that
it is indeed a bottleneck of a transition process. The problem is
exacerbated by the fact that the replica method weights all states with the
Boltzmann weight so it might miss the rare saddle points of low energy in
favour of more abundant high energy ones. In this paper we develop an
alternative technique which is based on the modified dynamical approach \cite
{Ioffe98} for the calculation of the barriers between the metastable states.
Before we discuss how to modify the dynamical theory so it does not neglect
rare processes such as transitions over the barriers we briefly review the
standard dynamical approach to the spin-glasses. In this approach one starts
from the Lagrangian formulation of the Langevin dynamics, averages over the
disorder and, making the saddle point approximation, arrives at a closed
system of equations for the sigle site spin-spin correlation function $%
D(t_{1},t_{2})$ and response function $G(t_{1},t_{2})$. The corrections to
the saddle point approximation are small in $1/N$ because interaction has a
range $N$. For example, in the case of the spherical p-spin interacting
model one gets a set of integro-differential equations which were solved
numerically (and partially analytically); the solution is made possible by
the fact that these equations are \emph{forward propagating} in time which
is in turn due to the causality of the Langevin equations and initial
conditions imposed in the past \cite{kucu,pspin}. By construction these mean
field equations describe the most probable evolution of the system in time
and ignore all rare processes such as transitions over the barriers.
Empirically we visualize the processes that give the main contribution to
the conventional dynamical theory as motion down the energy landscape or
small fluctuations near the bottom of the valley.
Consider now a typical dynamical process that corresponds to the transition
between two close metastable states in a free energy landscape shown in
Fig.~1. The system initially is in the state 1 and we want to find the
probability of the transition to the state 2. The path from point 1 to point
2 consists of the uphill motion from state 1 to the unstable stationary
point 3 and the motion from unstable point 3 to stable state 2. Only the
first part of the motion corresponds to the rare process, and, therefore,
the probability of the transition between the states 1 and 2 is determined
by the uphill motion from 1 to 3. In the Lagrangian formulation of the
Langevin dynamics the uphill motion can be described as an instanton which
action gives the probability of the process. Note that to find this solution
one needs to ``force'' the system to go upward, i.e. one needs to apply a
boundary condition in future (at point 3 in our example) that destroys
causality of the theory. This complicates enormously the dynamical equations
describing the instanton motion compared to the dynamical equations
describing typical processes (such as motion from 3 to 2).
The main result of this paper is that an instanton motion can be
mapped into a usual motion going back in time. In our example it means that
the uphill motion from point 1 to point 3 can be mapped into the usual
downhill motion from point 3 to point 1. This allows one to solve the usual
\emph{forward propagating} equations instead of solving the complicated
equations describing the instanton motion. In general, this method does not
allow one to find the barrier beween one given state and another but it
allows one to find some instanton processes which, hopefully, correspond to
typical barriers in a system. In the simple example of a spherical
Sherrington-Kirkpatrick model which we consider here it gives the barrier
that separates doubly degenerate ground states that differ by the sign of
the magnetization; this barrier is physically important because it controls
the decay of the ground state magnetization in this system. In this model
there are only two locally stable states and $N$ locally unstable ones,
naturally one expects that these unstable states are saddle points and the
trajectory which connects two ground states must go through one of them but
these general qualitative arguments do not indicate which of these saddle
points should be used. Qualitatively the problem is that even in this simple
model the lowest saddle point might not connect two different minima but
connect one minima to itself (false pass). Our approach proves that it is
sufficient to climb up to the lowest of them in order to get from one ground
state to another.
\begin{figure}[h]
\unitlength1.0cm
\par
\begin{center}
\begin{picture}(6,2.4)
\epsfxsize=6.0 cm
\epsfysize=2.4 cm
\epsfbox{pic1.eps}
\end{picture}
\end{center}
\caption{Energy landscape illustrating a transition between states 1 and 2
via saddle point 3. Uphill motion from 1 to 3 is a rare process which is
missed by conventional mean field dynamical equations for the spin glasses,
the action corresponding to this process is large in $N$. Downhill motion
from 3 to 2 does not cost any action and is described by the conventional
dynamic equations, its action is zero.}
\end{figure}
Further, we confirm that in case when the free energy landscape of the
problem is known explicitly, the probability of the transition obtained as
an action of the instanton solution is $e^{-\Delta F/T}$ where $\Delta F$ is
the difference between the free energy at the end of the instanton
trajectory (point 3) and the free energy of the stationary state (point 1).
The free energy at the unstable fixed point 3 should be understood as $%
F=E-TS,$ where $E$ is the energy and $S$ is the entropy defined as a
logarithm of the configuration space restricted to the direction
perpendicular to the trajectory. Of course, there is no much need to calculate
the transition probability if the energy of the saddle point in known exactly and
it is established that the saddle point is dynamically accessible; the
advantage of the method is that it can be also used in the cases where energy
lanscape can not be found explicitly, e.g. in all problems in which the
disorder average was performed first.
The paper is organized as follows: In Section \ref{2} we prove that an
instanton equation of motion can be mapped into a usual equation of motion
reversed in time. In Section \ref{3} we consider an instanton transition in
the spherical SK model. Section \ref{4} summarizes our results.
\section{Instanton equations}
\label{2}
We start from the Langevin equation describing the overdamped relaxation of
the system with energy $\mathcal{H}$
\[
\Gamma _{0}^{-1}\partial _{t}\sigma _{i}=-{\frac{{\delta \beta \mathcal{H}}}{%
{\delta \sigma _{i}}}}+\xi _{i}.
\]
Here $\xi _{i}$ is the Langevin noise with the correlator
\begin{equation}
\langle \xi _{i}(t_{1})\xi _{j}(t_{2})\rangle =2\Gamma _{0}^{-1}\delta
_{i,j}\,\delta (t_{1}-t_{2})
\end{equation}
Below we shall choose the time units so that $\Gamma _{0}=1$. Using the
standard approach \cite{Fischer93} we get a path integral formulation of the
problem with the Lagrangian
\begin{equation}
\mathcal{L}=\sum_{i}-\hat{\sigma}_{i}^{2}-i\hat{\sigma}_{i}\left( \partial
_{t}\sigma _{i}+{\frac{{\delta \beta \mathcal{H}}}{{\delta \sigma _{i}}}}%
\right) +{\frac{1}{2}}{\frac{{\delta ^{2}\beta \mathcal{H}}}{{\delta \sigma
_{i}^{2}}}}. \label{lagr}
\end{equation}
The Green functions are defined by
\begin{eqnarray}
\mathcal{G}_{i,j}(t_{1},t_{2}) &\equiv &\left[
\begin{array}{cc}
\hat{D}_{i,j}(t_{1},t_{2}) & G_{i,j}^{\dagger }(t_{1},t_{2}) \\
G_{i,j}(t_{1},t_{2}) & D_{i,j}(t_{1},t_{2})
\end{array}
\right] \nonumber \\
&=&\int D\sigma D\hat{\sigma}\,\,e^{\int_{t_{i}}^{t_{f}}\mathcal{L}%
\,dt}\,\,\left[
\begin{array}{c}
i\hat{\sigma}_{i}(t_{1}) \\
\sigma _{i}(t_{1})
\end{array}
\right] [i\hat{\sigma}_{j}(t_{2}),\sigma _{j}(t_{2})],
\end{eqnarray}
and the dynamical action $A$ is defined by
\begin{equation}
e^{A}=\int D\sigma D\hat{\sigma}e^{\int_{t_{i}}^{t_{f}}\mathcal{L}\,dt}.
\label{prob}
\end{equation}
Usually one fixes only initial boundary conditions. In this case the Green
function G is casual, the anomalous Green function $\hat{D}$ is zero, and
the action vanishes. But if one considers rare processes as transitions over
the barriers, then one should also fix the final boundary conditions. In
that case the Green function G does not need to be causal, and the action
and the Green function $\hat{D}$ do not necessarily vanish.
Now we construct a mapping between an uphill motion (with a negative action
monotonically decreasing in time) and the downhill motion (with zero action)
going back in time. Consider arbitrary instanton process ($\sigma (t),%
\widehat{\sigma }(t)$) and impose both initial and final boundary
conditions. Applying the transformation
\begin{equation}
\lbrack i\hat{\sigma},\sigma ]\to [i\hat{\sigma}+\vec{\partial}_{t}\sigma
,\sigma ] \label{trsigma}
\end{equation}
to the Lagrangian (\ref{lagr}) we get
\begin{eqnarray}
\mathcal{L} &\to &\mathcal{L}_{n}=\sum_{i}-\hat{\sigma}_{i}^{2}-i\hat{\sigma}%
_{i}\left( -\partial _{t}\sigma _{i}+{\frac{{\delta \beta \mathcal{H}}}{{%
\delta \sigma _{i}}}}\right) +{\frac{1}{2}}{\frac{{\delta ^{2}\beta \mathcal{%
H}}}{{\delta \sigma _{i}^{2}}}} \nonumber \\
&&+\beta \left( H(t_{f})-H(t_{i})\right) \label{lagr1}
\end{eqnarray}
Note that this Lagrangian differs from the original Lagrangian (\ref{lagr})
only by the sign of time derivative and constant boundary term. Therefore
inverting time in the Lagrangian (\ref{lagr1}) one can make the Lagrangians (%
\ref{lagr},\ref{lagr1}) equivalent. Therefore, the Lagrangian (\ref{lagr1})
describes normal downhill motion formally inverted in time.
The Green function defined as averaged with respect to Lagrangian (3)
\begin{equation}
\mathcal{G}(t_{1},t_{2})=\left\langle \left[
\begin{array}{c}
i\hat{\sigma}(t_{1}) \\
\sigma (t_{1})
\end{array}
\right] [i\hat{\sigma}(t_{2}),\sigma (t_{2})]\right\rangle _{\mathcal{L}}
\end{equation}
and the Green function defined as averaged with respect to the Lagrangian (%
\ref{lagr1})
\begin{equation}
\mathcal{G}_{n}(t_{1},t_{2})=\left\langle \left[
\begin{array}{c}
i\hat{\sigma}(t_{1}) \\
\sigma (t_{1})
\end{array}
\right] [i\hat{\sigma}(t_{2}),\sigma (t_{2})]\right\rangle _{\mathcal{L}_{n}}
\end{equation}
are related by
\begin{equation}
\mathcal{G}(t_{1},t_{2})=\left[
\begin{array}{cc}
1 & \overrightarrow{\partial }_{t1} \\
0 & 1
\end{array}
\right] \mathcal{G}_{n}(t_{1},t_{2})\left[
\begin{array}{cc}
1 & 0 \\
\overleftarrow{\partial }_{t2} & 1
\end{array}
\right] . \label{transform}
\end{equation}
Because the Green function $\mathcal{G}_{n}$ corresponds to the normal
downhill process inverted in time, it should have the form
\begin{equation}
\mathcal{G}_{n}=\left[
\begin{array}{cc}
0 & G_{n}^{\dagger } \\
G_{n} & D_{n}
\end{array}
\right] .
\end{equation}
Therefore the transformation (\ref{transform}) in components is
\begin{equation}
D(t_1,t_2)=D_n(t_1,t_2) \label{trcom4}
\end{equation}
\begin{equation}
G(t_1,t_2)=G_n(t_1,t_2)+\partial_2 D_n(t_1,t_2) \label{trcom2}
\end{equation}
\begin{equation}
\hat D(t_1,t_2)=\partial_1 G_n(t_1,t_2)+\partial_2 G_n^\dagger(t_1,t_2)
+\partial_1\partial_2 D_n(t_1,t_2). \label{trcom1}
\end{equation}
Note that the response function $G_n$ should be purely advanced due to
formal inversion of time with respect to the downhill motion.
Thus, in order to construct the Green functions for the instanton processes
one should find the Green functions for the corresponding normal process,
invert time, and apply the transformation (\ref{transform}).
\begin{figure}[h]
\unitlength1.0cm
\par
\begin{center}
\begin{picture}(6,4)
\epsfxsize=6.0 cm
\epsfysize=4.0 cm
\epsfbox{pic2.eps}
\end{picture}
\end{center}
\caption{}
\end{figure}
Now we show that the action can be expressed through the energies and
configuration spaces of the initial and final states: Suppose that initially
the system is at the stable state $s$, and we want to find the probability
to escape this state going trough the unstable fixed point $u$ which is the
final state of the instanton motion. Note that in all the above
considerations we assumed completely fixed boundary conditions for $\sigma
_{i}.$ But physically, initial and final states correspond to some regions
in configuration states which we denote as $\Gamma _{s}$ and $\Gamma _{u}$
respectively. To emphasize the difference between the process with
completely fixed boundary conditions and the processes with physical
boundary conditions we refer to the former as elementary processes.
According to Eqs.(\ref{lagr},\ref{lagr1}) the probability of an elementary
process of motion from $s$ to $u$ ($w_{su}$) is related to the elementary
process of motion from $u$ to $s$ ($w_{us}$) through
\begin{equation}
w_{su}=w_{us}\,e^{\beta (E(t_{f})-E(t_{i}))}. \label{wrel}
\end{equation}
The probability to escape the state $s$ is
\begin{equation}
W_{su}=\Gamma _{u}w_{su}. \label{Wsu}
\end{equation}
On the other hand the probability to go from the unstable state $u$ to the
stable state $s$ is 1, therefore
\begin{equation}
W_{us}=w_{us}\Gamma _{s}=1. \label{Wus}
\end{equation}
Combining Eqs.(\ref{wrel},\ref{Wsu},\ref{Wus}) for the probability to escape
the stable state $u$ we get
\begin{equation}
W_{su}=\exp [-\beta (E(t_{f})-E(t_{i}))+S(t_{f})-S(t_{i})], \label{prob1}
\end{equation}
where $S=\ln \Gamma $ is the entropy. Note that at the stable point the
entropy $S(t_{i})$ is just the equilibrium entropy corresponding to this
state. The entropy of the final state $S(t_{f})$ can not be defined
thermodynamically because it corresponds to the configuration space of the
unstable state. Defining the free energy as $F=E-TS$ one can write ($\ref
{prob1}$) as
\begin{equation}
W_{su}=e^{-\beta F(t_{f})+\beta F(t_{i})}. \label{actiong}
\end{equation}
\section{Instanton transition in the spherical SK model.}
\label{3}
The Hamiltonian of the spherical SK model is
\begin{equation}
\mathcal{H}={\frac{1}{2}}\sum \sigma _{i}J_{ij}\sigma _{j},
\end{equation}
with the spherical constraint $\sum_{i}\sigma _{i}\sigma _{i}=N$ imposed.
The Hamiltonian becomes diagonal in the basis of eigenvectors of the matrix $%
J_{i,j}$
\begin{equation}
\mathcal{H}={\frac{1}{2}}\sum_{\mu }\epsilon _{\mu }\,s_{\mu }^{2},
\end{equation}
where
\begin{eqnarray}
\sum_{j}J_{i,j}\sigma _{j}^{\mu } &=&\epsilon _{\mu }\sigma _{i}^{\mu }, \\
\sigma _{i} &=&\sum_{\mu }s_{\mu }\,\sigma _{i}^{\mu }.
\end{eqnarray}
The Lagrangian corresponding to this model is
\begin{equation}
\mathcal{L}(s,\lambda )=-\sum_{\mu }\left[ \hat{s}_{\mu }^{2}+i\hat{s}_{\mu
}(\partial _{t}+\epsilon _{\mu }+\lambda )s_{\mu }\right] +{\frac{1}{2}}%
(N+2)\lambda ,
\end{equation}
where $\lambda $ is the time-dependent Lagrange multiplier field which
appears in the equation of motion due to the constraint and we take $\beta
=1 $ for convenience. The functional integral over the variables $s_{\mu },%
\widehat{s}_{\mu }$ should be performed with the weight that includes $%
\delta -$function that ensures the constraint. Using the integral
representation of this $\delta -$function
\begin{equation}
\delta (\sum_{\mu }s_{\mu }^{2}-N)=\int D\phi \,e^{i\int dt\,\,\phi
\,\,(\sum_{\mu }s_{\mu }^{2}-N)},
\end{equation}
we get the following Lagrangian
\begin{eqnarray}
\mathcal{L}(s,\hat{s},\lambda ,\phi ) &=&\sum_{\mu }\left( -\hat{s}_{\mu
}^{2}-i\hat{s}_{\mu }(\partial _{t}+\epsilon _{\mu }+\lambda )s_{\mu }+i\phi
\,\,s_{\mu }^{2}\right) \nonumber \\
&&+{\frac{1}{2}}(N+2)\lambda -i\phi N.
\end{eqnarray}
At low temperatures the condensation into the lowest eigenvalue $\mu =0$
eventually takes place. Therefore we introduce the condensate $S_{0}$ and
integrate over $s_{\mu }$ with $\mu \ge 1$ getting
\[
\mathcal{L}=-\hat{S}_{0}^{2}-i\hat{S}_{0}(\partial +\epsilon _{0}+\lambda
)S_{0}+i\phi \,\,S_{0}^{2}+{\frac{1}{2}}(N+2)\lambda -i\phi N
\]
\begin{equation}
-{\frac{1}{2}}\sum_{\mu \ge 1}Tr\ln \mathcal{G}_{\mu }, \label{splag}
\end{equation}
where the matrix Green function
\begin{equation}
\mathcal{G}_{\mu }=\left[
\begin{array}{cc}
\hat{D}_{\mu }(t_{1},t_{2}) & G_{\mu }^{\dagger }(t_{1},t_{2}) \\
G_{\mu }(t_{1},t_{2}) & D_{\mu }(t_{1},t_{2})
\end{array}
\right]
\end{equation}
satisfies the equation
\begin{equation}
\left[
\begin{array}{cc}
-2 & \partial_{t1}+\lambda(t_1)+\epsilon_\mu \\
-\partial_{t1}+\lambda(t_1)+\epsilon_\mu & \phi(t_1)
\end{array}
\right] \mathcal{G}_\mu=\delta(t_1-t_2). \label{spgrf}
\end{equation}
The number of cites $N$ is large, therefore we will perform the integrals
with the weight $\exp (\int \mathcal{L}dt)$ where $\mathcal{L}$ is given by (%
\ref{splag}) in the saddle point approximation. Taking the variation with
respect to $\phi ,\lambda ,\hat{S}_{0},S_{0}$ and transforming the variables
via $-2i\phi \to \phi ,\,\,\,i\hat{S}\to \sqrt{N}\hat{S}$ we get the saddle
point equations
\begin{eqnarray}
{\frac{1}{N}}\sum_{\mu \ge 1}D_{\mu }+S_{0}^{2} &=&1, \label{sadlpf1} \\
{\frac{1}{N}}\sum_{\mu \ge 1}G_{\mu }+\hat{S}_{0}S_{0} &=&0, \label{sadlpf2}
\\
(\partial +\epsilon _{0}+\lambda )S_{0}-2\hat{S}_{0} &=&0, \label{sadlpf3}
\\
(-\partial +\epsilon _{0}+\lambda )\hat{S}_{0}+\phi S_{0} &=&0,
\label{sadlpf4}
\end{eqnarray}
where
\[
D_{\mu }(t)=D_{\mu }(t,t),\,\,\,\,G_{\mu }(t)=G_{\mu }(t,t+\delta ).
\]
Note that Eqs.(\ref{sadlpf1}-\ref{sadlpf4}) contain only the equal time
Green functions. Therefore it is convenient to write the equations directly
for the equal time Green functions instead of (\ref{spgrf}):
\begin{eqnarray}
\partial G_{\mu } &=&2\hat{D}_{\mu }+\phi D_{\mu }, \label{sadlpf5} \\
\partial \hat{D}_{\mu } &=&2(\lambda +\epsilon _{\mu })\hat{D}_{\mu }+\phi
\,(1+2G_{\mu }), \label{sadlpf6} \\
\partial D_{\mu } &=&-2(\lambda +\epsilon _{\mu })D_{\mu }+2(1+2G_{\mu }),
\label{sadlpf7}
\end{eqnarray}
We assume that the system is initially at the equilibrium. This corresponds
to the stationary solution of Eqs.(\ref{sadlpf1}-\ref{sadlpf7})
\begin{eqnarray}
{\frac{1}{N}}\sum_{\mu \ge 1}D_{\mu }+S_{0}^{2} =1, \label{stp1} \\
D_{\mu } ={\frac{1}{{\lambda _{s}+\epsilon _{\mu }}}}, \label{stp2} \\
\lambda _{s} =-\epsilon _{0}, \label{stp3} \\
G_{\mu } =\hat{D}_{\mu }=\hat{S}_{0}=\phi =0,
\end{eqnarray}
where $\lambda _{s}$ is the value of $\lambda $ for this stable stationary
solution. Eq.(\ref{stp3}) follows from Eq.(\ref{sadlpf3}) because we assumed
that there is a nonzero condensate density $S_{0}.$ Note that there are two
equilibrium states with $S_{0}>0$ and $S_{0}<0.$
Our goal is to find the probability of the instanton transition from one
state to the other. For definiteness, assume that the initial state is one
with $S_{0}>0.$ Obviously, $S_{0}$ first decreases to zero during the
instanton process and then it becomes negative. Only the first uphill part
of the motion gives the contribution to the action, therefore we will
consider only this uphill part of the trajectory. The end point of this
uphill motion corresponds to an unstable fixed point of Eqs.(\ref{sadlpf1}-%
\ref{sadlpf7}). The condensate density $S_{0}$ is zero at this point
therefore from Eqs.(\ref{sadlpf1}-\ref{sadlpf7}) we get the following
equations corresponding to this stationary (but unstable) solution:
\begin{eqnarray}
{\frac{1}{N}}\sum_{\mu \ge 1}D_{\mu } &=&1, \\
D_{\mu } &=&{\frac{1}{{\lambda _{u}+\epsilon _{\mu }}}}, \\
G_{\mu } &=&\hat{D}_{\mu }=\hat{S}_{0}=\phi =0,
\end{eqnarray}
where $\lambda _{u}$ is the value of $\lambda $ at this unstable stationary
solution. Physically, it is natural to expect that that this solution
corresponds to the condensate in the first eigenstate $\mu =1.$ Indeed,
taking $\lambda _{u}=-\epsilon _{1}+{\frac{1}{{S_{1}^{2}}}},$ where $S_{1}$
is the condensate at the first eigenstate $\mu =1,$ we get
\begin{eqnarray}
{\frac{1}{N}}\sum_{\mu \ge 2}D_{\mu }+S_{1}^{2} &=&1, \\
D_{\mu } &=&{\frac{1}{{\lambda _{u}+\epsilon _{\mu }}}}.
\end{eqnarray}
These equations are similar to Eqs.(\ref{stp1}-\ref{stp3}) for the stable
fixed point with the difference that the system condenses into the first
eigenstate.
Now we need to find the trajectory connecting the fixed points mentioned
above. It was shown in the previous section that an uphill trajectory can be
mapped into a downhill trajectory going back in time. To show this,
according to Eqs.(\ref{trsigma},\ref{trcom2}), one should take
\begin{eqnarray}
G_{\mu } &=&{\frac{1}{2}}\partial D_{\mu }, \label{sptr1} \\
\hat{S}_{0} &=&\partial S_{0}, \label{sptr2}
\end{eqnarray}
where $D_{\mu },S_{0}$ should satisfy the downhill equations with inverse
time
\begin{eqnarray}
{\frac{1}{N}}\sum_{\mu \ge 1}D_{\mu }+S_{0}^{2} &=&1, \label{spdown1} \\
(-\partial +\epsilon _{0}+\lambda )S_{0} &=&0, \label{spdown2} \\
-\partial D_{\mu } &=&-2(\lambda +\epsilon _{\mu })D_{\mu }+2
\label{spdown3}
\end{eqnarray}
Indeed, taking
\begin{eqnarray}
\phi &=&\partial \lambda , \\
\hat{D}_{\mu } &=&{\frac{1}{4}}\partial ^{2}D_{\mu }-{\frac{1}{2}}D_{\mu
}\partial \lambda ,
\end{eqnarray}
along with Eqs.(\ref{sptr1},\ref{sptr2}), one can show that Eqs.(\ref
{sadlpf1}-\ref{sadlpf7}) are reduced to Eqs.(\ref{spdown1}-\ref{spdown3}).
The trajectory connecting the stable and unstable saddle points can be found
analytically; (see Appendix A) the result is
\begin{eqnarray}
\lambda (t)+\epsilon _{0} &=&(\lambda _{u}+\epsilon _{0})f[2\,|\lambda
_{u}+\epsilon _{0}|\,(t_{0}-t)] \\
D_{\mu }(t) &=&{\frac{1}{{\lambda _{u}+\epsilon _{\mu }}}}f[2\,|\lambda
_{u}+\epsilon _{0}|\,(t_{0}-t)] \nonumber \\
&&+{\frac{1}{{\epsilon _{\mu }}}}f[2|\lambda _{u}+\epsilon _{0}|(t-t_{0})],
\end{eqnarray}
where
\begin{equation}
f[x]={\frac{1}{{e^{x}+1}}},
\end{equation}
where $t_{0}$ is an arbitrary finite time which reflects the translational
invariance in time.
The last step is to calculate the action which determines the probability of
the instanton process. Note, that to find it, one should take $Tr\ln $ of
the operator containing $\lambda ,\phi $ which are functions of time. The
necessary calculation is presented in Appendix B, and the answer, simplified
with the help of Eqs.(\ref{sadlpf1}-\ref{sadlpf7}), is
\begin{equation}
A/N={\frac{1}{2}}\int dt\left( \phi +{\frac{2}{N}}\sum_{\mu \ge 1}{\frac{{%
G_{\mu }}}{{D_{\mu }}}}\right) .
\end{equation}
Using that $\phi =\partial \lambda $ and $G_{\mu }={\frac{1}{2}}\partial
D_{\mu }$ we get
\begin{equation}
A/N={\frac{1}{2}}\left( \lambda +{\frac{1}{N}}\sum_{\mu \ge 1}\ln D_{\mu
}\right) _{t=t_{i}}^{t=t_{f}}.
\end{equation}
Note that at the fixed points $D_{\mu }=1/(\epsilon _{\mu }+\lambda ),$
therefore one can write
\begin{equation}
A=-\Bigl[F(\lambda (t_{f}))-F(\lambda (t_{i}))\Bigr], \label{delf}
\end{equation}
where $F(\lambda )$ is the equilibrium free energy of this model
\begin{equation}
F(\lambda )/N=-\lambda /2+{\frac{1}{{2N}}}\sum_{\mu \ge 1}\ln (\epsilon
_{\mu }+\lambda ). \label{fener}
\end{equation}
The result (\ref{delf}) is in the agreement with the general result (\ref
{actiong}). As we mentioned in Sec.\ref{2} the free energy at the unstable
fixed point cannot be defined thermodynamically. But in this simple model
one can formally eliminate the unstable direction (i.e. impose the
constraint $S_0 =0$) and then it becomes possible to define the free energy
thermodynamically. That is why we got the difference of the equilibrium free
energies in (\ref{fener}). Note that the Lagrange multiplier $\lambda $
changes during the instanton process on $\lambda (t_{f})-\lambda
(t_{i})=\epsilon _{1}-\epsilon _{0}.$ The typical distance between
neighboring energy levels at the edges of energy level distribution is of
order of $1/\sqrt{N}.$ Therefore one can expand (\ref{fener}) in $\Delta
\lambda $ getting
\begin{equation}
A=-{\frac{N}{2T}}S_{0}^{2}(\epsilon _{1}-\epsilon _{0}),
\end{equation}
where we restored the temperature $T$. Note that this action is of the order
of $\sqrt{N}.$
\section{Discussion and conclusions.}
\label{4}
We developed the method that simplifies the calculation of the probablity of
rare processes such as transitions over the barriers. Our method is based on
the Lagrangian approach to the dynamics in which rare processes correspond
to the instantons. Generally, in order to obtain the probability of a
particular transition between two given states one need to apply a boundary
condition in future and in the past, this destroys the causality of the
theory: the response function $G=\langle \hat{s}s\rangle $ becomes non
causal and the anomalous correlation function $\hat{D}=\langle \hat{s}\hat{s}%
\rangle $ appears. This complicates the description of the instanton
processes.
The main result of this paper is that an instanton process can be mapped
into a usual process going back in time. So, knowing the correlation
functions of the corresponding normal process one can construct the
instanton correlation functions. We showed that this mapping gives the
sensible probability to escape a free energy well, $e^{-\Delta F/T}$, where $%
\Delta F$ is the depth of the free energy well. The free energy at the end
of the instanton trajectory cannot be defined thermodynamically because at
this point the system is at the unstable equilibrium, instead it should be
defined by $F=E-TS$, where $E$ is the energy and $S$ is the statistical
entropy at the end of the instanton trajectory, i.e. entropy constrained to
the states orthogonal to the descending direction.
We applied this approach to the spherical SK model which usual dynamical
properties were studied in Ref.\cite{cug}. This model has just two ground
states, corresponding to the energy level $\epsilon _{0}$, and no other
metastable states. Although the relaxation towards each of these two states
is exponential, model exhibits aging behavior when the system relaxes from a
random spin configuration to the equilibrium. We considered the instanton
transition from one ground state to the other. In accordance with our
general result the equations describing the instanton process in this model
can be transformed into the usual equations with inverted time. This
transformation allows one to find analytically the instanton trajectory. The
probability of this transition was found to be $e^{-S_{0}^{2}N(\epsilon
_{1}-\epsilon _{0})/2T}$ where $\epsilon _{1}$ is the energy of the first
(unstable) level. It shows that although the system has $N$ saddle points
and a complicated phase space the path connecting two ground states with
opposite magnetization might go via the saddle point with the lowest energy.
The typical distance between the neighboring energy levels at the edge of
the energy spectrum is of order $1/\sqrt{N},$ therefore the action is of
order $\sqrt{N}.$ Note that the distance between the energy levels $\epsilon
_{1}-\epsilon _{0}$ is different for different samples, therefore the
transition probability is not a self-averaging quantity. Therefore in this
problem it would be very difficult to get the correct answer for the
transition probability in any technique which involves averaging at the
beginning of calculations.
We hope that this method can be used to find the barriers between the
metastable states in more complicated spin glasses like $p>2$ spin models or
SK model. The first attempt of application of the instanton method to SK
model was done in Ref.\cite{Ioffe98} It is a more complicated problem because
there are many metastable states in these glasses and therefore the
averaging should be done at the beginning of calculations. The scaling of
the action $\sqrt{N}$ which we got for $p=2$ model is probably specific for
the spherical model because the barriers in spin glasses with exponential
number of states are due to the nonlinearity of the dynamical equations
which is absent in $p=2$ spherical model.
\section*{Appendix A}
In this Appendix we will find the trajectory connecting the unstable
fixed point with the stable one. It is natural to invert time in Eqs.(\ref
{spdown1}-\ref{spdown3}) so that they will describe the usual downhill
motion:
\begin{eqnarray}
{\frac{1}{N}}\sum_{\mu }\tilde{D}_{\mu }+\tilde{S}_{0}^{2} &=&1,
\label{downmot1} \\
(\partial +\epsilon _{0}+\tilde{\lambda})\tilde{S}_{0} &=&0,
\label{downmot2} \\
\partial \tilde{D}_{\mu } &=&-2(\tilde{\lambda}+\epsilon _{\mu })\tilde{D}%
_{\mu }+2, \label{downmot3}
\end{eqnarray}
where tilde means invertion of time with respect to the instanton motion,
for example $\tilde{S}_{0}(t)=S_{0}(-t).$ We need to find the solution of
these equatiosn corresponding to the downhill trajectory that begins from
the unstable fixed point and ends at the stable one. Therefore the initial
boundary condition is ($t=-\infty $)
\begin{eqnarray}
{\tilde D}_{\mu }(-\infty ) &=&{\frac{1}{{\epsilon _{\mu }+\lambda_{u}%
}}},\,\,\,\,\mu \ge 1, \label{bcon} \\
S_{0} &=&0, \\
\tilde{\lambda}(-\infty ) &=&-\epsilon _{1}+{\frac{1}{{\ S_{1}^{2}}}}, \\
\tilde{\lambda}(-\infty ) &=&\lambda _{u},
\end{eqnarray}
and the final one ($t=\infty $) is
\begin{eqnarray}
\tilde{D}_{\mu }(\infty ) &=&{\frac{1}{{\epsilon _{\mu }-\epsilon _{0}}}}%
,\,\,\,\,\mu \ge 1, \\
\tilde{\lambda}(\infty ) &=&-\epsilon _{0}, \\
S_{0} &\ne &0.
\end{eqnarray}
The solution of the Eq.(\ref{downmot3}) satisfying to the boundary condition
(\ref{bcon}) is
\begin{equation}
\tilde{D}_{\mu }(t)=2\int_{-\infty }^{t}dt^{\prime }e^{-2\int_{t^{\prime
}}^{t}[\tilde{\lambda}(t^{\prime \prime })+\epsilon _{\mu }]dt^{\prime
\prime }} \label{atd}
\end{equation}
Using Eqs.(\ref{downmot2},\ref{downmot3}) one can write Eq.(\ref{downmot1})
in the form
\begin{equation}
\tilde{\lambda}+\epsilon _{0}=-{\frac{1}{N}}\sum_{\mu }(\epsilon _{\mu
}-\epsilon _{0})D_{\mu }+1, \label{alam}
\end{equation}
which will be more convenient for us. For simplisity let us take $\epsilon
_{0}=0$ further in this Appendix. Substitution of Eq.(\ref{atd}) into Eq.(%
\ref{alam}) gives
\begin{equation}
\tilde{\lambda}(t)={\frac{2}{N}}\sum_{\mu }\int_{-\infty }^{t}dt^{\prime
}e^{-2\int_{t^{\prime }}^{t}[\tilde{\lambda}(t^{\prime \prime })+\epsilon
_{\mu }]dt^{\prime \prime }}\tilde{\lambda}(t^{\prime }).
\end{equation}
Introducing the function
\begin{equation}
F(t)=\Gamma (t)\tilde{\lambda}(t), \label{agamma}
\end{equation}
where $\Gamma (t)$ is defined by (up to the multiplication by a constant)
\begin{equation}
{\frac{{\Gamma (t)}}{{\Gamma (t^{\prime })}}}=e^{2\int_{t^{\prime }}^{t}%
\tilde{\lambda}(t^{\prime \prime })dt^{\prime \prime }},
\end{equation}
one can write Eq.(\ref{alam}) as a linear integral equation on $F(t)$
\begin{equation}
F(t)={\frac{2}{N}}\int_{-\infty }^{t}dt^{\prime }F(t^{\prime })\sum_{\mu
}e^{-2\epsilon _{\mu }(t-t^{\prime })},
\end{equation}
which can be easily solved giving
\begin{equation}
F(t)=A\,\,e^{2\lambda _{u}t},
\end{equation}
where $A$ is an arbitrary constant. Now using Eq.(\ref{agamma}) one can
write
\begin{equation}
e^{2\lambda _{u}(t-t^{\prime })}={\frac{{\lambda (t)}}{{\lambda (t^{\prime })%
}}}e^{2\int_{t^{\prime }}^{t}\tilde{\lambda}(t^{\prime \prime })dt^{\prime
\prime }},
\end{equation}
then taking a logarithm and differentiating with respect to $t$ we get
a differential equation on $\tilde{\lambda}$
\begin{equation}
\partial_t \tilde{\lambda}(t)=2\tilde{\lambda}(t)[\tilde{\lambda}_{u}-\tilde{%
\lambda}(t)],
\end{equation}
which can be solved giving
\begin{equation}
\tilde{\lambda}(t)=\tilde{\lambda}_{u}f[2\,|\tilde{\lambda}_{u}|\,(t-t_{0})],
\label{sollam}
\end{equation}
where
\begin{equation}
f[x]={\frac{1}{{e^{x}+1}}},
\end{equation}
and $t_{0}$ is an arbitrary time which reflects the translational invariance
in time. In case $\epsilon _{0}\ne 0$ one can easily generalize Eq.(\ref
{sollam}) to
\begin{equation}
\tilde{\lambda}(t)+\epsilon _{0}=(\tilde{\lambda}_{u}+\epsilon _{0})f[2\,|%
\tilde{\lambda}_{u}+\epsilon _{0}|\,(t-t_{0})].
\end{equation}
Knowing $\tilde{\lambda}(t)$ one can find $\tilde{D}_{\mu }$
\begin{eqnarray}
\tilde{D}_{\mu }(t) &=&{\frac{1}{{\lambda _{u}+\epsilon _{\mu }}}}%
f[2\,|\lambda _{u}+\epsilon _{0}|\,(t-t_{0})] \nonumber \\
&&+{\frac{1}{{\epsilon _{\mu }}}}f[2\,|\lambda _{u}+\epsilon
_{0}|\,(t_{0}-t)].
\end{eqnarray}
\section*{Appendix B}
The main problem in calculation of the action is to find
\begin{equation}
A_\mu\equiv{\frac{1}{2}}Tr \ln \left[
\begin{array}{cc}
-2 & \partial+\lambda+\epsilon_\mu \\
-\partial+\lambda+\epsilon_\mu & \phi
\end{array}
\right]. \label{appb1}
\end{equation}
Note that taking the variational derivative of (\ref{appb1}) with respect to
$\lambda(t)$ and $\phi(t)$ we get respectively $G_\mu(t,t)=1/2+G_\mu(t)$ and
${\frac{1}{2}} D_\mu (t).$ The idea of our method of calculation of (\ref
{appb1}) is to find a functional which gives the same functions ($G_\mu+{%
\frac{1}{2}}$ and ${\frac{1}{2}} D_\mu$) when one takes the variations with
respect to $\lambda$ and $\phi.$ Up to boundary terms this functional should
be equal to the action, and these boundary terms can be found from the
requirement that the action should be zero for any downhill trajectory.
Now let us find the functional mentioned above: Eqs.(\ref{sadlpf5}-\ref
{sadlpf7}) have the following invariant
\begin{equation}
D_\mu\hat D_\mu-(1+G_\mu)G_\mu=c, \label{appb2}
\end{equation}
where $c$ is an arbitrary constant. But initially $G_\mu=\hat D_\mu=0,$
therefore $c=0.$ The condition (\ref{appb2}) can be satisfied automatically
by introducing the new variables
\begin{equation}
D_\mu=\eta_\mu\eta_\mu, \,\,\,\, G_\mu=\eta_\mu\hat \eta_\mu-1/2, \,\,\,\,
\hat D_\mu=\hat \eta_\mu\hat \eta_\mu-{\frac{1}{{4 \eta_\mu^2}}}.
\end{equation}
The new variables $\eta_\mu,\hat\eta_\mu$ satisfy the following equations
\begin{eqnarray}
\partial\eta_\mu=-(\lambda+\epsilon_\mu)\eta_\mu+2\hat\eta_\mu,
\label{appbe1} \\
\partial\hat\eta_\mu=(\lambda+\epsilon_\mu)\hat\eta_\mu+\phi\,\eta_\mu -{%
\frac{1}{{2 \eta_\mu^3}}} . \label{appbe2}
\end{eqnarray}
These equations can be obtained by taking the variations of the functional
\begin{equation}
\Gamma_\mu=-\left[\hat\eta_\mu^2-\hat\eta_\mu(\partial+\epsilon_\mu+\lambda)%
\eta_\mu -{\frac{1}{2}}\phi\,\eta^2_\mu-{\frac{1}{{4\eta_\mu^2}}}\right]
\end{equation}
with respect to $\hat\eta$ and $\eta.$ Note that the variational derivatives
of (\ref{appb1}) and $\Gamma_\mu$ with respect to $\lambda,\phi$ are the
same if we take $\Gamma_\mu$ at the saddle point with respect to
$\hat\eta,\eta,$ i.e. Eqs.(\ref{appbe1},\ref{appbe2}) should be
satisfied. Therefore
the action $A_\mu$ can be written as
\begin{equation}
A_\mu=\Gamma_\mu+T\left.\Bigl(\hat
D_\mu(t),D_\mu(t),G_\mu(t),\lambda(t),\phi(t)\Bigr) \right|_{t=t_i}^{t=t_f},
\label{appb3}
\end{equation}
where $T$ is an unknown function. The initial and final times $t_i$ and $t_f
$ correspond to the fixed points of Eqs.(\ref{sadlpf1}-\ref{sadlpf7}),
therefore the abnormal functions $\hat D_\mu,G_\mu,\phi$ should be zero at
these points and we can simplify (\ref{appb3})
\begin{equation}
A_\mu=\Gamma_\mu+T\left.\Bigl(D_\mu(t),\lambda(t)\Bigr)
\right|_{t=t_i}^{t=t_f}.
\end{equation}
Using the saddle point equations (\ref{sadlpf1}-\ref{sadlpf7}) and Eqs.(\ref
{appbe1},\ref{appbe2}) one can simplify the total action $A$ getting
\begin{eqnarray}
A/N={\frac{1}{2}}\int dt\left(\phi+{\frac{1}{N}}\sum_\mu{\frac{{2G_\mu}}{{%
D_\mu}}} -{\frac{1}{{2N}}}\sum_\mu\partial \ln D_\mu\right) \nonumber \\
+\sum_\mu T\left.\Bigl(D_\mu(t),\lambda(t)\Bigr) \right|_{t=t_i}^{t=t_f}.
\label{appb5}
\end{eqnarray}
But this action should be zero for any downhill trajectory, therefore the
third and forth terms in (\ref{appb5}) should cancel each other, and we
finally get
\begin{equation}
A/N={\frac{1}{2}}\int dt\left(\phi+{\frac{1}{N}}\sum_\mu{\frac{{2G_\mu}}{{%
D_\mu}}}\right).
\end{equation}
|
\section{Introduction} \label{sec_introd}
During more than 25 years, the origin of gamma ray bursts (GRBs) has
been, perhaps, the deepest and most persistent problem in
astrophysics. However, with the advent of the Compton Gamma Ray
Observatory (CGRO) and its Burst and Transient Source Experiment
(BATSE) in 1991, a new phase in the research of GRBs started. In seven
years of operation, BATSE has accumulated a database of more than 2000
observations. The angular distribution of these bursts is isotropic
within the statistical limits, and the paucity of faint bursts implies
that we are seeing to near the edge of the source population (e.g.
Meegan et al. 1992, Fishman \& Meegan 1995). Both effects, isotropy
and non-homogeneity in the distribution, strongly suggest a
cosmological origin of the phenomenon. In support of this conclusion,
absorption lines (Fe II and Mg II) in the optical counterpart of GRB
970508 have been detected with a redshift of $z=0.835$. Along with
the absence of Lyman-$\alpha$ forest features in the spectra, these
results imply that the burst source is located at 0.835 $\le z \le $
2.3 (Metzger et al. 1997).
The energy required to generate cosmological bursts is as high as
$10^{51}$ erg s$^{-1}$. The very short timescale observed in the time
profiles indicate an extreme compactness that implies a source
initially opaque (because of $\gamma \gamma $ pair creation) to
$\gamma$-rays. The radiation pressure on the optically thick source
drives relativistic expansion, converting internal energy into kinetic
energy of the inflating shell. Baryonic pollution in this expanding
flow can trap the radiation until most of the initial energy has gone
into bulk motion with Lorentz factors of $\Gamma
\ge 10^2 - 10^3$. The kinetic energy, however, can be partially
converted into heat when the shell collides with the interstellar
medium or when shocks within the expanding source collide with one
another.
The randomized energy can be then radiated by
synchrotron radiation and inverse Compton scattering yielding
non-thermal bursts with timescales of seconds. This fireball scenario
has been developed by Cavallo and Rees (1978), Paczy\'nski (1986),
Goodman (1986), M\'esz\'aros and Rees (1993), M\'esz\'aros, Laguna and
Rees (1993) and others. A comprehensive review is presented by
M\'esz\'aros (1997).
The fireball model is a robust astrophysical scenario independent of
the mechanism assumed for the original energy release. A popular
mechanism is the merger of two collapsed stars in a binary system, for
instance, two neutron stars or a neutron star and a black hole (see
Narayan, Paczy\'nski \& Piran 1992, and references threin), although
other processes have been suggested (e.g. Usov 1992, Carter 1992,
Melia and Faterzzo 1992, Woosley 1993).
One important prediction of the fireball model, as well as by
any explosive mechanism, is that individual burst
profiles should be inherently asymmetric under time reversal, with a
shorter rise time than the subsequent decay time. This is a natural
consequence of a sudden particle energy increase (e.g. produced by a
shock) and the slower radiative dissipation of the energy excess.
Time asymmetry in GRBs light curves has been discussed by several
authors (e.g. Mitrofanov et al. 1994, Link et. al. 1993, Nemiroff et
al. 1994). In particular, Nemiroff et al. (1994) showed that in the
sample formed by those bursts with count rates greater than 1800
counts s$^{-1}$ and durations longer than 1s detected by BATSE until
1993 March 10, there is a significant asymmetry in the bursts profiles
in the sense that most bursts rise in a shorter time than they decay,
in agreement with what is expected from a general fireball model. The
most recent and complete study was made by Link and Epstein (1996).
They took 631 GRBs from the BATSE 3B catalog, including both faint and
bright bursts, and confirmed the global asymmetry in the burst
profiles showing that about two thirds of the events display fluxes
that rise faster than the subsequent fall. About 30\% of the bursts,
however, presented a peculiar asymmetry in the temporal profiles, with
slower risings than decays.
In this paper we focus on this subsample of {\it peculiar asymmetric
bursts} (PABs), which seems at first sight to conflict with some
predictions of the simplest scenarios for fireballs. In particular,
we shall discuss whether there are reasons to consider this subsample
of GRBs as representative of a class of sources with different
physical properties than other bursts. The structure of the paper is
as follows. In Section 2 we define the sample and present the results
of the symmetry analysis. We provide tables with the full results for
PABs in order to allow identification of specific events. In Section 3
we study the sky distribution of the sample, while in Section 4 we
investigate the level of positional coincidence (possible
repetition) that PABs show. Finally, we discuss the implications of
these results for theoretical models of GRBs.
\section{Sample and symmetry analysis}
We have studied the sample of 631 bursts from the BATSE 3B catalog
whose global symmetry properties were discussed by Link and Epstein (1996).
This sample contains both faint and bright bursts, spanning a 200-fold
range in peak flux. PREB plus DISC data types at 64 ms time
resolution, with four energy channels, were used in the analysis.
The time asymmetry of the individual burst profiles
was examined with the skewness function introduced by Link et al. (1993)
and used in
Link and Epstein's (1996) paper. This function is defined as
\begin{equation}
{\cal A} \equiv \frac { < \left( t -<t> \right)^3> }
{< \left( t -<t> \right)^2 >^{3/2}}.
\end{equation}
Here, angle brackets denote an average over the data sample, performed
as
\begin{equation}
<g(t)> \equiv \frac{ \sum_i (c_i - c_{th}) g(t_i)}
{\sum _i (c_i - c_{th})},
\end{equation}
where $c_i$ is the measured number of counts in the $i$th bin, $t_i$
is the time of the $i$th bin and $c_{th}$ is a threshold level
defined as,
\begin{equation}
c_{th}=f(c_p -b) + b.
\end{equation}
Here $c_p$ stands for the peak (maximum) count rate, $b$ is the
background, and $f<1$ is a fraction that will be fixed for the data
set. Fixing $f$ ensures that ${\cal A}$ is calculated to the same
fraction of the peak flux relative to the background. Larger values
of $f$ emphasize the structure of the peak over the surrounding
foothills. The normalization of ${\cal A}$ makes it independent of
background, duration, and amplitude. It is equal to 0 in the case of
symmetric bursts, greater than 0 for a burst whose peak rises more
quickly than it falls and smaller than 0 in the opposite case. It is
equal to 2 for an exact FRED (from {\it fast rise and exponential
decay}) and to $-2$ for an exact anti-FRED. Four fixed values of $f$
were analyzed: $f_1$=0.1, $f_2$=0.2, $f_3$=0.5 and $f_4$=0.67. The
only requirement for a burst to be tested in each of the four
$f$-values is that the number of bins whose $c_i$ exceeds or equals
$c_{th}$ is at least three. Consequently, the size of the sample
differs for each choice of $f$. The error bars in ${\cal A}$ represent
1$\sigma$ deviations, calculated by randomizing the number of counts
according to Poisson statistics and computing the variance of the
asymmetry parameter for many trials.
In Tables 1 - 4 we show the results of the skewness analysis for those
bursts that presented PAB behavior (i.e. ${\cal A} < 0$). Each table
contains the peak flux, trigger number, burst type, value of the symmetry
parameter for each $f_i$, and a classification of the bursts profiles
according to the following scheme: S for single-peaked or spike-like
bursts, M for multiply-peaked bursts, and C for complex or chaotic
events. Regarding the burst types, events
for which the ${\cal A}$ is negative for all $f$ are labelled
``1'', whereas events for which the errors in ${\cal A}$
allow positive ${\cal A}$ for at least one value of $f$ are
denoted ``2''.
In Fig. 1 we show specific examples of these profiles.
We found that 91 out of 631 bursts (14.4\%) are PABs, {\sl i.e.} do not
present
positive skewness for any $f$.\footnote{40
bursts out of these 91 are type 1.} Only 28.5\% of these bursts are
single-peaked. The rest are multiply-peaked or complex events.
Notice that most of S-type bursts are in Table 4. This is consistent
with the analysis technique: a fast burst, typically lasting a couple
of seconds, will have few points above the higher cut-offs and then,
data for ${\cal A}_{f_2,f_3,f_4}$ will not be computed.
As we can see from Tables 1 - 4 as well as from Fig. 1, PABs exhibit a
variety of temporal morphologies. If all of these events are
produced by a single mechanism, then there should be a very wide range
of boundary and initial conditions in the sources in order to generate
such a plurality of profiles. With the aim of searching for
differences between PABs and the more common bursts, we have computed
average values of the hardness ratio and durations of type 1 PABs.
These values are compared with similar estimates for those bursts with
${\cal A} > 0$ at all levels in Table 5. Due to the small number of
bursts and the variety in their features, dispersions are so large
that no conclusions can be drawn. However, in the
light of current data, it is clear that no significant correlation
is found between hardness ratio, or duration, with burst morphology.
\section{Sky distribution}
One of the most important results of BATSE is the discovery of that
GRBs are isotropically distributed on the sky
(see, however, Balazs et al. 1998).
With the recent
detection of high-redshifted absorption lines in the optical
counterparts of individual bursts (Djorgovski et al. 1997, Metzger et
al. 1997, van Paradjis et al. 1997) Galactic models appear to be
finally ruled out. However, one could ask whether the distribution of
PABs exhibits the same level of isotropy than that of the whole
sample. It could be the case, for instance, that PABs have a different
origin than other GRBs, and consequently, display a distinct
distribution on the sky
(e.g. there could be a statistically significant concentration
of PABs in the supergalactic plane or within any superstructure).
In order to quantify the isotropy we followed the method developed by
Briggs (1993). The dipole moment toward the Galactic center is $<\cos
\theta > $, the mean of $\cos \theta_i$, where $\theta_i$ is the angle
between the $i$th burst and the Galactic center. An excessively large
value of $<\cos \theta >$ indicates a significant dipole moment towards
the Galactic center. The quantity $<\sin^2 b -1/3 >$ tests for a
concentration in the Galactic plane or in the Galactic poles. The
expected mean values of the two statistics are zero for an isotropic
distribution and, if they are asymptotically gaussian distributed,
i.e. if for a large number of bursts in the sample ($N$)
they are gaussian distributed,
the variances $\sigma^2$ are $1/3N$ and $4/45 N$
respectively.
Briggs et al. (1996) noted that because the CGRO is in a low-Earth
orbit, about one-third of the sky is blocked by the Earth causing a
portion of the Galactic equator to be observed about 20\% less than the
poles. An additional effect is different exposure times between the
Galactic south and north poles. These effects must be taken into
account when computing the expected values of the statistics (Briggs
et al. 1996). Location errors on particular bursts, however, have no
impact on the isotropy characteristics because they are small compared
with the large scale of anisotropies we are testing against.
Table 6 shows that the distribution of all PABs (Fig. 2) is consistent
with perfect isotropy. The same is
true for sub-samples
of PABs. Some entries in Table 6 show small deviations from isotropy
(quadrupole). However, the small number of events make the asymptotic
gaussian distribution no longer valid, and one should compare with the
study of Briggs et al. (1996) (see their Fig. 4a and b).
Comparing with the values of $\sigma$ that arise from the previous
cited figures of the Briggs et al. work, we find,
consequently, that there is no detectable anisotropy in the sky
distribution of PABs and
we see that the 1$\sigma$ deviation from isotropy
contains the values of all entries in Table 6.
\section{Time-space clustering}
Several time-space clustering analysis of different GRB-samples have
come to contradictory conclusions about whether some GRBs repeat or not
(e.g. Quashnock \& Lamb 1993, Narayan \& Piran 1993, Wang \&
Lingenfelter 1995, Petrosian \& Efron 1995, Meegan et al. 1995). The
most complete study on the subject until now, carried out by Tegmark
et al. (1996), is based on the analysis of the angular power spectrum
of 1120 bursts from BATSE 3B catalog. These authors found that the
number of bursts that can be labelled as repeaters (considering just
one repetition) is not larger than 5\% at 99\% confidence.
The recent study of by Gorosabel et al. (1998), which
combined data from different satellites, shows that at most 15.8\% of
the events detected by WATCH recur in the BATSE sample (at 94\%
confidence level). Despite the discussion in the literature, it seems
clear that only a small fraction of the total number of GRBs
could repeat over timescales of up to a few years.
However, if PABs have a different physical origin than other bursts,
this subclass of bursts might exhibit time-space clustering. In fact,
we find that 48 out of 91 PABs (52.7\%) have companions within their
location error boxes in the sample of 631 bursts. If we consider just
bursts separated by less than 4$^o$, we find 40 possible repeaters
(44\% of the PAB-subsample); typically, the separation is about
2.5$^o$.
To estimate the statistical significance of these results, we have
made a numerical study as follows. We have simulated 1500 sets of 91
random positions for PABs. In order to do this, we have made rotations
on the celestial sphere sending a particular PAB with coordinates
$(l,b)$ to a new position $(l',b')$, which is obtained from the
previous by seting $l'=l+ R_1 \, 360^o$ and $b'=b + R_2 \, 90^o$,
and using appropriate spherical boundary conditions. Here, $R_1$ and
$R_2$ are different random numbers (between 0 and 1) which never
repeat. Doing this for each event we get a new set of simulated
PAB-positions. For this set we then compute the positional coincidence
level with respect to the fixed $631-91$ GRBs coordinates. As in the
real case, we shall assign a positional coincidence when two or more
bursts are separated by less than 4$^o$. After making 1500 operations
of this type (a larger number of simulations does not significantly
modify the results) we can obtain the mean value of the expected
number of positional coincidences and its $\sigma$. We obtain that for
91 GRBs, the average level of positional coincidences is 42.9 $\pm$
4.7, which is entirely compatible with the observed result for PABs
within 1$\sigma$.
We have repeated the process for the subset of 26 single-peaked PABs
(those denoted by an S in Tables 1 - 4). These events represent about
4\% of the whole sample and about 28.5\% of the PAB subsample. 15 out
of 26 bursts of this kind ($\sim$60\%) present companions within error boxes
of less than 4 degrees.
We find that the average
simulated positional coincidence level is 13.3 $\pm$ 2.5.
That is, the real coincidence level is also compatible with the random
result to within 1$\sigma$ and no particular association appears obvious.
If we now take positional coincidences separated
by less than 1$^o$, we find that 3 out of 26 single-peaked PABs have
companions. Repeating the simulations in this case yields an expected chance
association of 1 $\pm$ 1 events. This means that the real positional
coincidence is only compatible with the random one to within 2$\sigma$.
The number of events is of course too scarce to draw any conclusion, but if
this is confirmed in a larger sample it would entail an excess of 3.8\%
repetitions above
the result expected from chance associations (something compatible with
already mentioned
Tegmark et al.'s analyses).
As we shall see in the next section, spikes with peculiar asymmetry
present problems for their interpretation within the standard
fireballs models.
\section{PABs and the fireball models}
As mentioned in the introduction, GRBs profiles with ${\cal A} <0$ are
not expected from the simplest versions of the fireball model (i.e. a
single expanding shell that acts as a gamma ray emitter during a brief
time at some fixed radius from the central site of the explosion, e.g.
Fenimore et al. 1996). However, one of the distinctive features of
the fireball scenario is that the same basic mechanism can generate a
variety of time profiles for different initial and boundary
conditions. We now discuss whether these changes can provide the main
types of PABs observed in the sample.
In Fig. 3 we show the profile of BATSE trigger \#2450, which has
negative skewness function for all values of $f$ (see Table 1). This
is a typical multiply-peaked burst, with a precursor at $t=0$ and a
series of peaks of increasing height that start about 35 s after the
first signal. Individual peaks, when analysed with appropriate
$f$, give ${\cal A}>0$. Events of this kind can be understood as the
effect of a mild baryon loaded fireball (M\'esz\'aros \& Rees 1993).
Even a small baryon contamination ($M_b \ge 10^{-9} M_\odot$) of the
expanding pair-photon fireball is enough to trap the $\gamma$-rays
until most of the initial energy is transformed into kinetic energy of
the baryons. The fireball expands by radiation pressure and becomes
optically thin to Thomson scattering when the optical depth drops
below unity at a radius given by (M\'esz\'aros \& Rees 1993),
\begin{equation}
r_p \sim 0.6 \times 10^{15} \theta^{-1} E_{51}^{1/2} \eta^{-1/2}\;\; {\rm cm},
\end{equation}
where $\theta$ takes into account the possibility of channeling of the
flow ($\theta \sim 1$ corresponds to spherical symmetry), $E_{51}$ is
the original energy release ($e^\pm, \gamma$) in units of 10$^{51}$
erg, and $\eta = E_0/M_0 c^2$ is the initial radiation to the rest
mass energy ratio in the
fireball. At $r=r_p$, the $\gamma$-rays still trapped in the fireball
can escape producing a burst (Cavallo \& Rees 1978, Pacz\'nski 1986,
Goodman 1986). As shown by M\'esz\'aros \& Rees (1993), this burst
should be rather weak, with an observed energy in gamma rays of,
approximately,
\begin{equation}
E_p^{obs} \sim 7 \times 10^{47} \theta^{1/3} E_{51}^{1/2} \eta_3^{11/6} \;\;
{\rm erg},
\end{equation}
where $\eta_3=10^{-3} \eta$. This
prompt, small burst will form a precursor that can last a few seconds.
When the expanding relativistic shell collides with the interstellar
medium, a shock wave is formed and the gas in the post-shock region
is heated up to thermal Lorentz factors of $\gamma \sim \eta$,
reconverting the kinetic energy of the shell into thermal energy of
the particles. The thermal energy is radiated through synchrotron and
inverse Compton processes at MeV to GeV energies. A non-uniform
ambient medium can naturally lead to a multiply-peaked burst (e.g.
Fenimore et al. 1996). Events of this class will have ${\cal A}<0$,
as in the case of trigger
\#2450, due to the effect of the prompt precursor. Hence, these ${\cal
A}<0$ events can be explained within the fireball model.
In other cases, the precursor can remain undetected but a multiply-peaked
PAB can arise from internal shocks in bursts with several shells
with different Lorentz factors
(e.g. Kobayashi et al. 1997, Daigne \& Mochkovitch 1998).
In the simulations carried out
by Kobayashi et al. (1997), bursts with negative skewness
can be produced through multiple shell collisions
(e.g. see Fig. 2f of their work).
Complex bursts, as the one shown in Fig. 1d, could be the result of
instabilities on the expanding shell surface once it shocks the
interstellar medium. Hydromagnetic instabilities in the contact discontinuity
can lead to local variations in the fields and the flow's Lorentz factor,
yielding very rapid changes in the time profiles
(e.g. Daigne \& Mochkovitch 1998). The resulting global morphology
could resemble that seen in some bursts with negative
skewness, such as \#2240.
Single-peaked bursts with ${\cal A}<0$, however, appear to be more
difficult to explain with the fireball model. The main problem is
that a single spike with slower rising than falling cannot be
generated through dissipative shocks. In Fig. 4 we show BATSE trigger
\#444 (see also Table 3 and Fig. 1a). We have attempted to fit this event
with the multiple shell
model developed by Kobayashi et al. (1997). The $\gamma$-ray emission
is produced when a shock results from the collision of two shells with
different velocities. The randomized kinetic
energy is then radiated through synchrotron and inverse Compton processes.
Notice that the better the fit for the rising profile, the worse the
model describes the fall. This is a straightforward consequence of
the fact that cooling times are longer than particle acceleration
times at the shock.
To better understand the meaning of the theoretical curves
in Fig. 4 we recall the predicted luminosity in the case of a two
shell interaction (Kobayashi et al. 1997),
\begin{equation}
{\cal L}(t) \propto \left\{
\begin{array}{l}
1- ( 1 + 2 \gamma_m^2 c t /R)^{-2},
\hspace{0.5cm} 0<t<\delta t_e /2 \gamma_m^2 \\
( 1 + (2 \gamma_m^2 t - \delta t_e) c /R)^{-2}
-( 1 + 2 \gamma_m^2 c t /R)^{-2 }, \\
\hspace{0.8cm} t>\delta t_e /2 \gamma_m^2
\end{array}
\right.
\end{equation}
where $\gamma_m$ is the Lorentz factor of the merged shell (depending
on the Lorentz factor and mass of each colliding shell), $\delta
t_e / 2 \gamma_m^2$ is the time at which the burst reach its maximum,
and $R$ is the radius at which the collision takes place.
Observational data of a given burst, its height and duration up to the
maximum in the number of counts, allow a parameterization of ${\cal
L}(t)$ with
\begin{equation}
B=\frac{2 \gamma_m^2 c }{R} .
\end{equation}
The shape of the pulse is asymmetric with a fast rise and a slower
decline unlike a spike event with ${\cal A}<0$. Attempts to fit such
a burst using eq. (6) are shown in Fig. 4.
Spike-like bursts with ${\cal A}<0$ are predicted, however, in some
extrinsic models for GRBs. Torres et al. (1998a,b) have shown that
microlensing effects produced upon the core of high redshifted AGNs by
compact extragalactic objects which violate the weak energy condition
at a
macroscopic level would yield GRB-like lightcurves with spike-type
profiles and negative skewness function. A similar burst
with ${\cal A}>0$ should be observed from several months up to a few
years later in the same position of the sky, provided the lens has an
absolute mass of the order of $1M_\odot$. If this interpretation turns
out to be correct, it could explain not just S-type PABs but also any
apparent excess of positional coincidences among these bursts at a level
compatible with current constraints on repetition over the whole sample.
It would appear that the
small group of spike bursts with negative skewness deserve further
study.
\section{Conclusions}
GRBs exhibit a very rich variety of temporal profiles. Most of them have
highly variable structure over timescales significantly shorter than
the overall duration of the event. The study of burst morphology by
Link and Epstein (1996) shows that a significant fraction of bursts
($\sim 1/3$) have time histories in which the flux rises more rapidly
than it decays (PABs). Here we have argued that most PABs can be
accommodated by fireball models. Isotropy and other average features,
common to the bulk of observed bursts,
are shared by PABs. But there is, however, a subclass of PABs, those which
consist of a single, prominent
peak with negative skewness, that appear to be inconsistent with the
fireball mechanism. These events represent $\sim 4$ \% of the total sample
and certainly merit further research in order to clarify their nature.
\section*{Acknowledgments}
This research has made use of data obtained
through the High Energy Astrophysics Science
Archive Research Center, provided by the NASA/Goddard
Space Flight Center
and also of
NASA/IPAC Extragalactic Database, which is operated by the Jet Propulsion
Laboratory, California Institute of Technology, under contract with
NASA.
Our work has been supported by the Argentine agencies CONICET
(D.F.T. and G.E.R - under grant PIP N$^o$ 0430/98 -),
ANPCT (G.E.R.), and FOMEC (L.A.A.).
We acknowledge G. Bosch for his help in producing Fig. 2 and S. Grigera
for an enlightening discussion on numerical issues.
|
\section{Introduction}\label{intro}
At the end of the twentieth century, the non-perturbative quantization
of gravity remains an elusive goal for theoretical researchers.
There is not even a consensus on how the problem should best be tackled.
For example, considering pure-gravity approaches, we have on the one hand
Euclidean path-integral methods, which are close to usual formulations
of (non-generally covariant)
quantum field theories and well-suited for numerical simulations.
On the other hand, in canonical quantization approaches it is -- at
least in principle --
easier to address questions about the behaviour of spatial
three-geometries, but the complicated structure of the constraints
tends to lead to computational difficulties.\footnote{{Alternatively,
one could embed quantum gravity in a larger, unified theory like
string theory or (the as yet non-existent) M-theory. However, these
are still far from giving us any detailed information about
the quantum gravity sector.}}
Unfortunately, very little is known about the relation between the
covariant and canonical approaches. In part this is due to
the `signature problem' of the path-integral formulations:
the sum over all space-time geometries is usually taken over
Riemannian, and not over the physical Lorentzian (pseudo-Riemannian)
four-metrics modulo diffeomorphisms. The problem of how to relate
the two sectors by an appropriately generalized Wick rotation
remains unresolved.
Our aim is to investigate the possible consequences of taking the
Lorentzian structure seriously within a path-integral approach.
In order to gauge the difficulties this involves and
to circumvent technical problems, we first addressed the
issue in two space-time dimensions, where there already exists a
well-understood theory of (Euclidean) quantum gravity, namely,
Liouville gravity. In \cite{al}, we proposed a new, Lorentzian
model of 2d quantum gravity, obtained by taking the continuum
limit of a state sum of
dynamically triangulated two-geometries. The Lorentzian aspects
of the model were two-fold: firstly, the sum was taken only over those
two-geometries which are generated by evolving a one-dimensional
spatial slice and which
allow for the introduction of a causal structure. Secondly,
the Lorentzian propagator was obtained by a suitable analytic
continuation in the coupling constant. The first aspect turned out to
be the crucial one, leading to a continuum theory of 2d quantum
gravity {\it inequivalent} to the usual Liouville gravity.
This was shown in \cite{al}, where both the loop-to-loop
propagator and various geometric properties of the model were
calculated explicitly. The Hausdorff dimension of the quantum
geometry is $d_{H}\!=\! 2$, and points to a much smoother behaviour
than that of the Euclidean case (where $d_{H}\!=\! 4$).
However, we must emphasize that $d_H \!=\! 2$ does {\it not} imply
a {\it flat} geometry.
The model of Lorentzian gravity defined in \cite{al} allows
for arbitrarily large fluctuations of the spatial volume from
one time-slice to the next.
This is illustrated by Fig.\ \ref{fluctuate}, which
shows a typical surface generated by the Monte Carlo simulations,
to be described in Sec.\ \ref{simulation}.
The length of the compact spatial slice
fluctuates strongly with time (pointing along the vertical axis).
Using the results of \cite{al}, one easily derives
that in the thermodynamic limit and for large times
the average spatial volume $L$ and fluctuations around $L$ behave like
\begin{formula}{add1.1}
\left\langle L \right\rangle = \frac{1}{\sqrt{\Lambda}}~~~~{\rm and}~~~~
\left\langle \Delta L \right\rangle =
\sqrt{\left\langle L^2\right\rangle - \left\langle L\right\rangle^2} = \frac{1}{\sqrt{2\Lambda}}
\end{formula}
respectively, for a given cosmological constant $\Lambda$.
This demonstrates that even in the continuum limit
the fluctuations are large, and of the same order of magnitude
as the spatial volume itself.
\begin{figure}
\centerline{\hbox{\psfig{figure=history1bw.ps,height=10cm}}}
\caption[fluctuate]{A typical discrete history of pure Lorentzian gravity
with volume N=1024.}
\label{fluctuate}
\end{figure}
We managed in \cite{al}
to further trace the difference between the two quantum theories
to the presence or absence of so-called baby universes.
These are outgrowths of the two-geometry giving it the structure
of branchings-over-branchings, which are known to dominate the
typical geometry contributing to the Euclidean state sum.
On the other hand, in the Lorentzian state sum, one can suppress
the formation of such branchings {\it with respect to the
preferred spatial slicing} (which is not present in the Euclidean
picture, where no directions are distinguished). There is also
a physical motivation for suppressing the generation of baby
universes, since the associated (discrete) geometries can
usually not be embedded isometrically in a smooth Lorentzian
space-time. If nevertheless one {\it did} decide to generalize the
evolution rules of the Lorentzian model to allow for such
branchings (and keep only a weaker notion of causality, c.f. \cite{al}),
one would rederive the usual Euclidean Liouville results.
In what follows, when talking about `the Lorentzian model', we will
mean the unmodified model without branching baby universes, i.e. the
model of 2d quantum gravity that does not lie in the same universality
class as Liouville gravity. We also would like to point out that
from the point of view of a canonical
quantization the Lorentzian model is much more natural. The
inclusion of topology changes of space into a canonical scheme
would require a so-called third quantization of geometry.
In Liouville gravity, matter couples strongly to geometry, perhaps
even too
strongly in the sense that the combined system becomes inconsistent
when the central charge of the (conformal) matter exceeds one.
Arguments have been presented which link the strong deformation of
geometry to the creation of baby universes \cite{adj}. It is therefore
conceivable that the Lorentzian model of gravity -- where baby
universes are absent -- has a weaker and less
pathological coupling to matter.
In order to understand the behaviour of the combined
gravity-matter system, we are considering here the coupling
of the gravitational model of reference \cite{al}
to an Ising model of spin-$\frac{1}{2}$, with nearest-neighbour
interaction $\sum_{<ij>}\sigma_{i}
\sigma_{j}$ between its spins $\sigma_{i}\!=\!\pm 1$.
In \cite{al} we made a careful analysis of the implications
of the Lorentzian signature for the sum over space-time metrics.
The most straightforward way of obtaining
the continuum limit consisted in performing the calculations
in the discretized model with purely imaginary coupling
(corresponding to Euclidean signature) and
only afterwards `rotating back' to the Lorentzian sector.
Moreover, it turned out that certain simple properties, like the
fractal dimension of space-time, were independent of the
analytic continuation.
We will apply the same philosophy in the present context
by analyzing the Ising model coupled to 2d gravity with
coupling constants corresponding to the Euclidean signature
sector.
Nevertheless we will continue to talk about `Lorentzian' gravity
coupled to matter, because the choice
of two-dimensional Euclidean geometries contributing to
the path integral is dictated by the requirement that
after the rotation to Lorentzian signature
they should be causal and non-singular.
The `matter observables' we will consider are the
critical exponents for the Ising model,
characterizing the underlying $c\!=\! 1/2$ fermionic continuum
model coupled to gravity, which are not expected to change
under the rotation to Lorentzian signature.
In order to determine
the universality class of the interaction between matter and
gravity it is therefore convenient to work
entirely within the Euclidean sector of our Lorentzian
gravity model.
For fixed regular two-dimensional lattices, and in the absence of
an external magnetic field, the Ising model can be solved exactly in
a variety of ways (see, for example, \cite{nm,mw,dombo1}). The
partition function (for the square lattice) was found by
Onsager. Its critical behaviour is characterized
by a logarithmic singularity of the specific heat and the
critical exponents near the Curie temperature, $\alpha\!=\! 0$,
$\beta\!=\! 0.125$, and $\gamma\!=\! 1.75$, for the specific heat, the
spontaneous magnetization and the susceptibility respectively.
For the case of the usual Euclidean 2d gravity, described by an
ensemble of planar random surfaces, coupling to Ising spins was
first considered in \cite{kazakov}, where an exact solution was
obtained with the help of matrix model methods.
It could be shown that in the presence of gravity, the matter
behaviour is `softened' to a third-order phase transition,
characterized by critical exponents $\alpha\!=\! -1$, $\beta\!=\! 0.5$, and
$\gamma\!=\! 2$ \cite{bk}. On the other hand, the geometry is
`roughened', as exemplified by the increase
from $-1/2$ (pure 2d Liouville gravity)
to $-1/3$ of the entropy exponent $\gamma_{string}$
for baby universes on manifolds of spherical topology.
It is not entirely straightforward to apply the
methods used to obtain these exact solutions to the Lorentzian
gravity model. For example, one can write down an expression for the
transfer matrix generalizing that of the Onsager solution by imposing
a length cutoff $l_{0}$ on the length of spatial slices.
However, a major stumbling block to understanding the behaviour of
its eigenvalues as $l_{0}\rightarrow\infty$ is the fact that as a
consequence of the gravitational degrees of freedom, transitions
between spatial slices of different length are allowed.
This makes the use of Fourier transforms problematic, which are
an essential ingredient of this and other algebraic solution schemes.
Moreover,
the Hilbert space dimension for the discrete, finite model is
given by $\sum_{l=1}^{l_{0}}2^{l}$, which grows rapidly with $l_{0}$.
In the absence of an analytic exact solution\footnote{A matrix model of
Lorentzian gravity coupled to Ising spin has
been formulated recently. Its analysis is the subject of a
forthcoming publication \cite{xxx}.}, one way to try to extract
information about the matter-coupled model is by performing
a series expansion of the partition function $Z$ at high or low temperature,
or of suitable derivatives of $Z$. For flat, regular lattice
geometries, these have been studied extensively since the early days of
the Ising model. It is well-known that the high-$T$ expansion, in
particular, that of the magnetic susceptibility $\chi$ at zero field
is well-suited for obtaining information about the critical
behaviour of the theory. We will show that the same is true for
the coupled gravity-Ising model, after taking into account some
peculiarities to do with the fact that we have an ensemble of fluctuating
geometries instead of a fixed lattice. In the limit of large lattice size
$N$, there is a well-defined expansion in terms of $u:=\tanh \beta $,
where the coupling $\beta$ is proportional to the inverse temperature,
whose coefficients can be determined by diagrammatic techniques.
Given a plausible ansatz for the singularity structure of the
thermodynamic functions, one can then extract estimates for the
critical point and critical exponents from the first few terms of such
an expansion. These results are corroborated
by performing a Monte Carlo simulation of
Lorentzian gravity coupled to the Ising model. Apart from
being in good agreement with the high-$T$ expansion, the
simulations also allow us to measure the quantum geometrical properties
of the model.
\section{The high-$T$ expansion}\label{expansion}
Recall the usual high-$T$ expansion of the Ising model on a fixed
lattice of volume $N$, with partition function
\begin{formula}{2.1}
Z(\beta,N) =\sum_{ \{\sigma_i =\pm 1\} } \mbox{e}^{\b
\sum\limits_{<ij>}
\sigma_i\sigma_j + H\sum\limits_{i} \sigma_{i}},
\hspace{2cm} \beta=\frac{J}{kT},
\end{formula}
where the sum is taken over all possible spin configurations, and
$J>0$ denotes the ferromagnetic Ising coupling. We will only consider
the case of vanishing magnetic field, $H\!=\! 0$. The Ising spins are located at
the lattice vertices, labelled by $i,j\in 1\ldots v$.
A convenient expansion parameter at high temperature is $u:=\tanh \beta$,
which we can use to re-express
\begin{formula}{2.2}
\mbox{e}^{\b\sigma_i \sigma_j}=(1+u\, \sigma_i \sigma_j)
\cosh \b.
\end{formula}
Substituting \rf{2.2} into \rf{2.1}, the partition function becomes
\begin{eqnarray}
Z(\beta,N)&=&(\cosh \b )^s \sum_{\{\sigma_i \} }
\Bigl[ 1+u\sum_{<ij>}
\sigma_i\sigma_j +u^2 \sum_{<ij>} \sum_{<kl>} (\sigma_i\sigma_j)
(\sigma_k\sigma_l)+\ldots\Bigr]\label{2.3} \\
&=:& 2^v (\cosh \b )^s (1+\sum_{n\geq 1} \Omega_n u^n ),
\label{2.4}
\end{eqnarray}
with $v$ denoting the number of vertices and $s$ the number of
nearest-neighbour pairs (i.e. the number of lattice links). Note that
the terms $\sim u^n$ in eq.\ \rf{2.3} are only non-vanishing if every
$\sigma_i$ in $\sigma_{i_1} \sigma_{i_2}\ldots \sigma_{i_n}$ appears
an even number of times. Representing spin pairs $(\sigma_i \sigma_j)$
by drawing a link between $\sigma_i$ and $\sigma_j$ on the lattice,
this is equivalent to the following statement: non-vanishing contributions
to $\Omega_n$ in eq.\ \rf{2.4}
correspond to figures of lattice links which are closed
polygons, with an even number of links meeting at each vertex.
The coefficient $\Omega_n$ simply counts the number of such figures at order $n$
that can be put down on a given lattice, and will depend on the lattice
geometry (triangular, square, etc.). It is a polynomial in the
variable $N$.
Because of the extensive nature of the free energy $F(N)\!=\! -kT \ln Z(N)$,
we must have that $(1+\sum \Omega_n u^n )\sim \mbox{e}^{N (\ldots )}$ in the
thermodynamic limit $N\to\infty$, and we can therefore
write for the partition function per unit volume
\begin{formula}{2.5}
\ln Z(\beta):=\frac{1}{N} \ln
Z(\beta,N)=\frac{s}{N}\, \cosh\b +\frac{v}{N}\, \ln 2
+\sum_{n\geq 1}\omega_{n}^{(0)}u^{n},
\end{formula}
where $\omega_{n}^{(0)}$ is obtained by taking the term linear in $N$
in $\Omega_{n}$ and setting $N\!=\! 1$. Note that both connected and
disconnected graphs contribute to $\omega_{n}^{(0)}$.
A similar relation can be obtained
for the magnetic susceptibility at zero field, $\chi (N)=k
T\frac{\partial^{2}}{\partial H^{2}}\ln Z(N)|_{H=0}$.
At high temperature, the
susceptibility per unit volume can be expressed as
\begin{formula}{2.6}
\chi=k T\ (1+\sum_{n\geq 1}\omega_{n}^{(2)}u^{n}).
\end{formula}
The coefficients $\omega_{n}^{(2)}$ are the exact analogues of
$\omega_{n}^{(0)}$ in eq.\ \rf{2.5}, where the counting now refers to
polygon graphs with two odd vertices (vertices with an odd number of
incoming links), and all other vertices even
(c.f. \cite{dombo2}, but beware of the
difference in notation for the number of vertices).
Since we are primarily interested in the bulk behaviour of the
gravity-matter system, we will in the following for simplicity
choose the boundary conditions to be periodic. That is, we will identify
the top and bottom spatial slices of the cylindrical histories introduced
in \cite{al}. Clearly this is not going to affect the local properties
of the model. As above, we will denote the discrete volume,
i.e. the number of triangles of a given two-dimensional geometry (with
torus topology), by $N$. It follows immediately that such a geometry
contains $N$
time-like links, $N/2$ space-like links, $N/2$ vertices and $3 N/2$
nearest-neighbour pairs.
In quantum gravity the volume $N$ becomes a dynamical variable.
For fixed topology, the only coupling constant appearing in the action
of pure 2d quantum gravity
is the cosmological constant, multiplying the volume term.
The partition function of the Ising model coupled
to 2d Lorentzian quantum gravity is given by
\begin{formula}{add2.1}
G(\l,t,\b) = \sum_{T \in {\cal T}_t} e^{-\l N_T} Z_T(\b)=
\sum_{T \in {\cal T}_t} e^{-\l N_T}
\sum_{ \{\sigma_i(T)\} } \mbox{e}^{\b
\sum\limits_{\left\langle ij\right\rangle \in T}
\sigma_i\sigma_j },
\end{formula}
where the sum is taken over all triangulations $T$ with the topology
of a torus and $t$ time-slices, $N_T$ is the number of triangles
in $T$, and $Z_T(\b)$ the Ising partition function
\rf{2.1} defined on $T$. Fortunately, the summation
over volumes in eq.\ \rf{add2.1} does not lead to additional
complications in the analysis of the thermodynamic properties
of the spin system, since the state sums for fixed and fluctuating
volume are simply related by a Laplace transformation.
Rewrite relation \rf{add2.1} as
\begin{formula}{add2.2}
G(\l,t,\b)= \sum_N e^{-\l N} \tilde{Z}(\b,N,t) :=
\sum_N e^{-\l N} \sum_{T\in {\cal T}_{N,t}} Z_T(\b),
\end{formula}
where ${\cal T}_{N,t}$ denotes the toroidal triangulations of
volume $N$ and length $t$ in the time direction.
Analogous to eq.\ \rf{2.5}, we expect
the matter part $f(\b)$ of the free energy density in the
gravitational ensemble to behave in the thermodynamic
limit ($N \to \infty$ and $t \propto \sqrt{N}$)
like\footnote{Note that with the conventions used
in definition \rf{2.1}, the ground
state energy is $-3\b N/2$ and the free energy density $f(\b)$ is
negative.}
\begin{formula}{add2.4}
\tilde{Z}(\b,N,t) \to e^{(\l_c-\b f(\b)) N+ o(N)}.
\end{formula}
(For simplicity, we have set the ferromagnetic coupling to $J\!=\! 1$.)
We can now reexpress eq.\ \rf{add2.2} as
\begin{formula}{add2.5}
G(\l,t,\b) = \sum_N e^{(\l_c(\b)-\l)N + o(N)},~~~~
\l_c(\b) = \l_c -\b f(\b),
\end{formula}
where $\l_c\equiv \l_{c}(\b \!=\! 0) \!=\!\ln 2$ is the critical cosmological
constant of pure gravity, which was determined in \cite{al}.
Interesting limiting cases are $\b \to 0$ where
$ -\b f(\b) = \frac{1}{2}\ln 2$, reflecting
the factor $2^v$ in eq.\ \rf{2.3} (each spin has two states),
and the strong coupling region
$\b \to \infty$ where $-\b f(\b) \to 3\b/2$ (only the
ground state of all spins aligned contributes to
the state sum \rf{2.1}).
The term proportional to the pure gravity cosmological
constant $\l_c$ appearing together with the free energy in \rf{add2.4}
has its origin in the sum over all triangulations,
\begin{formula}{add2.3a}
\sum_{T\in {\cal T}_{N,t}}1 = e^{\l_c N+ o(N)}.
\end{formula}
A calculation of $\tilde{Z}(\b,N,t)$ not only
determines the thermodynamic
properties of the spin system in the presence of gravity, but at
the same time describes
gravitational aspects of the coupled system, for example, the
{\it critical cosmological constant} $\l_c(\b)$. Conversely,
knowledge of $\l_c(\b)$ determines the spin partition function
in the infinite volume limit. --
The analogue of the high-$T$ expansion \rf{2.4} in the presence of gravity
is given by
\begin{formula}{2.7}
\tilde{Z}(\beta,N,t)=(\cosh
\b)^{\frac{3N}{2}}\ 2^{\frac{N}{2}}\sum_{T\in {\cal T}_{N,t}}
(1+\sum_{n\geq 1} \tilde{\Omega}_n (T) u^n ).
\end{formula}
We may reexpress the critical cosmological
constant of the combined system as
\begin{formula}{add2.6}
\l_c(\b)=\l_c+\frac{3}{2} \ln \cosh \b +\frac{1}{2} \ln 2 + \tilde{f}(u),
\end{formula}
where $\tilde{f}(u)$ is defined in the thermodynamic limit by
\begin{formula}{add2.8}
\frac{\sum\limits_{T\in {\cal T}_{N,t}}
(1+\sum\limits_{n\geq 1} \tilde{\Omega}_n (T) u^n )}
{\sum\limits_{T\in {\cal T}_{N,t}}1}
= e^{N \tilde{f} (u)}.
\end{formula}
The coefficients $\tilde{\Omega}$ of the power series
now depend on the triangulation $T$.
When counting diagrams of a given type and order $n$, we must keep in
mind that the vertex neighbourhoods do not look all the same, as they
do in the case of a regular lattice, but that the distribution of
coordination numbers (numbers of links meeting at a vertex) is subject
to a probability distribution. The coefficients in the high-$T$ expansion
therefore count the {\it average} occurrence of a certain diagram type in
the ensemble of triangulations of a fixed volume $N$, for large $N$.
Starting to evaluate the series in \rf{2.7} order by order, one
immediately notices a qualitative difference from the regular case.
If we had considered a regular triangular lattice (coordination
number 6), the first non-trivial contribution to the counting of even
diagrams would have appeared at $n\!=\! 3$,
where one obtains $\Omega_3(N)\!=\! N$, coming from closed triangle graphs.
However, when looking at all two-dimensional random lattices contributing
to the sum over geometries in the gravity case, there are geometries
which have one or several `pinches'. A pinch is a spatial slice of
minimal length $l\!=\! 1$, which consists of a single link and a single
vertex (see Fig.\ \ref{fig1}).
Pinches occur even if the total volume of the two-geometry is kept fixed,
since in the presence of gravity the length of spatial slices is a
fluctuating dynamical variable.
\begin{figure}
\centerline{\hbox{\psfig{figure=pinch1.ps,height=7cm}}}
\caption[fig1]{A two-dimensional geometry with a `pinch' of
length 1.}
\label{fig1}
\end{figure}
For the gravitationally coupled Ising model, the lowest-order contribution
to the power series in $u$ in \rf{2.7} occurs therefore already at
order $n\!=\! 1$. Clearly, such pinching contributions will be present at
all orders, in both connected and disconnected diagrams, on top of the
ordinary `bulk' contributions, coming from diagrams which do not wind
around the spatial direction of the torus in a non-trivial way. The
former have no analogue on regular lattices.
Fortunately, it turns out that the pinch contributions are
irrelevant, in the sense that they contribute at a lower order
of $N$, whereas the bulk contributions in $\tilde{\Omega}_n$ go like
$N^k$, $k\geq 1$.
This can be seen most easily by considering the Laplace-transformed
partition function. Let us begin by evaluating
the zeroth-order term of eq.\ \rf{2.7},
\begin{formula}{2.8}
G(\tilde{\lambda},t)=
\sum_N \mbox{e}^{-\tilde{\lambda} N} \sum_{T\in {\cal T}_{N,t}} 1:=
\sum_N \mbox{e}^{-\lambda N} (\cosh \b)^{\frac{3N}{2}}\
2^{\frac{N}{2}} \sum_{T\in {\cal T}_{N,t}} 1,
\end{formula}
where for notational brevity we have defined
an `effective cosmological constant'
$\tilde{\lambda}\!=\!\lambda -\frac{3}{2}\ln \cosh \b -
\frac{1}{2} \ln 2$, in accordance with eq.\ \rf{add2.6}.
The left-hand side of \rf{2.8} can be computed
as
\begin{formula}{2.9}
G(\tilde{\lambda},t)=\oint \frac{dx}{2\pi i x}\; G(x,y=\frac{1}{x};
\mbox{e}^{-\tilde{\lambda}};t),
\end{formula}
given the explicit form of the propagator derived in ref.\ \cite{al}, to
which we also refer for details of notation.
The term proportional to $u^{1}$ in the Laplace transform of
\rf{2.7} is
\begin{formula}{2.10}
\sim u^{1}:\hspace{0.8cm} \sum_{N} \mbox{e}^{-\tilde{\lambda} N}
\sum_{T\in {\cal T}_{N,t}^{l=1}} 1=G(\tilde{\lambda},t)\,
\tilde{\Omega}_1^{\rm norm}(\tilde{\lambda}),
\end{formula}
where the second summation is over triangulations with a single `pinch'
of spatial length $l\!=\! 1$. To arrive at the last expression on the
right-hand side, the factor $G(\tilde{\lambda},t)$ has been pulled
out. In terms of quantities derived in \cite{al},
the normalized coefficient $\tilde{\Omega}_1^{\rm norm}$
is most easily computed as
\begin{formula}{2.11}
\tilde{\Omega}_1^{\rm norm}(\tilde{\lambda})=
\frac{\sum\limits_{{\tilde
t}=1}^{t-1}G_{\tilde\lambda}(x,l\!=\! 1;\tilde{t})
G_{\tilde\lambda}(l\!=\! 1,y;t-\tilde{t})}{G_{\tilde\lambda}(x,y;t)}
\Big|_{x=y=0}.
\end{formula}
We are interested in the behaviour of this expression in the
thermodynamic limit, which is tantamount to letting the cosmological
constant approach its critical value,
$\tilde{\lambda}\to \tilde{\lambda}_c$. In this limit,
\rf{2.11} yields simply a constant,
$\tilde{\Omega}_1^{\rm norm} \buildrel{a\rightarrow
0}\over\longrightarrow 2$. This is a general feature of
configurations with one or several pinches. For example, generalizing
to geometries with a single pinch of length $l$ gives a coefficient
$2l$ in the large-volume limit. As an example of a more complicated
configuration, the normalized coefficient for
histories with one pinch of length $l_{1}$ and a second one of
length $l_{2}$ becomes in this limit\footnote{Let us take the
opportunity to correct some misprints in equation (29) of \cite{al},
which has been used in deriving
formula \rf{2.12}. The correct equation reads
$$
G_\l(l_1,l_2;t) = \frac{F^{2t} (1-F^2)^2B_t^{l_1+l_2}}
{l_2 B_t^2 A_t^{l_1+l_2}}
\sum_{k=0}^{\min (l_1,l_2)-1} \frac{(l_1+l_2-k-1)!}{(l_1-k-1)!(l_2-k-1)!k!}
\left(- \frac{A_tC_t}{B_t^2}\right)^k,
$$
where $F,A_t,B_t$ and $C_t$ are defined in \cite{al}.}
\begin{formula}{2.12}
\sim u^{l_{1}+l_{2}}:\hspace{0.6cm}
\buildrel{a\rightarrow 0}\over\longrightarrow\,
3\sum_{k=0}^{{\rm min}(l_{1},l_{2})-1}(-1)^{k}
\frac{(l_{1}+l_{2}-k-1)!}{(l_{1}-k-1)! (l_{2}-k-1)! k!}.
\end{formula}
By contrast, let us now calculate the first bulk contribution, which
occurs at order $u^{3}$. The contribution to $\tilde{\Omega}_3$
is simply $N$, from counting the number of triangle graphs in the
2d geometry.
Taking the Laplace transform, we obtain
\begin{formula}{2.13}
\sum_{N} \mbox{e}^{-\tilde{\lambda} N} N \equiv
-\frac{\partial}{\partial \tilde{\lambda}} G(\tilde{\lambda},t)
\equiv \left\langle N \right\rangle\, G(\tilde{\lambda},t).
\end{formula}
Evaluating the expectation value of $N$ in the continuum limit,
one finds
\begin{eqnarray}
\left\langle N \right\rangle\ &=&- G(\tilde{\lambda},t)^{-1}\,
\frac{\partial}{\partial \tilde{\lambda}}
G(\tilde{\lambda},t)\\
\buildrel{a\rightarrow 0}\over\longrightarrow
&&-\frac{4\, (1-\mbox{e}^{-2 T\sqrt{\Lambda}}-T\sqrt{\Lambda}
(1-\mbox{e}^{-2 T\sqrt{\Lambda}}))}{a^{2}\Lambda
(1-\mbox{e}^{-2 T\sqrt{\Lambda}})}\,
\buildrel{T\;{\rm large}}\over\longrightarrow\,
\frac{4 T}{a^{2}\sqrt{\Lambda}}.\label{2.14}
\end{eqnarray}
(We are using the notation of \cite{al}, with $T$ and $\Lambda$
the continuum length of the two-geometry in `time'-direction and
the renormalized cosmological constant.)
This diverges exactly the way one would expect from a volume term.
It reiterates the conclusion of \cite{al} that all macroscopic
metric variables
scale canonically in the Lorentzian gravity model.
We conclude that in the thermodynamic limit, pinching terms will
be suppressed since their number is proportional to $N^{0}$,
whereas the (connected) bulk diagrams behave like $\sim N^{1}$.
For large $N$, the pinch contributions must therefore factorize
according to
\begin{formula}{2.15}
(1+\sum_{n\geq 1} \tilde{\Omega}_n (T) u^n )=
(1+N^{0} \sum_{m\geq 1} p_m (T) u^m )\,
(1+N\, \sum_{n\geq 1} \tilde{\omega}_n (T) u^n +O(N^{2})).
\end{formula}
Taking the logarithm, we see that the sum
$ (1+ \sum p_m (T) u^m )\sim N^{0}$
will only contribute a constant term to the free energy, which
does not affect the universal behaviour of the model.
We will make no attempt to calculate it explicitly. Similar
considerations apply to the high-$T$ expansion of the
magnetic susceptibility in the presence of gravity. The pinch
contributions factorize, and we will only need to compute the
multiplicity $\tilde{\omega}_{n}^{(2)}$
of bulk polygon graphs with two odd vertices per triangle in
\begin{formula}{2.16}
\chi\sim (1+\sum_{n\geq 1}\tilde{\omega}_{n}^{(2)}u^{n}).
\end{formula}
\begin{figure}
\centerline{\hbox{\psfig{figure=slice1.ps,height=6cm}}}
\caption[fig2]{The triangles contributing to the weight at the
vertex $i$.}
\label{fig2}
\end{figure}
Our next step will be to derive the probability distribution of
the coordination numbers in the Lorentzian gravity model, in the
thermodynamic limit as
the cosmological constant $\l \to \l_c \!=\! \ln 2$.
Recall that when generating an interpolating space-time between
an initial and a final spatial geometry, the geometry of
each space-time `sandwich' with $\Delta t\!=\! 1$ is independent of the
previous one in the sense that there are no local constraints
on how the numbers $k_i\geq 1$ of time-like future-pointing links
can be chosen at each vertex $i$ \cite{al}. Having reached a spatial
slice at time $t$, we can generate the space-time between
$t$ and $t\!+\! 1$ proceeding from `left to right'.
To each vertex $i$ at time $t$ we associate $k_i$ time-like
links (ending at vertices of the subsequent spatial slice at $t+1$)
and the space-like link to the right of the vertex.
There are therefore exactly $k_i$ triangles associated with the
vertex $i$, contributing with a weight factor $e^{-k\l}$ to the
action, as illustrated in Fig.\ \ref{fig2}. Since
the assignment of the order $k_i$ of outgoing time-like links to the
vertex is completely independent of the $k$-assignments of
other vertices, the probability distribution for $k$ outgoing
future-directed links is given by
\begin{formula}{2.19}
p_{\l}(k) = e^{-k\l} (e^{\l}-1).
\end{formula}
Strictly speaking, the argument leading to eq.\ \rf{2.19} is only correct
in the continuum limit in which `extreme pinching' to vanishing
spatial length $l\!=\! 0$ does not occur (for off-critical $\l$,
relation \rf{2.19} must be modified to account for the fact that moves
changing the torus topology are forbidden). Fortunately this is the only
case we are interested in, and the final probability distribution is
therefore obtained by setting $\mbox{e}^{-\l}=1/2$ in \rf{2.19}, yielding
\begin{formula}{2.20}
p(k)\equiv p_{\l_{c}}(k) = \frac{1}{2^k}.
\end{formula}
For reasons of symmetry, the distribution of incoming time-like
links at $i$ (originating at the slice at $t\!-\! 1$) is of course identical.
Given relation \rf{2.20}, we can now compute the
probability distribution $\tilde{p}(j)$ of
the vertex order, i.e. of the
{\it total} number $j$ of links meeting at a vertex
(incoming {\it and} outgoing time-like and space-like links),
\begin{formula}{2.21}
\tilde{p}(j)=\frac{j\!-\! 3}{2^{j-2}},\hspace{1cm} j\geq 4.
\end{formula}
With the distribution \rf{2.20} in hand, we can now embark on the actual
counting of diagrams contributing to the susceptibility coefficients
$\tilde{\omega}_{n}^{(2)}$ in \rf{2.16}.
We will only quote the results up to order $n\!=\! 5$. Further details
of the counting procedure will appear elsewhere. The average numbers
of diagrams per triangle (i.e. per unit volume) are listed in the
table below.
\begin{center}
\vspace{.5cm}
\begin{tabular}{|r||r|r|r|r|}
\hline
{\rule[-3mm]{0mm}{8mm} $n$} & open & closed & disconnected & total\\
\hline\hline
{\rule[-3mm]{0mm}{8mm} 1} & $\frac{3}{2}$ & 0 & 0 & $\frac{3}{2}$ \\
\hline
{\rule[-3mm]{0mm}{8mm} 2} & $8 \frac{1}{2}$ & 0 & 0 & $8 \frac{1}{2}$ \\
\hline
{\rule[-3mm]{0mm}{8mm} 3} & $43 \frac{1}{2}$ & 0 & 0 & $43 \frac{1}{2}$ \\
\hline
{\rule[-3mm]{0mm}{8mm} 4} & $214 \frac{5}{6}$ & 14 & $-17$ & $211
\frac{5}{6}$\\ \hline
{\rule[-3mm]{0mm}{8mm} 5} & \hspace{.8cm} $1038 \frac{1}{6}$ &
\hspace{.8cm} $134 \frac{5}{18}$ &
$-174 \frac{17}{18}$ & \hspace{.8cm} $997 \frac{1}{2}$\\ \hline
\end{tabular}
\vspace{.5cm}
\end{center}
\noindent
Open graphs are connected graphs without any self-intersections.
Closed graphs are connected graphs which are not open.
The disconnected graphs consist of two or more components and
contribute with a minus sign.
In order to double-check our results at order 4 and 5, where the
counting becomes slightly involved, we have performed a numerical
check on the coefficients $\tilde{\omega}_{n}^{(2)}$ listed above.
This was done by computer-generating histories of
length $\Delta t\sim 100$, with an initial spatial slice of length
$\Delta l\!=\! 200$, and counting diagrams of a given type.
The results are given in the table below and in very good agreement with the
exact calculation. They are based on a total of $\sim 3\times 10^{5}$
vertices at order 4 and $\sim 9\times 10^{5}$ vertices at order 5.
We have not listed the counting of disconnected
graphs separately, since it follows closely the counting of
closed connected graphs.
\begin{center}
\vspace{.5cm}
\begin{tabular}{|r||c|c|}\hline
{\rule[-3mm]{0mm}{8mm} $n$} & open & closed \\
\hline\hline
{\rule[-3mm]{0mm}{8mm} 4} & $214.642 \pm 0.179$ & $13.996 \pm 0.007$\\ \hline
{\rule[-3mm]{0mm}{8mm} 5} & \hspace{.4cm} $1037.770 \pm 0.751$
\hspace{.4cm} &
\hspace{.4cm} $134.197 \pm 0.098$ \hspace{.5cm}\\ \hline
\end{tabular}
\vspace{.5cm}
\end{center}
\subsection{Evaluation of results}
In order to evaluate the results from the high-$T$ expansion, we
assume a simple behaviour of the susceptibility of the
form
\begin{formula}{4.1}
\chi (u)\sim A\, \Bigl( 1-\frac{u}{u_c}\Bigr)^{-\gamma}+B
\end{formula}
near the critical point $u_c$, with analytic functions $A$ and $B$.
Using the ratio method
(see, for example, \cite{guttmann}), we have fitted the
susceptibility coefficients to
\begin{formula}{4.2}
r_{n}=\frac{\tilde{\omega}_n^{(2)}}{\tilde{\omega}_{n-1}^{(2)}}=
\frac{1}{u_c} \Bigl( 1+\frac{\gamma \mi1}{n}\Bigr).
\end{formula}
Plotting the ratios $r_{n}$ linearly against $1/n$ for
$n\in 1\ldots n_{\rm max}$, we have
extracted the following estimates for the critical point
$u_c$ and the critical susceptibility exponent $\gamma$:
\begin{center}
\vspace{.5cm}
\begin{tabular}{|c||c|c|}\hline
{\rule[-3mm]{0mm}{8mm} $n_{\rm max}$ }
& critical point & critical
exponent \\ \hline\hline
{\rule[-3mm]{0mm}{8mm} 3} & \hspace{.2cm} $u_c=0.2488$
\hspace{.2cm} &
$\gamma =1.820$\\ \hline
{\rule[-3mm]{0mm}{8mm} 4} & $u_c=0.2462$ &
$\gamma =1.789$\\ \hline
{\rule[-3mm]{0mm}{8mm} 5} & $u_c=0.2458$ &
$\gamma =1.783$\\ \hline
\end{tabular}
\vspace{.5cm}
\end{center}
The estimates for the critical exponent should be compared to
the exact values for $\gamma$ for the Ising model on a fixed, regular
lattice and on dynamically triangulated lattices (Ising spins
coupled to Euclidean quantum gravity), which are $\gamma^{\rm reg}\!=\! 1.75$
and $\gamma^{\rm dt}\!=\! 2$ respectively.
The data from the high-$T$ expansion clearly favour
$\gamma =1.75$ in our model. Indeed, the estimates for $\gamma$ are remarkably
close to this value, given that we are working only up to order 5
in the expansion parameter $u\!=\!\tanh \b$.
The conclusion that the critical exponents of the Ising model coupled
to Lorentzian quantum gravity coincide with those found on regular
lattices is also supported by the
Monte Carlo simulations we have performed.
However, before turning to a detailed
description of the simulations we would like to illustrate how well
the high-$T$ expansion works even at this rather low order.
We will compare the $\b$--dependent
cosmological constant $\l_c(\b)$ defined in eq.\ \rf{add2.6}, which can
be measured directly in the Monte Carlo simulation, with the same
quantity obtained from the high-$T$ expansion.
Recall that in the thermodynamic limit
$\l_c(\b)$ is essentially given by the spin free energy,
eq.\ \rf{add2.5},
which can be computed in the small-$\b$ expansion.
We have determined the density $\tilde{f}(u)$,
defined in eq.\ \rf{add2.8}, by counting closed
polygon graphs in the high-$T$ expansion up to order 6.
Inserting this into formula \rf{add2.6} leads to
\begin{formula}{add2.sc}
\l_c^{\rm high-T}(\b) = \l_c +\frac{1}{2} \ln 2 + \frac{3}{2}\ln \cosh \b +
u^3 +\frac{5}{3} u^4 +\frac{35}{9} u^5 +\frac{263}{27} u^6.
\end{formula}
In Fig.\ \ref{fig3} we show the data points
for $\l_c(\b)-3\b/2$ as measured
by the Monte Carlo simulation\footnote{The subtraction of $3\b/2$
has been performed to ensure a finite limit as $\b \to \infty$.
It corresponds to using the action
$\b \sum_{< ij >} (\sigma_i\sigma_j-1)$
in eq.\ \rf{2.1}, whose ground state has energy zero rather
than $-3 \b N/2$.}. Since $\l_c \!=\! \ln 2$ in pure gravity,
the data should approach $\frac{3}{2}\ln 2$ for
$\b \to 0$ and $\l_c \!=\!\ln 2$ for $\b \to \infty$, both of which
are well satisfied.
In order to quantify the effect of the $u$-expansion,
we have plotted both the zeroth-order expression
$F_1(\b) = \l_c +\frac{1}{2}\ln 2+\frac{3}{2}\ln \cosh \b - \frac{3}{2}\b $,
and the improved sixth-order expression
$F_2(\b) = \l^{\rm high-T}_c(\b) -\frac{3}{2}\b$. The latter
agrees well with the measured Monte Carlo values right up to the
neighbourhood of the critical Ising coupling
$\b_c$. At the critical point $\b_c$ the measured function $\l_c(\b)$
exhibits a cusp. This reflects the singular part
contained in $\l_c(\b)$ which of course cannot be captured by
simply plotting the analytic function \rf{add2.6}.
\begin{figure}
\centerline{\hbox{\psfig{figure=lam01.ps,height=8cm,angle=-90}}}
\vspace{36pt}
\centerline{\hbox{\psfig{figure=lam03.ps,height=8cm,angle=-90}}}
\caption[fig3]{The critical cosmological constant as a function of the
Ising coupling $\beta$, as measured
by Monte Carlo simulations ($t\!=\! 32$, $N\!=\! 2048$),
and compared to the corresponding
high-$T$ expansions $F_1(\b)$ and $F_2(\b)$ at order 0 and order 6.}
\label{fig3}
\end{figure}
\section{The Monte Carlo simulation}\label{simulation}
Monte Carlo simulations have been used successfully in the study
of Euclidean 2d quantum gravity. The formalism known as `dynamical
triangulations' provides a regularization of the functional integral
well-suited for such simulations, allowing in addition
for a straightforward matter coupling of Gaussian
fields as well as of spin degrees of freedom.
Extensive computer simulations of the combined
gravity-matter systems have been performed, leading to results
in perfect agreement with exact results derived from Liouville theory
and matrix model calculations.
The Lorentzian model resembles the dynamically triangulated model
in that its dynamics is associated with the fluctuating connectivity
of the triangulations contributing to the path integral.
This allows us to take over many of the techniques from the computer
simulations of the dynamically triangulated models.
We must specify the update of both the geometry and the
matter fields, the latter being standard: for a given triangulation we
update the spin configurations
by the same spin cluster algorithms used for dynamical triangulations.
This presents no problems since our configurations form a subset
of the full set of dynamical triangulations used in Euclidean
quantum gravity (on the torus). During the update of geometry, we want
to keep the number of time-slices fixed while allowing any space-like
fluctuations compatible with the model. A local change of geometry or
`move' which is clearly ergodic (i.e.\ can generate any of the
allowed configurations when applied successively) is shown in
Fig.\ \ref{move}.
It consists in deleting the two triangles adjacent to
a given space-like link (if the resulting configuration is allowed).
Its inverse is a `split' of a given vertex and two neighbouring time-like
links into two, thereby creating a new space-like link, as well
as two new triangles.
This is a special case of the so-called `grand canonical move' sometimes
used in dynamical triangulation simulations \cite{adfo,jkp,book}, and
does not preserve the total volume of space-time.
\begin{figure}
\centerline{\hbox{\psfig{figure=move.ps,height=3cm,angle=-90}}}
\caption[move]{The move used in the Monte Carlo updating of
the geometry.}
\label{move}
\end{figure}
Detailed balance equations for the move can be derived
from standard considerations \cite{book}. Let $N_V$ denote
the number of vertices ($N_V = N/2$, where $N$ is the
number of triangles), and $v$ a specific vertex.
For pure Lorentzian gravity without matter,
the equation for detailed balance reads
\begin{formula}{mc1}
P(N_V) \frac{ P(N_V \!\to\! N_V\!+\!1)}{N_V k_{in} k_{out}}=
P(N_V\!+\!1) P(N_V\!+\!1\! \to \!N_V),
\end{formula}
where $P(N_V)= \frac{e^{-2\l N_V}}{N_V !}$ is the probability
distribution for labelled triangulations, and $k_{in}$ and $k_{out}$
count the incoming and outgoing time-like links at $v$ (see Fig.\
\ref{move}). We are still free to choose $P(N_V \!\to\! N_V\!+\!1)$
and $ P(N_V\!+\!1\! \to \!N_V)$ such that condition \rf{mc1} is
satisfied. Once a transition probability $ P(N_V\!+\!1\! \to \!N_V)$,
say, has
been chosen, it will be tested during the simulation against the
uniform probability distribution between 0 and 1 as follows. Choose a
random number $r \in ]0,1]$. Then, if the move is {\it allowed} (i.e.\
if the resulting triangulation belongs to the allowed class of
configurations) it is accepted if $P(N_V\!+\!1\! \to \!N_V) > r$. If
it is not allowed, one proceeds to the next move.
It is straightforward to generalize the updating of geometry to
include Ising spins. The spin Hamiltonian is included in $P(N_V)$,
which now becomes a function of both $N_V$ and the spin
configurations. When inserting a vertex $v$, one has
to specify at the same time a spin associated with $v$.
The choice of spin up or down
is made with probability 1/2, and the final result tested as in the
case of the pure geometry update.
We have performed the computer simulation for surfaces with
toroidal topology and for system sizes of $N$=2048, 4050, 8192,
16200 and 32768 triangles, and with a number $t$=32, 45, 64, 90 and
128 of time-slices respectively. Since the moves are not
volume-preserving, fixing the system size to $N$ is implemented
as follows: we allow the volume to fluctuate within a certain, not too
wide range, and collect for every sweep the first
configuration with volume $N$. The volume
fluctuations are controlled by adding a term
$\delta\lambda\, (\Delta N)^2$ to the action, where $\Delta N$ is the
deviation of the volume from its desired value $N$. This term does not
affect the ensemble of configurations collected, since for all of
them $\Delta N\!=\! 0$. We find that $\left\langle \Delta N \right\rangle^{-1}\sim
\sqrt{\delta\lambda}$. Finally, one checks that the results obtained do not
depend on the chosen, allowed range of volume fluctuations.
A sweep is a set of approximately $N_V$ accepted moves.
For each $\beta$-value used in the multi-histogramming analysis
we perform $1.25\times 10^6$ sweeps ($0.75-1.00\times 10^6$ for $N=32768$).
Measurements are made every $5$ sweeps and errors are computed by
data binning.
\subsection{Numerical results for the spin system}
The determination of the critical properties of the
Ising spin system coupled to Lorentzian gravity
proceeds in two steps (see \cite{abt} for a recent, more
complete discussion in the context of 2d Euclidean quantum gravity).
We first locate the critical $\b$-value where the system undergoes
a transition from a magnetized (at large $\b$)
to an unmagnetized phase. Next, we perform simulations
in the neighbourhood of the critical value $\b_c$ and
use finite-size scaling to determine the critical exponents.
Finite-size scaling is also very useful for determining
the location of the critical coupling $\b_c$ itself, since
a number of standard observables show a characteristic behaviour for
$\b$ close to $\b_c$.
The following are some of the observables we have used, together with their
expected finite-size behaviour (see \cite{abt} for a full list):
\begin{formula}{chi}
\chi = N (\left\langle m^2\right\rangle -\left\langle |m| \right\rangle^2) \sim N^{\gamma/\n d_H}~~~~~~~~~~~~
{\rm (susceptibility)}
\end{formula}
\begin{formula}{lnm}
D_{\ln |m|}= N \Big( \left\langle e \right\rangle - \frac{\left\langle e |m|\right\rangle}{\left\langle |m| \right\rangle} \Big)
\sim N^{1/\n d_H}~~~~~~(D_{\ln |m|} \equiv \frac{d \ln |m|}{d \b})
\end{formula}
\begin{formula}{lnm2}
D_{\ln m^2} = N \Big(\left\langle e \right\rangle - \frac{\left\langle e m^2\right\rangle}{\left\langle m^2 \right\rangle} \Big)
\sim N^{1/\n d_H} ~~~~~(D_{\ln m^2} \equiv \frac{d \ln m^2}{d \b}),
\end{formula}
where $\gamma$ and $\n$ are the critical exponents of
the susceptibility and of the divergent spin-spin correlation length,
and $d_H$ is the Hausdorff or fractal dimension of space-time.
For a flat space-time (where of course $d_H\!=\! d\!=\! 2$), we have
$\n d$ = 2, whereas for
the Ising model coupled to Euclidean quantum gravity $\n d_H$ = 3
(and $d_H \approx 4$). The internal energy density $e$
and the magnetization $m$ of the spin system are defined by
\begin{formula}{e-m}
e = \frac{-1}{N Z_N(\b)}\frac{d Z_N(\b)}{d\b},~~~~m =
\frac{1}{N Z_N(\b,H)}\frac{d Z_N(\b,H)}{dH}\; \Big|_{H=0}.
\end{formula}
In order to find the critical point $\b_c$, we can use the fact that
the pseudo-critical coupling $\b_c(N)$ at volume $N$ is expected to
behave like
\begin{formula}{200}
\b_c(N) \sim \b_c + \frac{c}{N^{1/\n d_H}}
\end{formula}
close to $\b_c\!=\!\b_c(N\!=\!\infty)$, with $c$ a constant.
The observables \rf{chi}-\rf{lnm2} all have well-defined peaks
which we used for a precise location of $\b_c(N)$,
with the help of multi-histogram techniques. Eq.\ \rf{200}
can now be used to extract $\b_c$ and $1/\n d_H$.
However, it is advantageous to determine first
$1/\n d_H$ from the peak values of \rf{lnm} and \rf{lnm2},
and then substitute this value into \rf{200}, thus
reducing the number of free parameters in the fit.
Afterwards, one can check that consistent values
for $1/\n d_H$ are obtained from the observables
\rf{lnm} and \rf{lnm2} at $\b_c$, using multi-histogramming.
In our simulations, $1/\n d_H$ extracted from the
peaks was so close to the Onsager value 1/2
that we did not hesitate to use $1/\n d_H\!=\! 1/2$ in \rf{200}. Leaving
it as a free parameter one obtains consistent results, but the error
in $\beta_c$ becomes larger.
Finally, we have determined $\gamma/\n d_H$ from \rf{chi}, both
from the peak values and at $\b_c$
(and again using multi-histogramming).
In the table below we have listed the $\b_c$-values extracted
from measuring $D_{\ln |m|}$ and $D_{\ln m^2}$, and assuming
that $1/\n d_H$ = 1/2. For comparison with the high-$T$ results,
we also give the corresponding critical values for the expansion
parameter $u$. Similar
results are obtained from the rest of the observables we have measured.
\vspace{12pt}
\begin{center}
\begin{tabular}{|c||c|c|}\hline
{\rule[-3mm]{0mm}{8mm} Observable} & $\b_c$, using
$\frac{1}{\n d_H} = \frac{1}{2}$ &
$u_c= \tanh \b_c$ \\ \hline\hline
{\rule[-3mm]{0mm}{8mm}$D_{\ln |m|}$} & 0.2522 (2)& 0.2470 (2) \\ \hline
{\rule[-3mm]{0mm}{8mm}$D_{\ln m^2}$} & 0.2520 (1)& 0.2468 (2)\\ \hline
\end{tabular}
\end{center}
\vspace{12pt}
\noindent
Column 1 of the following table contains the values of $1/\n d_H$
extracted from the peak position for all three observables \rf{chi}--\rf{lnm2}
by using relation \rf{200} (with free parameters $\beta_c$, $1/\n d_H$ and
$c$). Column 2 and 3 give $1/\n d_H$ extracted directly from
\rf{lnm} and \rf{lnm2} by using the peak values of the observables
and their values at $\b_c$.
\vspace{12pt}
\begin{center}
\begin{tabular}{|c||c|c|c|}\hline
{\rule[-3mm]{0mm}{8mm}Observable}
& $\frac{1}{\n d_H}$ from peak pos.&
\hspace{.3cm}$\frac{1}{\n d_H}$ at peak \hspace{.3cm} &
\hspace{.3cm} $\frac{1}{\n d_H}$ at $\b_c$ \hspace{.3cm}
\\ \hline\hline
{\rule[-3mm]{0mm}{8mm} $\chi$} & 0.52 (2) & ---& ---\\ \hline
{\rule[-3mm]{0mm}{8mm} $D_{\ln |m|}$} &
0.47 (2)& 0.531 (2) & 0.521 (3) \\ \hline
{\rule[-3mm]{0mm}{8mm} $D_{\ln m^2}$}
& 0.53 (1) & 0.531 (1) & 0.520 (3) \\ \hline
\end{tabular}
\end{center}
\vspace{12pt}
\noindent
Lastly, we give the value of $\gamma/\n d_H$ extracted from the
susceptibility \rf{chi},
\vspace{12pt}
\begin{center}
\begin{tabular}{|c||c|c|}\hline
{\rule[-3mm]{0mm}{8mm} Observable} & value at peak& value at $\b_c$
\\ \hline\hline
{\rule[-3mm]{0mm}{8mm}$\chi$} & 0.883 (1) & 0.899 (2)\\ \hline
\end{tabular}
\end{center}
\vspace{12pt}
Comparing the estimates for $\b_c$ from the high-$T$ expansion and
the Monte Carlo simulation, one finds good agreement.
The results of the simulation strongly indicate
that the critical exponents are given by the Onsager values
$\n d_H \!=\! 1/2$ and $\gamma\!=\! 1.75$. Again, this corroborates
the conclusion already reached by means of the high-$T$ expansion.
Further evidence that the system belongs to the Onsager and not
the Euclidean gravity universality class comes from measuring the
magnetization exponent $\b_m/\n d_H$ and the specific heat exponent
$\alpha/\n d_H$.
Their Onsager values are $1/16$ and $0$, whereas in Euclidean gravity
they are $1/6$ and $-1/3$.
In our model, the magnetization exponent determined from $\left\langle |m|
\right\rangle_{\b=\b_c}\sim N^{-\b_m/\n d_H}$ was found to be
$\b_m/\n d_H = 0.070(1)$, favouring the Onsager value 0.0625
over the Euclidean gravity value $0.166\bar{6}$.
The specific heat exponent was obtained from the finite-size scaling
of the values of the specific heat peaks $C_V\sim N^{\alpha/\n
d_H}$. A power fit yields $\alpha/\n d_H=0.0861(7)$ at $\chi^2/{\rm
dof}=11.6$ whereas a logarithmic fit gives $\chi^2/{\rm dof}=1.57$,
supporting the conjecture that $\alpha\!=\! 0$.
We do not have independent measurements of the critical parameters
$\n$ and $d_H$ from the spin sector alone, but we will determine the
Hausdorff dimension $d_H$ in the next subsection from an
analysis of the geometry of Lorentzian quantum gravity coupled to
Ising spins.
\subsection{Numerical results for the geometry}
As is well-known from both analytical studies \cite{analytic,ajw,aaa}
and numerous Monte Carlo simulations (\cite{numerical,ajw} and
references in \cite{book}), finite-size scaling is a powerful tool for
determining the fractal space-time structure of
two-dimensional Euclidean quantum gravity.
The same technique can be used to investigate the geometric
properties of two-dimensional Lorentzian quantum gravity.
In a given triangulation, we define the distance between two vertices
as the minimal length of a connected path of links between them.
In 2d Euclidean quantum gravity this notion of distance
becomes proportional to the true geodesic
distance between the vertices in the infinite-volume limit.
We will assume this is also true for the present model.
All diffeomorphism-invariant correlation
functions of matter fields must be functions of this geodesic
distance. Both the geodesic distance and the fractal dimension
appear in the expression for the volume
\begin{formula}{300}
N(r) \sim r^{d_H} ~~~{\rm for}~~~ r \ll N^{1/d_H},
\end{formula}
which denotes the number of vertices (or triangles)
inside a ball (a disc in dimension 2) of link-radius $r$.
If $n_v(r)$ denotes
the number of vertices at distance $r$ from a fixed vertex $v$,
relation \rf{300} implies that
\begin{formula}{301}
n_v(r) \sim r^{d_H-1} ~~~{\rm for}~~~ r \ll N^{1/d_H} .
\end{formula}
Finite-size scaling for an observable $A$ then leads to a scaling
of the correlation function integrated over all points at
distance $r$ from a vertex according to
\begin{formula}{302}
\left\langle A(r) A(0)\right\rangle_N \sim N^{1-1/d_H-\Delta_A} F_A(x),~~~~x= r/N^{1/d_H}.
\end{formula}
The factor $N^{1-1/d_H}$ comes from the integration over points
at distance $r$ from vertex $v$, using \rf{301}, while $\Delta_A$ is the genuine
dynamical exponent of the correlator.
By measuring correlation functions for various volumes $N$, one can
determine $d_H$ and the critical exponents. We will concentrate
here on the Hausdorff dimension $d_H$.
One first rescales the height of the measured distributions
$\left\langle A(r) A(0)\right\rangle_N$ to a common value. However,
the distributions measured for different $N$ will
still have different width as functions of $r$. By appropriately
rescaling $r$, they can then be made to overlap in a
single, universal function $F_A(x)$.
From a technical point of view it is important to work with
the shifted variable
\begin{formula}{303}
x= \frac{r+a_A}{N^{1/d_H}},
\end{formula}
where the shift $a_A$ may depend on the observable $A$. Using
eq.\ \rf{303} takes into
account in an efficient way the major part of the short-distance
lattice artifacts, as has been discussed carefully in \cite{abj,ajw,aaa}.
Applying standard procedures from Euclidean 2d quantum gravity then
leads to the results summarized in Table \ref{haus}.
The observables appearing in Table \ref{haus} are:
(i) the number $n_v(r)$
of vertices at a given (link-)distance $r$ from a fixed vertex $v$,
which may be
viewed as the correlation function of the unit operator in quantum
gravity \cite{analytic};
(ii) the number $s_{up}(r)$ of spins at distance $r$ from a vertex $v$
which are aligned with the spin at $v$;
(iii) the number $s_{down}(r)$ of spins
with orientation opposite to the spin at $v$; (iv) the
spin-spin correlation function $s(r)$ between vertices
separated by a geodesic distance $r$;
(v) the function $S(r)$, obtained by
integrating $s(r)$ over all vertices at distance $r$ from a vertex $v$;
(vi) the distribution $SV(l)$ of spatial volumes,
with $l$ denoting the length of a given time-slice.
For the shift $a_A$, we obtained the estimate
$-4<a_A<-1$. Unfortunately our statistics
was not good enough to determine it with more accuracy.
However, the fact that it is non-vanishing justifies its introduction
in the first place. After a
suitable normalization, we expect the volume distribution to behave like
\begin{formula}{304}
SV(l) \sim f(l/N^{1/d_H}).
\end{formula}
Fig.~\ref{figSV} demonstrates clearly that for the Ising model at
$\beta\!=\! \beta_c$, $SV(l)$ scales as anticipated when we set $d_H\!=\! 2.0$.
Scaling the Ising distributions at $\beta\!=\!\beta_c(N)$,
the pseudo-critical point of the magnetic
susceptibility, or considering pure gravity leads to similar results.
\begin{figure}
\centerline{\hbox{\psfig{figure=SV.ps,height=8cm,angle=-90}}}
\caption[figSV]{The distribution of spatial volumes $SV(l)$ at
$\beta\!=\!\beta_c$, rescaled according to eq. \protect\rf{304}.
We have set $d_H\!=\! 2.0$.}
\label{figSV}
\end{figure}
\begin{table}[ht]
\begin{center}
\begin{tabular}{|c||c|c|c|}
\hline
\multicolumn{1}{|c|}{}&
\multicolumn{3}{|c|}{\rule[-3mm]{0mm}{8mm} {$d_H$} }\\
\multicolumn{1}{|c|}{\rule[-3mm]{0mm}{8mm} {Observable} } &
\multicolumn{1}{c}{$c=\frac{1}{2}$, $\b=\b_c$}&
\multicolumn{1}{c}{$c=\frac{1}{2}$, $\b=\b_c(N)$}&
\multicolumn{1}{c|}{\hspace{.2cm} $c=0$\hspace{.2cm}}\\
\hline\hline
{\rule[-3mm]{0mm}{8mm} $n_v$ } & 2.00(5) & 2.00(5) & 2.03(4) \\
{\rule[-3mm]{0mm}{8mm} $s_{up}$ } & 1.92(5) & 2.08(2) & ---\\
{\rule[-3mm]{0mm}{8mm} $s_{down}$ } & 2.20(3) & 2.07(2) & ---\\
{\rule[-3mm]{0mm}{8mm} $S$ } & 2.10(3) & 2.20(5) & ---\\
{\rule[-3mm]{0mm}{8mm} $s$ } & 2.05(7) & 2.05(5) & ---\\
{\rule[-3mm]{0mm}{8mm} $SV$ } & 2.00(4) & 2.00(3) & 2.00(3)\\
\hline
\end{tabular}
\end{center}
\caption[haus]{The Hausdorff dimension $d_H$, obtained from
the Ising model scaling at $\b_c$ (column 1), at $\b_c(N)$
(column 2) and from pure gravity (column 3).}
\label{haus}
\end{table}
We conclude from Table \ref{haus} that the Hausdorff dimension of
2d Lorentzian quantum gravity is close to the flat-space value
$d_{H}\!=\! 2$. This is clearly different from the Euclidean situation
which is characterized by $d_{H}\!=\! 4$ for pure gravity and
$d_H \geq 4$ in the presence of Ising spins.
The results for the Lorentzian gravity-matter system are particularly
convincing for the purely geometric observables $n_v(r)$ and $SV(l)$,
which basically coincide with the corresponding measurements obtained
in Lorentzian pure gravity.
\section{Conclusions}
We have presented compelling evidence, coming from a high-$T$
expansion as well as Monte Carlo simulations, that the critical
exponents of the Ising model coupled to Lorentzian gravity are
identical to the exponents in flat space. This is in contrast with the
situation in Euclidean gravity (i.e.\ Liouville gravity), where the
exponents change\footnote{The exponents of the Ising model coupled to
2d Euclidean quantum gravity are equal to those of the 3d spherical
model. It is not understood whether this is a coincidence. More
generally, the exponents of the $(m,m+1)$ minimal conformal model
coupled to 2d Euclidean quantum gravity agree with the critical
exponents of the spherical model in $2m/(m+1)$ dimensions.}.
Similarly, the fractal dimension of space-time is unchanged in the
Lorentzian model after coupling it to a conformal field theory (the
Ising model at the critical point). In Euclidean gravity the fractal
properties of space-time are in general a function of the central
change of the conformal field theory. From the evidence collected so
far, we conclude that matter and geometry couple weakly in Lorentzian
gravity and strongly in Euclidean gravity.
For the case of the Ising model, this difference can be explained in
more detail in geometric terms. As mentioned earlier, it has been
shown in \cite{al} that the difference between Euclidean and
Lorentzian gravity is related to the presence or absence of baby
universes. On the other hand, it is by now well understood that baby
universes are at the source of the strong coupling between spins and
geometry. This can happen because the spin configuration of a baby
universe can be flipped relative to that of the `parent' universe at
almost no cost in energy since the `baby' and the `parent' are
connected only by a thin tube. The geometry of two-dimensional
Euclidean quantum gravity is very fractal, with many `pinches' at all
scales, leading to typical spin configurations that look very
different from those on flat space-time. Moreover, the presence of
Ising spins on the surface effectively enhances the fractal baby
universe structure since it is exactly the lowest energy spin
configurations (apart from the ground state) that involve baby
universes. The interaction becomes so strong that it tears the
surface apart when we couple more than two Ising spins to the
two-dimensional geometry. This is the origin of the famous $c\!=\! 1$
barrier of two-dimensional Euclidean quantum gravity.
Once the creation of baby universes is disallowed, as in the case of
the Lorentzian model, the coupling between matter and geometry becomes
weak, and the matter theory has the same critical exponents as in flat
space-time. This happens although the typical space-time geometry is
by no means flat, a fact we have already emphasized in the
introduction, and which is illustrated by Fig.\ \ref{fluctuate}. On
the contrary, our model allows for maximal fluctuations of the spatial
volume which can jump from (essentially) zero to infinity in a single
time step. However, as we have been able to demonstrate, such violent
fluctuations of the two-geometry are still not sufficient to induce a
change in the critical exponents of the Ising model. From the above
arguments it seems likely that Lorentzian gravity can avoid the $c\!=\!
1$ barrier. This question is presently under investigation.
\subsection*{Acknowledgements}
We would like to thank C.\ Kristjansen for comments on a preliminary
version of this article.
|
\section{Introduction}
Numerous experiments have shown that perturbative Electroweak
interactions are
described by the Weinberg-Salam theory of a broken isospin-hypercharge
gauge symmetry. Within that theory the vacuum is
instrumental in breaking the isospin-hypercharge $SU(2)_I \mb{\times} U(1)_Y$
symmetry to a residual electromagnetic $U(1)_Q$ theory. Physically,
the symmetry is broken by the coupling between this vacuum and
isospin-hypercharge gauge fields. This coupling induces mass for the
W and Z components of the gauge fields, whilst the photon does not
couple, remains massless and represents the residual gauge theory.
However, in describing the above symmetry breaking Weinberg-Salam
theory also makes definite predictions about the vacuum
structure. Owing to the $SU(2)_I \mb{\times} U(1)_Y$ gauge symmetry, vacua are
predicted to be degenerate and collectively form the vacuum {\em
manifold}. This structure then implies the existence of
non-perturbative solutions, for example the electroweak strings and
the sphaleron. Thus in describing electromagnetism in
conjunction with a W and Z sector, Weinberg-Salam theory {\em implies}
the existence of further non-perturbative solutions related to the
vacuum structure.
By explicit calculation the relevant vacuum manifold $M$ is predicted
to be a three-sphere. This three-sphere is related to the gauge
structure by the familiar relation
\begin{equation}
\label{M}
M = S^3 \cong \frac{SU(2)_I \mb{\times} U(1)_Y}{U(1)_Q}.
\end{equation}
Now a reasonable question is: what is special about this relation? For
instance how does $SU(2)_I \mb{\times} U(1)_Y/U(1)_Q$ differ from
$SO(4)/SO(3)$, which is also isomorphic to a three sphere.
Mathematically, the answer to this question is: $SO(4)/SO(3)$ and
$SU(2)_I \mb{\times} U(1)_Y/U(1)_Q$ describe different {\em metrical
structures} on the three-sphere. Both describe homogenous metrics, but
they differ in the symmetry properties of the metric. Essentially
$SO(4)/SO(3)$ is associated with a homogenous and isotropic metric on
the three-sphere, whilst the other gives a homogenous and {\em
anisotropic} metric.
Within this paper we examine how this mathematical structure relates
to the vacuum structure. It turns out that the anisotropic $SU(2)_I \mb{\times}
U(1)_Y/U(1)_Q$ metric is naturally induced by the gauge
sector, with the degree of anisotropy specified by the gauge coupling
constants. In addition the isotropic $SO(4)/SO(3)$ metric is
naturally induced by the scalar sector of Weinberg-Salam theory.
Thus Eq.~(\ref{M}) is interpreted as appertaining to the gauge sector
geometry of Weinberg-Salam theory, whilst $S^3 \cong SO(4)/SO(3)$
appertains to the scalar sector.
We then use this framework to interpret the non-perturbative solution
spectrum in terms of the vacuum geometry. Such solutions are usually
specified in terms of their boundary conditions on the vacuum
manifold. Electroweak strings define embedded circles, whilst
the sphaleron defines an embedded two-sphere. With respect to the
geometry we show that these boundary conditions relate to totally
geodesic submanifolds of the three-sphere with respect to {\em both}
the scalar and gauge metrics. We also show that the scattering of
electroweak strings relate to the holonomy of their boundary
conditions with respect to these metrics.
\section{Electroweak Symmetry Breaking}
We start by quickly running through the symmetry breaking mechanism in
electroweak theory. We use this discussion to make explicit some of
the mathematical structure required for this paper.
The usual approach is taken, whereby a Lagrangian describes the
interaction of the gauge fields $W^\mu \in su(2)$ and $Y^\mu \in
u(1)$ with a scalar field $\Phi \in {\bf C}^2$. Minimisation of the
scalar potential yields a vacuum with less gauge invariance than the
original symmetry. Then orthogonal rotation of the $W^\mu$ and
$Y^\mu$ fields gives the $W, Z$ and photon basis of mass eigenstates
with respect to this vacuum.
The scalar-gauge sector of Weinberg-Salam theory is described by the
Lagrangian
\begin{equation}
\label{lag}
\calL = -\frac{1}{4} \inprod{W^{\mu \nu} + Y^{\mu \nu}}{W_{\mu \nu} +
Y_{\mu \nu}} +
(\calD^\mu \Phi)^\dagger (\calD_\mu \Phi)
- \la(\Phi^\dagger \Phi - v^2)^2,
\end{equation}
with the field tensors
\bse
\begin{eqnarray}
W^{\mu \nu} &=& \partial^\mu W^\nu - \partial^\nu W^\mu + [W^\mu,
W^\nu],\\
Y^{\mu \nu} &=& \partial^\mu Y^\nu - \partial^\nu Y^\mu,
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
the covariant derivative
\begin{equation}
\calD^\mu = \partial^\mu + W^\mu + Y^\mu,\\
\end{equation}
and $\inprod{\cdot}{\cdot}$ the $su(2) \oplus u(1)$ inner product.
There is some freedom in the choice of non-degenerate inner products
$\inprod{\cdot}{\cdot}$ on $su(2)_I \oplus u(1)_Y$, constrained to be
invariant under the adjoint action of $SU(2)_I \mb{\times} U(1)_Y$. We
parameterise the possible inner products in the following way
\begin{equation}
\label{inprod}
\inprod{X}{Y} = -\frac{1}{g^2} ( 2\mb{{\rm tr}} XY - \mb{{\rm tr}} X \mb{{\rm tr}} Y )
- \frac{1}{g'^2} \mb{{\rm tr}} X \mb{{\rm tr}} Y,
\end{equation}
with real, positive parameters $g$ and $g'$. Choosing a basis
$\{\frac{1}{2}i \si_a\}$ for $su(2)$, with $\si_a$ the Pauli spin
matrices, and $\frac{1}{2} i {\bf 1}_2$ for $u(1)$, we see that the
unit norm generators are
\begin{equation}
\norm{\frac{1}{2}ig\si_a}=\norm{\frac{1}{2}ig'{\bf 1}_2}=1.
\end{equation}
With these generators the covariant derivative explicitly takes the
form
\begin{equation}
\calD^\mu = \partial^\mu + \frac{1}{2} ig \si_a W^\mu_a
+ \frac{1}{2} ig' Y^\mu,
\end{equation}
yielding the familiar interpretation of $g$ and $g'$ as the isospin
and hypercharge gauge coupling constants.
Symmetry breaking is seen through minimisation of the
Lagrangian~(\ref{lag}), one solution of which is the vacuum solution
\begin{equation}
\Phi(x)=\Phi_0=v\left(\begin{array}{c} 0\\
1\end{array} \right),\ \ \ \ W^\mu=Y^\mu=0.
\end{equation}
The other minima of~(\ref{lag}) collectively give rise to the vacuum
manifold of degenerate equivalent solutions
\begin{equation}
\label{3sphere}
M = SU(2)_I \mb{\times} U(1)_Y \cdot \Phi_0 = \{\Phi : \Phi^\dagger \Phi =
v^2\},
\end{equation}
a three-sphere.
Around the vacuum $\Phi(x)=\Phi_0$ the gauge field mass eigenstates
are given by $W^\mu_1$, $W^\mu_2$, and
\bse
\begin{eqnarray}
\label{gena}
\left( \begin{array}{c} Z^\mu \\ Q^\mu \end{array} \right)
&=& \left( \begin{array}{cc} \cos \theta_w & -\sin \theta_w\\
\sin \theta_w & \cos \theta_w \end{array} \right)
\left( \begin{array}{c} W^\mu_3 \\ Y^\mu \end{array} \right),\\
\label{genb}
\left( \begin{array}{c} \al X_Z \\ eX_Q \end{array} \right)
&=& \left( \begin{array}{cc} \cos \theta_w & -\sin \theta_w\\
\sin \theta_w & \cos \theta_w \end{array} \right)
\left( \begin{array}{c} \frac{1}{2}ig\si_3 \\ \frac{1}{2}ig'{\bf 1}_2
\end{array} \right),
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
an orthogonal transformation of the fields, so that the covariant
derivative becomes
\begin{equation}\calD^\mu = \partial^\mu + \sum_{i=1}^{2}\frac{1}{2} ig \si_i W^\mu_i
+ \al X_Z Z^\mu + e X_Q Q^\mu.
\end{equation}
Interpreting $Q^\mu$ as the photon constrains $X_Q \Phi_0 =0$,
yielding $\tan \theta_w = g'/g$. This massless gauge field $Q^\mu$ is
associated with the a residual $U(1)_Q$ electromagnetic gauge
symmetry
Explicitly the generators are
\begin{equation}
\label{generator}
X_Z = \frac{1}{2}i\left( \begin{array}{cc} \cos 2\theta_w & 0 \\
0 & -1 \end{array} \right),\ \ \ \ \ \
X_Q = i\left( \begin{array}{cc} 1 & 0 \\ 0 & 0 \end{array} \right),
\end{equation}
with $\al = \sqrt{g^2+g'^2}$, $e=gg'/\al$.
The massive gauge field generators $\{\frac{1}{2}i\si_1,
\frac{1}{2}i\si_2, X_Z\}$ form an orthonormal basis for a vector
space $\calM$, such that
\begin{equation}
\label{reddec}
su(2)_I \oplus u(1)_Y = u(1)_Q \oplus \calM,
\end{equation}
with the orthogonality determined by the inner product
$\inprod{\cdot}{\cdot}$. This relation draws particular comparison to
the isomorphism $M \cong SU(2)_I \mb{\times} U(1)_Y/U(1)_Q$, associating the
space of massive generators with the tangent space to the vacuum
manifold at $\Phi_0$.
\section{Vacuum Geometry}
For both the scalar and gauge sectors we now explicitly calculate
their associated metrics. We also calculate the associated
isometry and isotropy groups for each, and from these groups specify the
corresponding geodesic structures.
Having obtained two inequivalent homogenous metrics on the vacuum
manifold we then compare their structure. We relate their isometry and
isotropy groups, and determine when curves are mutually geodesic with
respect to both metrics.
\subsection{Scalar Sector}
The structure associated with the scalar sector is a vector space of
scalar field values ${\bf C}^2$ equipped with a real Euclidean inner
product Re$[\Psi^\dagger \Phi]$. Regarding the vacuum manifold as
embedded within the vector space of scalar field values,
a natural metric may be induced on the vacuum manifold by specifying
its form on each tangent space to be that of the Euclidean inner
product. Such a metric is isotropic and homogenous. Its geodesics are
the great circles.
Explicitly, regard the tangent space to $M$ at $\Phi \in M$ as an
${\bf R}^3$ subspace of ${\bf C}^2$
\begin{equation}
\label{tansp}
T_\Phi M = \{\Psi \in {\bf C}^2 : {\rm Re}[\Psi^\dagger \Phi]=0\},
\end{equation}
with a corresponding metric induced from the real Euclidean inner
product
\begin{equation}
\label{metric-scalar}
g(T_1,T_2)_\Phi = {\rm Re}[T_1^\dagger T_2], \ \ \ T_1, T_2 \in
T_\Phi M.
\end{equation}
This metric has an $SU(2)_I \mb{\times} SU(2)_K$ group of isometries,
\begin{equation}
g(aT_1,aT_2)_{a\Phi}=g(T_1, T_2)_{\Phi},\ \ \ \
a \in SU(2)_I \mb{\times} SU(2)_K.
\end{equation}
The $SU(2)_I$ represents the usual
left isospin $SU(2)$ actions upon ${\bf C}^2$ with generators
$\{\frac{1}{2}i \si_a\}$, whilst $SU(2)_K$ acts upon ${\bf C}^2$ with
the generators $\{-\frac{1}{2} \si_2 K, \frac{1}{2} i
\si_2 K, -\frac{1}{2}i {\bf 1}_2\}$, where $K$ is the complex
conjugation operator. This $SU(2)_K$ contains the hypercharge
$U(1)_Y$, and some other additional symmetries of the scalar sector
which are not symmetries of the full gauge theory.
The isotropy group of $SU(2)_I \mb{\times} SU(2)_K$ upon $M$ at the point
$\Phi_0$ is the subgroup $SU(2)_{I-K}$ such that
\begin{equation}
SU(2)_{I-K} \cdot \Phi_0 = \Phi_0,
\end{equation}
which is generated by $\{\frac{1}{2}i \si_1+\frac{1}{2} \si_2 K,
\frac{1}{2}i \si_2-\frac{1}{2} i\si_2 K,\frac{1}{2}i
\si_3+\frac{1}{2}i {\bf 1}_2\}$. This gives the isomorphism
\begin{equation}
\label{iso-scalar}
M \cong \frac{SU(2)_I \mb{\times} SU(2)_K}{SU(2)_{I-K}}.
\end{equation}
One should be aware that $SU(2) \mb{\times} SU(2)$ is the compact covering
group of $SO(4)$, just as $SU(2)$ is the compact covering group of
$SO(3)$. Thus Eq.~(\ref{iso-scalar}) is an expression of the more
familiar relation $S^3 \cong SO(4)/SO(3)$.
Given the above isotropy and isometry properties of the metric we can
use the isomorphism~(\ref{iso-scalar}) to calculate the
geodesics upon the vacuum manifold with respect to the scalar sector
metric $g(\cdot,\cdot)$. This
geodesic structure follows from some results of differential
geometry~\cite[chapter X]{Nomi2}. Specifically, these results
examine the geodesic structure on the coset space, but this may
be simply carried back to $M \subset {\bf C}^2$ to give the results
that we require. We summarise the full result in an appendix and give
only the answer here.
We firstly need an inner product upon $su(2)_I \oplus su(2)_K$ which we
shall define as
\begin{equation}
\label{inprodsu2}
\inproda{X_I+X_K}{Y_I+Y_K} = - 2\mb{{\rm tr}} X_IY_I - 2\mb{{\rm tr}} X_KY_K,
\end{equation}
with an obvious notation. This then induces
\begin{equation}
\label{redsu2}
su(2)_I \oplus su(2)_K = su(2)_{I-K} \oplus \calN,
\end{equation}
where $\calN$ has an orthogonal basis $\{\frac{1}{2}i
\si_1-\frac{1}{2} \si_2 K, \frac{1}{2}i\si_2+\frac{1}{2} i\si_2
K,\frac{1}{2}i \si_3-\frac{1}{2}i {\bf 1}_2\}$.
The geodesic structure is then
\begin{quote}\em
the geodesics on $M$ with respect to the metric~$g(\cdot,\cdot)$,
which pass through $\Phi_0$ are:
\begin{equation}
\label{geo-scalar}
\ga_X=\{\exp(Xt)\Phi_0: t \in {\bf R}\},
\end{equation}
with $X \in \calN$.
\end{quote}
The above geodesic structure is the main result of this
section. Essentially we have derived the geodesic structure to consist
of the great circles embedded in a three-sphere. This result is
as expected, since we are merely
embedding the three-sphere within a Euclidean space. However,
it is the method which is of importance. The same approach may be
adopted for the gauge sector, where the result is intuitively less
clear. Also using the same formalism allows direct comparison between
the metrical structures associated with the gauge and scalar sectors.
We conclude this section by exploring some consequences of the two
Eqs.~(\ref{inprodsu2}, \ref{redsu2}). The space of geodesic generators
$\calN$ is related to the tangent space of $M$ at $\Phi_0$:
\begin{equation}
T_{\Phi_0}M = \calN \cdot \Phi_0.
\end{equation}
Considering a general $\Phi = a \Phi_0\in M$ with $a \in
SU(2)_I \mb{\times} SU(2)_K$, there is a more general association
\begin{equation}
T_{\Phi}M = \mb{{\rm Ad}}(a)\calN \cdot \Phi.
\end{equation}
For consistency the Euclidean inner product on $T_{\Phi}M$
is equivalent to the inner product on $\mb{{\rm Ad}}(a)\calN$ induced by
$\inproda{\cdot}{\cdot}$ of Eq.~(\ref{inprodsu2})
\begin{equation}
g(X_1 \Phi, X_2 \Phi)_{\Phi} = \inproda{X_1}{X_2},\ \ \ X_i \in
\mb{{\rm Ad}}(a) \cdot \calN.
\end{equation}
This property is essential to the derivation of
Eq.~(\ref{geo-scalar}), and motivates the choice made in
Eq.~(\ref{inprodsu2}).
\subsection{Gauge Sector}
The main structure associated with the gauge sector is the inner
product $\inprod{\cdot}{\cdot}$ of Eq.~(\ref{inprod}). It specifies
several important related features of the gauge sector. First of all
it defines the gauge kinetic term in Eq.~(\ref{lag}), introducing the
gauge coupling constants as the relative scales. Secondly it
stipulates the embedding of the massive gauge generators $\calM$ in
Eq.~(\ref{reddec}) to be perpendicular to $u(1)_Q$. Finally it renders
the photon $X_Q$, Z-field $X_Z$ and W-field generators mutually
orthonormal.
We shall use this inner product to define the gauge-sector
metric. The definition is achieved by associating the massive
generators $\calM$ with tangent spaces to the vacuum manifold in a
manner completely analogous to that in the last section. Then the
natural inner product $\inprod{\cdot}{\cdot}$ on the massive
generators $\calM$ induces a metric on the vacuum manifold. We
find the corresponding isometry group of this metric to be
the gauge group $SU(2)_I \mb{\times} U(1)_Y$ and the isotropy group to be the
residual symmetry $U(1)_Q$. Thus the metric is homogenous, but
anisotropic. Its geodesics are rather complicated in structure.
Explicitly, observe that the tangent space~(\ref{tansp}) may be
expressed
\begin{equation}
T_{\Phi_0}M = \calM \cdot \Phi_0.
\end{equation}
More generally, the corresponding tangent space at $\Phi=a \Phi_0 \in
M$ is, for any $a \in SU(2)_I \mb{\times} U(1)_Y$,
\begin{equation}
\label{tgt-iso}
T_\Phi M = aT_{\Phi_0} M = \mb{{\rm Ad}}(a)\calM \cdot \Phi.
\end{equation}
Transitivity over $M$ guarantees a natural isomorphism between any
tangent space and some $\mb{{\rm Ad}}(a) \calM$.
Using the isomorphism implied by Eq.~(\ref{tgt-iso}), the inner
product $\inprod{\cdot}{\cdot}$ associates a corresponding
metric on $M$
\begin{equation}
\label{metric-gauge}
h(X_1 \Phi, X_2 \Phi)_\Phi = \inprod{X_1}{X_2},\ \ \ X_1, X_2 \in
\mb{{\rm Ad}}(a)\calM.
\end{equation}
The precise form is parameterised by the hypercharge
and isospin coupling constants.
This metric has an $SU(2)_I \mb{\times} U(1)_Y$ group of isometries
\begin{equation}
h(aT_1,aT_2)_{a\Phi}=h(T_1, T_2)_{\Phi},\ \ \ \
a \in SU(2)_I \mb{\times} U(1)_Y.
\end{equation}
More precisely, by Eq.~(\ref{tgt-iso}) the action of $a \in SU(2)_I \mb{\times}
U(1)_Y$ upon $h(\cdot,\cdot)$ is
\begin{equation}
h(aT_1,aT_2)_{a\Phi} = \inprod{\mb{{\rm Ad}}(a)X_1}{\mb{{\rm Ad}}(a)X_2} =
\inprod{X_1}{X_2} = h(T_1,T_2)_{\Phi}.
\end{equation}
The above isometries represent the maximal subgroup of $SU(2)_I \mb{\times}
SU(2)_K$ leaving $\inprod{\cdot}{\cdot}$ invariant.
The isotropy group of this isometry group at the point
$\Phi_0$ in $M$ is the subgroup $U(1)_Q$ such that
\begin{equation}
U(1)_Q \cdot \Phi_0 = \Phi_0,
\end{equation}
giving the isomorphism
\begin{equation}
\label{iso}
M \cong \frac{SU(2)_I \mb{\times} U(1)_Y}{U(1)_Q}.
\end{equation}
Thus we recover the familiar relation for the vacuum manifold, but
now explicitly associated with the gauge sector metrical structure.
As for the scalar sector the importance of isomorphism~(\ref{iso}) is
that we may use the isotropy and isometry properties of
the metric to calculate the geodesics upon the vacuum manifold with
respect to the gauge sector metric. Again we give
only answer and refer to the full result summarised in the appendix.
The structure is
\begin{quote}\em
the geodesics on $M$ with respect to the metric $h(\cdot,\cdot)$, passing
through $\Phi_0$ are:
\begin{equation}
\label{geo-gauge}
\ga_X=\{\exp(Xt)\Phi_0: t \in {\bf R}\},
\end{equation}
with $X \in \calM$.
\end{quote}
A short calculation yields the geodesics through $\Phi_0$ associated
with the following generators in $\calM$
\begin{equation}
X = \frac{1}{2}i c_1 g\si_1 + \frac{1}{2}ic_2 g\si_2 + d X_Z, \ \ \ \
c_1^2 + c_2^2 + d^2 =1
\end{equation}
to be
\begin{equation}
\label{geodesics}
\ga_X(t) = v e^{-id \tan^2 \theta_w t/2}
\left(\begin{array}{c} c \sin t/2 \\
\cos t/2 - id \sin t/2 \end{array} \right),
\end{equation}
where $c=c_2 + ic_1$. This structure is rather complicated, for instance
the closed geodesics from a discrete set such that
\bse
\begin{eqnarray}
d \tan^2 \theta_w &\in& {\bf Q},\ \ \ \ c \neq 0, \\
d &\in& {\bf Q},\ \ \ \ c =0.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
Geodesics through other points $\Phi = a \Phi_0 \in M$ may be
simply obtained by acting correspondingly on Eq.~(\ref{geodesics}).
There are a couple of points to observe about the geodesic
structure. Firstly, their exists a totally geodesic two sphere through
$\Phi_0$ defined by those geodesics satisfying $d=0$. Secondly, in the
direction perpendicular to the tangent space of this sphere there
exists a closed geodesic satisfying $c=0$. Because of homogeneity
this is also true for any point on the vacuum manifold.
The above metric is homogenous but anisotropic.
Its anisotropy is parametrised by the weak mixing angle
$\theta_w$, becoming isotropic when $\theta_w$ vanishes. Thus one may
interpret the gauge metric $h(\cdot, \cdot)$ as a continuous and
homogenous deformation of the isotropic Euclidean scalar metric
$g(\cdot, \cdot)$. At each point this deformation leaves a two-sphere
and a circle unaffected, but
deforms relative to the two. In some sense this structure is analogous
to {\em squashing} the three-sphere; having the geometry of an
ellipsoid, but with the deformation homogeneous so as to respect the
homogeneity of~(\ref{iso}).
We conclude this part of the discussion by observing that the
above structure is rather special to
the three-sphere. It is related to the Hopf fibration. In the Hopf
fibration picture the metric on the $S^1$
fibres has a different length scale to that on the $S^2$ base
space. One should note that the only other spheres to have a similar
structure are the seven-sphere and the fifteen-sphere.
\subsection{Scalar-Gauge Geometry}
In summary, we found two inequivalent metrics on the electroweak
vacuum manifold
associated with the scalar and gauge sectors. We shall now determine
how the structure of these metrics relate to each other.
Comparing the respective symmetry groups determines those symmetries
which are shared. These shared symmetries define two
submanifolds whose geodesics are mutually geodesic with respect to the
two metrics. These correspond to the two-sphere and the circle
mentioned in the discussion of the gauge sector metric.
The scalar and gauge metrics, $g(\cdot,\cdot)$ and $h(\cdot,\cdot)$,
have the following isometry group decompositions with
respect to their isotropy groups
\bse
\begin{eqnarray}
su(2)_I \oplus su(2)_K &=& su(2)_{I-K} \oplus \calN,\\
su(2)_I \oplus u(1)_Y &=& u(1)_Q \oplus \calM,
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
where the group structure is related by
\bse
\begin{eqnarray}
u(1)_Q &\subset& su(2)_{I-K},\\
u(1)_Y &\subset& su(2)_K,\\
\calM &\subset& su(2)_I \oplus su(2)_K.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
Also, the tangent space to $M$ at $\Phi_0$ is related to $\calM$ and
$\calN$ by
\begin{equation}
T_{\Phi_0} M = \calM \cdot \Phi_0 = \calN \cdot \Phi_0.
\end{equation}
It is important to understand how the metrics $g(\cdot,\cdot)$ and
$h(\cdot,\cdot)$ relate to each another. By bilinearity of the
metrics, at the point $\Phi_0 \in M$,
\begin{equation}
g(T_1, T_2)_{\Phi_0} = h(AT_1, AT_2)_{\Phi_0},\ \ \
A \in GL(T_{\Phi_0}M),
\end{equation}
relating the metrics by a linear map of the tangent space. The
eigenspaces of $A$ can be found by explicitly calculating $A$, and
diagonalising it, This yields
\begin{equation}
\label{tdecom}
T_{\Phi_0}M = T_{\Phi_0}^Z M \oplus T_{\Phi_0}^W M,
\end{equation}
with
\bse
\begin{eqnarray}
T_{\Phi_0}^Z M &=& \calM_Z \cdot \Phi_0,\ \ \ \ \calM_Z = {\bf R}
\cdot X_Z,\\
T_{\Phi_0}^W M &=& \calM_W \cdot \Phi_0,\ \ \ \ \calM_W = {\bf R}
\cdot \frac{1}{2}i\si_1 \oplus {\bf R} \cdot \frac{1}{2}i\si_2.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
Then the metrics are related such that
\begin{eqnarray}
\label{relation}
g(X_Z\Phi_0 + X_W\Phi_0,Y_Z\Phi_0 + Y_W\Phi_0)_{\Phi_0}&=&\nonumber \\
\la_Zh(X_Z\Phi_0,Y_Z\Phi_0)_{\Phi_0}
&+& \la_Wh(X_W\Phi_0,Y_W\Phi_0)_{\Phi_0},
\end{eqnarray}
with an obvious notation, and $\la_Z=\al^2 v^2/\cos 2\theta_w$,
$\la_W=g^2v^2$. This may be easily
generalised to all $\Phi \in M$ by considering the action of
$SU(2)_I \mb{\times} U(1)_Y$ on Eq.~(\ref{relation}).
Decomposition~(\ref{tdecom}) also describes the geodesic structure
of $M$ with respect to $g(\cdot,\cdot)$ and $h(\cdot,\cdot)$ in a
rather nice way. Applying Eqs.~(\ref{geo-scalar}) and
(\ref{geo-gauge}), the submanifolds
\bse
\begin{eqnarray}
\label{subM}
M^Z = \exp(\calM_Z)\Phi_0,\ \ \ \ \ T_{\Phi_0}M^Z = T^Z_{\Phi_0} M \\
M^W = \exp(\calM_W)\Phi_0,\ \ \ \ T_{\Phi_0}M^W = T^W_{\Phi_0} M
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
are the {\em only} totally geodesic submanifolds of $M$ with respect
to {\em both} metrics $g(\cdot,\cdot)$ and $h(\cdot,\cdot)$. No other
submanifold of $M$ has this property.
One should note that the $\mb{{\rm Ad}}(U(1)_Q)$-irreducible subspaces
of $\calM$ with respect to the inner product $\inprod{\cdot}{\cdot}$
are
\begin{equation}
\calM = \calM_Z \oplus \calM_W,
\end{equation}
as found in the decomposition above. These also relate to the mass
eigenstates of the massive gauge fields. This property is in fact very
general~\cite{metom}.
\section{Physical Implications}
In summary of the previous section we found two homogenous metrics on
the vacuum manifold associated with the scalar and gauge sectors of
Weinberg-Salam theory. The scalar sector induces an isotropic metric,
whilst the gauge sector induces an anisotropic metric. There is a
unique totally geodesic two sphere with respect to both
metrics. Geodesic curves with respect to both metrics consist of
the geodesics in this two sphere, and one other circular path whose
tangent vector is orthogonal to the tangent plane of this two sphere.
Given this structure one might inquire as to how this relates to the
spectrum of non-perturbative solitonic-type solutions present in
Weinberg-Salam theory. It transpires that the electroweak strings
correspond to the mutually geodesic paths, whilst the
sphaleron corresponds to the mutually geodesic two-sphere. This
approach has the added bonus of interpreting the scattering of
electroweak strings in terms of the holonomy of their respective
geodesic. Also the dynamical stability of the Z-string, as the
weak mixing angle approaches $\pi/2$, is seen to correspond to extreme
anisotropy of the gauge metric.
\subsection{Electroweak Strings}
\label{sec-Electroweak}
Electroweak strings correspond to Nielsen-Olesen vortices embedded in
Weinberg-Salam theory~\cite{Vach92}. As such their boundary conditions
define circular paths on the vacuum manifold. Thus one might expect
that their spectrum and properties should correspond to the geometry
of the vacuum manifold. This is what we find. Their boundary conditions
correspond to the paths that are mutually geodesic with respect to
{\em both} metrics.
Formally, an electroweak string is defined by the embedding
\begin{eqnarray}
SU(2)_I \mb{\times} U(1)_Y &\rightarrow& U(1)_Q\nonumber \\
\cup \ \ \ &\ & \cup\\
U(1) &\rightarrow& 1,\nonumber
\end{eqnarray}
with the general Ansatz
\bse
\begin{eqnarray}
\Phi(r, \theta) &=& f_{\rm NO}(r) e^{X\theta}\Phi_0, \\
\ul{A}(r,\theta) &=& \frac{g_{\rm NO}(r)}{r} X \ul{\hat{\theta}},
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
where $X\in su(2)_I \oplus u(1)_Y$ is the vortex generator.
One may consider only $X \in \calM$, since these minimise the
magnetic energy~\cite{me2}. Thus one considers only
Ans\"atze with boundary conditions geodesic with respect to
the scalar metric $h(\cdot, \cdot)$.
The above vortex Ansatz is a solution provided that~\cite{Barr92}\\
(i) The scalar field must be single-valued. Hence the boundary
conditions describe a closed geodesic with $e^{2\pi
X}\Phi_0=\Phi_0$.\\
(ii) The Ansatz is a solution to the equations of motion; then fields
in the vortex do not induce currents perpendicular (in Lie algebra
space) to it~\cite{Vach94}. This may be equivalently phrased
as~\cite{me2}:
$X$ is a vortex generator if Re$[(X\Phi_0)^\dagger X^\perp \Phi_0] =
0$ for all $X^\perp$ such that $\inprod{X^\perp}{X}=0$.
Condition (ii) can be conveniently restated in terms of the
corresponding metrics:
\begin{quote}\em
$X$ is a vortex generator if the associated tangent vector
$T=X\Phi_0$ satisfies $g(T, T^\perp)=0$ for all $T^\perp$ such that
$h(T, T^\perp)=0$.
\end{quote}
Referring to the discussion around Eq.~(\ref{tdecom}), we see that $T$
must lie in one of the eigenspaces of the linear map relating the two
metrics. Namely $X$ is an element of either $\calM_Z$ or $\calM_W$,
the two $\mb{{\rm Ad}}(U(1)_Q$-irreducible subspaces of $\calM$ in the
decomposition
\begin{equation}
\calM = \calM_Z \oplus \calM_W.
\end{equation}
It is interesting that the geodesics defined from $\calM_Z$
and $\calM_W$ are the only geodesics which are {\em simultaneously
geodesic with respect to both metrics}. From a mathematical point of
view this is because these geodesics define submanifolds of the vacuum
manifold with coincident scalar and gauge metrics (to an
overall factor). From a physical point of
view, this may be interpreted as a minimisation of both the scalar and
gauge sectors of the action integral.
From the above it is a fairly trivial exercise to work out the forms
of the vortices, obtaining results in agreement with
refs.~\cite{Vach92, Barr92}.
\subsection{Semi-local Vortices}
When the weak mixing angle becomes $\pi/2$ the isospin gauge symmetry
$SU(2)_I$ becomes a global symmetry whilst the hypercharge symmetry
$U(1)_Y$ remains local. Such a model is interpreted as a complex
doublet scalar field with a gauged phase. For suitable scalar
potentials it admits dynamically stable semi-local vortex
solutions~\cite{vach91},
interpreted as the corresponding limit of a Z-string. By continuity
the Z-string is dynamically stable in a region close to weak mixing angle
$\pi/2$~\cite{Vach92}. We wish to point out that this is
related to the gauge metric becoming {\em extremely} anisotropic.
As the weak mixing angle tends to $\pi/2$ the isospin coupling $g$ tends to
zero. For $g=0$, the inner product $\inprod{\cdot}{\cdot}$ becomes ill
defined. It is well defined only on the subalgebra $u(1)_Y$, where it
takes the form
\begin{equation}
\label{ip1}
\inprod{X}{Y}_{\pi/2} = - \frac{1}{g'^2}\mb{{\rm tr}} X \mb{{\rm tr}} Y.
\end{equation}
By Eqs.~(\ref{metric-gauge}, \ref{subM}), the gauge metric is only
defined upon the one-dimensional submanifold $M_Z \subset M$, where
\begin{equation}
h_{\pi/2}(X_1 \Phi, X_2 \Phi)_\Phi = \inprod{X_1}{X_2}_{\pi/2},\ \ \
X_1, X_2 \in \calM_Z.
\end{equation}
This is interpreted as the limit of {\em extreme} anisotropy; this
anisotropy picking out the submanifold $M_Z \subset M$ upon which the
metric is well defined.
Physically, the submanifold $M_Z$ represents those points that may be
reached by a gauge transformation from $\Phi_0$. Other points within
the vacuum manifold may only be reached by a global transformation.
This property is related to the stability of the vortex. To decay, the
vortex solution must deform out of $M_Z$. Such a process costs large
gradient energies that may not be compensated for by a gauge
field~\cite{vach91}.
\subsection{Combination Electroweak Vortices}
When the weak mixing angle vanishes the hypercharge gauge symmetry
$U(1)_Y$ becomes a global symmetry whilst the isospin symmetry
$SU(2)_I$ remains local. Such a model is interpreted as a gauged
complex doublet scalar field with a global phase. Then Weinberg-Salam
theory represents the symmetry breaking $SU(2)_I \rightarrow {\bf 1}$.
Correspondingly there is a two-parameter family of embedded vortices
representing Z-strings, W-strings, and combination W-Z strings.
We wish to point out here that this is related to the gauge metric
becoming isotropic, coinciding with the scalar metric and thus all
geodesics of the scalar sector define electroweak strings.
When the weak mixing angle vanishes, so that $g'=0$, the inner
product $\inprod{\cdot}{\cdot}$ becomes well defined only on the
subalgebra $su(2)_I$,
\begin{equation}
\label{ip0}
\inprod{X}{Y}_0 = - \frac{1}{g^2} (2 \mb{{\rm tr}} XY - \mb{{\rm tr}} X \mb{{\rm tr}} Y).
\end{equation}
Analogous to Eq.~(\ref{metric-gauge}), the metric is then defined from
$\inprod{\cdot}{\cdot}_0$ to be
\begin{equation}
h_0(X_1\Phi,X_2\Phi)_\Phi = \inprod{X_1}{X_2}_0,\ \ \ X_1, X_2 \in
su(2)_I.
\end{equation}
Then the scalar and gauge metrics coincide upto a factor of the
isospin coupling constant $g$
\begin{equation}
g(T_1,T_2) = \frac{1}{g}h_0(T_1,T_2).
\end{equation}
Thus the isometry
group of the gauge metric increases to $SU(2)_I \mb{\times} SU(2)_K$, with this
$SU(2)_K$ now representing a set of global $SU(2)$ symmetries of the
gauge theory only apparent at vanishing weak mixing angle.
In terms of the electroweak vortex spectrum, condition (ii) is
satisfied for all vortex generators $X \in su(2)_I$, since with
respect to the two metrics all vortices are trivially coincident.
Hence the spectrum of vortex solutions becomes a continuous family
defined by elements $X \in su(2)_I$.
It is an interesting question to enquire what happens to these
solutions as the weak mixing angle moves off zero. By continuity one
might expect some sort of perturbed solution to exist. However it is
difficult to see what its boundary conditions would be since
necessarily they can only be a geodesic of either the gauge metric or
the scalar metric, not both.
\subsection{Non-Abelian Aharanov-Bohm Scattering}
It is known that the Aharov-Bohm scattering of particles off a vortex,
or one vortex off another, is controlled by the holonomy of a vortex's
boundary conditions~\cite{Wilz}. In this context the term holonomy was
used to indicate non-trivial parallel transport of either a vortex or
a charged particle in a circuit around a vortex. We show here that
this holonomy refers precisely to the holonomy with respect to the
gauge metric.
Associated with the magnetic flux of a vortex is the Wilson line
integral
\begin{equation}
\label{wilson}
U(\theta) = P \exp \left( \int_{0}^{\theta} {\bf A} \cdot d{\bf l}
\right) \subset G,
\end{equation}
at infinite radius. The function $U(2\pi)$ dictates the parallel
transport of matter fields around a vortex, such that a fermion
doublet $\Psi$ is transported to $U(2\pi)\Psi$. Diagonalisation of $U$
then associates components $\Psi_i$ with phase shifts $\xi_i$.
Non-trivial fermionic components $\Psi_i$ interact
with the vortex by an Aharanov-Bohm cross section
\begin{equation}
\deriv{\si}{\theta} = \frac{1}{2\pi k}
\frac{\sin^2(\xi_i/2)}{\sin^2(\theta/2)},
\end{equation}
whilst trivial components $e^{2 \pi \xi_k}=1$ interact by an Everett
cross section~\cite{Ever}. The above holds for parity symmetric
theories only. When charges for the left and right fermion fields
differ one must also include the effects of induced fermionic zero
modes
Substitution of the boundary conditions for the Z and W-string,
determined in sec.~(\ref{sec-Electroweak}), gives the Wilson line
integral for electroweak strings. For the Z-string
\begin{equation}
U_Z(\theta) = \exp(-2 X_Z \theta),
\end{equation}
whilst for the W-string
\begin{equation}
U_W(\theta) = \exp(i(\si_1 \sin \eta +
\si_2 \cos \eta) \theta).
\end{equation}
Vortex boundary conditions restrict $U(2 \pi)\in U(1)_Q$, where
explicitly $U_Z(2\pi)=\exp(-4 \pi \cos^2 \theta_w X_Q)$ and $U_W(2
\pi)={\bf 1}$. For the scattering of a fermion doublet $\Psi$ off a
Z-string this yields phase shifts
$\xi_1=-2 \pi \cos^2 \theta_w$ and $\xi_2=2\pi$. Thus upper components
interact with Z-strings by an Aharanov-Bohm cross section, whilst
lower components interact by an Everett cross section. Fermions
interact with W-strings only by an Everett cross section.
In \cite{Wilz} the Wilson line integral is related to holonomy,
referring to non-trivial parallel transport around the
vortex. Referring to our appendix, in particular the result
\begin{quote}\em
the parallel transport of ${\bf u} \in T_{\bf v}M$ along
$\tilde{\ga}_X(t)$ is
${\bf u}' = D(\exp(Xt)){\bf u}.$
\end{quote}
we see that this holonomy refers precisely to the {\em parallel
transport with respect to the gauge sector metric around its closed
geodesics}. For vortices relevant paths are geodesic with
respect to both metrics.
One should be aware that parity violation in the standard model means
that one must take into account the fermionic zero modes. This is done
in ref.~\cite{Gano93}.
\subsection{The Sphaleron}
Finally, we point out that the existence of the sphaleron solution in
Weinberg-Salam theory is related to the presence an embedded two
sphere in the vacuum manifold that is totally geodesic with respect to
both the gauge and scalar metrics. The dipole moment of a sphaleron is
also related to this embedding.
For $\theta_w=0$ the Ansatz
\bse
\begin{eqnarray}
\label{sph-a}
\Phi({\bf r})&=&f_{\rm sph}(r) \exp({\frac{i\pi}{2} \hat{r}_a
\si_a})\Phi_0,\\
\label{sph-b}
A_a({\bf r}) &=& g_{\rm sph}(r) \frac{i}{2} \epsilon_{abc}\hat{r}_b
\si_c,
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
constitutes a solution to Weinberg-Salam theory for suitable profile
functions $f_{\rm sph}$, $g_{\rm sph}$. It is unstable because
the boundary conditions define a topologically trivial map $S^2
\rightarrow SU(2)$. By continuity this solution is expected to persist
to non-zero $\theta_w$, with perturbations being produced in the
fields. Such a solution is referred to as the Sphaleron~\cite{Mant}.
The scalar field asymptotically maps onto the submanifold $M^W \subset
M$. Thus, as with electroweak strings, the boundary conditions
define a totally geodesic submanifold of the vacuum manifold with
respect to {\em both} the scalar and gauge metrics. One should note
that $M^W$ is the only two-dimensional submanifold of the vacuum
manifold defined so.
To prevent an electric monopole component at non-zero weak mixing angle,
the gauge field (\ref{sph-b}) deforms such that $\frac{i}{2}\si_3
\rightarrow X_Z$. This implies a preferred axis in the
$\mb{\underline{\hat{r}}}^3$-direction, with the configuration rotationally symmetric
about it. Inducing, to lowest order, dipolar perturbations
in the electromagnetic field
\begin{equation}
\label{dipole}
\de Q_a = \frac{\epsilon_{abc} \mu_b \hat{r}_c}{4 \pi r^3} \/ X_Q,
\end{equation}
with $\ul{\mu}$ parallel to $\mb{\underline{\hat{r}}}^3$, as found by substitution into
the field equations~\cite{Mant}.
\section{Discussion}
In this final section we briefly discuss some extensions to this work
and make some comments that may warrant further note.
\\
(i) {\bf The General Case}\\
The group theory in this paper can be extended to the general symmetry
breaking $G \rightarrow H$ in a fairly straightforward manner. Hence,
in general, one may expect two metrics on the vacuum manifold relating
to the scalar and gauge sector. As in Weinberg-Salam theory,
embedded vortices will be geodesic with with respect to both metrics,
and the Aharonov-Bohm scattering will relate to the holonomy of the
gauge sector metric.\\
(ii) {\bf Simplicity of Electroweak Theory}\\
Crucial to establishing the scalar and gauge metrics on the vacuum
manifold was establishing the isomorphisms with the
coset spaces $SU(2)_K \mb{\times} SU(2)_I / SU(2)_{I-K}$ and $SU(2)_I \mb{\times}
U(1)_Y/U(1)_Q$. The first of which is a symmetric space, and the
second is a non-symmetric homogenous space, interpreted as deformed
from the first. This structure is in fact quite special, and
electroweak theory constitutes the smallest dimensional example of
this. \\
\noindent
(iii) {\bf Energetics and Curvature}\\
From the metrical structure one has a corresponding curvature of the
vacuum manifold. It seems sensible that the energy of embedded
vortices should be associated with the sum of the curvatures of
the scalar and gauge metrics on the submanifold of the vacuum manifold
associated with the scalar boundary conditions of the
vortex. Coefficients of this sum should be related to those of the
Landau potential, and the value of this sum should be related to the
stability of the vortex.\\
(iv) {\bf Insensitivity to the Form of the Theory}\\
In relating the solitonic spectrum and properties to the metrical
structure of the theory one moves away from the specific details of
the Lagrangian. Thus the metrical approach relates more to the general
symmetry features of the theory rather than the specific model of
symmetry breaking.
\bigskip
{\noindent{\Large{\bf Acknowledgements.}}}
\nopagebreak
\bigskip
\nopagebreak
This work was supported in part by PPARC. N.L. acknowledges
EPSRC for a research studentship and King's College, Cambridge for
a junior research fellowship. This work was supported in part by
the European Commission under the Human Capital and Mobility program,
contract no. CHRX-CT94-0423.
\bigskip
\bigskip
{\noindent{\Large{\bf Appendix}}}
\par
\nopagebreak
\bigskip
\nopagebreak
\noindent
We provide here a quick summary of the results of \cite[chapter
X]{Nomi2} that are relevant to this work.
Consider a manifold $G/H$, where $H \subset G$ are compact Lie
groups. Then an important and relevant decomposition is the reductive
decomposition of the Lie algebras
\begin{equation}
\calG = \calH \oplus \calM,
\end{equation}
which satisfies
\bse
\begin{eqnarray}
\mb{{\rm Ad}}(H) \calH \subseteq \calH,\\
\mb{{\rm Ad}}(H) \calM \subseteq \calM.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
Here $\calM$ may be associated with the tangent space to $M$ at the
trivial coset. One should note that strictly speaking an inner product
$\inprod{\cdot}{\cdot}$ on $\calG$ is required to define the reductive
decomposition.
Given such a decomposition we can associate a $G$-invariant connection
on $G/H$ having the following properties:\\
(i)\ \ All geodesics in $M$ emanating from the trivial coset are of
the form
\begin{equation}
\ga_X(t) = \exp(Xt)H,
\end{equation}
with $X \in \calM$.\\
(ii)\ \ Considering $\exp(Xt)\subset G$, the parallel transport of a
tangent vector $Y \in \calM$ along the curve $\exp(Xt)H$ is
\begin{equation}
Y' = \mb{{\rm Ad}}(\exp(Xt))Y.
\end{equation}
(iii)\ \ The corresponding $G$-invariant metric on $G/H$ is associated
with the inner products on $\calG$ such that at the trivial coset
\begin{equation}
g(X, Y)_H = \inprod{X}{Y},
\end{equation}
identifying tangent vectors with elements of $\calM$.
By applying the above result one may examine the case with a vector
space $V$ such that $G$ acts on it by the $D$-representation. Then
$H_{\bf v}$ is the isotropy subgroup at ${\bf v} \in V$, and we
associate the manifold
\begin{equation}
M = D(G){\bf v} \cong \frac{G}{H}.
\end{equation}
Given an inner product $\inprod{\cdot}{\cdot}$ on $\calG$ we
associate a $G$-invariant connection on $M$ from the decomposition
$\calG = \calH_{\bf v} \oplus \calM$. Denoting the tangent space at
${\bf v}$ to $M$ by $T_{\bf v}M = \calM {\bf v}$, the
following properties are apparent\\
(i)\ \ All geodesics emanating from ${\bf v}$ are of the form
\begin{equation}
\tilde{\ga}_X(t) = D(\exp(Xt)){\bf v}.\\
\end{equation}
(ii) The parallel transport of ${\bf u} \in T_{\bf v}M$ along
$\tilde{\ga}_X(t)$ is
\begin{equation}
{\bf u}' = D(\exp(Xt)){\bf u}.
\end{equation}
(iii)\ \ The corresponding $G$-invariant metric on $M$ takes the
value at ${\bf v}$
\begin{equation}
g(X{\bf v}, Y{\bf v})_{\bf v} = \inprod{X}{Y}.
\end{equation}
|
\section{Introduction}
Rovibrationally excited H$_2$ molecules have been observed in
many astrophysical objects (for recent studies, see Weintraub et al. 1998;
van Dishoeck et al. 1998; Shupe et al. 1998; Bujarrabal et al. 1998;
Stanke et al. 1998).
The rovibrational levels
of the molecule may be
populated by ultraviolet pumping, by X-ray pumping, by the formation
mechanism, and by collisional excitation in shock-heated gas (Dalgarno
1995). The excited level populations are then modified by
collisions followed by quadrupole emissions.
The main colliding partners apart from H$_2$ are H and He.
Although He is
only one tenth as abundant as H, collisions with He may have a
significant influence in many astronomical environments depending
on the density, temperature and the
initial rotational and vibrational excitation of
the molecule. Collisions with He and H$_2$ are particularly important
when most of the hydrogen is in molecular form, as in dense molecular clouds.
To interpret observations of the radiation
emitted by the gas, the collision cross sections and
corresponding rate coefficients
characterizing the collisions must be known.
Emissions from excited rovibrational levels
of the molecule provide important
clues regarding the physical state of the gas, dissociation,
excitation and formation properties of H$_2$. Here we investigate
the collisional relaxation of vibrationally excited H$_2$ by He.
Rovibrational transitions in H$_2$ induced by collisions
with He atoms have been
the subject of a large number of theoretical calculations
in the past (Alexander 1976, 1977; Alexander and McGuire 1976;
Dove et al. 1980;
Eastes and Secrest 1972; Krauss and Mies 1965; McGuire and Kouri 1974;
Raczkowski et al. 1978)
and continue
to attract experimental (Audibert et al. 1976;
Michaut et al. 1998) and theoretical attention
(Flower et al. 1998; Dubernet \& Tuckey 1999; Balakrishnan et al. 1999).
Recent theoretical calculations
are motivated by the
availability of more accurate representations
of the interaction potentials
and the possibility of performing
quantum mechanical calculations with few
approximations. The potential energy surface determined by
Muchnick and Russek (1994) was used by Flower et al. (1998) and by
Balakrishnan et al. (1999) in recent
quantum mechanical calculations of rovibrational transition
rate coefficients for temperatures ranging from
100 to 5000K. Flower et al. presented their results
for vibrational levels $v=0,1$ and 2 of
ortho- and para-H$_2$. Balakrishnan
et al. (1999) reported similar results for $v=0$ and 1.
Though both authors have adopted similar close-coupling
approaches for the scattering calculations, Flower et al.
used a harmonic oscillator approximation for H$_2$ vibrational
wave functions in evaluating the matrix elements of the
potential while the calculations of Balakrishnan et al. made use of
the H$_2$ potential of Schwenke (1988) and the corresponding
numerically determined wave functions. The results of the two calculations
agreed well for pure rotational transitions but
some discrepancies were seen
for rovibrational transitions. We believe this
may be due to the different choice of vibrational
wave functions.
The sensitivity of
the rate coefficients to the choice of the H$_2$ wave function
was noted previously and differences could be
significant for excited vibrational levels. We find this to be
the case for transitions involving $v\geq 2$.
Thus, in this article, we report
rate coefficients for transitions from $v=2$ to 6 initial states
of H$_2$ induced by collisions with He atoms
using numerically exact quantum mechanical calculations.
We also report results of
quasiclassical trajectory (QCT) calculations and
examine the suitability of classical mechanical calculations in predicting
rovibrational transitions in H$_2$.
\section{Results}
The quantum mechanical calculations were performed using the
nonreactive scattering program MOLSCAT developed by Hutson and Green
(1994) with the He-H$_2$ interaction potential of Muchnick and Russek
(1994) and the H$_2$ potential of Schwenke (1988).
We refer to our earlier paper (Balakrishnan, Forrey \& Dalgarno, 1999) for
details of the numerical implementation.
Different basis sets were used in the calculations for transitions from
different initial vibrational levels. We use the
notation [$v_1$--$v_2$]($j_1$--$j_2$) to represent the basis set
where the quantities within
the square brackets give the range
of vibrational levels and those in braces give the range of
rotational levels coupled in each of the vibrational levels.
For transitions from $v=2,3$ and 4 we used, respectively,
the basis sets [0--3](0--11) \& [4](0--3),
[0--3](0--11) \& [4](0--9) and [3--5](0--11) \& [1,6](0--11). For
$v=5$ and 6 of para H$_2$ we used, respectively,
[4--6](0--14) \& [3,7](0--8)
and [5--7](0--14) \& [4,8](0--8).
During the calculations, we found that
the $\Delta v=\pm 2$ transitions are weak with cross sections
that are typically
orders of magnitude smaller
than for the $\Delta v=\pm 1$ transitions.
Thus, for $v=5$ and 6 of ortho-H$_2$, we have only included
the $\Delta v=\pm 1$ vibrational levels with $j$=0--13 in the basis set to
reduce the computational effort.
The basis sets were chosen as a compromise between
numerical efficiency and accuracy and could introduce some truncation
errors for transitions to levels which lie at the outer edge
of the basis set. Our convergence tests show that truncation errors are small.
Rovibrational transition cross sections
$\sigma_{vj,v'j'}$ where the pairs of numbers $vj$ and $v'j'$ respectively denote
the initial and final rovibrational quantum numbers, were computed for
kinetic energies ranging from 10$^{-4}$ to 3 eV.
Sufficient total angular momentum
partial waves were included in the calculations to secure convergence of the
cross sections.
The quasiclassical calculations were carried out
using the standard classical trajectory method
as described by Lepp, Buch and Dalgarno (1995) in which the
procedure of Blais and Truhlar (1976) was adopted for the final
state analysis. Because rovibrational transitions are rare
at low velocities, useful results could be obtained only for collisions
at energies above 0.1 eV. The results are averages over 10000 trajectories.
The quantum mechanical calculations were performed using MOLSCAT (Hutson \& Green, 1994)
suitably adapted for the present system with the potential
represented by a Legendre polynomial expansion in which we retained
nonvanishing terms of orders 0 to 10, inclusive.
Calculation of rovibrational transition rate coefficients over
a wide range of temperatures requires the determination of scattering
cross sections at the energies spanned by the
Boltzmann distribution at each temperature. This is a computationally demanding
problem especially when quantum mechanical calculations are required.
For many systems, QCT calculations offer a
good compromise between accuracy and computational effort. However,
the validity of classical mechanics is in question especially for
lighter systems and at lower temperatures where quantum mechanical effects such as
tunneling are important. Due to the small masses of the atoms involved,
the present system offers an excellent opportunity to test the reliability
of QCT calculations in predicting rovibrational transitions.
Though such attempts have been made in the past
(Dove et al. 1980) the classical mechanical and quantum mechanical calculations
were done in different energy regimes and a one-to-one comparison
was not possible.
We carried out quantum mechanical
and QCT calculations of rovibrational transition cross
sections for the present system over a wide range of energies.
In Figure 1 we compare the results for
pure rotational de-excitation transitions ($\Delta j=j'-j=-2$)
with $j=2$ in $v=0$ to 5.
There are striking similarities and
differences between the quantum mechanical and QCT
results.
They agree quite well at higher energies and have a similar
energy dependence with both
calculations predicting the same maximum for the cross sections before they
fall off. In contrast, the agreement becomes less satisfactory
at lower energies with the QCT cross sections rapidly
decreasing to zero as the energy is decreased. The quantum mechanical
results pass through minima
and subsequently increase with further decrease of the kinetic
energy. This is a purely quantum mechanical effect
that has important
consequences for the low temperature rate coefficients (Balakrishnan et al. 1998).
The quantum mechanical
cross sections eventually conform to an
inverse velocity dependence as the relative translational
energy is decreased to zero and the corresponding rate coefficients are
finite in the limit of zero temperature, in accordance with Wigner's threshold law.
The results illustrate that QCT calculations may be
used reliably to calculate rotational transitions for energies higher than 0.5 eV, but
at lower energies,
quantum mechanical calculation
must be employed. The energy regime for the validity of QCT results
is more restricted for rovibrational transitions
as illustrated in Figure 2 in which
we show cross sections for rovibrational transitions
$v,j\rightarrow v-1,j+2$ with $j=5$ in
$v=3$ to 5. The results indicate that
the QCT method is inadequate to calculate rovibrational
transition cross sections at impact energies below 1 eV. The QCT
results exhibit a sharp fall below a threshold near an energy of 1 eV whereas
the quantum mechanical results vary smoothly.
Similar results hold for other
transitions. The higher collision energies required for the validity
of the QCT method for rovibrational transitions compared to
pure rotational transitions is due in part to the much smaller
cross sections for rovibrational transitions making the results
more sensitive to the details of the dynamics.
Rate coefficients $k_{vj,v'j'}(T)$ were calculated for the temperature range
$T=100$ to 4000 K by averaging the cross sections over
a Boltzmann distribution of relative velocities.
Flower et al. (1998) have reported rate coefficients for rovibrational
transitions from $v=0-2$ for $T=1000, 2000$ and 4500 K.
A comparison of some of the de-excitation
rate coefficients from the $v=2$ level calculated in this paper and
those reported by Flower {\it et al.} is given in Table 1 for $T=1000$ K.
The
agreement is good for rovibrational transitions involving $|\Delta j|=0,2$ and 4
but larger differences, by factors between 2 and 4, are seen for
transitions involving larger values of $|\Delta j|$.
Our rate coefficients are generally greater than those of Flower {\it et al.} (1998)
for transitions where the discrepancy is large.
Our rate coefficient calculations extend those of
Flower {\it et al.} (1998) to include the $v=4,5$ and 6 vibrational levels. The
preferential formation of H$_2$
in these vibrational levels has been discussed by Dalgarno (1995).
In Tables 2 and 3, we present our comprehensive results for
rovibrational de-excitation transitions from different rotational
levels in para- and ortho-H$_2$ from the $v=2$ level as
functions of temperature. The corresponding excitation rate coefficients
may be obtained from detailed balance
\begin{equation}
k_{v'j',v,j}(T)=\frac{(2j+1)}{(2j'+1)} \exp{[(\epsilon_{v'j'}-
\epsilon_{vj})/k_BT]} k_{vj,v'j'}(T)
\end{equation}
where $k_B$ is the Boltzmann constant and $\epsilon_{vj}$ is the
rovibrational energy of the molecule in the $v,j$ level.
Similar results
for $v=3$ to 6 are presented in Tables 4-11. Tables 2-11 reveal
some interesting aspects of energy transfer.
It can be seen that for temperatures less than 1000 K,
rate coefficients for rovibrational transitions
involving $\Delta j=-4\Delta v$
predominate over
other transitions where changes in both $v$ and $j$ occur.
This is clearly seen for ($vj,v'j'$)
transitions with $\Delta v=v'-v=-1$
for $j=3-7$ and the effect becomes stronger with increasing
rotational excitation but less important as $T$ increases.
This is an example of quasi-resonant scattering
(Stewart et al. 1988; Forrey et al. 1999). A detailed
study of this process in the limit of zero temperature
and its
correspondence with classical mechanics has been carried out recently (Forrey et al. 1999).
The efficient conversion of vibrational energy into rotational energy
may produce a significant population of high rotational levels
in environments where molecular hydrogen is subjected to an intense flux
of X-rays or ultraviolet photons.
\section{Acknowledgments}
This work was supported by the
National Science Foundation (NSF), Division of Astronomy. MV was supported
by the NSF through the Research Experience for Undergraduates
program at the Smithsonian Astrophysical Observatory.
\clearpage
|
\section{Introduction}
\label{sec:intro}
In Supersymmetric (SUSY)~\cite{ref:SUSY} models each elementary
particle is accompanied by a
supersymmetric partner whose spin differs by half a unit.
Most of the searches for these supersymmetric particles (``sparticles'')
are performed within the Minimal Supersymmetric extension of the
Standard Model\ (MSSM)~\cite{ref:MSSM}, assuming $R$-parity\ conservation.
$R$-parity~\cite{ref:rparity} is a new multiplicative quantum number defined as
$R_{p}=(-1)^{2S+3B+L}$ where $S$, $B$ and $L$ are the spin, baryon and
lepton number of the particle, respectively.
$R$-parity\ discriminates between ordinary
and supersymmetric particles: \Rp\ = +1 for the Standard Model\ particles
and \Rp\ = --1 for their supersymmetric partners.
$R$-parity\ conservation implies that supersymmetric particles are
always pair produced and always decay through
cascade decays to ordinary particles and the lightest supersymmetric
particle (LSP). In this context, the LSP is often assumed to be
the lightest neutralino, $\neutralino$, which is then expected to be
stable and to escape detection due to its weakly interacting nature.
The characteristic signature of the supersymmetric $R$-parity\
conserving decays is therefore missing energy.
In this paper, the possible direct manifestations of $R$-parity\
breaking couplings via processes with
distinct signatures are studied.
If $R$-parity\ is violated, sparticles can decay directly to Standard Model\
particles. Therefore, the signatures sought in the analyses of
this paper differ from the
missing energy signatures of $R$-parity\ conserving processes.
With the MSSM particle content, $R$-parity\ violating interactions are
parametrised with a gauge-invariant superpotential that includes the
following Yukawa coupling terms~\cite{ref:dreiner1}:
\begin{eqnarray}
{\sl W}_{RPV} =
\lambda_{ijk} L_i L_j {\overline E}_k
+ \lambda^{'}_{ijk} L_i Q_j {\overline D}_k
+ \lambda^{''}_{ijk} {\overline U}_i {\overline D}_j {\overline D}_k,
\label{lagrangian}
\end{eqnarray}
where $i,j,k$ are the generation indices of the superfields
$L, Q,E,D$ and $U$. $L$ and $Q$ are lepton and quark left-handed doublets,
respectively.
$\overline E$, $\overline D$ and $\overline U$ are right-handed
singlet charge-conjugate superfields for the charged
leptons and down- and up-type quarks, respectively.
The interactions corresponding to these superpotential terms are
assumed to respect the gauge symmetry SU(3)$_{\rm C}$ $\times$
SU(2)$_{\rm L}$ $\times$ U(1)$_{\rm Y}$ of the Standard Model.
The \lb$_{ijk}$ are
non-vanishing only if $i < j$, so that at least two different
generations are coupled in the purely leptonic vertices.
The \lbpp$_{ijk}$ are
are non-vanishing only for for $ j < k $.
The \lb\ and \lbp\ couplings both violate lepton number ($L$) conservation
and the \lbpp\ couplings violate baryon number ($B$) conservation.
There are nine \lb\ couplings for the triple lepton vertices, 27 \lbp\
couplings for the lepton-quark-quark vertices and nine \lbpp\ couplings for the
triple quark vertices. There are therefore a total of 45 new
$R$-parity\ violating couplings.
In the constrained MSSM framework~\footnote{
The constrained MSSM implies a common gaugino mass
and a common sfermion mass at the GUT scale.},
there are five initial parameters completely determining all sparticle masses
and couplings.
Recently, supersymmetric models with $R$-parity\ violation (RPV) have
attracted considerable theoretical and phenomenological interest
(see for instance~\cite{ref:dreiner1}).
Indeed, there exist no theoretical or experimental arguments excluding
$R$-parity\ violation~\cite{ref:opal_rpv_gauginos,ref:aleph_rpv_lle,
ref:aleph_rpv_lqd}.
Therefore, it is
important to consider the phenomenology of possible $R$-parity\
violating scenarios. The branching ratios of some of the $R$-parity\
violating decay modes can be comparable or even larger than
$R$-parity\ conserving modes. For example, this could be the case
for the scalar top
quark (``stop") decay modes to third-generation fermions.
From the experimental point of view, there are several
upper bounds\footnote{All quoted limits are given for a sparticle mass
of 100~GeV.}
on the $R$-parity\ violating Yukawa couplings, \lb, \lbp\ and \lbpp.
A list of upper limits on individual couplings
can be found
in~\cite{ref:bhatta,ref:barger1,ref:agashe,ref:godbole,ref:ellis}.
Most of the upper limits on the couplings are of
${\cal{O}} (10^{-2})$, but there also exist
some more stringent limits.
For instance, \lbp$_{111} < 10^{-4}$ from
neutrinoless double beta decay~\cite{ref:mohapatra},
\lbpp$_{112} < 10^{-6}$~\cite{ref:goity} from double
nucleon decay and \lbpp$_{113} < 10^{-4}$~\cite{ref:goity} from
limits on
$n-\overline{n}$ oscillation.
Most of the couplings are constrained by
experimental results but
most of these upper bounds are still high compared to the sensitivity
attainable with direct searches at LEP (of ${\cal{O}} (10^{-5})$).
Furthermore the simultaneous presence
of the couplings \lbpp\ ($B$-violating) and \lbp\ ($L$-violating) is
forbidden since it would allow fast squark-mediated proton decay at
tree level. The experimental non-observation of proton decay
places strong bounds on the product of these two couplings, i.e.,
\lbp $\times$ \lbpp $<$ 10$^{-10}$~\cite{ref:smir}.
Although pair production is not required with $R$-parity\ violation,
only searches for $R$-parity\ violating decays of pair-produced scalar
fermions (``sfermions"), such as the charged and neutral scalar leptons
and scalar top quark,
are presented in this paper.
Their production is fully determined by gauge couplings and their masses.
Supersymmetric particles can also be singly produced and, for example,
indirect limits from the OPAL two-fermion pair-production
cross-section measurements are given in~\cite{opal_2fermion}.
Two different scenarios are probed. In the first scenario, the decays
of sfermions via the lightest neutralino, $\tilde{\chi}^0_1$, are considered,
where $\tilde{\chi}^0_1$ is
treated as the LSP and assumed to decay via an $R$-parity\ violating
interaction.
These are denoted ``indirect decays''.
SUSY cascade decays via particles other than the LSP are not
considered.
In the second scenario,
``direct'' decays
of sparticles to Standard Model\ particles are investigated.
In this case, the sparticle considered is assumed
to be the LSP, such that $R$-parity\ conserving decay modes do not
contribute.
In both scenarios, it is assumed that only one of the
45 Yukawa-like couplings is non zero at a time.
Only values of the Yukawa-like \lb-couplings larger than
$ {\cal{O}} (10^{-5})$ are relevant to this analysis.
For smaller couplings, the lifetime of sparticles would be
sufficiently long
to produce a secondary decay vertex, clearly detached
from the primary vertex, or even outside the detector. These topologies
have not been considered in this paper, but decays outside the
detector have been treated elsewhere \cite{ref:stable-part}.
In this paper, the data produced in ${\mathrm e}^+ {\mathrm e}^-$ collisions at LEP and
collected with the OPAL detector
during 1997 at a
centre-of-mass energy of
$\sqrt{s} \simeq$183~GeV are analysed. These data correspond
to an integrated luminosity of
about 56~pb$^{-1}$.
The production and $R$-parity\ violating decays
of $\tilde{\ell}$, $\tilde{\nu}$ via \lb\ and \lbp\ and $\stopx$ via \lbp\ and
\lbpp\ are described
in Section~\ref{sec:production},
together with the possible signal topologies
resulting from these processes. The
signal and background Monte Carlo simulations used in the
different analyses are described in Section~\ref{sec:MC},
and a short description of the
OPAL detector follows in Section~\ref{sec:opaldet}.
Sections~\ref{sec:multileptons}, \ref{sec:2jets2leptons},
\ref{sec:jetsleptons}
,\ref{sec:multijetsemiss} and \ref{sec:multijetsnoemiss}
describe the specific analyses optimised to search for $R$-parity\ violating
processes. The physics interpretation is given in
Section~\ref{sec:results} which presents
cross-section limits and interpretations in the MSSM.
\section{Sparticle Production and Decays}
\label{sec:production}
In this section, the production and decay modes of different sfermion
species are discussed.
The decay modes that result from \lb, \lbp\ and \lbpp\
couplings are presented.
Table~\ref{tab:relation} summarises
the production and decay mechanisms as well as the coupling
involved in the decay,
the final state topologies searched for,
and the analysis names as used in
Sections~\ref{sec:multileptons}, \ref{sec:2jets2leptons},
\ref{sec:jetsleptons},
\ref{sec:multijetsemiss} and \ref{sec:multijetsnoemiss}.
In the indirect decays, the particles
resulting from the $\tilde{\chi}^0_1$ decay are put in parentheses.
\begin{table}[htbp]
\begin{center}
\begin{tabular}{|ll|l|l|c|}
\hline
Production and Decay & & Coupling & Topology & Analysis \\
\hline
\hline
$ \tilde{\ell}^+ \tilde{\ell}^- \rightarrow $ &
$ \nu \ell $ $ \nu \ell $ & \lb\ direct &
2 $\ell$ + $E_{miss}$
& (A) \ref{sec:2leptons}\\
\hline
$ \tilde{\nu} \tilde{\nu} \rightarrow $ $\nu \chin $ $ \nu \chin \ra $ &
$\nu (\nu \ell^+ \ell^-)$ $\nu (\nu \ell^+ \ell^-)$ &
\lb\ indirect &
4 $\ell$ + $E_{miss}$ &
(B) \ref{sec:4leptons} \\
$ \tilde{\nu}\snu \rightarrow $ &
$ \ell^+ \ell^- $ $ \ell^+ \ell^- $ & \lb\ direct &
4 $\ell$ &
(C) \ref{sec:4leptons} \\
\hline
$ \tilde{\ell}^+ \tilde{\ell}^- \rightarrow $
$ \ell^+ \chin \ell^- \chin$ $\ra$ &
$\ell^+ (\nu \ell^+ \ell^-)$ $\ell^- (\nu \ell^+ \ell^-)$ & \lb\ indirect &
6 $\ell$ + $E_{miss}$ &
(D) \ref{sec:6leptons} \\
\hline
\hline
$ \stoppair \rightarrow $ &
$ e^+ q $ $ e^- q $ & \lbp\ direct &
2 e + 2 jets &
(E) \ref{sec:stopemu} \\
$ \stoppair \rightarrow $ &
$ \mu^+ q $ $ \mu^- q $ & \lbp\ direct &
2 $\mu$ + 2 jets &
(E) \ref{sec:stopemu} \\
$ \stoppair \rightarrow $ &
$ \tau^+ q $ $ \tau^- q $ & \lbp\ direct &
2 $\tau$ + 2 jets &
(F) \ref{sec:stoptau} \\
\hline
\hline
$ \tilde{\tau}^+ \tilde{\tau}^- \rightarrow $
$ \tau^+ \chin \tau^- \chin$ $\ra$ &
$\tau^+ (\ell q q)$ $\tau^- (\ell q q)$ & \lbp\ indirect &
$\tau$ + jets &
(F) \ref{sec:jetstau} \\
&
$ \tau^+ (\nu q q)$ $ \tau^- (\ell q q)$ & \lbp\ indirect &
$\tau$ + jets &
(F) \ref{sec:jetstau} \\
&
$ \tau^+ (\nu q q)$ $ \tau^- (\nu q q)$ & \lbp\ indirect &
$\tau$ + jets &
(F) \ref{sec:jetstau} \\
$ \tilde{\ell}^+ \tilde{\ell}^- \rightarrow $
$ \ell^+ \chin \ell^- \chin$ $\ra$ &
$ \ell^+ (\ell q q)$ $ \ell^- (\ell q q)$ & \lbp\ indirect &
$\ell$ + jets &
(G) \ref{sec:jetsemu} \\
&
$ \ell^+ (\nu q q)$ $\ell^- (\ell q q)$ & \lbp\ indirect &
$\ell$ + jets &
(G) \ref{sec:jetsemu} \\
&
$ \ell^+ (\nu q q) $ $\ell^- (\nu q q)$ & \lbp\ indirect &
$\ell$ + jets &
(G) \ref{sec:jetsemu} \\
\hline
\hline
$ \tilde{\nu} \tilde{\nu} \rightarrow $ $\nu \chin $ $ \nu \chin \ra $ &
$ \nu (\nu q q)$ $\nu (\nu q q )$ & \lbp\ indirect &
4 jets + $E_{miss}$ &
(H) \ref{sec:multijetsemiss} \\
\hline
$ \tilde{\nu} \tilde{\nu} \rightarrow$ &
$ q q \; q q$ & \lbp\ direct &
4 jets &
(I) \ref{sec:multijetsnoemiss-sl} \\
$ \tilde{\ell}^+ \tilde{\ell}^- \rightarrow $ &
$q q \; q q$ & \lbp\ direct &
4 jets &
(I) \ref{sec:multijetsnoemiss-sl} \\
$ \tilde{q}\bar{\tilde{q}} \rightarrow $ &
$q q \; q q$ & \lbpp\ direct &
4 jets &
(I) \ref{sec:multijetsnoemiss-sq} \\
\hline
\end{tabular}
\end{center}
\caption{\it
List of production and decay mechanisms
of the channels that are covered by the various analyses
described in this paper. The couplings and decay type searched for in
each analysis and the corresponding topologies are described in the
second and third columns, respectively.
The corresponding section number is indicated
in the last column.}
\label{tab:relation}
\end{table}
The charged lepton, $\ell^\pm$, is either an electron or a muon.
Different analyses are applied when the charged lepton is an electron or a
muon (denoted ``electron channel" and ``muon channel") or when
it is a tau (denoted ``tau channel"). Each analysis is optimised
regarding the number of jets or charged leptons expected in the final states.
If the mass of the scalar charged lepton (``slepton")
is less than the beam energy, sleptons may be
pair produced in electron-positron collisions
through $s$-channel processes involving a $\Zboson$
or a $\gamma$. Scalar electrons (``selectrons," $\tilde{\mathrm e}$)
may also be produced through $t$-channel neutralino exchange. This may
enhance their production cross-section compared to those for the
scalar muons (``smuons," $\tilde{\mu}$) and scalar taus (``staus," $\tilde{\tau}$).
Similarly, neutral scalar leptons (``sneutrinos'') may be
pair-produced via the $s$-channel or through $t$-channel chargino exchange.
Sleptons and sneutrinos may decay directly to Standard Model\ particles through
the \lb$_{ijk} L_i L_j {\overline E}_k$ operator. The possible
decays are:
\begin{center}
$\tilde{\ell}^-_{iL} \rightarrow \overline{\nu}_j \ell^-_k$, \quad
$\tilde{\ell}^-_{jL} \rightarrow \overline{\nu}_i \ell^-_k$, \quad
$\tilde{\ell}^-_{kR} \rightarrow \nu_i \ell^-_j, \nu_j \ell^-_i$
\end{center}
\begin{center}
$\tilde{\nu}_{i} \rightarrow \ell^+_j \ell^-_k$, \quad
$\tilde{\nu}_{j} \rightarrow \ell^+_i \ell^-_k$
\end{center}
where $\tilde{\ell}^-_{iL}$ denotes a left-handed slepton of the $i^{th}$
generation and $\tilde{\ell}^-_{kR}$ denotes a right-handed slepton of the $k^{th}$
generation.
If the slepton or sneutrino decays directly via
the \lbp$_{ijk} L_i Q_j {\overline D}_k$
operator\footnote{
Right-handed sleptons cannot decay via the operator
\lbp$_{ijk} L_i Q_j {\overline D}_k$.}, the decay modes are:
\begin{center}
$\tilde{\ell}^-_{iL} \rightarrow \overline{u}_j d_k$, \quad
$\tilde{\nu}_{iL} \rightarrow \overline{d}_j d_k$
\end{center}
where $d_{k}$ denotes a down-type quark of the $k^{th}$ generation,
$u_{j}$ denotes an up-type quark of the $j^{th}$ generation and
$d_{j}$ denotes a down-type quark of the $j^{th}$ generation.
Sleptons and sneutrinos may also decay indirectly
to $\tilde{\chi}^0_1$ plus the corresponding charged or neutral
lepton\footnote{Decays like $\tilde{\ell} \rightarrow \tilde{\chi}^0_2 \ell$ or
$\tilde{\ell} \rightarrow \chargino \nu$ are not considered here but the
appropriate branching ratios are taken into account for
interpretation of the results.}:
\begin{center}
$\tilde{\ell} \rightarrow \tilde{\chi}^0_1 \ell$, \quad
$\tilde{\nu} \rightarrow \tilde{\chi}^0_1 \nu$
\end{center}
The $\tilde{\chi}^0_1$ may
subsequently decay violating $R$-parity\ with a \lb, \lbp\ or \lbpp\ coupling
through an intermediate slepton or sneutrino.
In the case of a non-vanishing \lb\ coupling,
the $\neutralino$ decays proceeding via the
\lb$_{ijk} L_i L_j {\overline E}_k$ operator are:
\begin{center}
$\tilde{\chi}^0_1 \rightarrow \ell^-_i \nu_j \ell^+_k$, \quad
$\tilde{\chi}^0_1 \rightarrow \ell^+_i \overline{\nu}_j \ell^-_k$, \quad
$\tilde{\chi}^0_1 \rightarrow \nu_i \ell^-_j \ell^+_k$, \quad
$\tilde{\chi}^0_1 \rightarrow \overline{\nu}_i \ell^+_j \ell^-_k$
\end{center}
In the case of a non-vanishing \lbp\ coupling, the
$\neutralino$ decays proceeding via the
\lbp$_{ijk} L_i Q_j {\overline D}_k$ operator are:
\begin{center}
$\tilde{\chi}^0_1 \rightarrow \ell^-_i u_j \overline{d}_k$, \quad
$\tilde{\chi}^0_1 \rightarrow \ell^+_i \overline{u}_j d_k$, \quad
$\tilde{\chi}^0_1 \rightarrow \nu_i d_j \overline{d}_k$, \quad
$\tilde{\chi}^0_1 \rightarrow \overline{\nu}_i \overline{d}_j d_k$,
\end{center}
In the case of a non-vanishing \lbpp\ coupling, the
$\neutralino$ decays proceeding via the
\lbpp$_{ijk} {\overline U}_i {\overline D}_j {\overline D}_k$ operator
are:
\begin{center}
$\tilde{\chi}^0_1 \rightarrow u_i d_j d_k$, \quad
$\tilde{\chi}^0_1 \rightarrow {\overline u}_i {\overline d}_j {\overline d_k}$
\end{center}
If the mass of the scalar top quark (``stop") is smaller than the beam energy,
stop quarks may be produced in pairs in e$^+$e$^-$ collisions
via $s$-channel $\Zboson$ or $\gamma$ exchange.
Due to the mixing of the left- and right-handed stop,
$ \stopl $ and $ \stopr $, the observable
$ \stopm = \stopl \cos \theta_{ \stopx\ } +
\stopr \sin \theta_{ \stopx\ } $
could become very light, even the lightest
supersymmetric particle.
The coupling of the $ \stopm $
to the $ \Zboson $
boson is determined by the mixing angle $ \theta_{ \stopx\ } $,
whose value
is determined by the top quark mass and the soft SUSY breaking parameters.
The $ \stopm $ decouples from the $\Zboson $ if
$\cos^2 \theta_{ \stopx } = \frac{4}{3} \sin^2 \bar{\theta}_{\rm{W}}$
($\theta_{ \stopx } \simeq 0.98 $ radian),
where
$\bar{\theta}_{\rm{W}}$ is the effective weak mixing angle.
For this value of $\theta_{\stopx}$,
$ \stoppair $
may only be produced via a virtual
$\gamma$ and the expected cross-section is therefore reduced.
For the purpose of $R$-parity\ violating searches, the stop quark is
assumed to be the lightest supersymmetric particle and only direct
decays are considered.
Only 9 of the 27 \lbp\ parameters are relevant:
$\lambda_{i3k}^{'}$, $i,k = 1,2,3$,
as the stop is contained in the
SU(2) doublet field but not in the down type singlet field.
If the stop decays via
the \lbp$_{ijk} L_i Q_j {\overline D}_k$ operator,
the decay modes are:
\begin{center}
$\stopx_{jL} \rightarrow \ell^+_i d_k$
\end{center}
If the stop decays via
the \lbpp$ {\overline U}_i {\overline D}_j {\overline D}_k $
operator, the decay modes are:
\begin{center}
$\stopx_{iR} \rightarrow \overline{d}_j \; \overline{d}_k$
\end{center}
Under the assumption of $R$-parity\ violation, the strength of the coupling
and the decay width of a sfermion are determined only by its mass and
the \lb, \lbp\ and \lbpp\ parameters
if the sparticle is the LSP.
If the sparticle is not the LSP, both the $R$-parity\ conserving and
the $R$-parity\ violating decay modes are accessible.
In the analyses described in this paper, tracks are required to come from the
interaction vertex. Analyses would become inefficient for decay lengths
larger than some centimeters. For very long lifetimes, the LSP decays
outside the detector, and in the case it is neutral,
the event topology would be exactly the same
as the \Rp\ conserving case.
For sleptons and sneutrinos, the decay widths
are given by~\cite{ref:barger2,ref:carena}:
\[
\Gamma (\tilde{\ell}^-_{i} \rightarrow \overline{\nu}_j \ell^-_k,
\tilde{\nu}_i \rightarrow \ell^+_j \ell^-_k) =
\frac{1}{16 \pi} \lambda^2_{ijk}
m_{\tilde{\ell}, \tilde{\nu}} \; , \qquad
\Gamma (\tilde{\ell}^-_{i} \rightarrow \overline{u}_j d_k,
\tilde{\nu}_i \rightarrow \bar{d}_j d_k) =
\frac{3}{16 \pi} \lambda^{'}{}^2_{ijk}
m_{\tilde{\ell},\tilde{\nu} } \; ,
\]
neglecting quark and lepton masses.
Similarly, the $R$-parity\ violating decay of the stop has a decay
width~\cite{ref:dreiner2} of:
\[
\Gamma (\stopl \rightarrow \ell^+_i d_k) =
\frac{1}{16 \pi} \lambda{'}{}^2_{i3k}
m_{\stopx\ }
\]
Under the conservative assumptions of a sparticle mass of
45 GeV\footnote{This number takes into account the indirect limits obtained
from the study of the Z$^0$ width at LEP1.}
and a decay length of 0.1 mm the analyses presented in this paper
would be sensitive to
\lb\ couplings larger than ${\cal O} (10^{-5})$.
\section{Monte Carlo Simulation}
\label{sec:MC}
Monte Carlo samples corresponding to the
charged slepton, sneutrino and stop
pair-production processes as well as
Monte Carlo samples used to estimate the background levels
due to Standard Model processes\ were simulated.
All generated events were processed
through the full simulation of the OPAL detector~\cite{ref:GOPAL},
and the same analysis chain was applied to simulated events
as to the data.
The simulation of the signal events has been done at $\sqrt{s}$~=183~GeV
with the Monte Carlo program SUSYGEN~\cite{ref:SUSYGEN}.
Charged and neutral sleptons decaying directly or indirectly
via \lb\ or \lbp\ have been produced
for the mass values of 45, 70 and 90 GeV. Five
masses (45, 60, 75, 80 and 90 GeV) were used for the sneutrino direct
decays via \lbp. Stop events were simulated at 6 different stop
masses (45, 55, 65, 75, 85 and 90 GeV).
Samples of 1000 or 2000 events were generated for each relevant coupling.
For the indirect decays, events were produced with
$\Delta m = m_{\tilde{\mathrm f}} - m_{\tilde{\chi}^0_1} = m_{\tilde{\mathrm f}}/2$.
Additional samples were simulated for $m_{\tilde{\mathrm f}}$ = 90~GeV and
$\Delta m = m_{\tilde{\mathrm f}} - m_{\tilde{\chi}^0_1} = 5$~GeV to account for changes in
the event topologies from the model parameters. The values of
$\Delta m$ were chosen to cover a large range for a limited number of
Monte Carlo events.
To estimate the systematic errors related to different
gaugino mixings, extra samples of pair-produced selectrons and
electron-sneutrinos were simulated with five
different sets of SUSY parameters.
Events were produced for each of the nine possible \lb\
couplings. Events were also simulated for each lepton flavour
corresponding to the first index of
\lbp. The quark flavour corresponding to the second and third index
of \lbp\ were fixed to the first and second generation, with a few
samples containing bottom qaurks for systematic checks.
\label{sec:higgs}
For the stop decaying via the \lbp\ coupling into a quark and a lepton
all nine combinations of quark and lepton flavours
in the final state were generated.
The production and decay of the stop is simulated as described
in~\cite{ref:stoppaper}.
The stops are hadronised to form colourless hadrons
and associated fragmentation particles,
according to the Lund string fragmentation scheme
(JETSET 7.4)~\cite{ref:JETSET1, ref:JETSET2}.
For the decay, a colour string was stretched between the spectator
quark and the quark from the stop decay. Further hadronisation
was also done using the Lund scheme.
The fragmentation function of Peterson~\cite{ref:peterson} has been used.
Events were simulated
with the mixing angle $\theta_{\stopx\ }$
set to zero.
The main sources of background arise from
Standard Model\ four-fermion,
two-photon and two-fermion (lepton-pair and multi-hadronic)
processes.
For two-photon processes, the PHOJET~\cite{ref:PHOJET} and
HERWIG~\cite{ref:herwig} generators have been used to simulate
hadronic final states.
The Vermaseren~\cite{ref:VERMASEREN} generator was
used to estimate the background contribution from all
two-photon ${\mathrm e}^+ {\mathrm e}^- \ell^+ \ell^-$
final states.
All other four-fermion final states, other than two-photon
${\mathrm e}^+ {\mathrm e}^- \ell^+ \ell^-$, were simulated
with grc4f~\cite{ref:grace4f}, which takes into
account all interfering four-fermion diagrams.
For the two-fermion final states, BHWIDE~\cite{ref:BHWIDE}
was used for the ee$(\gamma)$
final state and KORALZ~\cite{ref:KORALZ} for the
$\mu \mu$ and the $\tau \tau$ states. The multi-hadronic events,
${\rm qq}(\gamma)$,
were simulated using PYTHIA~\cite{ref:JETSET1}.
For small contributions to background final states with six or more
primary fermions, no Monte Carlo generator exists. These final states
are therefore not included in the background Monte Carlo
samples. Consequently, the background could be slightly
underestimated, which would lead to a conservative approach when
calculating upper bounds applying background subtraction.
\section{The OPAL Detector}
\label{sec:opaldet}
A complete description of the OPAL detector can be found
in Ref.~\cite{ref:OPAL-detector} and only a brief overview is given here.
The central detector consists of
a system of tracking chambers
providing charged particle tracking
over 96\% of the full solid
angle\footnote
{The OPAL coordinate system is defined so that the $z$ axis is in the
direction of the electron beam, the $x$ axis is horizontal
and points towards the centre of the LEP ring, and
$\theta$ and $\phi$
are the polar and azimuthal angles, defined relative to the
$+z$- and $+x$-axes, respectively. The radial coordinate is denoted
as $r$.}
inside a 0.435~T uniform magnetic field parallel to the beam axis.
It is composed of a two-layer
silicon microstrip vertex detector, a high precision drift chamber,
a large volume jet chamber and a set of $z$ chambers measuring
the track coordinates along the beam direction.
A lead-glass electromagnetic (EM)
calorimeter located outside the magnet coil
covers the full azimuthal range with excellent hermeticity
in the polar angle range of $|\cos \theta |<0.82$ for the barrel
region and $0.81<|\cos \theta |<0.984$ for the endcap region.
The magnet return yoke is instrumented for hadron calorimetry (HCAL)
and consists of barrel and endcap sections along with pole tip detectors that
together cover the region $|\cos \theta |<0.99$.
Four layers of muon chambers
cover the outside of the hadron calorimeter.
Electromagnetic calorimeters close to the beam axis
complete the geometrical acceptance down to 24 mrad, except
for the regions where a tungsten shield is present to protect
the detectors from synchrotron radiation.
These include
the forward detectors (FD) which are
lead-scintillator sandwich calorimeters and, at smaller angles,
silicon tungsten calorimeters (SW)~\cite{ref:SW}
located on both sides of the interaction point.
The gap between the endcap EM calorimeter and the FD
is instrumented with an additional lead-scintillator
electromagnetic calorimeter, called the gamma-catcher.
To be considered in the analyses, tracks in the central detector and clusters
in the electromagnetic calorimeter were required to satisfy the normal quality
criteria employed in OPAL's analysis of Standard Model (SM) lepton
pairs~\cite{ref:leptpairs}.
\section{Multi-lepton Final States}
\label{sec:multileptons}
This section describes the searches for purely leptonic final states
that may result from pair production of neutral or charged sleptons,
involving subsequent direct or indirect \lb\ decays
(see Table~\ref{tab:relation}).
\subsection{Event and Track Selection}
\label{sec:preselections}
The event preselection and lepton identification are
described in~\cite{ref:slept161}.
Multi-hadronic, cosmic and Bhabha
scattering events were vetoed~\cite{ref:slept161}.
At the preselection level, it was also required that
the ratio of the number of tracks satisfying the
quality criteria described in~\cite{ref:leptpairs}
to the total number of reconstructed tracks be greater than 0.2
to reduce backgrounds from beam-gas and beam-wall events.
The visible energy, the visible mass and the total transverse momentum
of the event were calculated using the method
described in~\cite{ref:OPAL-Higgs}.
Finally, the number of good charged tracks
was required to be at least two.
Only tracks with $|\cos \theta| <$ 0.95 were considered
for lepton identification. A track was considered ``isolated''
if the total energy of other charged tracks within
$10\degree$ of
the lepton candidate was less than 2~GeV.
A track was selected as an electron candidate if one of the following three
algorithms was satisfied: {\it (i)} the output probability of the
neural net algorithm described in
\cite{ref:NN} was larger than 0.8;
{\it (ii)} the electron selection
algorithm as described in \cite{ref:elecbarrel}
for the barrel region or in \cite{ref:elecendcap} for
the endcap region was satisfied;
{\it (iii)} $0.5 < E/p < 2.0$, where $p$ is
the momentum of the electron candidate
and $E$ is the energy of the electromagnetic calorimeter cluster
associated with the track.
A track was selected as a muon candidate according to the
criteria employed in OPAL's
analysis of Standard Model\ muon pairs~\cite{ref:leptpairs}.
That is, the track had associated activity in the muon chambers or hadron
calorimeter strips or it had a high momentum but was associated with only a
small energy deposit in the electromagnetic calorimeter.
Tau candidates were selected by requiring that there were at most three
tracks within a 35$^\circ$ cone.
The invariant mass computed using all good tracks and EM clusters
within the above cone had to be less than 3~GeV.
For muon and electron candidates, the momentum was estimated
from the charged track momentum measured in the central detector, while
for tau candidates the momentum was estimated from the vector sum of
the measured momenta of the charged tracks within the tau cone.
Tracks resulting from photon conversion
were rejected using the algorithm described in
\cite{ref:conversion}. For the two- and six-lepton final states,
the large background from two-photon
processes was reduced by requiring that
the total energy deposited in each silicon tungsten calorimeter be
less than 5~GeV, be less than 5~GeV in each forward calorimeter, and
be less than 5~GeV in each side of the gamma-catcher.
In addition to the requirement that there be no
unassociated electromagnetic cluster with an energy larger
than
25~GeV in the event, it was also required that there be no
unassociated
hadronic clusters with an energy larger than 10~GeV.
\subsection{Final States with Two Leptons plus Missing Energy}
\label{sec:2leptons}
Final states with two charged leptons and missing energy may
result from direct slepton decays via a \lb\ coupling.
The analysis
was optimised to retain good signal efficiency
while reducing the background, mainly due to
$\ell \ell \nu \nu$ final states from $\WW$ production and to
two-photon processes.
The following criteria were applied in addition to those described in
Section~\ref{sec:preselections}.
\begin{description}
\item[(A1)]
Events had to contain
exactly two identified
and oppositely-charged leptons, each with a transverse
momentum with respect to the beam axis greater than 2~GeV.
\item[(A2)]
The background from two-photon processes and ``radiative return" events
(${\mathrm e}^+ {\mathrm e}^- \ra {\mathrm Z} \gamma$, where the $\gamma$ escapes
down the beam pipe) was reduced by requiring
that the polar angle of the missing momentum, $\thmiss$,
satisfy \mbox{$\cosmiss < 0.9$}.
\item[(A3)]
To reduce further the residual background from
Standard Model lepton pair events,
it was required that $\Mvis /\sqrt{s} < 0.80$,
where $\Mvis$ is the event visible mass.
\item[(A4)]
The acoplanarity angle\footnote
{The acoplanarity angle, $\mbox{$\phi_{\mathrm{acop}}$}$,
is defined as 180$\degree$ minus the angle
between the two lepton momentum vectors
projected
into the $x-y$ plane.}
($\mbox{$\phi_{\mathrm{acop}}$}$) between the two leptons was required to be
greater than 10$\degree$ in order to reject Standard Model leptonic events,
and smaller than 175$\degree$ in order to reduce the
background due to photon conversions.
The acoplanarity angle distribution is shown
in Figure~\ref{fig:multilepton}~(a)
after cuts (A1) to (A3).
The acollinearity angle\footnote
{The acollinearity angle, $\mbox{$\theta_{\mathrm{acol}}$}$,
is defined as 180$\degree$ minus the space-angle
between the two lepton momentum vectors.}
($\mbox{$\theta_{\mathrm{acol}}$}$) was also required to be greater than 10$\degree$
and smaller than 175$\degree$.
\item[(A5)]
Cuts on $a_t^{\mathrm {miss}}$
and $p_t^{\mathrm {miss}}$ were applied;
$a_t^{\mathrm {miss}}$ is the component
of the missing momentum vector perpendicular to the
event thrust axis in the
plane transverse to the beam axis and $p_t^{\mathrm {miss}}$
is the missing transverse momentum.
The cuts on $a_t^{\mathrm {miss}}$ and $p_t^{\mathrm {miss}}$ are complementary
and reject some two-photon events with high transverse momentum.
The full description of these cuts
can be found in~\cite{ref:slept161}.
\end{description}
In order to maximise the detection efficiencies,
events were accepted if they passed the above selection criteria or if
they passed the selection of $\WW$ pair
events~\cite{ref:wwpaper} where both W's decay leptonically.
The preselection and detector status criteria described in
Section~\ref{sec:preselections} were imposed in both cases.
There are 75 events selected with
79.7 events expected from all Standard Model processes considered
(75.2 from $\WW$ events).
\begin{description}
\item[(A6)]
At this stage the background from two-photon processes and
$\WW$ production was reduced by categorizing the events
in different classes
according to the flavour of the leptons expected in the final state,
as can be seen in Table~\ref{tab:2leptons}.
Events were further selected by applying cuts on the momentum
of the two leptons as described in~\cite{ref:slept161} in both the right-
and left-handed slepton searches.
\end{description}
\begin{table}[htbp]
\begin{center}
\begin{tabular}{|r||c||c|c|c|}
\hline
Final State & Eff. (\%) & Selected Events & Tot. bkg MC & 4-f \\
\hline
\hline
$ee + E_{T {\mathrm miss}}$ & 58--76 & 11 & 13.8 & 13.5 \\
$\mu\mu + E_{T {\mathrm miss}}$ & 57--81 & 10 & 11.3 & 11.0 \\
$\tau\tau + E_{T {\mathrm miss}}$ & 30--50 & 10 & 15.5 & 12.5 \\
$ee$ or $e\mu$ or $\mu\mu + E_{T {\mathrm miss}}$ & 65--80 & 39 & 52.2 & 51.0 \\
$ee$ or $e\tau$ or $\tau\tau + E_{T {\mathrm miss}}$ & 58--71 & 39 & 51.9 & 48.6 \\
$\mu\mu$ or $\mu\tau$ or $\tau\tau + E_{T {\mathrm miss}}$ & 58--72 & 40 & 52.0 & 48.6 \\
\hline
\end{tabular}
\end{center}
\caption{\it
Detection efficiencies (in \%), events selected and background
predicted for the lepton-pair plus missing energy channel and
for slepton masses between 45 and 90~GeV.
The deficit of events selected in the data compared to the background
expectations is interpreted as a downward statistical fluctuation.
The number of events in the last three rows are largely correlated,
as many final states are shared.
}
\label{tab:2leptons}
\end{table}
The detection efficiencies are summarised in Table~\ref{tab:2leptons}.
The efficiencies are quoted for slepton
masses between 45 and 90~GeV.
Detection efficiencies were
estimated separately for right- and left-handed $\tilde{\mathrm e}$, $\tilde{\mu}$ and
$\tilde{\tau}$. The first three lines of Table~\ref{tab:2leptons} refer to
left-handed sleptons while the other lines refer to right-handed
sleptons.
Indeed, due
to the structure of the corresponding \lb\ term in the
Lagrangian of equation (\ref{lagrangian}), these
particles are expected to yield different final states.
The
expected background from all Standard Model
processes considered is normalised to the data luminosity of 56.5 pb$^{-1}$.
As can be seen in Table~\ref{tab:2leptons},
most of the background remaining comes from 4-fermion processes,
expected to be dominated by $\WW$ doubly-leptonic decays.
Due to beam-related backgrounds and to
incomplete modelling of two-photon
processes, there is poor agreement between the data
and Monte Carlo expectation in the early stages of some of the analyses.
When the two-photon processes have been effectively reduced
after specific cuts (for instance, a cut on the missing
transverse momentum), the agreement between data and Monte Carlo is good.
\subsection{Final States with Four Leptons with or without Missing Energy }
\label{sec:4leptons}
Final states with four charged leptons and no missing energy
may result from direct sneutrino decays while the final states with
missing energy may result from indirect sneutrino decays via a \lb\ coupling.
Two analyses have been developed and optimised separately for these two final
states.
No specific cut on the
lepton flavour present in the final state was applied.
To be independent of the type of decay and \lb\ coupling the two
analyses were at the end combined.
The following criteria were applied to select a possible signal in the
four leptons plus missing energy topology:
\begin{description}
\item[(B1)]
Events were required to have at least three charged tracks with a transverse
momentum with respect to the beam axis greater than 1.0~GeV.
\item[(B2)]
The event transverse momentum calculated without the hadron calorimeter
was required to be larger than 0.07 $\times \rs$.
This distribution
is shown in Figure~\ref{fig:multilepton}~(b)
after cut (B1) has been applied.
\item[(B3)]
Events had to contain at least three identified
isolated leptons each with a transverse
momentum with respect to the beam axis greater than 1.5~GeV.
\item[(B4)]
It was also required that $\Evis /\sqrt{s} < 1.1$,
where $\Evis$ is the event visible energy.
\item[(B5)]
The total leptonic energy, defined as the sum of the
energies of all identified leptons,
was required to be greater than $0.5 \times \Evis$.
\item[(B6)]
The background from two-photon processes and ``radiative return" events
(${\mathrm e}^+ {\mathrm e}^- \ra {\mathrm Z} \gamma$, where the $\gamma$ escapes
down the beam pipe) was reduced by requiring
that the polar angle of the missing momentum direction, $\thmiss$,
satisfies \mbox{$\cosmiss < 0.9$}.
\item[(B7)]
To reduce further the total background from Standard Model lepton
pair events, it was required that the energy sum of the two most
energetic leptons be smaller than 0.75~$\times \Evis$.
\end{description}
To select final states without missing energy, the following
requirements were imposed:
\begin{description}
\item[(C1)]
Events had to contain at least three identified
isolated leptons each with a transverse
momentum with respect to the beam axis greater than 1.5~GeV.
\item[(C2)]
It was also required that $0.65 < \Evis /\sqrt{s} < 2.0$.
\item[(C3)]
The total leptonic energy, defined as the sum of the
energy of all identified leptons,
was required to be greater than $0.65 \times \Evis$.
This distribution
is shown
in Figure~\ref{fig:multilepton}(c), after cuts (C1) have been
applied.
\item[(C4)]
To reduce the residual four-fermion background,
pairs were formed with the four most energetic tracks, and the invariant
mass was computed for each pair. Events were selected if
one of the three possible pairings satisfies
$ |m_{i,j} - m_{k,l}|/(m_{i,j} + m_{k,l}) < 0.4 $, were
$m_{i,j}$ is the invariant mass of the pair $(i,j)$. Only pairs
with invariant mass $m_{i,j}$ greater than 20 GeV were used in the computation.
\item[(C5)]
To reduce further the total background from Standard Model lepton
pair events, it was required that the energy sum of the two most
energetic leptons be smaller than 0.75~$\times \Evis$.
\end{description}
The two analyses were then combined. Events passing either set of criteria
were accepted as candidates.
Detection efficiencies
range from 34\% to
80\% for direct sneutrino decays,
and from 13\% to
58\% for indirect sneutrino decays,
for sneutrino masses between 45 and 90~GeV.
The expected background is
estimated to be 2.5 events.
There is one candidate event selected in the data.
\subsection{Final States with Six Leptons plus Missing Energy}
\label{sec:6leptons}
An analysis has been designed to select events with
six charged leptons and missing energy in the final state. These
topologies may for example result
from indirect slepton decays with a \lb\ coupling.
The following criteria were applied:
\begin{description}
\item[(D1)]
To reduce the background from
two-photon and di-lepton processes,
it was required that $0.1 < \Evis /\sqrt{s} < 0.7$.
\item[(D2)]
The event longitudinal momentum
was required to be smaller than 0.9 $\times {p_{\mathrm{vis}}}$,
where ${p_{\mathrm{vis}}}$ is the event total momentum.
\item[(D3)]
The event transverse momentum calculated without the hadron calorimeter
was required to be larger than 0.025 $\times \rs$.
This distribution
is shown in Figure~\ref{fig:multilepton}(d)
after cuts (D1) and (D2) have been applied.
\item[(D4)]
Events with less than five charged tracks
with a transverse momentum with respect
to the beam axis larger than 0.3~GeV were rejected.
\item[(D5)]
Events had to contain at least three well-identified
isolated leptons; at least two of them with a transverse
momentum with respect to the beam axis greater than 1.5~GeV, and the
third one with a transverse
momentum with respect to the beam axis greater than 0.3~GeV.
\item[(D6)]
The total leptonic energy,
was required to be greater than $0.2 \times \Evis$.
\end{description}
Detection efficiencies
range from 40\% to 88\% for indirect selectron decays,
from 59\% to 93\% for indirect smuon decays and
from 33\% to 70\% for indirect stau decays, for
slepton masses between 45 and 90~GeV.
The total background expectation is 1.7 events.
There is one candidate event selected in the data.
\subsection{Inefficiencies and Systematic Errors}
\label{sec:syserr}
Variations in the efficiencies were estimated with events
generated with $\Delta m$ = 5~GeV, as described in
Section~\ref{sec:MC}.
The inefficiency due to forward detector false vetoes caused by
beam-related backgrounds or detector noise was estimated from a study
of randomly triggered beam crossings to be
3.2\%. The quoted efficiencies
take this effect into account.
The systematic errors on the number of signal events expected that
have been considered are:
the statistical error on the determination of the efficiency from the
Monte Carlo simulation (typically less than 2\%);
the systematic error on the integrated luminosity of 0.4\%;
the uncertainty due to the interpolation of the efficiencies,
estimated to be 4.0\% and the lepton identification uncertainty,
estimated to be 2.4\% for the muons, 3.9\% for the electrons and
4.7\% for the taus. The systematic error arising from the modelling of
the variables used in the multi-lepton final state selections is
smaller than the lepton identification uncertainties.
The systematic error due to the trigger efficiency is
negligible because of the high lepton transverse momentum requirement.
The total systematic error was calculated by summing in quadrature
the individual errors and is incorporated into the limit calculation
using the method described in Ref.~\cite{ref:cousins}.
The systematic error on the number of expected background events from
SM processes has a negligible effect when computing limits.
\section{Final States with Two Jets and Two Leptons }
\label{sec:2jets2leptons}
\subsection{Electron and Muon Channels}
\label{sec:stopemu}
In this section, the analysis for the selection of the final state of two electrons
or two muons plus two jets and no missing energy is described.
These final states may result
from the direct decay of pair-produced stops via a \lbp\ coupling.
In contrast to the purely leptonic final states described in the previous
section, the topologies searched for in this analysis involve hadronic jets;
more stringent cuts are needed to obtain a purer lepton sample.
Particles are considered as electrons or muons if they are either identified by
the selection algorithms described in~\cite{ref:elecbarrel}
and~\cite{ref:elecendcap}, or by an algorithm used for selecting
semileptonic W decays, as
described in~\cite{ref:wwpaper}.
Events were preselected by requiring the following criteria to be
satisfied (the same criteria were also used for the analysis presented
in Section~\ref{sec:jetsleptons}):
The fraction of good tracks had to be greater than 0.2,
to reduce beam-gas
and beam-wall background events.
Events with fewer than
seven good charged tracks were not considered in order to reduce the
background from Bhabha scattering.
Events had to contain at least one identified
electron or muon with a momentum
greater than 3~GeV, to
reduce the background from final states with
low energy leptons (e or $\mu$).
To reduce background from two-photon processes, it was required that
the visible energy
normalised to the centre-of-mass energy,
$R_{\rm vis} = E_{\rm vis}/ \sqrt{s} >$~0.3.
The following cuts are then applied:
\begin{description}
\item [(E1)]
The visible energy had to be close to the centre-of-mass energy,
$0.75 < R_{\rm vis} < 1.25$.
Figure~\ref{fig:ivor_stop}(a) shows the visible energy distribution.
\item [(E2)]
It was required that four jets be reconstructed using the Durham~\cite{ref:durham} algorithm,
with $y_{34} > 0.001$, where $y_{34}$ is the cut parameter between 3
and 4 jets.
Both hadronic and leptonic objects are used in the jet reconstruction.
Figure~\ref{fig:ivor_stop}(b) shows the $y_{34}$ distribution.
\item [(E3)]
Events had to contain at least one pair of identified oppositely-charged
lepton candidates of the same flavour.
\item [(E4)]
To make use of the signal topology of two leptons and
two jets,
where a lepton and a jet stem from the same object,
a five-constraint (5C) kinematic fit was
performed for the two possible
combinations of each lepton with each jet.
The kinematic constraints are: the vector sum of all
momenta has to be
equal to zero, the total energy of all objects has to
be equal to the
centre-of-mass energy and the masses of the two
reconstructed particles
have to be equal.
From the three most energetic leptons of the same
flavour,
the two most isolated\footnote{
The most isolated lepton is the one with the largest
angle to the closest track.}
were selected and the rest of the event was
reconstructed as two jets.
The combination with the highest fit
probability was
selected. The probability for the fit, based on the
$\chi^2$, was
required to be larger than 0.01.
\item [(E5)]
The momentum of the most energetic lepton had to be
greater than 15~GeV and the momentum of the second most
energetic lepton had to be greater than 10~GeV.
\item [(E6)]
It was required that there be no charged track
within $15^{\circ}$ of the most energetic
lepton candidate.
\end{description}
\begin{figure}
\begin{center}
\epsfig{file=pr275_02.eps,width=15.0cm}
\caption[]{\sl
Stop search (Analysis E): (a) Visible energy R$_{vis}$
after the preselection and (b) jet resolution
$y_{34}$ after cut (E1).
Data are shown as points and the sum of all Monte Carlo background
processes is shown as the solid line.
The dashed histogram
shows signal Monte Carlo events for direct decays of $\stopm$ with
$m_{\stopm} = 85$~GeV and for \lbp$_{ij3}, (i=1,2)$.
The scale of the signal MC is arbitrary.
The arrows point into the regions accepted by the cuts.
}
\label{fig:ivor_stop}
\end{center}
\end{figure}
These cuts yield an efficiency of more than 50\% for a stop mass
of 65~GeV, which rises to approximately 65~\% for masses above 85~GeV.
No candidate event is selected in the data.
The expected background is 0.9 events for final states with two electrons
and 0.6 events for final states with two muons.
The largest background results in both cases from WW events.
The following systematic errors have been considered:
\begin{enumerate}
\item
The statistical error from the limited size of the Monte Carlo samples.
\item
The error due to the interpolation of efficiencies for mass values
between the generated stop masses, which was estimated to be less than 4\%.
\item
A 4\% error due to the lepton identification
for the electron
and a 2\% error for the muon channel.
\item
The fragmentation of the stop has been simulated using the fragmentation
function from Peterson {\it et al.} with the $\epsilon$ parameter
extrapolated from measurements of charm and bottom~\cite{ref:opalstop}.
To check the model dependence of the fragmentation,
it has also been performed using the function from
Bowler~\cite{ref:bowler}.
No significant change in the efficiency due to the difference in the
fragmentation function has
been found. The difference is at most 0.5\%, where a variation of the
$\epsilon$ parameter of the $\stopx\ $ in the Peterson {\it et al.} scheme
is included. This error on $\epsilon_{\stopx}$
is propagated from the error of $\epsilon_{\rm b}$ and the error on the b-quark
mass as described in detail in Ref.~\cite{ref:opalstop}.
\item
The signal events have been produced for a zero mixing angle between the two
stop eigenstates. The mixing angle describes the coupling between the stop
and the Z$^0$, and therefore the energy distribution of the initial state
radiation depends on this mixing angle.
To check the dependence of the detection efficiency on this angle,
events have been generated with $\theta_{\stopx\ } = 0.98$,
where the stop decouples from the Z$^0$.
The change in efficiency is less than 0.5\% for the two extreme cases.
\item
The Fermi motion of the spectator quark in the stop-hadron influences
its measured mass. The Fermi motion has been increased from 220~MeV to
520~MeV and the efficiency changes by no more than 1\%, which is taken as a
systematic error.
\item
The systematic error on the measured luminosity is 0.4\%.
\item
The systematic error due to the uncertainty in the trigger efficiency was
estimated to be negligible, because of the requirement of at least seven good
tracks.
\end{enumerate}
The systematic error on the expected number of
background events has been
estimated to be less than 20\% for all cases by varying
the cut values
by the experimental resolution.
\subsection{\boldmath The channel $\tilde{t}\tilde{\bar{t}}\rightarrow
\subsection{Tau Channel}
\label{sec:stoptau}
This section describes the analysis used to search for the final state
consisting of two $\tau$-leptons and
two jets, which may result from the direct decay of a stop
via a coupling $\lambda^\prime$.
The backgrounds come predominantly from
($Z/\gamma)^*\rightarrow q\bar{q}(\gamma)$ and SM four-fermion processes.
The selection begins with the identification of $\tau$ lepton candidates,
identical
to that in~\cite{ref:smpaper}, using three algorithms
designed to identify electronic, muonic and hadronic $\tau$-lepton decays.
An average of 2.3 $\tau$ candidates per signal event are identified.
The original $\tau$ lepton direction is approximated
by that of the visible decay products.
The following
requirements,
similar to those described in Ref.~\cite{ref:mssmpaper} up to (F4),
are then imposed:
\begin{description}
\item[(F1)]
Events are required to contain at least nine charged tracks, and must have at
least two $\tau$ lepton
candidates, including at least one pair where
each $\tau$ has electric charge $|q|=1$ and the charges sum to zero.
Pairs not fulfilling these requirements are not considered further.
\item[(F2)]
Events must have no more than a total of 20 GeV of energy
deposited in the forward detector, gamma catcher, and
silicon-tungsten calorimeter; a missing momentum vector
satisfying $|\cos\theta_{\rm miss}| < 0.97$,
a total transverse momentum of at least 2\% of $\sqrt{s}$, and a scalar sum
of all track and cluster transverse momenta larger than 40~GeV.
\item[(F3)]
Events must contain at least
three jets reconstructed using
the cone algorithm as in~\cite{ref:smpaper}\footnote{Here, single
electrons and muons from $\tau$ lepton decays are allowed to be recognised
as low-multiplicity ``jets''.}, and no energetic isolated photons\footnote{
An energetic
isolated
photon is defined as an electromagnetic cluster
with energy larger than 15~GeV and no track within a cone
of $30^\circ$ half-angle.}.
\item[(F4)]
Events must contain no track or cluster with
energy exceeding $0.3\sqrt{s}$.
\end{description}
For events surviving these requirements, the hadronic part of the event
corresponding to each surviving $\tau$ lepton candidate pair, composed
of those tracks and clusters not having been
identified as belonging to the pair (henceforth referred to as the
``rest of the event'' or RoE),
is then split into two jets
using the Durham~\cite{ref:durham}
algorithm. Two
pairings between the two $\tau$ candidates and the jets are
possible.
The invariant masses $m_{\tau j}$ of the two resulting $\tau$-jet systems
within each
pairing are then calculated
using only
the $\tau$ lepton and jet momentum directions and requiring energy and
momentum conservation. The pairing scheme
with the smaller difference between
$m_{\tau j1}$ and
$m_{\tau j2}$
is then chosen.
In order for a $\tau$ candidate pair to be considered
further, the following
requirements on $m_{\tau j1}$ and
$m_{\tau j2}$ are imposed, consistent
with the hypothesis of the decay of two heavy objects of identical mass:
\begin{description}
\item[(F5)]
Both $m_{\tau j1}$ and $m_{\tau j2}$ must be
at least 30 GeV.
\item[(F6)]
The difference in invariant masses must be no more than
30\% of their sum, i.e.
$|m_{\tau j1}-m_{\tau j2}|/|m_{\tau j1}+m_{\tau j2}|\leq 0.3$.
\end{description}
The distribution of
$|m_{\tau j1}-m_{\tau j2}|/|m_{\tau j1}+m_{\tau j2}|$ is shown in
Fig.~\ref{fig:twotaufourjet}~(a) for the data,
the backgrounds, and
for a signal sample with $m_{\tilde{t}}=75$ GeV.
The resolution
on $m_{\tau j}$ is typically below 5 GeV, except very close to
the kinematic limit.
A likelihood method similar to that described in~\cite{ref:opal_rpv_gauginos}
is then
applied to those events satisfying the above requirements,
in order to select a final $\tau$ candidate pair for each
event from those surviving, and to suppress further the remaining background.
Distributions of two of the input variables as well as that of
${\cal L}$ are shown in
Figures~\ref{fig:twotaufourjet}~(b) to (d).
In each event, the $\tau$-candidate pair with the highest value of ${\cal L}$
is chosen, and the following
requirement is then made:
\begin{description}
\item[(F7)] ${\cal L}>0.93$
\end{description}
Two events survive the selection while the background, almost all
from four-fermion processes, is
estimated to be
2.07 events for an integrated luminosity of 55.8 pb$^{-1}$.
The reconstructed $\tau$-jet masses
are 78.9 and 87.9 GeV for the first selected event and 71.7
and 67.2 GeV for the second one.
The detection efficiencies for
stop masses between 55 and 90 GeV range from 30 to 40\%,
while that for 45 GeV is approximately 22\%.
These efficiencies are affected by the following
relative uncertainties:
Monte Carlo statistics, typically 2.5 to 3.5\%;
uncertainty in the tau-lepton preselection efficiency, 1.2\%;
uncertainty in the modelling of the other preselection variables, 2.0\%;
uncertainties in the modelling of the likelihood input variables, 10.0\%;
uncertainties in the modelling of fragmentation and hadronisation, 6.0\%;
and uncertainty on the integrated luminosity, 0.5\%~\cite{ref:lumino}.
Taking these uncertainties as independent and adding them in quadrature
results in a total relative systematic uncertainty of 12.3\%
The systematic uncertainty in the number of expected background events was
estimated to be 18\%.
\section{Final States with more than Two Jets and at Least two Charged Leptons}
\label{sec:jetsleptons}
\subsection{Indirect Selectron and Smuon Decays}
\label{sec:jetsemu}
This section describes the event selection for final states from the indirect
decay of selectrons and smuons via the coupling \lbp.
The final state consists of two leptons of the same flavour from
the sleptons plus the decay products of the two $\chin$'s.
These will be two jets plus a neutral or charged lepton for each $\chin$.
This results in seven different final states for each slepton flavour, as
shown in Table~\ref{tab:sleptons_ivor}.
Electrons and muons are identified as described in Section~\ref{sec:stopemu}.
To identify taus in the final states an Artificial Neural Net based on
tracks~\cite{tauID} is used, rather than the selection presented in
Section~\ref{sec:stoptau} designed specifically for events with two
$\tau$'s.
The preselection is the same as described in Section~\ref{sec:2jets2leptons}.
The selection cuts are as follows:
\begin{description}
\item [(G1)]
A cut on the visible energy scaled by the centre-of-mass energy
in the range $0.5 < E_{\mathrm vis}/\sqrt{s} < 1.2$,
depending on the expected number of neutrinos,
is applied.
In addition a cut on the angle of the missing momentum
with respect to the beam direction
at $|\cos \theta| <$~0.95
is performed,
if some missing momentum is expected.
\item [(G2)]
The jets in the event have been reconstructed using the Durham algorithm.
The jet resolution $y_{45}$
at which the number of jets changes from 4 to 5 jets,
is required to be greater than 0.002.
This cut takes into account the high multiplicity of the signal events.
\item [(G3)]
To reduce the background from W pair production
for events with missing momentum,
a single-constraint kinematic fit has been performed.
The inputs to the fit are the momenta of the lepton and
the neutrino,
taking the missing momentum to be the momentum of the
neutrino, and the
rest of the event reconstructed into 2 jets.
The lepton is taken to be the
most energetic muon or electron in the case of smuon or
selectron production,
respectively.
The invariant mass is
calculated (a) for the lepton and the neutrino system
and (b) for the two
jet system, letting the masses of both systems be
independent.
The reconstructed mass of at least one system has to be
outside a mass
window of 70~GeV$< m < 90$~GeV, or the probability for
the fit has
to be less than 0.01.
\item [(G4)]
For the topologies with no charged lepton from the
$\chin$ decay,
the background from W pair production is reduced further
by a kinematic fit
on the invariant mass of two pairs of jets, when
reconstructing the
whole event into 4 jets.
This kinematic fit assumes energy and momentum
conservation and
the same mass for both jet pairs.
From the three possible jet pairings, the one with the
highest fit
probability is chosen.
The reconstructed mass of the jet pairs has to be
outside a mass
window of 70~GeV$< m < 90$~GeV, or the probability for
the fit has
to be less than 0.01.
\item [(G5)]
At least two leptons of the flavour of the slepton
have to be identified.
To have sensitivity also to small mass differences
between the slepton and
the
$\chin$, the required momentum has to be greater than
4~GeV
for both muons
in the smuon case and the required energy greater than 4~GeV and 3~GeV
for the two
electrons
in the selectron case, respectively.
\item [(G6)]
In addition to the leptons required in (G5), also the leptons from
the $\chin$ decay have to be identified. If two additional charged leptons are
expected, both have to be identified, if they have a different flavour than
the slepton.
If two taus are expected, only
one, being different from the leptons in cut (G5), has to be identified.
If a total of four leptons of the same flavour is expected,
including those in cut (G5), only three of them have to be identified.
If only one additional lepton is expected, it has to be identified.
The energy or momentum of the most energetic lepton has to be above
a cut value varying between 8 and 15~GeV, depending on the topology.
If a total of four leptons is required, for the second most energetic
an energy or momentum larger than a cut value varying between 3~GeV and 4~GeV,
depending on the topology, is required.
\item [(G7)]
To make use of the isolation of the leptons in the signal, one or two
of the identified leptons, depending on the expected topology,
are required to be isolated.
The isolation criterion is that
there be no charged track within a cone of
half opening angle $\phi$, such that $|\cos \phi| =$~0.99,
around the track of the lepton.
\end{description}
These selections give efficiencies between 45 and 85\% for final states
without taus, and around 30\% for final states with taus, all for
slepton masses greater than 70~GeV.
The expected backgrounds and the numbers of events observed for each
final state
are shown in Table~\ref{tab:sleptons_ivor}.
\begin{table}[tbp]
\begin{center}
\begin{tabular}{|l||c|c|}
\hline
Final State & Selected Events & Tot. bkg MC \\
\hline
\hline
$\tilde{\mu}^+ \tilde{\mu}^- \ra $ & & \\
\hline
$ \mu^+ \mu^- eqq eqq $ & 2 & 0.69 \\
$ \mu^+ \mu^- \mu qq \mu qq $ & 1 & 0.67 \\
$ \mu^+ \mu^- \tau qq \tau qq $ & 1 & 1.10 \\
$ \mu^+ \mu^- eqq \nu qq $ & 3 & 1.05 \\
$ \mu^+ \mu^- \mu qq \nu qq $ & 1 & 0.95 \\
$ \mu^+ \mu^- \tau qq \nu qq $ & 0 & 0.58 \\
$ \mu^+ \mu^- \nu qq \nu qq $ & 0 & 0.91 \\
\hline
\hline
$\tilde{e}^+ \tilde{e}^- \ra $ & & \\
\hline
$ e^+ e^- eqq eqq $ & 1 & 0.29 \\
$ e^+ e^- \mu qq \mu qq $ & 1 & 0.37 \\
$ e^+ e^- \tau qq \tau qq $ & 3 & 1.09 \\
$ e^+ e^- eqq \nu qq $ & 1 & 0.52 \\
$ e^+ e^- \mu qq \nu qq $ & 2 & 1.10 \\
$ e^+ e^- \tau qq \nu qq $ & 3 & 0.81 \\
$ e^+ e^- \nu qq \nu qq $ & 0 & 1.13 \\
\hline
\end{tabular}
\end{center}
\caption{\it
Number of events remaining after the selection cuts and the
expected backgrounds from all Standard Model processes.
The main contribution to the total
background comes
from $\WW$ leptonic decays (4-fermion processes); multi-hadronic events
contribute up to 30\% and other processes are negligible.}
\label{tab:sleptons_ivor}
\end{table}
\subsubsection*{Systematic Errors}
For the lepton identification, a systematic error of 4\% was estimated for the
electrons, 3\% for the muons and 3\% for the taus.
For the interpolation of the efficiency between the generated mass points,
a systematic error of 4\% has been assigned.
From the studies on the fragmentation in Section~\ref{sec:2jets2leptons}
the systematic error for this analysis is estimated to be less than 1\%.
The systematic error on the measured luminosity is 0.4\%.
The systematic error due to the uncertainty in the trigger efficiency
is negligible, because of the requirement of at least seven good
tracks.
The statistical error on the determination of the efficiency from the MC
samples has also been treated as a systematic error.
The systematic error on the expected number of background events has been
estimated to be less than 20\% for all cases.
\subsection{Stau Indirect Decays}
\label{sec:jetstau}
If requirements (F5) and (F6) described in
Section~\ref{sec:stoptau} are suppressed, then the
same analysis as that for the stop search in the tau channel
can be used to search for the indirect decay of staus
via the coupling \lbp,
where now the final state consists
of two $\tau$ leptons plus four jets and two additional leptons.
In this case, the reference distributions are regenerated in light of
the different topology of this signal, and the minimum required value
of the resulting likelihood discriminant ${\cal L}$
(cf. (F7)) is relaxed to 0.9.
No events survive the selection
while the background expectation rises slightly to 2.27 events.
The detection efficiencies range from 12\% for final states with two
taus, four quarks plus missing energy and $m_{\tilde{\tau}} =$~45~GeV, to 54\%
for final states with two
taus, four quarks plus two electrons and $m_{\tilde{\tau}} =$~70~GeV.
The systematic uncertainties are evaluated in the same way as for
the stop search as described in Section~\ref{sec:stoptau}, and are
similar in magnitude.
\section{Final States with Four Jets plus Missing Energy}
\label{sec:multijetsemiss}
Indirect decays of sneutrinos via $\lambda^\prime$ coupling
can lead to final states with four jets and large missing energy
due to the four undetected neutrinos. The dominant backgrounds come from
four-fermion processes and radiative or mis-measured two-fermion events.
The selection procedure is described below:
\begin{description}
\item[(H0)] The event has to be classified as a multi-hadron final-state as
described in \cite{LEP2MH}.
\item[(H1)] The visible energy of the event is required to be less than
0.75$\sqrt{s}$.
\item[(H2)] To reject two-photon and radiative two-fermion events
the transverse momentum should be larger than 10 GeV,
the total energy measured in the forward calorimeter, gamma-catcher
and silicon tungsten calorimeter should be less than 20 GeV,
and the missing momentum should not point along the beam
direction ($| \cos \theta_{\mathrm{miss}}| <$ 0.96).
\item[(H3)] The events are forced into four jets using the Durham
jet-finding algorithm, and rejected if the jet resolution parameter
$y_{34}$ is less than 0.0008.
\item[(H4)] An additional cut is applied
against semi-leptonic four-fermion events, vetoing on
isolated leptons being present in the event.
The lepton identification is based on
an Artificial Neural Network routine (ANN)~\cite{tauID},
which was originally designed to identify tau leptons
but is efficient for electrons and muons, as well.
If at least one lepton candidate is found,
with ANN output larger than 0.97,
the event is rejected.
\item[(H5)] Finally, a likelihood selection
is employed to classify the remaining events
as two-fermion, four-fermion or signal processes.
The method and the likelihood variables are described
in~\cite{ref:opal_rpv_gauginos}, with the restiction that the
minimum number of charged tracks and the minimum number of
electromagnetic clusters in a jet are replaced by the aplanarity of
the event~\cite{dpar}.
The event is rejected if its likelihood output is less than 0.9.
\end{description}
Figure~\ref{fig:gabi_fig} shows experimental plots
for the data, the estimated background and simulated signal
events.
After all cuts, 5 events are selected in the data sample, while
8.17$\pm$0.31$\pm$1.32 events
are expected from Standard Model processes, of which 75\% originate
from four-fermion processes. The signal detection efficiency varies
between 5\% and 34\% for sneutrino masses between 45 -- 90 GeV
for $\lambda^\prime_{121}$ and $\lambda^\prime_{123}$ couplings
if the mass difference is one half of the sneutrino mass.
For a small mass difference ($\approx$5 GeV),
the efficiency is more than doubled.
The small efficiency for light sneutrino masses is the result of
initial-state radiation and the larger boost of the particles, which make the
event similar to the QCD two-fermion background.
The expected signal rates are affected by the following
uncertainties: Monte Carlo statistics, 3.3 -- 13.9\%;
statistical and systematic uncertainties on the luminosity measurement,
0.3 and 0.4\%;
uncertainties on modelling of the kinematic variables, 6.7\%; and
on the lepton veto, 1.0\%.
The background estimate has the following errors: Monte Carlo
statistics, 3.7\%;
modelling of the hadronisation process estimated by comparing different event
generators, 5.3\%; uncertainty on the lepton veto,
1\%; and modelling of the kinematic variables, 14.9\%.
The inefficiency due to the forward energy veto is found to be 1.8\%.
\section{Final States with Four Jets without Missing Energy}
\label{sec:multijetsnoemiss}
Direct decays of sleptons (squarks) via $\lambda^\prime$
($\lambda^{\prime\prime}$) coupling
can result in final states with four well-separated,
high multiplicity hadronic jets and large visible energy.
The background comes from q$\bar{\mathrm q}(\gamma)$ events with hard gluon
emission and four-fermion processes, predominantly
W$^+$W$^-$ $\rightarrow$ qqqq.
The analysis closely follows our published selection
for H$^+$H$^-$ $\rightarrow$ qqqq \cite{tauID}.
First, well-defined four-jet events are
selected; then a set of variables are combined using a likelihood technique.
The preselection consists of the following steps:
\begin{description}
\item[(I0)] The event has to be classified as a multi-hadron final-state as
described in \cite{LEP2MH}.
\item[(I1)] To reduce the radiative two-fermion background,
the effective centre-of-mass energy of the
event,$\sqrt{s'}$~\cite{sprime},
is required to be greater than 150 GeV.
\item[(I2)] To ensure that the events are well-contained, the
visible energy should be greater than 0.7$\sqrt{s}$.
\item[(I3)] The events are forced into four jets using the Durham
jet-finding algorithm, and rejected if the jet resolution parameter
$y_{34}$ is less than 0.0025. Moreover, all jets must contain at least one
charged particle.
\item[(I4)] A four-constraint kinematic fit, applied to the jet four-momenta
requiring energy and momentum conservation (4C-fit),
should yield a $\chi^2$-probability larger than $10^{-5}$.
\item[(I5)] To test the compatibility
with pair-produced equal mass objects and
to obtain the best possible di-jet mass
resolution, the jet four-momenta are refitted requiring energy and momentum
conservation and equal di-jet masses (5C-fit).
The event is kept if at least one of the three di-jet combinations
has a $\chi^2$-probability larger than $10^{-5}$.
\end{description}
To separate the signal from the background events surviving the above selection
a likelihood technique is applied. Three event classes are defined: signal,
two-fermion and four-fermion.
\subsection{Sleptons}
\label{sec:multijetsnoemiss-sl}
We have used the H$^+$H$^-$ $\rightarrow$
c$\bar{\mathrm s}\bar{\mathrm c}$s
MC samples to produce the signal reference histograms.
This is possible because of the similarities between charged Higgs
and smuon, stau, muon- and tau-sneutrino decays.
Since selectrons and electron-sneutrinos can also be produced in
$t$-channel-exchange processes,
their event properties (especially the angular distributions) are different,
and we have used dedicated MC samples
with $\lambda^{\prime}_{121}$ and $\lambda^{\prime}_{123}$ couplings
to produce these reference histograms.
The following variables were used as input to the likelihood calculation:
\begin{list}{$\bullet$}{\itemsep=0pt \parsep=0pt \topsep=-5pt \leftmargin=30pt}
\item the cosine of the polar angle of the thrust axis;
\item the cosine of the smallest jet-jet angle;
\item the difference between the largest and smallest jet energy
after the 4C-fit;
\item the smallest di-jet mass difference after the 4C-fit;
\item the cosine of the
di-jet production angle multiplied by the sum of the jet charges
for the combination with the highest $\chi^2$-probability given by
the 5C-fit.
\end{list}
Events were accepted if their likelihood output was larger than 0.5, 0.55 and
0.6 for selectrons, electron-sneutrinos and other sleptons,
respectively.
The numbers of selected data and expected background events are listed in
Table~\ref{table:4jet} for the different selections.
Since the background is dominated by W$^+$W$^-$ production (82--87\%),
the mass distributions are peaked around the W$^\pm$ boson mass.
No excess (unexpected accumulation) was observed in the data.
Figure~\ref{fig:gabi_fig2}a shows, as an example, the mass distribution
of the selected events for the data,
the estimated background and simulated selectron events.
\begin{table}[ht]
\begin{center}
\begin{tabular}{|l|c|c|}
\hline
& Data & Background \\ \hline \hline
Preselection & 454 & 445.4$\pm$2.3 \\ \hline \hline
Selectron & 55 & 55.4$\pm$0.8 \\ \hline
Electron-sneutrino & 41 & 49.1$\pm$0.7 \\ \hline
Other sleptons & 50 & 48.8$\pm$0.7 \\ \hline
Squarks & 7 & 8.8$\pm$0.3 \\ \hline
\end{tabular}
\end{center}
\caption{\it
The numbers of selected data and expected background events
in the four-jet channel after the preselection and at the end of the
different selections. Only the statistical error is indicated.}
\label{table:4jet}
\end{table}
The di-jet mass resolution using the 5C-fit is 0.6--1.6
GeV, depending on the sparticle mass and decay.
Events in a 2$\sigma$ mass window around the test
mass were selected. The efficiencies vary between 11.3\% and 34.3\%
within such a mass window for sparticle masses between 50 and 75 GeV,
depending on the sparticle mass and decay.
The signal detection efficiency is subject to the following inefficiencies and
systematic errors:
the statistical error due to the limited number of Monte Carlo events,
4.4--17.7\%;
the uncertainty on modelling
the kinematic variables used in the analysis, 3\%;
and additionally for the smuon, muon-sneutrino, stau and tau-sneutrino
selection, the inefficiency due to the differences between the slepton
and the charged Higgs boson simulation, 0--12\%.
The background estimate has the following uncertainties:
the statistical error due to the limited number of Monte Carlo events,
1.5\%;
the statistical and systematic error on the luminosity measurement, 0.3 and
0.4\%;
the uncertainty on modelling the SM background processes, estimated by comparing
different event generators, 2\%; and
the kinematic variables used in the analysis, 4.9\%.
\subsection{Squarks}
\label{sec:multijetsnoemiss-sq}
Squarks are expected to hadronize resulting in a final state with six jets,
from which the two spectator jets have small energy, at least for heavy squarks,
and therefore it is still possible to reconstruct the squark pair events into
four jets.
To produce the signal reference histograms, we have used
dedicated squark samples generated by SUSYGEN
with $\lambda^{\prime\prime}_{121}$ and $\lambda^{\prime\prime}_{123}$
couplings. Since jets originating from squark decays are
narrower than the ones coming from Standard Model sources,
in addition to the five input variables used in the slepton searches,
two new variables are introduced:
\begin{list}{$\bullet$}{\itemsep=0pt \parsep=0pt \topsep=-5pt \leftmargin=30pt}
\item the smallest boosted jet thrust;
\item the highest jet mass.
\end{list}
The events are rejected if their likelihood output is less than 0.95.
Figures~\ref{fig:gabi_fig2}b-d show experimental plots
for the data, the estimated background and simulated signal
events.
The numbers of selected data and expected background events are listed in
Table~\ref{table:4jet}.
Since the background is dominated by W$^+$W$^-$ production (93.3\%),
the mass distribution is peaked around the W$^\pm$ boson mass.
No unexpected accumulation of events is observed in the data.
The di-jet mass resolution using the 5C kinematic fit is 0.45--1.2
GeV, depending on the squark mass and decay. A systematic shift of the
reconstructed mass (up to +2.2 GeV for squark masses of 45 GeV)
is observed, which is taken into
account when applying the 2$\sigma$ mass window.
The signal detection efficiencies within the mass windows vary
between 14.1\% and 29.8\% for squark masses of 45--90 GeV.
The signal detection efficiency is subject to the following inefficiencies and
systematic errors:
the statistical error due to the limited number of Monte Carlo events,
4.9--7.8\%; and
the uncertainty on modelling
the kinematic variables used in the analysis, 13.2\%.
The effect of
different fragmentation and hadronization models has been tested
comparing SUSYGEN and a special stop generator~\cite{ref:stopgen} used in
OPAL stop searches~\cite{ref:stoppaper}.
It was found that SUSYGEN produces wider
(more SM-like) jets,
and our efficiency would be more than a factor of two higher
for events generated by the stop generator.
Thus our efficiency estimates
using SUSYGEN are considered to be conservative.
The background estimate has the following uncertainties:
the statistical error due to the limited number of Monte Carlo events,
3.6\%;
the statistical and systematic error on the luminosity measurement, 0.3 and
0.4\%;
the uncertainty on modelling the SM background processes,
estimated by comparing different event generators, 20.4\%; and
the kinematic variables used in the analysis, 23.8\%.
The result of the slepton and squark
analyses is combined with previous searches performed at
$\sqrt{s}$=130--172 GeV for pair-produced, equal mass scalar particles (charged
Higgs bosons)~\cite{ch172} in order to increase the sensitivity for low mass
sleptons and squarks.
These previous searches are assumed to be equally efficient
for slepton, squark and charged Higgs search.
This hypothesis has been tested using slepton (squark) Monte Carlo
samples generated at $\sqrt{s}=172$ GeV for several $\lambda^\prime$
($\lambda^{\prime\prime}$) couplings with sparticle masses of 45, 55 and 70 GeV.
The efficiencies are found to be consistent within the statistical errors
except for the squark samples, where a relative 20\% increase
in the efficiency is observed.
Conservatively, this gain is not taken into account.
\section{Interpretation}
\label{sec:results}
No significant excess of events in the
data with respect to the expected background
has been observed for all analyses listed in Table~\ref{tab:relation}.
Production cross-section and mass limits have therefore been computed.
These limits also take into account
indirect limits obtained from the study of the Z$^0$ width at LEP1 and
therefore concern only sparticle masses above 45 GeV.
Two approaches are used to present sfermion production limits.
In the first one, upper limits
on production cross-sections as functions of the sfermion
masses are calculated with minimal model assumptions.
These upper limits in general do not depend
on the details of SUSY models, except for the assumptions
that the sparticles are pair-produced and that only one \lb-like coupling
at a time is nonzero, as stated in Section~\ref{sec:intro}.
In the second approach, limits on the sfermion masses were calculated
in the framework of the Constrained MSSM where mass limits are derived
using the following parameters:
$m_0$, the common sfermion mass at the GUT scale;
$M_2$, the SU(2) gaugino mass parameter
at electroweak scales\footnote{We assume that
$M_1$, the U(1) gaugino mass at electroweak scales,
is related to $M_2$ by the usual gauge unification
condition: $M_1 = \frac{5}{3} \tan^2 \theta_{\mathrm W} M_2$.};
$\mu$, the mixing parameter of the two Higgs doublets and
$\tan \beta= v_2/v_1$, the ratio of the vacuum expectation values for
the two Higgs doublets.
For the indirect sfermion decays, we have used the branching ratios
for the decay $\tilde{f} \rightarrow f \tilde{\chi}^0_1$ predicted
by the MSSM, and we have conservatively assumed no
experimental sensitivity to any other decay mode.
The branching ratio for direct decay is always treated as equal
to 1, as we allow
only one \lb\ coupling to be different from zero at a time.
The MSSM mass exclusion
plots presented in the following sections are computed for
$\tan \beta =$~1.5 and $\mu =$~--200~GeV. This choice of parameters
is rather conservative as sfermion production cross-sections generally
increase for larger values of $\tan \beta $ or $|\mu|$.
In the indirect decay of a sfermion, $\tilde{f} \rightarrow f \tilde{\chi}^0_1$,
via a \lbp\
coupling, the $\neutralino$ decays either as:
\begin{equation}
\tilde{\chi}^0_1 \rightarrow \ell^-_i u_j \overline{d}_k \; , \;\;\;\;
\tilde{\chi}^0_1 \rightarrow \ell^+_i \overline{u}_j d_k \; ,
\label{ref:decays1}
\end{equation}
or as:
\begin{equation}
\tilde{\chi}^0_1 \rightarrow \nu_i d_j \overline{d}_k \; , \;\;\;\;
\tilde{\chi}^0_1 \rightarrow \overline{\nu}_i \overline{d}_j d_k
\label{ref:decays2}
\end{equation}
This leads to final states
with two fermions from the sfermion decay plus the $\tilde{\chi}^0_1$ decay products:
\begin{enumerate}
\item
Four jets and two charged leptons if both $\tilde{\chi}^0_1$ decay via
(\ref{ref:decays1})
\item
Four jets and missing energy if both $\tilde{\chi}^0_1$ decay via
(\ref{ref:decays2})
\item
Four jets, one charged lepton and one neutrino if one $\tilde{\chi}^0_1$ decays via
(\ref{ref:decays1}) and the other via (\ref{ref:decays2}).
\end{enumerate}
The relative branching ratios
of the neutralino into a final state with a charged or a neutral
lepton
depends on the mass of the sneutrinos, the mass of the
sleptons and on the components of the gaugino (Wino or Higgsino).
To avoid a dependence of the results on the MSSM parameters, the
branching ratio of $\tilde{\chi}^0_1$ to charged leptons and jets (\ref{ref:decays1})
was varied between
0 and 1. The branching ratio of $\tilde{\chi}^0_1$ to neutrinos
and jets (\ref{ref:decays2})
was varied accordingly between 1 and 0. The combination of these two
branching ratios fixes the branching ratio for one $\tilde{\chi}^0_1$ decaying
via (\ref{ref:decays1}) and the other via (\ref{ref:decays2}).
A likelihood ratio method~\cite{likelihood} was used
to determine an upper limit for the
cross-section. This method combines the individual
analyses looking for the different final states possible for one given
\lbp\ coupling and assigns greater weight
to those with a higher expected sensitivity, taking into account the expected
number of background events.
This results in a cross-section limit as a function of the
branching ratio and the sfermion mass. By taking the
worst limit at each sfermion mass, a limit independent of the branching ratio
is determined.
For the direct decays, the final states are fully
determined by the indices of the coupling considered.
In the following sections, cross-sections limits are shown for the
various direct and indirect decays studied in this paper, see
Table~\ref{tab:relation}.
In each cross-section plot,
only the curve corresponding to the worst cross-section limit is
shown amongst all possible cross-section limits resulting from the
couplings considered. The coupling yielding the worst
cross-section limit is indicated in each plot.
Generally, the best excluded cross-section comes from final states
with a maximum number of muons and no taus, while the worst results come from
final states with many taus, due to their lower detection efficiency.
In the MSSM framework, the exclusion regions for the indirect
decays are valid for $\Delta m = m_{\tilde{\ell}} - m_{\tilde{\chi}^0_1} \ge 5$~GeV except
for the indirect decays of staus via \lbp\ which are
valid for $\Delta m = m_{\tilde{\ell}} - m_{\tilde{\chi}^0_1} \ge 22.5$~GeV. In this
particular case there is not enough sensitivity to place limits in
the small $\Delta m$ region.
The exclusion region for the direct decays is independent of
$\Delta m$.
All limits presented here are quoted at the 95\%~C.L.
The inefficiencies due to different angular distributions
(possible for selectron or electron sneutrino pair production via the
$t$-channel) of produced sfermions and decay products
were estimated for five different MSSM parameter sets, representing different
neutralino field contents (gaugino/higgsino) and couplings,
and calculated separately for each analysis.
The selection efficiencies may vary by up to 10\%.
In interpreting the results,
a conservative approach was adopted by choosing the lowest
efficiencies in the limit calculation.
The systematic and the statistical
errors were added in quadrature and then subtracted when using the
number of background events.
\input{pr275_19}
\input{pr275_20}
\input{pr275_21}
\input{pr275_22}
\input{pr275_23}
\subsection{Selectron Limits}
Figure~\ref{fig:cross_selectron_lb} shows
upper limits on the cross-sections of pair-produced $\tilde{\mathrm e}$ followed
by a decay via a \lb\ coupling:
for (a) the direct decay of a right-handed $\tilde{\mathrm e}_R$,
(b) the direct decay of a left-handed $\tilde{\mathrm e}_L$ and
(c) the indirect decay of a $\tilde{\mathrm e}_R$.
The production cross-section for left-handed sfermion is always
larger that for right-handed sfermions, therefore we have
conservatively quoted results for right-handed sfermions only.
For all cases,
the worst upper limit on the cross-section is 0.36~pb.
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_06.eps ,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Selectron decays via a \lb\ coupling: Upper limits at the 95\% C.L.
on the pair-production
cross-sections
for (a) the direct decay of a right-handed $\tilde{\mathrm e}_R$,
(b) the direct decay of a left-handed $\tilde{\mathrm e}_L$ and
(c) the indirect decay of a $\tilde{\mathrm e}_R$.
Only the worst limit curve is shown and the \lb\ corresponding to it is
indicated.}
\label{fig:cross_selectron_lb}
\end{figure}
Figure~\ref{fig:cross_selectron_lbp} shows
upper limits on the cross-sections of pair-produced $\tilde{\mathrm e}$ followed
by a decay via a \lbp\ coupling:
for (a) the indirect decay of a $\tilde{\mathrm e}_R$ in the electron
channel,
(b) the indirect decay of a $\tilde{\mathrm e}_R$ in the muon channel and
(c) the indirect decay of a $\tilde{\mathrm e}_R$ in the tau channel. For all cases,
the weakest upper limit on the cross-section is 2.5~pb.
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_07.eps ,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Indirect selectron decays via a \lbp\ coupling:
Upper limits at the 95\% C.L. on the pair-production cross-sections
for (a) the indirect decay of a $\tilde{\mathrm e}_R$ in the electron
channel,
(b) the indirect decay of a $\tilde{\mathrm e}_R$ in the muon channel and
(c) the indirect decay of a $\tilde{\mathrm e}_R$ in the tau channel.
}
\label{fig:cross_selectron_lbp}
\end{figure}
Figure~\ref{fig:cross_sele_4jets} shows
upper limits on the cross-sections of pair-produced $\tilde{\mathrm e}$ directly
decaying via a \lbp\ coupling to a four-jet final state. The peak
structure visible in the figure at approximately the mass of the
W-boson comes from irreducible background due to WW pair-production.
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_08.eps,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Direct selectron decays via a \lbp\ coupling:
Upper limits at the 95\% C.L.
on the pair-production cross-sections of $\tilde{\mathrm e}$.
}
\label{fig:cross_sele_4jets}
\end{figure}
In the MSSM, the $\tilde{\mathrm e}$ pair-production cross-section
is enhanced by the presence of the $t$-channel diagram.
Figure~\ref{fig:mssm_selectron}(a)
shows the 95\% C.L. exclusion limits for right-handed selectrons
decaying directly or indirectly via a \lb\ coupling.
In the region where the $\tilde{\chi}^0_1$ is heavier than
the $\tilde{\mathrm e}$, only direct decays are possible.
When the $\tilde{\chi}^0_1$ is lighter than the $\tilde{\mathrm e}$, the
indirect decays are expected to be dominant.
For indirect decays via a \lb\ coupling, a right-handed
selectron with a mass smaller than 84~GeV
is excluded at the 95\% C.L. in the case of a low-mass
$\tilde{\chi}^0_1$. For direct decays via a \lb\ coupling,
a right-handed selectron with a
mass smaller than 84~GeV
is excluded at the 95\% C.L.
Figure~\ref{fig:mssm_selectron}~(b)
shows the 95\% C.L. exclusion limits for selectrons
decaying via a \lbp\ coupling. The exclusion refers to right-handed
selectrons for the indirect decays and to left-handed selectrons for
direct decays.
In the case of indirect decay,
a right-handed selectron with a mass smaller than 72~GeV
is excluded at the 95\% C.L. in the case of a low-mass
$\tilde{\chi}^0_1$ and a left-handed selectron with a mass smaller than 76~GeV
is excluded in the case of direct decays.
\begin{figure}[htbp]
\centering
\epsfig{file=pr275_09.eps, width=17.0cm}
\caption[]{\sl
Selectron:
MSSM exclusion region for $\tilde{\mathrm e}^+ \tilde{\mathrm e}^-$
production in
the $(m_{\tilde{\mathrm e}}, m_{\tilde{\chi}^0_1})$ plane at 95\% C.L.
for (a) a \lb\ coupling and (b) a \lbp\ coupling.
For the direct and indirect decays via \lb\ and the indirect decays
via \lbp\, the exclusion region for $\tilde{\mathrm e}_R \tilde{\mathrm e}_R$ is shown. For
the direct decays via \lbp\, the exclusion is shown for the only
possible case of $\tilde{\mathrm e}_L \tilde{\mathrm e}_L$.
The kinematic limit is shown as the dashed line.
The gap between the excluded regions for direct and indirect decays
corresponds to $\Delta m = m_{\tilde{\ell}} - m_{\tilde{\chi}^0_1} < 5$~GeV.}
\label{fig:mssm_selectron}
\end{figure}
\subsection{Smuon Limits}
Figures~\ref{fig:cross_smuon_lb} and
\ref{fig:cross_smuon_lbp}
show upper limits on the cross-sections for pair-produced $\tilde{\mu}$.
The weakest upper limit on the cross-section is 0.30~pb for the
\lb\ couplings and 0.48~pb for the \lbp\ couplings.
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_10.eps ,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Smuon decays via a \lb\ coupling: Upper limits at the 95\% C.L.
on the pair production
cross-sections of $\tilde{\mu}$
for (a) the direct decay of a right-handed $\tilde{\mu}_R$,
(b) the direct decay of a left-handed $\tilde{\mu}_L$ and
(c) the indirect decay of a $\tilde{\mu}_R$.
Only the worst limit curve is shown and the \lb\ corresponding to it is
indicated.}
\label{fig:cross_smuon_lb}
\end{figure}
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_11.eps ,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Smuon decays via a \lbp\ coupling:
Upper limits at the 95\% C.L. on the pair-production cross-sections of $\tilde{\mu}$
for (a) the indirect decay of a $\tilde{\mu}_R$ in the electron
channel,
(b) the indirect decay of a $\tilde{\mu}_R$ in the muon channel and
(c) the indirect decay of a $\tilde{\mu}_R$ in the tau channel.
}
\label{fig:cross_smuon_lbp}
\end{figure}
Figure~\ref{fig:cross_smu_4jets} shows
upper limits on the cross-sections of pair produced $\tilde{\mu}$ directly
decaying via a \lbp\ coupling to a four-jet final state.
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_12.eps ,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Smuon direct decays via a \lbp\ coupling:
Upper limits at the 95\% C.L. on the pair-production cross-sections of $\tilde{\mu}$.
}
\label{fig:cross_smu_4jets}
\end{figure}
In the MSSM,
for indirect decays via a \lb\ coupling, a right-handed
smuon with a mass smaller than 74~GeV
is excluded at the 95\% C.L. in the case of a low-mass
$\tilde{\chi}^0_1$, see Figure~\ref{fig:mssm_smuon}.
For direct decays via a \lb\ coupling,
a right-handed smuon with a
mass smaller than 66~GeV
is excluded at the 95\% C.L.
For indirect decays via a \lbp\ coupling, a right-handed smuon with a
mass smaller than 50~GeV
is excluded at the 95\% C.L. in the case of a low-mass
$\tilde{\chi}^0_1$: for direct decays a left-handed smuon with a mass smaller
than 64~GeV is excluded.
\begin{figure}[htbp]
\centering
\epsfig{file=pr275_13.eps, width=17.0cm}
\caption[]{\sl
Smuon:
MSSM exclusion region for $\tilde{\mu}^+ \tilde{\mu}^-$
production in
the $(m_{\tilde{\mu}}, m_{\tilde{\chi}^0_1})$ plane at 95\% C.L.
for (a) a \lb\ coupling and (b) a \lbp\ coupling.
For the direct decays via \lbp\, the exclusion region is shown for
the case $\tilde{\mu}_L \tilde{\mu}_L$. In the other cases, the exclusion regions
for $\tilde{\mu}_R \tilde{\mu}_R$ are shown.
The kinematic limit is shown as the dashed line. }
\label{fig:mssm_smuon}
\end{figure}
\subsection{Stau Limits}
Figures~\ref{fig:cross_stau_lb}
to \ref{fig:mssm_stau}
show the exclusion plots for pair-produced $\tilde{\tau}$.
The weakest upper limit on the cross-section is 0.30~pb for the
\lb\ couplings
and 0.45~pb for the \lbp\ couplings.
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_14.eps ,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Stau decays via a \lb\ coupling: upper limits on the pair-production
cross-sections
for (a) the direct decay of a right-handed $\tilde{\tau}_R$,
(b) the direct decay of a left-handed $\tilde{\tau}_L$ and
(c) the indirect decay of a $\tilde{\tau}_R$.
Only the worst limit curve is shown and the \lb\ corresponding to it is
indicated.}
\label{fig:cross_stau_lb}
\end{figure}
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_15.eps ,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Stau decays via a \lbp\ coupling:
Upper limits on the pair-production cross-sections
for the indirect decay of a $\tilde{\tau}_R$ in the electron
channel.
The indirect decay of a $\tilde{\tau}_R$ in the muon channel and
the indirect decay of a $\tilde{\tau}_R$ in the tau channel yield identical
results.
Only the worst limit curve is shown and the \lbp\ corresponding to it is
indicated.}
\label{fig:cross_stau_lbp}
\end{figure}
Pair-produced $\tilde{\tau}$ directly
decaying via a \lbp\ coupling to a four-jet final state yield
identical results as shown for the $\tilde{\mu}$ case,
see Figure~\ref{fig:cross_smu_4jets}.
In the MSSM, for indirect decays via a \lb\ coupling, a right-handed
stau with a mass smaller than 66~GeV
is excluded at the 95\% C.L. in the case of a low-mass
$\tilde{\chi}^0_1$. For direct decays via a \lb\ coupling, a right-handed stau with a
mass smaller than 66~GeV
is excluded at the 95\% C.L.
For indirect decays via a \lbp\ coupling, a right-handed stau with a
mass smaller than 66~GeV
is excluded at the 95\% C.L. in the case of a low-mass
$\tilde{\chi}^0_1$.
For direct decays, a left-handed stau with a mass smaller than 63~GeV
is excluded.
\begin{figure}[htbp]
\centering
\epsfig{file=pr275_16.eps, width=17.0cm}
\caption[]{\sl
Stau:
MSSM exclusion region for $\tilde{\tau}^+ \tilde{\tau}^-$
production in
the $(m_{\tilde{\tau}}, m_{\tilde{\chi}^0_1})$ plane at 95\% C.L.
for (a) a \lb\ coupling
and (b) a \lbp\ coupling.
For direct decays via \lbp\, the exclusion region for $\tilde{\tau}_L
\tilde{\tau}_L$ is shown. In the other cases, exclusion regions for $\tilde{\tau}_R
\tilde{\tau}_R$ are shown.
The kinematic limit is shown as the dashed line. }
\label{fig:mssm_stau}
\end{figure}
\subsection{Sneutrino Limits}
Figures~\ref{fig:cross_snu_lb} and
\ref{fig:cross_snu_lbp}
show the exclusion plots for pair produced $\tilde{\nu}$.
The weakest upper limit on the cross-section is 0.52~pb for the
\lb\ couplings and 1.8~pb for the \lbp\ couplings.
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_17.eps ,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Sneutrino decays via a \lb\ coupling: Upper limits at the 95\% C.L.
on the pair-production
cross-sections
for (a) the direct decay,
(b) the indirect decay of $\tilde{\nu}_{\mu}$ (or $\tilde{\nu}_{\tau}$) and
(c) the indirect decay of $\tilde{\nu}_e$.
Only the worst limit curve is shown and the \lb\ corresponding to it is
indicated.}
\label{fig:cross_snu_lb}
\end{figure}
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_18.eps ,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Sneutrino decays via a \lbp\ coupling:
Upper limits at the 95\% C.L. on the pair-production cross-sections.
Only the worst limit curve is shown and the \lbp\ corresponding to it is
indicated.}
\label{fig:cross_snu_lbp}
\end{figure}
Figure~\ref{fig:cross_sneu_4jets} shows
upper limits on the cross-sections of pair-produced
$\tilde{\nu}$ decaying directly via a \lbp\ coupling to a four-jet final state.
The searches for $\tilde{\nu}_{\mu}$ and $\tilde{\nu}_{\tau}$ yield identical limits.
\begin{figure}[htbp]
\centering
\begin{tabular}{c}
\epsfig{file=pr275_19.eps ,width=15.0cm} \\
\end{tabular}
\caption[]{\sl
Sneutrino direct decays via a \lbp\ coupling:
Upper limits at the 95\% C.L. on the
pair-production cross-sections of $\tilde{\nu}_{\mathrm e}$.
The search for $\tilde{\nu}_{\mu}$ and $\tilde{\nu}_{\tau}$ yield identical limits
to the ones shown in Figure 12.
}
\label{fig:cross_sneu_4jets}
\end{figure}
In the MSSM, the $\tilde{\nu}_e$ pair-production cross-section
is enhanced by the presence of the $t$-channel diagram.
Figure~\ref{fig:mssm_sneutrino}(a)
shows the 95\% C.L. exclusion limits for $\tilde{\nu}_e$
decaying directly or indirectly via a \lb\ coupling.
For indirect decays via a \lb\ coupling, an electron sneutrino
with a mass smaller than 87~GeV
is excluded at the 95\% C.L. in the case of a low-mass
$\tilde{\chi}^0_1$. For direct decays via a \lb\ coupling,
an electron sneutrino with a
mass smaller than 88~GeV
is excluded at the 95\% C.L.
Figure~\ref{fig:mssm_sneutrino}~(b)
shows the 95\% C.L. exclusion limits for electron sneutrinos
decaying indirectly via a \lbp\ coupling.
In this case, an electron sneutrino with a mass smaller than 86~GeV
is excluded at the 95\% C.L. in the case of a low-mass
$\tilde{\chi}^0_1$.
For direct decays, a sneutrino with a mass smaller than 80~GeV is
excluded.
MSSM exclusion plots for $\tilde{\nu}_{\mu}$ and $\tilde{\nu}_{\tau}$ are not shown
because of their very small cross-section.
For direct $\tilde{\nu}_{\mu}$ decay via a \lb\ coupling
a lower mass limit of 66 GeV is derived. For direct $\tilde{\nu}_{\mu}$ decay
via a \lbp\ coupling a lower mass limit of 58 GeV is obtained.
\begin{figure}[htbp]
\centering
\epsfig{file=pr275_20.eps, width=17.0cm}
\caption[]{\sl
Sneutrino:
MSSM exclusion region for $\tilde{\nu}_e \tilde{\nu}_e$
production in
the $(m_{\tilde{\nu}_e}, m_{\tilde{\chi}^0_1})$ plane at 95\% C.L.
for (a) a \lb\ coupling and (b) a \lbp\ coupling.
The kinematic limit is shown as the dashed line. }
\label{fig:mssm_sneutrino}
\end{figure}
\subsection{Stop Limits}
For the stop search in the electron and muon channel,
no events satisfy the final selection cuts.
A cross-section limit of 0.15~pb was derived for the pair-production
of stops decaying directly via \lbp$_{13k}$ or \lbp$_{23k}$,
in the mass region 45~GeV $< m_{\stopx} <$90~GeV. The
excluded cross-section as a function of the stop mass is shown in
Fig.~\ref{fig:limit_stop}~(a).
If one assumes a stop production cross-section as predicted by the MSSM,
masses lower than 82~GeV can be excluded for any mixing
angle $\theta_{\stopx}$ under
the assumptions made above.
For the stop search in the tau channel, two events have satisfied the
final selection cuts.
A cross-section limit of 0.24~pb was derived for the pair-production
of the stops decaying directly via \lbp$_{i3k}$,
in the mass region 45~GeV $< m_{\stopx} <$ 90~GeV. The
excluded cross-section as a function of the stop mass is shown in
Fig.~\ref{fig:limit_stop}~(b).
In the tau channel,
masses lower than 73~GeV can be excluded for any mixing
angle $\theta_{\stopx}$.
More detailed exclusion limits
are given in Table~\ref{tab:limit_stop}.
For the stop decays via \lbpp couplings, 7 events satisfied the
selection cuts. A cross-section limit of approximately 0.3~pb was
derived for a stop mass up to $\approx$~75 GeV degrading slightly
in the range of the W mass as shown in Fig.~\ref{fig:limit_stop_4jets}.
\begin{table}[htb]
\begin{center}
\begin{tabular}{|c||c|c|}
\hline
Limits & $\theta_{\stopx} = 0$ rad & $\theta_{\stopx} = 0.98$ rad \\
\hline
\hline
$ \stopm \rightarrow {\rm e} +$ q & 86 GeV & 82 GeV \\
\hline
$ \stopm \rightarrow \mu +$ q & 86 GeV & 82 GeV \\
\hline
$ \stopm \rightarrow \tau +$ q & 81 GeV & 73 GeV \\
\hline
$ \stopm \rightarrow $ qq & 79 GeV & 76 GeV \\
\hline
\end{tabular}
\caption[]{\sl
Mass limits for stop for the two extreme values of the mixing
angle in the electron, muon and tau channels as well as in the 4-jet
channel.}
\label{tab:limit_stop}
\end{center}
\end{table}
\begin{figure}
\centering
\begin{tabular}{c}
\epsfig{file=pr275_21.eps,width=15cm} \\
\end{tabular}
\caption[]{\sl
Stop direct decays via a \lbp\ coupling:
Cross-section limits at the 95\% C.L.
in the electron and muon channels (a) and
in the tau channel (b). Also shown are the maximum (dashed-dotted line)
and minimum (dashed line)
cross-sections predicted by the MSSM,
corresponding to a mixing angle of 0 rad and 0.98 rad (decoupling
limit).
}
\label{fig:limit_stop}
\end{figure}
\begin{figure}
\centering
\begin{tabular}{c}
\epsfig{file=pr275_22.eps,width=15cm} \\
\end{tabular}
\caption[]{\sl
Stop direct decays via a \lbpp\ coupling:
Upper limits at the 95\% C.L. on the production cross-section.
Also shown are the maximum (dashed-dotted line) and minimum (dashed line)
cross-sections predicted by the MSSM,
corresponding to a mixing angle of 0 rad and 0.98 rad. }
\label{fig:limit_stop_4jets}
\end{figure}
\section{Conclusions}
\label{sec:conclusions}
We have performed a search for pair produced
sfermions with $R$-parity\ violating
decays using the data collected by the OPAL detector at
$\sqrt{s} \simeq 183$~GeV corresponding to a luminosity of
approximately 56
pb$^{-1}$.
Direct and indirect $R$-parity\ violating decay modes of $\tilde{\ell}$, $\tilde{\nu}$
via the Yukawa-like
\lb\ and \lbp\ couplings as well as
direct $R$-parity\ violating decay modes of
$\stopx$ via \lbp\ and
\lbpp\ were considered.
No significant excess of events has been observed in the data.
Upper limits on the pair production cross-sections for
sfermions
have been computed assuming that only $R$-parity\ violating decays occur.
These cross-section limits, within the MSSM frame used,
depend only on the mass of the sfermion and
not on other SUSY parameters.
Mass limits were derived in the framework of the
constrained Minimal Supersymmetric Standard Model
whenever the predicted cross-sections were sufficiently large.
|
\section{Introduction}
Recently, a great experimental material has been obtained for anomalous $f$%
-systems demonstrating so-called non-Fermi-liquid (NFL) behavior \cite
{Maple,Proc}. Manifestations of the NFL behavior are unusual temperature
dependences of magnetic susceptibility ($\chi (T)\sim T^{-\zeta },\zeta <1$%
), electronic specific heat ($C(T)/T$ is proportional to $T^{-\zeta }$ or $%
-\ln T$) and resistivity ($\rho (T)\sim T^\mu ,\mu <2$), etc. Such a
behavior is observed not only in alloys where disorder is present (U$_x$Y$%
_{1-x}$Pd$_3$, UPt$_{3-x}$Pd$_x$, UCu$_{5-x}$Pd$_x$, CeCu$_{6-x}$Au$_x$, U$_x
$Th$_{1-x}$Be$_{13}$), but also in some stoichiometric compounds, e.g., Ce$_7
$Ni$_3$ \cite{CeNi}, CeCu$_2$Si$_2,$CeNi$_2$Ge$_2$ \cite{Steg}. The latter
situation is most interesting from the physical point of view \cite{Alt}.
There are a number of theoretical mechanisms proposed to describe the NFL
state: two-channel Kondo scattering \cite{Tsv,twoch,Col}, ``Griffiths
singularities'' in disordered magnets \cite{Castr,And}, strong spin
fluctuations near a quantum magnetic phase transition \cite{Jul,Ioffe,Col1}
etc. Most of modern treatments of the NFL problem have a
semiphenomenological character. The only microscopic model where formation
of the NFL state is proven - the one-impurity two-channel Kondo model -
seems to be insufficient, since important role of intersite interactions is
now a matter of common experience \cite{Maple}.
In the present paper we start from the standard microscopic model of a
periodical Kondo lattice. Main role in the physics of the Kondo lattices
belongs to the interplay of the on-site Kondo screening and intersite
exchange interactions. This interplay results in the mutual renormalization
of the characteristic energy scales: the Kondo temperature $T_K$ and
spin-fluctuation frequency $\overline{\omega }$. We shall demonstrate that
during the renormalization process ``soft'' boson branches can be formed,
presence of singularities of spin spectral function being of crucial
importance. Scattering of electrons by such soft collective excitations just
leads to the formation of the NFL state (cf. \cite{Col1}).
In Sect. 2 the renormalization group (scaling) equations are presented. In
Sect.3 we consider the antiferromagnetic (AFM) state with account of the
spin-wave damping and the paramagnetic state with simple spin-diffusive
dynamics. It turns out that in these cases a NFL-like behavior (in a
restricted temperature interval) is possible, especially in the case of
quasi-two-dimensional (2D) spin fluctuations. In Sect. 4 we consider the
general problem of singularities of the scaling function. In Sect.5 we show
that the NFL behavior up to lowest temperatures can be naturally obtained
provided that we take into account magnon-like excitations in the case of a
complicated spin dynamics. The excitation picture required is characteristic
for real $f$-systems where several excitation branches exist. In Sect.6 we
discuss various physical properties and possible relation to experimental
data.
\section{The scaling equations}
To describe a Kondo lattice, we use the $s-f$ exchange model
\begin{equation}
H=\sum_{{\bf k}\sigma }t_{{\bf k}}c_{{\bf k}\sigma }^{\dagger }c_{{\bf k}%
\sigma }-I\sum_{i\alpha \beta }{\bf S}_i\mbox {\boldmath $\sigma $}_{\alpha
\beta }c_{i\alpha }^{\dagger }c_{i\beta }+\sum_{{\bf q}}J_{{\bf q}}{\bf S}_{%
{\bf -q}}{\bf S}_{{\bf q}}+H_a
\end{equation}
where $t_{{\bf k}}$ is the band energy, ${\bf S}_i$ and ${\bf S}_{{\bf q}}$
are spin-density operators and their Fourier transforms, $I$ is the $s-f$
exchange parameter, $J_{{\bf q}}$ are the intersite exchange parameters, $%
\sigma $ are the Pauli matrices, $H_a$ is the anisotropy Hamiltonian which
results in occurrence of the gap $\omega _0$ in the spin-wave spectrum. In
Refs.\cite{kondo,kondo1}, the interplay of the Kondo effect and intersite
interactions was investigated by the renormalization group approach. The
latter starts from the second-order perturbation theory with the use of the
equation-of-motion method (within the diagram technique in the pseudofermion
representation for the spin operators, such an approximation corresponds to
the one-loop scaling). The relevant variables are the effective
(renormalized) parameter of $s-f$ coupling $g_{ef}(C)=-2\rho I_{ef}(C)$ ($%
C\rightarrow -0$ is a flow cutoff parameter, $\rho $ is the bare density of
electron states at the Fermi level), characteristic ``exchange''
spin-fluctuation energy $\overline{\omega }_{ex}(C)$, gap in the spin-wave
spectrum $\omega _0(C)$, and magnetic moment $\overline{S}_{ef}(C)$. In the
magnetically ordered phase, $\overline{\omega }_{ex}$ is the magnon
frequency $\omega _{{\bf q}},$ which is averaged over the wavevectors ${\bf %
q=}2{\bf k}${\bf \ }where ${\bf k}$ runs over the Fermi surface; in the case
of dissipative spin dynamics (paramagnetic phase) $\overline{\omega }_{ex}$
is determined by the second moment of the spin spectral density.
Here we write down the set of scaling equations with account of the magnon
damping $\overline{\gamma }(C)$:
\begin{eqnarray}
\partial g_{ef}(C)/\partial C &=&\Lambda ,\partial \ln \overline{S}%
_{ef}(C)/\partial C=-\Lambda /2 \label{gl} \\
\partial \ln \overline{\omega }_{ex}(C)/\partial C &=&-a\Lambda /2,\partial
\ln \omega _0(C)/\partial C=-b\Lambda /2 \label{sl} \\
\partial \ln \overline{\gamma }(C)/\partial C &=&-c\Lambda /2 \label{gaml}
\end{eqnarray}
with
\[
\Lambda =\Lambda (C,\overline{\omega }_{ex}(C),\omega _0(C))=\frac{%
g_{ef}^2(C)}{|C|}\eta \left( \frac{\overline{\omega }_{ex}(C)}{|C|},\frac{%
\omega _0(C)}{|C|},\frac{\overline{\gamma }(C)}{|C|}\right) ,
\]
$a=1-\alpha $ for the paramagnetic (PM) phase, $a=1-\alpha ^{\prime },b=1$
for the antiferromagnetic (AFM) phase, $a=2(1-\alpha ^{\prime \prime }),b=2$
for the ferromagnetic (FM) phase; $\alpha ,\alpha ^{\prime },\alpha ^{\prime
\prime }$ are some averages over the Fermi surface (see Ref.\cite{kondo}).
For the staggered AFM ordering in the 2D and 3D cubic lattices where
\[
J_{{\bf q}}=2J_1\sum_{i=1}^d\cos q_i+4J_2\sum_{i>j}^d\cos q_i\cos q_j
\]
with $J_1$ and $J_2$ being the exchange integrals between nearest and
next-nearest neighbors ($|J_1|\gg |J_2|$), one can derive by using expansion
in small $q$%
\begin{equation}
\alpha ^{\prime }\simeq 2(d-1)\frac{J_2}{J_1}\left| \left\langle \exp (i{\bf %
kR}_2)\right\rangle _{t_{{\bf k}}=0}\right| ^2 \label{alph}
\end{equation}
where ${\bf R}_2$ runs over the next-nearest neighbors, the electron
spectrum $t_{{\bf k}}$ is referred to the Fermi energy $E_F=0$. Similar
equations can be obtained for the Coqblin-Schrieffer model \cite{kondo} and
in the case of anisotropic $s-f$ coupling \cite{kondo1}.
Using (\ref{gl})-(\ref{gaml}) we obtain the explicit expressions
\begin{eqnarray}
\overline{\omega }_{ex}(C) &=&\overline{\omega }_{ex}\exp (-a[g_{ef}(C)-g]/2)
\label{w+g} \\
\omega _0(C) &=&\omega _0\exp (-b[g_{ef}(C)-g]/2) \nonumber \\
S_{ef}(C) &=&S\exp (-[g_{ef}(C)-g]/2) \nonumber \\
\overline{\gamma }(C) &=&kg_{ef}^2(C)\overline{\omega }\exp
(-c[g_{ef}(C)-g]/2) \nonumber
\end{eqnarray}
In the last equation of (\ref{w+g}), we have taken into account that the
magnon damping is proportional to $g^2,$ $\overline{\gamma }=kg^2\overline{%
\omega }$ (the factor $k$ is determined by the bandstructure and magnetic
ordering), and the scaling equations should contain only renormalized
quantities.
\section{The scaling behavior in the presence of damping}
As demonstrated in Ref.\cite{kondo}, the singularities of the scaling
function $\eta $ can result in occurrence of a NFL behavior in a restricted
region due to fixing of the argument of the function $\eta $ at the
singularity during the scaling process, so that $\overline{\omega }(C)\simeq
|C|,$%
\begin{equation}
g_{ef}(\xi )-g\simeq 2(\xi -\lambda )/a,\lambda \equiv \ln (D/\overline{%
\omega }), \label{linnn}
\end{equation}
(the scale $D$ is defined by $g_{ef}(-D)=g,\xi \equiv \ln |D/C|$). This
region becomes not too narrow only provided that the bare coupling constant $%
g=-2I\rho $ is very close to the critical value $g_c$ for the magnetic
instability ($|g-g_c|/g_c\sim 10^{-4}\div 10^{-6}$). Here we consider the
scaling process with account of a not too small magnon damping (only very
small damping was introduced in Ref.\cite{kondo} to provide existence of the
magnetic-non-magnetic ground-state transition at $g=g_c$).
The scaling function $\eta $ in the FM and AFM phases for simple magnetic
structures reads (for simplicity, magnetic anisotropy is neglected in this
Section)
\begin{equation}
\eta \left( \overline{\omega }_{ex}/|C|,\overline{\gamma }/|C|\right) =\text{%
Re}\left\langle \left( 1-(\omega _{{\bf k-k}^{\prime }}^{}+i\gamma _{{\bf k-k%
}^{\prime }}^{})^2/C^2\right) ^{-1}\right\rangle _{t_k=t_{k^{\prime }}=0}
\label{etafm}
\end{equation}
For an isotropic three-dimensional (3D) ferromagnet integration in (\ref
{etafm}) for $\gamma =$ const and quadratic spin-wave spectrum yields
\begin{equation}
\eta (x,z)=\frac{1+z^2}{4x}\ln \frac{(1+x)^2+z^2}{(1-x)^2+z^2}+\frac{1+x}z%
\arctan \frac z{1+x}-\frac{1-x}z\arctan \frac z{1-x} \label{intfm}
\end{equation}
where $x=\overline{\omega }_{ex}/|C|,z=$ $\overline{\gamma }/|C|.$ Note that
last two terms in (\ref{intfm}) play a role similar to that of the
``incoherent'' contribution to the function $\eta ,$ which was treated in
Ref.\cite{kondo}.
For a 3D antiferromagnet integration in (\ref{etafm}) for the linear
spin-wave spectrum gives
\begin{equation}
\eta (x,z)=\frac 12\text{Re }[(1+iz)\ln (1+x+iz)+(1-iz)\ln (1+x-iz)]
\end{equation}
where we take into account the intersubband damping only,
\begin{eqnarray}
\gamma _{{\bf q}} &=&\pi 2I^2\overline{S}(J_{{\bf q}}-J_{{\bf Q}})\rho
^2\lambda _{{\bf q+Q}}, \label{dintr} \\
\lambda _{{\bf q}} &=&\rho ^{-2}\sum_{{\bf k}}\delta (t_{{\bf k}})\delta (t_{%
{\bf k+q}}).
\end{eqnarray}
The damping (\ref{dintr}) can be put nearly constant at not too large $q$
(the threshold value determined by the AFM gap can be neglected due to
formal smallness in $I$). In the 2D case we obtain in the same approximation
\begin{eqnarray}
\eta (x,z) &=&\frac{\nu ^3(x,z)}{\nu ^4(x,z)+z^2} \\
\nu ^2(x,z) &=&\frac 12[1-x^2+\sqrt{(1-x^2)^2+4z^2}] \nonumber
\end{eqnarray}
which modifies somewhat the result of Ref.\cite{kondo}.
The function $\lambda _{{\bf q}}$ determines, in particular, the factor $k$
in (\ref{w+g}). For a parabolic electron spectrum we obtain
\begin{equation}
\lambda _{{\bf q}}=\frac{\theta (1-x)}{zx}\times \left\{
\begin{tabular}{ll}
$1/6,$ & $d=3$ \\
$(4\pi )^{-1}(1-x^2)^{-1/2},$ & $d=2$%
\end{tabular}
\right. \label{lam}
\end{equation}
where $x=q/2k_F,$ $\theta (x)$ is the step function, $z$ is the electron
concentration (with both spin projections).
Main Kondo renormalization of the magnon damping comes from its
proportionality to the factor of $\overline{S}.$ Spin fluctuations can give
correction to this factor, as well as for the magnon frequency \cite{kondo}.
For simplicity, we restrict ourselves in numerical calculations to the AFM
case with $\delta \gamma _{{\bf q}}/\gamma _{{\bf q}}=\delta \omega _{{\bf q}%
}/\omega _{{\bf q}}=\delta \overline{S}/S,$ so that $c=1$ in (\ref{gaml}), (%
\ref{w+g}). The corresponding scaling trajectories are shown in Fig.1. One
can see that, unlike Ref.\cite{kondo}, the ``linear'' behavior, although
being somewhat smeared, is pronounced in a considerable region of $\xi $ for
not too small $|g-g_c|$, especially in the 2D case. In the 3D case the
linear region (\ref{linnn}) is followed by a quasi-linear behavior with
\begin{equation}
g_{ef}(\xi )\simeq A(\xi -\lambda ),\overline{\omega }(C)\propto |C|^{aA/2},%
\overline{S}_{ef}(C)\propto |C|^{A/2}, \label{lina}
\end{equation}
where $A<2/a.$ To investigate the latter behavior in more details, it is
instructive to consider also the case of a paramagnet with pure dissipative
dynamics. In the case of spin-diffusion behavior we have (cf. Ref.\cite
{kondo})
\begin{equation}
\eta ^{PM}(\frac{\overline{\omega }}C)=\sum_{{\bf q}}\lambda _{{\bf q}%
}\left[ 1+({\cal D}q^2/C)^2\right] ^{-1},\overline{\omega }=4{\cal D}k_F^2
\label{etpm}
\end{equation}
where ${\cal D}$ is the spin diffusion constant. As demonstrate numerical
calculations (see Fig.2), for $g\leq g_c$ the one-impurity behavior $%
1/g_{ef}(\xi )=1/g-\xi $ is changed at $\xi \simeq \lambda $ by the behavior
(\ref{lina}) with
\begin{equation}
A\simeq [g_{ef}(\lambda )]^2\Psi (0),g_{ef}(\lambda )\simeq g/(1-\lambda
g),\Psi (0)=\eta (1)\sim 0.5 \nonumber
\end{equation}
In the 3D case where $\eta ^{PM}(x)=\arctan x/x$ the quasi-linear NFL-like
behavior $g_{ef}(\xi )$ takes place in a rather narrow region. However, in
the 2D case we obtain
\begin{equation}
\eta ^{PM}(x)=\left[ \frac{1+(1+x^2)^{1/2}}{2(1+x^2)}\right] ^{1/2}
\end{equation}
and the NFL-like region becomes more wide due to a more slow decrease of $%
\eta ^{PM}(x)$ at $x\rightarrow \infty .$ Note that such regions are not
observed for $g>g_c.$
\section{Singularities of the scaling function}
To get further insight into the NFL-behavior problem, we perform an analysis
of singularities of the scaling function $\eta ,$ which is more general than
in Ref.\cite{kondo}. In the absence of damping we can write down
\begin{equation}
\eta \left( \overline{\omega }_{ex}/|C|,\omega _0/|C|\right) =\sum_{{\bf q}%
}\lambda _{{\bf q}}\left( 1-\omega _{{\bf q}}^2/C^2\right) ^{-1}
\label{etafm1}
\end{equation}
The singularities of $\eta $ correspond to the Van Hove singularities in the
magnon spectrum and to the boundary points $q=0$ and $q=2k_F.$
For $q\rightarrow 2k_F$ the magnon spectrum has the Kohn anomaly
\begin{equation}
\delta \omega _{{\bf q}}\propto \left\{
\begin{tabular}{ll}
$(q-2\dot k_F)\ln |q-2k_F|,$ & $d=3$ \\
$\sqrt{q^2-4k_F^2}\theta (q-2k_F),$ & $d=2$%
\end{tabular}
\right.
\end{equation}
Taking into account the dependences (\ref{lam}), which hold qualitatively in
a general situation, we obtain for $v=C^2-\overline{\omega }^2\rightarrow 0$%
\begin{equation}
\eta (v)\propto \left\{
\begin{tabular}{ll}
$\ln \ln |v|,$ & $d=3$ \\
$\theta (v)v^{-1/2},$ & $d=2$%
\end{tabular}
\right.
\end{equation}
Note that the singularity of the form $\ln v$ obtained for the $3D$ case in
Ref.\cite{kondo} (see also previous Section), is a consequence of a
simplified ``Debye'' model of the magnon spectrum. In fact, such a
dependence corresponds qualitatively to an intermediate asymptotics at
approaching the singularity. In any case, a small damping of spin
excitations should be introduced to cut the singularity. In the calculations
below we put the damping parameter $\delta =1/100$ (see Ref.\cite{kondo}).
In fact, pronounced extrema of $\eta ,$ but not singularities themselves
turn out to be important for the scaling behavior discussed below.
For $q\rightarrow 0$ we have $\lambda _{{\bf q}}\propto q^{-1}.$ Near the
points of minimum (maximum) in the magnon spectrum we have $\omega _{{\bf q}%
}^2-\omega _m^2\propto \pm q$ $^2,$ and for $v=C^2-\omega _m^2\rightarrow 0$
we obtain
\begin{equation}
\eta (v)\propto \left\{
\begin{tabular}{ll}
$\pm \ln |v|,$ & $d=3$ \\
$\mp \theta (\mp v)|v|^{-1/2},$ & $d=2$%
\end{tabular}
\right.
\end{equation}
so that $\eta \rightarrow -\infty $ near the band bottom and $\eta
\rightarrow +\infty $ near the band top. The Van Hove singularities in the
magnon band at $\omega =\omega _c$ for $q\neq 0$ yield weaker singularities
in $\eta (v)$ ($|v|^{1/2}$ for $d=3$ and a finite jump for $d=2$) and will
not be treated below.
\section{The scaling picture and NFL behavior in many-sublattice magnets}
Using the results of the previous Section we can propose a rather realistic
and universal mechanism of the NFL behavior. Suppose that the spin
excitation spectrum contains several branches which make additive
contributions to the function $\eta .$
As a simple example we can consider a two-sublattice ferrimagnet with the
localized-system Hamiltonian
\begin{equation}
H_f=\sum_{{\bf q}}(J_{{\bf q}}{\bf S}_{{\bf -q}}{\bf S}_{{\bf q}}+J_{{\bf q}%
}^{\prime }{\bf S}_{{\bf -q}}^{\prime }{\bf S}_{{\bf q}}^{\prime }+%
\widetilde{J}_{{\bf q}}{\bf S}_{{\bf -q}}{\bf S}_{{\bf q}}^{\prime }),
\end{equation}
the $s-f$ exchange interaction being taken into account only at one
sublattice (spins without primes). Similar to (\ref{etafm1}) we obtain
\begin{equation}
\eta =\sum_{{\bf q,}i=1,2}\lambda _{{\bf q}}\frac{\omega _{{\bf q}i}}{B_{%
{\bf q}}}\left( 1-\frac{\omega _{{\bf q}i}^2}{C^2}\right) ^{-1}
\end{equation}
where
\begin{eqnarray}
B_{{\bf q}} &=&\{[S(J_{{\bf q}}-J_0)+S^{\prime }(J_{{\bf q}}^{\prime
}-J_0^{\prime })-(S+S^{\prime })\widetilde{J}_0]^2-4SS^{\prime }\widetilde{J}%
_{{\bf q}}^2\}^{1/2}, \\
\omega _{{\bf q}1,2} &=&B_{{\bf q}}\mp |S(J_{{\bf q}}-J_0)-S^{\prime }(J_{%
{\bf q}}^{\prime }-J_0^{\prime })+(S-S^{\prime })\widetilde{J}_0| \nonumber
\end{eqnarray}
are the acoustical and optical modes.
The dependence
\begin{equation}
\Psi (\xi )=\sum_iz_i\eta _i[(\overline{\omega }_{ex,i}/D)e^\xi ,(\omega
_{0,i}/D)e^\xi ] \label{psiop}
\end{equation}
is shown in Fig.3 for the case of two excitation modes in a 3D
antiferromagnet (the expressions for the function $\eta _i$ in the case of
one mode with inclusion of anisotropy are given in Ref.\cite{kondo1}). The
only property of the function $\Psi $, which will be important below, is the
occurrence of the {\it second }zero with decreasing $|C|$ (or increasing $%
\xi $). This property follows immediately from existence of the ``positive''
singularity in $\Psi $ near the maximum frequency of the lower branch and of
the ``negative'' singularity near the minimum frequency of the upper branch
(e.g., for $\omega _1(q=2k_F)<\omega _2(q=0)$). One can expect that this is
a general property of many-sublattice magnets. The singularities can be also
connected with the crystal-field excitations \cite{FLow}.
As demonstrate both numerical calculations and analytical treatment, in some
interval of the bare coupling parameter, $g_{c1}<g<g_{c2},$ the argument of
the function $\Psi $ becomes fixed at the second zero, $C=C_0,$ during the
scaling process which is described by Eq.(\ref{gl}). This can be illustrated
for the simple case where $a=b$ for all the modes (e.g., for an
antiferromagnet in the nearest-neighbor approximation we have $a=b=1$). On
substituting (\ref{w+g}) into (\ref{gl}) we obtain
\begin{equation}
\partial (1/g_{ef})/\partial \xi =-\Psi (\xi -a[g_{ef}-g]/2) \label{gl1}
\end{equation}
Then we derive
\begin{equation}
g_{ef}(\xi \rightarrow \infty )\simeq (2/a)(\xi -\xi _0)-1/[\Psi ^{\prime
}(\xi _0)\xi ^2] \label{lin}
\end{equation}
where $\xi _0\sim \ln (D/\overline{\omega })$ is the second zero of the
function $\Psi (\xi ).$ Note that the first zero does not work since the
unrestricted increase of $g_{ef}$ corresponds to a decrease of the argument
of the function $\Psi $ in (\ref{gl1}), so that $\Psi \rightarrow +0.$
According to (\ref{w+g}) we obtain
\begin{equation}
\overline{\omega }_{ex}(C),\omega _0(C)\propto |C|,\overline{S}%
_{ef}(C)\propto |C|^{1/a} \label{lin1}
\end{equation}
Within the approach used, the behavior (\ref{lin}),(\ref{lin1}) takes place
up to $C=0$ $(\xi =\infty ).$ Although the one-loop scaling equations
themselves may become invalid with increasing $g_{ef},$ the tendency to the
formation of the ``soft'' magnon mode seems to be physically correct. The
scaling picture for three possible cases is shown in Fig.4. One can see that
the interval $[g_{c1},g_{c2}]$ where the NFL behavior occurs is not too
small, unlike Ref.\cite{kondo}.
In a more general case, where $a\neq b$ and the exponents in (\ref{sl})
differ for different frequencies, the linear $C$-dependence takes place only
for the total characteristic frequency $\overline{\omega }$ (e.g., for an
anisotropic antiferromagnet with one mode we have $\overline{\omega }%
^2=\omega _0^2+\overline{\omega }_{ex}^2$), and the behavior $g_{ef}(\xi )$
and $\overline{S}_{ef}(\xi )$ is more complicated. As follows from (\ref
{alph}), for the AFM state with small next-nearest exchange interactions the
case of small $|a-1|$ and $b=$ $1$ is realized.
\section{Discussion of physical properties and conclusions}
Consider the temperature dependence of the magnetic susceptibility. In the
spin-wave region we have for an AFM structure with the wavevector ${\bf Q}$%
\begin{equation}
\chi =\lim_{q\rightarrow 0}\langle \langle S_{{\bf q}}^x|S_{-{\bf q}%
}^x\rangle \rangle _{\omega =0}=(J_0-J_{{\bf Q}})^{-1}\propto \overline{S}/%
\overline{\omega }
\end{equation}
One can assume that the spin-wave description of the electron-magnon
interaction is adequate not only in the AFM phase, but also for systems with
a strong short-range AFM order (e.g., for 2D and frustrated 3D systems at
finite temperatures). Anomalous $f$-systems demonstrate indeed pronounced
quasi-2D spin fluctuations, see, e.g., Refs.\cite{Col1,Stock}. Using the
scaling arguments we can replace $\overline{\omega }\rightarrow \overline{%
\omega }(C),$ $\overline{S}\rightarrow \overline{S}_{ef}(C)$ with $|C|\sim
T, $ which yields
\begin{equation}
\chi (T)\propto T^{-\zeta },\zeta =(a-1)/a
\end{equation}
According to (\ref{alph}), the non-universal exponent $\lambda $ is
determined by details of magnetic structure and can be both positive and
negative. For a qualitative discussion, we can still use Fig.4 and treat the
difference $a-1$ as a perturbation. The increase of $\chi (T\rightarrow 0)$
(which is usually called NFL behavior) takes place for $a>1$ and, as follows
from (\ref{w+g}), corresponds to an increase of magnetic anisotropy
parameter with lowering $T$ (see Ref.\cite{kondo1}). Such a correlation may
be verified experimentally.
The temperature dependence of electronic specific heat can be estimated from
the second-order perturbation theory, $C_{el}(T)/T\propto 1/Z(T)$ where $%
Z(T) $ is the residue of the electron Green's function at the distance $T$
from the Fermi level (cf. Ref.\cite{IKFTT}). Then we have
\begin{equation}
C_{el}(T)/T\propto g_{ef}^2(T)\overline{S}_{ef}(T)/\overline{\omega }%
_{ex}(T)\propto \chi (T)\ln ^2T
\end{equation}
The dependence $C_{el}(T)/T\propto \chi (T)$ has been recently obtained
experimentally for a wide class of NFL systems \cite{And}. One can expect
that the accuracy of the experimental data is insufficient to pick out the
factor of $\ln ^2T.$ Generally, the temperature behavior of magnetic
characteristics ($\overline{S}$ and $\overline{\omega }$), which depend
exponentially on the coupling constant, is decisive for our NFL mechanisms.
The transport relaxation rate determining the temperature dependence of the
resistivity owing to scattering by spin fluctuations in AFM phase is given
by \cite{afm}
\begin{equation}
\frac 1\tau =\frac{2\pi }{v_F^2}I^2\overline{S}^2(J_0-J_{{\bf Q}})\rho
\langle ({\bf v_{k+Q}-v}_{{\bf k}})^2\rangle _{t_{{\bf k}}=0}\sum_{{\bf %
q\simeq Q}}\lambda _{{\bf q}}\left( -\frac{\partial N_{{\bf q}}}{\partial
\omega _{{\bf q}}}\right)
\end{equation}
Then we obtain
\begin{equation}
\ \frac 1\tau \propto g_{ef}^2(T)\overline{S}_{ef}(T)/\overline{\omega }%
_{ex}(T)T^2\propto T^2C_{el}(T)/T\propto T^{2-\zeta } \label{taul}
\end{equation}
Considering electron-electron scattering as the main scattering mechanism
one can expect another temperature dependence, namely
\begin{equation}
\frac 1\tau \propto \left[ \rho /Z(T)\right] ^2T^2\propto T^2\left[
C_{el}(T)/T\right] ^2\propto T^{2-2\zeta } \label{tau2}
\end{equation}
It would be interesting to compare experimental dependences on one hand, of
the resistivity and, on the other hand, of the susceptibility and specific
heat, to choose between dependences (\ref{taul}) and (\ref{tau2}). Validity
of the relation (\ref{taul}) would be a verification of the scenario for NFL
state formation proposed here.
The temperature dependences of specific heat, magnetic susceptibility and
resistivity in the case of a NFL-like behavior considered in Sect.3 differ
from those discussed above by the value of $\zeta =(a-1)A/2$ (for the second
``linear'' region).
To conclude, phenomenon of the NFL behavior seems to have a complicated
nature, so that it is hardly possible to propose an unified picture for all
the cases. The mechanisms considered above are not based on disorder
effects, but describe naturally the NFL state in ideal crystals. At the same
time, the damping makes the quasi-NFL behavior considered in Sect.3 more
pronounced and in a sense plays the role of disorder.
The damping is not important for the NFL behavior mechanism considered in
Sect.5. This NFL picture is ``true'' (i.e. holds up to lowest temperatures)
within the lowest-order scaling approach; treatment of higher-order
corrections to the scaling equations would provide additional information.
Unlike previous phenomenological works, existence of peculiar long-range
critical fluctuations near the quantum phase transition is not needed for
this mechanism, but local reconstruction of electronic states owing to the
Kondo effect is essential, the concrete form of spin spectral function being
of crucial importance. More detailed investigations of the NFL behavior for
complicated spectral functions, in particular with account of incoherent
contributions, would be also of interest.
The research described was supported in part by Grant No.99-02-16279 from
the Russian Basic Research Foundation.
\newpage\
{\sc Figure captions}
Fig.1. The scaling trajectories $g_{ef}(\xi )$ in isotropic 2D
antiferromagnets (solid lines, $g=0.154<g_c,g=0.155>g_c$) and 3D
antiferromagnets (dashed lines, $g=0.139<g_c,g=0.140>g_c$) with $%
k=0.5,a=c=1,\lambda =\ln (D/\overline{\omega })=5.$
Fig.2. The scaling trajectories $g_{ef}(\xi )$ in 2D (solid lines) and 3D
(dashed lines) paramagnets with $a=1/2,\lambda =5.$ The bare coupling
parameters are $g=0.135$ and $g=0.145,$ higher curves corresponding to
larger $g.$ The values of $g_c$ in 2D and 3D cases are 0.148 and 0.152.
Fig.3. The scaling functions $\Psi (\xi )$ defined by (\ref{psiop}) in the
case of a 3D antiferromagnet with two excitation modes The parameters are $%
z_1=0.4,z_2=0.6,$ $\ln (D/\overline{\omega }_2)=4,\overline{\omega }_2/%
\overline{\omega }_1=3,$ $\omega _{0,1}/\overline{\omega }_{ex,1}=0.2,\omega
_{0,2}/\overline{\omega }_{ex,2}=0.6$ ($\overline{\omega }_i^2=\overline{%
\omega }_{ex,i}^2+\omega _{0,i}^2$).
Fig.4. The scaling trajectories $g_{ef}(\xi )$ in a 3D antiferromagnet with
the parameters of Fig.3 and $a_i=b_i=1.$ The bare coupling parameters are $%
g=0.158\simeq g_{c2}$ (upper solid line with the asymptotics), $%
g=0.154\simeq g_{c1}$ (lower solid line), and $g_{c1}<g=0.156<g_{c2}$
(dashed line).
|
\section{INTRODUCTION}
This is the fourth 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],
which is based on our approach to investigation
of nonlinear problems -- general, with additional structures (Hamiltonian,
symplectic or quasicomplex), chaotic, quasiclassical, quantum, which are
considered in the framework of local (nonlinear) Fourier analysis, or wavelet
analysis. 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 this part we consider
spin orbital motion.
In section 3 we consider generalization of our approach
from part 1 to variational formulation in the biorthogonal bases of compactly
supported wavelets.
In section 4 we consider the different variational
multiresolution approach which gives us possibility
for computations in each scale separately.
\section{Spin-Orbital Motion}
Let us consider the system of equations for orbital motion
and Thomas-BMT equation for classical spin vector [9]:
$
\mathrm{d} q/\mathrm{d} t={\partial H_{orb}}/{\partial p}, \quad
{\mathrm{d} p}/{\mathrm{d} t}=-{\partial H_{orb}}/{\partial q}
$, $\quad\mathrm{d} s/\mathrm{d} t=w\times s$,
where
\begin{eqnarray}
H_{orb}&=&c\sqrt{\pi^2+m_0c^2}+e\Phi,\nonumber\\
w=&-&\frac{e}{m_0 c \gamma} (1+\gamma G)\vec B\\
&+&\frac{e}{m_0^3 c^3\gamma}\frac{G(\vec\pi\cdot\vec B)
\vec\pi}{(1+\gamma)}\nonumber\\
&+&\frac{e}{m_0^2 c^2\gamma}\frac{G +\gamma G+1}{(1+\gamma)}
[\pi\times E],\nonumber
\end{eqnarray}
$q=(q_1,q_2,q_3), p=(p_1,p_2,p_3)$ are canonical position and momentum,
$s=(s_1,s_2,s_3)$ is the classical spin vector of length $\hbar/2$,
$\pi=(\pi_1,\pi_2,\pi_3)$ is kinetic momentum vector.
We may introduce in 9-dimensional phase space $z=(q,p,s)$ the Poisson brackets
$
\{f(z),g(z)\}=f_qg_p-f_pg_q+[f_s\times g_s]\cdot s
$
and the Hamiltonian equations are
$
{\mathrm{d} z}/{\mathrm{d} t}=\{z,H\}
$
with Hamiltonian
\begin{equation}
H=H_{orb}(q,p,t)+w(q,p,t)\cdot s.
\end{equation}
More explicitly we have
\begin{eqnarray}
\frac{\mathrm{d} q}{\mathrm{d} t}&=&\frac{\partial H_{orb}}{\partial p}+\frac{\partial(w\cdot
s)}{\partial p}\nonumber\\
\frac{\mathrm{d} p}{\mathrm{d} t}&=&-\frac{\partial H_{orb}}{\partial q}-\frac{\partial(w\cdot
s)}{\partial q}\\
\frac{\mathrm{d} s}{\mathrm{d} t}&=&[w\times s]\nonumber
\end{eqnarray}
We will consider this dynamical system also in another paper
via invariant approach, based on consideration of Lie-Poison structures on
semidirect products.
But from the point of view which we used in this paper we may consider the
similar approximations as in the preceding parts and then we also arrive to
some type of polynomial dynamics.
\section{VARIATIONAL APPROACH IN BIORTHO\-GONAL WAVELET BASES}
Because integrand of variational functionals is represented
by bilinear form (scalar product) it seems more reasonable to
consider wavelet constructions [10] which take into account all advantages of
this structure.
The action functional for loops in the phase space is [11]
\begin{equation}
F(\gamma)=\displaystyle\int_\gamma pdq-\int_0^1H(t,\gamma(t))dt
\end{equation}
The critical points of $F$ are those loops $\gamma$, which solve
the Hamiltonian equations associated with the Hamiltonian $H$
and hence are periodic orbits. By the way, all critical points of $F$ are
the saddle points of infinite Morse index, but surprisingly this approach is
very effective. This will be demonstrated using several
variational techniques starting from minimax due to Rabinowitz
and ending with Floer homology. So, $(M,\omega)$ is symplectic
manifolds, $H: M \to R $, $H$ is Hamiltonian, $X_H$ is
unique Hamiltonian vector field defined by
$
\omega(X_H(x),\upsilon)=-dH(x)(\upsilon),\quad \upsilon\in T_xM,
\quad x\in M,
$
where $ \omega$ is the symplectic structure.
A T-periodic solution $x(t)$ of the Hamiltonian equations
$
\dot x=X_H(x)
$ on M
is a solution, satisfying the boundary conditions $x(T)$ $=x(0), T>0$.
Let us consider the loop space $\Omega=C^\infty(S^1, R^{2n})$,
where $S^1=R/{\bf Z}$, of smooth loops in $R^{2n}$.
Let us define a function $\Phi: \Omega\to R $ by setting
\begin{equation}
\Phi(x)=\displaystyle\int_0^1\frac{1}{2}<-J\dot x, x>dt-
\int_0^1 H(x(t))dt, \quad x\in\Omega
\end{equation}
The critical points of $\Phi$ are the periodic solutions of $\dot x=X_H(x)$.
Computing the derivative at $x\in\Omega$ in the direction of $y\in\Omega$,
we find
\begin{eqnarray}
&&\Phi'(x)(y)=\frac{d}{d\epsilon}\Phi(x+\epsilon y)\vert_{\epsilon=0}
=\\
&&\displaystyle\int_0^1<-J\dot x-\bigtriangledown H(x),y>dt\nonumber
\end{eqnarray}
Consequently, $\Phi'(x)(y)=0$ for all $y\in\Omega$ iff the loop $x$ satisfies
the equation
\begin{equation}
-J\dot x(t)-\bigtriangledown H(x(t))=0,
\end{equation}
i.e. $x(t)$ is a solution of the Hamiltonian equations, which also satisfies
$x(0)=x(1)$, i.e. periodic of period 1. Periodic loops may be represented by
their Fourier series:
$x(t)=\sum e^{k2\pi Jt}x_k, \ x_k\in R^{2k}$,
where $J$ is quasicomplex structure. We give relations between
quasicomplex structure and wavelets in our other paper.
But now we
need to take into account underlying bilinear structure via wavelets.
We started with two hierarchical sequences of approximations spaces [10]:
\begin{eqnarray}
&&\dots V_{-2}\subset V_{-1}\subset V_{0}\subset V_{1}\subset V_{2}\dots,\\
&&\dots \widetilde{V}_{-2}\subset\widetilde{V}_{-1}\subset
\widetilde{V}_{0}\subset\widetilde{V}_{1}\subset\widetilde{V}_{2}\dots,\nonumber
\end{eqnarray}
and as usually,
$W_0$ is complement to $V_0$ in $V_1$, but now not necessarily orthogonal
complement.
New orthogonality conditions have now the following form:
\begin{equation}
\widetilde {W}_{0}\perp V_0,\quad W_{0}\perp\widetilde{V}_{0},\quad
V_j\perp\widetilde{W}_j, \quad \widetilde{V}_j\perp W_j
\end{equation}
translates of $\psi$ $\mathrm{span}$ $ W_0$,
translates of $\tilde\psi \quad \mathrm{span} \quad\widetilde{W}_0$.
Biorthogonality conditions are
\begin{equation}
<\psi_{jk},\tilde{\psi}_{j'k'}>=
\int^\infty_{-\infty}\psi_{jk}(x)\tilde\psi_{j'k'}(x)\mathrm{d} x=
\delta_{kk'}\delta_{jj'},
\end{equation}
where
$\psi_{jk}(x)=2^{j/2}\psi(2^jx-k)$.
Functions $\varphi(x), \tilde\varphi(x-k)$ form dual pair:
$
<\varphi(x-k),\tilde\varphi(x-\ell)>=\delta_{kl}$,
$<\varphi(x-k),\tilde\psi(x-\ell)>=0$.
Functions $\varphi, \tilde\varphi$ generate a multiresolution analysis.
$\varphi(x-k)$, $\psi(x-k)$ are synthesis functions,
$\tilde\varphi(x-\ell)$, $\tilde\psi(x-\ell)$ are analysis functions.
Synthesis functions are biorthogonal to analysis functions. Scaling spaces
are orthogonal to dual wavelet spaces.
Two multiresolutions are intertwining
$
V_j+W_j=V_{j+1}, \quad \widetilde V_j+ \widetilde W_j = \widetilde V_{j+1}
$.
These are direct sums but not orthogonal sums.
So, our representation for solution has now the form
\begin{equation}
f(t)=\sum_{j,k}\tilde b_{jk}\psi_{jk}(t),
\end{equation}
where synthesis wavelets are used to synthesize the function. But
$\tilde b_{jk}$ come from inner products with analysis wavelets.
Biorthogonality yields
\begin{equation}
\tilde b_{\ell m}=\int f(t)\tilde{\psi}_{\ell m}(t) \mathrm{d} t.
\end{equation}
So, now we can introduce this more complicated construction into
our variational approach. We have modification only on the level of
computing coefficients of reduced nonlinear algebraical system.
This new construction is more flexible.
Biorthogonal point of view is more stable under the action of large
class of operators while orthogonal (one scale for multiresolution)
is fragile, all computations are much more simpler and we accelerate
the rate of convergence. In all types of Hamiltonian calculation,
which are based on some bilinear structures (symplectic or
Poissonian structures, bilinear form of integrand in variational
integral) this framework leads to greater success.
In particular cases we may use very useful wavelet packets from Fig.~1.
\begin{figure}[ht]
\centering
\epsfig{file=tha137a.eps, width=82.5mm, bb= 0 200 599 600, clip}
\caption{Wavelet packets.}
\end{figure}
\section{Evaluation of Nonlinearities Scale by Scale.Non-regular approximation.}
We use wavelet function $\psi(x)
$, which has $k$ vanishing moments
$
\int x^k \psi(x)\mathrm{d} x=0$, or equivalently
$x^k=\sum c_\ell\varphi_\ell(x)$ for each $k$,
$0\leq k\leq K$.
Let $P_j$ be orthogonal projector on space $V_j$. By tree algorithm we
have for any $u\in L^2({\bf R})$ and $\ell\in{\bf Z}$, that the wavelet
coefficients of $P_\ell(u)$, i.e. the set $\{<u,\psi_{j,k}>, j\leq\ell-1,
k\in{\bf Z}\}$ can be compute using hierarchic algorithms from the set of
scaling coefficients in $V_\ell$, i.e. the set $\{<u,\varphi_{\ell,k}>,
k\in{\bf Z}\}$ [12]. Because for scaling function $\varphi$ we have in general
only $\int\varphi(x)\mathrm{d} x=1$, therefore we have for any function $u\in L^2({\bf
R})$:
\begin{equation}
\lim_{j\to\infty, k2^{-j}\to x} \mid 2^{j/2}<u,\varphi_{j,k}>-u(x)\mid=0
\end{equation}
If the integer $n(\varphi)$ is the largest one such that
\begin{equation}
\int x^\alpha \varphi(x)\mathrm{d} x=0 \qquad \mbox{for}\qquad 1\leq\alpha\leq n
\end{equation}
then if $u\in C^{(n+1)}$ with $u^{(n+1)}$ bounded we have for $j\to\infty$
uniformly in k:
\begin{equation}
\mid 2^{j/2}<u,\varphi_{j,k}>-u(k2^{-j})\mid=O(2^{-j(n+1)}).
\end{equation}
Such scaling functions with zero moments are very useful for us from the point
of view of time-frequency localization, because we have for Fourier component
$\hat\Phi(\omega)$ of them, that exists some $C(\varphi)\in{\bf R}$, such
that for $\omega\to0$ $\quad\hat\Phi(\omega)=1+C(\varphi)$ $\mid\omega\mid^{2r+2}$
(remember, that we consider r-regular multiresolution analysis).
Using such type of scaling functions lead to superconvergence properties for
general Galerkin approximation [12].
Now we need some estimates in each scale for non-linear terms of type $u\mapsto
f(u)=f\circ u$, where f is $C^\infty$ (in previous and future parts we consider
only truncated Taylor series action). Let us consider non regular space of
approximation $\widetilde V$ of the form
\begin{equation}\label{eq:til1}
\widetilde V=V_q\oplus\sum_{q\leq j\leq p-1} \widetilde{W_j},
\end{equation}
with $\widetilde{W_j}\subset W_j$. We need efficient and precise estimate of
$f\circ u$ on $\widetilde V$. Let us set for $q\in{\bf Z}$ and $u\in L^2({\bf
R})$
\begin{equation}
\prod f_q (u)=2^{-q/2}\sum_{k\in{\bf
Z}}f(2^{q/2}<u,\varphi_{q,k}>)\cdot\varphi_{q,k}
\end{equation}
We have the following important for us estimation (uniformly in q) for $u,
f(u)\in H^{(n+1)}$ [12]:
\begin{equation}\label{eq:til2}
\|P_q\left(f(u)\right)-\prod f_q(u)\|_{L^2}=O\left(2^{-(n+1)q}\right)
\end{equation}
For non regular spaces (\ref{eq:til1}) we set
\begin{equation}
\prod f_{\widetilde V}(u)=\prod f_q(u)+\sum_{\ell=q,p-1} P_{\widetilde {W_j}}
\prod f_{\ell+1}(u)
\end{equation}
Then we have the following estimate:
\begin{equation}
\Vert P_{\widetilde V}\left(f(u)\right)-\prod f_{\widetilde V}(u)\Vert_{L^2}=O(2^{-(n+1)q})
\end{equation}
uniformly in q and $\widetilde V$ (\ref{eq:til1}).
This estimate depends on q, not p, i.e. on the scale of the coarse grid, not on
the finest grid used in definition of $\widetilde V$. We have for total error
\begin{eqnarray}
&&\Vert f(u)-\prod f_{\widetilde V}(u)\Vert=
\Vert f(u)-P_{\widetilde V}(f(u))\Vert_{L^2}\nonumber\\
&&+\Vert P_{\widetilde V}(f(u)-\prod f_{\widetilde V}(u))\Vert_{L^2}
\end{eqnarray}
and since the projection error in $\widetilde V$:
$
\Vert f(u)-P_{\bar{V}}\left(f(u)\right)\Vert_{L^2}
$
is much smaller than the projection error in
$V_q$ we have the improvement (20) of (\ref{eq:til2}).
In concrete calculations and estimates it is very useful to consider
approximations in the particular case of
c-structured space:
\begin{eqnarray}
\widetilde{V}=&&V_q+\sum^{p-1}_{j=q}span\{\psi_{j,k},\\
&& k\in[2^{(j-1)}-c,
2^{(j-1)}+c]
\quad\mbox{\rm mod}\quad 2^j\}\nonumber
\end{eqnarray}
We are very grateful to M.~Cornacchia (SLAC), W.~Herrmannsfeldt (SLAC),
Mrs. J.~Kono (LBL) and
M.~Laraneta (UCLA) for their permanent encouragement.
|
\section{Introduction}
One of the most exciting proposals of the last years in
string theory,
consists of the
possibility that the five distinct superstring theories in ten
dimensions plus
supergravity in eleven dimensions
be different vacua in the moduli space of a single underlying
eleven--dimensional
theory, the so--called M--theory \cite{Schwarz}.
In this respect,
Ho\v{r}ava and Witten proposed that
the strong--coupling limit of $E_8\times E_8$ heterotic
string theory can be obtained from M--theory. They used the low--energy
limit of M--theory,
eleven-dimensional
supergravity, on a manifold with boundary (a $S^1/Z_2$ orbifold),
with the
$E_8$ gauge multiplets at each of the
two ten--dimensional boundaries (the orbifold fixed planes)
\cite{Horava-Witten}.
Some phenomenological implications of the strong--coupling limit
of $E_8\times E_8$ heterotic string theory
have been studied by compactifying the eleven--dimensional M--theory
on a Calabi--Yau manifold times
the eleventh segment (orbifold) \cite{Witten}.
The resulting four--dimensional effective theory can reconcile
the observed Planck scale $M_{Planck}= 1.2 \times 10^{19}$ GeV
with the phenomenologically favored GUT scale
$M_{GUT}\approx 3\times 10^{16}$
GeV in a natural manner, providing an attractive framework
for the unification of couplings \cite{Witten,Banks-Dine}.
An additional phenomenological virtue of the M--theory limit
is that there can be a QCD axion whose high energy axion potential is
suppressed enough so that the strong CP problem can be solved
by the axion mechanism \cite{Banks-Dine,Choi}.
About the issue of supersymmetry
breaking, the possibility of generating it by the
gaugino condensation on the hidden boundary has been studied
\cite{Horava,Nilles-Olechowski-Yamaguchi,Lalak-Thomas,Lukas-Ovrut-Waldram2,Quiros,Choi2}
and also some interesting features of the
resulting soft supersymmetry--breaking terms were discussed.
In particular, gaugino masses turn
out to be of the same order as squark
masses \cite{Nilles-Olechowski-Yamaguchi} unlike the
weakly--coupled heterotic string case where gaugino masses are much smaller
than squark masses \cite{Beatriz}.
The analysis of the soft supersymmetry--breaking terms under the more general
assumption that supersymmetry is spontaneously broken by
the auxiliary components of the bulk moduli superfields in the model
(dilaton $S$ and modulus $T$)
was carried out in \cite{Mio,Lukas-Ovrut-Waldram2}.
It was examined in particular how the soft terms
vary when one moves from the weakly--coupled heterotic string
limit to the strongly--coupled limit. The conclusion being
that there can be a sizable difference between both limits.
As a consequence, the study
of the low--energy
($\approx M_W$) sparticle spectra \cite{Mio,Lopez,Bailin,Kawamura,Savoy,Alejandro,Chi}
gives also rise to qualitative differences.
However, all the above mentioned analyses of the
phenomenology of $N=1$ heterotic
M--theory vacua, were carried out only in the context of
the standard embedding of the spin connection into one of the
$E_8$ gauge groups. Although in the case of the weakly--coupled
heterotic string is simple to work with the standard
embedding \cite{Candelas} and more involved the analyses of non--standard
embedding
vacua \cite{Distler,Kachru}, in the strongly--coupled case, as
emphasized in \cite{five-branes2}, the standard embedding is not
particularly special.
Thus the analysis of the non--standard embedding
is very interesting since more general gauge
groups and matter fields may be present.
Recently, this analysis has been considered
in M--theory compactified on Calabi--Yau
\cite{Benakli,Lalak,five-branes,Benakli2}
and orbifold \cite{Stieberger} spaces and several issues,
as the four--dimensional effective action and scales, studied.
Concerning the latter
it was pointed out in the past, in the context of perturbative
strings,
that the size of the compactification scale might be of order $1$ TeV
without any obvious contradiction with experimental facts
\cite{Antoniadis} (for a different point of view, see \cite{Caceres}).
Recently, going away from perturbative vacua, it was
realized that the
string scale may be anywhere between the weak scale and the Planck
scale \cite{Lykken}.
This is the case of the type I string where several interesting
low string scale scenarios were proposed\footnote{
To trust them would imply to assume that
Nature is trying to mislead us with an apparent gauge coupling
unification at the scale $M_{GUT}\approx 3\times 10^{16}$ GeV.
In this sense, a reasonable doubt about those scenarios is
healthy.}: a $1$ TeV scale scenario,
where the size of the extra dimensions
may even be as large as a millimetre \cite{Dimopoulos}
and an intermediate scale ($\approx 10^{11}$ GeV)
scenario where some phenomenological issues \cite{Benakli2} and the
hierarchy $M_W/M_{Planck}\approx 10^{-16}$ \cite{Nosotros} can be explained.
A $1$ TeV string scale scenario, with the scale of extra dimensions
not smaller than $1$ TeV
may also arise in the context of type II strings \cite{Antoniadis2}.
Whether or not all these scenarios are possible in the context of M--theory
was analyzed recently in \cite{Benakli2} with interesting results.
On the other hand, more general vacua, still preserving $N=1$ supersymmetry,
may appear when $M5$--branes are included in the
computation \cite{Witten}.
These five--branes are non--perturbative objects,
located at points throughout the orbifold interval.
They have $3+1$ uncompactified dimensions
in order to preserve Lorenz invariance and $2$ compactified dimensions.
The appearance of anomalous $U(1)$ symmetries related to their
presence
was studied in \cite{Binetruy}.
Modifications to the four--dimensional effective action
were discussed in the context of orbifold
compactifications of heterotic M--theory \cite{Stieberger}
and investigated in great detail in Calabi--Yau compactifications
by
Lukas, Ovrut and Waldram \cite{five-branes,five-branes2}. The latter also
considered the
issue of supersymmetry breaking by gaugino condensation and soft terms.
It is worth remarking that in the presence of five--branes,
model building turns out to be extremely interesting.
In particular, to obtain three generation
models with realistic gauge groups, as for example $SU(5)$, is not
specially difficult \cite{Donagi}.
An important question in the study of string phenomenology is whether or
not the soft terms, and in particular the scalar masses, are universal
\cite{Jerusalen}. In this respect, let us recall the
situation
in the case of Calabi--Yau compactifications of
the weakly--coupled heterotic string. There the soft terms are in
general
non--universal due to the presence of off--diagonal matter K\"ahler
metrics
induced by the existence of different moduli ($T_i$) \cite{Hangbae}.
This can be avoided assuming that supersymmetry is broken in the
dilaton field ($S$) direction
\cite{Kaplunovsky-Louis,Brignole-Ibanez-Munoz2},
although {\it small} non--universality may arise due to string--loop
corrections \cite{Nir}\footnote{It is worth noticing
that supergravity--loop corrections may also
induce non-universality \cite{Sik}.}.
Concerning supersymmetry breaking of heterotic M--theory in a general
direction,
the situation
is similar to the weakly--coupled case: the existence in general
of different moduli, $T_i$, will give rise to non--universality. Moreover,
supersymmetry breaking in the dilaton direction will induce now
{\it large} non--universal soft scalar masses due to the presence of
$S$ together with $T_i$ in the matter field K\"ahler metrics \cite{Mio}.
In \cite{five-branes2} an improvement
to the problem of
non--universality in heterotic M--theory was
proposed: model building with Calabi--Yau spaces with only one K\"ahler
modulus $T$. Such spaces exist and, as mentioned above,
in the presence of non--standard
embedding and five--branes, to construct three--generation models
might be relatively easy.
In this paper we will
assume that the
standard model arises from the heterotic M--theory
compactified on a Calabi--Yau
manifold with only one field $T$, and then we will
study the
associated phenomenology.
In particular, we
will analyze first in detail the different scales of the theory,
eleven--dimensional
Planck mass, compactification scale, orbifold scale, and how they are
related taking into account higher order corrections to the
formulae. In this respect, we will study whether or not large internal
dimensions are natural in the context of the heterotic M--theory.
We will also perform a systematic analysis of the soft
supersymmetry--breaking terms, under the general assumption that
supersymmetry is spontaneously broken by the bulk moduli fields.
In section 2 we will concentrate on
standard and non--standard embedding without the presence of
five--branes. First, we will summarize results about the standard
embedding
case concerning the four--dimensional effective action, which will be
very
useful for the detailed analysis of soft terms and scales of the
theory.
We will also
compute the value of the $B$ parameter.
Then we will turn our thoughts to the study of the non--standard
embedding.
Basically, the same formulae than in the standard embedding case can be
used but with the value of one of the parameters which appears in
the formulae in a different
range.
This will give rise to
different possibilities for the scales of the theory and we will see
that to lower these scales is in principle possible in some
special limits. However, the necessity of a fine--tuning
or the existence of a hierarchy problem renders this
possibility unnatural. Likewise,
a different pattern of soft terms arises. For example,
the possibility of scalar masses larger than gaugino masses,
something which is forbidden in the standard embedding case and
if possible,
difficult to obtain, in the weakly--coupled heterotic string, is
now allowed.
In section 3 non--perturbative objects as five--branes are included in
the vacuum. Thus the whole analysis is modified. New parameters
contribute to the soft terms and their study is more involved. However,
an interesting pattern of soft terms arises. Again, scalars
heavier
than gauginos can be obtained but now more easily than in the non--standard
embedding case.
Concerning the scales, $M_{GUT}$ can easily be obtained. Extra
possibilities
to lower this scale arise but again they are unnatural.
Finally we leave the conclusions for section 4.
\section{Standard and non--standard embedding without five--branes}
\subsection{Four--dimensional effective action and scales}
Following Witten's investigation \cite{Witten}, the solution of the equations
of motion, preserving N=1 supersymmetry, of eleven-dimensional M--theory
\cite{Horava-Witten} compactified on
\begin{equation}
M_4\times S^1/Z_2\times X\ ,
\label{space}
\end{equation}
where $X$ is a six--dimensional
Calabi-Yau manifold and $M_4$ is four--dimensional
Minkowski space, can be analyzed by expanding it in powers of the
dimensionless parameter \cite{Banks-Dine}
\begin{equation}
\epsilon_1 = \frac{\pi\rho}{M_{11}^{3} V^{2/3}}\ ,
\label{epsilon1}
\end{equation}
where $M_{11}$ denotes the eleven-dimensional Planck mass, $V$ is the
Calabi-Yau volume and $\pi \rho$ denotes the length of the eleventh
segment.
To zeroth order in this expansion, analogously to the case of
the
weakly--coupled heterotic string theory, two model--independent
bulk moduli superfields $S$ and $T$ arise in the four--dimensional
effective supergravity of the compactified M--theory. Their scalar
components can be identified as
\begin{eqnarray}
S+\bar S &=&\frac{1}{\pi (4\pi)^{2/3}} M_{11}^6 V\ ,
\nonumber \\
T+\bar T &=& \frac{6^{1/3}}{(4\pi)^{4/3}}
M_{11}^{3} V^{1/3} \pi\rho\ .
\label{dilaton-modulus}
\end{eqnarray}
Notice that with these definitions the expansion parameter
(\ref{epsilon1}) can be written as
\begin{equation}
\epsilon_1 \approx
\frac{T+\bar T}{S+\bar S}\ .
\label{epsilon1bis}
\end{equation}
To higher orders, there appear gauge and matter superfields associated
to the observable and hidden sector gauge groups,
$G_O\times G_H \subset E_8\times E_8$,
where $G_O$($G_H$) is located at the boundary
$x^{11}=0$($x^{11}=\pi\rho$) with $x^{11}$ denoting the orbifold
coordinate.~From now on, we will use as our
notation the subscript $O$($H$) for quantities and functions of the
observable(hidden) sector. On the other hand, the internal space
becomes
deformed and is no longer as in (\ref{space}), a simple product of
$S^1/Z_2\times X$.
The effective supergravity obtained from this
M--theory
compactification was computed to the leading order
in \cite{Banks-Dine,Li-Lopez-Nanopoulos1,Dudas-Grojean,Nilles-Olechowski-Yamaguchi}. The order $\epsilon_1$ correction to the leading order
gauge
kinetic functions and K\"ahler potential was also computed in
\cite{Banks-Dine,Nilles-Stieberger,Choi,Nilles-Olechowski-Yamaguchi}
and
\cite{Lukas-Ovrut-Waldram} respectively. The final result
is specified by the following K\"ahler potential, $K$,
gauge kinetic functions, $f_O$, $f_H$, and superpotential $W_O$:
\begin{eqnarray}
K &=& -\ln (S+\bar{S}) -3\ln (T+\bar{T})
+\frac{3}{T+\bar{T}}
\left(1+\frac{1}{3}\epsilon_O\right) C_O^p \bar{C}_O^p
\nonumber\\ &&
+\frac{3}{T+\bar{T}}
\left(1+\frac{1}{3}\epsilon_H\right) C_H^p \bar{C}_H^p \ ,
\label{kahler}
\\
f_{O} &=& S+\beta_O T\ , \quad f_{H}=S+\beta_H T\ ,
\label{kinetic}
\\
W_O &=& d_{pqr}C_O^pC_O^qC_O^r\ ,
\label{superpotential}
\end{eqnarray}
with
\begin{equation}
\epsilon_O=\beta_O
\frac{T+\bar T}{S+\bar S}\ ,
\quad \epsilon_H=\beta_H
\frac{T+\bar T}{S+\bar S}\ .
\label{epsilonO}
\end{equation}
$d_{pqr}$ are constant coefficients, $C_O^p(C_H^p)$ are the
observable(hidden)
matter fields and the model--dependent integer coefficients
$\beta_O={1\over 8\pi^2}\int\omega\wedge[{\rm tr}
(F_O\wedge F_O)-\frac{1}{2}{\rm tr}(R\wedge R)]$,
$\beta_H={1\over 8\pi^2}\int\omega\wedge[{\rm tr}
(F_H\wedge F_H)-\frac{1}{2}{\rm tr}(R\wedge R)]$,
for the K\"ahler form $\omega$ normalized as
the generator of the integer (1,1)
cohomology\footnote{
Usually $\beta$ is considered to be an arbitrary real number.
For $T$ normalized as (\protect\ref{dilaton-modulus}),
it is required to be an integer \cite{Choi}.}.
Taking into account that the real parts of the gauge kinetic functions
in (\ref{kinetic}) multiplied by $4\pi$ are the inverse gauge coupling
constants $\alpha_O$ and $\alpha_H$, using (\ref{dilaton-modulus})
one can write \cite{Witten,Li-Lopez-Nanopoulos1,Benakli}
\begin{eqnarray}
\alpha_O &=& \frac{1}{2\pi(S+\bar S)(1+\epsilon_O)}
\label{alphaO}
\\
&=& \frac{(4\pi)^{2/3}}{2 M_{11}^{6} V_O}\ ,
\label{alphaO'}
\end{eqnarray}
with
\begin{equation}
V_O=V(1+\epsilon_O)\ ,
\label{volumeO}
\end{equation}
the observable--sector volume, and
\begin{eqnarray}
\alpha_H &=& \frac{1}{2\pi(S+\bar S)(1+\epsilon_H)}
\label{alphaH}
\\
&=& \frac{(4\pi)^{2/3}}{2 M_{11}^{6} V_H}\ ,
\label{alphaH'}
\end{eqnarray}
with
\begin{equation}
V_H=V(1+\epsilon_H)\ ,
\label{volumeH}
\end{equation}
the hidden--sector volume.
On the other hand, using (\ref{volumeO}),
the M-theory expression of the four--dimensional Planck scale
\begin{eqnarray}
M_{Planck}^2 &=& 16\pi^2\rho M_{11}^9 <V>\ ,
\label{Planck}
\end{eqnarray}
where $<V>$ is the average volume of the Calabi--Yau space
\begin{eqnarray}
<V> &=& \frac{V_O+V_H}{2}
\label{averagevolume}
\end{eqnarray}
and (\ref{dilaton-modulus}) one also finds
\begin{eqnarray}
V_O^{-1/6} &=& \left(\frac{V}{<V>}\right)^{1/2}
\left(\frac{6^{1/3}M_{Planck}^2}{2048 \pi^4}\right)^{1/2}
\left(\frac{4}{S+\bar S}\right)^{1/2}
\left(\frac{2}{T+ \bar T}\right)^{1/2}
\left(\frac{1}{1+\epsilon_O}\right)^{1/6}
\nonumber\\
&=&
\left(\frac{V}{<V>}\right)^{1/2}
3.6\times 10^{16}
\left(\frac{4}{S+\bar S}\right)^{1/2}
\left(\frac{2}{T+ \bar T}\right)^{1/2}
\left(\frac{1}{1+\epsilon_O}\right)^{1/6}
{\rm GeV}
\ ,
\nonumber\\
\label{gut}
\end{eqnarray}
which is a very useful formula (also in the context of five--branes)
as we will see below in order to
discuss whether or not the GUT scale or smaller scales
are obtained in a natural way. In this respect,
let us now obtain the connection between the different scales of the
theory:
the eleven--dimensional Planck mass, $M_{11}$, the
Calabi--Yau compactification scale, $V_O^{-1/6}$,
and the orbifold scale, $(\pi \rho)^{-1}$.
It is straightforward to obtain from (\ref{alphaO'}) the following
relation:
\begin{eqnarray}
\frac{M_{11}}{V_O^{-1/6}} &=& \left(2(4\pi)^{-2/3}\alpha_O\right)^{-1/6}\ .
\label{relation}
\end{eqnarray}
Likewise, using (\ref{Planck}) and (\ref{alphaO'})
we arrive at
\begin{eqnarray}
\frac{V_O^{-1/6}}{(\pi \rho)^{-1}} &=& \left(\frac{V}{<V>}\right)
\left(\frac{M_{Planck}}{(2048\pi^4)^{1/2}V_O^{-1/6}}\right)^2
\left(8192\pi^4\alpha_O^3\right)^{1/2}(1+\epsilon_O)
\nonumber\\
&=&
\left(\frac{V}{<V>}\right)
\left(\frac{2.7\times 10^{16}{\rm GeV}}{V_O^{-1/6}}\right)^2
\left(8192\pi^4\alpha_O^3\right)^{1/2}(1+\epsilon_O)
\ .
\label{relation2}
\end{eqnarray}
\subsection{Standard embedding}
Here we will summarize results found in the literature about the
standard embedding case, including the computation of the
soft terms under the general assumption of dilaton/modulus
supersymmetry
breaking,
and we will discuss in detail the issue of the scales
in
the theory.
Let us recall first that in this case, where the spin connection of the
Calabi--Yau space is embedded in the gauge connection of one of the
$E_8$ groups, the following constraint must be
fulfilled:
\begin{eqnarray}
\beta_O + \beta_H &=& 0\ ,
\label{constraint}
\end{eqnarray}
with\footnote{Non--standard embedding cases may also
fulfil this condition. Since their effective action, as we will
discuss below, is the same as
in the standard embedding the results of this subsection
can be applied to any model obtained from standard or non--standard embedding
with $\beta_O>0$. We leave the study of the case $\beta_O=0$ for the
next subsection.}
\begin{eqnarray}
\beta_O > 0\ .
\label{mayorcero}
\end{eqnarray}
Thus $\epsilon_O=-\epsilon_H>0$
implying that the observable--sector volume $V_O$ given by
(\ref{volumeO}) is larger than the hidden--sector volume $V_H$
given by (\ref{volumeH}) and therefore the gauge coupling of the observable
sector (\ref{alphaO'})
will be weaker than the gauge coupling of the hidden sector (\ref{alphaH'}).
Besides, since $V_H$ must be a positive quantity, one has to impose the
bound $\epsilon_O<1$. Altogether one gets
\begin{equation}
0 < \epsilon_O
<
1\ .
\label{epsilonbound}
\end{equation}
Notice that using (\ref{alphaO}) one can write
$\epsilon_O$ as
\begin{eqnarray}
\epsilon_O
=
\frac{4-(S+\bar S)}
{(S+\bar S)}\ ,
\label{nueva}
\end{eqnarray}
where we have already assumed that
the gauge group of the observable sector
$G_O$ is the one of the standard model or some unification gauge group
as $SU(5)$, $SO(10)$ or $E_6$, i.e. we are using
$\left(2\pi\alpha_O\right)^{-1}=4$ in
order to reproduce the LEP data about
$\alpha_{GUT}$ ($\alpha_O$ in our notation).
For a further study of scales and soft terms and comparison with
vacua in the presence of five--branes, we show $\epsilon_O$ versus $S+\bar S$
in Fig.~\ref{epsilon}.
\begin{figure}[htb]
\begin{center}
\epsfig{file= regiones.eps, width=8cm, height=8cm}
\end{center}
\vspace{-1.5cm}
\caption{$\epsilon_O$ versus $S+\bar S$.\label{epsilon}}
\end{figure}
The region between the lower bound $\epsilon_O=-1$ and $\epsilon_O=0$
corresponds to the non--standard embedding case and will be discussed
in subsection 2.3. The region with $\epsilon_O>1$ will be discussed
in the context of five--branes in section 3.
From (\ref{epsilonbound}), (\ref{nueva}) and (\ref{epsilonO})
one obtains that the dilaton and moduli fields are bounded (see also
Fig.~\ref{epsilon})
\begin{equation}
0 < \beta_O(T+ \bar T)
<
2\ , \quad
2
<
(S+ \bar S)
<
4\ .
\label{bounds}
\end{equation}
Finally, the average volume of the
Calabi--Yau space (\ref{averagevolume}) turns out to be equal to the
lowest order value
\begin{eqnarray}
<V> &=& V\ ,
\label{constraint2}
\end{eqnarray}
and (\ref{gut}) simplifies as
\begin{eqnarray}
V_O^{-1/6}
&=&
3.6\times 10^{16}
\left(\frac{4}{S+\bar S}\right)^{1/2}
\left(\frac{2}{T+ \bar T}\right)^{1/2}
\left(\frac{1}{1+\epsilon_O}\right)^{1/6}
{\rm GeV}\ ,
\label{gut2}
\end{eqnarray}
whereas (\ref{relation2})
becomes
\begin{equation}
\frac{V_O^{-1/6}}{(\pi \rho)^{-1}} =
7\ \left(\frac{2.7\times 10^{16}{\rm GeV}}{V_O^{-1/6}}\right)^2
(1+\epsilon_O)
\ ,
\label{relation3}
\end{equation}
where $\left(2\pi\alpha_O\right)^{-1}=4$ has been used.
This also allows us to write (\ref{relation}) as
\begin{eqnarray}
\frac{M_{11}}{V_O^{-1/6}} &=& 2 \ .
\label{relationn}
\end{eqnarray}
With all these results we can start now the study of scales and soft
terms in the theory.
\subsubsection{Scales}
The four--dimensional effective theory from heterotic M--theory can
reconcile the observed $M_{Planck}=1.2\times 10^{19}$ GeV with
the phenomenologically favored GUT scale $M_{GUT}\approx 3\times 10^{16}$
GeV in a natural manner \cite{Witten,Banks-Dine}.
This is to be compared to the weakly--coupled heterotic string where
$M_{string}=\left(\frac{\alpha_{GUT}}{8}\right)^{1/2} M_{Planck}
\approx 8.5\times 10^{17}$ GeV.
We will revisit this issue in detail taking into account the
higher order corrections studied above to the zeroth order formulae.
Identifying $M_{GUT}\approx 3\times 10^{16}$ with $V_O^{-1/6}$
one obtains from (\ref{relation3}), with $\epsilon_O$
constrained by (\ref{epsilonbound}), and (\ref{relationn}):
$M_{11}\approx 6\times 10^{16}$ GeV and
$(\pi\rho)^{-1}\approx (2.5-5.3)\times 10^{15}$ GeV, i.e.
the following pattern $(\pi\rho)^{-1}<V_O^{-1/6}<M_{11}$.
On the other hand, to obtain
$V_O^{-1/6} \approx 3\times 10^{16}$ GeV when $\beta_O>0$
is quite natural. This can be seen from
(\ref{gut2}) since (\ref{bounds}) implies that $T+\bar T$ and
$S+\bar S$ are essentially of order one.
Let us discuss this point in more detail.
Using (\ref{epsilonO}) and (\ref{nueva})
it is interesting to write (\ref{gut2}) as
\begin{eqnarray}
V_O^{-1/6}
&=&
3.6\times 10^{16}
\left(\frac{\beta_O}{2 \epsilon_O}\right)^{1/2}
\left(1+\epsilon_O\right)^{5/6}
{\rm GeV}\ .
\label{gut3}
\end{eqnarray}
This is shown in Fig.~\ref{scales1} where
$V_O^{-1/6}$ versus $\epsilon_O$ is plotted.
\begin{figure}[htb]
\begin{center}
\epsfig{file= scales1.eps, width=8cm, height=8cm}
\end{center}
\vspace{-1.0cm}
\caption{$\log M_{11}$, $\log V_O^{-1/6}$ and $\log (\pi\rho)^{-1}$
versus $\epsilon_O$
in the cases $\beta_O=1$ (for $0<\epsilon_O<1$)
and $\beta_O=-1$ (for $-1<\epsilon_O<0$). The straight line
indicates the phenomenologically favored GUT scale,
$M_{GUT} = 3\times 10^{16}$ GeV.\label{scales1}}
\end{figure}
The right hand side of the figure ($0<\epsilon_O<1$)
corresponds to the case $\beta_O>0$ whereas the left hand side
($-1<\epsilon_O<0$) corresponds to the case $\beta_O<0$, i.e.
the non--standard embedding situation that will be analyzed in the
next subsection.
For the moment we will concentrate on the case $\beta_O>0$ and,
in particular, in Fig.~\ref{scales1} we are showing an example with
$\beta_O=1$.
$(\pi\rho)^{-1}$ and
$M_{11}$ are also plotted in the figure using (\ref{relation3}) and
(\ref{relationn})
respectively. Most values of $\epsilon_O$ imply
$V_O^{-1/6} \approx 5\times 10^{16}$ GeV which is quite close
to the phenomenologically favored value. For example,
for $\epsilon_O=1/4$, which corresponds to $S+\bar S=16/5$ and
$T+\bar T=4/5$, we obtain
$V_O^{-1/6} = 6.1\times 10^{16}$ GeV
and for the limit\footnote{One may worry that the M--theory expansion would
not work in these cases where $\epsilon_O$ is of order one since
$\epsilon_1$ in (\ref{epsilon1bis}) is of order one also. However,
as argued in \cite{Mio} any correction which is $n$-th order
in $\epsilon_1$ accompanies at least $(n-1)$-powers of
$\epsilon_2=1/M_{11}^3 \pi\rho V^{1/3}\approx 1/2\pi^2(T+\bar T)$,
the generalization of the
string world--sheet coupling to the membrane world--volume coupling,
and thus is suppressed by
$\left(\epsilon_1\epsilon_2\right)^{n-1}\approx
\left(\alpha_O/\pi\right)^{n-1}$. This allows the M--theory expansion
to be valid even when $\epsilon_1$ becomes of order one.
This has been explicitly checked in \cite{Wyllard}.}
$\epsilon_O=1$,
which corresponds to $S+\bar S=T+\bar T=2$, we obtain the lowest
possible
value
$V_O^{-1/6} = 4.5\times 10^{16}$.
These qualitative results can only be modified in the limit
$\epsilon_O\rightarrow 0$, i.e. $(T+\bar T)\rightarrow 0$,
since then $V_O^{-1/6}\rightarrow \infty$. Notice that in this case
$(\pi\rho)^{-1}>V_O^{-1/6}$ (see Fig.~\ref{scales1}).
This limit is not interesting not only because $V_O^{-1/6}$ is too
large but also because we are effectively in the weakly coupled
region with a very small orbifold radius.
The results for $\beta_O\neq 1$ can easily be deduced from
the figure since
$V_O^{-1/6}(\beta_O\neq 1)=\beta_O^{1/2} V_O^{-1/6}(\beta_O=1)$,
$M_{11}(\beta_O\neq 1)=2 V_O^{-1/6}(\beta_O\neq 1)$
and
$(\pi\rho)^{-1}(\beta_O\neq 1)=\beta_O^{3/2}(\pi\rho)^{-1}(\beta_O=1)$.
Notice that for models with those values of $\beta_O$ we are in the limit of
validity if we want to obtain
$V_O^{-1/6} \approx 3\times 10^{16}$ GeV.
For example, For $\epsilon_O=1$ with $\beta_O=4$,
$V_O^{-1/6} = 9\times 10^{16}$ GeV.
Let us finally remark that, from the above discussion, it is
straightforward to deduce that large internal dimensions,
associated with the radius of the Calabi--Yau and/or the radius of
the orbifold, are not allowed.
\subsubsection{Soft terms}
Since the soft supersymmetry--breaking terms
depend on $\epsilon_O$ as we will see below, result
(\ref{epsilonbound}) simplifies their
analysis. Applying the standard
(tree level) soft term formulae
\cite{Soni-Weldon,Brignole-Ibanez-Munoz}
for the above supergravity model given by
(\ref{kahler}), (\ref{kinetic}) and (\ref{superpotential}), one can
compute the soft terms straightforwardly \cite{Mio}
\begin{eqnarray}
M &=& \frac{\sqrt{3}Cm_{3/2}}{1+\epsilon_O}\left(\sin\theta e^{-i\gamma_S} +
\frac{1}{\sqrt{3}}\epsilon_O\cos\theta e^{-i\gamma_T}\right)\ ,
\nonumber \\
m^2 &=& V_0 + m_{3/2}^2
-
\frac{3C^2m_{3/2}^2}{\left(3+\epsilon_O\right)^2}\left[
\epsilon_O\left(6+\epsilon_O\right)sin^2\theta +
\left(3+2\epsilon_O\right)\cos^2\theta\right.
\nonumber \\ &&
\left.
- 2\sqrt{3}\epsilon_O\sin\theta\cos\theta \cos(\gamma_S-\gamma_T)\right]\ ,
\nonumber \\
A &=&
-\frac{\sqrt{3}Cm_{3/2}}{3+\epsilon_O}\left[\left(3-2\epsilon_O\right)
\sin\theta
e^{-i\gamma_S} + \sqrt{3}\epsilon_O\cos\theta e^{-i\gamma_T} \right] \ .
\label{softterms}
\end{eqnarray}
Here $M$, $m$ and $A$ denote gaugino masses, scalar masses and
trilinear
parameters respectively. The bilinear parameter,
$B$, depends on the specific mechanism which could generate the
associated $\mu$ term \cite{Brignole-Ibanez-Munoz}.
Assuming that the source of the $\mu$ term is a
bilinear piece, $\mu (S,T)H_1H_2$, in the superpotential and/or
a bilinear piece, $Z(S,\bar S,T,\bar T)H_1H_2 +\ h.c.$, in the K\"ahler potential
the result is
\begin{eqnarray}
B &=& \hat \mu ^{-1} \left(\frac{T+\bar T}{3 + \epsilon_O}\right)
\left\{ \frac{\bar W(\bar S,\bar T)}{|W(S,T)|}
(S+\bar S)^{-1/2}(T+\bar T)^{-3/2}\mu m_{3/2}C \left(-3\cos
\theta e^{-i\gamma_T}\right. \right.
\nonumber\\&&
-\sqrt{3}\sin\theta e^{-i\gamma_S}+
\frac{6\cos\theta e^{-i\gamma_T}}{3+\epsilon_O}
+\frac{2\sqrt{3}\epsilon_O\sin\theta e^{-i\gamma_S}}{3+\epsilon_O}\nonumber
-\frac{1}{C}
\nonumber\\&&
\left. +\frac{F^S}{m_{3/2}C}\partial_S\ln\mu+
\frac{F^T}{m_{3/2}C}\partial_T\ln\mu\right)
\nonumber \\ &&
+(2m_{3/2}^2+V_O)Z
-m_{3/2}^2C \left[ \sqrt{3}\sin\theta(S+\bar S)(\partial_{\bar S}Z
e^{i\gamma_S}
-\partial_S Z e^{-i\gamma_S})\right.
\nonumber\\&&
\left. +\cos\theta(T+\bar T)(\partial_{\bar T}Z
e^{i\gamma_T}-\partial_T Z e^{-i\gamma_T})\right]
\nonumber \\ &&
+ Zm_{3/2}^2C \left(
\frac{6\cos\theta e^{-i\gamma_T}}{3+\epsilon_O}
+\frac{2\sqrt{3}\epsilon_O\sin\theta e^{-i\gamma_S}}{3+\epsilon_O} \right)
\nonumber\\ &&
-m_{3/2}^2C^2\left[ 3\sin^2\theta (S+\bar S)^2 \partial_{\bar S}\partial_S Z
+ \cos^2\theta(T+\bar T)^2 \partial_{\bar T}\partial_T Z \right.
\nonumber\\ &&
+
\left(\frac{S+\bar S}{3+\epsilon_O}\right)
\left(6\epsilon_O\sin^2\theta\partial_{\bar S}Z +
6\cos^2\theta\frac{T+\bar T}{S+\bar S}\partial_{\bar T}Z\right)
\nonumber \\ &&
+\sqrt{3}\sin\theta\cos\theta \left(e^{i(\gamma_S-\gamma_T)}\partial_{\bar S}
\partial_T Z+ e^{-i(\gamma_S-\gamma_T)}\partial_{\bar T}\partial_S Z\right)
(S+\bar S)(T+ \bar T)
\nonumber \\ &&
+\frac{\sqrt{3}\sin\theta\cos\theta}{3+\epsilon_O}\left(
6 e^{i(\gamma_S-\gamma_T)}(S+\bar S)\partial_{\bar S} Z \right.
\nonumber\\ &&
\left.\left.\left.
+
2\epsilon_O e^{-i(\gamma_S-\gamma_T)}(T+\bar T)\partial_{\bar T}
Z\right)
\right]
\right\}\ ,
\label{Bterm}
\end{eqnarray}
with the effective $\mu$ parameter given by:
\begin{eqnarray}
\hat \mu &=& \left(\frac{T+\bar T}{3+ \epsilon_O}\right)
\left(\frac{\bar W(\bar S,\bar T)}{|W(S,T)|}
(S+\bar S)^{-1/2}(T+\bar T)^{-3/2}
\mu \right.
\nonumber\\&&
\left. +m_{3/2}Z-\bar
F^{\bar S}\partial _{\bar S}Z-\bar F^{\bar T}\partial _{\bar T}Z\right)
\ ,
\label{mu}
\end{eqnarray}
where $\partial_S(\partial_T)\equiv
\frac{\partial}{\partial S}(\frac{\partial}{\partial T})$.
The part of (\ref{Bterm}) depending on $\mu$ was first computed in
\cite{Bailin,Kokorelis}.
We are using here the parameterization
introduced in \cite{Brignole-Ibanez-Munoz2}
in order to know what fields, either $S$ or $T$, play the predominant role
in the process of SUSY breaking
\begin{eqnarray}
F^S &=& \sqrt 3 m_{3/2}C(S+\bar S)\sin\theta e^{-i\gamma_S}\ ,
\nonumber \\
F^T &=& m_{3/2}C(T+\bar T)\cos\theta e^{-i\gamma_T}\ ,
\label{fterms}
\end{eqnarray}
with $m_{3/2}$ for the gravitino mass, $C^2=1+V_0/3m_{3/2}^2$
and $V_0$ for the (tree--level) vacuum energy density.
As mentioned in the introduction, the structure of these soft terms is
qualitatively different from
that of the weakly--coupled heterotic string
found in \cite{Brignole-Ibanez-Munoz2}, implying interesting
low--energy ($\approx M_W$) phenomenology \cite{Mio,Bailin,Chi}.
In particular, in Fig.~1 of \cite{Mio} the dependence on
$\theta$ of the soft terms for different values of $\epsilon_O$
in the range (\ref{epsilonbound}) is shown. For any value of
$\theta$, gauginos are heavier than scalars. We will come back to
discuss this point in more detail below.
\subsection{Non--standard embedding}
Although in the non--standard embedding case, there is no requirement
that the spin connection be embedded in the gauge connection, the
form of the effective action is still the same as in the standard--embedding
case, i.e. determined by
(\ref{kahler}), (\ref{kinetic}) and (\ref{superpotential}).
Also constraint (\ref{constraint})
is still valid.
However, the relevant difference now is that the possibility
\begin{eqnarray}
\beta_O < 0\ ,
\label{menorcero}
\end{eqnarray}
is allowed \cite{Benakli,Lalak,five-branes}.
This is the case for example of two
Calabi--Yau
models in \cite{Kachru}. One of them has $E_6$ as observable gauge
group
and three families with
$\beta_O=-8$ \cite{Benakli}, and the other has $SU(5)$ as observable
gauge group with $\beta_O=-4$ \cite{Bailin} .
We will revisit then the previous computations taking into account this
novel fact. In this sense we are concentrating in this subsection
on non--standard embedding models with $\beta_O<0$. The study of those with
$\beta_O>0$ is included in the previous subsection.
Since $\epsilon_O=-\epsilon_H<0$,
the volume $V_O$ in (\ref{volumeO}) is now smaller than
$V_H$ in (\ref{volumeH})
and therefore the
gauge coupling of the observable sector
(\ref{alphaO'}) will be stronger than the one of the
hidden sector\footnote{In the context of supersymmetry
breaking by gaugino condensation this scenario has several advantageous
features with respect to the standard embedding scenario. For a discussion
about this point see \cite{Lalak}.} (\ref{alphaH'}).
Besides, since $V_O$ must be a positive quantity, one has to impose the
bound $\epsilon_O>-1$. Altogether one gets
\begin{equation}
-1 < \epsilon_O
< 0\ ,
\label{epsilonbound'}
\end{equation}
which corresponds, using (\ref{nueva}), to the bound
(see also Fig.~\ref{epsilon})
\begin{eqnarray}
(S+\bar S)
> 4\ .
\label{bound1'}
\end{eqnarray}
Note that $\epsilon_O$
can approach the limit $-1$ only for very large values of
($S+\bar S$) and therefore of ($T+\bar T$).
\subsubsection{Scales}
Let us now study how the scales are modified in these models with respect
to those with $\beta_O>0$
studied in the previous subsection. We can use again
(\ref{gut3}), but now with $-1<\epsilon_O<0$.
This is shown in the left hand side
of the Fig.~\ref{scales1}.
Unlike the models of the previous subsection where always
$V_O^{-1/6}$ was bigger than the GUT scale
$3\times 10^{16}$ GeV for any $\beta_O>0$, in these non--standard
embedding models such a value can be obtained. For example in the
case shown in the figure, $\beta_O=-1$, with
$\epsilon_O=-0.35$
which, using (\ref{nueva}) and (\ref{epsilonO}),
corresponds to $S+\bar S=6.15$ and
$T+\bar T=2.15$, we obtain
$V_O^{-1/6} = 3\times 10^{16}$ GeV.
For other values of $\beta_O$ this is also possible. Notice that, as
discussed in the previous subsection, the
figure for $V_O^{-1/6}$ will be the same adding the constant
$\log |\beta_O|^{1/2}$. So still there will be lines, corresponding
to $V_O^{-1/6}$, intersecting with the
straight
line corresponding to
$M_{GUT} = 3\times 10^{16}$ GeV.
In this sense, if we want to obtain models with the
phenomenologically favored GUT scale, the non--standard embedding
is more compelling than cases with $\beta_O>0$.
On the other hand,
in the previous subsection
we obtained the lower bound $10^{16}$ GeV for all scales of
the theory (see the right hand side of Fig.~\ref{scales1}),
far away from any direct experimental detection. Now we want
to study this issue in the non--standard embedding.
In fact, it was first pointed out in \cite{Benakli2}
that the possibility
of lowering the scales of the theory with even
an extra dimension as large as a millimetre
in some special limits is allowed in M--theory. We will analyze this
in detail clarifying whether or not such limits may be naturally
obtained.
~From (\ref{gut3}), clearly
in the limit $\epsilon_O\rightarrow -1$ we are able to obtain
$V_O^{-1/6}\rightarrow 0$ and therefore, given (\ref{relation3}) also
$(\pi\rho)^{-1}\rightarrow 0$ (see the left hand side of Fig.~\ref{scales1}).
Thus to lower the scale $V_O^{-1/6}$ down to the experimental bound
(due to Kaluza--Klein excitations) of $1$ TeV
is possible in this limit. However, this is true only
for values of $\epsilon_O$ extremely close to $-1$.
For example,
for $\epsilon_O=-0.999$ which, using (\ref{nueva}) and (\ref{epsilonO}),
corresponds to $S+\bar S=4000$ and
$T+\bar T=3996$, we obtain\footnote{Since
the unification scale has been lowered, the value
$(2\pi\alpha_O)^{-1}=4$
should be accordingly modified. However, the results are
not going to be essentially modified by this small change.}
$V_O^{-1/6} = 8\times 10^{13}$ GeV
and
$(\pi\rho)^{-1} = 10^{11}$ GeV.
For $\epsilon_O=-0.999999$ corresponding to $S+\bar S=4\times 10^{6}$ and
$T+\bar T=4\times 10^{6}-4$, we obtain the intermediate scale
$V_O^{-1/6} = 2.5\times 10^{11}$ GeV, i.e. $M_{11}=5\times 10^{11}$ GeV,
with
$(\pi\rho)^{-1} = 3\times 10^{6}$ GeV. This is an interesting
possibility
since
an intermediate scale $\approx 10^{11}$ GeV was proposed in
\cite{Benakli2} in order to solve some phenomenological problems and in
\cite{Nosotros} in order to solve the $M_W/M_{Planck}$ hierarchy.
In any case,
it is obvious that the smaller the scale the larger the amount of
fine--tuning becomes. The experimental lower bound for the scale $V_O^{-1/6}$,
$1$ TeV, can be obtained with
$\epsilon_O=10^{-16}-1$, i.e.
$S+\bar S=4\times 10^{16}$ and
$T+\bar T=4\times 10^{16}-4$. Then one gets
$V_O^{-1/6}=1181.5$ GeV with $(\pi\rho)^{-1} = 3.2\times 10^{-9}$ GeV.
Since only gravity is free to propagate in the orbifold, this extremely
small value is not a problem from the experimental point of view.
In any case,
it is clear that low scales are possible
but the fine--tuning needed renders the situation highly
unnatural. Another problem related with the limit
$\epsilon_O\rightarrow -1$ will be found below when studying soft
terms,
since $|M|/m_{3/2}\rightarrow \infty$. Thus a extremely small gravitino
mass is needed to fine tune the gaugino mass $M$ to the $1$ TeV scale
in order to avoid the gauge hierarchy problem.
There is a value of $\beta_O$ which is in principle allowed and has not been
analyzed yet. This is the case $\beta_O=0$. As we will see in a moment,
to lower the scales a lot in this context is again possible.
Since $\epsilon_O$ in (\ref{epsilonO}) is vanishing and using
(\ref{nueva}),
$S+\bar S=4$, eq. (\ref{gut2}) can be written as
\begin{eqnarray}
V_O^{-1/6}
&=&
3.6\times 10^{16}
\left(\frac{2}{T+ \bar T}\right)^{1/2}
{\rm GeV}\ ,
\label{ggut2}
\end{eqnarray}
This is plotted in Fig.~\ref{scales2} together with
$(\pi\rho)^{-1}$ and $M_{11}$. We see that the value
$V_O^{-1/6} = 3\times 10^{16}$ GeV is obtained
for the reasonable value $T+\bar T=2.88$.
\begin{figure}[htb]
\begin{center}
\epsfig{file= scales2.eps, width=8cm, height=8cm}
\end{center}
\vspace{-1.0cm}
\caption{$\log M_{11}$, $\log V_O^{-1/6}$ and $\log (\pi\rho)^{-1}$
versus $\log (T+\bar T)$ for the case $\beta_O=0$.
The straight line
indicates the phenomenologically favored GUT scale,
$M_{GUT} = 3\times 10^{16}$ GeV.\label{scales2}}
\end{figure}
On the other hand,
the larger $T+\bar T$ the smaller
$V_O^{-1/6}$ becomes. In this way,
for
$T+\bar T=2.6\times 10^{11}$ GeV
one gets the intermediate scale $V_O^{-1/6}=10^{11}$ GeV, i.e.
$M_{11}=2\times 10^{11}$, with
$(\pi\rho)^{-1}=0.2$ GeV.
The lower bound for $V_O^{-1/6}$ is obtained
with $T+\bar T=4\times 10^{19}$ GeV. Then
one gets $V_O^{-1/6}=8\times 10^{6}$ GeV and
$(\pi\rho)^{-1}=10^{-13}$ GeV. Smaller values of
$V_O^{-1/6}$ are not allowed since experimental results on the
force of gravity constrain
$(\pi\rho)$ to be less than a millimetre.
Thus, although low scales are allowed for the particular value
$\beta_O=0$, clearly a hierarchy problem between
$S+\bar S$ and $T+\bar T$ is introduced.
\subsubsection{Soft terms}
Since as mentioned above, the form of the effective action is still the
same as in the standard--embedding case,
in order to analyze the soft terms
(\ref{softterms}) is still valid but for values of $\epsilon_O$ given
by (\ref{epsilonbound'}). This implies, as we will see below,
that the structure of these soft terms be qualitatively different
from that of the standard embedding case.
In what follows, given the current experimental limits,
we will assume $V_0=0$
and $\gamma_S=\gamma_T=0\ ({\rm mod}\ \, \pi)$.
More specifically, we will set $\gamma_S$ and $\gamma_T$
to zero and allow $\theta$ to vary in a range $[0, 2\pi)$.
We show in Fig.~\ref{nonstandardsoft} the dependence on $\theta$
of
the soft terms $M$, $m$, and $A$ in units of the gravitino mass for
different values of $\epsilon_O$.
Notice that for
$\theta \in [\pi, 2\pi)$ the corresponding figures could have easily been
deduced
since the shift $\theta\rightarrow \theta + \pi$ in (\ref{softterms})
implies $m\rightarrow m$, $M\rightarrow -M$ and $A\rightarrow -A$.
However we prefer to plot them explicitly to compare with the case
which will be
studied in the next section
where due to the inclusion of five--branes
this
symmetry is broken (see e.g. Fig.~\ref{susyfivesoft}).
Several comments are in order.
First of all, some (small) ranges of $\theta$ are forbidden
by having a negative scalar mass-squared.
About the possible range of soft terms,
the smaller the value of $\epsilon_O$, the larger the range becomes.
For example,
for
$\epsilon_O=-1/3$,
those ranges are
$0.1<|M|/m_{3/2}<2.65$, $0<m/m_{3/2}<1.35$ and $0.18<|A|/m_{3/2}<2.41$,
whereas
for $\epsilon_O=-3/5$,
they are
$0<|M|/m_{3/2}<4.58$, $0<m/m_{3/2}<1.67$ and $0.25<|A|/m_{3/2}<3.12$.
\begin{figure}[htbp]
\hspace{-0.7cm}
\epsfig{file= figure1.eps, width=16cm, height=16cm}
\vspace{-1.0cm}
\caption{Soft parameters in units of $m_{3/2}$ versus $\theta$ for
different values of $\epsilon_O$ in the non--standard embedding case. Here
$M$, $m$
and $A$ are the gaugino mass, the scalar mass and the trilinear parameter
respectively.\label{nonstandardsoft}}
\end{figure}
This result is to be compared with the one of \cite{Mio} where the
standard-embedding soft terms were analyzed. From Fig.~1 of that paper,
where cases $\epsilon_O=1/7,1/3,3/5,1$ where shown, we see that the
forbidden ranges of $\theta$ are now substantially decreased
For example, whereas the dilaton--dominated case, $|\sin\theta| = 1$
is always allowed, in the standard--embedding situation it may be
forbidden
depending on the value of $\epsilon_O$. Also we see that the range
of soft terms is now substantially increased. In the limit
$\epsilon_O\rightarrow -1$, $0.3<|A|/m_{3/2}<4.58$,
$0<m/m_{3/2}<2.26$ and $|M| \rightarrow \infty$.
In order to discuss the SUSY spectra further, it is worth
noticing that although gaugino masses are in general larger than
scalar masses, for values of $\epsilon_O$ approaching $-1$ there are two
narrow ranges of values of $\theta$ where the opposite situation
occurs.
This can be seen in Fig.~\ref{nonstandardsoft}
for the cases $\epsilon_O=-3/5, -9/10$.
Let us remark that $M/m_{3/2}$ and $m/m_{3/2}$ are then very
small
and therefore $m_{3/2}$ must be large in order to fulfil e.g. the
low--energy
bounds on gluino masses \footnote{This is a similar situation to
that of the weakly--coupled orbifold scenario (O-II) of
\cite{Brignole-Ibanez-Munoz2}. There for untwisted particles, in the limit
$\sin\theta\rightarrow 0$, scalar and gaugino masses vanish at tree
level:
then string loop effects become important and tend to make scalars
heavier than gauginos.}.
These ranges of $\theta$ can be seen in more detail
in Fig.~\ref{nonstandardcociente},
where the ratio $m/|M|$ versus $\theta$ is plotted
for different values of $\epsilon_O$.
This result is to be compared with the one of Fig.~3 in
\cite{Mio}.
There, the standard embedding case is analyzed and
$r\equiv m/|M|< 1$
for any value of
$\theta$.
Fig.~\ref{nonstandardcociente} also allows us to study some properties of the low--energy
($\approx M_W$)
spectra
independently of the details of the electroweak breaking
using the formula (see e.g. \cite{Mio})
\begin{eqnarray}
&&
\hspace{-1.0cm}
M_{\tilde g}:m_{\tilde Q_L}:m_{\tilde u_R}:m_{\tilde d_R}:m_{\tilde L_L}:
m_{\tilde e_R}
\nonumber \\
&&
\hspace{-1.0cm}
\approx
1:\frac{1}{3}\sqrt {7.6+r^2}:\frac{1}{3}\sqrt {7.17+r^2}:
\frac{1}{3}\sqrt {7.14+r^2}:\frac{1}{3}\sqrt{0.53+r^2}:\frac{1}{3}
\sqrt{0.15+r^2}\ ,
\label{masas}
\end{eqnarray}
where $\tilde g$ denote the gluino,
$\tilde l$ all the sleptons and $\tilde q$
first and second generation squarks. Most values of $\theta$ imply
$r<1$ and from (\ref{masas}) we obtain
\begin{eqnarray}
M_{\tilde g}\approx m_{\tilde q}>m_{\tilde l}\ .
\label{gluinos1}
\end{eqnarray}
This is also the generic (tree--level) result in Calabi--Yau compactifications
of the weakly--coupled heterotic
string, which can be recovered from (\ref{softterms}) by taking the
limit
$(T+\bar T)<<(S+\bar S)$, i.e. $\epsilon_O\rightarrow 0$.
Then $M=\sqrt {3} m=\sqrt {3} m_{3/2} \sin\theta$ \cite{Brignole-Ibanez-Munoz2}
implying $r=1/\sqrt 3$ except in the limit $\sin\theta\rightarrow 0$
which is not well defined.
Only in this limit one might expect $r>1$,
similarly to what happens in the
orbifold case, due to string loop corrections. This is something to be
checked \cite{NUEVA}.
However, when $\epsilon_O$
is approaching $-1$ the above situation (\ref{gluinos1})
concerning Fig.~\ref{nonstandardcociente} can be reversed since
\begin{eqnarray}
M_{\tilde g}<m_{\tilde q}\approx m_{\tilde l}\ ,
\label{gluinos2}
\end{eqnarray}
for
$\theta$ in the narrow ranges where $r>1$.
Finally, notice that in the case $\beta_O=0$, i.e. $\epsilon_O=0$,
which corresponds to the straight line $m/|M|=1/\sqrt 3$,
the phenomenologically interesting sum--rule
$\sum_{i=1}^3 m_i^2=M^2$ \cite{Scheich}
is trivially fulfilled. Moreover, it
is also fulfilled for any other possible value
\footnote{We thank T. Kobayashi for putting forward this fact to us.}
of $\epsilon_O$
when $\theta=\pi/6$ (and $\pi/6 + \pi$) since then the
$\epsilon_O$ contribution in (\ref{softterms}) is cancelled and one
obtains $m=m_{3/2}/2$ and
$|M|=\sqrt 3 m_{3/2}/2$.
See also in Fig.~\ref{nonstandardcociente}
how all graphs intersect each other at those angles.
Since this result is independent on the value of $\epsilon_O$, it
happened also for $0<\epsilon_O<1$ (see Fig.~3 in \cite{Mio}).
\begin{figure}[htbp]
\begin{center}
\epsfig{file= menosb.eps, width=8cm, height=8cm}
\end{center}
\vspace{-1.5cm}
\caption{$m/|M|$ versus $\theta$ for different values of $\epsilon_O$
in the non--standard embedding case.\label{nonstandardcociente}}
\end{figure}
\section{Vacua with five-branes}
In the previous section, we studied the phenomenology of heterotic
M--theory vacua obtained through standard and non--standard
embeddings.
Here we want to analyze (non--perturbative) heterotic M--theory vacua
due to the presence of five--branes \cite{Witten}.
As mentioned in the introduction, five--branes are non--perturbative objects,
located at points, $x^{11}=x_n (n=1,...,N)$, throughout the orbifold interval.
The modifications to the four--dimensional effective action determined by
(\ref{kahler}), (\ref{kinetic}) and (\ref{superpotential}),
due to their presence, have recently been investigated by
Lukas, Ovrut and Waldram \cite{five-branes,five-branes2}. Basically,
they are due to existence of moduli, $Z_n$, whose
${\rm Re}(Z_n)\equiv z_n=x_n/\pi\rho \in (0,1)$
are the five--brane positions in the normalized orbifold coordinates.
Then, the effective supergravity obtained from heterotic M--theory
compactified
on a Calabi--Yau manifold in the presence of five--branes is now determined by
\begin{eqnarray}
K &=& -\ln (S+\bar{S}) -3\ln (T+\bar{T})+ K_5 + \frac{3}{T+\bar{T}}
\left(1+\frac{1}{3}
e_O\right) H_{pq}C_O^p \bar{C}_O^q,
\nonumber\\
f_{O} &=& S+
B_O
T\ , \quad f_{H}=S+
B_H
T\ ,\nonumber \\
W_O &=& d_{pqr}C_O^pC_O^qC_O^r\ ,
\label{kahlerbranas}
\end{eqnarray}
with
\begin{equation}
e_O=b_O
\frac{T+\bar T}{S+\bar S}\ .
\label{Es}
\end{equation}
Here
$K_5$ is the K\"ahler potential for the five--brane moduli
$Z_n$, $H_{pq}$ is some $T$--independent metric and
\begin{eqnarray}
b_O &=& \beta_O + \sum_{n=1}^N(1-z_n)^2\beta_n\ ,
\nonumber \\
B_O &=& \beta_O + \sum_{n=1}^N(1-Z_n)^2\beta_n\ ,
\nonumber \\
B_H &=& \beta_H + \sum_{n=1}^N(Z_n)^2\beta_n\ ,
\label{Bes}
\end{eqnarray}
with $\beta_O$, $\beta_H$ the instanton numbers and $\beta_n$
the five--brane charges. The former, instead of constraint (\ref{constraint}),
must fulfil the following constraint:
\begin{equation}
\beta_O + \sum_{n=1}^N\beta_n+\beta_H=0\ .
\label{Besbis}
\end{equation}
It is worth noticing that in addition to the observable and hidden
sector
gauge groups, $G_O$ and $G_H$, there appear gauge groups
from the five--branes. Thus the total gauge group at low energies is in fact
$G_O\times G_H\times G_1\times ...\times G_N$ \cite{five-branes,five-branes2}.
We are assuming here
that $G_O$ arises from one of the $E_8$
groups. In this sense the five--brane sectors are considered
hidden sector interacting with the observable sector only through
bulk supergravity.
Let us also remark that we are considering, as in the previous section,
a compactification on
a Calabi--Yau manifold with only one K\"ahler modulus
$h_{1,1}=1$. As discussed in the introduction, such Calabi--Yau spaces
exist and their phenomenological properties are extremely interesting.
Assuming for simplicity that $<Z_n>=<z_n>$, i.e. $<B_O>=<b_O>$, the set
of eqs. (\ref{alphaO})--(\ref{volumeH}) is still valid with the
modification
$\epsilon_{O,H}\rightarrow e_{O,H}$, where
\begin{eqnarray}
e_H=b_H\frac{T+\bar T}{S+\bar S}\ ,
\label{lala}
\end{eqnarray}
with
\begin{eqnarray}
b_H = \beta_H + \sum_{n=1}^N(z_n)^2\beta_n\ .
\label{Besbis2}
\end{eqnarray}
Following the
analysis
of subsection 2.2
we can write $e_O$
as
\begin{eqnarray}
e_O
=
\frac{4-(S+\bar S)}
{(S+\bar S)}\ ,
\label{nueva2}
\end{eqnarray}
and therefore
Fig.~\ref{epsilon} is still valid substituying $\epsilon_O$ by $e_O$.
We can obtain different bounds on $e_O$
depending
on the sign of both $b_O$ and $b_H$:
\noindent {\it i)} $b_H\geq 0$, $b_O\leq 0$
Then $e_H$ is positive and $e_O$ negative. Since
$V_O=V(1+e_O)$ must be positive we need
\begin{eqnarray}
-1<e_O\leq 0\ .
\label{cota1}
\end{eqnarray}
\noindent {\it ii)} $b_H\geq 0$, $b_O>0$
Now since $e_O$ is positive $V_O$ will always be positive and
therefore
the only bound is
\begin{eqnarray}
0<e_O\ .
\label{cota2}
\end{eqnarray}
\noindent {\it iii)} $b_H<0$, $b_O>0$
In this case $e_H$ is negative. Since
$V_H=V(1+e_H)$ must be positive we need
$e_H>-1$. On the other hand, using (\ref{Es}) and (\ref{lala}),
$e_O=e_H\frac{b_O}{b_H}$
and as a consequence the following bounds are obtained
\begin{eqnarray}
0<e_O<\frac{b_O}{|b_H|}\ .
\label{cota3}
\end{eqnarray}
\noindent {\it iv)} $b_H<0$, $b_O\leq 0$
Now both $e_O$ and $e_H$ are negative. To avoid negative volumes
we need $e_O>-1$ and $e_H>-1$. As discussed above we can write
$e_O=e_H\frac{b_O}{b_H}$ and therefore two possibilities arise:
If $|b_O|\geq |b_H|$
\begin{eqnarray}
-1<e_O\leq 0\ .
\label{cota5}
\end{eqnarray}
If $|b_O|<|b_H|$
\begin{eqnarray}
-\frac{b_O}{b_H}< e_O\leq 0\ .
\label{cota4}
\end{eqnarray}
For instance, for the example studied in \cite{five-branes}
where there are four five--branes at
$(z_1,z_2,z_3,z_4)=(0.2,0.6,0.8,0.8)$
with charges $(\beta_1,\beta_2,\beta_3,\beta_4)=(1,1,1,1)$
and instanton number $(\beta_O,\beta_H)=(-1,-3)$, one obtains
$b_O=-0.12$, $b_H=-1.32$ and then one should apply
(\ref{cota4}) with the result $-0.09\ler e_O\ler 0$.
The
standard and non--standard
embeddings studied in section 2 are particular cases of this more
general
analysis with five--branes.
For $\beta_n=0$ in (\ref{Bes}) and (\ref{Besbis}), i.e. no five--branes,
we have $b_{O}=\beta_{O}=-\beta_H=-b_H$
and $e_O=\epsilon_O$. If $\beta_O<0$,
from {\it i)}
we recover the non--standard embedding case, $-1<\epsilon_O<0$. This is the
part of the graph corresponding to $S+\bar S>4$ shown in Fig.~\ref{epsilon}.
For $\beta_O>0$, from {\it iii)} we recover the standard embedding
(and also some non--standard
embedding) case since $0<\epsilon_O<1$
This is the part of the graph corresponding to $2<S+\bar S<4$ shown in
Fig.~\ref{epsilon}.
It is worth noticing that the values $0<(S+\bar S)<2$, corresponding to
$\epsilon_O>1$,
which are not possible in the absence of five--branes,
are allowed in their presence since
$e_O>1$ is possible (cases {\it ii)}
and {\it iii)}).
\subsection{Scales}
In the presence of five--branes (\ref{constraint2}) is no longer true since
(\ref{volumeO}) and (\ref{volumeH}) are modified in the following way:
$V_O=V(1+e_O)$ and $V_H=V(1+e_H)$. Therefore
\begin{equation}
<V>=V \left( 1+ \frac{e_O+e_H}{2}\right)
\ .
\label{medio}
\end{equation}
Then the relevant formulae to study the relation between the different
scales of the theory are (\ref{gut}) and
(\ref{relation2}) with the modification $\epsilon_O\rightarrow e_O$.
Notice that (\ref{relation}) is not modified.
Similarly to the case without five--branes, to obtain
$V_O^{-1/6}\approx 3\times 10^{16}$ GeV when
$T+\bar T$ and $S+\bar S$
are of order one is quite natural. This can be seen
from (\ref{gut})
\begin{eqnarray}
V_O^{-1/6} &=&
\left(\frac{1}{1+\frac{e_O}{2}\left(1+\frac{b_H}{b_O}\right)}\right)^{1/2}
3.6\times 10^{16}
\left(\frac{4}{S+\bar S}\right)^{1/2}
\left(\frac{2}{T+ \bar T}\right)^{1/2}
\left(\frac{1}{1+e_O}\right)^{1/6}
{\rm GeV}
\nonumber\\
\label{puf}
\end{eqnarray}
where (\ref{medio}) with (\ref{lala}) and (\ref{Es}) has been used.
Using (\ref{Es}) and (\ref{nueva2}) it is interesting to write
(\ref{puf}) as
\begin{eqnarray}
V_O^{-1/6}
&=&
\left(\frac{1}{1+\frac{e_O}{2}\left(1+\frac{b_H}{b_O}\right)}\right)^{1/2}
3.6\times 10^{16}
\left(\frac{b_O}{2 e_O}\right)^{1/2}
\left(1+e_O\right)^{5/6}
{\rm GeV}\ .
\label{niidea}
\end{eqnarray}
In the left hand side of the Fig.~\ref{scales1five}a
we show an example of the case {\it i)}, $b_O=-7/4$ and $b_H=5/4$,
corresponding to $-1<e_O<0$. In the right hand side
we show an example of the case {\it ii)}, $b_O=b_H=1/2$,
corresponding to $e_0>0$. Both are interesting examples since
they cover the whole range of validity of $e_O$ and will be used below
to study the soft terms. Unlike the standard and non--standard
embedding
cases with $\beta_O>0$ shown in the right hand side of Fig.~\ref{scales1},
now the line corresponding to $V_O^{-1/6}$ intersects easily
the straight line corresponding to the GUT scale. This is obtained
for $e_O=0.46$ which, using
(\ref{nueva2}) and (\ref{Es}),
corresponds to $S+\bar S=2.73$ and
$T+\bar T=2.54$.
Of course this effect is due essentially
to the extra factor appearing in (\ref{niidea}), coming from the
average
volume,
with respect to (\ref{gut3}).
\begin{figure}[htb]
\hspace{-0.7cm}
\epsfig{file= scale.eps, width=16cm, height=8cm}
\vspace{-1.0cm}
\caption{$\log M_{11}$, $\log V_O^{-1/6}$ and $\log (\pi\rho)^{-1}$
versus (a) $e_O$ and (b) $\log e_O$
in the cases $b_O=b_H=1/2$ (for $0<e_O$)
and $b_O=-7/4$, $b_H=5/4$ (for $-1<e_O<0$). The straight line
indicates the phenomenologically favored GUT scale,
$M_{GUT} = 3\times 10^{16}$ GeV.\label{scales1five}}
\end{figure}
Only in some special limits one may lower the scales.
As in the case without fivebranes, fine--tuning
$e_O\rightarrow -1$
we are able to obtain $V_O^{-1/6}$ as low as we wish.
The numerical results will be basically similar
to the ones of non--standard embedding in subsection 2.3.1
Moreover, $e_O>1$ is possible in the presence of five--branes.
Therefore with $e_O$ sufficiently large we may get $V_O^{-1/6}$
very small. This is shown in Fig.~\ref{scales1five}b.
For example, with $\log e_O=56.1$ the experimental lower bound
$(\pi\rho)^{-1}=10^{-13}$ GeV
is obtained for $V_O^{-1/6}=8\times 10^{6}$ GeV, corresponding
to $S+\bar S=3.1\times 10^{-56}$ and
$T+\bar T=8$.
Clearly we introduce a hierarchy problem.
\begin{figure}[htb]
\begin{center}
\epsfig{file= scales3.eps, width=8cm, height=8cm}
\end{center}
\vspace{-1.0cm}
\caption{$\log M_{11}$, $\log V_O^{-1/6}$ and $\log (\pi\rho)^{-1}$
versus $\log (T+\bar T)$ for the case $b_O=0$.
The straight line
indicates the phenomenologically favored GUT scale,
$M_{GUT} = 3\times 10^{16}$ GeV.\label{scales2five}}
\end{figure}
Finally,
Let us analyze the special case $b_O=0$. The analysis will be
similar
to the one of the case $\beta_O=0$ without five--branes in subsection
2.3.1.
Since $e_O$ in (\ref{Es}) is vanishing and using (\ref{nueva2}),
$S+\bar S=4$, eq. (\ref{gut}) can be written as
\begin{eqnarray}
V_O^{-1/6}
&=&
\left(\frac{1}{1+\frac{1}{8}b_H(T+\bar T)}\right)^{1/2}
3.6\times 10^{16}
\left(\frac{2}{T+ \bar T}\right)^{1/2}
{\rm GeV}\ ,
\label{media}
\end{eqnarray}
where (\ref{lala}) has been used. Depending on the value of $b_H$
we obtain different results. If $b_H=0$, eq. (\ref{ggut2}) is recovered
and therefore we obtain the same results as in the case
$\beta_O=\beta_H=0$ without five--branes (see Fig.~\ref{scales2}).
If $b_H>0$ the results are qualitatively similar, the larger
$T+\bar T$ the smaller $V_O^{-1/6}$ becomes. However, notice that now
for large $T$
we have a factor $(T+\bar T)^{-1}$ and then not so large values of
$T+\bar T$ as in Fig.~\ref{scales2} are needed in order to lower the scales.
This is shown in Fig.~\ref{scales2five} for the case $b_H=1$.
Then $V_O^{-1/6}=1$ TeV can be obtained for $T+\bar T=10^{14}$
with the size of
the orbifold $(\pi\rho)^{-1}=5\times 10^{-12}$ GeV close to its experimental
bound of $1$ millimetre.
In any case, still a large hierarchy between $S+\bar S$ and
$T+\bar T$ is needed.
Finally, for $b_H<0$ we are in the case {\it iv)} and therefore we
have
the constraint
$0<(T+\bar T)<4/|b_H|$, implying that $V_O^{-1/6}$ around the GUT scale
can be obtained but lowering it to intermediate or $1$ TeV values is
not
possible.
\subsection{Soft terms}
Let us now concentrate on the computation of soft terms. The above
supergravity model (\ref{kahlerbranas}) gives rise to the following
gaugino masses, scalar masses and trilinear parameters:
\begin{eqnarray}
M &=& \frac{1}{(S+\bar S)(1+\frac{B_O T + \bar B_O \bar T}{S+ \bar S})}
\left(F^S + F^T B_O + T F^n {\partial_n} {B_O} \right)\ ,
\nonumber \\
m^2 &=& V_0 + m_{3/2}^2 -\frac{1}{(3+e_O)^2}
\left[e_O(6+e_O) \frac{|F^S|^2}{(S+\bar S)^2} \right.
\nonumber\\ &&
+3(3+2e_O)
\frac{|F^T|^2}{(T+\bar T)^2}
-\frac{6e_O}{(S+\bar S)(T+\bar T)} {\rm Re\,} F^S \bar F^{\bar T}
\nonumber\\&&
+ \left(\frac{e_O}{b_O}(3+e_O)
\partial_n\partial_{\bar m} b_O - \frac{e_O^2}{b_O^2}
\partial_n b_O \partial_{\bar m} b_O \right) F^n\bar F^{\bar m}
\nonumber \\ &&
\left. -
\frac{6e_O}{b_O}\frac{\partial_{\bar n} b_O }{S+\bar S}
{\rm Re\,} F^S \bar F^{\bar n}
+ \frac{6e_O}{b_O}\frac{\partial_{\bar n} b_O}{T+\bar T}
{\rm Re\,}
F^T \bar F^{\bar n}
\right]\ ,
\nonumber\\
A&=&-\frac{1}{3+\epsilon_O}\left[
(3-2e_O)\frac{F^S}{S+\bar S}+
3e_O \frac{F^T}{T+\bar T} \right.
\nonumber\\ &&
+ \left. \left( \frac{3e_O}{b_O}\partial_n b_O
-(3+ e_O){\partial _n} K_5
\right) F^n \right]\ ,
\label{softfive}
\end{eqnarray}
where $\partial_n\equiv \frac{\partial}{\partial Z_n}$ and for
the moment we write explicitly the $F$--terms of the dilaton ($F^S$),
modulus ($F^T$) and five--branes ($F^n$). They must fulfil the relation
\begin{equation}
V_0=\frac{|F^S|^2}{(S+\bar S)^2}+\frac{3|F^T|^2}{(T+\bar T)^2}+
\bar F^{\bar n} F^m
\partial_{\bar n} \partial_m K_5 -3m_{3/2}^2 \ .
\label{potencial}
\end{equation}
Assuming, as in subsection~2.2.2, that the source of the $\mu$ term
is a bilinear piece in the superpotential and/or the K\"ahler
potential,
the result is
\begin{eqnarray}
B&=& \hat\mu^{-1}\left(\frac{T+\bar T}{3+e_O}\right)\left\{\frac{\bar W(\bar S,\bar T)}{|W(S,T)|}\ (S+\bar S)^{-1/2}(T+\bar T)^{-3/2}e^{K_5/2}\mu\right.
\nonumber \\
&&\left[ F^S\left(\frac{-1}{S+\bar S}+\right.\right.
\left.\partial_S\log\mu+\frac{2e_O}{3+e_O}\frac{1}{S+\bar S}\right)
+F^T\left(\frac{-3}{T+\bar T}+\right.\partial_T\log\mu
\nonumber \\
&&\left.+\frac{6}{3+e_O}\frac{1}{T+\bar T}\right)+
F^n\left(\partial_nK_5+\partial_n\log\mu-\frac{2\partial_nb_O}{3+e_O}\right.
\left.\frac{T+\bar T}{S+\bar S}\right)
-\left.m_{3/2}\right]
\nonumber\\
&&+
(2m_{3/2}^2+V_0)Z-m_{3/2}
\left(\bar F^{\bar S}\partial_{\bar S}Z+\right.
\left.\bar F^{\bar T}\partial_{\bar T}Z+\bar F^{\bar n}\partial_{\bar n}Z\right)
\nonumber\\
&&+m_{3/2}
\left[F^S\left(\partial_SZ+\frac{2e_O}{3+e_O}\right.\right.
\left.\frac{1}{S+\bar S}Z\right)+
F^T\left(\partial_TZ+\frac{6}{3+e_O}\right.
\left.\frac{1}{T+\bar T}Z\right)
\nonumber\\
&&
+F^n\left(\partial_nZ-\frac{2\partial_nb_O}{3+e_O}\right.
\left.\left.\frac{T+\bar T}{S+\bar S}Z\right)\right]-
|F^S|^2
\left(\partial_S\partial_{\bar S}Z+\frac{2e_O}{3+e_O}\frac{\partial_{\bar S}Z}{S+\bar S}\right)
\nonumber\\
&&-
|F^T|^2
\left(\partial_T\partial_{\bar T}Z+\frac{6}{3+e_O}\frac{\partial_{\bar T}Z}{T+\bar T}\right)
-\bar F^{\bar n}F^m
\left(\partial_m\partial_{\bar n}Z-\frac{2\partial_mb_O}{3+e_O}\right.
\left.\frac{T+\bar T}{S+\bar S}\partial_{\bar n}Z\right)
\nonumber\\
&&-\bar F^{\bar S}F^T
\left(\partial_T\partial_{\bar S}Z+\frac{6}{3+e_O}\frac{\partial_{\bar S}Z}{T+\bar T}\right)-
\bar F^{\bar T}F^S
\left(\partial_S\partial_{\bar T}Z+\frac{2e_O}{3+e_O}\frac{\partial_{\bar T}Z}{S+\bar S}\right)
\nonumber\\
&&-\bar F^{\bar S}F^n
\left(\partial_n\partial_{\bar S}Z-\frac{2\partial_nb_O}{3+e_O}\right.
\left.\frac{T+\bar T}{S+\bar S}\partial_{\bar S}Z\right)-
\bar F^{\bar n}F^S
\left(\partial_S\partial_{\bar n}Z+\frac{2e_O}{3+e_O}\frac{\partial_{\bar n}Z}{S+\bar S}\right)
\nonumber\\
&&-\bar F^{\bar T}F^n
\left(\partial_n\partial_{\bar T}Z-\frac{2\partial_nb_O}{3+e_O}\right.
\left.\frac{T+\bar T}{S+\bar S}\partial_{\bar T}Z\right)-
\bar F^{\bar n}F^T
\left.\left(\partial_T\partial_{\bar n}Z+\frac{6}{3+e_O}\frac{\partial_{\bar
n}Z}{T+\bar T}\right)\right\}\ ,
\label{bzeta}
\nonumber \\
\end{eqnarray}
with the effective $\mu$ parameter given by:
\begin{eqnarray}
\hat\mu&=&\left(\frac{T+\bar T}{3+e_O}\right)
\left(\frac{\bar W(\bar S,\bar T)}{|W(S,T)|}(S+\bar S)^{-1/2}(T+\bar T)^{-3/2}e^{K_5/2}\mu\right.
+ m_{3/2}Z
\nonumber \\
&&\left.-\bar{F}^{\bar S}\partial_{\bar S}Z-\bar{F}^{\bar T}\partial_{\bar T}Z-\bar{F}^{\bar n}\partial_{\bar n}Z\right)
\end{eqnarray}
Due to the possible contribution of several $F$--terms associated with
five--branes, which can have in principle off--diagonal
K\"ahler
metrics, the numerical computation of the soft terms turns out to be
extremely
involved. In order to get an idea of their value and also to study
the deviations with respect to the case without five--branes we can
do some simplifications. One possibility is to assume that
five--branes are present but only the
$F$--terms
associated with the dilaton and the modulus contribute to supersymmetry
breaking,
i.e. $F^n=0$. Then, assuming as before $<Z_n>=<z_n>$
and using parametrization (\ref{fterms}), eq. (\ref{softfive}) reduces to
eq. (\ref{softterms})
with $e_O$ instead of $\epsilon_O$.
Under these simplifying assumptions, Figs.~\ref{nonstandardsoft}
and \ref{nonstandardcociente} in section 2
(and Figs. 1 and 3 in \cite{Mio}) are also valid in this case
since, as discussed above
the
range of allowed values of $e_O$ includes those of $\epsilon_O$,
i.e. $-1<e_O<1$. The relevant difference with respect to the case
without
five--branes is that now values with $e_O\geq 1$ are allowed.
We
plot this possibility
in Figs.~\ref{nonsusyfivesoft} and \ref{nonsusyfivecociente}
for some values of $e_O$.
Fig.~\ref{nonsusyfivesoft} shows that the range of $\theta$ forbidden
by having a negative scalar mass--squared, is large and quite stable
against variations of $e_O$. This
pattern of soft terms is qualitatively different from that without
five--branes analyzed in subsection 2.3 for the non--standard embedding
and in Fig.~1 of \cite{Mio} for the standard embedding.
However, the fact that always scalar masses are smaller than gaugino masses
in the latter case (see Fig.~3 in \cite{Mio}) is also true here. In
Fig.~\ref{nonsusyfivecociente}
we show the ratio $m/|M|$ versus $\theta$ for different values
of $e_O$. Although the larger the value of
$e_O$, the larger the ratio becomes, there is an upper bound
$m/|M|=1$ (obtained for $\theta =0$). As discussed in (\ref{gluinos1})
we will obtain at low--energies,
$M_{\tilde g}\approx m_{\tilde q}>m_{\tilde l}$.
Notice that the sum--rule $3m^2=M^2$ is also fulfilled for
$\theta=\pi/6$ (and $\pi/6 + \pi$).
\begin{figure}[htb]
\hspace{-0.7cm}
\epsfig{file= figure4.eps, width=16cm, height=8cm}
\vspace{-1.0cm}
\caption{Soft parameters in units of $m_{3/2}$ versus
$\theta$ for
different values of $e_O$ when five--branes are present without
contributing to supersymmetry breaking.
\label{nonsusyfivesoft}}
\end{figure}
\begin{figure}[htb]
\begin{center}
\epsfig{file= ratiomas.eps, width=8cm, height=8cm}
\end{center}
\vspace{-1.5cm}
\caption{$m/|M|$ versus $\theta$
for different values of $e_O$
when five--branes are present without contributing to supersymmetry
breaking.\label{nonsusyfivecociente}}
\end{figure}
Another possibility to simplify the numerical computation of the soft
terms is to assume that there is only one five--brane in the model.
For example,
parametrizing consistently with (\ref{potencial})
\begin{eqnarray}
F^S &=& \sqrt{3} m_{3/2} C (S+\bar S)\sin\theta\cos\theta_1
e^{-i\gamma_S} \ ,
\nonumber\\
F^T &=& m_{3/2} C (T+\bar T)\cos\theta\cos\theta_1 e^{-i\gamma_T}\ ,
\nonumber \\
F^1 &=& \sqrt{3} m_{3/2} C
({{\partial}_1{\partial}_{\bar 1 }K_5})^{-1/2}sin\theta_1
e^{-i\gamma_1}\ ,
\label{ftermss}
\end{eqnarray}
where $\theta_1$ is the new goldstino angle associated to the $F$--term
of the five--brane $F^1$, we obtain from (\ref{softfive})
\begin{eqnarray}
M&=&\frac{ \sqrt{3} m_{3/2}C}
{\left(1+\frac{B_O T + \bar B_O \bar T}{S+ \bar S}\right)}
\left( \sin\theta\cos\theta_1
e^{-i\gamma_S}+\frac{1}{\sqrt{3}}B_O \frac{e_O}{b_O}
\cos\theta\cos\theta_1
e^{-i\gamma_T} \right.
\nonumber \\&&
\left. -\frac{2T}{S+\bar S}(1-Z_1)\beta_1
({{\partial}_1{\partial}_{\bar 1 }K_5})^{-1/2}
sin\theta_1 e^{-i\gamma_1} \right) \ ,
\nonumber \\
m^2&=&V_o + m_{3/2}^2-\frac{3 m_{3/2}^2C^2}{(3+e_O)^2}\left\{ e_O
(6+e_O)\sin^2\theta
\cos^2\theta_1 \right.
\nonumber \\&&
+(3+2e_O)\cos^2\theta\cos^2\theta_1
- 2\sqrt{3}e_O\sin\theta\cos\theta\cos^2\theta_1\cos(\gamma_S
-\gamma_T)
\nonumber \\&&
+ ({{\partial}_1{\partial}_{\bar 1 }K_5})^{-1}
\sin^2\theta_1 \left( (3+e_O)
\beta_1\frac{e_O}{2b_O} -\left[(1-z_1)
\beta_1 \frac{e_O}{b_O}\right]^2 \right)
\nonumber \\&&
+6(1-z_1)\beta_1\frac{e_O}{b_O}
({{\partial}_1{\partial}_{\bar 1 }K_5})^{-1/2}
\sin\theta\sin\theta_1\cos\theta_1\cos(\gamma_1
-\gamma_S)
\nonumber \\&&
\left. -2\sqrt{3}(1-z_1)\beta_1\frac{e_O}{b_O}
({{\partial}_1{\partial}_{\bar 1 }K_5})^{-1/2}
\cos\theta\sin\theta_1\cos\theta_1\cos(\gamma_T
-\gamma_n)\right\}\ ,
\nonumber\\
A&=&-\frac{\sqrt{3} m_{3/2}C}{3+e_O}\left[ (3-2e_O)
\sin\theta\cos\theta_1 e^{-i\gamma_S} + \sqrt{3}e_O\cos\theta\cos\theta_1
e^{-i\gamma_T} \right.
\nonumber \\&&
\left. -
({{\partial}_1{\partial}_{\bar 1 }K_5})^{-1/2}
sin\theta_1 e^{-i\gamma_1}
\left((3+e_O)
{{\partial}_1 K_5}
+ 3(1-z_1)\beta_1 \frac{e_O}{b_O} \right) \right] \ .
\label{una}
\end{eqnarray}
\begin{figure}[htb]
\hspace{-0.7cm}
\epsfig{file= figure6.eps, width=16cm, height=16cm}
\vspace{-1.0cm}
\caption{Soft parameters in units of $m_{3/2}$ versus $\theta$ for
different values of $e_O$ and
goldstino angle $\theta_1$
when one five--brane is
contributing to supersymmetry breaking.\label{susyfivesoft}}
\end{figure}
Similarly one can obtain the $B$ parameter using (\ref{bzeta}).
Unfortunately, the numerical analysis of this simplified case is not
straightforward. All soft terms depend not only on the new goldstino
angle $\theta_1$ in addition to
$m_{3/2}$, $\theta$
and $e_O$, but also on other free parameters.
For example, although gaugino masses can be further
simplified with the assumption
$<Z_n>=<z_n>$, i.e. $<B_O>=<\bar B_O>=<b_O>$, and $<T>=<\bar T>$
\begin{eqnarray}
M&=&\frac{ \sqrt{3} m_{3/2}C}
{1+e_O}
\lbrace \sin\theta\cos\theta_1 e^{-i\gamma_S}
+\frac{1}{\sqrt{3}} e_O
\cos\theta\cos\theta_1 e^{-i\gamma_T}
\nonumber \\&&
-\frac{e_O}{b_O}(1-z_1)\beta_1
({{\partial}_1{\partial}_{\bar 1 }K_5})^{-1/2}
sin\theta_1 e^{-i\gamma_1} \rbrace\ ,
\label{gauginos}
\end{eqnarray}
still
they have an explicit dependence on
$z_1$ and
$\partial_1 \partial_{\bar 1} K_5$.
Notice that, for a given model,
$\beta_O$ and $\beta_1$ are known and therefore $b_O$ can be computed
from (\ref{Bes}) once $z_1$ is fixed.
Something similar occurs for the
$A$ parameter, where $z_1$, $\partial_1 K_5$ and
$\partial_1 \partial_{\bar 1} K_5$ appear explicitly, and
for the scalar masses, where
$z_1$ and
$\partial_1 \partial_{\bar 1} K_5$ also appear.
Thus in order to compute soft terms when a five--brane is present and
contributing to supersymmetry breaking we have to input these values.
Fortunately, $z_1$ is in the range $(0,1)$ and, although
$K_5$ is not known,
since
it depends on $z_1$,
we expect
$\partial_1 K_5$, $\partial_1 \partial_{\bar 1} K_5 = {\cal O}(1)$.
So we can consider the following representative example:
$z_1=1/2$ and $\partial_1 K_5=\partial_1\partial_{\bar 1}K_5=1$.
Since, still we have to input the value of $b_O$, we choose the
interesting
example with $\beta_O=-2$ and $\beta_1=1$ which implies $b_O=-7/4$.
In this way, using (\ref{Besbis}) and (\ref{Besbis2}), $b_H$ is also fixed,
$b_H=5/4$, and from the result (\ref{cota1}) in case {\it i)} we know that the
whole range of allowed (negative) values of $e_O$ can be
analyzed.
\begin{figure}[htb]
\hspace{-0.7cm}
\epsfig{file= figure7.eps, width=16cm, height=8cm}
\vspace{-1.0cm}
\caption{$m/|M|$ versus $\theta$ for different values of
$e_O$ and
goldstino angle $\theta_1$
when one five--brane is
contributing to supersymmetry breaking.\label{susyfivecociente}}
\end{figure}
\begin{figure}[htb]
\hspace{-0.7cm}
\epsfig{file= figure8.eps, width=16cm, height=8cm}
\vspace{-1.0cm}
\caption{(a) Soft parameters in units of $m_{3/2}$ and (b)
$m/|M|$ versus $e_O$,
when only the five--brane contributes to supersymmetry breaking with
$\theta_1=\pi/2$.\label{susyonlyfivesoft}}
\end{figure}
Taking
into account again the current experimental limits, we will assume
$V_0=0$
and
$\gamma_S=\gamma_T=\gamma_1=0\ ({\rm mod}\ \, \pi)$. More specifically
we will set $\gamma_S$, $\gamma_T$ and $\gamma_1$ to zero and allow
$\theta$ and $\theta_1$ to vary in a range $[0,2\pi)$. We show in
Fig.~\ref{susyfivesoft}
the dependence on $\theta$ of the soft terms in units of the
gravitino mass for some values of $e_O$ and $\theta_1$.
Due to the contribution of
$\theta_1$ to the soft terms (\ref{una}) the shift
$\theta\rightarrow \theta + \pi$ does not imply, as in
Figs.~\ref{nonstandardsoft} and \ref{nonsusyfivesoft},
$m\rightarrow m$, $M\rightarrow -M$ and $A\rightarrow -A$.
However, this is still true for $\theta_1\rightarrow \theta_1 + \pi$.
Besides, under the shifts $\theta_1\rightarrow \pi - \theta_1$
and $\theta\rightarrow \theta + \pi$, $m$, $M$ and $A$ remain
invariant.
Therefore from the analysis of the figures corresponding to
$\theta_1\in [0,\pi/2]$ the rest of the figures can easily be deduced.
In particular, in Fig.~\ref{susyfivesoft} we plot the values of $\theta_1$,
$\pi/4$ and $\pi/3$.
For a fixed
value of $\theta_1$, as in the case
without five--branes shown in
Fig.~\ref{nonstandardsoft} (with $\epsilon_O$ instead of $e_O$),
the smaller the value of $e_O$ the larger the range of soft terms becomes.
However,
unlike Fig.~\ref{nonstandardsoft},
in Fig.~\ref{susyfivesoft} we see a remarkable fact: scalar masses
larger
than gaugino masses can easily be obtained. This happens not only for narrow
ranges
of $\theta$ but even for the whole range (see the figure
with $e_O=0$ and $\theta_1=\pi/3$). We can see this in more detail in
Fig.~\ref{susyfivecociente} where $m/|M|$ versus $\theta$ is plotted.
For example, $m/|M|>1$
can be obtained, for any value of $\theta$, for values
of $e_O$ close to zero in the case $\theta_1=\pi/3$. Moreover, larger values
of
$\theta_1$ allow larger ranges of $e_O$ with
$m/|M|>1$ for any value of
$\theta$.
For example, in the limiting case where supersymmetry is only broken
by the $F$--term of the five--brane $F^1$, i.e. $\theta_1=\pi/2$,
$m/|M|>1$ for $e_O>-0.65$. This case is shown in Fig.~\ref{susyonlyfivesoft},
where the soft terms and $m/|M|$ versus $e_O$ are plotted. Of course,
the figures are independent on $\theta$. Let us remark that
scalar masses larger than gaugino masses are not easy to obtain,
as discussed below (\ref{gluinos1}), in the
weakly--coupled heterotic string.
It is worth noticing that the sum rule studied above,
$3m^2=M^2$ for $\theta=\pi/6$ (and $\pi/6+\pi$) is no longer true
when five--branes contributing to supersymmetry breaking are
present.
\begin{figure}[htb]
\hspace{-1.0cm}
\epsfig{file= figure9.eps, width=16cm, height=16cm}
\vspace{-1.0cm}
\caption{The same as Fig.~\ref{susyfivesoft} but for positive values
of $e_O$.\label{susyfivesoft+}}
\end{figure}
Finally, let us mention that a different value for $\partial_1 K_5$
that
the one considered here, will only modify the expression of the $A$
parameter, since it is the only one which depends on that derivative.
On the other hand, one can check that
a different value for $\partial_1\partial_{\bar 1} K_5$
will not modify the qualitative results concerning gaugino and scalar
masses.
Notice also that $\partial_1\partial_{\bar 1} K_5$ appears
always in the denominator in the expressions of the soft terms.
So, the larger the value of $\partial_1\partial_{\bar 1} K_5$
the smaller the deviation with respect to the case without five--branes
becomes.
One can also check that
different values for $z_1$, $\beta_O$ and $\beta_H$
will not modify the qualitative results concerning gaugino and scalar
masses.
\begin{figure}[htb]
\hspace{-0.7cm}
\epsfig{file= figure10.eps, width=16cm, height=8cm}
\vspace{-1.0cm}
\caption{The same as Fig.~\ref{susyfivecociente}
but for positive values of $e_O$.\label{susyfivecociente+}}
\end{figure}
\begin{figure}[htb]
\hspace{-0.7cm}
\epsfig{file= figure11.eps, width=16cm, height=8cm}
\vspace{-1.0cm}
\caption{The same as Fig.~\ref{susyonlyfivesoft}
but for positive values of $e_O$.\label{susyonlyfivesoft+}}
\end{figure}
We have not analyzed yet positive values of $e_O$ in the presence of
a five--brane contributing to supersymmetry breaking. In order to carry
it out we choose now $\beta_O=1$ and $\beta_1=-2$ which implies
$b_O=b_H=1/2$. From (\ref{cota2}) in case {\it ii)} we deduce that the
whole range of allowed (positive) values of $e_O$ can be studied. We
show in Fig.~\ref{susyfivesoft+} the soft terms for the values $e_O=1/3, 3/5$.
For a fixed
value of $\theta_1$, the
larger
the value of $e_O$ the larger the range of soft terms becomes, unlike
the case without five--branes (see Fig.~1 in \cite{Mio}).
On the other hand, scalar masses larger than gaugino masses can also
be easily obtained as in the case with negative values of $e_O$
studied above. This is plotted in more detail in
Fig.~\ref{susyfivecociente+}
for different
values of $e_O$. For example, for $e_O=1/3$ and $\theta_1=\pi/3$,
$\theta\approx 3\pi/2$ one obtains $m/|M|\approx 10$.
Using (\ref{masas}) this result implies a relation of the type
(\ref{gluinos2}),
$m_{\tilde l}\approx m_{\tilde q}\approx 3.5 M_{\tilde g}$.
In Fig.~\ref{susyonlyfivesoft+} we show the limiting case where
supersymmetry
is only broken by the $F$--term of the five--brane.
\section{Conclusions}
In the present paper we have tried to perform a systematic
analysis
of the soft supersymmetry breaking terms arising in a Calabi--Yau
compactification
of the heterotic M--theory, as well as a detailed study of the
different scales of the theory.
Since, as discussed in the introduction,
Calabi--Yau manifolds with only
one K\"ahler modulus $T$ are very interesting from the phenomenological
point of view, not only because of their simplicity but also because
they might give rise to three--family models
with an improvement with respect to the
non--universality problem, we have concentrated on these spaces.
The soft
terms in the standard and non--standard embedding cases depend
explicitly on
the
gravitino mass $m_{3/2}$, the goldstino angle $\theta$ and the
parameter $\epsilon_O$ (see (\ref{epsilonO})).
The only difference between both cases is the range of values where
$\epsilon_O$ is valid. $-1<\epsilon_O<1$ in the latter and $0<\epsilon_O<1$
in the former. This will give rise to different patterns of soft
terms.
In particular, scalar masses larger than gaugino masses in the
non--standard
embedding case are allowed (see Fig.~\ref{nonstandardcociente}),
for narrow ranges of $\theta$, unlike
the standard embedding situation.
This has obvious implications for low--energy ($\approx M_W$)
phenomenology, as discussed in (\ref{gluinos1}) and (\ref{gluinos2}).
The presence of non--perturbative objects as five--branes in the vacuum
modifies the previous analysis substantially. Even if the five--branes
do not contribute to supersymmetry breaking
the soft terms are modified. Basically, the soft terms are given
by the same formulae than in the standard and non--standard embedding
but with a new parameter $e_O$ (instead of $\epsilon_O$),
see (\ref{Es}), which is valid not only for $-1<e_O<1$ but also
for $e_O\geq 1$. However, although the pattern of soft terms is different,
scalar masses larger than gaugino masses are not possible
(see Fig.~\ref{nonsusyfivecociente})
as in the
standard embedding situation. Other parameters appear when we allow
the
five--branes to contribute to supersymmetry breaking. In particular,
at least, a new goldstino angle $\theta_1$ must be included in the
computation of soft terms. In this way, scalar masses larger than
gaugino
masses can be obtained in a natural way (see
Figs.~\ref{susyfivecociente}
and \ref{susyfivecociente+}).
Depending on the values of
$e_O$ and $\theta_1$, scalars could be heavier than gauginos
even in the whole range of $\theta$.
As discussed below (\ref{gluinos1}) this might be possible in the
weakly--coupled heterotic string only for $\sin\theta\rightarrow 0$.
Concerning the scales of the theory, we have discussed in detail the
relations between the eleven--dimensional Planck mass, the
Calabi--Yau compactification
scale and the orbifold scale, taking into account higher order
corrections
to the formulae. Identifying
the compactification scale with the GUT scale, it is easier
to obtain the phenomenologically favored value,
$M_{GUT} \approx 3\times 10^{16}$ GeV, in the
non--standard embedding than in the standard one
(see Fig.~\ref{scales1}). On the other hand, to lower this scale
(and therefore the eleven--dimensional Planck scale which is around two times
bigger)
to intermediate values $\approx 10^{11}$ GeV or $1$ TeV values or
to obtain the radius of the orbifold as large as a millimetre
is in principle
possible in some special limits. In particular, in the non--standard
embedding when $\epsilon_O\rightarrow -1$ and also in the case
$\beta_O=0$ (see Fig.~\ref{scales2}).
However, the necessity of a fine--tuning in the former
and the existence of a hierarchy problem in the latter
render these possibilities
unnatural.
In the presence of five--branes, $M_{GUT}$ can be obtained more
easily (see Fig.~\ref{scales1five}a).
Although new possibilities arise in order to lower this scale,
in particular when $e_O$ is very large (see Fig.~\ref{scales1five}b),
again at the cost of introducing a hughe hierarchy problem.
\bigskip
\noindent {\bf Note added}
\noindent As this manuscript was prepared, refs. \cite{Lii} and \cite{Kubo}
appeared. The former discusses some of the issues presented in this
work,
particularly the scenario of subsection 2.3.
The latter also discusses soft terms in the presence of five--branes,
however, its numerical analysis concentrates on the dilaton limit with
$0<e_O<2/3$.
\bigskip
\noindent {\bf Acknowledgments}
\noindent The work of D.G. Cerde\~no has been supported by a Universidad
Aut\'onoma de Madrid grant.
The work of C. Mu\~noz has been supported
in part by the CICYT, under contract AEN97-1678-E, and
the European Union, under contract ERBFMRX CT96 0090.
|
\section{Introduction}
Realization of Bose-Einstein condensation (BEC) in trapped atomic
gases~\cite{Anderson,Bradley,Davis,Fried} has created an interdisciplinary
subfield of physics called coherent matter wave.
Developments of optical imaging techniques such as absorption imaging and
a nondestructive {\it in situ} method~\cite{Andrews96} have enabled us to
observe interference between matter waves~\cite{Andrews97} and dynamics
of BEC~\cite{AndrewsPRL,Hall}.
In a remarkable recent experiment, Hau {\it et al.}~\cite{Hau}
demonstrated that it is possible to make photons propagate through BEC at
a speed of 17 m/s, which suggests that the index of refraction of BEC is
14 orders of magnitude greater than that of a glass fiber.
In the present paper we predict yet another unique feature of BEC
concerning correlated photon scattering.
Various aspects of the optical response of BEC have been
predicted~\cite{Svistunov,Politzer,Lewen93,Java94,Lewen94,You94,Java95,You95,Morice,JavaRuo95,You96}:
a broad resonance~\cite{Java94,You94}, an extra structure in the
spectrum~\cite{Java95,JavaRuo95}, response to intense short
pulses~\cite{Lewen93,You95}, non-Lorentzian line shapes~\cite{You94,You96},
and a refractive index of a homogeneous gas~\cite{Morice}.
In the above work, kinetic motion of atoms is either ignored or not
explicitly taken into account because its time scale is usually much
slower than that of optical processes.
However, optical signatures of BEC are expected to appear also in the time
scale of kinetic motion, since the characteristic feature of BEC is the
degeneracy of the kinetic degrees of freedom as well as the atomic
internal degree of freedom.
The aim of the present paper is to study correlations of photons scattered
from trapped BEC that is irradiated by a weak far-off-resonant laser
field, and to show that they exhibit some characteristic features
associated with kinetic motion of the atoms in the trap.
By scattering a photon, the condensate suffers a recoil, which affects the
scattering of the subsequent photons.
As a consequence a correlation arises between a pair of photons with
respect to their scattering angles and time intervals.
We show that the presence of the condensate enhances the probability of a
pair of photons to be scattered into either the same direction or
symmetrical directions about laser propagation, and the enhancement occurs
periodically with respect to the time interval, where the period is set
by the frequency of the trap.
The interaction between atoms cannot be ignored, for the profile of the
condensate is significantly altered even for a few hundred
atoms~\cite{Ruprecht,Dalfovo96}.
While in Ref.~\cite{You96} effects of the atom-atom interaction on the
photon scattering are taken into account only through the atomic density
profile, the dynamical effect of the interaction on the optical response
has not been discussed.
We take into account the atom-atom interaction using the Bogoliubov
approximation~\cite{Bogoliubov}, and show that the correlations between
the scattered photons are deteriorated by the interaction.
This paper is organized as follows.
Section \ref{s:source} formulates the problem and derives an expression of
the electric-field operator far from BEC irradiated by a laser field.
Section \ref{s:noninteraction} studies coherence properties of scattered
photons.
Section~\ref{s:interaction} discusses effects of the atom-atom
interaction on the scattering of photons, and Sec.~\ref{s:conclusion}
concludes this paper.
\section{Formulation of the Problem}
\label{s:source}
We consider a situation in which trapped atoms are irradiated by a laser
field which is sufficiently detuned from the resonant frequency of the
atoms so that each photon is not scattered by the atoms more than once.
Photons are scattered from the atoms and detected by photon counters.
Quantum-statistical properties of the scattered photons can be
studied through the electric-field operator $\hat {\bf E}({\bf r}, t)$ in
the Heisenberg representation.
We derive here an expression of $\hat {\bf E}({\bf r}, t)$ at the position
far away from the trapped atoms compared with the spread of their wave
functions.
The Hamiltonian for the photon field is given by
\begin{equation}
\hat H_{\rm f} = \sum_{{\bf k}\sigma} \hbar \omega_{\bf k} \hat a_{{\bf k}
\sigma}^\dagger \hat a_{{\bf k} \sigma},
\end{equation}
where $\hat a_{{\bf k} \sigma}^\dagger$ and $\hat a_{{\bf k} \sigma}$ are
the creation and annihilation operators of a photon with wave vector $\bf
k$ and polarization $\sigma$.
The electric-field operator is expanded in terms of the creation and
annihilation operators as
\begin{equation}
\hat{\bf E}({\bf r}) = i \sum_{{\bf k} \sigma} \sqrt{\frac{\hbar
\omega_k}{2 \varepsilon_0 V}} \left( \bbox{\epsilon}_{{\bf k} \sigma} e^{i
{\bf k} \cdot {\bf r}} \hat a_{{\bf k} \sigma} - \bbox{\epsilon}_{{\bf k}
\sigma}^* e^{-i {\bf k} \cdot {\bf r}} \hat a_{{\bf k} \sigma}^\dagger
\right),
\end{equation}
where $\bbox{\epsilon}_{{\bf k} \sigma}$ is the polarization vector of
photons.
The internal states of two-level atoms are described by the Hamiltonian
\begin{equation} \label{H_AI}
\hat H_{\rm ai} = \sum_{j = 1}^N \frac{\hbar \omega_A}{2} \hat
\sigma_{jz},
\end{equation}
where $N$ is the number of atoms, $\hbar \omega_A$ is the transition
energy of the internal levels, and the operator $\hat \sigma_{jz}$ is the
Pauli operator which takes on the eigenvalue 1 (or $-1$) when the $j$th
atom is in the excited (or ground) internal level.
The Hamiltonian corresponding to the kinetic degrees of freedom of the
atoms is given by
\begin{equation} \label{H_AK}
\hat H_{\rm ak} = \sum_{j = 1}^N \left[ -\frac{\hbar^2}{2m} \nabla_j^2 +
V_t({\bf R}_j) \right] + \sum_{j \neq j'}^N \frac{2 \pi \hbar^2 a}{m}
\delta({\bf R}_j - {\bf R}_{j'}),
\end{equation}
where $m$ is the mass of the atom and $V_t({\bf R})$ is the trap
potential.
The last term in Eq.~(\ref{H_AK}) describes the Fermi's contact type of
interaction between hard-core atoms, which is characterized by the s-wave
scattering length $a$.
We assume the electric-dipole interaction between the atoms and the photon
field which is described by the Hamiltonian
\begin{equation} \label{H_ED}
\hat H_{\rm ed} = -\sum_{j = 1}^N \hat{\bf D}_j \cdot \hat{\bf E}({\bf
R}_j).
\end{equation}
The dipole operator denoted as $\hat{\bf D}_j$ has the form
\begin{equation}
\hat{\bf D}_j \equiv {\bf d} \hat \sigma_{j+} + {\bf d}^* \hat
\sigma_{j-},
\end{equation}
where $\bf d$ is the electric-dipole matrix element, and $\hat
\sigma_{j+}$ and $\hat \sigma_{j-}$ are the raising and lowering operators
of the internal state of the $j$th atom.
The Hamiltonian for the atom-laser interaction is given by
\begin{eqnarray}
\hat H_L & = & \sum_{j = 1}^N \Bigl[ {\cal E}_L (\bbox{\epsilon}_L \cdot
{\bf d}) e^{-i (\omega_L t - {\bf k}_L \cdot {\bf R}_j)} \hat \sigma_{j+}
\nonumber \\
& & + {\cal E}_L^* (\bbox{\epsilon}_L \cdot {\bf d})^* e^{i (\omega_L t -
{\bf k}_L \cdot {\bf R}_j)} \hat \sigma_{j-} \Bigr],
\end{eqnarray}
where ${\cal E}_L$, $\bbox{\epsilon}_L$, $\omega_L$, and ${\bf k}_L$
denote the amplitude, the polarization vector, the frequency, and the wave
vector of the laser field, respectively.
Since the optical processes of the internal states of the atoms are much
faster than those of the kinetic motion, we may adiabatically eliminate the
dynamics of the internal state.
When the detuning $\delta \equiv \omega_L - \omega_A$ is much larger than
both the atom-laser field coupling and the collective linewidth $\Gamma$
of the atoms~\cite{Java94,You94}, i.e.,
\begin{equation}
|\delta| \gg \frac{|{\cal E}_L {\bf d}|}{\hbar}, \;\;\; |\delta| \gg
\Gamma,
\end{equation}
the probability that the atoms are in the excited state $|e \; \rangle$ is
much smaller than that in the ground state $|g \; \rangle$.
In this case, the atomic internal state may be assumed to be in the ground
state of the Hamiltonian $\hat H_{\rm ai} + \hat H_{L}$, which is given,
up to the first order in $|{\cal E}_L {\bf d}| / \hbar \delta$, by
\begin{equation} \label{groundstate}
\prod_{j = 1}^N \left[ e^{i(\omega_L t - {\bf k}_L \cdot {\bf R}_j) / 2}
|g \; \rangle_j + \frac{|{\cal E}_L {\bf d}|}{\hbar \delta} e^{-i(\omega_L
t - {\bf k}_L \cdot {\bf R}_j) / 2} |e \; \rangle_j \right].
\end{equation}
Taking the expectation value of $\hat H_{\rm ed}$ in Eq.~(\ref{H_ED})
with respect to the atomic internal state (\ref{groundstate}), we obtain
the effective Hamiltonian describing the interaction between photons and
kinetic motion of the atoms via the atomic internal levels as
\begin{eqnarray}
\hat H_{\rm ed}' & = & \sum_{j = 1}^N \frac{|{\cal E}_L {\bf d}|}{\hbar
\delta}
\Bigl[ e^{i(\omega_L t - {\bf k}_L \cdot {\bf R}_j)} {\bf d} \nonumber \\
& & +
e^{-i(\omega_L t - {\bf k}_L \cdot {\bf R}_j)} {\bf d}^* \Bigr] \cdot
\hat{\bf E}({\bf R}_j).
\end{eqnarray}
In the following discussions we will employ the effective Hamiltonian
$\hat H_{\rm eff} = \hat H_{\rm f} + \hat H_{\rm ak} + \hat H_{\rm ed}'$
to describe the dynamics of photons and kinetic motion of the atoms.
The equation of motion for the electric-field operator in the Heisenberg
representation is obtained from the Hamiltonian $\hat H_{\rm eff}$ as
\begin{equation} \label{eomE}
\left( \nabla^2 - \frac{\partial^2}{c^2 \partial t^2} \right)
\hat{\bf E}({\bf r}, t) = \frac{i}{\hbar} [ \hat H_{\rm ed}'(t), \nabla^2
\hat{\bf A}({\bf r}, t) ],
\end{equation}
where the vector potential is given by
\begin{eqnarray}
\hat{\bf A}({\bf r}, t) & = & \sum_{{\bf k} \sigma} \sqrt{\frac{\hbar}{2
\varepsilon_0 \omega_k V}} \Bigl[ e^{i {\bf k} \cdot {\bf r}}
\bbox{\epsilon}_{{\bf k} \sigma} \hat a_{{\bf k} \sigma}(t) \nonumber \\
& & + e^{-i {\bf
k} \cdot {\bf r}} \bbox{\epsilon}_{{\bf k} \sigma}^* \hat a_{{\bf k}
\sigma}^\dagger(t) \Bigr].
\end{eqnarray}
The equation of motion (\ref{eomE}) can be solved, giving
\begin{equation}
\hat{\bf E}({\bf r}, t) = \hat{\bf E}_{\rm vac}({\bf r}, t) + \hat{\bf
E}_{\rm s}({\bf r}, t),
\end{equation}
where $\hat{\bf E}_{\rm vac}({\bf r}, t) = e^{\frac{i}{\hbar} \hat H_{\rm
f} t} \hat{\bf E}({\bf r}) e^{-\frac{i}{\hbar} \hat H_{\rm f} t}$ is the
vacuum field and
\begin{eqnarray} \label{sourcef}
\hat{\bf E}_{\rm s}({\bf r}, t) & = & -\frac{i}{4 \pi \hbar} \int d{\bf
r}' dt' \frac{[ \hat H_{\rm ed}'(t'), \nabla^2 \hat{\bf A}({\bf r}', t')
]}{|{\bf r} - {\bf r}'|} \nonumber \\
& & \times \delta(t - t' - |{\bf r} - {\bf r}'| / c)
\end{eqnarray}
describes the source field scattered from the atoms.
Evaluating the commutator and carrying out the integral with respect to
$t'$, the positive frequency part of the source field (\ref{sourcef})
becomes
\begin{eqnarray} \label{sf1}
\hat{\bf E}_{\rm s}^{(+)}({\bf r}, t) & = & -\frac{1}{4 \pi \varepsilon_0 V}
\frac{|{\cal E}_L {\bf d}|}{\hbar \delta} \int d{\bf r}' \frac{e^{-i
\omega_L t_r}}{|{\bf r} - {\bf r}'|} \nonumber \\
& & \times \sum_{\bf k} {\bf k} \times ({\bf d}
\times {\bf k}) e^{-i {\bf k} \cdot {\bf r}'} \sum_{j = 1}^N e^{i ({\bf
k}_L + {\bf k}) \cdot {\bf R}_j(t_r)}, \nonumber \\
\end{eqnarray}
where $t_r \equiv t - |{\bf r} - {\bf r}'| / c$ and ${\bf R}_j(t_r) \equiv
e^{\frac{i}{\hbar} \hat H_{\rm eff} t_r} {\bf R}_j e^{-\frac{i}{\hbar}
\hat H_{\rm eff} t_r}$.
Since $|{\cal E}_L {\bf d}| / \hbar \delta \ll 1$, it is sufficient to
keep only the terms first order in $|{\cal E}_L {\bf d}| / \hbar \delta$,
which amounts to disregarding the multiple scattering and the
dipole-dipole interaction.
Consequently, we can make an approximation
\begin{equation} \label{rapp}
{\bf R}_j(t_r) \simeq e^{\frac{i}{\hbar} \hat H_{\rm ak} t} {\bf R}_j
e^{-\frac{i}{\hbar} \hat H_{\rm ak} t} \equiv {\bf R}_j(t),
\end{equation}
where we ignore the time $|{\bf r} - {\bf r}'| / c$ it takes photons to
propagate from the atoms to the detector as it is much smaller than the
time scale of the kinetic motion of the atoms.
The integrand of the source field (\ref{sf1}) contributes only when ${\bf
r}'$ falls within the spread of the atomic wave function.
Therefore, if we choose the origin of the coordinate system at the center
of the trap, the distance $|{\bf r}|$ from the atom to the detector is
much larger than $|{\bf r}'|$, and hence $t_r \simeq t - (r / c) \left( 1
- {\bf r} \cdot {\bf r}' / r^2 \right)$ and $1 / |{\bf r} - {\bf r}'|
\simeq 1 / r$.
We thus obtain
\begin{eqnarray} \label{sourceE}
\hat {\bf E}_{\rm s}^{(+)}({\bf r}, t) & = & -\frac{1}{4 \pi \varepsilon_0 r V}
\frac{|{\cal E}_L {\bf d}|}{\delta \hbar} e^{-i \omega_L t} \int d{\bf
r}' \sum_{\bf k} {\bf k} \times ({\bf d} \times {\bf k}) \nonumber \\
& & \times e^{ikr} e^{i (k_L
\tilde{\bf r} - {\bf k}_L) \cdot {\bf r}'} \sum_{j = 1}^N e^{i ({\bf k}_L
+ {\bf k}) \cdot {\bf R}_j(t)} \nonumber \\
& = & -\frac{k_L^2}{4 \pi \varepsilon_0 r} \frac{|{\cal E}_L {\bf
d}|}{\delta \hbar} {\bf n} \times ({\bf d} \times {\bf n}) e^{-i \omega_L
t + i k_L r} \nonumber \\
& & \times \sum_{j = 1}^N e^{i ({\bf k}_L - k_L {\bf n}) \cdot {\bf
R}_j(t)} \nonumber \\
& \equiv & {\bf F}({\bf r}, t) \sum_{j = 1}^N e^{i \Delta {\bf k}
\cdot {\bf R}_j(t)},
\end{eqnarray}
where ${\bf n} \equiv {\bf r} / |{\bf r}|$ and $\Delta {\bf k} \equiv {\bf
k}_L - k_L {\bf n}$.
The expression of the source-field operator (\ref{sourceE}) consists of
the factor ${\bf F}({\bf r}, t)$ which depends on the spatial
configuration of the system and the operator that transfers a recoil
momentum $\hbar \Delta {\bf k}$ to the atoms.
\section{First and Second-Order Coherence of photons scattered from a
noninteracting condensate}
\label{s:noninteraction}
In this section we consider an ideal Bose gas, where the atom-atom
interaction is absent.
We write the eigenfunctions of the Hamiltonian $\hat H_{\rm ak}$ as
$f_n({\bf R})$, where $n$ is an index for the eigenstates with $n = 0$
referring to the ground state.
It is convenient to discuss the problem in the second-quantized
formalism.
We expand the atomic field operator as
\begin{equation} \label{psi}
\hat \psi({\bf R}) = \sum_n f_n({\bf R}) \hat b_n,
\end{equation}
where $\hat b_n$ annihilates an atom in the $n$th eigenstate, and
satisfies the Bose commutation relation $[\hat b_n, \hat b_{n'}^\dagger]
= \delta_{nn'}$.
The Hamiltonian $\hat H_{\rm ak}$ can then be rewritten as
\begin{equation}
\hat H_{\rm ak} = \sum_n \hbar \omega_n \hat b_n^\dagger b_n,
\end{equation}
where $\hbar \omega_n$ is the $n$th eigenenergy.
The second-quantized form of the operator $\sum_{j = 1}^N e^{i \Delta {\bf
k} \cdot {\bf R}_j (t)}$ in Eq.~(\ref{sourceE}) is given by
\begin{eqnarray} \label{D}
\hat{\cal D}(\Delta {\bf k}, t) & \equiv & \int d{\bf R} \hat
\psi^\dagger({\bf R}, t) e^{i \Delta {\bf k} \cdot {\bf R}} \hat \psi({\bf
R}, t) \nonumber \\
& = & \sum_{nn'} \langle n' | e^{i \Delta {\bf k} \cdot {\bf R}} | n
\rangle \hat b_{n'}^\dagger \hat b_n e^{i (\omega_{n'} - \omega_n) t},
\end{eqnarray}
where $\hat\psi({\bf r}, t) \equiv e^{\frac{i}{\hbar} \hat H_{\rm ak} t}
\hat\psi({\bf R}) e^{-\frac{i}{\hbar} \hat H_{\rm ak} t}$, and the bracket
$\langle n' | e^{i \Delta {\bf k} \cdot {\bf R}} | n \rangle$ is a
shorthand notation of $\int d{\bf R} f_{n'}^*({\bf R}) e^{i \Delta {\bf k}
\cdot {\bf R}} f_n({\bf R})$.
The source-field operator (\ref{sourceE}) is expressed as $\hat{\bf
E}^{(+)}({\bf r}, t) = {\bf F}({\bf r}, t) \hat{\cal D}(\Delta {\bf k},
t)$.
At zero temperature all $N$ atoms are Bose-Einstein condensed in the
ground-state level of the trap potential.
We assume an isotropic harmonic trap $V_t({\bf R}) = m \omega^2
R^2 / 2$ for simplicity.
The expectation value of the electric field at the position $\bf r$ becomes
\begin{eqnarray} \label{Eavg}
\langle \hat{\bf E}^{(+)}({\bf r}, t) \rangle & = & {\bf F}({\bf r}, t) N
\langle 0 | e^{i \Delta {\bf k} \cdot {\bf R}} | 0 \rangle \nonumber \\
& = & {\bf F}({\bf r}, t) N e^{-|d \Delta {\bf k}|^2 / 4},
\end{eqnarray}
where $d \equiv \left( \hbar / m \omega \right)^{1/2}$ denotes the
characteristic length scale of the trap.
The exponential factor in Eq.~(\ref{Eavg}) is a decreasing function of the
scattering angle.
Suppose that we detect the number of scattered photons in a configuration
schematically illustrated in Fig.~\ref{f:setup}.
\begin{figure}[tb]
\begin{center}
\leavevmode\epsfxsize=86mm \epsfbox{fig1.ps}
\end{center}
\caption{
Schematic of a scattering experiment.
The trapped Bose-Einstein condensate (BEC) is irradiated by a weak and
far off-resonant laser.
The scattered photons are detected by the detectors set around the BEC.
The positions of the detectors are denoted as ${\bf r}_i$ and the
scattering angles as $\theta_i$ ($i = 1, 2$).
}
\label{f:setup}
\end{figure}
The position of the $i$th detector is denoted as ${\bf r}_i$ and the
scattering angle as measured from the direction of laser propagation is
denoted as $\theta_i$.
The number of photons counted by the detector is proportional to the
normally-ordered intensity of the field~\cite{Loudon}
\begin{equation}
I({\bf r}, t) \equiv \langle \hat{\bf E}^{(-)}({\bf r}, t) \cdot \hat{\bf
E}^{(+)}({\bf r}, t) \rangle.
\end{equation}
Substituting Eq.~(\ref{sourceE}) into this gives
\begin{equation} \label{I}
I({\bf r}, t) = |{\bf F}({\bf r}, t)|^2 N \left[ 1 + (N - 1)
e^{-|d \Delta {\bf k}|^2 / 2} \right].
\end{equation}
The first term on the right-hand side is proportional to the number of
atoms $N$ and the angular dependence lies solely in $|{\bf F}({\bf r},
t)|^2$.
The second term is proportional to $N(N - 1)$ and contains a
forward-scattering factor of $e^{-|d \Delta {\bf k}|^2 / 2}$.
In fact, in experiments of Ref.~\cite{Andrews96}, a forward scattering
that satisfies $|d \Delta {\bf k}| \lesssim 1$ was observed at the onset
of BEC.
The degree of first-order coherence is defined by~\cite{Loudon}
\begin{equation}
g^{(1)}({\bf r}_1, t_1; {\bf r}_2, t_2) \equiv \frac{\langle \hat{\bf
E}^{(-)}({\bf r}_1, t_1) \cdot \hat{\bf E}^{(+)}({\bf r}_2, t_2)
\rangle}{\sqrt{I({\bf r}_1, t_1) I({\bf r}_2, t_2)}}.
\end{equation}
Substituting Eq.~(\ref{sourceE}) into this gives
\begin{eqnarray} \label{g1}
& & |g^{(1)}({\bf r}_1, t_1; {\bf r}_2, t_2)| \nonumber \\
& = & \frac{e^{-|d \Delta {\bf
k}_1|^2 / 4 - |d \Delta {\bf k}_2|^2 / 4} \left| e^{d^2 \Delta {\bf k}_1
\cdot \Delta {\bf k}_2 e^{-i \omega (t_1 - t_2)} / 2} + N - 1
\right|}{\left [ 1 + (N - 1) e^{-|d \Delta {\bf k}_1|^2 / 2}
\right]^{\frac{1}{2}} \left[ 1 + (N - 1) e^{-|d \Delta {\bf k}_2|^2 / 2}
\right]^{\frac{1}{2}}}, \nonumber \\
\end{eqnarray}
where $\Delta {\bf k}_i \equiv {\bf k}_L - k_L {\bf r}_i / |{\bf r}_i|$
($i = 1$ and $2$).
When the number of atoms is sufficiently large to satisfy $N e^{-|d \Delta
{\bf k}_i|^2 / 2} \gg 1$ for $i = 1$ and $2$, the degree of the
first-order coherence becomes $|g^{(1)}({\bf r}_1, t_1; {\bf r}_2, t_2)|
\simeq 1$.
The scattered photon field is therefore first-order coherent
when the number of atoms in the condensate is sufficiently large.
On the other hand, when $N e^{-|d \Delta {\bf k}_i|^2 / 2} \ll 1$ for both
$i = 1$ and $2$, Eq.~(\ref{g1}) reduces to
\begin{equation} \label{g1r}
|g^{(1)}({\bf r}_1, t_1; {\bf r}_2, t_2)| \simeq \exp\left(-\frac{d^2}{4}
|\Delta {\bf k}_1 e^{-i \omega (t_1 - t_2)} - \Delta {\bf k}_2 |^2
\right).
\end{equation}
When $t_1 = t_2$, the coherence (\ref{g1r}) is lost exponentially with
increasing the difference in the scattering angles $|\Delta {\bf k}_1 -
\Delta {\bf k}_2|$.
This is because two photon paths corresponding to atoms kicked in
different directions do not interfere with each other.
The coherence is periodic with respect to the time interval $t_1 - t_2$
with the period of $2\pi \omega^{-1}$.
We note that the same expression as Eq.~(\ref{g1r}) can be obtained by
substituting $N = 1$ in Eq.~(\ref{g1}), and therefore the correlation
described above cannot be ascribed to BEC.
Figure~\ref{f:g1} shows the degree of first-order coherence for $t_1 =
t_2$, $|g^{(1)}({\bf r}_1, t; {\bf r}_2, t)|$, where $N = 10^3$, $d = 1
\mu{\rm m}$, and $2 \pi / k_L = 589 {\rm nm}$ (the 3s-3p transition of
${}^{23}{\rm Na}$) are assumed.
\begin{figure}[tb]
\begin{center}
\leavevmode\epsfxsize=86mm \epsfbox{fig2.ps}
\end{center}
\caption{
The angular dependence of the degree of first-order coherence
$g^{(1)}({\bf r}_1, t; {\bf r}_2, t)$ of photons scattered from $10^3$
BEC atoms.
The wavelength of the laser field $2 \pi / k_L = 589 {\rm nm}$ (the 3s-3p
transition of ${}^{23}{\rm Na}$) and the characteristic length of the
trap $d = 1 \mu{\rm m}$ are assumed.
The wave vectors ${\bf k}_1$ and ${\bf k}_2$ are assumed to be in the same
plane, and $\theta_1$ and $\theta_2$ denote their angles as measured from
${\bf k}_L$.
}
\label{f:g1}
\end{figure}
The wave vector ${\bf k}_1$ and ${\bf k}_2$ are assumed to be in the same
plane, and $\theta_1$ and $\theta_2$ denote the angles of ${\bf k}_1$ and
${\bf k}_2$ as measured from ${\bf k}_L$ (see Fig.~\ref{f:setup}).
One can see that $g^{(1)} \simeq 1$ for $|d \Delta {\bf k}| \lesssim 1$,
and therefore, the photons scattered in the forward directions have
first-order coherence.
There are peaks along $\theta_1 = \theta_2$, which can be seen from
Eq.~(\ref{g1r}).
The degree of second-order coherence~\cite{Loudon} is defined by
\begin{eqnarray} \label{g2}
& & g^{(2)}({\bf r}_1, t_1; {\bf r}_2, t_2) \equiv \nonumber \\
& & \frac{{\displaystyle
\sum_{\alpha, \beta = x, y, z}} \langle \hat E_\alpha^{(-)}({\bf r}_1,
t_1) \hat E_\beta^{(-)}({\bf r}_2, t_2) \hat E_\beta^{(+)}({\bf r}_2, t_2)
\hat E_\alpha^{(+)}({\bf r}_1, t_1) \rangle}{I({\bf r}_1, t_1) I({\bf
r}_2, t_2)}. \nonumber \\
\end{eqnarray}
Substituting Eq.~(\ref{sourceE}) into this yields
\begin{eqnarray} \label{g2_2}
& & g^{(2)}({\bf r}_1, t_1; {\bf r}_2, t_2) = \nonumber \\
& & \Bigl\{ N(N - 1)(N - 2)(N -
3) e^{-|d \Delta {\bf k}_1|^2 / 2 - |d \Delta {\bf k}_2|^2 / 2} \nonumber
\\
& & + N(N - 1)(N - 2) \Bigl[ 2 e^{-|d \Delta {\bf k}_1|^2 / 2 - |d \Delta
{\bf k}_2|^2 / 2} \nonumber \\
& & \times {\rm Re} \Bigl( e^{-d^2 \Delta {\bf k}_1 \cdot \Delta
{\bf k}_2 e^{i \omega (t_2 - t_1)} / 2} + e^{d^2 \Delta {\bf k}_1 \cdot
\Delta {\bf k}_2 e^{-i \omega (t_2 - t_1)} / 2} \Bigr) \nonumber \\
& & + e^{-|d \Delta {\bf k}_1|^2 / 2} + e^{-|d \Delta {\bf k}_2|^2 / 2}
\Bigr] \nonumber \\
& & + N(N - 1) \Bigl[ 1 + e^{-d^2 |\Delta {\bf k}_1 / 2 - \Delta {\bf k}_2
e^{i \omega (t_2 - t_1)}|^2} \nonumber \\
& & + e^{-d^2 |\Delta {\bf k}_1 + \Delta {\bf
k}_2 e^{i \omega (t_2 - t_1)}|^2 / 2}
+ 2 e^{-|d \Delta {\bf k}_1|^2 / 2} \nonumber \\
& & + 2 e^{-|d \Delta {\bf k}_2|^2 / 2}
\cos\bigl( d^2 \Delta {\bf k}_1 \cdot \Delta {\bf k}_2 \sin \omega (t_2
- t_1) \bigr) \Bigr] + N \Bigr\} \nonumber \\
& & \Bigr/ \left\{ N^2 \left[ 1 + (N - 1) e^{-|d \Delta {\bf k}_1|^2 / 2}
\right] \left[ 1 + (N - 1) e^{-|d \Delta {\bf k}_2|^2 / 2} \right]
\right\}. \nonumber \\
\end{eqnarray}
If the number of atoms is sufficiently large to satisfy $N e^{-|d \Delta
{\bf k}_i|^2 / 2} \gg 1$ for both $i = 1$ and $2$, the degree of
second-order coherence becomes $g^{(2)}({\bf r}_1, t_1; {\bf r}_2, t_2)
\simeq 1$ as well as $|g^{(1)}({\bf r}_1, t_1; {\bf r}_2, t_2)| \simeq 1$.
In general, $r$th-order coherence~\cite{Loudon} becomes unity for all $r$,
and therefore, in this large $N$ limit the scattered photons are coherent
in all orders.
On the other hand, if $N e^{-|d \Delta {\bf k}_i|^2 / 2} \ll 1$ for both
$i = 1$ and $2$, Eq.~(\ref{g2_2}) reduces to
\begin{eqnarray} \label{g2r}
g^{(2)}({\bf r}_1, t_1; {\bf r}_2, t_2) & \simeq & 1 + e^{-d^2 |\Delta {\bf
k}_1 - \Delta {\bf k}_2 e^{i \omega (t_2 - t_1)}|^2 / 2} \nonumber \\
& & + e^{-d^2 |\Delta
{\bf k}_1 + \Delta {\bf k}_2 e^{i \omega (t_2 - t_1)}|^2 / 2}.
\end{eqnarray}
When $t_1 = t_2$, Eq.~(\ref{g2r}) becomes $g^{(2)} \simeq 2$ for either
$\Delta {\bf k}_1 \simeq \Delta {\bf k}_2$ or $\Delta {\bf k}_1 \simeq
-\Delta {\bf k}_2$.
This means that a pair of photons tends to be scattered either in
the same direction or in the symmetrical directions with respect to
laser propagation.
These correlation in scattered photons can be interpreted as follows.
In the case of $\Delta {\bf k}_1 \simeq \Delta {\bf k}_2$, the first
photon kicks an atom from the condensate to the state $e^{i \Delta {\bf
k}_1 \cdot {\bf R}} f_0({\bf R})$, and so does the subsequent photon,
where $f_0({\bf R})$ is the ground state of the trap potential.
Consequently, the Bose-enhancement factor of this process is
\begin{equation}
\langle (N - 2)_0, 2_{\Delta {\bf k}_1} | \hat b_{\Delta {\bf
k}_2}^\dagger \hat b_0 \hat b_{\Delta {\bf k}_1}^\dagger \hat b_0 | N_0,
0_{\Delta {\bf k}_1} \rangle \simeq \sqrt{2 N(N - 1)},
\end{equation}
where $\hat b_{\bf k}^\dagger \equiv \sum_n \langle n | e^{i {\bf k} \cdot
{\bf R}} | 0 \rangle \hat b_n^\dagger$ and $|m_{\bf k} \rangle \equiv \hat
b_{\bf k}^{\dagger m} / \sqrt{m!} |0 \rangle$.
Therefore the probability that this process occurs is enhanced by a factor
of two.
In the case of $\Delta {\bf k}_1 \simeq -\Delta {\bf k}_2$, there is a
process in which the subsequent photon kicks back the atom excited by
the first photon to the condensate, in addition to the process in which
two atoms are excited to the states $e^{i \Delta {\bf k}_1 \cdot {\bf R}}
f_0({\bf R})$ and $e^{i \Delta {\bf k}_2 \cdot {\bf R}} f_0({\bf R})$.
The Bose-enhancement factor of each process is
\begin{equation}
\langle N_0, 0_{\Delta {\bf k}_1} | \hat b_0^\dagger \hat b_{\Delta {\bf
k}_1} \hat b_{\Delta {\bf k}_1}^\dagger \hat b_0 | N_0, 0_{\Delta {\bf
k}_1} \rangle \simeq N
\end{equation}
and
\begin{eqnarray}
& & \langle (N - 2)_0, 1_{\Delta {\bf k}_1}, 1_{\Delta {\bf k}_2} | \hat
b_0^\dagger \hat b_{\Delta {\bf k}_1} \hat b_{\Delta {\bf k}_1}^\dagger
\hat b_0 | N_0, 0_{\Delta {\bf k}_1}, 0_{\Delta {\bf k}_2} \rangle
\nonumber \\
& & \simeq \sqrt{N(N - 1)},
\end{eqnarray}
and hence the probability is enhanced by a factor of two for large $N$.
Thus the above correlations are due to the quantum-statistical effect of
bosons.
In fact, the degree of second-order coherence reduces to $g^{(2)} = 1$
when $N = 1$, and a photon pair scattered from a single atom shows no
correlation.
We note that the presence of the condensate is required for the above
discussion.
The process in which the condensate participates is accompanied by the
Bose-enhancement factor of $\sqrt{N}$, and the probability that the
process occurs is $N$ times greater than that of the process between other
levels.
Therefore, scattering of photons by the condensate can clearly be
distinguished from that by the non-condensed atoms.
When the condensate is not present and the atoms are thermally
distributed, occupation numbers of any states are smaller than one, and
Bose-enhancement effects cannot be observed.
The degree of second-order coherence (\ref{g2_2}) for $t_1 = t_2$ is shown
in Fig.~\ref{f:g2}, where $N = 10^3$ and $d = 1 \mu{\rm m}$, and ${\bf
k}_1$ and ${\bf k}_2$ are assumed to be on the same plane.
\begin{figure}[tb]
\begin{center}
\leavevmode\epsfxsize=86mm \epsfbox{fig3.ps}
\end{center}
\caption{
The angular dependence of the degree of second-order coherence
$g^{(2)}({\bf r}_1, t; {\bf r}_2, t)$ of photons scattered from $10^3$
BEC atoms.
The wavelength of the laser field $2 \pi / k_L = 589 {\rm nm}$ (the 3s-3p
transition of ${}^{23}{\rm Na}$) and the characteristic length of the
trap $d = 1 \mu{\rm m}$ are assumed.
The wave vectors ${\bf k}_1$ and ${\bf k}_2$ are assumed to be in the same
plane, and $\theta_1$ and $\theta_2$ denote their angles as measured from
${\bf k}_L$.
}
\label{f:g2}
\end{figure}
We find that $g^{(2)} \simeq 1$ when the scattering angles are small, and
$g^{(2)}$ grows with increasing $\theta_1$.
While the peak of $g^{(2)}$ at the same scattering angle $\theta_1 =
\theta_2$ is $g^{(2)} \simeq 2$ for large $\theta_1$, the peak at the
opposite angle $\theta_1 = -\theta_2$ first grows and then decreases to
$g^{(2)} \simeq 1$ as $\theta_1$ increases.
This is due to the fact that $\Delta {\bf k}_1 = -\Delta {\bf k}_2$ is not
compatible with the energy conservation $|{\bf k}_1| = |{\bf k}_2| =
\omega_L / c$ (the recoil energy is much smaller than the photon energy),
and the third term in Eq.~(\ref{g2r}) cannot become unity even for
$\theta_1 = -\theta_2$.
In other words, the energy conservation prohibits the second photon from
exactly kicking the atom back to the condensate.
Figure \ref{f:g2t} shows the time dependence of the degree of second-order
coherence (\ref{g2_2}).
\begin{figure}
\begin{center}
\leavevmode\epsfxsize=86mm \epsfbox{fig4.ps}
\end{center}
\caption{
The time and angular dependence of the degree of second-order coherence
$g^{(2)}({\bf r}_1, t_1; {\bf r}_2, t_2)$ of photons scattered from $10^3$
BEC atoms.
The scattering angle of the first photon is fixed at $\theta_1 = \pi /
12$.
The wavelength of the laser field $2 \pi / k_L = 589 {\rm nm}$ (the 3s-3p
transition of ${}^{23}{\rm Na}$) and the characteristic length of the
trap $d = 1 \mu{\rm m}$ are assumed.
The wave vectors ${\bf k}_1$ and ${\bf k}_2$ are assumed to be in the same
plane, and $\theta_1$ and $\theta_2$ denote their angles as measured from
${\bf k}_L$.
}
\label{f:g2t}
\end{figure}
The parameters are the same as those in Fig.~\ref{f:g2}, and the
scattering angle of the first photon is fixed at $\theta_1 = \pi / 12$.
We find that the second-order coherence is periodic in $t_2 - t_1$ with
the period given by $\pi \omega^{-1}$.
This periodic behavior reflects the fact that a recoil atom oscillates in
the trap.
The coherence disappears when the atom is kicked out of the condensate,
and revives when the atom returns to the condensate.
We note that $g^{(2)}$ goes to below unity before and after the appearance
of the peaks.
This indicates that photons are suppressed to be scattered into $\theta_2
\simeq \pm \theta_1$ in these time intervals.
This comes from the terms proportional to $N(N - 1)(N - 2)$ in the
numerator of Eq.~(\ref{g2_2}), which shows destructive interference of
probability amplitudes of two-photon processes in which one photon kicks
an atom and another photon does not disturb the atoms.
The degree of first-order coherence $g^{(1)}$ can be measured by
superposing the photon field in the directions of ${\bf k}_1$ and ${\bf
k}_2$ with a delay time $t_2 - t_1$ in the path of the ${\bf k}_1$
photon.
Taking a delay of order $\omega^{-1} \sim 1$ ms while retaining the
coherence might be feasible by reducing the speed of light using an
ultracold atomic gas~\cite{Hau}.
On the other hand, a measurement of $g^{(2)}$ does not require the
coherent delay, since it measures a correlation between detected photons.
\section{Effects of atom-atom interactions}
\label{s:interaction}
In this section, we take into account the atom-atom interaction on the
coherence properties of photons scattered from BEC within the Bogoliubov
approximation~\cite{Bogoliubov}.
The second-quantized form of the kinetic part of the Hamiltonian is given
by
\begin{eqnarray} \label{secondH}
\hat K & = & \int d{\bf R} \hat \psi^\dagger({\bf R}) \left[
-\frac{\hbar^2}{2m} \nabla^2 + V_t({\bf R}) - \mu \right] \hat \psi({\bf
R}) \nonumber \\
& & + \frac{U_0}{2} \int d{\bf R} \hat \psi^\dagger({\bf R}) \hat
\psi^\dagger({\bf R}) \hat \psi({\bf R}) \hat \psi({\bf R}),
\end{eqnarray}
where $\hat \psi({\bf R})$ is the field operator of the atoms (\ref{psi})
and $U_0 \equiv 4 \pi \hbar^2 a / m$, $a$ being the s-wave scattering
length.
In the Bogoliubov approximation, the atomic field operator is
divided into a mean field part and fluctuations from it as
\begin{equation} \label{psiB}
\hat \psi({\bf R}) = \sqrt{N} \psi_g({\bf R}) + \delta \hat \psi({\bf
R}),
\end{equation}
where $\psi_g({\bf R})$ is the normalized single-particle wave function
which obeys the Gross-Pitaevskii (GP) equation~\cite{GP}
\begin{equation} \label{GP}
\left[ -\frac{\hbar^2}{2m} \nabla^2 + V_t({\bf R}) + U_0 N |\psi({\bf
R})|^2 \right] \psi({\bf R}) = \mu \psi({\bf R}),
\end{equation}
where $\mu$ is determined so that the solution of Eq.~(\ref{GP}) satisfies
the normalization condition $\int d{\bf R} |\psi({\bf R})|^2 = 1$.
Equation (\ref{GP}) guarantees that if we substitute Eq.~(\ref{psiB}) into
Eq.~(\ref{secondH}), terms linear in $\delta \hat\psi$ or
$\delta \hat\psi^\dagger$ in Eq.~(\ref{secondH}) vanish identically.
Because the interaction between atoms is weak, we may expect only small
deviations from the mean field at zero temperature, so that we may keep
only terms up to quadratic in $\delta \hat\psi$ and $\delta
\hat\psi^\dagger$:
\begin{eqnarray} \label{H2}
\hat K & \simeq &
N \int d{\bf R} \psi_g^*({\bf R}) \biggl[
-\frac{\hbar^2}{2m} \nabla^2 + V_t({\bf R}) - \mu \nonumber \\
& & + \frac{U_0 N}{2}
|\psi({\bf R})|^2 \biggr] \psi_g({\bf R}) \nonumber \\
& & + \int d{\bf R} \biggl[ \delta \hat\psi^\dagger({\bf R}) \biggl(
-\frac{\hbar^2}{2m} \nabla^2 + V_t({\bf R}) - \mu \nonumber \\
& & + 2 U_0 N |\psi({\bf
R})|^2 \biggr) \delta \hat\psi({\bf R}) \nonumber \\
& & + U_0 N \psi_g^{*2}({\bf R})
\delta \hat\psi^2({\bf R}) + U_0 N \psi_g^2({\bf R}) \delta
\hat\psi^{\dagger 2}({\bf R}) \biggr]
\end{eqnarray}
To diagonalize the quadratic terms, we write $\delta \hat \psi({\bf R})$
as
\begin{equation}
\delta \hat \psi({\bf R}) = \sum_\lambda \left[
u_\lambda({\bf R}) \hat \beta_\lambda + v_\lambda^*({\bf R}) \hat
\beta_\lambda^\dagger \right].
\end{equation}
Requiring the coefficients of the terms $\hat \beta_\lambda^2$ and $\hat
\beta_\lambda^{\dagger 2}$ to vanish, we find that $u_\lambda({\bf R})$
and $v_\lambda({\bf R})$ should satisfy
\begin{mathletters} \label{Beq}
\begin{eqnarray}
& & \left[ -\frac{\hbar^2}{2m} \nabla^2 + V_t({\bf R}) - \mu + 2 U_0 N
|\psi_g({\bf R})|^2 \right] u_\lambda({\bf R}) \nonumber \\
& & + U_0 N \psi_g({\bf R})^2
v_\lambda({\bf R}) = \hbar \omega_\lambda u_\lambda({\bf R}), \\
& & \left[ -\frac{\hbar^2}{2m} \nabla^2 + V_t({\bf R}) - \mu + 2 U_0 N
|\psi_g({\bf R})|^2 \right] v_\lambda({\bf R}) \nonumber \\
& & + U_0 N \psi_g^{*2}({\bf
R}) u_\lambda({\bf R}) = -\hbar \omega_\lambda v_\lambda({\bf R}).
\end{eqnarray}
\end{mathletters}
The quadratic part of the Hamiltonian (\ref{H2}) can then be diagonalized
as
\begin{equation} \label{HB}
\hat K = \sum_\lambda \hbar \omega_\lambda \hat \beta_\lambda^\dagger
\hat \beta_\lambda,
\end{equation}
where $\hat \beta_\lambda^\dagger$, $\hat \beta_\lambda$ are the creation
and annihilation operators of the Bogoliubov quasiparticles and $\hbar
\omega_\lambda$ denotes the energy of elementary excitations.
The constant terms are irrelevant to later discussions and are therefore
omitted in Eq.~(\ref{HB}).
The second quantized form of the operator $\sum_{j = 1}^N e^{i \Delta {\bf
k} \cdot {\bf R}_j(t)}$ that gives the momentum to an atom reads in the
Bogoliubov approximation as
\begin{equation} \label{DB}
\hat{\cal D}_B(\Delta {\bf k}, t) \equiv \int d{\bf R} \hat
\psi^\dagger({\bf R}, t) e^{i \Delta {\bf k} \cdot {\bf R}} \hat
\psi({\bf R}, t),
\end{equation}
where $\hat \psi({\bf R}, t) = e^{\frac{i}{\hbar} \hat K t} \hat
\psi({\bf R}) e^{-\frac{i}{\hbar} \hat K t}$.
Because $|u_\lambda({\bf R})| \gg |v_\lambda({\bf R})|$~\cite{Dalfovo97},
we may neglect $v_\lambda({\bf R})$ in Eq.~(\ref{psiB}).
The operator (\ref{DB}) then becomes
\begin{eqnarray} \label{DB2}
\hat{\cal D}_B(\Delta {\bf k}, t) & = & N \langle \psi_g | e^{i \Delta
{\bf k} \cdot {\bf R}} | \psi_g \rangle \nonumber \\
& & + \sqrt{N} \sum_\lambda \Bigl(
\langle \psi_g | e^{i \Delta {\bf k} \cdot {\bf R}} | u_\lambda \rangle
\hat \beta_\lambda e^{-i \omega_\lambda t} \nonumber \\
& & + \langle u_\lambda |
e^{i \Delta {\bf k} \cdot {\bf R}} | \psi_g \rangle \hat
\beta_\lambda^\dagger e^{i \omega_\lambda t} \Bigr) \nonumber \\
& & + \sum_{\lambda \mu} \langle u_\lambda | e^{i \Delta {\bf k} \cdot
{\bf R}} | u_\mu \rangle \hat \beta_\lambda^\dagger \hat \beta_\mu
e^{i (\omega_\lambda - \omega_\mu) t},
\end{eqnarray}
where we use the shorthand notation as in Eq.~(\ref{D}).
In the Bogoliubov approximation, the orthonormality condition is expressed
as
$\langle u_\lambda | u_\mu \rangle - \langle v_\lambda | v_\mu \rangle =
\delta_{\lambda \mu}$, and the completeness relation is given by
$\sum_\lambda (|u_\lambda \rangle \langle u_\lambda | - |v_\lambda \rangle
\langle v_\lambda |) = 1 - |\psi_g \rangle \langle \psi_g |$.
In the present approximation of neglecting $v_\lambda$, they reduce to
$\langle u_\lambda | u_\mu \rangle = \delta_{\lambda \mu}$ and
$\sum_\lambda |u_\lambda \rangle \langle u_\lambda | + |\psi_g \rangle
\langle \psi_g | = 1$, respectively.
Using the operator (\ref{DB2}), the source-field operator is expressed as
$\hat{\bf E}_{\rm s}^{(+)}({\bf r}, t) = {\bf F}({\bf r}, t) \hat{\cal
D}_B(\Delta {\bf k}, t)$.
The expectation values of the electric field and the intensity are
calculated to be
\begin{equation}
\langle \hat{\bf E}^{(+)}({\bf r}, t) \rangle = {\bf F}({\bf r}, t) N
\langle \psi_g | e^{i \Delta {\bf k} \cdot {\bf R}} | \psi_g \rangle,
\end{equation}
\begin{equation}
I({\bf r}, t) = |{\bf F}({\bf r}, t)|^2 N \left( 1 + N |\langle \psi_g
| e^{i \Delta {\bf k} \cdot {\bf R}} | \psi_g \rangle|^2 \right).
\end{equation}
These expressions are different from the noninteracting counterparts
(\ref{Eavg}) and (\ref{I}) in that the expectation value with
respect to the ground state of the noninteracting atoms $\langle 0
| e^{i \Delta {\bf k} \cdot {\bf R}} | 0 \rangle$ is replaced by that of
the ground state of the GP equation $\langle \psi_g | e^{i \Delta {\bf k}
\cdot {\bf R}} | \psi_g \rangle$, and $N - 1$ by $N$.
The latter arises from the assumption of the Bogoliubov theory that
the ground state is a coherent state rather than a Fock state.
Although this is an artifact, the error is of order $1 / N$ and
negligible for $N \gg 1$.
The former reflects the fact that the ground-state wave function of
interacting atoms expands for the repulsive interaction $a > 0$, and
contracts for the attractive interaction $a < 0$, compared with that of
noninteracting atoms.
The degree of first-order coherence is calculated to be
\begin{eqnarray} \label{g1int}
& & |g^{(1)}({\bf r}_1, t_1; {\bf r}_2, t_2)| = \nonumber \\
& & \frac{\langle e^{-i \Delta
{\bf k}_1 \cdot {\bf R}} \hat U_B(t_1 - t_2) e^{i \Delta {\bf k}_2 \cdot
{\bf R}} \rangle_g + N \langle e^{-i \Delta {\bf k}_1 \cdot {\bf R}}
\rangle_g \langle e^{i \Delta {\bf k}_2 \cdot {\bf R}} \rangle_g}{\left( 1
+ N |\langle e^{i \Delta {\bf k}_1 \cdot {\bf R}} \rangle_g|^2
\right)^{\frac{1}{2}} \left( 1 + N |\langle e^{i \Delta {\bf k}_2 \cdot
{\bf R}} \rangle_g|^2 \right)^{\frac{1}{2}}}, \nonumber \\
\end{eqnarray}
where $\hat U_B(t) \equiv \sum_\lambda |u_\lambda \rangle e^{-i
\omega_\lambda t} \langle u_\lambda |$, and $\langle \cdots \rangle_g$
denotes the expectation value with respect to the ground state of the GP
equation (\ref{GP}).
To evaluate the expectation value that includes $\hat U_B$, we need to
find $u_\lambda$ which we obtain by numerically diagonalizing
Eqs.~(\ref{Beq}a) and (\ref{Beq}b) using the method discussed in
Ref.~\cite{Edwards}.
Figure~\ref{f:g1int} shows the degree of first-order coherence
(\ref{g1int}) for $t_1 = t_2$.
\begin{figure}
\begin{center}
\leavevmode\epsfxsize=86mm \epsfbox{fig5.ps}
\end{center}
\caption{
The angular dependence of the degree of first-order coherence
$g^{(1)}({\bf r}_1, t; {\bf r}_2, t)$ of photons scattered from $10^3$
interacting BEC atoms.
The s-wave scattering length is assumed to be $a = 2.75$ nm.
The wavelength of the laser field $2 \pi / k_L = 589 {\rm nm}$ (the 3s-3p
transition of ${}^{23}{\rm Na}$) and the characteristic length of the
trap $d = 1 \mu{\rm m}$ are assumed.
The wave vectors ${\bf k}_1$ and ${\bf k}_2$ are assumed to be in the same
plane, and $\theta_1$ and $\theta_2$ denote their angles as measured from
${\bf k}_L$.
}
\label{f:g1int}
\end{figure}
The parameters are taken to be $N = 10^3$, $d = 1 \mu{\rm m}$, and $a =
2.75$ nm.
The ground state of the GP equation (\ref{GP}) is obtained
numerically~\cite{Dalfovo96}.
The sharp dips in Fig.~\ref{f:g1int} correspond to the points at which the
sign of $g^{(1)}$ changes (note that Fig.~\ref{f:g1int} displays the
absolute value of $g^{(1)}$), while the curves in the noninteracting case
are smooth (see Fig.~\ref{f:g1}).
To understand the oscillatory behaviors of $g^{(1)}$, we compare in
Fig.~\ref{f:wf} the profile of the interacting ground-state
wave function with $N = 10^3$ with the corresponding noninteracting one.
\begin{figure}
\begin{center}
\leavevmode\epsfxsize=86mm \epsfbox{fig6.ps}
\end{center}
\caption{
The profiles of the ground-state wave functions.
The solid curve shows the wave function of interacting $10^3$ atoms which
is obtained by solving the GP equation (\protect\ref{GP}),
and the dashed curve shows the noninteracting one.
}
\label{f:wf}
\end{figure}
As we can see from the figure the density profile near the origin becomes
flat due to the repulsive interaction which gives rise to the oscillatory
behavior of $e^{i\Delta {\bf k} \cdot {\bf R}}$.
The validity of this interpretation can be confirmed analytically in the
limit of strong interaction $N a / d \gg 1$.
In this limit the kinetic energy may be neglected in comparison with the
trap potential and the interaction energy (Thomas-Fermi approximation),
and the wave function is given by
\begin{equation}
\psi_{\rm TF}(R) = \sqrt{\frac{15}{8 \pi R_0^5} (R_0^2 - R^2)},
\end{equation}
where $R_0 \equiv d (15 N a / d)^{1 / 5}$~\cite{Baym}.
In this Thomas-Fermi limit, the expectation value becomes
\begin{eqnarray} \label{TF}
\langle \psi_g | e^{i \Delta {\bf k} \cdot {\bf R}} | \psi_g \rangle & = &
\frac{15}{2 |R_0 \Delta {\bf k}|^5} \bigl[ (6 - 2 R_0 |\Delta {\bf k}|)
\sin R_0 |\Delta {\bf k}| \nonumber \\
& & - 6 R_0 |\Delta {\bf k}| \cos R_0 |\Delta {\bf
k}| \bigr],
\end{eqnarray}
which shows oscillations of $g^{(1)}$.
In contrast the Gaussian wave function gives a monotonically decreasing
nonzero value $\langle e^{i \Delta {\bf k} \cdot {\bf R}} \rangle = e^{-|d
\Delta {\bf k}|^2 / 4}$.
The degree of second-order coherence is calculated to be
\begin{eqnarray}
& & g^{(2)}({\bf r}_1, t_1; {\bf r}_2, t_2) = \nonumber \\
& & \Bigl[ N^4 |\langle e^{i
\Delta {\bf k}_1 \cdot {\bf R}} \rangle_g \langle e^{i \Delta {\bf k}_2
\cdot {\bf R}} \rangle_g|^2 \nonumber \\
& & + N^3 \Bigl( 2 {\rm Re} \langle e^{-i \Delta
{\bf k}_1 \cdot {\bf R}} \rangle_g \langle e^{-i \Delta {\bf k}_2 \cdot
{\bf R}} \rangle_g \nonumber \\
& & \times \langle e^{i \Delta {\bf k}_2 \cdot {\bf R}} \hat
U_B(t_2 - t_1) e^{i \Delta {\bf k}_1 \cdot {\bf R}} \rangle_g \nonumber \\
& & + 2 {\rm Re} \langle e^{-i \Delta {\bf k}_1 \cdot {\bf R}} \rangle_g
\langle e^{i \Delta {\bf k}_2 \cdot {\bf R}} \rangle_g \langle e^{-i
\Delta {\bf k}_2 \cdot {\bf R}} \hat U_B(t_2 - t_1) e^{i \Delta {\bf k}_1
\cdot {\bf R}} \rangle_g \nonumber \\
& & + |\langle e^{i \Delta {\bf k}_1 \cdot {\bf R}}
\rangle_g|^2 + |\langle e^{i \Delta {\bf k}_2 \cdot {\bf R}} \rangle_g|^2
\Bigr) \nonumber \\
\label{g2int}
& & + N^2 \Bigl( 1 + |\langle e^{i \Delta {\bf k}_2 \cdot {\bf R}} \hat
U_B(t_2 - t_1) e^{i \Delta {\bf k}_1 \cdot {\bf R}} \rangle_g|^2 \nonumber
\\
& & + |\langle e^{-i \Delta {\bf k}_2 \cdot {\bf R}} \hat U_B(t_2 - t_1)
e^{i \Delta {\bf k}_1 \cdot {\bf R}} \rangle_g|^2 \nonumber \\
& & + 2 |\langle e^{i \Delta {\bf k}_1 \cdot {\bf R}} \rangle_g|^2 + 2
{\rm Re} \langle e^{-i \Delta {\bf k}_2 \cdot {\bf R}} \rangle_g \nonumber
\\
& & \times \langle e^{-i \Delta {\bf k}_1 \cdot {\bf R}} \hat
U_B^\dagger(t_2 - t_1)
e^{i \Delta {\bf k}_2 \cdot {\bf R}} \hat U_B(t_2 - t_1) e^{i \Delta {\bf
k}_1 \cdot {\bf R}} \rangle_g \Bigr) \nonumber \\
& & + N \Bigr] \Bigr/ \Bigl[ N^2 \left( 1 + N |\langle e^{i \Delta {\bf
k}_1 \cdot {\bf R}} \rangle_g|^2 \right) \left( 1 + N |\langle e^{i \Delta
{\bf k}_2 \cdot {\bf R}} \rangle_g|^2 \right) \Bigr]. \nonumber \\
\end{eqnarray}
Figure \ref{f:g2int} shows the degree of second-order coherence
(\ref{g2int}) for $t_1 = t_2$, where the parameters are the same as in
Fig.~\ref{f:g1int}.
\begin{figure}
\begin{center}
\leavevmode\epsfxsize=86mm \epsfbox{fig7.ps}
\end{center}
\caption{
The angular dependence of the degree of second-order coherence
$g^{(2)}({\bf r}_1, t; {\bf r}_2, t)$ of photons scattered from $10^3$
interacting BEC atoms.
The s-wave scattering length is assumed to be $a = 2.75$ nm.
The wavelength of a laser field $2 \pi / k_L = 589 {\rm nm}$ (the 3s-3p
transition of ${}^{23}{\rm Na}$) and the characteristic length of the
trap $d = 1 \mu{\rm m}$ are assumed.
The wave vectors ${\bf k}_1$ and ${\bf k}_2$ are assumed to be in the same
plane, and $\theta_1$ and $\theta_2$ denote their angles as measured from
${\bf k}_L$.
}
\label{f:g2int}
\end{figure}
One can see the complicated structure that arises from the oscillations
of $\langle e^{i \Delta {\bf k}_1 \cdot {\bf R}} \rangle_g$.
Figure~\ref{f:g2tint} shows the dependence on the time interval $t_2 -
t_1$ of $g^{(2)}({\bf r}_1, t_1; {\bf r}_2, t_2)$, where the parameters
are the same as in Fig.~\ref{f:g1int}, and the scattering angle of the
first photon $\theta_1$ is fixed at $\pi / 12$.
\begin{figure}
\begin{center}
\leavevmode\epsfxsize=86mm \epsfbox{fig8.ps}
\end{center}
\caption{
The time and angular dependence of the degree of second-order coherence
$g^{(2)}({\bf r}_1, t_1; {\bf r}_2, t_2)$ of photons scattered from $10^3$
interacting BEC atoms.
The scattering angle of the first photon is fixed at $\theta_1 = \pi /
12$.
The s-wave scattering length is assumed to be $a = 2.75$ nm.
The wavelength of the laser field $2 \pi / k_L = 589 {\rm nm}$ (the 3s-3p
transition of ${}^{23}{\rm Na}$) and the characteristic length of the
trap $d = 1 \mu{\rm m}$ are assumed.
The wave vectors ${\bf k}_1$ and ${\bf k}_2$ are assumed to be in the same
plane, and $\theta_1$ and $\theta_2$ denote their angles as measured from
${\bf k}_L$.
}
\label{f:g2tint}
\end{figure}
We find that the peaks appear periodically with their heights
monotonically decreasing.
The decay of the peaks can be understood as follows.
The atom is kicked out of the condensate upon receipt of recoil momentum,
and oscillates in the trap potential while encountering the condensate
periodically.
If the atom-atom interaction is not present, the Gaussian wave function of
the scattered atom retains its shape during the oscillations, and the
peaks revive completely as shown in Fig.~\ref{f:g2t}.
On the other hand, when the atom-atom interaction is present, the
wave function undergoes distortion by colliding the condensate.
As a result, the atom cannot be kicked back completely to the condensate
by scattering a photon, and therefore the correlation deteriorates.
\section{Conclusions}
\label{s:conclusion}
We have studied the coherence properties of photons scattered from the BEC
irradiated by a weak and far-off-resonant laser field, and have shown that
the degree of second-order coherence $g^{(2)}$ exhibits unique features
of trapped BEC.
That is, the probability of two photons being scattered either in the same
direction or in the symmetrical direction with respect to laser
propagation is enhanced, and these correlations appear periodically with
the period of atomic motion in the trapping potential.
These correlations in space and time are manifestations of the combined
effect of bosonic stimulation and quantized kinetic motion in the trap.
We have also taken into account effects of the atom-atom interaction, and
have confirmed that such correlations can be observed even in the presence
of the atom-atom interaction with appearance of an additional structure
(sharp dips in Fig.~\ref{f:g1int}) resulting from the repulsive nature of
the interaction.
\section*{ACKNOWLEDGMENT}
One of the authors (H.S.) acknowledges support by the Japan Society for
the Promotion of Science for Young Scientists.
|
\section{Introduction}
Plastic scintillation detectors have been used in nuclear and
high energy physics for many decades
\cite{Birks}.
Their advantages and disadvantages are recognized. Among their
benefits are fast response, ease of manufacture and versatility.
Their main drawbacks are radiation resistance and cost.
Many research projects have concentrated on improving the
fundamental properties of plastic scintillators
\cite{SCIFI93,SCIFI97}, but little
attention has focussed on their cost. Currently
available plastic scintillating materials are high quality
products whose cost is relatively expensive, and because of
that, their use in very large detectors has not been a
feasible option.
For instance,
MINOS (Main Injector Neutrino Oscillation Search)
will require 400,000 Kg of plastic scintillator for its
detector
\cite{MINOS}. With the price of cast
scintillator at approximately \$40 per Kg, such a detector would not
be affordable. However,
recent studies using
commercial polystyrene pellets as the base material for extrudable
plastic scintillators have allowed the MINOS collaboration
to consider a less expensive alternative in building
a plastic scintillation detector. Furthermore, the D0 experiment at
Fermilab has been able to use extruded plastic scintillator to
build and upgrade their Forward and Central Preshower Detectors.
In this case,
the driving force was not the cost of the material,
since only 2,000 Kg of plastic
scintillator were needed, but the opportunity to use a particular
shape (triangular bar)
that would have been expensive to machine out of cast plastic
scintillator sheets
\cite{D01,D02}.
\section{Extruded Plastic Scintillators}
Several factors contribute to the high cost of plastic
scintillating sheets and wavelength shifting fibers.
The main reason is
the labor-intensive nature of the manufacturing process.
The raw materials, namely styrene, vinyltoluene, and the
dopants, need to be highly pure.
These purification steps often take place just prior to the
material utilization.
Cleaning and assembly of the molds for the polymerization process
is a detail-oriented operation that adds to the overall timeline.
The polymerization cycle lasts several days. It consists of a high
temperature treatment to induce full conversion from monomer
to polymer, followed by a controlled ramp-down to room temperature to
achieve
a stress-free material.
Finally, there are machining charges for sheets and tiles,
and drawing charges for fibers that cannot be overlooked.
In order to significantly lower the cost of plastic scintillators,
extruded plastic scintillation materials need to be considered.
In an extrusion process, polymer pellets or powder must
be used. Commercial polystyrene pellets are readily available,
thus eliminating monomer purification and polymerization charges.
In addition, extrusion can manufacture nearly any shape,
increasing detector geometry options. There are, however, some
important disadvantages. The extruded plastic scintillator is
known to have poorer optical quality than the cast material.
The main cause is high particulate matter content
within the polystyrene pellets. General purpose
polystyrene pellets are utilized in numerous products but
few of them have strict optical requirements.
A way to bypass the short attenuation length problem is to extrude
a scintillator shape and use a wavelength shifting (WLS) fiber
as readout. Our first approach was a two-step process
that involved adding dopants to commercial polystyrene pellets to
produce scintillating polystyrene pellets, which were then used
to extrude a scintillator profile with a hole in the middle
for a WLS fiber (Figure~\ref{rect}).
The goal in the first step was to prepare scintillating pellets
of acceptable optical quality in a factory environment.
In addition to careful selection of the raw materials, the
manufacturing concerns dealt with possible
discoloration of the scintillating pellets because of
either residues present in the equipment or degradation
of the polymer pellets and the dopants in the processing device.
The latter could be induced by the presence of oxygen
at the high temperatures and pressures which constitute
the typical operating conditions.
\begin{figure}[]
\begin{picture}(432,108)
\put(144,36){\framebox(60,30)}
\put(174,51){\circle{5}}
\put(144,67){\line(3,1){100}}
\put(204,36){\line(3,1){100}}
\put(204,66){\line(3,1){100}}
\end{picture}
\vskip -.5 cm
\caption[]{
\label{rect}
\small Rectangular scintillator profile with a hole in the middle
for a WLS fiber.}
\end{figure}
After producing the first batch of scintillating pellets, samples
were cast to perform light yield and radiation
degradation studies.
Samples of standard cast scintillators such
as BC404 and BC408, and samples prepared through bulk polymerization
at Fermilab were also included in the studies (Table I).
All the samples had similar dopant composition and were cut as 2-cm
cubes. The light yield measurements were performed using a
$^{207}$Bi source (1 MeV electrons).
The light yield results (Table I) showed no significant
difference among them. The Bicron samples are made of
poly(vinyltoluene) instead of polystyrene which accounts
for the 20\% increase in light output \cite{Birks}.
The samples for radiation damage studies were placed in stainless
steel cans and connected to a vacuum pump for two weeks to remove
dissolved air and moisture. The cans were then
back-filled with nitrogen and irradiated with a $^{60}$Co source
at the Phoenix Memorial Laboratory of the University of Michigan.
The irradiations took place
at a rate of approximately 15 KGy/h to a total dose of 1 KGy.
After irradiation and annealing,
the extruded scintillator cubes showed a 5\%
decrease in light yield which is similar to the losses observed
in regular scintillator of this composition.
Based on these tests, there was no sign of degradation in the
scintillating pellets. The material was then used to produce extruded
scintillator of different profiles with a hole in the middle
for a WLS fiber.
\vspace{12pt}
\begin{center}
\small{
TABLE I. Relative light yield of samples with similar compositions
but from different manufacturing processes.}\\
\begin{tabular}{cccc} \hline\hline
\mbox{{\hspace{0.5in}}Scintillator{\hspace{0.5in}}} &
\mbox{{\hspace{0.5in}}Bicron$^{a}${\hspace{0.5in}}} &
\mbox{{\hspace{0.5in}}extruded{\hspace{0.5in}}} &
\mbox{{\hspace{0.5in}}bulk polymerized{\hspace{0.5in}}} \\
\hline
404&1.0&0.80&0.78\\
408&1.0&0.85&0.77\\
\hline\hline
\end{tabular}
\noindent
$^{a}$Bicron scintillator has a poly(vinyltoluene) matrix which
yields 20\% more light than a polystyrene one \cite{Birks}.
\end{center}
\subsection{Selection of Raw Materials}
There are many manufacturers and grades
of polystyrene pellets. Most of them fall under the category of general
purpose polystyrene. Only a few offer optical
quality polystyrene pellets. Needless to say,
there is a substantial difference in price. Nonetheless,
the first plastic scintillating pellets were prepared using
an optical grade polystyrene from Dow, labeled XU70251, which
was later superseded by XU70262 (Dow 262).
The price for Dow 262 is about \$4.5 per Kg.
After confirmation by the initial tests that high quality extruded plastic
scintillators were feasible, the quest began to replace the
costly optical grade pellets with general purpose material.
Various samples of different polystyrene grades were received
from Dow, Fina, Nova, BASF, Huntsman, etc. These samples had been
selected based on price, availability and melt flow rate for
ease of extrusion. These materials were cast into cylinders up to
3 inches long. Transmittance measurements were performed using
a Hewlett-Packard 8452 spectrophotometer. The materials tested
were compared to cast samples of Dow 262 pellets.
Often polystyrene contained additives that absorbed at long wavelengths
such as the Fina pellets illustrated in
Figure~\ref{trans}. This absorption would diminish the amount
of light produced by the dopants that need to be added
to make a particular scintillator.
Other features observed were long absorption tails and haziness caused by
additives and debris in the pellets.
There were a couple of materials that were repeatedly tested
and showed high clarity and lack of absorptions at long wavelengths.
Dow Styron 663 (Dow 663) was chosen
as the general purpose polystyrene grade to conduct our extrusion
studies. Its price ranges from \$1.3/Kg to \$1.7/Kg depending, among
other things, on the quantity ordered.
\begin{figure}[]
\centerline{\epsfxsize 3.5 truein \epsfbox{pladalmau1210fig2.eps}}
\vskip 0.2 cm
\caption[]{
\label{trans}
\small Transmittance data of commercial polystyrene pellets.}
\end{figure}
A variety of organic fluorescent compounds can be used
as primary and secondary dopants in plastic scintillator
applications.
The primary dopant is commonly used at a 1--1.5\% (by weight)
concentration. The secondary dopant or wavelength shifter
in scintillator is
utilized at a concentration of 0.01--0.03\% (by weight).
The goal was to prepare a blue-emitting scintillator that could
be readout with a green WLS fiber. Most green fibers are doped
with K27, and thus the emission of the scintillator would have to match
as best as possible the absorption of K27 in the fiber.
The selection of dopants was based on these spectroscopic
requirements as well as price and ease of manufacture.
{\em para}-Terphenyl (\$200--225/Kg) and PPO (\$100--160/Kg) were
considered as primary dopants.
POPOP and bis-MSB (both at \$0.5--1/g) were
tested as secondary dopants. The final choice for the extruded plastic
scintillator was PPO and POPOP in Dow 663.
Figure~\ref{ppo} plots the transmittance spectrum of an extruded
scintillator sample of this type.
\subsection{Manufacturing Techniques}
The majority of the extruded scintillator prepared has used
Method 1, a
two-step process conducted at two separate facilities.
Figure~\ref{meth1} depicts the flow chart for this method.
The first step was carried out at a company whose
function was to add the dopants to the polystyrene pellets.
(In the plastics industry, this trade is typically referred
to as a color or compounding business.)
Prior to the coloring run, polystyrene pellets were purged for several
days with an inert gas, generally argon, to remove dissolved oxygen
and moisture. The coloring step was a batch process where polystyrene
pellets and dopants were tumble-mixed for 15 min.\ and then added to the
hopper of an extruder. Each batch prepared 45 Kg of mixture.
A silicone oil was used as a coating aid to achieve better distribution
of the dopants on the pellet surface.
An argon flow was also added to the hopper to minimize the presence of
oxygen in the extruder. The die at the extruder head generated several
strings of material which were cut yielding the scintillating
pellets. At the end, the scintillating pellets
collected in many containers during
the run were blended to homogenize the material.
These pellets could now be used to produce plastic scintillators through
several procedures --- namely extrusion, casting and injection molding.
In this case, the scintillating pellets were
taken to an extrusion company to extrude the desired scintillator profile.
\begin{figure}[t]
\centerline{\epsfxsize 3.5 truein \epsfbox{pladalmau1210fig3.eps}}
\vskip 0.2 cm
\caption[]{
\label{ppo}
\small Transmittance data of extruded plastic scintillator.}
\end{figure}
\begin{figure}
\begin{picture}(432,288)
\put(30,255){\framebox(120,30){polystyrene pellets}}
\put(30,215){\framebox(120,30){dopants and additives}}
\put(200,235){\framebox(120,30){mixture batch}}
\put(270,195){\framebox(120,30){inert gas (argon)}}
\put(200,145){\framebox(120,30){processing device}}
\put(364,145){\framebox(90,30){extrusion}}
\put(320,160){\vector(1,0){44}}
\put(230,235){\vector(0,-1){60}}
\put(290,195){\vector(0,-1){20}}
\put(170,270){\line(0,-1){40}}
\put(170,250){\vector(1,0){30}}
\put(150,270){\line(1,0){20}}
\put(150,230){\line(1,0){20}}
\thicklines
\put(200,85){\framebox(120,40){\bf scintillating pellets}}
\put(160,15){\framebox(200,50){\bf extrude to form scintillating profile}}
\put(260,85){\vector(0,-1){20}}
\put(260,145){\vector(0,-1){20}}
\thinlines
\put(360,92){\framebox(80,25){cast}}
\put(80,92){\framebox(80,25){injection mold}}
\put(320,105){\vector(1,0){40}}
\put(200,105){\vector(-1,0){40}}
\end{picture}
\caption[]{
\label{meth1}
\small Two-step process: batch coloring and extrusion (Method 1).}
\end{figure}
Using this batch process, there is also the possibility of directly
extruding the scintillator profile and thus \mbox{by-passing}
the pelletizing
step. This is the route that the MINOS collaboration has chosen to
investigate. This variation of Method 1 can be less expensive since
all the work is done in one facility. It also reduces the heat history
of the product by removing its exposure to another high temperature
cycle and minimizes the chance of optical degradation.
The drawback is in the batch work since the polymer and the dopants
still need to be weighed for each mixture, and in the tumble-mixing step
which is susceptible to contamination and prone to errors.
An alternative to these operations is given by Method 2
which is summarized in
Figure~\ref{meth2}. Method 2 is a continuous in-line coloring and
extrusion process. It emphasizes the most direct pathway from
polystyrene pellets to the scintillator profile with the least
handling of raw materials.
In this situation, the purged polystyrene pellets and dopants are
metered into the extruder at the correct rate for the required
composition of the scintillator. An argon flow is still used
at the hopper. Coating agents are no longer needed.
The appropriate die profile gives rise to the
extruded scintillator form of choice. If the die can produce
strands, these can also be pelletized and the scintillating
pellets used in other processes.
Method 2 has been tested and produces plastic scintillator of
high quality and homogeneity.
Although it is a simple concept, the equipment needed to accurately
meter small quantities of powders such as the dopants and
to achieve a good distribution of the powders in the molten
polymer is not widely available. The difficulty in testing this
process was finding a facility with the adequate instrumentation.
\begin{figure}
\begin{picture}(432,288)
\put(144,250){\framebox(120,30){polystyrene pellets}}
\put(64,210){\framebox(120,30){dopants and additives}}
\put(224,210){\framebox(120,30){inert gas (argon)}}
\put(144,160){\framebox(120,30){processing device}}
\put(304,160){\framebox(100,30){scintillating pellets}}
\put(374,120){\framebox(80,25){cast}}
\put(374,90){\framebox(80,25){injection mold}}
\put(374,60){\framebox(80,25){extrusion}}
\thicklines
\put(104,80){\framebox(200,50){\bf extrude to form scintillating profile}}
\put(204,160){\vector(0,-1){30}}
\thinlines
\put(204,250){\vector(0,-1){60}}
\put(164,210){\vector(0,-1){20}}
\put(244,210){\vector(0,-1){20}}
\put(264,175){\vector(1,0){40}}
\put(345,160){\line(0,-1){87}}
\put(345,133){\vector(1,0){29}}
\put(345,103){\vector(1,0){29}}
\put(345,73){\vector(1,0){29}}
\end{picture}
\vskip -1.7 cm
\caption[]{
\label{meth2}
\small Continuous in-line coloring and extrusion process (Method 2).}
\end{figure}
\section{Light Yield of Extruded Plastic Scintillators}
Light yield studies have been performed on many samples of extruded
plastic scintillators. Although a variety of shapes and sizes
is available, the measurements have mostly been carried out on
11.5-cm long rectangular extrusions (1cm x 2cm) with a hole in the
middle for a green WLS fiber.
Each extrusion is tightly wrapped in Tyvek for this test.
The WLS fiber utilized is BC91A
(0.835 mm diameter, 1.5 m long) with one mirrored end.
The light yield test setup uses an electron spectrometer with
$^{106}$Ru source whose 3-MeV beam is momentum selected.
There is a small trigger counter in front of the extruded sample.
The photomultiplier tube used is a Hamamatsu R2165
which has excellent single photo-electron resolution.
The fiber is held at a fixed position from the PMT surface
to minimize fluctuations among measurements.
The light yield is determined from the following calculation:
\[Light Yield = \frac{Mean-Pedestal}{Gain}\]
\vspace{12pt}
\noindent
where the mean and the gain are defined as:
\[Mean = \frac{\sum_{i}^{n} v_{i} x_{i}}{\sum_{i}^{n} v_{i}}\]
\[Gain = First Peak - Pedestal\]
\noindent
where v$_{i}$ is the number of entries for each ADC value, x$_{i}$.
The data are fitted to locate the position of the first and second
peaks, and the pedestal.
Figure~\ref{light} presents the light yield distribution of an
extruded sample and the fit for the first and second electron peaks.
The results from a series of light yield measurements are
listed in Table II.
RDN 262 extrusions were prepared by the two-step batch process (Method 1)
using Dow 262 optical grade pellets.
Leistritz 262 and 262P samples were produced by the continuous
procedure (Method 2) using Dow 262 polymer.
Leistritz 663 samples were also prepared by Method 2 but used
general purpose polystyrene pellets (Dow 663).
Although the samples are from different runs, their light output
is similar. The Leistritz 262 samples show a slightly lower
light yield but their profile is smaller than that of the
remaining samples.
These samples were collected early in the extrusion run
when the profile was not completely to specification.
These results indicate that there is
no major difference in light yield between Method 1 and Method 2.
This test proves that the
continuous in-line coloring and extrusion process (Method 2) yields
a homogeneous part with the right concentration of dopants.
This aspect is less of a concern in Method 1 since the first step
includes batch tumble-mixing and post-blending of the
scintillating pellets before the scintillating profile is extruded.
In addition, these numbers confirm that Dow 663 (general purpose
polystyrene pellets) can replace the optical grade pellets
initially utilized.
More measurements are underway to compare these extrusions
to samples of commercial plastic scintillator sheets
which have been cut to the same profile.
\begin{figure}[]
\centerline{\epsfxsize 3.5 truein \epsfbox{pladalmau1210fig6.eps}}
\vskip 0.2 cm
\caption[]{
\label{light}
\small Light yield distribution of extruded plastic
scintillator (RDN 262) with WLS fiber. Solid curve is 2-gaussian
fit to first and second peaks.}
\end{figure}
\vspace{12pt}
\begin{center}
\small{
TABLE II. Light yield of extruded plastic
scintillator samples.}\\
\begin{tabular}{lcccc} \hline\hline
\mbox{{\hspace{0in}}Scintillator{\hspace{0.5in}}} &
\mbox{{\hspace{0.25in}}No. of samples{\hspace{0.25in}}} &
\mbox{{\hspace{0.25in}}Light Yield{\hspace{0.25in}}} &
\mbox{{\hspace{0.25in}}St. Dev.{\hspace{0.25in}}} &
\mbox{{\hspace{0.25in}}Characteristics{\hspace{0.25in}}}\\
\hline
RDN 262&30&2.05&0.09&Dow 262, Method 1\\
Leistritz 262&10&1.81&0.14&Dow 262, Method 2\\
Leistritz 262P&10&2.02&0.10&Dow 262, Method 2\\
Leistritz 663&15&2.22&0.07&Dow 663, Method 2\\
\hline\hline
\end{tabular}
\end{center}
\section{Conclusions}
Research on extruded plastic scintillator was driven
by the high cost of cast plastic scintillator.
The goal was to use commercially available polystyrene
pellets, in particular from a general purpose grade,
and standard extrusion equipment to lower the price
of plastic scintillators.
Extruded plastic scintillator strips have
been manufactured and tested.
The estimated price for extruded scintillator ranges
from \$3.5/Kg to \$6/Kg. About 50\% of the cost
is due to raw materials with the remaining 50\%
due to processing.
The results indicate
that the extruded scintillator profile with
a WLS fiber as readout is a valid system for
scintillation detectors.
The MINOS experiment will build a very large detector
using this technology.
The D0 experiment is assembling the Central and Forward
Preshower detectors using extruded scintillating
triangular strips.
|
\section{Introduction}
It has become increasingly clear that including the circulating
supercurrents outside the vortex core in a $d$-wave superconductor
can lead to a qualitative understanding - in the low magnetic field
limit, near $H_{c1}$ - of the effect of a magnetic field {\bf H} on
various properties of a superconductor in the vortex state.
The $\sqrt{H}$ dependence of the specific heat
predicted by Volovik\cite{Volo1,Volo2} and Kopnin and Volovik\cite{%
Kopn1,Kopn2} was verified experimentally.\cite{Moler1,Moler2,Junod}
A detailed analysis of the data for {\bf H} oriented perpendicular
to the CuO$_2$ planes including impurity scattering was given by
K\"ubert and Hirschfeld\cite{Kub1} (see also Barash {\it et al.}%
\cite{Barash}).
Transport properties were considered by K\"ubert and Hirschfeld.%
\cite{Kub2} Subsequently the analysis has been extended by Hirschfeld%
\cite{Hirsch1} to include the field {\bf H} in the CuO$_2$ plane and
he demonstrated a four-fold symmetry in the electronic thermal
conductivity. Very recently results have been obtained by Vekhter
{\it et al.}\cite{Vek1,Vek2} for the effect of an in-plane magnetic field on the
quasiparticle density of states (QDOS) in a tetragonal two dimensional
$d$-wave superconductor. The present work extends this previous study
to the orthorhombic $d$-wave case and also includes in the calculations
the effect of impurity scattering in both, Born's and resonant scattering
limits.
While most high $T_c$ oxides are tetragonal, YBa$_2$Cu$_3$O$_{6.95}$ (YBCO),
which is perhaps the most extensively studied of all the oxides, is
orthorhombic because its crystal structure includes CuO chains oriented
along the $b$-axis, as well as two CuO$_2$ planes per unit cell. Optical
measurements reveal that the carrier density in the chains in optimally
doped YBCO is of the same order of magnitude as it is in the planes and
the ratio of the associated $b$- to $a$-direction plasma frequency is
of the order of 2.\cite{Tanner} Similar large in-plane anisotropies
are observed in the infrared optical conductivity,\cite{Basov} d.c.\
resistivity,\cite{Gagnon1,Fried} zero temperature penetration depth,%
\cite{Zhang,Bonn} and thermal conductivity.\cite{Yu,Gagnon2}
Various approaches have been taken to treat band anisotropy including
a single tight binding band with different nearest neighbor hopping
in $a$- and $b$-direction,\cite{Odonov1,Odonov2} two band\cite{Atkin1,%
Atkin2,Odonov3,Odonov4} as well as three band\cite{Odonov5,Atkin3} models
which include the single CuO chain layer as well as two CuO$_2$ plane
layers per unit cell with transverse hopping between them. All these
models give the same qualitative features and describe reasonably well
the $a$-$b$ band anisotropy of YBCO. Additionally, in an orthorhombic structure the
superconducting gap will be a mixture\cite{Odonov6} of dominant
$d_{x^2-y^2}$ symmetry as well as subdominant $s_0$ and $s_{x^2+y^2}$
symmetric functions because they all belong to the same irreducible
representation of the $C_{2v}$ crystal point group.
The simplest approach which contains the physics of the more complex
calculations is a single infinite band with distinct effective masses
along the $a$ $(m_a)$ and $b$ $(m_b)$ directions\cite{Kim,Wu,Schur,Ju}
with a $(d+s)$-symmetric gap function on the cylindrical Fermi surface.%
\cite{Maki,Modre} For simplicity we will adopt throughout this paper
this approach and calculate the effect of an in-plane magnetic field {\bf H} on
the QDOS assuming a distribution of non overlapping vortex cores and account only
for the circulating supercurrents around the vortices. We will also
treat the effect of impurities in both, Born's and
resonant scattering limits. Impurities were not considered in the
previous calculations in tetragonal symmetry by Vekhter {\it et al.}\cite{Vek1}
which were confined to the pure limit. Thus, our results with impurities
in the tetragonal case are the first such results.
\section{Formalism}
We begin with a single free electron band in two dimensions with an
ellipsoidal Fermi surface and the single particle energy
$\varepsilon_{\bf k}$ in the normal state given by the dispersion
relation $(\hbar = 1)$
\begin{equation}
\varepsilon_{\bf k} = {k_x^2\over 2m_x}+{k_y^2\over 2m_y}-
\varepsilon_F \label{e1}
\end{equation}
where ${\bf k} = (k_x,k_y)$ is the momentum vector and $\varepsilon_F$
the Fermi energy. In Eq.~(\ref{e1}) $m_x$ and $m_y$ are effective
masses with $m_x > m_y$, {\it i.e.} chains are along the $k_y$-axis
which increases the electronic transport in that direction. A parameter
\begin{equation}
\alpha = {m_x-m_y\over m_x+m_y} \label{e2}
\end{equation}
is introduced and this is the single parameter which characterizes the
band structure anisotropy in our simple model. Consistent with the
above assumption for the energy band dispersion relation is a mixed
$(d+s)$-symmetric gap on the Fermi surface (an ellipse) of the form
\begin{equation}
\Delta_{\bf k} = \Delta_0\left (\cos 2\varphi +s\right ),
\label{e3}
\end{equation}
with $\Delta_0$ the gap amplitude. In Eq.~(\ref{e3}) $\varphi$ is a
polar angle measured from the $k_x$-axis in the two dimensional copper
oxide Brillouin zone giving the angle of {\bf k} (momentum on the
Fermi surface) as shown in Fig.~\ref{f1}. The first term in Eq.~(\ref{e3})
gives the dominant $d$-wave part of the gap which, on its own, would have
nodes on the main diagonal at $\varphi = \pm\pi/4$. The number $s$ is
a measure of the subdominant $s$-wave part. Both parts are allowed to
exist in an orthorhombic system such as YBCO from group theoretical
considerations. The $s$-component moves the nodes from
$\varphi = \pm\pi/4$ to $\varphi = \tan^{-1}\sqrt{(1+s)/(1-s)}$ off the
main diagonal of the Brillouin zone. For positive values of $s$ the
critical $\varphi$ defining the node is larger than $\pi/4$ while the opposite
holds for negative values of $s$.
In order to work with Eqs.~(\ref{e1}) and (\ref{e3}) it is convenient
to transform to new coordinates ${\bf p} = (p_x,p_y)$ in which the
Fermi surface is a circle (see Fig.~\ref{f1}).\cite{Kim} The required
transformation is given by
\begin{equation}
p_i = k_i\sqrt{m_x+m_y\over 2m_i},\qquad i = x,y
\label{e4}
\end{equation}
which leads to (Fig.~\ref{f1})
\begin{equation}
\tan\phi = \sqrt{m_y\over m_x}\tan\varphi,
\label{e5}
\end{equation}
where $\phi$ is now the angle of the momentum vector {\bf p} in the
transformed frame. Clearly, the electron dispersion relation (\ref{e1})
reduces to
\begin{equation}
\varepsilon_F = {p^2_{F,x}+p^2_{F,y}\over 2\bar{m}} =
{p^2_F\over 2\bar{m}},
\label{e6}
\end{equation}
with $\bar{m} = (m_x+m_y)/2$ the average band mass. Applying the same
transformation to the gap (\ref{e3}) gives the order parameter in the
{\bf p}-frame\cite{Wu,Schur,Ju}
\begin{equation}
\Delta_{\b p} = \Delta_0\left({\alpha+\cos 2\phi\over 1+\alpha
\cos 2\phi}+s\right) \equiv \Delta_0 f(\phi).
\label{e7}
\end{equation}
We show in Fig.~\ref{f2} the gap $\Delta_{\bf p}$ in the {\bf p}-frame for
several values of $\alpha$ (the band anisotropy) and $s$ (the gap
anisotropy). Frame (a) is for $\alpha = 0$ and $s=0$ and
is the well known pure $d$-wave case for which
plus and minus lobes have equal absolute magnitude and the zeros are
on the diagonal. In frame (b) $\alpha = 0.4$ and $s=0$, {\it i.e.} the
gap in the {\bf k}-frame (laboratory frame) is of pure $d$-wave
symmetry but the band structure orthorhombicity leads to a transformed
gap with larger positive than negative lobes and the nodes are shifted
upwards from $\pm\pi/4$ and similarly for the other symmetry related
angles. This trend is even more pronounced in frame (c) where
$\alpha = s = 0.4$. In this case $\alpha$ and $s$ add constructively to
further increase the positive lobes as compared to the negative ones,
and the nodes are shifted to yet larger angles. Finally, frame (d)
applies to $\alpha = 0.4$ and $s=-0.4$. For this set of parameters the
nodes in the {\bf p}-frame remain on the diagonal but at the same time
the negative lobes are dominant over the positive ones. We conclude that
a negative value for $s$ can in some sense counteract some of the effects
of a positive $\alpha$, but there is no complete cancelation between
these two parameters which play quite distinctive roles. The nodes can be
made to remain unshifted but the size of the lobes is changed over their
$\alpha = s = 0$ values.
In the laboratory frame ({\bf k}, before transformation) the
thermodynamic Green's function takes on the form
\begin{equation}
G({\bf k},\omega_n;{\bf r}) = -{\left(i\omega_n-{\bf v}_F({\bf k})
{\bf q}_s\right)\tau_0 +\varepsilon_{\bf k}\tau_3+\Delta_{\bf k}\tau_1
\over \left(\omega_n+i{\bf v}_F({\bf k}){\bf q}_s\right)^2+
\varepsilon^2_{\bf k} +\Delta^2_{\bf k}},
\label{e8}
\end{equation}
with $\varepsilon_{\bf k}$ and $\Delta_{\bf k}$ given by Eqs.~(\ref{e1})
and (\ref{e3}) respectively. In Eq.~(\ref{e8}) the $\tau$'s are Pauli
$2\times 2$ matrices and $i\omega_n$ is the $n$'th Matsubara frequency,
$\omega_n = (2n+1)\pi T, n = 0,\pm1,\pm2,\ldots$, and $T$ is the
temperature. The Doppler shift caused by the presence of circulating
supercurrents outside the vortex core (the region which we assume to
dominate the physics) is given by ${\bf v}_F({\bf k}){\bf q}_s$, where
${\bf v}_F({\bf k})$ is the electron Fermi velocity on the ellipsoidal
Fermi surface in {\bf k}-space (see Fig.~\ref{f1}) and which is
given by
\begin{equation}
{\bf v}_F({\bf k}) = \nabla_{\bf k}\varepsilon_{\bf k} =
{k_{F,x}\over m_x}{\bf\hat{x}}+{k_{F,y}\over m_y}{\bf\hat{y}},
\label{e9}
\end{equation}
with ${\bf \hat{x}}$ and ${\bf \hat{y}}$ unity vectors in the direction of
the $k_x$- and $k_y$-axis. Applying the transformation (\ref{e4}) to get
${\bf v}_F({\bf k})$ in the {\bf p}-frame leads to
\begin{equation}
{\bf v}_F({\bf p}) = {1\over\sqrt{\bar{m}}}\left(
{p_{F,x}\over\sqrt{m_x}}{\bf\hat{x}} +
{p_{F,y}\over\sqrt{m_y}}{\bf\hat{y}}\right).
\label{e10}
\end{equation}
The momentum of the superfluid currents outside the vortex core (placed
at the position ${\bf r} = 0$ in the laboratory frame) is assumed to be
inversely proportional to the distance $\vert{\bf r}\vert = r$ from the
core. For a magnetic field {\bf H} in the CuO$_2$ plane making an
angle $\gamma$ with the $k_x$-axis the supercurrents are in a plane
perpendicular to {\bf H} and therefore to the $k_x$-$k_y$ plane.
Assuming circular supercurrents, an approximation discussed in the paper
by Vekhter {\it et al.},\cite{Vek1} and denoting the vortex winding
angle by $\beta$ one obtains for the $x$ and $y$ component of the
superfluid momentum ${\bf q}_s$
\begin{eqnarray}
q_{s,x} &=& -\vert {\bf q}_s\vert \sin\beta\sin\gamma,\nonumber\\
q_{s,y} &=& -\vert {\bf q}_s\vert \sin\beta\cos\gamma,
\label{e11}
\end{eqnarray}
so that
\begin{equation}
{\bf v}_F({\bf p}){\bf q}_s =
{p_F\over m_x}\vert{\bf q}_s\vert\sin\beta\sqrt{1+\alpha}\left (
\sqrt{1+\alpha\over 1-\alpha}\sin\phi\cos\gamma-\cos\phi\sin\gamma
\right).
\label{e12}
\end{equation}
If one measures the distance $r$ in units of the intervortex distance
$R = a^{-1}\sqrt{\phi_0\over\pi H}$, where $\phi_0$ is the flux quantum
and $a$ a constant of the order of unity, we obtain
\begin{equation}
{\bf v}_F({\bf p}){\bf q}_s =
{E_H\over\rho}\sin\beta\sqrt{1+\alpha}\left (
\sqrt{1+\alpha\over 1-\alpha}\sin\phi\cos\gamma-\cos\phi\sin\gamma
\right),
\label{e13}
\end{equation}
with the normalized distance $\rho = r/R$ and the magnetic energy
\begin{equation}
E_H = {a\over 2}\bar{v}_F\sqrt{\pi H\over\phi_0} = \nu\Delta_0,
\label{e14}
\end{equation}
with $\bar{v}_F$ a suitably defined effective Fermi velocity\cite{Vek1}
which can also be made to contain the additional band factor
$\sqrt{1+\alpha}$ of Eq.~(\ref{e13}) and $\nu$ is a parameter which
measures the magnetic energy $E_H$ in units of the gap. Consequently,
we can write
\begin{equation}
{\bf v}_F({\bf p}){\bf q}_s = {E_H\over\rho}h(\phi,\gamma)\sin\beta,
\label{e15}
\end{equation}
with
\begin{equation}
h(\phi,\gamma) =
\sqrt{1+\alpha\over 1-\alpha}\sin\phi\cos\gamma-\cos\phi\sin\gamma.
\label{e16}
\end{equation}
The QDOS at zero frequency obtained from the Green's function (\ref{e8})
which includes the circulating supercurrents about the vortex core
and which must also be averaged over the vortex unit cell is given by
\begin{equation}
N(0,\gamma) = \int\limits^{2\pi}_0\!{d\phi\over 2\pi}\,
\int\limits^{2\pi}_0\!{d\beta\over\pi}\,
\int\limits^R_0\!{r dr\over R^2}\,\Re{\rm e}\left\{
{\vert {\bf v}_F({\bf p}){\bf q}_s\vert\over
\sqrt{({\bf v}_F({\bf p}){\bf q}_s)^2-\Delta^2_0f^2(\phi)}}
\right\}.
\label{e17}
\end{equation}
By the square root is meant that branch of the complex function which corresponds
to a positive imaginary part.
Eq.~(\ref{e17}) can be reduced to an one dimensional
integral form
\begin{equation}
N(0,\gamma) = {1\over 2\pi}\int\limits^{2\pi}_0\!d\phi\,
{\rm min}\left( 1,{\nu^2\vert h(\phi,\gamma)\vert^2\over
f^2(\phi)} \right).
\label{e18}
\end{equation}
One good approximation to this integral is obtained by noting that the
integrand in Eq.~(\ref{e18}) is dominated by the nodal region and, thus,
we get approximately
\begin{equation}
N(0,\gamma) \simeq 2{\nu\over\pi}\sum\limits_{\rm nodes}
\left\vert {h(\phi_n,\gamma)\over f'(\phi_n)}\right\vert,
\label{e19}
\end{equation}
where the $\phi_n$ are the nodes of the transformed gap (\ref{e7}) and
$f'(\phi_n)$ is the derivative of this gap function at its nodes. These
can be found at
\begin{equation}
\tan\phi_n = \pm\sqrt{{(1+\alpha)\over(1-\alpha)}
{(1+s)\over(1-s)}},
\label{e20}
\end{equation}
while in the laboratory frame, as we have noted before, the gap has its
nodes at $\tan\varphi_n = \pm\sqrt{(1+s)/(1-s)}$. Furthermore, the
derivative $f'(\phi_n)$ at the gap nodes can be worked out to be
\begin{equation}
f'(\phi_n) = 2(1+s)\sqrt{(1-s^2)(1-\alpha^2)},
\label{e21}
\end{equation}
and hence an approximation for $N(0,\gamma)$ follows from Eqs.~(\ref{e18}),
(\ref{e21}) and the definition of $h(\phi,\gamma)$ given in Eq.~%
(\ref{e15}) as
\begin{equation}
N(0,\gamma) = {\nu\over\pi}\sum\limits_{\rm nodes}
{\left\vert\sqrt{1+\alpha\over 1-\alpha}\sin\phi_n\cos\gamma-
\cos\phi_n\sin\gamma\right\vert\over
(1+s)\sqrt{(1-s^2)(1-\alpha^2)}},
\label{e22}
\end{equation}
with distinct contributions from only two of the four nodes.
For one set of two nodes we get
\begin{eqnarray}
\cos\phi_n &=& \sqrt{{1\over 2}{(1-\alpha)(1-s)\over 1+\alpha s}},
\nonumber\\
\sin\phi_n &=& \sqrt{{1\over 2}{(1+\alpha)(1+s)\over 1+\alpha s}},
\label{e23}
\end{eqnarray}
while for the other ones $\phi_n\to\pi-\phi_n$. We therefore arrive at the
approximate analytic expression
\begin{eqnarray}
N(0,\gamma) &=& {\sqrt{2}\nu\over\pi}{1\over\sqrt{(1-s^2)(1-\alpha^2)}}
{1\over 1+s}\sqrt{1+\alpha\over 1-\alpha}\nonumber\\
&&\times\left\{\left\vert(1+\alpha)\sqrt{1+s\over1+\alpha s}\cos\gamma-
(1-\alpha)\sqrt{1-s\over 1+\alpha s}\sin\gamma\right\vert\right .
\nonumber\\
&&+\left.\left\vert(1+\alpha)\sqrt{1+s\over 1+\alpha s}\cos\gamma+
(1-\alpha)\sqrt{1-s\over 1+\alpha s}\sin\gamma\right\vert\right\}
\nonumber\\
&=& {2\sqrt{2}\nu\over\pi}{1\over\sqrt{(1-s^2)(1-\alpha^2)}}
{1\over 1+s}\sqrt{1+\alpha\over 1-\alpha}\nonumber\\
&& \times{\rm max}\left(
\left\vert(1+\alpha)\sqrt{1+s\over 1+\alpha s}\cos\gamma\right\vert,
\left\vert(1-\alpha)\sqrt{1-s\over 1+\alpha s}\sin\gamma\right\vert
\right),
\label{e24}
\end{eqnarray}
where we have restored factors previously absorbed in the magnetic energy
$E_H$ that appeared explicitely in Eq.~(\ref{e13}). In the tetragonal,
pure $d$-wave case $(\alpha = s = 0)$ expression (\ref{e24}) reduces
properly to the known result\cite{Vek1}
\begin{equation}
N(0,\gamma) = {2\sqrt{2}\nu\over\pi}{\rm max}\left(\vert\sin\gamma\vert,
\vert\cos\gamma\right\vert),
\label{e25}
\end{equation}
in which case Eq.~(\ref{e25}) is known to be an excellent
approximation to the one dimensional integral of Eq.~(\ref{e18}).
When band anisotropy is included through a finite value of $\alpha$ or
a subdominant $s$-wave part, characterized by the parameter $s$,
the analytic expression (\ref{e24}) is not quite as good
an approximation as (\ref{e25}) was but it is
still acceptable as can be seen in Fig.~\ref{f3}. In this figure we
compare with our exact numerical results for $N(0,\gamma)$ based on the
one dimensional integral of Eq.~(\ref{e18}). Here only band anisotropy
is included with an effective mass anisotropy parameter $\alpha = 0.2$.
The dashed line is the exact result (\ref{e18}), the solid line is based
on Eq.~(\ref{e24}). The agreement is good. Note the two-fold symmetry
in the pattern obtained, although, since $s=0$, the gap in the laboratory
frame has pure $d$-wave symmetry with nodes on the main diagonal.
It is the band anisotropy that breaks the four-fold symmetry of the
tetragonal case. However, it is
very important to realize that because of $\alpha\ne 0$,
{\it i.e.} band orthorhombicity, the
sharp kinks in the curve for $N(0,\gamma)$ vs.\ $\gamma$ no longer fall
at the position of the nodes. This would be the case in a tetragonal
system but is not so for the ellipsoidal Fermi surface of Fig.~\ref{f1}.
We see from Eq.~(\ref{e24}) that the kink occurs at a critical angle
$\gamma_c$ given by (note that $s=0$):
\begin{equation}
\tan\gamma_c = {1+\alpha\over 1-\alpha},
\label{e26}
\end{equation}
which is $\pi/4$ only for the case $\alpha = 0$ (no anisotropy in the
electronic band structure). This is an important result because it is
often believed that the sharp structure in $N(0,\gamma)$ as a function
of $\gamma$ - the magnetic field orientation - gives the position of
the nodes. This is not true when band anisotropy exists.
The geometry of the situation is shown in Fig.~\ref{f4} which depicts
the laboratory frame while the result (\ref{e24}) was obtained in the
transformed, {\bf p}-frame where the Fermi surface is a circle. In this
figure ${\bf k}_n$ gives the position of a gap node on the ellipsoidal
Fermi surface $\varepsilon_{\bf k} = 0$, in Eq.~(\ref{e1}). The Fermi
velocity at that point, ${\bf v}_F({\bf k}_n)$, points in the direction
$\tan^{-1}\left({1+\alpha\over 1-\alpha}\tan\varphi\right)$. If the
magnetic field {\bf H} is placed in that direction the circulating
supercurrents will be in a plane perpendicular to the two dimensional
Brillouin zone and oriented along the tangent to the Fermi surface at
${\bf k} = {\bf k}_n$ as shown in the figure by the thick shaded line.
Thus, there will be no
Doppler shift. In the laboratory frame $\tan\varphi = \sqrt{(1+s)/(1-s)}$
from Eq.~(\ref{e3}) and we get immediately the result
\begin{equation}
\tan\gamma_c = {1+\alpha\over 1-\alpha}\sqrt{1+s\over 1-s}
\label{e27}
\end{equation}
for a finite subdominant $s$-wave part to the gap. This same result
follows from Eq.~(\ref{e24}) when the expression in the absolute
value switches sign. For finite values of $s$, the angle $\gamma_c$
coincides with the position of the nodes in the $(d+s)$-symmetric
gap in the laboratory frame only if $\alpha = 0$. We can also see
from (\ref{e27}) that a negative value of the subdominant $s$-wave part
({\it i.e.} a negative value of $s$) partially compensates for the effect
of $\alpha$ but does not cancel it out completely.
In Fig.~\ref{f4New} we show further results for the zero frequency
QDOS $N(0,\gamma)$ as a function of magnetic field orientation $\gamma$
for increasingly anisotropic bands. The parameter $\alpha$ is increased
from 0.2 in the solid curve to 0.4 (dashed curve), 0.6 (dotted
curve), and 0.8 (dashed-dotted curve). It is clear that the amplitude
of the predicted oscillations in the QDOS increases considerably with increasing
effective mass anisotropy $(\alpha)$ and that consequently this anisotropy
will be easier to detect as the orthorhombicity of the band structure
increases, {\it i.e.} as $\alpha$ increases. There is nearly a factor of
6 between the value at minimum and at maximum in the dashed-dotted curve.
The pattern of oscillation is also changed as $\alpha$ increases. The
intermediate secondary maximum, around $\gamma = 1.5\,{\rm rad}$ in the
solid curve, is not present any longer in the dashed-dotted curve.
Instead there is a minimum.
A fit to data of the form (\ref{e24}) could in principle yield information
on the two most important parameters $\alpha$ and $s$ but the position
of the sharp structure in the QDOS by itself does not give directly the
position of the nodes or the ratio of the subdominant $s$-wave admixture
to the dominant $d$-wave component. It is the direction of {\bf H}
giving no Doppler shift which determines the position of the minimum
in $N(0,\gamma)$. Before leaving Eq.~(\ref{e24}) we note that
$N(0,\gamma)$ is proportional to $\nu = E_H/\Delta_0$ and this is the same
dependence on the magnetic field as has been found in the tetragonal
case. There is no explicit change in this quantity due to anisotropy so that
the scaling
variable for the QDOS and for the specific heat stays unchanged. This
holds for the pure case but will be modified when disorder is considered.
In that case scaling breaks down\cite{Moler2} and specific details
play a role.
The finite frequency QDOS in a pure orthorhombic superconductor with a
$(d+s)$-symmetric gap is also of interest. It follows from the formula
\begin{equation}
N(\omega,\gamma) = \int\limits^{2\pi}_0\!{d\beta\over\pi}\,
\int\limits^1_0\!d\rho\,\rho
\int\limits^{2\pi}_0\!{d\phi\over 2\pi}\,
\Re{\rm e}\left\{
{\tilde{\omega}(\omega)-{\bf v}_F({\bf p}){\bf q}_s\over
\sqrt{(\tilde{\omega}(\omega)-{\bf v}_F({\bf p}){\bf q}_s)^2
-\tilde{\Delta}^2_{d+s}(\omega,\phi)}}
\right\},
\label{e28}
\end{equation}
which has been written in a form which remains valid even when impurity
scattering is included. Here $\rho = r/R$. For the pure case
$\tilde{\omega}(\omega) =
\omega$ and $\tilde{\Delta}_{d+s}(\omega,\phi) = \Delta_0f(\phi)$.
Results of the numerical evaluation of Eq.~(\ref{e28}) (an approximate
analytic formula similar to Eq.~(\ref{e24}) can also be derived but
is not given here as it has the same form as already presented in the
work of Vekhter {\it et al.}\cite{Vek1,Vek2}) gives the results shown
in Fig.~\ref{f5} for four directions of the magnetic field.
Here we present numerical results for $N(\omega,\gamma)$
as a function of $\omega/\Delta_0$ for $\alpha = 0.2$ and $\nu=0.3$.
For the magnetic field in the direction of $\gamma = 0^\circ$ (antinodal
direction in pure $d$-wave) the solid curve applies and an $\omega^2$
behavior is obtained at small $\omega$, just as in the tetragonal case.\cite{%
Vek1,Vek2} But the linear dependence expected in the nodal direction
in the tetragonal case now occurs in some other direction where the
Doppler shift at the nodal position is zero, namely at $\gamma_c$
as defined by Eq.~(\ref{e27}) which is where the superfluid momentum
${\bf q}_s$ is perpendicular to the Fermi velocity. For $\alpha=0.2$,
$\gamma_c\simeq 56^\circ$ the dashed-dotted curve indeed displays
linear dependence.
\section{Impurity Effects}
We now include in the calculations elastic impurity scattering. In this
case the full self consistent Green's function (\ref{e8}) enters the
impurity term and we need to solve two coupled equations, one for the
renormalized Matsubara frequency $\tilde{\omega}(\omega)$ and the other
one for the gap $\tilde{\Delta}_{d+s}(\omega,\phi)$. On the real
frequency axis they are:
\begin{mathletters}
\label{e29}
\begin{eqnarray}
\tilde{\omega}(\rho,\beta,\omega) &=& \omega+i\pi t^+
\Omega(\rho,\beta,\omega), \label{e29a}\\
\tilde{\Delta}_{d+s}(\rho,\beta,\omega,\phi) &=&
\Delta_{d+s}(\phi)+i\pi t^+ D(\rho,\beta,\omega), \label{e29b}
\end{eqnarray}
\end{mathletters}
with $\Delta_{d+s}(\phi) = \Delta_{\bf p}$ of Eq.~(\ref{e7}),
the pure system gap in the transformed frame. Here $\rho$ and $\beta$
are associated with the vortex and $\phi$ is the polar angle defining
the direction of {\bf p}. The use of Eqs.~(\ref{e29}) confines our
analysis to low temperatures where the temperature dependence of the
gap function can be neglected. Moreover, these equations are only
valid in Born's approximation, or weak impurity scattering limit. A
slightly more complicated set of equations needs to be solved in the
unitary limit (strong impurity scattering), namely
\begin{mathletters}
\label{e30}
\begin{eqnarray}
\tilde{\omega}(\rho,\beta,\omega) &=& \omega+i\pi \Gamma^+
{\Omega(\rho,\beta,\omega)\over\Omega^2(\rho,\beta,\omega)+
D^2(\rho,\beta,\omega)}, \label{e30a}\\
\tilde{\Delta}_{d+s}(\rho,\beta,\omega,\phi) &=&
\Delta_{d+s}(\phi)+i\pi \Gamma^+ {D(\rho,\beta,\omega)\over
\Omega^2(\rho,\beta,\omega)+D^2(\rho,\beta,\omega)}. \label{e30b}
\end{eqnarray}
\end{mathletters}
In the tetragonal case $(\alpha = 0)$, also considered here, the
symmetry is described by the C$_{4v}$ point group and the $(d+is)$-symmetry
of the order parameter is the only possible mixed symmetry allowed
under the symmetry operations of this group. In this case Eqs.~(\ref{e29b})
and (\ref{e30b}) are to be modified to give
\begin{equation}
\tilde{\Delta}_s(\rho,\beta,\omega) =
\Delta_s+i\pi t^+ D(\rho,\beta,\omega), \label{e29c}
\end{equation}
in Born's limit and
\begin{equation}
\tilde{\Delta}_s(\rho,\beta,\omega) =
\Delta_s+i\pi \Gamma^+ {D(\rho,\beta,\omega)\over
\Omega^2(\rho,\beta,\omega)+D^2(\rho,\beta,\omega)}. \label{e30c}
\end{equation}
in the resonant scattering limit. $\Delta_s$ is the $s$-wave contribution
to the order parameter in the clean limit and is equal to zero in
pure $d$-wave symmetry. The renormalized gap function is then written as:
\begin{equation}
\tilde{\Delta}_{d+is}(\rho,\beta,\omega,\phi) =
\Delta_d(\phi)+i\tilde{\Delta}_s(\rho,\beta,\omega).
\label{e30d}
\end{equation}
Finally, the functions $D(\rho,\beta,\omega)$ and $\Omega(\rho,\beta,\omega)$
are defined as
\begin{mathletters}
\label{e31}
\begin{eqnarray}
\Omega(\rho,\beta,\omega) &=& \left\{\begin{array}{ll}
\left\langle
{\tilde{\omega}(\rho,\beta,\omega)-{\bf v}_F({\bf p}){\bf q}_s\over
\sqrt{\left(\tilde{\omega}(\rho,\beta,\omega)-{\bf v}_F({\bf p}){\bf q}_s
\right)^2-\tilde{\Delta}^2_{d+s}(\rho,\beta,\omega,\phi)}}
\right\rangle_\phi & \quad {\rm orthorhombic,}\\
\left\langle
{\tilde{\omega}(\rho,\beta,\omega)-{\bf v}_F({\bf p}){\bf q}_s\over
\sqrt{\left(\tilde{\omega}(\rho,\beta,\omega)-{\bf v}_F({\bf p}){\bf q}_s
\right)^2-\tilde{\Delta}^2_s(\rho,\beta,\omega)-\Delta^2_d(\phi)}}
\right\rangle_\phi & \quad {\rm tetragonal,}
\end{array}
\right . \label{e31a}\\
D(\rho,\beta,\omega) &=& \left\{\begin{array}{ll}
\left\langle
{\tilde{\Delta}_{d+s}(\rho,\beta,\omega,\phi)\over
\sqrt{\left(\tilde{\omega}(\rho,\beta,\omega)-{\bf v}_F({\bf p}){\bf q}_s
\right)^2-\tilde{\Delta}^2_{d+s}(\rho,\beta,\omega,\phi)}}
\right\rangle_\phi & \quad {\rm orthorhombic,}\\
\left\langle
{\tilde{\Delta}_s(\rho,\beta,\omega)\over
\sqrt{\left(\tilde{\omega}(\rho,\beta,\omega)-{\bf v}_F({\bf p}){\bf q}_s
\right)^2-\tilde{\Delta}^2_s(\rho,\beta,\omega)-\Delta^2_d(\phi)}}
\right\rangle_\phi & \quad {\rm tetragonal,}
\end{array}
\right . \label{e31b}
\end{eqnarray}
\end{mathletters}
with $\langle\cdots\rangle_\phi$ the Fermi surface average in the
{\bf p}-frame. Finally, Eq.~(\ref{e28}) still applies for the QDOS and
it is only at this later stage that an average over vortex variables
is carried out. Eqs.~(\ref{e29}) or (\ref{e30}) need to be solved
self consistently for different values of $\rho$, $\beta$, and
$\omega$. The two remaining parameters are $t^+$ or $\Gamma^+$ which are
related to the strength of impurity scattering in the normal state
where $\pi t^+ = \pi\Gamma^+ = 1/2\tau_{imp}$ and $\tau_{imp}$ is the
impurity scattering
time in the normal state. It should also be noted at this point that the
impurity scattering potential
in the laboratory frame is now anisotropic. The role
of anisotropic impurity scattering in anisotropic superconductors has been
studied by Hara\'{n} and Nagi.\cite{Nagi} According to their results the
influence of the impurity concentration on superconductivity varies with
the anisotropy of the scattering potential, qualitative changes in, say
the temperature dependence of the penetration depth or the energy
dependence of the optical conductivity, cannot be expected.
A first result with impurities is given in Fig.~\ref{f6} for Born
scattering with $t^+ = E_H/16$ in a tetragonal system with a pure
$d$-wave gap. Here, $\Delta_0 = 24\,{\rm meV}$, $\nu = 0.1$, and what is
plotted is the zero frequency value of the QDOS, $N(0,\gamma)$, as
a function of the orientation of the in-plane magnetic field {\bf H}
which makes an angle $\gamma$ with the ${\bf\hat{x}}$-axis. The solid
line includes impurities and is compared with the dotted line for the
pure case. We note that the minima are somewhat filled and rounded
by the impurities as we might have expected and the maxima are slightly
higher and broader. Impurities increase the QDOS in the low energy
region. Close examination of the two curves also shows that they both
have four-fold symmetry. This is to be expected, but it is pointed out
that, even for a tetragonal system, the introduction of impurities
introduces a finite $s$-wave component to the gap, which is now of
$(d+is)$-symmetry, in Eqs.~(\ref{e29c})
or (\ref{e30c}) because it is the self consistent Green's function
that enters the definition of (\ref{e31}) and this leads to a nonzero
value of $D(\rho,\beta,\omega)$.
Our final result remains four-fold symmetric, {\it i.e.} it
retains the full tetragonal symmetry.
The effect of impurity scattering is much larger in the unitary limit.
Results are shown in Fig.~\ref{f7}. Again the underlying system has
tetragonal symmetry and a pure $d$-wave gap. The impurity scattering
parameter is $\Gamma^+ = E_H/16$ with a gap amplitude $\Delta_0 =%
24\,{\rm meV}$ and $\nu = E_H/\Delta_0 = 0.1$ in Eqs.~(\ref{e30}).
The solid line represents our results for the value of the QDOS
$N(0,\gamma)$ as a function of the magnetic field orientation $\gamma$
in the CuO$_2$ plane. Notice the scale on the vertical axis and that
the size of $N(0,\gamma)$ is roughly three times its value for the
Born limit (Fig.~\ref{f6}). This changes quite a lot when the
frequency is increased. The dashed curve applies to
$\omega/\Delta_0 = 0.05$, the dotted one to $\omega/\Delta_0 = 0.1$,
and the dashed-dotted one to $\omega/\Delta_0 = 0.15$. The percent anisotropy
in the QDOS is reduced with increasing $\omega$.
Results for the orthorhombic case are shown in Figs.~\ref{f8} and
\ref{f9}. In these figures $\nu = E_H/\Delta_0 = 0.1$, the band
anisotropy $\alpha = 0.41$, and the $s$-wave admixture to the gap
is characterized by $s = -0.25$ taken from a previous fit to experimental
data in YBCO as discussed by Wu {\it et al.}\cite{Wu} and
Sch\"urrer {\it et al.}\cite{Schur} Reasonable values for the parameters
in our admittedly simplified band structure model, namely $\alpha =%
(m_x-m_y)/(m_x+m_y)$ and the gap anisotropy, namely $s$, can be set
in comparison with penetration depth data on YBCO at optimum doping.
The measured ratio of the zero temperature penetration depth depends
on the anisotropic effective masses and sets the value of $\alpha = 0.41$
which follows from $\lambda_a/\lambda_b = 1600\,$\AA$/1030\,$\AA.\cite{%
Schur,Maki,Modre,Bonn1} Furthermore, in our model the low temperature
slopes of the penetration depth are approximated by
\begin{eqnarray}
{d\lambda_{xx}\over dT} &\simeq& {2\ln(2)\over\Delta_0}(1-\alpha-s),
\nonumber\\
{d\lambda_{yy}\over dT} &\simeq& {2\ln(2)\over\Delta_0}(1+\alpha+s).
\end{eqnarray}
The experiments of Bonn {\it et al.}\cite{Bonn1} give $s = -0.25$ as an
estimate of the orthorhombicity in the gap. It is these values that we
have used to estimate the expected anisotropy in the QDOS for {\bf H}
in the CuO$_2$ plane. As we have already stressed in the introduction,
the band structure in YBCO is much more complex than the model treated
here but the simplicity of our model allows us to get insight into the
role of orthorhombicity and develop a qualitative picture of its
effect. There are no adjustable parameters left.
For the orthorhombic case the pattern in Fig.~\ref{f8} for $N(0,\gamma)$
vs. $\gamma$ is two-fold symmetric even in the pure case. As before,
Born scattering (solid curve with $t^+ = E_H/16$) simply smooths out
slightly the minima seen in the pure case curve (dotted line). On the
other hand, unitary scattering, solid gray curve for $\Gamma^+ =%
E_H/16$, pushes the curve up by a factor of about two over the pure
limit case and the pattern is also significantly modified and smoother.
The small maxima around $\gamma = 1.5\,{\rm rad}$ and symmetry related
points in the pure curve (dotted) are completely washed out by unitary
impurity scattering. The amplitude of the variations in value of $N(0,\gamma)$
is also considerably reduced.
The curves described so far apply to untwinned single crystals. Three
more curves are shown in Fig.~\ref{f8}. The dashed-dotted curve
applies to the case of twinned samples and was obtained from the clean
limit (dotted curve) by averaging it with a similar curve displaced by
$90^\circ$. This corresponds to a simple average of equal numbers of
twins with ${\bf\hat{x}}$- and ${\bf\hat{y}}$-axis interchanged. It is
clear that this average greatly reduces the predicted anisotropy and
experiments aimed at discovering this anisotropy are best made on
untwinned samples. This is confirmed in preliminary data from
Junod's group\cite{Junod2} who indeed find no significant in-plane
anisotropy in the specific heat on twinned single crystals. At first
sight the noise in their experiment is below, but of the order of the
anisotropy predicted in our calculations. There are many reasons why
the actual anisotropy might in fact be even less than the amount
shown in our pure twinned curve (dashed-dotted). To mention only one:
YBCO is more three dimensional than many of the high $T_c$
oxides and this implies one more integration in the definition of the
QDOS (\ref{e28}) and this can be expected to reduce the anisotropy
further. The two other curves in Fig.~\ref{f8} include impurities and
show even less anisotropy than does the pure case. The dashed curve applies
to Born scattering with $t^+ = E_H/16$ and the dashed gray curve to
resonant scattering with $\Gamma^+ = E_H/16$.
In Fig.~\ref{f9} we present finite frequency results for $N(\omega,\gamma)$
vs. $\gamma$ (the magnetic field orientation in the CuO$_2$ plane) for
four finite frequency values, namely $\omega/\Delta_0 = 0.05$
(dashed-dotted curve), $\omega/\Delta_0 = 0.1$ (dashed curve),
$\omega/\Delta_0 = 0.15$ (dash-double dotted curve), and
$\omega/\Delta_0 = 0.2$ (dotted curve). The results are to be compared
with the solid curve which applies to the case $\omega = 0$ and has
already been shown in Fig.~\ref{f7}. It is clear that the two
pronounced minima in the solid curve become shallower with increasing
$\omega$ and are already quite small for the dotted curve with
$\omega/\Delta_0 = 0.2$. This complicated pattern should be reflected
in the accompanying pattern for the temperature dependence of the
specific heat.
More insight into the frequency dependence of the anisotropy can be
obtained from the frequency dependence $N(\omega,\gamma)$ which is shown
for several fixed directions $\gamma$ in Fig.~\ref{f10}, frame (a)
and (b). These apply respectively to the tetragonal and orthorhombic
case. In frame (a) $\gamma = 0^\circ$ (antinodal direction) and
$\gamma = 45^\circ$ (nodal direction) are considered. For the pure
case the solid curve which applies for $\gamma = 0^\circ$, shows an
$\omega^2$ dependence at small $\omega$ while the dotted curve for
$\gamma = 45^\circ$ is linear as expected\cite{Vek1,Vek2} and falls
below the value along the antinodal direction. The nodal direction
is indeed the direction of the minimum in $N(0,\gamma)$ of Fig.~\ref{f6}.
As the frequency $\omega$ is increased the solid and dotted curves
rapidly come together and even cross so that the nodal direction
has a larger value of $N(\omega,\gamma)$ than the antinodal direction
for $\omega$ larger than the crossover frequency. The anisotropy is
now very small and will not be observable in specific heat experiments.
Dependencies of $N(\omega,\gamma)$ vs. $\omega$ are changed only
slightly with the introduction of Born scattering impurities; the
dotted curve is for $\gamma = 0^\circ$ and the dashed-dotted one for
$\gamma = 45^\circ$, and $t^+ = E_H/16$. On the other hand, the gray
curves for resonant scattering with $\Gamma^+ = E_H/16$ are very
different and are approximately constant at low $\omega$
with little difference between the
two directions. It is obvious that little anisotropy remains
in the QDOS. Similar results are given in frame (b) for
the orthorhombic case with $\alpha=0.41$, $s=-0.25$, and $\nu = 0.1$,
{\it i.e.} the same value of $E_H$ to gap amplitude is used as in
frame (a). The most important difference to be noted is that the
critical angle, at which an $\omega$ dependence is found in the clean
limit, is at the angle $\gamma = 61.3^\circ$ and not at $\gamma = 45^\circ$.
Also, the anisotropy is not washed out quite as much in the resonant
scattering case when compared to the tetragonal case. Further in this
case at sufficiently high frequencies the two curves for $\gamma = 0^\circ$
and $\gamma = 61.3^\circ$ cross and the curve for $\gamma = 0^\circ$
falls below the one for $\gamma = 61.3^\circ$ thus reversing the pattern
of maxima and minima shown in Fig.~\ref{f9} for lower values of
$\omega$. We also note that as $\omega$ increases out of zero the QDOS
$N(\omega,\gamma)$ decreases for both $\gamma$ values given in frame (a)
(tetragonal case) while for the orthorhombic case in frame (b) the reverse
is true. This effect is also seen in Figs.~\ref{f7} and \ref{f9}. In
the tetragonal case the solid line of Fig.~\ref{f7} for which
$\omega = 0$ is above the others for $\omega\ne 0$ while in Fig.~\ref{f9},
the orthorhombic case, it is below.
\section{Conclusions}
We have calculated the effect of the circulating vortex supercurrents on the
QDOS of a two dimensional $d$-wave superconductor. For the magnetic field
{\bf H} placed in the CuO$_2$ plane the QDOS varies as a function of the
angle $\gamma$ between {\bf H} and the ${\bf\hat{x}}$-axis. We have
included in our calculations an $a$-$b$ effective mass anisotropy of
about 2 with a view of modeling the band structure of YBCO which is
orthorhombic because of
the existence of CuO chains, and for which the charge carrier oscillator
strength on the chains is approximately of the same order as in the
CuO$_2$ planes. In addition, the gap on the Fermi surface was assumed
to posses a subdominant $s$-wave as well as a dominant $d$-wave character.
Consideration of penetration depth data fixes the value of this admixture so that
we have no adjustable parameters. Impurities are also included in our
calculations in both, Born's and unitary limits. It is found
that the four-fold pattern predicted for the dependence of the zero
frequency value of the QDOS in a pure tetragonal system is now changed
to a two-fold variation due to the orthorhombicity in the gap or in the
band structure. The introduction of impurities leaves the symmetry of
the pattern unchanged but can reduce its amplitude. It is also found that
the pattern of variation of $N(\omega,\gamma)$ with $\gamma$ is altered
as the frequency is increased, although it remains four-fold for
tetragonal systems and two-fold for the orthorhombic case.
An important result is that the angle at which minima occur in $N(0,\gamma)$
vs. $\gamma$ does not correspond to the angle at which the zeros occur in
the gap as would be the case in a tetragonal system. Instead, the
minima occur when the normal state quasiparticle Fermi velocity in the
nodal direction is perpendicular to the direction of the vortex
supercurrents, {\it i.e.} parallel to the direction of the external
magnetic field.
Specific heat measurements in clean samples are most favorable for
observing the predicted two-fold pattern of anisotropy. To observe it
in YBCO detwinned samples are best because an average over equal numbers
of twins greatly reduces the predicted anisotropy for detwinned samples.
Consideration of more realistic tight binding band structure models
are in progress. Our very simple effective mass model, however, has
allowed us to understand the important changes orthorhombicity brings
to the magnetic field orientation dependence of the QDOS for a two
dimensional CuO$_2$ plane with {\bf H} in the plane. It has helped us
to determine the experimental conditions most favorable to the
observation of the predicted anisotropies.
\section*{Acknowledgments}
We have benefited from discussions with our colleagues P.J.\ Hirschfeld,
E.\ Nicol, I.\ Vekhter, and W.C.\ Wu.
Work supported in part by the Natural Sciences and Engineering Research
Council of Canada (NSERC), the Canadian Institute for Advanced
Research (CIAR), and by
Fonds zur F\"orderung der wissenschaftlichen Forschung (FWF), Vienna,
Austria under contract No. P11890-NAW.
|
\section*{Introduction}
More than 20 years have passed since the Lee-Weinberg bound on the stable
heavy neutrino was proposed \cite{lw}. It applies to an absolutely stable
particle which needs a conserved quantum number. The decaying particle
cosmology was propsed around the same time \cite{dkt}. But the decaying
particle cosmology has been extensively applied for the gravitino
\cite{ekn}. In this talk, we focus on the decay of the lightest
supersymmetric particle (LSP), $\chi$. For $\chi$ to decay, the
R-parity must be broken.
In supergravity, there can be nonrenormalizable interactions for
the R-parity violation, but we neglect these compared to the
renormalizable ones. There can be also R-violating bilinear terms,
of the form $L_iH_2$, which can be rotated away at the superpotential
level. However, in the presence of soft terms some bilinear terms
cannot be rotated away. But for decay processes, this subtle point is
not important. Thus we consider the R-violating trilinear terms,
\begin{equation}
W=\frac{1}{2}\lambda_{ijk}L^iL^jE^{ck}+\lambda^\prime_{ijk}L^iQ^jD^{ck}
+\frac{1}{2}\lambda^{\prime\prime}D^{ci}D^{cj}D^{ck}
\end{equation}
which are allowed by (i) supersymmetry and (ii) gauge symmetry. Therefore,
it is reasonable to consider the phenomenology of R-parity violation.
With the above superpotential present, the lepton number and/or baryon
number are broken. That is the reason for requiring the
$R=(-1)^{3B+L+2S}$ conservation. Anyway, with the R-parity violation, we
expect neutrino oscillation, proton decay, and other abnormal processes.
From laboratory experiments, the single bounds on the couplings are
not very strong, e.g. $\lambda_{121}<0.05, \lambda_{122}<0.05,
\lambda^\prime_{111}<0.001,\lambda^\prime_{112}<0.02,
\lambda^{\prime\prime}_{112}<10^{-6}$, and $\lambda^{\prime\prime}_{113}
<10^{-5}$. However, a combined bound from lower limit of the
proton decay lifetime is very strong
\begin{equation}
\lambda^\prime\lambda^{\prime\prime}<10^{-24}.
\end{equation}
If a universal strength for the R-violating couplings are assumed,
then these couplings are very small. For example, if a singlet
scalar field $\phi$ carrying odd R-parity develops a VEV,
$<\phi>\equiv \epsilon M_P$, then
the R-parity is spontaneously broken. At low energy, these effects
are represented by $\lambda,\lambda^\prime,$ and $\lambda^{\prime\prime}$
couplings which carry negative R-parity. These may arise from
nonrenormalizable interactions containing $\phi$ field. In this case,
these couplings can be of the same order.
Thus, to allow reasonably large R-violating couplings, one usually
fobid $\lambda^\prime$ or $\lambda^{\prime\prime}$ completely.
In general, laboratory experiments give upper bounds on the
couplings. However, the cosmological bounds depend on the region
of couplings. It is schematically shown in Fig. 1. In this talk,
I am interested in the region $\tau_\chi\sim 10^{-6}$ s.
\begin{figure}[b!]
\centerline{\epsfig{file=kim1.eps,height=5cm, width=10.5cm}}
\vspace{10pt}
\caption{A schematic view of allowed windows for the $\chi$ lifetime.
Our interest is near the point marked by the grey upward arrow.}
\label{fig1}
\end{figure}
\section*{The Lifetime of LSP}
The neutralino is a mixture of the neutral gauginos and neutral
Higgsinos,
\begin{equation}
\chi=N_1\tilde B+N_2\tilde W+N_3\tilde H_1^0+N_4\tilde H_2^0.
\end{equation}
The decay can occur, for example, through $\chi\rightarrow Q+\tilde
q$ where the virtual $\tilde q$ transforms to $L+D^c$. For this
process, the interaction is given by
\begin{equation}
-\frac{\lambda^\prime_{ijk}}{M^2_{\tilde f}}\left(
gN_2+\frac{1}{3}g_1N_1\right)\int d^4\theta U^{aj}\chi
E^iD^{ck}_{a}+{\rm h.c.}
\end{equation}
${}$For the $\chi\rightarrow u^c+d+e^+$ decay, there exist 9 diagrams.
The possible diagrams increase very rapidly if one considers colored
final states. Kinematically excluding $tt^c$ final states, there are
1,080 diagrams. Thus, we assume the {\it universal sfermion mass}
to simplify the expression. Then the $\chi$ decay rate is given by
\begin{eqnarray}
&\Gamma_{\rm tot}=\frac{g^2}{256\pi^3}\frac{m_\chi^5}{M^4_{\tilde f}}\cdot
\Large\{\sum|\lambda_{ijk}|^2\left(\frac{1}{8}N_2^2+\frac{3}{8}
N_1^2\tan^2\theta_W\right)+\sum_{j\ne 3}|\lambda^\prime_{ijk}|^2\left(
\frac{3}{4}N_2^2+\frac{7}{12}N_1^2\tan^2\theta_W\right)\nonumber\\
&+\sum_{ik}|\lambda^\prime_{i3k}|^2\left(
\frac{3}{8}N_2^2+\frac{7}{24}N_1^2\tan^2\theta_W-\frac{1}{2}
N_1N_2\tan\theta_W\right)+\sum_{k\ne 3}|\lambda^{\prime\prime}_{ijk}|^2
N_1^2\tan^2\theta_W \Large\}
\end{eqnarray}
where the overall coefficient in front of the curly bracket is
$1.8\times 10^{15}\ {\rm s}^{-1}$ for $M_{\tilde f}=1$~TeV and
$m_\chi=30$~GeV.
\section*{Cosmological Bound on $\tau_\chi$}
\subsection*{The decoupling temperature $T_D$}
The last moment when the heavy particles are in equilibrium with
photons is the decoupling temperature $T_D$. For the neutralino,
it is determined by a competition between the universe expansion rate
and $\chi$ destruction rate. The destruction rate is a function of
sfermion mass for the $t$-channel process $\chi+\chi\rightarrow f+f^c$.
If there exists a significant Higgsino component, i.e.
$(N_1^2+N_2^2)/|N^2_{3\ {\rm or}\ 4})|>(M_Z^2/M^2_{\tilde f})^2$, then
the $s$-channel $Z$ exchange diagram dominates. It is the easiest way
to estimate the decoupling temperature. However, we also include the
sfermion exchange diagrams in the numerical estimation. Nevertheless,
the decoupling temperature does not depend on these parameters very
much. Let us introduce a decoupling factor $d$
\begin{equation}
d=\frac{m_\chi}{T_D}.
\end{equation}
The decoupling factor d is given for various parameter sets in
Ref.~\cite{kkp}. For example, $M_{\tilde f}=300$~GeV, $m_\chi=60$~GeV,
$\tan\beta=10$, and the gaugino dominated neutralino
($N_1=0.9-1$, $N_2=0.1-0.5$, and $N_3=N_4=0$) give $d=21-22$. For a wide
range of parameters, $d$ is in the range 15--30.
In particular, we note the following. Higgsino dominated $\chi$ gives
a little bit smaller decoupling temperature and hence a lower
$\chi$ density. The $t$ and $u$ channel processes are important
for $N_{3,4}>0.1$ if $M_{\tilde f}>300$~GeV. However, the decoupling
temperature is rather insensitive to $\tan\beta$. In the numerical
analyses, we further assumed the relations $m_{A,h_1,h_2}
/m_\chi=1/3$ and $\Gamma_{A,h_1,h_2}=1/500$. We note, however, that
the cross section and the decoupling temperatures are not sensitive to
these assumptions.
In the next section, we are interested in relatively light neutralino
case, which excludes the possibility of producing $t,W,Z,H$ in the
$\chi\chi$ annihilation.
\subsection*{The $\chi$ lifetime bound}
If $\chi$ decays after 1 s, it affects the nucleosynthesis in a
way of destructing the already manufactured light elements.
Note that our $\chi$ lifetime is $10^{-15}/|\lambda|^2$~s and
the cosmic time scale in the radiation dominated era is
$t=0.3 N^{-1/2}M_P/T^2$.
Requiring that $E_\chi<$~(energy
of relativistic particles), we obtain
$\tau_\chi< 4.8\times 10^{10}$~(GeV/$m_\chi)^2$~s, which gives
a $|\lambda|$ bound of order $10^{-13}$.
A more stringent bound comes from the dissociation of light elements,
in particular the D dissociation \cite{ekn,lindley}.
The $\chi$ decay products degrade very rapidly by scattering with
background photons. We are interested in the photons with
$E>2.225$~MeV. But $e^+e^-$ production from scattering on the
background radiation turned out to be important also \cite{lindley}.
In this analysis, another important temperature parameter $T_*$
is introduced. At $T_*$, the rates for the Compton scattering and the
$e^+e^-$ production are comparable. The critical photon energy $E_*$
corresponding to $T_*$ is given by $E_*T_*= $(1/50)~MeV$^2$. Then we
note that $10^{-9}$ (with respect to the photon number) photons
with $\omega>25T_*$ can be used to produce $e^+e^-$. In this case,
the $e^+e^-$ scatter off the background radiation to transfer the
energy to photons which can dissociate $D$. Therefore, the maximum
allowable lifetime $\tau_{\rm max}$ is determined by the condition that
if it decayed later than $\tau_{\rm max}$, the probability for photon
to scatter with $e$ is negligible, namely it mostly scatter off with
$D$ to dissociate it. Therefore, $\chi$ must decay before
$\tau_{\rm max}$. The equation for $\tau_{\rm max}$ is
\begin{equation}
2\frac{n_\chi}{n_e}\frac{\epsilon}{\epsilon_0}\int\sum (\epsilon_0,T)
e^{-t/\tau_{\rm max}}\frac{dt}{\tau_{\rm max}}=1
\end{equation}
where
\begin{equation}
\sum(\epsilon,T)=\int^{\omega_1}_{Q_D}\frac{\sigma_D}{\sigma_{KN}+
\sigma_{pp}}2\left(1-\frac{\omega}{\omega_m}\right)\frac{d\omega}{\omega_m}.
\end{equation}
Here $\epsilon$ is the photon energy in the fraction of $\chi$ mass,
$\omega$ is the photon frequency, $\sigma_D$, $\sigma_{KN}$, and
$\sigma_{pp}$ are the deuteron dissociation cross section, Klein-Nishina
cross section, and $pp$ production cross section, respectively. The
maximum scattered photon energy is $12\gamma T$.
We calculated $\tau_{\rm max}$ from the above formula to obtain
\begin{equation}
\tau_{\rm max}=(2.2-2.5)\times 10^6\ {\rm s}.
\end{equation}
For $\delta_B\simeq 5\times 10^{-10}$ and 60~GeV photino-like (bino-like)
neutralino, the combination of R-violating couplings is bounded as
\begin{eqnarray}
&\sum 0.12(0.05)|\lambda_{ijk}|^2
+\sum 0.31(0.07)|\lambda^\prime_{ijk}(j\ne 3)|^2
+\sum 0.04(0.04)|\lambda^\prime_{i3k}|^2\nonumber\\
&+\sum 0.23(0.12)|\lambda^{\prime\prime}_{ijk}(i<j,k\ne 3)|^2
>7.7\times 10^{-24}
\end{eqnarray}
\section*{Conclusion}
Thus the sum of the R-violating couplings is bounded from below to
a few times $10^{-12}$, the details of which depends on the
nature of the neutralino. We studied the lifetime region around
$10^6$~s in Fig.~1. Of course, our bound is most meaningful if
the R-violating couplings are of the same order. It is interesting
to note that there is a allowed lifetime window between $10^3$~s
and $10^6$~s.
\begin{figure}[b!]
\centerline{\epsfig{file=kim2.eps,height=9cm, width=10cm}}
\vspace{10pt}
\caption{The maximum allowable lifetime of $\chi$ for $M_{\tilde f}=300$~GeV,
$N_1=0.99$, $N_2=0.14$, and $N_3=N_4=0$.}
\label{fig2}
\end{figure}
|
\section{\bf Introduction.}
Let $V$ be a $(n+1)$-dimensional vector space over ${\mathbb C}$, and let ${\PP^n}=\PP(V)$:
it is possible to show (cf. \cite{JSW}) that a Tango bundle $F$ on ${\PP^n}$ is contained in the following exact sequence:
$$0\rightarrow Q(-1) \rightarrow \frac{\wedge^2V} W\otimes {\mathcal O}_{\PPn} \rightarrow F(1)\rightarrow 0;$$
here $Q$ is the quotient bundle (cf. \cite{OSS}) on ${\PP^n}$ and $W\subseteq \wedge^2 V$ is a linear subspace such that:
\begin{equation}\label{ciao}
\begin{cases}
\dim_{{\mathbb C}}\PP(W) =m-1 \\
\PP(W)\cap \GG(1,n) =\emptyset
\end{cases}
\end{equation}
where $m= {\frac{(n-2)~(n-1)} 2}$ and
$\GG(1,n)$ is the Grassmannian of the lines in ${\PP^n}=\PP(V)$: hence $W$ does not contain any decomposable bivectors.
Moore \cite{Moo} has shown that $F$ is uniquely determined by the subspace $W\subseteq \wedge^2V$ and so by a point of the variety $\GG(m-1,N-1)$, with $N= {\frac{n~(n+1)}2}$; furthermore if $W$ is invariant under the action of a group $G\subseteq\PP\GL(n+1)$ then the Tango bundle, associated to $W$, is $G$-invariant too, i.e. $G\subseteq \Sym F$.
\vskip 1 cm
\section{\bf Action of $\SL(2)$.}
Let $U$ be a $2$-dimensional vector space over ${\mathbb C}$ and let us consider the complex projective space ${\PP^n}=\PP(S^nU)$: in this way, we have a natural action of $\SL(2) = \SL(U)$ over ${\PP^n}$.
We want to find a subspace $W\subseteq\wedge^2 S^nU$, $\SL(2)$-invariant and that satisfies (\ref{ciao}). For this purpose we prove the following:
\begin{prop}\label{daphne} The decomposition of $\wedge^2S^nU$ into irreducible representations is given by $S^{2(n-1)}U\oplus S^{2(n-3)}U\oplus S^{2(n-5)}U\oplus\dots$;
moreover if
$W=S^{2(n-3)}U\oplus S^{2(n-5)}U\oplus\dots$,
then $W$ satisfies (\ref{ciao}).
\end{prop}
This proposition immediately implies that for any $n\in {\mathbb N}$, such a subspace $W$ defines a $\SL(2)$-invariant Tango bundle $F$ on ${\PP^n}$, which is
described by the exact sequence:
$$0\rightarrow Q(-1) \rightarrow S^{2(n-1)}U\otimes {\mathcal O}_{\PPn} \rightarrow F(1)\rightarrow 0.$$
Before proceeding with the proof of the proposition, we prove the following lemma:
\begin {lemma}
Let $\{ v_0,\dots,v_n\} $ be a basis of $V$ and $\omega\in\GG (1,n)\subseteq\wedge^2V$ a non-vanishing decomposable bivector, then:
$$\omega = x_{i_0,j_0}(v_{i_0}\wedge v_{j_0})+ \sum_{i+j>i_0+j_0}x_{i,j}(v_i\wedge v_j)$$
where $x_{i,j}\in{\mathbb C}$ and $x_{i_0,j_0}\neq 0$.
\end {lemma}
\noindent
\begin{remark} In order to simplify the notations, we will often write $v_{i,j}$ instead of $v_i\wedge v_j$.
\end{remark}
\begin{proof}
We proceed by induction on $n$.
For $n=1$, there is nothing to prove.
Let us suppose now $n>1$ and let $\omega =v\wedge v'$ where $v=\sum
x_i v_i$ and $v'=\sum y_i v_i$.
Let $z_{i,j}=x_iy_j-x_jy_i$ then
$$\omega=\sum^{i<j}_{i+j\ge k_0} z_{i,j}v_{i,j},$$
where $k_0=\min\{k|z_{i,j}=0 \text{ if } i+j=k\}$.
If there exist $i_0,j_0\neq 0$ such that $i_0+j_0=k$, and
$z_{i_0,j_0}\neq 0$ then, since
$z_{0,i_0}=z_{0,j_0}=0$, it easily follows $x_0=y_0=0$: thus the lemma
is true by induction.
Otherwise, if such $i_0,j_0$ do not exist, then:
$$\omega=z_{0,k}v_{0,k}+\sum^{i<j}_{i+j>k_0} z_{i,j}v_{i,j}.$$
\end{proof}
\vskip .4 cm
\begin{proof}[Proof of proposition \ref{daphne}]
\
Let $V=S^nU$ and let $\{x,y\}$ be a basis of $V$: if $v_0=x^n,\dots, v_n=y^n$, then $\{v_0\dots v_n\}$ is a basis of $V$.
The weights of $S^nU$ are $\{n, n-2,\dots,-n\}$ (cf. \cite{Ful}, pag. 146--153) and since the weights of $\wedge^2 S^nU$ are given by the sums of couples of different weights of $S^nU$, it easily follows:
$$\wedge^2S^nU = S^{2(n-1)}U\oplus S^{2(n-3)}U\oplus S^{2(n-5)}U\oplus\dots$$
Indeed if $W= S^{2(n-3)}U\oplus S^{2(n-5)}U\oplus\dots$, then $\dim_{{\mathbb C}} W=m$.
Let us prove now that $W$ does not contain any decomposable bivector, as required.
We suppose that there exists $\omega \in W\cap \GG(1,n)$, such that $\omega\neq 0$; by the previous lemma, we get:
$$\omega = x_{i_0,j_0}~v_{i_0,j_0}+ \sum_{i+j>i_0+j_0}x_{i,j}~v_{i,j}$$
where $x_{i_0,j_0}\neq 0$. We want to show that, in this case, there
exists a vector of weight $2(n-1)$ in $W$: this contradicts with the
fact that $S^{2(n-1)}U\cap W = \{0\}$.
Let
$Y=\begin{pmatrix}0 & 0\cr 1 & 0\end{pmatrix}, H=\begin{pmatrix}1 & 0\cr 0 &-1\end{pmatrix}\in {\mathcal {SL}}(2)$,
and let $\tilde Y, \tilde H$ be the corresponding endomorphisms of $\wedge^2 S^nU$.
If we suppose $v_{n+1}=0$, we have:
$$\tilde Y(v_{i,j})=(n-i)~v_{i+1,j}+(n-j)~v_{i,j+1}\qquad \text{for any }i,j=0,\dots,n.$$
Hence if $k=(2n-1)-i_0-j_0$, then $x_{i_0,j_0}~\tilde Y^{(k)}(v_{i_0}\wedge v_{j_0})=\tilde Y^{(k)}(\omega)\in W$. On the other hand it results that $ \tilde Y^{(k)}(v_{i_0,j_0})=m~v_{n,n-1}$, where $m$ is a positive integer: this implies
that $v_{n,n-1}\in W$ and since $\tilde H (v_{n,n-1})=-2(n-1)v_{n,n-1} $, we see that $W$ contains a vector of weight $2(n-1)$.
\end{proof}
\begin{remark}
Moore \cite{Moo} has shown that the Tango bundles on $\PP^4$ have all
symmetry groups isomorphic to $\PP \Or(3)$ and that $\PP\GL(5)$ acts
transitively on the moduli space of the Tango bundles ${\mathcal
M}_{\PP^4}(0,2,2)$.
In higher dimensions the situation is different: in fact, with the
help of the software Macaulay 2 \cite{Mac},
it has been possible to prove that on $\PP^5$ the generic Tango bundle
has a discrete symmetry group and that there exist Tango bundles with
the symmetry group isomorphic to ${\mathbb C}^*$ (for istance the one defined
by
$W=<v_{0,5}+5v_{2,3},v_{1,4}+3v_{2,3},v_{0,4}-2v_{1,3},v_{2,5}+v_{3,4},
v_{0,3}+3v_{1,2},2v_{2,5}-3v_{3,4}>$).
The algorithm needed to calculate the dimension of the orbit of a
subspace $W_0\subseteq\wedge^2 V$ (where $n=5$)
under the action of $\PP \GL(6)$ was communicated to the author by
G. Ottaviani. We describe the
fundamental steps of it:
\begin{enumerate}
\item Let us choose $m$ as a $(6\times 15)$-matrix whose rows represent the generators of
the subspace $W_0\subseteq\wedge^2 V$;
\item We denote by $g = \{g_{i,j}\}$ a generic $(6\times
6)$-matrix and let's define $m'=m*\wedge^2 g$:
\end{enumerate}
\noindent $m'$ represents the image $g W_0$ of the matrix $g\in \PP
\GL(6)$ by the map $\eta:\PP \GL(6)\rightarrow \GG(6,\wedge^2 V)$;
By the Plucker embedding $\phi:\GG(m,\wedge^2 V)\hookrightarrow \PP^{5004}$,
the dimension of the orbit of $W_0$ is equal to the dimension of the
ideal generated by the minors $6\times 6$ of $m'$, but its calculation
is, computationally, too difficult.
Therefore in order to make the computation easier, we first calculate the derivative $d(\phi\circ \eta)$
at the identity matrix and then we compute
the dimension of its image: this number is exactly the dimension
of the orbit. We proceed as follows:
\begin{enumerate}
\stepcounter{enumi}
\stepcounter{enumi}
\item Let $v_1(g),\dots,v_6(g)$ be the rows of $m'$, and let
$v_i(g)_{g_{i,j}}= \frac {\partial v_i(g)} {\partial g_{i,j}}$.
\end{enumerate}
\noindent In order to compute the derivative $d(\phi\circ \eta)$,
we remind that, for any $I\subseteq\{1,\dots,15\}$ such that $\#I=6$,
we have:
$$\frac \partial {\partial g_{i,j}} \det \begin{pmatrix}v_0^I(g)\cr
\vdots\cr v_6^I(g)\end{pmatrix} =
\det \begin{pmatrix}v_0^I(g)_{g_{i,j}}\cr
v_1^I(g)\cr \vdots\cr v_6^I(g)\end{pmatrix}
+ \dots +
\det \begin{pmatrix}v_0^I(g)\cr\vdots\cr
v_5^I(g) \cr v_6^I(g)_{g_{i,j}}\end{pmatrix}
$$
where $v_i^I(g)$ denotes the vector composed by the components of
$v_i(g)$ with index in $I$.
\begin{enumerate}
\stepcounter{enumi}
\stepcounter{enumi}
\stepcounter{enumi}
\item Let's define $M_{i,j}^k = \begin{pmatrix}v_1(\Id_6)\cr\vdots\cr
v_k(\Id_6)_{g_{i,j}} \cr\vdots\cr v_6(\Id_6)\end{pmatrix}$;
\item let
$p_{i,j}$ be the sum of the vectors in $\PP^{5004}$ defined by the minors
of $M_{i,j}^k$ with $k=1,\dots,6$;
\item The rank of the matrix $\begin{pmatrix}p_{1,1}\cr
p_{1,2}\cr\vdots \cr p_{6,6}\end{pmatrix}$ is the dimension of the
orbit of $W_0$.
\end{enumerate}
\end{remark}
\vskip 1 cm
\section{\bf Weighted Tango Bundles.}
We have shown that for any $n$, there exists a Tango bundle $F$ on $\PP(S^nU)$ that is invariant under the ${\mathbb C}^*$-action defined by:
$$\begin{pmatrix} t^n & & &\cr
& t^{n-2} & &\cr
& &\ddots &\cr
& & & t^{-n} \end{pmatrix}\in\PP\GL(n+1)
\qquad\text{for any }t\in{\mathbb C}^* $$
This map induces an embedding of ${\mathbb C}^*$ in $\Sym F$ and so it is possible to study the pull-backs over ${\mathbb C}^{n+1}\setminus 0$ of such bundles (cf. \cite{AO, AO1}).
Let us fix $\alpha, \gamma\in {\mathbb N}$ such that $\gamma>n\alpha$ and let
$f_0,\dots,f_n\in{\mathbb C}[x_0,\dots,x_n]$ homogeneous polynomial of degree:
$$\deg f_k = \gamma + (n -2k)~\alpha \qquad\text{for each }k=0,\dots,n $$
and without common roots.
Let $\phi=(f_0,\dots,f_n)$ and let us take into account the following diagram:
$$\begin{CD}
{\mathbb C}^{n+1}\setminus 0 @>\phi>> S^nU\setminus 0\\
@V\pi_1 VV @VV\pi_2 V \\
\PP^n && \PP^n
\end{CD}$$
According to \cite{AO, Hor}, there exists an algebraic vector bundle $F_{\alpha,\gamma}$ on ${\PP^n}$ such that $\pi_1^*F_{\alpha,\gamma}=\phi^*\pi_2^* F$. Furthermore, since $Q$ is an homogeneous bundle \cite{OSS}, there exists $Q_{\alpha,\gamma}$ such that $\pi_1^*Q_{\alpha,\gamma}=\phi^*\pi_2^* Q$.
Such a bundle is contained in the weighted Euler sequence:
\begin{equation}\label{eulpes}
0\rightarrow {\mathcal O}_{\PPn}(-\gamma)\rightarrow S^n{\mathcal U}\rightarrowQ_{\alpha,\gamma}\rightarrow 0
\end{equation}
where ${\mathcal U}={\mathcal O}_{\PPn}(-\alpha)\oplus{\mathcal O}_{\PPn}(\alpha)$.
In general, we will call {\it weighted quotient bundle of weights $\alpha$ and $\gamma$} any bundles $Q_{\alpha,\gamma}$ contained in a sequence (\ref{eulpes}).
On the other hand $F_{\alpha,\gamma}$ is contained in the exact sequence:
\begin{equation}\label{tanpes}
0\rightarrow Q_{\alpha,\gamma}(-\gamma)\rightarrow {\mathcal V}\rightarrow F_{\alpha,\gamma}(\gamma)\rightarrow 0
\end{equation}
where ${\mathcal V}=S^{2(n-1)}{\mathcal U}$ and $Q_{\alpha,\gamma}$ is the pull-back over ${\mathbb C}^{n+1}\setminus 0$ of the quotient bundle $Q$ defined by the map $\phi$.
Also in this case, we will call {\it weighted Tango bundle of weights $\alpha$ and $\gamma$} any bundles $F_{\alpha,\gamma}$ contained in the sequence (\ref{tanpes}), where $Q_{\alpha,\gamma}$ is any weighted quotient bundle of weights $\alpha$ and $\gamma$.
By these sequences, it immediately follows that $c_1(F_{\alpha,\gamma})=0$ and that $c_i(F_{\alpha,\gamma})=c_i(\alpha,\gamma)$ for any $i=2,\dots,n-1$ (i.e. the Chern classes do not depend on the map $\phi$).
\begin{prop} A weighted Tango bundle $F_{\alpha,\gamma}$ is stable if and only if $\gamma>2(n-1)\alpha$.
\end{prop}
\begin{proof} Let $\gamma>2(n-1)\alpha$. By the Hoppe criterion \cite{Ho}, it suffices to show that $\HH^0(\wedge^qF_{\alpha,\gamma})=0$ for any $q=1,\dots,n-2$.
By the sequence:
$$0\rightarrow S^{k-1}S^n{\mathcal U}(-\gamma)\rightarrow S^{k}S^n{\mathcal U}\rightarrow S^kQ_{\alpha,\gamma}\rightarrow 0$$
obtained raising the sequence (\ref{eulpes}) to the $k$-th symmetric power, we see that:
$\HH^i(S^kQ_{\alpha,\gamma}(t))=0$ for any $i=1,\dots,n-2$ and $t\in{\mathbb Z}$.
On the other hand by (\ref{tanpes}), we have the long exact sequence:
$$0\rightarrow S^qQ_{\alpha,\gamma}(-q\gamma)\rightarrow \dots \rightarrow S^kQ_{\alpha,\gamma}(-k\gamma)\otimes \wedge^{q-k} {\mathcal V}\rightarrow\dots$$
$$\dots\rightarrowQ_{\alpha,\gamma}(-\gamma)\otimes\wedge^{q-1}{\mathcal V}\rightarrow\wedge^q{\mathcal V}\rightarrow\wedge^qF_{\alpha,\gamma}(q\gamma)\rightarrow 0$$
This sequence immediately implies that $\HH^0(\wedge^qF_{\alpha,\gamma})\subseteq\HH^0(\wedge^q {\mathcal V}(-q\gamma))$, and since
$$\max\{t\in{\mathbb Z}|{\mathcal O}_{\PPn}(t)\subseteq \wedge^q {\mathcal V}(-q\gamma)\}=q((2n-q-1)\alpha-\gamma)<0$$
we have that $\HH^0(\wedge^qF_{\alpha,\gamma})=0$ for any $q=1,\dots,n-2$, and so $F_{\alpha,\gamma}$ is stable.
Let us prove now that the condition is necessary.
By the sequences:
$$0\rightarrow {\mathcal O}_{\PPn}(-3\gamma)\rightarrow S^n{\mathcal U}(-2\gamma)\rightarrow Q_{\alpha,\gamma}(-2\gamma)\rightarrow 0$$
$$0\rightarrow Q_{\alpha,\gamma}(-2\gamma)\rightarrow {\mathcal V}(-\gamma)\rightarrow F_{\alpha,\gamma}\rightarrow 0$$
it follows that if $\gamma\le 2(n-1)\alpha$, then $\HH^0(F_{\alpha,\gamma})\neq 0$ and so $F_{\alpha,\gamma}$ cannot be stable.
\end{proof}
\vskip 1 cm
\section{\bf Small deformations of $F_{\alpha,\gamma}$.}
Let $E$ be a vector bundle on ${\PP^n}$: we will indicate with $(\kur E,e)$ the Kuranishi space of $E$ (cf. \cite {FK}), where $e\in \kur E$ is the point corresponding to the bundle $E$.
We are finally ready to introduce the main result of this paper:
\begin{prop}\label{teo}
Let $F_{\alpha,\gamma}^o$ be a weighted Tango bundle of weights $\alpha$ and $\gamma$.
Every small deformation of $F_{\alpha,\gamma}^o$ is still a weighted Tango bundle and its Kuranishi space is smooth at the point corresponding to $F_{\alpha,\gamma}^o$.
\end{prop}
\noindent
\vskip 0.5 cm
Before proceeding with the proof of the proposition, let us look at some preliminaries:
\begin{lemma}\label{koonteng}
Let $Q_{\alpha,\gamma}^o$ be a weighted quotient bundle.
Every small deformation of $Q_{\alpha,\gamma}^o$ is still a weighted quotient bundle and the Kuranishi space of $Q_{\alpha,\gamma}^o$ is smooth at the point corresponding to its isomorphism class.
\end{lemma}
\begin{proof}
The proof of this lemma is very similar to the proof of prop.~3.1 of \cite{AO}.
\end{proof}
\begin{lemma}\label{l1}
Let $F_{\alpha,\gamma}$ and $F_{\alpha,\gamma}'$ be two isomorphic weighted Tango bundles, defined by the sequences:
$$0\rightarrow Q_{\alpha,\gamma}(-\gamma) \rightarrow {\mathcal V}\rightarrow F_{\alpha,\gamma}(\gamma) \rightarrow 0$$
$$0\rightarrow Q_{\alpha,\gamma}'(-\gamma) \rightarrow {\mathcal V}\rightarrow F_{\alpha,\gamma}'(\gamma) \rightarrow 0$$
where $Q_{\alpha,\gamma}$ and $Q_{\alpha,\gamma}'$ are weighted quotient bundles.
Then $Q_{\alpha,\gamma}$ and $Q_{\alpha,\gamma}'$ are isomorphic.
\end{lemma}
\begin{proof}
By joining together the sequences (\ref{eulpes}) and (\ref{tanpes}), we get:
$$0\rightarrow {\mathcal O}_{\PPn}(-2\gamma){\stackrel{\phi}\rightarrow} S^n{\mathcal U}(-\gamma)\rightarrow {\mathcal V}\rightarrowF_{\alpha,\gamma}(\gamma)\rightarrow 0.$$
By proposition 1.4 of \cite{BS} and by the fact that $-2\gamma<-\gamma-n\alpha$, the last sequence is the minimal resolution of $F_{\alpha,\gamma}(\gamma)$: hence $Q_{\alpha,\gamma}(-2\gamma)=\coker \phi$ is directly defined by this resolution.
\end{proof}
\begin{lemma} \label{l2}
Every isomorphism between two weighted Tango bundles $F_{\alpha,\gamma} \rightarrow F_{\alpha,\gamma}'$ is induced by an isomorphism of sequences:
$$\begin{CD}
0@>>> Q_{\alpha,\gamma}(-\gamma) @>>> {\mathcal V} @>>> F_{\alpha,\gamma}(\gamma) @>>> 0 \\
&& @VVV @VVV @VVV \\
0@>>> Q_{\alpha,\gamma}'(-\gamma) @>>> {\mathcal V} @>>> F_{\alpha,\gamma}'(\gamma) @>>> 0 \\
\end{CD}$$
\end{lemma}
\begin{proof}
By the sequence
$$0\rightarrow {\mathcal O}_{\PPn}(-2\gamma)\otimes {\mathcal V}\rightarrow S^n{\mathcal U}(-\gamma)\otimes {\mathcal V}\rightarrowQ_{\alpha,\gamma}(-\gamma)\otimes {\mathcal V}\rightarrow 0,$$
and since
$$h^1(S^n{\mathcal U}(-\gamma)\otimes {\mathcal V})=h^2({\mathcal O}_{\PPn}(-2\gamma)\otimes {\mathcal V})=0,$$
we get $h^1(Q_{\alpha,\gamma}(-\gamma)\otimes {\mathcal V})=0;$ hence the lemma is proven.
\end{proof}
\begin{lemma} \label{l3}
Two morphisms $f,f'\in \Hom(Q_{\alpha,\gamma}(-\gamma),{\mathcal V})$ give the same element of ${\it Quot}_{{\mathcal V}|\PP^n}$ if and only if there exists an invertible $h\in\End(Q_{\alpha,\gamma}(-\gamma))$ such that
$$\ f=f'\circ h.$$
\end{lemma}
\begin{proof}
It follows from the definition of ${\it Quot}_{{\mathcal V}|\PP^n}$, (cf. \cite{Hu}).
\end{proof}
\
\begin{proof}[Proof of proposition \ref{teo}]
\
For brevity's sake, we will write $\widetilde F_o$ instead of $F_{\alpha,\gamma}^{o} $ and $\widetilde Q_o$ for $Q_{\alpha,\gamma}^{o}$.
Let also $\sigma_0\in \Hom(\widetilde Q_o(-\gamma),{\mathcal V})$ be such that $\widetilde F_o=\coker \sigma_0$.
Let ${\mathcal Q}$ be the sub-variety of the irreducible component of ${\it Quot}_{{\mathcal V}|\PP^n}$ composed by all the quotients of the maps $0\rightarrow Q_{\alpha,\gamma}(-\gamma){\stackrel {\sigma}\longrightarrow} {\mathcal V}$ for some weighted bundle $Q_{\alpha,\gamma}$ and containing the point $\sigma_0$ corresponding to $\widetilde F_o$: the morphisms $\Phi:({\mathcal Q},\sigma_0)\longrightarrow(\kur {\widetilde Q_o},q_0)$ and $\Psi:({\mathcal Q},\sigma_0)\longrightarrow (\kur {\widetilde F_o},f_0)$ are canonically defined.
A generic fiber of $\Phi$ is given by all the cokernels of the morphisms $Q_{\alpha,\gamma}(-\gamma)\rightarrow {\mathcal V}$ with a fixed $Q_{\alpha,\gamma}$, and so, by lemma \ref{l3}, its dimension is constantly equal ($\alpha$ and $\gamma$ are fixed) to
$h^0(Q_{\alpha,\gamma}^*(\gamma)\otimes {\mathcal V})-h^0(\End Q_{\alpha,\gamma})$.
Hence, since lemma \ref{koonteng} implies that $\dim_{q_0}(\kur {\widetilde Q_o})=h^1(\End \widetilde Q_o)$, we get:
$$\dim_{\sigma_0}{\mathcal Q}=h^0(\widetilde Q_o^*(\gamma)\otimes {\mathcal V})-h^0(\End \widetilde Q_o)+ h^1(\End \widetilde Q_o)$$
Let us study now the morphism $\Psi:\mathcal Q \longrightarrow \kur {\widetilde F_o}$:
if $\Sigma=\{ \sigma\in {\it Quot}_{{\mathcal V}|\PP^n}| F_{\sigma} \simeq \widetilde F_o \}$,
then it results $\Psi^{-1}(f_0)\subseteq \Sigma$ and by lemma \ref{l1}, \ref{l2} and \ref{l3}, it follows:
$$\dim_{\sigma_o}\Sigma = h^0(\End {\mathcal V})-\dim \{\varphi \in \End {\mathcal V}|\varphi\cdot\sigma_0 =\sigma_0 \}-h^0(\End \widetilde Q_o).$$
By the sequence:
$$0\rightarrow\widetilde F_o^* (-\gamma)\otimes {\mathcal V} \rightarrow \End {\mathcal V} \rightarrow \widetilde Q_o^*(\gamma)\otimes {\mathcal V} \rightarrow 0$$
obtained tensoring the dual sequence of (\ref{tanpes}) with ${\mathcal V}$, we have that:
$$\dim \{\varphi \in \End {\mathcal V}|\varphi\cdot\sigma_0 =\sigma_0\}=h^0(\widetilde F_o^*(-\gamma) \otimes {\mathcal V})$$
and so:
$$\dim_{\sigma_0} \Psi^{-1}(f_0)\le \dim_{\sigma_o}\Sigma = h^0(\End {\mathcal V})-h^0(\widetilde F_o^* (-\gamma)\otimes {\mathcal V})-h^0(\End \widetilde Q_o).$$
Hence:
$$h^1(\End \widetilde F_o)\ge \dim_{f_0}(\kur {\widetilde F_o})\ge h^1(\End \widetilde Q_o)+h^1(\widetilde F_o^* (-\gamma)\otimes {\mathcal V}).$$
To prove the proposition it suffices to show that $$h^1(\End \widetilde F_o)\le h^1(\End \widetilde Q_o)+h^1(\widetilde F_o^* (-\gamma)\otimes {\mathcal V}).$$
In fact this implies that $h^1(\End \widetilde F_o)=\dim_{f_0}(\kur {\widetilde F_o})$, i.e. $\kur {\widetilde F_o}$ is smooth at the point $f_0$, and that $\dim_{f_0}(\kur {\widetilde F_o})= \dim_{\sigma_0}{\mathcal Q}-\dim \Psi^{-1}(f_0)$, i.e. $\Psi$ is surjective.
By the exact sequence:
$$0\rightarrow \widetilde Q_o(-2\gamma)\otimes \widetilde F_o^*\rightarrow \widetilde Q_o(-\gamma)\otimes {\mathcal V}\rightarrow \End \widetilde Q_o\rightarrow 0$$
and by the vanishing of $\HH^1(\widetilde Q_o(-\gamma)\otimes {\mathcal V})$ and $\HH^2(\widetilde Q_o(-\gamma)\otimes {\mathcal V})$, we have that $\HH^1(\End \widetilde Q_o)=\HH^2(\widetilde Q_o(-2\gamma)\otimes \widetilde F_o^*)$.
Hence by the sequence:
$$0\rightarrow \widetilde Q_o(-2\gamma)\otimes \widetilde F_o^*\rightarrow \widetilde F_o^*(-\gamma)\otimes {\mathcal V}\rightarrow \End \widetilde F_o\rightarrow 0$$
and for what we have seen, we get the sequence of cohomology groups:
$$\dots\rightarrow \HH^1(\widetilde F_o^*(-\gamma)\otimes {\mathcal V})\rightarrow\HH^1(\End \widetilde F_o)\rightarrow \HH^1(\End \widetilde Q_o)\rightarrow\dots$$
In particular
$h^1(\End \widetilde F_o)\le h^1(\End \widetilde Q_o)+h^1(\widetilde F_o^* (-\gamma)\otimes {\mathcal V})$, as required.
\end{proof}
\vskip 1 cm
Theorem \ref{princess} easily follows from the previous proposition.
In fact if $\gamma\ge 2(n-1)\alpha$, we can consider the canonical algebraic map $\mathcal Q\longrightarrow \mathcal M(0,c_2,\dots,c_{n-1})$. The image of this map is a smooth quasi projective set composed uniquely by weigthed Tango bundles and it is an open neighborhood of $F_{\alpha,\gamma}^o$ in $\mathcal M(0,c_2,\dots,c_{n-1})$.
\vskip 2 cm
|
\section{Introduction}
Quasi-stellar objects (QSOs or quasars) are valuable probes
of the high-redshift Universe (Schneider 1998). Their
most distant representatives are now measurable out to
redshifts of $z\sim 5$ (Schneider, Schmidt \& Gunn 1991,
Sloan Digital Sky Survey press release 1998).
In Big Bang cosmologies, these redshifts correspond to times
when the Universe itself was just $\sim$1~Gyr old (see Fig. 1).
\begin{figure}[h]
\plotfiddle{figure1.ps}{3.2in}{0.0}{55.0}{55.0}{-225.0}{-478.0}
\end{figure}
\begin{quotation}
\noindent Fig. 1 --- Redshift versus age of the Universe
in Big Bang cosmologies. The three solid curves correspond to
$H_o = 65$~km~s$^{-1}$~Mpc$^{-1}$ and $\Omega_{\Lambda} = 0$ with
$\Omega_{M} = 0$, 0.3 or 1. The dotted curve corresponds
to $\Omega_{\Lambda} = 0.7$ and $\Omega_{M} = 0.3$.
The ``error'' bars show the range of ages possible for $H_o$ between
50 and 80~km~s$^{-1}$~Mpc$^{-1}$ (see Carroll \& Press 1992).
\end{quotation}
Understanding the elemental abundances in these distant,
early-epoch environments is a major goal of quasar research.
Some of the first spectroscopic studies noted simply that quasar
environments contain the usual array of ``metals''
(elements C, N, O and heavier) produced by stellar
nucleosynthesis (Shklovskii 1965, Burbidge \& Burbidge 1967).
More quantitative estimates of the abundances
came later from theoretical work on the broad
emission lines, culminating in the important
review by Davidson \& Netzer (1979 --- hereafter DN79,
also Baldwin \& Netzer 1978, Shields 1976). Those studies inferred
solar or slightly higher metal abundances, with large uncertainties.
The past two decades have seen considerable progress.
Today we have a better theoretical understanding of quasar
environments, and greater abilities to both observe and
model a range of abundance diagnostics.
We also have renewed motivation from the growing
evidence that links quasars to galaxies. See, for example, Kormendy
{\it et~al.} (1998), Magorrian {\it et~al.} (1998) for black hole--host galaxy
mass correlations, Chatzichristou, Vanderriest \& Jaffe (1999),
Hines {\it et~al.} (1999), McLeod, Rieke \& Storrie-Lombardi (1999),
Boyce, Disney \& Bleaken (1999), McLure {\it et~al.} (1998),
Aretxaga, Terlevich \& Boyle (1998), Carballo {\it et~al.} (1998),
Bahcall {\it et~al.} (1997), Miller, Tran \& Sheinis (1996),
McLeod \& Rieke (1995) for direct observations
of QSO hosts, Cavaliere \& Vittorini (1998),
Shaver {\it et~al.} (1998), Terlevich \& Boyle (1993),
Boyle \& Terlevich (1998), Osmer (1998)
for arguments based on QSO number-density evolution,
McCarthy (1993), Saikia \& Kulkarni (1998), Haas {\it et~al.} (1998),
Brotherton {\it et~al.} (1998a) for radio galaxy--radio quasar unification
schemes, and Turner (1991), Haehnelt \& Rees (1993),
Loeb \& Rasio (1994), Katz {\it et~al.} (1994), Haehnelt, Natarajan
\& Rees (1998), Haiman \& Loeb (1998), Taniguchi, Ikeuchi
\& Shioya (1999) for theoretical
links between QSOs and galaxy evolution.
If quasars reside, as expected, in galactic nuclei
or dense proto-galactic clumps, their abundances could yield
unique constraints on the evolution of those environments.
For example, quasar abundances can indirectly probe the
star formation that came before QSOs, possibly the first
stars forming in massive collapsed structures after the Big Bang.
Other studies of high-redshift galaxies and metal enrichment,
involving, for example, the ``Lyman-break'' objects (Steidel
{\it et~al.} 1998, Connolly {\it et~al.} 1997) or the
damped-\hbox{Ly$\alpha$}\ or \hbox{Ly$\alpha$}\ ``forest'' absorbers
in QSO spectra (Pettini {\it et~al.} 1997, Lu, Sargent \& Barlow 1998,
Rauch 1998), probe more extended structures.
The quasar results should therefore provide an important
piece to the overall puzzle of high-redshift star formation
and galaxy evolution.
Here we review the status and implications of quasar abundance
work. We regret that many interesting related topics must be
excluded; in particular, we will consider the quasars themselves
to be simply light sources surrounded by emitting and absorbing
gas. We discuss three abundance diagnostics that are
readily observable in QSOs at all redshifts: the broad emission
lines (BELs), the broad absorption lines (BALs) and the
intrinsic narrow absorption lines (NALs). We include
just these ``intrinsic'' spectral features to probe the
abundances near QSO engines ---
excluding measures of the extended host galaxies, nearby
cluster galaxies or cosmologically intervening gas.
We begin with separate discussions of each abundance probe (\S\S2--3),
followed by a summary of the overall results (\S4).
We then consider the plausible enrichment schemes, making a case
for normal chemical evolution by stars in galactic
nuclei (\S5). Within that scheme, we use results from galactic
studies (\S6) to derive further implications of the QSO abundances
(\S7). We close with a brief outline for future work (\S8).
In several sections below we will present results of
photoionization calculations performed with the
numerical code Cloudy (version 90.05, Ferland {\it et~al.} 1998).
This code is freely available on the world wide web
(http://www.pa.uky.edu/$\sim$gary/cloudy/). Finally, we will define
solar abundances according to the meteoritic results in
Grevesse \& Anders (1989).
\newpage
\section{Emission Line Diagnostics}
\subsection{Overview}
Quasars are surprisingly alike in their emission-line spectra
(Osmer \& Shields 1999 and refs. therein);
for example, the range of intensity ratios is far less than
in galactic nebulae. Figure 2
shows a composite UV spectrum that is fairly typical of QSOs
without strong BALs. The object-to-object similarities span the
full range of QSO redshifts, $0.1\la z\la 5$, more than 4 orders
of magnitude in luminosity, and billions of years in cosmological
look-back time. The emission lines are either insensitive
to the metal abundances, or QSOs have similar abundances across
enormous ranges in other parameters. We will argue that
the truth involves a bit of both explanations.
We will focus on the BELs in the rest-frame UV because they are
present and relatively easy to measure in all QSOs at all redshifts.
Furthermore, unlike the narrow emission lines, there is no ambiguity
about their close physical connection to QSO engines (DN79).
\begin{figure}[h]
\plotfiddle{figure2.ps}{2.9in}{0.0}{75.0}{75.0}{-235.0}{-198.0}
\end{figure}
\begin{quotation}
\noindent Fig. 2 --- Normalized mean spectrum of 13 QSOs at $z>4$
(from Shields {\it et~al.} 1997). Prominent BELs are labeled.
\end{quotation}
\subsection{Origin of the Broad Emission Lines}
Quasar emission-line research is an example of the ``inverse problem''
in astrophysics. We know the answer --- the observed spectrum of a
quasar, and we are trying to understand the question --- the conditions
that created it. Any model of the line-forming regions will
have uncertainties related to uniqueness, but these can be
minimized by considering the astrophysical context and by
limiting the models to essential properties. The essential
properties of the BEL region (BELR) are as follows:
1) The BELR is photoionized.
The main evidence for photoionization is that the
emission-line spectra change in response to changes in
the continuum, with lag-times corresponding to characteristic
radii of the BELR (Peterson 1993). The shape of the
ionizing continuum is a fundamental parameter and is in
itself an area of active
research (e.g. Zheng {\it et~al.} 1997; Korista, Baldwin \& Ferland 1997a,
Brunner {\it et~al.} 1997, Laor 1998).
We will present calculations using simple power-laws between
1~\hbox{$\mu$m}\ and 100~keV, and describe results that do not depend
strongly on the continuum shape.
2) The BELR spans a range of distances from the central
object. The line variability or ``reverberation'' studies just mentioned
find different lag-times for different ions. Highly ionized species
tend to lie closer to the continuum source. Overall,
the radial distances scale with luminosity, such that
$R\approx 0.1(L/10^{46} {\rm ergs\ s}^{-1})^{1/2}$~pc is a typical
value (Peterson 1993).
3) The BELR has a wide range of densities and ionization states.
The range in ionization follows simply from the lines detected,
from OI~$\lambda$ 1303 to at least NeVIII~$\lambda$ 774 (Hamann {\it et~al.}
1998). The range in density comes mainly from
the estimated radii and photoionization theory (e.g.
Ferland {\it et~al.} 1992). Clouds\footnote{We use the term
``cloud'' loosely, referring to some localized part of the
BELR but not favoring any particular model or geometry (see Arav
{\it et~al.} 1998, Mathews \& Capriotti 1985).} with densities from $10^8$
to $>$$10^{12}$~\hbox{cm$^{-3}$}\ may be present. Any given object could
have a broad mixture of BELR properties (Baldwin {\it et~al.} 1995, 1996).
4) The BELR probably has large column densities.
Large columns, typically $N_H\ga 10^{23}$~\hbox{cm$^{-2}$} ,
were originally used in BELR simulations
to produce a wide range of ionizations in single clouds
(Kwan \& Krolick 1981 --- hereafter KK81,
Ferland \& Persson 1989). These large columns might
not apply globally because we now know that different lines
form in different regions. In our calculations below, we will
truncate the clouds at the hydrogen recombination front,
with the result that different clouds/calculations can have
different total column densities. However, the
truncation depths are in all cases large enough to
include the full emission regions of the relevant lines.
5) Thermal velocities within clouds are believed to dominate the
local line broadening and radiative transfer. The observed
line-widths are thus due entirely to bulk motions of the gas.
This issue is important because {\it i)}
continuum photoexcitation (``pumping'') can overwhelm other
excitation processes if the local line broadening
(e.g. micro-turbulence) is large, and {\it ii)} the line optical depths
and thus photon escape probabilities (see below) vary inversely
with the amount of line broadening. The interplay between these
factors makes it hard to predict the behavior of a given line without
explicit calculations. Shields, Ferland \& Peterson
(1995) plot some examples for the particular case of low column
density clouds. One argument against significant
micro-turbulence involves the \hbox{Ly$\alpha$} /\hbox{H$\beta$}\ intensity ratio.
Simple recombination theory predicts a ratio of about 34
(Osterbrock 1989 --- hereafter O89)
while the observed value is far smaller, closer to 10 (Baldwin
1977a). This discrepancy is worsened by micro-turbulence
(Ferland 1999). The solution probably requires severely
trapped \hbox{Ly$\alpha$}\ photons resulting from large optical depths at
thermal line widths (see also Netzer {\it et~al.} 1995).
\subsection{Strategies for Abundance Work.}
There is much that is unknown about QSO line-forming regions.
We do not, for example, have a clear picture of the overall
geometry or the spatial variations of key parameters;
but we do not need this information for abundance work.
The emission lines from photoionized clouds are controlled
fundamentally by the energy balance and microphysics.
The strategy for abundance studies is to identify line
ratios that have significant abundance sensitivities and minimal
dependences on other unknown or uncertain parameters.
For example, we can minimize the sensitivity to
large-scale geometric effects by comparing lines that form
as much as possible in the same gas.
Detailed simulations are often needed to identify
useful line ratios and quantify their parameter sensitivities.
Simple analytic expressions can be used for some
applications and they can help, in any case, provide physical
insight into the emission-line behaviors.
Below we review some of the basic principles of photoionization
and emission-line formation. See O89 and Mihalas (1978) for
further reviews, Davison \& Netzer (1979 --- hereafter DN79),
KK81, Ferland \& Shields 1985, and Netzer (1990 --- hereafter N90)
for applications to QSOs, and Ferland {\it et~al.} (1998) for
more on the numerical simulations and input atomic data.
\subsection{Basics of Abundance Analysis}
\subsubsection{Collisionally-Excited Lines}
Collisionally-excited lines form by the internal
excitation of an ion following electron impact. Their emissivities,
or energy released per unit volume and time, follow from the
statistical equilibrium of the energy levels. For example, the
equilibrium (detailed balance) equation for a 2-level atom is,
\begin{equation}
n_l n_e q_{lu}\ =\ n_u(\beta A_{ul}+ n_e q_{ul}) \ \ \ \
[{\rm cm}^{-3}\ {\rm s}^{-1}]
\end{equation}
where $n_e$ is the electron density,
$\beta$ is the probability for line photons
escaping the local region ($0\leq\beta\leq1$), $A_{ul}$ is the
spontaneous decay rate, $n_u$ and $n_l$ are the number densities
in the upper and lower states, and $q_{lu}$ and
$q_{ul}$ are the upward and downward collisional rate coefficients,
respectively. Note that $\beta\sim\tau^{-1}$ when $\tau\gg 1$, where
$\tau$ is the line-center optical depth (Frisch 1984).
For most applications the ions are mainly in their ground state
and $n_l$ is approximately the ionic density. The line
emissivity is,
\begin{equation}
\epsilon_{coll}\ =\ n_u\beta A_{ul} h\nu_o\ =\ n_l\beta A_{ul} h\nu_o
\left({{n_e q_{lu}}\over{\beta A_{ul}+n_e q_{ul}}}\right)
\ \ \ \ [{\rm ergs\ cm}^{-3}\ {\rm s}^{-1}]
\end{equation}
where $\nu_o$ is the line frequency.
This emissivity has a strong temperature dependence because
$q_{ul}\propto T^{-1/2}$ and $(q_{lu}/q_{ul}) =
(g_u/g_l)\exp{(-h\nu_o/kT)}$, where $g_u$ and $g_l$ are the
statistical weights. In the high
density limit we have,
\begin{equation}
\epsilon_{coll}\ =\ n_l\beta A_{ul} h\nu_o
\,{{g_l}\over{g_u}}
\,\exp{\left(-{{h\nu_o}\over{kT}}\right)}
\end{equation}
and the levels are said to be ``thermalized.'' Line thermalization,
where $\epsilon_{coll}$ no longer depends on the transition strength,
additionally requires $\tau\gg 1$. ($A_{ul}$ and $\tau$ are
both proportional to the oscillator strength, which
therefore drops out of the factor $\beta A_{ul}\approx
A_{ul}/\tau$ in Eqn. 3 if $\tau\gg 1$.)
At low densities we have,
\begin{equation}
\epsilon_{coll}\ =\ n_l n_e q_{lu} h\nu_o\ \propto\
n_l n_e T^{-1/2}\,\exp{\left(-{{h\nu_o}\over{kT}}\right)}
\end{equation}
Note that $\epsilon_{coll}$ scales here like the density squared,
compared to the linear dependence in Equation 3.
The critical density, $n_{crit}$, between these two limits is the
density where the two terms in the denominator of Equation 2 are equal,
\begin{equation}
n_{crit}\ =\ {{\beta A_{ul}}\over{q_{ul}}}\approx
{{A_{ul}}\over{\tau q_{ul}}}\
\end{equation}
where the approximate relation holds only if $\tau\gg 1$.
Physically $n_{crit}$ is the density where the upper level is as likely
to be de-excited by collisions as by radiative decays. Note that
significant optical depths have the effect of
lowering $n_{crit}$. Also note that
transitions with very different oscillator strengths (but similar
collision strengths) will have similar $n_{crit}$ in
the limit $\tau\gg 1$ (because $A_{ul}/\tau$ is independent of
oscillator strength).
\subsubsection{Recombination Lines}
The most prominent recombination lines belong to HI, HeI and
HeII, with HI \hbox{Ly$\alpha$}\ being typically strongest. These lines form
by the capture of free electrons into excited states, followed
by radiative decays to lower states. In
the simplest case, where every photon escapes freely and
competing processes are unimportant, the emissivity is,
\begin{equation}
\epsilon_{rec}\ =\ n_{i}\,n_e\,\alpha_{rad}\,h\nu_o\ \propto\
n_{i}\,n_e T^{-1} \ \ \ \
[{\rm ergs\ cm}^{-3}\ {\rm s}^{-1}]
\end{equation}
where $\alpha_{rad}$ is the radiative recombination coefficient
into the upper energy state and $n_{i}$ is the number density of
parent ions. The temperature dependence is approximate and
derives from $\alpha_{rad}$ (see O89).
\subsubsection{Deriving Abundance Ratios}
These two types of lines can be combined to form three
types of ratios for abundance analysis.
The general idea is that for any element $a$ in ion stage $i$,
the observed line
intensity, $I(a_i)$, is proportional to the density in that
ion, $n(a_i)$, times a function of the overall gas
density and temperature, $F(a_i,T,n)$, such that
$I(a_i) = n(a_i) F(a_i,T,n)$. The ionic abundance ratios
are then given by,
\begin{equation}
{{n(a_i)}\over{n(b_j)}}\ =\ {{I(a_i)}\over{I(b_j)}}\, {{F(b_j,T,n)}\over{F(a_i,T,n)}}
\end{equation}
Abundance studies require line pairs for which the ratio of the two
functions $F$ is nearly constant or has limiting behaviors
that still allow for abundance constraints.
The last step is to convert the ionic abundances
into elemental abundances, which we express logarithmically
relative to solar ratios as\footnote{Our notation here
is based on the usual definition of logarithmic abundances
normalized to solar ratios, $[a/b]\equiv \log(a/b) - \log(a/b)_{\odot}$.},
\begin{equation}
\left[{a\over b}\right]\ = \
\log\left({{n(a_i)}\over{n(b_j)}}\right)\ +\
\log\left({{f(b_j)}\over{f(a_i)}}\right)\ +\
\log\left({b\over a}\right)_{\odot}
\end{equation}
where $f(a_i)$ is the fraction of element $a$ in ion stage $i$,
etc. The middle term on the right hand side is the ionization
correction ($IC$), which can be deduced from numerical simulations
or set to zero (in the log) based on the similarity of the species
(Peimbert 1967). Another strategy is to compare summed
combinations of lines from different ion stages so that
$IC$ tends to zero on average (Davidson 1977).
Ratios of pure recombination lines are simplest
because they are least sensitive to the temperature and
density. In principle, we could derive the He/H abundance from
these ratios. However, in practice, all of the strong
HI and HeI recombination lines in QSOs,
most notably \hbox{Ly$\alpha$} , are affected by collisions
and thermalization effects. Moreover, because H$^o$,
He$^+$ and He$^{+2}$ have different ionization energies,
they need not be co-spatial in the BELR and their levels
of ionization depend on the different numbers of
photons available to produce each ion (Williams 1971).
As a result, the H and He recombination spectra are most
useful as indicators of the shape of the ionizing continuum
(e.g. Korista {\it et~al.} 1997a). We do not expect substantial
deviations from solar He/H abundances anyway, based on
normal galactic chemical evolution, and the BEL
data are grossly consistent with that expectation.
The second possible ratio involves collisional to recombination
lines. These ratios have strong temperature dependences (compare
Eqns. 3 and 4 to Eqn. 6). Nonetheless, they can still be used
for abundance work if the temperature sensitivities are
quantified by explicit calculations. For example, there is an upper
limit on the line ratio NV~$\lambda$ 1240/HeII~$\lambda$ 1640 related to the
maximum temperature attained in photoionized BELRs.
That upper limit sets a firm lower limit on the N/He abundance
(\S2.6.3 below).
The last ratio, and the one most often used, involves two
collisionally-excited lines. Roughly a dozen collisionally-excited
BELs are routinely measured in the UV spectra of quasars, so there
are a variety of possibilities. The ideal collisionally-excited line
pair would have similar excitation energies, so their
ratio has a small $h\Delta\nu_o /kT$ and thus a small temperature
dependence (Eqns. 3 and 4).
Similar values of $n_{crit}$ and similar ionization energies
further minimize the sensitivities to density and BELR structure.
Well-chosen ratios that meet these criteria can sometimes
provide abundance estimates without recourse
to detailed simulations (e.g. Shields 1976, \S2.6.1 below)
\subsection{Photoionization Simulations}
A photoionized cloud is essentially a large-scale fluorescence
problem. Energy comes into the cloud via continuum radiation,
is converted into kinetic energy by the photo-ejection of
electrons, and then leaves the cloud by various emission processes
-- mainly line radiation. The lines are thus the primary coolants;
their total intensity depends on energy conservation and not
at all on particular cloud properties.
In general situations, e.g. dense environments like
BELRs, individual line strengths can be governed by a number
of competing processes and by feedback related to the
cloud structure and energy balance. Detailed
calculations are needed to simultaneously consider a complex
network of coupled processes. Here we describe some basic
results for the line formation and ionization structure
in realistic BELR clouds.
\subsubsection{Parameters of Photoionization Equilibrium}
The fundamental parameters in photoionization simulations
are the shape and intensity of the ionizing continuum, and
the gas' space density, column density, and chemical composition.
The flux of hydrogen-ionizing photons at the illuminated
face of a cloud is,
\begin{equation}
\Phi(H)\ \equiv\ \int_{\nu_{LL}}^{\infty}{{f_{\nu}}\over{h\nu}}
\,d\nu \ \ \ \ [{\rm photons\ cm}^{-2}\ {\rm s}^{-1}]
\end{equation}
where $f_{\nu}$ is the energy flux density and $\nu_{LL}$ is
the frequency corresponding to 1 Ryd. A dimensionless ionization
parameter $U\equiv\Phi(H)/cn_H$ is often used instead, where $c$
is the speed of light and $n_H$ is the total hydrogen density
(H$^o$ + H$^+$).
$U$ is proportional to the level of ionization and has the
advantage of stressing homology relations between clouds with
the same $U$ but different $\Phi(H)$ and $n_H$. This simplification
is appropriate if we are interested in just the gross
ionization structure or in emission lines that are not
collisionally suppressed. More generally, we can use either
$\Phi(H)$ or $U$ as long as the density is also specified.
\subsubsection{A Computed Structure}
Figure 3 shows the ionization structure of a typical BELR cloud
photoionized by a power-law spectrum with $\alpha = -1.5$,
where $f_{\nu}\propto \nu^{\alpha}$.
The hydrogen recombination front occurs at a depth of
$\sim$10$^{12}$~cm, while the He$^{+2}$--He$^+$ front is near
10$^{11}$~cm. Note that there is significant ionization
beyond the nominal H$^o$--H$^+$ front, due to penetrating X-rays
and Balmer continuum photoionizations out of the $n=2$ level
in H$^o$ (KK81). Some important low-ionization lines like
FeII form in that region. The ionization fractions in plots like
Figure 3 help us identify ions,
such as O$^{+5}$, N$^{+4}$ and He$^{+2}$, that are roughly
co-spatial and thus good candidates for abundance comparisons.
\begin{quotation}
\centerline{XXXXXX INSERT FIGURE HERE XXXXXX}
\noindent Fig. 3 --- Ionization structure for a
nominal BELR cloud with $n_H = 10^{10}$~\hbox{cm$^{-3}$} , $\log U = -1.5$ and
solar abundances.
\end{quotation}
\subsubsection{An Example: the CIV~$\lambda$ 1549 Equivalent Width}
CIV~$\lambda$ 1549 is one of the strongest collisionally-excited lines
in quasar spectra. The left panel of Figure 4 shows how its
predicted equivalent width changes with the
density ($n_H$) and ionizing flux ($\Phi(H)$, see Korista {\it et~al.}
1997b for many more similar plots). Powerful selection effects are
clearly at work; the line radiates efficiently over just a
narrow range of parameters. Varying $\Phi(H)$ is equivalent to
moving the cloud closer or farther from the continuum source.
The line is weak at large values of $\Phi(H)$, because
carbon is too highly ionized, and at low values of $\Phi(H)$,
because carbon is too neutral. The line strength also changes with
the gas density. When the density is above $n_{crit}$, the line is
collisionally suppressed and other permitted lines take over
the cooling. When the density is low,
the line weakens as the many forbidden and semi-forbidden lines
become efficient coolants and the gas temperature declines. The line
is most prominent at $n_H\approx 10^{10}$~\hbox{cm$^{-3}$}\ and $\log U\approx -1.5$,
which are the canonical BELR parameters deduced over twenty
years ago from analysis of the CIV emission (DN79).
\begin{quotation}
\centerline{XXXXXX INSERT FIGURE HERE XXXXXX}
\noindent Fig. 4 --- Predicted equivalent width (EW)
of CIV~$\lambda$ 1549 as a function
of the cloud density, $n_H$, and incident ionizing
flux, $\Phi(H)$. The equivalent
width here is dimensionless (line flux/$\nu_o f_{\nu_o}$ in the continuum)
and applies for the hypothetical case of global covering factor
unity. Flux ratios for NV~$\lambda$ 1240/HeII~$\lambda$ 1640 and NV/CIV are
also shown. Other parameters are the same as Fig. 3.
\end{quotation}
It is important to remember that these selection effects exist
whenever we observe an emission line. Baldwin {\it et~al.} (1995)
showed that a typical quasar BEL spectrum might result simply
from selection effects operating in BELRs with
a wide range of cloud properties (e.g. density and distance
from the QSO). Numerical simulations can identify pairs of
lines with similar selection behaviors so that their ratios
are insensitive to the ranges or specific values of the parameters.
\subsubsection{Line Dependence on Continuum Shape}
Figure 5 shows a series of calculations using different
incident spectral shapes.
The actual shape of the ionizing continua
in QSOs is a complicated issue, but the UV to X-ray slopes
are roughly consistent with $\alpha\sim -1.5$, near the center
of the range shown (see Laor 1998,
Korista {\it et~al.} 1997a for recent discussions). The results
in Figure 5 mainly reflect the conservation of energy in the cloud.
Harder spectra (less negative $\alpha$) provide more heating per
photoionization, leading to higher temperatures. The increased
heating requires more line cooling via
collisionally-excited lines like CIV. The ratio of a
collisionally-excited line to a recombination line,
such as CIV/\hbox{Ly$\alpha$} , is proportional to the cooling per recombination
or equivalently the heating per photoionization (DN79).
Such ratios therefore have a strong continuum-shape dependence.
The strengths of collisionally-excited lines relative to the
adjacent continuum (i.e. their equivalent widths) also depend
on the spectral slope because of the temperature sensitivity
and because the continuum below the lines might be very
different from that controlling the ionization. Ratios of
collisionally-excited lines, such as NV/CIV, can similarly
depend on the spectral shape if their ionization or
excitation energies are different. In dense BELRs, these
simple behaviors can be moderated by other effects.
For example, the \hbox{Ly$\alpha$}\ equivalent width increases
with spectral hardening at fixed $U$ (Fig. 5) because it
has a significant collisional (temperature-sensitive)
contribution.
\begin{quotation}
\centerline{XXXXXX INSERT FIGURE HERE XXXXXX}
\noindent Fig. 5 --- Predicted line flux ratios, gas
temperatures ($T_4 = T/10^4$~K in the O$^{+2}$ zone,
i.e. weighted by the O$^{+2}$
fraction), and dimensionless equivalent
widths in \hbox{Ly$\alpha$}\ (EW, as in Fig. 4) are plotted for clouds
photoionized by different power law spectra.
Other parameters are the same as Fig. 3. The lines
are CIII~$\lambda$ 977, NIII~$\lambda$ 991, OVI~$\lambda$ 1034,
NV~$\lambda$ 1240, CIV~$\lambda$ 1549, HeII~$\lambda$ 1640,
OIII]~$\lambda$ 1664, NIII]~$\lambda$ 1750 and CIII]~$\lambda$ 1909.
\end{quotation}
\subsubsection{Line Dependence on Abundances}
The left-hand panel of Figure 6 shows a series of calculations
for clouds with different metallicities, $Z$ (scaled from solar
and preserving solar ratios). The strengths of the
collisionally-excited lines relative to \hbox{Ly$\alpha$}\ change little
with $Z$. In particular, CIV/\hbox{Ly$\alpha$}\ varies negligibly for
$0.1\la Z\la 30$~\Zsun\ (see also Hamann \& Ferland 1993a,
hereafter HF93a). We have already noted that these ratios
are more sensitive to the continuum shape (\S2.5.4).
Their lack of sensitivity to $Z$ can be traced to feedback
in the energy balance. As the metal abundances grow,
the line cooling increases. The growing metallicities,
which might otherwise increase the metal line strengths, are
thus balanced in real clouds by lower temperatures ---
with the result that the total metal line flux stays
constant. This feedback is especially important for
strong lines, like CIV, that by themselves control a large
fraction of the cooling. Weak lines respond better
to abundance changes. At low metallicities ($Z\la 0.02$~\Zsun )
none of the metal lines are important coolants and their overall
strengths do scale with $Z$.
Another factor in the line
behaviors at high $Z$ is the increasing bound-free
continuum absorption by metal ions.
The metals absorb a larger fraction of
the far-UV flux at high $Z$, such that the H and He recombination
lines become somewhat weaker. This effect
dominates the high-$Z$ rise in OVI/HeII and NV/HeII in
Figure 6.
The right-hand panel in Figure 6 shows the same line
ratios as before, but in this
case nitrogen is scaled such that
N/H~$\propto Z^2$ (where N/H is solar at $Z=$~\Zsun ).
This selective scaling is based on the expected secondary
nucleosynthesis of nitrogen (\S6 below).
Shields (1976) noted that this abundance pattern should
occur in QSOs by analogy with its direct observation
in galactic HII regions. Figure 6 shows that
it leads to a strong metallicity dependence for line ratios
involving nitrogen. This strong dependence is possible
because the N lines do not control the cooling.
\newpage
\begin{quotation}
\centerline{XXXXXX INSERT FIGURE HERE XXXXXX}
\noindent Fig. 6 --- Predicted line flux ratios for
photoionized clouds
with different metallicities, $Z$. The metals are
scaled together (preserving solar ratios) in the left-hand
panel, while nitrogen is selectively scaled like
$Z^2$ in the right panel.
Other parameters are the same as Fig. 3. See Fig. 5
for line notations.
\end{quotation}
\subsection{Abundance Diagnostics and Results}
\subsubsection{Intercombination Lines}
Shields (1976) proposed using various collisionally-excited
intercombination (semi-forbidden) lines to derive
metal-to-metal abundance ratios in QSOs. He emphasized the strengths
of NIII]~$\lambda$ 1750 and NIV]~$\lambda$ 1486 compared
to OIII]~$\lambda$ 1664, CIII]~$\lambda$ 1909 and CIV~$\lambda$ 1549
as potential diagnostics of the overall
metallicity. As noted above, the metallicity dependence stems
from the expected $Z^2$ scaling of N via secondary
nucleosynthesis (also \S6 below). Shields selected
lines with similar ionization and excitation energies, so
that their ratios are insensitive to the uncertain
temperature, ionization and geometry.
Comparisons with the measured line ratios in QSOs
(see also Davidson 1977, Baldwin \& Netzer 1978, DN79,
Osmer 1980, Uomoto 1984) suggested that N/C and N/O are
often solar or higher, consistent with solar or higher
metallicities. Gaskell, Shields \& Wampler (1981) extended this
analysis to SiIII]~$\lambda$ 1892 and other lines to show that the
refractory elements cannot be substantially depleted by dust
in BELRs.
One drawback of the
intercombination lines is that most of them
are weak and therefore difficult (or impossible) to measure.
Nonetheless, the best recent measurements (Wills {\it et~al.}
1995, Laor {\it et~al.} 1995, Boyle 1990, Baldwin {\it et~al.} 1996)
support the earlier results.
It is now possible to gather even more data for these lines
at a range of redshifts. A note of caution is that the
strong feature generally attributed to CIII]~$\lambda$ 1909 can have
large contributions from other lines (Laor {\it et~al.} 1995, 1997,
Baldwin {\it et~al.} 1996), so
that ratios like NIII]/CIII] might systematically
underestimate N/C if line blending is not accounted for.
A more serious concern is that
the early theoretical work did not consider the range
of high densities now believed to be present in the
BELR (\S2.2). The intercombination lines probably
form at or near their critical densities (typically
$3\times 10^9$ to 10$^{11}$~\hbox{cm$^{-3}$}\ for $\beta = 1$ in Eqn. 5).
Lines with different $n_{crit}$ could have different degrees of
collisional suppression. (For example, the calculated results
using $n_e\approx n_H = 10^{10}$~\hbox{cm$^{-3}$}\
in Figs. 5 and 6 favor large NIII]/CIII] at a given N/C abundance
because CIII] is collisionally suppressed above its
$n_{crit}\approx 3\times 10^9$~\hbox{cm$^{-3}$} .) If there is a range
of densities, lines with different $n_{crit}$ might form
in different regions (even if they have similar ionizations),
leading to a geometry dependence.
Nonetheless, line ratios involving similar $n_{crit}$ and
similar sensitivities to other parameters, such as
NIII]/OIII], could still be
robust abundance indicators when they are measurable.
More theoretical work is needed to explore the parameter
sensitivities and selection effects that can influence
these lines in complex BELRs.
\subsubsection{Permitted Lines}
There are several possibilities for abundance diagnostics
among the permitted UV lines. Figure 6 shows that
NIII~$\lambda$ 991/CIII~$\lambda$ 977
and NV~$\lambda$ 1240/CIV~$\lambda$ 1549 should be good tracers of N/C.
Another possibility is NV/HeII~$\lambda$ 1640, or perhaps
NV/(CIV+OVI~$\lambda$ 1034). The NV, OVI and HeII lines form in
overlapping regions (Fig. 3), as do NIII and CIII, so their
flux ratios should be insensitive to the global BELR structure.
Also, as noted above, the N lines are not important coolants
and thus responsive to abundance changes.
There are practical problems with most of these lines, however;
NV is blended with \hbox{Ly$\alpha$} , CIII, NIII and HeII are weak, and CIII, NIII
and OVI lie in the ``forest'' of intervening \hbox{Ly$\alpha$}\ absorption lines.
Nonetheless, improvements in the quality of data (for example,
high resolution and high signal-to-noise spectra in the \hbox{Ly$\alpha$}\
forest) are permitting increasingly accurate measurements of
these lines in large QSO samples.
\subsubsection{NV/HeII and NV/CIV}
Some of the first studies of large QSO samples
noted that NV~$\lambda$ 1240 is often stronger than
predicted by photoionization
models using solar abundances (Osmer \& Smith 1976 and 1977).
The NV/HeII and NV/CIV ratios have since received particular
attention as abundance diagnostics (Hamann \& Ferland 1992,
HF93a, Hamann \& Ferland 1993b --- hereafter HF93b,
and Ferland {\it et~al.} 1996 --- hereafter F96).
Figure 7 shows the measured
ratios in these lines for QSOs at different redshifts
(from HF93b and Hamann {\it et~al.} 1997a, with some new data and
modifications based mainly on Wills {\it et~al.} 1995 and Baldwin {\it et~al.}
1996). NV/HeII is the ratio of a collisional to recombination line,
with the expected strong temperature dependence (\S2.4).
Calculations similar to (but more exhaustive than) those shown in
Figures 4--6, indicate that NV/HeII reaches a maximum value
linked to the maximum temperature in photoionized clouds (F96).
The maximum NV/HeII ratio is $\sim$2--3 for solar N/He abundances,
depending on how ``hard'' a continuum shape one considers
realistic for QSOs. (Beware that the highest ratios in Fig. 4
occur for parameters where both lines are growing weak,
cf. the EW(CIV) plot or Korista {\it et~al.} 1997b.)
Nominal BELR parameters predict NV/HeII near
unity for solar N/He (Figs. 4--6). These
predictions fall well below most of the measured ratios (Fig. 7),
implying that QSOs typically have super-solar N/He.
The ad hoc (high) temperatures that
would be needed to explain the observed NV/HeII ratios with
solar N/He are inconsistent with photoionization equilibrium, and
would lead to strong far-UV emission lines. The fact that these
far-UV line strengths are not seen sets an upper limit on the
temperature and supports the result for super-solar N/He (F96).
The NV/CIV lines are collisionally excited with similar
energies, so the temperature dependence
is smaller than NV/HeII. Nominal BELR
parameters predict NV/CIV of order 0.1 for solar N/C (Figs.
4--6, also HF93a,b). Comparisons with the data in Figure 7
thus indicate super-solar N/C for most QSOs.
The two NV ratios together therefore imply that
1) quasar metallicities are often solar or higher,
especially in high redshift, high luminosity objects,
and 2) nitrogen (e.g. N/C) is typically enhanced compared
to solar ratios (F96, HF93a,b).
The conclusion for enhanced N/C is based largely
on NV/CIV, but we note that the scaling of $N\propto Z^2$ leads to
self-consistent estimates of $Z$ based on NV/HeII and NV/CIV (Fig. 7).
The actual $Z$ values are uncertain, but the main point is
that many observed ratios require $Z\ga$~\Zsun .
Figures 6 and 7 combined suggest that the nominal
metallicity range is $1\la Z\la 10$~\Zsun\
for standard photoionization parameters and $N\propto Z^2$.
\begin{figure}[h]
\plotfiddle{figure7.ps}{4.0in}{0.0}{60.0}{60.0}{-246.0}{-516.0}
\end{figure}
\begin{quotation}
\noindent Fig. 7 --- Measured NV/HeII and NV/CIV flux
ratios versus redshift (left panels) and
continuum luminosity (right). The upper and lower ranges
might be undersampled (especially for NV/HeII at redshifts $>$1)
because limits on weak lines (e.g. HeII) were often
not available from the literature. The two asterisks in each
panel represent mean values measured by Osmer {\it et~al.} (1994)
for ``high'' and ``low'' luminosity QSOs at redshift $>$3.
The solid curves are predictions
based on chemical evolution models (discussed in \S6 below).
\end{quotation}
HF93b noted that the observed NV ratios tend to be higher in
more luminous sources (Fig. 7).
Most BELs exhibit the well-known ``Baldwin effect," that is,
lower equivalent widths at higher luminosities (Baldwin 1977b).
This effect is well established in CIV and appears to be even
stronger in OVI (Zheng {\it et~al.} 1995,
Kinney, Rivolo \& Koratkar 1990, Osmer \& Shields 1999).
Surprisingly, NV does not show this effect (Osmer {\it et~al.} 1994,
Laor {\it et~al.} 1995, Francis \& Koratkar 1995)
even though its ionization is intermediate
between CIV and OVI and its electron structure is identical.
We proposed that the peculiar NV behavior is due
to generally higher metallicities and more enhanced N abundances
in more luminous QSOs. The recent theoretical study of the
Baldwin effect by Korista, Baldwin \& Ferland (1998) gives
quantitative support to that conclusion. In \S7 we will
argue that this proposed metallicity--luminosity trend in
QSOs could naturally result from a mass--metallicity
correlation among their host galaxies.
The abundance results based on NV have been questioned
by Turnshek {\it et~al.} (1988, 1996) and Krolik \& Voit (1998), who
argue that the NV BEL forms largely by resonance scattering in
an outflowing BAL region. NV might be selectively enhanced
by this mechanism because it can scatter both the continuum
and the underlying \hbox{Ly$\alpha$}\ emission line. However, explicit
calculations of the line scattering (Hamann, Korista \& Morris
1993, Hamann \& Korista 1996, Hamann,
Korista \& Ferland 1999a) do not support this scenario.
For example, 1) the amount of NV scattering estimated by
Krolik \& Voit (1998) is too large by a factor of
$\sim$3 on average, because BALRs do not
generally have the right velocity/optical depth structure
to scatter all of the incident \hbox{Ly$\alpha$}\ photons. In particular,
NV BALs are not usually black across the \hbox{Ly$\alpha$}\ emission line.
2) It is difficult for NV ions in high-velocity
BAL winds to scatter \hbox{Ly$\alpha$}\ photons into simple emission
profiles with observed half-widths of typically 2000 to 2500~\rm{\hbox{km s$^{-1}$}} .
For example, isotropic scattering of the \hbox{Ly$\alpha$}\ flux would produce
BEL half-widths of $\sim$6000~\rm{\hbox{km s$^{-1}$}}\ (the velocity separation
between the NV and \hbox{Ly$\alpha$}\ lines). Anisotropic scattering
(e.g. in BALRs with equatorial or bipolar geometries) would lead
to strong orientation effects and systematically broader BEL
profiles in BAL versus non-BAL QSOs. These differences are
not observed (Weymann {\it et~al.} 1991). 3) It is not clear
why, in individual spectra, the NV emission profiles
should closely resemble those of other BELs if the former is
produced by scattering in a high-velocity BAL wind while the
latter are collisionally excited in a separate
region (i.e. the usual BELR --- whose velocity field is not mostly
radial based on the reverberation studies, T\"urler \&
Courvoisier 1997, Korista {\it et~al.} 1995). Finally, 4)
large scattering contributions to NV would minimally require
much larger global BALR covering factors (the fraction
of the sky covered by the BALR as seen from the central QSO)
than expected from their observed detection frequency
in (randomly oriented) QSO samples. Goodrich (1997) and
Krolik \& Voit (1998) argue that larger global covering
factors could occur, but that issue is not settled.
Another concern is that complex BELR geometries
might cause the NV/HeII and NV/CIV abundance indicators
to fail --- but they would fail in opposite directions.
Specifically,
clouds that are truncated at different physical depths (see Fig. 3)
could produce strong HeII with little or no NV and CIV emission,
or strong HeII and NV with little or no CIV. For a given
abundance set, this type of truncation could therefore either
lower the observed NV/HeII ratio or increase NV/CIV.
Comparing the data to simulations that do not
take truncation into account (Figs. 4--6) might then
lead to underestimated N/He abundances or overestimated
N/C. However, we have already shown that these two line ratios
yield similar metallicities when compared to the non-truncated
simulations, so we are not likely being
mislead by complex BELR geometries. Moreover, the
NV/HeII ratio provides in any case a secure lower limit on N/He.
\subsubsection{FeII/MgII}
The broad FeII emission lines pose unique problems
because the atomic physics is complex and many
blended lines contribute to the spectrum,
particularly at the wavelengths $\sim$2000--3000~{\it Astron. Astrophys.} \ and
$\sim$4500--5500~{\it Astron. Astrophys.} . Nonetheless, FeII is worth the effort
because a delay of $\sim$1~Gyr in the Fe enrichment,
relative to $\alpha$ elements such as O, Mg or Si,
might provide a ``clock'' for constraining
the ages of QSOs and the epoch of their first star formation
(see \S\S6--7 below, also HF93b).
A series of important papers on FeII emission
(Osterbrock 1977, Phillips 1977,1978, Grandi 1981) culminated
with Wills \& Netzer (1983) and Wills, Netzer \& Wills (1985,
hereafter WNW). They performed sophisticated calculations
showing that the large observed FeII fluxes,
e.g. FeII(UV)/MgII~$\lambda$ 2799, require that either Fe is
several times overabundant (compared to solar ratios)
or some unknown process dominates the FeII excitation.
One process that might selectively enhance FeII emission
is photoexcitation by \hbox{Ly$\alpha$}\ photons (Johansson \& Jordan 1984).
The absorption of \hbox{Ly$\alpha$}\ radiation can pump electrons from
the lower (metastable) energy levels of Fe$^+$ into specific
high-energy states, leading to
fluorescent cascades. WNW discounted
this mechanism because it appeared insignificant in their
simulations, but Penston (1987) noted that
\hbox{Ly$\alpha$}\ pumping is known to be important in some emission-line
stars, such as the symbiotic star RR Tel, and therefore
might be important in QSOs. More recent FeII simulations
using better atomic data and exploring a wider range of
physical conditions
(Sigut \& Pradhan 1998, Verner {\it et~al.} 1999) suggest that \hbox{Ly$\alpha$}\
can be important in some circumstances, but it is not yet
clear if those circumstances occur significantly in QSOs.
Recent observations have renewed interest in
this question by showing that the FeII(UV)/MgII emission
fluxes can be larger than the WNW predictions even at
$z>4$, with the tentative conclusion that Fe/Mg is at
least solar (and thus the objects are at least $\sim$1~Gyr old,
Taniguchi {\it et~al.} 1997, Yoshii, Tsujimoto \& Kawara
1998, Thompson, Hill \& Elston 1999 and refs. therein).
New theoretical efforts,
such as Sigut \& Pradhan (1998) and Verner {\it et~al.} (1999),
are needed to test these conclusions and quantify the
uncertainties. However, a better way to measure
Fe/$\alpha$ might be with the intrinsic NALs (see below).
\section{Absorption Line Diagnostics}
\subsection{Overview: Types of Absorption Lines}
Quasar absorption lines can have a variety of intrinsic
or cosmologically intervening origins.
We exclude from our discussion the
damped-\hbox{Ly$\alpha$}\ absorbers and the ``forest'' of many narrow \hbox{Ly$\alpha$}\
systems with weak or absent metal lines because
they form in cosmologically intervening gas (Rauch 1998).
The remaining metal-line systems can be divided into
two classes according to
their broad or narrow profiles. This division is
a gross simplification, but still useful
because it distinguishes the clearly intrinsic broad
lines from the many others of uncertain origin. Here
we briefly characterize the two (broad and narrow) line types.
\subsubsection{Broad Absorption Lines (BALs)}
Broad absorption lines
are blueshifted with respect to the emission lines and have
velocity widths of at least a few thousand \rm{\hbox{km s$^{-1}$}}\
(for example, Fig. 8). They appear in 10 to 15\% of
optically-selected QSOs and clearly identify
high-velocity winds from the central engines.
The precise location of the absorbing gas is unknown,
but there is little doubt that it is intrinsic --- originating
within at least a few tens of parsecs from the QSOs.
See recent work by Weymann {\it et~al.} (1991), Barlow {\it et~al.} (1992),
Korista {\it et~al.} (1993), Hamann {\it et~al.} (1993),
Voit, Weymann \& Korista (1993), Murray {\it et~al.} (1995),
Arav (1996), Turnshek {\it et~al.} (1997),
Brotherton {\it et~al.} (1998b), and the reviews by
Turnshek (1988, 1994), Weymann, Turnshek \& Christianen (1985)
and Weymann (1994, 1997).
\begin{figure}[h]
\plotfiddle{figure8.ps}{2.5in}{0.0}{65.0}{65.0}{-205.0}{-200.0}
\end{figure}
\begin{quotation}
\noindent Fig. 8 --- Spectrum of the BALQSO PG~1254+047
(emission redshift $z_e = 1.01$) with
emission lines labeled across the top and possible BALs marked
below at redshifts corresponding to the 3 deepest minima in the
CIV trough. Not all of the labeled lines are detected. The smooth
dotted curve is a power-law continuum fit extrapolated to short
wavelengths (from Hamann 1998).
\end{quotation}
\subsubsection{Narrow Absorption Lines (NALs)}
A practical definition of NALs would limit their full
widths at half minimum (FWHMs) to
less than the velocity separation of important
doublets (e.g. $<$500~\rm{\hbox{km s$^{-1}$}}\ for CIV, $<$1930 \rm{\hbox{km s$^{-1}$}}\ for SiIV
or $<$960 \rm{\hbox{km s$^{-1}$}}\ for NV), because it is our ability to
resolve these doublets that makes their analysis
fundamentally different from the BALs (\S3.2.2 below).
NALs can form in a variety of locations, ranging from
very near QSOs, as in ejecta like the BALs, to unrelated gas
or galaxies at cosmological distances (Weymann {\it et~al.} 1979).
It is not yet known what fraction of
NALs at any velocity shift meet our definition of intrinsic
(\S1). Several studies have noted a statistical excess
of NALs within a few thousand \rm{\hbox{km s$^{-1}$}}\ of the emission redshifts.
These are the so-called ``associated'' or {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\
absorbers (with redshifts close to the emission
redshift, Weymann {\it et~al.} 1979 and 1981, Young {\it et~al.} 1982,
Foltz {\it et~al.} 1986 and 1988, Anderson {\it et~al.} 1987).
Their strengths and frequency of occurrence
appear to correlate with the QSO luminosities or radio
properties, suggesting some physical relationship (also M\"oller
{\it et~al.} 1994, Aldcroft, Bechtold \& Elvis 1994, Wills {\it et~al.} 1995,
Barthel, Tytler \& Vestergaard 1997).
These correlations may extend to NALs at blueshifts
of 30,000~\rm{\hbox{km s$^{-1}$}}\ or more (Richards \& York 1998).
Nontheless, we might expect a larger
fraction of intrinsic NALs nearer the emission redshift
and, if they are ejected from QSOs, they should appear
at {$z_a<z_e$}\ rather than {$z_a>z_e$} .
Several tests have been developed
to help identify intrinsic NALs, including 1)
time-variable line strengths, 2) multiplet ratios that imply
partial line-of-sight coverage of the background light
source(s), 3) high gas densities inferred from excited-state
absorption lines, and 4) well-resolved line profiles that are
smooth and broad compared to both thermal line widths
and to the velocity dispersions expected in intervening clouds
(e.g. Bahcall, Sargent \& Schmidt 1967, Williams {\it et~al.} 1975,
Young {\it et~al.} 1982, Barlow \& Sargent 1997, Hamann {\it et~al.} 1997b
-- hereafter H97b, Hamann {\it et~al.} 1997c, Petitjean \& Srianand 1999,
Ganguly {\it et~al.} 1999, and refs. therein).
These criteria might not be definitive individually,
but they sometimes appear in combination.
Figure 9 shows a {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ NAL system that is clearly intrinsic
based on time-variable line strengths, partial line-of-sight
coverage and relatively broad profiles.
High metallicities might be another indicator of intrinsic
absorption (\S3.4 below), but that criterion would bias abundance
studies; we would like to determine
the intrinsic versus intervening nature independent
of the abundances. The other (non-abundance)
tests indicate that intrinsic NALs can have
velocity shifts out to $\ga$24,000~\rm{\hbox{km s$^{-1}$}}\ and a wide range of
FWHMs down to $\la$30~\rm{\hbox{km s$^{-1}$}} . See the references above and
the reviews of {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ systems
by Weymann {\it et~al.} (1981) and Foltz {\it et~al.} (1988).
\subsection{General Abundance Analysis}
Abundance estimates from absorption lines are, in principle,
more straightforward than for emission lines because the
absorption strengths are not sensitive to the temperatures or
space densities. Moreover, absorption
lines yield direct measures of the column densities in different
ions. We need only apply appropriate ionization corrections
to convert the column densities into relative abundances.
For example, the abundance ratio of any two elements $a$ and
$b$ can be written,
\begin{equation}
\left[{a\over b}\right]\ = \
\log\left({{N(a_i)}\over{N(b_j)}}\right)\ +\
\log\left({{f(b_j)}\over{f(a_i)}}\right)\ +\
\log\left({b\over a}\right)_{\odot}
\end{equation}
which is identical to Equation 8 except that the $N$ here are the
column densities. Once again we define the ionization correction
as $IC\equiv \log(f(b_j)/f(a_i))$.
Abundance studies would ideally compare lines with similar
ionizations to minimize $IC$ and reduce the sensitivity
to potentially complex geometries. Unfortunately, the lines
available often require significant ionization corrections.
In particular, we are often forced to compare
highly-ionized metals (such as CIV) to HI (\hbox{Ly$\alpha$} ) to derive the
metallicity. We must therefore use ionization models.
\begin{figure}[h]
\plotfiddle{figure9a.ps}{2.7in}{0.0}{38.0}{38.0}{-163.0}{-285.0}
\plotfiddle{figure9b.ps}{2.7in}{0.0}{38.0}{38.0}{-115.0}{-30.0}
\end{figure}
\begin{quotation}
\noindent Fig. 9 --- Spectra of the {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ absorber
in UM675 ($z_e = 2.15$) showing its time-variability in two
epochs (top panel) and broad, smooth profiles at higher
spectral resolution (9~\rm{\hbox{km s$^{-1}$}} , bottom panels,
from Hamann {\it et~al.} 1995, 1997b).
\end{quotation}
The usual assumption is that the gas is photoionized
by the QSO continuum flux. Collisional ionization would
lead to lower derived metallicities because it creates
less HI (and more HII) for a given level of ionization
in the metals (cf. Figs. 4 and 5 in Hamann {\it et~al.} 1995,
also Turnshek {\it et~al.} 1996). However, collisional ionization
has been generally dismissed for BAL regions (BALRs)
because 1) it would
be energetically hard to maintain, 2) it would
produce excessive amounts of line emission (because
of the much higher temperatures), and 3) it is hard to
reconcile with the observed simultaneous variabilities in BAL
troughs across a wide range of velocities (Weymann {\it et~al.} 1985,
Junkkarinen {\it et~al.} 1987, Barlow 1993). In contrast, the
strong radiative flux known to be present in QSOs
provides a natural ionization source. We will assume
that photoionization dominates in both BALRs and
intrinsic NALRs.
\begin{figure}[h]
\plotfiddle{figure10.ps}{4.6in}{0.0}{95.0}{95.0}{-293.0}{-294.0}
\end{figure}
\begin{quotation}
\noindent Fig. 10 --- Ionization fractions
in optically thin clouds photoionized at different $U$ by a
power-law spectrum with $\alpha =$~$-$1.5. The HI fraction
appears across the top. The curves for the metal ions
are labeled above or below their peaks whenever possible. The
notation here is HI = H$^o$, CIV = C$^{+3}$, etc.
\end{quotation}
Estimates of $IC$ generally come from plots
like Figure 10, which shows the ionization fractions of
HI and various metal ions $M_i$ in photoionized clouds
(see \S2.5 and Ferland {\it et~al.} 1998 for general descriptions
of the calculations).
Ideally, we would compare column densities in different ions of
the same element to obtain abundance-independent constraints
on the ionization and thus $IC$. Otherwise, column densities
in different elements can also constrain $IC$ with assumptions
about the relative abundances. Note that the results in
Figure 10 are not sensitive to the particular abundances used
the calculations (in this case solar), so the figure is useful
for general abundance/ionization estimates (see Hamann 1997
--- hereafter H97).
The model clouds are optically thin in the ionizing UV continuum,
which means that gradients in the ionization are negligible
across the cloud and the ionization fractions do not
depend on the total column densities. This simplification
appears to be appropriate for most intrinsic
absorption line systems (based on their measured column densities),
although shielding by many far-UV BALs might affect the ionization
structure downstream in BALRs (Korista {\it et~al.} 1996,
Turnshek 1997, H97). Also, systems with
low-ionization lines like FeII or MgII can be optically
thick at the HI Lyman edge (Bergeron \& Stasi\'nska
1986, Voit {\it et~al.} 1993, Wampler, Chugai \& Petitjean 1995) and
may require calculations with specific
column densities that match the data.
\subsubsection{Ionization Ambiguities}
The main theoretical uncertainties involve the shape of the
ionizing spectrum, the frequent
lack of ionization constraints (too few lines measured),
and the possibility of inhomogeneous (multi-zone) absorbing
media. H97 addressed these issues
by calculating $IC$ values for a wide range of conditions
in photoionized clouds. He noted that, whenever there is or might
be a multi-zone with a range of ionizations, we can still
make conservatively low estimates of the metal-to-hydrogen
abundance ratios by assuming each metal line forms
where that ion is most favored --- that is, at the peak of its
ionization fraction $f(M_i)$ in Figure 10. We can also place
firm lower limits on the metal-to-hydrogen ratios by
adopting the minimum values of $IC$, which correspond to
minima in the $f$(HI)/$f$(M$_i$) ratios (see also Bergeron \&
Stasi\'nska 1986). The lower limits are robust, even though they
come from 1-zone calculations, because different or additional
zones can only mean that larger $IC$ values are appropriate for
the data. Figure 11 plots several minimum metal-to-hydrogen $IC$s
for optically thin clouds photoionized by different power-law
spectra. The results in this plot simply get added to the
logarithmic column density ratios (Eqn. 10) to derive minimum
metallicities. Note that some important metal-to-metal ratios also
have minimum $IC$ values, such as PV/CIV and FeII/MgII
(Hamann 1998, Hamann {\it et~al.} 1999b).
\subsubsection{Column Densities and Partial Coverage}
The final critical issue is deriving accurate column
densities from the absorption troughs. In the simplest case,
the line optical depths are related to the observed intensities by,
\begin{equation}
I_{v}\ =\ I_o\,\exp{(-\tau_{v})}
\end{equation}
where $I_v$ and $I_o$ are the observed and intrinsic
(unabsorbed) intensities, respectively, and $\tau_{v}$ is the
line optical depth, at each velocity shift $v$. The
column densities follow from the optical depths by,
\begin{equation}
N\ =\ {{m_ec}\over{\pi e^2 f\lambda_o}}\,\int\tau_v\ dv
\end{equation}
\newpage
\begin{figure}[h]
\plotfiddle{figure11.ps}{4.0in}{0.0}{100.0}{100.0}{-320.0}{-273.0}
\end{figure}
\begin{quotation}
\noindent Fig. 11 --- Minimum metal ion to HI ionization
corrections ($IC$) normalized to solar abundances (the last two
terms in Eqn. 10) are plotted for optically thin clouds photoionized
by power-law spectra with different indices ($\alpha$). The notation
is the same as Fig. 10. The curves have been shifted vertically by
+1 for OVI, by $-$2 for PV, and by $-$1 for AlII and AlIII.
The curves for nitrogen ions are dash-dot.
\end{quotation}
\noindent where $f$ is the oscillator strength and $\lambda_o$ is
the laboratory wavelength of the line.
Column density derivations can involve
line profile fitting or direct integration over the observed
profiles (via Eqns. 11 and 12, Junkkarinen {\it et~al.} 1983, Grillmair
\& Turnshek 1987, Korista {\it et~al.} 1992,
Savage \& Sembach 1991, Jenkins 1996, Arav {\it et~al.} 1999 ---
hereafter A99). Very optically
thick lines are not useful because the inferred values of
$\tau_{v}$ are far too sensitive to uncertainties
in $I_{v}$. In other cases,
the analysis might still be compromised by 1) unresolved
absorption-line components or 2) unabsorbed flux that fills in
the the bottoms of the observed troughs. If either of these
possibilities occurs, the derived optical depths and
column densities become lower limits and the derived
abundances become incorrect. Errors from the first possibility
can always be reduced or avoided by higher resolution spectroscopy.
\newpage
\begin{quotation}
\centerline{XXXXXX INSERT FIGURE HERE XXXXXX}
\noindent Fig. 12 ---
Schematic showing possible ``partial coverage'' geometries. Partial
line-of-sight coverage occurs when light rays like C, which pass through
absorption-line clouds (indicated by filled ellipsoids), are combined
with rays like A, B or D, which do not. Ray A represents reflected
light from a putative scattering region. Ray B simply misses the
absorption-line region. Ray D passes through the nominal absorbing
zone but suffers no absorption because the region is porous.
\end{quotation}
The second possibility, of filled-in absorption troughs,
is actually an asset for identifying intrinsic NALs
(\S3.1). We will refer to this filling-in generally as
``partial coverage'' of the background emission source.
Figure 12 shows several geometries that might produce
partial coverage and filled-in troughs\footnote{The situation
can be potentially more complicated if the absorber itself is
a source of emission. The analysis discussed below remains the
same, however.}. When partial coverage occurs,
the observed intensities depend on both the optical depth
and the line-of-sight coverage fraction, $C_f$,
at each velocity,
\begin{equation}
I_v\ =\ (1-C_f)\,I_o\, +\, C_f\,I_o\,\exp{(-\tau_{v})}
\end{equation}
where $0\leq C_f\leq 1$ and the first term on the right
side is the unabsorbed (or uncovered) contribution.
Measured absorption lines can thus be
shallow even when the true optical
depths are large. In the limit $\tau_{v}\gg 1$, we have,
\begin{equation}
C_f\ =\ 1 - {{I_{v}}\over{I_o}}
\end{equation}
Outside of that limit,
we can compare lines whose true optical depth ratios are fixed by
atomic physics, such as the HI Lyman lines or doublets like
CIV~$\lambda$\lam 1548,1550, SiIV~$\lambda$\lam 1394,1403, etc., to
determine uniquely both the coverage fractions and the true
optical depths across the line profiles
(H97b, Barlow \& Sargent 1997, A99, Srianand \&
Shankaranarayanan 1999, Ganguly {\it et~al.} 1999).
For example, a little algebra shows that for doublets with
true optical depth ratios of $\sim$2 (as in CIV,
SiIV, etc.) the coverage fraction at each absorption velocity is,
\begin{equation}
C_f\ =\ {{I_1^2 - 2I_1 + 1}\over{I_2 - 2I_1 +1}}
\end{equation}
where $I_1$ and $I_2$ are the observed
line intensities, normalized by $I_o$, at the same velocity in
the weaker and stronger line troughs, respectively.
The corresponding line optical depths are $\tau_2 = 2\tau_1$ and,
\begin{equation}
\tau_1\ =\ \ln\left({{C_f}\over{I_1 + C_f - 1}}\right)
\end{equation}
It is a major strength of the NALs that we can
resolve key multiplet lines and use this analysis
to measure the coverage fractions and thus derive reliable
column densities and abundances. It is a great weakness
of the BALs that this analysis is usually not
possible because the lines are blended. We will
argue below that BAL studies so far have been seriously
compromised by unaccounted for partial-coverage effects.
The only drawback of partial coverage for the NALs is that
there might be a range of coverage fractions in multi-zone
absorbing media. There is already evidence in some cases
for coverage fractions that differ between ions or
change with velocity across the line profiles
(Barlow \& Sargent 1997, Barlow, Hamann \& Sargent 1997,
H97b). Variations in $C_f$ with velocity
can always be dealt with by analyzing limited velocity intervals
in the line profiles (see also Arav 1997).
But one can imagine complex geometries where ionization-dependent
coverage fractions would jeopardize the simple
analysis described above, in particular for comparisons
between high and low ionization species like CIV and HI.
Abundance ratios based on disparate species like these might
require specific models of the ionization-dependent coverage.
On the other hand, this worst-case scenario is not known
to occur, and there is no reason to believe it would lead
to generally overestimated metallicities anyway.
\subsection{Broad Absorption Line Results}
One common characteristic of BAL spectra is that the
metallic resonance lines like CIV~$\lambda$ 1548,1951, SiIV~$\lambda$ 1394,1403,
NV~$\lambda$ 1239,1243 and OVI~$\lambda$ 1032,1038 are typically
strong (deep) compared to \hbox{Ly$\alpha$}\ (e.g. Fig. 8). This result,
and the fact that low-ionization lines like MgII~$\lambda$ 2796,2804
and FeII (UV) are usually absent, indicates that the BALR ionization
is generally high (Turnshek 1984, Weymann {\it et~al.} 1981,1985).
However, quantitative studies of the ionization have repeatedly
failed to explain the measured line strengths with solar
abundances. These difficulties were first noted by
Junkkarinen (1980) and Turnshek (1981, also
Weymann \& Foltz 1983), who
showed that photoionization models with
power-law ionizing spectra and solar abundances
underpredict the metal ions, especially SiIV, by large
factors relative to HI. A straightforward
conclusion is that the metallicities are
well above solar. Turnshek (1986, 1988) and
Turnshek {\it et~al.} (1987) estimated
metal abundances (C/H) of 10 to 100 times solar,
and provided tentative evidence for some extreme
metal-to-metal abundance ratios such as P/C~$\ga$~100
times solar.
Better data in the past ten years have
done nothing to change these startling results
(e.g. Turnshek {\it et~al.} 1996). The early concerns about
unresolved line components (Junkkarinen {\it et~al.} 1987, Kwan 1990)
have gone away, thanks to spectroscopy with
the Keck 10 m telescope at resolutions ($\sim$7~\rm{\hbox{km s$^{-1}$}} ) close
to the thermal speeds (Barlow \& Junkkarinen 1994,
Junkkarinen 1998). The previously tentative detections of
PV~$\lambda$ 1118,1128 absorption, which led to the large
P/C abundance estimates, have now been confirmed in
two objects by excellent wavelength coincidences, by
the predicted weakness of nearby lines like FeIII~$\lambda$ 1122,
and in one case by the probable presence of PIV~$\lambda$ 951
absorption (Junkkarinen {\it et~al.} 1997, Hamann 1998, Fig. 8).
The commonality of PV absorption is not yet known
(see also Korista {\it et~al.} 1992, Turnshek {\it et~al.} 1996), but its
relative strength in just the two cases is surprising because
the solar P/C ratio is only $\sim$0.001.
More complex theoretical analyses, considering
a range of ionizing spectral shapes or
multiple ionization zones, also do not change the main result
for metallicities and P/C ratios well above solar
(Weymann {\it et~al.} 1985, Turnshek {\it et~al.} 1987, 1996 and 1997,
Korista {\it et~al.} 1996). H97 used the analysis
in \S3.2 to determine how high the abundances
must be given the measured column densities and a
photoionized BALR. He showed that average BALR column densities
require [C/H] and [N/H]~$>$~0 and [Si/H]~$>$~1.0 for any range of
ionizations and reasonable spectral shapes. The conservatively
low (but not quite minimum) values of $IC$ indicate
[C/H] and [N/H]~$\ga$~1.0 and [Si/H]~$\ga$~1.7.
The results for individual BAL
systems can be much higher. In PG1254+047 (Fig. 8, Hamann 1998)
the inferred minimum abundances are [C/H] and [N/H]~$\ga$~1.0,
[Si/H]~$\ga$~1.8 and [P/C]~$\ga$~2.2.
However, we will now argue that all these BAL abundance
results are incorrect, because
partial coverage effects have led to generally underestimated
column densities.
\subsubsection{Uncertainties and Conclusions}
There is now direct evidence for partial coverage in some BALQSOs
based on widely separated lines of the same ion (A99) and resolved
doublets in several narrow BALs and BAL components
(Telfer {\it et~al.} 1999, Barlow \& Junkkarinen 1994, Wampler {\it et~al.} 1995,
Korista {\it et~al.} 1992 --- confirmed by Junkkarinen 1998).
Although most this evidence applies to narrow features,
it is noteworthy that there are no
counterexamples to our knowledge --- where narrow line
components associated with BALs indicate complete coverage
(also Junkkarinen 1998).
There is also circumstantial evidence for partial coverage
in BAL systems. Namely, 1)
spectropolarimetry indicates that BAL troughs can be filled in by
polarized flux (probably from an extended scattering
region) that is not covered by the BALR (Fig. 12, Goodrich \& Miller
1995, Cohen {\it et~al.} 1995, Hines \& Wills 1995, Schmidt \& Hines 1999).
2) Some BAL systems have a wide range of lines with suspiciously
similar strengths or flat-bottom troughs that do not reach zero
intensity (Arav 1997). 3) Voit {\it et~al.} (1993) made
a strong case for low-ionization BALRs
being optically thick at the Lyman limit,
which implies large optical depths in \hbox{Ly$\alpha$} , yet the \hbox{Ly$\alpha$}\
troughs are not generally black in these systems.
4) The larger column densities that follow assuming partial
coverage and saturated BALs
($N_H\ga 10^{22}$~\hbox{cm$^{-2}$} , Hamann 1998) are consistent with
the large absorbing columns inferred from X-ray observations
of BALQSOs (Green \& Mathur 1996, Green {\it et~al.} 1997,
Gallagher {\it et~al.} 1999).
More indirect evidence comes from the abundance results
themselves. Voit (1997) noted that the derived overabundances
tend to be greater for rare elements like P than for
common elements like C. This is precisely what would occur
if line saturation is not taken into account. The
surprising detections of PV might actually be a signature of
line saturation (and partial coverage) in strong lines like CIV,
rather than extreme abundances (Hamann 1998).
This assertion is supported by the one known NAL system with
PV~$\lambda$\lam 1118,1128 absorption, where
the doublet ratios in CIV, NV and SiIV clearly indicate
$\tau\gg$1 (Barlow {\it et~al.} 1997, Barlow 1998).
We conclude that BAL column densities have been generally
underestimated and the true BALR abundances are not
known. Observed differences between BAL profiles that resemble
simple optical depth effects are probably caused by a mixture
of ionization, coverage fraction and optical depth differences
in complex, multi-zone BALRs.
This conclusion paints a grim picture for BAL abundance work,
but it might still be possible to derive accurate
column densities and therefore abundances for some BALQSOs or
some portions of BAL profiles (Wampler {\it et~al.} 1995, Turnshek 1997,
A99). Most needed are spectra at shorter rest-frame
wavelengths to measure widely separated lines of the same ion
and thereby diagnose the coverage fractions and true optical depths
(\S3.2.2, Arav 1997, A99).
\subsection{Narrow Absorption Line Results}
In contrast to the BALs,
intrinsic NALs might be the best abundance probes we
have for QSO environments. Resolved measurements of NAL
multiplets allow us to measure both the coverage fractions
and true column densities (\S3.2.2). The NALs also
allow separate measurements of important lines that are
often blended in BAL systems, such as NV~$\lambda$ 1239,1243--\hbox{Ly$\alpha$} ,
OVI~$\lambda$ 1032,1038--\hbox{Ly$\beta$}\ and many others.
We therefore have potentially many more constraints on
both the ionization and abundances.
Early NAL studies did not have the quality of data needed
to derive column densties and abundances, but several
groups noted a tendency for larger NV/CIV
line strength ratios in {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ systems compared to {$z_a\kern -3pt \ll\kern -3pt z_e$}\
(Weymann {\it et~al.} 1981, Hartquist \& Snijders 1982,
Bergeron \& Kunth 1983, Morris {\it et~al.} 1986,
Bergeron \& Boiss\'e 1986). This trend is probably
not due simply to higher ionization in {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ absorbers,
because recent studies show that {$z_a\kern -3pt \ll\kern -3pt z_e$}\
systems typically have strong OVI lines and
therefore considerable high-ionization gas; NV appears
to be weak relative to both CIV and OVI
at {$z_a\kern -3pt \ll\kern -3pt z_e$}\ (Lu \& Savage 1993, Bergeron {\it et~al.} 1994,
Burles \& Tytler 1996, Kirkman \& Tytler 1997,
Savage, Tripp \& Lu 1998). The lower NV/OVI and NV/CIV line
ratios at {$z_a\kern -3pt \ll\kern -3pt z_e$}\ could be caused by an underabundance of
nitrogen (relative to solar ratios) in metal-poor intervening
gas (Bergeron {\it et~al.} 1994, Hamann {\it et~al.} 1997d,
Kirkman \& Tytler 1997). This would be the
classic abundance pattern involving secondary nitrogen
(Vila-Costas \& Edmunds 1993).
Relatively higher N abundances and thus stronger NV
absorption lines should occur naturally in metal-rich
environments near QSOs (see \S\S6--7 below).
The first explicit estimates of {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ metallicities
were by Wampler {\it et~al.} (1993), M\"oller {\it et~al.} (1994), Petitjean
{\it et~al.} (1994) and Savaglio {\it et~al.} (1994) for QSOs at redshifts of
roughly 2 to 4. These studies found that {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ systems often
have $Z\ga$~\Zsun , which is at least an order of magnitude
larger than the {$z_a\kern -3pt \ll\kern -3pt z_e$}\ systems measured in the same
data. Several of the metal-rich {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ systems have doublet
ratios implying partial coverage and thus, very likely, an intrinsic
origin (Wampler {\it et~al.} 1993, Petitjean {\it et~al.} 1994).
The location of the other {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ absorbers is not known,
but Petitjean {\it et~al.} (1994) noted a marked change from
[C/H]~$\la$~$-$1 to [C/H]~$\ga$~0 at a blueshift of
$\sim$15,000~\rm{\hbox{km s$^{-1}$}}\ relative to the emission lines. If high
abundances occur only in intrinsic systems, then these results
suggest that most {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ NALs are intrinsic (also M\"oller
{\it et~al.} 1994).
More recent studies support these findings.
Petitjean \& Srianand (1999) measured $Z\ga$~\Zsun\ and
[N/C]~$>0$ in an intrinsic (partial coverage) {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ absorber.
For {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ systems of unknown origin,
Savage {\it et~al.} (1998) estimated roughly solar metallicities
and Tripp {\it et~al.} (1997) obtained [N/C]~$\ga 0.1$ and, very
conservatively, [C/H]~$\ga -0.8$. (The lower limit on [C/H]
for the latter system is $-$0.2 when more likely ionizing
spectral shapes are used in the calculations.)
Savaglio {\it et~al.} (1997) revised the
metallicities downward slightly from their 1994 paper to
$-1<$~[C/H]~$<0$, based on better data. Those systems
are of special interest because of their
high redshift ($z_a\approx 4.1$).
Wampler {\it et~al.} (1996) estimated $Z\sim 2$~\Zsun\ (based
on a tentative detection of OI~$\lambda$ 1303)
for the only other {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ systems studied so far at $z>4$.
H97, H97b and Hamann {\it et~al.} (1995, 1997e, 1999b)
used the analysis outlined in \S3.2 to
determine metallicities or establish lower limits
for several {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ systems, including some mentioned above and
some that are clearly intrinsic by the indicators in \S3.1.
The results generally confirm the previous estimates
and show further that, even when there are
no constraints on the ionization (for example, when only
\hbox{Ly$\alpha$}\ and CIV lines are measured), the column densities can
still require $Z\ga$~\Zsun . A quick survey of those results
suggests that bona fide intrinsic systems,
and most others with $Z\ga$~\Zsun , have [N/C]~$\ga 0.0$.
\subsubsection{Uncertainties and Conclusions}
Most of the studies mentioned above would benefit from
better data (higher signal-to-noise ratios and higher
spectral resolutions) and more ionization constraints
(wider wavelength coverage),
but the frequent result for $Z\ga$~\Zsun\ is convincing.
Unlike the BALs, there are no obvious systematic effects
that might lead to higher abundance estimates for {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\
systems compared to {$z_a\kern -3pt \ll\kern -3pt z_e$} . The possibility of ionization-dependent
coverage fractions presents an uncertainty for those systems
with partial coverage, but we do not expect that to cause
systematic overestimates of the metallicities (\S3.2.2).
We conclude that many {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ NALs and, more importantly, all
of the ``confirmed'' intrinsic systems, have $Z\ga$~0.5\Zsun\
and usually $Z\ga$~\Zsun . The upper limits on $Z$ are
uncertain. The largest estimate for a well-measured system
is $Z\sim 10$~\Zsun\ (Petitjean {\it et~al.} 1994),
but those data are also consistent with metallicities as low as
solar because of ionization uncertainties (H97).
There are mixed and confusing reports in the literature regarding
metal-to-metal abundance ratios, most notably N/C. In contrast
to Franceschini \& Gratton (1997), we find no tendency for
sub-solar N/C in {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ systems. In fact, there is the general
trend for stronger NV absorption at {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ compared to {$z_a\kern -3pt \ll\kern -3pt z_e$}\
systems, and the most reliable abundance data suggest solar or
higher N/C ratios whenever $Z\ga$~\Zsun .
The only serious problem is in interpreting the abundance
results for systems of unknown origin. High metallicities
might correlate strongly with absorption near QSOs,
but the metallicities cannot define the absorber's location.
For example, Tripp {\it et~al.} (1996) estimated $Z\ga$~\Zsun\ and
[N/C]~$\ga$~0 for a {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ system where the lack
of excited-state absorption in CII$^*$~$\lambda$ 1336
(compared the measured CII~$\lambda$ 1335)
implies that the density is low, $\la$7 \hbox{cm$^{-3}$} , and thus
the distance from the QSO is large, $\ga$300~kpc.
(The relationship between density and distance follows
from the flux requirements for photoionization,
\S2.5.1.) Super-solar metallicities at these large distances
are surprising. At $\ga$300~kpc from the QSO, we might have
expected very low intergalactic or halo-like abundances.
The solution might be that the absorbing gas was enriched much
nearer the QSO and then ejected (Tripp {\it et~al.} 1996).
Unfortunately, the excited-state lines used for density and
distance estimates are not generally available for {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ systems
(because they have low ionization energies, e.g. CII$^*$ and
SiII$^*$). Of the six {$z_a\kern -1.5pt \approx\kern -1.5pt z_e$}\ absorbers known to be far ($\ga$10~kpc)
from QSOs based on these indicators,
three of them clearly have {$z_a>z_e$}\ and are probably not intrinsic
for that reason (Williams {\it et~al.} 1975, Williams \& Weymann 1976,
Sargent {\it et~al.} 1982, Morris {\it et~al.} 1986,
Barlow {\it et~al.} 1997). Only one has a
metallicity estimate --- the system with $Z\ga$~\Zsun\ at
$\ga$300~kpc distance studied by Tripp {\it et~al.} (1996).
\section{General Abundance Summary}
The main abundance results are as follows.
1) There is a growing consensus from the BELs and
NALs for $Z\ga$~\Zsun\ in QSOs out to $z>4$.
The upper limits on the metallicities are not well known,
but none of the data require $Z>10$~\Zsun .
Solar to a few times solar appears to be typical. Based
on very limited data, there is no evidence for a decline at
the highest redshifts.
2) A trend in the NV/HeII and NV/CIV BEL ratios suggests that
the metallicities are generally higher in more luminous QSOs.
3) The BELs and NALs both suggest that the relative nitrogen
abundance (e.g. N/C and N/O) is typically solar or higher.
We will argue below (\S6) that this result corroborates the
evidence for $Z\ga$~\Zsun\ (because of the likely secondary
origin of nitrogen at these metallicities).
4) There is tentative evidence for super-solar Fe/Mg abundances
out to $z>4$ based on the FeII/MgII BEL strengths. Again,
based on limited data, there is no evidence for a decline in this
ratio at the highest redshifts.
5) The extremely high metallicities and large P/C ratios
derived so far from the BALs
are probably incorrect. In further support of that conclusion,
we note that BELR simulations using the nominally derived
BAL abundances (including large enhancements in P and
other odd-numbered elements like Al, Shields 1996) are
inconsistent with observed BEL spectra (based on unpublished
work in collaboration with G. Shields).
\section{Enrichment Scenarios}
Several scenarios have been proposed for the production of
heavy elements near QSOs, including 1) the normal evolution
of stellar populations in galactic nuclei (Hamann \& Ferland 1992,
HF93b), 2) central star clusters with enhanced supernova (and
perhaps nova) rates due to mass accreted onto stars as they plunge
through QSO accretion disks (Artimowitz, Lin \& Wampler 1993),
3) star formation inside QSO accretion disks (Silk \& Rees 1998,
Collin 1998), and 4) nucleosynthesis without stars inside
accretion disks (Jin, Arnett \& Chakrabarti 1989, Kundt 1996).
\subsection{Occam's Razor: The Case for Normal Galactic Chemical
Evolution}
The first scenario listed above, for normal galactic
chemical evolution, is most compelling because 1) it is the only
one of these processes known to occur and 2) it is sufficient to
explain the QSO data. In particular, the stars in the centers of
massive galaxies today are (mostly) old and metal rich
(Bica, Arimoto \& Alloin 1988,
Bica, Alloin \& Schmidt 1990,
Gorgas, Efstathiou \& Arag\' on Salamanca 1990, Bruzual {\it et~al.}
1997, Vazdekis {\it et~al.} 1997, Jablonka, Alloin \& Bica 1992,
Jablonka, Martin \& Arimoto 1996,
Feltzing \& Gilmore 1998, Worthy, Faber \& Gonzalez 1992,
Kuntschner \& Davies 1997,
Sansom \& Proctor 1998, Ortolani {\it et~al.} 1996,
Sil'chenko, Burenkov \& Vlasyuk 1998,
Idiart, de Freitas Pacheco, \& Costa 1996,
Fisher, Franx \& Illingworth 1995, Bressan, Chiosi \& Tantalo 1996).
The exact ages are uncertain, but there is growing evidence for most
of the star formation in massive spheroids
(ellipticals and the bulges of large spiral galaxies) occurring
at redshifts $z\ga2$--3, especially (but not only) for galaxies
in clusters (see also Renzini 1998,1997, Bernardi {\it et~al.} 1998,
Bruzual \& Magris 1997, Ellis {\it et~al.} 1997, Tantalo, Chiosi
\& Bressan 1998, Ivison {\it et~al.} 1998, Kodama \& Arimoto 1997,
Ziegler \& Bender 1997, Kauffmann 1996,
Van Dokkum {\it et~al.} 1998, Mushotzky \& Loewenstein 1997
Spinrad {\it et~al.} 1997, Stanford {\it et~al.} 1998, Heap {\it et~al.} 1998,
Barger {\it et~al.} 1998a,b). The star-forming (Lyman-break or
\hbox{Ly$\alpha$} -emission) objects
measured directly at $z\ga 3$ might be galactic or proto-galactic
nuclei in the throes of rapid evolution (Friaca \& Terlevich 1999,
Baugh {\it et~al.} 1998,
Steidel {\it et~al.} 1998 and 1999, Connolly {\it et~al.} 1997, Lowenthal {\it et~al.} 1997,
Trager {\it et~al.} 1997, Hu, Cowie \& McMahon 1998,
Franx {\it et~al.} 1997, Madau {\it et~al.} 1996, Giavalisco, Steidel \& Machetto
1996). These objects are more numerous than QSOs and some
have been measured at $z>5$ (Dey {\it et~al.} 1998,
Hu {\it et~al.} 1998, Weymann {\it et~al.} 1998), beyond the highest known
QSO redshift of $z\approx 5.0$ (Sloan Digital Sky Survey
press release 1998). On the theoretical
side, recent cosmic-structure simulations show that proto-galactic
condensations can form stars and reach solar or higher
metallicities at $z\ga 6$ (Gnedin \& Ostriker 1997).
Quasars might form in the most massive and most dense
of these early-epoch star-forming environments
(Turner 1991, Loeb 1993, Haehnelt \& Rees 1993, Miralda-Escude \&
Rees 1997, Haehnelt {\it et~al.} 1998, Spaans \& Carollo 1997). They might
also form preferentially in globally dense cluster environments,
based on the higher detection rates of star-forming galaxies near
high-$z$ QSOs (Djorgovski 1998).
The gas in these environments might have been long
ago ejected via galactic winds, consumed by central black holes
or diluted by subsequent infall, but its signature
remains in the old stars today. The mean stellar
metallicities\footnote{It is worth noting here that, because
of a significant time-delay in the iron enrichment,
O/H and Mg/H are better measures of the overall ``metallicity''
than Fe/H (see \S6 and Wheeler {\it et~al.} 1989).}
in the cores of massive low-redshift galaxies are
typically $\langle Z_{stars}\rangle \sim 1$--3~\Zsun\
(see refs. listed above).
Individual stars are distributed about the
mean with metallicities reflecting the gas-phase abundance
at the time of their formation. If the interstellar gas is well-mixed
and the abundances grow monotonically (as expected in simple
enrichment schemes, \S6), the gas-phase metallicity, $Z_{gas}$,
will always exceed $\langle Z_{stars}\rangle$. Only the most
recently formed stars will have metallicities as high as the gas.
Therefore, the most metal-rich stars today should reveal the
gas-phase abundances near the end of the last major
star-forming epoch.
In the bulge of our own Galaxy, the nominal value of
$\langle Z_{stars}\rangle$ is 1~\Zsun\ and the tail of the
distribution reaches $Z_{stars}\ga 3$~\Zsun , with even higher
values obtaining near the Galactic center (Rich 1988 and 1990,
Geisler \& Friel 1992, McWilliam \& Rich 1994, Minniti {\it et~al.}
1995, Tiede, Frogel \& Terndrup 1995, Terndrup, Sadler \& Rich
1995, Idiart {\it et~al.} 1996, Castro {\it et~al.} 1996, Bruzual {\it et~al.} 1997).
The gas-phase metallicity should therefore have been
$Z_{gas}\ga 3$~\Zsun\ after most of the Bulge star formation
occurred. Simple chemical evolution models
indicate more generally that $Z_{gas}$ should be $\sim$2
to 3 times $\langle Z_{stars}\rangle$ in spheroidal systems
like galactic nuclei (Searle \& Zinn 1978, Tinsley 1980, Rich
1990, Edmunds 1992, de Fretas Pacheco 1996). Thus the observations
of $\langle Z_{stars}\rangle \sim 1$--3~\Zsun\ suggest that gas
with $Z_{gas}\sim 2$--9~Z$_{\odot}$ once existed in these
environments.
We might therefore expect to find
$2\la Z\la 9$~\Zsun\ in QSOs, as long as 1) most of the
local star formation occurred before the QSOs ``turned on'' or
became observable and 2) the metal-rich gas produced by that
star formation was not substantially diluted or ejected.
These expectations are consistent with the abundance estimates
reported above (\S4). More exotic enrichment schemes
are therefore not needed to explain the QSO data.
\section{More Insights from Galactic Chemical Evolution}
If we assume that QSO environments were indeed enriched by
normal stellar populations, then we can use the results from galactic
abundance and chemical evolution studies to interpret the QSO data.
Here we describe some relevant galactic results (see Wheeler,
Sneden \& Truran 1989 for a general review).
\subsection{The Galactic Mass-Metallicity Relation}
One important result from galaxy studies
is the well-known mass--metallicity
relationship among ellipticals and spiral bulges (Faber 1973,
Faber {\it et~al.} 1989,
Bender, Burstein \& Faber 1993, Zaritsky, Kennicutt \& Huchra 1994,
Jablonka {\it et~al.} 1996, Coziol {\it et~al.} 1997). This relationship is
attributed to the action of galactic winds;
massive galaxies reach higher metallicities because they have
deeper gravitational potentials and are better able to retain
their gas against the building thermal pressures
from supernovae (Larson 1974,
Arimoto \& Yoshii 1987, Franx \& Illingworth 1990). Low-mass
systems eject their gas before high $Z$'s are attained.
Quasar metallicities should be similarly tied to
the gravitational binding energy of the local star-forming
regions and, perhaps, to the total masses of their host galaxies
(\S7.1 below).
\subsection{Specific Abundance Predictions}
Another key result is the abundance behaviors of N and
Fe relative to the $\alpha$ elements such as O, Mg and Si.
HF93b constructed 1-zone infall models of galactic chemical
evolution to illustrate these behaviors in different
environments. Figure 13 plots the results for two scenarios
at opposite extremes. Both use the same nucleosynthetic
yields, but the ``Giant Elliptical'' model has much faster evolution
rates and a flatter IMF (more favorable to high-mass stars)
compared to the ``Solar Neighborhood'' (or spiral disk) case.
The Giant Elliptical evolves passively (without further star
formation) after $\sim$1~Gyr because the gas is essentially
exhausted. The parameters used in these calculations were
based on standard galactic infall models (e.g.
Arimoto \& Yoshii 1987, Matteucci \& Tornamb\'e 1987,
Matteucci \& Francois 1989, Matteucci \& Brocato 1990,
K\"oppen \& Arimoto 1990). However, the results are only
illustrative and not intended to match entire galaxies.
For example, evolution like the Giant Elliptical model might
occur in just the central cores of extreme high-mass
galaxies (cf. Friaca \& Terlevich 1998).
\begin{figure}[h]
\plotfiddle{figure13.ps}{2.3in}{0.0}{50.0}{50.0}{-200.0}{-475.0}
\end{figure}
\begin{quotation}
\noindent Fig. 13 --- Logarithmic gas-phase abundance ratios
normalized to solar for the two evolution models
discussed in \S6.2 (adapted from HF93b). Two scenarios
for the N enrichment are shown (thin solid lines); one
with secondary only and the other with secondary+primary
(causing a plateau in N/O at low $Z$).
\end{quotation}
\subsection{Fe/$\alpha$ as a Clock}
At early times the abundance evolution
is controlled by short-lived massive
stars, mainly via type II supernovae (SN~II's).
The $\alpha$ elements, such as O and Mg, come almost
exclusively from these objects, but Fe has a large delayed
contribution from type Ia supernovae (SN~Ia's) --- whose
precursors are believed to be intermediate mass stars in close
binaries (Branch 1998). The predicted time delay is roughly
1~Gyr based on the IMF-weighted stellar lifetimes (Fig. 13,
Greggio \& Renzini 1983, Matteucci \& Greggio 1986).
The actual delay is uncertain, but recent estimates
are in the range $\sim$0.3 to 3~Gyr (Matteucci 1994, Yoshii,
Tsujimoto \& Nomoto 1996, Yoshii {\it et~al.} 1998).
Because this delay does not depend on any of the global
evolution time scales (e.g. the star formation rate, etc.),
Fe/$\alpha$ can serve as an absolute ``clock'' for
constraining the ages of star-forming environments (Tinsley 1979,
Thomas, Greggio \& Bender 1998).
Observations of
metal-poor Galactic stars suggest that the baseline value of
[Fe/$\alpha$] due to SN~II's alone is nominally
$-$0.7 to $-$0.4 (Israelian, Garcia
\& Rebolo 1998, Nissen {\it et~al.} 1994, King 1993,
Gratton 1991, Magain 1989, Barbuy 1988, also
de Freitas Pacheco 1996),
which is slightly larger than the prediction in Figure 13.
The subsequent increase caused by SN~Ia's is a factor of a few
or more. Note that the increase in Fe/$\alpha$ should be
larger in rapidly-evolving spheroidal systems because 1) by the
time their SN~Ia's ``turn on,'' there is relatively little gas
left and each SN~Ia has a greater effect, also 2) their rapid early
star formation means that the SN~Ia's occurring later are more
nearly synchronized. The net result can
be substantially super-solar Fe/$\alpha$ in the gas (even
though Fe/$\alpha$ is sub-solar in most stars).
\subsection{Nitrogen Abundances}
Nitrogen also exhibits a delayed enhancement, although not
on a fixed time scale like Fe/$\alpha$. Nitrogen's selective
behavior is due to secondary CNO nucleosynthesis, where N
forms out of pre-existing C and O. Studies
of galactic HII regions indicate that secondary processing
dominates at metallicities above $\sim$0.2 \Zsun ,
resulting in N/O scaling like O/H (or N~$\propto Z^2$) in that
regime. At lower metallicities, primary N can be more
important based on an observed plateau in [N/O] at roughly $-$0.7
(see Tinsley 1980, Vila-Costas \& Edmunds 1993, Thurston, Edmunds
\& Henry 1996, Van Zee, Skillman \& Salzer 1998,
Kobulnicky \& Skillman 1998, Thuan, Izotov \& Lipovetsky 1995,
Izotov \& Thuan 1999, but see also Garnett 1990, Lu {\it et~al.} 1998).
The models in Figure 13 show two N/O behaviors, for secondary
only and secondary+primary, where the latter has a low-$Z$ plateau
forced to match the HII region data. Notice that the secondary growth
in N/O can be shifted down considerably from the simple theoretical
relation [N/O]~=~[O/H] because of the delays related to stellar
lifetimes. We therefore have a strong prediction, based on both
observations and these simulations, that measured values of
[N/O]~$\ga$~0 imply $Z\ga$~\Zsun\ --- especially in fast-evolving
spheroidal systems. This prediction was exploited above in
the analysis of QSO BELs (\S2.6, Shields 1976).
\section{Implications of QSO Abundances}
\subsection{High-Redshift Star Formation}
We can conclude from the previous sections that QSOs are
associated with vigorous star formation,
consistent with the early-epoch evolution of
massive galactic nuclei or dense proto-galactic clumps (\S5).
However, QSO abundances provide new constraints.
For example, the general result for $Z\ga$~\Zsun\
suggests that most of the enrichment and local star formation
occurs before QSOs ``turn on'' or become observable.
The enrichment times can be
so short in principle (Fig. 13, HF93b) that the star
formation might also be coeval with QSO formation. In any
event, the enrichment times cannot be much longer that
$\sim$1~Gyr for at least the highest redshift objects
(depending on the cosmology, Fig. 1).
If the QSO metallicities representative of a well-mixed
interstellar medium, we can conclude further that the star
formation was extensive. That is, a significant fraction
of the initial gas must be converted into stars and stellar
remnants to achieve $Z_{gas}\ga$~\Zsun . The exact fraction
depends on the IMF. A solar neighborhood IMF (Scalo 1990,
as in the Solar Neighborhood model of \S6) would lead to
mass fractions in gas of only $\la$15\% at $Z\sim$~\Zsun ,
and would not be able to produce $Z_{gas}$ above a few \Zsun\
at all. Flatter IMFs (favoring massive stars) could reach
$Z_{gas}\ga$~\Zsun\ while consuming less of the gas.
For example, the gas fraction corresponding to $Z_{gas}\sim$~\Zsun\
in the Giant Elliptical model of \S6.2 is nearly 70\%.
Figure 7 in \S2.6.3 illustrates the main star formation characteristics
required by the QSO data. The solid curves on the right-hand side
of that figure show theoretical BEL ratios from photoionization simulations
that use nominal BELR parameters and abundances from the two
chemical evolution models in Figure 13 (see HF93b for more details).
The evolution is assumed to
begin with the Big Bang and the conversion of time into redshift
assumes a cosmology with $H_o = 65$~km~s$^{-1}$~Mpc$^{-1}$,
$\Omega_{M} = 1$ and $\Omega_{\Lambda} = 0$ (Fig. 1).
The main results are that the
Solar Neighborhood evolution is too slow and, in any case,
does not reach high enough metallicities or nitrogen enhancements
to match most of the high-redshift QSOs. Much shorter time
scales and usually higher metallicities, as in the Giant Elliptical
simulation, are needed.
A trend in the NV BELs suggests further that the
metallicities are typically higher in more luminous QSOs
(\S2.6.3). That result needs confirmation, but it could
result naturally from a mass--metallicity relationship
among QSO host galaxies that is similar (or identical to)
the well-known relation in low-redshift galaxies (\S6, HF93b).
By analogy with the galactic relation, the most luminous
and metal-rich QSOs might reside in the most dense
or massive host environments. This situation would be
consistent with studies showing that QSO luminosities,
QSO masses, and central black hole masses in galactic nuclei
all appear to correlate with the mass of the
surrounding galaxies (McLeod {\it et~al.} 1999, McLeod \& Rieke 1995,
Bahcall {\it et~al.} 1997, Magorrian {\it et~al.} 1998, Laor 1998,
also Haehnelt \& Rees 1993). A direct application of
the galactic mass-metallicity relation suggests
that metal-rich QSOs reside in galaxies
(or proto-galaxies) that are minimally as massive
(or as tightly bound) as our own Milky Way.
\subsection{Fe/$\alpha$: Timescales and Cosmology}
One of the most interesting predictions from galactic
studies (\S6) is that Fe/$\alpha$ ratios in QSOs
might constrain the epoch of their first star
formation and perhaps the cosmology. In particular,
large Fe/$\alpha$ ratios (solar or higher) would suggest
that the local stellar populations are at least
$\sim$1~Gyr old. At the highest QSO redshifts ($z\sim 5$), this age
constraint would push the epoch of first star formation beyond
the limits of current direct observation,
to $z>6$ (Fig. 1). The $\sim$1~Gyr constraint
would also be difficult to
reconcile with $\Omega_M\approx 1$ in Big Bang cosmologies.
Conversely, measurements of low Fe/$\alpha$ would suggest that
the local stellar populations are younger than $\sim$1~Gyr
(although we could not rule out the possibility that only
SN~II's contributed to the enrichment for some reason).
Some BEL studies have already suggested that Fe/$\alpha$ is
above solar in $z>4$ QSOs (\S2.6.4).
\subsection{Comparisons to Other Results}
Quasar abundances should be viewed in the context of other
measures of the metallicity and star formation at high redshifts.
Damped-\hbox{Ly$\alpha$}\ absorbers in QSO spectra,
which probe lines of sight through large intervening galaxies
(probably spiral disks, Prochaska \& Wolfe 1998),
have mean (gas-phase) metallicities of order 0.05~\Zsun\ at
$z\ga 2$ (Lu {\it et~al.} 1996, Pettini {\it et~al.} 1997,
Lu, Sargent \& Barlow 1998, Prochaska \& Wolfe 1999).
The \hbox{Ly$\alpha$}\ forest absorbers, which
presumably probe much more extended and tenuous inter-galactic
structures (Rauch 1998), typically have metalicities $<$0.01~\Zsun\
at high redshifts (Rauch, Haehnelt \& Steinmetz 1997,
Songalia \& Cowie 1996, Tytler {\it et~al.} 1995).
The much higher metal abundances near QSOs are consistent
with the rapid and more extensive evolution
expected in dense environments (Gnedin \& Ostriker 1997).
Perhaps this evolution is similar to that occurring in the many
star-forming objects that are now measured directly
at redshifts comparable to, and greater than, the QSOs (see
refs. in \S5.1).
The detections of strong dust and molecular gas emissions from
QSOs support the evidence from their high abundances that
considerable local star formation preceded the QSO epoch. The
dust and molecules, presumably manufactured by stars, appear
even in QSOs at $z\ga4$ (Isaac {\it et~al.} 1994,
Omont {\it et~al.} 1996, Guilloteau {\it et~al.} 1997).
\section{Future Prospects}
We now have the observational and theoretical abilities
to test and dramatically extend all of the QSO abundance work
discussed above. The most pressing needs are to
1) develop more independent abundance
diagnostics, and 2) obtain more and better data to
compare diagnostics in
large QSO samples --- spanning a range of redshifts, luminosities,
radio properties, etc. Absorption line studies
will benefit generally from higher spectral resolutions and wider
wavelength coverage, providing more accurate column densities
and more numerous constraints
on the coverage fractions, ionizations and abundances (\S3).
BEL studies should include more of the
weaker lines, such as OVI~$\lambda$ 1034, CIII~$\lambda$ 977,
NIII~$\lambda$ 991 and the intercombination lines, whenever
possible (\S2). Theoretical analysis of the
FeII/MgII emission ratios, in particular, is needed to test the
tentative conclusion for high Fe/Mg abundances. This and other
BEL results should be tested further by examining the same lines
(or same elements) in intrinsic NAL systems. The steady
improvement in our observational capabilities at all wavelengths
will provide many more diagnostic opportunities.
Below are some specific issues that new studies might address.
1) More data at high redshifts will
constrain better the epoch and extent of early star formation
associated with QSOs.
2) Reliable measurements of Fe/$\alpha$ will further constrain
the epoch of first star formation and, perhaps, the cosmology
via the $\sim$1~Gyr enrichment clock.
3) Better estimates of the metal-to-metal ratios generally
will reveal more specifics of the star formation histories
via comparisons to well-studied galactic environments
and theoretical nucleosynthetic yields.
4) Abundances for QSOs spanning a wide range of luminosities and
redshifts will isolate any evolutionary (redshift) trends and test
the tentative luminosity--$Z$ relationship. This relationship might
prove to be a useful indicator of the total masses or
densities of the local stellar populations by analogy with the
mass--$Z$ trend in nearby galaxies.
5) The range of QSO metallicities at a given redshift and luminosity
will help constrain the extent of star formation
occurring before QSOs
turn on or become observable. Are there any low metallicity QSOs?
6) Combining the QSO abundances with direct imaging studies
of their host galaxies should test ideas about the chemical enrichment
and help us interpret data at the highest redshifts
where direct imaging is (so far) not possible. For example, are QSOs
in large galaxies (e.g. giant ellipticals) more metal-rich than others?
7) Correlations between the abundances and other properties of
QSOs, such
as radio-loudness or UV--X-ray continuum shape, might reveal new
environmental factors in the enrichment or systematic
uncertainties in our abundance derivations.
8) Observations with wide wavelength coverage would allow us to
compare abundances derived from the narrow emission lines (in the
rest-frame optical) to BEL and NAL data
in the same objects. These diverse diagnostics
might provide crude abundance maps of QSO host galaxies.
9) How do QSO abundances compare to their low-redshift
counterparts, the Seyfert galaxies and active galactic
nuclei (AGNs)? Low-redshift metallicities might be
less than the QSOs due to recent mergers or gaseous infall.
\bigskip\bigskip\bigskip
\noindent ACKNOWLEDGEMENTS
\noindent We are grateful to G Burbidge for his patience and encouragement.
We also thank KT Korista and JC Shields for helpful comments on this
manuscript, and TA Barlow, N Arav and VT Junkkarinen for useful
discussions. GF thanks the Canadian Institute for Theoretical Astrophysics
for their hospitality during a sabbatical year, and acknowledges support
from the Natural Science and Engineering Research Council of Canada
through CITA. The work of FH was supported by NASA grant NAG 5-3234.
Research in nebular astrophysics at the University of Kentucky is supported
by the NSF through grant 96-17083 and by NASA through its ATP (award NAG
5-4235) and LTSA programs.
\newpage
\section{Literature Cited}
\parskip=0pt
\leftskip=0.5in
\parindent=-0.5in
\def {\it Annu. Rev. Astron. Astrophys.} {{\it Annu. Rev. Astron. Astrophys.} }
\def {\it Ap. J.} {{\it Ap. J.} }
\def {\it Ap. J. Suppl.} {{\it Ap. J. Suppl.} }
\def {\it Astron. J.} {{\it Astron. J.} }
\def {\it Astron. Astrophys.} {{\it Astron. Astrophys.} }
\def {\it MNRAS} {{\it MNRAS} }
\def {\it PASP} {{\it PASP} }
Aldcroft TL, bechtold J, Elvis M. 1994, {\it Ap. J. Suppl.} 93, 1
Anderson SF, Weymann RJ, Foltz CB, Chaffee FH.
1987, {\it Astron. J.} 94, 278
Arav N. 1996, {\it Ap. J.} 465, 617
Arav N. 1997, in Mass Ejection From AGN, eds. R Weymann, I Shlosman,
N Arav, ASP Conf. Series, 128, 208
Arav N, Barlow TA, Laor A, Sargent WLW, Blandford RD. 1998, {\it MNRAS} 297, 990
Arav N, Korista KT, de Kool M, Junkkarinen VT, Begelman MC. 1999,
{\it Ap. J.} in press (A99)
Aretxaga I, Terlevich RJ, Boyle BJ. 1998, {\it MNRAS} 296, 643
Arimoto N Yoshii Y. 1987, {\it Astron. Astrophys.} 173, 23
Artymowicz P, Lin DNC, Wampler EJ. 1993, {\it Ap. J.} 409, 592
Bahcall JN, Kirhakos S, Saxe DH, Schneider DP. 1997, {\it Ap. J.}
479, 642
Bahcall JN, Sargent WLW, Schmidt M. 1967, {\it Ap. J.} 149, L11
Baldwin JA. 1977a, {\it MNRAS} 178, 67P
Baldwin JA. 1977b, {\it Ap. J.} 214, 679
Baldwin JA, \& Netzer H. 1978, {\it Ap. J.} 226, 1
Baldwin JA, Ferland GJ, Korista KT, Carswell RF,
Hamann F, {\it et~al.} 1996, {\it Ap. J.} 461, 664
Baldwin JA, Ferland GJ, Korista KT, Verner D. 1995, {\it Ap. J.} 455, L119
Barbuy B. 1988, {\it Astron. Astrophys.} 191, 121
Barger AA, Aragon-Salamanca A, Smail I, Ellis RS, Couch WJ,
{\it et~al.} 1998a, {\it Ap. J.} 501, 522
Barger AA, Cowie LL, Trentham N, Fulton E, Hu EM, {\it et~al.} 1998b,
preprint (astro-ph/9809299)
Barlow TA. 1993, Ph.D. Dissertation, University of California
-- San Diego
Barlow TA. 1998, private comm.
Barlow TA, Hamann F, Sargent WLW. 1997, in
Mass Ejection From AGN, eds. R Weymann, I Shlosman, N Arav,
ASP Conf. Series, 128, 13
Barlow TA, Junkkarinen VT. 1994, BAAS, 26, 1339
Barlow TA, Junkkarinen VT, Burbidge EM, Weymann RJ,
Morris SL, {\it et~al.} 1992, {\it Ap. J.} 397, 81
Barlow TA, Sargent WLW. 1997, {\it Astron. J.} 113, 136
Barthel PD, Tytler DR, Vestergaard M. 1997, in
Mass Ejection From AGN, eds. R Weymann, I Shlosman, N Arav,
ASP Conf. Series, 128, 48
Baugh CM, Cole S, Frenk CS, Lacey CG. 1998, {\it Ap. J.} 498, 504
Bender R, Burstein D, Faber SM. 1993, {\it Ap. J.} 411, 153
Bergeron J, Boiss\'e P. 1986, {\it Astron. Astrophys.} 168, 6
Bergeron J, Kunth D. 1983, {\it MNRAS} 205, 1053
Bergeron J, Petitjean P, Sargent WLW, Bahcall JN, Boksenberg A,
{\it et~al.} 1994, {\it Ap. J.} 436, 33
Bergeron J Stasi\'nska G. 1986, {\it Astron. Astrophys.} 169, 1
Bernardi M, Renzini A, da Costa LN, Wegner G, Victoria M,
{\it et~al.} 1998, preprint (astro-ph/9810066)
Bica E, Alloin D, Schmidt AA. 1990, {\it Astron. Astrophys.} 228, 23
Bica E, Arimoto N, Alloin D. 1988, {\it Astron. Astrophys.} 202, 8
Boyce PJ, Disney MJ, Bleaken DG. 1999, {\it MNRAS} 302, 39
Boyle B. 1990, {\it MNRAS} 243, 231
Boyle BJ, Terlevich RJ. 1998, {\it MNRAS} 293, 49
Branch D. 1998, {\it Annu. Rev. Astron. Astrophys.} 36, 17
Bressan A, Chiosi C, Tantalo R. 1996, {\it Astron. Astrophys.} 311, 425
Brotherton MS, Van Breugel W, Smith RJ, Boyle BJ, Shanks T. 1998b,
{\it Ap. J.} 505, 7
Brotherton MS, Wills BJ, Dey A, Van Breugal W, Antonucci R. 1998a,
{\it Ap. J.} 501, 110
Brunner H, Mueller C, Friedrich P, Doerrer T, Staubert R, {\it et~al.}
1997, {\it Astron. Astrophys.} 326, 885
Bruzual G, Barbuy B, Ortolani S, Bica E, Cuisinier F. 1997,
{\it Astron. J.} 114, 1531
Bruzual G, Magris G. 1997, in the STScI Symp on the Hubble Deep Field,
(astro-ph/9707154)
Burbidge G, Burbidge M. 1967, in Quasi-Stellar Objects,
WH Freeman and Co.
Burles S, Tytler D. 1996, {\it Ap. J.} 460, 584
Carroll SM, Press WH. 1992, {\it Annu. Rev. Astron. Astrophys.} 30, 499
Carballo R, Sanchez SF, Gonzalas-Serrano JI, Benn CR, Vigotti M.
1998, {\it Astron. J.} 115, 1234
Castro S, Rich RM, McWilliam A, Ho LC, Spinrad H, {\it et~al.} 1996,
{\it Astron. J.} 111, 2439
Cavaliere A, Vittorini V. 1998, in The Young Universe, eds. S D'Odorico,
A Fontana, E Giallongo, ASP Conf. Ser., 146, 28
Chatzichristou ET, Vanderriest C, Jaffe W. 1999, {\it Astron. Astrophys.} 343, 407
Cohen MH, Ogle PM, Tran HD,
Vermeulen RC, Miller JS, {\it et~al.} 1995, {\it Ap. J.} 448, L77
Collin S. 1998, preprint
Connolly AA, Szalay AS, Dickenson M, SubbaRao MU, Brunner RJ. 1997,
{\it Ap. J.} 486, L11
Cozial R, Contini T, Davoust E, Consid\`ere S. 1997, {\it Ap. J.} 481, L67
Davidson K. 1977, {\it Ap. J.} 218, 20
Davidson K, Netzer H. 1979, Rev. Mod. Phys., 51, 715, (DN79).
de Freitas Pacheco JA. 1996, {\it MNRAS} 278, 841
Dey A, Spinrad H, Stern D, Graham JR, Chaffee FH.
1998, {\it Ap. J.} 498, L93
Djorgovski SG. 1998, in Fundamental Parameters of Cosmology,
ed. Y Giroud-H\`eraud, (Gif sur Yvette:Editions Fronti\`eres),
in press
Edmunds MG. 1992, in Elements and the Cosmos, eds. MG Edmunds, R Terlevich,
(Cambridge Univ. Press:New York), p.289
Ellis R, Smail I, Dressler A, Couch WJ, Oemler WJ, {\it et~al.}
1997, {\it Ap. J.} 483, 582
Faber SM. 1973, {\it Ap. J.} 179, 423
Faber SM, Wegner G, Burstein D, Davies RL, Dressler A,
{\it et~al.} 1989, {\it Ap. J. Suppl.} 69, 763
Feltzing S, Gilmore G. 1998, in Galaxy Evolution: Connecting the
Distant Universe with the Local Fossil Record, p. 71
Ferland GJ. 1999, in Quasars as Standard Candles for Cosmology,
eds. J Baldwin, GJ Ferland, KT Korista, in press
Ferland GJ, Baldwin JA, Korista KT, Hamann F, Carswell
RF, {\it et~al.} 1996, {\it Ap. J.} 461, 683
Ferland GJ, Korista KT, Verner DA, Ferguson JW, Kingdon JB, {\it et~al.}
1998, {\it PASP} 110, 761
Ferland GJ, Persson SE. 1989, {\it Ap. J.} 347, 656
Ferland GJ, Peterson BM, Horne K, Welsch WF,
Nahar SN. 1992, {\it Ap. J.} 387, 95
Ferland GJ, Shields GA. 1985, in Astrophysics of Active Galaxies
and Quasi-Stellar Objects, ed. JS Miller, Univ. Sci. Books, p. 157
Fisher D, Franx M, Illingworth G. 1995, {\it Ap. J.} 448, 119
Foltz CB, Chaffee F, Weymann RJ, Anderson SF.
1988, in QSO Absorption Lines: Probing the
Universe, eds. JC Blades, DA Turnshek, CA Norman (Cambridge:
Cambridge Univ Press), p. 53
Foltz CB, Weymann RJ, Peterson BM, Sun L, Malkan MA, {\it et~al.}
1986, {\it Ap. J.} 307, 504
Franceschini A, Gratton R. 1997, {\it MNRAS} 286, 235
Francis PJ, Koratkar A. 1995, {\it MNRAS} 274, 504
Franx M, Illingworth GD. 1990, {\it Ap. J.} 359, L41
Franx M, Illingworth GD, Kelson DD, Van Dokkum PG,
Tran K. 1997, {\it Ap. J.} 486, 75
Friaca ACS, Terlevich RJ. 1998, {\it MNRAS} 298, 399
Friaca ACS, Terlevich RJ. 1999, {\it MNRAS} in press
Frisch H. 1984, in Methods in Radiative Transfer, ed. W Kalkofen
(Cambridge:Cambridge Univ. Press), 65
Gallagher SC, Brandt WN, Sambruna RM, Mathur S, Yamasaki N. 1999,
{\it Ap. J.} in press
Ganguly R, Eracleous M, Charlton JC, Churchill CW. 1999, {\it Astron. J.}
in press
Garnett DR. 1990, {\it Ap. J.} 363, 142
Gaskell CM, Shields GA, and Wampler EJ. 1981, {\it Ap. J.} 249, 443
Geisler D, Friel DE. 1992, {\it Astron. J.} 104, 128
Giavalisco M, Steidel CC, Macchetto FD. 1996, {\it Ap. J.} 470, 189
Gnedin NY, Ostriker JP. 1997, {\it Ap. J.} 486, 581
Goodrich RW, Miller JS. 1995, {\it Ap. J.} 448, L73
Gorgas J, Efstathiou G, Arag\'on Salamanca AA. 1990, {\it MNRAS} 245, 217
Grandi SA. 1981, {\it Ap. J.} 251, 451
Gratton RG. 1991, in Evolution of Stars: The Photospheric
Abundance Connection, eds. G. Michaud, AV Tutukov,
IAU Symp. 145, (Montreal:Univ. Montreal), 27
Green PJ, Aldcroft TL, Mathur S, Shartel N. 1997, {\it Ap. J.} 484, 135
Green PJ, Mathur S. 1996, {\it Ap. J.} 462, 637
Greggio L, and Renzini A. 1983 {\it Astron. Astrophys.} 118, 217
Grevesse N, Anders E. 1989, in Cosmic Abundances of Matter,
AIP Conf. Proc. 183, ed. CI Waddington (New York:AIP), 1
Grillmair CJ, Turnshek DA. 1987, in QSO Absorption Lines: Probing the
Universe, Poster Papers, eds. JC Blades, C Norman, DA Turnshek,
p. 1
Guilloteau S, Omont A, McMahon RG, Cox P, Petitjean P. 1997,
{\it Astron. Astrophys.} 328, L1
Haas M, Chini R, Maisenheimer K, Stickel M, Lemke D, {\it et~al.} 1998,
{\it Ap. J.} 503, L109
Haehnelt MG, Natarajan P, Rees MJ. 1998, {\it MNRAS} 300, 817
Haehnelt MG, Rees MJ. 1993, {\it MNRAS} 263, 168
Haiman Z, Loeb A. 1998, {\it Ap. J.} 503, 505
Hamann F. 1997, {\it Ap. J. Suppl.} 109, 279 (H97)
Hamann F. 1998, {\it Ap. J.} 500, 798
Hamann F, Barlow TA, Beaver EA, Burbidge EM, Cohen
RD, {\it et~al.} 1995, {\it Ap. J.} 443, 606
Hamann F, Barlow TA, Junkkarinen V. 1997c, {\it Ap. J.} 478, 87
Hamann F, Barlow TA, Cohen RD, Junkkarinen V, Burbidge EM.
1997b, {\it Ap. J.} 478, 80 (H97b)
Hamann F, Barlow TA, Cohen RD, Junkkarinen V, Burbidge EM.
1997e, in Mass Ejection From AGN, eds. R Weymann, I Shlosman,
N Arav, ASP Conf. Series, 128, 187
Hamann F, Beaver EA, Cohen RD, Junkkarinen V, Lyons RW, {\it et~al.} . 1997d,
{\it Ap. J.} 488, 155
Hamann F, Chaffee R, Weymann RJ, Barlow TA, Junkkarinen VT. 1999b,
in prep.
Hamann F, Cohen RD, Shields JC, Burbidge EM, Junkkarinen VT, {\it et~al.}
1998, ApJ, 496, 761
Hamann F, Ferland GJ. 1992, ApJL, 391, L53
Hamann F, Ferland GJ. 1993a, Rev. Mex. Astr. Astrof., 26, 53
Hamann F, Ferland GJ. 1993b, {\it Ap. J.} 418, 11, (HF93b)
Hamann F, Korista KT. 1996, {\it Ap. J.} 464, 158
Hamann F, Korista KT, Ferland GJ. 1999a, in prep.
Hamann F, Korista KT, Morris SL. 1993, {\it Ap. J.} 415, 541
Hamann F, Shields JC, Cohen RD, Junkkarinen VT, Burbidge EM.
1997a, in Emission Lines From Active Galaxies: New Methods and Techniques,
IAU Col. 159, eds. BM Peterson, F-Z Cheng, AS Wilson, ASP Conf. Ser.,
113, 96
Hartquist TW, Snijders MAJ. 1982, Nature, 299, 783
Heap SR, Brown TM, Hubeny I, Landsman W, Yi S, {\it et~al.} 1998, {\it Ap. J.} 492, L131
Hines DC, Low FJ, Thompson RI, Weymann RJ, Storrie-Lombardi LJ. 1999,
{\it Ap. J.} 512, 140
Hines DC, Wills BJ. 1995, {\it Ap. J.} 448, L69
Hu EM, Cowie LL, McMahon RG. 1998, {\it Ap. J.} 502, L99
Idiart TP, De Freitas Pacheco JA, Costa RDD.
1996, {\it Astron. J.} 112, 2541
Isaac KG, McMahon RG, Hills RE, Withington S. 1994, {\it MNRAS} 269, 28
Isotov YI, Thuan TX. 1999, {\it Ap. J.} in press
Israelian G, L\'opez RJG, Rebolo R. 1998, {\it Ap. J.} 507, 805
Ivison RJ, Dunlop JS, Hughes DH, Archibald EN, Stevens JA, {\it et~al.}
1998, {\it Ap. J.} 494, 211
Jablonka P, Alloin D, Bica E. 1992, {\it Astron. Astrophys.} 260, 97
Jablonka P, Martin P, Arimoto N. 1996, {\it Astron. J.} 112, 1415
Johansson S, Jordan C. 1984, {\it MNRAS} 210, 239
Jenkins EB. 1996, {\it Ap. J.} 471, 292
Jin L, Arnett WD, Chakrabarti SK. 1989, {\it Ap. J.} 336, 572
Junkkarinen VT. 1980, Ph.D. Dissertation, University of California
-- San Diego
Junkkarinen VT. 1998, private comm.
Junkkarinen VT, Beaver EA, Burbidge EM, Cohen RC,
Hamann F, {\it et~al.} 1997, in
Mass Ejection From AGN, eds. R Weymann, I Shlosman, N Arav,
ASP Conf. Series, 128, 220
Junkkarinen VT, Burbidge EM, Smith HE. 1983, {\it Ap. J.} 265,
51
Junkkarinen VT, Burbidge EM, Smith HE. 1987, {\it Ap. J.} 317,
460
Katz N, Quinn T, Bertschinger E, Gelb JM. 1994, {\it MNRAS} 270, L71
Kauffmann G. 1996, {\it MNRAS} 281, 487
King JR. 1993, {\it Astron. J.} 106, 1206
Kinney AL, Rivolo AR, Koratkar AR. 1990, {\it Ap. J.} 357, 338
Kirkman D, Tytler D. 1997, {\it Ap. J.} 489, L123
Kobulnicky HA, Skillman ED. 1998, {\it Ap. J.} 497, 601
Kodama T, Arimoto N. 1997, {\it Astron. Astrophys.} 320, 41
K\" oppen J, Arimoto N. 1990, {\it Astron. Astrophys.} 240, 22
Korista KT, Alloin D, Barr P, Clavel J, Cohen RD, {\it et~al.} 1995,
{\it Ap. J. Suppl.} 97, 285
Korista KT, Baldwin JA, Ferland GJ. 1998, {\it Ap. J.} 507, 24
Korista KT, Baldwin JA, Ferland GJ, Verner D. 1997b, {\it Ap. J. Suppl.} 108, 401
Korista KT, Ferland GJ, Baldwin BA. 1997a, {\it Ap. J.} 487, 555
Korista KT, Hamann F, Ferguson J, Ferland GJ. 1996,
{\it Ap. J.} 461, 641
Korista KT, Voit GM, Morris SL, Weymann RJ. 1993, {\it Ap. J. Suppl.} 88, 357
Korista KT, Weymann RJ, Morris SL, Kopko M, Turnshek
DA, {\it et~al.} 1992, {\it Ap. J.} 401, 529
Kormendy J, Bender R, Evans AS, Richstone D. 1998,
{\it Astron. J.} 115, 1823
Krolik J, Voit GM. 1998, {\it Ap. J.} 497, L5
Kundt W. 1996, {\it Astrophys. Sp. Sci.} 235, 319
Kundtschner H, Davies RL. 1997, {\it MNRAS} 295, L29
Kwan J. 1990, {\it Ap. J.} 353, 123
Kwan J, Krolik J. 1981, {\it Ap. J.} 250, 478, (KK81)
Laor A. 1998, in Quasars as Standard Candles for Cosmology, eds.
J Baldwin, GJ Ferland, KT Korista, in press
Laor A, Bahcall JN, Jannuzi BT, Schneider DP, Green RF. 1995, {\it Ap. J. Suppl.} 99, 1
Laor A, Jannuzi BT, Green RF, Boroson TA. 1997, {\it Ap. J.} 489, 656
Larson RJ. 1974, {\it MNRAS} 169, 229
Loeb A. 1993, {\it Ap. J.} 404, L37
Loeb A, Rasio FA. 1994, {\it Ap. J.} 432, 52
Lowenthal JD, Koo DC, Guzman R, Gallego J, Phillips AC. 1997, {\it Ap. J.} 481, 673
Lu L, Sargent WLW, Barlow TA. 1998, {\it Astron. J.} 115, 55
Lu L, Sargent WLW, Barlow TA, Churchhill CW, Vogt SS. 1996, {\it Ap. J. Suppl.} 107, 475
Lu L, Savage BD. 1993, {\it Ap. J.} 403, 127
Madau P, Ferguson HC, Dickenson ME, Giavalisco M, Steidel CC,
{\it et~al.} 1996, {\it MNRAS} 283, 1388
Magain P. 1989, {\it Astron. Astrophys.} 209, 211
Magorrian J, Tremaine S, Richstone D, Bender R, Bower G,
{\it et~al.} 1998, {\it Astron. J.} 115, 2285
Mathews WG, Capriotti ER. 1985, in Astrophysics of Active Galaxies
and Quasi-Stellar Objects, ed. JS Miller, Univ. Sci. Books, p. 183
Matteucci F. 1994, {\it Astron. Astrophys.} 288, 57
Matteucci F, Brocato E. 1990, {\it Ap. J.} 365, 539
Matteucci F, Francois P. 1989, {\it MNRAS} 239, 885
Matteucci F, Greggio L. 1986, {\it Astron. Astrophys.} 154, 279
Matteucci F, Tornamb\`e A. 1987. {\it Astron. Astrophys.} 185, 51
McCarthy PJ. 1993, {\it Annu. Rev. Astron. Astrophys.} 31, 639
McLeod KK, Rieke GH 1995, {\it Ap. J.} 454, L77
McLeod KK, Rieke GH, Storrie-Lombardi LJ. 1999, {\it Ap. J.} 511, L67
McLure RJ, Dunlop JS, Kukula MJ, Baum SA,
O'Dea CP, {\it et~al.} 1998, preprint (astro-ph/9809030)
McWilliam A, Rich RM. 1994, {\it Ap. J. Suppl.} 91, 749
Mihalas D. 1978, Stellar Atmospheres, WH Freeman \& Co.,
(San Francisco:Univ. Chicago Press)
Miller J, Tran H, Sheinis A. 1996, BAAS, 28, 1031
Minniti D, Olszewski EW, Liebert J, White SD, Hill JM,
{\it et~al.} 1995, {\it MNRAS} 277, 1293
Miralda-Escud\'e J, Rees MJ. 1997, {\it Ap. J.} 478, L57
M\"oller P, Jakobsen P, Perryman MAC. 1994, {\it Astron. Astrophys.} 287, 719
Morris SL, Weymann RJ, Foltz
CB, Turnshek DA, Shectman S, {\it et~al.} 1986, {\it Ap. J.} 310, 40
Murray N, Chiang J, Grossman SA, Voit GM. 1995, {\it Ap. J.} 451, 498
Mushotzky RF, Loewenstein M. 1997, {\it Ap. J.} 481, L63
Netzer H. 1990, in Active Galactic Nuclei, eds. RD Blandford,
H Netzer, L Woltjer, (Berlin:Springer), 57
Netzer H, Brotherton MS, Wills BJ, Han M, Baldwin JA, {\it et~al.}
1995, 448, 27
Netzer H, Wills BJ. 1983, {\it Ap. J.} 275, 445
Nissen PE, Gustafsson B, Edvardsson B, Gilmore G. 1994, {\it Astron. Astrophys.} 285, 440
Omont A, Petitjean P, Guilloteau S, McMahon RG, Solomon PM.
1996, Nature, 382, 428
Ortolani S, Renzini A, Gilmozzi R, Marconi G, Barbuy B, {\it et~al.}
1996, in Formation of the Galactic Halo...Inside and Out,
Eds. H Morrison, A Sarajedini, ASP Conf. Ser., 92, 96
Osmer PS. 1980, {\it Ap. J.} 237, 666
Osmer PS. 1998, in The Young Universe, eds. S D'Odorico, A Fontana,
E Giallongo, ASP Conf. Ser., 146, 1
Osmer PS, Porter AC, Green RF. 1994, {\it Ap. J.} 436, 678
Osmer PS, Shields, JC. 1999, in Quasars as Standard Candles for Cosmology,
eds. J Baldwin, GJ Ferland, KT Korista, in press
Osmer PS, Smith MG. 1976, {\it Ap. J.} 210, 276
Osmer PS, Smith MG. 1977, {\it Ap. J.} 213, 607
Osterbrock DE. 1977, {\it Ap. J.} 215, 733
Osterbrock DE. 1989, Astrophysics of Gaseous Nebulae and
Active Galactic Nuclei, University Science Press
Peimbert M. 1967, {\it Ap. J.} 150, 825
Penston M. 1987, {\it MNRAS} 229, 1P
Peterson BM. 1993, {\it PASP} 105, 1084
Petitjean P, Rauch M, Carswell RF. 1994, {\it Astron. Astrophys.} 291, 29
Petitjean P, Srianand R. 1999, {\it Astron. Astrophys.} in press
Pettini M, King DL, Smith LJ, Hunstead RW. 1997, {\it Ap. J.}
486, 665
Phillips MM. 1977, {\it Ap. J.} 215, 746
Phillips MM. 1978, {\it Ap. J.} 226, 736
Prochaska JX, Wolfe AM. 1998, {\it Ap. J.} 507, 113
Prochaska JX, Wolfe AM. 1999, {\it Ap. J.} in press
Rauch M. 1998, {\it Annu. Rev. Astron. Astrophys.} 36, 267
Rauch M, Haehnelt MG, Steinmetz M. 1997, {\it Ap. J.} 481, 601
Renzini A. 1997, {\it Ap. J.} 488, 35
Renzini A. 1998, in The Young Universe, eds. S D'Odorico, A Fontana,
E Giallongo, ASP Conf. Ser., 146, 298
Rich RM. 1988, {\it Astron. J.} 95, 828
Rich RM. 1990, {\it Ap. J.} 362, 604
Richards GT, York DG, Yanny B, Kollgaard RI, Laurent-Muehleisen SA,
{\it et~al.} 1999, {\it Ap. J.} in press
Saikia D, Kulkarni AR. 1998, {\it MNRAS} 298, L45
Sansom AE, Proctor RN. 1998, {\it MNRAS} 297, 953
Savage BD, Sembach KR. 1991, {\it Ap. J.} 379, 245
Savage BD, Tripp TM, Lu L. 1998, {\it Astron. J.} 115, 436
Savaglio S, Cristiani S, D'Odorico S, Fontana A, Giallong E, {\it et~al.}
1997, {\it Astron. Astrophys.} 318, 347
Savaglio S, D'Odorico S, M\"o ller P. 1994, {\it Astron. Astrophys.} 281, 331
Scalo JM. 1990, in Windows on Galaxies, eds. G Fabbiano,
JS Gallagher, A Renzini, (Dordrecht:Kluwer Academic Publishers)
p. 125
Schmidt GD, Hines DC. 1999, {\it Ap. J.} in press
Schneider DP. 1998, in Science with NGST, eds. EP Smith, A Koratkar,
ASP Conf. Ser., 133, 106
Schneider DP, Schmidt M, and Gunn JE. 1991, {\it Astron. J.} 102, 837
Searle L, Zinn R. 1978, {\it Ap. J.} 225, 357
Shaver PA, Hook IM, Jackson CA, Wall JV, Kellerman KI. 1998,
in Highly redshifted Radio Lines, eds. C Carilli, S Radford,
K Menton, G. Langston, (PASP:San Francisco), in press
Shields GA. 1976, {\it Ap. J.} 204, 330
Shields GA. 1996, {\it Ap. J.} 461, L9
Shields JC, Ferland GJ, Peterson, BM. 1995, {\it Ap. J.} 441, 507
Shields JC, Hamann F, Foltz CB, Chaffee FH. 1997, in Emission
Lines in Active Galaxies, eds. BM Peterson, F-Z Cheng,
ASP Conf. Ser., 113, 118
Shklovskii IS. 1965, Sov. Astr., 8, 635
Sigut TAA, Pradhan AK. 1998, {\it Ap. J.} 499, L139
Sil'chenko OK, Burenkov AN, Vlasyuk VV. 1998, {\it Astron. Astrophys.} 337, 349
Silk J, Rees MJ. 1998, {\it Astron. Astrophys.} 331, L1
Songalia A, Cowie LL. 1996, {\it Astron. J.} 112, 335
Spaans M, Corollo CM. 1997, {\it Ap. J.} 482, L93
Spinrad H, Dey A, Stern D, Dunlop J, Peacock J,
{\it et~al.} 1997, {\it Ap. J.} 484, 581
Srianand R, Shankaranarayanan S. 1999, {\it Ap. J.} in press
Stanford SA, Eisenhardt PR, Dickenson M. 1998,
{\it Ap. J.} 492, 461
Steidel CC, Adelberger KL, Dickenson M, Giavalisco M,
Pettini M, {\it et~al.} 1998, {\it Ap. J.} 492, 428
Steidel CC, Adelberger KL, Giavalisco M, Dickenson M,
Pettini M. 1999, {\it Ap. J.} in press
Taniguchi Y, Arimoto N, Murayama T, Evans AS, Sanders DB,
{\it et~al.} 1997, in Quasar Hosts, eds. DL Clements, I Perez-Fouron,
ESO-IAC Conf. Pro., p. 127
Taniguchi Y, Ikeuchi S, Shioya Y. 1999, {\it Ap. J.} 514, L12
Tantalo R, Chiosi C, Bressan A. 1998, {\it Astron. Astrophys.} 333, 419
Telfer RC, Kriss GA, Zheng W, Davidsen AF, Green RF. 1999,
{\it Ap. J.} in press
Terndrup DM, Sadler EM, Rich RR. 1995, {\it Astron. J.} 110, 1774
Terlevich RJ, Boyle BJ. 1993, {\it MNRAS} 262, 491
Thomas D, Greggio L, \& Bender R. 1998, {\it MNRAS} in press
Thompson KL, Hill GJ, Elston R. 1999, {\it Ap. J.} in press
Thuan TX, Izotov YI, Lipovetsky VA. 1995, {\it Ap. J.} 445, 108
Thurston TR, Edmunds MG, Henry RB. 1996, {\it MNRAS} 283, 990
Tiede GP, Frogel JA, Terndrup DM. 1995, {\it Astron. J.} 110, 2788
Tinsley B. 1979, {\it Ap. J.} 229, 1046
Tinsley B. 1980, Fund. of Cosmic Phys., 5, 287
Trager SC, Faber SM, Dressler A, Oemler A. 1997, {\it Ap. J.} 485, 92
Tripp TM, Lu L, Savage BD. 1996, {\it Ap. J. Suppl.} 102, 239
Tripp TM, Lu L, Savage BD. 1997, {\it Ap. J. Suppl.} 112, 1
T\"urler M, Courvoisier TJ-L. 1997, {\it Astron. Astrophys.} 329, 863
Turner EL. 1991, {\it Astron. J.} 101, 5
Turnshek DA. 1981, Ph.D. Dissertation, University of Arizona
Turnshek DA. 1984, 280, 51
Turnshek DA. 1986, in Quasars, eds. G. Swarup, VK Kapahi, IAU Symp.
No. 119, (Dordrecht: Reidel), p. 317
Turnshek DA, Briggs FH, Foltz CB, Grillmair CJ, Weymann RJ.
1987, in QSO Absorption Lines: Probing the
Universe, Poster Papers, eds. JC Blades, C Norman, DA Turnshek,
p. 1
Turnshek DA. 1988, in QSO Absorption Lines: Probing the
Universe, eds. JC Blades, C Norman, DA Turnshek, (Cambridge:
Cambridge Univ Press), p. 17
Turnshek DA. 1994, in QSO Asborption Lines, ed. G Meylan,
(Springer-Verlag), p. 223
Turnshek DA. 1997, in Mass Ejection From AGN, eds. R Weymann,
I Shlosman, N Arav, ASP Conf. Series, 128, 193
Turnshek DA, Kopko M, Monier E, Noll D, Espey B, {\it et~al.}
1996, {\it Ap. J.} 463, 110
Turnshek DA, Monier EM, Christopher JS, Espey BR. 1997,
{\it Ap. J.} 476, 40
Tytler D, Fan XM, Burles S, Cottrell L, Davis C, {\it et~al.} 1995,
in QSO Absorption Lines, ed. G Meylan, (Garching:ESO), 289
Uomoto A. 1984, ApJ 284, 497
Van Dokkum PG, Franx M, Kelson DD, Illingworth GD.
1998, {\it Ap. J.} 504, L17
Van Zee L, Skillman ED, Salzer JJ. 1998, {\it Ap. J.} 497, L1
Vazdekis A, Peletier RF, Beckman JE, Casuso
E. 1997, {\it Ap. J. Suppl.} 111, 203
Verner EM, Verner DA, Korista KT, Ferguson JW, Hamann F, {\it et~al.}
1999, {\it Ap. J.} in press
Vila-Costas MB, Edmunds MG. 1993, {\it MNRAS} 265, 199
Voit GM, 1997, in Mass Ejection From AGN, eds. R Weymann,
I Shlosman, N Arav, ASP Conf. Series, 128, 200
Voit GM, Weymann RJ, Korista KT. 1993, ApJ 413, 95
Wampler EJ, Bergeron J, Petitjean P. 1993, {\it Astron. Astrophys.} 273, 15
Wampler EJ, Chugai NN, Petitjean P. 1995, {\it Ap. J.} 443, 586
Warren SJ, Hewett PC, Osmer PS. 1994, {\it Ap. J.} 421, 412
Weymann RJ. 1994, in QSO Asborption Lines, ed. G Meylan,
(Springer-Verlag), p. 213
Weymann RJ. 1997, in Mass Ejection From AGN, eds. R Weymann,
I Shlosman, N Arav, ASP Conf. Series, 128, 3
Weymann RJ, Foltz C. 1983, in Quasars and Gravitational Lenses,
24th Liege Int. Astrophy. Coll., (Univ. de Liege:Belgium), p. 538
Weymann RJ, Turnshek DA, Christiansen WA. 1985,
in Astrophysics of Active Galaxies and Quasi-Stellar Objects,
ed. J Miller, (Mill Valley, CA: University Science Books), 185
Weymann RJ, Morris SL, Foltz CB, Hewett PC. 1991,
{\it Ap. J.} 373, 23
Weymann RJ, Carswell RF, Smith MGA. 1981, {\it Annu. Rev. Astron. Astrophys.} 19, 41
Weymann RJ, Stern D, Bunker A, Spinrad H, Chaffee FH, {\it et~al.}
1998, {\it Ap. J.} 505, L95
Weymann RJ, Williams RE, Peterson BM, Turnshek
DA. 1979, {\it Ap. J.} 218, 6 19
Wheeler JC, Sneden C, Truran JW. 1989, {\it Annu. Rev. Astron. Astrophys.}
27, 279
Williams RE. 1971, {\it Ap. J.} 167, L27
Williams RE, Strittmatter PA,
Carswell RF, Craine ER. 1975, {\it Ap. J.} 202, 296
Williams RE, Weymann RJ. 1976, {\it Ap. J.} 207, L143
Will BJ, Netzer H. 1983, {\it Ap. J.} 275, 445
Wills BJ, Netzer H, and Wills D. 1985, {\it Ap. J.} 288, 94
Wills BJ, Thompson KL, Han M, Netzer H, Wills D, {\it et~al.} 1995, {\it Ap. J.} 447, 139
Worthey G, Faber SM, Gonzalez J. 1992, {\it Ap. J.} 398, 69
Yoshii Y, Tsujimoto T, Kawara K. 1998, {\it Ap. J.} 507, L113
Yoshii Y, Tsujimoto T, Nomoto K. 1996, {\it Ap. J.} 462, 266
Young P, Sargent WLW, and Boksenberg A. 1982, {\it Ap. J. Suppl.} 48, 455
Zaritsky D, Kennicutt RC, Huchra JP. 1994, {\it Ap. J.} 420, 87
Zheng W, Kriss GA, Davidsen AF. 1995, {\it Ap. J.} 440, 606
Zheng W, Kriss GA, Telfer RC, Grimes JP, Davidsen AF. 1997, {\it Ap. J.} 475, 469
Ziegler BL, Bender R. 1997, {\it MNRAS} 291, 527
\end{document}
|
\section{Introduction}
In their search for an optical counterpart for the gamma ray burst
GRB980425, Galama et al. (1998a) detected SN 1998bw in the galactic
arm of ESO 184-G82(EOP 184-82), which Tinney et al. (1998) determined
to have a red shift of $0.0085 \pm 0.0002$. The supernova's light
curves rose sharply after the burst, and its spatial coordinates were
well within the burst's error box, strongly suggesting a connection
between the two events. The probability of their independence was
estimated by Galama et al. (1998a) at $1.1 \times 10^{-4}$. However,
BeppoSAX also detected a fading x-ray
source (generally thought to be the hallmark of the burst counterpart)
at a position inconsistent with SN 1998bw (Pian et al. 1998, Piro et
al. 1998, Pian et al. 1999), so the relationship between SN 1998bw and
GRB980425 is unclear.
Further observations showed that SN 1998bw is
positionally coincident with a second BeppoSAX x-ray source which has
faded by a factor of two in brightness from 26 April to 10 November 1999,
which is consistent with x-ray emission from a supernova plus the galaxy
(Pian et al. 1999).
SN 1998bw has peculiar and unique properties other than a possible
association with a Gamma-Ray Burst. Its spectrum is unique (although
two Type Ic supernovae have somewhat similar spectra; see Iwamoto
et al. 1998) and displays ejection velocities measured from the blue wings
of the Ca II line as high as $60, 000 km \cdot s^{-1}$ (Kulkarni et al.
1998). Its emissions at radio wavelengths increased much more quickly
than other supernovae, and it is also the most luminous supernova to date at
radio wavelengths (Kulkarni et al. 1998). These coincidences of unusual
properties greatly strengthen the connections between GRB 980425 and SN
1998bw. In general, a concensus has emerged that the burst is related to
the supernova, and this has inspired much research detailing connections
between the two phenomena.
Due to the unique and pivotal nature of SN 1998bw, it is imperative that
the light curve be tracked in a wide range of optical bands as long as
possible. Galama et al. (1998a) tracked the U, B, V, R, and I light curves
for 58 days after the burst, and these showed a typical peak as generally
seen for supernovae of many types. In the interests of recording
as much data as possible for such an unprecedented event, we followed up
on their results with further observations in the B, V, and I filters.
\section{Observations}
The data were obtained using the Yale 1-m telescope, at the Cerro Tololo
Inter-American Observatory in Chile, from 27 June through 28 October 1998.
Our series of observations commenced as soon as the refurbishing of the
Yale 1-m telescope had been compeleted and the CCD camera installed. The
images' pixel size was 0.30'', with a field of view of $10.2' \times
10.2'$. Our exposure times were always 300 seconds per image, using B,
V, and I filters. Our typical seeing had a FWHM of $1.2''$. We obtained
139 measures of the brightness of SN 1998bw.
The images were first processed with the normal procedure for overscan
correction, bias subtraction and flat fielding. Our photometric analyses
were made with IRAF's program ``APPHOT''. We used apertures of only three
pixel radius to minimize interference from the parent galaxy, which was
a sufficient size because of the high signal-to-noise ratio. Our
background annuli were constructed with inner and outer radii of thirty
and forty pixels centered on the star, with the sky background taken as
the mode within this annulus. The deduced sky background for the supernova
is close to that deduced for isolated stars.
For each image, observed magnitudes were recorded for SN 1998bw and for
five comparison stars. We used the updated magnitudes provided
by Galama et al. (1998b), choosing numbers 4, 6, 7, 9, and 10 because they
were relatively bright. Comparison star \#2 was excluded because its data
was just eratic enough to arouse suspicion that it might be a small
amplitude variable star, although is was not definitively identified
as such.
We were able to estimate our magnitude uncertainties by comparing the
standard stars with each other in a variety of images. These errors
were determined to be substantially dependent on the apparent magnitudes
of the stars due to the normal Poisson variations, which is why we used
the brightest ones for our data analysis. Monitoring the differences
between the standard stars in each image also enables us to catch
photometric problems with the standard stars (due to cosmic rays, bad
columns, etc.). A few nights had large uncertainty due to clouds or bad
seeing. Our results are that in general we have systematic uncertainty of
0.02 mag added in quadrature with the statistical errors reported by
IRAF. For the supernova, the statistical errors are generally
substantially smaller than our systematic errors in the early portions of
our light curve. The comparison stars are fainter than the supernova, yet
our use of the average of five stars as our `standard' improves the
accuracy of this `standard' to $\sim 0.01$ mag. In all, the uncertainties
in our supernova magnitudes typically range from 0.02 to 0.04 mag.
SN 1998bw appears in the spiral arm of its host galaxy, so we must
consider the effects of the galaxy light in our photometry. Fortunately,
the supernova was quite bright during the entire duration of our study and
the contribution of light from the spiral arm is minimal. To be quantitative,
we have measured the surface brightness of the center of the spiral arm
on both sides of the supernova and compared this with the total brightness
within our photometric aperture centered on the supernova. Images of the
galaxy from before the supernova show the brightness along the spiral arm
to be uniform along the position of the supernova, so we know that there
are no significant knots or stars at the supernova position. At the
beginning of our light curve, the contamination from galaxy light in
our photometry aperture varied from $0.5 \%$ to $0.8 \%$ for the three
filters. So we have a systematic error which is smaller than our quoted
uncertainties that will make the supernova slightly fainter than
tabulated. Ideally, we should wait several years for the transient to
fade to invisibility, then get further images with our same equipment and
subtract off the galaxy light; but in the meantime the systematic error
is known to be small.
For each image, we compared the instrumental magnitude of the supernova to
the average instrumental magnitude of the five standard stars. This
difference was then applied to the average of the standard stars' actual
magnitudes taken from Galama et al. (1998b) to determine the actual magnitude
of SN 1998bw. Our results are plotted in Fig. 1 (along with Galama's
earlier results) and tabulated in Table 1.
We are impressed with the remarkable linearity of our portion of the light
curves. The best fit lines to our data are displayed in Figure 1, and we
see no significant systematic deviation from perfect lines at any time
or in any color. Our limits on systematic deviations are $<0.05$ for our entire
123 day observation time. In the B filter, the light is declining at
$0.0141 \pm 0.0002$ magnitudes per day, which corresponds to a radioactive
half-life
of $53.4 \pm 0.8$ days. For V, these figures are $0.0194 \pm 0.0003$
magnitudes per day, which a corresponding radioactive half-life of $40.9
\pm 0.7$ days. For I, the figures are $0.0181 \pm 0.0003$ magnitudes per
day and a half-life of $41.6 \pm 0.7$ days.
The B light curve has a somewhat slower decay than in the V and I bands.
For an extinction of $A_{v} = 0.2$ (Galama et al. 1998), the
supernova's B-V was 0.82 mag at the beginning of our observation
period and around 0.3 mag toward the end. The extinction corrected V-I rose
by a small amount, from 0.53 to 0.60 mag, during the same time period.
With our B, V, and I light curves, we can approximate the
bolometric light curve for radiation from the ultraviolet to the infrared.
We have done this by first correcting for galactic extinction ($A_{v} = 0.20$),
converting our magnitudes into $f_{\nu}$, adopting a power law spectrum
from B
to V and from V to I, adopting a Rayleigh-Jeans spectrum for lower
frequencies than I, adopting a Wien spectrum for higher frequencies than
B, and integrating the spectrum. For JD2450996 (68 days after the burst)
we get a bolometric flux of $1.1 \times 10^{-11} erg \cdot cm^{-2}
\cdot s^{-1}$ or a bolometric luminosity of $2.0 \times 10^{42} erg
\cdot s^{-1}$. For JD2451098 (170 days after the burst)
we get a bolometric flux of $2.1 \times 10^{-12} erg \cdot cm^{-2}
\cdot s^{-1}$ or a bolometric luminosity of $3.9 \times 10^{41} erg
\cdot s^{-1}$. For the conversion
to luminosity, we adopted a velocity of $2550 km \cdot s^{-1}$ and a
Hubble Constant of $65 km \cdot s^{-1} \cdot Mpc^{-1}$, for a distance of
39 Mpc. These calculated luminosities have
significant uncertainties arising from the extinction ($\sim 10\%$),
bolometric correction ($\sim 30\%$), and distance ($\sim 20\%$), so that
overall errors perhaps as large as $\sim 50\%$ might be present. The
effective half-life for this decline is 44 days.
\section{Comparison with Other Supernova}
SN1998bw has many unique and extreme properties, however at first
look, its light curve appears to be that of a normal supernova. Is the
light curve unique? We will compare our light curve with those of Type
Ia, Ib, Ic, and II supernovae in turn.
For the majority of Type Ia events, the B light curve fades by
1.1 magnitudes in the first 15 days afters peak (Hamuy 1996a), while
SN 1998bw has the same drop. The usual slope of the late time B light
curve is $0.01516 \pm 0.0024 mag/day$ for Type Ia events (Barbon et al.
1984) and is easily compatible with that of SN 1998bw. However, the
decline rates in V and I differ substantially between most of the Type Ia
events and SN 1998bw (0.0184 versus 0.024 and 0.0181 versus 0.041
magnitudes per day respectively) from 70-80 days after peak. A second
important difference is that almost all Type Ia events display a prominent
bump in the I band light curves from 20-50 days after peak (Hamuy et al.
1996a, Riess et al. 1999), while SN 1998bw does not show any sign of such
a bump. A third difference is that SN 1998bw has a peak absolute magnitude
of $-18.88 \pm 0.05$ (Galama et al. 1998a) whereas the majority of Type Ia
events have peak absolute magnitudes of $- 19.26$ (Hamuy et al. 1996b).
However, uncertainties in distance and extinction can perhaps be as large
as a third of a magnitude, so this third difference may not be
significant. A fourth difference is that the extinction corrected color
at peak of SN 1998bw is $B-V = 0.47 \pm 0.07$ mag (Galama et al. 1998a) while
the usual value for Type Ia events is $0.00 mag$ (Hamuy et al. 1996b). So
in all, the light curve of SN 1998bw looks similar to that of Type Ia
events, yet detailed parameters are quantitatively different.
Perhaps a closer match can be found with the anomalous Type Ia SN 1991bg
(Leibundgut et al. 1993; Fillipenko et al. 1992). This event had a
substantially redder peak color ($B-V \sim 0.8$), a much lower peak
absolute magnitude ($M_{B} = -16.62$), and no bump in the I band light
curves. While the detailed light curve (and spectrum) of SN 1991bg is
still different from that SN 1998bw, we note that many of the properties
are more like those of SN 1991bg than of normal Type Ia events.
Type Ib and Ic light curves have not been characterized as closely as
those of Type Ia supernovae. Nevertheless, enough is known (e.g., Uomoto
\& Kirshner, 1986; Ensman and Woosley 1988; Clocchiatti et al. 1997) to
find similarities and differences with SN 1998bw. The overall light
curve of Type Ib and Ic events is the same for SN 1998bw with similar
decline rates over the first 15 days. The late time decline rate of
Type Ic events vary, apparently with two classes, as slow and fast
decliners. The 60-180
day decline rates of roughly 0.016 magnitudes per day are seen for
the Type Ic events SN1983N and SN1983V (Clocchiatti et al. 1997) which is
comparable to that for SN 1998bw. The color evolution for Type Ic events
is similar to that of SN 1998bw, both at peak and at late times.
The peak absolute magnitude of Type Ic events vary about $M_{B} \sim
-17.5$, yet are all significantly fainter than SN 1998bw. However,
with the few well measured Type Ic events having a wide scatter, the
luminosity of SN 1998bw may not be unusual. In all, SN 1998bw appears
to have a light curve within the class of Type Ic events.
Type II supernovae vary greatly in their light curve shape and color
(Patat et al. 1993). The colors, peak absolute magnitudes, decline rate
of SN 1998bw from peak, and late time decline rate are all within the
normal range for the IIL subclass. Nevertheless, there are some subtle
distinctions; such as a total lack of any indication of a plateau in the I
band and the switch to the late time decline rate only $\sim 30$ days
after peak.
In all, the light curve of SN1998bw is fully consistent with those
of Type Ic supernovae, in keeping with the spectral classification and
physical models.
\section{Comparison with Models}
The decay rate of the tail is so close to an exponential that we suggest
that this is no coincidence. In addition, the measured decline rate
corresponds to that expected from the decay of radio active cobalt (with a
half-life of 78.5 days) as modified by the effects due to the expansion of
the shell (Colgate and McKee 1969). Hence, it is reasonable to take our
light curve as strong evidence that the late time light curve of SN 1998bw
is being powered by the decay of cobalt, with the difference in slope
caused by the leakage of gamma radiation from the shell.
Three detailed models have been presented seeking to explain the
light curve of SN1998bw. Iwamoto et al. (1998) and Iwamoto (1999) model
the event as an extremely energetic explosion of a massive star stripped
down to its carbon/oxygen core. Woosley, Eastman, and Schmidt (1998)
independently present a similar model with similar results.
H\"{o}flich, Wheeler, and Wang (1998) and Wheeler, H\"{o}flich, and Wang
(1999) present a model with an aspherical explosion in the nondegenerate
C/O core of a massive star. An asymmetric event can account for the
observed polarization. All three models account for the early light
curve, the early colors, and the early spectrum with generally
acceptable accuracy.
Iwamoto et al. (1998) and Iwamoto (1999) present
predictions for the late-time V light curve of SN1998bw (see Figure 1 of
Iwamoto 1999) as a perfectly straight line in a log-log plot of flux
versus time since the burst, for a predicted power law with slope -2.75.
This power law prediction does not agree with the observed exponential
decline. However, K. Nomoto (1999, private communication) has presented a
more detailed light curve prediction which shows a more complicated shape
(neither a power law nor an exponential) than presented in Iwamoto (1999).
The model V magnitude declines by 2.80 magnitudes from 27 June to 28
October with deviations from a simple exponential curve defined by the end
points of up to 0.22 mag. For comparison, our data shows a decline of
2.26 mag over this same time and maximum departures from a simple
exponential decline of $<0.05 mag$.
One way to distinguish the three models is by the explosion energy
and the ejected $^{56}Ni$ mass. The spherically symmetric models have
energies and nickel masses of around $3 \times 10^{52} erg$ and $0.7
M_{\odot}$, while the aspherical models have $2 \times 10^{51} erg$ and
$0.2 M_{\odot}$. Wheeler, H\"{o}flich, and Wang point out
that if the late time light curve tracks the radioactive decay line then
the ejected nickel mass can be determined, and that this might prove the
simplest discriminant between models.
One possible method to measure the nickel mass from our light
curve is to scale the luminosity in the tail from another supernova of
known late-time luminosity and nickel mass. The best case for comparison
might be SN1987A which has a well measured light curve (Hamuy et al.
1988), a well known distance (50 kpc; McCray 1993), and a well known
nickel mass ($0.069 \pm 0.003 M_{\odot}$; McCray 1993). On day 170 after
the core
collapse, SN1998bw had an extinction corrected V magnitude of 17.60, while
SN1987A had an extinction corrected magnitude of 4.39. If SN1987A were
placed at 39 Mpc, then its V magnitude should appear 1.25 mag fainter than
we observed for SN1998bw. This implies a nickel mass 3.2 times larger, or
that SN1998bw has $0.22 M_{\odot}$. The dominant uncertainty arises from
the distance to SN1998bw, for which peculiar velocities of up to $400 km
\cdot s^{-1}$ and uncertainties in the Hubble Constant of up to $10 km
\cdot s^{-1} \cdot Mpc^{-1}$ yield
nickel mass uncertainties of $0.09 M_{\odot}$.
The procedure in the previous paragraph can only be approximate,
in particular since there might be substantial leakage of the gamma rays
from the expanding nebula. Such leakage could explain why our observed
decline is steeper than that associated with $^{56}Co$ decay (Clocchiatti
\& Wheeler 1997). Such leakage would lower the late-time luminosity
and lower the deduced nickel mass. The V light curve of SN1987A declined
with the $^{56}Co$ rate whereas
the V light curve of SN1998bw declined at a roughly twice the rate. The
effect of leakage on our previous derived nickel mass can only be
estimated within specific models, yet it is likely that our $0.22 \pm 0.09
M_{\odot}$ value must be regarded as a lower limit.
We expect that our late-time light curve will provide a set of
observations useful for refining and constraining individual models of the
unique SN1998bw. In particular, our data might constrain the quantity of
ejected $^{56}Ni$ so as to decide between symmetric and asymmetric models.
Another challenge to models is to explain the near perfect exponential
shape of the light curve even though the slope is not that of the
$^{56}Co$ decay.
We will continue to monitor the brightness of SN1998bw in 1999.
However, the background light from its host galaxy is increasingly a
problem for exact photometry.
|
\section{figure captions}
Fig.1 The spin wave dispersions obtained by (a) LMTO calculations and
(b) the tight binding approach (model A)
along the symmetry directions $\Gamma$X,
XM and MR shown as a function of doping.
Fig. 2 The doping dependence of the exchange couplings $J_1$,
$J_2$,
$J_4$ and $J_8$ between atoms at ($a$ 0 0), ($a$ $a$ 0), (2$a$ 0 0) and
(3$a$ 0 0),
where $a$ is the lattice parameter.
Fig. 3 The (a) minority-spin and (c) majority-spin $d$ partial density of
states
within models A, B and C. The spin wave dispersions along
$\Gamma$X as a function of doping within models (b) B
and (d) C are shown along with (e) the combined contributions
of models B and C. $y$ refers to the hole concentration in the majority-spin
$e_g$ band with reference to its half-filled case. $z$ is the electron
concentration in the minority-spin $t_{2g}$ band. $x$ is the net
concentration
of the doped holes and is given by $x=y-z$.
Fig. 4 The dependence of the
spin wave energies on the $e_g$ hole doping $y$ along $\Gamma$X
within model D. The hopping between oxygen atoms and the
$t_{2g}$ orbitals on the Mn atom have
been left out of the model.
(a) $pd\sigma$=-2.02~eV
and (b) $pd\sigma$=-2.25~eV.
Fig. 5 The variation of the exchange couplings $J_1$, $J_2$,
$J_4$ and $J_8$ with the $e_g$ hole doping $y$.
Open circles are for the case (model B) including the
hopping between oxygen atoms and $pd\sigma$=-2.02~eV.
Open and filled squares are for the cases (model D) without the hopping
between oxygen atoms: $pd\sigma$=-2.02~eV (open squares)
and $pd\sigma$=-2.25~eV (filled squares). The $t_{2g}$ orbitals on the
Mn atom have been left out of the basis set.
\end{document}
|
\section{The quon algebra}
We assume that $b_m$,$b^{\dagger}_m$, $m=-j,...+j$ are (2j+1) operators which
satisfy the $q$--mutation relations (or the quon algebra) defined by
\cite{green}, i.e.,
\begin{equation}
[b_m,b^{\dagger}_{m^{\prime}}]_{q}=
b_m ~{b^{\dagger}}_{m^{\prime}}-q ~{b^{\dagger}}_{m^{\prime}} ~b_m =
~{\delta}_{m,m^{\prime}}.
\label{comu}
\end{equation}
This relation is a deformation of the Bose and Fermi algebras and
interpolates between those algebras when $q$ goes from +1 to -1.
The Fock space is constructed as usual from the application of the
creation operators, ($b_\alpha^\dagger$), on the vacuum state,
defined as $b_\alpha|0>=0$ for all $\alpha$. It has been
shown\cite{fivel,zagier} that the squared norm of any polynomials of the
creation operators, $b_\alpha^\dagger$,
acting on the vacuum state is positive definite.
In what follows we derive some important results in order to construct
the $su(2)$ generators. We have at our disposal
the transition number operator \cite{green} $N_{\alpha\beta}$,
which has the usual commutation relations:
\begin{equation}
[N_{\alpha\beta},b^{\dagger}_{\mu}]={\delta}_{\beta\mu}~b^{\dagger}_{\alpha}~,~
[N_{\alpha\beta},b_\mu]=-{\delta}_{\alpha\mu}~b_{\beta}~~. \label{tran}
\end{equation}
The transition number operator is an infinite series in the $b^{\dagger}_{\mu}$'s
and $b_\mu$'s and it has been obtained in closed form in
\cite{zagier,mel}. The general structure of this operator is \cite{mel} :
\begin{equation}
N_{\alpha\beta}~=~b^{\dagger}_{\alpha} b_{\beta}~+~\sum_{n=1}^{\infty}~\sum_{(i_1 ...i_n)}~
\sum_{\pi}~c_{\pi(i_1),...,\pi(i_n), i_1 ...i_n} ~
(Y_{\alpha\pi(i_1), ...\pi(i_n)})^{\dagger} ~ Y_{\beta i_1 ...i_n}
\label{nger}
\end{equation}
where the summation over $\pi$ means that we consider all different
permutations of the indices $i_1,i_2, ...,i_n$, including their repetitions.
$\pi(i_1),\pi(i_2), ...\pi(i_n)$ corresponds to each of these permutations
and $Y^\dagger$ is the adjoint of $Y$.
The operators $Y_{\beta i_1 ...i_n}$ are obtained through the recursion
relations \cite{mel,dori}:
\begin{equation}
Y_{k i_1 ...i_{n+1}}~=~Y_{k i_1 ...i_n}~b_{i_{n+1}}~-~
q^{n+1}~b_{i_{n+1}}~Y_{k i_1 ...i_n}
\label{recur}
\end{equation}
with
\[
Y_{ki}~=~b_k b_{i}~-~q~b_i b_k
\]
The coefficients $c_{\pi(i_1),...,\pi(i_n), i_1 ...i_n}$
are only functions of the deformation parameter $q$ \cite{mel,dori}.
The first terms in the series are:
\begin{equation}
N_{\alpha \beta} = b^{\dagger}_{\alpha} b_{\beta} + (1-{q}^2)^{-1}
\sum_m(b^{\dagger}_mb^{\dagger}_{\alpha} - q b^{\dagger}_{\alpha}b^{\dagger}_m)(b_{\beta}b_m-q b_m b_{\beta})
+... ~~.
\end{equation}
From the recursion relation, eq.(\ref{recur}), and eq.(\ref{tran}),
one can show that
\[
[N_{\alpha \beta}~,(Y_{{\alpha}^{\prime} i_1 ...i_n})^{\dagger} ]=
\delta_{\beta i_1} (Y_{{\alpha}^{\prime} \alpha i_2 ...i_n })^{\dagger}+
\delta_{\beta i_2} (Y_{{\alpha}^{\prime} i_1 \alpha i_3 ...i_n})^{\dagger}
+...+
\delta_{\beta i_{n-1}} (Y_{{\alpha}^{\prime} i_1 i_2 ...\alpha i_n})^{\dagger}
\]
\begin{equation}
+~ \delta_{\beta i_n}
(Y_{{\alpha}^{\prime} i_1 i_2 ...i_{n-1}\alpha})^{\dagger}~+~
\delta_{\beta \alpha^\prime } (Y_{ \alpha i_1 i_2 ...i_n} )^{\dagger}.
\label{ncom}
\end{equation}
As a subsidiary result, it follows from eq.(\ref{ncom}) that;
\[
[N_{\alpha \beta}~,(Y_{\alpha^\prime \pi(i_1) ...\pi(i_n) })^{\dagger}
Y_{\beta^\prime i_1 ... i_n } ] =
\delta_{\beta \alpha^\prime }~
(Y_{\alpha \pi(i_1) ...\pi(i_n) })^{\dagger}
~ Y_{ \beta^\prime i_1 ... i_n }
\]
\begin{equation} ~-~
\delta_{\alpha \beta^\prime }~
(Y_{ \alpha^\prime \pi{i_1} ...\pi{i_{n+1}}})^{\dagger}
~ Y_{\beta i_1 ... i_{n+1} } \label{NYY}.
\end{equation}
In what follows we calculate the commutator
$[N_{\alpha \beta},N_{\alpha^{\prime} \beta^{\prime}}]$. For this
purpose, we substitute
$N_{\alpha^{\prime} \beta^{\prime} }$ given in eq.(\ref{nger})
into the commutator above, which yields
\[
[N_{\alpha \beta}, N_{\alpha^{\prime} \beta^{\prime}}] =
[N_{\alpha \beta},
~{b^{\dagger}}_{\alpha^{\prime}} ~b_{\beta^{\prime}}]~
\]
\[
+~\sum_{n=1}^{\infty}~\sum_{(i_1 ...i_n)}~
\sum_{\pi}~c_{{\alpha^{\prime}}\pi(i_1),...,\pi(i_n),
{\beta^{\prime}} i_1 ...i_n} ~
[N_{\alpha ,\beta},(Y_{{\alpha^{\prime}} \pi(i_1), ...\pi(i_n)})^{\dagger} ~
Y_{{\beta^{\prime}} i_1 ...i_n}].
\]
Using eqs.(\ref{tran}) and (\ref{NYY}), we obtain an important
relation obeyed by the transition number operator:
\begin{equation}
[N_{\alpha \beta},N_{\alpha^{\prime} \beta^{\prime}}] =
{\delta}_{\beta {\alpha}^{\prime}}~N_{\alpha\beta^{\prime}}
-{\delta}_{\alpha {\beta}^{\prime}} N_{{\alpha}^{\prime}\beta } ~~.
\label{rip}
\end{equation}
The above commutation relation shows that the
transition number operators are the generators of an $su(2j+1)$ algebra.
\section{ $su(2)$ tensor operators}
In order to see that the $b^{\dagger}_{\mu}$ is an irreducible tensor
operator, we have to show that
\[
[J_{0} ,b^{\dagger}_{\mu} ]= \mu~b^{\dagger}_{\mu} ~~,
\]
\begin{equation}
[J_{\pm},b^{\dagger}_{\mu} ]= \sqrt{(j\mp \mu)(j\pm \mu +1) } b^{\dagger}_{\mu \pm 1}~~,\label{tens}
\end{equation}
where $J_{0} ,J_{\pm}$ are usual $su(2)$ generators,
\[
[J_{0},J_{\pm}]= {\pm} J_{\pm} ~~,
\]
\begin{equation}
[J_{+} , J_{-} ]= 2 J_{0} ~~,
\end{equation}
which have to be built from the same $b^{\dagger}_{\mu}$ operators.
Notice that the label $m$ corresponds to the projection of
$j$ and we are restricted to a single $j$--level. Operators
belonging to different levels commute with each other.
Thus we may define the $su(2)$ generators as:
\[
J_0 = \sum_{\nu=-j}^{+j} \nu~N_{\nu \nu}
\]
\begin{equation}
J_{\pm} = \sum_{\nu=-j}^{+j} \sqrt{(j\mp\nu)(j\pm\nu+1)} N_{\nu\pm 1~\nu} \label{ma}
\end{equation}
where the index $\nu$ takes the values $-j, -j+1,...j-1,j$ .
It is then simple to verify that the $b^{\dagger}_{\mu}$ operators
satisfy eqs.(\ref{tens}) and then form a genuine irreducible
tensor operator of the $su(2)$ algebra:
\[
[J_0,b^{\dagger}_{\mu}]= \sum_{\nu} \nu [N_{\nu \nu},b^{\dagger}_{\mu}]= \sum_{\nu}{\delta}_{\nu\mu}
b^{\dagger}_{\mu} = \mu b^{\dagger}_{\mu}
\]
and
\[
[J_{\pm},b^{\dagger}_{\mu}]= \sum_{\nu} \sqrt{(j\mp\nu)(j\pm\nu+1)} [N_{\nu\pm 1~\nu},b^{\dagger}_{\mu}] =
\sum_{\nu} \sqrt{(j\mp\nu)(j\pm\nu+1)} {\delta}_{\nu\mu}{b^{\dagger}}_{\nu\pm 1} =
\]
\[
\sqrt{(j\mp\mu)(j\pm\mu+1)} {b^{\dagger}}_{\mu\pm 1} ~~.
\]
To finish our demonstration we still have to show that the
operators defined in eqs.(\ref{ma}), are indeed
generators of the usual $su(2)$ algebra. This proves the
consistency of our approach. From eq.(\ref{ma}), we can write
\begin{equation}
[J_0,J_{\pm}]= \sum_{\nu,{\nu}^{\prime}}{\nu}
\sqrt{(j\mp{\nu}^{\prime})(j\pm{\nu}^{\prime}+1)}
[N_{\nu \nu} ,N_{{\nu}^{\prime}\pm 1 ~{\nu}^{\prime} } ]
\end{equation}
\[ = \sum_{\nu~{\nu}^{\prime}}
\nu~ \sqrt{(j\mp {\nu}^{\prime})(j\pm {\nu}^{\prime}+1)}
({\delta}_{\nu~ {\nu}^{\prime}\pm 1} N_{\nu{\nu}^{\prime}}
-{\delta}_{\nu~{\nu}^{\prime}} N_{{\nu}^{\prime}\pm 1~\nu } ) = \pm J_\pm ~~.
\]
One can show in an analogous way that $[J_+ , J_- ]=2~J_0 $
and therefore the usual $su(2)$ generators are consistent
with the quon algebra, eq.(\ref{comu}).
\section{ Concluding Remarks}
Thus, the tensor coupling
has to be done with the usual Clebsch-Gordan coefficients.
In the Introduction we have mentioned, as an application of the
{\it quon} statistics, the description of many body states by means of
bosons, as it is done, for example in the IBM \cite {ibm}. In that model
the so called $s$ and $d$ bosons are introduced
and in more sophisticated versions higher angular momentum bosons
are also needed. It follows from the above demonstration that we
can now define deformed $s$ and $d$ bosons, which behave like {\it quons}
but keep the same angular momentum coupling rules and coefficients as
in the non-deformed case. Of course, many other applications are
possible.
\vskip 0.35in
This work has been partially supported by CNPq.
\newpage
|
\section{Introduction}
\label{intro}
When studying observables in the process
$\mbox{$\mathrm{e}^+\mathrm{e}^-$} \rightarrow \mathrm{Hadrons}$, it is
found that the perturbative QCD predictions
have to be complemented by non-perturbative
corrections of the form $1/Q^p$, where
$Q$ is the centre-of-mass energy $E_{CM}$,
and the power $p$ depends on the particular
observable. For fully inclusive observables
such as the total cross section
$p=4$, however, for less inclusive ones
such as event shape variables the power is much smaller,
typically $p=1$. So for those observables non-perturbative
effects can be sizeable, e.g., at LEP1
energies corrections of 5-10\% are found. Since
these variables are extensively used for \mbox{$\alpha_s$}\
determinations, non-perturbative effects have to
be well understood. Until recently they have been
determined from QCD-inspired Monte Carlo (MC) models of
hadronization, however, this method leads to
model dependence and thus limitations
in the precision of the \mbox{$\alpha_s$}\ measurements. The
new approach of power law corrections to event
shapes, pioneered by Dokshitzer and Webber \cite{Webber1},
could lead to improvements in this respect.
The event shape variables studied are Thrust,
C-parameter, Heavy Jet Mass and Total and Wide Jet
Broadening.
These are infrared
and collinear safe variables, and perturbative
predictions are known up to second order in
\mbox{$\alpha_s$}, as well as the resummations of leading and
next-to-leading logarithms to all orders.
The results presented in the following are mostly
based on the study of Ref.~\cite{Hasko}, where
data from \mbox{$\mathrm{e}^+\mathrm{e}^-$}\ annihilations at $E_{CM}=14$ GeV
up to 161 GeV have been analyzed. In addition,
some preliminary results from the LEP2 runs up to
189 GeV have been employed.
\section{Power Corrections}
\label{powers}
Power corrections are supposed to have their origin in
infrared divergences (renormalons) in the perturbative
expansions when the
overall energy scale $Q$ approaches the Landau pole
$\Lambda$. The first approaches were based on the
assumption of the existence of a universal
non-singular behaviour of an effective strong
coupling at small scales, parametrized by a non-perturbative
parameter
\begin{equation}
\alpha_0(\mu_I) = \frac{1}{\mu_I} \int_{0}^{\mu_I} dk\; \mbox{$\alpha_s$}(k) \; ,
\end{equation}
with $\Lambda \ll \mu_I \ll Q$, which separates
the perturbative from the non-perturbative region, $\mu_I =2$ GeV,
typically. $\alpha_0(\mu_I)$ is assumed to be universal.
\subsection{Mean Values}
\label{mean}
Using the Ansatz described above, the following prediction is
obtained for the mean value of an event shape variable $f$~:
\begin{equation}
\langle\, f\, \rangle = \langle\, f\, \rangle^{pert} \;+\; \langle\, f\, \rangle^{pow} \; ,
\end{equation}
where $\langle\, f\, \rangle^{pert}$ is the full second order prediction of the form
\begin{equation}
\mbox{$\alpha_s$}(\mu^2) A_f +
\mbox{$\alpha_s$}^2(\mu^2) \left[ B_f + A_f b_0 \ln\frac{\mu^2}{s} \right] \; .
\end{equation}
Here $\mu^2$ is the renormalization scale, $s=E^2_{CM}$, and $b_0 = (33-2n_f)/(12\pi)$,
$n_f$ being the number of active flavours.
The power correction term is given by $\langle\, f\, \rangle^{pow} = a_f {\cal P}$,
where $a_f$ is 2 for Thrust, 1 for Heavy Jet Mass and $3\pi$ in case of the
C-parameter. $\cal P$ is a universal function of the form
$
{\cal P} \approx \;{\cal M}\; \mu_I\; \alpha_0(\mu_I) / Q \;
$
(up to a constant and corrections of order \mbox{$\alpha_s$}\ and $\mbox{$\alpha_s$}^2$).
The {\it Milan factor} ${\cal M} \approx 1.8$ \cite{milanfactor}
takes into account two-loop effects. Recently it has been found
\cite{newbroad} that
in the case of the Jet Broadening variable the power correction is of a more
complicated type compared to above, namely of the form $1/(Q \sqrt{\mbox{$\alpha_s$}(Q)})$.
\begin{figure}[htb]
\begin{center}
\vspace{-1cm}
\includegraphics[width=7cm]{run_mean.eps}
\vspace{-1cm}
\caption{Mean values of event shape variables as a function of the
centre-of-mass energy.
\label{fig:delphimeans}}
\end{center}
\vspace{-1cm}
\end{figure}
DELPHI \cite{delphi} have measured mean values for Thrust,
Wide Jet Broadening and
Heavy Jet Mass from the LEP1 and LEP2 data and combined
their results with measurements from
low energy \mbox{$\mathrm{e}^+\mathrm{e}^-$}\ experiments in order to extract $\mbox{$\alpha_s$}(M_Z)$ and $\alpha_0$ from
a fit of the power law Ansatz to these data. The fits are displayed in
Fig.\ \ref{fig:delphimeans}. Very good fits are obtained with $\mbox{$\alpha_s$}(M_Z)$
between 0.118 and 0.120, and $\alpha_0(2 \mathrm{GeV})$ between 0.40 and 0.55. Similar results have been found in the analysis of Ref.~\cite{newbroad}.
\subsection{Distributions}
\label{distributions}
\begin{figure}[htb]
\begin{center}
\vspace{-1cm}
\includegraphics[width=7cm]{bw_dist.eps}
\vspace{-1cm}
\caption{Fits to distributions of the Wide Jet Broadening
at several centre-of-mass energies. \label{fig:bw_dist}}
\end{center}
\vspace{-1cm}
\end{figure}
For distributions of event shape observables it has been shown
\cite{dokdis} that the non-perturbative corrections lead to a shift
in the distribution, i.e,
\begin{equation}
\frac{1}{\sigma_{tot}} \frac{d \sigma(f)^{corr}}{d f} =
\frac{1}{\sigma_{tot}} \frac{d \sigma(f - \Delta f)^{pert}}{d f}
\end{equation}
where in the cases of Thrust, Heavy Jet Mass and C-parameter
the shift $\Delta f$
is given by exactly the same terms as the correction for the mean values, i.e.,
$\Delta f = a_f {\cal P}$. An improved calculation \cite{newbroad} for the Jet
Broadening variable has shown that in this case the distribution is not only
shifted, but also squeezed, since the shift is of the form
$\Delta B \propto {\cal P} \ln(1/B)$.
The perturbative distribution is obtained from a matching of the
full next-to-leading
order prediction to the resummation of all leading and
next-to-leading logarithms $\ln f$.
Theoretical uncertainties on the \mbox{$\alpha_s$}\ and $\alpha_0$ determinations
are estimated
from variations of the renormalization scale and the scheme applied for
matching
the fixed order and resummed calculations. Central values are given for
$\mu^2=s$.
In Fig.\ \ref{fig:bw_dist} the fits to the Wide Jet Broadening are displayed,
for various centre-of-mass energies. Good fits are obtained for the power law
Ansatz as well as for the more traditional approach of hadronization
corrections from
MC models. Similar fits to Thrust and Heavy Jet Mass work well at
high energies,
however, some deviations are found at very small energies.
There probably the limit of
applicability of the power law approach is reached. Furthermore,
mass effects could play a role there.
In Fig.\ \ref{fig:distsum} a summary of the results can be found.
A combination of the results gives
$\mbox{$\alpha_s$}(M_Z) = 0.1082 \pm 0.0021$,
$\alpha_0(2 \mathrm{GeV}) = 0.504 \pm 0.042$.
For $\alpha_0$ universality is found at the level of 20\%.
The value of the
strong coupling results to be lower than the one obtained from
a similar fit when using MC models
for the hadronization corrections, namely $\mbox{$\alpha_s$}(M_Z) = 0.1232 \pm 0.0040$.
This difference has still to be understood.
\begin{figure}[htb]
\begin{center}
\vspace{-1cm}
\includegraphics[width=7cm]{distsum.eps}
\vspace{-1cm}
\caption{Summary of the power law fits to distributions of event shapes.
\label{fig:distsum}}
\end{center}
\vspace{-1cm}
\end{figure}
\section{Non-Perturbative Shape Functions}
\label{shapefunctions}
Recently it has been shown \cite{Sterman} that all leading
power corrections of the type $1/(fQ)$, f being the event shape variable, can
be resummed when folding the perturbative distribution with a
non-perturbative \textit{shape function}.
The form of the shape function depends on the observable,
and new non-perturbative
parameters are introduced. Fits have been tried for
Thrust and Heavy Jet Mass. For the former a good fit quality for a
large energy range is obtained, and $\mbox{$\alpha_s$}(M_Z)$ values close to
the world average are found. However, in case of the latter no satisfactory
fits could be achieved. This should be followed up in the future analyses.
\section{Conclusions}
\label{conclusions}
Significant progress has been made in the understanding
of power corrections to event shape variables in \mbox{$\mathrm{e}^+\mathrm{e}^-$}\ annihilations.
Universality of the non-perturbative parameter $\alpha_0$ is observed
at the level of 20\%. Some open questions remain such as the difference
of \mbox{$\alpha_s$}\ values obtained with power laws and MC corrections. Also the
effects of quark or hadron masses should be studied.
Power law predictions for other
variables such as the differential two-jet rate as well as the energy-energy
correlations are awaited for.
A new approach based on non-perturbative shape functions looks very promising,
but some further investigations are required.
\section{Acknowledgements}
\label{thanx}
I would like to thank H.~Stenzel for providing me with the
results of his power law studies, and G.~Salam and G.P.~Korchemsky
for helpful discussions.
|
\section{Introduction}
Density correlations in the vicinity of equilibrium planar interfaces
have been extensively studied by numerous authors \cite{e1,e2,w,s,dj,z}.
Several methods, including the density functional theory
\cite{e1,e2}, the capillary wave model \cite{w,s},
and the eigenstate expansion of the fluctuations \cite {e2,dj,z} were used to calculate the
correlation function.
Most existing approaches have to confront a problem of divergence,
caused
by the vanishing energy cost of rigid shifts of the interface in an
infinite system without external potential.
We will focus in this work on
the Hamiltonian second derivative $\left.{\d^2 \H[\r] \over\d \r(z_1) \d \r(z_2)}\right|_{\r=\r_0}\!\!\!\!$
eigenstate expansion method (sometimes also called ``field theoretical method'', \cite{e2,z}).
In the framework of this method the divergence manifests itself
in the term $\phi_0(z_1)\phi_0(z_2)/\lambda_0$
where $\phi_0(z)$ and $\lambda_0$ are the ``zero'' eigenstate and eigenvalue;
$\phi_0(z)\equiv d\rho_0(z)/dz$ where $\rho_0(z)$ is the equilibrium density profile.
The eigenvalue $\lambda_0$ is zero for vanishing external
localizing field,
The traditional physical explanation for the divergence is the following:
the density fluctuations corresponding to the interface
shifts, to the first order of shift amplitude, have component
only along the zero eigenstate,
$\Delta\rho(z)= \rho_0(z+\Delta z)-\rho_0(z)=d\rho(z)/dz \Delta z +{\cal O}(\Delta z)^2$,
and
free wandering of the interface as a whole results in the ambiguity
in defining the density-density correlation function.
In the framework of the capillary wave model (\cite{w}) this ambiguity
is overcome by renormalizing the equilibrium density profile
by taking into account the free wandering
mode. As a result, average width of the renormalized interface diverges
for vanishing external field, and the zero-order term contribution to the
density-density correlation function tends to zero.
However, after such renormalization, one loses information about instantenous
and local
fluctuations of the interface.
In an attempt to resolve this ambiguity,
we would like to focus our attention on the question of divergence of the zero eigenstate term
The main motivation of our approach lies in the following.
It is true that in an infinite system without a confining external field the
energy cost of rigid shifts of the planar two-phase interface is zero,
${\cal H}[\rho(z_0+\Delta z)]={\cal H}[\rho(z_0)]$. Yet this does not mean that the amplitude
$a_0$ of the
density fluctuation proportional to the zero eigenstate,
$\delta \rho (z)=a_0 \rho'_o(z)$ can grow infinitely without a free-energy
penalty. Physically it follows from the fact, that although the position of the interface is undetermined,
the actual values of the density near the interface
cannot go significantly beyond the density of either of the bulk phases.
For all the realistic Hamiltonian functionals describing stable
systems, this non-divergence is controlled by
a positive coefficient in front of the highest power of the density.
For the Ginzburg-Landau
functional
\begin{equation}
\label{GL}
{\cal H}[\rho]=\int (\{{d\over dz}\rho(z)\}^2+\{1-\rho^2(z)\}^2) dz
\end{equation}
the latter is $+\rho^4(z)$. The divergence that appears when only the
harmonic (second-order) terms of the
expansion of the Hamiltonian are used is the direct consequence of
neglect of all higher-order terms.
Hence, to eliminate the divergence of the zero-eigenstate term in the correlation
function,
it is natural to try using
higher-order terms of the expansion of the Hamiltonian around the
equilibrium density profile (see \cite{j} where the equilibrium profile was calculated using
higher-order terms).
It turns out that fourth-order terms are sufficient to keep
the zero-eigenstate contribution finite.
We propose an approximate method to consider these fourth order terms;
this method is formally introduced in Section II. Concrete examples for
the Ginzburg-Landau Hamiltonian for one- and three- dimensional systems
are presented in Sections III and IV.
However, after eliminating the divergence of the zero-order eigenstate term, it
is natural to ask,
which eigenstate or combination of eigenstates of the second
derivative matrix describe the macroscopic shifts of the interface
$\Delta\rho(z)= \rho_0(z+\Delta z)-\rho_0(z)$?
The expansion coefficient of this density fluctuation along the zero
eigenstate remains finite
even for infinite shifts,
\begin{equation}
\label{gen1}
\int_{-\infty}^{+\infty}\Delta\rho(z)\rho_0'(z)dz \rightarrow \rho_0(+\infty)
[\rho_0(+\infty)-\rho_0(-\infty)]
\end{equation}
for $\Delta z\rightarrow +\infty$.
This statement could serve as another argument for a finite average value of the
amplitude of fluctuation along
$\rho_0'(z)$. The same is true for all other bound (localized) eigenstates.
Yet the projection of the shift $\rho_0(z+\Delta z)-\rho_0(z)$ onto low-lying
continuum states
diverges as $\Delta z \rightarrow \infty$. For the Ginzburg-Landau Hamiltonian
the mean-field
equilibrium density profile is $\rho_0(z)=\tanh(z)$, and the
first non-localized eigenstate of the second derivative matrix is proportional to
$(3 \tanh(z)-1)/2$. The integral
\begin{equation}
\label{gen}
\int_{-\infty}^{+\infty}(\tanh(z+\Delta z)-\tanh(z))(3/2 \tanh(z)-1/2)dz \rightarrow 2 \Delta z
\end{equation}
diverges linearly for $\Delta z \rightarrow \infty$.
It means that since macroscopic shifts have no energy cost, the appropriate
linear combination of low-lying continuum states with some of the
coefficients growing proportionally to the
magnitude of the shift also has no energy cost. The finiteness
of the average values of all the continuum spectrum amplitudes
is another artifact of second-order truncation; in particular,
it is a result of the neglect of the mixing of different harmonics in the third
and
higher-order terms.
Consequently, our approximation is convergent only because, after going to the
higher-order expansion in the zero term, we stopped short of going beyond the harmonic
approximation for terms containing the lowest-lying continuum states.
However, the main contribution to the correlations near the interface comes from the
bound states, while continuum states are more relevant for the correlations in the
bulk phases.
A possible merit of our approximation is the improvement in the accuracy
of
density correlation calculations
in the vicinity of the interstate. To compare our results to experimental data, theoretical calculations should be supplemented by {\it a priori} knowledge of
the macroscopic localization of the interface.
For a three-dimensional system the situation is similar. For
square-gradient energy functionals, the perturbation
$\Delta\rho(x,y,z)=\rho_0(z+f(x,y))-\rho_0(z)$ of an initially flat interface
has the energy cost
\begin{equation}
\label{3di}
\Delta F = \int_{-\infty}^{+\infty}
[\rho'_0(z)]^2 dz\int_{-\infty}^{+\infty}\int_{-\infty}^{+\infty} |\nabla f(x,y)|^2
dx dy
\end{equation}
For long-wavelength fluctuations, $f(x,y)\sim\exp[i(k_x x+k_y y)]$,
$|\vec k|\ll 1$, the energy cost of such fluctuations vanishes as
$|k|^2$. However, considering the expansion of the density fluctuations over
the system of eigenstates (now in 3D) of the Hamiltonian second derivative matrix,
$\psi_i(\vec r)\equiv 1/L\; \phi_i(z) \exp(ik_x x)\exp(ik_y y)$, the divergent
contribution comes not from the terms with localized $\phi_i$, but from the bottom
of the continuum
of $z$-coordinate eigenstates. To illustrate that, in section IV we
calculate the convergent
contribution to the correlation function from the $\phi_0(z) \exp(ik_x x)\exp(ik_y y)$
eigenstate.
\section{Formalism}
A density-density correlation function $g(z_1,z_2)$
is defined as a thermal average of a product of density fluctuations
$\Delta \rho(z)=\rho(z)-\rho_0(z)$
around the equilibrium density
profile $\rho_0(z)$:
\begin{equation}
\label{cor}
g(z_1,z_2)\equiv \langle\Delta\rho(z_1)\Delta\rho(z_2)\rangle.
\end{equation}
For simplicity, in this section we consider a one-dimensional case, $\rho=\rho(z)$.
Following \cite{z} we express it as a functional integral over all possible density
profiles,
\begin{equation}
\label{def}
g(z_1,z_2)=(1/Z)\int { \cal D}\rho \; \Delta\rho(z_1)\Delta\rho(z_2) \exp\{ -{{\cal H}[\rho]}\}
\end{equation}
where ${{\cal H}[\rho]}\equiv H[\rho]/k_b T$ is a reduced Hamiltonian functional,
and the partition function $Z$ serves as the normalization constant,
\begin{equation}
\label{nor}
Z=\int { \cal D}\rho \; \exp\{ -{{\cal H}[\rho]}\}
\end{equation}
In the mean-field approximation that we will use through this work,
the equilibrium density profile
$\rho_0(z)$
is determined as the one that
minimizes the Hamiltonian,
\begin{equation}
\label{eqi}
\left.{\d \H[\r] \over\d \r(z)}\right|_{\r=\r_0}\!\!\!\!=0.
\end{equation}
To evaluate (\ref{def}),
we proceed by calculating the eigenstates of the integral operator
with the kernel $\left.{\d^2 \H[\r] \over\d \r(z_1) \d \r(z_2)}\right|_{\r=\r_0}\!\!\!\!$,
\begin{equation}
\int \left.{\d^2 \H[\r] \over\d \r(z_1) \d \r(z_2)}\right|_{\r=\r_0}\!\!\!\! \phi_i(z_2)\,dz_2=\lambda_i\phi_i(z_1).
\label{eig}
\end{equation}
There is always a special eigenstate corresponding to $\phi_0(z)=d\rho_0(z)/dz$
which has zero eigenvalue $\lambda_0=0$ since
\begin{equation}
\label{zer}
\int \left.{\d^2 \H[\r] \over\d \r(z_1) \d \r(z_2)}\right|_{\r=\r_0}\!\!\!\! {d\rho_0(z_2)\over dz_2}\, dz_2={d\over dz}\left.{\d \H[\r] \over\d \r(z)}\right|_{\r=\r_0}\!\!\!\! \equiv0.
\end{equation}
The Hamiltonian is usually real and contains only even powers
of differential operators; it makes the integral operator in Eq.~(\ref{eig})
Hermitian. The
system of eigenstates $\phi_i$ is complete
and orthogonal; we also assume that it is normalized with unit weight function
and all eigenfunctions $\phi_i$ are made real.
An arbitrary density fluctuation $\Delta\rho(z)= \rho(z)-\rho_0(z)$
can be expanded over the complete set of functions $\{\phi_i(z)\}$:
\begin{equation}
\label{exp}
\Delta\rho(z)=\sum_{i=0}^{\infty} a_i \phi_i(z), \; \; a_i=\int \Delta\rho(z)\phi_i(z)dz
\end{equation}
Using the expansion(\ref{exp}), the functional integral (\ref{def}) can be expressed
as
\begin{equation}
\label{cor2}
g(z_1,z_2)={1\over Z}\sum_{i=0}^{\infty} \sum_{j=0}^{\infty} \int\ldots \int
a_i a_j \phi_i(z_1) \phi_j(z_2) \exp \{-{\cal H}[\rho_0+\sum_{k=0}^{\infty}a_k \phi_k]\}
\prod_{m=0}^{\infty}da_m
\end{equation}
where the normalization constant
\begin{equation}
\label{cor3}
Z=\int\ldots \int
\exp \{-{\cal H}[\rho_0+\sum_{k=0}^{\infty}a_k \phi_k]\}
\prod_{m=0}^{\infty}da_m.
\end{equation}
We assume here that the integrals in both numerator and denominator
are convergent. As we mentioned in the Introduction, it is not true
for the single specific direction in the space of coefficients $\{a_i\}$,
corresponding to
the rigid macroscopic shifts of the interface. However, as a result of the
approximations made below, this divergence will not affect the further
calculations.
Traditionally, ${\cal H}[\rho_0+\sum_{k=0}^{\infty}a_k \phi_k]$
is expanded to the second order around the
equilibrium density profile, the orthogonality conditions for the $\phi_i$ are
used,
and the corresponding Gaussian integrals factorized \cite{e2,z}.
As a result, the familiar expression for the
density-density correlation function is
recovered:
\begin{equation}
\label{trad}
g(z_1,z_2)=\sum_{i=0}^{\infty}\langle a_i^2\rangle\phi_i(z_1) \phi_i(z_2),
\end{equation}
where $\langle a_i^2\rangle={1/ \lambda_i}$.
As we already mentioned, the zero term
diverges since $\lambda_0=0$.
However, as we discussed in the Introduction, from physical considerations
$\langle a_0^2\rangle$ must
have finite value. It is indeed the case
if in Eqs.~(\ref{cor2}), (\ref{cor3}) one goes to higher than second order in
the expansion of ${\cal H}[\rho]$.
In our case, the expansion of
${\cal H}[\rho]$ up to the fourth order in the density fluctuation around the
equilibrium profile is sufficient:
\begin{eqnarray}
\label{big}
\nonumber
&{\cal H}[\rho_0+{\displaystyle\sum_{k=0}^{\infty}}a_k \phi_k]\approx &\\
\nonumber
&{\displaystyle\sum_{k=0}^{\infty}a_k \int }\left.{\d \H[\r] \over\d \r(z)}\right|_{\r=\r_0}\!\!\!\! \phi_k(z)
dz+&\\
&{1\over2}{\displaystyle \sum_{k,l=0}^{\infty}a_k a_l\int\int}\left.{\d^2 \H[\r] \over\d \r(z_1) \d \r(z_2)}\right|_{\r=\r_0}\!\!\!\!\phi_k(z_1)
\phi_l(z_2) {\displaystyle \prod_{j=1}^2 }dz_j +&\\
\nonumber
&{1\over3!}{\displaystyle \sum_{k,l,m=0}^{\infty}a_k a_l a_m
\int\int\int} \left.{\d^3 \H[\r] \over\d \r(z_1)\d \r(z_2)\d \r(z_3)}\right|_{\r=\r_0}\!\!\!\! \phi_k(z_1)\phi_l(z_2)\phi_m(z_3){\displaystyle \prod_{j=1}^3}
dz_j +&\\
\nonumber
&{1\over4!}{\displaystyle \sum_{k,l,m,n=0}^{\infty}a_ka_l a_m a_n
\int\int\int\int} \left.{\d^4 \H[\r] \over\d \r(z_1)\d \r(z_2)\d \r(z_3)\d \r(z_4)
\phi_k(z_1)\phi_l(z_2)\phi_m(z_3)\phi_n(z_4) {\displaystyle \prod_{j=1}^4 dz_j}.&
\end {eqnarray}
The first order term is identically zero, the second-order terms are used in the
traditional
formalism, and the third and fourth-order terms are essential for our treatment.
If we substitute this expansion into the expressions for the correlations
function (\ref{cor2}), (\ref{cor3}), for non-pathological forms of the
Hamiltonian
functional ${\cal H}[\rho]$, it will produce a finite value for
$\langle a_0^2\rangle$. However, in their complete form Eqs.~(\ref{cor2}),
(\ref{cor3}), (\ref{big}) are
hardly tractable. Assuming that we are far enough from a critical point, we
use the following approximation to evaluate
$\langle a_0^2\rangle$.
We assume that the second-order expansion works well enough
for all $\langle a_i^2\rangle={1/ \lambda_i}$ with $i\neq 0$
and drop from Eq.~(\ref{big})
all terms
that do not contain $a_0$. In the remaining terms we replace
all combinations of $a_i$, $a_ia_j$, $a_ia_ja_k$,
$\{i,j,k\}\neq0$ by their average values obtained with
the second-order expansion:
\begin{eqnarray}
\label{av}
\nonumber
&\langle a_i \rangle=0,&\\
&\langle a_i a_j \rangle=\delta_{ij}{1\over \lambda_i},&\\
\nonumber
&\langle a_i a_j a_k \rangle=0.&\\
\nonumber
\end{eqnarray}
This approximations allow us to express $\langle a_0^2\rangle$ in the form of the
following integral:
\begin{equation}
\label{e}
\langle a_0^2\rangle={\int x^2 \exp\{-\alpha x^4 -\beta x^2 -\gamma x -\delta x^3
\}dx \over
\int \exp\{-\alpha x^4 -\beta x^2 -\gamma x -\delta x^3\} dx},
\end {equation}
with coefficients $\alpha,\;\beta,\;\gamma$ given by
\begin{eqnarray}
\label{coeff}
\nonumber
&\alpha={1\over4!}{\displaystyle
\int\int\int\int} \left.{\d^4 \H[\r] \over\d \r(z_1)\d \r(z_2)\d \r(z_3)\d \r(z_4)
\phi_0(z_1)\phi_0(z_2)\phi_0(z_3)\phi_0(z_4){\displaystyle \prod_{j=1}^4 }dz_j,&\\
&\beta={1\over4}{\displaystyle \sum_{i=1}^{\infty}
{1\over\lambda_i}\int\int\int\int \left.{\d^4 \H[\r] \over\d \r(z_1)\d \r(z_2)\d \r(z_3)\d \r(z_4)
\phi_i(z_1)\phi_i(z_2)\phi_0(z_3)\phi_0(z_4) \prod_{j=1}^4 }dz_j,&\\
\nonumber
&\gamma={1\over2}{\displaystyle \sum_{i=1}^{\infty}{1\over\lambda_i}
\int\int\int \left.{\d^3 \H[\r] \over\d \r(z_1)\d \r(z_2)\d \r(z_3)}\right|_{\r=\r_0}\!\!\!\!
\phi_i(z_1)\phi_i(z_2)\phi_0(z_3) \prod_{j=1}^3 }dz_j,&\\
\nonumber
&\delta={1\over3!}{\displaystyle
\int\int\int \left.{\d^3 \H[\r] \over\d \r(z_1)\d \r(z_2)\d \r(z_3)}\right|_{\r=\r_0}\!\!\!\!
\phi_0(z_1)\phi_0(z_2)\phi_0(z_3) \prod_{j=1}^3 }dz_j.&\\
\nonumber
\end{eqnarray}
\section{Ginzburg-Landau Hamiltonian}
To illustrate our approach let us a consider a simple
1D system with the Ginzburg-Landau Hamiltonian (\ref{GL}). The
equilibrium density profile for symmetric boundary conditions
$\rho(-l)=-1$, $\rho(+l)=1$, $l\rightarrow \infty$ is
$\rho_0(z)=\tanh(z)$. The eigenvalue equation (\ref{eig}) takes the
differential form
\begin{equation}
\label{shr}
-2{d^2\over dz^2}\phi_i(z)-4\phi_i(z)+12\tanh^2(z)\phi_i(z)=\lambda_i\phi_i(z),
\end{equation}
and the third and fourth-order functional derivatives of ${\cal H}[\rho]$ are:
\begin{equation}
\left.{\d^3 \H[\r] \over\d \r(z_1)\d \r(z_2)\d \r(z_3)}\right|_{\r=\r_0}\!\!\!\!=24\tanh(z_1)\delta(z_1-z_2)\delta(z_2-z_3)
\end{equation}
\begin{equation}
\left.{\d^4 \H[\r] \over\d \r(z_1)\d \r(z_2)\d \r(z_3)\d \r(z_4)=24\delta(z_1-z_2)\delta(z_2-z_3)\delta(z_3-z_4).
\end{equation}
In fact, the expansion (\ref{big}) is now exact for ${\cal H}[\rho]$
since
all higher-order variational derivatives are identically equal to zero.
To proceed further we need to calculate
$\alpha$, $\beta$, $\gamma$, and $\delta$ as defined in
Eq.~(\ref{coeff}). From parity consideration,
the coefficients $\gamma$ and
$\delta$ are zero.
The calculation of $\alpha $ is straightforward:
\begin{equation}
\label{a}
\alpha= {9\over 16}
\int_{-\infty}^{+\infty}{dz \over \cosh^{8}(z)}
={18\over35}.
\end{equation}
The coefficient ${9\over 16}$ appears from the normalization condition:
\begin{equation}
\label{norm}
\int_{-\infty}^{+\infty}\phi_i^2(z)dz=1.
\end{equation}
To evaluate $\beta$ we need to know the eigenstates $\phi_i(z)$ of
Eq.~(\ref{shr}):
\begin{equation}
\label{b}
\beta={9\over 2}\sum_{i=1}^{\infty}\int_{-\infty}^{+\infty}
{\phi_i^2(z)\over\lambda_i}
{dz \over \cosh^{4}(z)}.
\end{equation}
The sum in Eq.~(\ref{b}) contains a contribution from one bound state,
($\phi_1(z)=\sqrt{3/2} \sinh(z)\cosh^{-2}(z)$, $\lambda_1=6$) and continuum states
($i\geq 2$)
To evaluate the sum over the continuum,
we use expressions obtained in \cite{z} (Eq.~(3.15 - 3.20)):
\begin{equation}
\label{cont}
{9\over 2}\sum_{i=2}^{\infty}\int_{-\infty}^{+\infty}
{\phi_i^2(z)\over\lambda_i}
{dz \over \cosh^{4}(z)}={9\over 4}\int_{-\infty}^{+\infty}
{dk \over 2 \pi}{1 \over (k^2+4)^2}{1 \over (k^2+1)}
\int_{-\infty}^{+\infty}{\phi_k(z) \over \cosh^{4}(z)} dz
\end{equation}
where the continuum eigenstates $\phi_k(z)$ are
\begin{equation}
\label{fk}
\phi_k(z)=\exp(ikz)[1+k^2+3ik\tanh(z)-3\tanh^2(z)].
\end{equation}
The normalization of $\phi_k(z)$ is taken into account in the $k$-dependent factor
of the outer integral in (\ref{cont}).
Both integrations in (\ref{cont}) are straightforward and finally we obtain for
$\beta$
\begin{equation}
\label{be}
\beta\approx {6\over35} + 0.022 \approx 0.193
\end {equation}
The contribution from the continuum to the value of $\beta$ is $\approx 13\%$.
We substitute these values of $\alpha$ and $\beta$ into Eq.~(\ref{e}), and, in the first order
in $\beta/ \sqrt{\alpha}$, obtain for
$\langle a_0^2\rangle$:
\begin{equation}
\label{et}
\langle a_0^2\rangle \approx {\int x^2 \exp\{-\alpha z^4\}(1-\beta z^2) dz \over
\int \exp\{- {\alpha}z^4\} (1-\beta z^2) dz}={1\over \sqrt{\alpha}} \; {\Gamma^2({3\over 4})-
{\beta \over \sqrt{\alpha}}{\pi \sqrt{2}\over 4} \over \pi \sqrt{2} -{\beta \over \sqrt{\alpha}}
\Gamma^2({3\over 4})} \approx 0.415
\end {equation}
It is interesting to note that if one ignores all cross-terms (
contribution from $\beta$) in
Eq.~(\ref{e}), $\langle a_0^2\rangle \approx 0.471$.
For the Ginzburg-Landau Hamiltonian, ${\beta \over \sqrt{\alpha}}\approx
0.27$ plays the role of a small parameter in
the approximations (\ref{av}) - (\ref{coeff}),
as well as in (\ref{et}).
Hence there is a certain degree of numerical justification of heuristic assumptions made in (\ref{av}) - (\ref{coeff}).
To obtain the final result for the density-density correlation function
(\ref{trad})
we use the following method. When the external potential $V(z)=cz$,
linear in the coordinate $z$ measured form the interface location
$z=0$,
is added to the Hamiltonian, the zero eigenvalue
becomes proportional to the coefficient in front of this term:
$\lambda_0 \propto c$.
In \cite{z} an expression for the correlation function
$G_c(z_1,z_2)$ (Eq. (4.17) in \cite{z})
is obtained in the presence of an external potential $V(z)$, with
$V(z)\rightarrow
c\tanh(z)$ when $c \rightarrow 0$.
It is shown that the zero eigenvalue $\lambda_0/c \rightarrow 1 $ for
$c \rightarrow 0$. Since for $i\geq 1$ all $\lambda_i$ go to constant limits when
the external potential is turned off, the ``truncated'' correlation function
$\bar g$
(without the zero eigenstate term) can be expressed as:
\begin{equation}
\label{gbar}
\bar g({z_1,z_2})\equiv\sum_{i=1}^{\infty}{\phi_i(z_1) \phi_i(z_2)\over \lambda_i}=
\lim_{c \rightarrow 0} {d \over d c}cG_c(z_1,z_2).
\end {equation}
To recover the ``non-truncated'' correlation function $g({z_1,z_2})$ we
add to $\bar g({z_1,z_2})$ from (\ref{gbar}) the correct contribution form the
zero-order term,
\begin{equation}
\label{gr}
g({z_1,z_2})= \bar g({z_1,z_2})+{3\over 4\cosh^2(z_1)\cosh^2(z_2)}
\langle a_0^2\rangle.
\end {equation}
with $\langle a_0^2\rangle$ given by (\ref{et}).
Sketches of the density-density correlation function (\ref{gr})
are presented in Figs.~1,2.
It is straightforward to demonstrate that far from the interface
(${z_1.z_2}\gg 1$), the density-density correlation function
(\ref{gr}) goes to the correct bulk phase limit,
\begin{equation}
\label{bulk}
g_{bulk}(z_1,z_2)={\exp(-2|z_1-z_2|)\over 8}.
\end{equation}
\begin{figure}
\centerline{\epsfxsize=12cm \epsfbox{cont.eps}}
\centerline{{\small {\bf Fig.~1}.
A contour plot of the density-density correlation function
$g(z_1,z_2)$.}}
\end{figure}
\begin{figure}
\centerline{\epsfxsize=12cm \epsfbox{3d.eps}}
\centerline{{\small {\bf Fig.~2}.
A 3D sketch of the density-density correlation function
$g(z_1,z_2)$.}}
\end{figure}
\section{3D Calculation}
The next logical step is to generalize this approach to
a more realistic 3D system.
For simplicity, we consider the same
Ginzburg-Landau Hamiltonian as in (\ref{GL})
\begin{equation}
\label{GL3D}
{\cal H}[\rho]=\int (|\nabla\r|^2+\{1-\rho^2(\vec r)\}^2 )d\vec r
\end{equation}
The equilibrium density profile for the symmetric boundary conditions
$\rho(z=-l)=-1$, $\rho(z=+l)=1$, $l\rightarrow \infty$ is the same
as in the 1D case,
$\rho_0(z)=\tanh(z)$. The
three-dimensional analog of the
eigenvalue equation (\ref{shr}) reads
\begin{equation}
\label{shr3D}
-2\Delta\psi_i(\vec r)-4\psi_i(\vec r)+12\tanh^2(z)\psi_i(\vec r)=\lambda_i\psi_i(\vec r),
\end{equation}
Since the ``potential'' term in (\ref{shr3D}) depends only on $z$,
the variables can be separated, $\psi_i(\vec r)=\phi_i(z)\xi_{i}(x,y)$.
We assume that in the $xy$ plane the system is confined to a square box of
size
$L$ with periodic boundary conditions.
Then the $xy$ components of the eigenstates of (\ref{shr3D})
are the normalized plane waves
$\xi_{\vec k}(x,y)=(1/L)\exp(ik_x x)
\exp(ik_y y)$ with $ k_{\{x,y\}}=2\pi n_{\{x,y\}}/L$, $ n_{\{x,y\}}=0,\pm 1,
\pm 2 \ldots$.
Expanding density variations over the complete set of
functions $\psi_{i\vec k}(\vec r)$ and taking the thermal average,
we obtain for the density-density correlation function (compare to (\ref{cor2})):
\begin{eqnarray}
\label{cor3D}
\nonumber
&g(\vec r_1,\vec r_2)={1\over Z}\displaystyle\sum_{i,\vec k_1}
\displaystyle\sum_{j,\vec k_2} \int \ldots \int
a_{i.\vec k_1}
a_{j\vec k_2} \psi_{i\vec k_1}^*(\vec r_1) \psi_{j\vec k_2}(\vec r_2)&\\
& \exp \{-{\cal H}[\rho_0+\displaystyle\sum_{m,\vec k_3} a_{m\vec k_3} \psi_{m\vec k_3}]\}
\displaystyle \prod_{n,\vec k_4}da_{n\vec k_4}.&
\end{eqnarray}
Similarly to (\ref{cor3}), $Z$ is a normalization constant:
\begin{equation}
\label{cor33D}
Z=\int \ldots \int
\exp \{-{\cal H}[\rho_0+\sum_{m,\vec k_1} a_{m\vec k_1} \psi_{m\vec k_1}]\}
\prod_{n,\vec k_2}da_{n\vec k_2}.
\end{equation}
Here and below we use the shorthand notation: for sums, products, and
subscript indexes symbol $\vec k$ means the set of $xy$ eigenstate
indexes $\{n_x,n_y\}$.
After expanding (\ref{cor3D}) to the second order in the
density variation,
an expression analogous to (\ref{trad}) is recovered:
\begin{equation}
\label{trad3D}
g(\vec r_1,\vec r_2)=\sum_{i,\vec k}\langle a_{i\vec k}^2\rangle\psi_{i\vec k}^*(\vec r_1)
\psi_{i\vec k}(\vec r_2),
\end{equation}
with $\langle a_{i\vec k}^2\rangle={1/ (\lambda_i+2k^2)}$.
As in the (\ref{trad}), a similar problem arises for the
$\lambda_0=0$ eigenstate:
for $L\rightarrow \infty$ the sum on $\vec k$ diverges
at the lower limit.
To avoid the divergence occurring in the $\lambda_0=0$ term, we suggest the same
recipe as in the
one-dimensional
case: to continue the expansion of the Hamiltonian to
the fourth order.
For simplicity, we neglect mixing of the eigenstates with different $\lambda_i$
in the fourth-order term. Similarly to the 1D case, where the relative
contribution of mixing is given by the ratio of
coefficients $\beta/\sqrt{\alpha}\approx 0.27 $ (\ref{a},\ref{be},\ref{et}),
the inclusion of this
mixing here will not affect convergence
but will slightly change the numerical values of the coefficients.
For all $\lambda_i \neq 0$, the second-order result (\ref{trad3D}) is sufficient,
hence we perform fourth-order expansion only for the subspace of
eigenstates with $\lambda_0=0$.
The contribution from the $\lambda_0=0$ eigenstate to the density-density correlation function can be expressed as:
\begin{eqnarray}
\label{cor3D0}
\nonumber
&g_0(\vec r_1,\vec r_2)={1\over Z_0}\displaystyle\sum_{\vec k_1}
\displaystyle\sum_{\vec k_2} \int \ldots \int
a_{0\vec k_1}
a_{0\vec k_2} \psi_{0\vec k_1}^*(\vec r_1) \psi_{0\vec k_2}^*(\vec r_2)&\\
& \exp \{-{\cal H}[\rho_0+\displaystyle\sum_{\vec k_3} a_{0\vec k_3} \psi_{0\vec k_3}]\}
\displaystyle \prod_{\vec k_4}da_{0\vec k_4},&\\
\nonumber
&Z_0=
\displaystyle \int \ldots \int
\exp \{-{\cal H}[\rho_0+\displaystyle \sum_{\vec k_1} a_{0\vec k_1} \psi_{0\vec k_1}]\}
\;\displaystyle \prod_{\vec k_2}da_{0\vec k_2}.&
\end{eqnarray}
Using the orthogonality conditions for $\xi_{\vec k}(z,y)$ and
Eq.~(\ref{coeff}) we obtain
\begin{eqnarray}
\label{cor3D1}
\nonumber
&g_0(\vec r_1,\vec r_2)={1\over Z_0}\displaystyle\sum_{\vec k_1}
\int \ldots \int
a_{0\vec k_1}^2
\psi_{0\vec k_1}^*(\vec r_1) \psi_{0\vec k_1}(\vec r_2)&\\
& \exp \{-2\displaystyle\sum_{\vec k_2} a_{0\vec k_2}^2 k_2^2
-{\alpha\over L^2}[\displaystyle\sum_{\vec k_3} a_{0\vec k_3}^2]^2 \}
\;\displaystyle \prod_{\vec k_4}da_{0\vec k_4},&\\
\nonumber
&Z_0=\displaystyle
\int \ldots \int
\exp \{-2\displaystyle \sum_{\vec k_1} a_{0\vec k_1}^2 k_1^2 -
{\alpha\over L^2}[\displaystyle \sum_{\vec k_2}
a_{0\vec k_3}^2]^2 \}
\;\displaystyle \prod_{\vec k_3}da_{0\vec k_3},&
\end{eqnarray}
where $\alpha$ is defined by (\ref{a}).
Direct evaluation of the functional integrals in (\ref{cor3D1}) is impossible;
however, a simple approximation will allow us to obtain a physically reasonable
expression for $g_0(\vec r_1,\vec r_2)$.
The main contribution to the integral in (\ref{cor3D1}) comes from
$a_{0\vec k}$ with small $|\vec k|$, so in the first approximation
it is natural to introduce
an upper cutoff $C$ in sums on $\vec k$.
In particular, we replace the infinite limits in all
the sums and products on $n_x$ and $n_y$ in (\ref{cor3D1}) by
the finite cutoff,
$|n_{x,y}|\leq \sqrt{C}L/2$. It corresponds to a system-size-independent
cutoff for a wavevector $\vec k$ with $|k_{x,y}|\leq \sqrt{C} \pi$.
We select $C$ in such a way that
the term $-2\sum_{\vec k} a_{0\vec k}^2 k^2$, quadratic in $a_{0\vec k}$,
can be neglected in the exponent, which decouples the integration over
$da_{0\vec k}$ from the summation on ${n_x,n_y}$.
Besides neglecting the quadratic term, we remove
the functions $\psi_{0\vec k}(\vec r)$ from (\ref{cor3D1}),
and call the remaining expression $A$.
\begin{equation}
\label{A}
A \equiv {\displaystyle \sum_{k_1}^N
\int \ldots \int
a_{k_1}^2
\exp \{-{\alpha\over L^2}[\displaystyle\sum_{k_2=1}^N a_{k_2}^2]^2 \}
\;\displaystyle \prod_{k_3}^N da_{k_3}\over
\displaystyle
\int \ldots \int
\exp \{-{\alpha\over L^2}[\displaystyle\sum_{k_4=1}^N a_{k_4}^2]^2 \}
\;\displaystyle \prod_{k_5}^N da_{k_5}},
\end{equation}
with $N=CL^2$. This expression can be evaluated by using
$N$-dimensional spherical coordinates, $\sum_{k}^N a_{k}^2 \equiv R^2$
\begin{equation}
\label{A1}
A = {\displaystyle \int R^2
\exp \{-{\alpha\over L^2}R^4\}
R^{N-1}\Omega _N dR \over
\displaystyle \int
\exp \{-{\alpha\over L^2}R^4\}
R^{N-1}\Omega _N dR }={L\over\sqrt{\alpha}}{\Gamma({N+2\over 4})\over
\Gamma({N\over 4})}={L\over 2}[\sqrt{N\over\alpha}+{\cal O}
({1\over\sqrt{N}})]
\end{equation}
Here $\Omega_N=2\pi^{N/2}/\Gamma(N/2)$ is the area of the N-dimensional
sphere.
By inspection, one can identify $A$ as the sum of the first N terms of
$\langle a^2\rangle $, $A=\sum_{l=1}^N\langle a^2\rangle $. Therefore
\begin{equation}
\label{A2}
\langle a^2\rangle={A\over N}={1\over 2} \sqrt{1\over \alpha C}
\end{equation}
Now we return to Eq.~(\ref{cor3D1}) and replace one sum
in the fourth-order term in the exponential by
$\sum_{i=1}^N \langle a^2\rangle$:
\begin{equation}
{\alpha\over L^2}[\displaystyle\sum_{\vec k} a_{0\vec k}^2]^2\approx
{\alpha\over L^2}[\displaystyle\sum_{\vec k} a_{0\vec k}^2]N \langle a^2 \rangle
={\sqrt{\alpha C}\over 2}[\displaystyle\sum_{\vec k} a_{0\vec k}^2]
\end{equation}
After this substitution, Eq.~(\ref{cor3D1}) becomes a product of
Gaussian integrals and its evaluation becomes trivial:
\begin{equation}
\label{g0}
g_0(\vec r_1,\vec r_2)\approx \phi_0(z_1) \phi_0(z_2)
\sum_{\vec k}\tilde a_{\vec k}^2 {\exp [i (k_x x+ k_y y)]\over L^2},
\end {equation}
where the ``improved'' average values $\tilde a_{\vec k}^2$ are given by
\begin{equation}
\label{g1}
\tilde a_{\vec k}^2\equiv{ \int a_{\vec k}^2 \exp[-a_{\vec k}^2(2k^2+\sqrt{\alpha C}/2)]
d a_{\vec k}\over \int \exp[-a_{\vec k}^2(2k^2+\sqrt{\alpha C}/2)]
d a_{\vec k}}={1\over 4 k^2+\sqrt{\alpha C}}
\end {equation}
Taking the limit $L\rightarrow\infty$ and replacing
the summation $\sum_{n_x,n_y}$ in
(\ref{g0}) by the integration
$(L/2\pi)^2\int \int dk_x dk_y$, we obtain
\begin{equation}
\label{gf0}
g_0(\vec r_1,\vec r_2)\approx \phi_0(z_1) \phi_0(z_2)
{1\over (4\pi)^2} \int_0^{\infty}{k dk\over k^2 + \sqrt{\alpha C}/4}
J_0(kr')=\phi_0(z_1) \phi_0(z_2){1\over (4\pi)^2} K_0[{
(\alpha C)^{1/4} r_{\perp} \over 2}].
\end {equation}
Here $J_0$ and $K_0$ are Bessel and Modified Hankel
functions of zero order and
$r_{\perp}\equiv\sqrt{(x_1-x_2)^2+(y_1-y_2)^2}$.
For large positive $q$ one has $K_0(q)=\sqrt{\pi/2q}\exp(-q)
[1+{\cal O} (1/q)]$. Consequently, we can identify $(\alpha C)^{1/4}/2$ with the
previously introduced upper cutoff for the wavevector $k$, {\it i.e.}, with
$\sqrt{C} \pi$.
It allows us to express the constant $C$ through the known parameters of
the system,
$C=\alpha/(4\pi)^4$.
Finally, we write for the zero-eigenstate contribution to the
correlation function:
\begin{equation}
\label{gf}
g_0(\vec r_1,\vec r_2)\approx \phi_0(z_1) \phi_0(z_2){1\over (4\pi)^2}
K_0[{\sqrt{ \alpha}\over 4\pi}\sqrt{(x_1-x_2)^2+(y_1-y_2)^2}].
\end {equation}
The contribution from other eigenstates, $\tilde g(\vec r_1,\vec r_2)$,
is obtained by straightforward integration,
\begin{equation}
\label{go}
\tilde g(\vec r_1,\vec r_2)= \sum_{j=1}\phi_j(z_1) \phi_j(z_2){1\over 2 (2\pi)^2}
K_0[\sqrt{ \lambda_j\over 2}\sqrt{(x_1-x_2)^2+(y_1-y_2)^2}]
\end {equation}.
\section{Conclusion}
We show that taking into account the fourth-order terms in the expansion of the
Hamiltonian
in the calculation of the density-density correlation function makes
the previously divergent (for $D\leq3$) zero-eigenstate term convergent.
We also note that the macroscopic shifts
of the interfacial profile are described not by the zero and other bound
eigenstates,
but by the combination of low-lying
continuum eigenstates of the Hamiltonian second-derivative matrix.
The inclusion of the convergent zero-order term allows us to
improve the accuracy of the calculation of the correlation function
in the vicinity of the interface.
Our approach could be relevant for the experimental results
obtained in the microgravity conditions using a local analytical method,
{\it e.g.}, scattering with a narrow beam, focused on the interface.
\section{Acknowledgments}
This work was done in the research group of Prof. B. Widom as part of a program supported by the U.S. National Science Foundation and the Cornell Center for
Material Research. I thank Prof. Widom for having suggested this problem,
for stimulating discussions during the course of the work, and for comments
on the manuscript.
|
\section{* MacroDefs}
\documentclass[12pt]{amsproc}
\usepackage{latexsym}
\def\publname{To appear in the Proceedings volume \\[1pt]
``Algebraic Geometry -- Hirzebruch~70'' \\[1pt]
Series: Contemporary Mathematics}
\copyrightinfo{}{}
\typeout{Macros}
\newtheorem{theorem}{Theorem \thesection.\!\!}
\newtheorem*{mtheorem}{Main Theorem}
\newtheorem{mntheorem}[theorem]{Main Theorem \thesection.\!\!}
\newtheorem{lemma}[theorem]{Lemma \thesection.\!\!}
\newtheorem{vanlemma}[theorem]{Vanishing Lemma \thesection.\!\!}
\newtheorem{proposition}[theorem]{Proposition \thesection.\!\!}
\newtheorem{corollary}[theorem]{Corollary \thesection.\!\!}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition \thesection.\!\!}
\newtheorem{remark}[theorem]{Remark \thesection.\!\!}
\newtheorem{example}[theorem]{Example \thesection.\!\!}
\newtheorem*{prexao}{Proof of Example 1.5}
\newtheorem*{procoo}{Proof of Corollary 4.5}
\newtheorem*{protheo}{Proof of Theorem 4.4}
\typeout{Private macros}
\DeclareMathOperator\Hom{Hom}
\DeclareMathOperator\coker{coker}
\DeclareMathOperator\Tor{Tor}
\DeclareMathOperator\pt{pt}
\def\A{{\mathcal A}}
\def\E{{\mathcal E}}
\def\F{{\mathcal F}}
\def\IH{{\mathcal IH}}
\def\H{{\mathcal H}}
\def\b{^{\scriptscriptstyle\bullet}}
\def\mal{\mathbin{\!\cdot\!}}
\def\CC{{\mathbb C}}
\def\TT{{\mathbb T}}
\def\QQ{{\mathbb Q}}
\def\ZZ{{\mathbb Z}}
\def\PP{{\mathbb P}}
\def\NN{{\mathbb N}}
\def\RR{{\mathbb R}}
\def\mm{{\mathbb m}}
\def\mm{{\mathfrak m}}
\def\cld{{\rm cld}}
\def\longto{\longrightarrow}
\def\mapdown#1{\Big\downarrow\rlap{$\vcenter{\hbox{$\scriptstyle#1$}}$}}
\def\mimapdown#1{\big\downarrow\rlap{$\vcenter{\hbox{$\scriptstyle#1$}}$}}
\def\mapup#1{\Big\uparrow\rlap{$\vcenter{\hbox{$\scriptstyle#1$}}$}}
\def\mimapup#1{\big\uparrow\rlap{$\vcenter{\hbox{$\scriptstyle#1$}}$}}
\begin{document}
\typeout{Topmatter}
\title[Equivariant Intersection Cohomology of Toric Varieties]
{Equivariant Intersection Cohomology\\ of Toric Varieties}
\author[G. Barthel]{Gottfried Barthel}
\address{G.B.: Fakult\"at f\"ur Mathematik und
Informatik, Universit\"at Konstanz, Fach D 203, D-78457 Konstanz}
\email{<EMAIL>}
\author[J.-P. Brasselet]{Jean-Paul Brasselet}
\address{J.-P.B.: IML - CNRS, Case 907 - Luminy, F-13288 Marseille Cedex~9}
\email{<EMAIL>}
\author[K.-H. Fieseler]{\\ Karl-Heinz Fieseler}
\address{K.-H.F.: Matematiska Institutionen, Box 480, Uppsala
Universitet, SE-75106 Uppsala}
\email{<EMAIL>}
\author[L. Kaup]{Ludger Kaup}
\address{L.K.: Fakult\"at f\"ur Mathematik und Informatik,
Universit\"at Konstanz, Fach D 203, D-78457 Konstanz}
\email{<EMAIL>}
\keywords{Toric varieties,
equivariant cohomology, intersection
cohomology, equivariant intersection cohomology,
minimal complexes, sheaves on fans, minimal extension
sheaves, equivariantly formal spaces.
}
\subjclass{14M25, 32S60, 52B20, 55N25}
\dedicatory{Respectfully dedicated to Friedrich Hirzebruch}
\begin{abstract} We investigate the equivariant intersection cohomology
of a toric variety. Considering the defining fan of the variety as a
finite topological space with the subfans being the open sets (that
corresponds to the ``toric'' topology given by the invariant open
subsets), equivariant intersection cohomology provides a sheaf (of
graded modules over a sheaf of graded rings) on that ``fan space''. We
prove that this sheaf is a {\em ``minimal extension sheaf''}, i.e.,
that it satisfies three relatively simple axioms which are known to
characterize such a sheaf up to isomorphism. In the verification of the
second of these axioms, a key role is played by ``equivariantly
formal'' toric varieties, where equivariant and ``usual''
(non-equivariant) intersection cohomology determine each other by
K\"unneth type formul\ae. Minimal extension sheaves can be constructed
in a purely formal way and thus also exist for non-rational fans. As a
consequence, we can extend the notion of an {\em equivariantly formal
fan} even to this general setup. In this way, it will be possible to
introduce ``virtual'' intersection cohomology for equivariantly formal
non-rational fans.
\end{abstract}
\maketitle
\tableofcontents
\typeout{Introduction}
\section*{Introduction}
\noindent The relation between algebraic geometry and topology is
especially close in the case of smooth compact toric varieties: The
canonical homomorphism from the Chow-theoretic intersection ring to
the integral homology intersection ring is an isomorphism. Even when
we allow quotient singularities, the result remains valid if we replace
integer with rational coefficients. \par
As is well known\footnote{For the basic theory of toric varieties,
we refer the reader to the pertinent monographs \cite{Ew}, \cite{Fu},
or \cite{Oda$_1$}.}, toric varieties admit a description in terms of
combinatorial-geometric objects: cones, fans, and polytopes. In the
case considered above, these combinatorial-geometric objects allow us
to compute the intersection ring explicitely. In the opposite direction,
properties of the intersection ring known from algebraic geometry
yield consequences for the describing geometric object. A most striking
application is Stanley's beautiful proof -- using the hard Lefschetz
theorem for projective toric varieties -- for the necessity of the
conditions that characterize the face numbers of simplicial polytopes,
conjectured by \hbox{McMullen} (see, e.g., \cite[\S5.6]{Fu}). That
example should suffice to explain the great interest of toric
varieties in studying the combinatorial geometry of fans -- as a
variation of the title of Hirzebruch's landmark book, one might call
this an {\it Application of Topological Methods from Algebraic Geometry
to Combinatorial Convexity}. \par
Stated in terms of the defining fan, a toric variety $X = X_{\Delta}$
is compact and $\QQ$-smooth (i.e., a rational homology manifold) if the
fan~$\Delta$ is {\it complete\/} and {\it simplicial}. When studying the
(rational) cohomology algebra of such a toric variety~$X$, a crucial
role is played by the (rational) {\it Stanley-Reisner ring\/}
$$
S_\Delta := \QQ[\Delta^1] / I \;,
$$
where $\Delta^1$ is the set of {\it rays\/} (i.e., one-dimensional
cones) of~$\Delta$, and $\QQ[\Delta^1]$ is the polynomial algebra
$\QQ\big[(t_\rho)_{\rho \in \Delta^1}\big]$ on free
generators~$t_{\rho}$ that are in one-to-one correspondence to the
rays~$\rho$, and where~$I$ is the homogeneous ideal
$$
I := \Bigl\langle \prod_{j=1}^k t_{\rho_{j}} \;;\; \sum_{j=1}^k
\rho_{j} \not\in \Delta \Bigr\rangle
$$
generated by those square-free monomials where the rays corresponding
to the factors do not span a cone of the fan. The theorem of Jurkiewicz
and Danilov \cite[3.3, p.~134]{Oda$_1$} describes the (rational)
cohomology algebra $H\b(X)$ as the quotient of $S_\Delta$ modulo the
ideal~$\tilde J$, the image of the homogeneous ideal
$$
J := \Bigl\langle \sum_{\rho \in \Delta^1} \chi(v_\rho) \mal t_\rho
\;;\; \chi \in \Hom(\TT, \CC^*) \Bigr\rangle \subset \QQ[\Delta^1]\;,
$$
where $v_\rho$ denotes the unique primitive lattice vector spanning
the ray~$\rho$, and $\TT$, the acting torus. We thus have isomorphisms
$$
H\b(X_{\Delta},\QQ) \,\cong\, S_\Delta/\tilde J \,\cong\,
\QQ[\Delta^1]\, /\, (I + J)\;. \leqno{(\dagger)}
$$ \par
The ideals~$I$ and~$J$ are most naturally interpreted using the
one-to-one correspondence between rays and $\TT$-invariant irreducible
divisors on~$X$: The monomials generating~$I$ correspond to sets of
divisors with empty intersection, and the generators of~$J$, to
$\TT$-invariant principal divisors. In fact, identifying a product of
generators with the intersection of the corresponding divisors, this
correspondence extends to isomorphisms between the rings
$\QQ[\Delta^1]/(I+J) \cong H\b(X)$ and the rational Chow ring of~$X$.
-- We note that the ideal~$I$, and hence the Stanley-Reisner ring,
depends only on the combinatorial structure of the fan. \par
The isomorphism $\QQ[\Delta^1]/(I+J) \cong H\b(X)$ is a morphism of
graded rings that multiplies degrees by two. As a consequence, the
odd-dimensional cohomology of~$X$ vanishes. Furthermore, it turns out
that the Betti numbers are explicitly computable in terms of
combinatorial data of the fan. \par
Besides of being a graded algebra that encodes the combinatorics of
the fan~$\Delta$, the Stanley-Reisner ring~$S_{\Delta}$ has a second,
more geometric, inter\-pre\-ta\-tion. To give it, we consider, for a
completely arbitrary fan~$\Delta$, the graded algebra $\A\b(\Delta)$ of
{\it (rational) $\Delta$-piecewise polynomial functions\/} on the
support $|\Delta|$ of the fan\footnote{Unless otherwise stated, cones
and fans are always considered as subsets of the {\it rational\/}
vector space $N_{\QQ}$ generated by the one parameter subgroups of the
torus.},
i.e., functions $f \colon |\Delta| \to \QQ$ whose restriction
to each cone $\sigma \in \Delta$ extends to a -- unique -- polynomial
function $f_{\sigma}$ on the linear span $V_{\sigma} := \sigma +
(-\sigma)$ of~$\sigma$. Since the algebra of polynomial functions
on~$V_{\sigma}$ is $S\b(V_{\sigma}^*)$, the symmetric $\QQ$-algebra of
the dual vector space $V_{\sigma}^*$, we arrive at the formal definition
$$
\A\b(\Delta) := \{f \colon |\Delta| \to \QQ \;;\; \forall\,{\sigma
\in \Delta}\;\; \exists\,{f_{\sigma} \in S\b(V_{\sigma}^*)}:
f|_{\sigma} = f_{\sigma}|_{\sigma}\} \;.
$$
In addition to this graded $\QQ$-algebra of piecewise polynomial
functions, we also have to consider the graded $\QQ$-algebra
$$
A\b = S\b(V^*) \cong \QQ\,[u_{1}, \dots, u_{n}]
$$
of (globally) polynomial functions on $V := M_\QQ$, where $u_{1}, \dots,
u_{n}$ denotes a basis of $V^* := M_\QQ$ (see section~0 for notations).
The obvious restriction homomorphism $A\b \to \A\b(\Delta)$,
$f \mapsto f|_{|\Delta|}$ of graded
algebras is injective if the fan~$\Delta$ contains at least one
$n$-dimensional cone, thus making $A\b$ a graded subalgebra of
$\A\b(\Delta)$. \par
In the sequel, the algebra $\A\b(\Delta)$ will be used without any
restriction on the fan. But assuming again that~$\Delta$ is complete and
simplicial for the moment, we obtain
a homomorphism $\QQ[\Delta^1] \to \A\b(\Delta)$ of graded $\QQ$-algebras
by associating to each generator~$t_{\rho}$ of $\QQ[\Delta^1]$ the
unique piecewise linear function on $|\Delta|$ that takes the value~1 at
the vector~$v_{\rho}$ and vanishes on all other rays. This homomorphism
is surjective, and its kernel coincides with the ideal~$I$, so we get
the identification
$$
S_\Delta \buildrel \cong \over \longrightarrow
\A\b(\Delta) \leqno{(*)}
$$
that provides the geometric interpretation of the Stanley-Reisner ring.
\par
Under that identification, the ideal $\tilde J$ in $S_\Delta$ corresponds
to the homogeneous ideal $\mm \cdot \A\b(\Delta)$ generated by the
(globally) linear functions. (In the (sub-)algebra $A\b$, the ideal
generated by linear functions is the unique homogeneous maximal ideal
$\mm = A^{>0}$ of polynomial functions vanishing at the origin.) Using
the above isomorphism~$(\dagger)$, we may look at $H\b(X)$ as quotient
modulo that ideal. We can thus describe the cohomology algebra of a
compact, $\QQ$-smooth toric variety $X_{\Delta}$ by an isomorphism
$$
H\b(X_{\Delta})
\,\cong\, (A\b/\mm) \otimes_{A\b} \A\b(\Delta) \;.
$$ \par
On the other hand, the ring $\A\b(\Delta)$ itself admits a direct
topological interpretation: There is a natural isomorphism
$$
H\b_{\TT}(X_\Delta) \buildrel\cong\over\longrightarrow
\A\b(\Delta)\leqno{(**)}
$$
with the {\it equivariant cohomology algebra\/} of $X_\Delta$ (see,
e.g., \cite{BriVe}, \cite{GoKoMPh}, and the first section of this
article). Hence, the two isomorphisms $(*)$ and $(**)$ together yield
a third, topological, interpretation of the Stanley-Reisner ring.
\par\medskip
The case of {\it non-simplicial\/} fans is much more elusive, as the
above results do not remain valid. In fact, the cohomology becomes
quite a delicate object that is difficult to compute. In particular,
the Betti numbers are not always determined by the combinatorial type
of the fan.\par
Fortunately, for compact toric varieties with arbitrary singularities,
intersection cohomology (with respect to ``middle perversity'') is known
to behave much better than ``usual''
cohomology in many respects, e.g., Poincar\'e duality still holds, there
is a Hard Lefschetz Theorem in the projective case, and intersection
cohomology Betti numbers are combinatorial invariants. With the crucial
role of the Stanley-Reisner ring for cohomology and with its various
interpretations in mind, when looking for a substitute on the
fan-theoretic side that could play a similar role for intersection
cohomology, we were lead to investigate the properties of the
{\it equivariant intersection cohomology\/} $IH\b_\TT(X_\Delta)$. This
graded vector space is endowed with a natural structure of a graded
module over the graded ring $H\b_\TT(X_\Delta)$.
The aim of this article is to find some analogue of the combinatorial
description provided by the above isomorphism~$(**)$ in the case of a
completely arbitrary fan~$\Delta$, replacing equivariant cohomology with
equivariant intersection cohomology and keeping possible generalizations
to the case of non-rational fans in mind. It is convenient to adopt a
sheaf-theoretic point of view: The finite family consisting of all
$\TT$-invariant open subsets of the toric variety~$X$ is a topology
on~$X$, and associating to such a ``toric'' open subset $U \subset X$ the
graded vector space $IH\b_\TT(U)$ yields a presheaf on this topological
space. Since open toric subvarieties of $X = X_{\Delta}$
are in one-to-one correspondence to subfans of the defining fan~$\Delta$,
the ``toric'' topology on~$X_{\Delta}$ corresponds to the ``fan topology''
on the finite set~$\Delta$, namely, the topology given by the collection
of the subfans $\Lambda \preceq \Delta$, together with the empty set, as
family of open sets. For such a subfan~$\Lambda$, we then understand the
graded $A\b$-module $IH\b_{\TT}(X_\Lambda)$ as the module of sections
over~$\Lambda$ of the {\it equivariant intersection cohomology
(pre-)sheaf} $\IH\b_\TT$ on the ``fan space''~$\Delta$.
To be more precise, we show that the presheaf
$$
\IH\b_\TT \colon \Lambda \mapsto \IH\b_\TT(\Lambda) :=
IH\b_\TT(X_\Lambda,\QQ)\;, \quad\hbox{for $\Lambda \preceq \Delta$}\;,
$$
is in fact a sheaf of modules over the {\it sheaf of rational piecewise
polynomial functions on the support of the fan\/}
$$
\A\b \colon \Lambda \mapsto \A\b (\Lambda)
:= \{ f: |\Lambda| \longrightarrow \QQ\;;\;
\hbox{$\Lambda$-piecewise polynomial}\, \}\; ,
$$
and we prove that the sheaf $\IH\b_\TT$ has the three properties stated
in the following definition. \par
We call a sheaf $\E\b$ of $\A\b$-modules a {\it minimal extension
sheaf\/} (of the constant sheaf~$\QQ$) if it satisfies the following
conditions\footnote{For a sheaf~$\F$ on~$\Delta$ and a cone~$\sigma$,
we simply write~$\F(\sigma)$ to denote the space of sections
$\F(\langle\sigma\rangle)$
over the affine subfan $\langle\sigma\rangle$ generated by~$\sigma$.}:
\smallskip
{\def\litem{\par\noindent \hangindent=36pt\ltextindent}
\def\ltextindent#1{\hbox to \hangindent{#1\hss}\ignorespaces}\it
\litem{\rm(N)} {\rm Normalization:\/} For the zero cone~$o := \{0\}$,
there is an isomorphism $\E\b(o) \cong \A\b(o)\; (= \QQ\b)$.
\litem{\rm(PF)} {\rm Pointwise Freeness:\/} For each cone $\sigma \in
\Delta$, the module~$\E\b(\sigma) := \E\b\bigl(\langle \sigma
\rangle\bigr)$ is {\it free} over~$\A\b(\sigma)$.
\litem{\rm(LME)} {\rm Local Minimal Extension $\bmod\ \mm$:\/} For each
cone $\sigma \in \Delta$, the restriction mapping
$$
\varphi_\sigma \colon \E\b(\sigma) \to \E\b(\partial\sigma)
$$
induces an isomorphism
$$
\overline \varphi_\sigma \colon
\frac {\E\b(\sigma)}{\mm \mal \E\b(\sigma)}
\buildrel \cong \over \longrightarrow
\frac {\E\b(\partial\sigma)} {\mm \mal \E\b(\partial\sigma)}
$$
of graded vector spaces.\par}
\smallskip
Restated in other words, a minimal extension sheaf~$\E\b$ on a
fan~$\Delta$ is characterized as follows: It is a sheaf
of graded $\A\b$-modules satisfying the equality $\E\b (o) = \QQ\b$ and
having the property that for each cone $\sigma \in \Delta$, the
module~$\E\b(\sigma)$ is free and of minimal rank over~$\A\b(\sigma)$
such that the restriction $\varphi_\sigma\colon
\E\b(\sigma) \to \E\b(\partial\sigma) $ is surjective.
\par
It is not difficult to see that such a sheaf~$\E\b$ can be constructed
recursively on the $k$-skeletons $\Delta^{\le k}$ of the fan~$\Delta$
in a purely formal way, starting from~$\E^{0}=\QQ$, and that the
resulting sheaf is {\em uniquely determined up to isomorphism}.\par
Using this notion of minimal extension sheaves, the central result of
the article can be stated as follows:
\begin{mtheorem}
The equivariant intersection cohomology sheaf $\IH\b_\TT$ on a fan
space~$\Delta$ is a minimal extension sheaf of~$\QQ\b$.
\end{mtheorem}
The fact that $\IH\b_{\TT}$ is a minimal extension sheaf has interesting
consequences that are discussed in \cite{BBFK$_{2}$}: In particular, it
leads to a simple proof of the inductive ``Local-Global-Formula'' for
the computation of intersection cohomology Betti numbers of compact
toric varieties, but also, {\it mutatis mutandis}, it opens a way to
introduce the analogous sheaf for arbitrary (not necessarily rational)
fans in real vector spaces. On the other hand, the ``minimal complexes''
of \cite{BeLu} occur now naturally as the ``cellular \v{C}ech complexes''
of the sheaf $\IH\b_\TT$.\par
\smallskip
The article is organized as follows: After fixing our (more or less
standard) notation, we first recall in section~1 the definition and some
basic properties of the equivariant cohomology of a toric variety; in
particular, we look at the {\it presheaf\/} $\H\b_{\TT}$ that it
defines on the fan space. For later use, we also touch upon the
equivariant Chern class of toric line bundles. In section~2, we
introduce the equivariant intersection cohomology and prove that it
vanishes in odd degrees, which implies that it actually defines a
{\it sheaf\/} $\IH\b_{\TT}$ on the fan space.\par
The characteristic properties (N), (PF), and (LME) of $\IH\b_{\TT}$,
formalized in the notion of a minimal extension sheaf, are discussed
in section~3. Whereas property~(N) is easily seen to hold, the other
two require considerably more work. Property~(PF) is established in
section~4 by proving that any {\it contractible affine toric variety
is equivariantly formal\/} (with respect to intersection cohomology):
The ``usual'' intersection cohomology $IH\b(X)$ is the quotient of the
equivariant theory $IH\b_\TT(X)$ modulo the homogeneous $A\b$-submodule
$\mm \mal IH\b_\TT(X)$. The constructions used in the proof are also
crucial for proving the remaining property (LME) in section~5. The
article concludes with a discussion of {\it ``equivariantly formal''
fans\/}, i.e., fans~$\Delta$ such that the corresponding toric variety
$X_{\Delta}$ is equivariantly formal. This property holds both for
complete fans and for ``affine fans'' consisting of a full dimensional
cone and its faces; more generally, it holds if and only if the
non-equivariant intersection cohomology vanishes in odd degrees.
\par \smallskip
It is our pleasure to thank Michel Brion, Volker Puppe, and the members
of the ``Algebraic Group Actions'' seminar at Warsaw University led by
Andrzej Bia{\l}ynicki-Birula for stimulating discussions and for useful
comments. We appreciate the pertinent remarks of the referee.
\setcounter{section}{-1}
\typeout{1. Preliminaries}
\section{Preliminaries}
\setcounter{theorem}{0}
\noindent We use the following {\it notation\/}: We denote with $\TT
\cong (\CC^*)^n$ the complex {\it algebraic torus\/} of dimension~$n$,
with $N := \Hom(\CC^*, \TT) \cong \ZZ^n$ the lattice of its one
parameter subgroups, and with $M := \Hom(\TT, \CC^*) \cong \ZZ^n$ the
dual lattice of characters. We recall that there are natural
isomorphism $\TT \cong N \otimes_{\ZZ} \CC^{*} \cong
\Hom_{\ZZ}(M, \CC^{*})$. Let~$\Delta$ be a fan in the {\it rational\/}
vector space $V := N_\QQ := N \otimes_\ZZ \QQ$. For two cones~$\sigma$
and~$\tau$ in~$\Delta$, we write $\tau \prec \sigma$ if~$\tau$ is a
{\it proper\/} face of~$\sigma$, and $\tau \prec_1 \sigma$ if~$\tau$ is
a {\it facet\/} (i.e., a face of codimension~1) of~$\sigma$. -- The
symbol $\Lambda \prec \Delta$ indicates that a proper subset $\Lambda$
of $\Delta$ actually is a {\it subfan.\/}
A cone $\sigma \in \Delta$ {\it generates\/} the {\it ``affine''} subfan
$\langle\sigma\rangle := \{\sigma\} \cup \partial\sigma$ consisting of
$\sigma$ and the {\it boundary subfan\/} $\partial\sigma := \{\tau \in
\Delta\,;\, \tau \prec \sigma\}$. Occasionally, if there is no risk of
confusion, we simply write~$\sigma$ instead of~$\langle\sigma\rangle$
for notational convenience. \par
For a cone $\sigma \in \Delta$, we denote with $V_\sigma := \sigma +
(-\sigma)$ the linear span of $\sigma$ in $V$, and with $\TT_\sigma$,
the subtorus of $\TT$ corresponding to the sublattice $N_\sigma := N
\cap V_\sigma$ as lattice of its one parameter subgroups (i.e., we have
$\TT_\sigma \cong N_{\sigma} \otimes_{\ZZ} \CC^{*}$). The associated
lattice of characters is $M_\sigma := M/N_\sigma^{\perp}$. Let us denote
with $V_{\sigma}^*$ the dual of $V_{\sigma}$ in $V^*:= M_\QQ := M
\otimes_\ZZ \QQ$. The symmetric $\QQ$-algebra $A\b_\sigma :=
S\b(V_{\sigma}^*)$ is naturally isomorphic to the algebra of polynomial
functions $f \colon V_\sigma \to \QQ$. The restriction
$f \mapsto f|_\sigma$ thus provides an isomorphism
$$
A\b_\sigma \buildrel \cong \over \longrightarrow
\A\b(\sigma) := \A\b(\langle\sigma\rangle) \leqno{(0.1)}
$$
with the algebra of sections over~$\langle\sigma\rangle$ of the
sheaf~$\A\b$ of piecewise polynomial functions described in the
introduction. This is compatible with the induced homomorphisms
$A\b_\sigma \to A\b_\tau$ and $\A\b(\sigma) \to \A\b(\tau)$ for each
face $\tau \prec \sigma$. -- In view of the cohomological interpretation
discussed in section~1, we endow the algebra~$A\b_\sigma$ -- just as we
did with $A\b := S\b(V^*)$ -- with the grading that doubles the usual
degrees, so linear polynomials (i.e., elements of $V_\sigma^*$) get the
degree~$2$, etc.
If $F\b$ is a graded $A\b$-module, we denote with $\overline F\b$ its
residue class module
$$
\overline F\b := F\b/(\mm \mal F\b) \,\cong\, \QQ\b
\otimes_{A\b} F\b \qquad
\hbox{(with $\mm := A^{>0}$ and $\QQ\b := A\b/\mm$),} \leqno{(0.2)}
$$
where $\mm$ is the unique homogeneous maximal ideal of $A\b$, and $\QQ\b$
is the graded algebra concentrated in degree zero with~$\QQ^{0} := \QQ$.
-- By means of the natural
epimorphism $A\b \to A\b_\sigma$ extending the projection $V^* \to
V^*_\sigma$ (and corresponding to the restriction of polynomial functions
from $V:=N_{\QQ}$ to $V_{\sigma}$), every $A\b_\sigma$-module
$F\b_\sigma$ also is an $A\b$-module, and there is a canonical
isomorphism $\overline F\b_\sigma \cong F\b_\sigma/(\mm_\sigma \mal
F\b_\sigma)$.\par
In the sequel, we shall freely use the following basic facts on finitely
generated graded $A\b$-modules~$F\b$: Given a family $(f_1, \dots, f_r)$
of homogeneous elements in $F\b$, it generates $F\b$ over $A\b$ if and
only if the system of residue classes $(\overline f_1, \dots, \overline
f_r)$ modulo $\mm \mal F\b$ generates the vector space ${\overline{F}}\b$.
If $F\b$ is free, then $(f_1, \dots, f_r)$ is part of a basis of $F\b$
over $A\b$ if and only if $(\overline f_1, \dots, \overline f_r)$ is
linearly
independent over~$\QQ$. Furthermore, every homomorphism $\varphi \colon
F\b \to G\b$ of graded $A\b$-modules induces a homomorphism $\overline
\varphi \colon {\overline F}\b \to {\overline{G}}\b$ of graded vector
spaces which is surjective if and only if~$\varphi$ is so. If $F\b$ is
free, then every homomorphism $\psi \colon {\overline F}\b \to
{\overline{G}}\b$ can be lifted to a homomorphism $\varphi \colon F\b \to
G\b$; if both $F\b$ and $G\b$ are free, then $\varphi$ is an isomorphism
if and only if that holds for~$\overline \varphi$.\par
For the affine toric variety $X_\sigma$ associated to~$\sigma$,
the torus~$\TT_{\sigma}$ is the {\it isotropy subtorus\/} at any point
in the unique closed orbit $B_{\sigma}$, and this orbit $B_{\sigma}$ is
$\TT$-isomorphic to the quotient torus $\TT/\TT_{\sigma}$, looked at as
a homogeneous space. Any splitting $\TT \cong {\TT\,}' \times \TT_\sigma$
of the torus~$\TT$ into the isotropy subtorus $\TT_\sigma$ and a
complementary subtorus~${\TT\,}' \cong \TT/\TT_{\sigma}$ extends to an
equivariant splitting of affine toric varieties: The choice of an
arbitrary base point~$x_{o}$ in the open dense orbit~$B_{o}$ determines
an embedding of~$\TT$ into~$X_{\sigma}$. Denoting with~$Z_\sigma$ the
closure of~$\TT_\sigma$ in~$X_\sigma$ with respect to this embedding, we
obtain an isomorphism
$$
{\TT\,}' \times Z_\sigma \buildrel\cong\over\longto X_\sigma\,,\;
(t,z) \mapsto tz \,. \leqno{(0.3)}
$$
The unique point $x_\sigma \in B_\sigma \cap Z_\sigma$ will sometimes be
referred to as the {\it distinguished point} in the orbit~$B_\sigma$. We
remark that~$Z_\sigma$, being the $\TT_{\sigma}$-toric variety
corresponding to~$\sigma$ considered as $N_{\sigma}$-cone, is
equivariantly contractible to its fixed point~$x_{\sigma}$, so~$X_\sigma$
has the closed orbit~$B_\sigma$ as equivariant deformation retract. -- We
refer to the above splitting~(0.3) as the {\it affine orbit splitting}.
\par
\goodbreak
\section {Equivariant Cohomology of Toric Varieties}
\setcounter{theorem}{0}
\noindent
Before proceeding to study the equivariant {\it intersection\/}
cohomology, we first look at the ``usual'' $\TT$-equivariant cohomology
$H\b_{\TT}(X)$ of a toric variety $X = X_\Delta$. We recall the
definition: With respect to a fixed identification $\TT \cong
(\CC^{*})^n$, a universal $\TT$-bundle is given by the principal
$(\CC^{*})^n$-bundle
$$
E\TT :=(\CC^\infty \setminus \{0\})^n \longrightarrow
B\TT := (\PP_{\infty})^n \;,
$$
the limit of the finite-dimensional approximations
$$
E_{m}\TT := (\CC^{m+1} \setminus \{0\})^n
\longrightarrow B_{m}\TT := (\PP_m)^n
$$
for $m \to \infty$. One considers the associated bundles
$$
X_\TT := E\TT \times_{\TT} X \longrightarrow B\TT
\qquad\hbox{and}\qquad X_{\TT,m} :=
E_{m}\TT \times_{\TT} X \longrightarrow B_{m}\TT
$$
with fibre~$X$ and defines the {\it (rational) equivariant cohomology
algebra\/} of~$X$ as follows:
$$
H\b_{\TT}(X) := H\b_{\TT}(X, \QQ) := H\b(X_{\TT}, \QQ).
$$
For the homogeneous part of some fixed degree $q \ge 0$, there is the
description
$$
H^q_\TT(X, \QQ) \cong \lim_{m \to \infty} H^q(X_{\TT,m}, \QQ) \,.
$$
The bundle projection $X_\TT \to B\TT$ makes $H\b_{\TT}(X)$ an algebra
over the cohomology algebra $H\b(B\TT) \cong \QQ[u_{1}, \dots, u_{n}]$
of the classifying space $B\TT = (\PP_{\infty})^n$ of~$\TT$, a polynomial
algebra on~$n$ free generators of degree~$2$.\par
To relate $H\b_{\TT}(X)$ with the combinatorial data encoded in the
defining fan~$\Delta$ for~$X$, we first recall the following result.
\begin{lemma} \label{isom1.1}
For each cone $\sigma$, there are natural isomorphisms
$$
\A\b(\sigma) \cong A\b_\sigma \cong H\b(B\TT_{\sigma}) \cong
H\b_\TT(X_{\sigma}) \;,
$$
i.e., they are compatible with maps defined by face relations $\tau
\prec \sigma$.
\end{lemma}
\begin{proof} The isomorphism on the left-hand side has been discussed
in~(0.1). The one in the middle is the special case $V = V_{\sigma}$,
$\TT = \TT_{\sigma}$ of the isomorphism of graded $\QQ$-algebras
$$
A\b = S\b(V^*) \buildrel \cong \over \longto H\b(B\TT)
$$
induced from the {\it Chern class homomorphism}
$$
c\colon M \longrightarrow H^2(B\TT)\,, \quad \chi \mapsto c_1(L_\chi)
$$
that associates to a character $\chi \in M \subset V^*=M_{\QQ}$ the
first Chern class of the line bundle $L_\chi := E\TT \times_{\TT}
\CC_\chi \to E\TT \times_\TT \{{\rm pt}\} = B\TT$. Here $\CC_\chi$
denotes the one-dimensional $\TT$-module with weight $\chi$.
For a cone $\sigma \in \Delta$ and the corresponding subtorus
$\TT_\sigma$ of $\TT$, the restriction mapping $f \mapsto f|_{V_\sigma}$
and the natural mapping $B\TT_{\sigma} \hookrightarrow B\TT$ induce a
commutative diagram
\[
\begin{array}{ccc}
A\b & \longto & A\b_\sigma \\[3pt]
\mimapdown{\cong} & & \mimapdown{\cong}\\[3pt]
H\b(B\TT) & \longto & H\b(B\TT_{\sigma})
\end{array}
\]
since we have $L_{\chi}|_{B\TT_{\sigma}} = L_{\chi|_{\TT_\sigma}}$
(the restriction of $L_\chi$ to $B\TT_{\sigma}$ is just the line bundle
associated to the character $\chi|_{\TT_\sigma}$) and Chern classes
are functorial, thus proving the naturality of the isomorphism in the
middle.
We now discuss the isomorphism $H\b(B\TT_{\sigma}) \cong
H\b_\TT(X_{\sigma})$ on the right hand side: The affine orbit splitting
$(X_{\sigma},\TT) \cong {\TT\,}' \times (Z_\sigma, \TT_{\sigma})$ of
(0.3) and the $\TT_{\sigma}$-equivariant contraction $Z_{\sigma} \simeq
\{x_{\sigma}\}$ onto the distinguished point induce isomorphisms
\begin{eqnarray*}
H\b(B\TT_\sigma) & \cong &
H\b(E\TT_\sigma \times_{\TT_\sigma} \{x_{\sigma}\}) \cong
H\b(E\TT_\sigma
\times_{\TT_\sigma} Z_\sigma) \\
& = & H\b_{\TT_{\sigma}}(Z_\sigma)
\cong H\b_\TT({\TT\,}' \times Z_\sigma) = H\b_\TT(X_\sigma)\;.
\end{eqnarray*}
The whole construction is natural with respect to some face relation
$\tau \prec \sigma$. This is easily seen from any splitting of the torus
in the form $\TT = \TT' \times \TT_\sigma = {\TT\,}' \times ({\TT\,}''
\times \TT_\tau) = ({\TT\,}' \times {\TT\,}'') \times \TT_\tau$, since
the choices of~${\TT\,}'$ and~${\TT\,}''$ do not play any role.
\end{proof}
\typeout{1.A Equivariant Cohomology etc.}
\section*{1.A Equivariant Cohomology as a Presheaf on the Fan}
\noindent On the fan space~$\Delta$ defined in the introduction, we
now consider the {\it presheaf\/} $\H\b_\TT$ of graded algebras that
is given by
$$
\H\b_\TT \colon \Lambda
\mapsto H\b_\TT(X_\Lambda) \quad\hbox{for $\Lambda \preceq
\Delta$}\,.
$$
Inverting the isomorphisms of Lemma~1.1 and using the fact that~$\A\b$
clearly is a sheaf on~$\Delta$, we obtain the following result:
\begin{corollary}
There is a homomorphism of presheaves $\H\b_\TT \to \A\b$
that is an isomorphism on the stalks, so $\A\b$ is the associated
sheaf to the presheaf~$\H\b_\TT$.\qed
\end{corollary}
Note here that the stalk $\F_\sigma$ of a presheaf $\F$ on $\Delta$
coincides with $\F(\sigma)$, since for each point~$\sigma$ of the fan
space, the basic open set~$\langle\sigma\rangle$ is its smallest open
neighbourhood.
In the simplicial case, that homomorphism turns out to be an
isomorphism:
\begin{theorem}\label{isom1.3}
For a simplicial fan $\Delta$, the homomorphism
$$
\H\b_\TT \longto \A\b
$$
of graded presheaves is an isomorphism, so $\H\b_\TT$ is a sheaf (and
actually flabby).
\end{theorem}
\begin{proof}
We proceed by induction on the number of cones in the fan~$\Delta$.
For $\Delta = \{o\}$, the assertion is obvious. For the induction
step, we choose a maximal cone $\sigma \in \Delta$ and consider the
Mayer-Vietoris sequences associated to $\Lambda := \Delta \setminus
\{ \sigma \}$ and $\langle\sigma\rangle$, both for $\H\b_\TT$ and
for $\A\b$. It suffices to prove that in the commutative~diagram
\[
\begin{array}{ccccccccc}
0 &\!\! \longto \!\! & \H^{2q}_\TT(\Delta) & \!\! \longto\!\!
& \H^{2q}_\TT(\Lambda) \oplus \H^{2q}_\TT(\sigma) & \!\! \buildrel
\alpha \over \longto\!\! & \H^{2q}_\TT(\Lambda \cap \sigma)
& \!\!\longto\!\! & \!\! \H^{2q+1}_\TT(\Delta)\\[3pt]
& & \mimapdown{\cong} & &
\mimapdown{\cong} & & \mimapdown{\cong} & & \\[3pt]
0 & \!\!
\longto\!\! & \A^{2q}(\Delta) & \!\! \longto\!\! & \A^{2q}(\Lambda)
\oplus \A^{2q}(\sigma) & \!\! \buildrel \beta \over \longto\!\! &
\A^{2q}(\Lambda \cap \sigma) & \!\!\longto\!\! & 0
\end{array}
\]
obtained from Corollary~1.2, the rows are exact, the vertical arrows
are isomorphisms, and $\H^{2q+1}_{\TT}(\Delta)$ vanishes. Applying
the induction hypothesis to the fans $\Lambda$ and $\Lambda \cap
\langle\sigma\rangle$, and Lemma~1.1 to~$\langle\sigma\rangle$, we see
that the assertion holds for the second and the third arrow,
and we obtain the leading~$0$ in the
upper row. Furthermore, since the fan $\Delta$ is simplicial, it is
not difficult to see that the sheaf~$\A\b$ is flabby;
hence, the map~$\beta$, and thus also~$\alpha$, is surjective. This
proves $\H^{2q+1}_{\TT}(\Delta) = 0$, as, by induction hypothesis, we
know that $\H^{2q+1}_{\TT}(\Lambda) \oplus \H^{2q+1}_{\TT}(\sigma)$
vanishes. By the Five Lemma, the first vertical arrow is an
isomorphism as well, thus proving our~claim.
\end{proof}
In the non-simplicial case, we still have some partial results which
are proved by similar arguments:
\begin{remark} \label{rem1.4} (i) If a fan $\Delta$ can be successively
built up from cones $\sigma_1, \dots, \sigma_r$ such that $\sigma_{i+1}$
intersects $\sigma_1 \cup \dots \cup \sigma_i$ in a single proper face,
then the homomorphism $\H\b_\TT \to \A\b$ is an isomorphism.
(ii) For an arbitrary toric variety, we have isomorphisms of sheaves
$$
\H^q_\TT \cong \A^q \quad\hbox{in all degrees $q \le 2$;}
$$
in particular, $\H^2_\TT$ is a sheaf, and there is an isomorphism
$$
H^2_\TT(X_{\Delta}) \cong \A^2(\Delta)\; .
$$
\end{remark}
On the other hand, this does not carry over to degree~$q=3$, as
follows from the next example:
\begin{example} \label{exam1.5} There is a three-dimensional toric
variety~$X$ with
$
H^3_\TT(X) \cong \QQ\, .
$
\end{example}
Since we know that $\H^3_\TT(\sigma) \cong \A^3(\sigma)$ vanishes on
each basic open set, the associated sheaf is the zero sheaf. Hence the
example implies the following statement:
\begin{corollary} \label{cor1.6} In the non-simplicial case, the presheaf
$\H\b_\TT$ is in general not a sheaf.
\end{corollary}
\begin{prexao}
We consider the fan~$\Delta$ generated by the four ``vertical'' facets
of a cube centred at the origin, and write it in the form $\Delta_{1}
\cup \Delta_{2}$, where~$\Delta_{1}$ is generated by two adjacent
facets, and~$\Delta_{2}$ is generated by the other two. According to
Remark~1.4, i), we have isomorphisms $\H\b_\TT(\Delta_{j}) \cong
\A\b (\Delta_{j})$ and $\H\b_\TT(\Delta_{1} \cap \Delta_{2}) \cong
\A\b (\Delta_{1} \cap \Delta_{2})$. We now show that the homomorphism
$$
\A^2(\Delta_{1}) \oplus \A^2(\Delta_{2}) \longrightarrow
\A^2(\Delta_{1} \cap \Delta_{2})
$$
in the appropriate Mayer-Vietoris sequence is not surjective. In fact,
its cokernel is one-dimen\-sio\-nal: The fan $\Delta_{1} \cap
\Delta_{2}$ is the union $\langle \tau_{1} \rangle \cup \langle
\tau_{2} \rangle$ of the subfans generated by the two opposite
two-dimensional ``vertical'' cones $\tau_{j}$ that are spanned by the
``outer'' vertical edges of the two adjacent facets. The vector space
$\A^2(\Delta_{1} \cap \Delta_{2}) = A^2_{\tau_1} \oplus A^2_{\tau_2}$
is four-dimensional. The restriction homomorphisms from $\A^2(\Delta_{j})$
to $\A^2(\Delta_{1} \cap \Delta_{2})$ both map onto the threee-dimensional
subspace consisting of all piecewise linear functions for which the
differences between the values at the top vertex and at the bottom
vertex on both edges agree. Applying that $\H\b_\TT(\Lambda) =
\A\b(\Lambda)$ holds for $\Lambda = \Delta_{j}$ and for $\Lambda =
\Delta_{1} \cap \Delta_{2}$, we get $H^3_\TT(X_{\Delta}) \cong \QQ$.
\hfill $\Box$
\end{prexao}
\typeout{1.B. Toric Line Bundles etc.}
\section*{1.B. Toric Line Bundles and their Equivariant Chern Class}
\noindent A line bundle $L \to X$ on a toric variety~$X$ is called a
{\it toric line bundle\/} if there is a $\TT$-action by bundle
automorphisms on the total space such that the bundle projection is
equivariant. We obtain an induced line bundle $L_{\TT} := E\TT
\times_{\TT} L$ on $X_\TT$, whose Chern class $c_1(L_{\TT}) \in
H^2(X_\TT) = H^2_\TT(X)$ is called the {\it equivariant Chern class\/}
of~$L$ on~$X$ and is denoted with $c_1^\TT(L)$. This class can be
considered as a lift of the ``usual'' Chern class $c_1(L) \in H^2(X)$
to $H^2_\TT(X) = H^2(X_\TT)$: Each fibre map $X \hookrightarrow
X_\TT$ of the bundle $X_\TT \to B\TT$ induces the same ``edge
homomorphism'' $H^2_\TT(X) \to H^2(X)$, and that edge homomorphism
maps $c_1^\TT(L)$ to $c_1(L)$.
Now let $\Delta$ be a {\em purely $n$-dimensional fan} (i.e., a fan
generated by its $n$-dimensional cones). For use in section~6, we have
to determine the $\Delta$-piecewise linear function $\psi := \psi_{L}
\in \A^2(\Delta)$ corresponding to the equivariant Chern class
$c_1^\TT(L) \in H^2_\TT(X)$ of a toric line bundle~$L$ on~$X$ with
respect to the isomorphism $H^2_\TT(X_\Delta) \to \A^2(\Delta)$. It is
convenient to introduce the notation $\check\TT := \TT \times \CC^*$,
$\check N := N \oplus \ZZ$, and $\check V := V \oplus \QQ$.
The total space of~$L$ is in a natural way a toric variety with acting
torus $\check\TT$, where the second factor acts on the fibres by scalar
multiplication. The associated principal $\CC^*$-bundle $L_0$, obtained
from $L$ by removing the zero section, is an invariant open subset. Let
us first describe the fan $\check\Delta_0$ in $\check V$ corresponding
to~$L_{0}$. The fan is determined by the following properties: The
projection $p \colon \check V \to V$ maps $|\check\Delta_0|$
homeomorphically onto $|\Delta|$, inducing a bijection between the
cones in $\check\Delta_0$ and $\Delta$. In order to assure local
triviality of the projection, we have to require the equality
$$
p(\check N_{\check\sigma}) = N_{p(\check\sigma)}
$$
for each cone $\check\sigma$ in $\check\Delta_0$. Then $L$ is the
$\check\TT$-toric variety given by the fan
$$
\check\Delta := \check\Delta_0 \cup \{ \check\sigma + \rho \;;\;
\check\sigma \in \check\Delta_0 \} \;,\quad \rho :=
\QQ_{\ge 0} \mal (0_{V},1_{\QQ})
$$
spanned by $\check\Delta_0$ and the ``vertical'' ray $\rho$ in $\check V$.
The support $|\check\Delta_0|$ is the graph of a piecewise linear function
$\psi := \psi_{L} \colon |\Delta| \to \QQ$ taking
integral values at lattice points. Vice versa, such a function clearly
determines a fan~$\check\Delta_0$ of the above type. We note
that~$\psi_{L}$ is the composition of the $\Delta$-piecewise linear
homeomorphism $(p|_{\check\Delta_0})^{-1} \colon |\Delta| \to
|\check\Delta_0|$ and the linear projection $\check V \to \QQ$ onto the
second factor.
\begin{remark} \label{rem1.7}
If, as above, the projection $p \colon \check V \to V$ maps
$|\check\Delta_0|$ homeomorphically onto $|\Delta|$, but if the condition
$p(\check N_{\check\sigma}) = N_{p(\check\sigma)}$ is not satisfied, then
$L_0$ is called a {\it toric Seifert bundle\/}. Replacing the lattice
$\check N = N \times \ZZ$ by $N \times \frac{1}{m}\ZZ$ for a suitable
$m\in \NN_{>0}$, we see that the above condition holds for that new
lattice. On the level of tori and toric varieties, that means passing
from $\TT \times \CC^*$ to $\TT \times \CC^*/C_m \cong \TT \times \CC^*$
and from $L_0$ to $L_0/C_m$, respectively, where the group $C_m \subset
\CC^*$ of $m$-th roots of unity acts on $L_0$ as subgroup of the second
factor in $\TT \times \CC^*$.
\end{remark}
\begin{lemma} \label{lem1.8}
For a purely $n$-dimensional fan~$\Delta$, the function $\psi_{L} \in
A^2(\Delta)$ is the image of the equivariant Chern class $c_1^\TT(L)$
of $L$ under the isomorphism $H^2_\TT(X_\Delta) \to \A^2 (\Delta)$
from {\/\rm Remark 1.4, (ii)}.
\end{lemma}
\begin{proof}
For a cone $\sigma \in \Delta^n$, let $\psi_\sigma \in M$ be the
character which coincides with $\psi_L$ on $\sigma \cap N$. Since the
map $\H^2_\TT \to \A^2$ is an isomorphism of sheaves, it suffices to
show that the ``local'' equivariant Chern class $c_1^\TT(L|_{X_\sigma})
\in H^2_\TT(X_\sigma)$ is mapped onto $\psi_\sigma \in M \cong
\A^2(\sigma)$. Observing that the inclusion $x_\sigma \hookrightarrow
X_\sigma$ of the fixed point $x_\sigma \in X_\sigma$ induces an
isomorphism $H^2_\TT(X_\sigma) \buildrel \cong \over \longto
H^2_\TT(x_\sigma)$, we may further restrict our attention
to the fibre $L_{x_\sigma}$ of $L$ over~$x_\sigma$. As a $\TT$-module,
this fibre is nothing but $\CC_{\psi_\sigma}$, and the character
$\psi_\sigma \in M \cong H^2(B\TT) \cong H^2_{\TT}(x_{\sigma})$ is the
Chern class of that bundle. This completes the proof.
\end{proof}
\typeout{2. Equivariant Intersection Cohomology Sheaf}
\section{The Equivariant Intersection Cohomology Sheaf}
\setcounter{theorem}{0}
\noindent For a non-simplicial fan~$\Delta$, the equivariant cohomology
presheaf $\H\b_\TT$ is no longer a sheaf (see Example~1.5), so
Theorem~1.3 fails to be true in the general case. The situation is much
better behaved for {\it intersection\/} cohomology, though we do not
have such a nice combinatorial interpretation as is given by the
Stanley-Reisner ring in the simplicial case. \par
First let us recall how to describe equivariant intersection
cohomology: Following the approach of F.~Kirwan\footnote{See \cite{Ki},
formula (2.12) and the surrounding text; for a more ``sophisticated''
approach, see \cite{Bry} and \cite{Jo}.} and using the notation
of section~1, one defines the $q$-th {\it(rational) equivariant
intersection cohomology group\/} of $X$ as the limit
$$
IH^q_\TT(X) := IH^q_\TT(X; \QQ) :=
\lim_{m \to \infty} IH^q(X_{\TT,m}; \QQ)
$$
and sets
$$
IH\b_\TT(X):= \bigoplus_{q=0}^\infty IH^q_\TT(X)\, .
$$
This construction provides a presheaf
$$
\IH\b_\TT
\colon \Lambda \mapsto \IH\b_\TT (\Lambda)
:= IH\b_\TT(X_\Lambda) \quad\hbox{for $\Lambda \preceq \Delta$}
$$
on the fan space~$\Delta$. In order to prove that it is in fact a
sheaf, we verify the following basic result:
\begin{vanlemma} \label{vanlem}
The equivariant intersection cohomology $IH_\TT^q(X)$ of a toric variety
$X$ vanishes in odd degrees~$q$.
\end{vanlemma}
\begin{proof} Let $\hat\Delta$ be a simplicial refinement of the
defining fan~$\Delta$ for~$X$, and denote with $\hat X \to X$ the
corresponding equivariant $\QQ$-resolution of singularities. By
theorem~1.3, the assertion holds for~$\hat X$, since then
$IH\b_\TT(\hat X) \cong H\b_\TT(\hat X) \cong \A\b_{\hat\Delta}$.
By the equivariant version of the Decomposition Theorem of Beilinson,
Bernstein,
Deligne, and Gabber as stated in \cite[p.~394]{Ki} (see also
\cite[5.3]{BeLu}
or \cite{BreLu}), we may interpret $IH^q_{\TT}(X)$ as a subspace of
$IH^q_{\TT}(\hat
X)$. That proves the assertion. \end{proof}
\begin{theorem} \label{sheaf2.2}
The presheaf $\IH\b_\TT$
on $\Delta$ is a sheaf of $\A\b$-modules.
\end{theorem}
\begin{proof} Since there are only finitely many open subsets
in~$\Delta$,
it suffices to verify the sheaf axioms for two open
subsets~$\Lambda_1,
\Lambda_2 \preceq \Delta$. We thus have to prove the exactness of the
sequence
$$
0 \longto \IH_\TT^q(\Lambda_1 \cup \Lambda_2) \longto
\IH_\TT^q(\Lambda_1)
\oplus \IH_\TT^q(\Lambda_2) \longto \IH_\TT^q(\Lambda_1 \cap
\Lambda_2)
\;.
$$
That follows from the Lemma~2.1: The exactness is obvious
if $q$ is odd; for even~$q$, the vector space $\IH_\TT^{q-1}(\Lambda_1
\cap \Lambda_2)$ vanishes and thus, the sequence is part of
the exact Mayer-Vietoris sequence for~$IH_\TT\b$.
As a consequence, $\IH\b_\TT$ is a sheaf of $\A\b$-modules, since
$\A\b$
is the associated sheaf to the presheaf $\H\b_\TT$, and each
$IH\b_\TT(X_\Lambda)$
is an $H\b_\TT(X_\Lambda)$-module. \end{proof}
\typeout{3. Minimal Extension Sheaves}
\section{Minimal Extension Sheaves}
\setcounter{theorem}{0}
\noindent We now proceed towards (re-)stating and verifying the three
properties that actually characterize the sheaf $\IH\b_\TT$ up to
isomorphism.
For the ease of notation, we write $E\b_\Lambda$ for the module
$\E\b(\Lambda)$ of sections of a sheaf $\E\b$ over a subfan $\Lambda
\preceq \Delta$; in particular, we do so for the residue class module
$\overline{E\b_\Lambda} := E\b_\Lambda/(\mm \mal E\b_\Lambda) \cong
\QQ\b \otimes_{A\b} E\b_\Lambda$, see~(0.2). Using this notation, we
restate the properties (N) ({\it Normalization\/}), (PF) ({\it
Pointwise Freeness\/}), and (LME) ({\it Local Minimal Extension
$\bmod\; \mm$\/}) from the introduction.
\begin{definition} \label{mes}
A sheaf $\E\b$ of graded $\A\b$-modules on the fan $\Delta$ is called
a {\it minimal extension sheaf\/}(of~$\QQ\b$) if it satisfies the
following conditions:
\begin{description}
\item[(N)] One has $E\b_o \cong A\b_o = \QQ\b$ for the zero cone~$o$.
\item[(PF)] For each cone $\sigma \in \Delta$, the module~$E\b_{\sigma}$
is {\it free} over~$A\b_{\sigma}$.
\item[(LME)] For each cone $\sigma \in \Delta$, the
restriction mapping $\varphi_\sigma \colon E\b_{\sigma} \to
E\b _{\partial \sigma}$ induces an isomorphism
$$
\overline \varphi_\sigma \colon \overline{E}\b_{\sigma} \buildrel
\cong \over \longrightarrow \overline{E}\b_{\partial\sigma}
$$
of graded vector spaces.
\end{description}
\end{definition}
Condition~(LME) implies that $\E\b$ is minimal in the set of all flabby
sheaves of graded $\A\b$-modules that satisfy the conditions~(N)
and~(PF), cf.\ Remark~3.3, whence the name ``minimal extension sheaf''.
-- Furthermore, let us note that on a simplicial subfan~$\Lambda$, the
restriction~$\E\b|_{\Lambda}$ of such a sheaf is isomorphic
to~$\A\b|_{\Lambda}$, so it is locally free of rank~one. In the
non-simplicial case, however, the rank of the stalks of $\E\b$ is not
constant, so $\E\b$ can not be a locally free $\A\b$-module.
We now show how to construct such a sheaf on an arbitrary fan space.
\begin{proposition} \label{prop3.1}
On every fan $\Delta$, there exists a minimal extension sheaf $\E\b$,
and it is unique up to isomorphism.
\end{proposition}
\begin{proof} We define the sheaf $\E\b$ inductively on the $k$-skeleton
subfans
$$
\Delta^{\le k} := \bigcup_{j \le k} \Delta^j \;,\quad
\Delta^j := \{\sigma \in \Delta \;;\; \dim\sigma = j\}\;,
$$
starting with $E_o \b := \QQ\b$ for $k=0$. Suppose that for some
$k > 0$, the sheaf~$\E\b$ has already been constructed on
$\Delta^{\le k-1}$, so in particular, for each $k$-dimensional cone
$\sigma$, the module $E\b_{\partial\sigma}$ is given. It thus suffices
to define $E\b_{\sigma}$ together with a restriction homomorphism
$E\b_{\sigma} \to E\b_{\partial \sigma}$ inducing an isomorphism on the
quotients modulo~$\mm$. This is achieved by setting
$$
E\b_{\sigma} := A\b_{\sigma} \otimes_{\QQ} \overline E\b_{\partial
\sigma}
$$
with the restriction map being induced by some $\QQ\b$-linear section
$s: \overline E\b_{\partial\sigma} \to E\b_{\partial \sigma}$ of the
residue
class map $E\b_{\partial \sigma} \to \overline E\b_{\partial
\sigma}$.
The unicity is proved similarly in an inductive manner; we refer to
our companion article \cite{BBFK$_{2}$} for details. \end{proof}
\begin{remark} \label{rem3.2}
A minimal extension sheaf $\E\b$ is flabby and vanishes in odd
degrees.
\end{remark}
\begin{proof} Since the fan space~$\Delta$ is covered by finitely many
affine fans, it suffices to prove that for each cone $\sigma \in
\Delta$, the restriction map $\varphi_\sigma$ is surjective. Using the
results on graded modules recalled in the ``Preliminaries'', that is a
consequence of condition~(LME). -- The vanishing of~$\E^{2q+1}$ follows
immediately from the same condition, since $A\b_\sigma$ and thus $E\b_o
\cong A\b_o = \QQ\b$ ``live'' only in even degrees.
\end{proof}
The properties of minimal extension sheaves are investigated in
\cite{BBFK$_{2}$}; let us quote just one result:
\begin{quote}
{\it The sheaf $\A\b$ is a minimal extension sheaf if and only
if the fan~$\Delta$ is simplicial.}
\end{quote}
Our aim here is to show that $\IH\b_\TT$ represents this unique
isomorphism class of minimal extension sheaves.
\begin{mntheorem} \label{mntheo}
The equivariant intersection cohomology sheaf $\IH\b_\TT$ on $\Delta$
is a minimal extension sheaf.
\end{mntheorem}
\begin{proof} The Normalization property is obviously satisfied, since
we have
$$
\IH\b_\TT(o) = IH\b_\TT(\TT) \cong \QQ\b\, .
$$
The Pointwise Freeness condition will be verified in Corollary~4.5, and
the Local Minimal Extension requirement, in Proposition~5.1.
\end{proof}
\typeout{4. Equivariantly Formal Toric Varieties etc.}
\section{Equivariantly Formal Toric Varieties and Pointwise Freeness
{of~$\IH\b_\TT$}.}
\setcounter{theorem}{0}
\noindent We now proceed to verify the condition
\begin{enumerate}
\item[(PF)] {\it For each cone~$\sigma$, the $A\b_{\sigma}$-module
$IH\b_{\TT}(X_{\sigma})$ is free\/.}
\end{enumerate}
This follows immediately from the isomorphism
$$
IH\b_{\TT}(X_{\sigma}) \cong A\b_{\sigma} \otimes IH\b(Z_{\sigma})
$$
obtained in the proof of Corollary~4.5 below, where $Z_{\sigma}$ is the
contractible affine $\TT_{\sigma}$-toric variety occuring in the orbit
splitting $X_{\sigma} \cong {\TT\,}' \times Z_{\sigma}$ as in~(0.3).
The crucial point is that~$Z_{\sigma}$ satisfies the conditions -- of
course, with respect to the acting torus~$\TT_{\sigma}$ -- stated in
the following result, a more general version of which can be found in
\cite{GoKoMPh}.
\begin{lemma} \label{lem4.1}
For a toric variety $X$, the following statements are equivalent:
\begin{enumerate}
\item[i)] The K\"unneth formula $IH\b_\TT(X) \cong H\b(B\TT) \otimes
IH\b(X)$ holds (which, as above, implies that $IH_\TT\b(X)$ is a
free $A\b$-module).
\item[ii)] With $\overline{IH\b_\TT(X)} :=
IH\b_\TT(X)/\big(\mm \mal IH\b_\TT(X)\big)$, each inclusion $X
\hookrightarrow X_\TT$ as a fibre induces an isomorphism
$$
\overline{IH\b_\TT(X)} \;\;\cong\;\; IH\b(X) \leqno{(4.1)}
$$
of graded vector spaces.
\item[iii)] The non-equivariant intersection cohomology $IH^q(X)$
vanishes in odd degrees~$q$.
\end{enumerate}
\end{lemma}
\begin{proof} Since condition~i) says that for intersection
cohomology, $X_\TT$ behaves like the product $X \times B\TT$, the
implication ``i) $\Rightarrow$ ii)'' is obvious, and ``ii)
$\Rightarrow$ iii)'' follows immediately from the Vanishing
Lemma~2.1. For the implication ``iii)
$\Rightarrow$ i)'', we observe that the assumption implies the
degeneration of the intersection cohomology Leray-Serre spectral
sequence
$$
E_2^{p,q} = H^p(B\TT) \otimes IH^q(X) \Rightarrow IH^{p+q}_\TT(X)
$$
associated to the fibering $X_\TT \to B\TT$ at the $E_{2}$-level:
Since $H^p(B\TT)$ vanishes for odd $p$, the spectral terms $E_2^{p,q}$
vanish for odd total degrees $k = p+q$, and consequently, the
differentials $d_2^{p,q} \colon E_2^{p,q} \to E_2^{p+2,q-1}$ are trivial.
By induction on~$r$, that holds also for every $r \ge 2$.
\end{proof}
Thus, for toric varieties satisfying these properties, the equivariant
and the non-equi\-variant theory determine each other in a simple way.
Following \cite{GoKoMPh}, we use the following terminology:
\begin{definition} \label{equifor}
A toric variety $X$ -- and its defining fan -- are called
{\em equivariantly formal\/} (for intersection cohomology), or
$IH_\TT$-formal for short, if and only if $X$ satisfies one --
and hence all -- of the above three conditions.
\end{definition}
We point out that there is an analogous notion for ``ordinary''
cohomology, and that a toric variety may be $IH_\TT$-formal, but not
$H_\TT$-formal (e.g., a compact toric threefold with $b_3 \ne 0$).
As we are mainly dealing with intersection cohomology, however, we use
``equivariantly formal'' only in the sense of ``$IH_\TT$-formal''.
The most important cases when toric varieties are equivariantly formal
are the compact case and the contractible affine case. From the theorem
of Jurkiewicz and Danilov cited in the introduction, it follows that a
rationally smooth compact toric variety is equivariantly formal. The
proof in the compact case is now an easy consequence of the famous
Decomposition Theorem:
\begin{proposition} \label{compa}
A compact toric variety $X$ is equivariantly formal.
\end{proposition}
\begin{proof} Let $\pi: \hat X \to X$ denote a toric $\QQ$-resolution
as in the proof of the Vanishing Lemma~2.1. By the ``classical'' (i.e,
non-equivariant) Decomposition Theorem of Beilinson, Bernstein,
Deligne and Gabber, we know that $IH\b(X)$ is isomorphic to a direct
summand of $H\b(\hat X)$, and, according to the theorem of
Jur\-kie\-wicz and Danilov, that module vanishes in odd degrees.
\end{proof}
We now state the analogous result for the contractible affine case
where it is considerably more difficult to handle:
\begin{theorem} \label{affine}
A contractible affine toric variety $X$ is equivariantly formal.
\end{theorem}
This result immediately yields the ``Pointwise Freeness''
property~(PF).
\begin{corollary} \label{cor4.4}
For each cone $\sigma \in \Delta$, the $A\b_{\sigma}$-module
$\IH\b_\TT(X_\sigma)$ is free.
\end{corollary}
\begin{procoo}
The affine orbit splitting $X_{\sigma} \cong {\TT\,}' \times Z_\sigma$
as in~(0.3) yields the isomorphism
$$
IH\b_\TT(X_\sigma) \cong IH\b_{{\TT\,}' \times \TT_{\sigma}}({\TT\,}'
\times Z_\sigma) \cong IH\b_{\TT_\sigma}(Z_\sigma) \;.\leqno{(4.2)}
$$
By the theorem, the contractible affine $\TT_{\sigma}$-toric
variety~$Z_{\sigma}$ is $IH\b_{\TT_{\sigma}}$-formal, i.e., we have
isomorphisms
$$
IH\b_{\TT_\sigma}(Z_\sigma) \;\cong \;
H\b (B\TT_{\sigma}) \otimes IH\b(Z_\sigma)
\;=\; A\b_\sigma \otimes IH\b (Z_\sigma)\; ,
$$
showing that $IH\b_{\TT_\sigma}(Z_\sigma)$ is a free
$A\b_\sigma$-module and thus proving the corollary.
\hfill$\Box$
\end{procoo}
\begin{protheo} Writing $X := X_{\sigma}$ for ease of notation, we
first notice that $X$ is a distinguished neighbourhood of its (unique)
fixed point~$x$. According to the attachment condition (see
\cite[V.4.2]{Bo}), we thus have an isomorphism
$$
IH\b(X) \cong \tau_{<n} IH\b(X^*)\, ,\leqno{(4.3)}
$$
where $X^* := X \setminus \{x\}$ is the ``punctured'' toric variety
obtained by removing the fixed point. Hence, it suffices to prove that
$IH^q(X^{*})$ vanishes in odd degrees $q < n$. The basic idea to reach
that aim is to pass from~$X^*$ to a projective toric variety~$Y$
having~$X$ as its ``affine cone'' and then to compare $IH\b(X^*)$
with~$IH\b(Y)$, keeping in mind that $Y$ is equivariantly formal.
Such a projective toric variety~$Y$ is obtained as quotient of~$X^{*}$
modulo the action of any one parameter subgroup of~$\TT$ having the
following property: The orbits of the induced $\CC^*$-action on~$X$
have the fixed point~$x$ as common ``source'', so an equivariant
contraction is provided by $(t,x) \mapsto t \mal x$ for $t \to 0$,
where~$t \in\; ]0,1] \subset \CC^{*}$ is the parameter. A one parameter
subgroup satisfies these conditions if and only if the representing
lattice vector $\alpha \in N = \Hom(\CC^*, \TT)$ lies in the relative
interior of~$\sigma$. If, in addition, the lattice vector~$\alpha$ is
primitive, then the induced $\CC^*$-action on~$X^{*}$ is free, possibly
up to some finite isotropy along lower-dimensional $\TT$-orbits.
Let~$F \subset \alpha(\CC^*)$ be the finite subgroup generated by these
isotropy groups. Then the equivariant mapping $X^{*} \to Y$ factors
through the quotient $X^{*}/F$, making $X^{*}/F \to Y$ a principal
$\CC^{*}$-bundle (cf.\ Remark~1.7). Replacing~$X^{*}$ with $X^{*}/F$
does not change the intersection cohomology: In fact, there is an
identification
$$
IH\b(X^*/F) \cong IH\b(X^*)^F = IH\b(X^*) \leqno{(4.4)}
$$
(see \cite[Lemma 2.12]{Ki$_{2}$}), the equality holding since the action
of $F$ on $IH\b(X^*)$ is trivial: It is induced from a -- necessarily
trivial -- action of the connected group $\TT$ on the rational vector
space $IH^*(X^*)$.
The quotient $Y$ is a toric variety for the torus $\overline \TT :=
\TT / \alpha(\CC^*)$, whose lattice of one parameter subgroups is
$\overline{N} := N/(\ZZ \mal \alpha)$. The defining fan for~$Y$ is
the image
$$
\Phi := p(\partial\sigma) := \{p(\tau) \;;\; \tau \prec \sigma\}
$$
of the boundary fan $\partial\sigma$ under the quotient mapping
$p \colon V \to W$ from $V = N_{\QQ}$ onto the quotient vector space
$W := \overline{N}_{\QQ} := N_{\QQ}/(\QQ \mal \alpha)$.
We now proceed to proving that $\tau_{<n}IH\b(X^*)$ vanishes in odd
degrees, thus showing that $X$ is equivariantly formal. By the
identification (4.4), we may assume without loss of generality
that~$F$ is the trivial subgroup and thus, that $X^* \to Y$ is a
principal $\CC^*$-bundle over the projective toric variety $Y$. We
next consider the toric line bundle $L \to Y$ obtained from $X^*
\to Y$ by adding a zero section opposite to the fixed point of $X$
(``section at infinity'') to that $\CC^*$-bundle: Its total
space~$L$ is the toric variety associated to the fan
$$
\Sigma := \partial\sigma \cup
\{ \tau + \QQ_{\ge 0}\mal(-\alpha)\;;\; \tau \prec \sigma \}
$$
in $V = N_{\QQ}$, with the projection $N \to \overline N$ inducing a
mapping of fans $\Sigma \to \Phi$ and thus, a toric morphism
$L \to Y$. The line bundle~$L \to Y$ is ample (see, e.g.,
\cite[2.12, p.~82]{Oda$_1$}), so a suitable tensor power is very
ample. If~$L$ is already very ample itself, then the one point
compactification of~$L$ is the projective cone over~$Y$ with respect
to the projective embedding determined by the sections of~$L$, the
corresponding complete fan being $\Sigma \cup \{ \sigma \}$.
Let us now look at the following commutative diagram whose top row is
part of the long exact $IH\b$-sequence associated to the pair
$(L,X^*)$:
\[
\begin{array}{ccccccccc}
\dots \longto\!\! & IH^{q-1}(X^*) &
\!\!\longto\!\! & IH^q(L,X^*) & \!\!\longto\!\! & IH^q(L) &
\!\!\longto\!\! & IH^q(X^*) & \!\!\longto \dots\;{} \\[3pt]
& & & \mimapup{\cong} & & \mimapup{\cong} & & & \\[3pt]
& & & IH^{q-2}(Y) & \!\!\buildrel \lambda\over\longto\!\! &
IH^q(Y) & & & \;
\end{array}
\]
Here the first vertical isomorphism is the Thom isomorphism for the
line bundle, the second one is induced by the bundle projection, and
$$
\lambda \colon IH\b(Y)[-2] \to IH\b(Y)
$$
is the homomorphism given by cup product multiplication with
$c_{1}(L) \in H^2(Y)$. The resulting long exact
sequence
$$
\dots \longto IH^{q-1}(X^*) \longto IH^{q-2}(Y)
\buildrel \lambda \over \longto IH^q(Y) \longto IH^q(X^*) \longto
\dots
$$
is the Gysin sequence associated to the bundle $X^{*} \to Y$.
Using the fact that~$Y$ is equivariantly formal and thus has vanishing
intersection cohomology in odd degrees, the Gysin sequence decomposes
into shorter exact sequences
$$
0 \longto IH^{q-1}(X^*) \longto IH^{q-2}(Y)
\buildrel \lambda \over \longto IH^q(Y) \longto IH^q(X^*) \longto 0
$$
if $q$ is even. Hence, it suffices to show that $\lambda$ is injective
for $q \le n$.
Since $L$ is ample, we may apply the hard Lefschetz theorem for
intersection cohomology: a suitable tensor power $L^{\otimes m}$ of
$L$ is very ample, and since $c_1(L^{\otimes m})= m c_1(L)$, the
assertion of the theorem holds for $L$ as well. Hence, we know that
$$
\lambda^k \colon IH^{n-1-k}(Y) \longto IH^{n-1+k}(Y)
$$
is an isomorphism for every $k \ge 0\,$; as a consequence, we see that
$$ \lambda \colon IH^{q-2}(Y) \longto IH^q(Y) $$
is injective for $q \le n$ and surjective for $q \ge n$.
\hfill $\Box$
\end{protheo}
For later use, we note that the properties of~$\lambda$ yield
an isomorphism
$$
\coker\,\lambda \cong \tau_{<n}IH\b(X^*) \,.\leqno{(4.5)}
$$
\typeout{5. Local Minimal Extension Property etc.}
\section{Local Minimal Extension Property of $\IH\b_\TT$.}
\setcounter{theorem}{0}
\noindent To complete the proof of our Main Theorem stating that the
sheaf $\IH\b_{\TT}$ is a minimal extension sheaf, we have to verify
the condition~(LME). We restate it as follows:
\begin{proposition} \label{prop5.1}
For each cone~$\sigma$, the restriction mapping
$$
IH\b_{\TT}(X_{\sigma}) \to IH\b_{\TT}(X_{\partial\sigma})
$$
induces an isomorphism between the quotients modulo the maximal
ideal~$\mm$ of~$A\b$.
\end{proposition}
\begin{proof} By the ``relative affine orbit splitting''
$(X_\sigma, X_{\partial\sigma}) \cong \TT' \times
(Z_\sigma, Z_{\partial\sigma})$ and the formula (4.2) as well as its
analogue $IH\b_{\TT}(X_{\partial\sigma}) \cong
IH\b_{\TT_{\sigma}}(Z_{\partial\sigma})$, we see that it is sufficient
to consider the case of an $n$-dimensional cone~$\sigma$. We use the
same notations as in the previous section; in particular, we write $X
= X_\sigma$ and $X^* = X^*_\sigma$. Furthermore, by the arguments of
the preceding section, we may replace $\TT$, $X_\sigma$, and
$X_{\partial \sigma}$ by $\TT/F$, $X_\sigma/F$, and
$X_{\partial \sigma}/F$, respectively, where $F \subset \TT$ is a
suitable finite subgroup, without changing the base ring $H\b(B\,\TT)
\cong H\b\bigl(B(\TT/F)\bigr)$ and the above homomorphism. Hence we
may assume that $X^* \to Y$ is a principal $\CC^*$-bundle.
First we collect in a big commutative diagram all the objects we have
to consider:
\[
\begin{array}{ccccc}
IH\b_\TT(X) & \longto & IH\b_\TT(X^*) &
{\mathrel{\mathop{\longleftarrow }\limits^{\cong}_{\pi_{\TT}}}}
& IH\b_{\overline \TT}(Y) \\[3pt]
\mimapdown{} & & \mimapdown{} & & \mimapdown{} \\[3pt]
\overline{IH\b_\TT(X)} & \longto & \overline{IH\b_\TT(X^*)} &
\buildrel{\overline\pi_{\TT}}\over \longleftarrow &
\overline {IH\b_{\overline \TT}(Y)} \\[3pt]
\mimapdown{\cong} & & \bigcap & & \mimapdown{\cong} \\[3pt]
IH\b(X) & \longto & IH\b(X^*) &
\buildrel{\pi}\over \longleftarrow & IH\b(Y) \\[3pt]
& \llap{$\scriptstyle\cong$} \searrow & \bigcup & & \mimapdown{} \\[3pt]
& & \tau_{<n}IH\b(X^*) & \buildrel \cong \over \longleftarrow &
\coker\, \lambda {\,.
\end{array}
\]
Here, $\lambda$ again denotes the homomorphism given by the cup product
with the Chern class $c_1(L)$ of the line bundle $L \to Y$ as in the
previous section.
Once having established the diagram, the proof is achieved as soon as
we have identified the quotient
$$
\overline{IH\b_\TT(X^*)} \;:= \; (A\b/\mm) \otimes_{A\b}
IH\b_\TT(X^*) \;\cong\;
{IH\b_\TT(X^*)} \bigm/ {\mm \mal IH\b_\TT(X^*)}
$$
-- this is just the image of the ``edge homomorphism''
$IH\b_\TT(X^*) \to IH\b(X^*)$ relating the equivariant and the
non-equivariant theory -- with $\tau_{<n}IH\b(X^*)$.
To that end, we consider the right hand side of the diagram, carefully
keeping track of the different module structures in the top row:
Whereas $IH\b_\TT(X)$ and $IH\b_\TT(X^*)$ are both modules over $A\b =
S\b(M_\QQ) \cong H\b(B\TT)$, we have to look at
$IH\b_{\overline \TT}(Y)$ as a module over the ring $B\b :=
S\b(\overline M_\QQ) \cong H\b(B\overline\TT)$, where $\overline M :=
\overline N^*$. In particular, the ``overlined'' modules in the
second row are quotients modulo the maximal ideals of the respective
base rings: For~$X$ and~$X^{*}$, this is the ideal $\mm = \mm_{A} :=
A^{>0}$, whereas for~$Y$, we have to consider
$$
\overline{IH\b_{\overline \TT}(Y)}
\;:=\; (B\b/\mm_B) \otimes_{B\b} IH\b_{\overline \TT}(Y) \;\cong\;
{IH\b_{\overline \TT}(Y)} \bigm/
{\mm_{B} \mal IH\b_{\overline \TT}(Y)}
$$
(with $\mm_{B} := B^{>0}$). As the projective $\overline \TT$-toric
variety~$Y$ is equivariantly formal, we may identify this graded
$\QQ\b$-module with $IH\b(Y)$. We may now consider~$B\b$ as a subring
of $A\b$ since $\overline M = \overline N^*$ is canonically isomorphic
to a submodule of $N^*=M$. Hence, writing the horizontal arrow
$\overline\pi_{\TT} \colon \overline {IH\b_{\overline \TT}(Y)} \to
\overline{IH\b_\TT(X^*)}$ as
$$
{IH\b_{\overline \TT}(Y)}
\bigm/ {\mm_{B} \mal IH\b_{\overline\TT}(Y)} \longto
{IH\b_\TT(X^*)}
\bigm/ {\mm_{A} \mal IH\b_\TT(X^*)} \;,
$$
we see that it is an epimorphism, being induced by the horizontal
isomorphism~$\pi_{\TT}$ in the top row. Thus, using the isomorphism in
the bottom row, we are done if we can show that the kernel of~$\pi$
equals the image of the ``hard Lefschetz homomorphism''~$\lambda$.
Before we do that let us give some further remarks on the big diagram:
The isomorphism
$$
IH\b_{\overline \TT}(Y) \;
{\mathrel{\mathop{\longto}\limits^{\cong}_{\pi_{\TT}}}}\;
IH\b_\TT(X^*) \leqno{(5.1)}
$$
in the top row is obtained as follows: The bundle projection
$p \colon X^* \to Y$ induces a compatible family of projections
$$
X^*_{\TT,m} \to Y_{\TT,m} \cong Y_{\overline\TT,m} \times
B_{m}\CC^{*} \to Y_{\overline\TT,m}
$$
between the finite-dimensional approximations of $X_{\TT}$ and
$Y_{\overline\TT}$. As these projections are {\it placid\/} maps,
there is a (unique) induced homomorphism
$$
\pi_{\TT} \colon IH\b_{\overline\TT}(Y) \to IH\b_{\TT}(X^*)
$$
(for a discussion, see \cite[\S4]{GoMPh} or \cite[3.3]{BBFGK}). We now
note that the bundle is locally trivial and that there is a finite
open affine covering of~$Y$ by toric subvarieties~$V_i$ such that the
restricted bundle $U_i := p^{-1}(V_i) \to V_i$ actually is trivial.
By~(4.2), we thus have isomorphisms $IH\b_\TT(U_i) \cong
IH\b_{\overline \TT}(V_i)$; gluing these by a Mayer-Vietoris argument
yields the result. -- We note that, by construction, the
isomorphism~(5.1) is a morphism of $B\b$-modules; it thus induces an
isomorphism of the quotients modulo the homogeneous maximal ideal
$\mm_{B} = B^{>0}$, while $\overline{ IH_\TT\b(X^*)}$ refers to the
bigger base ring $A\b$ and thus is a quotient of
$\overline{IH_{ \overline \TT}\b(Y)}$.
We recall why we obtain the other isomorphisms occuring in the diagram:
By the attachment condition, the restriction mapping $IH\b(X) \to
IH\b(X^*)$ factors through the inclusion of $\tau_{<n}IH\b(X^*)$,
inducing the oblique isomorphism. The two vertical isomorphisms follow
from the fact that both~$X$ and~$Y$ are equivariantly formal. The
lower horizontal isomorphism has been obtained in~(4.5) in the
previous section.
We now continue with the proof of Proposition 5.1. First of all, we
lift the ``hard Lefschetz map'' $\lambda \colon IH\b(Y)[-2] \to
IH\b(Y)$ to a map in equivariant intersection cohomology: We make the
line bundle $L \to Y$ a $\overline \TT$-toric line bundle by choosing
some sublattice~$N_{0}$ complementary to~$\ZZ \mal \alpha$, thus
providing a direct sum decomposition $N = \ZZ \alpha \oplus N_0$. From
section~1.B, we recall that the $\overline\TT$-equivariant
Chern class $c_1^{\overline \TT}(L) \in H^2_{\overline\TT}(Y)$ of~$L$
is a lifting of the ``usual'' Chern class $c_1(L) \in H^2(Y)$. It follows
that the cup product with $c_1^{\overline\TT}(L)$ yields a homomorphism
$\lambda_{\overline \TT} \colon IH\b_{\overline \TT}(Y)[-2] \to
IH\b_{\overline \TT}(Y)$ that is a lifting of the mapping $\lambda
\colon IH\b(Y)[-2] \to IH\b(Y)$ given by the cup product with $c_1(L)$.
We further recall that $\IH\b_{\overline\TT}$ is a sheaf on the defining
fan~$\Phi$ for~$Y$, and it is a module over the ``structure sheaf''
${}_\Phi\A\b$ corresponding to the fan $\Phi$. Using the canonical
isomorphism $H^2_{\overline\TT}(Y) \to \A^2(\Phi)$, we may identify the
equivariant Chern class~$c_1^{\overline \TT}(L) \in
H^2(Y_{\overline \TT})$ with the $\Phi$-piecewise linear function $\psi
:= \psi_{L} \in A^2(\Phi)$ (see Remark~1.4, (ii), and Lemma~1.8). \par
Now the quotient projection $N \to \overline N$ induces an isomorphism
$$
{}_\Phi\A\b(\Phi) \buildrel \cong \over \longto
{}_\Delta\A\b(\partial \sigma)\,,
$$
and the image of $\psi \in \A^2(\Phi)$ coincides with the restriction
to $\partial \sigma$ of the ``global'' linear form $f \in A^2$, the
projection $f \colon N_\QQ = \QQ \mal \alpha \oplus (N_0)_{\QQ} \to
\QQ$ mapping $\alpha$ to $-1$ and having $(N_0)_{\QQ}$ as kernel, cf.\
section~1.B.
On the other hand, we have an isomorphism of polynomial rings $A\b =
B\b[f]$ and thus, the equality $\mm = \mm_A = (\mm_B, f) := \mm_B + A\b
\mal f$ for the homogeneous maximal ideals. The proof of the assertion
that the restriction homomorphism $IH\b_{\TT}(X) \to IH\b_{\TT}(X^*)$
induces an isomorphism on the quotients with respect to the submodules
generated by~$\mm_A$ is now obtained as follows: Under the inverse of the
isomorphism~(5.1), the submodule $\mm_{A} \mal IH\b_\TT(X^*) =
(\mm_{B},f) \mal IH\b_\TT(X^*)$ is mapped onto $(\mm_{B},\psi) \mal
IH\b_{\overline \TT}(Y)$. We thus have an isomorphism of quotients
$$
\frac{IH\b_\TT(X^*)}{\mm_{A} \mal IH\b_\TT(X^*)}
\;\cong\; \frac{IH\b_{\overline \TT}(Y)} {(\mm_{B},\psi) \mal
IH\b_{\overline \TT}(Y)} \;\;.\leqno{(5.2)}
$$
As explained above, the mapping $IH\b_{\overline \TT}(Y)[-2] \to
IH\b_{\overline \TT}(Y)$ given as multiplication by $\psi \in
\A^2(\Phi)$ lifts the ``Hard Lefschetz homomorphism'' $\lambda \colon
IH\b(Y)[-2] \to IH\b(Y)$ to the equivariant theory. Since~$Y$, as a
projective $\overline \TT$-toric variety, is equivariantly formal, we
may eventually rewrite the right hand side of the above isomorphism
(5.2) as follows:
$$
\frac{IH\b_{\overline \TT}(Y)} {(\mm_{B},\psi) \mal
IH\b_{\overline \TT}(Y)} \;\cong\;
\frac{IH\b(Y)}{\lambda\big(IH\b_{\ }(Y)\big)}
= \coker\,\lambda \;.
$$
The proof is now achieved using the isomorphisms
(4.5), (4.3), and the fact that the contractible affine toric
variety~$X$ is $IH\b_{\TT}$-formal.
\end{proof}
\typeout{6. Some Results on Equivariantly Formal Fans}
\section{Some Results on Equivariantly Formal Fans}
\setcounter{theorem}{0}
\noindent
We recall from Lemma 4.1 that in the case of an equivariantly formal
toric variety, the equivariant intersection cohomology determines the
``usual'' intersection cohomology in a straightforward way, namely,
as the quotient by the $A\b$-submodule generated by the homogeneous
maximal ideal $\mm$ of the ring~$A\b$. Since
the equivariant intersection cohomology sheaf $\IH\b_\TT$ is a minimal
extension sheaf, we thus have an isomorphism $IH\b(X_\Delta) \cong
\overline{\E\b(\Delta)} := \E\b(\Delta)/\mm\mal\E\b(\Delta)$, where
as usual~$\E\b$ denotes a minimal extension sheaf on~$\Delta$.
It is convenient to call a (rational) fan~$\Delta$ {\it equivariantly
formal} if the toric variety~$X_\Delta$ has that property. In
Proposition~6.1, ii) below, we shall see that equivariant formality
can be characterized by the freeness of the $A\b$-module
$IH_{\TT}\b(X)$. Using the
isomorphism $IH_{\TT}\b(X_\Delta) \cong \E\b(\Delta)$ together with
the fact that minimal extension sheaves exist on non-rational fans as
well, this allows to introduce the notion of a (``virtually'')
equivariantly formal fan in the non-rational
case, thus eventually leading to a notion of ``virtual'' intersection
cohomology for arbitrary equivariantly formal fans.
So far, we know that complete fans and $n$-dimensional affine fans are
of this type. In order to study further examples, we first collect
some properties pertinent to equivariantly formal fans.
\begin{proposition} \label{prop6.1}
\hangindent=1.8\parindent\hangafter=1
{\rm i)} The $A\b$-module $\IH\b_\TT(\Delta)$ is torsion-free if and
only if we have $\Delta$ is purely $n$-dimensional.
\begin{enumerate}
\item[ii)] The $A\b$-module $\IH\b_\TT(\Delta)$ is free if and only
if $\Delta$ is equivariantly formal.
\item[iii)] An equivariantly formal fan $\Delta$ is purely $n$-dimensional.
\item[iv)] If $\Delta$ has an equivariantly formal subdivision
$\hat\Delta$, then $\Delta$ itself is equivariantly formal.
\end{enumerate}
\end{proposition}
We note explicitly that, as a consequence of ii), a notion of
``virtual'' equivariant formality may be defined even for not
necessarily rational fans in $\RR^n$ via minimal extension sheaves
-- no toric varieties are needed.
\begin{proof} i) ``$\Longleftarrow$'': Since $\IH\b_\TT$ is a sheaf,
we see that we have a natural inclusion
$$
\IH\b_\TT (\Delta) \;\subset\;\bigoplus_{\sigma
\in \Delta^{\max}}\IH\b_\TT(\sigma)
$$
of $A\b$-modules. By Corollary~4.5, each $\IH\b_\TT(\sigma)$ is a
free $A\b_\sigma$-module. For $\dim\sigma=n$, we have $A\b \cong
A\b_\sigma$, so the right hand side is a free $A\b$-module. Moreover,
every submodule of a torsion-free module is again torsion-free.
\noindent ``$\Longrightarrow$'': If $\sigma \in \Delta$ is a maximal
cone of dimension $d<n$, let $\Delta' := \Delta \setminus \{\sigma\}$.
The product of some non-zero polynomial function $h \in A\b_\sigma$
vanishing on $\partial\sigma$ (such a function can be obtained as a
product $h = \prod_{\tau \prec_1 \sigma}\ell_\tau$ of non-zero linear
functions $\ell_\tau \in A^2_\sigma$ with $\ell_\tau|_\tau = 0$) and
of a non-zero section in $\IH_\TT\b(\sigma)$ yields a non-zero section
$f \in \IH\b_\TT(\sigma)$ (recall that $\IH\b_\TT(\sigma)$ is a free
$A\b_\sigma$-module!) that vanishes on $\partial\sigma$. We extend it
trivially outside of $\sigma$ and thus get a non-trivial torsion
element, since it is ``killed'' by every non-zero global linear
function in~$A^2$ vanishing on~$\sigma$.
\noindent ii): ``$\Longleftarrow$'': This is clear by Lemma~4.1, i).
\noindent ``$\Longrightarrow$'': We only sketch the argument, leaving
details for future exposition: Consider the intersection cohomology
version of the {\it Eilenberg-Moore spectral sequence\/} (see, e.g.,
\cite[\S~7.2.1]{MCl}) that computes the intersection cohomology of the
pull back of a bundle. Here we look at the bundle $X_\TT \to B\TT$ and
take as map the inclusion of a one point set $\{\pt\}$ into $B\TT$.
Since $B\TT$ is simply connected, the spectral sequence converges: We
have $E_2^{p,q} \Longrightarrow IH^{p+q}(X)$ with
$$
E_2^{p,q} \cong \Tor^{p,q}_{H\b(B\TT)}\big(H\b(\pt),IH\b_\TT(X)\big)
\cong \Tor^{p,q}_{A\b}\big(\QQ\b,IH\b_\TT(X)\big)\, .
$$
Here $A\b$, $\QQ\b$, and $IH\b_\TT(X)$, respectively, are considered
as differential graded algebras resp.\ modules with trivial
differential, and we can compare with the classical Tor functors of
commutative algebra: If $M\b$, $N\b$ are graded $A\b$-modules, the
corresponding Tor-modules are again in a natural way graded
$A\b$-modules:
$$
\Tor_p^{A\b}(M\b,N\b) =\bigoplus_{q=0}^\infty \Tor^q_p(M\b,N\b)
$$
\noindent such that
$$
\Tor^{p,q}_{A\b}(M\b,N\b) \cong \Tor^q_{-p}(M\b,N\b)\, .
$$
Since $IH\b_\TT(X)$ is a free $A\b$-module, we obtain that
$$
E_2^{0,\bullet} \cong \QQ\b \otimes_{A\b} IH\b_\TT(X) \cong
\overline{IH\b_\TT(X)}
$$
and $E_2^{p,q}=\{0\}$ for $p \not= 0$. So in particular, we have
$E_2^{p,q}=\{0\}$ for $p+q$ odd, and hence also $IH^{p+q}(X)=\{0\}$
in that case.
\noindent iii) This follows immediately from i) and ii), since free
modules are torsion-free.
\noindent iv) According to the Decomposition Theorem of Beilinson,
Bernstein, Deligne, and Gabber, we know that $IH\b(X_\Delta)$ is a
direct summand of $IH\b(X_{\hat \Delta})$, thus inheriting the
property that the ``usual'' intersection cohomology vanishes in odd
degrees.
\end{proof}
The following example shows that the condition $\Delta^{\max} =
\Delta^n$ of i) is not sufficient for equivariant formality:
\begin{example} \label{exam6.2} Let $\Delta$ be a fan consisting of two
equivariantly formal subfans $\Delta_1$ and $\Delta_2$ intersecting
in a single cone $\tau$. If the codimension of $\tau$ is at least~$2$,
then $\Delta$ is not equivariantly formal.
\end{example}
\begin{proof} We intend to prove that the intersection cohomology
Betti number $Ib_{3}$ is non-zero. With $X := X_{\Delta}$ and $X_i :=
X_{\Delta_i}$, we consider the following part of the exact
Mayer-Vietoris sequence:
$$
0 \!\to\! IH^1(X_{\tau}) \!\to\! IH^2(X) \!\to\! IH^2(X_{1}) \oplus
IH^2(X_{2}) \!\to\! IH^2(X_{\tau}) \!\to\! IH^3(X) \!\to\! 0 \;.
$$
The zeroes at both ends are due to the fact that the toric varieties
$X_{1}$ and $X_{2}$ are equivariantly formal.
The ``affine orbit splitting'' (0.3) provides an isomorphism $X_\tau
\cong (\CC^*)^k \times Z_{\tau}$ with $x := \hbox{codim}\,\tau$,
where $Z_{\tau}$, as a contractible affine $\TT_{\tau}$-toric variety,
is known to be equivariantly formal. By the K\"unneth formula, we have
$IH\b(X_\tau) \cong H\b\big((\CC^*)^k\big) \otimes_\QQ IH\b(Z_{\tau})$;
in particular, we get $Ib^1(X_\tau) = k$ and $Ib^2(X_\tau) = \binom{k}{2}
+ Ib^2(Z_{\tau})$. By the results of \cite[\S4]{BBFK}, the Betti number
$Ib^2(X_{\Phi})$ of an arbitrary $d$-dimensional toric variety given by
a non-degenerate fan~$\Lambda$ (i.e., such that $\Lambda$ spans $V$) is
determined by the number $a := \#\Lambda^{(1)}$ of rays, namely, we
have $Ib^2(X_{\Lambda}) = a-d$. Denoting with $a_{i} :=
\#\Delta_{i}^{(1)}$ and $a_{3} := \#\tau^{(1)}$ the respective number
of rays, we clearly have $\#\Delta^{(1)} = a_{1} + a_{2} - a_{3}$.
We thus obtain $Ib^2(X_{i}) = a_{i}-n$, $Ib^2(X) = a_{1}+a_{2}-a_{3}-n$,
and $Ib^2(Z_{\tau}) = a_{3}-(n-k)$. Since the Euler characteristic of
an exact sequence vanishes, we obtain $Ib^3(X) = k(k-1)/2 > 0 $.
\end{proof}
We now state a necessary ``topological'' condition for a fan to be
equivariantly formal, thus obviously providing many examples of toric
varieties not having that property.
\begin{proposition} \label{prop6.3} If $\Delta$ is an equivariantly
formal fan, then the complement $N_{\RR} \setminus |\Delta|$ of the
support $|\Delta|$ is connected.
\end{proposition}
This is a consequence of the following inequality:
\begin{lemma} \label{lem6.4} The intersection cohomology Betti number
$Ib^{2n-1}(X_{\Delta})$ satisfies the inequality
$$
Ib^{2n-1}(X_{\Delta}) \ge b_0(N_{\RR} \setminus |\Delta|)-1 \;.
$$
\end{lemma}
\begin{proof} It is known (see, e.g., \cite[3.5]{KaFi}) that the
natural
homomorphism
$$
IH^{2n-1}(X_{\Delta}) \longto H_1^\cld(X_{\Delta})
$$
is surjective (and even an isomorphism, though we do not need this
stronger result). To investigate the target, let $\tilde\Delta$ be a
completion of the fan. The set of cones $\tilde\Delta \setminus
\Delta$ defines a closed invariant subvariety $\tilde{A}$ of the
compact toric variety $\tilde{X} := X_{\tilde\Delta}$ that has the
same number $b_0(\tilde{A})$ of connected components as $N_{\RR}
\setminus |\Delta|$. Combining that with the isomorphism
$H_1^\cld(X_{\Delta}) \cong H_1(\tilde{X}, \tilde{A})$ and with the
exact sequence
$$
\dots \longto H_1(\tilde{X}, \tilde{A}) \longto H_0(\tilde{A})
\longto H_0(\tilde{X}) \longto H_0(\tilde{X}, \tilde{A}) = 0
$$
eventually yields the following chain of inequalities
$$
Ib^{2n-1}(X_{\Delta}) \ge b^{2n-1}(X_{\Delta}) \ge
b_{1}(X_{\Delta}) = b_{1}(\tilde X, \tilde A) \ge
b_{0}(\tilde A) - 1 = b_{0}(N_{\RR} \setminus |\Delta|)
$$
that proves the assertion.
\end{proof}
Suppose a fan $\Delta$ is obtained from an equivariantly formal fan
$\Delta_0$ by adding some $n$-dimensional cone together with its
faces. Then is natural to ask if $\Delta$ is again equivariantly
formal. The above example~6.2 shows that $|\Delta_0| \cap \sigma$
should not be of too small dimension.
Some partial positive results are given by the following propositions.
\begin{proposition} \label{prop6.5} If an equivariantly formal fan
$\Delta_0$ and an $n$-dimensional cone $\sigma$ intersect in a
single simplicial cone $\tau \in \Delta$ that is a facet of $\sigma$,
then the enlarged fan
$$
\Delta := \Delta_0 \cup \langle \sigma\rangle
$$
is equivariantly formal.
\end{proposition}
\begin{proof} We have to prove that the toric variety $X := X_\Delta$
has vanishing intersection cohomology in each odd degree~$q$ if $X_0
:= X_{\Delta_0}$ has. To that end, we look at the exact sequence
$$
\dots \longto IH^{q-1}(X_0) \longto IH^{q}(X, X_0)
\longto IH^{q}(X) \longto IH^{q}(X_0) = 0
$$
where the final term vanishes since $q$ is odd. It clearly suffices to
prove that $IH^{q}(X, X_0)$ vanishes. We may identify this relative
group with $IH^{q}(X_{\sigma}, X_{\tau})$ by excision. We thus consider
the analogous exact sequence
$$
\dots \longto IH^{q-1}(X_{\tau}) \longto IH^{q}(X_{\sigma}, X_{\tau})
\longto IH^{q}(X_{\sigma}) = 0 \,.
$$
As $\tau$ is simplicial, there is an isomorphism $X_{\tau} \cong
(\CC^{n-1}/F) \times \CC^{*}$, where $F$ is a finite subgroup of
$(\CC^{*})^{n-1}$ acting diagonally on $\CC^{n-1}$. By the K\"unneth
formula, we thus have isomorphisms $IH\b(X_{\tau}) \cong
\QQ\b \otimes_{\QQ} H\b(\CC^{*}) \cong H\b(S^1)$, so $IH^{q}(X_{\tau})$
vanishes for each $q > 1$. For $q \ge 3$, these facts immediately yield
$IH^{q}(X_{\sigma}, X_{\tau}) = 0$. The remaining case $q=1$ follows
from the fact that the restriction mapping $IH^0(X_{\sigma}) \to
IH^0(X_{\tau})$ is an isomorphism. \end{proof}
If the cone $\sigma$ is simplicial, we may even allow that it meets
$\Delta$ in several facets:
\begin{proposition} \label{prop6.6} If an equivariantly formal fan
$\Delta_0$ and an $n$-dimensional simplicial cone $\sigma$ intersect
in a subfan $\Lambda$ generated by facets $\tau_{i} \prec_1 \sigma$
for $i=1, \dots, \ell$, (i.e., each $\tau_{i}$ is a cone of
$\Delta_0$), then the enlarged fan
$$
\Delta := \Delta_0 \cup \langle \sigma \rangle
$$
is equivariantly formal.
\end{proposition}
\begin{proof} As in the proof of Proposition 6.5, we have to show
that $IH\b(X,X_0) \cong IH^q(X_{\sigma}, X_{\Lambda})$ vanishes in
odd degrees. There is an isomorphism
$$
(X_{\sigma}, X_{\Lambda}) \cong \big(\CC^{n-\ell} \times
(\CC^\ell, \CC^\ell \setminus \{ 0\})\big)\bigm/ F \;,
$$
where $F$ is a finite subgroup of $(\CC^{*})^n$ acting diagonally on
$\CC^n$, such that $\TT \cong (\CC^{*})^n/F$. As passing to the
quotient by $F$ does not influence the rational (intersection)
cohomology (see \cite[Thm. II.19.2]{Bre}), we obtain
$$
IH^q(X_{\sigma}, X_{\Lambda}) \cong H^q(X_{\sigma}, X_{\Lambda})
\cong H^q(\CC^\ell, \CC^\ell \setminus \{ 0 \}) =0
\quad\hbox{for}\quad q \not = 0,2\ell \;,
$$
thus proving the assertion.\end{proof}
Certain fans obtained from complete ones by removing a few
$n$-dimensional cones are equivariantly formal. This is implied by
the following result, stated in \cite[Th.~4.2]{Oda$_2$} and
rephrased here in terms of equivariant formality:
\begin{theorem} \label{ishi} {\rm [Ishida]} The subfan supported
by the complement of the star of a single ray in a complete fan is
equivariantly formal.
\end{theorem}
\begin{corollary} \label{cor6.8}
\hangindent=1.8\parindent\hangafter=1
{\rm i)} A purely $n$-dimensional fan with convex support is
equivariantly formal.
\begin{enumerate}
\item[ii)] A purely $n$-dimensional fan with open convex complement is
equivariantly formal. In particular, if $X$ is a complete toric
variety and $x_\sigma$, a fixed point, then $X \setminus \{x_\sigma\}$
is equivariantly formal
\end{enumerate}
\end{corollary}
\begin{proof} We choose a vector $v$ such that either $-v$ lies in the
interior of the support (case~i), or that $v$ lies in the interior of
the complement (case~ii), and complete the given fan~$\Delta$ by adding
the new ray $\rho := \QQ_{\ge0}v$ together with all cones of the form
$\rho + \tau$, where $\tau$ is a cone in the boundary of the support.
The ``old'' fan then is the complement of the star of the ``new'' ray.
\end{proof}
Proposition 6.3 above shows that the completeness hypothesis in the
second part of~ii) cannot be omitted: If~$\Delta$ is a non-complete fan
and $\sigma \in \Delta$, an $n$-dimensional cone, then the support of
the subfan $\Delta \setminus \{\sigma\}$ that is defining $X_\Delta
\setminus \{x_\sigma\}$ has a non-connected complement, since $|\Delta
\setminus \{\sigma\}|$ is obtained from $|\Delta|$ by removing the
interior of~$\sigma$, which is separated by the facets from the
exterior and thus, from the complement of~$|\Delta|$.
\medskip
Using Propostition 6.2 and some homological algebra, we shall prove in
the companion article \cite{BBFK$_{2}$} the ``topological''
characterization of equivariantly formal fans given in the theorem
below. To formulate it, we use the following notation: For a purely
$n$-dimensional fan~$\Delta$, we denote with $\partial \Delta$ the
subfan supported by the topological boundary of~$|\Delta|$ (this is
the subfan generated by those $(n-1)$-dimensional cones that are
contained in only one $n$-dimensional cone). Fix a euclidean norm in the
real vector space $N_\RR$, and denote with~$S$ resp.\ with $\partial S$
the intersection of the unit sphere with the support of the fan
$\Delta$ resp.\ of $\partial \Delta$. Furthermore let $Z := S \cup
c(\partial S)$ be the compact topological space obtained by patching
together $S$ and the real cone $c(\partial S)$ over $\partial S$ along
their respective boundaries.
\begin{theorem} \label{kach} A purely $n$-dimensional fan~$\Delta$ is
equivariantly formal if and only if the following conditions hold:
The pair $(S, \partial S)$ has the
real homology of an $(n-1)$-ball modulo its boundary, and $Z$ is a real
homology manifold outside the vertex of the cone $c(\partial S)$.
The last condition is satisfied e.g. if $S$ is a manifold with
boundary.
\end{theorem}
\typeout{References}
|
\section{Introduction}
Periodic orbits provide the skeleton of the dynamics of classical
Hamiltonian systems.
Generic dynamical systems display a mixed phase space,
consisting of islands of stability residing in chaotic seas.
The periodic orbits are neither grouped in families, like in
integrable systems, nor are they well isolated and all unstable,
as for chaotic systems. One rather finds also stable periodic orbits
surrounded by islands of regular behaviour.
These islands of stability look locally like
an almost integrable system, altogether with
a KAM structure of invariant tori and chains
of periodic orbits, remnants of rational tori
of a supposedly contiguous integrable situation
\cite{Lichtenberg:Lieberman:1992}.
In a semiclassical treatment
of the corresponding quantum system, clusters of
proximate orbits display a collective behaviour.
The bifurcations at the centre
of the island and their semiclassical treatment
have been addressed in a number of recent works,
both for the generic variants
\cite{Ozorio:Hannay:1987,Ozorio:1988,Sieber:1996a,%
Schomerus:Sieber:1997,Sieber:Schomerus:1998}
as well as for classically
non-generic, but semiclassically still relevant
cases \cite{Schomerus:1997a,Schomerus:1998}.
In \cite{Tomsovic:Grinberg:1995} the class of near-integrable
systems has been addressed, and a uniform semiclassical
approximation for the most frequently encountered
broken rational tori
(consisting of a stable and an unstable periodic orbit)
was presented. We will call these tori the `simple' tori.
Interestingly, the semiclassical approximation works reasonably
well even beyond the point where
the stable orbit becomes unstable \cite{Main:Wunner:1999}.
The same configuration of a stable and an unstable orbit is also typical
close (in parameter space) to most types of period-$n$-tupling
bifurcations at the centre of a stability island. For bifurcation number
$n\ge 5$,
for the island-chain scenario with $n=4$,
and also (as a consequence of a more complicated bifurcation scenario)
for $n=3$,
two satellite orbits are expelled from the centre.
At a certain distance from the bifurcation
the satellite orbits can be treated as isolated from
the central orbit;
however,
a collective semiclassical treatment of the two satellites is often still
necessary, and this can be achieved by
using the abovementioned approximation for the simple torus.
Although encountered less frequently, there are situations where
a broken torus consists not of two, but a higher number
of periodic orbits. In the islands of stability
tori of this type appear especially
at larger distance from a bifurcation.
In this work we study the twofold-broken rational torus,
consisting of two stable and two unstable periodic orbits.
It is described by a normal form which is obtained from the normal form
of the simple torus
by including the second harmonic in an angular coordinate.
From the normal form we construct a uniform approximation that can be used to
improve semiclassical trace formulae. Indeed, the relevance of this
configuration is much enhanced in the semiclassical context:
Here one has to consider also `ghost' orbits with complex coordinates
\cite{Kus:Haake:1993b}, and
even a simple broken torus can be affected by nearby ghosts, making a treatment
as a `pre-formed' twofold-broken torus advisable.
This is illustrated in a model system, a periodically
driven angular momentum vector (the kicked top), where we find a
configuration of four period-three orbits
which can be regarded as a twofold-broken rational torus.
A reduction of the error of the trace formula by a factor of about $2-3$
is found even when two of these satellites are ghosts.
An even higher accuracy gain is attained for so-called
`inverse-$\hbar$ spectroscopy'.
\section{Normal forms and uniform approximations}
We restrict the analysis to two-dimensional area-preserving maps.
(The results are also applicable to autonomous Hamiltonian systems with
two degrees of freedom.)
The quantum version of the map is generated by
the unitary Floquet operator $F$ which acts on
the vectors of a Hilbert space, mapping the space onto itself.
The semiclassical trace formula relates the traces $\mbox{tr}\,F^n$
to the classical periodic orbits.
Isolated orbits of primitive period $n_0$ give an additive contribution
\cite{Gutzwiller:1971,Gutzwiller:1990,Tabor:1983,Junker:Leschke:1992}
\begin{equation}
\label{eq:statphaseint}
C=A
\exp\left[\i\frac S\hbar - \i\frac \pi2\mu\right]
\end{equation}
with amplitude
\begin{equation}
A=
n_0
|2-\tr
M|^{-1/2}
\label{eq:amplitudes}
\end{equation}
to all traces $\mbox{tr}\,F^n$ with $n=n_0 r$ and integer repetition number $r$.
Besides the primitive period,
three classical quantities of the ($r$-th return of the) periodic orbit
enter, the action $S$,
the trace of the linearized $n$-step map
$M$, and the Maslov index $\mu$.
The expression (\ref{eq:statphaseint}) is derived by a stationary-phase
approximation and becomes inaccurate when orbits lie close together.
Then a collective treatment of the region $\Omega$
inhabited by the proximate orbits becomes
necessary. This can be achieved by introducing normal forms for a phase
function $\Phi$
and an amplitude function $\Psi$ into the more general expression
\cite{Ozorio:Hannay:1987,Ozorio:1988,Sieber:1996a}
\begin{equation}
\label{eq:pqint}
C_{\Omega}=\frac 1 {2\pi\hbar}
\int_{\Omega}\d\varphi'\;\d I\;
\Psi(\varphi',I)\exp\left[\frac \i \hbar
\Phi(\varphi',I)-\i \frac \pi 2 \nu\right].
\end{equation}
Here $I$, $\varphi$ are canonical polar (or cylinder) coordinates,
and $\nu$ is the Morse index.
The famous Poincar{\'e}--Birkhoff theorem states
that a perturbation of an integrable system causes tori with rational
winding number
to break into chains of alternating stable and unstable periodic points.
However, it does not give a quantitative criterion for the number of
distinct orbits that lie on this chain.
The simple broken torus is described by the normal form
\begin{equation}
S(I,\varphi')=S_0+I\varphi'-a I^2-b \cos\varphi'
\end{equation}
for the generating function $S$, with constants $S_0$, $a$, and $b$. The
corresponding map $(I,\varphi)\to(I',\varphi')$,
implicitly given by
\begin{equation}
\varphi=\frac{\partial S}{\partial I},\quad
I'=\frac{\partial S}{\partial \varphi'},
\end{equation}
is the well-known standard map.
This normal form has been used in \cite{Tomsovic:Grinberg:1995}
to obtain a uniform semiclassical
approximation for the simple broken torus,
smoothly interpolating
between the two non-commuting classical and integrable limits
($\hbar\to 0$ and $b\to 0$, respectively).
A more complete picture can be obtained when one
includes the second
harmonic in the angular variable $\varphi'$ and works with the extended
normal form
\begin{equation}
S(I,\varphi')=S_0+I\varphi'-a I^2 -b\cos\varphi'-c\cos(2\varphi'+2\varphi_0).
\label{eq:nform2}
\end{equation}
The strategy of including higher-order terms in normal forms has been
pursued before, with two different incentives. Firstly, the inclusion
of higher orders is a tool to equip a normal form with a sufficient number
of independent parameters. This can be necessary to account for all
classical properties (stabilities and actions)
of the periodic orbits described by the normal form.
Secondly, the higher orders describe additional periodic orbits,
and hence more complicated configurations than the usual normal forms
\cite{Schomerus:1997a,Schomerus:1998,Sadovskii:Shaw:1995,%
Sadovskii:Delos:1996,Main:Wunner:1997}.
A semiclassical description often succeeds only when all orbits
of an extended normal form are treated collectively.
Presently we aim at the inclusion of additional periodic orbits.
The periodic orbits satisfy the fixed point conditions
\begin{eqnarray}
&&I=\frac{\partial S}{\partial\varphi'}=
I+b\sin\varphi'
+2c\sin(2\varphi'+2\varphi_0),
\\
&&\varphi'= \frac{\partial S}{\partial I}=\varphi'-2a I,
\end{eqnarray}
resulting in $I=0$ and
\begin{eqnarray}
&&
b\sin\varphi' +2c\sin(2\varphi'+2\varphi_0)=0.
\end{eqnarray}
This condition amounts to finding the roots
of a fourth-order polynomial in $\sin\varphi'$.
Depending on the parameters $b$, $c$, and $\varphi_0$ there are
either four real solutions or two real and two complex solutions.
For $|b|<2|c|$ there are always four real solutions.
For $|b|>|c|$ only two real solutions are found.
When $|b/c|$ is fixed in the range $(1,2)$ and $\varphi_0$ is varied
one finds tangent bifurcations with two real solutions on one side
of the bifurcation and four real solutions on the other side.
The real solutions correspond to conventional periodic orbits
while the complex solutions are `ghosts'.
They are of no consequence for the classical dynamics but
can be important for the semiclassical description of the quantum system,
as has been shown in \cite{Kus:Haake:1993b} and as we shall
see once more below.
The joint contribution of the
orbits on the twofold-broken torus
is found by introducing into Eq.\ (\ref{eq:pqint})
the normal form
\begin{equation}
\Phi=S_0-a I^2-b\cos\varphi'-c\cos(2\varphi'+2\varphi_0)
\label{eq:phi}
\end{equation}
[cf.\ Eq.\ (\ref{eq:nform2})] for the phase function and
\begin{equation}
\Psi=1+d\cos(\varphi'+\varphi_1)+e\cos(2\varphi'+2\varphi_0)
\label{eq:psi}
\end{equation}
for the amplitude function.
The
stationary-phase limit of Eq.\ (\ref{eq:pqint}) is a sum
of four additive contributions of form
(\ref{eq:statphaseint}), each representing one orbit.
The four parameters $S_0$, $b$, $c$, $\varphi_0$ are determined
by matching the phases of each contribution to
the actions $S$ of the periodic orbits. The parameters $a$, $d$,
$e$, and $\varphi_1$ are fixed by the stability amplitudes
$A$. This strategy works also when two of the orbits are ghosts:
Phases and amplitudes become then complex,
but are related by complex conjugation,
and the number of real independent parameters remains unchanged.
Eqs.\ (\ref{eq:pqint},\ref{eq:phi},\ref{eq:psi})
represent a uniform approximation of the
joint contribution of the twofold-broken torus.
The integration over $I$ is readily carried out, which leaves us with a
one-dimensional strongly oscillating integral over the
coordinate $\varphi'$. Numerically it is most conveniently evaluated by
the method of steepest descent, for which the integration contour
is deformed into the complex plain.
On the new contour
the integrand decreases
exponentially.
The new contour can also visit stationary points with
complex coordinates, i.\,e., ghost orbits.
\section{Numerical results}
We shall illustrate our findings,
and especially the relevance of ghosts,
for a situation encountered in a
model system, the dynamics
of a periodically driven angular momentum vector ${\bf J}$
(the kicked top \cite{Bibel}).
The components of ${\bf J}$ obey the usual commutation rules
$[J_x,J_y]=iJ_z$ (and cyclic permutations).
The total angular momentum ${\bf J}^2=j(j+1)$ is conserved,
restricting the dynamics to the irreducible representations of the
angular-momentum algebra. The Hilbert space dimension is $2j+1$.
The effective Planck's constant is $1/(j+1/2)$ and the classical limit
is attained for $j\to \infty$.
We work here with a Floquet operator of the explicit form
\begin{eqnarray}
F&=&\exp\left[-\i k_z\frac {J_z^2}{2j+1}-\i p_z J_z\right]
\exp\left[-\i p_y J_y\right] \nonumber
\\
&&\times
\exp\left[-\i k_x\frac {J_x^2}{2j+1}-\i p_x J_x\right]
.
\end{eqnarray}
The dynamics consists of a sequence of linear
rotations by angles $p_i$ alternating with torsions
of strength $k_i$. We hold the $p_i$ fixed
($p_x=0.3$, $p_y=1.0$, $p_z=0.8$) while varying
the control parameter
$k\equiv k_z=10k_x$.
Complete semiclassical spectra of this system throughout the full
transition from integrable ($k=0$) to well-developed chaotic behaviour
($k\approx 10$) have been
presented in \cite{Schomerus:Haake:1997}. The system has also been used
to illustrate
uniform semiclassical approximations for various kinds of bifurcations
\cite{Schomerus:Sieber:1997,Sieber:Schomerus:1998,Schomerus:1997a}.
\begin{figure}
\epsfxsize6cm
\centerline{\rotate[r]{\epsffile{fig1.ps}}}
\caption{
Phase space portrait of the kicked top with $k=4.0$.
The phase space is the unit sphere, parameterized here with the
azimuthal angle $q$ and the $z$-component $p$ (in conventional spherical
coordinates, $q=\phi$ and $p=\cos\theta$).
The circle indicates the central period-one orbit, the
other symbols indicate the positions of the four
period-three satellites (all orbits are unstable).
}
\label{fig:phasespace}
\end{figure}
\begin{figure}
\epsfxsize8cm
\centerline{\epsffile{fig2.ps}}
\caption{
Integration contours in the complex
$\varphi'$-plane on both sides of the
tangent bifurcation in the kicked top at $k=3.7856$.
For each $k$ the normal-form coefficients
of the phase function $\Phi$
have been determined
from the actions of the periodic orbits.
The contours pass through the saddle points of the
normal form (circles) and proceed in the direction
of steepest descent (constant phase of $\Phi$).
}
\label{fig:contours}
\end{figure}
We concentrate on a particular
configuration of period-three orbits which suggests a treatment
as a twofold-broken torus. The configuration comes about in a sequence of three
bifurcations: At $k=1.9715\ldots$ a pair of period-three satellites
is born in close vicinity to
a period-one orbit in the centre of a stability island.
At $k=1.9753$ a period-tripling bifurcation takes place where the stable
satellite collides with the central orbit.
On the other side of the
bifurcation the satellites form a simple broken torus around the centre.
This sequence of
bifurcations can be described by extended normal forms
\cite{Schomerus:1997a,Schomerus:1998}, and the close neighbourhood of the two
bifurcations is not exceptional (see e.\,g.\ \cite{Mao:Delos:1992}).
As the control parameter $k$ is increased
further the satellites move away from the centre. At $k=3.7856$ another pair
of period-three satellites appears in a tangent bifurcation.
For $k<5$ the
four satellites form a configuration that resembles a twofold-broken torus.
Figure \ref{fig:phasespace} displays a phase space portrait for $k=4$, and
Figure \ref{fig:contours} shows steepest-descent contours in the
complex $\varphi'$-plane on both sides of the final bifurcation.
\begin{figure}
\epsfxsize8cm
\centerline{\epsffile{fig3.ps}}
\caption{
Semiclassical error of ${\mbox tr}\, F^3$ for the kicked top
($k=3.0$) as
a function of $j$ in the two
approximations i), ii) which are explained in
the text.
}
\label{fig:trace}
\end{figure}
\begin{figure}
\epsfxsize8cm
\centerline{\epsffile{fig4.ps}}
\caption{
Collective peak of the four satellites and the central orbit in the
action spectrum $|T(S)|^2$ for $k=3.0$. The semiclassical approximations
i) and ii) (see text) are compared to
the result of an exact quantum-mechanical computation.
}
\label{fig:actionspectrum}
\end{figure}
We evaluated $\mbox{tr}\,F^3$ at $k=3.0$ with the quantum number $j$
ranging from $1$ to $50$.
At the given value of the control parameter only one of the pairs of
satellites mentioned above has real
coordinates, and its
distance to the centre of the stability island is already quite large.
The other two satellites are still ghosts,
but their bifurcation is not far away.
This leaves us with the choice between two semiclassical approximations:
i) We can group the two real satellites together with the central orbit
and treat the ghosts separately, or
ii) we can group the four satellites as a twofold-broken torus
and treat the central orbit separately.
(In principle, we could complicate matters even more and
group all the orbits together, but this is rather impractical and
beyond the scope of the present work.)
The semiclassical evaluation of the trace involves also five other orbits.
The error of the two approximations is shown in
Figure \ref{fig:trace}. The error of approximation ii) is about a
factor $2-3$ smaller than that of approximation i). Although the
accuracy gain is not dramatic, this result favours clearly
a treatment of the satellites as a pre-formed twofold-broken torus.
A somewhat more demanding test is aided by `inverse-$\hbar$ spectroscopy'
\cite{Tomsovic:Grinberg:1995,Wintgen}.
We consider
the discrete truncated Fourier analysis
\begin{equation}
T(S)=\frac{1}{32}\sum_{j=1}^{32}\mbox{tr}\,F^3(j)\exp[-\i (j+\case 12)S]
\end{equation}
of the trace with respect to the quantum number $j$.
The `action spectrum' $|T(S)|^2$ displays peaks at the actions of the
periodic orbits that contribute to $\mbox{tr}\, F^3$.
Since accidental action degeneracies do not occur in the present example,
the quality of the semiclassical approximations i) and ii) can now be assessed
directly, without interference of the remaining orbits.
Figure \ref{fig:actionspectrum} shows the collective peak of the four satellites
and the
central orbit. Approximation ii) agrees almost perfectly with the exact result,
while approximation i) overestimates the peak-height distinctively.
Of course, the uniform approximation presented here is not restricted to
the situation where two of the orbits are ghosts,
but is valid for four real orbits on the broken torus as
well. Indeed we find an improvement in the semiclassical
accuracy of $\mbox{tr}\,F^3$ over the full range $2.8<k<5$.
\section{Conclusion}
In this paper a uniform approximation for a broken rational torus
consisting of four periodic orbits has been presented.
The approximation was tested in a model system where the phase-space is mixed,
and the tori are grouped around (but not too close to) a central orbit.
It can be expected that the approximation will be even more
useful in studies of globally near-integrable systems.
\ack
This work was supported by the DFG (Sonderforschungsbereich 237) and the
European Community
(Program for the Training and Mobility of Researchers).
\section*{References}
|
\section{INTRODUCTION
\label{sec:axp1}}
Anomalous X-ray pulsars (AXPs)
are a sub-class of X-ray pulsars distinguished by
pulse periods lying in a narrow range between
$6\,{\rm s}$ and $12\,{\rm s}$,
low X-ray luminosities,
soft X-ray spectra,
no detected optical counterparts,
no detected
orbital Doppler shifts of pulse arrival times,
and associations with shell-type supernova remnants
(\markcite{mer95b}Mereghetti \& Stella 1995;
\markcite{van95}van Paradijs, Taam \& van den Heuvel 1995).
The nature of AXPs is uncertain; possibilities include
an accreting neutron star in a binary, with a low-mass
white dwarf or He-burning star as a companion
(\markcite{mer95b}Mereghetti \& Stella 1995;
\markcite{bay96}Baykal \& Swank 1996;
\markcite{mer98}Mereghetti, Israel \& Stella 1998),
an isolated neutron star accreting from a residual disk
following a phase of common-envelope evolution
(\markcite{van95}van Paradijs et al.\ 1995;
\markcite{gho97}Ghosh, Angelini \& White 1997),
a non-accreting massive white dwarf
(\markcite{mor88}Morini et al.\ 1988;
\markcite{pac90}Paczy\'{n}ski 1990;
\markcite{uso94}Usov 1994),
and a non-accreting magnetar
(\markcite{tho96}Thompson \& Duncan 1996;
\markcite{hey98}Heyl \& Hernquist 1998;
\markcite{kou98}Kouveliotou et al.\ 1998).
The two AXPs with
well-sampled timing histories,
1E 1048.1$-$5937 and 1E 2259$+$586,
are observed to spin down irregularly:
the rotation frequency $\Omega$ of each object decreases
linearly with time $t$ on average, but there are `bumps'
superimposed on this average trend
during which $\dot{\Omega}=d\Omega / dt < 0$
fluctuates by a factor of between two and five
every five to ten years
(\markcite{mer95a}Mereghetti 1995;
\markcite{bay96}Baykal \& Swank 1996;
\markcite{bay98}Baykal et al.\ 1998;
\markcite{oos98}Oosterbroek et al.\ 1998;
and references therein).
In accreting-star models of AXPs,
the average spin-down trend is
attributed to the accretion torque acting on a
near-equilibrium rotator,
and the bumps are the result of white torque
noise (e.g.\ due to disk inhomogeneities)
as in ordinary binary X-ray pulsars
(\markcite{bay96}Baykal \& Swank 1996).
In isolated-star models, the spin-down trend
is attributed to magnetic-dipole braking,
and the bumps are analogous to glitches
observed in rotation-powered pulsars like Vela
(\markcite{uso94}Usov 1994;
\markcite{hey98}Heyl \& Hernquist 1998).
In this Letter, we present unequivocal new evidence
that AXPs are non-accreting magnetars.
In \S\ref{sec:axp2},
it is shown that as a magnetar spins down it wobbles
anharmonically, with a period of five to ten years,
due to an effect called `radiative precession'.
The spin-down signature of the wobble,
calculated theoretically in \S\ref{sec:axp3},
matches closely the bumpy timing
histories of 1E 1048.1$-$5937 and 1E 2259$+$586.
The theory is used to predict $\Omega(t)$
over the next 20 years for both objects
and yields a statistical relation
between bump recurrence time
and average spin-down rate for the AXP population
as a whole.
Implications for the internal and magnetospheric
structures of magnetars, including AXPs and
soft gamma-ray repeaters (SGRs),
are explored in \S\ref{sec:axp4}.
\section{RADIATIVE PRECESSION
\label{sec:axp2}}
A magnetar is a triaxial body in general.
Hydromagnetic stresses arising from non-radial gradients
of the superstrong internal magnetic field,
e.g.\ between the magnetic poles and equator if the field
is dipolar,
deform the star, inducing matter-density perturbations
$\delta\rho\sim B_{\rm in}^2/\mu_0 c_{\rm s}^2$
and hence a fractional difference
$\epsilon\sim \delta\rho R^5 / I_1
\approx 2\times 10^{-9} (B_{\rm in}/10^{10}\,{\rm T})^2$
between the principal moments of inertia
(\markcite{gol70}Goldreich 1970;
\markcite{dec80}de Campli 1980;
\markcite{mel98}Melatos 1998).
Here,
$B_{\rm in}$ is the characteristic magnitude of the
internal magnetic field, $c_{\rm s}$ is the isothermal
sound speed ($c_{\rm s}\approx 3^{-1/2}c$),
and $R$ is the stellar radius;
one has $B_{\rm in}\approx B_0$ if the internal
field is confined to the stellar crust
and $B_{\rm in}\gtrsim B_0$ if it
is generated deep inside the star,
e.g.\ in a convective dynamo
(\markcite{tho93}Thompson \& Duncan 1993).
In a rotation-powered pulsar with
$B_0\lesssim 10^9\,{\rm T}$,
the hydromagnetic deformation is comparable to the
elastic deformation arising from shear stresses in the
crystalline stellar crust,
and the principal axes of inertia are oriented
arbitrarily with respect to the axis ${\bf m}$
of the external magnetic dipole
(\markcite{gol70}Goldreich 1970;
\markcite{dec80}de Campli 1980;
\markcite{mel98}Melatos 1998).
In a magnetar, however, the hydromagnetic deformation is
much larger, and ${\bf m}$ is approximately parallel to
one of the principal axes (${\bf e}_3$, say);
the alignment is not exact
due to the complicated structure
of the internal field near its generation site,
cf.\ the non-axisymmetric, magnetically modified Taylor
columns seen in simulations of the geodynamo
(\markcite{gla96}Glatzmaier \& Roberts 1996).
Provided that the rotation axis ${\bf \Omega}$ is
not parallel to ${\bf m}\approx {\bf e}_3$,
as is usually the case for rotation-powered pulsars,
the star precesses about ${\bf e}_3$
with period
$\tau_{\rm pr}
\sim 2\pi/\epsilon\Omega
\approx 85(B_{\rm in}/10^{10}\,{\rm T})^{-2}
(\Omega/1\,{\rm rad\,s^{-1}})^{-1}\,{\rm yr}$.
The Eulerian precession is not free. It couples
to a component of the vacuum magnetic-dipole
torque associated with the asymmetric inertia
of the near-zone radiation fields
outside the magnetar;
the electromagnetic energy density (and hence inertia)
of the near-zone radiation fields is greater at the
magnetic poles than at the equator by an amount
$\sim B_0^2/\mu_0$,
resulting in an oscillatory, precession-inducing
torque
(\markcite{gol70}Goldreich 1970;
\markcite{goo85}Good \& Ng 1985;
\markcite{mel98}Melatos 1998).
The near-field torque exceeds the familiar
braking torque ($\propto\Omega^3$)
by a factor $c/\Omega R\gg 1$ and acts on a
commensurately shorter time-scale,
$\tau_{\rm nf}\sim\tau_0 \Omega R/c
\approx
6 (B_0/10^{10}\,{\rm T})^{-2}
(\Omega/1\,{\rm rad\,s^{-1}})^{-1}\,{\rm yr}$,
where
$\tau_0=\mu_0 c^3 I_1/2\pi B_0^2 R^6 \Omega^2
\approx
2\times 10^5 (B_0/10^{10}\,{\rm T})^{-2}
(\Omega/1\,{\rm rad\,s^{-1}})^{-2}{\rm yr}$
is the characteristic electromagnetic braking time.
Since the near-field torque is directed along
${\bf \Omega}\times {\bf m}$, it does not
change $\Omega$ for a spherical star.
Nor does it change $\Omega$ for an aspherical star,
provided $\epsilon$ is large enough
to give $\tau_{\rm pr}\ll\tau_{\rm nf}$,
so that the near-field torque averages to zero over
one precession period.
When the dominant deformation is hydromagnetic,
however,
one finds
$\tau_{\rm pr}/\tau_{\rm nf}
\approx
14 (B_0/B_{\rm in})^2$,
close to unity (i.e.\ strong coupling)
provided $B_{\rm in}$
is moderately larger than $B_0$ as expected
(\markcite{tho93}Thompson \& Duncan 1993).
Under these circumstances,
the star executes an anharmonic wobble,
called radiative precession,
with period $\tau_{\rm pr}$ ($\approx\tau_{nf}$),
and the near-field torque changes $\Omega$ on
the precession time-scale
(\markcite{mel98}Melatos 1998).
\section{THEORY OF BUMPY SPIN-DOWN
\label{sec:axp3}}
\subsection{Solution of Euler's Equations of Motion:
Past and Future $\Omega (t)$
\label{sec:axp3a}}
We now show that the timing signature of radiative
precession matches the observed bumpy spin-down of AXPs
by solving numerically Euler's equations of motion for a
biaxial, dipole magnet rotating {\em in vacuo},
\begin{eqnarray}
\dot{\Omega}_1
& = &
-\epsilon\Omega_2\Omega_3
+ \Omega_0^{-2}\tau_0^{-1}\cos\chi
[ a\Omega^2(-\Omega_1\cos\chi+\Omega_3\sin\chi) +
b\Omega_2(\Omega_1\sin\chi+\Omega_3\cos\chi) ],
\label{eq:axp1}
\\
\dot{\Omega}_2
& = &
\epsilon\Omega_1\Omega_3
+ \Omega_0^{-2}\tau_0^{-1}
[ -a\Omega^2\Omega_2 +
b(-\Omega_1\cos\chi+\Omega_3\sin\chi)
(\Omega_1\sin\chi+\Omega_3\cos\chi) ],
\label{eq:axp2}
\\
\dot{\Omega}_3
& = &
- \Omega_0^{-2}\tau_0^{-1}\sin\chi
[ a\Omega^2(-\Omega_1\cos\chi+\Omega_3\sin\chi) +
b\Omega_2(\Omega_1\sin\chi+\Omega_3\cos\chi) ].
\label{eq:axp3}
\end{eqnarray}
In (\ref{eq:axp1})--(\ref{eq:axp3}),
subscripts denote components along the
principal axes of inertia,
$\chi$ is the angle between ${\bf m}$ and ${\bf e}_3$,
and we have $a=0.33$, $b=0.094c/\Omega_0 R$,
and $\Omega_0=\Omega(t=t_0)$,
where $t_0$ is an arbitrary origin;
for derivations of the equations, see
\markcite{gol70}Goldreich (1970) and
\markcite{mel98}Melatos (1998).
Terms in (\ref{eq:axp1})--(\ref{eq:axp3})
proportional to $\epsilon$ give rise to Eulerian precession,
terms proportional to $b\tau_0^{-1}$ are associated with
the near-field torque,
and terms proportional to $a\tau_0^{-1}$ cause
secular braking.
A biaxial, dipole magnet rotating {\em in vacuo}
is an idealized model of a hydromagnetically
deformed magnetar.
In reality, such a body is triaxial
(if it is indeed rigid),
has high-order and/or off-centered multipoles contributing
to the near-zone magnetic field,
and is surrounded by a plasma magnetosphere.
The values of $a$ and $b$ reflect,
in a coarse way,
the magnetization state of the stellar interior
and the distribution of magnetospheric currents
(\markcite{mel98}Melatos 1998).
In Figures \ref{fig:axp1} and \ref{fig:axp2},
we plot the computed $\Omega(t)$ from
(\ref{eq:axp1})--(\ref{eq:axp3})
on top of the X-ray timing histories of
1E 1048.1$-$5937 and 1E 2259$+$586 respectively,
extending the theoretical curves 20 years beyond the
present as a testable prediction.
The timing history of 1E 2259$+$586 is sufficiently
well-sampled that only one good fit is possible.
In the case of 1E 1048.1$-$5937, the solid curve is the
favored fit, with $\tau_{\rm pr}\approx 18\,{\rm yr}$,
but an alternative fit, with similar average slope and
$\tau_{\rm pr}\approx 6\,{\rm yr}$,
is also acceptable.
Multiple good fits are hard to find.
Euler's equations of motion contain just three unknown
parameters: $\epsilon$, $\tau_0$, and $\chi$.
It is significant that the theory agrees with
the observations as well as it does
despite its idealized nature ---
particularly as only two of the three parameters
are truly free, since one needs strong torque
coupling
($\tau_{\rm pr}\sim\tau_{\rm nf}$,
or equivalently
$\epsilon\Omega_0\tau_0\sim c/\Omega_0 R$)
in order to get any bumps at all.
Moreover, the best agreement with observations is
achieved for parameter values that are consistent
with general physical considerations.
If AXPs are hydromagnetically deformed magnetars
with
$B_{\rm in}\gtrsim B_0\gtrsim 3\times 10^{10}\,{\rm T}$,
one expects
$\epsilon\sim 10^{-8}$,
$\Omega_0\tau_0\sim 10^{11}$,
and $\chi$ relatively small
(cf.\ $\chi\approx 11^\circ$ for the Earth),
as discussed above.
The theoretical curves do not pass exactly through every
available data point, and more departures are
expected in the future,
e.g.\ the slope between the last two data points
in Figure \ref{fig:axp1} is $\approx 0.6$ times
the solid-curve theoretical slope at that epoch.
Indeed, a formal estimate of the
chi-square for the solid curve in Figure \ref{fig:axp1},
taking published timing uncertainties at face value,
implies that the fit is poor:
one finds a chi-square of $\approx 8\times 10^3$ with 10 degrees
of freedom, notably inferior
to multiple-glitch models, for example
(\markcite{hey98}Heyl \& Hernquist 1998).
This is because a biaxial, dipole magnet is an
over-idealized model of an AXP, as noted above;
the chi-square likelihood improves dramatically for
a more realistic model with just two extra parameters,
e.g.\ a non-dipolar near-zone magnetic field
and a triaxial ellipsoid of inertia.
However, our aim in this paper is not to model
$\Omega(t)$ in detail, but to account for key
gross features of the data ---
the average spin-down rate,
bump recurrence time, and bump amplitude ---
with a simple physical theory.
In this regard, the agreement with observations
is good.
One data point in the timing history of 1E 2259$+$586,
at $t-t_0=15.4\,{\rm yr}$,
represents a spin-up event
lasting at most $0.8\,{\rm yr}$
at the $1\sigma$ level of uncertainty
(\markcite{bay96}Baykal \& Swank 1996).
Spin-up cannot be explained
by radiative precession
because (i) $\dot{\Omega}$ is always negative,
even while
$|\dot{\Omega}|$ decreases by up to a factor of five
during a bump,
and (ii)
there is no natural way to explain a $0.8\,{\rm yr}$
time-scale.
If taken at face value, the spin-up event must
be a different phenomenon,
e.g.\ a discontinuous change of internal structure
analogous to the glitches of rotation-powered
pulsars like Vela
(\markcite{uso94}Usov 1994;
\markcite{hey98}Heyl \& Hernquist 1998).
\subsection{Predicted Population Statistics
\label{sec:axp3b}}
Numerical studies show that
the bump recurrence time and average spin-down rate
satisfy
$\tau_{\rm pr}\propto
B_{\rm in}^{-2}\Omega^{-1} f(\chi)$
and
$\langle\dot{\Omega}\rangle\propto
B_0^2\Omega^3 g(\chi)$,
with $1\leq f(\chi), \, g(\chi) \leq 10$.
In other words,
the greater the magnetic field of an AXP,
the shorter is its precession period and the greater
is its spin-down rate, with
\begin{equation}
\langle\dot{\Omega}\rangle \approx
-2\times 10^{-4} (B_0/B_{\rm in})^2
(\Omega/1\,{\rm rad\,s^{-1}})^2
(\tau_{\rm pr}/ 1\,{\rm yr})^{-1}
\,{\rm rad\,s^{-1}\,yr^{-1}}.
\label{eq:axp4}
\end{equation}
In addition,
one finds that $\dot{\Omega}$ increases
from
$\dot{\Omega}\approx \langle\dot{\Omega}\rangle$ to
$\dot{\Omega}\approx 0$
over a time
$\approx 0.25\tau_{\rm pr}$
during the course of a bump,
yielding a bump amplitude
\begin{equation}
\Delta\Omega_{\rm pr}/\Omega \approx
5\times 10^{-5} (B_0/B_{\rm in})^2
(\Omega/1\,{\rm rad\,s^{-1}})~;
\label{eq:axp5}
\end{equation}
this number is similar for all AXPs.
Both the relations (\ref{eq:axp4}) and (\ref{eq:axp5})
will be subject to observational testing
as the population of AXPs with measured timing histories
swells over time.
Note that they are statistical relations,
with scatter expected about an overall trend.
The detailed structure of the magnetic field
inside an AXP --- which does not enter into the
idealized theory presented here, except through $\chi$ ---
is likely to differ from object to object,
affecting
$\tau_{\rm pr}$, $\langle\dot{\Omega}\rangle$,
and $\Delta\Omega_{\rm pr}$ significantly,
as the broad ranges of $f(\chi)$ and $g(\chi)$ attest.
\section{DISCUSSION
\label{sec:axp4}}
Why is radiative precession not observed in
rotation-powered pulsars with $B_0\lesssim 10^9\,{\rm T}$?
These objects are deformed hydromagnetically like
magnetars, with an added elastic deformation,
and they spin down electromagnetically.
Young pulsars like the Crab, with
$\tau_0\approx 10^3\,{\rm yr}$ and
$\Omega R/c\approx 10^{-2}$,
ought to precess with period
$\tau_{\rm pr}\approx 10\,{\rm yr}$,
yet there is no clean evidence of bumpy spin-down
in radio timing data, nor of concomitant
changes in pulse profile (e.g.\ relative height and
separation of conal components) and polarization
characteristics (e.g.\ position-angle swing).
\markcite{lyn88}Lyne, Pritchard \& Smith (1988) reported
a quasi-sinusoidal variation
in Crab timing residuals with a period of $\approx 20$
months but judged it likely to be an artifact
of an overlooked glitch.
(See also \markcite{mel98}Melatos 1998.)
The only reliable detection of pulsar precession
to date has been the general-relativistic geodetic
precession of the binary pulsar PSR B1913$+$16
(\markcite{wei89}Weisberg, Romani \& Taylor 1989;
\markcite{kra98}Kramer 1998).
One possible explanation is that radiative precession
is damped in pulsars with $B_0\lesssim 10^9\,{\rm T}$.
Frictional dissipation inside the star, due to
time-dependent elastic strains in the crust
and/or imperfect crust-core coupling, is thought to
occur on a time-scale $\lesssim 1\,{\rm yr}$
(\markcite{lin93}Link, Epstein \& Baym 1993),
rapidly aligning ${\bf \Omega}$ with ${\bf e}_3$.
(In the case of the Earth, dissipation restricts
${\bf \Omega}$ to fluctuating within just
$1''$ of ${\bf e}_3$ under the
action of solar and lunar tides;
see Bur\v{s}a \& P\v{e}\v{c} 1993).
In a magnetar, where the magnetic energy exceeds
the mechanical energy of rotation,
the stiffening effect of the superstrong magnetic field
may hinder the elastic strains and/or sheared fluid flows
responsible for internal damping.
A second possible explanation is that conduction
currents in the magnetosphere of a pulsar
with $B_0\lesssim 10^9\,{\rm T}$,
where electron-positron pair production is plentiful,
modify the electromagnetic torque in such a way as
to suppress the precession-inducing near-field
component.
In a magnetar, where it is thought that pair production
is quenched, e.g.\ by positronium formation
(\markcite{uso96}Usov \& Melrose 1996;
J.\ Heyl 1999, private communication)
or photon splitting
(\markcite{bar98}Baring \& Harding 1998),
the vacuum magnetic-dipole torque,
with its unmodified near-field component,
is a closer approximation to reality.
Both explanations imply that radiative precession will
not be suppressed in SGRs, where one has
$B_0\sim 10^{11}\,{\rm T}$
(\markcite{kou98}Kouveliotou et al.\ 1998).
What does radiative precession teach us about strongly
magnetized neutron stars themselves?
Firstly, the close agreement
between theory and observation implies that AXPs are indeed
non-accreting magnetars, and that accretion is not needed
to explain fluctuations in $\dot{\Omega}$, contrary to
claims in the literature
(e.g.\ \markcite{mer95a}Mereghetti 1995).
Secondly, if SGRs are magnetars too,
as indicated by recent X-ray timing data
(\markcite{kou98}Kouveliotou et al.\ 1998),
they ought to exhibit bumpy spin-down just like AXPs,
perhaps punctuated by brief, glitch-like spin-up events
during the gamma-ray bursts themselves.
This constitutes a direct test of the magnetar model for SGRs.
Thirdly, the fact that bumpy spin-down is seen in AXPs
implies that $B_{\rm in}$ is at most a few times $B_0$
to ensure strong coupling,
i.e.\ $\tau_{\rm pr}\sim \tau_{\rm nf}$,
as discussed above.
This is indirect evidence that the magnetic field of
these objects is generated relatively near the stellar
surface.
Fourthly, radiative precession
places an upper limit on the fluid viscosity $\eta$
of a newly born magnetar; if $\eta$ is too high,
${\bf \Omega}$ aligns with ${\bf e}_3\approx{\bf m}$
before the stellar crust
crystallizes, and there is no precession
(\markcite{mel98}Melatos 1998).
The upper limit obtained in this way ---
that the viscous damping time
is greater than the crust crystallization time of
$\sim 1\,{\rm yr}$ ---
validates certain semi-quantitative calculations of $\eta$ for
pulsars with $B_0\lesssim 10^9\,{\rm T}$ by
\markcite{cut87}Cutler \& Lindblom (1987).
We remark in closing that the predicted hydromagnetic
deformation ($\epsilon\sim 10^{-8}$)
constitutes a misaligned mass quadrupole
which generates gravitational radiation.
However, the radiation is too weak to be detected by
planned gravitational-wave interferometers like LIGO
and VIRGO, because AXPs are slow rotators and the
gravitational-wave amplitude is a strongly rising
function of $\Omega$.
\acknowledgments
I am indebted to Lars Bildsten for bringing the existence
of spin-down irregularities in AXPs to my attention,
and for subsequent discussions.
I also thank Jon Arons, Peter Goldreich, and the referee,
Jeremy Heyl, for comments.
This work was supported by NASA Grant NAG5--3073,
NSF Grant AST--95--28271,
and by the Miller Institute for Basic Research in Science
through a Miller Fellowship.
|
\section{Introduction}
It is widely accepted that the formation of stars occurs through
gravitational collapse of
the interstellar gas clouds. One of the possible mechanisms of
triggering the star and/or stellar cluster formation process is
a high-velocity (supersonic) cloud-cloud collisions.
In a such collision a dense gaseous slab is formed
with two plane-parallel
shock fronts propagating away from the interface (Usami et al.
1995). Then the slab grows in mass becoming unstable
against gravitational instability which causes its fragmentation.
Newly formed the gaseous clumps, in turn, collapse further
and evolve into stars and/or stellar clusters.
The structure of a self-gravitating, isothermal shock pressure-bounded
slab of gas which usually is formed in the cloud-cloud
collisions was investigated by Ledoux (1951) and later by
Elmegreen \& Elmegreen (1978) among others. Lubow \& Pringle (1993),
LP93 thereafter, have
also derived an analytic dispersion relation for the linear
perturbations and investigated, in detail, the full unstable branch
in such a slab of gas.
Possible existence of WIMP matter, one of the
form of dark matter which itself is a dominant mass component
of the universe, is strongly motivated both by
standard models of particle physics and cosmology.
It has been established that the mass density associated
with the luminous matter (stars, hydrogen clouds, x-ray gas in
clusters, etc.)
cannot account for the observed dynamics on galactic
scales and above (Trimble, 1987), therefore, revealing the existence
of large amounts of dark matter in this universe.
It has been argued that the dark matter could be
anything from novel weakly interacting elementary
particles (massive neutrinos, neutralinos) to baryonic matter
in some invisible form (brown dwarfs, primordial black holes,
cold molecular gas, etc.).
Recently, it has been shown (Tsiklauri 1998) that
classical Jeans theory of the stability of self-gravitating
interstellar gas clouds is significantly modified by
taking into account the presence of background
weakly interacting massive particles.
It has been found (Tsiklauri 1998) that the presence of
WIMP matter yields an unavoidable reduction of
the Jeans length, the Jeans mass and the Jeans time (time-scale of
the collapse via gravitational instability).
Resuming aforesaid, and in the light of recent results of Tsiklauri
(1998),
a revision of a conventional theory (e.g. LP93) of the stability of
self-gravitating, isothermal shock pressure-bounded
slab of gas against linear perturbations in the
case of presence of WIMP matter, or generally speaking any type of
{\it microscopic} non-baryonic dark matter that is coupled
with the slab only
via gravitational interaction, seems to be of a considerable
importance for full understanding of the star and/or stellar
cluster formation process.
\section{The model}
Equations that govern dynamics of two self-gravitating
fluids (a slab of gas and WIMP matter) inter-coupled only
via gravitational interaction can be written in the following
way:
$$
{{\partial \rho_i}\over{\partial t}} + \nabla (\rho_i \vec V_i)=0,
\eqno(1)
$$
$$
{{\partial \vec V_i}\over{\partial t}} +
(\vec V_i \cdot \nabla) \vec V_i = - \nabla \phi -
{{\nabla P_i}\over{\rho_i}}, \eqno(2)
$$
$$
P_i = K_i \rho_i^{\gamma_i}, \eqno(3)
$$
$$
\Delta \phi = 4 \pi G \sum_i \rho_i, \eqno(4)
$$
where notation is standard: $\vec V_i$, $P_i$, $\rho_i$ and $\phi$
denote velocity, pressure, mass density and gravitational potential
of the fluids. The subscript $i=G,W$ denotes two components,
baryonic gas and WIMP matter respectively.
Now, writing every physical quantity, for brevity commonly
denoted by $\vec f(x,z)$, in a form of
$\vec f(x,z)= \vec f^0 + \vec f^\prime(z) \exp[{i( \omega t -kx)}]$
and doing usual linearization of the Eqs.(1)--(4)
we can obtain following a set of coupled
differential equations for the perturbations
$$
{{d^2 P_G^\prime}\over{dz^2}}+
\left[{{{\omega^2}\over{C_{sG}^2}} -k^2+
{{4 \pi G \rho_G^0}\over{C_{sG}^2}}}\right]P_G^\prime=
-{{4 \pi G \rho_W^0}\over{C_{sW}^2}} P_W^\prime, \eqno(5)
$$
$$
{{d^2 P_W^\prime}\over{dz^2}}+
\left[{{{\omega^2}\over{C_{sW}^2}} -k^2+
{{4 \pi G \rho_W^0}\over{C_{sW}^2}}}\right]P_W^\prime=
-{{4 \pi G \rho_G^0}\over{C_{sG}^2}} P_G^\prime, \eqno(6)
$$
$$
{{d^2 \phi^\prime}\over{dz^2}}-k^2 \phi^\prime=
4 \pi G \left({{{P_G^\prime}\over{C_{sG}^2}}+
{{P_W^\prime}\over{C_{sW}^2}}}\right). \eqno(7)
$$
Here, $C_{sG}$ and $C_{sW}$ denote speeds of sound for the
baryonic gas and WIMP matter respectively.
It is worthwhile to note that Eqs.(5) and (6) resemble closely
to the equations describing coupled pendulums
which arise in the non-modal study of linear perturbations
in shear flows (cf. Chagelishvili, Rogava \& Tsiklauri 1996, 1997).
The fundamental (normal) oscillatory modes of the Eqs.(5) and (6) are
given by the following expression:
$$
q_\pm^2= {{1}\over{2}}
\left[{q_G^2 +q_W^2 \pm
\sqrt{(q_G^2-q_W^2)^2+4 \alpha_G \alpha_W}}\right], \eqno(8)
$$
where
$$
q_G^2=\left[{{{\omega^2}\over{C_{sG}^2}} -k^2+
{{4 \pi G \rho_G^0}\over{C_{sG}^2}}}\right], \eqno(9)
$$
and
$$
q_W^2=\left[{{{\omega^2}\over{C_{sW}^2}} -k^2+
{{4 \pi G \rho_W^0}\over{C_{sW}^2}}}\right], \eqno(10)
$$
are the eigenfrequencies of the system. Also,
$\alpha_G=4 \pi G \rho_G^0/C_{sG}^2$
and $\alpha_W=4 \pi G \rho_W^0/C_{sW}^2$.
A general solution of the Eqs.(5)--(7) can be readily
found. Following LP93 we consider only symmetric modes
$dP_G^\prime/dz=dP_W^\prime/dz=0$ at $z=0$.
There are two distinct, general solutions of the Eqs.(5)--(7)
depending on the sign of $q_-^2$.
For the $q_-^2>0$ we obtain
$$
P_G^\prime = A_- \cos(q_-z)+A_+ \cos(q_+z), \eqno(11)
$$
$$
P_W^\prime = {{q_-^2-q_G^2}\over{\alpha_W}}A_- \cos(q_-z)+
{{q_+^2-q_G^2}\over{\alpha_W}}A_+ \cos(q_+z), \eqno(12)
$$
$$
\phi^\prime=C \cosh(kz)-
\left({\beta_G+{{q_-^2-q_G^2}\over{\alpha_W}} \beta_W}\right)
{{A_-}\over{k^2+q_-^2}} \cos(q_-z)-
$$
$$
\left({\beta_G+{{q_+^2-q_G^2}\over{\alpha_W}} \beta_W}\right)
{{A_+}\over{k^2+q_+^2}} \cos(q_+z), \eqno(13)
$$
here we introduced notation $\beta_G=4 \pi G / C_{sG}^2$
and $\beta_W=4 \pi G / C_{sW}^2$.
For the $q_-^2<0$ we obtain
$$
P_G^\prime = A_- \cosh(\sqrt{-q_-^2}z)+A_+ \cos(q_+z), \eqno(14)
$$
$$
P_W^\prime = {{-q_-^2-q_G^2}\over{\alpha_W}}A_-
\cosh(\sqrt{-q_-^2}z)+ {{q_+^2-q_G^2}\over{\alpha_W}}A_+
\cos(q_+z), \eqno(15)
$$
$$
\phi^\prime=C \cosh(kz)-
\left({\beta_G+{{-q_-^2-q_G^2}\over{\alpha_W}} \beta_W}\right)
{{A_-}\over{k^2-q_-^2}} \cosh(\sqrt{-q_-^2}z)-
$$
$$
\left({\beta_G+{{q_+^2-q_G^2}\over{\alpha_W}} \beta_W}\right)
{{A_+}\over{k^2+q_+^2}} \cos(q_+z), \eqno(16)
$$
Note that in the case of presence of the background WIMP matter
there are two fundamental (normal) modes $q_\pm$ (see Eq.(8)).
Whereas, in the case with no WIMP matter there is only one oscillatory
mode, namely $q_G$ (see Eq.(32) in LP93).
This is the presence of WIMP matter that causes
appearance of a novel unstable mode described by Eqs.(14)--(16).
As our next step towards derivation of the dispersion relation
for linear perturbations, following LP93, we impose relevant boundary
conditions. The first boundary condition is that the Lagrangian
pressure perturbation is zero at the slab boundary, namely that
$$
P_i^\prime + \xi_z {{dP_i^0}\over{dz}}=0, \eqno(17)
$$
at $z=a$ (with $a$ being slab half-thickness),
where $\vec \xi$ vector is the Lagrangian
displacement vector of the perturbation and $i=W,G$. In this case,
the displacement is simply related to $\vec V^\prime$ by
$\vec \xi = - i \vec V^\prime/ \omega$.
Using relation $\nabla P_i^0= - \rho_i^0 \nabla \phi^0=
-4 \pi G a \rho_i^0(\rho_G^0+ \rho_W^0)$ for the
unperturbed physical quantities, Eq. (17), at $z=a$,
reduces to the two following conditions
$$
{{P_i^\prime}\over{\rho_i^0}}= \xi_z 4 \pi G
(\rho_G^0+ \rho_W^0) a, \eqno(18)
$$
The other boundary condition is the one for
$\phi^\prime$ (LP93):
$$
{{d \phi_e^\prime}\over{dz}} - {{d \phi^\prime}\over{dz}}=
4 \pi G (\rho_G^0+ \rho_W^0) \xi_z,
$$
evaluated at $z=a$, where $\phi_e^\prime$ is the
perturbed gravitational potential in the region
$|z|>a$ and is given by
$\phi_e^\prime=\phi^\prime(a) \exp[-k(|z|-a)]$,
where it has been assumed that $k>0$.
Thus, we obtain
$$
{{d \phi^\prime}\over{dz}} \biggl|_{z=a}= -k \phi^\prime(a)
- 4 \pi G (\rho_G^0+ \rho_W^0) \xi_z. \eqno(19)
$$
\subsection{The case when $q_-^2 > 0$}
Applying boundary condistions (18) and (19) to
the general solutions in the case when $q_-^2 > 0$
[Eqs.(11)--(13)] yields
$$
{{A_+}\over{C}}=-[{\cosh(ka)+\sinh(ka)}] \times
\Biggl[{{D_+q_+}\over{k}} \sin(q_+a) -\chi {{D_-q_-}\over{k}}
\sin(q_-a)+
$$
$$
\left\{{{{1}\over{\rho_G^0ak}}-D_+}\right\} \cos(q_+a)-
\chi \left\{{{{1}\over{\rho_G^0ak}}-D_-}\right\} \cos(q_-a)
\Biggr]^{-1}. \eqno(20)
$$
Here we have introduced following notation:
$$
D_-=\left({\beta_G+{{q_-^2-q_G^2}\over{\alpha_W}} \beta_W}\right)
{{1}\over{k^2+q_-^2}},
$$
$$
D_+= \left({\beta_G+{{q_+^2-q_G^2}\over{\alpha_W}} \beta_W}\right)
{{1}\over{k^2+q_+^2}},
$$
and
$$
\chi = {{\cos(q_+a)}\over{\cos(q_-a)}}
{{[1- \delta (q_+^2 -q_G^2)/ \alpha_W]}
\over{[1- \delta (q_-^2 -q_G^2)/ \alpha_W]}}; \;\;\;
\delta \equiv {{\rho_G^0}\over{\rho_W^0}}.
$$
Another condition which relates $A_+$ and $C$
can be obtained using $z$-component of the Eq.(2) for
the perturbations in baryonic gas and
$\vec \xi = - i \vec V^\prime/ \omega$ (LP93).
Doing this yields
$$
{{A_+}\over{C}}= \rho_G^0 ak \sinh(ka)
\Biggl[
{{\omega^2 \cos(q_+a)}\over{4 \pi G(\rho_G^0+ \rho_W^0)}} -
{{\chi \omega^2 \cos(q_-a)}\over{4 \pi G(\rho_G^0+ \rho_W^0)}} -
$$
$$
\chi q_-a(1-\rho_G^0D_-) \sin(q_-a)
+ q_+a(1-\rho_G^0D_+) \sin(q_+a)
\Biggr]^{-1}. \eqno(21)
$$
Now, combining Eqs.(20) and (21) allows us to obtain
the dispersion relation for the perturbations for
the case when $q_-^2 > 0$
$$
{{\sinh(ka)+\cosh(ka)}\over{\sinh(ka)}}=-
{{b_1^+ \cos(q_+a)+b_1^- \cos(q_-a)+b_2^+ \sin(q_+a) +b_2^-
\sin(q_-a)} \over{b_3^+ \cos(q_+a)+b_3^-
\cos(q_-a)+b_4^+ \sin(q_+a) +b_4^- \sin(q_-a)}}.
\eqno(22)
$$
where
$b_1^+=1-D_+ \rho_G^0 ak$,
$b_1^-=- \chi (1 - D_- \rho_G^0 ak)$,
$b_2^+=D_+ \rho_G^0 q_+ a$,
$b_2^-= - \chi D_- \rho_G^0 q_-a$,
$b_3^+= \omega^2 /[4 \pi G (\rho_G^0+ \rho_W^0)]$,
$b_3^- = - \chi \omega^2 / [4 \pi G (\rho_G^0+ \rho_W^0)]$,
$b_4^+=q_+a(1-\rho_G^0D_+)$,
$b_4^-=-\chi q_-a (1-\rho_G^0D_-)$.
\subsection{The case when $q_-^2 < 0$}
We now apply boundary conditions (18) and (19) to
the general solutions in the case when $q_-^2 < 0$
[Eqs.(14)--(16)]. Thus we obtain
$$
{{A_+}\over{C}}=-[{\cosh(ka)+\sinh(ka)}] \times
\Biggl[{{D_+q_+}\over{k}} \sin(q_+a) +\chi^u
{{D_-^u\sqrt{-q_-^2}}\over{k}} \sinh(\sqrt{-q_-^2}a)+
$$
$$
\left\{{{{1}\over{\rho_G^0ak}}-D_+}\right\} \cos(q_+a)-
\chi^u \left\{{{{1}\over{\rho_G^0ak}}-D_-^u}\right\}
\cosh(\sqrt{-q_-^2}a) \Biggr]^{-1}. \eqno(23)
$$
Here we have introduced following notation:
$$
D_-^u=\left({\beta_G+{{-q_-^2-q_G^2}\over{\alpha_W}}
\beta_W}\right) {{1}\over{k^2-q_-^2}},
$$
and
$$
\chi^u = {{\cos(q_+a)}\over{\cosh(\sqrt{-q_-^2}a)}}
{{[1- \delta (q_+^2 -q_G^2)/ \alpha_W]}
\over{[1- \delta (-q_-^2 -q_G^2)/ \alpha_W]}}
$$
Again using the condition which relates $A_+$ and $C$
that is obtained using $z$-component of the Eq.(2) for
the perturbations in baryonic gas and
$\vec \xi = - i \vec V^\prime/ \omega$, we get
$$
{{A_+}\over{C}}= \rho_G^0 ak \sinh(ka)
\Biggl[
{{\omega^2 \cos(q_+a)}\over{4 \pi G(\rho_G^0+ \rho_W^0)}} -
{{\chi^u \omega^2 \cosh(\sqrt{-q_-^2}a)}
\over{4 \pi G(\rho_G^0+ \rho_W^0)}} +
$$
$$
\chi^u \sqrt{-q_-^2}a(1-\rho_G^0D_-^u) \sinh(\sqrt{-q_-^2}a)
+ q_+a(1-\rho_G^0D_+) \sin(q_+a)
\Biggr]^{-1}. \eqno(24)
$$
Finally, using Eqs.(23) and (24) we to obtain
the dispersion relation for the perturbations in
the case when $q_-^2 < 0$
$$
{{\sinh(ka)+\cosh(ka)}\over{\sinh(ka)}}=-
{{b_1^+ \cos(q_+a)+b_{1u}^- \cosh(\sqrt{-q_-^2}a)+b_2^+
\sin(q_+a) + b_{2u}^- \sinh(\sqrt{-q_-^2}a)}
\over{b_3^+ \cos(q_+a)+b_{3u}^- \cosh(\sqrt{-q_-^2}a)+
b_4^+ \sin(q_+a) +b_{4u}^- \sinh(\sqrt{-q_-^2}a)}}.
\eqno(25)
$$
where
$b_{1u}^-=- \chi^u (1 - D_-^u \rho_G^0 ak)$,
$b_{2u}^-= \chi^u D_-^u \rho_G^0 \sqrt{-q_-^2}a$,
$b_{3u}^- = - \chi^u \omega^2 / [4 \pi G (\rho_G^0+ \rho_W^0)]$,
$b_{4u}^-= \chi^u \sqrt{-q_-^2}a (1-\rho_G^0D_-^u)$.
\section{Discussion}
In this paper we have derived two dispersions relations for
linear perturbations in a pressure-bounded, self-gravitating
gas slab which is gravitationally coupled with the
background weakly interacting massive particles.
In the original paper LP93 authors found that
the slab alone is unstable at sufficiently long wavelength and
the growth rate is of the order of $\sqrt{G\rho}$, with $\rho$
being mass density of the slab, but at high external pressures
the nature of the instability is quite different from the
classical Jeans instability. Moreover, they have developed an
analytic model which reproduces numerical results of
Elmegreen \& Elmegreen (1978). Interesting result of these authors
is that the critical wavenumber for the onset of the instability
is always of the order of the slab thickness, regardless of the
level of self-gravity and the external pressure.
Present analysis revealed that taking into account presence
of the putative WIMP dark matter significantly modifies
results of LP93 with major changes being appearance of
two distinct fundamental (normal) oscillatory modes $q_{\pm}$
and the existence of an additional unstable mode when
$q_-^2 < 0$. Obtained dispersion relations Eqs.(22) and (25)
need detailed investigation in various limiting cases.
Especially interesting would be doing analysis of the
low-frequency modes, for which LP93 found an
unexpected result that the critical wavenumber for the onset
of the instability is always of the order of the slab thickness.
\section{Acknowledgements}
I would like to thank Dr. Steven Lubow of Space Telescope Science
Institute (Baltimore) for useful comments on his previous
published work.
|
\section{Introduction}
Theoretical prediction of two inertial ranges, consequence of
both energy and enstrophy
concervation laws by the two- dimensional Euler equations, was and still
is
one of the most remarkable achievements of statistical hydrodynamics
[1]. A direct
and most important outcome of these conservation laws is the fact that if
a fluid is stirred by a random (or non-random) forcing, acting on a scale
$l_{f}=1/k_{f}$, the produced energy is spent on creation of the large-scale
($l>l_{f}$)
flow which cannot be dissipated in the limit of the large Reynlds number
$\nu\rightarrow 0$. This means that the dissipation terms are irrelevant
in the inverse cascade range. Since the dissipation contributions are one of
the most difficult obstacles on the road toward turbulence theory
(see below), one can
hope that in two dimensions the situation is greatly simplified. This hope is
supported by recent numerical and physical experiments showing that as long as
the integral scale $L_{i}\propto t^{\frac{3}{2}}$ is much smaller than the
size of the system, the velocity field at the scales $L_{i}>>l>>l_{f}$ is a
stationary close-to-gaussian process characterized by the structure functions
\begin{equation}
S_{n}=\overline{(u(x+r)-u(x))^{n}}\equiv \overline{(\Delta u)^{n}}\propto
(Pr)^{\frac{n}{3}}
\end{equation}
\noindent where the pumping rate $P$ is defined below [2]-[4]. Moreover, both
numerical and physical experiments were not accurate enough to measure
\begin{equation}
s_{2n+1}=\frac{S_{2n+1}}{S_{2}^{\frac{2n+1}{2}}}<<1
\end{equation}
\noindent which were too small. This means that the observed
probability density
$P(\Delta u)$ was very close to symmetric one. This experimental fact
differs from the
outcome of the measurements in three dimensions
where $s_{n}$'s are very large
when $n$ is not small. Thus, the absence of strong (may be any)
intermittency
in two-dimensional turbulence and proximity of the statistics of velocity
field to gaussian makes the problem seem tractable.
\noindent The equations of motion are (density $\rho\equiv 1$):
\begin{equation}
\partial_{t}v_{i}+v_{j}\partial_{j}v_{i}=-\partial_{i}p+\nu\nabla^{2}v_{i}+f_{i}
\end{equation}
\noindent and
\begin{equation}
\partial_{i}v_{i}=0
\end{equation}
\noindent where $\bf f$ is a forcing function mimicking the
large-scale turbulence production mechanism and in a statistically
steady state the mean pumping rate $P
=\overline{{\bf f\cdot v}}$. In the inverse cascade range the
dissipation terms in (3) will be irrelevant. Neglecting it and multiplying (3
) by $v_{i}$ we
obtain readily
\begin{equation}
E=\frac{1}{2}\overline{v^{2}}=Pt
\end{equation}
\noindent Thus, in this case the energy linearly grows with time.
In this paper we define the force correlation function as:
\begin{equation}
<f_{i}({\bf k})f_{j}({\bf k'})>\propto
P(\delta_{ij}-\frac{k_{i}k_{j}}{k^{2}})
\frac{\delta(k-k_{f})}{k}\delta({\bf k+k'})\delta(t-t')
\end{equation}
\noindent so that
\begin{equation}
\overline{(f(x+r)-f(r))^{2}}\propto P(1-Cos(k_{f}r))
\end{equation}
\noindent
\noindent It will be clear below that the forcing term enters the equations
for the probability density of velocity differences exclusively through the
expression (7) and in the limit $k_{f}r<<1$ it contribution is
$O((k_{f}r)^{2})$
which is a well-known fact. In the energy range we are interested in this work
$k_{f}r>>1$ and the oscillating contribution can be neglected leading to
disappearence of the forcing scale from equation for the PDF. Thus the general
expression for the structure functions is:
\begin{equation}
S_{n}(r)\propto (Pr)^{\frac{n}{3}}(\frac{r}{L_{i}(t)})^{\delta_{n}}
\end{equation}
\noindent where the exponents $\delta_{n}$ denote possible deviations from the
Kolmogorov scaling. If a statistically steady state exist in the limit
$L_{i}>>l>>l_{f}$, then all $\delta_{n}=0$ since $L_{i}\propto
t^{\frac{3}{2}}$. This would be prooof of `` normal'' (Kolmogorov) scaling
in the inverse cascade range, provided one can show that the PDF $P(\Delta u)$
in the inertial range is independent on its counterpart in the interval
$l\approx l_{f}$. This is the subject of the present paper which is organized
as follows. In the next Section the equations for the generating functions
are introduced. Section 3 is devoted to a short analysis of the Polyakov
theory of Burgers turbulence some aspects of which are used in this paper.
Some physical considerations, which are basis for the developing theory,
are presented in Section 4. S
In Sections 5 and 6 the equations for the transvers and longitudinal
probability density functions are derived and solved. Summary and discussion
are presented in Section 7.
Now we would like to recall some well-known properties of velocity
correlation functions in incompressible fluids, needed below.
Consider two points
${\bf x}$ and ${\bf x'}$ and define ${\bf r}={\bf x-x'}$. Assuming that
the $x$-axis is paralel to the displacement vector ${\bf r}$, one
can find that in the two-dimensional flow $d=2$
for the separation $r$ in the
inertial range [5]-[7]:
\begin{equation}
\frac{1}{r^{d+1}}\partial_{r}r^{d+1}S_{3}=\frac{12}{d}{\cal E}
\end{equation}
\noindent giving
\begin{equation}
S_{3}=\overline{(\Delta u)^{3}}
\equiv\overline{(u(x')-u(x))^{3}}\approx \frac{12}{d(d+2)}r
\end{equation}
\noindent and
\begin{equation}
S^{t}_{3}=\overline{(\Delta v)^{3}}
\equiv\overline{(v(x')-v(x))^{3}}\approx 0
\end{equation}
\noindent where $u$ and $v$ are the components of velocity field paralel
and perpendicular to the $x$-axis (vector ${\bf r})$. The relations (9)-(11)
resulting from equations of motion (3) are dynamic properties of the velocity
field. Kinematics also gives something interesting:
\begin{equation}
\frac{1}{r^{d-1}}\frac{d}{dr}r^{d-1}S_{2}=(d-1)S_{2}^{t}\equiv\overline{(\Delta
v)^{2}}
\end{equation}
\noindent and in two dimensions we have:
\begin{equation}
S_{3t}\equiv \overline{\Delta u (\Delta v)^{2}}=\frac{1}{3}\frac{d}{dr}S_{3}
\end{equation}
\section{Equation for Generating Function}
\noindent
We consider the $N$-point generating function:
\begin{equation}
Z=<e^{\lambda_{i}\cdot {\bf v(x_{i})}}>
\end{equation}
\noindent where the vectors ${\bf x_{i}}$ define the positions of the points
denoted $1\leq i \leq N$.
Using the incompressibility condition,
the equation for $Z$ can be written:
\begin{equation}
\frac{\partial Z}{\partial t}+\frac{\partial^{2} Z
}{\partial \lambda_{i,\mu}\partial x_{i,\mu}}=I_{f}+I_{p}
\end{equation}
\noindent with
\begin{equation}
I_{f}=\sum_{j} <{\bf \lambda_{j}\cdot f(x_{j})}e^{\lambda_{i}u(x_{i})}>
\end{equation}
\begin{equation}
I_{p}=-\sum_{j}\lambda_{j}<e^{\lambda_{i}u(x_{i})}\frac{\partial p(x_{j})}{\partial x_{j}}>
\end{equation}
\noindent The dissipation contributions have been neglected here as
irrelevant.
In what follows we will be mainly interested
in the probability density function of the two-point velocity
differences which is ontained from (7)-(10), setting
$\bf{\lambda_{1}+\lambda_{2}}=0$ (see Ref. [8] and the theory developed
below),
so that
\begin{equation}
Z=<exp{(\bf{\lambda\cdot U})}>
\end{equation}
\noindent where
\begin{equation}
{\bf U}={\bf u(x')-u(x)}\equiv \Delta {\bf u}
\end{equation}
\noindent
The moments of the two-point velocity differences which
in
homogeneous and isotropic turbulence can depend only on
the absolute values of two vectors
(velocity difference ${\bf v(x')-v(x)}$ and displacement
${\bf r\equiv x'-x}$) and the angle $\theta$ between them with $\theta=\pi/2$
and $\theta=0$ corresponding to transverse and longitudinal structure
functions, respectively.
\noindent
It is easy to show [5]- [6] that the
general form of the second-order
structure function in the inertial range is:
\begin{equation}
S_{2}(r,\theta)= \frac{2+\xi_{2}}{2}D_{LL}(r)(1-\frac{\xi_{2}}{2+\xi_{2}}cos^{2}(\theta))
\end{equation}
\noindent with $D_{LL}(r)=<(u(x)-u(x+r))^{2}>$.
More involved relation can
be written for the fourth-order moment:
\begin{equation}
S_{4}(r,\theta)=D_{LLLL}(r)cos^{4}(\theta)-3D_{LLNN}(r)sin^{2}(2\theta)+
D_{NNNN}(r)sin^{2}(\theta)
\end{equation}
\noindent where $D_{LLNN}=<(v(x)-v(x+r))^{2}(u(x)-u(x+r))^{2}>$
and $v$ and $u$ are the components of the velocity field perpendicular
and parallel to the $x$-axis, respectively. In general,
in the llimit $cos(\theta)\equiv s\rightarrow \pm 1$, corresponding to the moments of the
longitudinal velocity differences
$S_{n}(r,s)\rightarrow S_{n}(r)cos^{n}(\theta)$.
This means that in this limit
$Z(\lambda,r,s)\rightarrow Z(\lambda s,r)\equiv Z(\lambda_{x},r)$.
The generating function can depend only on three variables:
$$\eta_{1}=r;~~ \eta_{2}=\frac{{\bf \lambda\cdot r}}{{\bf r}}\equiv
\lambda cos(\theta);~~ \eta_{3}=\sqrt{\lambda^{2}-\eta_{2}^{2}};$$
In these variables:
\begin{equation}
Z_{t}+[\partial_{\eta_{1}}\partial_{\eta_{2}}+\frac{d-1}{r}\partial_{\eta_{2}}
+\frac{\eta_{3}}{r}\partial_{\eta_{2}}\partial_{\eta_{3}}+\frac{(2-d)\eta_{2}}{r\eta_{3}}\partial_{\eta_{3}}-\frac{\eta_{2}}{r}\partial^{2}_{\eta_{3}}]Z=
I_{f}+I_{p}
\end{equation}
\noindent where
\begin{equation}
I_{p}=
\lambda_{i}<(\partial_{2,i} p(2)-\partial_{1,i} p(1))e^{\bf \lambda\cdot U}>
\end{equation}
\noindent and
\begin{equation}
I_{f}=(\eta_{2}^{2}+\eta_{3}^{2})P(1-Cos(k_{f}r))Z
\end{equation}
\noindent where, to simplify notation we set $\partial_{i,\alpha}\equiv
\frac{\partial}{\partial x.\alpha}$ and $v(i)\equiv v({\bf x_{i}})$.
\noindent In two dimensions
the equation for the generating function becomes with $P=1$
(the subscript $o$ is omitted hereafter):
\begin{equation}
[\partial_{\eta_{1}}\partial_{\eta_{2}}+\frac{1}{r}\partial_{\eta_{2}}+
\frac{\eta_{3}}{r}\frac{\partial^{2}}{\partial_{\eta_{2}}\partial{\eta_{3}}}
-\frac{\eta_{2}}{r}\frac{\partial^{2}}{\partial \eta{_3}^{2}}-
(\eta_{2}^{2}+\eta_{3}^{2})]Z=I_{p}
\end{equation}
\noindent The generating function can be written as:
\begin{equation}
Z=<e^{\eta_{2}\Delta u + \eta_{3}\Delta v}>
\end{equation}
\noindent so that any correlation function
\begin{equation}
<(\Delta u)^{n}\Delta v)^{m}>=\frac{\partial^{n}}{\partial
\eta_{2}^{n}}\frac{\partial^{n}}{\partial \eta_{3}^{m}}Z(\eta_{2}=\eta_{3}=0)
\end{equation}
\noindent Neglecting the pressure term $I_{p}$ and differentiating (25) once
over $\eta_{2}$ we obtain immediately
\begin{equation}
\frac{1}{r}\frac{d}{dr}rS_{2}=S_{2}^{t}
\end{equation}
\noindent Second differentiation (again neglecting $I_{p}$) gives:
\begin{equation}
\frac{1}{r}\frac{d}{dr}rS_{3}-\frac{2}{r}S_{3t}-2=0
\end{equation}
\noindent Combined with (13) this expression gives
\begin{equation}
\frac{1}{r^{3}}\frac{d}{dr}r^{3}S_{3}-6=0
\end{equation}
\noindent which is nothing but the Kolmogorov relation, derived in 2d without
contributions from the pressure terms. It follows from (25) that it is
reasonable
to look for a scaling solution $Z(\eta_{2},\eta_{3},r)=Z(X_{2},X_{3})$ where
$X_{i}=\eta_{i}r^{\frac{1}{3}}$.
\section{Polyakov's theory of Burgers turbulence}
The dissipation-generated
contributions $O(\nu \nabla^{2} \overline{u_{i}u_{j}})\neq 0$ in the limit
$\nu\rightarrow 0$. This is a consequence of the ultra-violet singularity
$\nabla^{2}\overline{u_{i}(x)u_{j}(x+r)}\rightarrow \infty$ when $r\rightarrow 0$ making the
theory (the closure problem) extremely difficult.
The expression for this ``dissipation anomaly'', part of the equation
for
the generating function, was developed by Polyakov for
the problem of the one-dimensional Burgers equation stirred by the random
force [8]. Theory of two-dimensional turbulence is free from the troubles
coming from the
ultra-violet (dissipation) singularities. Still, here we review some of
the aspects of
Polyakov's theory which
we believe are
of general interest and which will be most helpful below.
Polyakov considered a one-dimensional problem [8]:
\begin{equation}
u_{t}+uu_{x}=f+\nu u_{xx}
\end{equation}
\noindent where the random force is defined by the correlation function
\begin{equation}
\overline{f(x,t)f(x+r,t')}=\kappa(r)\delta(t-t')
\end{equation}
\noindent The equation for generating function, analogous to (14), is written
readily:
\begin{equation}
Z_{t}+\lambda_{j}\frac{\partial}{\partial
\lambda_{j}}\frac{1}{\lambda_{j}}\frac{\partial Z}{\partial r}=
\kappa(r_{ij})\lambda_{i}\lambda_{j}Z+D
\end{equation}
\noindent where
\begin{equation}
D=\nu \lambda_{j}<u''(x_{j},t)e^{\lambda_{k}u(x_{k},t)}>
\end{equation}
\noindent In the limit $r_{ij}\rightarrow 0$ the force correlation function
$\kappa(r_{ij})=O(1-r_{ij}^{2})$ which imposes scaling properties of the
velocity correlation functions. In general, the generating function depends
on both velocity differences $U_{-}=\Delta u=u(x_{i})-u(x_{j})$ and sums
$U_{+}=u(x_{i})+u_(x_{j})$ which makes the problem very difficult.
Defining Galilean invariance as independence of the correlation functions on
``non-universal'' single-point $u_{rms}^{2}=\overline{u^{2}}$,
Polyakov assumed that if all $|U_{-}|<<u_{rms}$ then $U_{-}$ and $U_{+}$ are
statistically independent and $\sum \lambda_{i}=0$.
In this case (see (8)), introducing
$\mu=\lambda_{2}-\lambda_{1}$ and the two-point generating function
\begin{equation}
Z(\mu)=<e^{\mu \Delta u}>
\end{equation}
\noindent the equation for $Z$ reads in a steady
state:
\begin{equation}
(\frac{\partial}{\partial \mu}-\frac{1}{\mu})\frac{\partial }{\partial r}Z=
-r^{2}\mu^{2}Z+D
\end{equation}
\noindent where
\begin{equation}
D=\mu \nu<(u''(x+r)-u''(x))e^{\mu\Delta u}>
\end{equation}
\noindent It is clear that the $O(r^2)$ forcing term imposes the scaling
variable $\xi=\mu r$ and $Z=F(\mu r)$ where $F$ is a solution of the following equation:
\begin{equation}
\xi F''-F'+\xi^{2}F=D
\end{equation}
\noindent The problem is in evaluation of the dissipation
contribution $D$.
On the first glance one can attempt to neglect $D$ and solve the resulting
equation. This is not so simple, however. The Laplace transform of (38) gives an
equation for the probability density $P=\frac{1}{r}\Phi(\frac{U}{r})
\equiv \frac{1}{r}\Phi(X)$
$$\Phi''+X^{2}\Phi'+X\Phi=0$$
\noindent Introducing
\begin{equation}
\Phi=Exp(-\frac{X^{3}}{6})\Psi
\end{equation}
\noindent gives
\begin{equation}
\Psi''=(\frac{X^{2}}{4}-2X)\Psi
\end{equation}
\noindent which is the Schrodinger equation for a particle in a potential
$U(X)=X^{4}/4-2X$ not having any positive solutions.
\noindent The positivity of
the probability density is a severe constraint on a possible solution of the
equation of motion. That is where the dissipation contribution $D$ comes to
the rescue. Polyakov proposed a self-consistent conjecture about the structure
of the dissipation term
\begin{equation}
D=(\frac{b}{\mu}+a)Z
\end{equation}
\noindent modifying the potential in the Schrodinger equation with the
coefficients $b$ and $a$
chosen to produce the zero-energy ground state corresponding to positive PDF.
According to Ref.(8) this expression is the only one satisfying the galileo
invariance of the small-scale dynamics.
\noindent The fact that the one or multi-dimensional advection
contributions to the equation for the generating function do not lead to
positive solutions for the PDF is a general phenomenon (see below). The
importance of Polyakov's theory is, among other things,
in realization that the dynamic closures for the remaining terms
must remove this problem. This
dramatically narrows the allowed classes of closures. Thus, the
expressions for, $D$ or the pressure terms (see below), combined with
advective contributions to equation for $Z$ can be correct only and only if
they lead
to positive
solutions for the PDF's in the entire range where $|\Delta u|<<u_{rms}$ and $r<<L_{i}$.
\section{Physical Considerations}
The problem of two-dimensional turbulence is simlified by the fact that the
dissipation contributions are irrelevant on the scales $l>>l_{f}$ we are
interested in. Moreover, since $u_{rms}$ grows with time,
the statistically steady small-scale velocity differences
$U_{-}=\Delta u$ with $r<<L(t)$ must be
decoupled from $U_{+}$ in (25). This means that the terms
\begin{equation}
\overline{(\Delta u)^{n}(\Delta v)^{m}}
\end{equation}
\noindent can eneter the equation for $P(\Delta u,r)$ while the ones,
involving
\begin{equation}
\overline{(\Delta u)^{n}(\Delta v)^{m}U_{+}^{p}}
\end{equation}
\noindent cannot. In principle, it can happen that the
$U_{-}U_{+}$-correlation functions can sum up into something time-independent.
However, at present we discard this bizarre possibility.
\noindent Next, the pressure gradients
\begin{equation}
\nabla p(x+r)-\nabla p(x)
\end{equation}
\noindent appearing in the equation (22)-(24) for $Z$
involve integrals
over entire
space. It is clear that, if the steady state exists,
the large- scale contribution to the pressure integrals,
depending on $L=L(t)$ cannot
contribute to the small-scale steady-state dynamics, described by (25).
That is why the pressure contributions to $I_{p}$ (23)
must depend exclusively on the local scale $r$. This leads us to
an assumption that the pressure gradients in (23) are local
in a sense the they can be expressed in terms of the velocity field at the
points $x$ and $x+r$. The application of these
considerations are presented below.
The theory of Burgers turbulence, dealt with the ``universal'' part of
the dynamics, i.e. with the moments of velocity difference $S_{n}$ with
$n<1$. The theory of
two-dimensional turbulence, we are interested in, must produce the
moments with $n<\infty$ and that is why the algebtaic expressions for the
PDF's, characteristic of Burgers dynamics, are irrelevant. In addition, we
expect the small-scale dynamics in 2d to be independent on the forcing
function. This makes this problem very different.
\section{Transverse Structure Functions}
Unlike the probability density function for the longitudinal
velocity differences
$P(\Delta u,r)$,
the transverse velocity difference probability density
is symmetric, i.e. $P(\Delta v,r)=P(-\Delta v,r)$.
We are interested in the equation (25) in the limit $\eta_{2}\rightarrow 0$.
Let us first discuss some of the general properties of incompressible
turbulence. Consider the forcing fucntion
$${\bf f}(x,y)=(f_{x}(x,y),0)$$
In this case the equation (25) is:
\begin{equation}
[\partial_{\eta_{1}}\partial_{\eta_{2}}+\frac{1}{r}\partial_{\eta_{2}}+
\frac{\eta_{3}}{r}\frac{\partial^{2}}{\partial_{\eta_{2}}\partial{\eta_{3}}}
-\frac{\eta_{2}}{r}\frac{\partial^{2}}{\partial \eta{_3}^{2}}-
\eta_{2}^{2}]Z=I_{p}
\end{equation}
\noindent Then, setting $\eta_{2}=0$ removes all
information about forcing
function from the equation of motion. Based on our general intuition and
numerical data we know that two flows strirred by a one-component
or by a two-component (statistically isotropic) forcing function are
identical at the scales $l>>l_{f}$, provided the total fluxes
generated by these
forcing functions are equal. This happens due to pressure terms
$$\Delta p=-\nabla_{i}\nabla_{j}v_{i}v_{j}$$
\noindent effectively mixing various components of the velocity field. This
universality, i.e. independence of the small-scale turbulence on the
symmetries of
the forcing, enables us to write an expression for the $I_{p}$
contribution to (25).
According to considerations, presented in a previous section, the pressure
gradients in the equation (25) are local and their dynamic
role is in mixing various components of velocity
field. Thus the only contribution to $I_{p}$, not vanishing in the limit
$\eta_{2}\rightarrow 0$, can be estimated as:
\begin{equation}
b\frac{\eta_{3}}{r}<\Delta u \Delta v e^{\eta_{2}\Delta u+\eta_{3}\Delta v}>=
b\frac{\eta_{3}}{r}
\frac{\partial}{\partial \eta_{2}}
<\Delta v e^{\eta_{2}\Delta u+\eta_{3}\Delta v}>
\end{equation}
\noindent Using a theorem (see Frisch (8), for example)
that for the random gaussian
process $\xi$ (see below)
\begin{equation}
<\xi F(\xi)>=\overline{\xi^{2}}<\frac{\partial F(\xi)}{\partial \xi}>
\end{equation}
\noindent we derive in the limit $\eta_{2}\rightarrow 0$
\begin{equation}
I_{p}\approx b\eta_{3}^{2}\frac{\overline{(\Delta v)^{2}}}{r}\frac{\partial
Z_{3}}{\partial \eta_{2}}
\end{equation}
\noindent
Substituting this into (25) and integrating over $\eta_{2}$ gives in the limit
$\eta_{2}\rightarrow 0$:
\begin{equation}
\frac{\partial Z_{3}}{\partial r}+\frac{Z_{3}}{r}+\frac{\eta_{3}}{r}\frac{\partial Z_{3}}{\partial
\eta_{3}}-\frac{\gamma}{r^{\frac{1}{3}}}\eta_{3}^{2}Z+
\Omega(\eta_{3})=\Gamma(\eta_{3})
\end{equation}
\noindent where $\gamma$ is undetermined parameter and an arbitrary function
$$\Gamma(\eta_{3})=Z_{3}/r+\Omega(\eta_{3})$$
\noindent with
$$-\Omega(\eta_{3})=
lim_{\eta_{2}\rightarrow 0}~\eta_{3}^{2}\int Z(\eta_{2},\eta_{3},r)d\eta_{2}$$
\noindent is chosen to satisfy
a trivial constraint
$Z_{3}(\eta_{3}=0,r)=1$ and the above mentioned universality.
\noindent
This gives:
\begin{equation}
\frac{\partial Z_{3}}{\partial r}+\frac{\eta_{3}}{r}\frac{\partial Z_{3}}{\partial
\eta_{3}}-\frac{\gamma}{r^{\frac{1}{3}}}\eta_{3}^{2}Z=0
\end{equation}
\noindent where $Z_{3}=Z(\eta_{2}=0,\eta_{3})$.
This equation is invariant under
$\eta_{3}\rightarrow -\eta_{3}$ - transformation. It is important that the
$O(\eta_{3}^{2})$ contribution to (50) comes from the pressure term but not
from the forcing, present in the original equation (25).
Seeking a solution
to this
equation in a scaling form
$Z_{3}(\eta_{3},r)=Z(\eta_{3}r^{\frac{1}{3}})\equiv Z(X)$ gives:
\begin{equation}
\frac{4X}{3}Z_{X}=\gamma X^{2}Z
\end{equation}
\noindent and
\begin{equation}
Z=Exp(\frac{3\gamma}{8}\eta_{3}^{2}r^{\frac{2}{3}})
\end{equation}
This generating function corresponds to the gaussian distribution of
transverse velocity differences $P(\Delta v)$ with the second-order structure
function
\begin{equation}
S_{2}^{t}(r)=\overline{(\Delta v)^{2}}=\frac{3\gamma}{4}r^{\frac{2}{3}}
\end{equation}
The equation (50) correseponds to a one-dimensional
linear Langevin equation for ``velocity field'' $V=v/(Pr^){\frac{1}{3}}$
\begin{equation}
v_{\tau}(x)=-v(x)+\phi(x,\tau)
\end{equation}
\noindent where $\tau\propto tr^{-\frac{2}{3}}P^{\frac{1}{3}}$ and
the non-local gaussian ``universal'' forcing $\phi(x,\tau)$,
generated by the
nonlinearity of the original equation is defined by the correlation function
\begin{equation}
\overline{\phi(k,\tau)\phi(k',\tau')}\propto
\delta(k+k')\delta(\tau-\tau')
\end{equation}
\noindent The generating function for the field $V$ is
$$z=<e^{XV}>$$
\noindent Since $\tau\propto tr^{-\frac{2}{3}}$ and $V\propto
vr^{-\frac{1}{3}}$
this equation is strongly non-local. It becomes local, however in the
wave-number space. This will be discussed later.
\noindent Now we can attempt to justify the relation (46). According to (23)
and taking into account that the $x$-axis is paralel to the displacement $r$
in the limit $\eta_{2}\rightarrow 0$
$$I_{p}\approx \eta_{3}<(\partial_{y}p(0)-\partial_{y'} p(r))
Exp(\eta_{3}\Delta v+\eta_{2}\Delta u)>$$
\noindent where
$$\partial_{y}p(0)-\partial_{y'} p(r)=\int
k_{y}(1-e^{ik_{x}r})[\frac{k_{x}^{2}}{k^{2}}u(q)u(k-q)+\frac{k_{y}^{2}}{k^{2}}v(q)v(k-q)+\frac{k_{x}k_{y}}{k^{2}}u(q)v(k-q)]d^{2}kd^{2}q$$
\noindent and the exponent is expressed simply as:
$$e^{\eta_{3}\Delta v +\eta_{2}\Delta u}=
Exp[\eta_{3}\int (1-e^{iQ_{x}r})v(Q)d^{2}Q +
\eta_{2}\int (1-e^{iQ_{x}r})u(Q)d^{2}Q]$$
\noindent
It will be come clear below that transverse velocity differences $\Delta v$
obey gaussian statistics and the longitudinal ones $\Delta u$
are very close to
gaussian. Then, substituting the above
expressions into $I_{p}$ and expanding
the exponent
we generate an infinite series involving various products of $u(q)$'s and
$v(q)$'s. In case of an incompressible, statistically isotropic gaussian
velocity field, we are dealing with, these products are split
into pairs:
$$<v_{i}(q)v_{j}(Q)>\propto
q^{-\frac{8}{3}}(\delta_{ij}-\frac{q_{i}q_{j}}{q^{2}})\delta(q+Q)$$
\noindent The $k_{y}$ integration is carried over the interval
$-\infty<k<\infty$ and in the isotropic case we are dealing with the only
non-zero terms are those involving even powers of $k_{y}$. These terms are
generated by the expansion of
$$e^{\eta_{2} \Delta u}$$
\noindent They however, , being $O(\eta_{2})$,
disappear in the limit $\eta_{2}\rightarrow 0$.
Thus:
$$I_{p}=\eta_{3}\int d^{2}kd^{2}q
k_{y}(1-e^{ik_{x}r})\frac{k_{x}k_{y}}{k^{2}}<u(q)v(k-q)
Exp(\eta_{3}\int (1-e^{iQ_{x}r})v(Q)d^{2}Q +
\eta_{2}\int (1-e^{iQ_{x}r})u(Q)d^{2}Q)>
$$
\noindent where the $O(\eta_{2})$ contribution to the exponent is temporarily
kept to make the transformation
$$\Delta u e^{\eta_{2}\Delta u}=\frac{\partial e^{\eta_{2}\Delta
u}}{\partial \eta_{2}}$$
\noindent to (46) possible. Only after that we set $\eta_{2}=0$.
This proves that the only contribution to the
equation for the probability density function comes from the $O(\Delta u
\Delta v)$ mixing components, involved in the pressure gradients. This
relation justifies the estimate (46).
\section{Longitudinal Velocity Differences}
\noindent The remarkable fact that in the limit $\eta_{2}\rightarrow 0$
all contributions to the equation (25) contain
$\frac{\partial}{\partial \eta_{2}}$ enables separation of variables:
integrating the resulting equation over $\eta_{2}$ gives the closed equation
for $Z_{3}(\eta_{3})$. The corresponding dynamic
equation is linear, meaning that transverse
velocity fluctuations do not directly contribute to the energy transfer
between different scales. This effect is possible only in 2d where the
$O((d-2)\frac{\partial}{\partial \eta_{3}})$
enstrophy production term in (22), not containing
$\frac{\partial}{\partial \eta_{2}}$, is equal to zero. This simplification,
combined with locality of the pressure-gradient
effects, allowed us to
derive a closed-form expression for $Z_{3}$.
The role of pressure in the dynamics of tranverse components of velocity field
is mainly restricted to control of the ``energy redistribution'' neccessary
for generation of
isotropic and incompressible velocity field. The longitudinal field dynamics
are much more involved. The advection (pressure excluding) part of
non-linearity tends to produce large
gradients of velocity field (``shock generation''
using the Burgers equation
phenomenology), manifesting itself in creation of a
constant energy flux in the wave-number space.
Pressure is the only factor preventing the shock
formation.
Interested in the longitudinal
correlation functions we set $\eta_{3}=0$. Then, the term in (25)
\begin{equation}
\frac{\eta_{2}}{r}\frac{\partial^{2}Z}{\partial \eta_{3}^{2}}=
\frac{\eta_{2}}{r}<(\Delta v)^{2}e^{\eta_{2}\Delta u}>\approx
\frac{\eta_{2}A_{2}^{t}}{r^{\frac{1}{3}}}Z_{2}+O(\eta_{2}^{2};~\eta_{3}^{2};
~\eta_{2}^{2}\eta_{3})
\end{equation}
\noindent The last relation is accurate since substituting this into (25),
differentiating once over $\eta_{2}$ and setting
both $\eta_{3}=\eta_{2}=0$ gives:
\begin{equation}
\frac{1}{r}\frac{\partial}{\partial r}rS_{2}-\frac{A_{2}^{t}}{r^\frac{1}{3}}=
\frac{\partial I_{p}(0,0)}{\partial \eta_{2}}
\end{equation}
\noindent Since $S_{2}(r)=A_{2}r^{\frac{2}{3}}$ this equation gives:
\begin{equation}
\frac{5}{3}A_{2}-A_{2}^{t}=
r^{\frac{1}{3}}\frac{\partial I_{p}(0,0)}{\partial \eta_{2}}
\end{equation}
\noindent which, according to (12) is exact since
$\frac{\partial I_{p}(0,0)}{\partial \eta_{2}}=0$ (see below).
Let us consider some general properties of the pressure term $I_{p}$ in the
limit $\eta_{3}\rightarrow 0$. We have:
\begin{equation}
I_{p}\approx \eta_{2}<(\frac{\partial p(2)}{\partial x_{2}}-\frac{\partial
p(1)}{\partial x_{1}})
Exp(\eta_{2}\Delta u +\eta_{3}\Delta v)>
\end{equation}
\noindent
Expanding the exponent and recalling that the isotropic and
incompressible turbulence $\overline{\Delta u}=\overline{\Delta v}=0$ and
$\overline{p(x)v_{i}(x')}=0$, we conclude that
\begin{equation}
I_{p}\approx \eta_{2}<(\frac{\partial p(2)}{\partial x_{2}}-\frac{\partial
p(1)}{\partial x_{1}})
(\eta_{2}\Delta u +\eta_{3}\Delta v)^{2}+...>=O(\alpha \eta_{2}^{3}+\beta
\eta_{2}^{2}\eta_{3}+...)
\end{equation}
\noindent It is clear that the relation (48), derived above for the case of
gaussian statistics, satisfied this
general property of the flow. Thus when $\eta_{3}\rightarrow 0$,
we approximate
\begin{equation}
I_{p}\approx c\eta_{2}^{3}Z+G
\end{equation}
\noindent where $c$ is a yet undetermined constant and $G$ denotes the
contributions to $I_{p}$, properly modifying numerical coefficients in the
equation (25). The presence of the $O(\eta_{2}^{3})$ distinguishes this
equation from the one for transverse PDF considered in the previous section.
There the assumed role of pressure was limited to the mixing of various
components of velocity field. That is why all we accounted for was $O(
\Delta v \Delta u)$ contributions to pressure. Here, in addition we
also consider $O(\eta_{2}^{3})$ contributions, responsible for prevention of
the shock formation.
The resulting equation is:
\begin{equation}
\frac{1}{r^{3}}\frac{\partial^{2}}{\partial \eta_{2}\partial r}r^{3}Z_{2}-
\frac{11}{5r^{\frac{1}{3}}}A_{2}^{t}\eta_{2}Z_{2}-3\eta_{2}^{2}Z_{2}-
c\eta_{2}^{3}Z_{2}=0
\end{equation}
\noindent The Laplace transform of gives equation for the probability density
$P(\Delta u,r)$:
\begin{equation}
cP_{UUU}-3P_{UU}+\frac{1}{r^{3}}\frac{\partial}{\partial r}r^{3}UP+
\frac{11 A_{2}^{t}}{5}P_{U}=0
\end{equation}
\noindent Seeking solution in a scaling form (the parameter $c$ will be
determined below)
\begin{equation}
P(U,r)=\frac{1}{r^{\frac{1}{3}}}F(\frac{U}{r^{\frac{1}{3}}})
\end{equation}
\noindent we obtain
\begin{equation}
cF_{xxx}-3F_{xx}+(b-\frac{x^{2}}{3})F_{x}+\frac{8}{3}xF=0
\end{equation}
\noindent
Where $b=\frac{11}{3}A_{2}$.
All, but one,
term in (65) change
sign wnen $x\rightarrow -x$. The $O(F_{xx})$ symmetry-breaking contribution
is neccessary for existence of the non-zero energy flux.
Assuming for a time being that, in accord with numerical and physical
experiments, that the flux is small (see relation (2)), we first neglect the
$O(F_{xx})$- contribution, find solution and then take it into account
perturbatively. The equation is:
\begin{equation}
cF^{o}_{xxx}+(b-\frac{x^{2}}{3})F^{o}_{x}+\frac{8}{3}xF^{o}=0
\end{equation}
\noindent with solution:
\begin{equation}
F^{o}=e^{\frac{x^{2}}{2A_{2}}}
\end{equation}
\noindent where $c=\frac{A_{2}^{2}}{3}$.
If $A_{2}>>1$, then the neglected $F_{xx}=O(1/A_{2})$ term is small.
This means
that the odd-order moments, computed with the PDF, which is a solution of
(65), must be small in a sense defined by the relation (2). At the same
time the even-order moments must be close to the gaussian ones.
Analytic solution of (65) is difficult. However, one can evaluate
all moments $\frac{S_{n}}{r^{\frac{n}{3}}}=A_{n}$ in terms of only one
parameter $A_{2}$:
\begin{equation}
S_{n+1}=-\frac{3}{n+10}(-\frac{A_{2}^{2}}{3}n(n-1)(n-2)S_{n-3}-3n(n-1)S_{n-2}-\frac{11}{3}A_{2}nS_{n-1})
\end{equation}
This relation gives: $A_{1}=0; A_{3}=3/2; A_{4}=3; A_{5}=12.43A_{2}; A_{6}=
15A_{2}^{3}-36; A_{7}=37.71A_{4}$ etc. These numbers can be tested in
numerical experiments. The one-loop renormalized perturbation expansions give
$A_{2}\approx 10$, while numerical simulations are consistent with
$A_{2}\approx 12$. Keeping these numbers in mind, it follows from (68)
that the accurate measurements of the odd-order moments is the
only way to verify
predictions of the present theory. The deviations of the
even-order moments from the gaussian ones are too small to be detected by both
physical and numerical experiments. It can be checked that the ratios
$$s_{2n+1}=\frac{S_{2n+1}}{S_{2n}^{\frac{2n+1}{2n}}}$$
\noindent vary in the interval $0.04-0.1$ for $2<n<10$ and $A_{2}\approx 10$.
With $A_{2}\approx 12$ these numbers decrease even more.
\section{Summary and Conclusions}
\noindent The experimentally observed gaussian or very close to it statistics
of transverse velocity differences was extremely puzzling since, on
the first glance,
it is incompatible with the non-trivial Kolmogorov scaling,
resulting from strong non-linearity of the problem. The most surpring and
interesting result, derived in this paper, is that due to the symmetries of
the problem the equation, governing probability density function of transverse
velocity differences, has one derivative less than the one corresponding to
the longitudinal differences. This means, in turn, that transverse
components of
velocity field are governed by a non-local linear, equation, driven by a
universal, non-local, solution-depending gaussian force. This reduction,
resembling the super-symmetry effects in field theory, is surprising if not
miraculous.
The non-local equation in the physical space, obtained above,
corresponds to the Langevin equation in the Fourier space:
\begin{equation}
v_{t}(k)+c_{\nu}P^{\frac{1}{3}}k^{\frac{2}{3}}v=f_{R}(k,t)
\end{equation}
\noindent where $c_{\nu}$ is an amplitude of
``effective'' (turbulent) viscosity and
\begin{equation}
\overline{f_{R}(k,t)f_{R}(k',t')}\propto k^{-1}\delta(k+k')\delta(t-t')
\end{equation}
\noindent used in [9]-[10]
in the renormalization group treatments of fluid turbulence.
\noindent
The irrelevance of the
dissipation terms in two-dimensional turbulence
makes the problem much more tractable than its
three-dimensional counterpart. Still, in order to close the equations
for probability density of velocity field
one needs an expression for the pressure contributions.
The situation is even more simplified by the fact
that the large-scale-dominated single-point variables are time-dependent and
must decouple from the steady-state small-scale dynamics. That is why one can
use an assumption about locality of the pressure gradient effects leaving only
the mixing $O(\Delta u \Delta v)$ contributions to the two-point pressure
difference. It can be tested by a mere accounting that all other contributions
to the expression for $I_{p}$ involve one or more $U_{+}$'s and leading to
the time-dependent result. This means that they must
disappear from the steady state equations (25) and (45).
The range of
possible
models for pressure is narrowed by a few dynamic and kinematic constraints and
by the fact that the resulting equation must give positive solution. A simple
calculation shows that the model for the pressure gradient terms, introduced
in this paper, is consistent
with the derived gaussian statistics.
The equations for PDF of longitudinal velocity differences do not
correspond to linear dynamics. Still, the derived solution only slightly
deviates from
gaussian. This is possible due to the relative smallness of the energy flux
in two dimensions.
The results presented here seem to agree with both physical and numerical
experiments. The obtained close-to-gaussian statistics justifies various
one-loop renormalized perturbation expansions giving $A_{2}\approx
10-12$. Using this number we realize that it is extremely difficult to
experimentally detect deviations from the gaussian statistics.
Still, some fine details of the present theory, related to the pressure
gradient-velocity correlation functions can be tested numerically.
In addition, measurements of a few odd-order moments can shed some light on
validity of the present theory.
The equations and solution presented here leave one question unanswered:
are these
${\bf the}$ solutions or not? Our experience with the Burgers
and 2d Navier-Stokes
equations teach us that it is very difficult to find a self-consistent closure
leading to the positive solution for the PDF's. Stretching this statement
a bit
we feel that a closure, satisfying dynamic constraints and leading
to a a plausable
solution has a great chances to be correct.
\noindent Absense of intermittency in a steady-state
developing inertial range discovered
in two-dimensional turbulence [2]-[4] seems to be a general
phenomenon observed in a drift-wave turbulence [11] and in a one-dimensional
model of a passive scalar advected by a compressible velocity field [12].
These observations support our understanding of intermittency as a phenomenon
originating from
interaction of the large and small-scale velocity fluctuations. In a
developing statistically steady
inertial range, were the integral scale is strongly time-dependent,
these interactions must be small for the small-scale steady state to exist.
At the later stages the finite size effects, destroying
time-independence of the small scale
dynamics, lead to formation of coherent structures and new dynamic phenomena
which are beyond the scope of the present theory.
\section{Acknowledment} I am grateful to A. Polyakov, M. Vergassola,
M.Chertkov, B.Shraiman, Y. Sinai and I. Kolokolov for many
interesting and
illuminating discussions.
\noindent {\bf references}
\\
1. R.H.Kraichnan, Phys.Fluids. {\bf 10}, 1417 (1967),
\\
2. L.M. Smith and V. Yakhot, Phys.Rev.Lett. {\bf 71}, 352 (1993)
\\
3. L.Smith and V. Yakhot, J. Fluid. Mech. {\bf 274}, 115 (1994)
\\
4. P. Tabeling and J. Paret, Phys. Fluids, {\bf 12}, 3126 (1998)
\\
5. L.D.Landau and E.M. Lifshitz, Fluid Mechanics, Pergamon Press, Oxford, 198
\\
6. A.S.Monin and A.M.Yaglom, ``Statistical Fluid Mechanics'' vol. 1, MIT Press,
Cambridge, MA (1971)
\\
7 U. Frisch, ``Turbulence'', Cambridge University Press, 1995
\\
8 . A.M. Polyakov, Phys.Rev. E, {\bf 52}, 6183 (1995)
Phys.Rev. E {\bf 52}, 6183 (1995)
\\
9. de Dominicis and P.C.Martin, Phys.Rev.A {\bf 19}, 419 (1979)
\\
10. V. Yakhot and S.A.Orszag, Phys.Rev.Lett. {\bf 57}, 1722 (1986)
\\
11. N. Kukharkin, S.A.Orszag and V.Yakhot, Phys.Rev.Lett., {\bf 75}, 2486
(1995)
\\
12. K. Gawedzki and M.Vergassola, COND-MAT/9811399, (1998)
\\
\end{document}
|
\section{Introduction}
One-dimensional and quasi-one dimensional quantum spin systems have
attracted much attention in recent years due to the large
number of experimental realizations of such systems
and the variety
of theoretical techniques, both analytical and numerical, available to
study the relevant models. Due to the presence of large quantum
fluctuations in low dimensions, these systems present unusual properties
such as a gap between a singlet ground state and excited
non-singlet states. Examples include spin ladder systems in
which a small number of one-dimensional spin-1/2 chains interact among
themselves\cite{ladder}. In this case, in a way very similar to the
Haldane spin-S problem\cite{haldane_gap}, it has been found that if the number
of
chains is even the system effectively behaves as an integer spin
chain with a gap in the low-energy spectrum, while it remains
massless for an odd number of chains. Some two-chain ladders
which exhibit a gap are
$SrCu_2O_3$\cite{azuma_srcuo} and
$Cu_2(C_5H_{12}N_2)_2Cl_4$\cite{chaboussant_cuhpcl}, and an example
of a gapless
three-chain ladder is $Sr_2Cu_3O_5$\cite{azuma_srcuo}. Thus far we
implicitly assumed that the boundary conditions in the transverse
direction were open boundary conditions (OBC). These boundary
conditions correspond to having all the chains lying in the same
plane. This is the situation encountered in experimental systems such
as $Sr_2Cu_3O_5$.
In contrast with OBC, periodic boundary
conditions (PBC) are frustrating for (2n+1) coupled chains.
As a consequence all the spin
excitations are gapped\cite{schulz_berlin,orignac_spintube} in the
case of periodic boundary conditions. They are also
gapped for 2n chains with PBC but the mechanism is related to singlet
formation as in the OBC case and not frustration. The PBC could be
achieved in an experimental system by having the coupled chains
forming a cylinder instead of lying in a plane.
A richer behaviour emerges when these gapped
or ungapped systems are placed in a magnetic field. Then it is possible for an
integer spin chain to be gapless and a half-odd-integer spin chain to
show a gap above the ground state for certain values of the
field\cite{oshikawa,cabra,cabra_unp,cithra}.
This has been demonstrated by several
methods such as bosonization \cite{schulz_berlin,affleck_houches}, perturbation
theory\cite{reigrotzki} or density-matrix renormalization group method
(DMRG)\cite{white,dmrg0,kawan1}. In particular, it has been shown that
spin-1/2 chains and ladders with a gap
undergo a continuous phase transition from a
commensurate zero uniform magnetization phase to an incommensurate
phase with non-zero magnetization\cite{cithra}, and
the magnetization of the system can exhibit plateaus at certain
non-zero values of the magnetic field\cite{cabra,mila_field}. Further,
a striking property of the quantum spin-chains in a uniform magnetic
field pointing along the direction of the axial symmetry
($z$-direction), is the topological quantization of the magnetization
under a changing of the magnetic field\cite{oshikawa}. It was shown
that starting from
a generalized Lieb-Schultz-Mattis (LSM) theorem\cite{lsm},
that translationally
invariant spin chains in an applied field can be gapful without
breaking translational symmetry, only when the magnetization per spin,
$m$, obeys $S-m=integer$, where $S$ is the maximum possible spin in
each unit cell of the Hamiltonian. Such gapped phases correspond to plateaus at
these quantized values of $m$. In Ref. \onlinecite{sen} the behavior of
the magnetization versus magnetic field has been investigated in
details using DMRG techniques
for three coupled spin 1/2 chains with both periodic and open
boundary conditions. Plateaus have been obtained at $m=1/2$ and
$m=3/2$ in agreement with Ref. \onlinecite{oshikawa}. Further, for the case of PBC
a small plateau at $m=0$ was also obtained \cite{cabra_unp,kawan1,sen}.
Finally, there seems to
exist some weak
evidence for a plateau at $m=1$ for PBC\cite{sen}. Strong coupling Low
Energy Hamiltonians for these two systems were also derived in
Ref.~\onlinecite{sen}.
In this paper, we investigate a three-chain ladder with PBC
(spin-tube) in presence of a uniform magnetic field
by using
bosonization and renormalization group techniques. We are concerned by
the transition region between the magnetization plateaus at $m=1/2$
and $m=3/2$. Our analysis
is based on the low-energy effective Hamiltonian (LEH)
derived for strong coupling between the rungs\cite{sen}. We identify the
LEH as an anisotropic
SU(3) spin chain with symmetry
breaking terms in a longitudinal magnetic field, and analyze its
low-energy
physics via bosonization and RG techniques.
This approach allows us to predict the behavior of the spin-spin
correlation functions in this transition region and the NMR relaxation
rate. This also allows for an investigation of the possibility of a
plateau at $m=1$.
The paper is organized as follows. In Sec.\ref{leh}, we recall the
derivation\cite{sen} of the LEH
and reduce it to an anisotropic SU(3) spin chain, while
that in OBC ladder reduces to an anisotropic SU(2) spin chain.
The difference
between PBC and OBC models becomes obvious in the language of
effective Hamiltonians.
In Sec.\ref{magn} we discuss the generalization of the LSM
theorem\cite{lsm} for SU(3) spin chains. We then review the
analysis of the magnetization process of isotropic SU(3) spin chain,
and discuss the difficulty\cite{sen} in obtaining the position of the
plateaus from the LEH.
We show there is no gap at $m=1$
in a simple numerical analysis using density matrix renormalization
group method as well.
The bosonized Hamiltonian is derived in
Sec.\ref{weakcoupling}. In section \ref{RG}, we analyze the
low-energy effective Hamiltonian in a weak-coupling limit by
calculating the one loop renormalization
group (RG) in the marginally perturbed SU(3) Wess-Zumino-Witten
model\cite{prokovskii} and discuss the
renormalization group flow. At weak coupling, the
flow is to an invariant surface, leading to
gapless excitations above the ground state with no breaking of the
discrete symmetry. Based on the weak coupling renormalization group
analysis and the usual continuity between weak coupling and strong
coupling in one-dimensional systems, we claim that the spin-tube is
described by a two component Luttinger liquid at low energy and long
wavelength.
In section\ref{correlations}, we discuss the effect of a variation of
the magnetic
field in that problem, and show that it does not affect the
two component Luttinger liquid behavior.
We discuss the analogy of this two component Luttinger liquid
with the S2 phase of
the bilinear-biquadratic spin 1 chain\cite{fath_biquadratic}.
Then, having established the equivalence with the
two component Luttinger liquid, we calculate the spin
correlation functions in the critical region and the
temperature dependence of the NMR longitudinal relaxation rate
$T_1^{-1}$. We also present a theoretical description of the plateau
at $m=1$ in the framework of bosonization. Comparing this description
with the numerical results of Ref. \onlinecite{sen} we conclude that
the presence of a plateau at $m=1$ is unlikely in the spin-tube. We verify
our results on the absence of plateaus at $m=1$ using DMRG, and
indicate the modifications of the spin-tube Hamiltonian that could
lead to a plateau.
Sec.\ref{conclusions} contains the
concluding remarks.
Technical details can be found in the
appendices.
\section{Low-energy effective Hamiltonian of the spin-tube}\label{leh}
The Hamiltonian of the three-chain ladder with periodic boundary
conditions (PBC) in the presence of an external
magnetic field is,
\begin{equation}
\label{ham}
H=J\sum_{i=1}^{N}\sum_{p=1}^{3} \vec S_{i,p} \vec S_{i+1,p}+ J_{\perp}
\sum_{i=1}^{N}\sum_{p=1}^{3} \vec S_{i,p} \vec S_{i,p+1} -\vec h
\sum_{i=1}^{N}\sum_{p=1}^{3} \vec S_{i,p},
\end{equation}
\noindent where $p$ (resp. $i$) is a chain (resp. site) index, $J$ is the
coupling along the chain, $J_{\perp}$ the transverse coupling and
the site $(i,4)$ is identified with the site $(i,1)$. The three-chain
ladder with periodic boundary conditions can be viewed as forming a tube
with an equilateral triangular cross section (see Fig.~\ref{fig:tube}).
We will refer to this system as a {\it spin-tube}.
In the rest of the paper we shall consider the model for
$J_{\perp} \gg J$, and the aim of this section is to recall briefly
the derivation of the low energy Hamiltonian \cite{sen} in this limit.
To begin with, for $J=0$, the system consists of independent
rungs. The eight states of a given rung fall into a spin-3/2
quadruplet
and two spin-1/2 doublets. In the absence of a magnetic field, the
spin 3/2 states on a given triangle
are all degenerate with energy $3/4 J_{\perp}$. These states are:
\begin{eqnarray}
|3/2;3/2\rangle & = & |\uparrow \uparrow \uparrow\rangle \nonumber \\
|3/2;1/2\rangle & = & \frac{1}{\sqrt 3}\lbrack |\uparrow \uparrow \downarrow\rangle +
|\uparrow \downarrow \uparrow\rangle +| \downarrow \uparrow \uparrow\rangle
\rbrack \nonumber \\
|3/2;-1/2\rangle & = & \frac{1}{\sqrt 3}\lbrack |\downarrow \downarrow \uparrow\rangle+
|\downarrow \uparrow \downarrow\rangle +| \uparrow \downarrow \downarrow\rangle \rbrack
\nonumber \\
|3/2;-3/2\rangle & = & |\downarrow \downarrow \downarrow\rangle
\end{eqnarray}
Also, in the absence of a magnetic field and on a given rung, the
two spin 1/2 doublets, corresponding to the left and right
chiralities
(-/+), are degenerate with energy
$-3/4 J_{\perp}$. These states are:
\begin{eqnarray}
|\uparrow +\rangle & = & \frac{1}{\sqrt 3} \lbrack |\downarrow \uparrow
\uparrow\rangle + j |\uparrow \downarrow \uparrow\rangle + j^2 |\uparrow \uparrow
\downarrow\rangle \rbrack \nonumber \\
|\downarrow +\rangle & = & \frac{1}{\sqrt 3} \lbrack |\uparrow \downarrow
\downarrow\rangle + j |\downarrow \uparrow \downarrow\rangle + j^2 |\downarrow \downarrow
\uparrow\rangle \rbrack \nonumber \\
|\uparrow -\rangle & = & \frac{1}{\sqrt 3} \lbrack |\downarrow \uparrow
\uparrow\rangle + j^2 |\uparrow \downarrow \uparrow\rangle + j |\uparrow \uparrow
\downarrow\rangle \rbrack \nonumber \\
\label{doublet}
|\downarrow -\rangle & = & \frac{1}{\sqrt 3} \lbrack |\uparrow \downarrow
\downarrow\rangle + j^2 |\downarrow \uparrow \downarrow\rangle + j |\downarrow \downarrow
\uparrow\rangle \rbrack
\label{doublets}
\end{eqnarray}
\noindent where $j=\exp(\frac{2\pi i}{3})$.
When an external magnetic field is switched on the degeneracy in the different multiplets is lifted.
The energy levels of the
state $|\uparrow \uparrow \uparrow\rangle$ (in the spin-3/2 multiplet)
and the spin-1/2 states
$|\uparrow +\rangle$, $|\uparrow -\rangle$ cross at $h_c=\frac{3}{2}
J_\perp$ (see. Fig.~\ref{fig:levels}). As a result for
$h<h_c$ one has a ground state
magnetization $\langle S^z \rangle=1/2$ and for $h>h_c$, $\langle S^z
\rangle=3/2$, i.e. $h_c$ is a transition point between two
magnetization plateaus. If a small coupling $J$ is
turned on, this transition
is expected to broaden between $h_{1/2,+}$ and $h_{3/2,-}$, where
$h_{3/2,-}-h_{1/2,+}$ is of the order of $J$. We expect that in this
interval $\langle S^z\rangle$ increases continuously with $h$.
In this limit the properties of
the system can be studied by perturbing with $H_1$ around the decoupled rung
hamiltonian $H_0$,
\begin{eqnarray}
H&=&H_0+H_1 \\
H_0&=&J_{\perp}\sum_{i=1}^{N}\sum_{p=1}^{3} \vec S_{i,p} \vec S_{i,p+1}
-h_c \sum_{i=1}^{N}\sum_{p=1}^{3}S_{i,p}^{z}\label{h0},\\
H_1&=& J\sum_{i=1}^{N}\sum_{p=1}^{3} \vec S_{i,p} \vec S_{i+1,p}
-(h-h_c) \sum_{i=1}^{N}\sum_{p=1}^{3}S_{i,p}^{z}.\label{h1}
\end{eqnarray}
At $h=h_c$ the groundstate of $H_0$ is $3^{N}$ fold degenerate,
the states $|\uparrow -\rangle_i$,
$|\uparrow +\rangle_i$, $|\uparrow \uparrow \uparrow\rangle_i$, to be
denoted respectively $|\tilde 1\rangle_i$, $|\tilde 2\rangle_i$,
$|\tilde 3\rangle_i$, spanning the low-energy subspace. $H_1$
lifts the degeneracy in the subspace, leading to an
effective Hamiltonian that can be derived by standard perturbation
theory. Since in the truncated subspace there are 3
states per triangle, it is natural to express the spin
operators in the basis given by Gell-Mann matrices $\lambda^{\alpha},\; \alpha= 1 \dots 8$.
(The conventions we use for the matrices can be found for instance in
Refs. \onlinecite{gellmann_matrix} and \onlinecite{gasiorowicz}).
By considering the action of the spin operators $S_{1,2,3}^+$ and
$S_{1,2,3}^z$ on each state of the truncated Hilbert space
the spin operators can be expressed in
terms of the matrices as,
\begin{eqnarray}
S_{i,p}^{+} & =& \frac{1}{2\sqrt{3}}\lbrack j^{p-1} (\lambda_i^{6}+i
\lambda_i^{7})+
j^{2(p-1)}(\lambda_{i}^{4}- i \lambda_{i}^{5})\rbrack \\
S_{i,p}^{z} &= &\frac{1}{3} \lbrack
\frac 5 6 I -\frac{\lambda_i^{8}}{\sqrt{3}}-
j^{2(p-1)}(\lambda_{i}^{1}+i\lambda_i^{2})-j^{(p-1)}(\lambda_{i}^{1}-i\lambda_i^{2}) \rbrack \label{si}
\end{eqnarray}
\noindent where $I$ is the identity matrix.
The total rung spin is given by:
\begin{eqnarray}\label{siz}
S_i^z=\left ( \frac{5}{6}I -\frac{\lambda_i^8}{\sqrt{3}}\right ).
\end{eqnarray}
The effective Hamiltonian to first order then becomes:
\begin{eqnarray}
H_{\text{eff.}}&=&\tilde{H}_0 + \tilde{H}_I, \label{heff1} \\
\tilde{H}_{0}& = &\frac{J}{4}
\sum_{i=1}^{N}\sum_{\alpha=1}^8\lambda_{i}^{\alpha}
\lambda_{i+1}^{\alpha} \label{h_0}\\
\tilde{H}_I&=&q\sum_{i=1}^{N}[\lambda_{i}^1\lambda_{i+1}^1+\lambda_{i}^2\lambda_{i+1}^2] + u \sum_{i=1}^{N} \lambda_i^{3} \lambda_{i+1}^3
+u^\prime \sum_{i=1}^N \lambda_{i}^8 \lambda_{i+1}^8
+ \frac{h_{\text{eff.}}}{\sqrt{3}}
\sum_{i=1}^{N} \lambda_i^{8} \label{h_1}
\end{eqnarray}
In our case, $q=-J/12$, $u=-J/4$, $u^\prime=-5J/36$
and $h_{\text{eff.}}=h-h_c-5J/9$; hereafter we choose our units so that
$J=1$.
The Hamiltonian (\ref{heff1}) is written as an isotropic SU(3) spin
chain $\tilde H_0$ and terms $\tilde H_I$ that break the symmetry.
This form will
be convenient later on when we study such questions as what regions of
parameter space are gapless and the behavior of correlations
functions there.
Another form of the Hamiltonian is convenient when one wishes to study
the plateau structure.
Introduce\cite{sen} a different basis of $SU(3)$
operators $T^{\pm}_{1}$,
$T^{\pm}_{2}$, $T^{\pm}_{3}$, and $T^z$ defined by:
\begin{eqnarray}
\label{eq:T-lambda}
T_1^{\pm}&=&(\lambda_1 \pm \imath\lambda_2)/2 \\
T_2^{\pm}&=&(\lambda_4 \pm \imath \lambda_5)/2 \\
T_3^{\pm}&=&(\lambda_6 \pm \imath \lambda_7)/2 \\
T^{z}&=& -2\frac{\lambda_8}{\sqrt{3}}
\end{eqnarray}
Then, to first order, and up
to a constant, the effective Hamiltonian reads:
\begin{eqnarray} \label{heff}
H_{eff} & = & \frac{J}{2} \sum_{i} \lbrack T^{+}_{i,2}T^{-}_{i+1,2}+
T^{-}_{i,2}T^{+}_{i+1,2}+ T^{+}_{i,3}T^{-}_{i+1,3}+
T^{-}_{i,3}T^{+}_{i+1,3}\rbrack+ \nonumber \\
& + & \frac{J}{3} \sum_i \lbrack T^{+}_{i,1}T^{-}_{i+1,1}+
T^{-}_{i,1}T^{+}_{i+1,1} \rbrack +\frac{J}{12} \sum_i
T^{z}_{i}T^{z}_{i+1} -
(\frac{1}{2}h - \frac{3}{4} J_{\perp} -\frac{5}{18} J) \sum_iT^{z}_i.
\end{eqnarray}
This is the Hamiltonian derived in Ref. \onlinecite{sen}.
In this form the underlying structure of
an anisotropic SU(3) spin chain in a
``$\lambda^8$ magnetic field'' is unexploited.
The correspondance between our notations and those of
Ref. \onlinecite{sen} can be found in table
\ref{correspondance1}.
The form of the Hamiltonian (\ref{heff1})
may help in relating
our model to integrable versions of the SU(3) spin chains. Isotropic
spin chains are known to be
integrable by Bethe Ansatz techniques\cite{uimin,sutherland}. The
magnetization process of SU(3)
spin chains with a magnetic field coupled to $\lambda^3$ has been
analyzed in the context of the bilinear-biquadratic spin 1 chain at
the integrable Uimin-Lai-Sutherland point
\cite{parkinson_uls_magfield,kiwata_uls_magfield,schmitt_uls_phase_diagram}
by solving numerically the Bethe Ansatz equations.
There exist also integrable anisotropic SU(3) spin-chains
\cite{maassarani_su(n)_xx}, but the chain described by the
Hamiltonian (\ref{heff1}) is not one of them.
To study the magnetization process and the correlation functions of
the chain we will therefore have to resort
to a combination of approximate methods such as bosonization,
renormalization group and strong coupling analysis.
We will, first, analyze the strong-coupling
effective Hamiltonian (\ref{heff}) to study the region of transitions
between plateaus of the magnetization, next we will discuss the
low-energy properties of the effective Hamiltonian (\ref{heff1}) via a
renormalization group analysis.
We conclude this section by contrasting the Open and Periodic
boundary conditions. The same strong coupling
analysis can be done for the OBC case. In contrast with the PBC
case, we have only a two fold degeneracy instead of three
at $J=0$ under a strong
field $h= \frac{7J_{\perp}}{8}$.
These two low energy states are
\begin{eqnarray}
|3/2; 3/2 \rangle &=& |\uparrow \uparrow \uparrow \rangle \nonumber \\
|1/2; 1/2 \rangle &=&\sqrt{\frac{2}{3}} \left(
\frac{1}{2} |\uparrow \uparrow \downarrow \rangle +
\frac{1}{2} |\downarrow \uparrow \uparrow \rangle-
|\uparrow \downarrow \uparrow \rangle.
\right)
\end{eqnarray}
The effective Hamiltonian to first order perturbation in $J/J_{\perp}$
becomes the well known spin-1/2 XXZ model,
\begin{equation}
H_{eff} = \frac{J}{4} \sum_i \left( \sigma ^x_i \sigma ^x_{i+1} +
\sigma^y_i \sigma^y_{i+1}
+ \Delta \sigma^z_i \sigma^z_{i+1} \right)-\left( \frac{h}{2}-
\frac{7J_{\perp}}{16} \right) \sum_i \sigma^z_i ,
\label{obc}
\end{equation}
where $\Delta = \frac{5}{18}$.
It is well known that this Hamiltonian has no gap for $\Delta \leq 1$
except at the saturated magnetization $\sigma^z = \pm 1$,
which corresponds to $m=1/2, 3/2$ in the original ladder model.
This agrees with a weak coupling analysis ($J >> J_{\perp}$)
based on bosonization \cite{schulz_berlin,cabra_unp}.
In the strong coupling analysis, one can clearly see the difference
between PBC and OBC in their effective Hamiltonian.
\section{Analysis of the magnetization process of an anisotropic
SU(3) chain}
We begin the analysis by discussing the generalization of
the Lieb, Schultz, Mattis
theorem to the SU(3) case\cite{lsm_theorem,affleck_lieb}. The theorem
allows us to predict
the possible locations of the plateaus. Comparing then these
predictions for a chain of SU(3) spins for the plateaus' location with
those
for 3 spin-1/2 chains we conclude that the mapping onto a
SU(3) spin chain does not introduce spurious plateaus, and should
therefore give physically correct results for $|h-h_c|\sim J$.
Next, we will discuss the magnetization process of an isotropic
SU(3) spin chain and address the question whether a cusp appears
in the magnetization versus magnetic
field curve (such a cusp is not related to magnetization plateaus).
A cusp has been observed in spin-1 chains with
bilinear-biquadratic exchange which can be mapped onto an SU(3)
spin chain for a special value of the biquadratic exchange.
We argue that in our case a cusp in the magnetization should not be expected.
Finally, We recall how the values of the magnetic field corresponding
to the plateaus\cite{sen} at $\langle S^z \rangle=1/2,3/2$ are
obtained from the anisotropic SU(3) spin chain.
\subsection{ The Lieb-Schultz-Mattis Theorem for a chain of SU(3)
spins}
Consider an anisotropic chain of SU(3) spins,
\begin{equation}\label{eq:generic_anisotropic_su3}
H=\sum_{i=1}^N \left[\sum_\alpha a_\alpha \lambda^\alpha_i
\lambda_{i+1}^\alpha + \beta (\lambda_i^3 \lambda_{i+1}^8+\lambda_i^8
\lambda_{i+1}^3) \right],
\end{equation}
where $a_1=a_2, a_4=a_5, a_6=a_7$, which we rewrite as,
\begin{eqnarray}
H & = & 2 \sum_i \left[ a_1
(T_{i,1}^+ T_{i+1,1}^- +T_{i,1}^- T_{i+1,1}^+)+a_4 (T_{i,2}^+
T_{i+1,2}^- +T_{i,2}^- T_{i+1,2}^+) +a_6 (T_{i,3}^+ T_{i+1,3}^- +T_{i,3}^-
T_{i+1,3}^+)+ \right .
\nonumber \\
\label{eq:hamiltonian_tuv}
& & \;\;\; + \left . 2a_3 \lambda_i^3\lambda_{i+1}^3 + 2a_8 \lambda_i^8
\lambda_{i+1}^8 + 2 \beta (\lambda_i^3 \lambda_{i+1}^8+\lambda_i^8
\lambda_{i+1}^3)\right] .
\end{eqnarray}
The anisotropic SU(3) chain we
are considering in this paper falls into this class of Hamiltonians.
The purpose of the LSM theorem is to classify possible gapless excitations
above the ground state.
Introduce the operators:
\begin{eqnarray}
U_3 & = & \exp\left(\imath \frac{2\pi}{N} \sum_{n=1}^N n \lambda_n^3\right)
\label{U3} \\
U_8 & = & \exp\left(\imath \frac{2\pi}{N\sqrt{3}}
\sum_{n=1}^N n \lambda_n^8\right)\label{U8}
\end{eqnarray}
\noindent and begin by studying the state $U_3| 0 \rangle$. We wish to
compare its energy expectation value with the vacuum's.
Consider therefore the
expression $\langle 0 | U_3^\dagger H U_3 -H | 0 \rangle$.
Using,
\begin{eqnarray}\label{eq:transformation_u3}
U_3^\dagger T_{n,1}^+ U_3& = & e^{2\imath
\frac{2\pi n}{N}} T_{n,1}^+ \nonumber \\
U_3^\dagger T_{n,2}^+ U_3 & = & e^{\imath
\frac{2\pi n}{N}}T_{n,2}^+ \nonumber \\
U_3^\dagger T_{n,3}^+ U_3 & = & e^{-\imath
\frac{2\pi n}{N}}T_{n,3}^+
\end{eqnarray}
we find upon developing it as a power
series in $1/N$ ($N$ is the size of the system), that the zeroth
order term vanishes, since it contains averages of the form,
$\imath \langle T_{n+1,p}^+ T_{n,p}^- -T_{n+1,p}^-
T_{n,p}^+\rangle$, which cancel when invariance under parity $m \to 2n+1-m$,
is used: $\langle T_{n+1,p}^+ T_{n,p}^- \rangle = \langle T_{n+1,p}^-
T_{n,p}^+\rangle$. Therefore, the expansion begins with a first
order term in $1/N$,
\begin{eqnarray}
\frac 1 N \times \left(-\frac {2\pi^2}{N} \sum_n \left[ 4 a_1 \langle
T_{n+1,1}^+
T_{n,1}^- + H. c. \rangle + a_2 \langle T_{n+1,2}^+
T_{n,2}^- + H. c. \rangle + a_3 \langle T_{n+1,3}^+
T_{n,3}^- + H. c. \rangle\right] \right) \nonumber
\end{eqnarray}
Translation invariance guarantees that the coefficient of $1/N$ in
this term is indeed finite. The translation operator $T$,
is defined by:
\begin{equation}
T^{-1}\lambda_n^\alpha T=\left \{\begin{array}{cc}
\lambda_{n+1}^\alpha & \mbox{ }\mathrm{if }\; n<N \\
\lambda_1^\alpha & \mbox{ }\mathrm{if }\; n=N, \\
\end{array}\right.
\end{equation}
and translation invariance, $T |0 \rangle= |0 \rangle$, (together
with the expression
of $U_3$, Eq.(\ref{U3})), implies:
$T U_3 |0 \rangle= T U_3 T^{-1} T |0 \rangle=e^{-2 \pi \imath
(\lambda_N^3-m_3)}| 0 \rangle$, where
$m_3=\frac 1 N \sum_{n=1}^N \lambda_n^3$.
Therefore, the state $U_3\mid 0 \rangle$ is orthogonal to the ground
state if $m_3$ is non-integer. This state has energy $O(1/N)$ above the
ground state, indicating either a ground state with a broken
symmetry or gapless excitations above the ground state for m
non-integer\cite{lsm,oshikawa}.
A gap in the excitation spectrum can only exist for $m_3=-1,0,1$ in the
absence of broken symmetry ground states.
For a ground state with $p$-site-periodicity
instead of one site translational symmetry,
\begin{equation}
T^p |0 \rangle = |0 \rangle,
\end{equation}
we obtain,
$T^p U_3 | 0 \rangle = e^{- 2 \pi \imath (\sum_{i=0} ^p
\lambda^3 _{N-i} - p m_3)}$ for a positive integer $p$.
A gap excitation can then
appear only for $m_3= q/p$ with an integer $q= -p, \cdots, p$.
If we now consider the action of $U_8$, we have:
\begin{eqnarray}\label{eq:transformation_u8}
U_8^\dagger T_{n,1}^+ U_8 & = & T_{n,1}^+ \nonumber \\
U_8^\dagger T_{n,2}^+ U_8 & = & e^{\imath
\frac{2\pi n}{N}}T_{n,2}^+ \nonumber \\
U_8^\dagger T_{n,3}^+ U_8 & = & e^{\imath
\frac{2\pi n}{N}}T_{n,3}^+
\end{eqnarray}
\noindent Once again, $\langle 0\mid U_8^\dagger H U_8 -H \mid
0\rangle = O(1/N)$, but this time: $TU_8\mid 0
\rangle=e^{-\imath \frac{2\pi}{\sqrt{3}}(\lambda_8^N-m_8)}U_8 \mid 0 \rangle$,
where $m_8=\frac 1 N \sum_{n=1}^N \lambda_n^8$.
This implies that a gapful excitation on a translational
invariant ground state is only possible for
$\frac{m_8}{\sqrt{3}}=n +1/3$, where $n$ is integer.
As in the case of $U_3$, a gap on $p$-site-periodic ground state
can exist for
$p \frac{m_8}{\sqrt{3}} =n + 1/3$.
As a consequence, two conditions have to be met to avoid having
gapless excitations above the translationally invariant ground state (p=1):
\begin{eqnarray}\label{eq:conditions_plateaus}
\frac{m_8}{\sqrt{3}}&=&n +1/3 \nonumber \\
m_3&=&n^\prime.\nonumber
\end{eqnarray}
The preceding discussion is quite general. In the specific case we are
considering, the magnetic excitations are described by
$S^z=5/6- \lambda_8/\sqrt{3}$. Therefore, we expect that the
magnetization plateaus are associated with the absence of excited
states above the ground states generated by $U_8$. This implies that
the only possible magnetization plateaus correspond to $\lambda_8$.
Clearly, such a conclusion could have been reached by considering
the LSM Theorem for a three chain system.
However, the LSM theorem for the SU(3) chain also indicates the possibility
of a gap for chiral excitations for $\langle \lambda_3 \rangle =
-1,0,1$. Because the Hamiltonian (\ref{h_0})--(\ref{h_1})
shows chiral symmetry, one has necessarily $\langle \lambda_3
\rangle=0$. Therefore, according to the LSM theorem, we are in a situation
where a gap in the chirality modes can be present and we should decide
whether this gap actually opens. Let us consider two simple limiting
cases.
For $\langle \lambda_8 \rangle/\sqrt{3}=1/3$,
i.e. $\langle S^z \rangle=1/2$, the only possible states on each site
of the SU(3) spin chain are $\mid \tilde 1 \rangle$ and$\mid \tilde 2
\rangle$. In that case, it can be easily seen that the
Hamiltonian (\ref{heff}) reduces to an effective XY chain \cite{sen},
implying gapless chiral excitations. On the other hand, if
$\langle \lambda_8 \rangle/\sqrt{3}=-2/3$, the only remaining mode on
each site is $|\tilde 3 \rangle$. One has indeed in this state
$\lambda_3=0$, but this time there is a gap in chiral excitations.
If there is a ground state with broken translational symmetry,
the situation is more complicated.
Therefore, the question of actual gap opening cannot be settled by the
LSM theorem alone.
\subsection{Comparison with the magnetization process of an isotropic
SU(3) spin chain}
The isotropic SU(3) spin chain is known to be integrable by the Bethe
Ansatz~\cite{uimin,sutherland}. It is also known that the bilinear
biquadratic spin-1 chain defined by the Hamiltonian:
\begin{equation}\label{eq:bilinear_biquad}
H=\sum_i \left[ \vec{S}_i.\vec{S}_{i+1} +\beta
(\vec{S}_i.\vec{S}_{i+1})^2 -h S^z_i \right]
\end{equation}
for $\beta=1$ (ULS point) can be mapped onto an isotropic SU(3) spin chain.
In the context of the bilinear-biquadratic spin chain, the
magnetization process of isotropic SU(3) spin chains has been
investigated in
detail by solving
numerically the Bethe Ansatz equations.
In that case, the magnetic field couples to $\lambda_3=S^z$.
A cusp in was obtained in
the magnetization as a function of the magnetic field\cite{parkinson_uls_magfield,kiwata_uls_magfield}. The
cusp is due to the fact that three different types of Bethe Ansatz
quasiparticles have respectively chemical potentials: $h,0,-h$. As a
result, when $h$ is large enough the band with the highest chemical
potential is emptied causing the cusp in the magnetization.
Also the effect of an
anisotropy $-D(S_i^z)^2 \propto \lambda^8$ on the bilinear biquadratic
spin-1 chain at the ULS point was studied \cite{schmitt_uls_phase_diagram}.
Rephrasing the results in the context
of SU(3) spin chains, it was found that when one applies a field that
couples to $\lambda^8$ (respectively $\lambda^3 +\frac
{\lambda^8}{\sqrt{3}}$, $\lambda^3 -\frac
{\lambda^8}{\sqrt{3}}$) the resulting average value of $\lambda^8$
(respectively $\lambda^3 +\frac
{\lambda^8}{\sqrt{3}}$, $\lambda^3 -\frac
{\lambda^8}{\sqrt{3}}$) shows no cusp as a function of the applied
magnetic field. The reason is that in this case, the Bethe ansatz
particles have chemical potentials $-h,-h,\;2h$ which prevent band
emptying effects.
In our case, similarly, the magnetic field couples to
$\lambda_8$. We thus conjecture that although
the anisotropy renders the system non-integrable, the absence of band
emptying effects should persist in the anisotropic case preventing any
cusp in the magnetization.
Also, on general grounds, the Hamiltonian (\ref{heff}) is invariant
under exchange of chiralities. Therefore, we expect to have $\langle
\lambda_3 \rangle=0$, whatever the applied magnetic field.
\subsection{Strong-coupling analysis of the effective Hamiltonian:
Magnetization plateaus}\label{magn}
In this section we will recall the evolution of the ground state
magnetization as a function of the external magnetic field. As noted
before, when the intrachain coupling $J$ is set to zero,
upon switching the external magnetic field on, we find
at increasing $h$ that
the ground state of a given rung undergoes a transition between the
spin-3/2 state, $|\uparrow \uparrow \uparrow\rangle$ and the spin-1/2 states,
$|\uparrow +\rangle$, $|\uparrow -\rangle$ at $h_c=\frac{3}{2} J_\perp$,
resulting in a sudden jump of the magnetization between $m=1/2$ and
$m=3/2$ (as shown in Fig.~\ref{fig:levels}).
If the coupling $J$ is non-zero but small this transition
is broadened between $h_{3/2,-}$ and $h_{1/2,+}$ which can be identified,
respectively, with the lower and upper critical fields of the saturation
plateaus of the magnetization at $m=3/2$ and $m=1/2$ in the
terminology of ref. \onlinecite{sen}. As noted there, it is easy to obtain
\begin{equation}
h_{3/2,-}=(\frac{3}{2} J_{\perp} +2J).
\end{equation}
by considering the condition for the ferromagnetic state $\mid \tilde
3 \tilde 3 \cdots \tilde 3 \rangle$ to be stable upon the
introduction of spin 1/2 states $| \tilde 1 \rangle$ or $|\tilde 2
\rangle$.
It is harder to determine the lower critical field $h_{1/2,+}$, below which
the magnetization plateau is at $m=1/2$. Indeed, on the
plateau $m=1/2$, there are two
possible states on each site, $| \tilde 1 \rangle$ or $|\tilde 2
\rangle$, so that this plateau is described by an
effective XY chain. As a consequence, the ground state wavefunction
for $m=1/2$, obtained from the Jordan-Wigner transformation is a
complicated linear combinations of states of the form $| \cdots
\tilde{2} \tilde{2} \tilde{1} \tilde{2} \tilde{1} \cdots \rangle$.
One has to consider the energy loss created by the introduction of a
$|\tilde 3\rangle$ state in that chain, and balance it with the energy
gained from the magnetic field. This problem bears some
similarity to the dynamics of a few holes
in a $t$-$J$ chain, which has an SU(2)$ \times $U(1) symmetry
instead of the U(1) $\times$ U(1) symmetry in our model.
The analogy with the $t$-$J$ model suggests
a two-component Luttinger liquid behavior of the system in a large
part of the phase diagram.
\section{bosonization and weak-coupling analysis}\label{weakcoupling}
We proceed now to study the long distance properties
of the effective Hamiltonian $H_{\text{eff.}}$ defined in Eq. (\ref{heff1}).
It is a sum of
an isotropic SU(3) spin chain Hamiltonian plus SU(3) symmetry
breaking terms,
\begin{eqnarray}
\label{htot}
H_{\text{eff.}}=\tilde H_0+\tilde H_1+\tilde H_2+\tilde H_3+\tilde H_h
\end{eqnarray}
\noindent with
\begin{eqnarray}
\tilde H_0&=&\frac{J}{4}
\sum_{i=1}^{N}\sum_{\alpha=1,\dots,8}\lambda_{i}^{\alpha}
\lambda_{i+1}^{\alpha} \label{su3ham} \\
\tilde H_1&=&q\sum_{i=1}^{N}\lbrack \lambda_{i}^1\lambda_{i+1}^1+
\lambda_{i}^2\lambda_{i+1}^2 \rbrack \label{lamb67}\\
\tilde H_2&=&u\sum_{i=1}^{N}\lambda_{i}^3\lambda_{i+1}^3 \label{lamb33} \\
\tilde H_3&=&u'\sum_{i=1}^{N}\lambda_{i}^8\lambda_{i+1}^8 \label{lamb88} \\
\tilde H_h&=&h_{eff}\sum_{i=1}^{N}\frac {\lambda_i^8}{\sqrt{3}} \label{lambh}
\end{eqnarray}
\noindent In our case, $q=-1/12 J$, $u=-1/4 J$,
$u'=-5/36 J$, and
$h_{eff}=\frac{h-h_c}{2}-\frac{5}{18}J$;
hereafter we choose our units so that $J=1$.
\subsection{Non-abelian bosonization of a SU(3) spin chain}
The $SU(3)$ invariant
Hamiltonian $\tilde H_0$ can be solved exactly by the Bethe
ansatz\cite{uimin,sutherland}. The solution shows that the SU(3) spin
chain has 2 branches of excitations, with dispersion
$\epsilon_j(k)=\frac J 4 \frac{2\pi}{\sin (\pi j/3)}[\cos(\pi j/3-
|k|)-\cos (\pi j/3)],\; \; j=1,2$. These excitations are gapless,
and for $|k| \to 0$, one has $\epsilon_1(k)=\epsilon_2(k)\simeq
\frac{2\pi}{3} \frac J 4 \mid k \mid$, i.e. the dispersion relation
assumes at long wavelength a massless relativistic form. Accordingly,
the low energy, long wavelength excitations of the $SU(3)$ spin chain
can be bosonized. More precisely, these excitations are
described\cite{aff_wz,affleck_su(n)} by the
SU(3) level 1 ($SU(3)_1$) Wess-Zumino-Novikov-Witten (WZNW)
model\cite{witten_wz}, perturbed by a marginally irrelevant $SU(3)$
invariant operator. A review of WZNW models can be found in
Ref. \onlinecite{tsvelikb}.
In Hamiltonian form, the $SU(3)_1$ model can be
written as:
\begin{equation}\label{eq:wznw_model_general}
H_{WZNW}=\frac{2\pi}{3}\sum_{a=1}^{3^2-1}
:J^a_R(x)J_R^a(x):+:J^a_L(x)J_L^a(x):
\end{equation}
where the right and left currents satisfy the following commutation
relations (Kac-Moody algebra at level 1):
\begin{equation}\label{eq:kac_moody_level1}
\lbrack J^{\alpha}_{R(L)}, J^{\beta}_{R(L)}\rbrack=if^{\alpha \beta \gamma}
\delta(x-y)J^{\gamma}_{R(L)}(y)+ \frac{i\delta_{\alpha \beta}}{2\pi}
\delta'(x-y).
\end{equation}
In Equation (\ref{eq:kac_moody_level1}), the $f^{\alpha \beta \gamma}$
are the structure constants of $SU(3)$.
The central charge is $C=\frac{1 \times (3^2-1)}{3+1}=2$, indicating
that the $SU(3)_1$ WZNW model can be described in terms of two free
bosonic fields.
As mentioned above, the $SU(3)$ spin chain is described asymptotically
by the $SU(3)_1$ model perturbed by a marginally irrelevant
SU(3) invariant term,
\begin{equation}
\label{generalham}
H \to H_{WZW}+g_0 \int \frac{dx}{2\pi} \Phi_0(x),
\end{equation}
where the marginal operator, $\Phi_0(x)=\sum_{\alpha=1}^{3} J_R^{\alpha}(x)J_L^{\alpha}( x)$,
couples the right and left currents.
The finite size correction to the
ground state energy of the $SU(3)$ chain can be obtained
from the Bethe Ansatz solution. These corrections are logarithmic and
are in agreement with those obtained from the continuum Hamiltonian
(\ref{generalham}).
This situation is very similar to the more familiar case of the
$SU(2)$ spin chain, which is described at low energy and long
wavelength by the marginally perturbed $SU(2)_1$ WZNW
model\cite{affleck_log_corr}.
In general, the magnitude of $g_0$ cannot be obtained from the lattice
Hamiltonian in the case of a $SU(N)$ spin chain (see the discussion of
the case $N=2$ in Ref. \onlinecite{affleck_log_corr}).
This is even more problematic when one adds perturbations to the $SU(3)$
invariant spin chain.
Another difficulty is that these perturbations
are not small in our case and strictly speaking
cannot be treated in perturbation theory.
However, in one dimension weak and strong
coupling behavior are often continuously
connected\cite{ziman_spin3/2,schulz_hubbard_exact,%
schulz_houches,kawakami_bethe_U<0} so that a weak coupling analysis
can provide very valuable information on the qualitative physics at
strong coupling.
Therefore, if we can find a weak coupling model that is described by
marginally perturbed $SU(3)_1$ WZNW model and if we add to it small
perturbations of the form (\ref{lamb67})--(\ref{lamb88}), we will be
able to make a reasonable guess of the low energy long wavelength
continuum theory associated with the Hamiltonian $H_{\text{eff.}}$.
By analogy with the Heisenberg model, we expect that the
difference between the weak and the strong coupling regime will reduce to a
renormalization of some parameters of the effective low energy
theory. For non-integrable models, these parameters can be obtained
numerically by calculating \emph{thermodynamic} quantities via
exact diagonalization methods
\cite{hayward_2chain,mila_zotos,ogata_tj}.
In our case, it is not difficult to see that the spin sector of the
$SU(3)$ Hubbard model \cite{affleck_su(n),assaraf_su(n)} is a good candidate for a
weak coupling model. This model is defined by the Hamiltonian:
\begin{equation}
H=-t \sum_{i,n=1,2,3}(c^\dagger_{i+1,n}c_{i,n} +
\text{h.c.}) + U\sum_{i,n \ne m} n_{i,n} n_{i,m}
\end{equation}
where $c_{i,n}$ annihilates a fermion of flavor $n \in
[1,2,3]$ on site $i$, and $n_{i,n}=c^\dagger_{i,n}c_{i,n}$.
The basic idea is that, starting from the
lattice Hamiltonian of the $SU(3)$ Hubbard model it is possible to
take the continuum
limit and then separate the spin excitations from the
charge excitations by means of weak coupling bosonization.
In the strong coupling limit, $U \to \infty$, a constraint of one
fermion per site is imposed,
\begin{equation}
\label{constraint}
\sum_{n}c^{\dagger}_{i,n}c_{i,n}=1.
\end{equation}
With one fermion
per site, the charge degrees of freedom are frozen out and one is left
only with SU(3) spin degrees of freedom.
Second order perturbation theory in $t$ then shows that the model can be mapped
onto an isotropic SU(3) spin chain with the lattice SU(3) spin operators
\begin{equation}\label{eq:abrikosov_su(3)}
\Lambda_i^{\alpha}=\sum_{n,m}
c^{\dagger}_{i,m}\lambda^{\alpha}_{n,m}c_{i,n},
\end{equation}
under the constraint (\ref{constraint}).
Under the hypothesis of
continuity, the same $SU(3)_1$ field theory should describe the weak
and strong coupling limits in the spin sector. The difference between
the weak and strong coupling limit corresponds to
the disappearance of the charge sector.
This reduction of the number of degrees of freedom
can be obtained in a consistent way by treating the constraint
(\ref{constraint}) within the
effective theory \cite{itoikato,itoimukaida}.
Thus, our strong coupling theory is the spin sector of the $SU(3)$
Hubbard model with a filling of one fermion per site and $U\to \infty$.
Let us discuss the weak coupling regime. The constraint
(\ref{constraint}) sets the Fermi momentum at
$k_F=\frac{\pi}{3}$ for the three fermion flavors. Since we are
interested in low-energy, long wavelength properties,
we linearize the spectrum for each flavor around the two Fermi points and
introduce
the right and left moving fermion modes in the continuum limit,
\begin{equation}
c_{i,n}^{\dagger} \simeq \sqrt{a}(e^{ik_Fx}\psi^{\dagger}
_{L,n}(x)+e^{-ik_Fx}\psi^{\dagger}_{R,n}(x)),
\end{equation}
\noindent where $x=ia$ and $a$ is the lattice spacing.
For $U=0$, the linearized Hamiltonian is:
\begin{equation}
H_{\text{linearized}}=-\imath v_F \sum_n \int dx (\psi^\dagger_{R,n}\partial_x \psi_{R,n} - \psi^\dagger_{L,n}\partial_x \psi_{L,n}).
\end{equation}
This Hamiltonian is conformal invariant and
can be rewritten in terms of the right and left
charge currents $J_{R(L)}=\sum_n \psi^\dagger_{R(L),n}\psi_{R(L),n}$,
and the eight SU(3) spin currents (right and left) $J^a_{R(L)}=\sum_n
\psi^\dagger_{R(L),n}\frac{\lambda^a_{n,m}}{2} \psi_{R(L),m}$.
One thus separates the charge and spin sectors:
\begin{equation}
H=H_{\text{charge}}+H_{\text{spin}}
\end{equation}
where the charge sector,
\begin{equation}
H_{\text{charge}}=v_F \int dx :J_R(x)J_R(x): + :J_L(x)J_L(x):
\end{equation}
and the spin sector is again described by the $SU(3)_1$ model discussed earlier,
\begin{equation}\label{ssuu}
H_{\text{spin}}=v_F \sum_a \int dx :J^a_R(x)J^a_R(x): + :J^a_L(x)J^a_L(x):
\end{equation}
The charge currents satisfy $U(1)$ Kac-Moody algebra, whereas the spin
currents satisfy the $SU(3)_1$ Kac-Moody algebra as can be checked
explicitly. When the interaction is weakly turned on,
$U/t\ll 1$, it
does not break spin charge separation but induces a $g_0 \propto
U$ term\cite{affleck_su(n)}.
We discussed thus far non-abelian bosonization in order
to stay close to the literature on $SU(3)$ spin chains. However, an
abelian bosonization approach to the isotropic $SU(3)$ spin chains
starting from the $SU(3)$ Hubbard model is perfectly
feasible. Such an
approach has been introduced for isotropic SU(N) spin chains
in Ref. \onlinecite{assaraf_su(n)}. It is
outlined in appendix \ref{app:abelian_hubbard_su(3)}.
In fact, for the rest of this section, we shall employ abelian
bosonization because it renders
the calculation of
correlation functions extremely easy even when the $SU(3)$ symmetry is
explicitly broken.
\subsection{Abelian bosonization approach}
Abelian
bosonization gives the following Hamiltonian for an $SU(3)$ invariant
spin chain
(or the spin sector of the $SU(3)$ Hubbard model):
\begin{eqnarray}
H_{su(3)} & = &\int \frac{dx}{2\pi} u\left[ (\pi \Pi_a)^2 +(\pi \Pi_b)^2 +
(\partial_x \phi_a)^2 +(\partial_x \phi_b)^2 \right]\nonumber \\
& + & \frac{2Ua}{(2
\pi a)^2}\int dx (\cos \sqrt{8}\phi_a +\cos \sqrt{2}(\phi_a +
\sqrt{3}\phi_b) + \cos \sqrt{2}(\phi_a -\sqrt{3}\phi_b) ) \nonumber \\
& - & Ua \int
\frac {dx}{\pi^2} \left[ (\partial_x \phi_a)^2 + (\partial_x \phi_b)^2
\right].
\end{eqnarray}
A derivation can be found in Appendix
\ref{app:abelian_hubbard_su(3)}.
The
free term corresponds to Eq.(\ref{ssuu}).
Under renormalization, $H_{su(3)}$ flows
to a fixed point Hamiltonian\cite{assaraf_su(n)}:
\begin{equation}
H^* = \int \frac{dx}{2\pi} u^*\left[ (\pi \Pi_a)^2 +(\pi \Pi_b)^2 +
(\partial_x \phi_a)^2 +(\partial_x \phi_b)^2 \right],
\end{equation}
where $u^*$ is given by the Bethe Ansatz as $u^*=\frac{2\pi}{3} \frac
J 4 $.
One can check using the expressions (\ref{eq:lambda_operators})
that this leads to a scaling dimension of $1$ for the
uniform component of $\Lambda^{\alpha}(x) \simeq a^{-1}\Lambda_i^{\alpha},\; x=ia $ ($\alpha=1 \cdots 8$),
and $2/3$ for the $2\pi/3$ component (see Eq. (\ref{eq:lambda_operators})).
These scaling dimensions coincide with
those obtained from non-abelian bosonization\cite{aff_wz,assaraf_su(n)}.
Turning now to the $SU(3)$ symmetry breaking terms, we find that
in the abelian bosonization representation they take the following form:
\begin{eqnarray}
\tilde H_1 & = & \int \frac {dx}{\pi^2} \left[ -\frac {qa}{4} (\pi
\Pi_a)^2 - \frac {qa} 2(\partial_x \phi_a)^2 +
\frac {qa}{12} (\partial_x \phi_b)^2 \right] +
\frac{2qa}{(2 \pi a)^2}\int dx \cos \sqrt{8} \phi_a + \frac{\sqrt{3}
qa}{(2\pi)^2} \int dx \partial_x \phi_b, \label{h11}\nonumber \\
\tilde H_2& = & \int \frac{dx}{\pi^2} \left[ \frac{5ua}{2}
(\partial_x \phi_a)^2 + \frac{ua}{6} (\partial_x \phi_b)^2 \right] +
\frac{2ua}{4(\pi a)^2} \int dx \cos (\sqrt{8}\phi_a) + \int
\frac{\sqrt{3}u}{(2\pi)^2} \partial_x \phi_b ,\label{h2}\\
\tilde H_3 & = & \int \frac{dx}{\pi^2} \left[
\frac{u'a}{6} (\partial_x \phi_a)^2 + \frac{5u'a}{2}(\partial_x \phi_b)^2 \right]
-\frac{2u'a}{3(2\pi a)^2} \int dx \cos (\sqrt{8}\phi_a)-
\nonumber \\
& &\; \; \; \; \; -\frac{4u^\prime a}{3(\pi a)^2} \int dx \cos \sqrt{2} \phi_a
\cos \sqrt{6} \phi_b + \frac{u^\prime \sqrt{2}}{3 \pi^2} \int
\partial_x \phi_b,\label{h3}\\
\tilde H_h&=& -(h-h_c)
\int dx \frac{\sqrt{2}}{\pi} (\partial_x \phi_b).\label{hh}
\end{eqnarray}
The physical interpretation of the terms proportional to $\partial_x
\phi_b$ is very simple. The bosonized Hamiltonian is derived under
the assumption that the magnetization per triangle is close to
$5/(6a)$. When the magnetization per triangle is \emph{exactly}
$5/(6a)$ the terms $\partial_x \phi_b$ do not appear in the
Hamiltonian. Therefore, the presence in the Hamiltonian of such terms
means that the magnetic field
needed to impose a magnetization of $5/(6a)$ per triangle is
renormalized away from its bare value. Also, since
the Hamiltonian preserves the
symmetry between $+$ and $-$ chiralities, it is invariant under the
transformation $\pi_a \to -\pi_a$, $\phi_a \to -\phi_a$. In
particular, this precludes the terms $\partial_x \phi_a$
in the Hamiltonian.
Assembling all terms we finally have the following
field theory describing the spin sector of the SU(3) Hubbard model in
the presence of symmetry breaking perturbations,
\begin{eqnarray}
H & = & v_F \sum_{i=a,b} \int \frac{dx}{2\pi} \lbrack
((\pi \Pi_i)^2+(\partial_x \phi_i)^2\rbrack+
+\frac{2g_1}{(2\pi a)^2}\int dx \cos \sqrt{8} \phi_a(x)\nonumber \\
& + &
\frac{4g_2}{(2\pi a)^2}\int dx \cos \sqrt{2} \phi_a(x)\cos \sqrt{6}
\phi_b(x) + \frac{g_4}{\pi^2} \int dx (\partial_x \phi_a)^2 +
\frac{g_5}{\pi^2} \int dx (\partial_x \phi_b)^2 +\frac{h}{\pi}\int
dx \partial_x \phi_b
\label{htott}
\end{eqnarray}
\noindent with $v_F=2ta\sin (k_Fa)=\sqrt{3}ta$,
$h=(\frac{\sqrt{3}}{4\pi}u+\frac{\sqrt{2}}{3\pi}u'+
\frac{\sqrt{2}}{2\pi}q)$, and $t$ the hopping
amplitude. In our units, $t=1$.
The notations can be made more compact by introducing the vectors
$\vec{\phi}=(\phi_a,\phi_b)$ and $\vec{\alpha}_1=(1,0)$ ,
$\vec{\alpha}_2=(1/2,\sqrt{3}/2)$ and
$\vec{\alpha}_3=(1/2,-\sqrt{3}/2)$, where,
\begin{eqnarray}
K_a &=&\left(\frac{1-\frac{qa}{2\pi v_F}}{1-\frac{Ua}{\pi
v_F}-\frac{qa}{\pi v_F}+\frac{u'a}{3\pi v_F}+\frac{5ua}{\pi
v_F}}\right)^{1/2}\nonumber \\
u_a &=&v_F \left[\left(1-\frac{qa}{2\pi v_F}\right)\left(1-\frac{Ua}{\pi
v_F}-\frac{qa}{\pi v_F}+\frac{u'a}{3\pi v_F}+\frac{5ua}{\pi
v_F}\right)\right]^{1/2}\nonumber \\
K_b &=&\left( \frac 1 {1-\frac{Ua}{\pi v_F}+\frac{5u'a}{\pi v_F}+\frac{ua}{3 \pi
v_F} +\frac{qa}{6\pi v_F}}\right)^{1/2}\nonumber \\
u_b &=&v_F\left(1-\frac{Ua}{\pi v_F}+\frac{5u'a}{\pi v_F}+\frac{ua}{3 \pi
v_F} +\frac{qa}{6\pi v_F}\right)^{1/2}.
\end{eqnarray}
The Hamiltonian can then be rewritten\cite{note_fusion_2d} as,
\begin{eqnarray}
H=\int \frac{dx}{2\pi}\left[u_a K_a (\pi
\Pi_a)^2+\frac{u_a}{K_a}(\partial_x \phi_a)^2 + u_b K_b(\pi
\Pi_b)^2 +\frac{u_b}{K_b}(\partial_x \phi_b)^2 \right]+
\sum_{i=1}^3 \frac{2 g_i}{(2\pi a)^2} \int dx \cos (\sqrt{8} \vec \alpha_i. \vec \phi) \nonumber
\end{eqnarray}
The interactions $(\partial_x \phi_{a,b})^2$ can render the
marginal operators $\cos (\sqrt{8}\phi_a)$,
$ \cos (\sqrt{2}\phi_a \pm \sqrt{6}\phi_b)$ marginally relevant and
cause the opening of a gap.
In such a case, the low energy properties of the system cannot be
described by two massless bosons. One can have either a massive and a
massless boson or two massive bosons.
This depends on the coupling constants $u$, $u'$, $m$, $q$, and the
magnetic field $h$. In order to explore this possibility in more
detail, one has to use renormalization group equations. This is the
subject of the forthcoming sections.
\section{The renormalization group flow in zero magnetic field}\label{RG}
In this section, we discuss the flow of the renormalization group
equation and the phase diagram that results from it. Qualitatively,
the renormalization group equations are similar to the
Kosterlitz-Thouless renormalization group
equations\cite{kosterlitz_renormalisation_xy,jose}. We expect
therefore to obtain a gapless phase corresponding to the
flow to a fixed hypersurface of the
6-dimensional space of coupling constants and one (or possibly many)
gapped phase where the coupling constants flow to infinity.
We also expect that the phase transition is of infinite
order\cite{kosterlitz_renormalisation_xy}.
Our task is therefore to determine the initial conditions and follow
the flow. This will allow us to conclude on the nature of
the ground state of the anisotropic SU(3) chain.
A straightforward application to the Hamiltonian (\ref{htott})
of the standard method
\cite{jose,giamarchi_logs} would be
inconvenient since one needs to expand to
third order of correlation functions in order to get the full one loop
RG equations\cite{nelson_fusion_2d}.
We will use instead operator product
expansion (OPE)
techniques\cite{kadanoff_gaussian_model,knops_sine-gordon}.
In our case, the algebra of operators $(\partial_x
\phi_a)^2$, $(\partial_x \phi_b)^2$ and $\cos (\sqrt{8} \vec \alpha_i .\vec
\phi)$ close under OPE (for details see Appendix \ref{app:derivation_ope}).
In particular,
\begin{eqnarray}
\label{opem}
\cos (\sqrt{8}\vec \alpha \vec \phi(x,\tau))\cos (\sqrt{8}\vec \alpha
\vec \phi(0,0)) & \simeq & \frac{-2a^4}{(x^2+(u\tau)^2)^2} \left
[ \sum_{p=a,b} (\alpha_i^p)^2
(x^2 (\partial_x \phi_p)^2 + \tau^2 (\partial_\tau \phi_p)^2 + 2 x\tau
\partial_x \phi_p \partial_\tau \phi_p \right],\nonumber \\
\cos(\sqrt{8}\vec \alpha \vec \phi(x,\tau))\cos(\sqrt{8}\vec \beta
\vec \phi(0,0))& & \simeq \frac 1 2
\left[ \left(\frac {a^2} {x^2+(u\tau)^2}\right)^{-K \vec \alpha.\vec
\beta}\cos \sqrt{8}(\vec \alpha +\vec \beta).\vec
\phi(0,0) + \right.\nonumber \\
& & \;\;\;\;\left. \left(\frac{a^2}{x^2+(u\tau)^2}\right)^{K \vec \alpha . \vec \beta}
\cos \sqrt{8}(\vec \alpha -\vec \beta).\vec \phi(0,0) \right],
\nonumber
\end{eqnarray}
lead to the following
RG equations (see appendix \ref{app:derivation_ope}):
\begin{eqnarray}
\frac{dy_1}{dl} & = & 2y_1y_4 - y_2^2/2 \nonumber \\
\frac{dy_2}{dl} & = & (1/2 y_4+3/2 y_5)y_2 - y_1y_2/2 \nonumber \\
\frac{dy_4}{dl} & = & y_1^2/2+y_2^2/4 \nonumber \\
\label{rge}
\frac{dy_5}{dl} & = & 3/4 y_2^2,
\end{eqnarray}
\noindent where we have denoted $y_i=g_i/\pi v_F$, with $v_F$ the Fermi
velocity. Note that is a fixed surface
$(g_1,g_2)=(0,0)$, because of three truly marginal operators
$\partial_x \phi_a \partial_x \phi_b$, $(\partial_x \phi_a)^2$ and $
(\partial_x \phi_b)^2$.
An alternative approach based on non-abelian bosonization can be used. In
this approach, having expressed the
Hamiltonian in terms of products of right and left moving currents
$J_R^a J_L^a$, an operator product expansion for
currents is derived\cite{itoi_rg_calculation}. Such an approach leads
to the same RG equations as the Abelian bosonization
approach.
The initial values of the running coupling constants (at the cut off
scale $a$)
for the spin sector of the SU(3) Hubbard model perturbed by $\tilde
H_{1,2,3}$ are given by,
\begin{eqnarray}
y_1(a)& = & \frac{g_1(a)}{\pi v_F}=(Ua +qa - \frac{u'a}{3} +ua)/\pi
v_F)
\nonumber \\
y_2(a)& = & \frac{g_2(a)}{\pi v_F} = (Ua +\frac{u'a}{3})/\pi v_F
\nonumber \\
y_4(a) & = & \frac{g_4(a)}{\pi v_F} = (-\frac{Ua}{2} +\frac{5ua}{2}+
\frac{qa}{2} + \frac{u'a}{6})/\pi v_F \nonumber \\
\label{starp}
y_5(a) & = & \frac{g_5(a)}{\pi v_F} = (-\frac{Ua}{2} +\frac{ua}{6} +
\frac{5u'a}{2} + \frac{qa}{12})/\pi v_F.
\end{eqnarray}
In the expression for the initial coupling constants (\ref{starp}) we
have $u=-J/4$, $q=-J/12$, $u'=-5J/36$, and we assume $J,U \ll t$.
Hereafter, we choose $J=4$
and put $v_F$ equal to unity, thus the numerical starting values are
given by
\begin{eqnarray}\label{eq:ini_cond}
y_1 & = & -0.365467+y_0 \nonumber \\
y_2 & = & -0.0589463 + y_0 \nonumber \\
y_4 & = & -0.878299 -y_0/2 \nonumber \\
y_5 & = & -0.5039907 -y_0/2.
\end{eqnarray}
Here $y_0 \propto U$.
These values are not small, so that the one loop RG equations are not
valid. However, numerically solving these RG equations with initial
conditions (\ref{eq:ini_cond}) shows that they
flow to a fixed point on the surface $(g_1,g_2)=(0,0)$
for any $y_0 \in [0,1]$ (see Figs. \ref{fig:rgflow_u=0} and
\ref{fig:rgflow_u=1}). At this fixed point, one has a renormalized
Hamiltonian with:
\begin{equation}
H=\int \frac {dx} {2\pi} \left[ u_a^* K_a^* (\pi \Pi_a)^2 +
\frac{u_a^*}{K_a^*} (\partial_x \phi_a)^2 + u_b^* K_b^* (\pi \Pi_a)^2 +
\frac{u_b^*}{K_b^*} (\partial_x \phi_a)^2 \right]
\end{equation}
This proves that certainly at weak coupling the long distance
properties of the system are described by a two component Luttinger
liquid. At strong coupling, i. e. in the case of the spin tube, this
is only an indication that the two component Luttinger liquid is
possible. In order to give a definitive proof, one should prove that
there is no singularity in the ground state energy as couplings
increase. Comparing Fig.~\ref{fig:rgflow_u=0}
and~\ref{fig:rgflow_u=1}, one can see that the magnitude of the fixed
point values of $g_4$ and $g_5$ depends on the strength of the
marginal SU(3) symmetric interaction $y_0$.
This fact, combined with the fact that the RG equations are only valid
at weak coupling precludes the use of the RG to give an accurate
estimate of $K_a^*$ and $K_b^*$. However, one can still determine from
the RG equations whether these quantities are larger or smaller than
one.
Although we stress that these figures should not be taken too
seriously,
we find using the expressions of $K_a$ and $K_b$ as a function of
$g_4$ and $g_5$ $K_a^*=1.9$ and $K_b^*=1.5$ i.e. both are larger than
1. Concerning the question
whether the two component Luttinger liquid persists at
large coupling, we can remark that the deviation from isotropy in our
case makes the interaction between the SU(3) spins less
antiferromagnetic. It is well known that in the case of the XXZ chain,
reducing the antiferromagnetic character of the spin spin interaction
(i.e. working at $J_z <J$ ) prevents the formation of a gap
\cite{haldane_xxzchain}. Therefore, it seems likely that no gap would
develop in the spectrum.
To test our conjecture, numerical work, especially
calculation of $K_a^*,K_b^*$ by exact diagonalization would prove
very valuable.
The existence of a two component Luttinger liquid phase has important
consequences. In particular, it implies a non-zero magnetic
susceptibility $\chi \propto K_b/u_b$, and a $T$ linear specific heat
of the form:
\begin{equation}
\label{eq:specific_heat}
C=\frac {\pi T}{6 u_a} +\frac{\pi T}{6 u_b} .
\end{equation}
The calculation of the correlation functions and NMR relaxation rate
are postponed to section \ref{correlations}.
\section{Strong magnetic field case: fixed point Hamiltonian and
correlation functions}\label{correlations}
\subsection{ Generic magnetic field}
\subsubsection{Renormalization group flow under a magnetic field}
Till now, we haven't taken into account the terms associated with the
magnetic field $h_b$, which can be treated as a
perturbation having fixed the external magnetic field at
$h_c=3/2J_{\perp}$. To see if the flow remains unchanged in
this case, let us reobtain the renormalization group equations with
finite $h$. The simplest way to address this problem is to perform a
Legendre transformation\cite{giamarchi_spin_flop} on the Hamiltonian
(\ref{htott}). The non-zero
average value of the field $\phi_b$ due to the finite magnetization can
be eliminated by a simple shift of the $\phi_b$ fields,
i.e. $\phi_{b}=\phi_{b}- \pi m_{b}x$, where $m_b=-\langle \partial_x
\phi_b \rangle /\pi $. One has the relation between $m_b$ and the
magnetization :
\begin{equation}
\label{eq:magnetization_mb}
m_b=-\sqrt{\frac 3 2}\left( \frac{\langle S^z \rangle}{a} -\frac 5 {6a}\right)
\end{equation}
The cosine terms, however,
are not invariant under this shift and the renormalization group
equations (\ref{rge}) for the couplings $g_1$, $g_2$ and $g_3$,
for a change of the length scale $a\rightarrow
e^{l} a$, now become (the details of the calculation can be found in
the appendix \ref{rghh}):
\begin{eqnarray}\label{betafh}
\frac{dy_1}{dl} & = & 2y_1y_4 -y_2^2 J_0(\pi m_b(l) a \sqrt{3})/2 \nonumber \\
\frac{dy_2}{dl} & = & (1/2 y_4+3/2 y_5)y_2 -y_1y_2 J_0(\pi m_b(l) a \frac{\sqrt{3}}2)/2 \nonumber \\
\frac{dy_4}{dl} & = & y_1^2/2+y_2^2 J_0(\pi m_b(l) a \frac{\sqrt{3}}2)/4 \nonumber \\
\frac{dy_5}{dl} & = & 3/4 y_2^2 J_0(\pi m_b(l) a \frac{\sqrt{3}}2),
\end{eqnarray}
\noindent where $J_0$ is the Bessel function that results from the use
of a sharp cutoff in the real space. One also has
$m_b(l)=m_b(0)e^l$. One can check that setting $m_b(0)=0$ one recovers
the equations (\ref{rge}).
If the RG equation for the magnetization is trivial,
the magnetic field on the other hand, satisfies a non-trivial RG
equation:
\begin{equation}
\frac{dh_b}{dl}=h_b+\frac{\sqrt{3}}{a \sqrt{8}}y_2^2 J_1(\pi \sqrt{6}
m_b(l) a).
\end{equation}
Let us discuss qualitatively the physics predicted by the
Eqs. (\ref{betafh}). One sees rather easily that for $m_b (l) a \ll
1$, the Bessel functions tend to zero, so that one is left with a
Sine-Gordon renormalization group equation for $y_1,y_4$.
Compared to the case of zero magnetization, we see that $y_4$ is more
negative and $y_1$ is smaller in absolute value. Therefore, we expect
that $y_1$ will be even more irrelevant in the presence of the finite
magnetization.
We conclude that the presence of a non-zero magnetization does not
affect the two component Luttinger liquid behavior.
The crossover scale can be roughly estimated as:
\begin{equation}
\label{lstar}
l^{\star} \simeq ln(\frac{v_F}{m_b a}).
\end{equation}
At this crossover scale, the flow of $y_5$ is completely
cut. This implies a variation of $K_a,K_b$ with the magnetization.
At the value
of $l$ given by (\ref{lstar}) the magnetic energy is of
the order of energy cut-off, therefore the magnetic field term
cannot be treated as a perturbation. When the initial magnetization
goes to
infinity the renormalization is stopped for smaller and smaller $l$.
The coupling constants $g_1,g_2,g_3$ then become
zero, while $g_4,g_5$ assume the values they have at the scale
$l^*$. Returning to the Hamiltonian (\ref{htott}), we see that it
becomes a quadratic Hamiltonian.
\subsubsection{Fixed point Hamiltonian}
Following the the preceding discussion, we conclude that
the asymptotic behavior of the three
chain system under a magnetic field is governed by the Hamiltonian:
\begin{eqnarray}
H^{\star}= \int \frac{dx}{2\pi} v_F \left[ (\pi \Pi_a)^2 + (\partial_x
\phi_a)^2 + (\pi \Pi_b)^2 +(\partial_x \tilde \phi_b)^2 \right]
+\frac{g^*_4}{\pi^2}\int dx (\partial_x \phi_a)^2
+\frac{g^*_5}{\pi^2}\int dx (\partial_x \phi_b)^2 \nonumber
\end{eqnarray}
\noindent where
$g_{4,5}^*$ are functions of the magnetic field.
The field $\tilde \phi_b$ is related to $\phi_b$ in the following way:
\begin{equation}
\phi_b = \tilde \phi_b+ \pi (m-\frac 5 {6a})\sqrt{\frac 3 2} x,
\label{shift}
\end{equation}
\noindent while the dual fields $\theta_a$ and $\theta_b$ are
not shifted. This condition guarantees that $\tilde \phi_b$ satisfies
periodic boundary conditions.
The fixed point Hamiltonian can be rewritten:
\begin{equation}
\label{hcrittt}
H^{\star}= \int \frac{dx}{2\pi} \left[ u^*_a K^*_a(\pi
\Pi_a)^2+\frac{u^*_a}{K^*_a}(\partial_x \phi_a)^2\right] +
\int \frac{dx}{2\pi} \left[ u^*_b K^*_b(\pi
\Pi_b)^2+\frac{u^*_b}{K^*_b}(\partial_x \tilde \phi_b)^2\right],
\end{equation}
\noindent where $u^*_iK^*_i=v_F; \; i=a,b$ and
$u^*_{a,b}/K^*_{a,b}=v_F +2 g_{4,5}^*/\pi$.
Both the velocities of excitations, $u_i$,
and the compactification radii, $K_i$, depend on the magnetic field $h$
through
$g_i^{\star}(h)$.
Therefore, the low energy properties of the system are described by
two decoupled $c=1$ conformal field theories with velocities and
compactification radii depending on the applied magnetic field.
This is valid at the level of perturbation theory for the
spin sector of the SU(3) Hubbard model. However, we are actually
interested in the SU(3) anisotropic spin chain for which perturbation
theory does not apply. In the latter case, we
expect, relying on the continuity between the weak and the strong
coupling regime, that the anisotropic SU(3) spin chain under magnetic
field will also be described by a two component Luttinger liquid.
However, the velocities and compactification radii cannot be
obtained by perturbation theory techniques.
Nevertheless, it is known that the velocities and compactification
radii can be obtained by calculating only thermodynamic quantities
using, for instance, exact diagonalization
techniques\cite{ogata_tj,mila_zotos}.
The problem of the determination of these exponents
in terms of measurable thermodynamic
quantities in the specific case of the anisotropic SU(3) spin chain
will be discussed in Appendix
\ref{app:exposants_bosonized}. The knowledge of the exponents then
permits the calculation of correlation functions. This is the subject
of the forthcoming section.
\subsubsection{Correlation functions}
In this section, we want to calculate the three Matsubara
correlation functions:
\begin{eqnarray}
\chi_{zz}(x,\tau)&=&\langle T_\tau S^z(x,\tau) S^z(0,0) \rangle \label{eq:totspin_corr}\\
\chi_{+-,p}(x,\tau)&=& \langle T_\tau S_p^+(x,\tau) S_p^-(0,0) \rangle \label{eq:locspin_corr}\\
\chi_{zz,p}(x,\tau)&= &\langle T_\tau S_p^z(x,\tau) S_p^z(0,0) \rangle \label{eq:locspin_corr_z},
\end{eqnarray}
where $p=1,2,3$ is a chain index.
The first correlation function is useful for neutron scattering
experiments, whereas the correlation functions
(\ref{eq:locspin_corr})
are useful for
the calculation of NMR relaxation rates.
The Matsubara correlation functions in Fourier space are given by:
\begin{equation}
\label{susc}
\chi_{ij}(q,\omega_n,T)= \int_0^\beta \;d\tau dx \exp^{\imath(\omega_n \tau - q x)}
\langle T_\tau \lbrack S^i(x,\tau),S^j(0,0)\rbrack \rangle_T,
\end{equation}
from which the finite temperature correlations are obtained by the
analytic continuation $\imath \omega_n \to \omega +\imath 0_+$.
We will first concentrate on the $T=0$
calculation, then explain how to extend the calculation to finite
temperature.
We begin with the calculation of $\chi_{zz}$. Using the equation
(\ref{siz}) we have:
\begin{equation}
\chi_{zz}=\frac 1 3 \langle T_\tau \Lambda^8(x,\tau) \Lambda^8(0,0)
\rangle +(\langle S^z\rangle)^2 .
\end{equation}
Using the bosonized expressions of the SU(3) spins,
Eq. (\ref{eq:lambda_operators}), and the usual expression for the bosonized
correlation
functions\cite{schulz_moriond}, we obtain\cite{note_logs_1}:
\begin{eqnarray}
\chi_{zz}& = & (\langle S^z\rangle)^2 +\frac {K_b} {3\pi^2}
\frac{(u_b\tau)^2-x^2}{(x^2+(u_b \tau)^2)^2} +
\frac {e^{\imath (\frac {2 \pi
x}{3a} -\pi (m -\frac 5 {6a}))x}} {6 (\pi a)^2} \left(\frac
{a^2}{x^2 +( u_a \tau)^2}\right)^{\frac {K_a}{6}} \left(\frac
{a^2}{x^2 +( u_b \tau)^2}\right)^{\frac {K_b}{2}} \nonumber \\
& &\;\;\;+ \frac {e^{\imath (\frac {2 \pi
x}{3a} +2\pi (m -\frac 5 {6a}) )x}} { 3(\pi a)^2}
\left(\frac
{a^2}{x^2 +( u_b \tau)^2}\right)^{\frac {2K_b}{3}} +\text{c.c},
\end{eqnarray}
where $m=\langle S^z \rangle /a$.
Turning to $\chi_{+-,p}$, it is easily seen using Eq. (\ref{si}) that
it is independent of
$p$ and equal to:
\begin{eqnarray}
\chi_{+-,p}(x,\tau)=\frac 1 {12} \left[ \langle T_\tau (\Lambda^1 +\imath
\Lambda^2)(x,\tau)(\Lambda^1 -\imath \Lambda^2)(0,0)\rangle + \langle
T_\tau (\Lambda^4 + \imath \Lambda^5) (x,\tau) (\Lambda^4 - \imath
\Lambda^5)(0,0) \rangle \right] \nonumber
\end{eqnarray}
Similarly, $\chi_{zz,p}$ is independent of $p$ (see Eq. (\ref{si})
and the contribution
not already included in $\chi_{zz}$ is of the form:
\begin{equation}
\langle T_\tau (\Lambda^6 + \imath \Lambda^7) (x,\tau) (\Lambda^6 - \imath
\Lambda^7)(0,0) \rangle.
\end{equation}
The expressions of the required correlators are obtained as:
\begin{eqnarray}\label{eq:general_correlation}
\langle T_\tau (\Lambda^n + \imath \Lambda^{n+1}) (x,\tau)(\Lambda^n - \imath
\Lambda^{n+1})(0,0) \rangle = \frac 2 {(\pi a)^2}\left[\left(\frac
{a^2}{x^2 +(u_a \tau)^2}\right)^{\nu_{n,1}} \left(\frac{a^2}{x^2 +(u_b
\tau)^2}\right)^{\nu_{n,2}} \right. \nonumber\\
\left. \times \cos (Q_n x + \Phi_n(\tau/x)) + \left(\frac
{a^2}{x^2 +(u_a \tau)^2}\right)^{\eta_{n,1}} \left(\frac{a^2}{x^2 +(u_b
\tau)^2}\right)^{\eta_{n,2}}
\cos (\frac {2\pi}{3a} + Q_n^\prime x +
\Psi_n(\tau/x)) \right]
\end{eqnarray}
where $n=1,4,6$.
The exponents are given by:
\begin{eqnarray}
\nu_{1,1}&=&\frac 1 {2K_a}+ \frac {K_a} 2 \; \nu_{1,2}=0 \\
\eta_{1,1}&=&\frac 1 {2K_a} \; \eta_{1,2}=\frac {K_b} 6 \\
\nu_{4,1}&=&\nu_{6,1}=\frac 1 {8 K_a} + \frac {K_a} 8 \;
\nu_{4,2}=\nu_{6,2}= \frac 3 {8K_b} + \frac {3 K_b} 8 \\
\eta_{4,1}&=&\eta_{6,1}=\frac 1 {8 K_a} + \frac {K_a} 8 \;
\eta_{4,2}=\eta_{6,2}=\frac 3 {8 K_b} + \frac {K_b} {24}.
\end{eqnarray}
It can be checked that for
$u_a=u_b$, $K_a=K_b=1$, one recovers the exponents of the isotropic SU(3) spin
chain\cite{aff_wz}, namely $\nu_{n,1}+\nu_{n,2}=1$ and
$\eta_{n,1}+\eta_{n,2}=2/3$.
One also has:
\begin{eqnarray}\label{eq:q_vectors}
Q_1&=&0, \;\; \;Q_1^\prime = -\pi \left(m -
\frac{5}{6a}\right)
\nonumber \\
Q_4&=&\frac 3 2 Q_1^\prime=-Q_6,\;\; \;\;\; Q_4^\prime = -\frac
{Q_1^\prime}{2}=-Q_6^\prime.
\end{eqnarray}
Recall $\langle S^z \rangle=5/6 +\sqrt{2/3} m_b$.
This allows the determination of all incommensurate modes.
Finally, the functions:
\begin{eqnarray}
\Phi_1\left(\frac \tau x\right)&=&2\arctan \left(\frac{u_a \tau}
x\right), \; \; \Psi_1\left(\frac \tau x\right)=0 \nonumber \\
\Phi_4\left(\frac \tau x\right)&=&\Phi_6\left(\frac \tau x\right)=\frac
1 2 \arctan \left(\frac{u_a \tau} x\right) +\frac 3 2 \arctan
\left(\frac{u_b \tau} x\right) \nonumber \\
\Psi_4\left(\frac \tau x\right)&=&-\Psi_6\left(\frac \tau x\right)=\frac
1 2 \left[\arctan \left(\frac{u_a \tau} x\right)-\arctan
\left(\frac{u_b \tau} x\right)\right]
\end{eqnarray}
All the preceding results are valid only at $T=0$. However, it is
useful also to
calculate the correlation functions for $T>0$,
in particular in order to obtain NMR relaxation rates.
To obtain the finite temperature Matsubara correlation functions,
we can use a conformal transformation since we have two decoupled
$c=1$ conformal field theories. The explicit expression of this
transformation is:
\begin{equation}\label{eq:conformal_mapping}
x+\imath u_i \tau \to \beta u_i \sinh \left( \frac {2\pi (x+ \imath
u_i \tau)} {\beta u_i} \right),
\end{equation}
where $i=a,b$.
Therefore, to obtain\cite{note_logs_2} the finite temperature Matsubara correlation
functions, one has to make the substitution:
\begin{eqnarray}
\label{eq:conf_substitution}
x^2+(u_i \tau)^2 &\to& (\beta u_i)^2 \left[ \cosh^2 \left(\frac {2\pi x} {\beta
u_i}\right) - \cos^2 \left( \frac {2\pi \tau} \beta \right) \right]
\nonumber \\
\arctan \left(\frac{u_i \tau} x \right) &\to&
\arctan\left(\frac{\tan\left(2\pi \frac \tau \beta\right)}{\tanh \left(
2 \pi \frac x {\beta u_i}\right)}\right).
\end{eqnarray}
With the help of the above results for the spin-spin correlation
functions we can evaluate the $T$ dependence of the NMR
longitudinal relaxation rate $T_1$,
\begin{equation}
\frac{1}{T_1} \propto \lim_{\omega_n \rightarrow 0}\int_0^\beta d\tau
e^{\imath \omega_n \tau} \langle T_\tau S^+_p(0,\tau) S_p^-(0,0) \rangle_T.
\end{equation}
We find,
\begin{eqnarray}
\frac 1 {T_1} \propto (a~ T^{\frac 1 {2K_a} +\frac{K_a} 2 -1} +b~ T^{\frac 1
{2K_a} +\frac {K_b} 6} +c ~T^{\frac{3 K_a+ K_b} {24} + \frac 1 {8 K_a} + \frac
3 {8 K_b} -1} +d ~T^{\frac 1 {8K_a} + \frac 3 {8K_b} + \frac {3 K_b +
K_a} 8 -1 } ) \nonumber
\end{eqnarray}
The low temperature exponent is the smallest
of the the four exponents above.
\subsubsection{Comparison with a spin-1 chain with biquadratic coupling}
In the case of a bilinear biquadratic spin-1 chain defined in
Eq. (\ref{eq:bilinear_biquad}), close to the
Uimin-Lai-Sutherland point ($\beta \simeq 1$),
a mapping onto an anisotropic SU(3) spin
chain is also possible\cite{fath_biquadratic}.
However, there are important differences.
First, the expression of the spin operators in terms of the
Gell-mann matrices is different from the ones obtained in
the spin-tube case.
For the spin-1 case, one has:
\begin{eqnarray}\label{eq:spin1_lambda}
S_n^x& = & \frac{\sqrt{2}}{2}\left(\lambda_n^4+\lambda_n^6\right) \nonumber
\\
S_n^y& = & \frac{\sqrt{2}}{2}\left(\lambda_n^5-\lambda_n^7\right) \nonumber
\\
S_n^z& = &\lambda_n^3.
\end{eqnarray}
These expressions should be contrasted with Eqs.(\ref{si})--(\ref{siz}).
\noindent Although the expressions (\ref{eq:spin1_lambda})
lead to incommensurate modes, the expression of the
correlation functions is different from the case of the spin tube.
Second, the expression of the Hamiltonian in terms of $\lambda$
matrices in the spin-1 case is different from the expression
(\ref{heff}). Namely, the Hamiltonian (\ref{eq:bilinear_biquad}) can
be rewritten in terms of Gell-Mann matrices as:
\begin{eqnarray}
\label{eq:bilin_biquad_su(3)}
H& = &\sum_i \left[\frac \beta 2 (\lambda^8_i \lambda^8_{i+1} +\lambda^1_i
\lambda^1_{i+1} + \lambda^2_i
\lambda^2_{i+1}) + (1-\frac \beta 2) \lambda_i^3
\lambda_{i+1}^3 \right.\nonumber \\
& +& \left. \frac 1 2 (\lambda^4_i \lambda^4_{i+1}+ \lambda^5_i \lambda^5_{i+1} +
\lambda^6_i \lambda^6_{i+1} + \lambda^7_i \lambda^7_{i+1}) + \frac
{1-\beta} 2 (\lambda^4_{i} \lambda^6_{i+1} + \lambda^6_{i}
\lambda^4_{i+1}- \lambda^5_{i} \lambda^7_{i+1} -\lambda^7_{i}
\lambda^5_{i+1})\right]
\end{eqnarray}
Finally, for $\beta<1$ the spin-1 bilinear biquadratic chain has a gap
and the two component Luttinger liquid can only be observed for a
large enough applied magnetic field.
Nevertheless, the two problems have in common the
presence of a gapless two component Luttinger liquid ground
state\cite{fath_biquadratic}, and
the formation of incommensurate modes under a magnetic field, so that
loosely speaking they belong to the same universality class.
This can be understood as a consequence of the fact that both models
can be related to anisotropic SU(3) spin chains.
One should note that the formation of incommensurate modes in the
presence of the magnetic field in the spin-tube
is not related to the presence of \emph{gapped} incommensurate modes
in the bilinear-biquadratic spin-1 chain
\cite{golinelli_incommensurate}. In the latter case, the
incommensurate modes originate from the fact that in the absence of
the biquadratic chain, the (gapped) modes of the spin-1 chain are at $q=0$ and
$q=\pi$ whereas at the ULS point the (gapless) modes are at $q=0$ and
$q=2\pi/3$. The presence of gapped incommensurate modes between these
two limits is merely a consequence of the continuity of the transition
between the Haldane gap phase and the gapless phase beyond the ULS
point. On the other hand, in the presence of the magnetic field, the
gapless modes of the spin-tube or those of the spin-1 chain simply
move away from $2\pi/3$ similarly to what happens in a single
spin-1/2 chain.
\subsection{ Is there a magnetization plateau for $\langle S^z
\rangle=1$ ?}
\subsubsection{ The Umklapp terms and quantization condition on the magnetization}
\medskip
In the presence of a magnetic field, one of the central issues is the
quantization condition on the total magnetization $\langle S^z\rangle$ for the
appearance of plateaus. This condition may be investigated by looking
at the bosonized expression for the
spin-operators (\ref{eq:lambda_operators}). After using the transformation
(\ref{transf}) to take into account a non-zero magnetization, one can
rederive an expression for the non-SU(3) symmetric perturbations.
Contrary to the case of zero magnetization, we cannot assume
\emph{a priori} that the ``$4k_F$'' terms are highly oscillating since
the phase $e^{4\imath k_F x}$ may be compensated by a phase arising
from the transformation (\ref{transf}). This can be interpreted as a
condition for Umklapp processes between the three different fermion flavors.
A systematic investigation indicates that the only possible Umklapp
terms originate from the term
$\lambda^3_i\lambda^3_{i+1}$
or $\lambda_i^8\lambda_{i+1}^8$ in the Hamiltonian (\ref{htot}). These
Umklapp terms are:
\begin{eqnarray}
\cos(\sqrt{8}\phi_a-2\sqrt{\frac{3}{2}} \tilde \phi_b +\alpha_1) \label{umk1}
\\
\cos(\sqrt{8}\phi_a+2\sqrt{\frac{3}{2}}\tilde \phi_b + \alpha_2 )\label{umk2}\\
\cos(4\sqrt{\frac{2}{3}}\tilde \phi_b + \alpha_3),\label{umk3}
\end{eqnarray}
\noindent The conditions that must be fulfilled so that these Umklapp
terms be present are:
\begin{equation}\label{eq:condition_umk1}
\langle \Lambda^3\rangle-\frac{\langle \Lambda^8\rangle}{\sqrt{3}}=\frac{1}{3} \mbox{ }\mbox{or}
\mbox{ }-\frac{2}{3},
\end{equation}
for the term (\ref{umk1}),
\begin{equation}\label{eq:condition_umk2}
\langle \Lambda^3\rangle+\frac{\langle \Lambda^8\rangle}{\sqrt{3}}=\frac{1}{3} \mbox{ }\mbox{or}
\mbox{ }-\frac{2}{3},
\end{equation}
for the term (\ref{umk2}), and
\begin{equation}\label{eq:condition_umk3}
\frac{\langle \Lambda^8\rangle}{\sqrt{3}}=-\frac{1}{6} \mbox{ }\mbox{or} \mbox{ }
\frac{1}{3},
\end{equation}
for the term (\ref{umk3}).
If we take into account the fact that $\langle \Lambda_3 \rangle=0$,
there are only two nontrivial conditions to obtain an Umklapp.
These are the condition (\ref{eq:condition_umk2}) that reduces to
$\langle S^z \rangle = 7/6$, and the condition (\ref{eq:condition_umk3}),
reducing to $\langle S^z \rangle=1$.
At these values of the magnetization the
operators (respectively) (\ref{umk2}) or (\ref{umk3})
can in principle open a gap in the excitations of the system
leading to a plateau in the magnetization curve. In agreement with
the LSM theorem, this implies a ground state that breaks
translation invariance.
For this to happen, these operators must be relevant, namely,
\begin{equation}\label{eq:umk2_relevant}
(K_a+9K_b/4)<1
\end{equation}
for the term (\ref{umk2}), and
\begin{equation}
\label{eq:umk3_relevant}
K_b<3/4
\end{equation}
for the term (\ref{umk3}).
If one of these conditions is satisfied, a gap exists
at $\langle S^z \rangle=5/6$ more likely than at $\langle S^z
\rangle=1$ or $\langle S^z \rangle=7/6$.
\subsubsection{A theory of the possible plateau at $\langle S^z
\rangle =1$}
There seems to be evidence for an extra plateau at $\langle S^z
\rangle=1$
in the magnetization curve of a 36 sites system at $J_\perp/J=3$
obtained by DMRG in Ref.~\onlinecite{sen}
(see Figure~10 of Ref.~\onlinecite{sen}).
However, this small plateau
observed at $\langle S^z \rangle =1 $ in the magnetization curve may
also be ascribed to the small system size\cite{diptiman_sen_pc}. Here,
we will investigate in more details the possibility of such a plateau in
the framework of the bosonized theory. In particular, we will try to give a
description of the behavior of correlation functions in the system.
Due to the presence of the Umklapp term (\ref{umk3}) the bosonized
Hamiltonian describing the low energy excitations of the spin tube at
$\langle S^z \rangle =1 $ is:
\begin{eqnarray}
\label{eq:hamiltonian_m=1}
H&=& H_a + H_b \nonumber \\
H_a &=& \int dx \frac {dx}{2\pi} \left[ u_a K_a (\pi \Pi_a)^2 +
\frac{u_a}{K_a} (\partial_x\phi_a)^2 \right] + \frac{2g_1}{(2\pi a)^2}
\int dx \cos \sqrt{8} \phi_a \nonumber \\
H_b&=& \int dx \frac {dx}{2\pi} \left[ u_b K_b (\pi \Pi_b)^2 +
\frac{u_b}{K_b} (\partial_x\phi_b)^2 \right] + \frac{2u'a}{3 (\pi a)^2}
\int dx \cos \left( 4 \sqrt{\frac 2 3} \phi_b + \frac{2\pi} 3\right)
\end{eqnarray}
If there is indeed a magnetization plateau at $\langle S^z
\rangle=1$, a gap opens in the excitation spectrum of the field
$\phi_b$. However, the Hamiltonian contains no terms coupling
$\phi_a$ and $\phi_b$, so that $\phi_a$ could remain ungapped.
This would lead to a single component
Luttinger liquid behavior on this plateau, and a power law decay of
some spin-spin correlation functions.
Since $u' <0$, the Umklapp term would impose $\sqrt{\frac 2 3}\langle
\tilde \phi_b \rangle = -\frac{2\pi} 3$.
If this is so, careful treatment of the expressions
(\ref{eq:lambda_operators}) of the bosonized forms of the $\Lambda$
operators is necessary when we have a gap,
on the $\langle S^z \rangle=1$ plateau.
To eliminate gapped states completely, one can
use the following expression (see appendix \ref{app:abelian_hubbard_su(3)}).
\begin{equation}
\Lambda^8 _i = \frac{1}{\sqrt{3}}(1 -3 c_{i,3}^\dag c_{i,3}),
\end{equation}
for $S^z _i$ instead of (\ref{eq:abrikosov_su(3)}) here we use
the constraint (\ref{constraint}). This equation indicates
we have only the gapful excitations $\phi_b$ in $\Lambda^8(x)$.
One has
\begin{eqnarray}
\label{eq:lambda_op_m=1}
(\Lambda^1+\imath \Lambda^2)(x)&=&\frac{e^{\imath \sqrt{2}\theta_a}}{\pi
a}\left[ 2\cos \sqrt{2} \phi_a + 2 C_1 \cos \left( \frac \pi {2a} x
+\frac \pi 6\right) \right] \nonumber \\
\Lambda^8(x)&=& \frac{3} {\pi a \sqrt{3}} e^{\imath \frac \pi a x} C_2
\end{eqnarray}
Where:
\begin{eqnarray}
C_1=\langle e^{\imath \sqrt{ \frac 2 3} ( \tilde \phi_b - \langle
\tilde \phi_b \rangle)}\rangle.
\nonumber \\
C_2=\langle e^{\imath 2 \sqrt{ \frac 2 3} ( \tilde \phi_b - \langle
\tilde \phi_b \rangle)} \rangle
\end{eqnarray}
We do not give the expressions for the other operators since they
show exponential decay
except for $\Lambda^3$,
never enter
spin correlation function.
We find then that translational symmetry is broken on the
$\langle S^z \rangle=1$ plateau, with a period of $4$ for the ground
state.
This is in agreement with the LSM theorem that rules out a
translationally invariant plateau at $\langle S^z \rangle=1$.
We also see that an additional period of $4$ will appear in the
correlation functions showing a power law decay.
The correlation functions for the operators of
Eq. (\ref{eq:lambda_op_m=1}) is:
\begin{eqnarray}
&&\langle T_\tau (\Lambda^1 +\imath \Lambda^2)(x,\tau) (\Lambda^1
-\imath \Lambda^2)(x',0)\rangle = \frac 1 {(\pi a)^2} \left[ 2C_1^2
\left\{ (1+\frac 1 2 e^{\imath \frac{\pi x'} a}) \cos \left[\frac \pi
{2a}(x-x')\right] \right. \right. \;\;\;\;\;\;\;\;\;\;\; \nonumber \\
&& \;\;\;\;\;\;\;\;\;\;\;-\left. \left. \frac{\sqrt{3}} 2 e^{\imath \frac {\pi x'} a} \sin
\left[\frac \pi {2a} (x-x')\right]\right\} \left(\frac {a^2} {(x-x')^2
+(u_a^* \tau)^2}\right)^{\frac 1 {2K_a}} \right. \nonumber \\
&&\;\;\;\;\;\;\;\;\;\;\;- \left. \left( \frac{a^2}
{(x-x')^2+(u_a^* \tau)^2}\right)^{\frac {K_a} 2 + \frac 1 {2K_a}}
\frac{(x-x')^2-(u_a^* \tau)^2}{(x-x')^2+(u_a^* \tau)^2}\right] \nonumber
\end{eqnarray}
Using the
expressions (\ref{si}) and (\ref{siz}) of the spins in terms of
$\lambda$ matrices we find that the correlations
$\langle S^+_n S^-_{n'}\rangle$
show an exponential decay whereas the correlations $\langle S^z_{n,p} S^z_{n',p} \rangle$ follow
a power law decay.
We have the following expressions for the equal time spin-spin correlation
functions:
\begin{eqnarray}
\label{eq:correlation_m=1}
&&\langle S_n^z \rangle = 1- \frac{C_2}{\pi} e^{\imath \pi n}
\nonumber \\
&&\langle S_{n,p}^z S_{n',p}^z \rangle - \langle S_{n,p}^z \rangle \langle
S_{n',p}^z \rangle=\frac 2 9
\langle (\Lambda_n^1 +\imath
\Lambda_n^2)(\Lambda_{n'}^1 -\imath \Lambda_{n'}^2) \rangle
\end{eqnarray}
Comparing with
figures (6) and (7) of Ref. \onlinecite{sen}, one sees that such behavior
is not obtained in numerical calculations. This leaves two options:
one is that the system size (36 sites) in Ref. \onlinecite{sen} is too small to
observe the finite but large correlation length. This is not unreasonable,
since $K_b$ could be only slightly smaller than $3/4$. The second
possibility is that the plateau at $\langle S^z \rangle
=1$ is an artifact of the small
system size. In Sec. \ref{dmrg_m=1} it will be shown using DMRG for systems
of up to 120 sites that it is the latter possibility that is obtained.
If we wish to obtain non-trivial plateaus smaller
values of Luttinger parameters $K_a$ or $K_b$ are needed.
This could be achieved by adding a sufficiently strong
antiferromagnetic Ising
term along the chain,
\begin{equation}
\sum_{i=1} ^N \sum_{p=1} ^3 S^z _{i,p} S^z_{i+1,p}
\end{equation}
or an extra coupling,
\begin{equation}
\sum_{i=1} ^N \sum_{p=1} ^3 \sum_{q=1} ^3\vec S_{i,p} \vec S_{i+1,q}.
\end{equation}
The extra plateaus would lie
at $\langle S^z \rangle = 5/6, \ 1, \ 7/6$ and are allowed by an
extended LSM theorem in the case of a periodic ground state. \\
\subsubsection{Zero gap for $\langle S^z \rangle =1$ and finite size
scaling of DMRG results}\label{dmrg_m=1}
We continue the density matrix renormalization group\cite{white,dmrg0}
study of the three leg ladder at $\langle S^z \rangle =1$ given by Tandon
\textit{et al.} \cite{sen}. Their results for finite chain length show
there might be a plateau at $\langle S^z \rangle =1$ but the system
size is too small to draw definitive conclusions\cite{diptiman_sen_pc}. In this
section, we show that the apparent plateau at $\langle S^z \rangle
=1$ is indeed a finite size artifact.
In finite size study, there is a finite energy gap
between any two nondegenerate energy levels. We therefore use $1/N$
scaling to show that the energy gap scales to zero in the thermodynamic
limit for low energy excitations with $\Delta S^{tot,z}=0$ and
$\Delta S^{tot,z}=1$, respectively.
The system is not dimerized since there is no gap in both
$\Delta S^{tot,z}=0$ and $\Delta S^{tot,z}=1$ excitations. Besides the
information that there is a certain kind of gapless excitations
obtained from LSM, numerical analysis gives information on whether
$\Delta S^{tot,z}=0$ and $\Delta S^{tot,z}=1$ excitations are gapped, and
shows whether the system is dimerized. We conclude from both the
DMRG calculation and the RG analysis that $\Delta S^{tot,z}=1$
excitations are gapless for $J_\perp$ much smaller than $J$ to
$J_\perp$ bigger than $J$ and there is no magnetization plateau at
$\langle S^z \rangle =1$.
We use periodic boundary condition for the rung and open boundary
condition for the other direction, the $i$ direction in
Eq.(\ref{ham}), in our DMRG calculation. We calculate the lowest
energy states for $S^z = N-2$, $N-1$, $N$, $N+1$, $N+2$ and denote
the lowest energy in each $S^z$ sector as $E_0(S_z)$ and the second
lowest energy as $E_1(S_z)$. We have set $h=0$ in Eq.(\ref{ham})
and we use $J_\perp / J =3$ to calculate $E_0(S_z)$ and $E_1(S_z)$.
In DMRG we keep $200$ optimized states and biggest truncation error is
$10^{-6}$. We make calculations for even chain length from $N=4$ to $N=40$.
For each length $N$ we calculate the $\Delta S^{tot,z}=0$ gap by
$E_1(S_z)-E_0(S_z)$, and the $\Delta S^{tot,z}=1$ gap by
$E_0(S_z-1)+E_0(S_z+1)-2 E_0(S_z)$. For $m=1$, we have plot
$E_1(S_z)-E_0(S_z)$ vis $1/N$ and $E_0(S_z-1)+E_0(S_z+1)-2 E_0(S_z)$
vis $1/N$ at magnetization $S_z=N$ in Fig.\ref{fig:gapscaling}. The
fittings to the second order of $1/N$ in the figure show that the gaps
scale to zero in form $E_1(S_z)-E_0(S_z)\sim 1/N$ and
$E_0(S_z-1)+E_0(S_z+1)-2 E_0(S_z) \sim 1/N$ when $N\to\infty$.
We can see the zero gap excitations also by analyzing the spectrum.
Around $\langle S^z \rangle =1$ at length $N$ and $S_z$ we will have
$3N/2-S_z$ doublets given in Eq.(\ref{doublets}). When $3N/2-S_z$
is odd the ground state is doubly degenerate due to the permutation
symmetry of the two kinds of the doublets and the two ground states
have parity $-$ respect to the $i$ to $N+1-i$ reflection symmetry.
When $3N/2-S_z$ is even, the ground state is unique, but such unique
ground states for $N$ and for $N+4$ have different reflection parity.
It suggests that there is no gap. There is no such a parity change
from $N$ to $N+4$ for ground states of gapped translational invariant
systems or dimerized system. It supports the basic picture in
previous sections: when we increase or decrease $S_z$, we put in or
take out gapless quasiparticles (doublets here), these gapless
quasiparticles have different parity on different energy levels.
These parity and degeneracy can be obtained by further detailed analysis
of the low energy excitations of the Luttinger liquid.\cite{parity}
We conclude that there is no plateau at $\langle S^z \rangle =1$ since
there is no gap for $\Delta S^{tot,z}=1$ excitations. To obtain
numerically the non-trivial plateaus, we need smaller value of
Luttinger parameters $K_a$ or $K_b$. If we add a sufficiently strong
Ising term along the chain direction (in $J$), the condition is
satisfied and we can obtain such plateaus at
$\langle S^z \rangle =5/6$, $1$, and $7/6$ as we predicted out in
previous subsections. These plateaus are
allowed by an extended LSM theorem in the case of periodic ground state
with a certain periodicity, as explained in section III.
DMRG calculation for $\langle S^z \rangle =5/6$ and $7/6$ asks for
more technique effort\cite{dmrg2}. Calculating numerically the
Luttinger liquid exponents\cite{dmrg3} is also important to fulfill
the understanding of trileg ladder in future studies.
\section{Conclusions}\label{conclusions}
In this paper, we have analyzed the strong coupling limit of the three
chain system with periodic boundary conditions in the presence of a
magnetic field,
using a mapping on an
anisotropic SU(3) chain. A straightforward extension of the LSM
theorem allowed us to locate the possible magnetization
plateaus. Then, we applied bosonization and renormalization group
techniques to show that for $1/2<m<3/2$, the system would be described
by a two component Luttinger liquid. This allowed us to obtain the
spin-spin correlation functions of the system, and to follow the
position of the various incommensurate modes in the spin
spin-correlation function as a function of the magnetization. Finally,
we have predicted the temperature dependence of the NMR relaxation
rate in this region.
We have also discussed the relation of the two component Luttinger
liquid we obtained with the two component Luttinger liquid phase
of the bilinear biquadratic spin-1 chain, and shown that despite
the similarity of the two problems, the two phases are not related to
each other.
We considered the evidence for a magnetization plateau
at $m=1$ in the framework of our bosonized description, and concluded
that if such a plateau exists, the ground state at $m=1$ should break
translational symmetry with a period of two lattice spacing. We
obtained the correlation functions of such a ground state as well as
the average value of the magnetization at each site, and found
that if such a plateau does exist the ground state would exhibit some
kind of antiferromagnetic order. The numerical simulation of
Ref.~\onlinecite{sen} shows no evidence for such antiferromagnetic
order thus pointing to an absence of plateau at $m=1$.
To clarify whether or not there is a plateau at $m=1$,
we calculate energy gap around $m=1$ using DMRG method
and fit the data
in the linear function of the inverse system size.
No gap was found in the thermodynamic limit,
therefore, we conclude that there is no plateau at $m=1$.
One obvious direction to extend our work is to calculate numerically
the Luttinger liquid exponents of the three chain ladder with periodic
boundary conditions. In appendix~\ref{app:exposants_bosonized},
indications can
be found on how these exponents could in principle be extracted. The study
of these exponents in the case of anisotropic
integrable SU(3) spin chains would also be interesting.
The present paper has been almost entirely concerned with the limit
$J_\perp\gg J$, and it would be interesting to investigate whether the
behavior we have obtained is valid also in the opposite limit,
$J_\perp \ll J$. It would also be interesting to study anisotropic
generalizations of the spin-tube model and check for plateaus at
$m=1$ as well as extend the analysis to higher n spin-tubes.
Finally, the relation between the model we have considered and
classical statistical systems such as Coulomb gases may be worth
analyzing in more details.
\section{Acknowledgments}
C.I., would like to thank I. Affleck
for kind hospitality in University of British Columbia extended to him.
E. O. acknowledges support from NSF under grant
NSF DMR 96-14999. R. Citro acknowledges financial support from
Universit\`a degli Studi di Salerno. The various authors are grateful
to I. Affleck, R. Chitra, M. Oshikawa and Paul Zinn-Justin for many useful
and enlightening
discussions. E. O. thanks D. Sen for e-mail correspondance on the plateaus
problem.
\vfill
\noindent $^{\star}$ {\footnotesize On leave from: Dipartimento di Scienze
Fisiche ``E. Caianiello'', Universit\`a di Salerno and Unit\`a INFM di
Salerno, 84081 Baronissi (Sa), Italy} \newline
\noindent $^{\star \star}$ {\footnotesize On leave from: Department
of Physics, Nihon University, Kanda Surugadai, Tokyo 101, Japan}
\noindent $^{\star \star \star}$ {\footnotesize On leave from:
Institute of Theoretical Physics, CAS, Beijing 100080, P R China}
\newpage
|
\section{Introduction}
The aim of this paper is to classify the structure of the solution set
of embedded Nielsen-Olesen vortices for general gauge theories. To do
this requires the concepts of embedded defects, first formally
introduced in \cite{Vach94}. In many ways this work improves on and
extends the initial treatment of vortex classification attempted in
\cite{me1}, and the companion paper to that \cite{me2} contains many
illustrative examples of the formalism discussed herewithin.
This first section is devoted to a quick review of gauge theories
coupled to a scalar field sector. This will serve to set the scene and
establish the notation for the later sections of this paper where we
shall classify vortex solutions.
We emphasize in particular how the symmetry breaking relates to the group
structure of gauge theory.
Fermions will be excluded from the discussion, so
we shall only have to consider the effects of scalar-gauge
interaction. This is appropriate because
fermions seem to merely modify the form of solution, introducing
fermionic zero modes {\em around} the background scalar-gauge
configuration, whereas here we are interested in the
background configuration itself, which is determined by the {\em
global} topographical features of the gauge theory and its symmetry
breaking.
\subsection{Yang-Mills Theories Coupled to a Scalar Field}
We are concerned with field theories whose basic dynamical variables
are gauge potentials $A_\mu$ and scalar fields $\Phi$.
Interaction of the scalar field $\Phi$ with the gauge potential
$A_\mu$ is specified by the gauge symmetry group $G$, a compact Lie
group acting upon the scalar field via the representation $D$ of
$G$. We consider theories of the form described by the
Lagrangian~(\ref{lag}) below. Once the gauge group and its action upon
the scalar field are specified the theory is completely determined
up to the strength of scalar field self couplings and the gauge
coupling constants.
The gauge potential $A_\mu$ lies in the Lie algebra of $G$, which we
denote by $\calG$. The field tensor of the gauge potential is
\begin{equation}
F_{\mu \nu} := \partial_\mu A_\nu - \partial_\nu A_\mu + [A_\mu,
A_\nu] \in \calG.
\end{equation}
The local actions of the gauge group upon the scalar field and gauge
potential take the form (for $g(x) \in G$)
\bse
\begin{eqnarray}
\label{eq-gauget-a}
\Phi(x) \mapsto \Phi'(x) &=& D(g(x)) \Phi(x), \\
\label{eq-gauget-b}
A_\mu(x) \mapsto {A'_\mu}(x) &=& g(x) A_\mu(x) g(x)^{-1}
- (\partial_\mu g(x)) g(x)^{-1}.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
The field tensor transforms under a similarity transformation
\begin{equation}
F_{\mu \nu}(x) \mapsto F_{\mu \nu}'(x) = g(x) F_{\mu \nu}(x)
g(x)^{-1}.
\end{equation}
The covariant derivative of $\Phi$,
\begin{equation}
\label{cd}
D_{\mu} \Phi = \left( \partial_\mu + d(A_\mu) \right) \Phi,\\
\end{equation}
transforms according to
\begin{equation}
D_{\mu} \Phi(x) \mapsto (D_{\mu} \Phi)'(x) = D(g(x)) D_{\mu} \Phi(x).
\end{equation}
Here $d(X)$ is the derived representation of $D$, describing
how $\calG$ acts on $\Phi$ by the relation $D(e^X) = e^{d(X)}$.
We need inner products on $\calG$ and on the
space $\calV$ of $\Phi$ values, both of which\footnote{Note that we are
using the same symbol for inner products on
$\calG$ and $\calV$; we hope it should be clear from
the context which inner product we are considering.}
must be invariant under the actions of
the group $G$: on $\calV$
$\inprod{D(g)\Phi}{D(g)\Phi}=\inprod{\Phi}{\Phi}$, while on $\calG$,
$\inprod{\mb{{\rm Ad}}(g)X}{\mb{{\rm Ad}}(g)X}=\inprod{X}{X}$, where $\mb{{\rm Ad}}(.)$ is the adjoint
action of $G$ on $\calG$ defined by $ge^Xg^{-1}=e^{\mb{{\rm Ad}}(g)X}$. Suitable
forms and properties of these are defined in Appendix B.
Assembling the above, the minimal gauge-invariant Lagrangian
describing the gauge picture interaction of a scalar field with a
gauge potential is:
\begin{equation}
\label{lag}
\calL[\Phi, A_\mu] = -\frac{1}{4} \inprod{F_{\mu \nu}}{F^{\mu \nu}}
+ \frac{1}{2} \inprod{D_\mu \Phi}{D^\mu \Phi} - V[\Phi].
\end{equation}
Here $V[\phi]$ is the scalar potential, describing self interaction of
the scalar field, and is constrained to be invariant under local
(gauge) transformations\footnote{Sensible quantisation also requires
the potential to be a fourth order (or less) polynomial.}.
The gauge coupling constants manifest themselves in an interesting way.
The Lie algebra $\calG$ has a natural decomposition into
commuting subalgebras
\begin{equation}
\label{ideals}
\calG = \calG_1 \oplus \cdots \oplus \calG_n,
\end{equation}
with each $\calG_f$ either simple or one-dimensional. Thus an $\mb{{\rm Ad}}(G)$
invariant inner product on $\calG$ has $n$ scales, related
to the norm on each $\calG_f$. With respect to this inner product the
unit norm generators are written $q_f X_f$; these $q_f$'s are
interpreted as the gauge coupling constants appertaining to $\calG_f$,
as explicitly illustrated in sec.~(\ref{Weinberg-Salam}).
Furthermore, taking the limit $q_f \rightarrow 0$ makes the symmetry
$\calG_f$ global. For a unit norm generator $q_f X_f$, one sees
that the effect of taking $q_f \rightarrow 0$ is to decouple the gauge
field in Eq.~(\ref{cd}), rendering $\calG_f$ global.
\subsection{Symmetry Breaking}
The form of the scalar potential $V$ may cause the fully symmetric theory
to be unstable.
The trivial background vacuum takes field values $\Phi(x)=0$ and
$A_\mu(x)=0$, and a perturbative quantum field theory around it has full
gauge symmetry $G$. However, this may {\em not} be the global energy
minimum among all background field configurations. In such a situation
the unstable background will decay to the stable background vacuum, where
$\Phi(x)=\Phi_0\ne 0$. Such a background does not respect the full
symmetries of the original theory, but only a compact subgroup $H$. In
this case, the theory may admit solitonic scalar-gauge configurations
that asymptotically tend to the stable background. In this paper we
are primarily interested in the spectrum and classification of these
configurations.
It is the form of the scalar potential $V[\Phi]$ that determines
whether, and how, the gauge symmetry is spontaneously or dynamically
broken. The stable background vacuum is the one in which $\Phi(x) =
\Phi_0$, $A_\mu(x) = 0$, where $\Phi_0$ is a {\em global} minimum of the
scalar potential. Then the quantum field theory with this vacuum is
obtained by perturbing around $\Phi_0$, and is described by the Lagrangian
\begin{equation}
\calL_H [\Psi, A_\mu] = \calL_G[\Phi_0 + \Psi, A_\mu],
\end{equation}
which has residual gauge symmetry
\begin{equation}
H = \{h \in G : D(h) \Phi_0 = \Phi_0\}.
\end{equation}
The quantum field theory contains a spectrum of massive
gauge bosons corresponding to those symmetries that are broken; these
gauge bosons have internal directions in ${\calM}$, where
\begin{equation}
\calG = \calH \oplus {\calM},
\end{equation}
the direct sum being defined by the inner product
$\inprod{\cdot}{\cdot}$.
Because the theory has gauge symmetry $G$ (about the trivial vacuum),
the global minimum of the scalar potential generally has degenerate
values, forming a compact manifold. This is the vacuum manifold,
\begin{equation}
\label{vacuum manifold}
M = D(G) \Phi_0 \cong G/H;
\end{equation}
it is the shape of this manifold that determines the non-perturbative,
solitonic spectrum of the scalar-gauge theory. The coset space $G/H$
is homogenous, and at the identity element has tangent space ${\calM}$. The
group $H$ acts as rotations on points of $G/H$ around the identity
element. This picture is echoed in the space $D(G) \Phi_0 \subset \calV$,
where the tangent space at $\Phi_0$ is $d({\calM})\Phi_0$ and $D(H)$ acts as
rotations of points of $D(G)\Phi_0$ around $\Phi_0$. One thus interprets
${\calM}$ as vectors that {\em move} $\Phi_0$, and $H$ as actions that rotate
{\em around} $\Phi_0$.
The set of all symmetries of the vacuum manifold form the isometry
group
\begin{equation}
I=\{a \in {\rm aut}(\calV) : a M = M\},
\end{equation}
which is generally larger than $G$. This has a corresponding induced
representation $\tilde{D}$ of $I$, reducing to $D$ of $G$, and a
subgroup
\begin{equation}
J=\{j \in I : {\tilde D}(j) \Phi_0 = \Phi_0\},
\end{equation}
containing $H$. The vacuum manifold may then be re-expressed
\begin{equation}
M \cong \frac{G}{H} \cong \frac{I}{J}.
\end{equation}
In some cases these additional elements of $I$ represent {\em
hidden} global symmetries of the theory; we discuss the $SU(2)
\rightarrow 1$ example in Sec.~(\ref{sec-su2}). In other cases,
however, the additional elements are {\em not} symmetries of the
theory. An example is the electroweak model, discussed in
Sec.~(\ref{Weinberg-Salam}), where $I=SU(2) \mb{\times} SU(2)$ is {\em
not} a symmetry of the model unless ${\sin} \theta_w = 0$. However $M$
is isomorphic to the symmetric space $I/J \cong S^3$.
In what follows we shall assume
that all hidden symmetries have been included in the symmetry group
$G$.
\section{Vortices}
\label{sec-2}
This is the pivotal section of the paper: here we describe how
vortices are classified; whereas the rest of the paper will
be used to explain and establish these results.
The tactic is to consider the archetypal Abelian-Higgs vortex, and
then to {\em embed} it into a larger theory --- one may then examine
the associated gauge freedom by moving it continuously around within the
larger theory. This will naturally partition the set of all vortex
solutions into {\em families} of gauge equivalent vortices. We will find
that this process is equivalent to a partitioning of the tangent space
to the vacuum manifold,
${\calM}$, in a very natural way; this equivalence being
provided by a natural association between vortices and vectors in
${\calM}$. Thus by explicitly partitioning ${\calM}$ into its constituent parts a
classification of vortices is achieved.
Our classification of vortices is on a level separate to the question of
stability. Classification relies on the group {\em actions} on the
tangent space, whereas stability relies on the {\em
topology}\footnote{Semi-local, or dynamical, stability is slightly more
subtle; we discuss this later} of the vacuum manifold.
\subsection{The Abelian-Higgs Vortex}
This is the archetypal model of a vortex, the Nielsen-Olesen vortex,
within the simplest theory that contains such vortices.
For simplicity, we consider only straight, static vortices, and impose
cylindrical symmetry.
The Abelian-Higgs model consists of a one-dimensional complex scalar
field coupled to a $U(1)$ gauge potential $A_\mu$, with a symmetry
breaking Landau potential. (Note that, to conform with our general
notation, we take the gauge potential $A_\mu$ to be pure imaginary.) The
Lagrangian is of the form
\begin{equation}
\calL[\Phi, A_\mu] = -\frac{1}{4} F_{\mu \nu}^\star F^{\mu \nu}
+ \frac{1}{2} D_\mu \Phi^\star D^\mu \Phi - V[\Phi],
\end{equation}
with
\bse
\begin{eqnarray}
F_{\mu \nu} &=& \partial_\mu A_\nu - \partial_\nu A_\mu,\\
D_\mu \Phi &=& \left( \partial_\mu + q A_\mu \right) \Phi,\\
V[\Phi] &=& \frac{1}{2}\la(\Phi^\star \Phi - \eta^2)^2.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
Here $\la > 0$ and $\eta$ are parameters of the potential, and $q$ the
gauge coupling constant. For $\eta^2 >0$ the trivial vacuum is
unstable and decays to a vacuum of the form $\Phi_0 = \eta
e^{i\chi}$. Perturbations around this comprise a field theory
consisting of a massive gauge field interacting with a massive scalar
field; this theory having no manifest gauge symmetry. All this
information is encoded within the symmetry breaking
\begin{equation}
U(1) \rightarrow 1.
\end{equation}
The gauge symmetry breaking leads to the existence of vortex
solutions. In the temporal gauge $A_0 =0$, they have the form
of the Nielsen-Olesen Ansatz:
\bse
\begin{eqnarray}
\label{eq-NOa}
\Phi(r, \theta) &=& \eta f_{\rm NO}(r) e^{in \theta},\\
\label{eq-NOb}
\ul{A}(r,\theta) &=& \frac{g_{\rm NO}(r)}{r} \left( \frac{in}{q}
\right) \ul{\hat{\theta}},
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
with boundary conditions
\begin{equation}
f_{\rm NO}(0)=g_{\rm NO}(0)=0
, \ \ {\rm and}\ \ f_{\rm NO}(\infty)=g_{\rm NO}(\infty) =1,
\end{equation}
Asymptotically the configurations wind around the possible degenerate
vacua, and then over the intervening space the fields continuously
interpolate. The $n= \pm 1$ solutions are stable because there is no
way they may be {\em continuously} deformed to the vacuum
$\Phi(x)=\Phi_0$ whilst keeping the asymptotic field within the vacuum
manifold. The asymptotic field configuration, which forms a circle in
the space of scalar field values, winds around the spatial circle at
infinity --- since this map has topological degree it may not be
continuously deformed to other inequivalent situations.
\subsection{Vortices in General Gauge Theories}
We now embed the archetypal $U(1) \rightarrow 1$ vortex
(\ref{eq-NOa},\ref{eq-NOb}) within a larger theory:
\begin{eqnarray}
G &\rightarrow& H\nonumber \\
\cup &\ & \cup \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ H \cap U(1) = 1\\
U(1) &\rightarrow& 1.\nonumber
\end{eqnarray}
The general vortex Ansatz is
\bse
\begin{eqnarray}
\label{eq-vortex-a}
\Phi(r, \theta) &=& f_{\rm NO}(r) D(e^{X\theta})\Phi_0, \\
\label{eq-vortex-b}
\ul{A}(r,\theta) &=& \frac{g_{\rm NO}(r)}{r} X \ul{\hat{\theta}}.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
where $X\in {\calM}$ is the generator of the embedded $U(1)$.
In this paper we shall classify vortex solutions of this form.
It should be remarked that the Ansatz
(\ref{eq-vortex-a}, \ref{eq-vortex-b}) is not the most general we
could assume. The lowest-energy configuration with these boundary
conditions may in some cases be distorted from the
pure embedded-vortex solution, though the distortions
are irrelevant for the classification.
Asymptotically the scalar field winds around a geodesic
$D(e^{X\theta})\Phi_0$ on the vacuum manifold $D(G) \Phi_0$, and
nearer the core the solution continuously interpolates over a two
dimensional surface within $\calV$ containing this geodesic.
There could be components of the scalar field $\Phi$ out of this
surface that vanish at infinity but are non-zero in the core
\cite{andy97}. However, such components do not affect the
classification of the vortex solutions, so we may ignore them here.
The asymptotic gauge field in the minimum-energy configuration is in
the direction of $X\in \cal M$, because if we were to add any
component in $\cal H$, the effect would be to leave the scalar field
unchanged but to add an extra term to the gauge-field
magnetic energy. Though one should note that when gauge field takes
components within more than one simple part of $G$ the various
components may have different radial dependence, and so near the
core there may be components in $\cal H$. Again, however, these
components in $\cal H$ do not affect the classification.
\subsubsection{Vortex Generators}
We shall always consider vortices in the temporal gauge, and assume
cylindrical symmetry, {\em i.e.} their field values are cylindrically
symmetric, and stationary in time. Non-trivial gauge transformations that
respect this gauge are the {\em global} (or rigid) gauge
transformations where $g(x)$ in Eq.~(\ref{eq-gauget-a},\ref{eq-gauget-b})
is independent of $x$.
By Eq.~(\ref{vacuum manifold}) we can use the gauge freedom of $G$ to
fix $\Phi_0$. Then vortex solutions are defined by just one variable,
the vortex generator $X\in {\calM}$. To classify vortex solutions, we must
answer the questions: which generators $X \in {\calM}$ define vortex
solutions? And which of those are gauge equivalent?
The remaining gauge freedom corresponds to global transformations by
elements $h\in H$, the residual symmetry group. Two vortices
described by Ans\"atze of the form (\ref{eq-vortex-a},
\ref{eq-vortex-b}) are gauge equivalent if and only if they are
related by such a transformation. Since
$\Phi_0$ is unaltered, the only effect is to transform the generator $X
\in {\calM}$:
\begin{equation}
X \mapsto \mb{{\rm Ad}}(h) X, \ \ \ {\rm with}\ h \in H,
\end{equation}
where $\mb{{\rm Ad}}$ is the adjoint representation of $G$ acting on $\calG$.
Geometrically, the scalar boundary
conditions of a vortex describe a geodesic $D(e^{X\theta})\Phi_0$
on the vacuum manifold, with the gauge potential being defined from
the tangent vector to the corresponding curve $e^{X\theta}$ in $G$. Global
gauge transformations defined by elements of $H$ keep the point $\Phi_0$
fixed, and rotate the geodesic around this point. Thus vortex generators
that may be rotated into one another by $\mb{{\rm Ad}}(H)$ define gauge equivalent
vortices.
There are two constraints that the vortex Ansatz must
obey~\cite{barr92}:\\
(i) Closure. The asymptotic scalar field forms a closed geodesic
with $D(e^{2 \pi X}) \Phi_0 = \Phi_0$. Equivalently, $e^{2 \pi X}\in
H$.\\
(ii) The Ansatz is a solution to the equations of motion. An argument
of \cite{Vach94} identifies this to be when: fields in the vortex
do not induce currents perpendicular (in Lie algebra space) to it. This
can be shown to be equivalent to (see appendix A):
\begin{quote}
\em if $X$ is a vortex generator, then for all $X^{\perp}$ such that
$\inprod{X}{X^\perp} = 0$ one has
$\inprod{d(X)\Phi_0}{d(X^\perp)\Phi_0}=0$.
\end{quote}
In other words the map $X \mapsto d(X)\Phi_0$ is {\em conformal} over
the classes of generators that define vortex solutions.
\subsubsection{Family Structure of Vortex Solutions}
Consider the action of $\mb{{\rm Ad}}(H)$ on ${\calM}$. In general, ${\calM}$ may be {\em
reducible}; let us write
\begin{equation}
\label{eq-decomp}
{\calM} = {\calM}_1 \oplus \cdots \oplus {\calM}_n
\end{equation}
where ${\calM}_i$ is (real) irreducible under the action of $\mb{{\rm Ad}}(H)$. In
Appendix B we show that the (real) inner products on $\calV$ and
$\calM$ are related upon this decomposition by:
\begin{eqnarray}
\label{crucial}
\inprod{d(X_i)\Phi_0}{d(Y_j)\Phi_0} &=& \la_i \la_j
\inprod{X_i}{Y_j}, \ X_i \in {\calM}_i, Y_j \in {\calM}_j, \\
{\rm where}\ \la_i &=&
\frac{\norm{d(X_i)\Phi_0}}{\norm{X_i}}\nonumber,
\end{eqnarray}
and $\la_i$ is constant upon its particular ${\calM}_i$.
Comparing Eq.~(\ref{crucial}) with condition (ii) above we see that
generally, vortex generators lie
only in the individual ${\calM}_i$'s, but for some pairs of values of gauge
coupling constants, defined by the condition,
\begin{equation}
\la_i(q_1,\cdots ,q_n) = \frac{\norm{d(X_i)\Phi_0}}{\norm{X_i}} =
\frac{\norm{d(X_j)\Phi_0}}{\norm{X_j}} = \la_j(q_1,\cdots ,q_n),
\end{equation}
the
solution space increases, encompassing {\em combinations} of vortices
with generators in ${\calM}_i$ and vortices with generators in ${\calM}_j$.
Then the vortex generators are those generators that close in ${\calM}_i
\oplus {\calM}_j$; such vortices we dub {\em combination vortices}.
Leaving aside the special case of combination vortices, we may
concentrate on vortices described by generators $X$ in one specific
irreducible subspace ${\calM}_i$. First, we show that every $X$ is
equivalent to one lying in a subspace that plays the same role as a
Cartan subalgebra in a Lie algebra.
For each ${\calM}_i$ one may define a {\em rank} analogous to the rank of a
Lie group: the rank $r$ of ${\calM}_i$ is the maximum number of linearly
independent commuting generators in ${\calM}_i$. Let $T_1,\ldots,T_r$ be
such a set, and consider the algebra, $\calT$ say, that they generate.
It is an Abelian subalgebra of $\calG$, and $T=\exp\calT$ is an Abelian
subgroup of $G$, a torus.
Then, in general we conjecture that vortex generators are given by the
following crucial result:
\begin{quote}\em
for ${\calM}_i$ of rank $r$, we can find a set of linearly independent
commuting generators $\{T_1, \ldots, T_r\}$ such that the $X \in {\calM}_i$
satisfying the closure condition may be written
\begin{equation}
\label{result}
X = \mb{{\rm Ad}}(h) \sum_{k=1}^{r} l_k T_k,
\end{equation}
with $h \in H$, and $l_k\in{\bf Z}$.
\end{quote}
In sec.~(\ref{sec-proof}) we prove the above result for two categories
of vacuum manifolds: symmetric spaces, and symmetric spaces that are
modified to admit semi-local vortices. Practically, these two
situations cover most cases of physical relevance, although we suspect
that the result is true generally.
\subsubsection{Gauge Equivalent Embedded Vortices}
Each of the vortices labelled by the set of coefficients $\{l_j\}$ has
an orbit of gauge equivalent solutions associated with it,
\bse
\begin{eqnarray}
\mb{{\rm Ad}}(H) \sum_{k=1}^{r} l_k T_k \cong H/H_{\{ l_j\}},
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \\
{\rm where}\ \ \ H_{\{ l_j\}} = \{h\in H :
\mb{{\rm Ad}}(h)\sum_{k=1}^{r} l_k T_k = \sum_{k=1}^{r} l_k T_k\}.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
Generally these orbits differ in structure depending upon which
element of the lattice of coefficients is considered.
In addition, these orbits may contain vortex generators
with different sets of coefficients $\{l_j\}$ that are gauge
equivalent. In fact,
$H$ acts as a discrete transformation group on the lattice $\ul L$. We
can construct this group as follows.
First, define the subgroup of $H$ that maps $\calT$ into itself:
\begin{equation}
N({\calT}) = \{h\in H: \mb{{\rm Ad}}(h){\calT}={\calT}\}.
\end{equation}
Then, introduce the subgroup which leaves each point of the lattice
unaltered:
\begin{equation}
B({\calT}) = \{h\in H: \mb{{\rm Ad}}(h)T_k=T_k, k=1,\ldots,r\}.
\end{equation}
The effective transformation group on the lattice $\ul L$ is the
quotient group $K=N({\calT})/B({\calT})$. Thus finally the set of
gauge-inequivalent vortex generators in ${\calM}_i$ is the quotient ${\ul
L}/K$, comprising equivalence classes of lattice points under the
induced action of $K$.
We conjecture that the effect of $K$ upon the lattice $\ul L$ allows
us to enumerate gauge inequivalent vortex generators by $l_1 \geq
\cdots \geq l_r$ for those $\calM_i$'s of dimension larger than
one. This conjecture is largely motivated by example (\ref{sec-su5}),
and is beyond the scope of the present work to prove.
This completes the classification of vortex generators in each of the
subspaces ${\calM}_i$. Of course, if the coupling constants allow then one
may also have combination vortex generators, as described above.
\section{Symmetric Spaces and the Conjugacy of Maximal Tori}
\label{sec-proof}
The goal of this section is to discuss the mathematics of, and prove,
the following crucial result upon vacuum manifolds that are either
compact symmetric spaces or compact symmetric spaces that are modified
to admit semi-local vortices:
\begin{quote}\em
for ${\calM}_i$ of rank $r$, we can find a set of linearly independent
commuting generators $\{T_1, \ldots, T_r\}$ such that the $X \in {\calM}_i$
satisfying the closure condition may be written
\begin{equation}
X = \mb{{\rm Ad}}(h) \sum_{k=1}^{r} l_k T_k,\nonumber
\end{equation}
with $h \in H$, and $l_k\in{\bf Z}$.
\end{quote}
Before establishing the above result for our cases of interest, we
shall discuss some basic properties of symmetric spaces that will be
useful as a background to the work in this paper~\cite{Helg}.
There are several ways of characterising when the vacuum manifold
$G/H$ is a symmetric space and we find the most convenient in the
context of this work to be:
$G/H$ is a symmetric space if the decomposition $\calG = \calH
\oplus \calM$ obeys
\begin{equation}
\label{symspace}
[\calM, \calM] \subseteq \calH.
\end{equation}
This has the consequence (Cartan, \cite{cartd}) that $M$ may be
written as $M \cong \exp(\calM)$, and then $G$ has the polar
decomposition into cosets $G = \exp(\calM)H$. Geodesics in
$\exp(\calM)$ are easily found, and those passing through $\bf 1$ may
be written as $\ga_X(t) = \exp(Xt)$ with $X \in \calM$.
Other important features of symmetric spaces may be found from the
above structure, such as a well defined composition of elements in $M$
relating to a symmetry operation upon geodesics. However these
features are not central to this paper and we shall not discuss them
further.
Cartan has proved that the compact symmetric spaces can be written as
$M = M_0 \mb{\times} M_+$; where $M_0$ is the toroidal part of $M$, written
$M_0=U(1)^k$, and $M_+ \cong G/H$, with $G$ semi simple, is the
rest. We shall establish the main result of this section firstly upon
$M_0$, and then $M_+$, before discussing the case when $M$ has been
modified from a symmetric space so as to admit semi-local vortices.
\subsection{Toroidal $M_0$}
\label{tor}
For the toroidal part of the vacuum $M_0$ we may write $M_0 \cong G/H$
with $G=U(1)^k$ and $H = {\bf 1}$. Trivially then $\calM_0 = \calG =
u(1)^k$ and then $[\calM_0, \calM_0] = 0 \subseteq \calH$.
Eq.~(\ref{result}) is trivial in such a case, since each $\calM_i$ is
one-dimensional.
\subsection{Non-toroidal $M_+$}
For the non-toroidal part of a symmetric vacuum $M_+$, Cartan has
shown that $M_+ \cong G/H$ where $G$ is a semi-simple Lie group and
$H$ is embedded in $G$ according to Eq.~(\ref{symspace}) above. We
now establish the main result of this section for such cases.
Firstly we establish the following, a variant of Hunt's lemma
\cite{Hunt}:
\begin{quote}\em
for $X, Y \in {\calM}_i$ there exists $h_0 \in H$ such that $[\mb{{\rm Ad}}(h_0)X, Y]
=0$.
\end{quote}
This result is proved by considering a function $f: H \to {\bf R}$
defined by
\begin{equation}
f(h) = \inprod{Ad(h)X}{Y}.
\end{equation}
Since this is a bounded real function on a compact domain, it attains it
maximum, say at a point $h_0$. Then for any
$K\in \calH$,
\begin{equation}
0=\left. \frac{d}{dt}f(e^{Kt}h_0)\right|_{t=0}=
\inprod{\mb{{\rm ad}}(K)\mb{{\rm Ad}}(h_0)X}{Y}=\inprod{K}{[\mb{{\rm Ad}}(h_0)X,Y]}.
\end{equation}
Finally, since $G \rightarrow H$ defines a symmetric space,
$[\mb{{\rm Ad}}(h_0)X, Y] \in \calH$ and therefore vanishes, proving the lemma.
From this result, in three steps we may establish our main result when
the vacuum is a symmetric space.
Firstly, there exists $Y \in
{\calM}_i$ such that the centraliser $\calC$ of $Y$ in ${\calM}_i$
\begin{equation}
\calC(Y) = \{ U \in {\calM}_i : [Y,U]=0 \},
\end{equation}
is precisely $\calT$. This follows by considering elements $Y$ that
define paths $\{\exp(\al Y): \al \in {\bf R}\}$ dense in $T =
\exp(\calT)$. Then for any $Z \in \calM_i$ such that $[Z, Y]=0$,
also $[Z, \calT]=0$ and thus, by maximality, $Z \in \calT$.
Next, we use the result established above. Thus for any $X \in
{\calM}_i$ there exists $h_0 \in H$ such that $\mb{{\rm Ad}}(h_0)X \in
\calC(Y)$. Hence, we have shown that there exists $h_0 \in H$ with
$\mb{{\rm Ad}}(h_0)X \in \calT$.
Finally, we note that within $\calT$, the generators
$X$ that satisfy the condition of closure, $e^{2\pi X}\in H$, obviously
form a lattice, $\ul L$ say, and we can choose the generators $T_j$ so
that this comprises precisely the linear combinations with integer
coefficients, $\sum_j l_j T_j$, $\l_j\in{\bf Z}$.
\subsection{$M$ Admits Semi-local Vortices}
Here we deal with the situation when $M \cong G/H$ admits semi-local
vortices, and has been modified from a symmetric space.
Specifically we are going to consider vacuum manifolds of the form
\begin{equation}
\label{nonsym}
M \cong \frac{K \mb{\times} A/Z}{H},
\end{equation}
where $K$ is semi-simple, $A$ is Abelian and connected, $Z$ is a
finite subgroup, and $H$ is isomorphic to a subgroup $L$ of $K$, such
that $K/L$ is a compact symmetric space.
Perhaps the easiest way to see the structure of Eq.~(\ref{nonsym}) is
to represent it by a two step symmetry breaking:
\begin{eqnarray}
\calG = \calK \oplus \calA &\rightarrow& \calL \oplus \calA = \calH'
\oplus \calC(L) \oplus \calA \nonumber \\
&\rightarrow& \ \calH \ = \calH' \oplus \calB,
\end{eqnarray}
where here $H'$ is the semi-simple part of $L$, $\calC(L)$ is the
connected centre of $\calL$, and $\calB \cong \calC(L)$.
Decomposing $\calG = \calH \oplus \calM$, we see that $\calM$ is
made up of two parts, and may be written as $\calM = \calP \oplus
\calB^\perp$, where
\bse
\begin{eqnarray}
\calK &=& \calL \oplus \calP,\\
\label{calB}
\calC(L) \oplus \calA &=& \calB \oplus \calB^\perp.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
From this decomposition the main result~(\ref{result}) of this section
can be established: it is trivially true upon $\calB^\perp$ and,
because $K/L$ is a symmetric space, it is true also on
$\calP$.
\section{The Stability of Vortices in General Gauge Theories}
The stability of the vortices is on a level separate from the family
structure discussed above, though clearly all vortices in the same
family have the same stability structure since all such vortices are
gauge equivalent.
Classical stability is determined by the energetics of the vortex
solution. If it is energetically favourable for a vortex solution to
continuously relax to the vacuum then that vortex is unstable. This is
the general situation for vortex solutions, and generally one needs
some property of the scalar-gauge theory to `prop-up' the solution, so
as to make the configuration a local {\em minimum} of the energy.
There are two ways this may be achieved: either through the topology
of the vacuum manifold, or through the dynamics of the
solution~\cite{Vach91}. We discuss each of these cases separately.
\subsection{Topological Stability}
Topological stability arises through the vacuum manifold not being
simply connected, so that
\begin{equation}
\label{eq-homotopy}
\pi_1 (G/H) \neq 0.
\end{equation}
Vortex families whose boundary conditions are the elements of the
non-trivial homotopy classes of Eq.~(\ref{eq-homotopy}) are imbued
with a {\em conserved} topological charge, which in appropriate cases
guarantees their stability. Physically, the vacuum manifold contains
loops that are incontractible, so that the vortex solutions corresponding
to such loops are topologically obstructed from continuously deforming
to the vacuum.
There are two distinct ways in which topological stability may occur:
either through an Abelian or a non-Abelian part of the symmetry breaking.
These two different situations also have contrasting properties for their
family structure, and the nature of their stability. Therefore we discuss
these two cases separately.
Abelian topological vortices are present for a symmetry breaking scheme
such as
\begin{equation}
G = G' \mb{\times} A/Z \rightarrow H \subseteq G',
\end{equation}
where $A$ is an Abelian subgroup, a torus
\begin{equation}
A= U(1)_1 \mb{\times} \cdots \mb{\times} U(1)_N
\end{equation}
and $Z$ is a finite group. Then $\pi_1(G/H)$ takes the form
\begin{equation}
\pi_1 (G/H) = P \mb{\times}{\bf Z}^N,
\end{equation}
where $P$ is a finite group. The Abelian topologically stable vortices
have boundary conditions that belong to non-trivial ${\bf Z}^N$ homotopy
classes.
The Lie algebra of $G$ decomposes under the adjoint action of $H$ into
\begin{equation}
\calG = \calH \oplus {\calM}' \oplus \calA.
\end{equation}
The Abelian subalgebra $\calA$ is generated by $N$ commuting generators
$\{T_k\}$. We can always choose these so that the elements satisfying the
closure condition are given by $\sum_k l_k T_k$, though only some of these
may actually correspond to stable vortices. Vortices with distinct
values of the quantum numbers $l_k$ are necessarily gauge inequivalent;
there is no non-trivial family structure.
Note that $\calA$ may be described as the intersection of ${\calM}$ and the centre
$\calC$ of the Lie algebra $\calG$.
Non-Abelian topological stability can arise from elements outside the
centre of the group; if for example the vacuum manifold $G/H$ has the form
of the quotient of a simply connected manifold $M$ by a discrete
group ${\bf Z}_n$, then $\pi_1(G/H) = \pi_1(M/{\bf Z}_n) = {\bf Z}_n$.
This gives the vortex generators in the relevant ${\calM}_i$ subspaces a
mod-$n$ topological charge, imparting stability to the corresponding
vortex solutions. Such stable vortices correspond to {\em fractional
quantum} embedded vortices, and the solutions have non-trivial family
structure because necessarily ${\rm dim}({\calM}_i)>1$.
\subsection{Dynamical Stability}
\label{dynamical}
Even when there is no topology to stabilise a vortex solution
there may be non-trivial dynamics making the decay modes of
the vortex energetically unfavourable: this is called dynamical
stability~\cite{Vach91}. Often this occurs when an Abelian symmetry
couples much more strongly than a non-Abelian symmetry; then the
decay modes cost large gradient energies that cannot be
compensated by inducing a non-Abelian gauge field.
Such vortices are solutions to gauge theories that have gauge coupling
constants close to a `semi-local limit', where the scalar-gauge theory
admits semi-local vortices. Preskill \cite{Pres92}, has shown that
semi-local vortices are solutions to gauge theories of the form
\begin{equation}
G_{\rm global} \mb{\times} U(1)_{\rm local} \rightarrow H, \ \ \ {\rm with}\
H \cap U(1)_{\rm local}={\bf 1},
\label{eq-22}
\end{equation}
where the suffices `global' and `local' represent non-gauge and gauged
symmetries, respectively. Requiring that this does not admit
topological vortices leads to the condition $H \not\subset
G_{\rm global}$. Note that Eq.~(\ref{eq-22}) need only be a sub-part
of a more general symmetry breaking.
Thus we define a `semi-local limit' of a gauge theory to be a limit of
the gauge coupling constants $\{ q_f \}$ in which part of the gauge
theory takes the form in Eq.~(\ref{eq-22}). Recall that the formal
limit of taking a coupling constant
$q_f \rightarrow 0$ decouples the gauge field and
makes the corresponding gauge symmetry global. Thus dynamically stable
vortices are generated by the corresponding generator of semi-local
vortices, but with the gauge coupling constants {\em close} to those
in the semi-local limit.
The above may be expressed succinctly in terms of the group theory,
describing which vortex generators give solutions that may be
dynamically stable. Recall that $\calC$ is the centre of the Lie
algebra $\calG$, and denote the projection of an element $X \in \calG$
into $\calM$ by ${\rm pr}_\calM(X)$. We then have the following:
\begin{quote}\em
vortex generators $X \in {\rm pr}_\calM(\calC)$ such that $X \not\in
\calC$ define embedded vortices that
are stable in a well defined semi-local limit of the model and are
thus dynamically stable for a region of parameter space around that
semi-local limit.
\end{quote}
We also have a useful subresult about the dimension of $\calM_i$'s for
such vortices:
\begin{quote}\em
vortices with a stable semi-local limit are {\em always} generated by
generators in one dimensional $\calM_i$'s.
\end{quote}
The proofs are given in Appendix C.
Together these give a nice interpretation of the role of semi-locality:
considering $\calM_i \subset {\rm pr}_\calM(\calC)$, then the
embedding $\calM_i \subset \calG$ is determined by the coupling
constants $\{ q_i\}$. Essentially the semi-local limit is for
coupling constants such that the embedding is $\calM_i \subset
u(1)_{\rm local}$. Then dynamical stability occurs when the embedding
is close to this.
\section{Examples}
This section illustrates principles discussed generally in the
preceding sections.
\subsection{$SU(2) \rightarrow 1$}
\label{sec-su2}
This is an examples of a symmetry breaking that contains hidden
symmetries.
Clearly, $\calH=0$ and $\calM= su(2)$. Then, since $\calM$ is a Lie
algebra, the relation $[\calM, \calM] = \calM$ holds.
However the vacuum manifold is a three-sphere embedded in ${\bf C}^2$,
with an isometry group
\begin{equation}
I = I(S^3) = SU(2)_l \mb{\times} SU(2)_g,
\end{equation}
which is larger than the gauge group. The $SU(2)_l$ represents the
left actions upon a complex doublet, with generators
represented by $X_a = -\frac{1}{2}i \si_a$, and the $SU(2)_g$ is a
hidden global symmetry with generators $Y_1= -\frac{1}{2} \si_2 K, \
Y_2=\frac{1}{2} i \si_2 K$, and $Y_3= -\frac{1}{2}i {\bf 1}_2$. Here
$K$ is the complex conjugation operator and $\si_a$ are the Pauli spin
matrices.
Then the stability group of $I$ upon a point of $M$ is
\begin{equation}
J = SU(2)_{l+g},
\end{equation}
the diagonal subgroup lying {\em between} $SU(2)_l$ and $SU(2)_g$.
Hence the actual symmetry breaking is
\begin{equation}
\label{eq-fullsb}
SU(2)_l \mb{\times} SU(2)_g \rightarrow SU(2)_{l+g},
\end{equation}
which contains the gauge symmetry breaking $SU(2)_l \rightarrow 1$,
and has additional global symmetry $SU(2)_g$. Defining $\calI = \calJ
\oplus \calN$ one can easily verify $[\calN, \calN] \subseteq \calJ$.
It is interesting to note that the above model is the Weinberg-Salam
model, with electroweak mixing angle $\sin{\theta_w}=0$.
\subsection{$SU(2) \mb{\times} U(1) \rightarrow U(1)$}
\label{Weinberg-Salam}
The gauge symmetry $SU(2) \mb{\times} U(1)$ acts on a two-dimensional complex
scalar field $\Ph$ by the fundamental representation. The generators
of $SU(2)$
are $X_a = -\frac{1}{2}i \si_a$ and the $U(1)$ generator is
$X_0 =\frac{1}{2} i {\bf 1}_2$.
The vacuum manifold is
\begin{equation}
M \cong \frac{SU(2) \mb{\times} U(1)}{U(1)} \cong \frac{SU(2) \mb{\times}
SU(2)}{SU(2)_{\rm diag}},
\end{equation}
thus the isometry group is $I= SU(2) \mb{\times} SU(2)$. This is in general
not a symmetry group of the theory unless ${\sin}\theta_w=0$.
The inner product on $su(2) \oplus u(1) \subset gl({\bf C}^2)$ is
obtained from Appendix B:
\begin{equation}
\inprod{X}{Y} = -2 s_1 \mb{{\rm tr}} XY - \frac{s_2}{4}\mb{{\rm tr}} X \mb{{\rm tr}} Y.
\end{equation}
To obtain the usual scalar-gauge coupling~(\ref{couples}), $s_1$ and
$s_2$ are related to the hypercharge $g'$ and isospin $g$ coupling
constants such that
\begin{equation}
\inprod{X}{Y} = -\frac{1}{g^2} \left\{ 2{\mb{{\rm tr}}}XY + (\cot^2 \theta_w-1)
{\mb{{\rm tr}}} X {\mb{{\rm tr}}} Y \right\},
\end{equation}
with $\tan \theta_w = g'/g$.
The unit norm generators are then $gX_a$ and $g'X_0$, and thus the
components of a gauge field $A^\mu$ couple to $\Ph$ as
\begin{equation}
\label{couples}
A^\mu \Phi = (gW^\mu_a X_a + g' Y^\mu X_0) \Phi.
\end{equation}
Taking the vacuum to be defined from $\Ph_0 = v (0\ 1)^\top$ the
decomposition $\calG = \calH \oplus \calM$ takes the form
\begin{equation}
\calH = \left( \begin{array}{cc} i \al & 0 \\ 0 & 0 \end{array}
\right)\ \ \ \ {\rm and}\ \ \ \
{\calM} = \left( \begin{array}{cc} -i\be \cos 2\theta_w & \ga \\ -\ga^* &
i\be \end{array} \right),
\end{equation}
with $\al, \be$ real and $\ga$ complex. Then under $\mb{{\rm Ad}}(H)$,
$\calM$ decomposes to the irreducible subspaces
$\calM = \calM_1 \oplus \calM_2$, with
\begin{equation}
\calM_1 =
\left( \begin{array}{cc} -i\be \cos 2 \theta_w & 0 \\ 0 & i \be
\end{array} \right) \ \ \ \ {\rm and}\ \ \ \
\calM_2 =
\left( \begin{array}{cc} 0 & \ga \\ -\ga^* & 0 \end{array}
\right).
\end{equation}
Additionally ${\rm pr}(u(1)_Y) = \calM_1$.
Thus we obtain a one-parameter family of gauge equivalent unstable
embedded vortices and a gauge invariant embedded vortex with a stable
semi-local limit. These are identified as the W-strings and Z-string
of the Weinberg-Salam model.
\subsection{$SU(5) \rightarrow SU(3) \mb{\times} SU(2) \mb{\times} U(1)$}
\label{sec-su5}
This is an example of a symmetry breaking where the irreducible
${\calM}_i$'s are of a non-trivial rank.
The gauge group $G=SU(5)$ acts on the
scalar field $\Ph \in \calG$ by the adjoint action. For vacuum
expectation value
\begin{equation}
\Ph_0 = iv \left( \begin{array}{ccc} \frac{2}{3} \bf{1}_3 & \vdots &
\bf{0} \\ \cdots & \cdots & \cdots \\ \bf{0} & \vdots & -\bf{1}_2
\end{array} \right),
\end{equation}
$G$ breaks to $H = SU(3)_c \mb{\times} SU(2)_I \mb{\times} U(1)_Y$:
\begin{equation}
\left( \begin{array}{ccc} SU(3)_c & \vdots &
\bf{0} \\ \cdots & \cdots & \cdots \\ \bf{0} & \vdots & SU(2)_I
\end{array} \right)
\mb{\times}
\left( \begin{array}{ccc} e^{\frac{2}{3}i \theta} {\bf 1_3} & \vdots &
\bf{0} \\ \cdots & \cdots & \cdots \\ \bf{0} & \vdots &
e^{-i \theta}{\bf 1}_2 \end{array} \right)
\subset SU(5).
\end{equation}
Reducing $\calG$ into $\calG = \calH \oplus {\calM}$, where
\begin{equation}
{\calM} = \left( \begin{array}{ccc} \bf{0}_3 & \vdots &
\underline{A} \\ \cdots & \cdots & \cdots \\ -\underline{A}^\dagger
& \vdots & \bf{0}_2
\end{array} \right),
\end{equation}
one finds this to be irreducible under the adjoint action of $H$.
However it is clear that ${\rm rank}({\calM})=2$, for example with
generators $T_1, T_2$ such that
\begin{equation}
T_i = \frac{1}{2}\left( \begin{array}{ccc} \bf{0}_3 & \vdots &
\underline{A}_i \\ \cdots & \cdots & \cdots \\
-\underline{A}_i^\dagger & \vdots & \bf{0}_2
\end{array} \right), \
{\rm with}\
\underline{A}_1 = \left( \begin{array}{cc} 1 & 0 \\
0 & 0 \\ 0 & 0 \end{array} \right),\ \
\underline{A}_2 = \left( \begin{array}{cc} 0 & 0 \\
0 & 1 \\ 0 & 0 \end{array} \right),
\end{equation}
then $[T_1, T_2]=0$ and there are no other linearly independent
generators that commute with both of these.
Then using Eq.~(\ref{result}) the vortex generators in ${\calM}$ are:
\begin{equation}
\label{eq-su5gen}
X = \mb{{\rm Ad}}(h) \left(l_1 T_1 + l_2 T_2 \right),
\end{equation}
with $l_i \in {\bf Z}$ and $h \in H$. These generators form a ${\bf
Z}^2$ lattice of solutions $(l_1, l_2)$, with an orbit of solutions
generated by $\mb{{\rm Ad}}(H)$ from each point of the lattice. However, there
is degeneracy since it is easy to find $h_1, h_2 \in H$ such that
\bse
\begin{eqnarray}
\mb{{\rm Ad}}(h_1) (l_1 T_1 + l_2 T_2) &=& -l_2 T_1 + l_1 T_2, \\
\mb{{\rm Ad}}(h_2) (l_1 T_1 + l_2 T_2) &=& -l_1 T_1 + l_2 T_2.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
Together these generate the discrete group of actions $\mb{{\rm Ad}}(K)$ that
takes the pair $(l_1, l_2)$ to the set of eight pairs $\{ (\pm l_1,
\pm l_2), (\mp l_1, \pm l_2)$, $(\pm l_2, \pm l_1), (\mp l_2, \pm
l_1) \}$.
Then the set of gauge inequivalent embedded vortex generators in $SU(5)$
are those in Eq.~(\ref{eq-su5gen}), but with $(l_1, l_2)$ restricted
to the octant $l_1 \geq l_2 \geq 0$.
To explicitly give an example of a vortex we give the $(1,0)$ solution:
\bse
\begin{eqnarray}
\Phi(r, \theta) &=& f_{\rm NO}(r) \mb{{\rm Ad}}(e^{T_1\theta})\Phi_0, \\
\ul{A}(r,\theta) &=& \frac{g_{\rm NO}(r)}{r} T_1 \ul{\hat{\theta}}.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
then, with a little calculation
\begin{equation}
\mb{{\rm Ad}}(e^{T_1\theta})\Phi_0= \Phi_0 - \frac{5v\sin \frac{\theta}{2}
A}{3}( \sin \frac{\theta}{2} {\bf 1}_5 + \cos \frac{\theta}{2}
T_1),
\end{equation}
where $A_{ij}=i(\de_{i1}\de_{j1} - \de_{i4}\de_{j4})$. A non-trivial
geodesic, that is certainly not contained within a plane through the
origin.
\section{Conclusions}
We conclude by briefly summarising our classification.
We consider minimal, gauge invariant theories~(\ref{lag}) of symmetry
breaking $G \rightarrow H$, where $H \subset G$ are compact Lie
groups. Then the following structure is required: writing $\calG =
\calH \oplus \calM$, one splits $\calM =\calM_1 \oplus \cdots \oplus
\calM_n$, the irreducible subspaces of $\calM$ under $\mb{{\rm Ad}}(H)$.
Then cylindrically symmetric, time independent vortex solutions are of
the form
\begin{eqnarray}
\Phi(r, \theta) &=& f_{\rm NO}(r) D(e^{X\theta})\Phi_0, \nonumber \\
\ul{A}(r,\theta) &=& \frac{g_{\rm NO}(r)}{r} X \ul{\hat{\theta}}. \nonumber
\end{eqnarray}
with $X \in \calM_i$ specified by the following result, which has been
rigorously established for $G/H$ a symmetric space or a symmetric
space modified to admit semi-local vortices:
\begin{quote}\em
for ${\calM}_i$ of rank $r$, we can find a set of linearly independent
commuting generators $\{T_1, \ldots, T_r\}$ such that the $X \in {\calM}_i$
satisfying the closure condition may be written
$$
X = \mb{{\rm Ad}}(h) \sum_{k=1}^{r} l_k T_k,
$$
with $h \in H$, and $l_k\in{\bf Z}$.
\end{quote}
Generally this classifies {\em all} vortices of the embedded
Nielsen-Olesen type, however for certain critical values of the
ratios of gauge coupling constants the solution set may increase
to include combination vortices also, with generators $X$ lying
between $\calM_i$ and $\calM_j$.
Stability of vortices may be of two types. Firstly, topological
stability given by the non-trivial first homotopy classes of
$G/H$. Secondly, dynamical stability, specified by the following
result:
\begin{quote}\em
vortex generators $X \in {\rm pr}_\calM(\calC)$ such that $X \not\in
\calC$ define embedded vortices that
are stable in a well defined semi-local limit of the model and are
thus dynamically stable for a region of parameter space around that
semi-local limit.
\end{quote}
Such vortex generators always lie in one-dimensional $\calM_i$'s.
We conjecture that the above classification also holds true in
general.
\bigskip
{\noindent{\Large{\bf Acknowledgements.}}}
\nopagebreak
\bigskip
\nopagebreak
This work was supported in part by PPARC. N.L. acknowledges
EPSRC for a research studentship and thanks A.C. Davis for
interesting discussions related to this work. This work was supported
in part by the European Commision under the Human Capital and Mobility
program, contract no. CHRX-CT94-0423.
\bigskip
\bigskip
{\noindent{\Large{\bf Appendix A}}}
\par
\nopagebreak
\bigskip
\nopagebreak
\noindent
The Lagrangian (\ref{lag}) gives field equations
\bse
\begin{eqnarray}
D^\mu D_\mu \Phi &=& -\pderiv{V}{\Phi}, \\
\inprod{D^\mu F_{\mu\nu}}{X} &=& \inprod{J_\nu}{X} =
\inprod{d(X)\Phi}{D_\nu \Phi} - \inprod{D_\nu \Phi}{d(X)\Phi}.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
The vortex Ansatz
\bse
\begin{eqnarray}
\Phi_{\rm vor}(r, \theta) &=& f_{\rm NO}(r) D(e^{X\theta})\Phi_0, \\
\ul{A}_{\rm vor}(r,\theta) &=& \frac{g_{\rm NO}(r)}{r} X
\ul{\hat{\theta}},
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
naturally splits $\calG$ globally into
$\calG = \calG_\mb{{\rm emb}} \oplus \calG_\mb{{\rm emb}}^\perp$ such that
$\inprod{\ul{A}_{\rm vor}(x)}{\calG_\mb{{\rm emb}}} \neq 0$ for all nonzero $x$.
In the scalar sector it provides a local decomposition
$\calV = \calV_\mb{{\rm emb}}(\theta) \oplus \calV_\mb{{\rm emb}}^\perp(\theta)$ where
$\calV_\mb{{\rm emb}}(\theta) = D(e^{X \theta})({\bf R}\Phi_0 + {\bf R}
d(X_\mb{{\rm emb}}) \Phi_0)$, with ${\bf R}$ the real numberline.
Substituting this vortex Ansatz into the field equations and requiring
it to be a solution yields
\bse
\begin{eqnarray}
\inprod{D^\mu D_\mu \Phi_{\rm vor}(x)}{\calV_\mb{{\rm emb}}^\perp(\theta)} &=&
0,\\
\inprod{J^\mu(x)}{\calG_\mb{{\rm emb}}^\perp} &=& 0,
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
from which one obtains,
\bse
\begin{eqnarray}
\inprod{\Psi}{\pderiv{V}{\Phi}{[\phi]}}
&=& 0,\ \ \Ps \in \calV_\mb{{\rm emb}}^\perp(\theta), \ph \in
\calV_\mb{{\rm emb}}(\theta), \\
\langle d(X^\perp)\Ph, \calV_\mb{{\rm emb}}(\theta) \rangle &=& 0,\ \
\Ph \in \calV_\mb{{\rm emb}}(\theta), X^\perp \in \calG_\mb{{\rm emb}}^\perp.
\end{eqnarray}
\setcounter{equation}{\the\subeqnno}\endgroup
The first condition constrains the scalar potential and may restrict
combination vortex solutions. The second we rephrase:
using the identity $\inprod{d(\calG)\Ph}{\Ph}=0$ for $\Ph \in \calV$,
\begin{eqnarray}
\inprod{d(X^\perp)\Ph}{\calV_\mb{{\rm emb}}(\theta)} =0
&\Leftrightarrow&
\inprod{d(X^\perp) D(e^{X\theta})\Ph_0}{D(e^{X \theta})d(X)\Phi_0}=0
\nonumber \\
&\Leftrightarrow&
\inprod{d(\mb{{\rm Ad}}(e^{-X \theta})X^\perp)\Ph_0}{d(X)\Ph_0} = 0, \\
&\Leftrightarrow&
\inprod{d(X^\perp)\Ph_0}{d(X)\Ph_0} = 0. \nonumber
\end{eqnarray}
The last step by virtue of $G_\mb{{\rm emb}}$ being a subgroup of $G$ implies
that $\mb{{\rm Ad}}(e^{X \theta})\calG_\mb{{\rm emb}}^\perp = \calG_\mb{{\rm emb}}^\perp$.
One should note that the above proof is a modified version of that
given in $\cite{me1}$, to encompass the situation when $\calV_\mb{{\rm emb}}$ is
defined locally. In this context our definition of an embedded defect
is more general than that discussed in \cite{Vach94}, where they
considered $\calV_{\rm emb}$ to be independent of $\theta$ -- which is
equivalent to assuming that $D(e^{X \theta})\Phi_0$ is a circle
centred on the origin within $\calV$. Generically, geodesics on
non-spherical homogenous spaces are not of this form --- an example of
which is given in sec.~(\ref{sec-su5}).
\bigskip
\bigskip
{\noindent{\Large{\bf Appendix B: Inner Product Structures on $\calV$
and $\calG$}}}
\par
\nopagebreak
\bigskip
\nopagebreak
\noindent
The inner product on $\calV$ is given by the (real) Euclidean inner
product. Considering an element ${\bf v} \in \calV$ as a column vector
the form we shall use is:
\begin{equation}
\inprod{{\bf v}_1}{{\bf v}_2} = {\rm Re} [ {\bf v}_1^\dagger {\bf
v}_2 ].
\end{equation}
This inner product is isometric under automorphisms of $\calV$.
To construct a non-degenerate Ad$(G)$-invariant inner
product on $\calG$, we start by splitting it into its
commuting subalgebras, $\calG_1\oplus\cdots\oplus\calG_n$,
where each $\calG_f$ is either simple or a one-dimensional
$u(1)$ algebra. For each $\calG_f$ there is a natural
invariant scalar product $\inproda{\cdot}{\cdot}_f$, unique up to a
factor~\cite{oni}. On a $u(1)$ algebra, we may set $\{X,X\}_f=1$,
where $X$ is the generator normalised so that
$e^{2\pi X}=1$. Upon a simple $\calG_f$ it is (proportional to) the
Killing form; regarding $\calG_f$ as a matrix algebra embedded in
gl$({\bf C}^p)$ for some $p$, the inner product may be defined by
\begin{equation}
\inproda{X}{Y}_f = -p {\rm Re}[{\rm tr}(X^{\dag} Y)].
\end{equation}
Then the most general Ad$(G)$-invariant scalar product on
$\calG$ is
\begin{equation}
\label{scaleip}
\inprod{X}{Y} = \sum_{f=1}^n {s_f}\inproda{X_f}{Y_f}_f,
\end{equation}
for arbitrary values of the scaling factors $s_f$, where $X =
X_1+\cdots+X_n$, $Y=Y_1+\cdots+Y_n$ and $X_f, Y_f \in \calG_f$. The
scales $s_f$ relate to the norm on each $\calG_f$; for $X_f \in
\calG_f$ such that $\inproda{X_f}{X_f}=1$ the unit norm
generator with respect to $\inprod{\cdot}{\cdot}$ is written $q_f
X_f$, with $q_f$ a function of $\{s_1,\cdots,s_n\}$.
These two inner products over their respective spaces are related by
the following result:
\result{Theorem}{
The (real) inner product on $\calV$ is related to the inner product
on $\calM$ by
\begin{eqnarray}
\inprod{d(X_i){\bf v}}{d(Y_j){\bf v}} &=& \la_i \la_j
\inprod{X_i}{Y_j}, \ X_i \in {\calM}_i, Y_j \in {\calM}_j,\nonumber \\
{\rm where}\ \la_i &=&
\frac{\norm{d(X_i){\bf v}}}{\norm{X_i}}. \nonumber
\end{eqnarray}
Each $\la_i$ is constant upon its particular ${\calM}_i$.
}\\
The theorem is original to this paper, although it is a fairly simple
application of linear algebra. It is a consequence of two inner
products having a similar invariance, and is most easily derived from:
\result{Lemma}{
Consider $\inprod{\cdot}{\cdot}_1$ and $\inprod{\cdot}{\cdot}_2$ two
(real) inner products on $\calV$. Then $\calV$ decomposes into
$\calV_1 \oplus \cdots \oplus \calV_n$ such that
\begin{equation}
\inprod{{\bf v}_i}{{\bf v}_j}_2 = \la_i \la_j
\inprod{{\bf v}_i}{{\bf v}_j}_1, \ {\bf v}_i \in {\calM}_i,
{\bf v}_j \in {\calM}_j.
\end{equation}
Each $\la_i$ is constant over its particular $\calV_i$.
}
\proof{
Choosing a basis for $\calV$ orthonormal with respect to
$\inprod{\cdot}{\cdot}_1$, the inner products can be written
$\inprod{{\bf u}}{{\bf v}}_1={\bf u}^{\top}{\bf v}$ and
$\inprod{{\bf u}}{{\bf v}}_2={\bf u}^{\top} A {\bf v}$, with ${\bf u},
{\bf v} \in \calV$, with $A$ a non-degenerate symmetric matrix. Denote
the eigenspaces of $A$ by $\calV_1 \oplus \cdots \oplus \calV_n$, with
corresponding eigenvalues $\la_i^2$, then for ${\bf v}_i \in \calV_i$
\begin{equation}
\inprod{{\bf v}_i}{{\bf v}_j}_2={\bf v}_i^\top A{\bf v}_j=\la_i \la_j
{\bf v}_i^\top {\bf v}_j.
\end{equation}
Finally it is clear that $\la_i$ is constant over $\calV_i$ because
it is an eigenvalue, but also $\la_i = \norm{{\bf v}_i}_2/\norm{{\bf
v}_i}_1$
}
Then from this lemma the theorem is easily proved:
\proof{
Firstly, let $\inprod{X}{Y}_1$ denote the usual $\mb{{\rm Ad}}(G)$
invariant inner product over $\calG$, but restricted to $\calM$; note
that it is $\mb{{\rm Ad}}(H)$ invariant.
Then one may induce a second inner product on $\calM$ from that on
$\calV$: this takes the form
$\inprod{X}{Y}_2 = \inprod{d(X){\bf v}}{d(Y) {\bf v}}$. Again this
inner product is $\mb{{\rm Ad}}(H)$ invariant over $\calM$.
From the lemma $\calM$ reduces to $\widetilde\calM_1 \oplus \cdots
\oplus \widetilde\calM_k$ such that for $X\in\widetilde\calM_i$,
$Y\in\widetilde\calM_j$
\begin{equation}
\label{eq}
\inprod{d(X){\bf v}}{d(Y){\bf v}}=\la_i\la_j\inprod{X}{Y}.
\end{equation}
Here $\la_i=\la(X_i)=\norm{d(X_i){\bf v}}/\norm{X_i}$ is
constant over each $\widetilde{\calM}_i$. However $\la(X)$ is also $\mb{{\rm Ad}}(H)$
invariant $\la(\mb{{\rm Ad}}(H)X)=\la(X)$, and thus also constant over each
${\calM}_i$. Hence each $\widetilde{\calM}_i$ is a direct sum of ${\calM}_i$'s and
Eq.~(\ref{eq}) is true for $X_i \in {\calM}_i$, $Y_j \in {\calM}_j$.
}
\bigskip
\bigskip
{\noindent{\Large{\bf Appendix C: Dynamical Stability Proofs}}}
\par
\nopagebreak
\bigskip
\nopagebreak
\noindent
For completeness we give the following proofs to results stated in
section (\ref{dynamical}).
\result{Theorem}{
Vortex generators $X \in {\rm pr}_\calM(\calC)$ such that $X \not\in
\calC$ define embedded vortices that
are stable in a well defined semi-local limit of the model and are
thus dynamically stable for a region of parameter space around that
semi-local limit.
}
\proof{
Consider $X \in {\rm pr}_\calM(\calC)$, such that it generates a
closed geodesic. Then $X = X_c + X_h$, with $X_c \in \calC$, such that
${\rm pr}_\calM (X_c)= X$, generating a $U(1)_c \subseteq C$ and
$X_h \in \calH$ generating a $U(1)_h \subset H$. These define a
decomposition of the symmetry breaking of the Lie algebras:
\begin{equation}
\calG = \calG' \oplus u(1)_c \rightarrow
\calH= \calH' \oplus u(1)_h,
\end{equation}
with $U(1)_c \cap H = 1$. It is now clear that the appropriate
semi-local limit to obtain a symmetry breaking of the form
Eq.~(\ref{eq-22}) is to make coupling constants appertaining to $G'$
vanish. The corresponding vortex generator is $X$. This completes the
proof.}
\result{Lemma}{
Vortices with a stable semi-local limit are {\em always} generated by
generators in one dimensional $\calM_i$'s.
}
\proof{
The proof that if $\calM_i \subseteq {\rm pr}_\calM(\calC)$ then
$\dim(\calM_i)=1$ relies on the following identification: if
one writes $\calM_{\rm ker}$ as the collection of one dimensional
$\calM_i$'s then necessarily $\calM_{\rm ker} = \{X \in
\calM : {\rm \mb{{\rm Ad}}}(H)X = X\}$. Now we are reduced to showing
${\rm pr}_\calM(\calC) \subseteq \calM_{\rm ker}$.
Consider $(X_c + X_h) \in {\rm pr}_\calM(\calC)$, with $X_c \in \calC$
and $X_h \in \calH$. Then $[\calH, X_c + X_h] \in
\calM$. But also $[\calH, X_c + X_h]=[\calH, X_h] \in \calH$.
Hence $[\calH, X_c +X_h]=0$, and equivalently $\mb{{\rm Ad}}(H)(X_c + X_h)=X_c +
X_h$, proving the result.}
|
\section{Introduction}
The picture that has emerged from a suite of multi-wavelength observations of
the Galactic center is that it contains a rotating, clumpy
molecular ring (also known as the Circumnuclear
Disk- CND) which is heated by a centrally concentrated source
(the IRS 16 cluster) of UV radiation. Within the ring's central cavity,
three ``Arms'' of ionized gas (Sgr A
West) are generally in orbital motion around the center, which is believed to
contain a $\sim$ 2.5$\times$10$^{6}$ \, \hbox{$\hbox{M}_\odot$} black hole (Ekart and Genzel 1996; Ghez et al.
1998).
The coincidence in the geometry and kinematics of
the southwestern edge of the molecular ring and ionized gas
suggests that the ionized gas is
dynamically coupled to the inner edge of
the circumnuclear ring. The western edge of the ring
(sometimes called the Western Arc) is
out of the plane of the sky and obscuring the background
stars and absorbing
Br $\gamma$ emission from the western arm causing the ratio of
H92$\alpha$ to Br $\gamma$ to increase (see recent reviews by Genzel, Hollenbach and
Townes 1994; Morris and Serabyn 1997).
Although the CND is often described as having
a circular geometry and kinematics,
(e.g. Morris and Serabyn 1997 and the references cited therein), the ring structure shows
deviations from circular geometry. A
prominent gap in the distribution of
molecular gas in the CND shows non-circular moving gas (e.g. Jackson et al. 1993).
The CND gap is located
where an ionized ``streamer'' appears to run along
the extension of the N. Arm beyond the CND. The thermal ionized gas in
the streamers show that the Northern Arm of Sgr A West is not coupled
dynamically
and spatially to the
molecular gas in the inner edge of the CND.
On a larger scale the non-thermal structure known as Sgr A East is thought to be a
shell-type explosive event, possibly a supernova remnant (SNR), surrounding
the thermal source Sgr A West.
The recent discovery of
OH(1720MHz) masers at the interface of the 50 \, \hbox{km}\,\hbox{s}^{-1}\ molecular cloud and Sgr A East
provides strong evidence that these two are physically interacting
with each other (e.g. Mezger et al. 1996; Zylka et al. 1992; Serabyn et
al. 1992;
Yusef-Zadeh et al. 1996).
The masers are detected to the southeast of
Sgr A East (sources A, D-G)
as well as
to the northwest of the CND. The maser spots
surrounding the Sgr A East shell have radial velocities
near 50 \, \hbox{km}\,\hbox{s}^{-1}\ close to the systemic
velocity of the SNR. The
expansion of a supernova into the Sgr A East molecular cloud (i.e.
the 50 \, \hbox{km}\,\hbox{s}^{-1}\ cloud) is considered to be responsible
for the production of the shocked OH(1720 MHz) maser emission.
The nature of the association between the two systems,
Sgr A East and Sgr A West and their corresponding molecular clouds,
is not well established.
Radio continuum studies have pointed out
the curious alignment of the western
edge of Sgr A East and the Western Arc of Sgr A West
(Ekers et al. 1983; Yusef-Zadeh and Morris 1987; Pedlar et al. 1989).
Sgr A East is
behind Sgr A West, a conclusion based on low-frequency observations
showing a hole in the
distribution of ionized gas toward the Arms of Sgr A West
(Yusef-Zadeh and Morris 1987; Pedlar et al. 1989).
These observations imply that
a segment of the Sgr A East shell must lie behind
the Galactic center by a distance that is estimated to
range between a few to hundreds of parsecs.
The relative location of Sgr A East with respect to the Galactic center
is of interest for a number of reasons. First, the separation between the two
systems is relevant to the role that the relativistic particles of
Sgr A East may play in
upscattering the CND's IR and UV photons to high energies.
The strength of this interaction depends
on the distance between Sgr A East and the CND. This process
was recently incorporated into the modeling the high-energy emission from
the Galactic center based on EGRET measurements (Melia et al. 1998a,b; Fatuzzo
et al. 1998).
In this model, Sgr A East is considered to be within 5pc of the
Galactic center. Second, the disturbance of the molecular
ring by Sgr A East also depends upon the separation between the
two objects.
Recent studies suggest that the N. Arm of Sgr A West
may delineate the inner edge of
an infalling ``tongue'' of neutral gas toward the Galactic center.
The obvious question that needs to be addressed is whether
Sgr A East is responsible for the observed deviation from circular motion
at the location of the gap in the CND and for the infalling cloud.
Lastly, the closeness of Sgr A East to the Galactic center and its unusual energetics
brings up the possibility that the sources responsible for the explosion are
the family of young, massive stars in the IRS 16 cluster.
This hypothesis may provide an insight into the
nature of massive star formation within the same parent cloud in the inner
5pc of the Galactic center.
A signature of an interaction between Sgr A East and the CND
should be shocked gas as the blast wave of Sgr A East compresses the gas
in the CND.
The unambiguous detection of shocked (as opposed to UV-heated) gas associated with
the Galactic center has been difficult. The
intensity ratios of the v=2--1 and 1--0 S(1) lines of H$_2$ (Gatley et al.
1984; Burton \& Allen 1992; Pak et al. 1996) are often
consistent with collisional excitation rather than fluorescence,
but the high density of the molecular gas at the Galactic center
and the intense UV radiation field in the region allow the line
ratios from UV-irradiated gas to resemble those of shock-heated gas
(Sternberg \& Dalgarno 1989). However, the presence of the OH 1720 MHz line,
and the absence of the 1665/1667 MHz lines provides a
clear diagnostic of shocked molecular gas, as the FIR
radiation field from warm dust in UV-heated clouds would pump the latter
transitions. Here we report preliminary results of our
new H$_2$ and [FeII] and 1.2 and 90cm radio continuum observations
and examine the interaction picture of Sgr A East with the CND.
\section{Observations}
We observed the
Galactic center in the
1--0 S(1) transition
of H$_2$ line using both the NICMOS on the HST
as well UNSWIRF on the AAT.
The UNSWIRF observations were carried
out using four overlapping frames with a size of
90$''$, a pixel size of 0.77$''$
and channel maps
separated by $\approx$39 \, \hbox{km}\,\hbox{s}^{-1}\ (Figs. 1,2 and 3).
Using camera 3 of NICMOS, a total of 12 overlapping frames each
with a 52$''$ size and a pixel size of 0.203$''$
was observed in both the line and the continuum.
Each frame was dithered by
16$''$ in each principal directions of
the detector, and the final image was constructed by
mosaicing all 12
frames, a portion of which is shown in Figure 2.
The 90cm continuum observations were carried out in
the BnA (Fig. 4) and A-array configurations (Figs. 5 and 6)
of
the Very Large Array of the National
Radio Astronomy Observatory\footnote{The National Radio Astronomy
Observatory is a facility of the National Science Foundation, operated
under a cooperative agreement by Associated Universities, Inc.}.
Camera 3 of NICMOS was also used to observe the [FeII] line emission
from the central 110$''\times160''$, a subsection of which is shown in Figure 8b.
The NICMOS images were produced by subtracting images in the 1\% filter
centered on the 1.64$\mu$m [FeII] line. The continuum was adjusted
to minimize the stellar residuals, but both positive and
negative residuals remain. Some of the residuals are due to differences
in intrinsic colors and locally patchy extinction.
These observations have provided
wealth of structural details with subarcsecond resolution capability of
the NICMOS, the velocity information from UNSWIRF, and the
ability to discriminate line from continuum emission that UNSWIRF gives.
The high resolution 1.2cm image were based on combined A and B array configurations
whereas the low-resolution image was based on C configuration.
The astrometry of the radio and IR
images was done by using the position of IRS 7 with respect to Sgr A$^*$
(Menten et al. 1995). A detailed account of
these
observations will be given elsewhere.
\section{Results}
\subsection{H$_2$ Filament and H$_2$ Gas inside the CND}
Figure 1 shows contours of 1--0 S(1) H$_2$ line emission
measured by UNSWIRF superposed on the greyscale 6cm
continuum emission from Sgr A West and East (Yusef-Zadeh and Morris 1987).
A segment of
the Sgr A East nonthermal shell is seen to the NE near
$\alpha = 17^h 42^m 32^s, \delta = -28^0 58'$
whereas
the spiral-shaped structure of Sgr A West is evident near the
center of the Figure.
The peaks
located in the
NE (N. lobe) and SW (S. lobe) parts of the
ring are
consistent with limb-brightening of the inner edges along the principal
axis.
An extended H$_2$ feature with a plume-like appearance
is noted within the ring. This new feature
appears to terminate at the position of IRS 7 (a more detailed account of
this feature will be given elsewhere).
The distribution of H$_2$ line emission appears like a ring
surrounding Sgr A West except
in the region near $\alpha = 17^h 42^m 30^s, \delta = -28^0 58' 45''$,
where
the HCN distribution indicates
non-circular motion.
at the location of the inner
gap (Jackson et al. 1993).
A long
filamentary structure with an extent of $\approx1'$ is
discovered to the NW of the CND. A gap is noticed along
the extent of the filament near
$\alpha = 17^h 42^m 29^s, \delta = -28^0 58' 30''$
which shows a similarity in its spatial distribution
to the iner gap of the CND.
The crosses in all the figures show the positions of OH (1720MHz) masers
and the star symbol is coincident with the position of Sgr A$^*$.
The C and B masers with velocities of 43 and 134 \, \hbox{km}\,\hbox{s}^{-1}\
are seen near the outer and inner gaps of H$_2$ distribution.
\begin{figure}
\caption{Contours of H$_2$ 1--0 S(1) emission from the CND
with a pixel size of 0.77$''$ is superposed
on a 6cm radio continuum image of thermal emission from Sgr A West
with a spatial resolution of 0.67$''\times0.4''$.
A segment of the
nonthermal Sgr A East shell is seen to the NE. The star symbol represents Sgr~A$^*$
and the crosses show the position of OH (1720 MHz) masers.}
\end{figure}
Figure 2 shows
the same contours as in Figure 1, but they are now superposed
on the greyscale H$_2$ image based on the NICMOS observations.
Because of the complex nature of Galactic center sources,
the stellar continuum in NICMOS images
has not been removed completely.
There is an overall agreement between the
NICMOS and UNSWIRF data in spite of the contamination
from
stellar continuum in NICMOS images.
The new H$_2$ filament appears strikingly sharp with a width of
about 0.5$''$ in the NICMOS image.
\begin{figure}
\caption{The same as Figure 1 except that the greyscale image is replaced by
the HST/NICMOS H$_2$ image. The greyscale image is convolved by a Gaussian
having a size twice the
pixel size of 0.203$''$. }
\end{figure}
Figure 3 shows the velocity distribution of H$_2$
emission based on one 90$''$ field
centered near the N. lobe as observed with UNSWIRF.
The crosses
indicate the position of the masers.
The N. lobe of the CND
appears prominantly at higher velocities near the 120 and
159 \, \hbox{km}\,\hbox{s}^{-1}\ channel maps.
The new H$_2$ filament appears as a coherent velocity
feature and is best seen in the 41.2 and 80.4 \, \hbox{km}\,\hbox{s}^{-1}
channel maps.
The C maser with a velocity of 43 \, \hbox{km}\,\hbox{s}^{-1}\ is probably related to the new H$_2$
filament, whereas the B masers,
with velocities of
about 134 \, \hbox{km}\,\hbox{s}^{-1}\, are associated with the N. lobe.
The overall velocity
structure of the new filament is also similar to that of the
inner edge of the CND near the Western Arc, suggesting that the new
filament is part of the CND but
outlining the outer edge of the CND. A ridge of weak H$_2$ emission
is also seen in the 41 and 80 \, \hbox{km}\,\hbox{s}^{-1}\ channel maps. This feature
appears to connect the two gaps, as
seen in
the new filament
and in the inner edge of the CND, suggesting that the H$_2$ features are indeed
associated with each other.
\begin{figure}
\caption{The H$_2$ velocity distribution of a field
centered on the
N. lobe of the CND is shown in reverse greyscale. The LSR channel velocities
are shown on top right corners and the crosses represent OH maser sources, C to the
right and B to the left. The high
and low values are shown in black and white,
respectively. The white features are
artefacts from imperfect continuum subtraction. }
\end{figure}
\subsection{Radio Continuum Features at 90 and 1.2cm}
Figure 4 shows the greyscale distribution of 90cm emission
superposed on contours of H$_2$ emission.
The hole at the center is
due to free-free absorption of the arms of Sgr A West against Sgr A East.
The 90cm image
also shows the ionized streamers in absorption at the position of the
inner and outer gaps.
The elongated structure of Sgr A East, which is
saturated in this image, delineates the bright shell of Sgr A East.
The western edge of the shell is weaker in its surface
brightness due to optical depth effect but what is clear is a deviation
from a shell-like geometry to the west of the CND.
We also note that the CND is clearly surrounded by
nonthermal emission (Pedlar et al. 1989) not only
to its western edge but also to its eastern edge. The prominent eastern shell of Sgr A East
lies near $\alpha = 17^h 42^m 38^s$ which is about 1.5$'$ east of the
eastern edge of the CND. Figure 5 presents a more close up view of the
90cm emission with a higher resolution
where contours of HCN emission (Wright et al. 1987)
appear to be surrounded by Sgr A East.
In addition, the 90cm emission
shows an elongated weak nonthermal feature within the hole.
The striking feature in this image is the continuum emission
crossing the bar of ionized gas
associated with Sgr A West.
This shows clearly a lack of free-free absorption of a segment of
the bar against
Sgr A East.
Figure 6 shows contours of 90cm emission superposed on a greyscale 1.2cm
continuum image of Sgr A West showing the extent of the elongated low-frequency
feature with respect to Sgr A West and Sgr A$^*$.
The presence of low-frequency 90cm emission from the direction of
the bar and Sgr A$^*$ is surprising because the
bar of ionized gas is known to be the
brightest feature with the highest ionized density in Sgr A West.
If Sgr A East lies behind Sgr A West, the bar should have
the largest optical depth at 90cm.
This suggests that the Sgr A East shell is
not totally behind Sgr A West as the elongated feature
which is probably a nonthermal feature associated with Sgr A East
crosses
the hole created by free-free absorption. In fact, the comparison
of the molecular and ionized material with the 90cm image,
as presented in Figures 5 and 6, gives the impression of
an anti-correlation between the molecular material and the
90cm emission. The lack of nonthermal emission from Sgr A East
versus the free-free absorption of thermal Sgr A West against
Sgr A East will be considered in more detail elsewhere.
Radio continuum emission at $\lambda$1.2cm traces
the distribution of thermal gas in
Sgr A West and is unlikely to be contaminated by the nonthermal Sgr A East
given its steep radio spectrum in radio wavelengths (Pedlar et al. 1989).
Figure 7 shows contours of
HCN emission superposed against the 1.2cm greyscale of Sgr A West.
This figure shows the prominent three arms of Sgr A West and a
faint Arm of ionized gas which appears to delineate the outer western edge
of the CND.
This new and extended ionized feature runs between
$\alpha = 17^h 42^m 27^s, \delta = -28^0 59' 50''$
and $\alpha = 17^h 42^m 28^s, \delta = -28^0 59' 15''$
and has a typical brightness between 0.5 to 1.5
mJy/beam
with a beam size of 1.58''$\times0.76''$. There is also a peak
of 90cm emission coincident with the new high-frequency 1.2cm
feature near
$\alpha = 17^h 42^m 26.5^s, \delta = -28^0 59' 30''$
with a flux density of 80 mJy in a beam of $9.5''\times5.1''$.
It appears that the western edge of the CND
is sandwiched by two ionized layers from east and west along the inner and outer
edges of the CND, respectively.
The eastern or the inner layer of the CND is traced by
the S. arm (i.e. the Western Arc)
and is ionized by the strong UV radiation
field originated by the IRS 16 cluster.
The western or the outer layer of ionized gas is identified as a new and weakly emitting
arm of ionized gas tracing the outer edge of the CND. The latter can either
be due to photoionization by the radiation field from nearby stars
lying outside the CND or by shocked gas as a result of an interaction
with Sgr A East as a J-type shock is driven into the CND.
\begin{figure}
\caption{ Contours of H$_2$ distribution superimposed on the 90cm emission
from Sgr A East
with a spatial resolution of 16.5$''\times10''$.
Sgr A West is absorbed against Sgr A East at this low frequency.}
\end{figure}
\begin{figure}
\caption{Contours of HCN emission (G\"usten et al. 1987; Wright et al. 1987)
with a spatial resolution of 5-7$''$
superposed on the
central region of the 90cm distribution
with a spatial resolution of 9.5$''\times5.1''$.}
\end{figure}
\begin{figure}
\caption{Contours of 90cm emission as seen in
figure 5 are superposed on
the greyscale distribution of 1.2cm emission
with a spatial resolution of 1.6$''\times0.8''$. The star marks the
position of Sgr A$^*$.}
\end{figure}
\begin{figure}
\caption{A greyscale distribution of ionized gas at 1.2cm
as shown in figure 6
superimposed on contours of HCN distribution}
\end{figure}
\subsection{Dark Patches}
We note prominent
dark patches
distributed throughout the H$_2$ images. These dark features
which are best seen in NICMOS images in Figure 2
are extended and appear to
be concentrated prominantly along
the inner and outer gaps of the CND and
along the
remarkably extended structure NW of the H$_2$ filament.
Although the lack of proper removal of stellar continuum
contributes to the absorption features in NICMOS images, all the
features noted here are extended, have been detected in continuum
images and show
atomic or
molecular couterparts. Estimates of the extiction toward these
patches will be given elsewhere.
An example of the dark feature in Figure 2 is
located near the N. lobe but to the south of the
inner gap of the CND near
$\alpha = 17^h 42^m 30^s, \delta = -28^0 58' 50''$.
This feature coincides with the well-known OI and
dust feature
within the CND (Davidson et al 1992; Jackson et al. 1993; Telesco et al.
1996). The OI feature is thought to lie
in the CND with H$_2$ density of 3$\times10^5$ cm$^{-3}$
and a mass of 300 \, \hbox{$\hbox{M}_\odot$}
suggesting
that the dark feature in this region is located at the Galactic center.
The strongest support for this suggestion
comes from the distribution
of radio continuum
emission at 1.2cm
and the distribution of [FeII] feature based on the VLA and HST/NICMOS
measurements.
Figure 8a shows the N. Arm of Sgr A West and a faint semi-circular shell
structure surrounding the northern tip of the N. Arm. This weak
shell-like ionized feature with a typical brightness of
0.2 mJy within a beam of 0.3$''\times0.2''$ coincides with the outer rim of
the OI/dust feature and the dark H$_2$ patch.
This shell of ionized feature suggests that the
the dark patch is photoionized externally.
Figure 8b shows the
distribution of [FeII] in the same region as shown in Figure 8a.
A dearth of [FeII] emission corresponding to the
northern half of the N. Arm is seen at the position of H$_2$
dark patch. Both the [FeII] and H$_2$ features are
surrounded by a rim of ionized gas in the
1.2cm image.
We note a lack of correlation along the Northern arm
in 1.2cm and [FeII] images,
as shown in figures 8a and b.
The northern half of the N. Arm also shows
a sudden drop by a factor of 2 in
its brightness to the north of
$\alpha = 17^h 42^m 29.7^s, \delta = -28^0 59'$ (Yusef-Zadeh and Wardle 1993).
Considering that free-free emission at 1.2cm is optically thin, the drop in the
emission measure as the Northern arm crosses the shell
is a strong support for the association of the OI/dust feature with
the N. arm.
\begin{figure}
\caption{(left) A greyscale distribution of ionized gas
at 1.2cm with a resolution of 0.3$''\times0.2''$
showing the Northern arm and
a shell of weak ionized gas around its northern tip.
(right) A greyscale distribution of
[FeII] ionized gas showing a lack of emission centered on
the northern half of the N. arm. }
\end{figure}
The coincidence of
OI/dust emitting feature with the dark features and
an ionized rim
surrounding them
suggests that they are all located within the CND.
Further support comes from
a drop in the emission measure of optically thin free-free emission
from the N. Arm at 1.2cm. A detailed examination of the
ionized gas in Sgr A West shows also a number of dramatic dark features
which can not be explained by a lack of short-spacing data. These
dark continuum features are likely to delineate the region of
cold dust and gas where the emission measure drops significantly.
Recent proper motion results show the direction of the ionized flow
supporting the infall picture of the N. Arm toward the Galactic center
(Yusef-Zadeh, Roberts and Biretta 1998; Zhao and Goss 1998). The large-scale
distribution of dark features in the H$_2$ images suggests that the 50 \, \hbox{km}\,\hbox{s}^{-1}\
could be responsible for the infall of neutral material toward the Galactic center
as evidenced by the disturbed molecular material in the CND at the location of the
inner and outer gaps.
The location of the dark patches along the N. Arm, near
Sgr A$^*$, suggests that the atomic and molecular gas
survives well within the cavity of ionized gas.
We also note the darkest feature lying close to the
IRS 16 cluster with very bright continuum emission. Because of
the strong stellar emission from the IRS 16 cluster, this dark feature may be
due to
artefacts from imperfect continuum subtraction.
However, it is interesting
to note that the dark feature coincides
with the highly blue shifted
H$_2$CO, OH, HI and HCO$^+$ absorption
features, with velocities of about --190 \, \hbox{km}\,\hbox{s}^{-1}\.
(Marr et al. 1993; Pauls et al. 1993; Yusef-Zadeh et al.1993;
Yusef-Zadeh 1994; Zhao et al. 1995). The kinematic and spatial distribution of the
absorbing gas suggest that
the gas is located
at the Galactic center
(see Liszt and Burton 1995 for an alternative interpretation).
\section{Discussion}
The new H$_2$ filament
running parallel to the nonthermal shell of Sgr A East
supports the picture that the impact of the nonthermal shell of
Sgr A East with a molecular cloud is responsible
for shocked H$_2$ emission from the filament.
Additional support for the shock hypothesis comes from the sheet-like
morphology and the location of maser source C
along the filament. The radial velocity of the
C maser is about 43 \, \hbox{km}\,\hbox{s}^{-1}\ which is within the
H$_2$ velocity range of velocities of 40 and 80 \, \hbox{km}\,\hbox{s}^{-1}\,
as shown in Figure 3. These so-called supernova masers are
thought to be excellent tracers of
shocks and are seen primarily in regions where
SNR's are interacting with molecular clouds (Wardle et al. 1998;
Green et al. 1998; Frail et al. 1996). The lack of any thermal continuum
or Br $\gamma$ emission at the location of the new filament is consistent
with the
C-type shocks being responsible for exciting the OH gas (Wardle et al. 1998).
The compelling evidence that the new filament traces the shocked molecular
gas by the expansion of the Sgr A East shell into the
cloud raises the question of the location of the filamentary shocked gas with
respect to the CND. It is possible that either the
shocked gas is associated with the 50 \, \hbox{km}\,\hbox{s}^{-1}\ cloud and it happens to be aligned
fortuitously along the outer edge of the CND or that the
outer edge of the CND is shocked externally as a result of its interaction
with Sgr A East. What follows is a series of arguments
in support of
the outer boundary of the CND being collisionally
excited by the
expansion of the nonthermal Sgr A East
into the CND.
First,
the anti-correlation of the streamers of
ionized gas and molecular gas at the positions of the inner and outer
H$_2$ gaps
show compelling morphological evidence for the association of
the streamers of Sgr A West and the new H$_2$ filament which
traces the shocked gas associated with the 50 \, \hbox{km}\,\hbox{s}^{-1}\ molecular cloud.
Second,
the inner and outer boundary where the gaps are observed show similar
velocity distributions suggesting that
they both must be part of the molecular ring.
A ridge of H$_2$
emission linking the new filament
to the inner edge of the CND furthers the support for the physical
association between them.
Third, the evidence that Sgr A East and the CND are physically
interacting with each other comes from detailed 90cm
observations
of the Galactic center.
In these images, the western and eastern edges of
the CND appear to be
engulfed by a shell of radio continuum emission at 90cm.
The emission
at these low-frequencies are due to nonthermal processes and
is expected to be associated with Sgr A East.
In addition, the evidence for low-frequency emission from the region of the
bar and absorption of the arms of Sgr A West add further
evidence that Sgr A East surrounds the CND.
The high frequency
radio emission delineating the outer edge of the Western Arc of the CND
could also be used as an argument for the interaction hypothesis if the
radiation field outside the CND is not strong enough to ionize the outer
edge of the CND. A lack of [FeII] emission from the northern half of
the N. arm and its correlation with
OI/dust emission and H$_2$ dark patches as well as a number of kinematically
disturbed features in the CND all could be used as implications of
a scenario in which a cold dust cloud plunges into the CND by an energetic
phenomena.
Lastly, the recombination line emission toward Sgr A East and
Sgr A$^*$ show stimulated thermal gas at velocities near 50 \, \hbox{km}\,\hbox{s}^{-1}\ (Anantharamaiah et al.
1998; Roberts and Goss 1993). It is possible
that the
thermal ionized gas is associated with the 50 \, \hbox{km}\,\hbox{s}^{-1}\ molecular cloud. If so,
a part of the 50 \, \hbox{km}\,\hbox{s}^{-1}\ molecular cloud with which Sgr A East is interacting
must be in front the of Galactic center.
\section{Summary}
The conclusion that
Sgr A East and the CND are interacting is inescapable based on
various lines of evidence and arguments given above.
A further test of
this model could be done by
measuring the intensity ratio of v=2--1 and 1--0 S(1) line emission from the
region already observed presented in v=1--0 S(1) line.
The implication of such an interaction is that the inverse Compton
scattering of IR and UV photons should not be ignored in modeling
the high-energy
emission from the Galactic center. Furthermore, the disturbance of the
gas in the inner edge of the CND is caused by the impact of Sgr A East.
These suggest that
Sgr A East is responsible for pushing
gas cloud into the
cavity. In this scenario, the N. Arm traces the
trajectory of the infalling cloud as it is ionized by the strong
UV radiation field of the IRS 16 cluster. The three-dimensional motion
of ionized gas based on proper motion and radial velocity
measurements (Yusef-Zadeh, Roberts and Biretta 1998; Zhao and Goss 1998)
is consistent with this picture.
The dark patches seen within the inner 30$''$ of Sgr A$^*$,
in the CND coincident with OI feature, in the inner and outer gaps and the
extended dark feature to the NW of the filament in NICMOS image of H$_2$ distribution
gives the impression that the 50 \, \hbox{km}\,\hbox{s}^{-1}\ molecular cloud may in fact be
responsible for feeding the Galactic center (see Coil and Ho 1998).
Lastly, this interaction
picture of Sgr A East and the CND may
shed some new light on the possibility
of the common origin of the IRS 16 cluster and the
progenitor of stars
responsible for the formation of Sgr A East.
\acknowledgments{ We thank Mel Wright for providing the HCN image and
M. Ashley, S. Ryder and J. Storey for assistance with using UNSWIRF.
Basic research in
radio astronomy at the Naval Research Laboratory is supported by the
Office of Naval Research.
TJWL is
supported by an NRL-National Research Council Associateship.}
|
\section{Introduction}
{}In recent years substantial progress has been made in understanding
orientifold compactifications. Various six dimensional orientifold vacua
were constructed, for instance, in \cite{PS,GP,GJ,bij}. Generalizations of
these constructions to four dimensional orientifold vacua have also been
discussed in detail in \cite{BL,Sagnotti,KS,Zw,ibanez,KST}. In many cases
the world-sheet approach to orientifolds is adequate and gives rise to
consistent anomaly free vacua in six and four dimensions. However, it was
pointed out in \cite{KST} that there
are cases where the perturbative orientifold description misses certain
non-perturbative sectors which are nonetheless present and must be taken
into account. In certain cases this inadequacy results in obvious
inconsistencies such as lack of tadpole and anomaly cancellation. Examples
of such cases were discussed in \cite{Zw,ibanez,KST}. In other cases,
however, the issue is more subtle as the non-perturbative states arise in
anomaly free combinations, so that they are easier to miss.
{}In certain four dimensional ${\cal N}=1$ supersymmetric cases
Type I-heterotic duality enables one to argue along the lines of
\cite{ZK,KS,KST} that the non-perturbative states become heavy
and decouple once the
corresponding orbifold singularities are appropriately blown
up. In particular, these are the ${\bf Z}_3$ \cite{Sagnotti,ZK},
${\bf Z}_7$, ${\bf Z}_3\otimes {\bf Z}_3$ and ${\bf Z}_6$ \cite{KS},
${\bf Z}_2\otimes {\bf Z}_2\otimes {\bf Z}_3$ \cite{223}, and
$\Delta(3\cdot3^2)$ (the latter is non-Abelian) \cite{class}
orbifold cases. As was argued in \cite{KST}, the same, however, does not
seem to be the case for
other tadpole free orientifolds such as the ${\bf Z}_2\otimes{\bf Z}_6$,
${\bf Z}_3\otimes{\bf Z}_6$ and ${\bf Z}_6\otimes{\bf Z}_6$ \cite{Zw}, and
${\bf Z}_{12}$ \cite{ibanez}
orbifold cases. The purpose of this paper is to understand
non-perturbative states in such orientifold compactifications. In
particular, in some cases we are able to obtain the massless states arising
in such sectors.
{}The origin of non-perturbative states in orientifold compactifications
can already be understood in six dimensions. Thus, in the K3 orbifold examples
of \cite{GJ} the orientifold projection is not $\Omega$, which we will use
to denote that in the smooth K3 case, but rather $\Omega J^\prime$, where
$J^\prime$ maps the $g$ twisted sector to its conjugate $g^{-1}$ twisted
sector (assuming $g^2\not=1$) \cite{pol}. Geometrically this can be viewed
as a permutation of two ${\bf P}^1$'s associated with each fixed point of
the orbifold \cite{KST}. (More precisely, these ${\bf P}^1$'s correspond to
the orbifold blow-ups.) This is different from the orientifold projection
in the smooth case where (after blowing up) the orientifold projection
does {\em not} permute the two ${\bf P}^1$'s. In the case of the
$\Omega J^\prime$ projection the ``twisted'' open string sectors corresponding
to the orientifold elements $\Omega J^\prime g$ are absent
\cite{blum,KST}. However, if the orientifold projection is $\Omega$, then
the ``twisted'' open string sectors corresponding
to the orientifold elements $\Omega g$ are present. In fact, these states
are non-perturbative from the orientifold viewpoint and are required for
gravitational anomaly cancellation in six dimensions. As we will see, in
certain cases Type I-heterotic duality will allow us to understand such
sectors.
{}In four dimensional orientifolds with ${\cal N}=1$ supersymmetry there
are always sectors (except in the ${\bf Z}_2\otimes {\bf Z}_2$ model of
\cite{BL} which is completely perturbative from the orientifold viewpoint)
such that there is only one ${\bf P}^1$ per fixed point, so only the
$\Omega$ orientifold projection is allowed. This results in
non-perturbative ``twisted'' open string sectors, which, as we have
already mentioned, decouple in certain cases once the appropriate blow-ups
are performed. In other cases we must include these states to obtain the
complete description of a given orientifold.
{}Some of the non-perturbative orientifolds have perturbative heterotic
duals. However, non-perturbative orientifolds with, say, D5-branes are
non-perturbative from the heterotic viewpoint. In this paper we are
therefore exploring some vacua in the region ${\cal D}$ in Fig.1 (which has
been borrowed from \cite{class}). Thus, the ${\bf Z}_6^\prime$ orbifold
compactification we discuss in section III is an example of a four
dimensional {\em chiral} ${\cal N}=1$ supersymmetric string vacuum which is
non-perturbative from both orientifold and heterotic points of view.
{}In the next section we give examples of six dimensional non-perturbative
orientifolds. These are illustrative as the gravitational anomaly
cancellation requirement in six dimensions is rather constraining which
makes it easier to discuss such vacua. Once we understand these six
dimensional examples, we move down to four dimensions in section III.
\section{Six Dimensional Examples}
{}In this section we will give an example of a non-perturbative six
dimensional orientifold with ${\cal N}=1$ supersymmetry. This vacuum
corresponds to the $\Omega$ orientifold of Type IIB on the ${\bf Z}_6$
orbifold limit of K3 (that is, K3 in this case is $T^4/{\bf
Z}_6$). Alternatively, this model can be viewed as a Type I
compactification on the same K3 with D5-branes. In the dual heterotic
picture these D5-branes are mapped to small instantons.
{}Thus, consider Type IIB on ${\cal M}_2=T^4/{\bf Z}_6$ (with trivial NS-NS
$B$-field), where the
generator $g$ of ${\bf Z}_6(\approx{\bf Z}_3\otimes {\bf Z}_2)$ acts on
the complex coordinates $z_1,z_2$ on
$T^4$ as follows:
\begin{equation}
g z_1=\omega z_1~,~~~g z_2=\omega^{-1} z_2~,
\end{equation}
where $\omega\equiv\exp(2\pi i/6)$. This theory has ${\cal N}=2$
supersymmetry in six dimensions.
{}Next, we consider the $\Omega$ orientifold of this theory. Note that, as
we have already mentioned in the previous section, the $\Omega$ projection
acts as in the smooth K3 case. This, in particular, implies that we must
first blow up the orbifold singularities before orientifolding.
{}Thus, after orientifolding the closed string sector contains only one
tensor multiplet (but no extra tensor multiplets), and 20 hypermultiplets
(plus the corresponding ${\cal N}=1$ supergravity multiplet). Here 2 of the
above hypermultiplets come from the untwisted sector, 2 more arise in the
${\bf Z}_6$ twisted sectors $g,g^{-1}$, $10(=2\times 5)$ hypermultiplets
come from the ${\bf Z}_3$ twisted sectors $g^2,g^{-2}$ (note that 5 linear
combinations of the original 9 points fixed under the ${\bf Z}_3$ action
are invariant under the ${\bf Z}_2$ twist), and, finally, 6 hypermultiplets
come from the ${\bf Z}_2$ twisted sector $g^3$ (note that 6 linear
combinations of the original 16 points fixed under the ${\bf Z}_2$ action
are invariant under the ${\bf Z}_3$ twist).
{}As to the open string sector, there are 32 D9-branes plus 32
D5-branes. This is required by the untwisted tadpole cancellation
conditions. In the following we will consider the case where all 32
D5-branes are sitting on top of each other at one of the orientifold
5-planes which we choose to be that at the origin of the orbifold (that is,
at $z_1=z_2=0$). We can view the usual (``untwisted'') 99 and 55 open
string sectors as the ``$\Omega$-twisted'' sector of the orientifold.
Similarly, the 59 sector can be viewed as ``$\Omega g^3$-twisted''
sector. These sectors are perturbative from the orientifold point of
view. That is, these sectors can be understood in the usual world-sheet
picture using the standard D-brane techniques. There are, however,
additional sectors in this model which do not possess such a simple
description. These are the ``$\Omega g^2$- and $\Omega g^{-2}$-twisted''
sectors, which can be viewed as the ${\bf Z}_3$ twisted 99 plus 55 sectors, as
well as the ``$\Omega g$- and $\Omega g^{-1}$-twisted'' sectors, which can be
viewed as the ${\bf Z}_3$ twisted 59 sector. These ${\bf Z}_3$ twisted open
string sectors are {\em non-perturbative} from the orientifold
viewpoint. Nonetheless, using Type I-heterotic duality we can (in this
particular example) understand these sectors and obtain the corresponding
massless states.
{}The key observation here is the following. First, recall the ${\bf Z}_3$
model of \cite{GJ}. This model is obtained as follows. Consider Type IIB on
${\cal M}_2=T^4/{\bf Z}_3$ (with trivial NS-NS $B$-field), where the
generator $\theta$ of ${\bf Z}_3$ (which is a subgroup of the above ${\bf
Z}_6$ orbifold group with the identification $\theta=g^2$) acts on
the complex coordinates $z_1,z_2$ on $T^4$ as follows:
\begin{equation}
\theta z_1=\alpha z_1~,~~~\theta z_2=\alpha^{-1} z_2~,
\end{equation}
where $\alpha\equiv\omega^2=\exp(2\pi i/3)$. Next, consider the $\Omega
J^\prime$ orientifold of this theory, where $\Omega$ is the same as above
(that is, it acts as in the smooth K3 case), whereas $J^\prime$, which was
discussed in the previous section, maps the $\theta$ twisted sector to the
$\theta^{-1}$ twisted sector \cite{pol}. After orientifolding the closed
string sector contains (along with the corresponding ${\cal N}=1$
supergravity multiplet) 10 tensor multiplets (one usual tensor multiplet
plus 9 extra tensor multiplets) and 11 hypermultiplets. In the open string
sector we have 32 D9-branes (but no D5-branes), and the action of the
orbifold group on the Chan-Paton charges is given by the corresponding
projective representation of ${\bf Z}_3$ in terms of the $16\times 16$
matrices\footnote{Throughout this paper we work with $16\times 16$
(rather than $32\times 32$) Chan-Paton matrices for we choose not to count the
orientifold images of the corresponding D-branes.}
$\{\gamma_1,\gamma_\theta,\gamma_{\theta^{-1}}\}$, where
$\gamma_1={\bf I}_{16}$ is the
$16\times 16$ identity matrix, $\gamma_{\theta^{-1}}=\gamma^{-1}_\theta$, and
(up to equivalent representations) $\gamma_\theta$ is given by
\begin{equation}\label{Z_3}
\gamma_\theta={\mbox{diag}}(\alpha{\bf I}_4,\alpha^{-1}{\bf I}_4,{\bf
I}_8)~.
\end{equation}
Note that the above form of $\gamma_\theta$ is dictated by the
corresponding twisted tadpole cancellation conditions \cite{GJ}.
The massless spectrum of the open string sector of this model consists of
gauge vector
supermultiplets of $U(8)\otimes SO(16)$ as well as the corresponding
charged hypermultiplets transforming in the irreps $({\bf 28},{\bf 1})$ and
$({\bf 8},{\bf 16})$ of the gauge group. Note that the above spectrum
satisfies the gravitational anomaly cancellation condition \cite{anomalies}
\begin{equation}\label{anom}
n_H-n_V=273-29 n_T~,
\end{equation}
where the number of tensor multiplets $n_T=10$, the number of
hypermultiplets $n_H=167$ (11 of which arise in the closed string sector,
while the other 156 come from the open string sector), and the number of
vector multiplets (all of which come from the open string sector) $n_V=184$.
{}The above model does not possess a perturbative heterotic dual as it
contains extra tensor multiplets. The reason for their appearance is that
the orientifold projection was chosen as $\Omega J^\prime$. On the other
hand, had we chosen the orientifold projection (after the appropriate
blow-ups) to be $\Omega$, we would have obtained a vacuum with no extra tensor
multiplets in the closed string sector. If, however, we choose the
Chan-Paton matrices to be the same as above (which in the 99 sector would
lead to the same massless spectrum), we would obtain an anomalous model
where the condition (\ref{anom}) is not satisfied. This signals that the
``naive'' orientifold (that is, world-sheet) approach in this case misses
some states whose existence is dictated by the anomaly cancellation
conditions.
{}Fortunately, in the above case we can actually understand the origin of the
``missing'' states by using Type I-heterotic duality. The point is that
the $\Omega$ orientifold of Type IIB on the (appropriately blown up)
orbifold $T^4/{\bf Z}_3$ is the same as Type I compactified on the same
space. The latter should be dual to a {\em perturbative} heterotic
compactification as we have no D5-branes in this case. In fact, we can
identify the corresponding heterotic vacuum as being the $T^4/{\bf Z}_3$
compactification with the gauge bundle defined by the orbifold action on
the ${\mbox{Spin}}(32)/{\bf Z}_2$ degrees of freedom given by the matrices
$\{{\bf W}_1,{\bf W}_\theta,{\bf
W}_{\theta^{-1}}\}\equiv\{\gamma_1,\gamma_\theta,\gamma_{\theta^{-1}}\}$.
Here the $16\times 16$ ${\bf W}$-matrices act on the
${\mbox{Spin}}(32)/{\bf Z}_2$ momentum states as the corresponding
rotations. (Alternatively, we can describe this action in terms of the
corresponding shifts of the ${\mbox{Spin}}(32)/{\bf Z}_2$ momentum lattice.)
The resulting heterotic model is perturbatively well defined (that is, it
satisfies all the corresponding consistency requirements such as modular
invariance). The untwisted sector of this model contains the ${\cal N}=1$
supergravity multiplet, one tensor multiplet and 2 neutral hypermultiplets
(these states are mapped to the untwisted closed string sector of the dual
Type I vacuum); it also contains the
gauge vector supermultiplets of $U(8)\otimes SO(16)$ as well as
the corresponding
charged hypermultiplets transforming in the irreps $({\bf 28},{\bf 1})$ and
$({\bf 8},{\bf 16})$ of the gauge group (these states are mapped to the
``untwisted'' open string sector of the dual Type I vacuum described above).
The twisted sector of the heterotic model contains 18 hypermultiplets
(neutral under the non-Abelian subgroups but fractionally\footnote{These
charges are fractional in the normalization (which is conventional in
discussing the orientifold models) where the fundamental of $SU(8)$ has
charge $+1$ under the corresponding $U(1)$ subgroup. In the following we
will omit the $U(1)$ charges which generically are fractional in the
``twisted'' open string sectors. In certain cases one can determine these
charges using the Type I-heterotic duality (as on the heterotic side these
charges can be computed in the corresponding twisted sectors). Nonetheless,
we choose not to do this here as various subtleties involved in such
computations might have obscured the main point of this paper.}
charged under the $U(1)$
subgroup) plus 9 additional hypermultiplets transforming in the irrep $({\bf
28},{\bf 1})$ of $U(8)\otimes SO(16)$ (these states are also fractionally
charged under the $U(1)$ subgroup). The above 18 hypermultiplets correspond
to the orbifold blow-up modes, and are mapped to the twisted closed string
sector hypermultiplets on the Type I side. Note that the 9 twisted
hypermultiplets (on the heterotic side) charged under the $SU(8)$ subgroup
have no match in the orientifold picture. That is, these are
non-perturbative from the orientifold viewpoint. Nonetheless, they must
exist in the corresponding Type I model. In fact, they are precisely the
states required to satisfy the anomaly cancellation condition
(\ref{anom}). These states (which are charged under the 99 gauge group) can
be thought of as arising in ${\bf Z}_3$ twisted 99 sector. Type I-heterotic
duality is then a tool which allows us to understand such states (by doing a
perturbative heterotic calculation) which do not possess a world-sheet
description in the orientifold language. For the later convenience we give
the massless spectrum of the ${\bf Z}_3$ model in Table I.
{}Let us now go back to the ${\bf Z}_6$ model we have defined in the
beginning of this section. Since the ${\bf Z}_3$ model with the $\Omega$
orientifold projection and the choice of the gauge bundle (\ref{Z_3})
is consistent, the ${\bf Z}_6$ model should also be
consistent as long as we make sure that the corresponding ${\bf Z}_2$ model
is consistent. That is, we must choose the action of the ${\bf Z}_2$ twist
on the Chan-Paton factors appropriately.
{}A consistent solution for the Chan-Paton matrices in the ${\bf Z}_2$
model was given in \cite{PS,GP}. Thus, consider Type IIB on ${\cal
M}_2=T^4/{\bf Z}_2$ (with trivial NS-NS $B$-field), where the generator $R$
of ${\bf Z}_2$ reflects the complex coordinates $z_1,z_2$ on $T^4$:
$Rz_{1,2}=-z_{1,2}$. Next, consider the $\Omega$ orientifold of this
theory. The closed string sector contains (along with the corresponding
${\cal N}=1$ supergravity multiplet) one tensor multiplet and 20
hypermultiplets (4 of which come from the untwisted sector, and the other
16 come from the ${\bf Z}_2$ twisted sector). In the open string sector we have
32 D9-branes plus 32 D5-branes. Let all of the D5-branes sit on top of each
other at the orientifold 5-plane located at the origin of the
orbifold. Then the twisted D9- and D5-brane Chan-Paton matrices
$\gamma_{9,R}$ and $\gamma_{9,R}$ are given by
\begin{equation}
\gamma_{9,R}=\gamma_{5,R}={\mbox{diag}}(i{\bf I}_8 , -i{\bf I}_8)~.
\end{equation}
The massless spectrum of the open string sector consists of gauge vector
supermultiplets of $U(16)_{99}\otimes U(16)_{55}$ as well as the
corresponding charged hypermultiplets transforming in the irreps $2\times
({\bf 120};{\bf 1})$ (99 sector), $2\times ({\bf 1};{\bf 120})$ (55 sector),
$({\bf 16};{\bf 16})$ (59 sector). This spectrum is summarized in Table I,
and it can be seen to satisfy the
anomaly cancellation condition (\ref{anom}).
{}The ``untwisted'' 99, 55 and 59 sector
matter content in the full ${\bf Z}_6$ model is the same as in the ${\bf
Z}_6$ model of \cite{GJ}. In particular, the orbifold action on the
Chan-Paton charges is given by the
corresponding twisted Chan-Paton matrices (which are identical for both D9-
and D5-branes in the present setup):
\begin{eqnarray}\label{CP1}
&&\gamma_{9,\theta}=\gamma_{5,\theta}=
{\mbox{diag}}(\alpha {\bf I}_4 , \alpha^{-1} {\bf I}_4, {\bf I}_8)~,\\
\label{CP2}
&&\gamma_{9,R}=\gamma_{5,R}={\mbox{diag}}(i{\bf I}_2, -i{\bf I}_2 ,
i{\bf I}_2, -i{\bf I}_2, i{\bf I}_4, -i{\bf I}_4)~.
\end{eqnarray}
Here $\theta=g^2$ and $R=g^3$ ($g$ is the generator of the ${\bf Z}_6$
orbifold group). Thus, the gauge group is $[U(4)\otimes U(4)\otimes
U(8)]_{99}$ in the 99 sector. Since we have placed all the D5-branes on
top of each other at the orientifold 5-plane located at the origin, the 55
sector gauge group $[U(4)\otimes U(4)\otimes
U(8)]_{55}$ is the same as in the 99 sector. The matter content of the
``untwisted'' 99, 55 and 59 sectors (these are the sectors that are
perturbative from the orientifold viewpoint) is given in Table I.
{}We are now ready to determine the ``twisted'' 99, 55 and 59 sector
matter content in the full ${\bf Z}_6$ model. The twisted 99 sector states
are straightforward to determine. We have already discussed the matter
arising in the twisted 99 sector in the ${\bf Z}_3$ model. To obtain the
corresponding states in the ${\bf Z}_6$ model we must project onto the
${\bf Z}_2$ invariant states. Note that the original 9 fixed points can be
organized into 5 linear combinations invariant under the ${\bf Z}_2$ twist
plus 4 linear combinations which pick up a minus sign under the ${\bf Z}_2$
twist. Thus, after taking into account the ${\bf Z}_2$ action on both the
fixed points and Chan-Paton charges, in the twisted 99 sector we have 5
copies of hypermultiplets in the irreps $({\bf 6},{\bf 1},{\bf 1})_{99}$
and $({\bf 1},{\bf 6},{\bf 1})_{99}$, and also 4 copies of hypermultiplets
in the irrep $({\bf 4},{\bf 4},{\bf 1})_{99}$.
{}The twisted 55 sector is a bit more subtle. Since we have placed all of
the D5-branes at the origin (which is a fixed point of both the ${\bf Z}_3$
and ${\bf Z}_2$ twists), it follows that we have only one copy of
hypermultiplets in irreps $({\bf 6},{\bf 1},{\bf 1})_{55}$
and $({\bf 1},{\bf 6},{\bf 1})_{55}$. (No hypermultiplets in the irrep
$({\bf 4},{\bf 4},{\bf 1})_{55}$ appear as the origin is invariant under
the ${\bf Z}_2$ twist.) In particular, the other 8 of the original 9 fixed
points of the ${\bf Z}_3$ twist play no role in this discussion as the
twisted 55 states arise due to the {\em local} geometry near the origin.
{}Finally, the twisted 59 sector contains one copy of hypermultiplets in
the irreps $({\bf 4},{\bf 1},{\bf 1};{\bf 4},{\bf 1},{\bf 1})_{59}$ and
$({\bf 1},{\bf 4},{\bf 1};{\bf 1},{\bf 4},{\bf 1})_{59}$. The multiplicity
of these states follows from considerations similar to those in the 55
sector. As to the gauge charges, they are dictated by the fact the twisted
states of this type (due to their origin discussed in the previous section)
can only be charged under the part of the original gauge group (in this
case $SO(32)$) broken by the corresponding (that is, ${\bf Z}_3$) twist.
{}The full massless spectrum of the ${\bf Z}_6$ model is summarized in Table
I (there we also give the massless spectra of the ${\bf Z}_2$ and ${\bf
Z}_3$ models). It is straightforward to check that the anomaly cancellation
condition (\ref{anom}) is satisfied in this model. Note that the twisted
99 and 55
sectors no longer exhibit the naive ``T-duality''. This is due to the fact
that these sectors do {\em not} arise via a straightforward orbifold
reduction of the corresponding ($SO(32)$) gauge theory (with ${\cal N}=2$
supersymmetry in six dimensions).
{}Thus, the ${\bf Z}_2$ orbifold model of \cite{PS,GP} is perturbative from
the orientifold viewpoint, but is non-perturbative in the heterotic
language. In contrast, the ${\bf Z}_3$ model in Table I is perturbative
from the heterotic viewpoint, but has no perturbative orientifold
description. The ${\bf Z}_6$ model is non-perturbative in both orientifold
and heterotic pictures. Here we should point out that the ${\bf Z}_4$ case
with the $\Omega$ projection (instead of the $\Omega J^\prime$ projection
as in \cite{GJ}) does not appear to be tractable in the above approach as
the action of the ${\bf Z}_4$ twist on the Chan-Paton charges in this case does
not correspond to a perturbative gauge bundle on the heterotic side. Thus,
to understand the ${\bf Z}_4$ case one would need a different approach.
{}We would like to end this section with the following comment. Following
\cite{bij} we could attempt to construct similar non-perturbative
orientifolds with non-trivial NS-NS $B$-field turned on. Unfortunately,
however, we would then lose control over the non-perturbative
(``twisted'') states. Thus, for instance, the ${\bf Z}_3$ model with a
non-zero NS-NS $B$-field of rank 2 does not appear to have a perturbative
heterotic dual.
Nonetheless, in the next section we will see that in four
dimensions one can find cases with non-zero NS-NS $B$-field which can be
understood within the present approach.
\section{Four Dimensional Examples}
{}In this section we extend the discussion in the previous section and
obtain non-perturbative orientifolds with ${\cal N}=1$ supersymmetry in
four dimensions.
{}Thus, consider Type IIB on ${\cal M}_3=T^6/{\bf Z}_6^\prime$ (with
trivial NS-NS $B$-field), where the generators $\theta$ and $R$ of
the ${\bf Z}_3$ and ${\bf
Z}_2$ subgroups of ${\bf Z}_6^\prime$ act on the complex coordinates on
$T^6$ as follows:
\begin{eqnarray}
&&\theta z_1 = z_1~,~~~\theta z_2 =\alpha z_2~,~~~\theta z_3=\alpha^{-1}
z_3 ~,\\
&&Rz_1 = -z_1~,~~~Rz_2 =-z_2~,~~~Rz_3=z_3 ~,
\end{eqnarray}
where $\alpha\equiv\exp(2\pi i/3)$. This theory has ${\cal N}=2$
supersymmetry in four dimensions.
{}Next, consider the $\Omega$ orientifold of this theory. Here $\Omega$
acts as in the smooth Calabi-Yau case. This, in particular, implies that we
must first appropriately blow up (some of) the orbifold singularities
before orientifolding.
{}After orientifolding the closed string sector contains the ${\cal N}=1$
supergravity multiplet, the dilaton plus axion supermultiplet, and also 46
chiral supermultiplets. In particular, 4 of these arise in the untwisted
sector, 18 come from the ${\bf Z}_3$ twisted sector, and ${\bf Z}_2$ and
${\bf Z}_6$ sectors each contribute 12 chiral supermultiplets.
{}As to the open string sector, there are 32 D9-branes and 32
D5-branes. This is required by the untwisted tadpole cancellation
conditions. In the following we will focus on the case where all 32
D5-branes (which wrap the $z_3$ complex direction) are sitting on top of
each other at the orientifold 5-plane located at $z_1=z_2=0$.
{}The ``untwisted'' 99, 55 and 59 sectors in this orientifold (that is, the
sectors with well defined perturbative description) have been obtained in
\cite{Zw,ibanez,KST}. However, in \cite{KST} it was pointed out that there
are also ``twisted'' 99, 55 and 59 sector contributions one must take into
account. This should be clear from the discussion in the previous
section. In fact, following the approach we have used for constructing the
six dimensional ${\bf Z}_6$ model in the previous section, we can also
obtain the massless states in the twisted open string sectors in the four
dimensional ${\bf Z}_6^\prime$ model as
well. Since the discussion in the four dimensional case parallels quite
closely that for the six dimensional example, we simply summarize the
massless spectrum of the ${\bf Z}_6^\prime$ model in Table II. Let us,
however, mention that the multiplicity 3 in the twisted 55 and 59 sectors
is due to the fact that there are 3 fixed points in the $z_3$ direction
which the D5-branes wrap. On the other hand, the original 9 fixed points
relevant in the discussion for the twisted 99 sector decompose into 6
linear combinations invariant under the ${\bf Z}_2$ twist plus 3 linear
combinations that pick up a minus sign under the ${\bf Z}_2$ twist. As to
the orbifold action on the Chan-Paton factors, in this case it is given by
the same Chan-Paton matrices (\ref{CP1}) and (\ref{CP2})
as in the six dimensional case.
{}Note that the four dimensional ${\bf Z}_6^\prime$
orbifold model (with trivial
NS-NS $B$-field) given in Table II is chiral. In fact, it is an example of
a chiral ${\cal N}=1$ supersymmetric string vacuum which is
non-perturbative from both the orientifold and heterotic viewpoints. The
spectrum of this model can be seen to be free of non-Abelian gauge
anomalies, and their cancellation in this case in non-trivial. In
particular, the ``untwisted'' open string sectors alone produce an anomaly
free spectrum. The ``twisted'' sector contributions are also anomaly
free. This is why the ``naive'' (that is, perturbative) tadpole cancellation
conditions for this orientifold give rise to an anomaly free
model. Nonetheless, this does not imply that the spectrum obtained via the
orientifold techniques is complete. In fact, the non-perturbative (that is,
``twisted'') sector contributions are non-trivial, and must be taken into
account.
{}Before we end this section, we would like to discuss a similar model with
a non-trivial NS-NS $B$-field turned on. Thus, consider the above ${\bf
Z}_6^\prime$
orbifold model in the presence of the non-zero NS-NS $B$-field in the
$z_1$ direction\footnote{As we have already mentioned in the previous section,
turning on the $B$-field in the $z_2$ and/or $z_3$ directions would result
in a loss of control over the twisted open string states as they would have
no perturbative description on the heterotic side.}. Thus, let
$T^6=T^2\otimes T^4$, where the two-torus $T^2$ is parametrized by the
complex coordinate $z_1$ (on which the ${\bf Z}_3$ twist $\theta$ acts
trivially), whereas the four-torus $T^4$ is parametrized by the other two
complex coordinates $z_2,z_3$. Turning on the $B$-field (or rank 2)
corresponding to $T^2$ results in rank reduction in both 99 and 55 sectors
by a factor of 2 \cite{Bij,bij}. On the heterotic side this corresponds to
turning on two Wilson lines on the two non-contractable cycles of $T^2$
such that they reduce the original $SO(32)$ gauge symmetry down to
$SO(16)$ \cite{ZK,BN}. In particular, the first Wilson line corresponding
to, say, the $a$-cycle of $T^2$ breaks $SO(32)$ down to $SO(16)\otimes
SO(16)$. Then the second Wilson line corresponding to the $b$-cycle of
$T^2$ breaks $SO(16)\otimes SO(16)$ to its diagonal $SO(16)_{diag}$
subgroup. This can be viewed as a freely acting ${\bf Z}_2$ orbifold which
permutes the two $SO(16)$ subgroups. In the $SO(16)_{diag}$ language the
${\bf Z}_3$ orbifold action on the ${\mbox{Spin}}(32)/{\bf Z}_2$ degrees of
freedom can then be described in terms of the $8\times 8$ matrices $\{{\bf
W}_1,{\bf W}_\theta,{\bf W}_{\theta^{-1}}\}$, where ${\bf
W}_1={\bf I}_8$, ${\bf W}_\theta={\mbox{diag}}(\alpha{\bf I}_2,
\alpha^{-1}{\bf I}_2,{\bf I}_4)$, and ${\bf W}_{\theta^{-1}}={\bf
W}_\theta^{-1}$. On the orientifold side we then identify the corresponding
$8\times 8$ Chan-Paton matrices via $\{\gamma_1,\gamma_\theta,
\gamma_{\theta^{-1}}\}\equiv\{{\bf W}_1,{\bf W}_\theta,{\bf
W}_{\theta^{-1}}\}$. Such an identification is in fact consistent with the
twisted tadpole cancellation conditions which, in particular, imply that we
must have ${\mbox{Tr}}(\gamma_\theta)=+2$ (instead of
${\mbox{Tr}}(\gamma_\theta)=+4$ as in the case without the $B$-field)
\cite{class}. Thus, we conclude that in this case the heterotic gauge
bundle is indeed perturbative, and we can therefore obtain, say,
the twisted 99 sector states via the corresponding heterotic computation.
{}The spectrum of the above ${\bf Z}_6^\prime$ model with the $B$-field can be
computed in complete parallel with the case without the $B$-field with the
following straightforward modifications. First, the corresponding twisted
Chan-Paton matrices (which we now write in the $8\times 8$ basis) are given by:
\begin{eqnarray}
&&\gamma_{9,\theta}=\gamma_{5,\theta}=
{\mbox{diag}}(\alpha {\bf I}_2 , \alpha^{-1} {\bf I}_2, {\bf I}_4)~,\\
&&\gamma_{9,R}=\gamma_{5,R}={\mbox{diag}}(i, -i,
i, -i, i{\bf I}_2, -i{\bf I}_2)~.
\end{eqnarray}
The gauge group is $[U(2) \otimes U(2) \otimes U(4)]_{99}\otimes
[U(2) \otimes U(2) \otimes U(4)]_{55}$ (all the D5-branes are sitting on
top of each other at the orientifold 5-plane located at $z_1=z_2=0$). The
spectrum of the open string sector can then be read off the spectrum in
Table II (corresponding to the case without the $B$-field) with the obvious
substitutions for the irreps of the corresponding gauge subgroups. Note,
however, that the multiplicities of (both untwisted as well as twisted)
59 sector states is now 2 times that shown in Table II. This doubling of
states is due to the rank-2 $B$-field, and was explained in detail in
\cite{bij}.
{}Before we end this section, let us point out that the approach we have
described in the previous sections can be straightforwardly applied to
study other four dimensional orientifolds with twisted (that is,
non-perturbative) sectors. More concretely, these are the ${\bf Z}_2\otimes
{\bf Z}_6$, ${\bf Z}_3\otimes {\bf Z}_6$ and ${\bf Z}_6\otimes
{\bf Z}_6$ cases whose perturbative (from the orientifold viewpoint)
massless spectra were discussed in \cite{Zw}. (The ${\bf Z}_3\otimes
{\bf Z}_6$ case was also discussed in \cite{ibanez}.) We will not consider
these models in detail in this paper but will come back to them in
\cite{new} where we intend to discuss a more general class of Type I
compactifications, namely, on Voisin-Borcea orbifolds. Note, however, that
the non-perturbative sectors in the ${\bf Z}_{12}$ model of \cite{ibanez}
are not tractable within the above approach as ${\bf Z}_{12}$
contains a ${\bf Z}_4$
subgroup. As we have already mentioned in the previous section, this case
requires techniques other than those employed in this paper.
\acknowledgments
{}I would like to thank Mirjam Cveti{\v c} for valuable discussions.
This work was supported in part by the grant
NSF PHY-96-02074,
and the DOE 1994 OJI award. I would also like to thank Albert and
Ribena Yu for financial support.
\newpage
\begin{figure}[t]
\epsfxsize=16 cm
\epsfbox{figure.eps}
\bigskip
\caption{A schematic picture of the space of four dimensional ${\cal N}=1$ Type I and
heterotic vacua. The region ${\cal A}\cup{\cal B}$ corresponds to perturbative Type I vacua.
The region ${\cal A}\cup{\cal C}$ corresponds to perturbative heterotic vacua. The vacua
in the region ${\cal A}$ are perturbative from both the Type I and heterotic viewpoints. The region ${\cal D}$ contains both non-perturbative Type I and heterotic vacua.}
\end{figure}
\begin{table}[t]
\begin{tabular}{|c|c|l|c|c|}
Model & Gauge Group & \phantom{Hy} Charged & Neutral
& Extra Tensor \\
& &Hypermultiplets & Hypermultiplets
&Multiplets \\
\hline
${\bf Z}_2$ & $U(16)_{99} \otimes U(16)_{55}$ &
$2 ({\bf 120};{\bf 1})_U$ & $20$
& $0$ \\
& & $2 ({\bf 1};{\bf 120})_U$ & & \\
& & $({\bf 16};{\bf 16})_U$ & & \\
\hline
${\bf Z}_3$ & $[U(8) \otimes SO(16)]_{99}$ & $({\bf 28},{\bf 1})_U$ & $20$
& $0$ \\
& & $({\bf 8},{\bf 16})_U$ & & \\
& & $9\times ({\bf 28},{\bf 1})_T$ & & \\
\hline
${\bf Z}_6$ & $[U(4) \otimes U(4) \otimes U(8)]_{99}\otimes$
& $({\bf 6},{\bf 1},{\bf 1};{\bf 1},{\bf 1},{\bf 1})_U$ & $20$ & $0$ \\
& $[U(4) \otimes U(4) \otimes U(8)]_{55}$&
$({\bf 1},{\bf 6},{\bf 1};{\bf 1},{\bf 1},{\bf 1})_U$ & & \\
& & $({\bf 4},{\bf 1},{\bf 8};{\bf 1},{\bf 1},{\bf 1})_U$ & & \\
& & $({\bf 1},{\bf 4},{\bf 8};{\bf 1},{\bf 1},{\bf 1})_U$ & & \\
& & $({\bf 1},{\bf 1},{\bf 1};{\bf 6},{\bf 1},{\bf 1})_U$ & & \\
& & $({\bf 1},{\bf 1},{\bf 1};{\bf 1},{\bf 6},{\bf 1})_U$ & & \\
& & $({\bf 1},{\bf 1},{\bf 1};{\bf 4},{\bf 1},{\bf 8})_U$ & & \\
& & $({\bf 1},{\bf 1},{\bf 1};{\bf 1},{\bf 4},{\bf 8})_U$ & & \\
& & $({\bf 4},{\bf 1},{\bf 1};{\bf 4},{\bf 1},{\bf 1})_U$ & & \\
& & $({\bf 1},{\bf 4},{\bf 1};{\bf 1},{\bf 4},{\bf 1})_U$ & & \\
& & $({\bf 1},{\bf 1},{\bf 8};{\bf 1},{\bf 1},{\bf 8})_U$ & & \\
& & $5\times ({\bf 6},{\bf 1},{\bf 1};{\bf 1},{\bf 1},{\bf 1})_T$ & & \\
& & $5\times ({\bf 1},{\bf 6},{\bf 1};{\bf 1},{\bf 1},{\bf 1})_T$ & & \\
& & $4\times ({\bf 4},{\bf 4},{\bf 1};{\bf 1},{\bf 1},{\bf 1})_T$ & & \\
& & $({\bf 1},{\bf 1},{\bf 1};{\bf 6},{\bf 1},{\bf 1})_T$ & & \\
& & $({\bf 1},{\bf 1},{\bf 1};{\bf 1},{\bf 6},{\bf 1})_T$ & & \\
& & $({\bf 4},{\bf 1},{\bf 1};{\bf 4},{\bf 1},{\bf 1})_T$ & & \\
& & $({\bf 1},{\bf 4},{\bf 1};{\bf 1},{\bf 4},{\bf 1})_T$ & & \\
\hline
\end{tabular}
\caption{The massless spectra of the six dimensional Type IIB orientifolds
on $T^4/{\bf Z}_N$ for $N=2,3,6$.
The semi-colon in the column ``Charged Hypermultiplets'' separates $99$ and
$55$ representations. The subscript ``$U$'' indicates that the
corresponding (``untwisted'') state is perturbative from the orientifold
viewpoint. The subscript ``$T$'' indicates that the
corresponding (``twisted'') state is non-perturbative from the orientifold
viewpoint. The $U(1)$ charges are not shown, and by ``neutral''
hypermultiplets we mean that the corresponding states are not charged
under the non-Abelian subgroups.}
\end{table}
\begin{table}[t]
\begin{tabular}{|c|c|l|c|}
Model & Gauge Group & \phantom{Hy} Charged & Neutral
\\
& &Chiral Multiplets & Chiral Multiplets
\\
\hline
${\bf Z}_6^\prime$ & $[U(4) \otimes U(4) \otimes U(8)]_{99}\otimes$
& $({\bf 1},{\bf 1},{\bf 28};{\bf 1},{\bf 1},{\bf 1})_U$ & $U:~~~4$ \\
& $[U(4) \otimes U(4) \otimes U(8)]_{55}$&
$({\bf 1},{\bf 1},{\overline {\bf 28}};
{\bf 1},{\bf 1},{\bf 1})_U$ & ${\bf Z}_3:~~~18$ \\
& & $({\bf 4},{\bf 4},{\bf 1};{\bf 1},{\bf 1},{\bf 1})_U$ & ${\bf
Z}_2:~~~12$ \\
& & $({\overline {\bf 4}},{\overline {\bf 4}},{\bf 1};
{\bf 1},{\bf 1},{\bf 1})_U$ & ${\bf
Z}_6:~~~12$ \\
& & $({\bf 4},{\bf 1},{\bf 8};
{\bf 1},{\bf 1},{\bf 1})_U$ & \\
& & $({\bf 1},{\overline {\bf 4}},{\overline {\bf 8}};
{\bf 1},{\bf 1},{\bf 1})_U$ & \\
& & $({\overline {\bf 6}},{\bf 1},{\bf 1};
{\bf 1},{\bf 1},{\bf 1})_U$ & \\
& & $({\bf 1},{\bf 6},{\bf 1};
{\bf 1},{\bf 1},{\bf 1})_U$ & \\
& & $({\overline {\bf 4}},{\bf 1},{\bf 8};
{\bf 1},{\bf 1},{\bf 1})_U$ & \\
& & $({\bf 1},{\bf 4},{\overline {\bf 8}};
{\bf 1},{\bf 1},{\bf 1})_U$ & \\
& & $({\bf 4},{\overline {\bf 4}},{\bf 1};
{\bf 1},{\bf 1},{\bf 1})_U$ & \\
& & Same as above with $99\leftrightarrow 55$ & \\
& & $({\bf 4},{\bf 1},{\bf 1};{\bf 4},{\bf 1},{\bf 1})_U$ & \\
& & $({\bf 1},{\bf 4},{\bf 1};{\bf 1},{\bf 1},{\bf 8})_U$ & \\
& & $({\bf 1},{\bf 1},{\bf 8};{\bf 1},{\bf 4},{\bf 1})_U$ & \\
& & $({\overline {\bf 4}},{\bf 1},{\bf 1};{\bf 1},{\bf 1},
{\overline {\bf 8}})_U$ & \\
& & $({\bf 1},{\overline {\bf 4}},{\bf 1};{\bf 1},{\overline {\bf 4}},
{\bf 1})_U$ & \\
& & $({\bf 1},{\bf 1},{\overline {\bf 8}};{\overline {\bf 4}},{\bf 1},
{\bf 1})_U$ & \\
& & $6\times ({\bf 4},{\overline {\bf 4}},{\bf 1};
{\bf 1},{\bf 1},{\bf 1})_T$ & \\
& & $3\times ({\overline {\bf 4}},{\bf 4},{\bf 1};
{\bf 1},{\bf 1},{\bf 1})_T$ & \\
& & $6\times ({\overline {\bf 6}},{\bf 1},{\bf 1};
{\bf 1},{\bf 1},{\bf 1})_T$ & \\
& & $3\times ({\bf 6},{\bf 1},{\bf 1};
{\bf 1},{\bf 1},{\bf 1})_T$ & \\
& & $6\times ({\bf 1},{\bf 6},{\bf 1};
{\bf 1},{\bf 1},{\bf 1})_T$ & \\
& & $3\times ({\bf 1},{\overline {\bf 6}},{\bf 1};
{\bf 1},{\bf 1},{\bf 1})_T$ & \\
& & $3\times ({\bf 1},{\bf 1},{\bf 1};
{\bf 4},{\overline {\bf 4}},{\bf 1})_T$ & \\
& & $3\times ({\bf 1},{\bf 1},{\bf 1};
{\overline {\bf 6}},{\bf 1},{\bf 1})_T$ & \\
& & $3\times ({\bf 1},{\bf 1},{\bf 1};{\bf 1},{\bf 6},{\bf 1})_T$ & \\
& & $3\times ({\overline {\bf 4}},{\bf 1},{\bf 1};{\overline {\bf 4}},
{\bf 1},{\bf 1})_T$ & \\
& & $3\times ({\bf 1},{\bf 4},{\bf 1};{\bf 1},{\bf 4},{\bf 1})_T$ & \\
\hline
\end{tabular}
\caption{The massless spectrum of the four dimensional Type IIB orientifold
on $T^6/{\bf Z}_6^\prime$.
The semi-colon in the column ``Charged Chiral Multiplets'' separates $99$ and
$55$ representations. The subscript ``$U$'' indicates that the
corresponding (``untwisted'') state is perturbative from the orientifold
viewpoint. The subscript ``$T$'' indicates that the
corresponding (``twisted'') state is non-perturbative from the orientifold
viewpoint. The $U(1)$ charges are not shown, and by ``neutral''
chiral multiplets we mean that the corresponding states are not charged
under the non-Abelian subgroups.}
\end{table}
|
\section{Introduction.}
The Matrix Theory \cite{bfss} has supposed an important advance in
uncovering the global properties of M Theory or strongly coupled IIA String
Theory. It has answered several questions as well as offered a number of new
problems to be solved. Most problems are related to the fact that the model
is formulated in the light cone and it is precisely the light cone direction
the one that is taken to define the string coupling. This mixes the degrees
of freedom in such a way that calculations are simplified, but somehow lack
the easy interpretation they had in perturbative string theory. Objects are
not easily recognized and are, in fact, complicated combinations of the
asymptotic states of the Lorentz invariant theory. The study of its
thermodynamics was carried out in \cite{us3,sath,sath2,amb,zh} but it is
difficult to directly relate those calculations to the ones in usual
perturbative string theory.
A partial remedy to this was set with the Matrix String Theory of
\cite{dvv,sfm} that put forward a non-perturbative formulation of String
Theory. The objects are still strings and the perturbative limit is not so
hard to find and analyze. In fact, the difference is in the way interactions
and non-perturbative objects appear. In the usual perturbative string theory
everything lies on the form of the world-sheet: its topological form
accounts for the different interaction terms while its singularities (fixed
points, boundary conditions...) are related to non-perturbative effects. In
the new approach, the world-sheet is always a circle and the dynamics is
defined by the Yang-Mills fields on it. Their interactions tell us about the
strings' interactions and their non-perturbative configurations are related
to D-branes.
Previous works studied the general thermal scenario and the relation
among the `problems of Hagedorn' in the different matrix string theories
\cite{us4} and the perturbative free string limit \cite{gri}, that is also
obtained here from another perspective. The question I have attempted to
address in this article is what does this approach add to thermodynamics
when energies are beyond $g^{-1}$. I have analyzed how should
non-perturbative Yang-Mills configurations be included in the Free Energy
and the result is a more dynamical view of string and D-brane thermodynamics
in the whole energy range.
The calculations support, in the appropriate limit, those done before (in
\cite{many}) and unveil the drastic consequences of including oscillating
membranes. The Hagedorn problem is worsened by the appearance of this even
more degenerate object and the final picture at very high energies seems to
be dominated by a single and very excited membrane at very low temperature.
\section{Review of the matrix strings from the finite temperature point of
view.}
Let us start with the Hamiltonian of the Matrix Model compactified on
$S^1(light-cone)\times S^1$. It is
\begin{equation}
H=\frac{1}{2}\frac{R_+}{ 2 \pi} \int_0^{2 \pi} d \sigma
tr \left\{ p_i^2+(DX_i)^2+ \theta^T D \theta+\frac{1}{R_9^2}\left( E^2 -\left[
X^i,X^j \right]^2 \right) + \frac{1}{R_9}\theta^T
\gamma_i \left[ X^i,\theta \right] \right\}
\label{hamil}
\end{equation}
It is in $\alpha'(=l_p^3 R_9^{-1})$ units. Taking the limit of type IIA
Matrix String \cite{dvv} means equating $R_9=\sqrt{\alpha'} g_s$ and
assuming that $g_s$ is small. In that limit, configurations with
non-commuting matrices are very heavy and we can consider them decoupled or
include them as solitonic objects.
The Hamiltonian of the perturbative objects is a free $U(1)^M$ gauge
theory that describes exactly the same fields as the Green-Schwarz
Hamiltonian plus a two-dimensional gauge field. This one is pure gauge and
can be taken to zero but for the solitonic configurations related to the
finite circle in which the theory is defined. We shall take those
configurations into account later.
We shall now relate every quantum number of the $U(1)^M$ SYM to its
counterpart in String Theory. The light-cone variables are defined as
\begin{eqnarray}
p^+=p_0+p_{11} \nonumber \\
p^-=p_0-p_{11}
\label{cone}
\end{eqnarray}
Let us remember that the light-cone Hamiltonian of strings is
\begin{equation}
H=\frac{1}{2} (p^++p^-)
\label{ham1}
\end{equation}
with
\begin{equation}
p^-=\frac{p_i^2}{2 p^+}+\frac{1}{2 \alpha' p^+}\sum_n \left( \alpha^i_{-n}
\alpha^i_{n}+ \tilde{\alpha}^i_{-n} \tilde{\alpha}^i_{n} \right)
\label{ham2}
\end{equation}
plus the fermionic part and with
\begin{equation}
\sum_n \alpha^i_{-n} \alpha^i_{n}=N_{n_i} n_i
\end{equation}
Here $p^+=\frac{M}{R}$ is discrete. The $n$ are the oscillation numbers.
Note that the second quantization of the string fields is already included
and is, of course, independent of the number $N$. This is the usual
treatment of multi-field systems in perturbative String Theory with a single
world-sheet. In the SYM, the discrimination in different strings is related
to the `twisted states'. One can consider diagonal matrices in which the
elementary fields are not periodic, but, rather, have the following
property:
\begin{eqnarray}
X_{i,i}(\sigma)&=&X_{i+1,i+1}(\sigma+2\pi) \nonumber \\
&\mbox{ ... }& \\
X_{i+k,i+k}(\sigma)&=&X_{i,i}(\sigma+2\pi) \nonumber
\end{eqnarray}
for a $k$-times-twisted state. This is allowed by the $Z^M$ symmetry that is
left even after taking the $g_s \rightarrow 0$ limit. The result is just the
same as if we considered that the world-sheet could have different lengths,
all multiples of the initial one. This is tantamount to rescaling the
prefactor of the Hamiltonian, that is the light-like momentum $p^+$. That is
why these twisted states represent the Kaluza-Klein momentum number of each
object in the eleventh dimension. They are equivalent in the $0+1$ SYM to
the bound states of the original D0-branes. The total Kaluza-Klein momentum
is therefore $M=\sum_k k M_k$, if $M_k$ is the number of $k$-times-twisted
strings. The sum in $M$, and particularly in each $M_k$ is the real second
quantization of strings because it allows to deal with several string
world-sheets at the same time. To find out the spectrum, each scalar field
has to be expanded in a Fourier series. The Hamiltonian of each term is a
harmonic oscillator with frequency
$\omega=2\pi n$ so their energy is
\begin{equation}
H=\frac{M}{2R_+} + \frac{1}{2} \frac{R_+}{M \alpha'}\left[ \sum_n n
\left( N_n + \frac{1}{2}\right)+\sum_{\tilde{n}} \tilde{n}
\left( \tilde{N}_{\tilde{n}} + \frac{1}{2}\right) \right] +
\frac{R_+ p_i^2}{2 M}
\end{equation}
It exactly coincides with (\ref{ham1}) and (\ref{ham2}). We have focused
on the bosonic part. The fermionic one is a fermionic harmonic oscillator
that cancels the zero point energy of the bosonic one. In the original model
of D0-branes moving in a circle, these oscillator modes are mapped by the
flip and the Taylor construction \footnote{I do not explicitly call T-duality
to this construction because in this case it is not. From this M-theory point
of view, $R_9$ and $R_+$ are just parameters and T- and S-dualities are just
changes of names and variables.} \cite{taylor} to the modes of the open strings in which the
second end is stuck in the D-particle after having wound several times around
the circle. $N$ is the total light-like momentum of the system.
We shall now derive the known partition function of free strings from
this discrete light-cone Hamiltonian. It is
\begin{equation}
Z=\exp \left[ -\beta {\cal F}(\beta) \right]
\end{equation}
where ${\cal F }$ is the free energy that can be calculated with
\begin{equation}
-\beta {\cal F}(\beta)=Z_1=\sum_{\mbox{\tiny osc}}\sum_{M} \int d^8
\vec{p} \exp \left\{ -\frac{\beta}{2} \left[ \frac{M}{R_+} + \frac{R_+}{M}
\left( \vec{p_T}^2 + m^2(n,N_n,{\tilde{n}},{\tilde{N}}_{\tilde{n}}) \right) \right]
\right\}
\end{equation}
if
\begin{equation}
m^2(n,N_n,\tilde{n},\tilde{N}_{\tilde{n}})= N+\tilde{N}
\end{equation}
That is the total oscillator number. The integral over the continuous
momenta is Gaussian and can be performed. Besides, the sum over the
oscillator states can be decomposed in mass levels. This yields
\begin{equation}
-\beta {\cal F}(\beta)=\sum_{N} \sum_{M} a_N \left( \frac{\beta R_+}{M}
\right)^{4}
\exp \left[ -\frac{\beta}{2} \left( \frac{M}{R_+} + \frac{R_+}{M} N+R_+ \omega
\right) \right].
\label{lc}
\end{equation}
The coefficients $a_N$ can be related with arguments of number theory to
\begin{equation}
a_N=\frac{1}{N!} \frac{d^N}{d x^N}\left[ \theta_4 (0,x)^{-8} \right]_{x=0}
\label{coeff}
\end{equation}
This is the usual relation between thermodynamics and the string
non-thermal partition function. This is usually obtained after the
integration of the constraint that relates the left and right moving
oscillators. Here, it is got as a count of the number of ways a sum
\begin{equation}
\sum n N_n+\tilde{n} \tilde{N}_{\tilde{n}}
\end{equation}
gives the same total $N$. The result is the same, no matter how different
they seem because of the properties of the Jacobi Theta functions. The
constraint for the discrete light-cone states has been implemented as
\begin{equation}
N=\tilde{N}+\omega M
\end{equation}
Expression (\ref{lc}) is the best we can get when
$R_+$ is finite. If we take the limit $R_+ \rightarrow \infty$, we can
obtain the usual form of the free energy. We have to substitute
\begin{eqnarray}
\frac{M}{R_+} & \longrightarrow & p_+ \nonumber \\
\sum_{M} & \longrightarrow & R_+ \int d p_+
\end{eqnarray}
The states with $\omega\neq 0$
decouple when taking the infinite-$M$ limit. Also, in order to obtain a
proper time representation, we need to change variables
\begin{equation} p_+ \rightarrow \frac{N \beta}{t}
\end{equation}
so that
\begin{equation}
-\beta {\cal F}(\beta)=R_+
\beta^{-4}\sum_{N} \left( \frac{1}{N\beta} \right)^{-5} a_N \int \frac{dt}{t}
t^{-5} e^{-\frac{\beta^2 N}{4 t}} e^{-t}
\end{equation}
In the right-hand side we can recognize the integral representation of the
Bessel function $K$, that we use to finally obtain
\begin{equation}
-\beta {\cal F}(\beta)=R_+ \beta^{-4} \sum_{N}
\left( \frac{2}{N\beta} \right)^{-5} a_N K_5(\beta \sqrt{2 N})
\end{equation}
This has been obtained using classical statistics. Quantum statistics can
be added now as a series in $r$ substituting $\beta$ with $\beta r$.
Supersymmetry reduces the series to the odd terms:
\begin{equation}
-\beta {\cal F}(\beta)=R_+ \beta^{-4} \sum_{N} \sum_{r,odd}
\frac{1}{r^{10}} \left( \frac{2}{N\beta} \right)^{-5} a_N K_5(\beta r \sqrt{2N})
\label{closed}
\end{equation}
This is exactly what results following the $S$-representation or analog
model, that consists in separately summing the contributions of each of the
fields that appear at each mass level of the string. From this, using the
integral representation of the $K$ function and the coefficients
(\ref{coeff}), it is direct to obtain the usual modular invariant form. It
was got in a different way in \cite{gri}.
\section{The electric flux as a non-perturbative correction.}
The last quantum number that we have got left to interpret is the
Kaluza-Klein momentum of the original D-particles related to the ninth
direction. It is mapped, in the matrix string limit, to the Ramond-Ramond
charge of the matrix D-particles \cite{dvv,bfm} (not the original, but the new
ones). These degrees of freedom, that were perturbative in the original matrix
theory, are now solitonic. The reason is that in every reduction of a YM action
to any number of dimensions, the number of degrees of freedom is $D-2$ except
in the reduction to $0+1$ (see \cite{us3}). In that case one more coordinate
appears in the free theory, which is essential to the interpretation of the
model as a theory in eleven dimensions. With the transformation to a $1+1$
theory that has been performed, that degree of freedom has vanished and only
left as a reminder a numerable set of non-equivalent topological sectors. Each
sector is characterized by the expectation value of momentum related to the
field $A_9$. The flux must be \begin{equation} \Phi=\frac{1}{2\pi}\int d \sigma tr E=n,
\label{charge} \end{equation} where $n$ is the Ramond-Ramond charge. That is the same as
to say $p_9=\frac{n}{R_9}$.
Like in the previous section, I shall now give an interpretation to
every quantum number in the picture of String Theory. Remember that for a
general background that included both D0-branes and fundamental strings one
should separate the whole matrix into diagonal boxes, each containing one
twisted state. If the electric field related to the $U(1)^{M_i}$ diagonal
subgroup of a given box takes an expectation value, then that box describes a
D0-brane; else, it is a fundamental string.
The term of
the Hamiltonian that contains the electric field, is, in string units:
\begin{equation}
H_E=\frac{R_+}{4 \pi} \int d\sigma \frac{E^2}{g_s^2}.
\end{equation}
If we consider simple, diagonal forms for $E$, the energy of a D-particle
state at rest is
\begin{equation}
H_E=\frac{R_+}{2} \frac{n^2}{g_s^2}.
\end{equation}
As before, twisted states of these configurations represent particles
with different light-cone momenta. Their energy is
\begin{equation}
H_E=\frac{R_+}{2k}\frac{n^2}{g_s^2}=\frac{M_{D0}^2}{2p^+}.
\end{equation}
The $k$ is in the denominator because -remember- $E$ is not a field but a
momentum. Both numbers, $n$ and $k$, are completely independent. One can
have, for example $k \neq 1$ for $n=1$. That is got with a $k$-times-twisted
state with, for instance, $\Phi=k^{-1}$ for each matrix element. This is
possible because what must have an integer flux number is the $trace$ of the
field, not necessarily each matrix component.
Each twisted state describes a D-particle. Bound states of D-particles in
String Theory cannot be seen like that, not even like twisted states in
this framework. Here, they are simply eigenstates of the ninth component of
the momentum with different eigenvalues or, in the dual system, field
configurations with different topological numbers (different $n$).
The coordinate fields can also in this case be expanded in a Fourier
series. However, there is a subtlety here that has to be noticed. It is related
to the equations of motion of the scalar fields in the presence of the electric
field. To see it clearest, let me use the YM theory in the open space, whose
properties are more familiar. One equation is:
\begin{equation}
\partial_\mu F^{\mu \nu}=0
\end{equation}
which for the electric component and after a Fourier transform means that
\begin{equation}
\vec{p} \vec{E}=0.
\end{equation}
This is the well known condition of transversality of the electromagnetic
waves. After the dimensional reduction, the only change is that both the
momentum and the electric field have only components in the remaining
directions. In our case, we only have one dimension and it is compactified.
The formula above implies that the electric field and the Kaluza-Klein momentum
exclude each other. In the previous section we studied the YM theory with
momentum (oscillator number) but without electric field. Now we are considering
the electric field so the theory is reduced to its zero Kaluza-Klein mode.
This is the reason why D0-branes are pointlike even when they form twisted
states and the size of the matrix that contains them ($M$ in my notation) may be
taken to infinity. Another explanation to the absence of oscillators
(world-sheet KK momentum) can be obtained going back through the construction
of W. Taylor
to the original Matrix Model. Then the electric field is transformed into
the momentum along the ninth direction and the KK modes (oscillators) are the
open strings
that stretch between the D-particles that form the one-dimensional lattice
that describes one D0-brane in a circle. The open strings are represented by
the massive gauge fields that appear when the symmetry is broken when the
positions of the D-particles acquire vacuum expectation value. Now, if we
consider the D-particle moving with a precise momentum, its conjugate
coordinate is completely unmeasurable and, therefore, it cannot have any
expectation value. This prevents the breaking of the gauge symmetry and the
appearance of any massive gauge field.
This explains why D0-branes are pointlike in this approach. One important
consequence is that, for twisted states to be formed, the D-particles must
coincide at one point, unlike what happens with the closed strings. This is
not exactly the explanation given by \cite{dvv}. Their interpretation only
included D-particles with $M (p^+)= n (\mbox{D0 number})$, which is a strange relation between the
Dirichlet charge and the light-cone momentum. The string-like excitations
were eliminated from the spectrum just because their light-cone momentum was
set to zero in the large-$N$ limit.
Before that, let us see how open strings can be attached to the D0-brane
in the perturbative limit. The mechanism, as pointed out in \cite{dvv} is
the same that causes interaction between different strings. When, at some
value of $\sigma$ the eigenvalues of all the spatial coordinates of the
D-particle coincide with those of a close string, some off-diagonal fields
get locally {\it nearly} massless. They are responsible for the attachment.
However, it seems to me that this case of D0-branes is a little bit subtler
than the strings' one. Firstly, whenever a closed string coincides at one
point with the position of the D-particle, there is not any exact
enhancement of symmetry. The reason is that the expectation value of $A_1$
is not proportional to the identity. The system is described by a matrix that
consists of two boxes. One is the D-particle and has an electric field. The other
is the matrix string. For simplicity, let us assume that both boxes have the
same size (both objects have the same light-cone momentum). In the D0-brane box
\begin{equation}
\partial_0 A_1\propto E =\mbox{constant}
\end{equation}
while in the string box the field can be set to $0$. Therefore
\begin{equation}
A_1 \propto n t (\sigma_3+Id)
\end{equation}
In the usual matrix string interaction, all fields coincide at one point and
there is an enhancement of symmetry. Here, the presence of the $\sigma_3$ field
slightly breaks that symmetry because off-diagonal fields do not become
completely massless. Let us compute its mass.
The interesting term of the Hamiltonian is
\begin{equation}
R_9[A^9,X^i]^2
\end{equation}
We change
\begin{equation}
(A^9, X^i) \rightarrow R_9^{-1/2} (A^9, X^i)
\end{equation}
for the fields to have appropriate dimensions and, taking into account that
\begin{equation}
<(A^9)^2>\propto R_9^2= g^2 \alpha'
\end{equation}
the corresponding mass of the mediating field is
\begin{equation}
m^2 \propto g
\end{equation}
We are supposing that $g$ is very small, so the interchange of these very
light fields is permitted. To deduce from here that the coupling constant of
this interaction between the D-particle and the string is $g$, it should be
necessary to generalize the calculation in \cite{bonelli}. Another feature
that makes this attachment different from perturbative string interactions
is that it has more drastic consequences. The string loses half the
supersymmetry due to the existence of a privileged point that breaks the
invariance under reparameterizations. The point is distinguished in the
final world-sheet that combines both objects (and goes from $0$ to $4 \pi$)
as a discontinuity in the electric field, that is,
\begin{eqnarray}
E(\sigma)=E \hspace{1cm} \mbox{if} \hspace{1cm} \sigma \in [0,2 \pi] \nonumber \\
E(\sigma)=0 \hspace{1cm} \mbox{if} \hspace{1cm} \sigma \in [2 \pi,4 \pi]
\label{conf}
\end{eqnarray}
The reason for that is that the twisted state crosses both boxes: the one
that represents the D-particle and that of the fundamental string.
We have written before that in a the world-sheet of a single object the
electric field must be constant because any other configuration is
energetically very costly. The reason why (\ref{conf}) is possible is that
the process by which the electric field of some $U(1)$ extends itself to the
total group that includes both the string and the D-particle is clearly
non-perturbative. However, it is not completely forbidden: it is just the
absorption of a string and its light-cone momentum ($p_+$) by a D-particle
when both interact.
In fact, the way the light-cone momentum is shared between the two
objects can be calculated. We have showed how the string is tied to the
D0-brane in all the transverse directions, but nothing has been said about
the longitudinal one. The difficulty is that the light-cone description we
have obliged the D-particle to be in the momentum representation. This
prevents the explicit appearance of a Dirichlet boundary condition. The only
reason for the string to be attached also in the ninth direction is
dynamical. We have to work with the covariant Hamiltonian one gets after
undoing the change of variables that took us to the light-cone. The free
part is
\begin{equation}
H=\sqrt{M^2+\vec{p}^2_T+ (p^9_{\mbox{total}}-p^9_{\mbox{o.s.}})^2}+
\sqrt{m^2_{\mbox{o.s.}}+(p^9_{\mbox{o.s.}})^2}
\end{equation}
where the subscript `o.s.' stands for open string, $M$ is the mass of the
D-particle and $m$ is the mass of the string. The minimum of the energy
occurs when
\begin{equation}
\frac{p_{\mbox{o.s.}}}{p_{\mbox d0}}=\frac{m}{M}
\end{equation}
This condition implies that the speeds of both objects coincide and
therefore, the dynamics of the open string is tied to that of the
D-particle. The link, however, is not very strong because the string can go
away with just the cost of its kinetic energy
\begin{equation}
\frac{(p^9_{\mbox{o.s.}})^2}{2m}.
\end{equation}
This is due to the fact that when an open string is attached to a
D0-brane, its two ends are in the same point and so the string can close and
be emitted at no cost.
It would be interesting to add these fields into the partition function
of (\ref{closed}). Once we throw away the part of the interaction and take
the $N\rightarrow \infty$ limit, the light-cone Hamiltonian is
\begin{equation}
H_{l.c.}=\frac{p_+^{\mbox{d0}}}{2}+\frac{p_+^{\mbox{o.s.}}}{2}+
\frac{1}{2p_+^{\mbox{d0}}} \left[ M^2+ (\vec{p}^{\mbox{d0}}_T)^2 \right] +
\frac{1}{2p_+^{\mbox{o.s.}}} \left[ M^2+ (\vec{p}^{\mbox{o.s.}}_T)^2 \right]
\end{equation}
with the constraint
\begin{equation}
\vec{p}_T^{\mbox{d0}}=\frac{M}{m} \vec{p}^{\mbox{o.s.}}_T \hspace{1cm} \hspace{1cm}
p_+^{\mbox{d0}}=\frac{M}{m} p_+^{\mbox{o.s.}}
\end{equation}
It is straightforward to follow the calculation in the previous section
and to obtain the following proper-time representation of the Free Energy:
\begin{equation}
-\beta F(\beta)=\sum_{\{ m, M \} } \beta \int \frac{dt}{t} t^{-5} \exp\left[
-\frac{t}{2} \left( M+m \right)^2 - \frac{1}{2t} \beta^2 \right]
\end{equation}
We can now substitute the mass spectra
\begin{eqnarray}
m^2=N \nonumber \\
M^2=\frac{n^2}{g^2}
\end{eqnarray}
where $N$ is the oscillator number of the attached string. This gives
\begin{equation}
-\beta F(\beta)=\sum_{\{n, N \} } b_N \beta \int \frac{dt}{t} t^{-5}
\exp\left[ - \frac{t}{2} \left( \frac{|n|}{g}+ \sqrt{N} \right)^2 - \frac{1}{2t}
\beta^2 \right]
\label{total}
\end{equation}
$b_N$ is the degeneracy associated to each mass level. This is the result
using classical statistics. The complete result has one more sum that
completes the exponential with $\beta$ to the usual Jacobi $\theta_3$ and
$\theta_2$ functions for bosons and fermions respectively.
This does not look very familiar. If we want to see the open string's
Free Energy to emerge from this mess, we have to use the Born-Oppenheimer
approximation. It consists in classifying the degrees of freedom in fast and
slow ones. This approach was used in \cite{us3, amb} to study the dynamics
of D0-branes. This case is different because the light-cone direction is not
the one defined by $g$. This means that we have positive as well as negative
Ramond-Ramond charges and the excitations of them include the whole open
string spectrum, not just the massless sector. Anyway, the approximation is
valid as long as the mass of the D-particles is large. Here, the fast modes
are the open strings and the slow ones are the movements of the D-particles.
To see the open string partition function, we have to freeze the movements
of the D0-branes. We cannot do it with the proper time representation so we
have to go back to the Hamiltonian and set, there, all the momenta to zero
(or to a fixed value). The Hamiltonian is then
\begin{equation}
H=\frac{|n|}{g}+ \sqrt{N}
\end{equation}
Index $n$ is fixed, we do not have to sum over it because we do not want
to allow the creation and annihilation of D-branes in our gas. To simplify,
consider just one D-brane. The Free Energy is simply
\begin{equation}
-\beta F_n(\beta)=\sum_{r,odd} \sum_{N=0}^\infty b_N
e^{-\frac{|n|}{g}\beta -\sqrt{N} \beta r }
\end{equation}
There is not an $r$ multiplying the D-particle mass because that part is,
in fact, not thermalized at all; it plays the role of a cosmological
constant, it is just the energy of the vacuum.
This representation of the Free Energy is not very recognizable so it is
better to put it in an integral form. The steps to follow are
\begin{eqnarray}
-\beta F_n(\beta)=e^{-\frac{|n|}{g}\beta} \sum \exp(-\sqrt{N} \beta r)=
e^{-\frac{|n|}{g}\beta} \sum b_N \sqrt{\frac{2}{\pi}} (\beta \sqrt{N}r)^{1/2}
K_{1/2}(\beta \sqrt{N}r)= \nonumber \\
= e^{-\frac{|n|}{g}\beta} \sum b_N \frac{1}{\sqrt{\pi}} \beta \int_0^\infty
\frac{d t}{t} t^{-1/2} \exp (- \frac{\beta^2r^2}{2 t} - N t) = \nonumber \\
= e^{-\frac{|n|}{g}\beta} \frac{1}{\sqrt{\pi}} \beta \int_0^\infty
\frac{d t}{t} t^{-1/2}
\theta_2(0,\frac{\beta^2}{2 \pi t}) \frac{\theta_2^4(0,i t)}{\eta^{12}(i t)}
\end{eqnarray}
The last expression is exactly the Free Energy of the open string theory
with Dirichlet boundary conditions in all directions, as obtained in the
S-representation. I have substituted the Laplace series
\begin{equation}
\sum_N b_N e^{-N t}
\end{equation}
with the appropriate Theta function. It is different from the one used for
closed strings because here the central charge breaks the supersymmetry and
the contribution of the fermions is changed: the supermultiplets are
shorter. I do not spend time with this because the procedure is known and
identical to that in the perturbative string theory.
In this approach, the partition function of the slow modes represents a
count of the `possible vacuums' of the system. This is the way the existence
and positions of D-branes are seen in perturbative string theory.
Once we have seen that expression (\ref{total}) reproduces in the
appropriate limit the Free Energy of the attached strings, let us study the
structure when all the terms are taken into account. I am specially
interested in the modular properties.
The first thing we have to do is to get rid of the square roots. We get
it in two steps; expanding the root and taking advantage of this trick:
\begin{eqnarray}
\exp\left( -t \sqrt{N} \frac{|n|}{g} \right) =
\sqrt{\frac{2}{\pi}} \left( -t \sqrt{N} \frac{|n|}{g}
\right)^\frac{1}{2} K_{1/2}\left( t \sqrt{N} \frac{|n|}{g}
\right)= \nonumber \\
= \frac{t^{1/2}}{\sqrt{\pi}} \int \frac{d s}{s} s^{-1/2} \exp \left[
-\frac{1}{4s} \left( t N \frac{n^2}{g^2} \right) - s t \right]
\end{eqnarray}
The cost is that we have added another integral. Altogether, the Free
Energy is
\begin{eqnarray}
-\beta F(\beta)=\frac{\beta}{\pi} \sum \int \frac{d s}{s} \frac{d t}{t}
s^{-1/2} t^{-11/2} b_N \times \nonumber \\ \times
\exp \left[ -\frac{1}{2t} \beta^2 (2m+1)^2- \frac{t}{2}\left(
\frac{n^2}{g^2} + N+ \frac{1}{2s}+ 2 s N \frac{n^2}{g^2}\right) \right]
\end{eqnarray}
We would like to take this mess and put some order, but, as far as I can
see, it is only partially possible. I have found two alternative ways to do
so:
\begin{eqnarray}
-\beta F(\beta)=\frac{\beta}{\pi} \sum_{n\neq 0} \int \frac{d s}{s} \frac{d t}{t}
s^{-1/2} t^{-11/2} \theta_2\left( 0,i \frac{\beta^2}{\pi t}\right)
\times \nonumber \\ \times f \left[ \left( 1+ 2s \frac{n^2}{g^2} \right) \frac{t}{2}
\right] \exp \left( -\frac{t}{2} \frac{n^2}{g^2} -\frac{t}{4s} \right)
\end{eqnarray}
and
\begin{eqnarray}
-\beta F(\beta)=\frac{\beta}{\pi} \sum \int \frac{d s}{s}
\frac{d t}{t} s^{-1/2} t^{-11/2} b_N \theta_4\left( 0,i
\frac{\beta^2}{\pi t}\right)
\times \nonumber \\ \times
\left[ \theta_3\left( 0,\frac{i t}{2 g^2} +\frac{i t s N}{g^2} \right) -1 \right]
\exp \left( - \frac{t}{2} N -\frac{t}{4s} \right)
\end{eqnarray}
To simplify, I have defined $f$ to be the open string partition function
\begin{equation}
f(t)=\frac{\theta_2^4(0,i t)}{\eta^{12}(i t)}
\end{equation}
The introduction of D-branes as rigid non-perturbative corrections to
string thermodynamics has always yielded the same result: the critical
Hagedorn temperature does not change, although the properties of the
transition and the even its existence do depend on the background. The
reason for this stability is that, in fact and as far as D-branes are solely
static objects, the critical temperature only depends on the properties of
the world-sheet field theory. It is, therefore, an interesting topic to
decide if the addition of the dynamics of D0-brane changes anything. To know
that from the Canonical Ensemble, we have to calculate the behaviour of the
integrand of (\ref{total}) when $t \rightarrow 0$. To study the exponent
\begin{equation}
\sum_{N,n} \exp \left( -\frac{t}{2}\frac{n^2}{g^2} - \frac{t}{2} N +
\sqrt{8 \pi^2 N} -t \sqrt{N} \left| \frac{n}{g} \right| \right)
\end{equation}
is enough. The coefficients $b_N$ have been substituted with their
asymptotic expression:
\begin{equation}
b_N \sim N^{-9/4} e^{\sqrt{8} \pi \sqrt{N}}
\end{equation}
It is possible to evaluate the sum in $N$ with a saddle point approximation.
The result of that is
\begin{eqnarray}
\sum_{n} \exp \left( -\frac{t}{2}\frac{n^2}{g^2} + \frac{t}{2}\frac{n^2}{g^2}
-\sqrt{8 \pi^2}\left| \frac{n}{g} \right| +\frac{4 \pi^2}{4 t} \right) =
\sum_{n} \exp \left( -\sqrt{8 \pi^2}\left| \frac{n}{g} \right| +
\frac{4 \pi^2 }{t} \right)
\end{eqnarray}
The only dependence on $t$ is the last fraction that diverges at $t=0$.
The critical temperature is found when that term is compared to the first
mode of the thermal $\theta_4$ function:
\begin{equation}
\frac{4 \pi^2 }{t}=\frac{\beta_c^2}{2 t} \Rightarrow \beta_c=\sqrt{8} \pi
\end{equation}
This is the usual Hagedorn temperature both for (type II) closed and open
strings so the conclusion is that the dynamics of the D-particles with the
excitations we have considered do not change it. Remember that the spectrum
we have used is the free one, but that does not mean it is the complete one
even when $g=0$. The open strings that stretch between a couple of
D-particles have been considered in what precedes as part of the interaction
even when their contribution to the energy does not vanish at zero coupling.
The origin of this confusion is the non-commutativity of the D-brane
coordinate fields. When the off-diagonal terms are taken into account, the
separation into individual objects gets darker and the inclusion of their
contribution is a hard task. Maybe this states that we have not considered
are responsible for the apparent lack of modular invariance of the
expression obtained.
In a very interesting work B. Sathiapalan \cite{sath} showed an aspect of
high temperature matrix theory that is related to this work. He used the $D=1+1$
field theory to deduce a `deconfinement-like' transition that he related to
Hagedorn. The energy of such transition would be non-perturbative so one may
question why the results and the methods of this section (and the following
ones) are so different. The answer is that the physical situations described are
different. The parameter that distinguishes both is the relative distance of the
objects or, if you want, the density of the gas. That separates two regions in
phase space according to whether the typical distance between objects is bigger
or smaller than $\alpha'$. If they are bigger, one must consider first the addition
of BPS non-perturbative objects like D-particles and D-membranes because their
energies are smaller than those of off-diagonal fields. This is what could be
called a dilute gas approximation and that is what is being carried out in this
work. If the density of the gas is such that typical distances are smaller than
the string length, then one should consider first the contribution of
off-diagonal modes. That is more difficult and up to now, only qualitative
information could be extracted in \cite{sath}. D-objects carry
non-perturbative information but their geometrical nature makes them more
tractable.
\section{Membranes as D2-branes.}
The flux that defines a membrane lying along the seventh and the eighth
directions is
\begin{equation}
\Phi=\frac{i}{4 \pi}\int d\sigma tr < \left[ X^7,X^8 \right] >=T_2 A .
\end{equation}
where $A$ is the area of the membrane. It is necessary for that trace not to
be null that $N$ (the size of the matrices) should be infinite. If some
fields with that commutation relation take some expectation values, a
membrane is `created' from the vacuum. It is important not to confuse these
non-commuting fields with the centre-of-mass coordinates of the membrane
whose dynamics is described by the $N\times N$ identities that can be added
without changing the commutation relation. That is, the fields that give the
D2-brane charge are in the $SU(N)$ part. The expectation values of the
non-commuting matrix fields are related to the extension of the membrane in
each direction. Notice that neither the centre of mass nor the lengths are
affected by any uncertainty relation but the usual quantum one. Both
non-commuting fields can independently have any expectation values.
As an example, we can choose the representation of the fields to be
equivalent to the position and momentum operators defined over a rectangle.
The correct normalization is such that their eigenvalues lie in the segment
$(3/2)^{1/3}[-1,1]$. In the usual quantum theory, the eigenvalues of the
coordinates and the momenta cannot be both continuous and bounded at the
same time; however, the reason has to do with the boundary conditions one
imposes to the eigenfunctions. There are not any boundary conditions in this
case, so that does not happen here. To fix notation let me write
\begin{eqnarray}
\hat{X}^7=X^7 x \nonumber \\
\hat{X}^8=X^8 y \\
\left[ x,y \right]=1. \nonumber
\label{mat}
\end{eqnarray}
Here $x$ and $y$ are operators (as matrices) that act over the Hilbert space
and $X^7$ and $X^8$ are numbers. However, as I have already written, they
are internal degrees of freedom and, as such, they are completely
independent of the position of the centre of mass. All excitations over this
background should be interpreted as the excitations of the membranes
(D2-branes in the perturbative limit).
It is now our task to study some of the possible excitations of these
physical configurations. We shall begin with the quantization of precisely
the non-commuting fields so that we know what are their possible expectation
values, that is, the possible areas for the membrane. This is so because the
eigenvalues of the operators are the points in the target space that are
occupied by the membrane. The Hamiltonian that concerns them is
\begin{equation}
H_{78}=p_7^2+p_8^2+\frac{1}{g_s^2} X_7^2 X_8^2,
\label{longitudinal}
\end{equation}
where I have taken $\alpha'=1$ and I have been loose about constants. A first
approximation to the energy of the fundamental state is to consider that
$p x \simeq \hbar$ and minimize the Hamiltonian. The value of $X^7$ and
$X^8$ that are obtained are:
\begin{equation}
X^7 \simeq X^8 \simeq g^{1/3} = l_p.
\end{equation}
This gives the very interesting result that the `smallest' possible
membrane is of the size of the eleven-dimensional Planck area. This also
increases the importance of the non-perturbative effects in string theory.
It is known that the energy density of D-objects is of order $g^{-1}$ in
string units; however, the result above means that the first membrane state
that is excited appears at an energy of order $g^{-1/3}$.
These excitations that I have already discussed are longitudinal along
the directions of the membrane. The transversal ones can be described by
fields like
\begin{equation}
X^6=f(x,y).
\end{equation}
If we take one, for instance, to be proportional to $y$, its Hamiltonian,
with the correct normalizations would be
\begin{equation}
H_6= \frac{1}{2 p^+} \left( p_6^2 + \frac{L_7^2}{g_s^2} X_6^2 \right).
\end{equation}
The coordinate is no longer free but rather, it oscillates in a harmonic
oscillator with frequency
\begin{equation}
\omega=\frac{L_7}{g_s}.
\end{equation}
That is precisely the one-dimensional tension of the membrane in the eighth
direction, that is, integrating out the other dimension. Explicitly:
\begin{equation}
T_1^{(8)}=\int_0^{L^7} T_2=\int_0^{L^7} \frac{1}{g_s}=\frac{L_7}{g_s},
\end{equation}
in $\alpha'$ units. We can now interpret these excitations as waves moving in
the eighth direction whose amplitude vibrates in the sixth. It is
straightforward to consider other excitations oscillating in other
directions and also moving in the seventh. The interpretation as waves is
quite clear because what the fluctuations actually do is to locally rotate
the membrane towards some direction and then leave it back where it was
lying.
In fact, we can make a Fourier expansion of the fields moving in each
direction. We must impose appropriately chosen boundary conditions. If we
choose them to be Neumann, the expansion is made in terms of these fields:
\begin{equation}
\Psi_n= \sqrt{2n} \sin \left( \pi n x \right)
\label{transversal}
\end{equation}
where $x$ is again the $N\times N$ matrix defined in (\ref{mat}). These
fields parameterize the waves moving in the seventh direction. Others do it
in the eighth and there are combinations that complete the Fourier
two-dimensional expansion of the membrane's vibrating modes. The mass these
fields acquire through the interaction with the background is
\begin{equation}
E_n^2=N_n n \frac{L^7}{g_s}.
\end{equation}
where $N_n$ is the excitation mode of the harmonic oscillator. Notice that,
in eleven-dimensional units,
\begin{equation}
\frac{L^7}{g_s}=\frac{L^7}{l_p^3}
\end{equation}
So that if we compactify the seventh direction and consider it the quantum
one (make a flip), the vibrations of the membrane in the other (eighth)
direction have the usual string spectrum proportional to the new $\alpha'(=l_p^3
(L^7)^{-1})$. Just like these, all possible vibrations of the membrane can
be constructed. These fields represent the second quantization of the
classical waves that deform the surface of the membrane. All of them are
excited only at non-perturbative energies so that we can conclude that in
the perturbative string limit the D-branes are completely rigid.
Now, I shall explain how perturbative open strings can be formed. Let us
recall what gauge symmetry we have left after the non-commutative fields
have acquired their expectation values. As I have written before, the
matrices can be chosen to be coordinate and momentum (derivative) operators
acting over a one-dimensional space. This gives the correct commutation
relation and their eigenvalues give the eigenvalues of our matrices. There
are many $SU(\infty)$ matrices that obey that algebra. The different
possibilities are related to the different representations (coordinate,
momentum or mixed) of the basis that is used to span the Hilbert space.
Changing the representation is equivalent to changing the gauge. That is the
freedom we keep. This is related to the invariance under world-volume
reparameterizations inside the membrane. Notice that there is no statistical
$Z_N$ symmetry involved, which is coherent with the fact that we only have
one object.
Let us see some cases in which this symmetry is enhanced. Consider two
membranes that cross at one line. The case is identical if there is only one
membrane that crosses itself. One world-volume representation
would be
\begin{eqnarray}
X^6_a(\sigma^1_a,\sigma^2_a)=\sigma^1_a \nonumber \\
X^6_b(\sigma^1_b,\sigma^2_b)=0 \nonumber \\
X^7_a(\sigma^1_a,\sigma^2_a)=0 \nonumber \\
X^7_b(\sigma^1_b,\sigma^2_b)=\sigma^1_b \\
X^8_a(\sigma^1_a,\sigma^2_a)=\sigma^2_a \nonumber \\
X^8_b(\sigma^1_b,\sigma^2_b)=\sigma^2_b \nonumber
\end{eqnarray}
where $a$ and $b$ index the membranes and $1$ and $2$ the directions. The
parameters $\sigma_{a,b}$ run inside $[-1,1]$, for example. The crossing
occurs at $\sigma^1_a=\sigma^1_b=0$. In the matrix case, there are two
appropriate representations of $\hat{X}^8$ as a momentum of the other
coordinates at that line. They are related to the different ways how the
momentum (derivative) can flow across the membrane. Figure \ref{mems}
explains it. Mathematically, one can define $X^8$ as a derivative (at
$\sigma^1=0$) in all these ways:
\begin{eqnarray}
\left. \partial_a f \right|_{\sigma=0}=\lim_{\Delta \sigma\rightarrow 0}
\frac{f(\sigma_a+\Delta \sigma_{a/b})-f(\sigma_a)}{\Delta\sigma_{a/b}}
\nonumber \\
\left. \partial_b f \right|_{\sigma=0}=\lim_{\Delta \sigma\rightarrow 0}
\frac{f(\sigma_b+\Delta \sigma_{b/a})-f(\sigma_b)}{\Delta\sigma_{b/a}}
\end{eqnarray}
\begin{figure}
\let\picnaturalsize=N
\def3in{3in}
\deftempe.ps{mems.ps}
\ifx\nopictures Y\else{\ifx\epsfloaded Y\else\input epsf \fi
\let\epsfloaded=Y
\centerline{\ifx\picnaturalsize N\epsfxsize 3in\fi
\epsfbox{tempe.ps}}}\fi
\caption{The three different parameterizations of two crossing membranes
(seen with little perspective).}
\label{mems}
\end{figure}
There are three different possibilities that correspond to the three ways
one can separate two crossing planes into two surfaces. These
representations are, in the matrix language, gauge equivalent matrices. This
gauge enhancement can be seen as responsible for perturbative
splitting-and-joining interactions between membranes.
A similar argument can be used to explain the attachment of the open
strings to the D-brane.
Consider, then, a matrix that can be `diagonalized' into two boxes: one is
the membrane and the other is a usual matrix string state. There can be an
enhancement of symmetry if at least at one particular value of $\sigma$ (the
original world-sheet) the coordinate of the string coincides with one of
the target space points occupied by the D-brane. That means that their
transverse coordinates must coincide and that the coordinates of the string
which are paralel to the membrane must be equal to one of the eigenvalues of
each non-commuting matrix.
In that case, it is possible to exchange each matrix (seen as a
derivative operator acting over the Hilbert space in the representation of
the operator of the other direction)
\begin{equation}
\left. \partial f \right|_i= f(i+1)-f(i-1)
\end{equation}
with another like this:
\begin{eqnarray}
\left. \partial f \right|_i= \left( f(i+1)-f(i-1) \right) \nonumber \\
+\delta_{\kappa,i+1} \times \left( f(\kappa)-f(i+1) \right)
-\delta_{\kappa,i-1} \times \left( f(\kappa)-f(i-1) \right)
\end{eqnarray}
if $\kappa$ is the index that marks the field of the string. We have not
cared much about the normalizations. This change is, in fact, equivalent to
interchanging the matrix indices $i$ and $\kappa$ so that it is a symmetry
for both matrices, $\hat{X}^7$ and $\hat{X}^8$ (this is important). This
symmetry has to exist in any representation of the matrices (any
parameterization), in particular, both where one of them is diagonal. I
point this out to remark the necessity for both eigenvalues of the position
of the string to coincide with a point inside the membrane. This is a
locally (in $\sigma$) conserved gauge symmetry. In this case the enhancement
is to a $Z_2$ symmetry (instead of $SU(2)$ from a $Z_2$, like in the case of
closed strings).
This symmetry allows perturbative pointlike interactions between strings
and D2-branes. Besides it allows the creation of a kind of `twisted states'
that are related to the attached open strings and the boundary closed string
states. Figure (\ref{twist}) is a graphic explanation. Wherever two lines
coincide a $Z_2$ twist is performed and one field takes the role of the
other. There can be two special points (or just one) that mark the two ends
of the open string. As can be seen in the figure, the field that comes out
of the membrane after the twist to substitute the string (straight lines
with slope) is the way the momentum flows back between the two ends of the
string. As a final result of the twist the fields that represent the two
positions inside the membrane (thick lines) are interchanged. This is
possible thanks to the same symmetry that has been described above. In a
membrane state without any open string, the interchange of the value of two
diagonal fields is not permitted because it would change the commutator with
the matrix in the other direction. In this case, the twist has not only
swapped the eigenvalues, but also the momentum-like matrix in such a way
that the total action is part of the complete reparameterization group.
Somehow, this is the recovery of part of the usual Weyl group that appears
in perturbative Matrix String Theory. Moreover, one can consider two
membranes so that the two special points are one in each of them. Then the
swap of indices strongly reminds the fact that stretching open strings are
represented in the usual SYM D-brane action as off-diagonal fields with the
two indices ($X^{ij}$ if $i$ and $j$ are the indices of the branes) and
therefore those that mediate their interchange.
When an open string is part of a twisted state, its only movement can be
to continuously change its endpoints. For that it is essential that $N$ is
infinite so that the spectrum of eigenvalues of the matrices in continuous.
The centre of mass is tied to the membrane and the only way out it to join
the two endpoints so that the twisted state can break without loss of
energy.
\begin{figure}
\let\picnaturalsize=N
\def3in{3in}
\deftempe.ps{twist.ps}
\ifx\nopictures Y\else{\ifx\epsfloaded Y\else\input epsf \fi
\let\epsfloaded=Y
\centerline{\ifx\picnaturalsize N\epsfxsize 3in\fi
\epsfbox{tempe.ps}}}\fi
\caption{Twisted state of an open string with two membrane points. As
$\sigma$ runs, the fields in the non-commuting $SU(N)$ part cannot oscillate
or move (horizontal lines) because their excitations are non-perturbative
but the open string field can go from one point to the other.}
\label{twist}
\end{figure}
Once a state like that has been excited, interactions with other
external strings are easier to understand because away from the fixed
points, the strings behave just like ordinary matrix strings and interact as
usual.
\section{Membranes with different charges and different geometries.}
A membrane with opposite charge to the one studied in the previous
section would be, for instance
\begin{eqnarray}
\hat{X}^7=X^7 x \nonumber \\
\hat{X}^8=- X^8 y \nonumber \\
\left[ x,y \right]=1,
\end{eqnarray}
so that
\begin{equation}
\left[ \hat{X}^7,\hat{X}^8 \right]=- X^7 X^8 \times Id.
\end{equation}
They are related by a change in the orientation of the base that
parameterizes them. A representation of the fields was
\begin{equation}
\hat{X^8}=\frac{\partial}{\partial X^7}
\end{equation}
for the previous case while for the anti-brane it is
\begin{equation}
\hat{X^8}=-\frac{\partial}{\partial X^7}.
\end{equation}
The conclusion from this is that it is possible to associate to every
membrane the vector product of the vectors of the basis over its surface. It
represents its orientation in the sense that it goes inwards in one face of
the membrane and outwards in the other. If the vectors of two membranes have
the same direction, they have the same charge; else, they are oppositely
charged.
This is also true looking directly at the supersymmetry algebra of the
membranes as BPS states, but the form it adopts is different. The
Ramond-Ramond charge can be seen as a central charge of the algebra in the
form of a constant anti-symmetric tensor with two indices. It is a tensor
that transforms covariantly under Lorentz transformations. One can, in
particular, consider as before a membrane lying in the seventh and the
eighth directions. Its charge has the form
\begin{equation}
Z^{78}=-Z^{87}=C.
\end{equation}
It is very easy to see that after a $180^o$ rotation, the charge changes to
\begin{equation}
Z'^{78}=-Z'^{87}=-C.
\end{equation}
This will have interesting consequences when discussing membranes of
different geometries, which is done in the following. It also explains, in a
very geometrical way why the coupling to the Ramond-Ramond forms is opposite
in each case. Let us consider, for simplicity, the coupling of the massless
level of the closed string to the D2-brane. It only couples to the graviton
and to the Ramond-Ramond three-form. The coupling to the graviton involves
only the energy-stress tensor and, therefore, it is universal and
independent of the charge. However, the form couples through the integral
\begin{equation}
\int d x_7 d x_8 A^{780}.
\end{equation}
Changing the orientation of the base that spans the plane means changing
the sign of the product $d x_7 d x_8$ or, equivalently, interchanging the
limits of one of the integrals. That is why
\begin{equation}
\int d x'_7 d x'_8 A^{780}=-\int d x_7 d x_8 A^{780}
\end{equation}
and the coupling to the form has the opposite sign.
Once we have studied plane membranes, it is not difficult to construct
the fields we need to describe any other membrane. The algorithm is: take a
parameterization of the surface and then equate two locally perpendicular
coordinates to any representation of the two non-commuting matrices ($x$ and
$y$) that have been used in the previous section. An easy example is a
cylinder. The fields we need are
\begin{eqnarray}
\hat{\phi}=\phi \times x \nonumber \\
\hat{X^8}=X^8 \times y
\end{eqnarray}
with
\begin{equation}
[x,y]=1.
\end{equation}
Then
\begin{eqnarray}
\hat{X^8}=X^8 \nonumber \\
\hat{X^7}=\sin \hat{\phi} \nonumber \\
\hat{X^6}=\cos \hat{\phi},
\end{eqnarray}
with an appropriate normalization. These three fields describe the
cylindrical membrane. We have needed one more parameter than in the plane
case because it extends itself in three dimensions. The proportion between
the expectation values of $X^6$ and $X^7$ is related to the excentricity of
the ellipse that appears cutting the cylinder perpendicularly to its height.
It is very interesting to see that even though the energy of this background
corresponds to that of a membrane with that form and surface ($T_2 \times
A$), the total D2-charge is zero because the trace of both commutators
($[\hat{X}^7,\hat{X}^8]$ and $[\hat{X}^6,\hat{X}^8]$) vanish and thus all
the supersymmetry is exact and this state cannot be considered BPS. What
happens is that, {\cal locally}, we can define a conserved charge, but when
one integrates over the surface (sums the trace up) the charge of each point
is cancelled against the diametrically opposed one. This is directly related
to the rotational properties of the charge that have been described a little
above. When the surface curves around itself, the orientation of the inner
face continuously turns so that the points that stand in front of each other
possess opposite charges. In that sense, a self-interaction can occur such
that a closed string gets out of the membrane at one point and is reabsorbed
at the diametrically opposed one. This string interchange has the same
properties as that of the brane-antibrane problem. In particular, if the
cylinder is small enough, some open-string states become tachyonic and the
system unstable. The conclusion is that this cylindrical membranes are only
semi-stable, and the larger their radii are, the more stable. If the radius
of the cylinder becomes of order $\alpha'$, non-perturbative interactions
involving off-diagonal fields that transport D2-charge annihilate the
membrane and the supersymmetry is recovered in all space.
This analysis can be directly generalized to the torus, the sphere and
any other closed surface. It is important to remark that these are curved
surfaces embedded in a flat space-time. If the membrane is toroidal but is
wrapping around a toroidal space, this is not valid. In that case,
`opposite' points are not related by a $180^o$ rotation, but rather by a
$\pi R$ translation. This, of course, does not change the orientation and
therefore the total charge does not vanish.
If we want to describe those wrapped membranes, we can perform a
T-duality of the compactified theory. Then the definition of the central
charge is
\begin{equation}
\Phi=\frac{i}{4 \pi} \int d\sigma d\sigma_7 d\sigma_8 tr
< \left[ A^7,A^8 \right] >=T_2 A
\end{equation}
and the potential term
\begin{equation}
V=\frac{1}{g_s^2}\int d\sigma d\sigma_7 d\sigma_8 tr
\left[ A^7,A^8 \right]^2.
\end{equation}
It is clear that a proper choice of the constant fields (proportional to
$x$ and $y$, for example) one can obtain configurations with non-zero
charge.
Another interesting situation is a background where two or more membranes
of the same charge coincide. There, we should be able to see an enhancement
of the symmetry to a local (in the world-volume) $SU(N)$ gauge symmetry.
This happens indeed. The system is described by a matrix that can be put in
the form of two independent boxes, each representing one membrane. When all
the expectation values of all the identity matrices that represent the
centres of mass of the membranes are identical, these two fields
\begin{equation}
\left( \begin{array}{cc} 0 & I \\ I & 0 \end{array}
\right) \hspace{1cm} \mbox{and} \hspace{1cm} \left(
\begin{array}{cc} 0 & i I \\ -i I & 0 \end{array} \right)
\end{equation}
become massless and generate a {\cal global} $SU(N)$ ($SU(2)$ in this case)
symmetry. The local symmetry is related to the freedom the boundary closed
strings (open strings in the other picture) have to form `twisted states' of
the kind described in the previous section. The states that connect the two
membranes are now massless and are responsible for the $SU(N)$ symmetry
restoration at every point in the world-volume. An even better way of
understanding it is to notice that, indeed, the possible choice of two
operators as derivative of the dual ones is local, that is, it can be done
for every point in the world-volume.
\section{Thermodynamical effects of membranes.}
As has already been commented, the study of the effects of static,
non-dynamical D-branes in string thermodynamics has been studied at length
in several articles \cite{many}. What I would like to add is a glimpse of
what happens to the systems when the thermodynamical energies are high
enough to create pairs of membranes and anti-membranes. In the traditional
scenarios, the branes were stable because they possess a conserved charge
and so cannot decay; however, when energies are of order $g^{-1}$ a finite
density of non-perturbative objects should appear even if the total charge
is zero. In a certain sense, their enormous mass only acts as a chemical
potential that obstructs their nucleation. This affects the system in a
different manner if we use the Canonical or Microcanonical Ensemble and if
we suppose the total energy to be finite or we deal with densities.
The picture is clearer if we suppose we are in the Microcanonical
Ensemble and the total energy is finite. In that case, the low energy
behaviour is `protected' from non-perturbative effects (that means:
unaffected by them) thanks to their mass. Only when the energies are high
enough should we consider them. We have seen that the spectrum of membranes
includes the quantization of the classical modes. The number of these modes
grows very fast with the energy but it is not possible (rather: we do not
know of any way) to calculate their degeneracy exactly; nevertheless, there
is a computation in \cite{ealvarez} that supplies an asymptotic form for the
density of mass states. The formula is
\begin{equation}
\Omega(m)=\exp (m^{4/3})
\end{equation}
This should approximately count the degeneracy of the modes in formulas
(\ref{longitudinal}) and (\ref{transversal}) and others alike. It only
considers toroidal membranes but it is enough to see the asymptotic
behaviour. They are the non-pertubative excitations that change the form of
the membrane and come as a quantization of its classical oscillation modes.
If we assume that the temperature at those energies is very near Hagedorn,
the density of states we should use is
\begin{equation}
\Omega(E)={\cal H}\left( E-\frac{1}{g^{1/3}}\right) \exp (E^{4/3})+ {\cal H}
\left( \frac{1}{g^{1/3}}-E \right) e^{\beta_H E}
\end{equation}
The term in the right corresponds to the gas of strings at the Hagedorn
temperature. It gives a behaviour shown in figure \ref{tempe}. I have
supposed that the membranes begin having their asymptotic behaviour soon
after their first modes are excited. The temperature can be seen to be
decreasing so that the specific heat is negative. The asymptotic temperature
for very high energies is zero. This is a consequence of the fact that
membranes as well as higher dimensional objects have senseless
thermodynamics. Nevertheless, the sense is recovered in this case thanks to
the large `chemical potential' that is the mass. The negative specific heat
is the result of a continuous series of relativistic mass thresholds being
opened. It is exactly the same phenomenon as Hagedorn, but with the
asymptotic temperature equalling zero. The huge mass degeneracy makes it
much more entropic to accumulate energy in the form of mass than in the form
of kinetic energy (temperature).
\begin{figure}
\let\picnaturalsize=N
\def3in{3in}
\deftempe.ps{tempe.ps}
\ifx\nopictures Y\else{\ifx\epsfloaded Y\else\input epsf \fi
\let\epsfloaded=Y
\centerline{\ifx\picnaturalsize N\epsfxsize 3in\fi
\epsfbox{tempe.ps}}}\fi
\caption{Temperature of a IIA string gas at non-perturbative energies.}
\label{tempe}
\end{figure}
The situation when one takes the thermodynamical limit ($V \rightarrow
\infty$ and $E \rightarrow \infty$) is worse because any individual object
can survey the whole energy range. In this case equipartition is broken from
the very beginning and, in principle, one membrane would absorb everything
around it until the universe were completely frozen. This picture, however,
has to be changed a little. Since we are dealing with the membrane as a
single object, the situation becomes similar to the appearance of black
holes at finite temperature. Black holes are always much more degenerate
than anything around them and classically one should expect them to grow and
grow until nothing else exists. Fortunately, this is changed when the first
quantum corrections are introduced: the black hole can and does reach
equilibrium with a gas of massless fields, its own Hawking radiation. This
happens because of two things: firstly, the degeneracy of the black hole,
large as it is, is not infinite and it always leaves some space for the rest
of the objects; and secondly, interaction makes these objects (the membrane
as well as the black hole) have a finite probability to emit energy in the
form of strings. Another source of instability for membranes is that we
assume as a natural fact that the total D2-charge of the universe is zero.
This means that pairs of membranes can decay into a very large number of
strings.
All this phenomena (emition, pair annihilation and scattering) heat the
gas that gets a finite temperature when the energy density is also finite.
Presumably this temperature should be lower than Hagedorn and go down to
zero when the density tends to infinity.
A more serious problem is trying to see the perspective that offers the
Canonical Ensemble. The simple attempt to calculate the Free Energy is
hopeless:
\begin{equation}
\beta F(\beta)=\int d E \Omega(E) e^{-\beta E}=\int d E e^{-\beta
E} e^{\beta E^{4/3}}=\infty
\end{equation}
It has not got any sense at any temperature. This is, of course, due to
the fact that the behaviour of the system is by no means regular at any
temperature.
\section{Speculations about higher dimensional objects and further comments.}
After an intense study of many years \cite{hag,us12} on the problem of
Hagedorn, many conclusions have been achieved but still many doubts remain.
The physical picture has evolved but, although some non-perturbative effects
were added \cite{many} the problem has remained more or less the same
because only the dynamics of the strings was studied. The emergence of a
fully eleven-dimensional M-theory with unwrapped free membranes seems to
change things beyond non-perturbative energies. Moreover, after we have seen
the drastic effects of the oscillating modes of membranes, one could wonder
what could happen when higher dimensional objects like five-branes enter the
scene. Any calculation is impossible at this moment because we do not know
the precise spectrum and can only guess whether all the classical modes of a
five-brane can be excited on the M-five-brane or not. The same happens with
the nine-brane. If we naively accept that they do, the result at extremely
high temperatures is the preeminence of the single nine-brane, that is, the
whole space-time oscillating in high frequency modes. As non-perturbative
deformations of space-time, black holes should have an important role to
play.
Another thing that I want to explain is why branes with different
dimensions behave so differently when, in the light of T-duality, they are
basically the same kind of objects. To understand this better, we look at
the construction of T-duality in \cite{taylor}. A D0-brane moving in a
space compactified in a circle becomes equivalent to a wrapped D1-brane
moving in the dual space. The clearest image is an intermediate one that
consists in noticing that a D0-brane in a circle is equivalent to a linear
lattice of them in an open space with a constant separation. This means
that, in fact, the wrapped D-string dynamics is equivalent to that of an
infinite number of D-particles. Following the argument the dynamics of a
wrapped D-membrane is reproduced by a bidimensional rectangular lattice of
D-particles. However, the D-string or the D-membrane that come out of this
description are rigid and must be wrapping a torus. The D-particle cannot
reproduce the oscillations of the higher dimensional objects. To put it
short: the dynamics of lower dimensional D-branes are fully contained in
that of the higher dimensional ones, but not vice versa.
\section{Conclusions.}
Throughout this work, I have tried to incorporate the advances in the
knowledge of non-perturbative corrections to String Theory that has yielded
the Matrix String Theory. The picture is more dynamical and clearer than had
been the previous images of Dirichlet branes. For objects with two
dimensions or less, the construction is powerful enough to allow quite a
complete treatment of the whole spectrum.
After the first, qualitative approach of the global thermal consequences
made in \cite{us4}, a more quantitative calculation was in need. After
reproducing the usual perturbative string limit, I have introduced the
effects of free D-particles. This includes the bound states among
themselves, with closed strings and, also, the open strings whose two ends
are tied to the same D-particle. The expression obtained is not explicitly
invariant under modular transformations, possibly because not all the open
string states were included; however it has a critical temperature that
precisely equals the Hagedorn one.
Regarding the membranes, the complexity of their spectrum makes it much
harder to obtain exact expressions. However it is possible to find out the
symmetry that fixes the open strings to the D2-brane and that is responsible
for their perturbative interaction. Apart from this usual perturbative
states, I have found the fields that represent the quantization of the
classical oscillations of the membrane as an extended object with a given
tension. The inclusion of these states changes drastically the
thermodynamics at non-perturbative energies and beyond. The canonical
description seems imposible while in the Microcanonical Ensemble, the system
there is dominated by a single large membrane extremely excited that tends
to absorb everything that comes in its way. This cools the universe so much
that the asymptotic temperature is zero.
Maybe the most graphic way to understand this cooling is in a more or
less cosmological manner. Take the universe to be filled with an enormous
energy and that we slowly decrease it. At the beginning we have a cold gas
in thermal equilibrium with the excited membranes. The number of membranes
tends to two because the total D2-charge must be zero. As we extract energy
from the system, it gets hotter because the membranes are more and more
unstable and tend to annihilate between themselves. Eventually they
completely dissolve in the string gas that heats up to the Hagedorn
temperature. In that moment the picture is the one drawn in \cite{us12}
with a fat string in equilibrium with the surrounding sea.
One of the reasons for this strange behaviour is the absence of modular
invariance, that is lost when the eleventh dimension is discovered. Maybe
there is another symmetry that involves higher dimensional objects
(five-branes) that might correct this in part.
\section*{Acknowledgements.}
I have to thank M. Laucelli Meana and M. A. R. Osorio for discussions on
this topic (and many others).
\newpage
|
\section{Introduction}
It was shown in~\cite{kor sai} that the famous Veneziano amplitude, from
which all the string theory starts, comes naturally from one of the
simplest solutions of the functional pentagon equation (FPE).
More generally, FPE is
intimately connected with the duality condition for scattering
processes.
From the viewpoint of the theory of integrable models, FPE is a rather
trivial equation whose solutions have transparent geometrical or
group-theoretic meaning~\cite[section~5]{kor sai}. It looks natural
to search for similar constructions with FPE replaced by
the functional tetrahedron equation (FTE). As the relations between
the pentagon and duality condition are like those between the tetrahedron
and local Yang--Baxter equation (LYBE), the duality condition is likely
to be replaced by LYBE.
In this paper, I find such FTE and LYBE solutions that are described
by formulas very similar to those describing Veneziano amplitude
in~\cite{kor sai}, including the fundamental property of M\"obius
invariance. They are what I mean by the tetrahedral analog of
Veneziano amplitude.
\section{A functional transformation for edge variables from
refactorization equation}
Consider the following ``refactorization equation''
for the product of three matrices:
\begin{eqnarray}
&&
\pmatrix{ a_1&b_1& 0\cr c_1&d_1& 0 \cr
0 & 0 & 1 }
\pmatrix{ a_2& 0 &b_2\cr 0 & 1 & 0 \cr
c_2& 0 &d_2 }
\pmatrix{ 1 & 0 & 0 \cr
0 &a_3&b_3\cr 0 &c_3&d_3 } =
\nonumber\\[0.5\normalbaselineskip]
&& =
\pmatrix{ 1 & 0 & 0 \cr
0 &a'_3&b'_3\cr 0 &c'_3&d'_3 }
\pmatrix{ a'_2& 0 &b'_2\cr 0 & 1 & 0 \cr
c'_2& 0 &d'_2 }
\pmatrix{ a'_1&b'_1& 0 \cr c'_1&d'_1& 0 \cr
0 & 0 & 1 },
\label{ieq pererazlozhenie}
\end{eqnarray}
($a_1,\ldots, d'_3$ are numbers) for the case when all six
submatrices $\pmatrix{a_i^{(\prime)} & b_i^{(\prime)}\cr
\noalign{\vskip0.05\baselineskip}
c_i^{(\prime)} & d_i^{(\prime)}}$ have the form
\begin{equation}
\pmatrix{a & b\cr c & d}=\pmatrix{\alpha & 1-\alpha\cr 1-\beta & \beta}.
\label{eq alpha}
\end{equation}
In other words, each of the six matrices in~(\ref{ieq pererazlozhenie})
transforms the vector $\pmatrix{1\cr 1\cr 1}$ into itself.
It is known from~\cite{pomi-iv,dokt,troe} that each side of
(\ref{ieq pererazlozhenie}) determines the other side to within
some ``gauge freedom'', and one can verify that the additional
conditions~(\ref{eq alpha}) are exactly good for fixing that freedom.
The fate of an {\em arbitrary\/}
vector $\pmatrix{p\cr q\cr r}$ under the action
of both sides of (\ref{ieq pererazlozhenie}) is more
complicated. We present it in Figure~\ref{fig pqr},
\begin{figure}[hb]
\begin{center}
\unitlength=1mm
\special{em:linewidth 0.4pt}
\linethickness{0.4pt}
\begin{picture}(100.00,40.00)
\put(0.00,0.00){\line(1,1){40.00}}
\put(30.00,40.00){\line(0,-1){40.00}}
\put(0.00,10.00){\line(1,0){40.00}}
\put(3.00,4.00){\makebox(0,0)[rb]{$y$}}
\put(3.00,11.00){\makebox(0,0)[cb]{$x$}}
\put(29.00,0.00){\makebox(0,0)[rb]{$z$}}
\put(38.00,11.00){\makebox(0,0)[cb]{$p$}}
\put(36.00,37.00){\makebox(0,0)[rb]{$q$}}
\put(29.00,40.00){\makebox(0,0)[rt]{$r$}}
\put(23.00,11.00){\makebox(0,0)[cb]{$u$}}
\put(20.00,21.00){\makebox(0,0)[rb]{$v$}}
\put(31.00,17.00){\makebox(0,0)[lc]{$w$}}
\put(11.00,9.00){\makebox(0,0)[lt]{$X_1$}}
\put(31.00,9.00){\makebox(0,0)[lt]{$X_2$}}
\put(31.00,29.00){\makebox(0,0)[lt]{$X_3$}}
\put(30.00,30.00){\circle*{1.00}}
\put(30.00,10.00){\circle*{1.00}}
\put(10.00,10.00){\circle*{1.00}}
\put(60.00,30.00){\line(1,0){40.00}}
\put(70.00,40.00){\line(0,-1){40.00}}
\put(60.00,0.00){\line(1,1){40.00}}
\put(70.00,30.00){\circle*{1.00}}
\put(70.00,10.00){\circle*{1.00}}
\put(90.00,30.00){\circle*{1.00}}
\put(60.00,31.00){\makebox(0,0)[lb]{$x$}}
\put(62.00,3.00){\makebox(0,0)[rb]{$y$}}
\put(69.00,0.00){\makebox(0,0)[rb]{$z$}}
\put(100.00,31.00){\makebox(0,0)[rb]{$p$}}
\put(96.00,37.00){\makebox(0,0)[rb]{$q$}}
\put(71.00,40.00){\makebox(0,0)[lt]{$r$}}
\put(80.00,31.00){\makebox(0,0)[cb]{$f$}}
\put(79.00,20.00){\makebox(0,0)[rb]{$g$}}
\put(69.00,20.00){\makebox(0,0)[rc]{$h$}}
\put(69.00,31.00){\makebox(0,0)[rb]{$Y_2$}}
\put(91.00,29.00){\makebox(0,0)[lt]{$Y_1$}}
\put(69.00,11.00){\makebox(0,0)[rb]{$Y_3$}}
\end{picture}
\end{center}
\caption{}
\label{fig pqr}
\end{figure}
where we denote the matrices entering~(\ref{ieq pererazlozhenie}), in their
order in that equation, by letters $X_1$, $X_2$, $X_3$, $Y_3$, $Y_2$,
$Y_1$. The meaning of the LHS of Figure~\ref{fig pqr} is that
$$
X_3 \pmatrix{p\cr q\cr r} = \pmatrix{p\cr v\cr w}, \qquad
X_2 \pmatrix{p\cr v\cr w} = \pmatrix{u\cr v\cr z}, \qquad
X_1 \pmatrix{u\cr v\cr z} = \pmatrix{x\cr y\cr z},
$$
while the meaning of the RHS is that
$$
Y_1 \pmatrix{p\cr q\cr r} = \pmatrix{f\cr g\cr r}, \qquad
Y_2 \pmatrix{f\cr g\cr r} = \pmatrix{x\cr g\cr h}, \qquad
Y_3 \pmatrix{x\cr g\cr h} = \pmatrix{x\cr y\cr z}.
$$
One can see that if, vice versa, all the values $x,y,z,\ldots$
in e.g.\ the LHS of Figure~\ref{fig pqr} are given, then matrices
$X_1,X_2,X_3$
of the form~(\ref{eq alpha}) are recovered unambiguously.
So, we can take some given values of nine numbers in the LHS,
get the triple of matrices $X_1,X_2,X_3$ from them, then get
$Y_1,Y_2,Y_3$ by (\ref{ieq pererazlozhenie}), and then get the
missing values $f,g,h$ in the RHS from $p,q,r$ using $Y_1,Y_2,Y_3$.
We will formulate this the following way: for any fixed ``outer''
variables $x$, $y$, $z$, $p$, $q$, $r$, the transformation
\begin{equation}
R=R(x,y,z,p,q,r)\;\colon\;\; (u,v,w)\mapsto (f,g,h)
\label{eq R}
\end{equation}
is given.
The transformations (\ref{eq R}) satisfy the {\em functional
tetrahedron equation\/} (FTE).
To explain this, note that equation (\ref{ieq pererazlozhenie})
can be naturally regarded as an equation in the direct sum of three
one-dimensional complex linear spaces, each of the matrices
acting nontrivially only in a direct sum of two
of them. One can consider similar relations in a direct sum of {\em four\/}
spaces (each of the matrices acting nontrivially again only in a direct sum
of two spaces). Let us picture in Figure~\ref{fig 4lines}
\begin{figure}[ht]
\begin{center}
\unitlength=1mm
\special{em:linewidth 0.4pt}
\linethickness{0.4pt}
\begin{picture}(100.00,40.00)
\put(0.00,5.00){\line(2,1){40.00}}
\put(0.00,10.00){\line(1,0){40.00}}
\put(30.00,0.00){\line(0,1){40.00}}
\put(15.00,0.00){\line(1,2){20.00}}
\put(60.00,30.00){\line(1,0){40.00}}
\put(70.00,40.00){\line(0,-1){40.00}}
\put(65.00,0.00){\line(1,2){20.00}}
\put(60.00,15.00){\line(2,1){40.00}}
\put(10.00,10.00){\circle*{1.00}}
\put(20.00,10.00){\circle*{1.00}}
\put(30.00,10.00){\circle*{1.00}}
\put(23.33,16.67){\circle*{1.00}}
\put(30.00,20.00){\circle*{1.00}}
\put(30.00,30.00){\circle*{1.00}}
\put(70.00,30.00){\circle*{1.00}}
\put(80.00,30.00){\circle*{1.00}}
\put(90.00,30.00){\circle*{1.00}}
\put(70.00,10.00){\circle*{1.00}}
\put(70.00,20.00){\circle*{1.00}}
\put(76.67,23.33){\circle*{1.00}}
\put(9.00,11.00){\makebox(0,0)[rb]{$1$}}
\put(21.00,9.00){\makebox(0,0)[lt]{$2$}}
\put(31.00,9.00){\makebox(0,0)[lt]{$4$}}
\put(31.00,19.00){\makebox(0,0)[lt]{$5$}}
\put(29.00,31.00){\makebox(0,0)[rb]{$6$}}
\put(22.50,17.90){\makebox(0,0)[rb]{$3$}}
\end{picture}
\end{center}
\caption{}
\label{fig 4lines}
\end{figure}
the spaces as straight lines, put matrices at their intersections,
and attach the results of matrix action upon some
4-vector to line segments like in Figure~\ref{fig pqr},
and then consider the transition from the LHS of Figure~\ref{fig 4lines}
to its RHS as a composition of ``elementary'' transformations~$R$
of type~(\ref{eq R}).
As was explained in the paper~\cite{troe} (and the reader will verify
it him-/herself easily), there exist two different compositions of four
$R$s both transforming the LHS of Figure~\ref{fig pqr} in its RHS.
The first of them starts with~$R_{356}$, by which we mean ``turning
inside out'' triangle~$356$, while the other---with~$R_{123}$.
We can write FTE in the same abstract form as in~\cite{troe}:
\begin{equation}
R_{123} \circ R_{145} \circ R_{246} \circ R_{356} =
R_{356} \circ R_{246} \circ R_{145} \circ R_{123},
\label{ieq1}
\end{equation}
but the sense of (\ref{ieq1}) is now different: $R$ is now a transformation
of variables belonging to the {\em edges\/} rather than of matrices
belonging to vertices.
To prove FTE (\ref{ieq1}) for edge variables, note that the variables
belonging to {\em inner\/} edges (i.e., say, edges $12$, $13$, \ldots,
$56$ in the LHS of Figure~\ref{fig 4lines}) are unambiguously recovered
if variables at {\em outer\/} edges and matrices at vertices are given.
The FTE for matrices, according to~\cite{troe}, does hold, while the
variables at outer edges are not changed by the transformations. Thus,
the variables at inner edges do not depend on the way of transformations
as well.
\section{M\"obius invariance}
The same way as we have traced the fate of vector $\pmatrix{p\cr q\cr r}$
under the action of LHS and RHS of (\ref{ieq pererazlozhenie})
in Figure~\ref{fig pqr}, we can trace the fate of two more vectors,
namely
\begin{equation}
\pmatrix{p_n\cr q_n\cr r_n} = \kappa \pmatrix{1\cr 1\cr 1} +
\lambda \pmatrix{p\cr q\cr r}
\quad \hbox{and} \quad
\pmatrix{p_d\cr q_d\cr r_d} = \mu \pmatrix{1\cr 1\cr 1} +
\nu \pmatrix{p\cr q\cr r},
\label{*}
\end{equation}
where $\kappa,\lambda,\mu,\nu$ are some constants (and subcripts
$n$ and $d$ stand for ``numerator'' and ``denominator'', see
formula~(\ref{eq ed}) below).
I do not draw here corresponding diagrams, differing from
Figure~\ref{fig pqr} only in that $n$ or $d$ is added to all small
letters.
Now let us do the following {\em gauge transformations\/} (in the
sense of~\cite{pomi-iv,dokt,troe}) on matrices $X_1,\ldots,Y_3$:
\begin{eqnarray*}
\hitrayakomanda 1 x y u v {}, \\
\hitrayakomanda 2 u z p w {}, \\
\hitrayakomanda 3 v w q r {}, \\
\hitrayakomanda 1 f g p q {\prime}, \\
\hitrayakomanda 2 x h f r {\prime}, \\
\hitrayakomanda 3 y z g h {\prime}.
\end{eqnarray*}
Here, of course, $x_d=\mu+\nu x$ etc., in analogy with $p_d$, $q_d$ and~$r_d$ in~(\ref{*}).
Denote the so obtained matrices $\tilde X_1, \ldots, \tilde Y_3$. Now
imagine a version of Figure~\ref{fig pqr} for these matrices with tildes.
One can say that the transformation of vectors corresponding to the
above gauge matrix transformation has brought all variables with
subscript~$d$ into~$1$, and hence the matrices with tildes have again
the form~(\ref{eq alpha}). As for the variables with subscript~$n$,
they turned into
\begin{equation}
x\to \tilde x={x_n\over x_d}={\kappa+\lambda x\over \mu+\nu x}
\quad \hbox{etc.}
\label{eq ed}
\end{equation}
We see from here that a linear-fractional (M\"obius) transformation
of variables $x,y,\ldots$ commutes with the transformation~$R$.
Clearly, the same conclusion could be made from the explicit formulas
for $R$ given in Section~\ref{sec explicit}. The M\"obius invariance
is an argument in support of the idea that $R$ really is an analog of
``pentagonal'' transformation from~\cite{kor sai} connected with
Veneziano amplitude.
\section{Connection between volume elements and the explicit form of
functional transformation}
\label{sec explicit}
Let us now vary the edge variables in Figure~\ref{fig pqr}, with
matrices $X_1,\ldots Y_3$ fixed. For instance, consider the variables
at outer edges of the LHS of that Figure as functions of three inner
variables $u,v,w$, and calculate the corresponding partial derivatives.
The reader will easily check that
\begin{equation}
\pder xu={x-v\over u-v},\quad
\pder xv={x-u\over v-u},\quad
\pder yu={y-v\over u-v}
\label{eq pder}
\end{equation}
and so on.
Using formulas of the type (\ref{eq pder}), it is not hard to obtain
the following relations for ``volume elements'':
\begin{equation}
{\rm d} x \wedge {\rm d} y \wedge {\rm d} z =
{x-y\over u-v}{z-u\over w-u} \,{\rm d} u \wedge {\rm d} v \wedge {\rm d} w
\label{eq diff1}
\end{equation}
from the LHS of Figure~\ref{fig pqr} and similarly
\begin{equation}
{\rm d} x \wedge {\rm d} y \wedge {\rm d} z =
{x-h\over f-h}{y-z\over g-h} \,{\rm d} f \wedge {\rm d} g \wedge {\rm d} h
\label{eq diff2}
\end{equation}
from its RHS. The equalness of the RHSs of (\ref{eq diff1}) and
(\ref{eq diff2}) can be called ``the relation between
${\rm d} u \wedge {\rm d} v \wedge {\rm d} w$ and ${\rm d} f \wedge {\rm d} g \wedge {\rm d} h$
got via ${\rm d} x \wedge {\rm d} y \wedge {\rm d} z$''.
Similarly, the equalness of the RHSs of relations
\begin{equation}
{\rm d} y \wedge {\rm d} z \wedge {\rm d} p =
{y-u\over v-u}{z-p\over w-u} \,{\rm d} u \wedge {\rm d} v \wedge {\rm d} w
\label{eq diff3}
\end{equation}
and
\begin{equation}
{\rm d} y \wedge {\rm d} z \wedge {\rm d} p =
{y-z\over g-h}{p-g\over f-g} \,{\rm d} u \wedge {\rm d} v \wedge {\rm d} w
\label{eq diff4}
\end{equation}
can be called ``the relation between
${\rm d} u \wedge {\rm d} v \wedge {\rm d} w$ and ${\rm d} f \wedge {\rm d} g \wedge {\rm d} h$
got via ${\rm d} y \wedge {\rm d} z \wedge {\rm d} p$''. There are four more pairs
of relations of the type (\ref{eq diff1}--\ref{eq diff4}) with
${\rm d} z \wedge {\rm d} p \wedge {\rm d} q$, ${\rm d} p \wedge {\rm d} q \wedge {\rm d} r$,
${\rm d} q \wedge {\rm d} r \wedge {\rm d} x$ and ${\rm d} r \wedge {\rm d} x \wedge {\rm d} y$
respectively in their LHSs.
Certainly, one can exclude the differentials from those relations
and obtain formulas giving explicitely the connection between
edge variables, i.e.\ the transformation~$R$, namely
\begin{eqnarray}
{x-y\over u-y}\,{u-z\over p-z}&=&{x-h\over f-h}\,{f-g\over p-g},
\label{eqnarray nachalo} \\
{y-x\over v-x}\,{v-r\over q-r}&=&{y-h\over g-h}\,{g-f\over q-f}, \\
{z-p\over w-p}\,{w-q\over r-q}&=&{z-g\over h-g}\,{h-f\over r-f}, \\
{x-v\over u-v}\,{u-w\over p-w}&=&{x-r\over f-r}\,{f-q\over p-q}, \\
{y-u\over v-u}\,{v-w\over q-w}&=&{y-z\over g-z}\,{g-p\over q-p}, \\
{z-u\over w-u}\,{w-v\over r-v}&=&{z-y\over h-y}\,{h-x\over r-x}.
\label{eqnarray konec}
\end{eqnarray}
\section{Local Yang--Baxter equation}
The local Yang--Baxter equation (LYBE) dealt with in this section
differs from the conventional Yang--Baxter equation, first, in its
continuous (instead of usual discrete) ``set of colours'' and, second
(and this is what makes it ``local'', or ``twisted''),
in that all six $R$-matrices (instead of which we will have, however,
functions of 4 complex variables) entering it are different
(in the usual Yang--Baxter equation, the LHS and RHS are made from the same
3 matrices, multiplied in different orders). Namely, our LYBE will
have the following form (for {\em real\/} $x,y,\ldots$):
\begin{eqnarray}
\int L(x,y,u,v)\, M(u,z,p,w)\, N(v,w,q,r) \,{\rm d} u \wedge {\rm d} v \wedge {\rm d} w
\nonumber\\
=\int N'(y,z,g,h)\,M'(x,h,f,r)\,L'(f,g,p,q) \,{\rm d} f \wedge {\rm d} g \wedge {\rm d} h.
\label{eq int}
\end{eqnarray}
In the same way as the duality relation in~\cite{kor sai}, the
equality~(\ref{eq int}) will hold if we require that the relation
hold obtained from (\ref{eq int}) by {\em removing the integration
signs}, with the triples of variables $u,v,w$ and $f,g,h$
connected by some dependence. For such dependence, we will take
the transformation $R$ from formula~(\ref{eq R}). Then, the following
construction of functions $L,\ldots,N'$ can be proposed.
Take the relation
\begin{equation}
{x-y\over u-v}{z-u\over w-u} \,{\rm d} u \wedge {\rm d} v \wedge {\rm d} w =
{x-h\over f-h}{y-z\over g-h} \,{\rm d} f \wedge {\rm d} g \wedge {\rm d} h
\label{eq diff}
\end{equation}
(see (\ref{eq diff1}, \ref{eq diff2})), and also the relations
(\ref{eqnarray nachalo}--\ref{eqnarray konec}) raised in arbitrary
degrees (the relations (\ref{eqnarray nachalo}--\ref{eqnarray konec})
are not independent, so one of those degrees can be set to zero).
Then multiply separately the LHSs and RHSs of all so obtained relations
(including (\ref{eq diff})). The obtained LHS and RHS will be exactly
the integrands in (\ref{eq int}), and from them the multipliers
$L,\ldots,N'$ depending on proper quadruples of variables are easily
extracted.
I leave for further work the problem of possible choices of integration
domains in~(\ref{eq int}) and integral regularization (if needed).
The explicit form of functions $L,\ldots,N'$ will also be presented
elsewhere. Let me just note that we can also regard all
the variables $x,y,\ldots$ as {\em complex}. In such case, we should
multiply the integrands (including differentials) by their complex
conjugates and integrate over some domains of {\em six\/} real dimensions.
This will be the tetrahedral analog of Virasoro--Shapiro amplitude.
\section{Discussion}
The LYBE of the form (\ref{eq int}), as well as the duality
equations from~\cite{kor sai}, are interesting because there
exists a hope to construct from them interesting ``exactly solvable''
functional integrals, perhaps connected with 3-dimensional statistical
physics. By the way, here I have presented the tetrahedral analog of
{\em one\/} of two models in~\cite{kor sai}, and it seems fascinatingly
interesting to construct the analog of the other model (and their
generalizations). Very interesting will be also to clarify the relations
between pentagon and tetrahedron equations, where, despite the presence
of the excellent work~\cite{kash ser}, many things are unclear.
\subsection*{Acknowledgements}
I am grateful to Satoru Saito for valuable discussions during my stay
at Tokyo Metropolitan University, in the course of which the idea started
to revisit, from the modern integrability theory viewpoint, the algebraic
structures from which string theory was born in its time. I am also
grateful to Sergei Sergeev for many discussions on tetrahedron and
pentagon equations. Finally, I am glad to thank Russian Foundation
for Basic Research for its (mostly moral) support under
grant no.~98-01-00895.
|
\section{Introduction}
The dynamics of a scalar field $\theta({\bf x},t)$ advected in a turbulent
velocity field ${\bf u}({\bf x},t)$ is of practical relevance in many fields of
current research such as air pollution or chemical reactions in the stratosphere
in connection with the ozone hole \cite{Edo96}. Especially for problems
in atmospheric physics, models of two-dimensional turbulent flows give a good
approximation of the dynamical processes and are frequently
used\cite{Lil89,Les87}. More recently, two-dimensional turbulence has become
experimentally accessible in mercury layers\cite{Som86}, thin salt water
layers\cite{Tab91,JerTab97,Goll97,Car96}, and soap
films\cite{GhaDer89,Mar98,Riv98,Rut98}. Two-dimensional turbulence is also
interesting because of its fundamentally different behavior compared to the
three-dimensional case. Since the enstrophy is a second inviscid invariant
beside the energy two cascades develop: starting from a fixed, intermediate
injection scale, energy is transported to larger spatial scales in an inverse
energy cascade and to smaller ones in an enstrophy cascade\cite{Kra67,Bat69}.
The scaling behavior of a passive scalar in a turbulent fluid was analyzed
mainly in three dimensions where three different regimes could be identified.
Depending on the Reynolds number of the underlying fluid turbulence and the
ratio of the kinematic viscosity to the scalar diffusivity one distinguishes the
viscous-convective Batchelor regime\cite{Bat59}, the inertial-convective
regime\cite{Obu49,Cor51}, and the inertial-diffusive regime. In 2-d the
situation is more complicated, since already the velocity field shows a variety
of scaling regimes. In particular, the inverse cascade process gives rise to
the formation of large scale vortices that change on very slow time scales only
\cite{Ben87} and can dominate the
dynamics of the passive scalar, at least on intermediate time scales
\cite{Bab87,Bas94}. The formation of coherent vortices can be suppressed
by a large scale dissipation mechanism. If this additional dissipation
is present a statistically stationary homogeneous and isotropic
turbulent flow field develops,
that can be characterized by its structure function.
We assume that a
passive scalar in such a flow field also develops a statistically
stationary state which can be characterized by its own structure function.
The approach used to analyse the structure function of the passive scalar is
geometric measure theory \cite{Con91,ConPro93,ConPro94,ProCon93}. This powerful
method allows to connect the structure function of the passive scalar to that of
the underlying flow field and thus to link the statistical behavior of both.
The result are scale resolved bounds on the scaling behavior. Upper bounds are
easiest to derive and often give very good results, see e.g. the favorable
comparison between theory and numerical simulations in \cite{GroLoh94}. The
derivation of lower bounds is possible\cite{ConPro94} but much more difficult
and will not be attempted here. So assuming the reliability of the upper bounds
we would like to see how the different regimes in ${\bf u}$ are reflected in the
scaling properties of the scalar field passively advected by the flow. Some
aspects of the 2-d case have been discussed previously \cite{ProCon93}, see
below. In addition, we would like to compare the predictions to the results of
experiments of Cardoso {\it et al.} \cite{Car96}, where certain discrepancies
to theory were noted. As we will see the discrepancies can be accounted for if
the experimentally measured structure function is substituted for the velocity
field.
The model we consider is that of a scalar field $\theta({\bf x},t)$
transported in the turbulent flow field ${\bf u}({\bf x},t)$ according to
\begin{eqnarray}
\label{sceq}
\frac{\partial\theta}{\partial t} + ({\bf u}\cdot{\bf \nabla})\theta
=\kappa{\bf\nabla}^2\theta + f_{\theta}\;.
\end{eqnarray}
$\kappa$ denotes the diffusivity. The force density $f_{\theta}$ models
external boundary conditions and the driving and assures a statistically
stationary field $\theta({\bf x},t)$. The scalar $\theta$ is assumed to be {\it
passive}, i.e. it does not affect the dynamics and the statistical properties
of the velocity field. We assume that in the presence
of a large scale dissipation mechanism a
homogeneous, isotropic, and stationary
turbulent state develops.
The ratio of the kinematic viscosity $\nu$ to the scalar
diffusivity
$\kappa$ defines the Prandtl number $Pr=\nu/\kappa$ (this is the
nomenclature used when $\theta$ is a temperature field; if it describes a
concentration then the corresponding ratio is known as the Schmidt number).
The scaling exponents $\zeta^{(\theta)}_n$ of the $n$-th order scalar
structure functions, defined as
\begin{eqnarray}
\label{strucdef}
D^{(\theta)}_{n}(r)=\langle|\theta({\bf x}+{\bf r},t)-\theta({\bf
x},t)|^n\rangle \sim r^{\zeta^{(\theta)}_n}\;,
\end{eqnarray}
can be obtained from an analysis of the fractal dimension $\delta_g^{(d)}$ of
$d$-dimensional scalar field graphs; $\langle\cdot\rangle$ denotes the
statistical ensemble average. The fundamentals of the geometric measure theory
approach were laid out by Constantin {\it et al.}
\cite{Con91,ConPro93,ConPro94} who derived the fractal dimension
$\delta_g^{(d)}$ ($d$ is the space dimension). Closely related to the present
investigation is the application to two-dimensional chaotic surface
waves\cite{ProCon93}. The $Pr$ dependence of 3-d passive scalar advection
within this approach was discussed in \cite{GroLoh94}. As in that work we will
aim at a rather direct relation between scaling exponents and velocity structure
functions.
The outline of the paper is as follows. In Sec. \ref{sec_basic} the basic
concepts of the evaluation of the fractal graph dimension are summarized. The
results of the mean-field approach\cite{GroMer92} for fully developed
two-dimensional turbulence in the direct enstrophy cascade range -- the scaling
behavior of the 2nd order velocity structure function $D_{2}(r)= \langle|{\bf
u}({\bf x}+{\bf r},t)-{\bf u}({\bf x},t)|^{2}\rangle$ -- are recalled. In a
second step we interpolate the scaling of $D_{2}(r)$ to the inverse energy
cascade range, where no analytical result is known. We obtain $D_2(r)$ from
Fourier transform of an energy spectrum as is found in many numerical
simulations. In Sec. \ref{sec_resu} the fractal dimension of the passive
scalar graph is derived over a broad range of Prandtl numbers, both in the
enstrophy inertial subrange (ISR) and in the energy ISR with the previous
relations for the structure function. We conclude with a summary, a discussion
of the relation to the findings in the quasi-two-dimensional dispersion
experiments by Cardoso {\it et al.}\cite{Car96}, and some remarks on open
questions.
\section{Basic concepts}
\label{sec_basic}
\subsection{Fractal dimension of the passive scalar graph}
{}From now on all considerations are made for the case of a two-dimensional
flow field. The graph of the scalar field is then a 2-d surface in 3-d space.
The Hausdorff dimension of this graph is obtained from the scaling behavior of
the Hausdorff volume $H(G(B_r^{(2)}))$ of the graph $G(B_r^{(2)})=\{({\bf
x},\theta)|{\bf x}\in B_r^{(2)},\, \theta=\theta({\bf x})\}$ over a disk of
radius $r$ (the 2-d ball $B_r^{(2)}$) \cite{Fal85},
\begin{eqnarray}
\label{scal}
H(G(B_r^{(2)}))\sim r^{\delta_g^{(2)}}\;.
\end{eqnarray}
In two dimensions the fractal dimension $\delta_g^{(2)}$ is connected to the
scaling exponent $\zeta^{(\theta)}_1$, cf. eq. (\ref{strucdef}), through the
inequality\cite{ConPro93}
\begin{equation}
\label{ineq1}
\delta_g^{(2)}\le 3-\zeta^{(\theta)}_1\;.
\end{equation}
We assume equality in (\ref{ineq1})\cite{ConPro93,GroLoh94}
and use the relation
$\delta_g^{(1)}=\delta_g^{(2)}-1$, where $\delta_g^{(1)}$ is the fractal
dimension of the level sets $\theta_0=\theta({\bf x})$. The relative
Hausdorff volume $H(G(B_r^{(2)}))/V(B_r^{(2)})$ is given by geometric measure
theory\cite{Fed69,Mor88} as
\begin{eqnarray}
\label{vol}
\frac{H(G(B_r^{(2)}))}{V(B_r^{(2)})}&=&
\frac{1}{V(B_r^{(2)})}\int_{B_r^{(2)}}\,
\sqrt{1+r^2|{\bf\nabla}\tilde{\theta}|^2}\,\mbox{d}^2{\bf x}\;, \nonumber \\
&\le&\sqrt{1+\frac{1}{\pi} \int_{B_r^{(2)}}\,|{\bf\nabla}\tilde{\theta}|^2
\,\mbox{d}^2{\bf x}}\;,
\end{eqnarray}
where the Cauchy-Schwartz inequality and $V(B_r^{(2)})=\pi\,r^2$ were used in
the last line. The passive scalar field $\theta({\bf x},t)$ is measured in
units of $\theta_{r.m.s.}=\sqrt{\langle\theta^2\rangle}$, thus leading to
dimensionless $\tilde{\theta}=\theta/\theta_{r.m.s.}$. Equation (\ref{vol}) is
a generalization of the well-known volume formula
$V=\int\sqrt{g}\,\mbox{d}^2{\bf y}$ to fractal sets, where $V$ is a
two-dimensional curved hyper surface embedded in the three-dimensional Euclidean
space and $g$ the determinant of the metric tensor $g_{ij}(y^1,y^2)$.
We now turn to the evaluation of $\delta_g^{(2)}$. The term
$|{\bf\nabla}\tilde{\theta}|^2$ can be replaced by means of (\ref{sceq})
by
\begin{equation}
|{\bf\nabla}\tilde{\theta}|^2=\frac{1}{2\kappa}(\kappa
\Delta\tilde{\theta}^2-({\bf u}\cdot{\bf \nabla})\tilde{\theta}^2)+
\frac{f_{\theta}\tilde{\theta}}{\kappa\theta_{r.m.s.}}\;.
\end{equation}
With this eq.~(\ref{vol}) becomes
\end{multicols}
\begin{eqnarray}
\label{vol1}
\frac{H(G(B_r^{(2)}))}{V(B_r^{(2)})}
\le\sqrt{1+\frac{1}{\pi} \int_{B_r^{(2)}}\,
\left\{\frac{1}{2\kappa}[-({\bf u}\cdot{\bf \nabla})\tilde{\theta}^2
+\kappa\Delta\tilde{\theta}^2]+
\frac{f_{\theta}\tilde{\theta}}{\kappa\theta_{r.m.s.}}\right\}
\,\mbox{d}^2{\bf x}}\;.
\end{eqnarray}
\begin{multicols}{2}
We will consider the three integrals under the square root separately
and denote them by $I_1$, $I_2$, and
$I_3$, respectively. In the three-dimensional case\cite{GroLoh94} the terms
$I_2$ and $I_3$ vanish in the large Reynolds number limit. They also satisfy
the inequality $I_2\le3\sqrt{I_3}$ which changes to $I_2\le2\sqrt{I_3}$ in the
two-dimensional case. $I_3$ can be estimated as
\begin{eqnarray}
\label{i3eq}
I_3=\frac{1}{\pi}\int_{B_r^{(2)}}\,\frac{f_{\theta}\tilde{\theta}}
{\kappa\theta_{r.m.s.}}\,\mbox{d}^2{\bf x}
=\frac{r^2\epsilon_{\theta}}{\kappa\theta_{r.m.s.}^2}
=\frac{\epsilon_{\theta}\epsilon_{\omega}^{-1/3}}{\theta_{r.m.s.}^2}
Pr\,\tilde{r}^2\;,
\end{eqnarray}
where the scalar dissipation rate
$\epsilon_{\theta}=\kappa\langle|{\bf\nabla}\theta|^2 \rangle$, the enstrophy
dissipation rate
$\epsilon_{\omega}=\nu\langle|{\boldmath\nabla}\omega|^2\rangle$, and
stationarity are used. In the case of a three-dimensional passive
scalar this term contains a factor $\nu^{1/2}$ and thus can be neglected.
In 2-d the smallest scales are given by the enstrophy dissipation rate
and this factor disappears. Hence $I_3$ cannot be neglected;
its importance is evidently controlled by Prandtl number $Pr$,
length scale $r$ and dimensionless prefactor
\begin{equation}
\alpha = \frac{\epsilon_{\theta}\epsilon_{\omega}^{-1/3}}
{\theta_{r.m.s.}^2} \,.
\label{alp_def}
\end{equation}
The term $I_2$ can still be neglected on
account of its subdominant scaling in $r$. We introduce dimensionless length
scales $\tilde{r}=r/\eta_{\omega}$ by means of the enstrophy dissipation length $\eta_{\omega}=\nu^{1/2}
\epsilon_{\omega}^{-1/6}$ since in 2-d turbulence it is the enstrophy cascade
that brings the energy to the smallest scales where viscosity dominates.
It follows from (\ref{vol1}) for $I_1$ by applying the Gauss Theorem and the
Cauchy-Schwarz inequality
\begin{eqnarray}
I_1&=& \frac{r}{\kappa}\oint_{\partial B_r^{(2)}}
\frac{\tilde{\theta}^2
({\bf u}-{\bf u}_0)\cdot{\bf n}}{u_r}\,\mbox{d}\,r\;,
\nonumber \\
&\le&\frac{r}{\kappa}
\sqrt{\oint_{\partial B_r^{(2)}}\frac{\tilde{\theta}^4}{u_r}\,\mbox{d}\,r}
\;
\sqrt{\oint_{\partial B_r^{(2)}}\frac{
(({\bf u}-{\bf u}_0)\cdot{\bf n})^2
}{u_r}\,\mbox{d}\,r}\;.
\end{eqnarray}
The quantity $u_r=2\pi r$ is the circumference. It is possible to add ${\bf
u}_0$, the velocity at the center of $B_r^{(2)}$, due to the assumed
homogeneity.
The first term on the right hand side contains the square root of the
passive scalar flatness. Since we are interested in the scaling properties
of $I_1$, it suffices to know that the scalar flatness is
a constant, independent of $r$. However, there do not seem to
be numerical or experimental data for the passive scalar flatness in 2-d.
Data for the velocity field from the experiments
\cite{JerTab97} and the numerical simulations \cite{SmiYak93}
suggest Gaussian behavior in the absence of coherent structures
in the regime of the inverse cascade.
More recent experiments suggest that this result also
extends into the region of the direct enstrophy cascade \cite{Tab99}.
However, since there are models where a Gaussian statistics for a random
velocity field causes non-Gaussian
scalar statistics \cite{Sigg94,Krai94}, this information is insufficient
to infer Gaussian statistics for the passive scalar. In the following
we will work with the Gaussian flatness value of three
for the passive scalar. It should be kept in mind that
deviations from this value will most
likely be scale dependent and will give rise to modifications
of the scaling exponents.
The second term is the longitudinal velocity structure
function $D_{\parallel}(r)$. Thus we find
\begin{equation}
\label{i1in}
I_1\le\frac{\sqrt{3}}{\kappa}r\sqrt{D_{\parallel}(r)}\;.
\end{equation}
Combining (\ref{scal}), (\ref{vol1}), (\ref{i3eq}), and
(\ref{i1in}) we end up with an inequality for
the fractal dimension $\delta^{(2)}_g$ of the passive scalar graph in two
dimensions,
\begin{eqnarray}
\label{fracdim}
\delta^{(2)}_g-2\le\frac{\mbox{d}}{\mbox{d}\,\ln\tilde{r}}
\ln\sqrt{1+\frac{\epsilon_{\theta}\epsilon_{\omega}^{-1/3}}{\theta_{r.m.s.}^2}
Pr\,\tilde{r}^2+\sqrt{3} Pr \tilde{r}\sqrt{\tilde{D_{\parallel}}}}
\;,
\end{eqnarray}
where $\tilde{D}_{\parallel}=D_{\parallel}/(\epsilon^{2/3}_{\omega}
\eta_{\omega}^{2})$. This inequality, relating the scaling exponent
$\delta^{(2)}_g$ to the longitudinal structure function of the underlying
turbulent flow field $\tilde{D_{\parallel}}$ is the main result of this section.
For most of the discussion that follows we will assume equality in
(\ref{fracdim}); in the three-dimensional case this is a very good assumption
\cite{GroLoh94}.
\subsection{Structure functions in two-dimensional turbulence}
To evaluate (\ref{fracdim}) we need information on the scaling behavior of the
2nd order longitudinal structure function $D_{\parallel}$. The longitudinal
structure function $D_{\parallel}(r)$
and transversal structure function $D_\perp(r)$ make
up the velocity structure function $D_2(r)$ and are connected by
incompressibility, $D_{\perp}=D_{\parallel}+r\frac{dD_{\parallel}}{dr}$.
Eliminating the transversal part then gives \cite{MonYag75,Lan87}
\begin{eqnarray}
\label{dlon}
D_{\parallel}(r)=\frac{1}{r^2}\int_0^r \rho\,D_2(\rho)\,\mbox{d}\,\rho\;.
\end{eqnarray}
As there are two
inertial ranges with several different scaling regimes, there is no analytical
expression for the structure function. As far as we are aware, the best that
can be achieved analytically is the structure function for the enstrophy cascade
as discussed by Grossmann and Mertens\cite{GroMer92}. They used a mean field
type approach for the fully developed, turbulent velocity field in the enstrophy
cascade, i.e. for spatial scales $\eta_{\omega}<r<r_{in}$. Separating small and
large scales one finds energy and enstrophy balance equations where terms
resulting from the small scale fluctuations act like an effective eddy viscosity
for the large scale components of $\omega$. Analytical expressions for the 2nd
order vorticity structure function $D^{(\omega)}_2(r)$ and the 2nd order
velocity structure function $D_2(r)$ can be found using the Batchelor
interpolation technique\cite{GroMer92,Bat51}. In dimensionless form they read
\end{multicols}
\begin{eqnarray}
\label{mergro1}
\tilde{D}_2(\tilde{r})&=&\frac{\tilde{D}_2^{(\omega)}(\infty)}{4}\,
\frac{\tilde{r}^{2}}{(1+a\,\tilde{r}^{2})^{1/3}}+
\left(\frac{Re^{\ast}}{2}-
\frac{\tilde{D}_2^{(\omega)}(\infty)}{4}\right)\,\tilde{r}^{2}
\;,
\end{eqnarray}
\begin{multicols}{2}
with the parameter $a=\frac{15}{592}$ and the asymptotic value
$\tilde{D}^{(\omega)}_2(\infty)=
D^{(\omega)}_2(\infty)/\epsilon_{\omega}^{2/3}=14.8$. This spectrum also
depends on the energy dissipation $\epsilon$, which when expressed in the length
and energy scales of the enstrophy cascade becomes the dimensionless parameter
$Re^{\ast}=\epsilon/(\epsilon_{\omega}^{2/3}\nu)$. The structure functions are
shown in Fig.~1. Besides the prominent $\tilde{r}^2$ behavior that follows
already by dimensional analysis one notes an intermediate scaling with
$\tilde{r}^{4/3}$; the range over which this scaling is observed depends on
$Re^\ast$ (see below). The corresponding longitudinal velocity structure
function $\tilde{D}_{\parallel}(\tilde{r})$ is given with eq. (\ref{dlon}) by
\begin{eqnarray}
\label{mergro2}
\tilde{D}_{\parallel}(\tilde{r})&=&
\left(\frac{Re^{\ast}}{8}-
\frac{\tilde{D}_2^{(\omega)}(\infty)}{16}\right)\,\tilde{r}^{2}+
\frac{3\tilde{D}_2^{(\omega)}(\infty)}{8 a}\nonumber\\
&\times&\left[
\frac{(1+a\,\tilde{r}^{2})^{5/3}-1}{5\,a\,\tilde{r}^{2}}-
\frac{(1+a\,\tilde{r}^{2})^{2/3}-1}{2\,a\,\tilde{r}^{2}}
\right]\;.
\end{eqnarray}
For the energy ISR no such analytical expression is known. We therefore combine
a model for the energy distribution in $k$-space with numerical transformations
to obtain the longitudinal structure function. Recent experiments on forced
two-dimensional turbulence\cite{JerTab97,Rut98}, a number of direct numerical
simulations\cite{SmiYak93,FriSul84,Ben86,Bor93,KevFar97,Bab97}, field theoretical
investigations\cite{Fal94} as well as cascade
models\cite{Schu94} support the existence of a Kolmogorov-like scaling for the
energy spectrum, $E(k)\sim\,k^{-5/3}$ for $(k<k_f)$, in the energy ISR and
$E(k)\sim\,k^{-\beta}$ with $\beta\ge3$ for $(k>k_f)$ for the enstrophy ISR. We
therefore start with the following model spectrum for the amplitudes
$\langle|{\bf u}_{\bf k}|^2\rangle$ of the velocity field in a Fourier
representation in a periodic box of size $L=2\pi$
\end{multicols}
\begin{eqnarray}
\langle|{\bf u}_{\bf k}|^2\rangle\!\sim\!
\left\{
\begin{array}{r@{\quad:\quad}l}
k^3 & \frac{2\pi}{L}\le k\le k_1, \\
k^{-2/3} & k_1<k\le k_f,\\
k^{-\beta} & k_f<k\le k_{\omega}=\frac{1}{\eta_{\omega}},
\;\beta\ge 2\, ,\\
k^{-\beta}\exp\left[-\left(\frac{k-k_{\omega}}{k_{\omega}}\right)^{2}\right]
& k_{\omega}<k\;.
\end{array}
\right.
\label{modspec}
\end{eqnarray}
\begin{multicols}{2}
Note the different scalings for $\langle|{\bf u}_{\bf k}|^2\rangle$ and the
energy spectrum $E(k)$ due to phase space factor, i.e.
$E(k)\!\sim\!k^{-\beta-1}$
corresponds to $\langle|{\bf u}_{\bf k}|^2\rangle\!\sim\!k^{-\beta}$.
The first range approximates finite system size effects where we have chosen a
slope of 3 in correspondence with results of numerical
experiments\cite{FriSul84,Bab97}. This is followed by the inverse energy cascade
range with a Kolmogorov-like scaling law. At the injection scale $k_f$ the
enstrophy cascade to larger values of $k$ starts, followed by the viscous
cutoff. The energy spectra with $\beta=3$ for three different values of the
injection wavenumber $k_f$ are shown in Fig. \ref{p5}.
The relation between velocity spectrum scaling and the velocity structure
function $D_2(r)$ assuming stationarity, homogeneity, and isotropy is given
by the volume average
\begin{eqnarray}
D_2(r)&=&\frac{1}{V}\int_{V}\,|\,{\bf u}({\bf x}+{\bf r})-{\bf u}({\bf x})\,|
^{2}\,\mbox{d}\!V,\nonumber \\
&=&\frac{1}{V}\int_{V}\,|\,\sum_{{\bf k}}{\bf u}_{\bf k}\,
\exp(i{\bf k}\cdot{\bf x})
[\exp(i{\bf k}\cdot{\bf r})-1]\,|^{2}\,\mbox{d}\!V,\nonumber \\
&=&2\,\sum_{{\bf k}}\,\langle|{\bf u}_{\bf k}|^{2}\rangle\,
(1-\cos{({\bf k}\cdot{\bf r})})\, .
\end{eqnarray}
By averaging over all directions (due to isotropy) in ${\bf k}$-space
the cosine gives rise to the Bessel function $\mbox{J}_0(kr)$,
\begin{equation}
\label{mod_struc}
D_2(r)=2\,\sum_k\,\langle|{\bf u}_{\bf k}|^{2}\rangle\,
(1-\mbox{J}_{0}(kr)) \,.
\end{equation}
The model spectrum (\ref{modspec}) is then substituted and the summation in
(\ref{mod_struc}) is evaluated numerically using a finite, geometrically
scaling set
of wave numbers. It should be mentioned here that the model does
not contain a spectral range
that would correspond to the intermediate $\tilde{r}^{4/3}$--scaling of
the structure function in the enstrophy ISR found in the analytical theory.
We will come back to this point in the discussion of our results.
\section{Results}
\label{sec_resu}
\subsection{Fractal dimension in the enstrophy ISR}
We first calculate the scaling behavior in the enstrophy ISR where the analytical
expression (\ref{mergro1}) is available. Inserting (\ref{mergro1}) in
(\ref{fracdim}) and neglecting the term $I_3$ for the moment, one notes that
$\delta_g^{(2)}$ depends on three quantities: the parameter $Re^{\ast}$, the
Prandtl number $Pr$, and the scale $\tilde{r}$ itself. The numerical result for
$\delta_g^{(1)}=\delta_g^{(2)}-1$ are shown in Fig. \ref{p2} for a Prandtl
number range varying over ten orders of magnitude and $Re^{\ast}=7.6$. The grey
shaded area denotes the range of scales where $\tilde{r}^{4/3}$ gives the main
contribution to the structure function. It is only in this range that we find
$1<\delta_g^{(1)}<2$. The range is bounded by
$\tilde{r}_1\le\tilde{r}\le\tilde{r}_2$ where $\tilde{r}_1$ is the crossover
scale from the viscous subrange (VSR) and $\tilde{r}_2$ is the crossover scale
to the $\tilde{r}^2$--scaling in the enstrophy ISR,
\begin{eqnarray}
\label{scaran}
\tilde{r}_1 &=&\frac{1}{\sqrt{3}}
(\tilde{D}_2^{(\omega)}(\infty))^2(Re^{\ast})^{-3/2}\;,
\nonumber \\
\tilde{r}_2 &=&\frac{1}{\sqrt{3}}
(\tilde{D}_2^{(\omega)}(\infty))^2(Re^{\ast}-
\tilde{D}_2^{(\omega)}(\infty)/2)^{-3/2}\;,
\end{eqnarray}
where $\tilde{D}_2^{(\omega)}(\infty)=14.8$ has to be taken. The larger
$Re^{\ast}$ the smaller the range of the $\tilde{r}^{4/3}$--scaling. It can be
observed only for $Re^{\ast}$ within the interval
\begin{eqnarray}
\label{rebounds}
7.4\!\approx\!\frac{\tilde{D}_2^{(\omega)}(\infty)}{2}
\le Re^{\ast}\le
\left(\frac{(\tilde{D}_2^{(\omega)}(\infty))^4}{3}\right)^{1/3}\!\approx\!25\;.
\end{eqnarray}
The lower bound follows from the positivity of the structure function by its
definition (cf. second term of (\ref{mergro1})). The upper bound is a result
of eq. (\ref{scaran}) and the constraint $\tilde{r}_1\ge1$. For $Re^{\ast}$
approaching 7.4 follows $\tilde{r}_2$ goes to infinity. The $\tilde{r}^{4/3}$--
scaling range is then extended over the whole enstrophy ISR. We see in Fig.
\ref{p3} that for increasing $Re^{\ast}$ the intermediate fractal scaling of the
graph is more and more suppressed and conclude that this behavior of
$\delta^{(1)}_g$ is due to the presence of the $\tilde{r}^{4/3}$--scaling range.
The above estimates give $\tilde{r}_1\!\approx\!6.0$ and
$\tilde{r}_2\!\approx\!1400$ for $Re^{\ast}=7.6$ and
$\tilde{r}_1\!\approx\!1.0$ and
$\tilde{r}_2\!\approx\!1.7$ for $Re^{\ast}=25.0$, respectively.
In the lower panel the corresponding
scaling exponent of the scalar structure function
$\zeta^{(\theta)}_1=2-\delta^{(1)}_g$ is plotted. The plateau of the structure
function $D_1^{(\theta)}$ for large Prandtl number and scales below the smallest
scales in the turbulent fluid ($r/\eta_\omega<1$) corresponds to the Batchelor
regime of chaotic scalar advection in a smooth fluid\cite{Bat59}.
For {\it small values of $Pr$} the diffusion $\kappa$ dominates the passive
scalar dynamics. The scalar field is smooth, $\delta^{(1)}_g=1$. The exponent
$\delta^{(1)}_g$ grows when the second term in the square root of eq.
(\ref{fracdim}) becomes dominant. By inserting the power law
$\tilde{D}_{\parallel}=\frac{\sqrt[3]{9}}{20}
(\tilde{D}_2^{(\omega)}(\infty)\tilde{r})^{4/3}$ for the enstrophy ISR at
$\tilde{r}=\tilde{r}_c$ one gets a crossover for
\begin{eqnarray}
\tilde{r}_c=\sqrt[10]{\frac{8000}{243}}Pr^{-3/5}(\tilde{D}_2^{(\omega)}
(\infty))^{-2/5}
\approx 0.48\, Pr^{-3/5}.
\label{rcsmooth}
\end{eqnarray}
By putting $\tilde{r}_c=\tilde{r}_2$ and using (\ref{scaran})
the maximum Prandtl number $Pr_s$ without fractal $\delta^{(1)}_g$ can
be estimated as
\begin{eqnarray}
Pr_s<2\sqrt{5}(\tilde{D}_2^{(\omega)}(\infty))^{-4}
(Re^{\ast}-\tilde{D}_2^{(\omega)}(\infty)/2)^{5/2}\;.
\end{eqnarray}
With $Re^{\ast}=7.6$ and $25.0$ this gives
$Pr_s\lesssim 2\cdot 10^{-6}$ and $10^{-1}$, respectively.
For {\it large values of $Pr$} one observes a transition to $\delta^{(1)}_g=2$
even when the velocity field is in the VSR. Again the second term of
(\ref{fracdim}) dominates because of its large prefactor $Pr$. Taking
$\tilde{D}_{\parallel}=\frac{Re^{\ast}}{8}\tilde{r}^2$ for the VSR gives
\begin{eqnarray}
\tilde{r}_c=\sqrt[4]{\frac{8}{3}}Pr^{-1/2}(Re^{\ast})^{-1/4}\;.
\end{eqnarray}
With $\tilde{r}_c=\frac{1}{10}\tilde{r}_1$ we get those $Pr_l$ which give
$\delta^{(1)}_g=2$ in the VSR over at least one decade of scales,
\begin{eqnarray}
Pr_l>200\sqrt{6}(\tilde{D}_2^{(\omega)}(\infty))^{-4}(Re^{\ast})^{5/2}\;.
\end{eqnarray}
For $Re^{\ast}=7.6$ and $25.0$ this results in $Pr_l\!\gtrsim\!2.0$ and $30.0$,
respectively.
The structure function of a passive scalar in the enstrophy ISR shows four
different regimes. For very small $\tilde{r}$ smoothness gives
$\delta_g^{(1)}=1$. This is followed by the Batchelor regime $\delta_g^{(1)}=2$
for sufficiently large $Pr$. The $\tilde{r}^{4/3}$--scaling discovered by
Grossmann and Mertens is reflected in a decrease of $\delta_g^{(1)}$ below $2$
near $r/\eta_\omega\approx 10^{1\pm1}$. For larger $\tilde{r}$ it goes back up
to $2$.
So far we neglected the term $I_3=\alpha \,Pr\,\tilde{r}^2$
(see (\ref{i3eq}) and (\ref{alp_def})) in our calculation.
Because of its $\tilde{r}^2$-scaling it dominates the structure
function for large $\tilde{r}$.
In \cite{ProCon93} this term was assumed to be subdominant.
Substituting the various
definitions it can be expressed as a ratio of two rates,
\begin{eqnarray}
\label{taylor}
\alpha=\frac{\kappa \langle|{\bf\nabla}\theta|^2 \rangle}
{(\nu \langle|{\bf\nabla}\omega|^2 \rangle)^{1/3}
\langle\theta^2 \rangle}
=\frac{r_{\theta}}{r_{\omega}}\,.
\end{eqnarray}
The rate $r_{\theta}=\epsilon_{\theta}/\theta^2_{r.m.s.}$ is
a scalar forcing rate. $r_{\omega}=\epsilon_{\omega}^{1/3}$
is the strain rate in the enstrophy cascade and characteristic of
the passive scalar advection by the vortices.
The case $\alpha>1$ then corresponds to $r_\theta>r_\omega$,
i.e. fast driving and slow advection. Then the scalar field
fills space and $\delta_g^{(1)}\sim 2$. In the other case,
$\alpha<1$, the advection dominates and the structure function
of the fluid is reflected in that of the scalar.
It is this latter case that was discussed in \cite{ProCon93}
for surface waves.
The size of $\alpha$ is determined by the experimental
situation and has to be taken from
measurements. All quantities that enter (\ref{taylor}) are experimentally
accessible; note that the enstrophy dissipation rate is related to velocity
gradients via $\epsilon_{\omega}=-8\langle(\partial_x
u_x)^3\rangle$\cite{GroMer92}.
Results for different $Pr$ with $\alpha=1$ are shown
in Fig.~\ref{p4}. The main effect
of an increasing $I_3$ is the suppression of the crossover scaling
and a transition for large $r$.
\subsection{Extension to the energy ISR}
The extension of $D_2(r)$ to the whole range of scales is done with
(\ref{mod_struc}) and the results for $\delta_g^{(1)}$ are given in
Fig.~\ref{p6} for three input model spectra (see Fig. \ref{p5}) which differ by
the injection wavenumber $k_f$. The smaller $k_f$ the longer is the enstrophy
ISR extended which results in a dominant range where $\delta_g^{(1)}=2$. On the
other hand, the larger $k_f$ the more dominant the inverse energy cascade range,
indicated as the grey shaded area in Fig. \ref{p6}. The corresponding
longitudinal velocity structure function $D_{\parallel}(r)$ is superimposed.
Note that the model spectrum has to be normalized to give
$\tilde{D}_{\parallel}=\frac{Re^{\ast}}{8}\tilde{r}^2$ in the VSR. In the
enstrophy ISR we find $\tilde{D}_{\parallel}(r)\sim \tilde{r}^2$ and in the
energy ISR $\tilde{D}_{\parallel}(r)\sim \tilde{r}^{2/3}$, leading to
$\delta_g^{(1)}=2$ and $\delta_g^{(1)}=\frac{5}{3}$, respectively.
As mentioned, the model spectrum does not show the
$\tilde{r}^{4/3}$--scaling predicted by \cite{GroMer92}. Therefore,
if $Re^{\ast}$ is in the range where a $\tilde{r}^{4/3}$--scaling appears
the $\delta_g^{(1)}$ values for $r\simeq\eta_{\omega}$ have to be replaced by
the ones in Figs. \ref{p2}, \ref{p3}, and \ref{p4}.
For very large values of $r$ we can replace $\mbox{J}_{0}(kr)$ by its asymptotic
form $\mbox{J}_{0}(kr)\approx\sqrt{\frac{2}{\pi kr}} \cos(kr-\frac{\pi}{4})$
resulting in $D_2(r)\approx 2\,\sum_k\,\langle|{\bf u}_{\bf k}|^{2}\rangle$ in
(\ref{mod_struc}). The constant asymptotic behavior of the structure function
corresponds with $\delta_g^{(1)}=\frac{3}{2}$ (cf. (\ref{fracdim})).
The model spectrum contains a free parameter $\beta$ which has no
agreed upon value. Numerical simulations \cite{Ben86,Bor93,KevFar97,Bab97}
suggest a range $\beta\in[2,4]$.
For $\beta=2$ we get
$\delta_g^{(1)}$ slightly below 2 in the enstrophy ISR which changes clearly to
$\delta_g^{(1)}=2$ for $\beta>2$ (cf. Fig. \ref{p7}). As expected, the value
of $\delta_g^{(1)}$ in the energy ISR is insensitive to a $\beta$--variation.
Again we have to discuss the additional influence of the $I_3$ term in
(\ref{vol1}). Will inverse cascade effects be suppressed in the large $Pr$
number case because of the dominance of
the $\tilde{r}^2$--scaling at large separations? In order to determine
the scale $\tilde{r}_c$ where $I_3\ge I_1$, we use the experimental
value for the Kolmogorov
constant $C_K$ \cite{JerTab97} and
assume a completely extended inverse cascade with no intermittency corrections.
Then
$D_2(r)=4\,C_K\,\epsilon^{2/3} \int^{\infty}_0(1-\mbox{J}_{0}(kr))k^{-5/3}\,
\mbox{d}\,k$ and $\tilde{D_2}=b_2\tilde{r}^{2/3}$. With
$C_K$ between 5.5 and 7, we find
$b_2$ between 31.5 and 40 for the energy ISR and thus finally
\begin{eqnarray}
\tilde{r}_{\alpha}\ge\left(\frac{9b_2}{8\alpha^2}\right)^{3/4}
\approx l\alpha^{-3/2}\;,
\label{ralength}
\end{eqnarray}
where $l$ lies between 14 and 17. The scale $\tilde{r}_{\alpha}$ is shifted
towards larger values for decreasing $\alpha$. A factor $\alpha\!\sim\!1$ can
suppress the scaling behavior in the energy ISR which was found above completely.
This fact is illustrated in Fig. \ref{p8}. Clearly the asymptotic state for
$\tilde{r}$ to infinity leads here to $\delta_g^{(1)}$ approaching 2.
\section{DISCUSSION}
\label{sec_sum}
Our main findings for a passive scalar in a
2-d turbulent flow field can be summarized as follows:
(1) There is a critical scale set by equation (\ref{rcsmooth}) below which the spectrum
is smooth, $\delta_g^{(1)}=1$, because of diffusion dominance.
(2) Between this scale and the injection scale $r_{in}$ the scaling
exponent $\delta_g^{(1)}=2$ in most cases.
(3) An exception is found for $Re^*$ in the interval set by (\ref{rebounds}),
where a scaling exponent $\delta_g^{(1)}<2$ is found. The limits of this
interval are given by (\ref{scaran}) and the deviation from 2 is controlled
by the parameter $\alpha$, eq. (\ref{taylor}).
(4) Beyond the injection length and up to a length set by
equation (\ref{ralength}),
the scalar field scales with the exponent $\delta_g^{(1)}=5/3$
as expected for the energy inertial subrange.
(5) Above the length scale set by (\ref{ralength}), the exponent again increases to
$2$.
What is most surprising is that the scaling derived within
geometric measure theory depends not only on the scaling
of the velocity field but also on two additional dimensionless numbers,
the Reynolds number $Re^*$ which causes the intermediate scaling
in the enstrophy viscous subrange and on $\alpha$ which suppresses
the velocity field induced scaling at large separations for
rapid driving.
At this point input from experiments on two-dimensional turbulence is
necessary to check and expand the theoretical results.
Cardoso {\it et al.} \cite{Car96} measured dispersion in a
quasi-two-dimensional turbulent flow and compared with results
for the energy inertial subrange. They observed a velocity
structure function with scaling $\tilde{r}^0$ and a fractal
dimension $\delta_g^{(1)}$ between $1.3$ and $1.5$ with an average of
about $1.4$. Substituting a velocity scaling function
$\tilde{D}_{\parallel}=C\tilde{r}^0$ in our main equation
(\ref{fracdim}) gives
\begin{eqnarray}
\delta^{(1)}_g\le 1+\frac{\mbox{d}}{\mbox{d}\,\ln\tilde{r}}
\ln\sqrt{1+5000(\alpha \tilde{r}^2+ \sqrt{3 C} \tilde{r})}
\;.
\end{eqnarray}
If the quadratic term can be neglected, i.e. if $\tilde{r}$ is small
enough, the inequality reads $\delta^{(1)}_g\le 3/2$. The experimental
results are indeed below but close to this limit, so that the
assumption that the distances are small is probably reasonable.
For larger separation there is a crossover to $\delta^{(1)}_g\le 2$,
and it would be interesting to see whether the experimental data
follow this behavior. For the energy inertial subrange
[and not too large separations, see (\ref{ralength})], the inequality
would be $\delta^{(1)}_g\le 5/3$, higher than the one for the
experimentally observed spectrum.
Further experiments or numerical studies to check the results from geometric
measure theory, especially the ones for the enstrophy cascade and for the
dependence on $\alpha$, are clearly needed. Perhaps it is possible to combine
the experiments on passive scalar mixing \cite{Car96,Goll97}
with the set-up for
extended, stationary inverse and direct cascades \cite{JerTab97,Tab99} in order
to measure the scaling behavior mentioned in (2). In order to check the
predictions for the enstrophy cascade in (1) the spatial resolution has to be
enlarged. Otherwise e.g. the existence of the intermediate
$\tilde{r}^{4/3}$--scaling of $\tilde{D}_2(\tilde{r})$ cannot be detected.
We remind the reader that this range is only well established for values of
$Re^{\ast}$ close to its lower threshold (see Fig. \ref{p1}).
Its localization with respect to $\tilde{r}$ prevents it from being seen
in the Fourier spectrum, as already discussed by Grossmann and
Mertens \cite{GroMer92}.
Another open question which calls for more input from numerical simulations
and experiments is that of the scalar flatness in 2-d. For a non-Gaussian scalar
statistics we would expect a scale--dependent flatness $F_{\theta}(\tilde{r})$
causing a further scale dependence of the third term in (\ref{fracdim}) and
thus leading to a modification of the present model.
The problem studied here has also interesting links to magnetohydrodynamics.
First steps towards using geometric measure theory in this context were
undertaken by Grauer and Marliani\cite{GraMar95}. In two dimensions there is a
direct relation between magnetic field advection and the scalar dynamics studied
here since the vector potential for the magnetic field has only a $z$-component.
Consequences of this relation are under investigation.
|
\section{Introduction}
\label{sec:introduction}
The Sunyaev-Zel'dovich (SZ) effect has been predicted for several
decades (Sunyaev and Zel'dovich 1972), but it has only recently become
a valuable observational tool. In the past few years, we have seen
the observational programs go from ones aimed at detecting the effect
to using it to map out massive clusters of galaxies
(e.g., Birkinshaw 1999). In many ways, this
is similar to the rapid evolution of CMB anisotropy experiments.
The SZ effect arises from hot gas between us and the surface of last
scattering distorting the thermal spectrum of the CMB. This could
describe a large variety of situations, but the SZ effect is usually
discussed in the context of the effects of low redshift (below $z\sim 10$)
gas that is heated mainly by gravitational infall in the course of structure
formation. That will be the focus of this paper as well.
There are two SZ effects which are discussed. The first is the so-called
thermal effect arising from Compton scattering, while the second is
known as the kinetic effect, arising from simple
Thomson scattering by a moving object.
The cool CMB photons traversing hot gas can undergo Compton scattering
from the hot electrons, gaining a small amount of energy in the exchange.
A typical fractional energy gain by the photons,
in the non-relativistic limit of low electron and photon
energies would be roughly ${kT_{gas} \over m_e c^2}$ per collision.
The cross-section for collision is simply the Thomson cross-section,
$\sigma_T$, leading to an optical depth for a gas of electron number
density $n_e$ and characteristic size $L$ of $\tau \sim n_e \sigma_T L$.
Thus, the CMB photons as a whole should get a fractional energy gain
roughly given by the Compton $y$ parameter, defined as the fractional
energy gain multiplied by the optical depth to scattering.
This scattering conserves photon number, so the net effect is to shift
the CMB spectrum up in energy, distorting the blackbody curve slightly.
This leads to a decrement at low frequencies, as more photons have been
scattered out of those energy bins than have been scattered in,
while there is an increment at high frequencies.
The exact spectral form thus depends on the slope of the blackbody
curve. The crossover
frequency is near $217$ GHz, with small relativistic corrections
(Rephaeli 1995) making the crossover frequency a weak function
of the gas temperature.
This is a distinctly non-thermal spectrum, and CMB experiments with
good frequency coverage will be able to easily distinguish this
from purely thermal fluctuations.
The spectral behavior is shown in Figure \ref{spectrum}. In particular,
note that the spectral behavior is slowly varying over the entire MAP
CMB satellite experiment
frequency coverage (22-90 GHz), and thus could be a difficult
foreground to remove. On the other hand, the PLANCK satellite will
have coverage from 30-850 GHz, allowing good sampling of the peculiar
spectral behavior of the thermal SZ effect.
\begin{figure}
\epsfysize=3.7 in
\centerline{\epsfbox{spectrum.ps}}
\caption{Spectral behavior of thermal SZ effect, where ${\Delta T \over
T_{CMB}}= f(\nu)y$,
as a function of frequency, in the non-relativistic limit.
Note that the high frequency area is far out in the Wien part
of the CMB blackbody curve, and therefore there is very little
signal to use to detect a temperature difference.}
\label{spectrum}
\end{figure}
The kinetic effect arises from a moving mass of hot gas between us and
the (mis-named in this case) surface of last scattering. The moving object
sees a significant dipole in the CMB, with a hot spot in the direction
of its motion. Thomson scattering will remove this dipole from the
photons that undergo scattering. Back in the rest frame of the CMB, the
gas will have created an apparent hot spot for observers that see the
gas moving toward them and a cool spot for observers with the gas moving
away from them.
The kinetic effect does not cause
a deviation from a blackbody spectrum; it rather leads to a blackbody
with a slightly different temperature. This means that the kinetic SZ
effect has exactly the same spectral behavior as the primary CMB anisotropy.
Thus, the kinetic effect is a foreground which is impossible to remove
by spectral methods and it is important to understand its expected
contributions to CMB anisotropy. Likewise, the primary CMB anisotropy
can be a major source of confusion for determining peculiar velocities
of clusters using the kinetic SZ effect.
The kinetic effect involves an energy shift per scattering which is
effectively just a redshift, and thus on the order
of $(v_{bulk}/c)$. Bulk peculiar velocities of clusters are expected
to be less than $\sim 1000$ ${\rm km\ s^{-1}}$ in any cosmology.
The thermal effect depends on the temperature of the electron gas,
with an energy shift per scattering on the order of ${kT_e \over m_e
c^2}$.
Assuming that the electrons
and ions share the same temperature (Fox and Loeb 1997),
the thermal energy in the gas is typically on the order
$m_H v_\sigma^2$, where $v_\sigma$ is the internal velocity dispersion
in the potential well of the cluster, which can be $\sim1000$ ${\rm km\ s^{-1}}$.
At most frequencies, the
thermal effect will be larger than the kinetic by a factor of ten or
more.
The peculiar spectral behavior
of the thermal effect means that the relative effects of the kinetic
and thermal effects change with frequency, with the only SZ signal
at the null of the thermal effect coming from the kinetic effect. It
has been suggested that the peculiar velocity field
of clusters could be mapped out in this way,
but the small magnitude of the kinetic SZ effect makes this
a very difficult experiment. Upper limits have been
placed on the peculiar velocities of clusters (Holzapfel {\em et~al.}\ 1997a),
but a clear detection of the kinetic effect has not yet been
observed.
\section{Clusters of Galaxies}
\label{sec:clusters}
Clusters of galaxies are excellent sources of the SZ
effect. Most of the baryonic mass of these objects is in the form
of diffuse inter-galactic gas, heated to the virial temperature of
the cluster. With cluster masses as large as $10^{15} M_\odot$
and radii of several Mpc, the gas can be $\sim 10$ keV.
It is a fairly good approximation to take clusters as isothermal spheres
with a truncation at the virial radius:
\begin{equation}
\rho(r) = {\rho_\circ \over 1+({r \over r_c})^2 } \quad {\rm for} \quad r<R_v
\end{equation}
where the virial radius corresponds to the radius at which the infalling
gas is shocked and converts much of its kinetic energy into thermal
energy.
Assuming that the electron number density is also of this form,
and that isothermality is a good approximation
(Markevitch {\em et~al.}\ 1998), then
we can derive a value for the central SZ effect:
\begin{equation}
{\Delta T_{\circ} \over T_{CMB}} = 2 \, f(\nu)\, {kT_{gas} \over m_e c^2}\, \sigma_T
\, n_\circ \, r_c \, \arctan{({R_v \over r_c})}
\end{equation}
where $f(\nu)$ in the above gives the spectral dependence, approaching -2 in
the Rayleigh-Jeans limit.
The SZ central decrement (at low frequencies) depends on the distribution
of electrons in the cluster, with more condensed clusters having a stronger
signal for a given mass.
For a massive galaxy cluster, $n_\circ \sim 0.01 \, {\rm cm^{-3}},
T \sim 10 \, {\rm keV}, r_c \sim 250 \, {\rm kpc}$, and the virial radius is
usually a factor of ten (or more)
larger than the core radius. Using these numbers,
we arrive at a central decrement of $\Delta T/T_{CMB} \sim 10^{-3}$.
However, the central density and core radius are expected to vary from
cluster to cluster in a poorly understood way, making the central decrement
a difficult observable to understand.
The central decrement is a redshift-independent probe of clusters. The
magnitude of the central decrement is sensitive to the degree of
concentration of the cluster, as well as the path length along the
line of sight through the cluster. For a given set of cluster properties,
though, the central decrement does not depend on the redshift.
An observable that is much easier to understand is the total integrated
SZ signal:
\begin{equation}
S \equiv S_{\circ} \int d\Omega \,{\Delta T \over T_{CMB}} = S_{\circ} \int d\Omega \int dl \, f(\nu)
{kT_{gas} \over m_e c^2} \sigma_T n_e(l)
\end{equation}
where $S_{\circ}$ gives the conversion to physical units such as Jy.
Writing the integral over solid angle as $dA=d\Omega D_A(z)^2$, with $D_A(z)$
the angular diameter distance, this becomes
\begin{equation}
S = S_{\circ} f(\nu) {kT_{gas} \over m_e c^2} \sigma_T {N_e \over D_A(z)^2}
\end{equation}
with $N_e$ the total number of electrons, which should be simply related
to the total mass. Note that we have assumed nothing about the distribution
of the mass, only that the gas is isothermal. Any experiment with a
beam size larger than the cluster will not be sensitive to the degree of
concentration of the gas. This is not the case for most observables of
a system. For example, X-ray bremsstrahlung emission is proportional
to $n_e^2 T^{0.5}$, making the integrated emission highly sensitive to the
degree of concentration of the gas.
Finally, it is important to note that the SZ effect, as a distortion
of the blackbody spectrum, is independent of redshift for a given
set of cluster properties and therefore all redshifts contribute equally to the
magnitude of the effect along a given line of sight. Of course, a
nearby cluster will intersect more lines of sight, and thus will have
a larger integrated effect over the sky, as seen by the inclusion of
the angular diameter distance in the expression for the total integrated
effect.
Clusters are typically a few arcminutes in extent, so
as long as the beam size is $\la 1'$
the SZ effect is essentially independent of redshift. Even in the case
of large beams, the redshift dependence is not particularly restrictive.
For comparison, the angular diameter distance is
smaller than the luminosity distance by a factor of $(1+z)^2$.
\section{Observations}
\label{sec:observations}
As for observations of intrinsic anisotropy in the CMB, observations
of the SZ effect have progressed over the last several years from low
S/N detections and upper limits to high confidence detections and
detailed images. The dramatic increase in the quality of the
observations is due to improvements both in low-noise detection
systems and in observing techniques, usually using specialized
instrumentation with which the systematics that often prevent one
from obtaining the required sensitivity are carefully controlled. Such
systematics include, for example, the spatial and temporal variations
in the emission from the atmosphere and the surrounding ground.
The SZ effect observations so far have been targeted mainly
toward X-ray luminous galaxy clusters. Large scale
non-targeted surveys for the SZ effect could, in principle,
provide a unique and powerful probe of the distant universe
as discussed in section~\ref{sec:non-targeted} The power of such
surveys is
due to the independence of the SZ effect on the redshift
to the galaxy cluster or filament causing the scattering.
Unfortunately, the
sensitivity of the current instrumentation used to measure the effect is not
high enough to allow a large region to surveyed in a reasonable
amount of time. This situation may change soon, as instruments
with the required sensitivity are being planned now.
The existing observations of the SZ
effect toward galaxy clusters can be roughly divided into
two classes:
1) single dish observations toward low
redshift galaxy clusters ($z < 0.2$) at resolutions of several
arcminutes, and 2) interferometric observations of distant
clusters ($z > 0.2$) at resolutions of roughly an arcminute and higher.
A recent review of the observations can be found in
Birkinshaw (1999). Here we discuss briefly a few of the
results to provide the reader with a measure of the
quality of the data presently available.
Motivation for obtaining measurements of the SZ effect toward
galaxy clusters is provided by the ability to combine
SZ effect and X-ray emission data to determine cluster distances,
and therefore the
Hubble constant, and also to determine the fraction of
the total cluster mass contained in the hot gas.
The distance measurements are compelling as the estimates
do not share any of the systematics inherent in more traditional
methods; the technique relies solely on an understanding
of the cluster gas. The SZ effect distances do, of course,
suffer from other systematics. The gas mass fraction determinations
are compelling as
the gas mass accounts for most of the known baryonic mass
of the cluster (Forman and Jones 1982, White {\em et~al.}\ 1993)
and therefore the gas mass fraction allows a
good measure of the baryonic mass fraction for the cluster.
Furthermore, the baryonic gas mass fraction for massive clusters should
reflect the universal value $\Omega_B / \Omega_M$. Thus,
we can solve for $\Omega_M$ by using the cluster gas mass
fraction and the value of $\Omega_B$ derived from
big bang nucleosynthesis (BBN)
calculations (Copi, Schramm, and Turner 1995)
constrained to fit the observed
primordial abundance of light elements (e.g., Burles and Tytler 1998).
As discussed in Birkinshaw (1999), the first measurements of
the SZ effect were made with single dish radio telescopes.
Successful detections were obtained, although the reported
results show considerable scatter, reflecting the difficulty of the
measurement (see earlier review by Birkinshaw 1991 and references therein).
These observations illustrated the need for telescopes and receiver
systems designed explicitly to minimize differential atmospheric
emission and ground pickup. Recent state-of-the-art single dish
observations at radio (Herbig {\em et~al.}\ 1995, Myers {\em et~al.}\ 1997) and at millimeter wavelengths
(Wilbanks {\em et~al.}\ 1994, Holzapfel {\em et~al.}\ 1997b) have resulted in significant detections of the effect
and limited mapping. Observations near the null of the thermal
effect have been made in an attempt to measure the peculiar velocities
of galaxy clusters. An upper limit of about 2000~${\rm km\ s^{-1}}$\ has been
set for the $z\sim0.2$ clusters Abell~2163 and Abell~1689
by Holzapfel {\em et~al.}\ (1997a), who also report a significant detection
of the increment at frequencies above the null (see also
Lamarre {\em et~al.}\ 1998).
Interferometric techniques have been used to produce high
quality images of the SZ effect (e.g., Jones {\em et~al.}\ 1993,
Grainge {\em et~al.}\ 1993, Carlstrom, Joy, and Grego 1996,
Grainge {\em et~al.}\ 1996,
Carlstrom {\em et~al.}\ 1998). The high stability of
interferometry is being exploited to
make these observations. Interferometry
also provides data on the cluster over a large
range of angular scales making it possible to
remove contaminating emission from individual radio sources.
Lastly, it is possible to produce two-dimensional
images of the SZ effect with interferometry.
We show
in Figure~\ref{fig:szimages} four of the 27 clusters
imaged with the cm-wave SZ effect experiment on
the OVRO\footnote{An array of 10.4 m mm-wave telescopes
located in the Owens Valley, CA and operated by Caltech.}
and BIMA\footnote{An array of
ten 6.1 m mm-wave telescopes located
at Hat Creek, California and operated by the
Berkeley-Illinois-Maryland-Association} interferometers
(Carlstrom {em et~al.}\ private
comm.).
\begin{figure}
\plotone{sz4panel.eps}
\caption{Images of the the SZ effect observed toward four
galaxy clusters with similar luminosities, but spanning redshifts from 0.17 to 0.83. For
all four clusters the
contour levels are integer multiples of --75~$\mu K$,
which corresponds to
1.5 to 5 times the noise level, depending on the cluster.
The solid white ellipse in the bottom left corner of
each panel represents the effective resolution of the image.
The comparable SZ decrement observed for each cluster
illustrates the independence on the SZ effect signal on redshift.
The data were taken with low-noise cm-wave receivers installed
on the OVRO and BIMA mm-wave interferometric arrays. }
\label{fig:szimages}
\end{figure}
In his recent review of the SZ effect,
Birkinshaw (1999) provides a summary of
Hubble constant determinations from the above single
dish and interferometric, as well as other, SZ effect
observations of clusters with
redshifts as high
as 0.55. The mean value of the derived Hubble constants
is $\sim 60 \ {\rm km\,s^{-1} Mpc^{-1}}$ with no
meaningful constraint on the deceleration parameter.
The scatter in the Hubble constant
measurements is of order $20\ km s^{-1} Mpc^{-1}$.
Birkinshaw
points out, however, that
many of the observational uncertainties, such
as the X-ray and radio absolute flux scales are correlated between
the observations.
Two groups have reported cluster gas fractions
based on SZ effect data
(Myers {\em et~al.}\ 1997, Grego 1999, Grego {\em et~al.}\ 1999).
Myers {\it et al.}\ use their single dish SZ effect observations
combined with $\beta$-models for the gas density distribution
constrained by X-ray imaging data, and with gas temperatures
determined by X-ray spectroscopy, to determine the
gas mass fractions $f_g$ for a sample of four clusters.
They report $f_g h = 0.061\pm 0.01$
for three nearby clusters, although a fourth
cluster in their sample gives $f_g h = 0.166 \pm 0.014$.
Grego {\it et al.}\ report the results for
18 clusters imaged with the OVRO and BIMA cm-wave
interferometric SZ system. They use the SZ effect data itself
to constrain the $\beta$-models and X-ray spectroscopy
for the gas temperatures. For the entire
sample spanning redshifts from 0.171 to 0.826
and assuming an open $\Omega = 0.3$ cosmology, they
find $f_g h = 0.077 ^{+0.006}_{-0.010}$
scaled to $r_{500}$, the radius at which the
overdensity of the cluster is 500. Using only clusters
with redshifts less than 0.25 so that the
cosmology assumed will have little effect on
the gas mass fractions derived, they find $f_g h = 0.085 ^{+0.011}_{-0.145}$,
again scaled to $r_{500}$.
Assuming that the baryonic mass fraction for clusters
equals the universal value and using
$\Omega_B h^2 = 0.019 \pm 0.001$ determined from
BBN and primordial abundance measurements (Burles and Tytler 1998),
these gas mass fractions imply $\Omega_M h \sim 0.25$.
Future improvements in SZ effect observations, such
as dedicated interferometric arrays and single dish
telescopes equipped with bolometric arrays, will
allow detailed imaging of many more than the
order 30 clusters imaged to date and should
allow definitive measurements of the Hubble constant
and deceleration parameter. Commensurate improvements
in the absolute calibration scale for both the
radio and X-ray observations are also needed
and will be provided, in part, by the AXAF/Chandra satellite
at X-ray wavelengths, and
by careful ground based efforts and by the
planned CMB satellite missions at radio wavelengths.
The expected increase
in sensitivity and imaging speed will make large
scale non-targeted surveys of the SZ effect possible,
allowing a complete inventory of clusters at high
redshifts.
\section{The SZ Effect at Higher Redshifts}
\label{sec:high_redshift}
The imprint of distant clusters on the CMB has two important implications.
As the SZ effect is the upscattering of CMB photons, it should come as no
surprise that this can be a contaminant in CMB experiments. The thermal SZ
effect can become larger than the CMB signal around $\ell \ga
2000$, depending on the cosmology and cluster properties.
The thermal effect
has a distinctive frequency signature, allowing a multi-frequency experiment
with sampling on both sides of the null (near 217 GHz) to separate it from
intrinsic CMB anisotropy.
The kinetic SZ effect cannot be spectrally distinguished from the primary CMB
anisotropy, and could be impossible to remove from the data.
Fortunately, the kinetic signal is expected to be
much lower than the thermal signal, and
thus never becomes dominant.
While Planck has frequency channels that
trace out the thermal SZ spectrum on both sides of the null, this
is not the case for MAP which only has frequency coverage up to
90 GHz. For the thermal SZ effect, this is still fairly far into the
Rayleigh-Jeans region, where the signal becomes independent of
frequency. Thus, there is very little leverage for inferring the
amount of SZ contamination in measurements of the CMB. Several
papers have investigated the expected anisotropy due to
high-redshift clusters (Cole and Kaiser 1989,
Atrio-Barandela and Muecket 1999,
Colafrancesco {\em et~al.}\ 1994). We discuss the contribution of
the SZ effect to the angular power spectrum in
section \ref{sec:angular}
The second major implication is that surveys could be done to search
for this effect to map out the cluster abundance as a function of
redshift. This is expected to be a powerful probe of cosmology
(e.g., Barbosa {\em et~al.}\ 1996, Colafrancesco {\em et~al.}\ 1997)
as discussed in section~\ref{sec:non-targeted}
However, both the level of CMB contamination and the evolution of the
cluster abundance depend on cosmology and cluster physics. It is expected
that the dependencies are different for the two observables, and combining
CMB measurements with deep surveys should together separate cosmological
evolution of the number density from evolution of the gas properties.
In this section, we will lay out the tools required to probe
high-redshift clusters. Both the comoving number density and
the physical properties of clusters are expected to evolve
with redshift. The number density evolution of clusters of
a given mass is fixed by the cosmology, while a model is
needed to relate mass and temperature as a function of redshift.
Beyond this, X-ray studies indicate that the distribution of gas
within clusters also evolves with mass and redshift,
and must
also be taken into account. We will deal with these points,
in turn, below.
\subsection{Cosmological Evolution of Cluster Abundances}
\label{sec:cosmological}
The prospect of using the SZ effect to detect high-redshift clusters
is very exciting. Presumably, an SZ experiment should be able to
find clusters at any redshift, given high sensitivity and large sky
coverage. The expected number of clusters at a given sensitivity
depends both on the cosmology and the history of the cluster gas.
In a low-density universe, structure stops forming when the expansion
starts to be driven by either curvature or a cosmological constant.
If we live in a
low-density universe, the structure that we now see, by and large,
also existed out to a redshift of $z \sim 1$ and should be detectable
out to those redshifts.
In a critical-density universe, the structure that
we now see probably formed very recently, and thus would not be
detectable at high redshift.
This means that a given patch of sky should
contain more clusters in a low-density universe.
Since the
SZ effect is redshift-independent, all of these clusters should
be detectable.
To illustrate this, we will show results generated by assuming that
the Press-Schechter (Press \& Schechter 1976) formula for the
evolution of the number density with redshift is correct. We will
adopt two models with one having $\Omega_M=1.0,\Omega_{\Lambda}
=0.0, h=0.5, \sigma_8=0.6$ and the other having $\Omega_M=0.3,
\Omega_{\Lambda}=0.7,h=0.6,\sigma_8=1.0$.
and the usual CDM shape parameter
for the power spectrum was taken to be $\Gamma=0.25$.
Both of these models reproduce the local cluster abundance
(Viana and Liddle 1999), within the rather large errors.
All work was done assuming a global baryon fraction $f_b = 0.2$.
This is in reasonable agreement with observations of X-ray emissions
(Evrard 1997, Mohr, Mathiesen, and Evrard 1999) and SZ decrements
(Myers {\em et~al.}\ 1997, Grego {\em et~al.}\ 1999) from clusters.
Comparison between expected yields from BBN
(Copi, Schramm, and Turner 1995) and abundances of light elements
requires that $\Omega_b h^2=0.02$ (Burles and Tytler 1998). While
the assumed baryon fraction is consistent with BBN for $\Omega_M=0.3$,
this is
not the case for $\Omega_M=1$. Thus, our results may overpredict
the expected SZ signal in high-density universes, if BBN is correct and
the measured baryon fractions are in error.
The Press-Schechter formula gives the number density of virialized objects
of a given mass,
specifically excluding any mass which is not virialized.
This would be missing SZ signal from structures which have been significantly
heated but are not yet virialized. Thus, the results here are most likely to be underestimates of the total SZ signal from clusters, although
the importance of this effect has yet to be convincingly demonstrated.
For all of the results shown here, a lower mass limit of $10^{12} M_{\odot}$
was chosen, since masses below this would probably have cooled on a timescale
much shorter than a Hubble time. The reason for this is twofold. First of
all, low-mass objects are cooler to start with, and thus have less
energy to radiate. Secondly, cooling becomes increasingly efficient per
unit mass at lower temperatures.
\subsection{Mass-Temperature Relation}
\label{sec:mass-temp}
Press-Schechter gives a prediction for the number densities of objects
of various masses, while the SZ effect is sensitive to both the number
of electrons along the line of sight and the gas temperature.
A conversion between mass and SZ signal is therefore required.
The gas temperature can be obtained
by assuming that the gas is in virial equilibrium.
Virial equilibrium requires that $kT \propto {G M_v \over R_v}$. The
exact constant of proportionality depends on the distribution of mass
inside the cluster.
We need an estimate of the virial radius to get a mass-temperature
relation. This is usually done by assuming the spherical collapse
model.
A region that is overdense relative to the
critical density will eventually pull out of Hubble flow, turn around
and collapse. The time of turnaround will be at exactly half the time
of eventual collapse. Assuming that the epoch of observation is the
epoch of formation, the turnaround radius of an observed cluster can
be easily calculated given a cosmology. Ignoring $\Lambda$, the
virial relation asserts that $2K=-W$,
where K and W are the total kinetic and gravitational potential
energy, respectively. Given that $K=0$ at turnaround
and energy is conserved, this indicates that $W_v=2 W_{ta}$
(where $ta$ subscripts indicate properties at turnaround).
This directly leads to $R_v={R_{ta} \over 2}$. Including the effects
of $\Lambda$ modify this relation slightly (Lahav {\em et~al.}\ 1991).
At the same time, the
background universe has been expanding, changing the relative overdensity
of the bound region. The mean overdensity of the bound region thus
depends on the cosmology.
Fitting functions are given by Bryan and Norman (1998) for the mean overdensity
relative to the critical density, with their fit for a universe with
$\Omega_M+\Omega_{\Lambda}=1$ given by
\begin{equation}
\Delta(z) = 18 \pi^2 + 82 x -39 x^2
\end{equation}
where $x=\Omega_M(z)-1$.
The viral relation tells us that
$kT_{gas} \propto GM/R_{vir}$, but the exact constant of proportionality
depends on the distribution of matter. For this, we turn to numerical
simulations. Using the normalization of Bryan and Norman (1998), and
a notation where $H(z) \equiv 100h \, E(z)$, we use
\begin{equation}
\label{eqn:mass-temp}
T_{gas} = 1.4 \times 10^7 (h^2 \Delta(z) E(z)^2)^{1 \over 3}
({M \over 10^{15} M_\odot})^{2 \over 3} \, \, {\rm K}
\end{equation}
For a critical density universe, equation \ref{eqn:mass-temp} becomes
very simple, with the temperature for a given mass
scaling simply as $(1+z)$. This is easy to understand, since all
lengths (including the virial radius) will scale as $(1+z)^{-1}$,
so $GM/R_{v} \propto (1+z)$.
\subsection{Evolution of Cluster Gas}
\label{sec:evolution}
Evolution of cluster gas must also be taken into account. The
sizes of clusters are expected to evolve with the redshift of
formation,
with higher temperatures, and
therefore a stronger SZ signal for a given mass at higher
redshift. For the total SZ effect flux
for a cluster, this gets diluted
by the decreasing angular extent, but the central decrement suffers
no such dilution. In fact, since the central decrement
$\propto \rho r T_{gas}$, and $T_{gas} r \propto M$, the central decrement
for a fixed mass would be expected to nearly scale with the average
gas density.
If clusters evolve self-similarly, the central decrement
(for $\Omega_M=1$) evolves as
\begin{equation}
\Delta T_o \propto kT_{gas}n_o r_c \propto M(1+z)^3 \quad ({\rm self - similar})
\end{equation}
since $kT \propto GM/R$ and $n_o \propto (1+z)^3$.
The gas doesn't have to evolve this way, and X-ray observations seem
to indicate that it does not
(Kaiser 1991, Mohr, Mathiesen, and Evrard 1999).
For example, if reionization dumped
a substantial amount of entropy into the gas, or if the gas gained
a significant amount of entropy during infall, it is reasonable that
the core may not be able to adiabatically contract to form a compact core,
and the core could be more diffuse than self-similarity would predict.
The outer regions should be largely unchanged, giving a relation for
the core radius:
\begin{equation}
r_c^2 \propto {\rho_{crit}(z)\over \rho_c} R_v^2
\end{equation}
The self-similar solution has $\rho_c \propto \rho_{crit}$, and
$r_c \propto R_v$. The core properties in that case are independent
of cluster mass. If there is instead a minimum core entropy, then a
gas with polytropic index 5/3 will have $\rho_c \propto T^{3 \over 2}$.
The core properties now depend on cluster mass, with lower masses having
less concentration.
Under these conditions, we obtain a different relation for the
central decrement:
\begin{equation}
\Delta T_o \propto M \rho_{crit}(z)^{0.5} T_{gas}^{3/4} \propto
M^{3/2} (1+z)^{9/4} \quad ({\rm constant\,entropy\,core})
\end{equation}
for a fixed mass and a flat matter-dominated cosmology.
This is significantly different, both in its
mass dependence and its redshift evolution. Thus, we would expect
both qualitative and quantitative differences in counts above
a given central decrement in the two gas evolution scenarios.
This is shown in the left panel of Figure \ref{dndz_cd}. In this figure,
we compare counts of clusters above a central decrement for self-similar
scaling with a core:virial radii ratio of 1:10 and a constant entropy core
model normalized such that a 9 keV cluster has a central electron
number density of $2.0 \times 10^{-3}\,{\rm cm^{-3}}$. Counts are shown per
square degree for both a flat matter-dominated model and a model
with a cosmological constant. All calculations are done assuming
an observing frequency of 30 GHz.
\begin{figure}
\centerline{
\epsfxsize=.51\textwidth \epsfbox{dndz_dt.ps} \hfil
\epsfxsize=.51\textwidth \epsfbox{dndz_s2.ps}}
\caption{Counts per square degree of clusters with (left) central decrements
above a threshold of $100 \mu K$ (at 30 GHz)
and (right) total flux density above 0.5 mJy. The upper curve (at high
z) for each cosmology corresponds to self-similar evolution while
the lower curve is for
evolution with a constant entropy core.}
\label{dndz_cd}
\end{figure}
The two models for evolution form an envelope for each cosmology, probably
bracketing the real evolution the gas follows. It can be seen that the
range is enormous, with gas evolution equally important to the cosmological
evolution.
Compare this with the right panel of Figure \ref{dndz_cd}, where we show
the counts per square degree for the same models, except now for all
clusters with $S>0.5$\,mJy. Recall that it was earlier stated that the total
flux is independent of the gas distribution, assuming a constant
baryon fraction, and isothermality.
The constant entropy core model does not have a constant
baryon fraction, with low-mass clusters having
a lower baryon fraction. This is easy to understand, as low-mass
clusters have shallower potential wells, making it more difficult
for pre-heated gas to fall in.
Even for this case, it can be seen that
the counts do not differ substantially from those expected from
self-similar evolution, and cosmology is clearly the defining
element.
\section{Probing High-Redshift Clusters}
\label{sec:probing}
Using the tools of the previous section, we can now estimate the
expected signatures of high-redshift clusters. These can be
studied in at least two ways: non-targeted SZ surveys and
through the angular power spectrum of the anisotropy imprinted
on the CMB. The next few years will see both of these methods
implemented.
\subsection{Non-Targeted SZ Surveys}
\label{sec:non-targeted}
Current observations of the SZ effect have been targeted at
clusters which were known to have significant X-ray emission.
Non-targeted surveys hold much promise for studying the evolution
of the cluster abundance.
Planned non-targeted SZ surveys are sensitive to both cosmology and gas
evolution.
Optical follow-up of SZ surveys will be conducted to obtain the redshift
distribution.
This redshift
distribution is vital for finding the epoch of structure
freeze-out, which is set by the redshift at which either curvature or
$\Lambda$ first becomes dynamically important. This redshift distribution
allows an opportunity to differentiate between gas evolution and
cosmological evolution of the abundance. This is shown in
the left panel of Figure \ref{dndz_cd},
showing the redshift distributions of all clusters with a
central decrement above a given value, for different gas evolution
histories and cosmologies. Shown in the right panel of
Figure \ref{dndz_cd} is the distribution
of all clusters with a total flux density above a given threshold
for two different cosmologies. The total flux density is
almost independent of the gas evolution history, and is thus a very
clean probe of cosmology.
Two things to note in Figure 3 are 1) the large number of
expected clusters, and 2) the sensitivity of a small beam
experiment to the gas evolution of clusters. The first point
is interesting since the cluster mass threshold used for
the plots have been selected to be within reach of
planned SZ survey experiments. The importance of the second
point will be shown in the next section where its effect on
the angular power spectrum is discussed.
CMB experiments are expected to, in some ways, act as SZ cluster
surveys. It is expected that satellite experiments should yield
a large catalog of clusters. This will provide excellent information
on the local abundance of clusters, but the catalog will not have
much information on the redshift evolution. To underscore this
point, we have compared expected cluster catalogs for full sky
coverage with a high flux cut-off with a 12 deg$^2$ survey with much better
sensitivity in Figure \ref{dn_surveys}.
This plot was done by assuming self-similar evolution, but
it can be seen in Figure \ref{dndz_cd} that gas evolution does not have
much of an effect on this sort of plot.
The full-sky survey has a flux density cut-off of
20 mJy at 30 GHz, roughly comparable to the best expected performance
of MAP or PLANCK. The high-sensitivity survey is given a flux density
cut-off of 0.3 mJy, which may be possible with upcoming planned
ground-based SZ surveys.
While the large survey will yield more clusters, there is
very little information beyond the local cluster abundance, which is
already constrained by X-ray surveys (e.g., Viana and Liddle 1999). It
is the evolution of the cluster abundance which is most important for
cosmology.
\begin{figure}
\epsfysize = 4.5 in
\centerline{\epsfbox{dndz_s.ps}}
\caption{Expected number counts for two cosmologies for a low-sensitivity
full-sky SZ survey such as PLANCK compared with possible upcoming
ground-based surveys as a function of redshift.}
\label{dn_surveys}
\end{figure}
\subsection{Angular Power Spectrum}
\label{sec:angular}
The power spectra of the thermal and kinetic effects are practically
identical, with the main difference being an offset in amplitude by a factor
of $\sim 10-20$. The power spectrum is easy to understand. At large angular
scales (low $\ell$), clusters are unresolved, acting like point sources. The
angular Fourier transform of a point source has a flat amplitude, thus
the $c_\ell$ spectrum is flat.
At small angular scales (large $\ell$), the clusters are resolved
and the structure of the cluster is the important. If we take clusters to have
$\rho \propto 1/(r_c^2 + r^2)$, then the angular SZ profile has $\Delta T
\propto 1/\sqrt{\theta_c^2+\theta^2}$. The angular
Fourier transform of this is proportional to $\exp(-\theta_c \ell) /\ell$,
giving $\ell^2 c_\ell
\propto \exp(-2 \theta_c \ell)$.
Thus, a plot of $\ell(\ell+1)c_\ell$ for a single cluster
would rise as $\ell^2$ at large angular scales, while
it is suppressed as an exponential at small scales. The transition between
these two regimes happens around the characteristic scale of a cluster,
the core radius, leading to a peak in the power spectrum near $\ell \sim 2000$.
Both the amplitude and position of the peak
of the angular power spectrum are sensitive to the chosen
cosmology and the evolution of cluster gas. The amplitude
at low $\ell$ is set by the total abundance of SZ effect-inducing
clusters. The local abundance is fairly well constrained from X-ray
observations, but the evolution with redshift is sensitive to cosmology.
Since the SZ effect is
redshift-independent, high-redshift clusters will contribute to CMB
anisotropy as far back as they exist. This is particularly noticeable in
the slope of the power spectrum at low $\ell$'s. A model with a cosmological
constant has a significant contribution from compact high-redshift sources,
which have angular Fourier transforms which remain constant out to higher
values of $\ell$. Thus, the $\Lambda CDM$ model stays close to a line going
as $\ell^2$ longer than the flat matter-dominated model.
\begin{figure}
\epsfysize = 4.5 in
\centerline{\epsfbox{c_l.ps}}
\caption{Angular power spectra $[{ \ell(\ell+1)c_\ell \over 2\pi}]^{1 \over 2}$ for two
cosmologies and the two gas evolution histories. The top curve
for each cosmology corresponds
to self-similar evolution while the lower curve is for evolution with
a constant entropy core}
\label{c_l}
\end{figure}
The ratio of high-$\ell$ to low-$\ell$ contributions is effectively a measure
of a weighted average of angular core radii. Self-similar evolution
leads to much smaller cores for higher
redshift clusters, and thus these power spectra have significantly
more signal at high $\ell$ than their isentropic counterparts.
Low-density universes
have many more distant clusters (and thus more small-cored clusters) than
high density universes, giving more power at high $\ell$.
These two mechanisms, gas evolution and cosmological evolution of
the cluster abundance, can both change the average angular core radius
and lead to comparable effects in the SZ power spectrum,
making a clean interpretation of an SZ power spectrum
difficult.
Shown in Figure \ref{c_l_src} is the effect of removing all bright
SZ sources above 20 mJy (at 30 GHz). While this is very effective
at low $\ell$, beating down the SZ signal by nearly a factor of 4, at
high $\ell$ this has little effect.
A large fraction of the signal at low $\ell$'s comes from these clusters.
The bulk of the high $\ell$ signal still comes from
distant clusters, undetected individually. This is problematic, since
it is exactly in this region where the SZ power spectrum starts to
dominate over the signal from the primary anisotropy.
\begin{figure}
\epsfysize = 4.5 in
\centerline{\epsfbox{c_l_src.ps}}
\caption{Angular power spectra $[{ \ell(\ell+1)c_\ell \over 2\pi}]^{1 \over 2}$ for a
$\Lambda CDM$ cosmology and the two gas evolution histories,
showing the effect of removing all sources with a flux density
greater than 20 mJy (at 30 GHz). The top curve (at high $\ell$)
for each cosmology corresponds
to self-similar evolution while the lower curve is for evolution with
a constant entropy core}
\label{c_l_src}
\end{figure}
The thermal SZ effect does not approach the magnitude of the expected
CMB anisotropy until $\ell \ga 2000$. MAP has very little sensitivity at
this scale, and most current CMB experiments are not sensitive at this
scale either. However, many interferometric experiments have sensitivity
at this scale, and upcoming experiments such as PLANCK should definitely
pick up this SZ signal.
Added to this is uncertainty in the baryon fraction.
Converting to BBN values would depress the expected
power from SZ clusters in a flat matter-dominated universe by
a factor of $0.2/\Omega_b$. While the evidence for BBN is strong,
the measured baryon fractions in clusters seem difficult to
dismiss, making the choice of baryon fraction an important,
but difficult one in the case of $\Omega_M=1$.
\section{Summary}
\label{sec:summary}
The contribution to the CMB angular power spectrum by the SZ effect
from galaxy clusters is discussed. At low $\ell$ the primary CMB
anisotropy dominates the power spectrum. The SZ contribution rises
roughly as $\ell^2$ until $\ell \sim 2000$, where individual clusters
are resolved. At these $\ell$'s, the SZ contribution dominates that of
the primary CMB anisotropy. Since the spectral distortion of the CMB
induced by the SZ effect is distinctly non-thermal, it can be
separated from intrinsic anisotropy for CMB observations which have
broad frequency coverage.
A measurement of the contribution to the CMB anisotropy power spectrum
from the SZ effect would provide a measure of the underlying
cosmology, independent of the values derived from the primary
anisotropy. Combining large-scale non-targeted SZ effect surveys
currently under consideration and the SZ signal detected by CMB
experiments should give an independent determination of
$\Omega_m,\Omega_{\Lambda},{\rm and}\, \sigma_8$.
In order to complement CMB experiments, it would be useful to have SZ
effect surveys which can at least partially resolve clusters.
As seen in Figures 3-6, it is important to have this
structural information in order to constrain the evolution of the
cluster gas. Since typical core radii will be $1'$ or less, it would
be preferable to have a survey with a beam of this size or smaller.
The SZ effect should be more constructively considered as a low-redshift
source of secondary fluctuations in the CMB rather than simply as a foreground,
since the signature can be used as a direct constraint on cosmology.
|
\section{Introduction}
It has been established recently~\cite{Izv,Goslar,PLB} that the
Hamiltonian of the simplest version of the interacting boson model
IBM-1~\cite{Iachello} possesses an additional symmetry, the so-called
parameter symmetry, that is a symmetry of the parameter space of the
model. The symmetry manifests itself in the existence of two sets of the
Hamiltonian parameters that generate
identical spectra.
The IBM-1 Hamiltonian has an algebraic structure
characterized by the U(6) algebra. The spectrum and the eigenfunctions
can be found analytically in three particular cases (the U(5), SU(3) and
SO(6) dynamical symmetry (DS)
limits). A non-trivial issue of the parameter symmetry is that it
establishes an equivalence between the exactly solvable IBM-1 DS limits
and transitional IBM-1 Hamiltonians of a general form.
\par
The parameter symmetry is associated with canonical transformations
of boson operators
linking different realizations of SU(3) and SO(6) subalgebras in the
U(6) algebra~\cite{Goslar,PLB}.
\par
In this paper we propose a generalization of the parameter symmetry
concept on the case of IBM-2, the proton-neutron version of IBM. After a
survey of the parameter symmetry of IBM-1, we turn the discussion to the
structure of the general IBM-2 Hamiltonian and derive the
IBM-2 parameter symmetry relations.
\section{Parameter Symmetry of IBM-1}
Within IBM-1 nuclear states are labelled
by a fixed total number $N$ of bosons of two types, $s$ and $d$,
with quantum numbers $l^{\pi }=0^{+}$ and $l^{\pi }=2^{+}$,
respectively~\cite{Iachello}. The U(6) algebra is generated by
36 bilinear combinations of boson operators: $s^{+}s$,
$d^{+}_{\mu }d_{\nu}$, $d^{+}_{\mu }s$, $s^{+}d_{\mu}$.
The general
IBM-1 Hamiltonian can be expressed as~\cite{Iachello}
\begin{equation}
\label{e}
\begin{array}{rl}
H(\{k_i\})\;=& H_0+k_1C_1[\mbox{U(5)}]+k_2C_2[\mbox{U(5)}]
+k_3C_2[\mbox{SO(5)}] \\
&+\: k_4C_2[\mbox{SO(3)}]+k_5C_2[\mbox{SO(6)}]+k_6C_2[\mbox{SU(3)}] \; ,
\end{array}
\end{equation}
where $C_1$ and $C_2$ stand for the first and the second rank
Casimir invariants of the algebras entering the reduction chains of the
U(6) algebra:
\begin{equation}
\label{d}
{\renewcommand{\arraystretch}{0.1}
\begin{array}{cclc}
& & \mbox{U(5)}\supset \mbox{SO(5)}\supset
\mbox{SO(3)} & \qquad \mbox{I} \\
& \nearrow & \\
\mbox{U(6)} & \rightarrow & \mbox{SU(3)}\supset \mbox{SO(3)}
& \qquad \mbox{II} \\
& \searrow & \\
& & \mbox{SO(6)}\supset \mbox{SO(5)}\supset \mbox{SO(3)} \;
& \qquad \mbox{III} \\
\end{array} }
\end{equation}
We define the Casimir operators as in the book~\cite{Frank}:
\begin{equation}
\label{fg}
\begin{array}{@{}ll}
C_1[\mbox{U(5)}]=n_d \; ,
&
C_2[\mbox{SO(5)}]=2(T^3{\cdot }T^3)+2(T^1\cdot T^1) \; ,
\\
C_2[\mbox{U(5)}]=n_d(n_d{+}4) \; ,
&
C_2[\mbox{SO(6)}]=N(N{+}4)-4\left(P^{+}\cdot P\right) \; ,
\\
C_2[\mbox{SO(3)}]=10(T^1\cdot T^1) \; ,
&
C_2[\mbox{SU(3)}]=2(Q\cdot Q)+\frac{15}{2}(T^1\cdot T^1) \; ,
\end{array}
\end{equation}
where generators of the groups entering reduction chains (\ref{d}) are
given in Table~1,
\begin{table}
\caption{Generators of U(6), U(5), SU(3), SO(6), SO(5) and SO(3) algebras}
\vspace{3mm}
\begin{center}
\begin{tabular}{|c|l|}
\hline
$\vphantom{\widetilde{\widetilde{\overline{\mbox{SO}}}}_{\pi \nu }}$
Algebra & Generators \\
\hline
\hline
$\vphantom{\widetilde{\widetilde{\overline{\mbox{SO}}}}_{\pi \nu }}$
U(6) &$[d^+{\times}s]^{(2)}_{\mu }$,
$[s^+{\times}\tilde d]^{(2)}_{\mu } $, $S=[s^+{\times}s]^{(0)}_0$,
$T^\lambda_\mu=[d^+{\times}\tilde d]^{(\lambda )}_{\mu }$, \
$\lambda =0,1,2,3,4\!$
\\ \hline
$\vphantom{\widetilde{\widetilde{\overline{\mbox{SO}}}}_{\pi \nu }}$
U(5) & $T^\lambda_\mu=[d^+{\times}\tilde d]^{(\lambda )}_{\mu }$, \ \
$\lambda =0,1,2,3,4$ \\ \hline
SU(3) & $Q_\mu =[d^+{\times}s+s^+{\times}\tilde d]^{(2)}_{\mu }
{-}\frac{\sqrt{7}}{2}[d^+{\times}\tilde d]^{(2)}_{\mu }$, \ \
$T^1_\mu=[d^+{\times}\tilde d]^{(1)}_{\mu }$ \\
\hline
SO(6) & $Q^0_\mu =[d^+{\times}s+s^+{\times}\tilde d]^{(2)}_{\mu }$, \ \
$T^\lambda_\mu=[d^+{\times}\tilde d]^{(\lambda )}_{\mu }$, \ \
$\lambda =1,3$ \\ \hline
SO(5) & $T^\lambda_\mu=[d^+{\times}\tilde d]^{(\lambda )}_{\mu }$, \ \
$\lambda =1,3$ \\
\hline
SO(3) & $T^1_\mu=[d^+{\times}\tilde d]^{(1)}_{\mu }$ \\
\hline
\hline
\end{tabular}
\end{center}
\end{table}
\begin{equation}
\label{def}
\begin{array}{ll}
n_d=(d^+ \cdot \tilde d) \; ,
& P=\frac12\left((\tilde d \cdot \tilde d)-(s\cdot s)\right) \;,
\end{array}
\end{equation}
$\tilde d_{\mu }=(-1)^{\mu }d_{-\mu }$, ${\left( t\cdot u\right) }$ and
${\left[ t\times u\right] _{\mu }^{(\lambda )}}$ are
scalar and tensor products, respectively, of spherical tensors $t$
and $u$.
\par
Dynamical symmetry limits correspond to the cases when
the Hamiltonian involves Casimir operators
belonging to one of the reduction chains (\ref{d}) only, and hence
the eigenvalues and eigenfunctions can be found
analytically. The spectrum of the IBM Hamiltonian in the case of a DS
limit is one of the typical nuclear
spectra~\cite{Iachello}: vibrational in the U(5)
DS limit ($k_5{=}k_6{=}0$), rotational in the SU(3) DS
limit ($k_1{=}k_2$${=}k_3$${=}k_5{=}$ 0) and $\gamma $-unstable in
the SO(6) DS limit ($k_1{=}k_2{=}k_6{=}0$). A transitional
nuclear Hamiltonian that does not possess any DS, is
conventionally believed to generate a spectrum different from those
corresponding to any of the DS limits.
\par
As we have shown in Refs.~\cite{Izv,Goslar,PLB}, the IBM-1 Hamiltonian
possesses a parameter symmetry, namely:
\par
{\it Hamiltonians $H(\{k_i\})$ and $H(\{k^{\prime }_i\})$ defined by
Eq.}~(\ref{e}) {\it have identical spectra of eigenvalues
if the corresponding parameter sets $\{k_i\}$ and
$\{k^{\prime }_i\}$ are related as
\begin{equation}
\label{i1}
\begin{array}{l}
H^{\prime }_0=H_0,\ k^{\prime }_1=k_1+2k_6,\ k^{\prime }_2=k_2+2k_6,\
k^{\prime }_3=k_3-6k_6,\\
k^{\prime }_4=k_4+2k_6,\ k^{\prime }_5=k_5+4k_6,\ k^{\prime }_6=-k_6
\end{array}
\end{equation}
in the case $k_6\ne 0$, or as
\begin{equation}
\label{i2}
\begin{array}{l}
H^{\prime }_0=H_0+10k_5N,\ k^{\prime }_1=k_1+4k_5(N{+}2),\
k^{\prime }_2=k_2-4k_5,\\
k^{\prime }_3=k_3+2k_5,\ k^{\prime }_4=k_4,\
k^{\prime }_5=-k_5,\ k^{\prime }_6=0
\end{array}
\end{equation}
in the case $k_6= 0$}.
\par
This statement was proved
(see Refs.~\cite{Izv,Goslar,PLB} for details)
by constructing a unitary transformation
$U$ such that
$H(\{k_i^{\prime }\}) =UH(\{k_i\})U^{-1}$.
\par
Thus, for any set of the IBM-1 parameters there is another set which
generates the identical spectrum. The only exception is the U(5) DS limit
when the two sets of the parameters coincide as is seen from (\ref{i2}).
\par
One of the most intriguing issues of the parameter symmetry is that it
establishes the equivalence of
the nuclear spectrum corresponding to a certain DS to
the spectrum of a transitional IBM-1 Hamiltonian. As follows from
Eqs.~(\ref{i1}), the rotational spectrum of the SU(3) DS limit
($k_1{=}k_2{=}k_3{=}k_5{=}0$) appears to be equivalent to the
spectrum of
the transitional Hamiltonian with the set of
parameters $\{k^{\prime }_i\}\equiv \{k^{\prime }_1{=}2k_6$,
$k^{\prime }_2{=}2k_6$, $k^{\prime }_3{=}{-}6k_6$,
$k^{\prime }_4{=}k_4{+}2k_6$, $k^{\prime }_5{=}2k_6$,
$k^{\prime }_6{=}{-}k_6\}$
that does not correspond to any DS. Similarly, it follows
from (\ref{i2}), that the $\gamma $-unstable spectrum of the SO(6) DS limit
($k_1{=}k_2{=}k_6{=}0$) can be obtained with the set of parameters
$\{k^{\prime }_i\}\equiv\{ k^{\prime }_6{=}0$,
$k^{\prime}_1{=}8(N{+}2)k_5$, $k^{\prime }_2{=}{-}8k_5$,
$k^{\prime}_3{=}k_3{+}4k_5$,
$k^{\prime }_4{=}k_4$, $k^{\prime }_5{=}{-}k_5\}$
corresponding to the \mbox{U(5)--SO(6)} transitional nuclear
spectrum. In Ref.~\cite{Kusnezov}, such transitional Hamiltonians were
referred to as the ones possessing a hidden symmetry.
\par
To reveal the origin of the parameter symmetry, we note
that there is an ambiguity in definition of boson operators within
IBM~\cite{PLB,Iachello,Frank,VanIsacker,discrete}. One
can apply to the boson operators gauge transformations
$R_s(\varphi_s)$ and $R_d(\varphi_d)$ defined as \cite{VanIsacker,Iachello}
\begin{equation}
\label{rotation2R}
\begin{array}{ll}
R_s(\varphi_s)\,s^{+}=\exp(\mbox{i}\varphi _s/2)\,s^{+} \; , \qquad &
R_s(\varphi_s)\,s=\exp(-\mbox{i}\varphi _s/2)\,s \; , \\
R_d(\varphi_d)\,d^{+}_{\mu }=\exp(\mbox{i}\varphi _d/2)\,d^{+}_{\mu } \; ,
\qquad &
R_d(\varphi_d)\,\tilde d_{\mu }=\exp(-\mbox{i}\varphi_d/2)\,\tilde d_{\mu } \; .
\end{array}
\end{equation}
Note that these transformations are the canonical ones, i.e.\ they do
not violate the boson commutation relations.
However, the structure of the IBM Hamiltonian implies severe
restrictions on the use of the transformations (\ref{rotation2R}). For
example, in the case of the general IBM Hamiltonian, one can only
apply to the Hamiltonian the gauge transformation
$R(\varphi)\equiv R_s(\varphi_s)\times R_d(\varphi_d)$ with
$\varphi\equiv(\varphi_s-\varphi_d)/2=0,\,\pi$ and arbitrary
$\tilde\varphi\equiv(\varphi_s+\varphi_d)/2$~\cite{VanIsacker,discrete}.
Similarly, in the case of the
transitional SO(6)--U(5) IBM Hamiltonian with $k_{6}=0$, one can use
the gauge transformation $R(\varphi)$ with
$\varphi=0,\,\displaystyle\frac{\pi}{2},\,\pi,\,\frac{3\pi}{2}$.
One can also apply to the boson operators the
particle-hole conjugation $\tilde R$~\cite{Dieperink,Frank,discrete}
defined as
\begin{equation}
\label{ph2R}
\begin{array}{ll}
\tilde{R}\,s^{+}=s \; , \qquad & \tilde{R}\,s=-s^{+} \; , \\
\tilde{R}\, d^{+}_{\mu }=\tilde d_{\mu } \; , \qquad &
\tilde{R}\,\tilde d_{\mu} =-d^{+}_{\mu } \; ,
\end{array}
\end{equation}
and operators $\tilde{R}(\varphi)\equiv \tilde{R}\times R(\varphi)$ that
are consistent with the Hamiltonian structure provided that
$\varphi=0,\,\displaystyle\frac{\pi}{2},\,\pi,\,\frac{3\pi}{2}$ in the case
$k_{6}=0$ or $\varphi=0,\,\pi$ in the case $k_{6}\ne 0$. The operators
${R}(\varphi)$ and $\tilde{R}(\varphi)$ comprise a point group studied
elsewhere~\cite{discrete}.
We use the following notations for operators subjected to the
transformations $R(\varphi)$ and $\tilde R(\varphi)$:
${^{\alpha}O}\equiv R(\varphi)\,O$ and
${^{-\alpha}O}\equiv \tilde R(\varphi)\,O$, $\alpha=\varphi/\pi$.
Hamiltonians ${^{\alpha}H}(\{k_i\})\equiv R(\varphi)\,H(\{k_i\})$ are,
of course, isospectral with the initial Hamiltonian $H(\{k_i\})$
[Hamiltonians
${^{-\alpha}H}(\{k_i\})\equiv \tilde R(\varphi)\,H(\{k_i\})$ may
be not isospectral with $H(\{k_i\})$; however, one can always find a
simple and not very restrictive constraint on the parameters $k_i$
that will guarantee the
isospectrality of ${^{-\alpha}H}(\{k_i\})$ and
$H(\{k_i\})$]. Thus we can use transformations (\ref{rotation2R}) [and
in some cases (\ref{ph2R})] to study parameter symmetries and hidden
symmetries of IBM.
For example, the Hamiltonian
${^{1}H}(\{k_i\})\equiv R(\pi)\,H(\{k_i\})$ is isospectral but not
equivalent to the initial Hamiltonian $H(\{k_i\})$. Using Table~1 and
expressions (\ref{e}), (\ref{fg}), (\ref{def}) and (\ref{rotation2R}),
one can obtain \cite{PLB} that ${^{1}H}(\{k_i\})=H(\{k^{\prime}_i\})$ where the
set of parameters $\{k^{\prime}_i\}$ is defined by the parameter
symmetry relation (\ref{i1}). Thus, the transformation $R(\pi)$
is equivalent to the parameter symmetry transformation (\ref{i1}).
Note that, as is seen from Eqs.~(\ref{fg})--(\ref{def}) and Table~1,
only the Casimir operator $C_{2}[\mbox{SU(3)}]$ and the SU(3) generator
$Q_\mu$ are changed under the transformation $R(\pi)$:
\begin{equation}
R(\pi)\,Q_\mu=
{^{1}Q}_{\mu }= -[d^{+}{\times }\,s\,+\,s^{+}{\times }\,
\tilde{d}]^{(2)}_{\mu }
-\frac{\sqrt{7}}{2}[d^{+}{\times }\,\tilde{d}]^{(2)}_{\mu }\; . \label{Qbar}
\end{equation}
The quadrupole operators $Q_{\mu }$ and ${^{1}Q}_{\mu }$ correspond to
different embeddings of the SU(3) subalgebra in the U(6) algebra
[see also \cite{Dieperink} for other realizations of SU(3)]. Using
parameter symmetry transformation (\ref{i1}) it is easy to express
the Casimir operator $C_{2}\!\left[{\rm {SU^{1}(3)}}\right] $ of
the ${\rm SU}^{1}(3)$ algebra
associated with the quadrupole operator (\ref{Qbar}) through
$C_{2}\left[\mbox{SU(3)}\right]$ and Casimir operators of other
algebras~\cite{Goslar,PLB}:
\begin{equation}
\begin{array}{rl}
C_{2}\!\left[{\rm{SU^{1}(3)}}\right] = & 2C_1[\mbox{U(5)}]
+2C_{2}[\mbox{U(5)}]-6C_{2}[\mbox{SO(5)}] \\
& +\ 2C_{2}[\mbox{SO(3)}]+4C_{2}[\mbox{SO(6)}]-C_{2}[\mbox{SU(3)}] \; .
\end{array}
\label{m1}
\end{equation}
\par
In the case $k_{6}=0$ we have ${^{1}H}(\{k_i\})=H(\{k_i\})$ and
the transformation $R(\pi)$ does not generate the parameter symmetry.
However, in this case we can apply the transformation $R(\pi/2)$ to the
Hamiltonian.
Using Table~1 and expressions (\ref{e}), (\ref{fg}), (\ref{def}) and
(\ref{rotation2R}),
we obtain \cite{PLB} that $^{\slantfrac{1}{2}}H(\{k_i\})=H(\{k^{\prime}_i\})$
where the
set of parameters $\{k^{\prime}_i\}$ is defined by the parameter
symmetry relation (\ref{i2}). Hence, the transformation $R(\pi/2)$
is equivalent to the parameter symmetry transformation (\ref{i2}).
With the help of the transformation $R(\pi/2)$ we obtain a new
monopole operator:
\begin{equation}
\overline{P}\equiv {^{\slantfrac{1}{2}}P}=R(\pi/2)\, P=
\frac{1}{2}\left((\tilde d{\cdot } \tilde d)+
(s{\cdot } s)\right) .
\label{Pbar}
\end{equation}
This monopole operator corresponds to an alternative embedding of the
SO(6) subalgebra in the U(6) algebra \cite{VanIsacker,Frank}.
Using the parameter symmetry relation (\ref{i2}) it is easy to
obtain~\cite{Goslar,PLB} the following expression for
the Casimir operator of the $\overline{\mbox{SO(6)}}$ algebra associated with
the monopole operator $\overline{P}$:
\begin{equation}
C_{2}\!\left[ \overline{\mbox{SO(6)}}\right] =
10N{+}4(N{+}2)\,C_{1}[\mbox{U(5)}]
{-}4C_{2}[\mbox{U(5)}]{+}2C_{2}[\mbox{SO(5)}]{-}C_{2}[\mbox{SO(6)}]\,.
\label{n1}
\end{equation}
\par
Note, that the Casimir operators of alternatively embedded
algebras SU$^1$(3) and $\overline{\rm SO}(6)$ are not independent from
the Casimir operators of other algebras and should not be
included into the general Hamiltonian (\ref{e}).
\par
The transformations $\tilde R(\varphi)$ do not generate new parameter
symmetries. However one more parameter symmetry relation can be
obtained in the case of IBM-1 that is not associated with the
transformations $R(\varphi)$ and $\tilde R(\varphi)$ (see~\cite{PLB}
for a more detailed discussion).
\par
Usually in applications the Hamiltonian parameters $\{k_i\}$ are obtained
by the fit to nuclear spectra. Due to the parameter symmetry, the fit
of the parameters appears to be ambiguous.
To discriminate between the two sets of parameters
giving rise to identical spectra,
it is natural to study electromagnetic
transitions.
In the consistent-$Q$ formalism (C$Q$F)~\cite{CQF}, both
monopole-monopole
(${P^{+}\cdot P}$) and quadru\-pole-quadru\-pole
($Q\cdot Q$) interactions are replaced in the Hamiltonian by a single
term ($Q^{\chi }\cdot Q^{\chi }$) where
the generalized quadrupole operator
\begin{equation}
\label{QHI}
Q^{\chi }_{\mu }=[d^{+}{\times}\,s +s^{+}{\times}\,\tilde d]^{(2)}_{\mu }
+\chi\,[d^{+}{\times}\, \tilde d]^{(2)}_{\mu }.
\end{equation}
Operator $Q^{\chi }$ is used for calculations of
$E2$-transition rates within C$Q$F.
Applying transformation $R(\pi)$ to the Hamiltonian
$H(\{k_i\})$, we find out that the only term in the new Hamiltonian
${^{1}H}(\{k_i\})=R(\pi)\,H(\{k_i\})$ that differs from the corresponding
term in the initial Hamiltonian $H(\{k^{\prime}_i\})$, is the
generalized quadrupole-quadrupole interaction:
\begin{equation}
\label{1Q1Q}
\left({^{1}Q}^{\chi }{\cdot }\,{^{1}Q}^{\chi }\right)=R(\pi)\,
\left({\vphantom{^{1}}Q^{\chi }}{\cdot }\,{Q^{\chi }}\right),
\end{equation}
where
\begin{equation}
\label{1Qchi}
{^{1}Q}^{\chi}_{\mu }\equiv R(\pi)\,{Q^{\chi }_{\mu }}=-Q^{-\chi }_{\mu } \; .
\end{equation}
The consistent transformation of the $E2$ transition operator
(\ref{QHI}) according to (\ref{1Qchi}) and of the generalized
quadrupole-quadrupole interaction in the Hamiltonian according to
(\ref{1Q1Q}), guarantees that the $E2$ transition rates remain unchanged.
Therefore in the general case $\chi\ne 0$ that corresponds to
$k_{6}\neq 0$, the $E2$ transition rates cannot be used to distinguish between
two sets of Hamiltonian parameters $\{k_i\}$ and $\{k'_i\}$ related by
the parameter symmetry (\ref{i1}), at least within the C$Q$F formalism.
if it is believed that the
C$Q$F ansatz is an adequate prescription for the electromagnetic
transition operator.
We note that in the general case $\chi\ne 0$
the type of the generalized quadrupole-quadrupole interaction [whether
it is of the form $(Q^{\chi}{\cdot }\,Q^{\chi })$ or
$(^1Q^{\chi}{\cdot }\,^1Q^{\chi })$]
is unambiguously determined by the set
of the Hamiltonian parameters $\{k_{i}\}$. We have shown in \cite{PLB} that
this is due to the fact that the generalized
quadrupole-quadrupole interaction
includes the monopole-monopole term (${P^{+}\cdot P}$).
In the case $\chi =0$ ($k_6=0$), we apply the transformation
$R(\pi/2)$ to the Hamiltonian and to the quadrupole operator
$Q^0_\mu$ to obtain
\begin{equation}
\label{Q0bar}
\overline{Q^{0}_{\mu }}\equiv ^{\slantfrac{1}{2}\!\!}Q^{0}_\mu\equiv
R(\pi/2)\,Q^{0}_{\mu } =-\mbox{i}[d^{+}{\times }\,s\
-\ s^{+}{\times }\,\tilde d]^{(2)}_{\mu }\, .
\end{equation}
However, in this case the generalized quadrupole-quadrupole
interaction is ambiguous. As we have shown in Ref.~\cite{PLB}, the
parameter symmetry relation (\ref{i2}) can be used to derive
\begin{equation}
\left(\vphantom{\overline{Q^0}}Q^{0}{\cdot }\,Q^{0}\right)=
-\left( \overline{Q^{0}}{\cdot }\,\overline{Q^{0}}
\right){+}10N{+}4(N{+}2)\,C_{1}[\mbox{U(5)}]{-}4C_{2}[\mbox{U(5)}]
{-}2C_{2}[\mbox{SO(5)}]\,. \label{Q0ambig}
\end{equation}
Thus in the case $k_{6}=0$ the IBM-1 Hamiltonian can be expressed either
through $\left(\vphantom{\overline{Q^0}} Q^{0}{\cdot }\,Q^{0}\right)$
or alternatively through
$\left( \overline{Q^{0}}{\cdot }\,\overline{Q^{0}}\right) $.
As a result, the definition of the $E2$ transition operator appears to be
ambiguous. Due to this ambiguity, the electromagnetic transition rates
cannot be used to distinguish among two sets of Hamiltonian parameters
$\{k_i\}$ and $\{k'_i\}$ related by the parameter symmetry (\ref{i2}).
The origin of the ambiguity of the generalized quadrupole-quadrupole
interaction is that the quadrupole-quadrupole interaction
$\left( Q{\cdot }\,Q\right)$ is not present in the Hamiltonian in the
case $k_6=0$ and the operators
$\left(\vphantom{\overline{Q^0}} Q^{0}{\cdot }\,Q^{0}\right)$ and
$\left( \overline{Q^{0}}{\cdot }\,\overline{Q^{0}}\right)$ within
C$Q$F replace the monopole-monopole term (${P^{+}\cdot P}$) in the
Hamiltonian (see Ref.~\cite{PLB} for more details).
\par
There is another possibility to distinguish among the two Hamiltonians
related by the parameter
symmetry in the case $k_{6}=0$. This possibility
stems from the $N$-dependence of the parameter symmetry~(\ref{i2}).
Since the relations~(\ref{i2}) involve the total
number of bosons $N$, the two sets of the parameters can generate
identical spectra for some particular nucleus only, the predictions
for the spectra of its isotopes or
isotones should differ. It is conventionally supposed
(see for example Ref.~\cite{Ndependence}) that the
spectra of neighboring even-even nuclei are described by the same set
of the IBM parameters, hence one can
discriminate between the parameter sets $\{k_i\}$ and $\{k'_i\}$ related
according to (\ref{i2}) by comparing the
spectra of the neighboring nuclei.
This is illustrated by Fig.~1 where the spectra of three Pt isotopes
are presented.
\begin{figure}
\centerline{\epsfig{file=fig.eps,width=0.98\textwidth}}
\caption{Few lowest levels of each $J^{\pi}$ of Pt isotopes. SO(6):
calculations within SO(6) DS limit with the parameters
suggested in Ref.~\protect\cite{Iachello}; PS: calculations with the
set of parameters obtained using
(\protect\ref{i2}) with $N=6$; Exp: experimental data of
Ref.~\protect\cite{ExpPt}.}
\end{figure}
The set of
parameters $k_{1}=k_{2}=k_{6}=0$, $k_{3}=50$~keV, $k_{4}=10$~keV, $
k_{5}=-42.75$~keV was suggested in \cite{Iachello}
for the description of $^{196}$Pt ($N$=6) within the SO(6) DS
limit of IBM.
The corresponding spectra are given in the left columns labelled by SO(6).
The set of parameters $k_{1}^{\prime }=-1368$~keV, $k_{2}^{\prime }=171$~keV,
$k_{3}^{\prime }=-35.5$~keV, $k_{4}^{\prime }=10$~keV, $k_{5}^{\prime }
=42.75$~keV and $k_{6}^{\prime }=0$ is obtained using (\ref{i2}) with $N=6$.
The corresponding spectra are given in the right columns labelled by PS.
The SO(6) and parameter symmetry spectra are, of course, identical in the
case of $^{196}$Pt but differ for other Pt isotopes.
\par
As is seen from (\ref{e}), (\ref{fg}) and (\ref{def}),
the transformation $R(\pi/2)$ or, equivalently, the parameter symmetry
transformation (\ref{i2}), changes the sign of the monopole-monopole
interaction $(P^{+}{\cdot }\,P)$ in the Hamiltonian. This sign change
manifests itself in the spectra of neighboring nuclei. It is usually
supposed that the pairing (monopole-monopole) interaction should be
attractive, i.e. $k_5<0$. Note that the set of parameters suggested in
Ref.~\cite{Iachello} with attractive pairing interaction fitted to
$^{196}$Pt, reproduces the experimental data on $^{192}$Pt and
$^{194}$Pt better (see Fig.~1) than the other set with
$k^{\prime }_5>0$.
The generalized
quadrupole-quadrupole interaction $(Q^{\chi }{\cdot }\,Q^{\chi })$
incorporates both quadrupole-quadrupole $(Q{\cdot }\,Q)$ and
pairing $(P^{+}{\cdot }\,P)$ interactions. The transformation
$R(\pi)$, or, equivalently, the parameter symmetry transformation
(\ref{i1}) changes only the sign of the quadrupole-quadrupole
interaction $(Q{\cdot }\,Q)$ in the Hamiltonian, as is seen from
Table~1 and Eqs.~(\ref{e}), (\ref{fg}) and (\ref{def}); the
monopole-monopole and other multipole-multipole terms are not effected
by the transformation $R(\pi)$. Contrary to that of the pairing
interaction, the sign of the quadrupole-quadrupole interaction
$(Q{\cdot }\,Q)$ is of no physical importance and is indistinguishable
in applications as we have shown above.
\section{ Parameter Symmetry of IBM-2}
IBM-2~\cite{IBM2_1,Iachello} is a proton-neutron version of
IBM. Within this model, $s$ and $d$ bosons are introduced
in the proton and neutron subspaces independently.
The symmetry algebra of the model is
U$_{\pi }$(6)$\otimes $U$_{\nu }$(6) generated by 72 bilinear operators
$s_{\rho }^{+}s_{\rho }$, $d^{+}_{\rho \mu }d_{\rho \nu}$,
$d^{+}_{\rho \mu }s_{\rho }$, $s^{+}_{\rho }d_{\rho \mu}$
($\rho= \pi $, $\nu $).
The general IBM-2 Hamiltonian $H$ consists of
proton part $H_{\pi }$, neutron part $H_{\nu }$ and
proton-neutron interaction $V_{\pi \nu }\,$,
\begin{equation}
\label{Hibm2}
H\left(\{k_i^\pi,k_i^\nu,k_i\}\right)=
H_{\pi} \left(\{k_i^\pi\}\right)
+H_{\nu}\left(\{k_i^\nu\}\right) +V_{\pi \nu }\left(\{k_i\}\right) ,
\end{equation}
and is characterized by 21 independent parameters
$\{k_i^\pi,k_i^\nu,k_i\}$~\cite{Iachello}.
The proton
and neutron
parts of the Hamiltonian,
$H_{\rho}\left(\{k_i^\rho\}\right)$, $\rho =\pi, \nu$,
are just the IBM-1 Hamiltonians and are given by (\ref{e}) with $H_0=0$.
It is desirable to express the proton-neutron interaction
$V_{\pi \nu}\left(\{k_i\}\right)$ as a superposition of Casimir operators of
combined proton-neutron subalgebras G$_{\pi \nu}$
of the U$_{\pi }$(6)$\otimes $U$_{\nu }$(6) algebra that
enter the reduction chains starting with
U$_{\pi }$(6)$\otimes $U$_{\nu }$(6) and ending with SO$_{\pi \nu }$(3).
The generators ${\cal G}_{\pi \nu }$ of the combined proton-neutron
algebras G$_{\pi \nu}$ are of the form
${\cal G}_{\pi \nu }={\cal G}_{\pi }+{\cal G}_{\nu }$ where
${\cal G}_{\pi }$ and ${\cal G}_{\nu }$ are generators of the
corresponding proton and neutron algebras, respectively. For example,
the generators of the SO$_{\pi \nu}$(6) algebra are
$Q^0_{\pi \mu }+Q^0_{\nu \mu }\,$,
$T^1_{\pi \mu }+T^1_{\nu \mu }\,$, and $T^3_{\pi \mu }+T^3_{\nu \mu }\,$.
The U$_{\pi }$(6)$\otimes $U$_{\nu }$(6) algebra has a number of
appropriate reduction chains.
There are three types of the reduction chains which include U(5),
SU(3) and SO(6) subalgebras~\cite{Frank}:
1. U(5) DS chains:
\begin{equation}
\label{chU5}
{\renewcommand{\arraystretch}{0.0}
\begin{array}{ccccccccc}
\mbox{U$_{\pi }$(6)} & \rightarrow & \mbox{U$_{\pi }$(5)} &
\rightarrow & \mbox{SO$_{\pi }$(5)} & \rightarrow &
\mbox{SO$_{\pi }$(3)} & & \\
& \searrow & & \searrow & & \searrow & & \searrow & \\
& & \mbox{U}_{\pi \nu }(6) & \rightarrow &
\mbox{U}_{\pi \nu }(5) & \rightarrow & \mbox{SO}_{\pi \nu }(5) &
\rightarrow & \mbox{SO}_{\pi \nu }(3) \\
& \nearrow & & \nearrow & & \nearrow & & \nearrow & \\
\mbox{U}_{\nu }(6) & \rightarrow & \mbox{U}_{\nu }(5) &
\rightarrow & \mbox{SO}_{\nu }(5) & \rightarrow &
\mbox{SO}_{\nu }(3) & &
\end{array} }
\end{equation}
2. SU(3) DS chains:
\begin{equation}
\label{chSU3}
{\renewcommand{\arraystretch}{0.0}
\begin{array}{ccccccc}
\mbox{U$_{\pi }$(6)} & \rightarrow & \mbox{SU$_{\pi }$(3)} &
\rightarrow & \mbox{SO$_{\pi }$(3)} & & \\
& \searrow & & \searrow & & \searrow & \\
& & \mbox{U}_{\pi \nu }(6) & \rightarrow &
\mbox{SU$_{\pi \nu }$(3)} & \rightarrow & \mbox{SO$_{\pi \nu }$(3)} \\
& \nearrow & & \nearrow & & \nearrow & \\
\mbox{U$_{\nu }$(6)} & \rightarrow & \mbox{SU$_{\nu }$(3)} &
\rightarrow & \mbox{SO$_{\nu }$(3)} & &
\end{array} }
\end{equation}
3. SO(6) DS chains:
\begin{equation}
\label{chSO6}
{\renewcommand{\arraystretch}{0.0}
\begin{array}{ccccccccc}
\mbox{U}_{\pi }(6) & \rightarrow & \mbox{SO}_{\pi }(6) & \rightarrow &
\mbox{SO}_{\pi }(5) & \rightarrow & \mbox{SO}_{\pi }(3) & & \\
& \searrow & & \searrow & & \searrow & & \searrow & \\
& & \mbox{U}_{\pi \nu }(6) & \rightarrow &
\mbox{SO}_{\pi \nu }(6) & \rightarrow &
\mbox{SO}_{\pi \nu }(5) & \rightarrow & \mbox{SO}_{\pi \nu }(3) \\
& \nearrow & & \nearrow & & \nearrow & & \nearrow & \\
\mbox{U}_{\nu }(6) & \rightarrow & \mbox{SO}_{\nu }(6) & \rightarrow &
\mbox{SO}_{\nu }(5) & \rightarrow & \mbox{SO}_{\nu }(3) & &
\end{array} }
\end{equation}
We note that the set of Casimir operators provided by the algebras
entering the reduction chains
(\ref{chU5})--(\ref{chSO6}), is not
complete enough to express the general IBM-2 Hamiltonian (\ref{Hibm2}).
The problem is partly solved by adding the $\overline{\mbox{SO}}(6)$
DS reduction chains~\cite{Frank} to the reduction chains
(\ref{chU5})--(\ref{chSO6}):
4. $\overline{\mbox{SO}}(6)$ DS chains:
\begin{equation}
\label{chbarSO6}
{\renewcommand{\arraystretch}{0.0}
\begin{array}{ccccccccc}
\mbox{U}_{\pi }(6) & \rightarrow & \overline{\mbox{SO}}_{\pi }(6)
& \rightarrow &
\mbox{SO}_{\pi }(5) & \rightarrow & \mbox{SO}_{\pi }(3) & & \\
& \searrow & & \searrow & & \searrow & & \searrow & \\
& & \mbox{U}_{\pi \nu }(6) & \rightarrow &
\overline{\mbox{SO}}_{\pi \nu }(6) & \rightarrow &
\mbox{SO}_{\pi \nu }(5) & \rightarrow & \mbox{SO}_{\pi \nu }(3) \\
& \nearrow & & \nearrow & & \nearrow & & \nearrow & \\
\mbox{U}_{\nu }(6) & \rightarrow & \overline{\mbox{SO}}_{\nu }(6)
& \rightarrow &
\mbox{SO}_{\nu }(5) & \rightarrow & \mbox{SO}_{\nu }(3) & &
\end{array} }
\end{equation}
The reduction chains (\ref{chU5})--(\ref{chbarSO6}) will be referred
to as standard DS reduction chains.
Contrary to the Casimir operators
$C_2[\overline{\mbox{SO}}_{\pi}(6)]$ and
$C_2[\overline{\mbox{SO}}_{\nu}(6)]$ [see Eq. (\ref{n1})],
the Casimir operator $C_2[\overline{\mbox{SO}}_{\pi \nu }(6)]$ is an
additional independent operator that can be used for the construction
of the general IBM-2 Hamiltonian.
However we still do not have a complete set of independent Casimir
operators. To obtain this set we should look for alternative
embeddings of the combined proton-neutron algebras. All the
alternative subalgebras can be obtained by applying all possible
transformations $R_\rho(\varphi_\rho)$ and
$\tilde R_\rho(\varphi_\rho)$ to the generators of all subalgebras in
the reduction chains (\ref{chU5})--(\ref{chbarSO6}). As a result, we
obtain alternative subalgebras ${\rm G}_\rho^{\alpha_\rho}$ and
${\rm G}_{\pi\nu}^{\alpha_\pi\alpha_\nu}$ with generators
${^{\alpha_\rho}{\cal G}_\rho}$ and
$\;{^{\alpha_\pi\alpha_\nu}{\cal G}_{\pi\nu}}
={^{\alpha_\pi}{\cal G}_\pi}+{^{\alpha_\nu}{\cal G}_\nu}$, respectively.
For example, the generators of the algebra
SO$^{0 1}_{\pi \nu}$(6) are $Q^0_{\pi \mu }-Q^0_{\nu \mu }$,
$T^1_{\pi \mu }+T^1_{\nu \mu }$ and $T^3_{\pi \mu }+T^3_{\nu \mu}$ [see
Eq.~(\ref{1Qchi}) for the expression of ${^{1}Q^0_{\nu\mu}}$]. The
$\overline{\mbox{SO}}_{\pi\nu}(6)$ algebra is the
${\rm
SO}_{\pi\nu}^{\!\!\!\!\!\!\!\slantfrac{1}{2}\!\!\!\!\!\!\!\slantfrac{1}{2}}(6)$
algebra in these notations. The SU$^{0 -1}_{\pi \nu}$(3) algebra
is equivalent to the
SU$^{*}_{\pi \nu }$(3) algebra introduced in Ref.~\cite{Dieperink}
for the description of triaxial shapes within IBM-2.
In such a way we obtain a large number of
alternative algebras. However, some of them are equivalent. For
example, any algebra ${\rm G}_{\pi\nu}^{\alpha_\pi\alpha_\nu}$ is equivalent
to its proton-neutron particle-hole counterpart
algebra ${\rm G}_{\pi\nu}^{-\alpha_\pi-\alpha_\nu}$ --- the
relative sign of $\alpha_\nu$ and $\alpha_\pi$ is only important, changing
the sign of both $\alpha_\nu$ and $\alpha_\pi$ we do not obtain a new
algebra. As follows from our analysis, there exist 2 different realizations of
U$_{\pi \nu }$(5), 2 different realizations of SO$_{\pi \nu }$(6),
2 different realizations of $\overline{\mbox{SO}}_{\pi \nu }$(6),
and 8 different realizations of SU$_{\pi \nu }$(3).
The alternative algebras provide us with Casimir operators that can be
used for the construction of $V_{\pi \nu}\left(\{k_i\}\right)$,
however not all of these Casimir operators are independent.
For example, the Casimir operators of alternative proton or neutron
algebras ${\rm G}_\rho^{\alpha_\rho}$ can be expressed through the Casimir
operators of untransformed algebras ${\rm G}_\rho$ [see Eqs.~(\ref{m1}) and
(\ref{n1})]; the rank-1 Casimir operator of any of
${\rm U}^{\alpha_\pi\alpha_\nu}_{\pi\nu}(5)$ algebras can be expressed
through the Casimir operators of ${\rm U}^{\alpha_\pi}_{\pi}(5)$ and
${\rm U}^{\alpha_\nu}_{\nu}(5)$:
$C_1[{\rm U}^{\alpha_\pi\alpha_\nu}_{\pi\nu}(5)]=
C_1[{\rm U}^{\alpha_\pi}_{\pi}(5)]+
C_1[{\rm U}^{\alpha_\nu}_{\nu}(5)]$, etc.
So, we should choose a set of independent rank-2 Casimir
operators of combined proton-neutron subalgebras.
We suggest to include in this set the Casimir operators of
U$_{\pi \nu }$(6), U$_{\pi \nu }$(5), SO$_{\pi \nu }$(6),
$\overline{\mbox{SO}}_{\pi \nu }$(6), SU$_{\pi \nu }$(3),
SU$^{01}_{\pi \nu }$(3) and SU$^{11}_{\pi \nu }$(3).
The Casimir operators of all the rest proton-neutron algebras can be
expressed through the ones included in the set, e.g.,
\begin{equation}
\label{CT9}
\begin{array}{rl}
C_2\left[\mbox{SU}^{0 -1}_{\pi \nu }(3)\right]\:
=&-\:C_2\,[\mbox{SU}^{01}_{\pi \nu }(3)]
+2C_2\left[\mbox{SU}_{\pi }(3)\right]
+2C_2\,[\mbox{SU}^1_{\nu }(3)] \\[2mm]
& \displaystyle
+\:\frac32\left\{C_2\left[\mbox{SO}_{\pi \nu }(3)\right]
-C_2\left[\mbox{SO}_{\pi }(3)\right]-C_2\left[\mbox{SO}_{\nu }(3)\right]
\vphantom{^2}
\right\} .
\end{array}
\end{equation}
The Casimir operator $C_2[\mbox{U}(6)]$ not
defined above can be expressed as
\begin{equation}
\label{C2U6}
C_2\left[\mbox{U}(6)\right]=
(S\cdot S)+\frac12(Q^{0}\cdot Q^{0})
+\frac12(\overline{Q^{0}}\cdot \overline{Q^{0}})
+\sum_{\lambda =0}^{4}(T^{\lambda }\cdot T^{\lambda }) \; .
\end{equation}
The proton-neutron interaction $V_{\pi \nu }\left(\{k_i\}\right)$
we express through the set of independent Casimir operators as
\begin{equation}
\label{Vpn}
\begin{array}{rl}
V_{\pi \nu }\left(\{k_i\}\right)\:
=&H_0+kC_2[\mbox{U}_{\pi \nu }(6)]
+k_2C_2[\mbox{U}_{\pi \nu }(5)]
+k_3C_2[\mbox{SO}_{\pi \nu }(5)] \\
&+\:k_4C_2[\mbox{SO}_{\pi \nu }(3)]
+k_5C_2[\mbox{SO}_{\pi \nu }(6)]
+k_6C_2[\overline{\mbox{SO}}_{\pi \nu }(6)] \\
&+\:k_7C_2[\mbox{SU}_{\pi \nu }(3)]
+k_8C_2[\mbox{SU}^{01}_{\pi \nu }(3)]
+k_9C_2[\mbox{SU}^{11}_{\pi \nu }(3)] \; .
\end{array}
\end{equation}
Note that the set of independent Casimir operators is not unique and,
as a result, alternative expressions for
$V_{\pi \nu }\left(\{k_i\}\right)$ can be suggested.
Another possible choice of the operators was used in Ref.~\cite{Leviatan2}.
It is seen that the construction of different realizations of boson
algebras plays an important role in IBM-2.
The incompleteness of boson Hamiltonians in the form of
superposition of Casimir invariants of different groups determined by
standard DS reduction chains, is a common property of systems of
two (or more) independent subsystems, e.g. it is also a property of the
vibron model of
triatomic molecules with the symmetry algebra
U$_1$(4)$\otimes $U$_2$(4)~\cite{LeviatanVM}.
The standard reduction chains (\ref{chU5})--(\ref{chbarSO6}) define
standard DS limits of IBM-2. As the Casimir operators of
SU$^{01}_{\pi\nu}$ and SU$^{11}_{\pi\nu}$ are present in the
Hamiltonian, we can also define non-standard SU$^{01}_{\pi\nu}$ and
SU$^{11}_{\pi\nu}$ DS limits of IBM-2 that are associated with non-standard
SU$^{01}$ and SU$^{11}$ DS reduction chains, respectively. The definition
of non-standard DS limits of the model is, of course, ambiguous
because of the ambiguity of definition of the complete set of Casimir
operators.
Applying all possible
transformations $R_\rho(\varphi_\rho)$ and
$\tilde R_\rho(\varphi_\rho)$ to all subalgebras in
the standard reduction chains (\ref{chU5})--(\ref{chbarSO6}), we obtain all
alternative reduction chains. Some of these reduction chains appear to
be equivalent to some of the others, some of them are equivalent to
some of the standard or non-standard DS reduction chains.
However the set of independent alternative
reduction chains can be easily defined. These independent alternative
reduction chains give rise to hidden symmetries of the model.
It is interesting that some of the hidden symmetries may be obtained
by means of transformations (e.g., by particle-hole transformations)
that are not isospectral.
Applying all possible
transformations $R_\rho(\varphi_\rho)$ and
$\tilde R_\rho(\varphi_\rho)$ to the general Hamiltonian (\ref{Hibm2}),
we obtain a general IBM-2 Hamiltonian that can be
(i)~identical to the initial Hamiltonian (\ref{Hibm2}),
(ii)~non-identical to but isospectral with the initial Hamiltonian
(\ref{Hibm2}), or (iii)~non-isospectral with the initial Hamiltonian
(\ref{Hibm2}). In the case (ii) we obtain standard parameter
symmetries of IBM-2 that are valid without restrictions on the
parameters of the model. The standard parameter symmetries are listed
in Table 2.
\begin{table}[t]
\caption{Standard parameter symmetry relations for the IBM-2 Hamiltonian}
\vspace{3mm}
\begin{center}
\begin{tabular}{|l|l|l|l|}
\cline{1-4}
A & B & C & D \\
\cline{1-4}
$\vphantom{\widetilde{\widetilde{H}}}$
$R_{\pi }(0){\times }R_{\nu }(0)$ &
$R_{\pi }(\pi ){\times }R_{\nu }(0)$ &
$R_{\pi }(0){\times }R_{\nu }(\pi )$ &
$R_{\pi }(\pi ){\times }R_{\nu }(\pi )$ \\
\cline{1-4}
$H_0$ & $H_0-10(k{+}2k_6)N$ & $H_0-10(k{+}2k_6)N$
& $H_0$ \\
\cline{1-4}
$k_1^{\pi}$ &
$k_1^{\pi}{+}2k_6^{\pi}{+}4(N_{\pi}{+}2)(2k_6{+}k)\!\!$ &
$k_1^{\pi}{+}4(N_{\pi}{+}2)(2k_6{+}k)$ &
$k_1^{\pi}{+}2k_6^{\pi}$ \\
$k_2^{\pi}$ &
$k_2^{\pi}{+}2k_6^{\pi}{-}8k_6{-}4k$ &
$k_2^{\pi}{-}8k_6{-}4k$ &
$k_2^{\pi}{+}2k_6^{\pi}$ \\
$k_3^{\pi}$ &
$k_3^{\pi}{-}6k_6^{\pi}{-}2k_5{+}2k_6{-}8k_7$ &
$k_3^{\pi}{-}2k_5{+}2k_6{-}8k_9$ &
$k_3^{\pi}{-}6k_6^{\pi}{-}8k_8$ \\
$k_4^{\pi}$ &
$k_4^{\pi}{+}2k_6^{\pi}$ &
$k_4^{\pi}$ &
$k_4^{\pi}{+}2k_6^{\pi}$ \\
$k_5^{\pi}$ &
$k_5^{\pi}{+}4k_6^{\pi}{+}2k_5{-}2k_6{+}8k_7$ &
$k_5^{\pi}{+}2k_5{-}2k_6{+}8k_9$ &
$k_5^{\pi}{+}4k_6^{\pi}{+}8k_8$ \\
$k_6^{\pi}$ &
${-}k_6^{\pi}$ &
$k_6^{\pi}$ &
${-}k_6^{\pi}$ \\
\cline{1-4}
$k_1^{\nu}$ &
$k_1^{\nu}{+}4(N_{\nu}{+}2)(k{+}2k_6)$ &
$k_1^{\nu}{+}2k_6^{\nu}{+}4(N_{\nu}{+}2)(k{+}2k_6)\!\!$ &
$k_1^{\nu}{+}2k_6^{\nu}$ \\
$k_2^{\nu}$ &
$k_2^{\nu}{-}8k_6{-}4k$ &
$k_2^{\nu}{+}2k_6^{\nu}{-}8k_6{-}4k$ &
$k_2^{\nu}{+}2k_6^{\nu}$ \\
$k_3^{\nu}$ &
$k_3^{\nu}{-}2k_5{+}2k_6{-}8k_7$ &
$k_3^{\nu}{-}6k_6^{\nu}{-}2k_5{+}2k_6{-}8k_9$ &
$k_3^{\nu}{-}6k_6^{\nu}{-}8k_8$ \\
$k_4^{\nu}$ &
$k_4^{\nu}$ &
$k_4^{\nu}{+}2k_6^{\nu}$ &
$k_4^{\nu}{+}2k_6^{\nu}$ \\
$k_5^{\nu}$ &
$k_5^{\nu}{+}2k_5{-}2k_6{+}8k_7$ &
$k_5^{\nu}{+}4k_6^{\nu}{+}2k_5{-}2k_6{+}8k_9$ &
$k_5^{\nu}{+}4k_6^{\nu}{+}8k_8$ \\
$k_6^{\nu}$ &
$k_6^{\nu}$ &
${-}k_6^{\nu}$ &
${-}k_6^{\nu}$ \\
\cline{1-4}
$k$ & $k$ & $k$ & $k$ \\
$k_2$ & $k_2$ & $k_2$ & $k_2$ \\
$k_3$ & $k_3{+}2k_5{+}2k_6{+}2k{+}8k_7$ &
$k_3{+}2k_5{+}2k_6{+}2k{+}8k_9$ & $k_3{+}8k_8$ \\
$k_4$ & $k_4$ & $k_4$ & $k_4$ \\
$k_5$ & ${-}k_5{-}k{-}8k_7$ & ${-}k_5{-}k{-}8k_9$ &
$k_5{-}8k_8$ \\
$k_6$ & ${-}k_6{-}k$ &
${-}k_6{-}k$ & $k_6$ \\
$k_7$ & $k_7$ & $k_8{+}k_9$ &
$k_8{+}k_9$ \\
$k_8$ & ${-}k_7{+}k_9$ &
$k_7{-}k_9$ & ${-}k_8$ \\
$k_9$ & $k_7{+}k_8$ &
$k_9$ & $k_7 {+}k_8$ \\
\cline{1-4}
\end{tabular}
\end{center}
\end{table}
In the case (iii) we do not immediately obtain parameter symmetries.
However, for any possible transformation $R_\rho(\varphi_\rho)$ [or
$\tilde{R}_\rho(\varphi_\rho)$] there always can be found some constraints on
the parameters $\{k^\pi_i,k^\nu_i,k_i\}$ such that the
Hamiltonians $H(\{k^\pi_i,k^\nu_i,k_i\})$ and
$R_\rho(\varphi_\rho)\,H(\{k^\pi_i,k^\nu_i,k_i\})$ [or
$\tilde{R}_\rho(\varphi_\rho)\,H(\{k^\pi_i,k^\nu_i,k_i\})$]
become isospectral
even if there is no isospectrality between these Hamiltonians in the
general case. Hence in the case (iii) we obtain additional
non-standard parameter symmetries that are valid only if the parameters fit
some relations. These additional symmetries are listed in Table~3 (the
constraining relations for the parameters are given in the first row).
All IBM-2 parameter symmetry relations with the only exception of the
parameter symmetry D (see Table~2), involve the total number of proton
bosons $N_\pi$ or/and the total number of neutron bosons $N_\nu$
($N=N_\pi+N_\nu$). Hence there is a principal possibility to distinguish
between few parameter sets giving rise to identical spectra by the
analysis of the spectra of neighboring isotopes or/and isotones.
\begin{table}
\epsfig{file=table3.eps,angle=90,height=0.99\textheight}
\end{table}
\section{Summary}
We have analyzed canonical transformations of boson operators consistent
with the structure of boson Hamiltonian in the cases of IBM-1 and IBM-2,
or, equivalently, different realizations of the symmetry algebra
of the model and its subalgebras. Analysis of alternatively embedded
subalgebras is of a particular importance in the case of IBM-2,
because it provides a regular way to construct the general IBM-2
Hamiltonian.
One can suppose that there should be no signals in physical applications
of switching from one realization of the symmetry
algebra to another equivalent realization.
However, it is not so. The existence of alternative realizations of the
symmetry algebra manifests itself as parameter symmetries of the model,
i.e. as existence of few sets of Hamiltonian parameters providing
identical spectra. The parameter symmetry is of physical importance
for applications of IBM since the parameters of the model are obtained
by the fit to experimental spectra. We have shown that in some cases one
can discriminate between the sets of parameters related by the parameter
symmetry by analyzing spectra of neighboring isotopes and/or isotones.
\par
The parameter symmetry is a common property of boson models.
For example, we have shown~\cite{VM} that it is present in the vibron
model (see e.g.~\cite{Frank}); we suppose that it can be found in
$sdg$-IBM and other algebraic models including fermion and boson-fermion
ones. There can exist other possibilities of discriminating between
isospectral parameter sets (see,
e.g., Ref.~\cite{Yoshida} where the proton-neutron interacting
boson-fermion model is discussed).
Before finishing the paper, we mention few recent papers related to the
present investigation. D.~Kusnezov~\cite{Kusnezov} discussed in detail
hidden symmetries, i.e. the particular cases of parameter symmetries
relating transitional Hamiltonians to the ones corresponding to DS
limits. He noted the relevance of hidden symmetries to the studies of
chaos. A more detailed discussion of this item can be found in
Ref.~\cite{Jolie}. In Ref.~\cite{d-parity} the so-called $d$ parity was
introduced for IBM Hamiltonian in the U(5)--SO(6) transitional case. The
$d$ parity operator commutes with the Hamiltonian and provides an
additional quantum number for qualification of the energy levels,
electromagnetic transitions, etc.
\par
We are thankful to R.~Bijker, D.~Bonatsos, C.~Daskaloyannis,
G.~F.~Filippov, F.~Iachello, R.~V.~Jolos, V.~P.~Karassiov, T.~Otsuka,
J.~Patera, N.~Pietralla, D.~L.~Pursey, V.~N.~Tolstoy and P.~Van~Isacker
for valuable discussions. The paper is supported partly by
the Competitive Center at St.~Petersburg State University,
the Russian Foundation of Basic Research, and
the European Community through project CI1*-CT94-0072.
|
\subsection*{Acknowledgements}
The authors would like to thank J. Fr\"ohlich for
stimulating discussions.
V.C. is grateful to B. I. Halperin for important remarks
and for hospitality at the Lyman Lab.
We hould like to thank L. Levitov and A. Chang for valuable
comments and information about their work.
|
\section{Introduction}
In optical relations among observed quantities, distances such as the
luminosity distance and the angular diameter distances play an
important role. They are clearly defined in the homogeneous
Friedmann-Lemaitre-Robertson-Walker model (Weinberg,\cite{rf:weintx}
Schneider et al.\cite{rf:sef}) owing to the simple nature of light
propagation in this case. In inhomogeneous
universes, however, their behavior is complicated, due to gravitational
lens effect which implies that light rays are deflected gravitationally
by an inhomogeneous matter distribution. On the other hand, we also use
distances to interpret the structure of gravitationally lensed
systems.
To correctly treat distances in inhomogeneous universes, it is
necessary first to have a reasonable formulation for the dynamics
describing local matter motion and
optics and clarify the validity condition of the formulation. A set of
fluid dynamical equations and the Poisson equation in the cosmological
Newtonian approximation was introduced and discussed by
Nariai\cite{rf:nari60} and Irvine\cite{rf:irv65} under the conditions
\begin{equation}
\label{eq:int1}
\vert \Phi \vert \ll 1, \ (v/c)^2 \ll 1, \ L/L_H \ll 1,
\end{equation}
where $\Phi, \ v, L$ and $L_H$ are the Newtonian gravitational potential,
matter velocity, the characteristic size of inhomogeneities and the
horizon size $\approx ct$,
respectively, and the spacetime is expressed as
\begin{equation}
\label{eq:int2}
ds^2 = -(1 +2\Phi)c^2 dt^2 +(1 -2\Phi) a^2(t)[d\chi^2 +\sigma^2(\chi)
d\Omega^2],
\end{equation}
where $a(t)$ is the scale-factor, $\sigma (\chi) = \sin \chi, \chi,
\sinh \chi$ for the background curvature $k = 1, 0, -1$, respectively,
and $d\Omega^2 = d\theta^2 + \sin^2 \theta \varphi^2$.
The above fluid dynamical equations can describe the nonlinear local
motion, while the gravitational field is linear with respect to
$\Phi$. The extension of the above cosmological Newtonian treatment to
a post-Newtonian treatment was performed by Futamase,\cite{rf:futa88}
Tomita\cite{rf:tpost88} and Shibata and Asada.\cite{rf:shib96}
Futamase showed that the condition
\begin{equation}
\label{eq:int3}
\epsilon^2 /\kappa \ll 1 \quad ( \epsilon^2 \sim \Phi \ {\rm and } \ \kappa
\sim L/L_H)
\end{equation}
is necessary for the higher-order expansion to be possible and
formulated the spatial averaging and the back-reaction to the background.
Moreover, Futamase and Sasaki\cite{rf:fs89} investigated the validity
of light propagation in the cosmological Newtonian iterative
approximation and discussed the distance problem.
In an empty region as the limiting inhomogeneous case, distances
exhibit behavior very different from those in homogeneous models
(the Friedmann distances). In the special case without tidal
shear from surrounding regions, the so-called Dyer-Roeder angular
diameter distance was derived by Zel'dovich,\cite{rf:zel64} Dashevskii
and Slysh\cite{rf:dash66}
and Dyer and Roeder.\cite{rf:dyroe72}\tocite{rf:dyroe73}
In a non-empty region with a constant matter density $\rho_m$ but no
tidal shear, we have the generalized
Dyer-Roeder distance with the clumpiness (or smoothing) parameter
$\alpha$, which is defined as
\begin{equation}
\label{eq:int4}
\rho_m/\rho_{\rm F} = \alpha = {\rm const}
\end{equation}
for the Friedmann density $\rho_{\rm F}$. The observational results derived
from the optical relations depend on whether we use
the Friedmann distances or the Dyer-Roeder distance. Quantitative
estimates for these difference and the effect of the cosmological
constant have been studied by Fukugita et al.\cite{rf:fuk92}
and Asada.\cite{rf:asad98} On the other hand, it is important to determine
what distances are most applicable and what value of the above
parameter $\alpha$ is best, in realistic inhomogeneous models.
Kasai et al.\cite{rf:kasa90} and Watanabe and Tomita\cite{rf:wt90}
numerically calculated the frequency distribution for generalized
distances in simple models in which particles are distributed randomly.
Recently, Tomita\cite{rf:tom98c} derived this distribution in more realistic
inhomogeneous models generated using the $N$-body simulation with the
CDM spectrum. The general result is that the average value of $\alpha$ is
nearly 1 and its dispersion decreases with the increase of the
redshift $z$, though it is $\sim 1$ for $z = 0.5$.
Another interesting topic is that involving the role of the shear
term and the Ricci and Weyl focusing terms (in the optical scalar
equation\cite{rf:sachs61})
in the behavior of distances. To this time, the shear
effect has been discussed by Weinberg,\cite{rf:wein76} Watanabe et
al.,\cite{rf:wts92} Watanabe and Sasaki,\cite{rf:ws90} and
Nakamura,\cite{rf:naka97} and its focusing effect has recently been
studied numerically by Hamana using a Monte Carlo simulation, taking
into account small-scale inhomogeneities.
In this review paper, basic optical relations and the definition of
distances are first given in \S 2, the lensing relations
in the generalized Dyer-Roeder distance are derived in \S 3 (by Asada),
the statistical behavior of distances analyzed in numerical
simulation is described in \S 4 (by Tomita), and the shear and focusing
effects are discussed in \S 5 (by Hamana).
\newpage
\section{Optical scalars and the definition of distances}
\subsection{Geometry of ray bundles}
Rays are expressed as $x^\mu = x^\mu(y^i, v)$, where $v$ is an
affine parameter. For each ray, the $y^i$ have constant values
($C^i$). The wave vector $k^\mu = \partial x^\mu/\partial v$ satisfies
the null condition and null-geodesic equation
\begin{equation}
\label{eq:opt1}
k^\alpha k_\alpha = 0, \quad {k^\alpha}_{;\beta} k^\beta = 0.
\end{equation}
For two rays with $y^i = C^i$ and $y^i = C^i + \delta C^i$, the
connection vector is
\begin{equation}
\label{eq:opt2}
\delta x^\mu = ({\partial x^\mu/\partial y^i}) \delta C^i.
\end{equation}
If we define a dot differentiation by
$(m^\alpha)^\cdot = {m^\alpha}_{;\beta} k^\beta$ \
for an arbitrary vector $m^\mu$, we obtain from Eq. (\ref{eq:opt2})
\begin{equation}
\label{eq:opt4}
(\delta x^\alpha)^\cdot = {k^\alpha}_{;\beta} \delta x^\beta,
\end{equation}
and then (Jordan et al.,\cite{rf:jord} Sachs\cite{rf:sachs61})
\begin{equation}
\label{eq:opt5}
(k_\alpha \delta x^\alpha)^\cdot = 0.
\end{equation}
Now let us consider the situation in which on a screen (at a point
P$_{\rm o}$) an observer sees the shadow formed by a source object
(at a point
P$_{\rm s}$). The connection vector $\delta_\bot x^\alpha$ vertical to
$u^\alpha$ and $k^\alpha$ at the point P$_{\rm s}$ is
\begin{equation}
\label{eq:opt6}
\delta_\bot x^\alpha = h^\alpha_\beta \delta x^\beta,
\end{equation}
where
\begin{equation}
\label{eq:opt7}
h^\alpha_\beta = \delta^\alpha_\beta - {k^\alpha k_\beta \over
(k^\gamma k_\gamma)^2} - {k^\alpha u_\beta +u^\alpha k_\beta \over
k^\gamma u_\gamma}.
\end{equation}
The $h^\alpha_\beta$ satisfy the relation
\begin{equation}
\label{eq:opt8}
h^\alpha_\beta h^\beta_\gamma = h^\alpha_\beta, \ h^\alpha_\beta u^\beta
= h^\alpha_\beta k^\beta = 0 \ {\rm and} \ h^\alpha_\alpha = 2.
\end{equation}
If $\bar{\delta} x^\alpha$ is the vector obtained by
parallel-transporting
$\delta_\bot x^\alpha$ from P$_{\rm s}$ along the ray, we have
\begin{equation}
\label{eq:opt9}
\bar{\delta} x^\alpha = \delta_\bot x^\alpha + \int (\delta_\bot
x^\alpha)^\cdot dv,
\end{equation}
and the connection vector $\bar{\delta}_\bot x^\alpha$ vertical to
$u^\alpha$ and $k^\alpha$ at the point P$_{\rm o}$ is
\begin{equation}
\label{eq:opt10}
\bar{\delta x^\alpha} = h^\alpha_\beta \bar{\delta} x^\beta =
\delta_\bot x^\alpha + \int h^\alpha_\beta
(\delta_\bot x^\beta)^\cdot dv.
\end{equation}
The length $\delta l$ and the angle $\alpha_{12}$ are expressed as
follows using the connection vector in the plane vertical to
$k^\alpha$ and the screen velocity $\bar{u}^\alpha$ :
\begin{equation}
\label{eq:opt11}
\delta l = (g_{\alpha\beta} \bar{\delta}x^\alpha
\bar{\delta}x^\beta)^{1/2}, \ \cos \alpha_{12} =
[(\bar{\delta}x^\alpha)_1 (\bar{\delta}x^\beta)_2]/[(\delta l)_1
(\delta l)_2].
\end{equation}
Here $\bar{u}^\alpha$ is parallel-transformed along the ray from
$u^\alpha$ at P$_{\rm s}$.
\subsection{Optical scalars}
Using Eqs.(\ref{eq:opt4}) and (\ref{eq:opt5}) we obtain
\begin{equation}
\label{eq:opt12}
(\delta_\bot x^\alpha)^\cdot = A^\alpha_\beta \delta_\bot x^\beta,
\end{equation}
where
\begin{equation}
\label{eq:opt13}
A_{\alpha \beta} = h^\gamma_\alpha h^\lambda_\beta k_{\gamma ; \lambda}.
\end{equation}
The tensor $A_{\alpha \beta}$ can be uniquely split as
\begin{eqnarray}
\label{eq:opt14}
A_{\alpha \beta} &=& A_{[\alpha \beta]} + \theta h_{\alpha \beta} +
\sigma_{\alpha \beta}, \cr
\omega &=& \Bigl[{1 \over 2}A_{[\alpha \beta]} A^{\alpha \beta}\Bigr]^{1 \over
2} = \Bigl[{1 \over 2}k_{[\alpha ;\beta]} k^{\alpha ;\beta}\Bigr]^{1 \over
2}, \cr
\theta &=& {1 \over 2} A^\gamma_\gamma = {1 \over 2} k^\gamma_{;\gamma},
\quad
\sigma_{\alpha \beta} = A_{(\alpha \beta)} - {1 \over
2}A^\gamma_\gamma h_{\alpha \beta}, \cr
\sigma &=& \Bigl({1 \over 2} \sigma_{\alpha \beta} \sigma^{\alpha
\beta}\Bigr)^{1 \over 2} = \Bigl\{{1 \over 2}[k_{(\alpha \beta)}
k^{\alpha \beta}
- {1 \over 2}(k^\gamma_{;\gamma})^2]\Bigr\}^{1 \over 2},
\end{eqnarray}
where $\theta, \sigma$ and $\omega$ are optical scalars representing
the expansion, shear and rotation, respectively, of ray bundles.
In geometric optics which we assume in the following, the rotation
vanishes, because $k^\mu$ is a gradient vector. By the transformation
of $v$, $\theta, \sigma$ and $\omega$ transform, but $\theta dv,
\sigma dv$ and $\omega dv$ are invariant (Jordan et al.,\cite{rf:jord}
Sachs\cite{rf:sachs61}).
The evolution equations for $\theta$ and $\sigma$ are
\begin{equation}
\label{eq:opt15}
{d\theta \over dv} = {1 \over 2} k^\mu_{;\mu\nu} k^\nu = -{1 \over 2}
{\cal R} - (\theta^2 + \sigma^2),
\end{equation}
\begin{equation}
\label{eq:opt16}
{d\sigma \over dv} = - {\cal C} -2 \theta \sigma,
\end{equation}
where ${\cal R} \equiv R_{\mu\nu} k^\mu k^\nu$, and ${\cal C}$ is
expressed in terms of the Weyl tensor as ${\cal C} \equiv
C_{\alpha\beta\gamma\lambda} k^\alpha k^{\gamma}\bar{t}^\beta
\bar{t}^{\lambda}$, with $t^\alpha$ a complex null vector satisfying
$t^\alpha t_\alpha = k^\alpha t_\alpha = 0, \bar{t}^\alpha
\bar{t}_\alpha = 1$. As can be seen from Eqs.~(\ref{eq:opt15}) and
(\ref{eq:opt16}) there two terms causing the focusing of ray
bundles. One is the Ricci focusing term ${\cal R}$, proportional to the
matter density, and the other is the Weyl focusing term ${\cal C}$,
connected with the shear.
\subsection{Definition of distances}
The length of the shadow of the interval between two rays in the
observer plane is given by
\begin{equation}
\label{eq:opt17}
(dl)^2 = g_{\alpha \beta}
\bar{\delta}_\bot x^\alpha \bar{\delta}_\bot x^\beta.
\end{equation}
Then we have
\begin{equation}
\label{eq:opt18}
{d (\delta l)\over \delta l} = A_{\alpha \beta} e^\alpha e^\beta dv = (\theta +
\sigma_{\alpha \beta} e^\alpha e^\beta) dv,
\end{equation}
where $e^\alpha \equiv \delta_\bot x^\alpha/ \delta l$ and $g_{\alpha
\beta} e^\alpha e^\beta = 1$. Next, let us consider the area of the
cross-section of a ray bundle given by
\begin{equation}
\label{eq:opt19}
\delta A = {1 \over 2} {\int_0}^{2\pi} (\delta l)^2 d\alpha.
\end{equation}
If the deviation of the cross-section from a circle is small, we
obtain
\begin{equation}
\label{eq:opt20}
d(\delta A) = (\delta A)^\cdot dv = dv \delta A {\int_0}^{2\pi}
{d(\delta l) \over \delta l} d\alpha
\end{equation}
or
\begin{equation}
\label{eq:opt21}
{d(\delta A) \over \delta A} = 2 \theta dv,
\end{equation}
because ${\int_0}^{2\pi}\sigma_{\alpha \beta} e^\alpha e^\beta
d\alpha$ vanishes.
Here we define two kinds of angular diameter distances (the linear
angular diameter distance $D_{\rm lA}$ and the area angular diameter
distance $D_{\rm aA}$) proportional to $\delta l$ and $(\delta A)^{1/2}$,
respectively, as
\begin{equation}
\label{eq:opt22}
{d D_{\rm lA} \over dv} = (\theta + \sigma_{\alpha \beta} e^\alpha
e^\beta) D_{\rm lA},
\end{equation}
\begin{equation}
\label{eq:opt23}
{d D_{\rm aA} \over dv} = \theta D_{\rm aA},
\end{equation}
where we consider ray bundles with $\theta = \infty$ at the
observer point P$_{\rm o}$. In the case of no shear, $D_{\rm lA}$
and $D_{\rm aA}$
are equal, but generally they are different. Their average values are
equal if the term $\sigma_{\alpha \beta} e^\alpha e^\beta$ is
cancelled out in the averaging process. In the situation that
$\theta= \infty$ at the source's point P$_{\rm s}$, we obtain the luminosity
distance $D_{\rm L}$ satisfying
\begin{equation}
\label{eq:opt24}
{d D_{\rm L} \over dv} = \theta D_{\rm L}.
\end{equation}
The relation between $D_{\rm aA}$ and $D_{\rm L}$ is proved to be
\begin{equation}
\label{eq:opt25}
D_{\rm L} = (1 + z)^2 D_{\rm aA}
\end{equation}
by Etheringen,\cite{rf:ether} where $z$ is the redshift
given by the
relation $1 + z = (u^\alpha k_\alpha)_{\rm source} \\
/(u^\alpha k_\alpha)_{\rm observer}$.
The solutions of Eqs. (\ref{eq:opt22}) and (\ref{eq:opt23})
are generally complicated, but they can easily be obtained in the
special case in which (1) the density is spatially constant, (2) there is
no shear, and (3) the affine parameter in the Friedman background can
be used. From Eqs. (\ref{eq:opt15}) and (\ref{eq:opt23}) we obtain
\begin{equation}
\label{eq:opt26}
{d^2 D_{\rm A} \over dv^2} = -{1 \over 2} {\cal R} D_{\rm A},
\end{equation}
where $D_{\rm A} = D_{\rm lA} = D_{\rm aA}$, ${\cal R} = 8\pi G \alpha \rho/ a^2 (t)
= 3 (1+z)^5 \alpha \Omega_0$, $\Omega_0$ is the total density
parameter, and $\alpha$ is the smoothing (clumpiness) parameter.
The affine parameter $v$ is related to $z$ as
\begin{equation}
\label{eq:opt27}
{dz \over dv} = (1 +z)^2 [(1+\Omega_0 z)(1+ z)^2 - \lambda_0 z(2+z)]^{1/2},
\end{equation}
where $\lambda_0$ is the normalized cosmological constant. The
boundary condition for $D_{\rm A}$ at epoch $z = z_1$ is
\begin{equation}
\label{eq:opt28}
D_{\rm A} = 0, \quad {d D_{\rm A} \over dz} = c (dt/dz)_{z=z_1} = {c \over
H_0}{a(z_1) \over a_0}.
\end{equation}
The solution of Eq. (\ref{eq:opt27}) is called the Friedmann
distance for $\alpha = 1$, the Dyer-Roeder distance for
$\alpha = 0$, and the generalized Dyer-Roeder distance for arbitrary
$\alpha$.
For the analyses of cosmological lens systems, we often use the lens
equation
\begin{equation}
\label{eq:opt29}
\mbox{\boldmath $\beta$} = \mbox{\boldmath $\theta$} -
4G{D_{OL} D_{LS} \over D_{OS}}
{\partial \over \partial \mbox{\boldmath $\theta$}}
\int d^2 \theta' \Sigma (\theta') \ln {\vert \theta -\theta'\vert
\over \theta_c},
\end{equation}
where {\boldmath $\beta$} and {\boldmath $\theta$} are
the angular position vectors (as seen by the observer) of the image
and source, respectively, relative to the lens, $\Sigma (\theta')$ is
the surface mass density of the lens on the lens plane, $\theta_c$
is an arbitrary constant angle,
and $D_{OL}, D_{OS}$ and $D_{LS}$ are the angular
distances between the lens and the observer, the source and the
observer, and the lens and the source, respectively. This equation
has so far been derived only from intuitive geometrical considerations
with use of the
thin lens approximation, but it is not clear what distances should be
used. In order to derive the lens equation from cosmological equations
of light propagation, Sasaki\cite{rf:sasaki93} used the equation of
geodesic deviation.
It is obtained from Eq. (\ref{eq:opt4}) for the connection vector:
\begin{equation}
\label{eq:opt30}
(\delta x^\mu)^{\cdot \cdot} = - R^\mu_{\nu\alpha\beta} k^\nu
k^\beta \delta x^\alpha.
\end{equation}
He solved this equation in the case of an ideal light path, which
is separated into three
regions: the homogeneous, shearless region I (between the observer and
the lens object), the region II including the lens object, and the
homogeneous, shearless region III (between the lens object and the
source). The light path in regions I and III is expressed by the
generalized Dyer-Roeder distance, and the deflection of light rays in
region III is determined using the thin lens approximation. The resulting
lens equation reduces to the usual one with the generalized
Dyer-Roeder distances. In the case that in the regions I and III there
are inhomogeneous matter distributions and the shear effect is not
negligible, however, the usual expression of the lens equation cannot
be used, as was shown by Sasaki.\cite{rf:sasaki93}
In the multi-lens-plane method,\cite{rf:sef}\cite{rf:sw88a}\tocite{rf:sw88b}
inhomogeneities as lens objects are
assumed to be only in the lens planes, and hence the use of the lens
equation in this method is consistent with the above assumption that
the regions I and III are homogeneous and shearless.
However the neglection of gravitational forces due to the difference
of the projected matter distribution from the real distribution may be
comparable with the neglection of weak forces from distant matter
distribution.
\bigskip
\section{Distances and lensing relations}
\subsection{Observation and distances in gravitational lensing}
There are some methods to determine cosmological parameters
by using gravitational lenses.
\cite{Refsdal64a,Refsdal64b,Refsdal66,PG,BN,FFK,rf:fuk92,rf:sef}
Most of them concern the following three typical
observational quantities:
(1) the bending angle, (2) the lensing statistics and (3) the time delay.
It is of great importance to clarify the determination of cosmological
parameters through their observations in the realistic universe.
In particular, it has been discussed that inhomogeneities of
the universe may affect the cosmological tests.
\cite{rf:fs89,rf:zel64,rf:dash66,rf:dyroe72,rf:dyroe73,rf:wts92,rf:sasaki93,rf:sw88a,rf:sw88b,Kantowski,Linder88,KFT,BS91}
In this section, we use the so-called Dyer-Roeder angular diameter
distance in order to take account of the inhomogeneities.
\cite{rf:dyroe72,rf:dyroe73,DR74}
We can consider this distance in two different cases, that of
the so-called filled beam in the Friedmann-Lemaitre-Robertson-Walker (FLRW)
universe, and that of the
so-called empty beam, when the right ray propagates through the empty region.
For comparison with the filled beam, the empty beam has been frequently
used and studied numerically in the literature (for instance, Fukugita et al.
\cite{FFK,rf:fuk92}).
However, it has not been clarified whether the observed quantities and/or
the cosmological parameters for the arbitrary case of
the clumpiness parameter are bounded between those for the filled beam
and the empty beam.
Moreover, numerical investigations have fixed redshifts of the lens and
the source, \cite{AA,rf:fuk92} though the effect of the clumpiness
on the observable depends on the redshifts of the lens and the source.
Therefore, it is important to clarify how the observation of gravitational
lensing depends on all the parameters (the density parameter,
cosmological constant, clumpiness parameter
and redshifts of the lens and the source). For this reason,
we derive the dependence on these parameters. \cite{Asada97,rf:asad98}
\subsection{Distance combinations in gravitational lenses}
\noindent
(1) bending angle
The lens equation is written as \cite{rf:sef}
\begin{equation}
\mbox{\boldmath $\beta$}=\mbox{\boldmath $\theta$}-{D_{\rm LS} \over D_{\rm OS}}
\mbox{\boldmath $\alpha$} .
\label{angle}
\end{equation}
Here, {\boldmath $\beta$} and {\boldmath $\theta$} are
the angular position vectors of the source and image, respectively,
and {\boldmath $\alpha$} is the vector representing the deflection angle.
The effective bending angle $(D_{\rm LS} / D_{\rm OS})\mbox{\boldmath $\alpha$}$
appears when we discuss the observations concerning the angle such as
the image separation and the location of the critical line.
\cite{BN,rf:sef}
Hence the ratio $D_{\rm LS} / D_{\rm OS}$ plays an important role in the discussion
of observations concerning the angle.
It has been argued that, in calculating the bending angle, the density
along the line of sight should be subtracted from the density of
the lens object. \cite{rf:sasaki93}
However, we assume that the density of the lens is much larger than
that along the line of sight, so that the effect of the clumpiness
on {\boldmath $\alpha$} can be ignored.
Thus, we consider only the ratio $D_{\rm LS} / D_{\rm OS}$ in the following.
\noindent
(2) lensing statistics
The differential probability of lensing events is \cite{PG,rf:sef}
\begin{equation}
d\tau=\sigma n_{\rm L} dl ,
\label{statistics}
\end{equation}
where
$n_{\rm L}$ is the number density of the lens, $dl$ is the physical
length of the
depth and $\sigma$ is the cross section, proportional to
$D_{\rm OL}D_{\rm LS} / D_{\rm OS}$.
Since $dl$ depends only on the cosmological parameters in the FLRW universe,
we investigate the combination $D_{\rm OL}D_{\rm LS} / D_{\rm OS}$ in order to
take account of the clumpiness of the matter.
\noindent
(3) time delay
The time delay between two images A and B is written as
\cite{Refsdal64b,rf:sef}
\begin{equation}
\Delta t_{\rm AB}={1+z_L \over c}{D_{\rm OL}D_{\rm OS} \over D_{\rm LS}}
\int^{\rm B}_{\rm A} d\mbox{\boldmath $\theta$}\cdot\Bigl(
{ \mbox{\boldmath $\alpha$}_{\rm A}+\mbox{\boldmath $\alpha$}_{\rm B} \over 2 }
-\mbox{\boldmath $\alpha$}(\mbox{\boldmath $\theta$}) \Bigr) ,
\label{delay}
\end{equation}
where {$\mbox{\boldmath $\alpha$}_{\rm A}$} and {$\mbox{\boldmath
$\alpha$}_{\rm B}$}
are the bending angles at the images A and B, respectively.
\subsection{Monotonic properties}
It is assumed that the affine parameter in the Dyer-Roeder distance
is the same as that in the FLRW universe,\cite{rf:dyroe72,rf:sasaki93}
namely Eq.~$(\ref{eq:opt27})$.
Since $\alpha$ represents the strength of the Ricci focusing
along the line of sight, the DR angular diameter distance is
a decreasing function of $\alpha$ for a fixed redshift,\cite{rf:dyroe73}
that is to say,
\begin{equation}
D_{\rm OL}(\alpha_1) > D_{\rm OL}(\alpha_2) \quad\mbox{for}\quad
\alpha_1 < \alpha_2 .
\label{drtime1}
\end{equation}
\noindent
(1) $D_{\rm LS}/D_{\rm OS}$
It has been shown that the distance ratio $D_{\rm LS}/D_{\rm OS}$ satisfies
\cite{Asada97}
\begin{equation}
{ D_{\rm LS} \over D_{\rm OS} }(\alpha_1) <
{ D_{\rm LS} \over D_{\rm OS} }(\alpha_2) \quad\mbox{for}\quad \alpha_1 < \alpha_2 .
\label{drbend1}
\end{equation}
This is shown as follows.
For fixed $z_{\rm S}$, $\Omega_0$ and $\lambda_0$, the ratio $D_{\rm
LS}/D_{\rm OS}$
can be considered as a function of $z_{\rm L}$, $X_{\alpha}(z_{\rm L})$.
We define $Y_{\alpha}(z_{\rm L})$ as $D_{\rm SL}/D_{\rm OS}$,
where $D_{\rm SL}$ is the Dyer-Roeder distance from the source to the lens.
Owing to the reciprocity, \cite{rf:ether} we obtain
\begin{equation}
Y_{\alpha}(z_{\rm L})={1+z_{\rm S} \over 1+z_{\rm L}}X_{\alpha}(z_{\rm L}) .
\label{reciprocity}
\end{equation}
Since $Y_{\alpha}$ depends on $z_{\rm L}$ only through $D_{\rm SL}$, it obeys
the equation
\begin{equation}
{d^2 \over d{v_{\rm L}}^2}Y_{\alpha}(z_{\rm L})+{3 \over 2}(1+z_{\rm L})^5
\alpha \Omega_0 Y_{\alpha}(z_{\rm L})=0 , \label{raychaudhuri2}
\end{equation}
where $v_{\rm L}$ is an affine parameter at the lens.
Let us define the Wronskian as
\begin{equation}
W(Y_{\alpha_1},Y_{\alpha_2})=Y_{\alpha_1}{d Y_{\alpha_2} \over
d v_{\rm L}}-Y_{\alpha_2}{d Y_{\alpha_1} \over d v_{\rm L}} .
\end{equation}
Then, we obtain
\begin{equation}
{d \over d v_{\rm L}}W(Y_{\alpha_1},Y_{\alpha_2}) < 0
\quad\mbox{for}\quad \alpha_1 < \alpha_2 .
\label{wronskian2}
\end{equation}
Since both $Y_{\alpha_1}$ and $Y_{\alpha_2}$ vanish at $z_L=z_{\rm S}$,
we obtain
\begin{equation}
W(Y_{\alpha_1}(z_{\rm S}),Y_{\alpha_2}(z_{\rm S}))=0 . \label{wronskian3}
\end{equation}
{}From Eqs.~$(\ref{wronskian2})$ and $(\ref{wronskian3})$, we find
\begin{equation}
W(Y_{\alpha_1},Y_{\alpha_2}) >0 , \label{wronskian4}
\end{equation}
where we used the fact that the affine parameter $v$ defined
by Eq.~$(\ref{eq:opt27})$ is an increasing function of $z$.
Equation ($\ref{wronskian4}$) can be rewritten as
\begin{equation}
{d \over d v_{\rm L}} \ln{Y_{\alpha_2} \over Y_{\alpha_1}} > 0 .
\label{wronskian5}
\end{equation}
Since $Y_{\alpha}$ always becomes $1+z_{\rm S}$ at the observer, we find
\begin{equation}
\ln{Y_{\alpha_2}(z_{\rm L}=0) \over Y_{\alpha_1}(z_{\rm L}=0)}=0 .
\label{wronskian6}
\end{equation}
{}From Eqs.~$(\ref{wronskian5})$ and $(\ref{wronskian6})$, we obtain
\begin{equation}
{Y_{\alpha_2} \over Y_{\alpha_1}}>1 . \label{proof2}
\end{equation}
Thus, from Eq.~$(\ref{reciprocity})$, Eq.~$(\ref{drbend1})$ is proved.
{}From Eqs.~$(\ref{angle})$ and $(\ref{drbend1})$, we see the image separation
as well as the effective bending angle {\it increases} with $\alpha$.
\noindent
(2) $D_{\rm OL}D_{\rm LS} / D_{\rm OS}$
Next let us prove that $D_{\rm OL}D_{\rm LS} / D_{\rm OS}$ increases monotonically
with $\alpha$.
We fix $\Omega$, $\lambda$, $z_{\rm L}$ and $z_{\rm S}$.
Then it is crucial to note that the distance from the lens
to the source can be expressed in terms of the distance function from
the observer, $D(z)$, as \cite{Linder88}
\begin{equation}
D_{\rm LS}={c \over H_0} (1+z_{\rm L}) D_{\rm OL}D_{\rm OS} \int^{v_{\rm S}
}_{v_{\rm L}} {dv \over D(z)^2} ,
\label{dls}
\end{equation}
where $H_0$ is the Hubble constant at present.
This can be rewritten as
\begin{equation}
{D_{\rm OL}D_{\rm LS} \over D_{\rm OS}}(\alpha)={c \over H_0}(1+z_{\rm L})
D_{\rm OL}\,^2 \int^{v_{\rm S}}_{v_{\rm L}} {dv \over D(z)^2} .
\label{drstat1}
\end{equation}
The right hand side of this equation depends on $\alpha$ only through
$D_{\rm OL} / D(z)$.
Following reasoning similar to that used in the proof of
Eq.~$(\ref{drbend1})$,
we obtain for $z_{\rm L} < z < z_{\rm S}$
\begin{equation}
{D_{\rm OL} \over D(z)}(\alpha_1) < {D_{\rm OL} \over D(z)}(\alpha_2)
\quad\mbox{for}\quad \alpha_1 < \alpha_2 .
\label{drstat2}
\end{equation}
{}From Eqs.~$(\ref{drstat1})$ and $(\ref{drstat2})$, we obtain
\begin{equation}
{D_{\rm OL}D_{\rm LS} \over D_{\rm OS}}(\alpha_1) <
{D_{\rm OL}D_{\rm LS} \over D_{\rm OS}}(\alpha_2)
\quad\mbox{for}\quad \alpha_1 < \alpha_2 .
\label{drstat3}
\end{equation}
Therefore, the gravitational lensing event rate {\it increases}
with $\alpha$.
\noindent
(3) $D_{\rm LS} / D_{\rm OL}D_{\rm OS}$
Finally, we investigate the combination of distances appearing
in the time delay.
Dividing Eq.~$(\ref{drtime1})$ by Eq. $(\ref{drbend1})$, we obtain
\begin{equation}
{D_{\rm OL}D_{\rm OS} \over D_{\rm LS}}(\alpha_1) >
{D_{\rm OL}D_{\rm OS} \over D_{\rm LS}}(\alpha_2)
\quad\mbox{for}\quad \alpha_1 < \alpha_2 .
\label{drtime2}
\end{equation}
Thus, the time delay {\it decreases} with $\alpha$.
As shown above, the three types of combinations of distances
are monotonic functions of the clumpiness parameter.
However, some of other combinations of distances are not monotonic
functions of $\alpha$, though these combinations may not be necessarily
related with the observation.
For instance, the combination $D_{\rm LS} / \sqrt{c D_{\rm OS} / H_0}$
is not a monotonic function of $\alpha$.
\subsection{Implications for cosmological tests}
We consider three types of the cosmological test which use
combinations of distances appearing in gravitational lensing.
Let us fix the density parameter in order to discuss constraints on
the cosmological constant.
\noindent
(1) $D_{\rm LS}/D_{\rm OS}$
The following relation holds
\begin{equation}
{D_{\rm LS} \over D_{\rm OS}}(\lambda_1) < {D_{\rm LS} \over
D_{\rm OS}}(\lambda_2) \quad\mbox{for} \quad \lambda_1 < \lambda_2 .
\label{lambend1}
\end{equation}
This is shown as follows.
Let us define
\begin{equation}
X_{\lambda}(z_{\rm L})={D_{\rm LS}(\alpha,\Omega_0,\lambda) \over
D_{\rm OS}(\alpha,\Omega_0,\lambda)}
\end{equation}
and
\begin{equation}
Y_{\lambda}(z_{\rm L})={D_{\rm SL}(\alpha,\Omega_0,\lambda) \over
D_{\rm OS}(\alpha,\Omega_0,\lambda)} .
\end{equation}
By the reciprocity,\cite{rf:ether} we obtain
\begin{equation}
Y_{\lambda}(z_{\rm L})={1+z_{\rm S} \over 1+z_{\rm L}}X_{\lambda}(z_{\rm L}) ,
\label{reciprocity2}
\end{equation}
which satisfies
\begin{equation}
{d^2 \over {dv_{\rm L}}^2}Y_{\lambda}(z_{\rm L})+{3 \over 2}
(1+z_{\rm L})^5 \alpha \Omega_0
Y_{\alpha}(z_{\rm L})=0 .
\label{raychaudhuri21}
\end{equation}
For $\lambda_i\,(i=1,2)$, the affine parameter $v_i$ satisfies
\begin{equation}
{dz_{\rm L} \over dv_i}=(1+z_{\rm L})^2 \sqrt{ \Omega_0 z_{\rm L}
(1+z_{\rm L})^2 -\lambda_i
z_{\rm L}(2+z_{\rm L})+(1+z_{\rm L})^2 } .
\label{affine21}
\end{equation}
We define the Wronskian as
\begin{equation}
W(Y_{\lambda_1},Y_{\lambda_2})=Y_{\lambda_1}{d Y_{\lambda_2} \over
d v_2}-Y_{\lambda_2}{d Y_{\lambda_1} \over d v_1} .
\label{wronskian21}
\end{equation}
Then, using Eq.~$(\ref{raychaudhuri21})$, we obtain
\begin{equation}
{d \over d z_{\rm L}}W(Y_{\lambda_1},Y_{\lambda_2}) < 0
\quad\mbox{for}\quad \lambda_1 < \lambda_2 .
\label{wronskian22}
\end{equation}
Since $Y_{\lambda}$ always vanishes at $z_{\rm L}=z_{\rm S}$, we also obtain
\begin{equation}
W(Y_{\lambda_1}(z_{\rm S}),Y_{\lambda_2}(z_{\rm S}))=0 .
\label{wronskian23}
\end{equation}
{}From Eqs.~$(\ref{wronskian22})$ and $(\ref{wronskian23})$, we find
\begin{equation}
W(Y_{\lambda_1},Y_{\lambda_2}) >0
\quad\mbox{for}\quad \lambda_1 < \lambda_2 ,
\label{wronskian24}
\end{equation}
which can be rewritten as
\begin{equation}
{d \over d z_{\rm L}} \ln{Y_{\lambda_2} \over Y_{\lambda_1}} > 0
\quad\mbox{for}\quad \lambda_1 < \lambda_2 .
\label{wronskian25}
\end{equation}
Since $Y_{\lambda}$ always becomes $1+z_{\rm S}$ at the observer, we have
\begin{equation}
\ln{Y_{\lambda_2}(z_{\rm L}=0) \over Y_{\lambda_1}(z_{\rm L}=0)}=0 .
\label{wronskian26}
\end{equation}
{}Finally from Eqs.~$(\ref{wronskian25})$ and $(\ref{wronskian26})$, we obtain
\begin{equation}
{Y_{\lambda_2} \over Y_{\lambda_1}} >1
\quad\mbox{for}\quad \lambda_1 < \lambda_2 .
\label{proof21}
\end{equation}
Thus, Eq.~$(\ref{lambend1})$ is proved.
Equations $(\ref{drbend1})$ and $(\ref{lambend1})$ imply that,
in a cosmological test using the bending angle, the cosmological constant
estimated by use of the distance formula in the FLRW universe is always
{\it less} than that etimated by use of the Dyer-Roeder distance
$(0 \leq \alpha <1)$.
\noindent
(2) $D_{\rm OL}D_{\rm LS}/D_{\rm OS}$
Multiplying Eq. $(\ref{lambend1})$ by
\begin{equation}
D_{\rm OL}(\lambda_1) < D_{\rm OL}(\lambda_2)
\quad\mbox{for}\quad \lambda_1 < \lambda_2 ,
\label{lamstat1}
\end{equation}
we obtain
\begin{equation}
{D_{\rm OL}D_{\rm LS} \over D_{\rm OS}}(\lambda_1) <
{D_{\rm OL}D_{\rm LS} \over D_{\rm OS}}(\lambda_2)
\quad\mbox{for}\quad \lambda_1 < \lambda_2 .
\label{lamstat2}
\end{equation}
Equation $(\ref{lamstat1})$ can be proved, for instance,
in the following manner:
The Dyer-Roeder distance is written as the integral equation
\cite{rf:sw88a,Linder88}
\begin{equation}
D(z;\alpha)=D(z;\alpha=1)+\sum^{\infty}_{i=1} \left[ {3 \over 2}
{c \over H_0} (1-\alpha) \Omega \right]^i \int^z_0 dy K_i(y,z)
D(y;\alpha=1) ,
\label{DRlam2}
\end{equation}
where $K_i(y,z)$ is defined as
\begin{equation}
K_1(x,y)=\left.{dv \over dz}\right|_{z=x} (1+x)^4 D(x,y;\alpha=1)
\label{K1}
\end{equation}
and
\begin{equation}
K_{i+1}(x,y)=\int^y_x dz K_1(x,z)K_i(z,y) .
\label{Ki}
\end{equation}
{}From Eqs. $(\ref{K1})$ and $(\ref{Ki})$,
it is shown that for $x<y$
\begin{equation}
K_i(x,y;\lambda_1) < K_i(x,y;\lambda_2)
\quad \mbox{for} \quad \lambda_1 < \lambda_2 ,
\label{Kilam}
\end{equation}
where we have used the relation
\begin{equation}
D(x,y;\alpha=1,\lambda_1) < D(x,y;\alpha=1,\lambda_2)
\quad \mbox{for} \quad \lambda_1 < \lambda_2 ,
\label{FLRWlam}
\end{equation}
applicable in the FLRW universe. Using Eqs. $(\ref{DRlam2})$,
$(\ref{Kilam})$ and $(\ref{FLRWlam})$,
and the positivity of $K_i$, we obtain Eq. $(\ref{lamstat1})$.
{}From Eqs. $(\ref{drstat3})$ and $(\ref{lamstat2})$, it is found that,
in a cosmological test using the lensing events rate,
the cosmological constant is always {\it underestimated}
by use of the distance formula in the FLRW universe.
\noindent
(3) $D_{\rm LS}/D_{\rm OL}D_{\rm OS}$
When the time delay is measured and the lens object is observed,
$D_{\rm OL}D_{\rm OS} / D_{\rm LS}$ can be determined from
Eq. ($\ref{delay}$).
On the other hand, when we denote the dimensionless distance between
$z_1$ and $z_2$ as $d_{12}=H_0 D_{12} / c$, which does not depend on
the Hubble constant, we obtain
\begin{equation}
{D_{\rm OL}D_{\rm OS} \over D_{\rm LS}}={c \over H_0}
{d_{\rm OL}d_{\rm OS} \over d_{\rm LS}} .
\label{delay2}
\end{equation}
Then, Eq. ($\ref{drtime2}$) becomes
\begin{equation}
{d_{\rm OL}d_{\rm OS} \over d_{\rm LS}}(\alpha_1) > {d_{\rm OL}d_{\rm OS} \over d_{\rm LS}}
(\alpha_2) \quad\mbox{for}\quad \alpha_1 < \alpha_2 .
\label{drtime3}
\end{equation}
Thus, from Eqs.($\ref{delay2}$) and ($\ref{drtime3}$), it is found that
$H_0$ estimated using the Dyer-Roeder distance {\it decreases}
with $\alpha$.
Thus, the Hubble constant can be bounded from below when we have little
knowledge on the clumpiness of the universe.
The lower bound is given by use of the distance
in the FLRW universe.
On the other hand, since the combination $D_{\rm LS} / D_{\rm OL}D_{\rm OS}$
is {\it not} a monotonic function of the cosmological constant,
the relation between the clumpiness of the universe and
the cosmological constant is not simple.
It should be noted that even the assumption of a spatially flat
universe $(\Omega+\lambda=1)$ does not change the above implications
for the three types of cosmological tests, since
the cosmological constant affects the Dyer-Roeder distance formula
only through the relation between $z$ and $v$, Eq. $(\ref{eq:opt27})$.
\subsection{Evolution of clumpiness}
We have taken the clumpiness parameter $\alpha$ as a constant
along the line of sight.
However, as a reasonable extension of the DR distance, $\alpha$ can be
considered as a function of the redshift in order to take account of
the growth of inhomogeneities of the universe.\cite{Linder88}
In proving the monotonic properties, it has never been assumed that
$\alpha$ is constant.
Hence, all the monotonic properties and the implications for cosmological
tests remain unchanged for the variable $\alpha(z)$.
That is to say, when $\alpha_1(z) < \alpha_2(z)$ is always satisfied
for $0<z<z_{\rm S}$, all we must to do is to replace parameters $\alpha_1$ and
$\alpha_2$ with functions $\alpha_1(z)$ and $\alpha_2(z)$ in
Eqs. $(\ref{drbend1})$, $(\ref{drstat3})$ and $(\ref{drtime2})$.
In particular, when $\alpha(z)$ is always less than unity on the
way from the source to the observer, both of the combinations of
distances appearing in (1) and (2) are less, while the combination
in (3) is larger than those in the FLRW universe.
Then, the decrease in the bending angle and the lensing event rate,
and the increase in the time delay hold even for a generalized DR
distance with variable $\alpha(z)$.
\newpage
\section{Average distances and the dispersions in inhomogeneous models}
In this section we describe the statistical behavior of angular diameter
distances in
inhomogeneous model universes at the stage of $0 < z < z_1 (= 5)$.
To derive the distances we use the light rays received by (or emitted
backwards from)
an observer at present (by solving the null-geodesic equation) in the
universes which were produced numerically.
\subsection{Model universes and lens models}
We consider three background models with
$(\Omega_0, \lambda_0) = (1.0, 0), (0.2, 0.8)$ and $(0.2, 0)$.
They are denoted as S, L and O, respectively, which represent the
standard model, a low-density flat model and an open model. The matter is
assumed to contain
particles consisting of galaxies and non-galactic clouds with equal
mass $m$, but generally different sizes. The inhomogeneous models are
given by the method of the $N$-body simulation (using Suto's tree
code)\cite{rf:su}
in periodic boxes with particle number $N = 32^3$.
The initial particle distributions were derived using Bertschinger's
software {\it COSMICS}\cite{rf:bert}
under the condition that their perturbations are given
as random fields with the spectrum of cold dark matter, their power
$n$ is 1, their normalization is specified as the dispersion
$\sigma_8 \ = \ 0.94$, and the Hubble constant is $H_0 = 100 h {\rm
Mpc^{-1} \ km \ s^{-1}}$, where $h = 0.5$ for $(1.0,0)$ and
$h = 0.7$ for other models with $\Omega_0 = 0.2$.
The box sizes for models S, L and O are
\begin{equation}
\label{eq:ba1}
L_0 \equiv a(t_0) l = 32.5 h^{-1}, \ 50 h^{-1}, \ 50 h^{-1}{\rm Mpc},
\end{equation}
and the particle masses are
\begin{equation}
\label{eq:ba2}
m (= \rho_{B0} {L_0}^3/N) = 2.90, \ 2.11, \ 2.11 \times 10^{11} h^{-1}
M_\odot,
\end{equation}
respectively, where $\rho_{B0}$ is the background mass density, $a(t)$
is the scale-factor, and $l$ is the comoving length.
The particle size $r_s \ (= a(t) x_s)$ is given in the form of
softening
radii, which have constant values when we calculate the gravitational
potential for lensing. For $r_s$ we consider the following two (lens)
models:
\noindent \mib{Lens \ model \ 1}. \quad
All particles in the low-density models ($\Omega_0 =
0.2$) have $r_s = 20 h^{-1}$kpc, $20 \%$ of the particles in
the flat model $(1.0, 0)$ have $r_s = 20 h^{-1}$kpc, and the
remaining particles have $r_s = 500 h^{-1}$kpc.
Thus, practically, particles with $\Omega_c = 0.2$ (which we call {\it compact
lens objects}) play the role of lens objects. Their number density is
much larger than the galactic density $\Omega_g \sim 0.02$.
\noindent \mib{Lens \ model \ 2}. \quad
$10 \%$ of the particles in the low-density model ($\Omega_0 = 0.2$)
and $2 \%$ of the particles in the flat model $(1.0, 0)$ have
$r_s = 20 h^{-1}$kpc, while the remaining particles have
$r_s = 500 h^{-1}$kpc. Therefore only galaxies corresponding to
$\Omega_g = 0.02$ play significant roles as lens objects, and the remaining
particles are regarded as diffuse clouds.
The background line-element is
\begin{equation}
\label{eq:ba3}
ds^2 = - c^2 dt^2 + a^2(t)(d\mib{x})^2/\Bigl[1 + K {1 \over 4}
(\mib{x})^2\Bigr]^2,
\end{equation}
and the Poisson equation and null-geodesic equation describing light rays
are given in \S 2 of a separate paper (by Tomita, Premadi and
Nakamura) of this volume.
\subsection{Angular diameter distances}
Here we treat the linear and area distances defined in \S 2
(Eqs. ($\ref{eq:opt22}$) and ($\ref{eq:opt23}$)).
Let us consider a pair of rays received by the observer with the
separation angle $\theta$. By solving null-geodesic equations,
the interval of the two rays at any epoch can be derived.
If $(\Delta \mib{x})_\perp$ is the component of the deviation vector
perpendicular to the central direction of the rays, the linear angular
diameter distance $D_{\rm lA}$ is given as
\begin{equation}
\label{eq:ba4}
D_{\rm lA} = a(t) (\Delta \mib{x})_\perp \Bigl[1 + {1 \over 4} K
(\mib{x})^2\Bigr]^{-1}/\theta,
\end{equation}
where the factor $(1-2\Phi)$ has been neglected, because
$\vert \Phi \vert \ll 1$ locally. The above expression can be
rewritten by use of $y^i \ (\equiv a_0 x^i/R_0)$ as
\begin{equation}
\label{eq:ba5}
D_{\rm lA} = {R_0 \over (1+z) F} (\Delta \mib{y})_\perp/\theta,
\end{equation}
where $F \equiv 1 -{1 \over 4}(R_0H_0/c)^2 (1 -\Omega_0 -\lambda_0)
(\mib{y})^2, \ a_0 = a(t_0) $ and \ $R_0 \equiv L_0/N^{1/3}$.
On the other hand, the area angular diameter distance $D_{\rm aA}$ is
given as follows using three rays (ray 1, ray 2 and ray 3) received by
the observer, such that on the observer plane the two lines between
ray 1 and ray 2, and between ray 1 and ray 3 are orthogonal and have
the same lengths (equal to the separation angle $\theta$).
If $(\Delta \mib{x})_{\perp(12)}, \ (\Delta \mib{x})_{\perp(13)}$ and
$(\Delta \mib{x})_{\perp(23)}$ are the components of the deviation vectors
(between ray 1 and ray 2, \ between ray 1 and ray 3 \ and between
ray 2 and ray 3 ) perpendicular to the central direction of
the rays, we obtain
\begin{equation}
\label{eq:ba6}
D_{\rm aA} = a(t) [(\Delta \mib{x})_{\perp(12)} \cdot (\Delta
\mib{x})_{\perp(13)}]^{1/2} \Bigl[1 + {1 \over 4} K (\mib{x})^2
\Bigr]^{-1}/\theta.
\end{equation}
Using $y^i \ (\equiv a_0 x^i/R_0)$, this reduces to
\begin{equation}
\label{eq:ba7}
D_{\rm aA} = {R_0 \over (1+z) F} \Big[{1 \over 2} \Delta y_{12} \Delta
y_{13} |(\Delta y_{12} - \Delta y_{13} + \Delta y_{23})(\Delta y_{12}
- \Delta y_{13} - \Delta y_{23})|\Big]^{1/4}/\theta,
\end{equation}
where $\Delta y_p \equiv |(\Delta \mib{y})_{\perp (p)}|$ with
$p = 12, 13, 23$.
In a previous paper (Tomita\cite{rf:tom98c}) we investigated the behavior of
$D_{\rm lA}$ for the
separation angle $\theta = 0.005 - 20$ arcsec in various model
universes, and found the dependence of distances on $\theta$ is small for
$\theta \leq 1.0$ arcsec. Here we fix the separation angle to $\theta
= 1.0$ arcsec and
consider the difference between the linear and area distances and
their dependence on the lens models 1 and 2.
In the present lensing simulation we performed the ray-shooting of 500
ray bundles to derive $D_{\rm lA}$ and $D_{\rm aA}$ for each set of two lens
models and three model universes. At the six epochs
$z = 0.5, 1, 2, 3, 4$ and $5$, we compared the calculated distances
with the Dyer-Roeder distance and determined the corresponding value
of the clumpiness parameter $\alpha$ as follows.
In Ref. 17) we calculated $\alpha$ for $0 \leq z \leq 5$ in the above
three model universes and found that the angular diameter distance
depends approximately linearly on $\alpha$ \ (cf. Figs. 3 $\sim$ 6 in
Ref. 17). For $|\alpha -1| \gg 1$ linearity does not hold
(cf. Eq. (3.35)), but most light rays are in the neighborhood of
$\alpha = 1$, as is verified below. Hence we determined $\alpha$
from the calculated distance
$D_{\rm A}~(= D_{\rm lA}$ or $D_{\rm aA})$ using the relation
\begin{equation}
\label{eq:d6}
\alpha = (D_{\rm A} - D_{\rm DR})/(D_{\rm F} - D_{\rm DR}),
\end{equation}
where $D_{\rm DR}$ is the limiting Dyer-Roeder distance with $\alpha = 0$,
and $D_{\rm F}$ is the calculated Friedmann distance in the homogeneous
case. This $D_{\rm F}$ is equal to the Dyer-Roeder distance with
$\alpha = 1$. Moreover, we consider the normalized distances defined by
\begin{equation}
\label{eq:d7}
d_{\rm A} = D_{\rm A}/D_{\rm F}.
\end{equation}
\begin{figure}[htb]
\parbox{.471\textwidth}
\epsfxsize=7.5cm
\epsfbox{dfig1.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm lA}$ in the lens model 1 and model S with $(\Omega_0, \lambda_0) =
(1.0, 0)$. Results for $z = 0.5, 1, 2, 3, 4$ and $5$ are denoted
by dot-long dashed, dot-short dashed, long dashed, short dashed,
dotted and solid lines, respectively.}}
\hspace{-2mm}
\parbox{.471\textwidth}{
\epsfxsize=7.5cm
\epsfbox{dfig2.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm lA}$ in the lens model 2 and model S with $(1.0, 0)$.
The lines have the same meaning as in Fig. 1.
}}
\end{figure}
\begin{figure}[htb]
\parbox{.471\textwidth}
\epsfxsize=7.5cm
\epsfbox{dfig3.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm aA}$ in the lens model 1 and model S with $(1.0, 0)$.
The lines have the same meaning as in Fig. 1. }}
\hspace{-2mm}
\parbox{.471\textwidth}{
\epsfxsize=7.5cm
\epsfbox{dfig4.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm aA}$ in the lens model 2 and model S with (1.0, 0).
The lines have the same meaning as in Fig. 1. }}
\end{figure}
\begin{figure}[htb]
\parbox{.471\textwidth}
\epsfxsize=7.5cm
\epsfbox{dfig5.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm lA}$ in the lens model 1 and model L with $(0.2, 0.8)$.
The lines have the same meaning as in Fig. 1. }}
\hspace{-2mm}
\parbox{.471\textwidth}{
\epsfxsize=7.5cm
\epsfbox{dfig6.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm lA}$ in the lens model 2 and model O1 with (0.2, 0.8).
The lines have the same meaning as in Fig. 1. }}
\end{figure}
\begin{table}
\caption{The average clumpiness parameter $\bar{\alpha}$ and its dispersion
$\sigma_\alpha$, and the average normalized distance $\bar{d}_{\rm A}$
and its dispersion $\sigma_d$ for two lens models
in model S with $(\Omega_0, \lambda_0) = (1.0, 0)$. \
}
\label{table:1}
\begin{center}
\begin{tabular}{cccccccccc} \hline \hline
& & &linear & $(D_{\rm lA})$& & &area &$(D_{\rm aA})$ & \\ \hline
lens & $z$ &$\bar{\alpha} $ & $\sigma_\alpha$ & $\bar{d}_{\rm A}
$ & $\sigma_d$&$\bar{\alpha} $ & $\sigma_\alpha$ & $\bar{d}_{\rm A}
$ & $\sigma_d$
\\ \hline
&0.5& $1.09$ & $0.43$ & 1.00 & 0.018 &$1.09$ & $0.36$ & 1.00 & 0.015\\
&1& $1.03$ & $0.25$ & 1.00 & 0.032 &$1.04$ & $0.22$ & 1.00 & 0.028\\
1&2& $1.02$ & $0.15$ & 0.99 & 0.051 &$1.03$ & $0.41$ & 0.99 & 0.045\\
&3& $1.01$ & $0.11$ & 0.99 & 0.063 &$1.02$ & $0.10$ & 0.99 & 0.056\\
&4& $1.01$ & $0.09$ & 0.99 & 0.073 &$1.02$ & $0.08$ & 0.99 & 0.064\\
&5& $1.01$ & $0.08$ & 0.99 & 0.080 &$1.02$ & $0.07$ & 0.99 & 0.071\\
\hline
&0.5& $1.09$ & $0.28$ & 1.00 & 0.012 &$1.07$ & $0.23$ & 1.00 & 0.009\\
&1& $1.03$ & $0.14$ & 1.00 & 0.019 &$1.03$ & $0.13$ & 1.00 & 0.017\\
2&2& $1.02$ & $0.09$ & 0.99 & 0.029 &$1.02$ & $0.08$ & 0.99 & 0.027\\
&3& $1.01$ & $0.07$ & 0.99 & 0.037 &$1.02$ & $0.06$ & 0.99 & 0.034\\
&4& $1.01$ & $0.06$ & 0.99 & 0.043 &$1.01$ & $0.05$ & 0.99 & 0.040\\
&5& $1.01$ & $0.05$ & 0.99 & 0.048 &$1.01$ & $0.04$ & 0.99 & 0.044\\
\hline
\end{tabular}
\end{center}
\end{table}
\begin{table}
\caption{The average clumpiness parameter $\bar{\alpha}$ and its dispersion
$\sigma_\alpha$, and the average normalized distance $\bar{d}_{\rm A}$
and its dispersion $\sigma_d$ for two lens models
in model L with $(\Omega_0, \lambda_0) = (0.2, 0.8)$. \
}
\label{table:2}
\begin{center}
\begin{tabular}{cccccccccc} \hline \hline
& & &linear & $(D_{\rm lA})$ & & &area & $(D_{\rm aA})$ & \\ \hline
lens & $z$ &$\bar{\alpha} $ & $\sigma_\alpha$ & $\bar{d}_{\rm A}
$ & $\sigma_d$&$\bar{\alpha} $ & $\sigma_\alpha$ & $\bar{d}_{\rm A}
$ & $\sigma_d$
\\ \hline
&0.5& $1.09$ & $1.43$ & 1.00 & 0.019 &$1.02$ & $0.91$ & 1.00 & 0.012\\
&1& $1.07$ & $1.15$ & 1.00 & 0.057 &$1.02$ & $0.69$ & 1.00 & 0.034\\
1&2& $1.02$ & $0.68$ & 1.00 & 0.114 &$1.03$ & $0.49$ & 0.99 & 0.083\\
&3& $1.01$ & $0.50$ & 1.00 & 0.157 &$1.03$ & $0.37$ & 0.99 & 0.118\\
&4& $1.00$ & $0.39$ & 1.00 & 0.188 &$1.02$ & $0.29$ & 0.99 & 0.138\\
&5& $1.00$ & $0.33$ & 1.00 & 0.211 &$1.02$ & $0.24$ & 0.99 & 0.155\\
\hline
&0.5& $1.08$ & $0.43$ & 1.00 & 0.006 &$1.06$ & $0.29$ & 1.00 & 0.004\\
&1& $1.04$ & $0.33$ & 1.00 & 0.016 &$1.03$ & $0.21$ & 1.00 & 0.010\\
2&2& $1.01$ & $0.20$ & 1.00 & 0.034 &$1.01$ & $0.13$ & 1.00 & 0.022\\
&3& $1.01$ & $0.15$ & 1.00 & 0.047 &$1.01$ & $0.10$ & 1.00 & 0.032\\
&4& $1.01$ & $0.12$ & 1.00 & 0.056 &$1.01$ & $0.08$ & 1.00 & 0.039\\
&5& $1.01$ & $0.10$ & 1.00 & 0.064 &$1.01$ & $0.07$ & 0.99 & 0.045\\
\hline
\end{tabular}
\end{center}
\end{table}
\bigskip
\begin{figure}[htb]
\parbox{.471\textwidth}
\epsfxsize=7.5cm
\epsfbox{dfig7.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm aA}$ in the lens model 1 and model O2 with $(0.2, 0.8)$.
The lines have the same meaning as in Fig. 1. }}
\hspace{-2mm}
\parbox{.471\textwidth}{
\epsfxsize=7.5cm
\epsfbox{dfig8.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm aA}$ in the lens model 2 and model O2 with (0.2, 0.8).
The lines have the same meaning as in Fig. 1. }}
\end{figure}
\begin{figure}[htb]
\parbox{.471\textwidth}
\epsfxsize=7.5cm
\epsfbox{dfig9.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm lA}$ in the lens model 1 and model O with $(0.2, 0)$.
The lines have the same meaning as in Fig. 1. }}
\hspace{-2mm}
\parbox{.471\textwidth}{
\epsfxsize=7.5cm
\epsfbox{dfig10.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm lA}$ in the lens model 2 and model O with (0.2, 0).
The lines have the same meaning as in Fig. 1. }}
\end{figure}
\begin{figure}[htb]
\parbox{.471\textwidth}
\epsfxsize=7.5cm
\epsfbox{dfig11.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm aA}$ in the lens model 1 and model O with $(0.2, 0)$.
The lines have the same meaning as in Fig. 1. }}
\hspace{-2mm}
\parbox{.471\textwidth}{
\epsfxsize=7.5cm
\epsfbox{dfig12.ps}
\caption{The percentage ($100 N(\alpha)/ N)$ of the distribution
of $\alpha$ in bins with the interval $\Delta \alpha = 0.4$, for
$D_{\rm aA}$ in the lens model 2 and model L with (0.2, 0).
The lines have the same meaning as in Fig. 1. }}
\end{figure}
\begin{table}
\caption{The average clumpiness parameter $\bar{\alpha}$ and its dispersion
$\sigma_\alpha$, and the average normalized distance $\bar{d}_{\rm A}$
and its dispersion $\sigma_d$ for two lens models
in model O with $(\Omega_0, \lambda_0) = (0.2, 0)$. \
}
\label{table:3}
\begin{center}
\begin{tabular}{cccccccccc} \hline \hline
& & &linear & $(D_{\rm lA})$ & & &area & $(D_{\rm aA})$ & \\ \hline
lens & $z$ &$\bar{\alpha} $ & $\sigma_\alpha$ & $\bar{d}_{\rm A}
$ & $\sigma_d$&$\bar{\alpha} $ & $\sigma_\alpha$ & $\bar{d}_{\rm A}
$ & $\sigma_d$
\\ \hline
&0.5& $1.02$ & $1.99$ & 1.00 & 0.020 &$1.07$ & $1.24$ & 1.00 & 0.012\\
&1& $1.07$ & $1.55$ & 1.00 & 0.046 &$1.12$ & $1.22$ & 1.00 & 0.037\\
1&2& $1.06$ & $1.06$ & 1.00 & 0.084 &$1.09$ & $0.84$ & 0.99 & 0.067\\
&3& $1.04$ & $0.89$ & 1.00 & 0.113 &$1.09$ & $0.74$ & 0.99 & 0.094\\
&4& $1.03$ & $0.82$ & 0.99 & 0.137 &$1.08$ & $0.68$ & 0.99 & 0.115\\
&5& $1.03$ & $0.75$ & 0.99 & 0.156 &$1.07$ & $0.62$ & 0.99 & 0.129\\
\hline
&0.5& $1.08$ & $0.61$ & 1.00 & 0.006 &$1.04$ & $0.44$ & 1.00 & 0.004\\
&1& $1.08$ & $0.49$ & 1.00 & 0.014 &$1.08$ & $0.41$ & 1.00 & 0.012\\
2&2& $1.04$ & $0.31$ & 1.00 & 0.025 &$1.06$ & $0.26$ & 1.00 & 0.021\\
&3& $1.03$ & $0.26$ & 1.00 & 0.033 &$1.06$ & $0.22$ & 0.99 & 0.028\\
&4& $1.02$ & $0.24$ & 1.00 & 0.040 &$1.05$ & $0.20$ & 0.99 & 0.034\\
&5& $1.02$ & $0.22$ & 1.00 & 0.046 &$1.05$ & $0.18$ & 0.99 & 0.038\\
\hline
\end{tabular}
\end{center}
\end{table}
\bigskip
As a result of statistical analysis for this ray-shooting, we
derived the average clumpiness parameter $\bar{\alpha}$, the average
normalized distance $\bar{d}_{\rm A}$, their dispersions
($\sigma_\alpha$ and $\sigma_d$) and the distribution ($N(\alpha)$) of
$\alpha$. Here, in order to study the frequency of ray pairs with
$\alpha$, we consider many bins with the interval $\Delta \alpha =
0.4$ and the centers $\alpha_i = 1.0 \pm 0.4 i \ (i = 0, 1, 2, ...)$.
The number of ray pairs with $\alpha_i - \Delta \alpha/2 \leq \alpha
\leq \alpha_i + \Delta \alpha/2$ is expressed as $N (\alpha_i)$ and the total
number of ray pairs is $N (= \sum_i N(\alpha_i))$.
In Tables I $\sim$ III, $(\bar{\alpha}, \sigma_\alpha,
\bar{d_{\rm A}}, \sigma_d)$ for $D_{\rm lA}$ and $D_{\rm aA}$
in the two lens models are shown for models S, L and O, respectively.
In Figs. 1 $\sim$ 12, the percentages of the distribution of $\alpha$,
that is, $100 N(\alpha)/ N$ for $D_{\rm lA}$ and $D_{\rm aA}$
are shown for the above three models.
The bar graphs in these figures have the same meaning as the line
graphs which were used in the previous paper.\cite{rf:tom98c}
In Figs. 1, 5 and 9, the distributions for six values of $z$ are
shown, and in the other figures those for only $z = 1, 3$ and $5$ are
shown to avoid confusion.
The following types of statistical behavior are found from these
tables and figures :
\medskip
\noindent (1) In all cases, both the average values $\bar{\alpha}$ and
$\bar{d}_{\rm A}$
are nearly equal to $1$, so that the average angular distance can be
regarded as the Friedmann distance.
\noindent (2) For each angular distance, the two kinds of
dispersions have different
behavior: $\sigma_\alpha$ increases in the order of O, L and S for the
same value of $z$, and $\sigma_\alpha$ in a given universe model
decreases with the
increase of $z$. On the other hand, $\sigma_d$ increases in the order
of L, O and S, and $\sigma_\alpha$ in a given universe model increases with the
increase of $z$. This behavior is connected with the situation that
the change in the Dyer-Roeder distance corresponding to the change in
$\alpha$ is small for $\alpha < 1$.
\noindent (3) Generally the dispersions for $D_{\rm lA}$ are larger
than those for $D_{\rm aA}$, and the ratios of two dispersions are
$\sim 1.2$ for model S and $\sim 1.6$ for models L and O.
These differences can be seen also by comparing Figs. 1 and 3, Figs. 2 and
4, ..., and Figs. 10 and 12.
\noindent (4) The dispersions in the lens model 2 are smaller than
those in the lens model 1. The ratios of the former to the latter are
$ \sim 2/3, 1/3, 1/3$ for universe models S, L and O, respectively.
These differences can be seen similarly by comparing Figs. 1 and 2,
Figs. 3 and 4, ..., and Figs. 11 and 12.
In the two lens models of model S and in the lens model 2 of models L
and O, the angular diameter distances can be regarded as the Friedmann
distance, because of the small dispersions. In the lens model 1 of
models L and O, however, we cannot always use the Friedmann distance,
because of comparatively large dispersions.
Using numerical ray-shooting in the $N$-body-simulating clumpy
cosmological models, we studied the statistical behavior of the
angular diameter distances $D_{\rm lA}$ and $D_{\rm aA}$ and
determined the clumpiness parameter
$\alpha$ by comparing it with the Friedmann distance $(\alpha = 1)$ and
the Dyer-Roeder distance $(\alpha > 0)$. Moreover, we studied the
behavior of the normalized distance $d_{\rm A}$. The results show that
all average values of $\alpha$ are nearly equal to 1, the
dispersions of the linear distance are slightly larger than those of
the area distance, and in the lens model 1 of
models L and O, the dispersions are not so small, and we cannot use
the Friedmann distance, while in the other cases we can use the
Friedmann distance because of small dispersions.
In the above averaging process, all light rays were taken into
account. If we consider only weakly deflected light rays as
contributing to weak lensing, the dispersion $\sigma_\alpha$ will
be slightly smaller than the values in the above tables. However, the
contribution of strong lensing to $\sigma_\alpha$ is small because of
its small frequency.
Finally, we touch on the estimate of lensing correction to source
magnitudes based on Tables I $\sim$ III. Using $\sigma_d$ in the area
distance it is given by
\begin{equation}
\label{eq:d8}
\Delta m (z) = {5 \over 2} \log [1 + 2 \sigma_d] = 2.18 \sigma_d
\end{equation}
for the source at $z$. Its values for the lens model $(1, 2)$ are
$\Delta m (0.5) = (0.033,
0.020), (0.026, 0.009)$ and $(0.026, 0.009),$
$\Delta m (1.0) =
(0.061, 0.037), (0.074, 0.022)$ and $(0.081, 0.026)$
\noindent for models S, L and O, respectively. This $\Delta m$ (for
the separation angle $\theta = 1$ arcsec) was derived independently of
$\Delta m$ (for the separation angle $\theta = 2$ arcsec), shown in the
\S 3 of Tomita, Premadi and Nakamura's paper in this Supplement,
but their values at $z = 1$ are roughly consistent.
The lensing correction to source magnitudes was also investigated by
Holz in different lens models and inhomogeneous models.\cite{rf:holz98}
\bigskip
\section{Shear effect on distances}
As we have shown in \S 2, the evolution of a cross sectional
area of a light ray bundle is determined by Ricci and Weyl focusing
along the trajectory of the ray bundle.
Ricci focusing is a convergency effect due to matter in the ray
bundle. On the other hand, Weyl focusing is a result of the tidal
shear on the ray bundle induced by the inhomogeneous distribution of
matter.
One of the main difficulties in deriving a distance-redshift relation
analytically for a realistic inhomogeneous universe lies in estimating
the effect of Weyl focusing on ray bundles.
Since the pioneering work of Gunn, \cite{gu67} there has been a great
deal of progress in constructing
an analytical approach to investigate statistical quantities of
realistic distances
(e.g., dispersion and skewness of probability distribution of image
magnifications).
Babul and Lee, \cite{ba91} among others, examined lensing magnification
effects on distances due to the large scale structures
($\geq 0.5h^{-1}$Mpc, where Hubble constant $H_0 = 100 h$km/sec/Mpc).
They found that the dispersion in image magnifications is negligible
even for sources at a redshift of 4.
At the same time, they pointed out that the dispersion is
very sensitive to the nature of the matter distribution on small
scales.
In their study, the effects of Weyl focusing were neglected based,
on a numerical study by Jaroszy\'nski et al.\ \cite{ja90} in which the
lensing magnification effects due to large scale structure ($\geq
1h^{-1}$Mpc) in cold dark matter models were examined,
and it was concluded that Weyl focusing has no significant effect on
image magnifications.
Frieman \cite{fr97} improved Babul and Lee's study to reflect the recent
developments in numerical and observational studies of large scale
structure.
Although he took small scale nonlinear structure into account,
Weyl focusing effects were neglected without any reasonable
basis.
Nakamura \cite{rf:naka97} examined the effects of shear on image
magnification
in the cold dark matter model universe with a linear density perturbation.
He found that the effect is sufficiently small and concluded that
Weyl focusing can be safely neglected for a light ray passing through
linear density inhomogeneities.
The above cited studies mainly focus on large scale inhomogeneities,
whereas the effects of small scale objects, such as galaxies and
clusters of galaxies, have not been satisfactorily taken into account.
It is, however, not clear whether the Weyl focusing effect due
to small scale inhomogeneities has a significant effect on
distances.
In this section, we discuss the Weyl focusing effect due
to small scale inhomogeneities, mainly following Hamana. \cite{ha99}
\subsection{Basic equations}
We first derive an evolution equation of lensing magnification from
the null geodesic equation, \cite{ha98} which is equivalent to the
optical scalar
equations,\cite{rf:sasaki93,se94} and is convenient to examine
gravitational lensing effects.
We rewrite the cosmological Newtonian metric (\ref{eq:int2}) as
\begin{equation}
\label{conf-metric}
ds^2 = a^2(\eta)\left[ -(1+2 \Phi) d\eta^2 + (1-2 \Phi) \gamma_{ij}
dx^i dx^j \right],
\end{equation}
where $\eta$ is a conformal time: $d\eta \equiv cdt$.
We write the above metric as $g_{\mu \nu} = a^2 \hat{g}_{\mu \nu}$.
Since the light cone structure is invariant
under the conformal transformation of the metric, in the following we
work in conformally related $\hat{g}$ world.
Let us consider an infinitesimal bundle of light rays
intersecting at the observer. We denote a connecting vector
which connects the fiducial light ray $\gamma$ to one of its neighbors
as $\xi^\mu$.
All gravitational focusing and shearing effects on the infinitesimal
light ray bundle are described by the geodesic deviation equation,
\begin{equation}
\label{geodev}
{{d^2 \xi^{\mu}} \over {d \lambda^2}} = -{R^{\mu}}_{\alpha \nu \beta}
\xi^{\nu} k^{\alpha} k^{\beta},
\end{equation}
where $k^{\alpha}=dx^{\alpha}/d\lambda$, and $\lambda$ is the affine
parameter along the fiducial light ray $\gamma$. We introduce a dyad
basis $e_{\mu}{}^A$ ($A$, $B$, $C,...=1,2$) in the two-dimensional screen
orthogonal to $k^{\mu}$ and parallel-propagated along $\gamma$.
The screen components of the connection vector are given by
\begin{equation}
\label{Y}
Y^A = e_{\mu}{}^A \xi^{\mu}.
\end{equation}
{}From the geodesic deviation equation (\ref{geodev}), one can
immediately find that $Y^A$ satisfies the Jacobi differential equation
\begin{equation}
\label{jaco}
{{d^2 Y_A}\over {d\lambda^2}} = {\cal{T}}_{A B} Y^B,
\end{equation}
where ${\cal{T}}_{AB}$ is the so-called optically tidal
matrix.~\cite{se94} From the metric (\ref{conf-metric}), up to first
order in $\Phi$, this matrix is given by
\begin{equation}
\label{tid}
\mbox{{\boldmath${\cal{T}}$}} =
-K \mbox{{\boldmath${\cal{I}}$}} - \left(
\begin{array}{ll}
{\cal{R}} + \mbox{Re}\left[{\cal{F}}\right] &
\mbox{Im}\left[{\cal{F}}\right] \\
\mbox{Im}\left[{\cal{F}}\right] &
{\cal{R}} - \mbox{Re}\left[{\cal{F}}\right]
\end{array}
\right),
\end{equation}
where {\boldmath${\cal{I}}$} is the $2 \times 2$ identity matrix, and
${\cal{R}}$ and ${\cal{F}}$ represent the Ricci and Weyl focusing
induced by the density inhomogeneities, respectively:
\begin{eqnarray}
\label{Ricci}
{\cal{R}} &=& \Delta^{(3)} \Phi ={{3 \Omega_0 H_0^2}
\over {2}} {{\delta} \over a},\\
\label{Wely}
{\cal{F}} &=& \Phi_{,11}-\Phi_{,22} + 2 i
\Phi_{,12} .
\end{eqnarray}
Here $\Delta^{(3)}$ is the Laplacian operator in the spatial section,
and $\delta$ is the density contrast defined by $\delta \equiv
(\rho-\rho_b)/\rho_b$, where $\rho_b$ is the mean matter density.
Owing to the linearity of (\ref{jaco}),
the solution of $Y^A$ can be written in terms of its initial value
$dY^A/d\lambda |_{\lambda=0} = \vartheta^A$ and the $\lambda$-dependent
linear transformation matrix ${\cal{D}}_{AB}$ can be written as
\begin{equation}
\label{D}
Y^A(\lambda) = {\cal{D}}^A{}_B(\lambda) \vartheta^B.
\end{equation}
Substituting the last equation into the Jacobi differential equation
(\ref{jaco}), we obtain
\begin{equation}
\label{evoD}
{{d^2{\cal{D}}_{AB}} \over {d \lambda^2}} = {\cal{T}}_{AC} {\cal{D}}_{CB}.
\end{equation}
Now, we derive an evolution equation of the lensing magnification
matrix relative to the smooth Friedmann distance from (\ref{evoD}).
First, we write (\ref{tid}) as
{\boldmath${\cal{T}}$}$ = ${\boldmath${\cal{T}}$}${}^{(0)} +
${\boldmath$\delta{\cal{T}}$}, with {\boldmath${\cal{T}}$}${}^{(0)} =
-K${\boldmath${\cal{I}}$}, and {\boldmath$\delta {\cal{T}}$} is the
second term in (\ref{tid}). In the homogeneous case,
{\boldmath$\delta{\cal{T}}$} is vanishing, and the solution of
{\boldmath${\cal{D}}$} is ${\cal{D}}_{AB}(\lambda) = D_f(\lambda)
\delta_{AB}$, where $D_f$ is, of course, the
standard angular diameter distance in the background Friedmann
universe.
It is natural to define the lensing magnification matrix relative to the
corresponding Friedmann universe as
\begin{equation}
\label{defM}
{\cal{M}}_{AB} (\lambda) \equiv {{{\cal{D}}_{AB} (\lambda)} \over {D_f
(\lambda)}} .
\end{equation}
Differentiating ${\cal{M}}_{AB}$ twice with respect to $\lambda$ and
using (\ref{evoD}), one finds
\begin{equation}
\label{difM}
{{d^2{\cal{M}}_{AB}} \over {d\lambda^2}} = -{2 \over {D_f}} {{d D_f}
\over {d \lambda}} {{d {\cal{M}}_{AB}} \over {d\lambda}} + \delta
{\cal{T}}_{AC} {\cal{M}}_{CB}.
\end{equation}
With the initial conditions {\boldmath${\cal{M}}$}$(\lambda) |_{\lambda=0} =
${\boldmath${\cal{I}}$} and $d${\boldmath${\cal{M}}$}
$(\lambda)/{d\lambda} |_{\lambda=0} = ${\boldmath${\cal{O}}$}, \cite{se94}
the last equation can be written in the integral form
\begin{equation}
\label{Mab}
{\cal{M}}_{AB} (\lambda) = \delta_{AB} + \int_0^{\lambda} d \lambda'
{{D_f(\lambda-\lambda') D_f(\lambda')} \over {D_f(\lambda)}} \delta
{\cal{T}}_{AC}(\lambda') {\cal{M}}_{CB} (\lambda').
\end{equation}
This is the general form of the evolution equation of the lensing magnification
matrix relative to the Friedmann distance in multiple gravitational
lensing theory.~\cite{rf:sef}
Note, in general, this equation is not an explicit equation for
${\cal{M}}_{AB}$, since it involves an integration over the
optical tidal matrix evaluated
on the light ray path, such that one first has to solve a null
geodesic equation.
Since, for almost all cases of cosmological interest, the
deflection angle is very small,\cite{rf:fs89}
we will neglect the deflection of
light rays.
\subsection{Order-of-magnitude estimate}
We now examine the magnitude of lensing effects on light ray bundles due
to randomly distributed virialized objects (e.g.\ galaxies and
clusters of galaxies) adopting an order-of-magnitude
estimate. \cite{rf:fs89,pe93}
As can be seen in Eq. (\ref{Mab}), if the magnitude of the
components of the matrix,
$\int d \lambda ' D_f(\lambda-\lambda') D_f(\lambda')/ D_f(\lambda) \delta
${\boldmath$\cal{T}$}$(\lambda')$ is small,
the magnitude of the lensing effects is dominated by these terms.
We, therefore, examine the magnitude of these terms.
For simplicity, we denote these terms as $\delta {\cal{M}}$.
Supposing that lensing objects are randomly distributed and that each
has a mass $M = 2 {\sigma_v}^2 l /G$,
where ${\sigma_v}$ is the one-dimensional velocity dispersion of the
lens objects, and $l$ is a characteristic comoving size of a lens
object.
Hence the mean comoving number density of the lens objects is
\begin{equation}
n_L ={{3 \Omega_L {H_0}^2} \over {16 \pi}} {1\over{\sigma_v}^{2}}
{1\over {l}},
\end{equation}
where
$\Omega_L$ is the density parameter of lens objects defined by
$\Omega_L \equiv \rho_L/(3 H_0^2/8 \pi G)$, with $\rho_L$ is the mean
density of the lens objects.
Thus, the mean comoving separation distance is
\begin{equation}
r_0 = \left({{ 16 \pi} \over {3 \Omega_L H_0^2}} \sigma_v^2 l
\right)^{1\over 3}.
\end{equation}
Then for a geodesic affine comoving distance of $\lambda$, the light
ray gravitationally encounters such objects
$N_g = \lambda/r_0$
times on average.
At each encounter, the contribution to the lensing magnification matrix is
\begin{eqnarray}
\Delta {\cal{M}} &=& 4\pi \left( {{\sigma_v} \over c} \right)^2
{{r_0} \over {b^2}} {{D_f(\lambda_d)
D_f(\lambda_s-\lambda_d)} \over {D_f(\lambda_s)}}\nonumber\\
&\sim& 4\pi \left( {{\sigma_v} \over c} \right)^2
{1 \over {r_0}} {{D_f(\lambda_d)
D_f(\lambda_s-\lambda_d)} \over {D_f(\lambda_s)}},
\end{eqnarray}
where $D_f(\lambda_i-\lambda_j)$ is the comoving angular diameter
distance, the subscripts $d$ and $s$ indicate the lens and source,
respectively, and $b$ is the comoving impact parameter.
In the above expression, we have assumed that the mean comoving impact
parameter is of order $r_0$.
Since the sign of each contribution will be random, the total
contribution to the lensing magnification matrix is
\begin{eqnarray}
\delta {\cal{M}} &\sim&
\Delta {\cal{M}} \sqrt{N_g}\nonumber\\
&\sim&
\sqrt{3 \over 4} \sqrt{\Omega_L} \left[ 4\pi
\left ({{\sigma_v} \over c}\right)^2 {1\over {l}} {c \over {H_0}}
\right]^{1\over 2} \left\langle{{{H_0} \over c} {{{D_f(\lambda_d)
D_f(\lambda_s- \lambda_d)}}
\over {D_f(\lambda_s)}}
}\right\rangle
\left[{{H_0} \over c} \lambda \right]^{1\over 2}\nonumber\\
\label{magoft}
&\sim& \sqrt{\Omega_L} \sqrt{\nu}
\left\langle{{{H_0} \over c} {{{D_f(\lambda_d)
D_f(\lambda_s- \lambda_d)}}
\over {D_f(\lambda_s)}}
}\right\rangle
\left[{{H_0} \over c} \lambda \right]^{1\over 2},
\end{eqnarray}
where $\nu$ is a compactness parameter of a lens object defined by
$\nu \equiv 4 \pi (\sigma_v/c)^2 l^{-1} \\
\times (c/H_0)$.
The contribution from direct encounters can be similarly
estimated by noting that the average number of encounters is
\begin{equation}
N_d ={{l^2 \lambda} \over {r_0^3}} = {{3 \Omega_L {H_0}^2}
\over {16 \pi}} {1\over{\sigma_v}^2} l \lambda,
\end{equation}
with each encounter contributing
\begin{equation}
\Delta {\cal{M}}_d = 4\pi \left({{\sigma_v} \over c} \right)^2
{1\over l} {{D_f(\lambda_d) D_f(\lambda_s-\lambda_s)} \over
{D_f{\lambda_s}}},
\end{equation}
with random sign.
The result turns out to be the same as that of gravitational distant
encounters, given by Eq. (\ref{magoft}).
The comoving affine distance $\lambda$ becomes $c/H_0$ at the source redshift
$z_s \sim 3$, and the averaged value of the distance combination over the
lens redshifts is of order
\begin{equation}
\left\langle {{H_0} \over c}
{{D_f(\lambda_d)D_f(\lambda_s-\lambda_d)}\over {D_f(\lambda_s)}}
\right\rangle \sim {\cal{O}}(0.1).
\end{equation}
Accordingly, we find that the magnitude of the total contribution of
the lensing effects to the lensing magnification matrix
scales as $\sim 0.1 \sqrt{\Omega_L} \sqrt{\nu}$ for the
source redshift $z_s \sim 3$. We have the relation
$\Omega_L \leq \Omega_0$ by definition, and $\Omega_0$ appears to
be less than unity. Thus $\Omega_L \leq 1$.
On the other hand, $\nu \le 1$ for
galaxies and clusters of galaxies.
Therefore a typical value of gravitational lensing effects on the
lensing matrix can be expected to be $ {\cal{O}}(0.1)$ or
smaller for a majority of random lines of sight.
The lensing magnification factor of a point like image is defined by
the determinant of the lensing magnification matrix.
Taking the determinant of the lensing magnification matrix
(\ref{Mab}), and expanding it in powers of $\delta{\cal{M}}$,
one can easily find that the leading term of Weyl focusing effects
is of order $\delta{\cal{M}}^2$.
On the other hand, that of the Ricci focusing term is of order
$\delta{\cal{M}}$.
Since we have seen that a typical value of $\delta {\cal{M}}$ is expected to
be ${\cal{O}}(0.1)$ or smaller, we
can conclude that, at least from a statistical point of view, Weyl
focusing has no significant effects on the image
magnifications or equivalently on the distances.
\subsection{Numerical investigation}
The above argument may sound too naive.
One of the authors (T.H.), numerically investigated Weyl focusing
effects on image magnifications by using the multiple
gravitational lens theory.\cite{ha99}
He focused on gravitational lensing effects due to small scale
virialized objects, such as galaxies and clusters of galaxies.
He considered a simple model of an inhomogeneous universe.
The matter distribution in the universe was modeled by randomly
distributed isothermal objects.
He found that, for the majority of the random lines of sight, Weyl
focusing has no significant effect, and the image magnification of
a point like source within a redshift of 5 is dominated by Ricci
focusing.
He also found that his result agrees well with the
order-of-magnitude estimate given above.
To summarize, we conclude that, except for a statistically very rare
kind of light ray, Weyl focusing has no significant effect on
image magnifications or equivalently on the distances.
\section{Concluding remarks}
Lensing observation in inhomogeneous universes was discussed
in \S 3, based on the so-called Dyer-Roeder distance, in which
one of the main assumptions is neglecting Weyl focusing.
Such a neglection seems correct in our universe, as was shown in
the \S 5.
The average angular distances in inhomogeneous model universes are the
Friedmann distances, as was shown in \S 4, but individual ray
bundles have various values of clumpiness parameters $\alpha$ because
of their dispersions. The observational quantities are sensitively
dependent on $\alpha$, as was shown in the \S 3, and so they may
have dispersions similar to $\alpha$.
The difference between linear and area angular diameter distances, which
is caused by the shear, is generally small in accord with the result
in \S 5, even though we considered small-scale
inhomogeneities, but the difference in the low-density models is
found to be larger than that in the Einstein-de Sitter model. This
implies that the shear effect is comparatively larger in the low-density
models.
\section*{Acknowledgements}
K.T. would like to thank Y.~Suto for helpful discussions about
$N$-body simulations. His numerical computations were performed on
the YITP computer system.
H.A.\ and T.H.\ would like to thank T.~Futamase for fruitful discussions.
H.A. would like to thank M.~Kasai, M.~Sasaki and T.~Tanaka
for useful conversations.
|
\section{Introduction}
It has been conjectured long ago that quark confinement in $QCD$ would arise
because the electric flux lines are squeezed into long flux
tubes by the condensation of the magnetic charge \cite{H1,M}.
This mechanism of confinement is known as the dual Meissner effect, since
it is dual, in the sense of electric/magnetic duality, to the confinement of
magnetic fluxes, that arises in a type-two superconductor, because of the
condensation of the electric charge \cite{H2}. \\
Since the electric flux lines would span two-dimensional surfaces
embedded into the four-dimensional space-time, the dual Meissner effect
leads to an effective theory of $QCD$ in terms of closed strings \cite{Po2}. \\
The absolute confinement of the electric flux requires indeed that the flux
line cannot break into an open string. \\
More analytically, the string functional integral would arise as the string
solution of the Migdal-Makeenko equation \cite{MM,J} in the large-$N$ limit of
$QCD$ \cite{H4, W}.
This is the string program, that has received recently a new revival,
most notably because of the implementation in the string setting of the
zig-zag symmetry \cite{Po2}. \\
A distinctive feature of the string program is that the existence of the
string is assumed as an ansatz for the solution of the Migdal-Makeenko equation.
In fact it is difficult to see how the strings would arise directly in terms
of the functional integral over the four-dimensional gauge connections. \\
A remarkable achievement in this direction was the representation of the
partition function
of the two-dimensional gauge theory as a sum over branched coverings
of the two-dimensional space-time \cite{GT}. These coverings are interpreted
as the string world sheets, giving evidence in favour
of the string solution of pure gauge theories. \\
Yet this representation is obtained from the exact result for the
two-dimensional partition function, without a direct link to gauge
configurations in the functional integral. \\
In an unrelated development, the cotangent bundle of the moduli
of holomorphic bundles on a Riemann surface has played a key role in the
Seiberg-Witten solution of the Coulomb branch of some four-dimensional
supersymmetric gauge theories \cite{SW}. \\
Branched covers appear in these solutions as the spectral curves of
the characteristic equation associated to a holomorphic one-form that
labels cotangent directions to the moduli of holomorphic bundles \cite{DW}. \\
The pre-potential, the unique holomorphic function
that determines the low-energy effective action of the supersymmetric
theory, is constructed by means of the spectral curve. \\
More precisely, certain submanifolds of the cotangent bundle, that
correspond to moduli of representations of the fundamental group of the
underlying two-dimensional base manifold, admit an integrable fibration
by Jacobians of branched coverings of the base two-dimensional manifold,
the Hitchin fibration, that in turn is equivalent to assign the pre-potential.
\\
While no direct link to physical four-dimensional fields may be attributed
to these coverings in the framework of the Seiberg-Witten solution,
a link to the string program would possibly arise, if the cotangent
bundle of unitary connections in two-dimensions could be embedded into
the four-dimensional $QCD$ functional integral.
Such an embedding was found in \cite{MB1}. \\
It was found there that the correct variables to define this embedding are
neither the four-dimensional gauge connections, $A$, nor their dual
variables,
$A^D$, but a partial mixing of them, that correspond to a partial
or fiberwise duality transformation \cite{MB1}. \\
The coordinates of the cotangent bundle of unitary connections, $T^*\cal{A}$,
appear naturally as the shift $A^D=A+\Psi$ is performed for two dualized
polarizations among the four components of the four-dimensional gauge
connection. \\
In addition, it was found in \cite{MB2}, that there is a dense embedding into
the $QCD$ functional integral, of an elliptic fibration of the
moduli space of parabolic $K(D)$ pairs into (an elliptic fibration of) the
quotient of the cotangent bundle by the action of the gauge group. \\
The last space admits a Hitchin fibration by the moduli of line bundles
over branched spectral covers, thus giving a dense embedding of these
objects into the $QCD$ functional integral. \\
While in \cite{MB2} the integrability properties of the Hitchin fibration
were used to reduce the problem of computing the functional integral in the
large-$N$ limit to the evaluation of the saddle-point of a certain effective
action that contains the Jacobian of the change of variables to the collective
field of the Hitchin fibration, in this paper we shall address the
following,
more qualitative issue, that relates to the string program. \\
What is the locus in the functional integral of the confining branch of
$QCD$, that is, what is the locus in the moduli space of parabolic $K(D)$
pairs, whose image by the Hitchin map contains only Riemann surfaces spanned
by closed strings ? \\
A partial answer to this question was given in \cite{MB3}.
In \cite{MB3} a physical interpretation of the occurrence of Hitchin
bundles in the fiberwise dual functional integral was given, in the light
of 't Hooft concept of Abelian projection \cite{H5}. \\
This interpretation identifies the branch points of the spectral covers as
magnetic monopoles and the parabolic points as electric charges. \\
Since confinement requires magnetic condensation and 't Hooft alternative
excludes electric condensation, the confining branch is the locus, in the
parabolic $K(D)$ pairs, whose image by the Hitchin map has no parabolic
singularity on the spectral cover \cite{MB3}, a not completely trivial
condition. \\
It should be noticed that this idea is in complete analogy with the
two-dimensional case \cite{GT}, in which the partition function is localized
on branched coverings of the base compact space-time, without parabolic points.
In fact the occurrence of parabolic points would imply the presence
in the vacuum to vacuum amplitudes of string diagrams with the topology of open strings,
a situation that it is appropriate to the Coulomb rather than the
confinement phase. This last statement may be exemplified thinking to a
sphere with two parabolic points as a topological cylinder, a vacuum diagram of
an open string theory. \\
We will find in this paper that the confinement locus is characterized
precisely by the condition that the residues of the Higgs current, $\Psi$, on the
parabolic divisor be nilpotent. \\
This condition turns out to be equivalent to the existence of a
(dense in the large-$N$ limit) hyper-Kahler reduction of the cotangent bundle
of unitary connections under the action of the gauge group. \\
The confining branch of $QCD$ is, therefore, the hyper-Kahler locus of the
Hitchin fibration of parabolic bundles, embedded in the $QCD$ path integral
as prescribed by fiberwise duality. \\
On the other side, this is precisely the locus for which spectral
covers with the topology of closed string diagrams, but not open ones,
occur in the functional integral. The dual mechanism of superconductivity
and the string interpretation are therefore compatible, as it should be, and
as it has been for long time believed \cite{Po2}. \\
One more comment.
It is a rather strange fact that the same or analogue
objects, that are used to construct the Seiberg-Witten solution of
four-dimensional $SUSY$ theories in the Coulomb branch, appear here as giving
rise to a physical string interpretation of the $QCD$ functional integral,
with an associated hyper-Kahler structure but no supersymmetry. \\
In fact we think that the explanation of this fact has much to do with
duality as opposed to supersymmetry. \\
The Seiberg-Witten solution starts from supersymmetry, through the
structure theorem for the low-energy effective action, as determined by the
pre-potential, and ends up with a non-linear geometric realization
of the Abelian electric magnetic/duality of the effective theory in the
Coulomb branch, in terms of a Legendre transformation of the pre-potential
\cite{SW}. \\
We start instead from the non-Abelian duality of the microscopic theory, as
defined by the functional integral, to gain, by
means of fiberwise duality and the embedding of parabolic
bundles, control over the large-$N$ limit \cite{MB2} and a mathematical
realization of the dual Meissner effect \cite{MB3} at the same time.
\section{ The nilpotent condition}
In this section we show that the spectral covers that are in the image by
the Hitchin map of parabolic $K(D)$ pairs have no parabolic divisor if and only
if the levels of the non-hermitian moment maps are nilpotent on each point of the
parabolic divisor. \\
This in turn is a necessary and sufficient condition for the moduli
space of parabolic $K(D)$ pairs to admit a hyper-Kahler structure.
In \cite{K,MB2} a special name was used to characterize this closed subspace:
parabolic Higgs bundles. \\
In any case the confinement criterium of this paper explains the physical
meaning of the hyper-Kahler structure, a mathematical condition whose
meaning was suspected to be physically relevant but not elucidated in
\cite{MB2}. Indeed, there it was argued that the two cases of the parabolic
$K(D)$ pairs and of the parabolic Higgs bundles present equivalent difficulties
from the point of view of solving the large-$N$ limit, in fact differing by
contributions of order of $\frac{1}{N}$. We now argue that parabolic Higgs
bundles correspond to the confining branch of $QCD$ in the fiberwise-dual
variables. \\
The functional integral for $QCD$ in \cite{MB2} is defined in terms of the
variables $(A_z, A_{\bar z}, \Psi_z, \Psi_{\bar z})$, obtained by means of a
fiberwise duality transformation from $(A_z, A_{\bar z}, A_u, A_{\bar u})$,
where $(z, \bar z, u, \bar u)$ are the complex coordinates on the product of two
two-dimensional tori, over which the theory is defined. \\
$(A_z, A_{\bar z}, \Psi_z, \Psi_{\bar z})$ define the coordinates of an
elliptic fibration of $T^* {\cal A}$, the cotangent bundle of unitary
connections on the $(z, \bar z)$ torus with the $(u, \bar u)$ torus as a base.
\\
The set of
pairs $(A, \Psi)$ that are solutions of the following differential equations
(elliptically fibered over the $(u, \bar u)$ torus)
is embedded into the space of parabolic $K(D)$ pairs \cite{MB3,MB2}:
\begin{eqnarray}
F_A-i \Psi \wedge \Psi &=& \frac{1}{|D|}\sum_p \mu^{0}_{p} \delta_p i dz
\wedge d\bar{z} \nonumber \\
\bar{\partial}_A \psi &=& \frac{1}{|D|}\sum_p \mu_{p} \delta_p dz
\wedge d\bar{z}\nonumber \\
\partial_A \bar{\psi} &=& \frac{1}{|D|}\sum_p \bar{\mu}_{p} \delta_p
d\bar{z} \wedge dz\;
\end{eqnarray}
where $\delta_p$ is the two-dimensional delta-function localized at $z_p$
and $(\mu^{0}_{p},\mu_{p},\bar{\mu}_{p})$ are the set of levels for the
moment maps \cite{MB2}.
The space of parabolic $K(D)$ pairs consists of a parabolic bundle with a
holomorphic connection $\bar{\partial}_{A}$ and a parabolic morphism ${\psi}$.
Eq.(1) defines a dense stratification of the functional integral over
$T^* {\cal A}$ because the set of levels is dense everywhere in function
space, in the sense of the distributions, as the divisor $D$ gets larger and
larger. \\\
According to Hitchin \cite{Hi}, there is a Hitchin fibration of parabolic
$K(D)$ pairs, defined by U(1) bundles over the following spectral cover:
\begin{eqnarray}
Det( \lambda1 - \Psi_z)=0
\end{eqnarray}
The spectral cover depends only from the eigenvalues of $\Psi_z$.
The condition that the spectral cover has no parabolic point is therefore
the condition that the eigenvalues of $\Psi_z$ have no poles.
We notice that the residues of the poles of $\Psi_z$ are determined
by the levels of the non-hermitian moment maps.
In fact $\Psi_z$ can be made meromorphic with residue at the point $p$
conjugated to
the level $\mu_p$ by means of a gauge transformation $G$ in the
complexification of the gauge group, that gauges to zero the connection
$\bar{A}_z$,
fiberwise:
\begin{eqnarray}
&&\bar{\partial} \psi- \frac{1}{|D|}\sum_p G \mu_{p}G^{-1} \delta_p
dz \wedge d\bar{z}=0
\nonumber \\
&&\partial \bar{\psi}- \frac{1}{|D|}\sum_p \bar{G}^{-1}\bar{\mu}_{p}\bar{G}
\delta_p d\bar{z} \wedge dz=0 \;
\end{eqnarray}
From this equation it follows that the residues of the eigenvalues of $\Psi_z$
are proportional to the eigenvalues of $\mu_{p}$.
If the eigenvalues of $\psi$ have no poles on the covering, $\mu_p$ must have
zero eigenvalues and therefore must be nilpotent and vice versa, that is the
conclusion looked forward. \\
There is however an apparent puzzle. Though the eigenvalues of $\psi$ cannot
have poles on the covering if the levels of the non-hermitian moment maps are
nilpotent, the traces of
powers of $\Psi_z$, that are expressed through symmetric polynomials in the
eigenvalues, certainly are meromorphic functions on the torus.
How can this happen if the eigenvalues of $\psi$ have no poles on the covering?
The answer is the following, as we have found by a direct check in the
$SU(2)$ case.
If $\mu_p$ is nilpotent, the eigenvalues of $\Psi_z$ have singularities that are
not parabolic but that look in the coordinates of the $z$ torus branched
singularities, for example $z^{-\frac{1}{2}}$. However we should remind the
reader that the eigenvalues of $\psi$ are really differentials on the covering.
Therefore $z^{-\frac{1}{2}}$ should be really interpreted as
$z^{-\frac{1}{2}}dz$, that is
$d(z^{\frac{1}{2}})$, that is, in fact, smooth on a simply branched covering.
There is no singularity on the covering. \\
Yet, symmetric powers of the eigenvalues of $\Psi_z$
may have meromorphic singularities on the torus.\\
It remains to show that if the residue of $\psi$ is nilpotent the quotient is
hyper-Kahler. This is a known result \cite{K}.
This concludes our proof. \\
In fact a slightly stronger statement holds.
If the residues of the Higgs field are nilpotent, Eq.(1) can be interpreted
as the vanishing condition for the moment maps of the action of the compact
$SU(N)$ gauge group on the pair $(A, \Psi)$ and on the cotangent space of
flags \cite{L}. The quotient under the action of the compact gauge group of
the set:
\begin{eqnarray}
&&F_A-i \Psi \wedge \Psi - \frac{1}{|D|}\sum_p \mu^{0}_{p} \delta_p i dz
\wedge d\bar{z}=0 \nonumber \\
&&\bar{\partial}_A \psi- \frac{1}{|D|}\sum_p n_{p} \delta_p
dz \wedge d\bar{z}=0 \nonumber \\
&&\partial_A \bar{\psi}- \frac{1}{|D|}\sum_p \bar{n}_{p} \delta_p
d\bar{z} \wedge dz=0 \;
\end{eqnarray}
with fixed eigenvalues of the hermitian moment map is, by a general result
\cite{Hi2}, the same as the quotient
defined by the complex moment maps:
\begin{eqnarray}
&&\bar{\partial}_A \psi- \frac{1}{|D|}\sum_p n_{p} \delta_p
dz \wedge d\bar{z}=0 \nonumber \\
&&\partial_A \bar{\psi}- \frac{1}{|D|}\sum_p \bar{n}_{p} \delta_p
d\bar{z} \wedge dz=0\;
\end{eqnarray}
under the action of the complexification of the gauge group.
\section{Conclusions}
Our conclusion is that if $QCD$ confines the electric charge the functional
integral in the fiberwise dual-variables defined in \cite{MB1, MB2}
must be localized on the hyper-Kahler locus of parabolic $K(D)$ pairs,
the parabolic Higgs bundles.
This space is characterized by a nilpotent residue of the Higgs current.
These are precisely the parabolic $K(D)$ pairs whose image by the
Hitchin map contain spectral covers arbitrarily branched, but with no
parabolic points. \\
The physical interpretation is that there is a monopole condensate in the
vacuum but no electric condensate and
only closed electric strings occur into vacuum to vacuum diagrams.
|
\section{Introduction}
Research involving mobile wireless ATM is advancing rapidly.
One of the earliest proposals for a wireless ATM
architecture is described in \cite{Raychaudhuri}.
In this paper, various alternatives for a wireless Media Access Channel
(MAC) are discussed and a MAC frame is proposed. The MAC
contains sequence numbers, service type, and a Time of Expiry (TOE)
scheduling policy as a means for improving real-time data traffic
handling. A related work which considers changes to Q.2931
\cite{Q2931} to support mobility is proposed in \cite{Yuan}.
A MAC protocol for wireless ATM is examined in \cite{McTiffin}
with a focus on Code Division Multiple Access (CDMA) in which
ATM cells are not preserved allowing a more efficient form of
packetization over the wireless network links. The ATM cells
are reconstructed from the wireless packetization method after
being received by the destination. The Rapidly Deployable Radio Network
Project (RDRN) architecture described in this paper maintains standard
ATM cells through the wireless links.
Research work on wireless ATM LANs have been described in \cite{Rednet}
and \cite{SWAN}. The mobile wireless ATM RDRN differs from these
LANs because the RDRN uses point-to-point radio communication over much
longer distances.
The system described in \cite{BAHAMA} and \cite{Veer} consists of Portable
Base Stations (PBS) and mobile users. PBSs are base stations which perform
ATM cell switching and are connected via Virtual Path Trees which are
preconfigured ATM Virtual Paths (VP). These trees can change based on
the topology as described in the {\em Virtual Trees Routing Protocol}
\cite{VTRP}. However, ATM cells are forwarded along the Virtual Path
Tree rather than switched, which differs from the ATM standard. An
alternative mobile
wireless ATM system is presented
in this paper which consists of a mobile PNNI architecture based on a
general purpose predictive mechanism known as Virtual Network Configuration
that allows seamless rapid handoff.
The objective of the Rapidly Deployable Radio Network (RDRN) effort
is to create an ATM-based wireless communication system that
will be adaptive at both the link and network levels to allow for rapid
deployment and response to a changing environment. The objective of the
architecture is to use adaptive point-to-point topology to gain the
advantages of ATM for wireless networks. A prototype of this system has
been implemented and will be demonstrated over a wide area
network. The system adapts to its environment and can automatically arrange
itself into a high capacity, fault tolerant, and reliable network.
The RDRN architecture is composed of two overlaid networks:
\begin{itemize}
\item a low bandwidth, low power omni-directional network for location
dissemination, switch coordination, and management which is the orderwire
network described in this paper,
\item a ``cellular-like'' system for multiple end-user access to the switch
using directional antennas for spatial reuse, and
and a high capacity, highly directional, multiple beam network for
switch-to-switch communication.
\end{itemize}
The network currently consists of two types of nodes, Edge Nodes (EN) and
Remote Nodes (RN) as shown in Figure \ref{scenario}. ENs where designed to
reside on the edge of a wired network and provide access to the wireless
network; however, EN also has wireless links.
The EN components include Edge Switches (ES) and optionally an ATM switch,
radio handling the ATM-based communications, packet radio for the low speed
orderwire running a protocol based on X.25 (AX.25), GPS receiver, and a
processor. Host nodes or remote nodes (RN) consist of the above, but do
not contain an ATM switch.
The ENs and RNs also include a phased array steerable antenna. The RDRN uses
position information from the GPS for steering antenna beams toward nearby
nodes and nulls toward interferers, thus establishing the high capacity links
as illustrated in Figure \ref{hwowov}. Figure \ref{hwowov} highlights an
ES (center of figure) with its omni-directional transmit and receive
orderwire antenna and an omni-directional receive and directional transmit
ATM-based links. Note that two RNs share the same $45^o$ beam from the
ES and that four distinct frequencies are in use to avoid interference.
The decision involving which beams to establish and which frequencies
to use is made by the topology algorithm which is discussed in a later
section.
\begin{figure*}[htbp]
\centerline{\psfig{file=figures/scenario.eps,width=6.0in}}
\caption{RDRN High-level Architecture.}
\label{scenario}
\end{figure*}
The ES has the capability of switching ATM cells among connected RNs or
passing the cells on to an ATM switch to wire-based nodes. Note that the
differences between an ES and RN are that the ES performs switching and
has the capability of higher speed radio links with other Edge Switches
as well as connections to wired ATM networks.
\begin{figure*}[htbp]
\centerline{\psfig{file=figures/hwoverview.eps,width=6.0in}}
\caption{RDRN Component Overview.}
\label{hwowov}
\end{figure*}
The orderwire network uses a low power, omni-directional channel, operating at
19200 bps, for signaling and communicating node locations to other network
elements. The orderwire aids link establishment between the ESs and between
the RNs and ESs, tracking remote nodes and determining link quality.
The orderwire operates over packet radios and is part of the Network
Control Protocol (NCP)\footnote{The Simple Network Management Protocol
(SNMP) Management Information Base (MIB) for the NCP operation as well as
live data from the running prototype RDRN system can be
retrieved from \htmladdnormallink{http:/\-/\-www.ittc.ukans.edu/\-$\sim$sbush/\-rdrn/\-ncp.html}{http://www.tisl.ukans.edu/~sbush/rdrn/ncp.html}.}.
An example of the user data and orderwire network topology is shown in
Figure \ref{owov}. In this figure, an ES serves as a link between
a wired and wireless network, while the remaining ESs act as wireless
switches. The protocol stack for this network is shown in
Figure \ref{hsstack}.
The focus of this paper is on the NCP and in particular on the orderwire
network and protocols. This includes protocol layer
configuration, link quality, hand-off, and host/switch assignment
along with information provided by the GPS system such as position
and time. The details of the user data network will be covered in
this paper only in terms of services required from, and interactions
with, the NCP.
Section \ref{WatmReqS} provides a more detailed description of the RDRN
system, with a focus on the requirements and interaction of each protocol
layer with the NCP. Operation of the NCP is described in Section \ref{initimp}.
A new concept known as Virtual Network Configuration (VNC) is explained
in Section \ref{vnc} along with an example application of
a Mobile Private Network-Network Interface (PNNI) enhanced with VNC. The
development and implementation of the NCP is described along with initial
timing results in Section \ref{emulation}. In Section \ref{NCPanal},
an analysis of NCP indicates the performance of NCP as the system is
scaled up. Finally, emulation results are presented in Section \ref{emres}.
\begin{figure*}[htbp]
\centerline{\psfig{file=figures/owov.eps,width=6.0in}}
\caption{Example Orderwire Topology.}
\label{owov}
\end{figure*}
\begin{figure}[htbp]
\centerline{\psfig{file=figures/protos.eps,width=3.25in}}
\caption{Wireless ATM Protocol Stack.}
\label{hsstack}
\end{figure}
\subsection{Bandwidth required for the Orderwire Network}
\label{owbw}
The traffic over the orderwire was analyzed to determine a relation between
the maximum update rate and the number of RNs. The protocol used for contention
resolution on the broadcast channel is the Aloha Protocol which is known to
have a maximum efficiency close to 18\%. Given the bandwidth of the orderwire
channel, size of an orderwire packet and this value for the efficiency, we
compute and plot the value for the maximum update rate (in packets per minute)
for a given number of RNs. The plot of Figure \ref{owcaphandreal} shows the
variation in update rate for between 5 and 30 RNs. This study gives us
an upper limit on the number of RNs that can be supported over the orderwire
given a minimum required update rate and handoff rate.
\begin{figure*}[htbp]
\centerline{\psfig{file=figures/owcap_hand_real.ps,width=6.0in}}
\caption{Orderwire Traffic Analysis {\bf without} Virtual Network Configuration.}
\label{owcaphandreal}
\end{figure*}
\subsection{ATM Network Configuration Layer}
This section briefly describes how ATM VCs are setup by the NCP.
As the orderwire network determines the topology
of all nodes in the wireless segment (e.g., RNs, ESs) in our architecture,
and establishes
link connectivity among adjacent nodes, setup is still required of the
actual ATM circuits on which wireless ATM are carried on the user
data overlay network. This is accomplished by providing standard
ATM signaling capabilities to RNs and ESs and using Classical IP over ATM
\cite{RFC1577} to associate ATM VCs to IP addresses. The Classical IP
over ATM implementation provided works for PVCs and SVCs (using ATMARP).
Since an ES may connect to multiple RNs (wireless connections) or
ATM switches (wired connections), it can be thought of as a software-based
ATM switch.
In this sense, an ES features ATM PNNI signaling while an RN features
ATM UNI signaling. By default, an RN creates one wireless-ATM protocol
stack and establishes an ATM VC signaling channel on such a stack; however,
the stack is initially in an inactive state (i.e., non-operational mode)
since there is no link connectivity to another node established yet.
Likewise, an ES creates a predefined number of wireless-ATM protocol
stacks -- acting like ports in an ATM switch -- and establishes ATM VC
signaling channels on all configured stacks which are also initialized
as inactive. Wireless-ATM protocol stacks are controlled
by a daemon, called the adaptation manager, which acts on behalf of the
orderwire network. The adaptation manager daemon not only controls the
stacks by setting their state to either active or inactive (default),
but also may modify configuration parameters of the stacks to
provide dynamic adaptation to link conditions.
Two possible scenarios illustrate the interactions
between the orderwire network and the wireless-ATM network.
In the first scenario the orderwire detects link connectivity between
an adjacent pair of nodes (e.g., RN-ES or ES-ES). In this case,
the orderwire network requests an
inactive stack from the adaptation manager daemon at each end and
associates them with a designated address. Upon establishment of
link connectivity, a requested wireless stack has its state set to
active and is ready to operate. Note that since the signaling channels
are preconfigured on the stacks in question, users on the wireless
establish end-to-end connections exactly as if they were connected
in a wired ATM network. The other scenario occurs when the orderwire
network detects a broken connection, at the link level, between
two connected nodes. This case is typical of an RN moving away from the
connectivity range of an ES. The orderwire network thus contacts the
adaptation manager daemon at each end to set the wireless stacks
in question to inactive. Since a wireless stack is never destroyed,
it can be reused in a future request from the orderwire to establish
connectivity to another pair of nodes.
\subsection{ATM Layer}
\label{ATM}
The protocol on the Edge Switch (ES) will remove
ATM cells from the AHDLC frames and switch them to
the proper port. It will also
pack ATM cells into an Adaptive HDLC frame to send to the radio.
The ATM Device Driver API and Adaptive Driver are detailed in
\cite{BushRDRN}. Note that standard ATM call setup signaling is used
and no AAL is precluded from use.
\subsection{Link Layer}
\label{link}
In this architecture ATM will be carried end-to-end. However, at the
edge between the wired (high-speed) network and wireless links,
multiple ATM cells will be combined into an HDLC-like frame. These
frames comprise the Adaptive HDLC (AHDLC) protocol.
The wireless data link layer
is adaptive to provide an appropriate trade-off between data rate and
reliability in order to support the various services.
For example, we may want to drop voice packets, which are
time sensitive, but retry data packets. The edge interface unit makes
this decision based on
knowledge of the requirements of each traffic stream, possibly based
on virtual circuit number.
For some types
of traffic, error correction may be achieved using
retransmission. Here, delay is increased for this class of traffic to
prevent cell losses. It is well known that even a few cell losses can
have a significant impact on the performance of TCP/IP
(Transmission Control Protocol/Internet Protocol), while TCP/IP
can cope with variable delays \cite{Caceres}. The
Adaptive High Level Data Link Control (AHDLC) protocol can
change in response to traffic requirements. ATM end-to-end provides
the following benefits:
\begin{enumerate}
\item Moderate cut-through, e.g. an IP segment may contain 8192
bytes or about 170 cells, while one ATM HDLC-like frame will contain on the
order of 3-20 cells
\item ATM is a standard protocol.
\item ATM can incorporate standardized Quality of Service (QoS) parameters which
could be based on the virtual circuit identifier.
\end{enumerate}
The link layer must also maintain cell order; this will be critical
during hand-off of an RN from one ES to another. Details of the
Adaptive HDLC protocol and frame structures can be found in
\cite{BushRDRN} and \cite{EwyRDRN} .
\subsection{Physical Layer}
\label{phys}
The physical layer includes all hardware components and the wireless
connections. This includes the high speed radios, orderwire packet
radios, ATM switch, antennas, and additional processor for
configuration and setup. This layer provides a raw pipe
for the data link layer described in the next section.
Directional beams from a single antenna are used to
obtain spatial reuse and Time Division Multiple Access (TDMA) is used
to provide access to multiple RNs within a beam.
The physical layer details can be found in \cite{EwyRDRN}.
The NCP sets up the physical layer wireless connections.
\subsection{Network Layer}
\label{ip}
This section of the architecture is concerned with the Internet
Protocol and how it relates to ATM and mobility. This layer provides a
well known and widely used network layer, whose primary purpose is to
provide routing between subnetworks and service for the
Transmission Control Protocol (TCP) and User Datagram Protocol (UDP)
transport layers. The relation between IP and ATM is still an open
issue. {\it Classical IP and ARP over ATM} \cite{RFC1577} (CLIP) is an
initial standard solution. However, it has several weak points such as
requiring a router to connect Logical IP Subnetworks (LIS) even when
they are directly connected at the ATM level, and requiring an ATM
Address Resolution Protocol (ARP) server to provide address resolution
for a single LIS. The
{\it Non-Broadcast Multiple Access (NBMA) Next Hop Resolution Protocol}
(NHRP)\cite{NBMA} provides a better solution but it is still in
draft form.
The RDRN architecture has implemented CLIP and supports both PVCs and
SVCs via ATMARP.
\section{Wireless ATM-Based Network Configuration Requirements}
\label{WatmReqS}
This section provides a brief overview of the high speed
protocol architecture for the RDRN wireless ATM network
\cite{BushRDRN}.
The purpose is to introduce the RDRN network
and more importantly to identify the requirements that each layer
will have for the network configuration protocol.
\section{Network Control Protocol Overview}
\label{initimp}
An initial implementation of the RDRN Network Control Protocol (NCP)
for the prototype system is presented next. The physical layer of the high
speed radio connection has a corresponding layer in the NCP, as
shown in Table \ref{prototab}. The following is a description and ordering
of events for the establishment of the wireless connections.
\begin{table*}[htbp]
\centering
\begin{tabular}{||l|l|l||} \hline
{\bf Protocol Layer} & {\bf Packet Types} & {\bf Packet Contents} \\ \hline \hline
Physical Layer & MYCALL & Callsign, Start-Up-Time \\ \hline
& NEWSWITCH & {\em empty packet} \\ \hline
& SWITCHPOS & GPS Time, GPS Position \\ \hline
& TOPOLOGY & Callsigns and Positions of each node \\ \hline
& USER\_POS & Callsign, GPS Time, GPS Position \\ \hline
& HANDOFF & Frequency, Time Slot, ES GPS Position \\ \hline
\end{tabular}
\caption{\label{prototab} Network Control Protocol Packets.}
\end{table*}
\subsection{Physical Layer of the Network Control Protocol}
\label{owphys}
At the physical level we will be using the orderwire to exchange
position, time and link quality information and to setup the wireless
connections. The process of setting up the wireless connections
involves setting up links between ESs and between ESs and RNs.
The network will have one master ES, which will run
the topology configuration algorithm \cite{cla} and distribute the resulting
topology information to all the connected ESs over point-to-point
orderwire packet radio links. In the current prototype the point-to-point link layer
for the orderwire uses AX.25 \cite{AX.25}.
The master ES is initially the first active ES,
and any ES has the capability of playing the role of the
master.
The first ES to become active initially broadcasts its
callsign and start-up-time in a {\bf MYCALL} packet, and listens for responses
from any other ESs. In this prototype system, the packet radio callsign is
assigned by the FCC and identifies the radio operator.
Since it is the first active ES, there would be no responses in a given time
period, say T. At the end of T seconds, the ES rebroadcasts its {\bf MYCALL}
packet and waits another T seconds. At the end of
2T seconds, if there are still no responses from other ESs, the
ES assumes that it is the first ES active and takes on
the role of the master. If the first two or more ESs start up
within T seconds of each other, at the end of the interval T, the ESs
compare the start-up times in all the received
{\bf MYCALL} packets and the ES with the oldest start-up time
becomes the master. In this system, accurate time stamps are provided by
the GPS.
Each successive ES that becomes active initially
broadcasts its callsign in a {\bf MYCALL} packet. The master on
receipt of a {\bf MYCALL} packet extracts the callsign of the source,
establishes a point-to-point link to the new ES and
sends it a {\bf NEWSWITCH} packet. The new ES on
receipt of the {\bf NEWSWITCH} packet over a point-to-point orderwire link, obtains
its position from its GPS receiver and sends its position to the master as a
{\bf SWITCHPOS} packet over the point-to-point orderwire link.
On receipt of a {\bf SWITCHPOS} packet, the master records the position
of the new ES in its switch position table, which is a table of ES positions, and
runs the topology configuration algorithm \cite{cla} to determine the best
possible interconnection of all the ESs. The master then distributes the
resulting information to all the ESs in the form of a {\bf TOPOLOGY} packet
over the point-to-point orderwire
links. Each ES then uses this information to setup the
inter-ES links as specified by the topology algorithm. The master
also distributes a copy of its switch position table to all the ESs
over the point-to-point orderwire links, which they can use in configuring RNs
as discussed below. This sequence of operations is illustrated in Figure
\ref{pap1} and Figure \ref{pap3}.
Also, the ES then uses the callsign information in the
switch position table to setup any additional point-to-point orderwire
packet radio
links corresponding to the inter ES links required to exchange any link
quality information. Thus this scheme results in
a point-to-point star
network of orderwire links with the master at the center of the star and also point-to-point
orderwire links between
those ESs that have a corresponding inter ES link, as shown in Figure
\ref{owov}.
In the event of failure of the master node which can be detected by listening
for the AX-25 messages generated on node failure, the remaining ESs exchange
{\bf MYCALL} packets, elect a new master node, and the network of ESs is
reconfigured using the topology configuration algorithm \cite{cla}.
\begin{figure}[htbp]
\centerline{\psfig{file=figures/pap1.eps,width=3.25in}}
\caption{State Diagram for Master EN.}
\label{pap1}
\end{figure}
\begin{figure}[htbp]
\centerline{\psfig{file=figures/pap3.eps,width=3.25in}}
\caption{State Diagram for EN not serving as Master.}
\label{pap3}
\end{figure}
Each RN that becomes active obtains its position from
its GPS receiver and broadcasts its position as a {\bf USER\_POS}
packet over the orderwire network. This packet is received by all
the nearby ESs. Each candidate ES then computes the distance between
the RN and all the candidate ESs which is possible since each ES has the
positions of all the other ESs from the switch position table. An initial
guess at the best ES to handle the RN is the closest ES. This ES
then feeds the new RN's position information along with the
positions of all its other connected RNs to a beamforming
algorithm that returns the steering angles for each of the beams on the ES
so that all the RNs can be configured. If the beamforming algorithm
determines that a beam and TDMA time slot
are available to support the new RN,
the ES
steers its beams so that all its connected RNs and the new
RN are configured. It also records the new RN's position in its user position
table which contains positions of connected RNs, establishes a
point-to-point orderwire link to the new RN and sends it a {\bf HANDOFF} packet
with link setup information indicating that the RN is
connected to it. If the new RN cannot be accommodated, the ES sends it
a {\bf HANDOFF} packet with the callsign of the next closest ES, to which
the RN sends another {\bf USER\_POS} packet over a point-to-point
orderwire link. This ES then uses the beamform algorithm to
determine if it can handle the RN. Figure \ref{pap2} shows the
states of operation and transitions between the states for a RN.
This scheme uses feedback from the beamforming algorithm together with
the distance information to configure the RN. It should be noted that the
underlying AX.25 protocol \cite{AX.25} provides error free transmissions over
point-to-point orderwire links. Also the point-to-point orderwire link can be
established from either end and the handshake mechanism for setting up such
a link is handled by AX.25. If the RN does
not receive a {\bf HANDOFF} packet within a given time it uses a
retry mechanism to ensure successful broadcast of its {\bf USER\_POS}
packet.
\begin{figure}[htbp]
\centerline{\psfig{file=figures/pap2.eps,width=3.25in}}
\caption{State Diagram for RN.}
\label{pap2}
\end{figure}
A point-to-point orderwire link is retained as long as a
RN is connected to a particular ES and a
corresponding high-speed link exists between them to enable exchange
of link quality information. The link can be torn down when the
mobile RN migrates to another ES in case of a
hand-off. Thus at the end of this network configuration process, three overlaid
networks are setup, namely, an orderwire network, an RN to ES network and
an inter-ES network. The orderwire network has links between the master ES and
every other active ES in a star configuration, links between ESs connected by
inter-ES links as well as links between RNs and the ESs to which they are connected, as
shown in Figure \ref{owov}.
Raw pipes for the user data links between RNs and appropriate ESs as well as for the
user data links between ESs are also set up.
\section{Virtual Network Configuration for a Rapidly Deployable Network}
\label{vnc}
In order to make RDRN truly rapidly deployable, configuration at all layers has
to be a dynamic and continuous process. Configuration can be a
function of such factors as load, distance, capacity and permissible
topology, all of which are constantly changing in a mobile environment.
A Time Warp \cite{Jefferson} based algorithm is used to anticipate
configuration changes and speed the reconfiguration process.
\subsection{Virtual Network Configuration Algorithm}
The Virtual Network Configuration (VNC) algorithm is an application of a
more general mechanism called Time Warp Emulation (TWE). Time Warp
Emulation is a modification of Time Warp \cite{Jefferson}.
The motivation behind TWE is to allow the actual components of a real-time
system to work ahead in time in order to predict
future behavior and adjust themselves when that behavior does not
match reality. This is accomplished by realizing that there are now two
types of {\em false} messages, those which arrive in the past relative
to the process's Local Virtual Time (LVT) and those messages which have
been generated which are time-stamped with the current real time, but
whose values exceed some tolerance from the component's current value.
The basic Time Warp mechanism is modified by adding a verification
query phase. This phase occurs when real time matches the receive time
of a message in the output queue of a process. In this phase, the physical
device being emulated in time is queried and the results compared with the value
of the message. A value exceeding a prespecified tolerance will cause a
rollback of the process.
\subsection{Virtual Network Configuration Overview}
The Virtual Network Configuration (VNC) algorithm can be explained
by an example. A remote node's direction, velocity, bandwidth used, number
of connections, past history and other factors can be used to approximate a new
configuration sometime into the future. All actual configuration
processes can begin to work ahead in time to where the remote node is
expected to be at some point in the future. If the prediction is
incorrect, but not far off, only some processing will have to be
rolled back in time. For example, the beamsteering process results may have to
be adjusted, but the topology and many higher level requirements will still
be correct. Working ahead and rolling back to adjust for error
with reality is an on-going process, which depends on the tradeoff
between allowable risk and amount of processing time allowed into the
future. As a specific example, consider the effects of hand-off on TCP
performance as described in \cite{Caceres}. In this work, throughputs
were measured for hand-off under various conditions and determined
to degrade badly.
\begin{comment}
\begin{table}[hbp]
\centering
\begin{tabular}{||l|l||} \hline
{\bf Type of Hand-off} & {\bf Bandwidth} \\ \hline \hline
No Hand-offs & 100\% \\ \hline
Overlapping hand-offs & 94\% \\ \hline
0-second rendezvous delay & 88\% \\ \hline
1-second rendezvous delay & 69\% \\ \hline
\end{tabular}
\caption{\label{hoff} Bandwidth Loss Due to Hand-off.}
\end{table}
\end{comment}
\subsection{Virtual Network Configuration Implementation}
The effort required to enhance the network configuration algorithm
to include Virtual Network Configuration is minimal. Three new fields are
added to each existing message in Table \ref{prototab}: antimessage toggle,
send time, and receive time. Physical processes include beamforming,
topology acquisition, table updates, and all processing required for
configuration. Each physical process is assigned a tolerance. When
the value of a real message exceeds the tolerance of a predicted
message stored in the send queue, the process is rolled back.
Also, an additional packet type was created for updating an approximation
of the Global Virtual Time (GVT). Because the system is composed of
asynchronously executing logical processes, each working ahead as
quickly as possible with its own local notion
of time, it is necessary to calculate the time of the system as a whole.
This system-wide time is the GVT. The difference between GVT and
current time is the amount of lookahead, $\Lambda$. Although $\mbox{GVT}
\ge t$ where $t$ is real time, $\Lambda$ is required because
it is used to control the efficiency and accuracy of the system. Since
the network configuration
system uses a master node as described in the physical layer setup, this
is a natural centralized location for a centralized GVT update method.
RNs transmit their LVT to the master, the master calculates an approximate
GVT and returns the result.
An estimate of the additional load on the orderwire packet radios
using VNC is shown in Figure \ref{vncbw}. It is assumed that virtual
messages are 65 bits
longer than real messages and there is one virtual message for each real
message. The figure shows the prototype 19,200 bps orderwire
link capacity as a function of the number of RNs, the position update
rate of each RN, and the hand-off rate. The capability of the orderwire
to support these rates without
VNC is discussed later in detail and is shown in
Figure \ref{owcaphandreal}. Comparing Figures \ref{vncbw} and
\ref{owcaphandreal}, it is apparent that the VNC slightly more than
doubles the orderwire load. However enough capacity remains to support
users with a reasonable position update rate and handoff rate with
this relatively low 19,200 bps orderwire bandwidth.
\begin{figure*}[htbp]
\centerline{\psfig{file=figures/owcap_hand_vnc.ps,width=6.0in}}
\caption{Orderwire Traffic Analysis {\bf with} Virtual Network Configuration.}
\label{vncbw}
\end{figure*}
\section{Summary}
\label{conclusion}
This paper described the design of a control and management network
for a mobile wireless ATM network. The orderwire system consists of a
packet radio network which overlays the mobile wireless ATM network and
receives GPS information. This information is used to control the
beamforming antenna subsystem which provides for spatial reuse.
This paper also proposed the design of the VNC algorithm which is a
novel concept for predictive configuration. A mobile ATM
PNNI based on VNC was also discussed.
As a prelude to the system implementation, results
of a Maisie simulation of the orderwire system were presented.
Finally, the Network Control Protocol was tested, initial problems
corrected, and initial performance results were obtained and
presented in this paper.
\subsection{Timing Results}
\label{timres}
This section summarizes the results of initial timing experiments that were
undertaken to examine the performance of the orderwire system. The
experiments involved determining the time required to transmit and process
each of the packet types listed in Table \ref{prototab} using the real packet
radios. These times represent the time to packetize, transmit, receive,
and depacketize each packet at the Network Control Protocol process.
Figure \ref{timepr} illustrates the physical configuration used for the
experiments involving the real packet radios. The results are presented in
Table \ref{timetab}.
Most of the overhead occurs during the initial system configuration which
occurs only once as long as ESs remain stationary.
With regard to a handoff, the 473 millisecond time to transmit and process
the handoff packet is on the same order of time as that required to compute
the beam angles and steer the beams. The following sections provide an
analysis and discuss the impact of scaling up the system on the configuration
time.
\begin{figure*}[htbp]
\centerline{\psfig{file=figures/timepr.eps,width=6.0in}}
\caption{Physical Setup for Packet Radio Timing.}
\label{timepr}
\end{figure*}
\begin{table}
\centering
\begin{tabular}{||l|c||} \hline
{\bf Event} & {\bf Time (ms)} \\ \hline \hline
USER\_POS & 677 \\ \hline
NEWSWITCH & 439 \\ \hline
HANDOFF & 473 \\ \hline
MYCALL & 492 \\ \hline
SWITCHPOS & 679 \\ \hline
TOPOLOGY & 664 \\ \hline
\end{tabular}
\caption{\label{timetab} Network Control Protocol Timing Results.}
\end{table}
\section{Development and Implementation}
\label{emulation}
The initial physical layer network control protocol design was done
using Maisie \cite{bagrodia},
a C-based parallel programming language. It facilitates creation
of entities which execute in parallel and the ability to easily
send and receive messages between entities.
A Maisie emulation of the entire network was developed which uses the actual
NCP code. This helped build confidence
that the design of the Network Control Protocol was correct.
The network control protocol code was initially tested with only the two
packet radios available. Since at least three packet radios are necessary
for a complete RN-ES-RN orderwire connection, the next step involved
emulating the packet radios via TCP/IP over Ethernet, and completing
the development of the code. The packet radio emulation also allowed
testing of various configurations that helped determine if the network control protocol
was scalable.
The physical layer of the Network Control Protocol is a single-unit
consumable resource system. There can be no
deadlock since there are no cycles. All message interactions take place
with a master switch,
except for the initial {\bf MYCALL} packet broadcast.
The GPS system was also emulated to provide the appearance of
mobility so that hand-offs of a host from one ES to another
could be tested. The GPS emulation is also an important component of
the Virtual Network Configuration Algorithm. The actual orderwire code is
used in these emulations.
\section{NCP Performance Analysis}
\label{NCPanal}
The analysis of the RDRN network configuration time using the protocols
proposed earlier, will be divided into three phases.
Phase I is the ES-ES configuration, Phase II
is the RN configuration, and Phase III is handoff configuration.
The specific numerical values used in this section were obtained from
Table \ref{timetab}.
\subsection{Phase I}
In Phase I the ES nodes act in a distributed manner to determine which ES
will become the master ES. The
master ES collects position information, determines the optimum
ES interconnections, and distributes the results back to the
ES nodes. The ES nodes determine the master ES by broadcasting {\bf MYCALL}
packets and collecting {\bf MYCALL} packets until the
MYCALL Timer expires with a prespecified time, $T$. The {\bf MYCALL}
packets contain the callsign and boot-time
of each ES. The ES with oldest boot-time is designated as the master.
$T$ should be chosen as the smallest value
which allows enough time for all {\bf MYCALL} packets to be received.
This would be approximately $0.492 * (N-1)$ seconds, where $N$ is the
total number of ES nodes. {\bf NEWSWITCH}
packets take on the order of 0.439 seconds to transfer, and therefore, it
will take $0.439 * (N-1)$
seconds to send these packets.
The ES nodes will respond with {\bf SWITCHPOS} packets which will take
another $0.679 * (N-1)$ seconds.
These events occur after each {\bf MYCALL} packet has been received,
and can occur before the MYCALL Timer has expired.
The next step in Phase I is to run the topology algorithm which is
based on a consistent labeling algorithm \cite{cla}. This algorithm
generates all fully connected topologies given ES node locations and
constraints on the antenna beams such that beams do not interfere with
one another. The information required by the topology module is the
GPS location of all ES nodes, transmit and receive beam widths, transmit
radius, the number of non-interfering frequency pairs, and an interference
multiplier. An interference multiplier of 1.0 assumes adaptive power
control, in which case it is assumed that beam power control will be
adjusted to exactly match the link distance. The interference multiplier
multiplied by a link's actual length will determine the range of
interference created by the link.
This takes on the order of $K_{top} \left[N^2+(L+1)^R \right]$ seconds
where $L$ is the number of available frequency pair
combinations with the addition of $1$ for no link. Assigning distinct
frequencies allows beams to overlap without interference.
$R$ is the number
of constrained links and $K_{top}$ is a constant. The constraints are
based on maximum beam length, beam widths, and number of frequencies
which can be supported.
The final step in Phase I is to distribute the topology information
to all ES nodes in {\bf TOPOLOGY} packets. This takes approximately
$0.664 + (0.1 * (N-1))$ seconds.
The time for Phase I to complete as a function of N is shown in Equation \ref{timep1}.
\begin{eqnarray}
P1(N) & = & \max\left[T, 0.439 * (N-1) + 0.492 * (N-1)\right] + \nonumber \\
& & K_{top}\left[N^2+(L+1)^R\right] + \label{timep1} \\
& & 0.664 + (0.1 * (N-1)) \nonumber
\end{eqnarray}
\subsection{Phase II}
Phase II is the RN configuration phase. Let U be the number of RNs
associated with a given ES. The first step is for the ES to receive
{\bf USER\_POS} packets from each RN. This takes $0.677 * U$ seconds.
The next step is to determine the optimum direction of the beams in order
to form a connection with the RNs. This algorithm execution is a linear
function of the number of RNs, which takes $K_{bf} * U$ seconds where
$K_{bf}$ is a constant. The algorithm is currently implemented in
MatLab and takes approximately 7.5 seconds to obtain reasonable convergence
of the beam direction to connect with four RNs.
The final step in Phase II is to generate a table of complex weights
for antenna beamforming and download this table to the hardware.
This is a function of the number of elements in the antenna
array, $K_{el}$, the number of beams, $B$, and the number of bits per
symbol, $M$. $K_{el}$ tables are created with $2^{M * B}$ entries per table.
This takes on the order of 2 seconds with 4 beams and 8 elements for
QPSK modulation on an OSF1 V4.0 386 DEC 3000/400 Alpha workstation.
The entire beamform and table generation module must be repeated for
every combination of transmitting RNs. A different table is used
depending on which RNs are currently transmitting data.
The complete time for Phase II is shown in Equation \ref{p2}.
\begin{equation}
P2(U) = 0.677 * U + \sum_{r=1}^{U} {U \choose r} \left( K_{bf} * r +
K_{el} * 2^{M * B} \right)
\label{p2}
\end{equation}
\subsection{Phase III}
Phase III, shown in Equation \ref{p3}, is the time required for the
orderwire to perform a hand-off.
The current network control code determines RN to ES associations based
on distance. When the distance between an RN and an ES other than its
currently associated ES becomes smaller than the distance between the
RN and its currently associated ES,
the current ES initiates a hand-off by sending a {\bf HANDOFF} packet.
This takes $0.473$ seconds. The RN will then initiate a point-to-point
orderwire connection with the new ES. Finally, Phase II must be run again
at the new ES, which is the reason for including the function $P2$.
\begin{equation}
P3_{RN} = 0.473 + P2(U+1)
\label{p3}
\end{equation}
\subsection{Orderwire Performance Emulation}
The emulation of the orderwire systems satisfies several goals. It
allows tests of configurations
that are beyond the scope of the prototype RDRN hardware. Specifically,
it verifies the correct operation of the RDRN Network Control Protocol
in a wide variety of situations.
The emulation helps to verify the correctness of the analytical
results obtained above.
As an additional benefit, much of the actual orderwire code was used by
the Maisie \cite{bagrodia} emulation allowing further validation of that
code.
The Edge Switch (ES) and Remote Node (RN) are modeled as a collection of
Maisie entities. This
is an emulation rather than a simulation because the Maisie code is
linked with the working orderwire code and also with the topology
algorithm.
There is an entity for each major component of the RDRN system
including the GPS receiver, packet radio, inter ES links, RN to ES links and
the Master, ES, and RN network configuration processors, as
well as other miscellaneous entities.
The input parameters to the emulation are shown in Tables \ref{emin_move},
\ref{emin_time}, \ref{emin_beam}.
\begin{table*}[htbp]
\centering
\begin{tabular}{||l|l||} \hline
{\bf Parameter } & {\bf Definition} \\ \hline \hline
NumRN & Number of Remote Nodes \\ \hline
NumES & Number of Edge Switches \\ \hline
ESDist & Inter Edge Switch spacing (forms rectangular area) \\ \hline
T & ES/ES MYCALL configuration time \\ \hline
maxV & Maximum RN speed for uniform distribution \\ \hline
S & Time to wait between node initial startups \\ \hline
ESspd & Initial Edge Switch speed \\ \hline
ESdir & Initial Edge Switch direction \\ \hline
RNspd & Initial Remote Node speed \\ \hline
RNdir & Initial Remote Node direction \\ \hline
\end{tabular}
\caption{\label{emin_move} NCP Emulation Mobility Input Parameters.}
\end{table*}
\begin{table*}[htbp]
\centering
\begin{tabular}{||l|l||} \hline
{\bf Parameter } & {\bf Definition} \\ \hline \hline
EndTime & Emulation end time (tenths of seconds) \\ \hline
VCCallTime & Inter High Speed Connection Setup Times \\ \hline
VCCallDuration & High Speed Connection Life Times \\ \hline
\end{tabular}
\caption{\label{emin_time} NCP Emulation Time Input Parameters.}
\end{table*}
\begin{table*}[htbp]
\centering
\begin{tabular}{||l|l||} \hline
{\bf Parameter } & {\bf Definition} \\ \hline \hline
UseRealTopology & Connect to MatLab and run actual program \\ \hline
Rlink & Maximum beam distance \\ \hline
Fmax & Number of non-interfering frequency pairs \\ \hline
Imult & Interference multiplier \\ \hline
Twidth & Transmitting Beam width \\ \hline
Rwidth & Receiving Beam width \\ \hline
\end{tabular}
\caption{\label{emin_beam} NCP Emulation Beam Input Parameters.}
\end{table*}
The RN VC setup process for connections over the inter ES antenna beams
is assumed to be Poisson. This represents ATM VC usage over the
physical link. The RN will maintain a
constant speed and direction until a hand-off occurs, then a new
speed and direction are generated from a uniform distribution.
This simplifies the analytical computation. Note that NCP packet transfer
times as measured in Table \ref{timetab} are used here.
\subsection{Orderwire Maisie Emulation Design}
The architecture for the RDRN link management and control is shown in
Figure \ref{ncsa}. The topology modules are used only on ES nodes capable of
becoming a master ES. The remaining modules are used on all ES nodes and RNs.
The beamform module determines an optimal steering
angle for the given number of beams which connects all RNs to be associated
with this ES. It computes an estimated signal to noise interference
ratio (SIR) and generates a table of complex weights which, once loaded,
will control the beam formation. Note that this table is not loaded until the table fill
trigger is activated. The connection table is used by the Adaptive HDLC and
ATM protocol stacks for configuration via the adaptation manager.
\begin{figure*}[htbp]
\centerline{\psfig{file=figures/linkd.eps,width=6.0in}}
\caption{Network Control System Architecture.}
\label{ncsa}
\end{figure*}
The emulation uses as much of the actual network control code as possible.
The packet radio driver, GPS driver, and Network Control Protocol state
machine are implemented in Maisie; tables, data structures, and decision
functions from working NCP code are used.
Figure \ref{ncpem} shows the structure of the Maisie entities. The entity
names are shown in the boxes and the message types are shown along the
lines. Direct communication between entities is represented as a solid line.
The dashed lines indicate from where entities are spawned.
\begin{figure*}[htbp]
\centerline{\psfig{file=figures/maisiencp.eps,width=6.0in}}
\caption{Emulation Design.}
\label{ncpem}
\end{figure*}
The RN entity which performs ATM VC setup (HSLRN entity in Figure \ref{ncpem})
generates calls as a Poisson process which the ES node (HSLES entity in
Figure \ref{ncpem}) will attempt to accept. If the EN moves out of range or the
ES has no beam or slot available the setup will be aborted. As the RN moves,
the ES will hand off the connection to the proper ES based on closest distance
between RN and ES.
\subsection{Effect of Scale on NCP}
The emulation was run to determine the effect on the NCP as the number
of ES and RN nodes increased. The dominate component of the configuration
time is the topology calculation run by the ES which is designated as the
master. Topology calculation involves searching through the problem space of
constraints on the directional beams for all feasible topologies and
choosing an optimal topology from that set as described in \cite{cla}.
The units on all values should be consistent with the GPS coordinate units,
and all angles are assumed to be degrees.
The beam constraint values are Maximum link distance 1000.0, Maximum
Frequencies 3, Interference Multiplier 1.0, Transmit Beam Width 10.0,
Receive Beam Width 10.0.
\begin{comment}
\begin{figure}[htbp]
\centerline{\psfig{file=figures/top.ps,width=3.25in}}
\caption{Topology Calculation Execution Time.}
\label{toptime}
\end{figure}
\end{comment}
The topology calculation is performed
in MatLab and uses the MatLab provided external C interface. Passing
information through this interface is clearly slow, therefore
these results do not represent the exact execution times of the prototype
system. However, they do provide a worse case test for the protocol.
A possible speedup may arise through the use of Virtual Network
Configuration, which will provide a mechanism for predicting values in
advance and also allows processing to be distributed.
Another improvement which may be considered is to implement a
hierarchical configuration. The network is partitioned into
a small number of clusters of nodes in such a way that nodes in each
group are as close
together as possible. The topology code is run as though these were
individual nodes located at the center of each group. This inter-group
connection will be added as constraints to the topology
computation for the intra-group connections. In this way the topology
program only needs to calculate small numbers of nodes which it does
relatively quickly.
\begin{comment}
\begin{figure}[htbp]
\centerline{\psfig{file=figures/config.ps,width=3.25in}}
\caption{Configuration Time.}
\label{configtime}
\end{figure}
\end{comment}
\subsection{MYCALL Timer}
The MYCALL Timer, set to a value of $T$ in the analysis section, controls
how long the system will wait to discover new ES nodes before completing the
configuration. If this value is set too low, new {\bf MYCALL} packets will
arrive after the topology calculation has begun, causing the system to
needlessly reconfigure.
If the MYCALL Timer value is too long, time will be wasted, which will
have a large impact on a mobile ES system. Table \ref{mycallsim} shows the
input parameters and Figure \ref{mycalltime} shows
the time required for all {\bf MYCALL} packets to be received as a
function of the number of ES nodes. These times are the
optimal value of the MYCALL Timer as a function of the number of
ES nodes because the these times are exactly the amount of time
required for all ESs to respond.
In order to prevent the possibility of an infinite loop of reconfigurations
from occurring, an exponential back-off on
the length of the {\bf MYCALL} Timer value is introduced.
As {\bf MYCALL} packets arrive after
T has expired, the next configuration occurs with an increased value
of T.
\begin{table*}[htbp]
\centering
\begin{tabular}{||l|l||} \hline
{\bf Parameter } & {\bf Value} \\ \hline
Number of RNs & $0$ \\ \hline
Number of ESs & $2$ thru $6$ \\ \hline
Inter-ES Distance & $20$ m (65.62 ft) \\ \hline
T & $20$ s \\ \hline
Maximum Velocity & $5$ m/s (11.16 mi/hr) \\ \hline
Initial ES Speed & $0$ m/s \\ \hline
Initial ES Direction & $0^o$ \\ \hline
Initial RN Speed & N/A \\ \hline
Initial RN Direction & N/A \\ \hline
Inter-VC Setup Time & $1200$ s \\ \hline
VC Call Duration & $600$ s \\ \hline
\end{tabular}
\caption{\label{mycallsim} MYCALL Timer Simulation Parameters.}
\end{table*}
\begin{figure}[htbp]
\centerline{\psfig{file=figures/mycall.ps,width=3.25in}}
\caption{MYCALL Packets Received.}
\label{mycalltime}
\end{figure}
\subsection{Link Usage Probability}
Multiple RNs may share a single beam using Time Division Multiplexing
(TDMA) within a beam. The time slices are divided into slots, thus
a $(beam, slot)$ tuple defines a physical link. The emulation was run
to determine the probability distribution of links used as a function
of the number of RNs. The parameters used in the emulation are shown in
Table \ref{chanusedsim} the results of which indicate the number of links
and thus the number of distinct $(beam, slot)$ tuples required. Figure
\ref{chanused} shows the link usage cumulative distribution function
for 4 and 7 RNs.
\begin{table*}[htbp]
\centering
\begin{tabular}{||l|l||} \hline
{\bf Parameter } & {\bf Value} \\ \hline \hline
Number of RNs & $4$ and $7$ \\ \hline
Number of ESs & $2$ \\ \hline
Inter-ES Distance & $20$ m (65.62 ft) \\ \hline
T & $20$ s \\ \hline
Maximum Velocity & $5$ m/s (11.16 mi/hr)\\ \hline
Initial ES Speed & $0$ m/s \\ \hline
Initial ES Direction & $0^o$ \\ \hline
Initial RN Speed & $5$ m/s (11.16 mi/hr)\\ \hline
Initial RN Direction & $0^o$ \\ \hline
Inter-VC Setup Time & $1200$ s \\ \hline
VC Call Duration & $600$ s \\ \hline
\end{tabular}
\caption{\label{chanusedsim} Link Usage Simulation Parameters.}
\end{table*}
\begin{figure}[htbp]
\centerline{\psfig{file=figures/chan.ps,width=3.25in}}
\caption{$(Beam, Slot)$ Usage.}
\label{chanused}
\end{figure}
\subsection{ES Mobility}
ES mobility is a more difficult problem and will be examined in more
detail as the research proceeds.
The parameters used in an emulation with mobile ES nodes are shown in Table
\ref{mobessim}.
As mentioned in the section on the MYCALL Timer, if a {\bf MYCALL} packet
arrives after this timer has expired, a reconfiguration occurs. This could
happen due to a new ES powering up or an ES which has changed position.
Figure \ref{mobestime} shows the times at which reconfigurations occurred
in a situation in which ES nodes were mobile.
Based on the state transitions generated from the emulation it is
apparent that the system is in a constant state of reconfiguration; no
reconfiguration has time to complete before a new one begins.
As ES nodes move, the NCP must notify RNs associated with an ES with the
new position of the ES as well as reconfigure the ES nodes.
To solve this problem, a tolerance, which may be associated with the
link quality,
will be introduced which indicates how far nodes can move
within in a beam before the beam angle must be recalculated,
which will allow more time between reconfigurations.
It is expected that this tolerance in addition to Virtual Network
Configuration will provide a solution to this problem.
\begin{table*}[htbp]
\centering
\begin{tabular}{||l|l||} \hline
{\bf Parameter } & {\bf Value} \\ \hline \hline
Number of RNs & $0$ \\ \hline
Number of ESs & $3$ \\ \hline
Inter-ES Distance & $20$ m (65.62 ft) \\ \hline
T & $20$ s \\ \hline
Maximum Velocity & $5$ m/s (11.16 mi/hr) \\ \hline
Initial ES Speed & $1$ m/s (2.23 mi/hr) \\ \hline
Initial ES Direction & $0^o$ \\ \hline
Initial RN Speed & N/A \\ \hline
Initial RN Direction & N/A \\ \hline
Inter-VC Setup Time & $1200$ s \\ \hline
VC Call Duration & $600$ s \\ \hline
\end{tabular}
\caption{\label{mobessim} Mobile ES Simulation Parameters.}
\end{table*}
\begin{figure}[htbp]
\centerline{\psfig{file=figures/mobes.ps,width=3.25in}}
\caption{Mobile ES Configuration Time.}
\label{mobestime}
\end{figure}
\subsection{Effect of Communication Failures}
The emulation was run with a given probability of failure on each
packet type of the Network Control Protocol. The following results
are based on the output of the finite state machine (FSM) transition
output of the emulation
and an explanation is given for each case.
A dropped {\bf MYCALL}
packet has no effect as long as at least one of the {\bf MYCALL}
packets from each ES is received at the master ES. This is the
only use of the AX.25 broadcast mode in the ES configuration. The
broadcast AX.25 mode is a one time, best effort delivery;
therefore, {\bf MYCALL} packets are repeatedly broadcast at the
NCP layer.
The Maisie emulation demonstrated that
a dropped {\bf NEWSWITCH} packet caused the
protocol to fail. This is because the master ES will wait until
it receives all {\bf SWITCHPOS} packets from all ES nodes for
which it had received {\bf MYCALL} packets. The {\bf NEWSWITCH}
packet is sent over the AX.25 in connection-oriented mode, e.g.
a mode in which corrupted frames are retransmitted;
the probability of loosing a packet in this mode is
very low.
A dropped {\bf SWITCHPOS} packet has the same effect as a dropped
{\bf NEWSWITCH} packet.
In order to avoid this situation, the NCP will re-send the {\bf NEWSWITCH}
if no response is received.
Finally, the Maisie emulation showed that
a lost {\bf TOPOLOGY} packet results in a partitioned network.
The ES which fails to receive the {\bf TOPOLOGY} packet is not joined with
the remaining ES nodes; however, this ES node continued to receive and
process {\bf USER\_POS} packets from all RNs. It therefore attempts to
form an initial connection with all RNs.
The solution for this condition is not to allow RN associations with
an ES node until the {\bf TOPOLOGY} packet is received. Because
{\bf MYCALL} packets are transmitted via broadcast AX.25, each ES node
can simply count the number of {\bf MYCALL} packets and estimate the time for
the master ES node to calculate the topology using the number of {\bf MYCALL}
packets as an estimate for the size of the network. If no {\bf TOPOLOGY}
packet is received within this time period, the ES node retransmits its
{\bf SWITCHPOS} packet to the master ES node in order to get a
{\bf TOPOLOGY} packet as a reply.
\subsection{Seamless Mobile ATM Routing}
This section discusses an incorporation of Virtual Network Configuration
(VNC) and the Network Control Protocol (NCP) as described in the previous
sections into the Private Network-Network Interface (PNNI) \cite{PNNI} to
facilitate seamless ATM hand-off. An attempt is made to minimize the changes
to the evolving PNNI standard. Figure \ref{pnniov} shows a high level view
of the PNNI Architecture. Terminology used in the PNNI Specification.
In this version of mobile PNNI, the standard PNNI route determination,
topology database, and topology exchange would reside within the NCP.
The NCP stack with VNC is shown in Figure \ref{vncstack}.
\begin{figure*}[htbp]
\centerline{\psfig{file=figures/pnni.eps,width=6.0in}}
\caption{PNNI Architecture Reference Model.}
\label{pnniov}
\end{figure*}
\begin{figure}[htbp]
\centerline{\psfig{file=figures/vncpnni.eps,width=3.25in}}
\caption{Virtual Network Configuration Stack.}
\label{vncstack}
\end{figure}
The enabling mechanism is the fact that VNC will cause the NCP
to create a topology which will exist after a hand-off occurs at
a time prior to the hand-off.
This will cause PNNI to perform its standard action of updating
its topology information immediately before the hand-off occurs.
Note that this is localized within a single Peer Group (PG).
The second enabling mechanism is a change to the PNNI signaling protocol.
In mobile PNNI, standard PNNI signaling is allowed to dynamically modify
logical links when triggered by a topology change. This is similar to a
{\bf CALL ABORT} message except that the ensuing {\bf RELEASE} messages
will be contained within the scope of the Peer Group (PG). This new
message will be called a {\bf SCOPED CALL ABORT} message.
When the topology changes due to an end system hand-off, a check is
made to determine which end system (RN) has changed logical nodes (LN).
An attempt is made to establish the same incoming VCs at the new LN as were
at the original LN and connections are established from the new LN to the
original border LNs of the Peer Group. This allows the RN to continue
transmitting with the same VCI as the hand-off occurs. The connections
from the original LNs to the border LNs are released after the hand-off
occurs. If the new LN is already using a VCI that was used at the original
LN, the {\bf HANDOFF} packet will contain the replacement VCIs to be used by
the end system (RN). There are now two branches of a logical link tree
established with the border LN as the root. After the hand-off takes
place the old branch is removed by the new {\bf SCOPED CALL ABORT} message.
Note that link changes are localized to a single Peer Group.
The fact that changes can be localized to a Peer Group greatly reduces
the impact on the network and implies that the mobile network should have
many levels in its PNNI hierarchy. In order to maintain cell order the
new path within the Peer Group is chosen so as to be equal to or longer
than the original path based on implementation dependent metrics.
Consider the network shown in Figure \ref{mobpnni}.
Peer Groups are enclosed in circles and the blackened nodes
represent the lowest level Peer Group Leader for each Peer Group.
End system A.1.2.X is about to hand-off from A.1.1 to A.2.2. The smallest
scope which encompasses the old and new LN is the LN A.
\begin{figure*}[htbp]
\centerline{\psfig{file=figures/mobpnni.eps,width=6.0in}}
\caption{Mobile PNNI Example.}
\label{mobpnni}
\end{figure*}
A.3.1 is the outgoing border node for LN A. A {\bf CALL SETUP} uses
normal PNNI operations to setup a logical link from A.3.1 to A.2.2.
After A.1.2.X hands off, a {\bf SCOPED CALL ABORT} message releases the
logical link from A.3.1 to A.1.1.
|
\section{Introduction}
\label{intro}
In recent years, there has been a great deal of experimental
and theoretical activity studying the magnetic properties of
insulating compounds that consist of arrays of well-isolated
one-dimensional magnetic sub-structures. An example is the compound
(VO)$_2$P$_2$O$_7$ (VOPO), which was initially thought to
be an excellent candidate for a spin-ladder compound based on
early experimental and theoretical work~\cite{eccle}. Inelastic
neutron scattering experiments~\cite{oldexp} on single crystal arrays
established that these early ideas were incorrect and the
structure consists, to a good approximation, of an array
of alternating antiferromagnetic chains that are weakly coupled
to each other in one direction perpendicular to the chain axis.
The spectrum of magnetic excitations was mapped out by these
inelastic neutron scattering experiments. The lowest lying
excitation seen is a triplet mode separated by a gap from
the singlet ground state of the system. This is identified
with the basic single particle excitation expected theoretically
in an alternating chain (see for instance Ref~\cite{uhschu} and
references therein). The experiment also saw an
additional triplet mode above this band in a large
part of the Brillouin zone. The origin of the second mode was unclear, and
there
was some speculation that it could be
ascribed to a triplet bound state of the elementary excitations
that is also expected to exist in these systems~\cite{uhschu}. Such a state
may be stabilized
by frustrating interactions~\cite{uhnorm}
It is more likely that the two
inequivalent magnetic
chains in VOPO have differing gap energies~\cite{kikuchi}; the second
mode would then be the basic triplet mode of the second set of chains.
New inelastic neutron scattering experiments on a single crystal of
(VO)$_2$P$_2$O$_7$ (VOPO)~\cite{ens} have been able to map out
the dispersion of the basic excitations of the system in greater
detail.
These experiments however also see {\em additional sharp low-energy modes}
with dispersions different from the modes seen previously.
These extra modes are, at first sight, extremely surprising
and it is tempting to take them to be a signal of some new, and
hitherto unanticipated features in the spectrum of the system (possibly
arising from frustrated couplings between alternating chains). However,
we argue that the real explanation for the new
modes is quite simple: both modes arise from a purely
geometric effect having to do with the actual positions of the
vanadium ions in the unit cell. The
two new modes may be thought of as shadows of the basic triplet modes
arising from umklapp scattering.
We begin by detailing the geometry involved and use a very simple
general argument to calculate the matrix element for the
umklapp scattering process that is responsible for producing
a shadow of the basic
triplet mode. We then suggest a
straightforward check of this explanation based on a
comparison of the experimentally observed intensities of the
basic mode and its shadow at various values of the momentum
transfer. It is important to emphasize at this stage that
this check is quite independent of any theoretical estimates of
the intensity of the basic mode as a function of momentum
transfer and relies only on relations between experimentally
observed intensity ratios; the calculation
of the intensity of the basic single particle triplet mode as a function
of $k$ ($2 \pi k/b \equiv k_b$ is the momentum transfer along
the chain direction, where $b$ is the unit cell dimension
along the chain axis, which is conventionally
labeled the b axis) is a separate problem that has been
addressed earlier for the simple alternating
chain~\cite{barnes} (these results
may also be used in conjunction
with our analysis to give approximate intensities of the shadow of the
single particle band, but we do not perform that
exercise here). A simple consequence of this
scenario is the prediction that the shadow band will {\em disappear for}
$h=0$ (here $2\pi h/a \equiv k_a$ is the momentum transfer along the
crystallographic
$a$
axis perpendicular to the alternating chain axis).
One also expects
that shadows of any bound-state mode will also be formed
by an analogous mechanism involving umklapp scattering.
We illustrate this by an explicit calculation, to leading
order in a strong-coupling expansion, of
the bound state contribution to the dynamic structure factor for
an alternating chain with the right geometry taken into account.
The calculated intensity ratios do provide an explicit example of the
general argument for the strength of the umklapp contribution.
We also briefly explore
the possibility that the magnetic interactions felt by even and
odd dimers (pairs of spins connected by the
stronger of the two antiferromagnetic interactions in an
alternating chain model) are slightly different. We expect that
this will change the strength of the shadow bands in a
significant way.
To get a feel for what to expect, we do a simple calculation, again within a
strong
coupling expansion, of the contribution of the basic
triplet mode to the dynamic structure factor for an alternating chain
with the right geometry and the small difference in magnetic
interactions felt by even and odd dimers. We see that this change
in the magnetic environments of even and odd dimers
leads to a weak intensity for the shadow mode even
at $h=0$ (in contrast to our
result for the simpler alternating chain of Fig~\ref{figchain1})
as well as a small splitting between the basic mode and
its shadow at $k=1/2 \, , \, 3/2$ in the fundamental Brillouin zone at
$h=0$. This shadow
at $h=0$, as well as the splitting at $k=1/2 \, , \, 3/2$
are a sensitive test of the difference in the magnetic environments
of even and odd dimers in the chain.
All experiments to date~\cite{oldexp,ens,agthesis} are
consistent with the absence of a shadow mode at $h=0$. However, in
the absence of any straightforward symmetry reason forcing the
even and odd dimers to be equivalent, the
possibility that more refined experiments with better
statistics will see a weak shadow is still open.
Lastly, it must be emphasized that our entire approach here ignores the
frustrating couplings between chains that has been argued to
exist based on the small, but experimentally detectable dispersion
seen as a function of $h$. These couplings
along the $a$ direction are important ingredients
of any quantitatively accurate calculation of the expected
neutron scattering intensity, but are not expected to change
significantly any of our conclusions regarding the shadow modes.
\section{Shadow bands due to umklapp scattering.}
\label{shadow}
We begin with a brief review of some of the relevant
details of the structure~\cite{nguyen} of VOPO:
Previous work~\cite{oldexp} on VOPO has established that
the compound may be thought of as an array of alternating
antiferromagnetic
chains (with the chains oriented along the crystallographic $b$
axis and the V$^{4+}$ ions forming the basic spin $1/2$
constituents of the chains) that are weakly coupled to each other in the $a$
direction, and essentially decoupled in the $c$ direction.
As mentioned in the introduction, we will, for the most part,
ignore the weak interchain coupling as it is not
expected to materially change any of our conclusions.\begin{figure}
\epsfxsize=5.0in
\centerline{\epsffile{figchain1.eps}}
\vspace{0.1in}
\caption{Geometry of the alternating chain. Note that the
staggering of the dimers in the $a$ direction is greatly
exaggerated in the figure; $v \approx 0.09 a$, while $w \approx 0.31 b$
with $a \approx 7.7$ angstroms and $b \approx 16.6$ angstroms.}
\label{figchain1}
\end{figure}
Each unit cell of VOPO contains eight V$^{4+}$ ions, comprising four dimers.
Each of the dimers belongs to a different chain. There is a pair of chains
near $z=0$,
and a second pair near $z=c/2$. The members of each pair are related to
each other via
a screw axis transformation and therefore there are at most two magnetically
distinct chains displaced from each other along the $c$ axis.
We can thus focus attention on one representative from each pair.
These two chains have similar (but not identical)
structures and the geometrical
effects we discuss here are very nearly identical for each type of chain.
Using the known structure, we
can draw out the actual positions of the
vanadium ions in the $a-b$ plane.
To a good approximation, this gives us
an alternating chain along the $b$ axis in which
successive dimers are slightly staggered along
the $a$ axis as shown in Fig~\ref{figchain1}.
There is an extremely tiny displacement
of the ions relative to each other in the $c$ direction; this
is small enough that we feel justified in ignoring it in our
analysis. Similarly we can ignore tilts of the dimer units away from
the $b$ axis.
The magnetic response of a single
chain may be modeled (modulo the possible
complications that form the subject matter
of section~\ref{magnetic}) by the simple alternating
chain Hamiltonian
\begin{equation}
{\cal H} = J \sum \limits_{i}[{\bf S}_{I}(i) \cdot {\bf S}_{II}(i)
+g{\bf S}_{II}(i) \cdot {\bf S}_{I}(i+1)] \; ,
\label{simpleH}
\end{equation}
where $J$ is the overall energy scale fixed by the microscopic
exchange constants in the system, $g$ represents the
ratio of the weak and the strong bonds of the alternating chain,
and the spins are labelled as in Fig~\ref{figchain1}. Note that
this Hamiltonian is invariant under translations of
$b/2$ along the $b$ axis. However,
a glance at Fig~\ref{figchain1} shows that
the staggering of the
positions of even and odd dimers in the $a$ direction
reduces the actual symmetry of the full structure to
translations by $b$ and not $b/2$ along the $b$ axis.
This of course implies that momentum conservation may
be violated during a neutron scattering event by
integer multiples of $2\pi/b$ along the chain axis. Note that
this is less stringent than the more usual
condition (which would be in force in the absence of any staggering
along the $a$ axis of even and odd dimers)
that momentum be conserved modulo integer multiples of
$4\pi/b$, and we believe that this simple fact is at the
root of the observed shadow bands. Thus, we expect that the extra modes
seen should be displaced by precisely $2\pi/b$ from the basic modes
of the alternating chain. This seems to be the case with the
experimental data~\cite{ens}. To clinch the identification,
we need to be able to make predictions for the intensities of the
extra modes relative to the basic modes and see how these compare
with the experimental numbers for the intensity ratios. This is what
we turn to next.
Let us begin our analysis by
writing down the usual spectral
representation for the dynamic structure factor of our system
at $T=0$:
\begin{eqnarray}
S_{zz}({\bf k}, \omega) & = & \sum \limits_{N} \delta(\omega -
E_N + E_0)\left |\langle \Phi_N| S^{z}(-{\bf k})|\Phi_0\rangle \right|^2
\; ,
\end{eqnarray}
where $|\Phi_0\rangle$ is the exact ground state of the
system, $|\Phi_N \rangle$ is an exact excited state labeled
by the index $N$, $E_0$ and $E_N$ are the energies of the
ground and the excited state respectively and $S^{z}(-{\bf k})$ is
defined as
\begin{equation}
S^{z}(-{\bf k}) = (\frac{b}{4L})^{1/2} \sum \limits_{j\,A} S^{z}_{A}(j)
e^{i{\bf k} \cdot {\bf x}_{j\, A}} \; ,
\end{equation}
where the subscript $A$ takes on values $I$ and $II$, $j$ refers to
the dimer index, $L$ is the length of
the chain and ${\bf x}_{j \, A}$ is the position of the
spin labeled by $j$ and $A$ (see Fig~\ref{figchain1}) (here and in
the rest of our discussion, we will exploit the rotational
invariance in spin space to focus only on the $zz$ component
of the dynamic structure factor).
It is convenient to formulate our analysis
in terms of operators that directly make reference to the
states of each strongly coupled dimer in the system. This
is achieved by transforming to the so-called `dimer boson'
representation~\cite{leuen,ssbhatt,cowley}.
Following Ref~\cite{ssbhatt}, we write
the spin operators as:
\begin{eqnarray}
S_{I}^{\alpha}(j) & = & \frac{1}{2}\left(s^{\dagger}(j)t_{\alpha}(j) +
t_{\alpha }^{\dagger}(j)s(j)-i\epsilon_{\alpha \beta
\gamma}t_{\beta}^{\dagger}(j)
t_{\gamma}(j)\right )~, \label{bonddefI} \\
S_{II}^{\alpha}(j) & = & \frac{1}{2}\left(-s^{\dagger}(j)t_{\alpha}(j) -
t_{\alpha }^{\dagger}(j)s(j)-i\epsilon_{\alpha \beta
\gamma}t_{\beta}^{\dagger}(j)
t_{\gamma}(j)\right )~,
\label{bonddefII}
\end{eqnarray}
where $\alpha$, $\beta$, and $\gamma$ are vector indices taking the values
$x$,$y$,$z$, repeated indices are summed over, and $\epsilon$ is the totally
antisymmetric tensor. $s^{\dagger}(j)$ and $t_{\alpha}^{\dagger}(j)$ are
respectively creation operators for singlet and
triplet bosons at `site' $j$ (in the bosonic language, each strongly
coupled dimer is thought of as a single site; the separation of
adjacent sites along the $b$ axis is then $b/2$).
The restriction that physical states of
a dimer are either singlets or triplets leads to the following constraint on
the boson occupation numbers at each site:
\begin{displaymath}
s^{\dagger}(j)s(j)+t_{\alpha}^{\dagger}(j)t_{\alpha}(j)=1~.
\end{displaymath}
The spin density is given by
\begin{displaymath}
\sigma_{\alpha}(j) = -i\epsilon_{\alpha \beta \gamma}t_{\beta}^{\dagger}(j)
t_{\gamma}(j)~.
\end{displaymath}
It is also convenient to define
\begin{displaymath}
\phi_{\alpha}(j)=s^{\dagger}(j)t_{\alpha}(j) +t_{\alpha }^{\dagger}(j)s(j)~.
\end{displaymath}
Note that as the constraint fixes the number of singlet particles
uniquely given the triplet occupation number, we may as
well think only in terms of triplet occupation numbers; we will
thus refer to any site which is occupied by a singlet as being
in the vacuum state.
Finally, it is useful to note that the alternating chain is
readily analyzed in the
limit of strong alternation (the so called strong coupling
limit, with $g \ll 1$)~\cite{uhschu,barnes}. In the language
we are using here, the lowest lying excitations in this limit
are single particle modes (with one bare triplet particle
excited above the ground state, which may be thought of
as vacuum). While
corrections are certainly introduced to this picture at
higher orders in $g$, it is still legitimate
to think of the basic triplet mode seen in the real system
as arising from the contribution of the fully renormalized single
particle excitation in the system (for a careful analysis
of this point for the closely related problem
of a spin-ladder, see Ref~\cite{sskd})
With these preliminaries out of the way, let us now go through the
extremely elementary general argument for the
strength of the umklapp matrix element responsible for the
shadow bands. We formulate this here only for contributions
to the spectral sum coming from the fully renormalized single
particle states of the system. The basic argument is nevertheless
expected to remain valid when applied to the contributions coming
from two-particle bound states; we will see this expectation
realized in an explicit calculation later.
Let us begin by writing the contribution of the fully renormalized
single particle states as
\begin{equation}
S_{zz}^{{\rm{1p}}}({\bf k}, \omega) = \sum \limits_{q = 0}^{4\pi/b}
\delta(\omega - \varepsilon_q)
\left| \langle q| S^{z}(-{\bf k}) |\Phi_0\rangle\right|^2 \; ,
\label{qsum1}
\end{equation}
where $|q\rangle$ is the exact, fully renormalized single particle
state of momentum $q$ in the chain direction (the state
of course has $z$ component of its spin equal to $0$; we will
not be very careful in this section about including this information
in our notation) and $\varepsilon_q$
is the energy of this state (with the ground state energy
set to zero). We may write this
state quite generally as
\begin{equation}
|q\rangle = (\frac{b}{2L})^{1/2} \sum \limits_{j}e^{iqx_j^{b}} | \Psi_1(j)
\rangle \; ,
\end{equation}
where $x_j^{b}$ is the $b$ component of the position vector of the
$j^{th}$ `site' (center of the $j^{th}$ dimer) in the chain and the
notation $|\Psi_1(j)\rangle$ is intended to denote a state that
differs from $|\Phi_0\rangle$ only locally in the vicinity of
site $j$. Furthermore, we can write $S^{z}(-{\bf k})$ as
\begin{equation}
S^{z}(-{\bf k}) =
(\frac{b}{2L})^{1/2}
(
e^{ik_a v/2} \sum \limits_{j \, {\rm odd}}e^{ik_bx_j^{b}} {\cal O}_{k_b}(j)
+ e^{-ik_a v/2} \sum \limits_{j \, {\rm even}}e^{ik_bx_j^{b}} {\cal
O}_{k_b}(j)
) \; ,
\label{szo}
\end{equation}
where the operator ${\cal O}_{k_b}(j)$ is defined as
\begin{equation}
{\cal O}_{k_b}(j) = \frac{1}{\sqrt{2}}(\cos(\frac{ k_bw}{2}) \sigma_{z}(j)
- i \sin(\frac{k_bw}{2}) \phi_{z}(j)) \; ,
\label{o12}
\end{equation}
with $w$ equal to the distance along the $b$ axis between the two
spins that form each dimer.
This now allows us to write the following expression for the
matrix element appearing in the spectral sum (\ref{qsum1}):
\begin{equation}
\left| \langle q| S^{z}(-{\bf k}) |\Phi_0\rangle\right|^2
=\frac{|{\cal M}(q,k_b)|^2}{4} |e^{-ik_av/2} + e^{i(k_b-q)b/2}e^{ik_av/2}|^2
{\tilde \delta}_{q \, , \, k_b} \; ,
\label{me1}
\end{equation}
where ${\tilde \delta}$ is defined as
\begin{equation}
{\tilde \delta}_{q \, , \, k_b} = \sum \limits_{n = - \infty}^{\infty}
\delta_{q \, , \, k_b + 2\pi n/b} \; ,
\end{equation}
and ${\cal M}$ is given as
\begin{eqnarray}
{\cal M}(q,k_b) & = & \sum \limits_{j} e^{-iqx_{j}^{b}}\langle \Psi_1(j)|{\cal
O}_{
k_b}(0)|\Phi_0\rangle \; .
\end{eqnarray}
Note that we can make the $k_b$ dependence of ${\cal M}$ explicit
by rewriting this as
\begin{equation}
{\cal M}(q,k_b) = \cos(\frac{ k_bw}{2}){\cal M}_{\sigma}(q) +
\sin(\frac{ k_bw}{2}){\cal M}_{\phi}(q) \; ,
\end{equation}
where
\begin{eqnarray}
{\cal M}_{\sigma}(q) &=& \frac{1}{\sqrt{2}}
\sum \limits_{j} e^{-iqx_{j}^{b}}\langle \Psi_1(j)|\sigma_{z}(0)|\Phi_0\rangle
\; , \nonumber \\
{\cal M}_{\phi}(q) & = & \frac{-i}{\sqrt{2}}
\sum \limits_{j} e^{-iqx_{j}^{b}}\langle \Psi_1(j)|\phi_{z}(0)|\Phi_0\rangle
\; .
\end{eqnarray}
Furthermore, previous work~\cite{barnes} has demonstrated that
${\cal M}_{\sigma}$
is identically zero for the single particle states of the
simple alternating chain Hamiltonian~(\ref{simpleH}).
Using all of this we can write the contribution of the fundamental
single-particle mode to the dynamic structure factor as
\begin{eqnarray}
S_{zz}^{{\rm{1p, basic}}}({\bf k}, \omega) & = & |{\cal M}_{\phi}(k_b)|^2
\sin^2(\frac{k_b w}{2}) \cos^2(\frac{k_a v}{2}) \delta(\omega -
\varepsilon_{k_b}) \; \; \; k_b \, \in \, (0,\frac{4\pi}{b})\;, \nonumber \\
& = & |{\cal M}_{\phi}(k_b-4\pi/b)|^2
\sin^2(\frac{k_b w}{2}) \cos^2(\frac{k_a v}{2}) \delta(\omega -
\varepsilon_{k_b - 4\pi/b}) \; \; \; k_b \, \in \,
(\frac{4\pi}{b},\frac{8\pi}{b})\;,
\end{eqnarray}
while the shadow contribution reads
\begin{eqnarray}
S_{zz}^{{\rm{1p, shadow}}}({\bf k}, \omega) & = &
|{\cal M}_{\phi}(k_b+2\pi/b)|^2
\sin^2(\frac{k_b w}{2}) \sin^2(\frac{k_a v}{2}) \delta(\omega -
\varepsilon_{k_b+2\pi/b}) \; \; \; k_b \, \in \,
(0,\frac{2\pi}{b})\;, \nonumber \\
& = &
|{\cal M}_{\phi}(k_b-2\pi/b)|^2
\sin^2(\frac{k_b w}{2}) \sin^2(\frac{k_a v}{2}) \delta(\omega -
\varepsilon_{k_b-2\pi/b}) \; \; \; k_b \, \in \,
(\frac{2\pi}{b},\frac{6\pi}{b})\;, \nonumber \\
& = &
|{\cal M}_{\phi}(k_b-6\pi/b)|^2
\sin^2(\frac{k_b w}{2}) \sin^2(\frac{k_a v}{2}) \delta(\omega -
\varepsilon_{k_b-6\pi/b}) \; \; \; k_b \, \in \,
(\frac{6\pi}{b},\frac{8\pi}{b})\;.\nonumber \\
&&
\label{shadrealcomp1}
\end{eqnarray}
Thus, we see quite generally that the intensity of the
shadow band should vanish as $k_a \rightarrow 0$ (modulo
the complications discussed in section~\ref{magnetic}). Moreover,
it is apparent from these expressions that the intensity of
the shadow is completely determined by the intensity of
the basic mode as a function of $k_b$. While there are, in
principle, a number of ways in which this may be checked against
the experimental data of Ref~\cite{ens}, it is probably best to
simply use the experimentally observed intensity of the fundamental
mode at fixed $k_a$ (for values of $k_b$ at which the `dimer coherence factor'
$\sin^2(k_bw/2)$ is large) to predict the intensity of
the shadow {\em at the same $k_a$} (this avoids complications
due to the weak two-dimensional couplings between chains that will
introduce additional dependence of the intensity on $k_a$). This prediction
can then be directly tested against the observed intensity of the
shadow after correcting for effects of the magnetic form factor
of the V$^{4+}$ ion. Note that this procedure makes no assumptions about
the form of ${\cal M}_{\phi}(k_b)$ for the alternating chain Hamiltonian
(\ref{simpleH}) and serves to separate the purely geometric
effect leading to the shadow from our approximate knowledge
of this function.
Let us conclude this section by noting that entirely analogous
arguments can be used to relate the expected intensity of the
shadow of a bound-state mode to the intensity of the
bound-state itself (of course, the analog of
${\cal M}_{\sigma}$ is no longer identically zero, but this
merely complicates the algebra a little). Instead of going through
the corresponding argument for the bound state modes in detail, we
choose to highlight the minor differences
involved by doing an approximate calculation of the intensity and
position of both the bound-state mode and its shadow to
leading order in a strong-coupling expansion. This is what we turn to
in the next section.
\section{Bound-state contributions within the strong-coupling expansion}
\label{bssc}
The first order of business is to work out the position in
the Brillouin zone and the energy of the $S=1$ bound state formed
from the physical (fully renormalized) triplet particles that
are the elementary excitations of the alternating chain (\ref{simpleH}).
To leading order in the strong coupling expansion, this is
particularly simple as the physical single-particle excitation
coincides with the bare triplet particle created by the triplet
boson operator as far as the energy levels are concerned. Following
the approach used in Ref~\cite{sskd}, it is easy to see~\cite{kdup} at
leading order that
the triplet bound state exists over two separate
intervals for the center of mass momentum $q_{cm}$
(note that
the center of mass momentum takes on values in the range $(0, 8\pi/b)$):
the first being $(4\pi/3b, 2\pi/b)$ and the second being $(6\pi/b,20\pi/3b)$.
The energy of the bound state (with the ground state energy set to
zero) is given to leading order as $\varepsilon_B(q_{cm})/J =
2 - g(4\cos^2(q_{cm}b/4)+1)/4$ (these results were first
obtained for a slightly more general Hamiltonian by Uhrig and
Schulz~\cite{uhschu}).
The next thing we need is the bound state wavefunction and the
ground state wavefunction correct to first order in $g$; note that it
does not suffice to know these eigenstates to leading (zeroth) order
in $g$ as it turns out that the bound state contributes to the
dynamic structure factor only at first order or higher in $g$.
We will write these eigenstates down in the basis of (bare) triplet
boson occupation numbers and polarizations. An extremely
elementary
calculation~\cite{kdup} gives us the following ground state, correct
to first order in $g$:
\begin{equation}
|\Phi_0\rangle = |0\rangle + \frac{g}{8}\sum \limits_{j}\left(
|(j)[0],(j+1)[0]\rangle -|(j)[-1],(j+1)[+1]\rangle-|(j)[+1],(j+1)[-1]\rangle
\right)\; ,
\end{equation}
where $|0\rangle$ represents the vacuum state for the
triplet bosons, the number in the square brackets gives the $z$ component
of the spin of the triplet boson and the number in
the parenthesis gives the site occupied by the boson (two such
pairs separated by a comma naturally denote a two-particle state
in the bosonic Fock space).
The zeroth order normalized bound state labeled by the
center-of-mass momentum $q_{cm}$ (and $z$ component of
spin equal to $0$) can be easily calculated~\cite{kdup} to
be
\begin{equation}
|q_{cm}[0]\rangle = \sum \limits_{j_2 > j_1}f_{q_{cm}}(j_1,j_2)
\left( |(j_1)[-1],(j_2)[+1]\rangle - |(j_1)[+1],(j_2)[-1]\rangle \right) \; ,
\end{equation}
where the bound-state wavefunction $f_{q_{cm}}$ is given as
\begin{equation}
f_{q_{cm}}(j_1,j_2) = (\frac{b}{2L})^{1/2}(\frac{e^{\kappa b}-1}{2})^{1/2}
e^{-\kappa|x^b_{j_2} - x^b_{j_1}|}e^{iq_{cm}(x^b_{j_2}+x^b_{j_1})/2} \; ,
\end{equation}
with $e^{-\kappa b/2} = 2 \cos(q_{cm}b/4)$.
Now, following the approach used in Ref~\cite{sskd}, it is quite
easy to see that the ${\cal O}(g)$ corrections to this will
involve states living in the zero, one, three and four boson
sectors of the Fock-space for the
bare triplet bosons, in addition to a ${\cal O}(g)$ correction to the
component in the two-boson sector. It is an elementary
exercise~\cite{kdup} to work out all these corrections except the
one in the two-boson sector (as this involves first working
out the effective dynamics of the physical particles to one higher
order in $g$). While the correction in the two-boson sector can
also be calculated without too much difficulty using the methods
referred to earlier, we do not bother to do this explicitly here as
it is quite clear that this correction term plays no role
in the leading order calculation of the bound state contribution to
the dynamic structure factor. In fact, for our purposes here,
it clearly suffices to work out the correction in the one boson
sector of the Fock space as this is the only correction that
affects our calculation.
This may be written down readily~\cite{kdup} as:
\begin{equation}
\delta^{(1)}|q_{cm}[0]\rangle = -\frac{g}{4}\sum \limits_{j}f_{q_{cm}}(j,j+1)
\left( |(j)[0]\rangle + |(j+1)[0]\rangle \right)\;,
\end{equation}
where $\delta^{(1)}|q_{cm}[0]\rangle$ is the first order correction
term in the one-boson sector.
With all this in place, we can begin our analysis of the
bound state contribution to the dynamic structure factor
by writing down the following spectral sum:
\begin{equation}
S^{\rm bs}_{zz}({\bf k},\omega)=\sum \limits_{q_{cm} = 0}^{8\pi/b}
\delta(\omega - \varepsilon_B(q_{cm}))
\left| \langle q_{cm}[0]| S^{z}(-{\bf k}) |\Phi_0\rangle\right|^2 \; .
\label{qcmsum1}
\end{equation}
Notice the different range of summation for $q_{cm}$ in comparison
with Eqn~(\ref{qsum1}) (it is of course understood that the sum is performed
only over those sub-intervals
in $q_{cm}$ that actually support the existence of a $S=1$
bound state).
We can now use (\ref{szo}) and (\ref{o12}) and calculate the
matrix element appearing in the spectral sum to first order in
$g$ using the ${\cal O}(g)$ wavefunctions calculated above.
While it is certainly possible to use the notation of section~\ref{shadow}
and
only quote the perturbative results for the analogs of
${\cal M}_{\sigma}$ and ${\cal M}_{\phi}$, we prefer to
put everything together and directly present results for
the leading contribution to the dynamic
structure factor.
Naturally,
these results are not expected to be quantitatively accurate. Rather,
they provide us with a non-trivial example of the general
argument of section~\ref{shadow} at work.\begin{figure}
\epsfxsize=4.5in
\centerline{\epsffile{figbs2.eps}}
\vspace{0.1in}
\caption{Intensity of the bound state mode $I_b$, and the intensity
of the shadow $I_s$ to leading
order in the strong coupling expansion. Note that the value of
$k_a$ is different for the two; in each case it is chosen to
maximize the intensity. The intensities are both normalized by
the average value of the intensity in the single-particle mode
at this order in the strong coupling expansion. The value
of $g$ used is $0.7$, which is approximately right for VOPO~{\protect
\cite{oldexp}}}
\label{figbs2}
\end{figure}
The bound state leads
to the following basic contribution to the dynamic structure factor
for $k_b \in (4\pi/3b,8\pi/3b)$ and again for $k_b \in (16\pi/3b,20\pi/3b)$:
\begin{eqnarray}
S_{zz}^{\rm bs, basic}({\bf k},\omega) & = &
\frac{g^2}{16}(1-4\cos^2(\frac{k_b b}{4}))\sin^2(\frac{k_b}{2}(w-\frac{b}{2}))
\cos^2(\frac{k_a v}{2}) \delta(\omega-\varepsilon_B(k_b))\;.
\end{eqnarray}
The shadow of the bound state gives for $k_b \in (0,2\pi/3b)$,
$k_b \in (10\pi/3b,14\pi/3b)$ and $k_b \in (22\pi/3b,8\pi/b)$:
\begin{eqnarray}
S_{zz}^{\rm bs, shadow}({\bf k},\omega) & = &
\frac{g^2}{16}(1-4\sin^2(\frac{k_b b}{4}))\cos^2(\frac{k_b}{2}(w-\frac{b}{2}))
\sin^2(\frac{k_a v}{2}) \delta(\omega-\varepsilon_B^{\rm s}(k_b))\;,
\end{eqnarray}
where $\varepsilon_b^{\rm s}(k_b)/J =2 - g(4\sin^2(k_b b/4)+1)/4$ gives
us the position of the shadow band.
The intensity in the bound state mode and in its
shadow is depicted are in Fig~\ref{figbs2}.
We thus see that the bound state mode also acquires a `shadow'
as anticipated earlier on the basis of the general argument.
Of course, as mentioned previously, this strong-coupling
calculation has very little quantitative significance. A
quantitative analysis of any experimental data on the bound
state mode and it's shadow would instead follow the
analog of the procedure outlined the end of section~\ref{shadow}
with the analog of ${\cal M}_{\sigma}$ included in the
analysis and the necessary changes made to allow for the
fact that the spectral sum is to be carried out
over a different range from the single particle case (note that
this type of analysis can, in principle, distinguish between
bound state and single particle triplet modes based on the
different intensity ratios between the modes and their shadows in
the two cases).
\section{A more complicated magnetic Hamiltonian?}
\label{magnetic}
In this final section we briefly consider a
possible complication that will affect our results at a
qualitative level:
namely the possibility that the magnetic interactions
felt by the even and the odd dimers are slightly different. This will
clearly change the intensities and the dispersions of the various
modes observed as the magnetic Hamiltonian will now be invariant
under translations of $b$ and not $b/2$ along the $b$ axis. Thus,
the staggering of the dimer positions along the $a$ axis will
no longer be the only thing breaking the larger symmetry of
translations by $b/2$. Clearly, in such a
situation, we expect that the `shadow'
band intensity will be non-zero even at $k_a = 0$ (indeed we
expect that it will depend quite sensitively on the
difference in the magnetic interactions of the even and
odd dimers). To get a feel for what to expect, let us
work out the intensity of the basic single particle
mode and its shadow to leading order within a
strong coupling expansion.
The Hamiltonian we have in mind can be parameterized as:
\begin{eqnarray}
{\cal H} &= &J \sum \limits_{j\, odd}[(1+g\lambda){\bf S}_{I}(j)
\cdot {\bf S}_{II}(j)
+g(1+\mu){\bf S}_{II}(j) \cdot {\bf S}_{I}(j+1)] \nonumber \\
&& + \;
J \sum \limits_{j\, even}[(1-g\lambda){\bf S}_{I}(j)
\cdot {\bf S}_{II}(j)
+g(1-\mu){\bf S}_{II}(j) \cdot {\bf S}_{I}(j+1)] \; ,
\label{compH}
\end{eqnarray}
where we have in mind that $\lambda$ and $\mu$ are both small
parameters that model the small differences in the
magnetic properties of the even and the odd dimers.
The calculation of the ${\cal O}(g)$ single-particle contribution to the
dynamic structure factor is quite
elementary and involves nothing
new. We will therefore be correspondingly brief.
We begin our analysis by noting that it is now more natural to
count states somewhat differently. We will restrict the
momentum carried by the single particle state to lie in the
range $(0, 2\pi/b)$ and allow for two distinct bands of
single particle states labeled by $+$ and $-$ subscripts.
The energies of these two bands are easily worked out to leading
order to
be
\begin{equation}
\varepsilon_{\pm}(q) = 1 \mp \frac{g}{2}(4 {\lambda}^2 + \cos^2(qb/2)
+{\mu}^2\sin^2(qb/2))^{1/2} \; .
\end{equation}
Moreover, it is quite elementary to see that the corresponding
eigenstates
to leading order in the strong coupling expansion are (again we choose
to write down the state with $S_z = 0$ as this is what we
need to calculate the $zz$ component of the dynamic structure factor):
\begin{equation}
|q_{\pm}[0]\rangle =(\frac{b}{2L})^{1/2}\sqrt{2}{\cal P}(q)( \sum
\limits_{j \, odd} e^{iqx^b_{j}}|j[0]\rangle
+y_{\pm}(q) \sum \limits_{j \, even} e^{iqx^b_{j}}|j[0]\rangle) \; ,
\end{equation}
where ${\cal P}(q) = 1/\sqrt{1+|y_{\pm}(q)|^2}$ and
$y_{\pm}(q)$ is given as
\begin{equation}
y_{\pm}(q) = \frac{2\lambda \pm (4 {\lambda}^2 + \cos^2(qb/2)
+{\mu}^2\sin^2(qb/2))^{1/2}}{\cos(qb/2) + i \mu \sin(qb/2)}
\end{equation}
One other thing we need is the
ground state for this model, correct to ${\cal O}(g)$.
This can also be worked out
quite easily to be~\cite{kdup}
\begin{eqnarray}
|\Phi_0\rangle & = &|0\rangle + \frac{g(1+\mu)}{8}\sum \limits_{j \,
odd}\left(
|(j)[0],(j+1)[0]\rangle -|(j)[-1],(j+1)[+1]\rangle-|(j)[+1],(j+1)[-1]\rangle
\right) \nonumber \\
&& + \; \frac{g(1-\mu)}{8}\sum \limits_{j \, even}\left(
|(j)[0],(j+1)[0]\rangle -|(j)[-1],(j+1)[+1]\rangle-|(j)[+1],(j+1)[-1]\rangle
\right) \nonumber \\
&&
\end{eqnarray}
We can now use all of this to work out the one particle
piece of the dynamic structure factor.\begin{figure}
\epsfxsize=4.5in
\centerline{\epsffile{figmag1.eps}}
\vspace{0.1in}
\caption{Position of the two single particle
bands when the interactions felt by the even and
odd dimers are different. As explained in the text, the
`basic mode' and it's `shadow' are actually both hybrids
made up of the two single particle bands in the problem.
We have chosen, for illustrative purposes, the values $\lambda = 0.1$
and $\mu = 0.1$. The parameter $g$ is set equal to $0.7$, which
is approximately correct for VOPO~{\protect \cite{oldexp}}.}
\label{figmag1}
\end{figure}
The resulting expressions
are quite messy for general ${\bf k}$ and not particularly illuminating.
We will
write them down here only for the special case of $k_a =0$, as this
is where we expect a real qualitative difference due to the
complications we have introduced into the problem:
\begin{eqnarray}
S_{zz}({\bf k}, \omega) &=& \sum \limits*12
_{\alpha =
\pm}\frac{\sin^2(k_bw/2)}{4}
\left[(1+\frac{2G_{\alpha}\cos(k_b b/2)}{G^2_{\alpha}+Q^2})
+ \frac{g}{2}(\cos(k_bb/2) + \frac{2Q^2G_{\alpha}}{G^2_{\alpha}+Q^2})\right]
\nonumber \\
&&~~~~~~~~~~~~~~~~~~~~~~\times\; \delta(\omega-\varepsilon_{\alpha}(k_b)) \; ,
\label{messy1}
\end{eqnarray}
where we have defined $Q^2$ as
\begin{equation}
Q^2(k_b) = \cos^2(k_b b/2) + \mu^2 \sin^2(k_b b/2) \; ,
\end{equation}
and $G$ as
\begin{equation}
G_{\pm}(k_b) = 2 \lambda \pm \sqrt{4\lambda^2 + Q^2(k_b)} \;.
\end{equation}
The details of the above expressions are not particularly important.
We only wish to use the above to arrive at some general qualitative
conclusions about the nature of the expected intensity at
various points in the Brillouin zone. The first of these is of course
that we have some non-zero intensity at the shadow positions even
at $k_a =0$. In this context, it is important to note that
both the `basic mode' and the `shadow' are actually
hybrids made up of the $+$ band and the $-$ band.
For small enough $\mu$ and $\lambda$, the intensity switches between
the two in such a manner that we have one approximately
continuous basic mode and another much weaker shadow mode that
is also approximately continuous.\begin{figure}
\epsfxsize=4.5in
\centerline{\epsffile{figmag2.eps}}
\vspace{0.1in}
\caption{Intensity at $k_a=0$ of the two single particle bands
when the interactions felt by the even and
odd dimers are different. For clarity, the $\sin^2(k_b w/2)$
modulation of the intensity is not included in the plots.
The values of $\lambda$, $\mu$ and $g$ are set as in Fig~{\protect
\ref{figmag1}}.}
\label{figmag2}
\end{figure}
These results are summarized in Fig~\ref{figmag1} and Fig~\ref{figmag2}.
Of course, the avoided level crossing between the two bands
leads to a small jump in position of both the basic and the
shadow mode situated near $\pi/b$ and $3\pi/b$. Thus the
intensity at the shadow positions and the gap introduced
by the avoided level crossing are
sensitive indicators of the difference between the
magnetic environments of even and odd dimers.
As mentioned earlier, all experiments to date~\cite{oldexp,ens,agthesis}
are consistent with the absence of extra modes at $k_a = 0$, but in
the absence of any straightforward symmetry reason,
more
experiments at $k_a = 0$ with better sensitivity and statistics are necessary
before one can completely rule out the existence
of such complications in the magnetic Hamiltonian of the
system.
\section{Conclusion}
The calculations presented here show that simple geometric effects
can lead to the presence of the shadow bands observed recently in VOPO.
These modes are similar to the `optic' modes that
can arise generally in coupled alternating-chain systems with more than
one dimer per unit cell, for example the `chain' layers in
Sr$_{14}$Cu$_{24}$O$_{41}$~\cite{matsuda}. However, in VOPO
the shadow modes arise from a single chain.
It should be possible to test our proposal by comparing the
experimentally observed
intensity ratios with our predictions.
Moreover, as mentioned earlier, the intensity ratios can
also distinguish between single-particle and bound state modes
The possible contribution of the triplet bound state
to the spin dynamics in VOPO remains an open question at this time.
\section{Acknowledgements}
KD would like to thank the Neutron Scattering Group at ORNL for hospitality,
and ORISE for financial support for a visit during which this work
was begun.
We thank T. Barnes, M. Enderle, B.C. Sales, H. Schwenk and D.A. Tennant
for numerous helpful discussions. Work at Princeton was supported
by NSF grant DMR-9809483.
The Oak Ridge National Laboratory is
managed for the U.S. D.O.E. by Lockheed Martin Energy Research
Corporation under contract
DE-AC05-96OR22464.
|
\section{Introduction}
Electroweak physics had its experimental beginning in inelastic
neutrino scattering neutral current measurements.
Particular measurements may be interpreted directly or
combined in global fits to constrain the Higgs mass and possibilities for
new physics. The increasingly precise magnetic moment measurement of the
muon at Brookhaven will limit nonstandard possibilities.\cite{roberts}
The $Z$ mass measurement
has developed a precision in the same league with $G^{\mu}_F$
and $\alpha_{EM}$, and the shape and decays show that we understand the decay
process as well as what states are available.\cite{renton}
For example there is room
for only three neutrinos. The various $Z$ asymmetries from
LEP \cite{duckeck} and SLC \cite{rowson} give the strongest indirect constraint
on the Higgs mass. With $e^+e^- \rightarrow hadrons$ measurements at
BES \cite{blondel} and Novosibirsk \cite{steinhauser} complementing or confirming
PQCD calculations,\cite{steinhauser} the precision of $\alpha_{EM}$ evolved to
the $Z$ mass has improved, and this constraint is becoming stronger.
The combination of $W$ mass \cite{straessner,demarteau}
and top quark mass \cite{konigsberg} is more of a check at the moment.
The Tevatron analyses for $W$ and top masses with existing data are
becoming mature, and substantial improvement will come with data from
the next run, which should start in 2000. Considerable improvement
on the $W$ mass is anticipated from the LEP collaborations with
recent data, and more data at higher energy is coming.
New strategies in neutrino scattering make neutral current measurements
interesting,\cite{johnson} and deep inelastic scattering at HERA is
becoming of interest from the electroweak point of view.\cite{brisson}
The absence of the Higgs particle in direct searches \cite{pepe}
is becoming as significant
an influence on what possibilities remain as the indirect limits.
Precision
electroweak studies are continuing on many fronts including $\tau$
studies,\cite{jessop} and the pending observation of $\nu_{\tau}$
interactions.\cite{rameika} None of these efforts has allowed us to break
out of the standard framework. I summarize results as presented, but
update to Moriond 99 numbers.\cite{lepewwg}
\section{BNL 821 Muon g-2}
The study of the magnetic moment of the muon at CERN was precise enough to
demonstrate the presence of hadronic corrections.\cite{cerngm2} The goal
of the ongoing program at Brookhaven is to become precise enough to demonstrate
the electroweak corrections. More accurate calculations of the hadronic
corrections are helping to make this realistic.\cite{steinhauser}
The experiment is a muon storage ring
consisting of a continuous, finely adjustable, iron dominated
super-conducting magnet. Mapping and adjusting the field has been an
ongoing program. The momentum of the muons is adjusted to minimize the
effect of the embedded electrostatic quadrupoles on muon spin precession.
Decays are observed at instrumented windows around the ring,
with high energy decay electrons acting to spin analyze the muons.
A measurement from an initial run with pion decay injection
approaches the CERN accuracy.\cite{g2prl}
The numbers are listed in Table \ref{tab3}.
This result was limited
by low intensity and detector effects, particularly due to pion injection,
as well as the field quality.
In two more recent runs, muon injection was established, detectors
improved, and the field much improved. The available data should produce
a $\pm1$ ppm measurement. The fringe field of the inflector magnet will
be improved in order for future runs to reach the goal of $~\pm0.3$ ppm for
both signs.
\begin{table}[!bht]
\caption{Muon g-2 measurements ($\times10^{-9}$).
\label{tab3}}
\vspace{0.2cm}
\begin{center}
\footnotesize
\begin{tabular}{|l|c|}
\hline
Measurement & Result \\
\hline
CERN & $1165923.5 \pm8.4$\\
BNL E821 initial run & $1165925 \pm15$\\
\hline
Standard Model Prediction & $1165916.3 \pm0.8$\\
\hline
\end{tabular}
\end{center}
\end{table}
\section{Measurements of the $Z$}
The precision $Z$ line shape has been an adventure story with significant
implications. The measurement is
\begin{equation}
m(Z) = 91.1867 \pm0.0021\ {\rm GeV/c}^2
\end{equation}\begin{equation}
\Gamma(z) = 2.4939 \pm0.0024\ {\rm GeV}.
\end{equation}
The mass is precise enough to rank with
the weak and electromagnetic couplings as precision input. That nothing
seems missing in $Z$ decay
places serious constraints on new physics possibilities.
Heavy flavor decay rates were a problem but the popular interpretation
was otherwise ruled out even
before the deviation went away.
Agreement continues in the tail of the $Z$ at LEP2.
There is one residual small discrepancy
seen at LEP and less at Moriond by SLC,
and that is in the asymmetry in b decays.
At 2.6 $\sigma$ or less, depending on input details, the effect is not
convincing.
The $Z$ asymmetries from LEP have had some updates in $\tau$ polarization,
and though some further fine tunings are expected, most of the analyses are
pretty much complete. SLD has gotten a lot more data recently, and the
preliminary results for the last two years data dominate the measurement.
The discrepancy between the SLD $A_{LR}$ and the average of the LEP
effective weak mixing continues but has become a lot less jarring,
as seen in Table \ref{tab1}. Note that the overall average is pulled
up a bit a by the LEP $b$ asymmetry, and down a bit by SLD $A_{LR}$.
A slight residual discrepancy seems historically appropriate.
\begin{table}[!bht]
\caption{Weak mixing measurements $(sin \Theta_{W eff})$ from $Z$
forward backward and polarization asymmetries.
\label{tab1}}
\vspace{0.2cm}
\begin{center}
\footnotesize
\begin{tabular}{|l|c|}
\hline
Measurement & Result \\
\hline
LEP lepton fb & $0.23117 \pm0.00054$\\
LEP $\tau$ pol. $A_{\tau}$ & $0.23202 \pm0.00057$\\
LEP $\tau$ pol. $A_e$ & $0.23141 \pm0.00065$\\
LEP $b$ fb & $0.23223 \pm0.00036$\\
LEP $c$ fb & $0.2321 \pm0.0010$\\
LEP jet charge fb & $0.2321 \pm0.0010$\\
SLD pol. $A_{LR}$ & $0.23109 \pm0.00029$\\
\hline
Average ($\chi^2$/DF 7.8/6) & $0.23157 \pm0.00018$\\
\hline
\end{tabular}
\end{center}
\end{table}
The impact of the effective weak mixing measurement is improving as
$\alpha_{EM}(m(Z))$ is better determined with PQCD calculations.
There is discrepancy with old SPEAR hadron rates but new points from
BES agree. Both PQCD and data driven calculations are agreeing and
improving.\cite{steinhauser}
\section{Measurements of the $W$}
The Tevatron Collider experiments have advanced the program begun at the
CERN $S\bar{p}pS$ collider, including $W$ mass measurements using
leptonic decay transverse mass. Updates at Moriond had D\O\ adding
plug electrons, and CDF adding the most recent electron sample. Beyond that,
further improvement will come with the next run, expected to start in 2000.
The LEP experiments have threshold $W$ mass measurements, and increasingly
precise direct reconstruction measurements. Possible QCD systematics in
the four quark mode are being confronted. The large sample at $\sqrt{s}=189$
GeV should allow a precision of $\pm40-45$ MeV/c$^2$ when fully analyzed;
the ALEPH and L3 analyses included this sample in the Moriond update.
Further data will continue to be collected through 2000. The measurements are
listed in Table \ref{tab2}.
\begin{table}[!bht]
\caption{$W$ mass measurements.
\label{tab2}}
\vspace{0.2cm}
\begin{center}
\footnotesize
\begin{tabular}{|l|c|}
\hline
Measurement & Result \\
\hline
LEP threshold & $80.400 \pm0.221$\\
LEP $qqqq$ & $80.485 \pm0.103$\\
LEP $\ell\nu qq$ & $80.318 \pm0.073$\\
Tevatron $\ell\nu$ & $80.448 \pm0.062$\\
\hline
Direct Average & $80.410 \pm0.044$\\
\hline
Indirect fit (LEPEWWG) & $80.364 \pm0.029$\\
\hline
\end{tabular}
\end{center}
\end{table}
Searches for nonstandard $W$ and $Z$ couplings are now dominated by LEP
measurements, although D\O\ makes a notable contribution. The large new
data sample at LEP should improve coupling limits by about a factor of three
when fully analyzed.
\section{Measurements of the Top Quark}
The impact of improving the $W$ precision will be limited by the precision
of the top quark mass measurement. The Tevatron data analyses are largely
complete with the two experiments in the different channels consistent,
as can be seen in Table \ref{tab4}.
A couple of years of new data should improve the precision by a factor of at
least two.
\begin{table}[!bht]
\caption{Top quark mass measurements. When two errors are given,
the first is statistical and the second systematic.
\label{tab4}}
\vspace{0.2cm}
\begin{center}
\footnotesize
\begin{tabular}{|l|c|}
\hline
Measurement & Result \\
\hline
CDF $\ell\nu qqqq$ & $175.9 \pm4.8 \pm5.3$\\
CDF $\ell\nu\ell\nu qq$ & $167.4 \pm10.3 \pm4.8$\\
CDF $qqqqqq$ & $186.0 \pm10.0 \pm5.7$\\
D\O\ $\ell\nu qqqq$ & $173.3 \pm5.6 \pm5.5$ \\
D\O\ $\ell\nu\ell\nu qq$ & $168.4 \pm12.3 \pm3.6$ \\
\hline
Average & $174.3 \pm5.1$\\
\hline
\end{tabular}
\end{center}
\end{table}
Detailed studies of top production and decay, including
limits on nonstandard decays, have begun. The expected increase in
statistics at the Tevatron once the upgraded collider and detectors get
going should have a salutary effect on these.
\section{Deep Inelastic Scattering}
The NuTeV group has revived the contribution of neutrinos to the
electroweak program. By using a carefully designed beam, a neutrino beam free
of anti-neutrinos and vice versa allows the difference in neutral to
charged current rates to be used to measure weak mixing. The new data
has similar statistics to CCFR but much improved systematics. The
electroweak physics implications are illustrated in Fig. \ref{radish}.
\begin{figure}[!bht]
\hspace*{.5in}\psfig{figure=better.eps,height=3.in}
\caption{$W$ Mass versus Top Mass showing $68\%$ CL allowed regions
for the combined direct measurements,
the SLC/LEP combined indirect measurement, NuTeV and CCFR with Higgs mass
predictions. Note the different slope for the different neutrino analyses.
\label{radish}}
\end{figure}
The HERA experiments still see a small excess at high $Q^2$, but
not so dramatic as it seemed a year ago. The data are sufficient to
see propagator effects in NC and CC events. The $t$ channel $W$ mass derived is
compatible with direct measurements, and an electroweak program
is getting started.
A neutrino beam at Fermilab for the NUMI project will have space available
for a near detector. There are some possibilities for PDF studies.\cite{morfin}
\section{The Higgs Search}
The Higgs particle of the minimal standard model has given no direct sign of
its presence. The lower limit on its mass is growing with the LEP energy as
searches are made for the process $e^+e^- \rightarrow Z^* \rightarrow ZH$.
Most signatures involve $H \rightarrow \bar{b}b$, so that $Z$ pairs give an
irreducible background. Fortunately
$Z \rightarrow \bar{b}b$ is now well understood.
The 189 GeV data has now been analyzed by the individual collaborations giving
limits as high as\begin{equation}
m(H) > 95.2\ {\rm GeV/c}^2 \ 95\%\ {\rm CL}.\end{equation} The combination of experiments
will give some improvement. With data at 200 GeV, a limit or
discovery reach up to $\sim109$ GeV/c$^2$ is in prospect.
The recent PDG global fits to electroweak data,\cite{erler} using the
data as of Vancouver 98, gives a most
favored Higgs mass is 107 GeV/c$^2$, so the search is covering quite
interesting territory. The limit is threatening the viability of
various popular scenarios for extending the standard model.
If the Higgs
is still missing at the end of the LEP program, with enough luminosity the
Tevatron detectors could extend the Higgs search. Eventually LHC detectors
will make the search comprehensive.
\section{Tau Physics}
The detailed study of $\tau$ decays involves precise decay parameters,
neutrino mass limits, and rare decay searches including lepton number violation
and CP violation in the $K^0_s\pi\nu$ angular distribution.
There is plenty of room for non-standard model physics.
The program being pursued at CLEO will be being joined at Babar and
Belle.
The E872 collaboration at Fermilab is searching for evidence of
$\nu_{\tau}$ interactions in emulsion at Fermilab. With part of
the data measured and analyzed, they have six candidates. Although this
corresponds well to expectations, systematic studies to eliminate
background possibilities are pending, but an announcement that
interactions have been observed is expected soon.
\section{Conclusions}
The simplest scenario for the standard model, with one residual Higgs
particle, remains viable. In the global fits, the strongest constraint
comes from measurements of $Z$ asymmetries. These dominate the thinness
of the indirect allowed region of Fig. \ref{radish}.
Some updates on these measurements
are pending, but more progress seems likely from $\alpha_{EM}(m(Z))$
improvement.
The $W$ mass measurement is improving considerably, and further improvement and
an improved top mass measurement, as will come with the next Tevatron run,
is needed to compete with the $Z$ asymmetries.
The direct Higgs search is beginning to cut into the indirect fit allowed
region. Perhaps a positive finding will come soon, but it seems like LHC will
be needed to create a contradiction. Perhaps a contradiction, which would
break us out of our mold, will come from one of the many electroweak studies
which do not directly contribute to the Higgs picture.
\section*{Acknowledgments}
I am grateful to all the speakers whose material I am summarizing,
and to our wonderful hosts. The work was supported in part by
the United States Department of Energy, Division of High Energy
Physics, Contract W-31-109-ENG-38.
\section*{References}
|
\section{Direct Neutrino Mass Searches}
From experiments we know that $m_{\nu_e} < {\rm few~eV}$. Even if
``formal" upper limits of 5 eV (95\% C.L.) \cite{Mainz} and 1.7 eV
(95\% C.L.) \cite{Troitsk} have been given in 1998, they are obtained with
negative measured square masses, what points to a systematic error. In fact
the Particle Data Book gives an upper bound of 15 eV \cite{PDB}. The bounds
on the effective Majorana $\nu_e$ mass from $\beta\beta o\nu$ decays are of
about 1 eV. The Heidelberg-Moscow $^{76}$Ge experiment quoted an upper
bound of 0.2 eV \cite{Heidelberg}, but there are still uncertanties related to
the evaluations of the nuclear form factors involved.
There has been no recent change in the upper bound of the $\nu_\mu$ mass,
$m_{\nu_\mu} < 0.17$ MeV \cite{PDB}, while for $\nu_\tau$ a preliminary analysis of data from
LEP, combining results of ALEPH and OPAL gives $m_{\nu_e} < 15~{\rm MeV}$
(95\% C.L.) \cite{Cerutti}. ALEPH and OPAL separately quoted upper
bounds of 18.2 MeV (95\% C.L.) and 27.6 MeV (95\% C.L.) respectively, while
new results of CLEO II give $m_{\nu_\tau} < 30~{\rm MeV}$ (95\% C.L.) \cite{CLEOII}.
Better bounds than these are derived from cosmological arguments if neutrinos are stable, and, if neutrinos are unstable, cosmology and astrophysics give bounds on
their lifetimes and decay modes.
\section{Expansion, Age and Content of the Universe.}
The Hot Big Bang (BB), the standard model of cosmology, establishes that the
Universe is homogeneous, isotropic and expanding from a state of extremely
high temperature $T$ and density $\rho$. The Hubble parameter $H$ (constant in space
but not in time) provides the proportionality between the velocity of recession
$v$ of faraway objects and their relative distance $d$, $v=Hd$ and $H=h$
100 km/sec Mpc. Most observational determination are converging to
$h=0.65 \pm 0.15$ \cite{h} (for example, $h= 0.67\pm 0.12$ comes from type Ia
Supernovae \cite{h} while $0.72 \pm 0.17$ comes from combining cepheids
studies and other methods \cite{Freedman98}), for the present value of the
expansion parameter. The lifetime of the Universe is counted from the
moment the expansion started, taken to be $t=0$. The cooling of white dwarfs
provides a lower bound to the present age of the Universe of $t_o > 10$ Gyr \cite{dwarfs} and the age of the oldest globular clusters gives $t_o = 13-25$ Gyr ($t_o = 11.5 \pm 1.3~{\rm Gyr}$ \cite{Chaboyer}, plus 1-2 Gyr for the formation of the
galaxy where the globular cluster is. The expansion rate, lifetime and the
content of the Universe are not independent.
In fact, $Ht_o = (h/0.75)(t_o/13$
Gyr) is a function of the densities of matter, radiation and vacuum in the
Universe. For an empty Universe $Ht_o=1$. The gravitational attraction of
matter and radiation slows down the expansion, so that $Ht_o<1$ in a matter
or radiation dominated Universe. In a vacuum dominated Universe $Ht_o>1$ instead, because gravitation is repulsive. In synthesis, a larger $H$ implies
a shorter $t_o$ and decreasing the matter or radiation content or increasing
the vacuum content of the Universe makes $t_o$ longer.
Densities $\rho_i$ are usualy given in units of the density of a spatially
flat Universe, the critical density $\rho_c = 10.5~{\rm h}^2~{\rm (keV/cm)}^3$
= 1.88 10$^{-29}~ h^2$ (g/cm$^3$) as $\Omega_i = \rho_i/\rho_c$. We call
$\rho_r$, the density of radiation (photons and other relativistic
particles) and $\rho_m$ the density matter (non-relativistic particles).
We will call here $\Omega_o = \Omega_m + \Omega_r$.
The vacuum energy provides a cosmological constant $\Lambda$, such that
$\rho_{\rm vac} = \Lambda/8\pi{\rm G} = \rho_\Lambda$. In our notation the total density of the Universe is $\Omega = \Omega_o + \Omega_\Lambda$.
Matter is much more
abundant than radiation at present, thus $\Omega_o \simeq
\Omega_m$. For a flat
matter dominated Universe,
$Ht_o = 2/3$. This means (using the relation given above) that
$t_o\geq 13$ Gyr (necesary to accomodate globular clusters) requires
$h \leq 0.50$, a very low value of $h$. Even using the absolute lower bound
$t_o \geq 10$ Gyr, we obtain $h \leq 0.65$, still values of h lower than many
present determinations. If $h$ is actually larger than 0.65, then we live in
a Universe with a non-zero cosmological constant or open or both. This tension
between $h$ and $t_o$ was until recently called the ``age crisis", but this
is not called a ``crisis" any longer.
One speaks of a ``crisis" only when a
paradigm is challenged. This paradigm was that of a flat matter dominated
Universe with $\Lambda=0$, until recently the model preferred by most
cosmologists due to its simplicity and aesthetic appeal. This paradigm has
now been changed, mostly by the Type Ia Supernovae (SN) data, which point to a
non-zero value of $\Lambda$, and, to a lesser extent, by data on the LSS of the Universe, which suggests $\Omega_m < 1$. Type Ia SN are white dwarfs which accrete mass from a comparison star and explode when reaching the Chandrasekar limit of about 1.4 solar masses, i.e. the maximum mass that can be supported
by the pressure of degenerate electrons. Two different groups \cite{Perlmutter9798} \cite{Garnavich} using Type Ia SN as ``calibrated" candles
(i.e. objects of known intrinsic luminosity) measured the curvature of the
relation between distances and velocities (of which the linear term is given
by the present value of the Hubble parameter $H$). Recession velocities are actually translated
into redshifts $z$. Thus, distances $d$ are given by $d=H^{-1}[z+O(z^2)]$.
At $z\simeq 0.5$, where most of the SN used are, the coefficient of the $z^2$ term depends
on the linear combination $\Omega_\Lambda - \Omega_m$ \cite{white}, and, in fact, the
confidence region given by both groups \cite{Perlmutter9798} \cite{Garnavich} lies along the line $\Omega_\Lambda-\Omega_m \simeq 0.5$,
in the $(\Omega_m,\Omega_\Lambda)$ plane. With 42 high-redshift supernovae,
Perlmutter {\it et al.} \cite{Perlmutter9798} found a non-zero positive cosmological constant with
probability larger than 99\% (0.9979 in their primary fit). In order to
determine $\Omega_\Lambda$ and $\Omega_m$ separately, one would wish to
combine this result with that of a complementary technique sensitive to a
different linear combination of the two quantities. In fact, an almost orthogonal linear combination is provided by the position of the
first acoustic peak
in the multipole expansion of the CMBR anisotropy, which depends on the total
energy density of the Universe, $\Omega= \Omega_\Lambda + \Omega_m$ (see below). Data already show
that this sum is not very different from one.
The crossing of $\Omega_\Lambda -
\Omega_m \simeq 0.5$ and $\Omega_\Lambda + \Omega_m \simeq 1$, suggests the values $\Omega_m \simeq 0.3$ and $\Omega_\Lambda \simeq 0.7$
(which would
saturate an earlier upper bound $\Omega_\Lambda < 0.7$ obtained for a flat
Universe from the frequency of gravitational lensing
of quasars \cite{quasars}).
In a few years,
with better data, $\Omega_m$ and $\Omega_\Lambda$ will be determined in this way to within 10\%. The satellite MAP, will be
launched by NASA in the year 2000 to measure the anisotropy of the CMBR (a European satellite, the Planck Surveyor, is expected for 2007).
The CMBR provides a snapshot of the Universe at the moment of recombination,
$t_{rec} \simeq 3\times 10^5$y, when atoms first became a stable. Photons, which
had a very short mean free path in the ionized matter before recombination,
interact for the last time and reach us from that ``surface of last scattering".
This radiation has the best black-body spectrum in the Universe with deviation
of less than 0.005\% and a temperature $T = 2.7277^\circ$K measured by the
COBE (Cosmic Background Explorer) Satellite \cite{Fixsen}. This radiation is
remarkably isotropic. Anisotropies are due to our motion with
respect to the CMBR rest frame (which generates a dipole anisotropy) and due to
the density inhomogeneities that triggered structure formation in the Universe.
Results from COBE and other experiments in balloons show temperature
anisotropies $(\delta T/T) \equiv (T-\bar T)/\bar T < 10^{-4}$, where $\bar T$
is the average temperature (given above). Once within the horizon, the
primordial density perturbations (in dark matter) set up sound or acoustic
oscillations in the fluid formed by photons, electrons and baryons before
recombination. In the surface of last scattering the peaks of
compression and rarefaction (for scales that are caught at extrema of their
oscillations) are seen as hot and cold spots respectively, both of which
appear as peaks in
the multipole expansion of the power spectrum of CMBR anisotropies. The
horizon size at recombination corresponds to an angle $\theta_H \simeq 1^\circ \sqrt{\Omega}$ in the present sky. This apparent angular size
depends on the geometry of the Universe: it is larger for a closed Universe
$(\Omega >1)$ and smaller in an open one $(\Omega < 1)$ \cite{Jungman}.
In multipole number $\ell \simeq 200^\circ/\theta$, the position of the
horizon is at $\ell_H \simeq (200 / \sqrt{\Omega})$. Angles larger
than $\theta_H$, $\ell < \ell_H$, correspond
to causally disconnected regions at the time of emission of the CMBR
photons, where no acoustic oscillations could have been set. Only values
$\ell \geq \ell_H$ correspond to scales within the horizon. Thus the
position of the first acoustic peak (a compression peak) should happen at
$\ell_H$, which depends on $\Omega$. Present data on
anisotropies, even if not conclusive, show the first peak at $\ell$'s not much
lower than 100 or higher than 300.
\section{Cosmic Energy Densities}
Maybe the most important cosmological constraint on stable neutrinos is the
mass bound coming from their cosmic energy density. In the same way we have
a cosmic background of photons, we expect the existence of a yet never seen
cosmic background of relic neutrinos. Knowing so well the CMBR temperature,
we know with great accuracy the number and energy density of the CMBR photons
which are the most abundant in the Universe by several orders of magnitude,
$n_\gamma = 2\zeta(3) T^3/\pi^2 = 412/{\rm cm}^3$, $\rho_\gamma =
\pi^2T^4/15 = 4.71\times 10^{-34}~({\rm g/cm}^3)$. We can compute the
expected abundance of neutrinos relative to photons. For light standard neutrinos (of mass $m_\nu < 1$ MeV and no lepton number asymmetry)
$n_{\nu_i} + n_{\bar\nu_i} = (3/11) n_\gamma = 112/{\rm cm}^3$ (including both
neutrinos and antineutrinos in equal numbers) for each light neutrino species.
The temperature of neutrinos is lower than that of photons
$T_\nu = (4/11)^{1/3} T = 1.9^\circ{\rm K} = 1.6~10^{-4}~{\rm eV}$. Knowing
$T_\nu$, we can compute the contribution of each relativistic $\nu$-species to the present radiation energy, usually parametrized as
$\rho_{\rm rad} = (\pi^2/30)~g_*(T)~ T^4$, where $g_*(T)$ is the effective
number of relativistic degrees of freedom. Every standard neutrino species
(with no lepton number asymmetry) adds 0.454 to $g^*$, while photons
contribute with 2. Photons and three relativistic neutino species add up to
$g_* = 3.362$ and $\Omega_{\rm rad} h^2 \simeq 4\times 10^{-5}(g_*/3.36)$.
If one or more standard neutrino species are non-relativistic at present
$(m_{\nu_i}>T_\nu)$, then their contribution to the present density of the
Universe is $\rho_\nu = \sum_i m_{\nu_i}(n_{\nu_i} + n_{\bar\nu_i}) =
\Omega_\nu\rho_c$, thus
\begin{equation}
\Omega_\nu h^2 = \frac{\sum_i m_{\nu_i}}{92~{\rm eV}}~.
\end{equation}
Only left-handed (non-relativistic) neutrinos (with no lepton number asymmetry) are considered here (for
Dirac neutrino masses $< 1~{\rm keV}$ this is correct, because the
contribution of the right-handed states is negligible). If $m_{\nu_i} =
0.4~h~{\rm eV}$ (0.3 eV with $h$ = 0.65) standard neutrinos would be as abundant as
luminous matter, namely the matter associated with typical stellar populations,
which is $\Omega_{\rm lum} \simeq 0.004~h^{-1}$ (0.6 $10^{-2}$ for $h$ = 0.65).
This matter is baryonic, but Big Bang Nucleosynthesis (BBNS) arguments
emply that the total density of baryons, $\Omega_B$, is larger. Estimates based on the sole
density of $D$ \cite{Burles}, whose primordial abundance is the best
known among the light elements \cite{Tytler}, give $\Omega_B =
(0.019 \pm 0.0024)~h^{-2}$, comparable to the density of a standard 2 eV
neutrino. Due to uncertainties in the observational
upper bound on the abundance $^4$He, only $D$ is used to obtain this range.
Also due to this uncertainty the limit on $N_\nu$,
the number of equivalent standard neutrino
species in equilibrium during NS, is uncler. At present
there are two estimates of primordial $^4$He. The lowest one,\cite{Olive} together
with data on $D$ would require $N_\nu < 3$, while the higher \cite{Izotov},
with a prior $N_\nu > 3$, implies $N_\nu < 3.2$ (95\% C.L.) \cite{Burles}.
The gravitationally dominant mass component of the Universe is ``dark", i.e.
it is not seen either in emission or absorption of any type of electromagnetic
radiation. This is the dark matter (DM). Recent measurements give
$\Omega_{\rm DM} > 0.15$ (an absolute lower bound, coming from satellites of
spiral galaxies), $\Omega_{\rm DM} = (0.19 \pm 0.06)$ B (with B $\simeq$ 1,
from the mass over light ratio of clusters), $\Omega_{\rm DM} = 0.44 \pm 0.11$
(from the baryon fraction in clusters, using BBNS), $\Omega_{\rm DM} \simeq
0.55 \pm 0.17$ (from the abundance of high-$z$ clusters). Notice
that none of these dynamical estimates reaches 1. Important for neutrinos
is a bound that depends on the total density of DM, coming from structure
formation arguments (presented in the following section), that say that most of the DM in the Universe
should be cold (i.e. non relativistic at temperatures of about $T \simeq 1$ keV, when galaxies should start forming), called CDM, and
only a small amount could be hot
(i.e. relativistic at $T \simeq 1$ keV, such as light neutrinos), called HDM. This bound is
$\Omega_\nu \leq (0.2-0.3)\Omega_{\rm m}$, where
$\Omega_{\rm m} \simeq \Omega_{\rm DM} = \Omega_{\rm CDM}+ \Omega_\nu$. With $\Omega_{\rm m} = 1$,
$\Omega_\nu \leq 0.2$ gives an often quoted bound $\sum_i m_{\nu_i} \leq
5~{\rm eV}$, for h = 0.5 (this low value of $h$ is necessary in a flat matter
dominated Universe to account for the age of the Universe, as mentioned
above). However, lower values of $\Omega_{\rm DM}$, as measurements seen now to point to, would lead to more
stringent bounds on $\Omega_\nu$.
We have spoken so far about neutrinos with no or negligible leton number asymmetry. However lepton asymmetries in neutrinos may be large (see the end of Sec. 4).
\section{Structure Formation of the Universe.}
The Universe looks lumpy at scales $\lambda \simeq 100$ Mpc, we see galaxies,
clusters, superclusters, voids, walls. But it was very smooth at the surface
of last scattering of the CMBR and later.
Inhomogeneities have been seen as anisotropies in the CMBR, so
the density contrast $\delta\rho/\rho \equiv (\rho(x)-\rho)/\rho$ (where $\rho$ is the average density)
cannot be much larger than $\delta T/T \simeq 10^{-4}$.
So inhomogeneities in density start small and grow through the Jeans (or
gravitational) instability; gravitation tends to further empty underdense
regions and to further increase the density of overdense regions. One can
follow analytically the evolution due to gravity of the density contrast
in the linear regime, where $\delta\rho/\rho < 1$. In a static fluid the rate
of growth of $\delta\rho/\rho$ is exponential, but in the Universe (an
expanding fluid) it slows down into either a power law, $\delta\rho/\rho
\sim a(t)$, in a matter dominated Universe, or it stops,
$\delta\rho/\rho \simeq$ constant, in a radiation or a curvature dominated
Universe (a matter dominated open Universe becomes curvature dominated for
$a(t) \geq \Omega_o/(1-\Omega_o)$). Here $a(t)$ is the scale factor of the Universe, which accounts for the Hubble expansion of the linear
dimensions of the Universe.
Perturbations have different physical linear dimensions $\lambda = a(t) \lambda_{\rm com}$, where $\lambda_{\rm com}$ are
linear dimensions measured in comoving coordinates (those
that expand with the Hubble flow). With the usual choice of $a = 1$
at present, $\lambda_{\rm {com}}$ are the present actual linear dimensions.
Since $a\sim t^\alpha$ with $\alpha < 1$ while
the horizon $ct$ grows linearly with $t$, the horizon increases
with time even in comoving coordinates, encompassing more material as
time goes. When $\lambda = ct$ we say
the perturbation of size $\lambda$ ``enters" into the horizon, we could better
say the perturbation is first encompassed by the horizon. This moment is
called ``horizon-crossing" and it happens at different times for different
linear scales $\lambda$, larger scales cross later.
Independently of the origin of the primordial fluctuation, it is convenient
to specify the spectrum of fluctuations at horizon-crossing,
$(\delta\rho/\rho)_{\rm hor}$. A spectrum scale
invariant at horizon-crossing, namely with $(\delta\rho/\rho)_{\rm hor}$ =
constant, is called a Harrison-Zel'dovich spectrum. COBE observations have shown the spectrum at horizon-crossing is in fact scale
invariant or very close to it.
After horizon-crossing, physical interactions act upon the inhomogeneities
and generate a ``processed" spectrum, which determines which structures
are formed first.
This question leads to the distinction
of three types of DM: hot (HDM), warm and cold (CDM) (i.e. relativistic, becoming non-relativistic and non-relativistic at temperatures of order keV).
Simulations have shown that CDM must be the most abundant form of matter,
because the ``processed" spectrum of perturbations generated in standard
CDM models reproduces the observations within 10\%. Standard CDM models,
make the simplest assumptions, namely $\Omega_{\rm CDM} + \Omega_{\rm B} \simeq
\Omega_o = 1,~\Lambda = 0$, scale invariant perturbations at horizon crossing,
and a scale independent ``biasing" by which only the highest peaks in the CDM
density distribution end up forming galaxies. There is only one feature in
the processed spectrum of CDM perturbations, a change of slope at the present
scale that corresponds to the horizon at the moment of matter-radiation
equality, $\lambda_{\rm eq}$.
The Universe is matter dominated at present, but due to the different
evolution with temperature $T$ of the density of matter and radiation,
$\rho_{\rm m} \sim T^3$, $\rho_{\rm r} \sim T^4$, the radiation was
dominant in the past, at $T>T_{\rm eq}$, where $T_{\rm eq}$
is the temperature of matter-radiation equality $\rho_{\rm r}(T_{\rm eq}) =
\rho_{\rm m}(T_{\rm eq})$, $T_{\rm eq} \simeq 5.8 {\rm eV}
\Omega_oh^2(3.36/g_*)$. $\Omega_o$ is the present matter density
(neglecting the present small radiation contribution) and
$g_*$ is the number of effective relativistic degrees of freedom
($g_* = 3.36$ with photons and three relativistic neutrino species). The
present physical size of the horizon then is
\begin{equation}
\lambda_{\rm eq} \simeq 10~ {\rm Mpc}
\left(\frac{g_*}{3.36}\right)^{1/2} \frac{1}{\Omega_oh^2}~.
\end{equation}
Perturbations with $\lambda <
\lambda_{\rm eq}$ enter into the horizon at $t < t_{\rm eq}$,
when the Universe is radiation dominated.
They cannot grow while the Universe is
radiation dominated, so they all start growing together at $t = t_{\rm eq}$
and they roughly have the same amplitude today, if they all start with the
same amplitude at horizon crossing. Perturbations with
$\lambda > \lambda_{\rm eq}$, instead,
enter into the horizon at $t > t_{\rm eq}$, when the Universe is matter
dominated, and, thus, start growing immediately. Consequently, perturbations
at larger scales enter later, have less time to grow and their amplitude is
smaller at present. Once $\lambda_{\rm eq}$ (the location of the change
slope) is fixed, the only remaining free parameter in the processed spectrum of CDM is an overall normalization, provided by the CMBR anisotropy measured by COBE at
large scales, $\theta > 20^\circ$. Density perturbations at these large
scales entered into the horizon very recently (so they did not grow much),
thus providing a measurement of $(\delta\rho/\rho)$ at horizon
crossing, $(\delta\rho/\rho)_{\rm hor}$
(for more details, see e.g. Ref. 21).
While both the shape and normalization so obtained are
almost right, they do not fit the observations \cite{Ostriker}.
The spectrum of standard CDM models has too much
power on small scales (large k$\sim \lambda^{-1}$), the scales of galaxy clusters and smaller.
Once the normalization given by COBE is fixed, there are
several possibilities to change the spectrum to agree with observations.
Because HDM tends to erase structure at small scales (while neutrinos are relativistic)
one of the solutions consists in adding to the CDM a bit of HDM, namely
neutrinos, in what are called mixed DM (MDM) or hot-cold DM (HCDM)
models \cite{Shafi}.
In particular, models with $\Omega_\nu=0.2$, what corresponds to
$\sum_i {m_{{\nu}_{i}}}$ = 5 eV, and the rest of $\Omega_o$ in CDM plus
some baryons, with $\Omega = 1$, $\Lambda = 0$ and a scale invariant
spectrum of fluctuations at horizon crossing work well. However
other possible variations of the standard CDM models also work well
to fit the LSS data,
for example that of a ``tilted''
primordial spectrum of fluctuations at horizon crossing, one that slightly
favors larger scales over smaller scales
(instead of the flat, scale invariant, Harrison-Zel'dovich spectrum) within
the COBE observational limits. This is called ``tilted'' CDM (TCDM).\cite{Cen}
A mixed model with
both some neutrinos and some ``tilt"
also does work, and in these models
$\Omega_\nu < 0.2$ (see, for example,\cite{DGT}).
Another family of solutions is obtained by realizing that
a shift towards larger scales of the only feature in the CDM spectrum, i.e.
$\lambda_{\rm eq}$, given in Eq. (2), the scale where the
slope in the spectrum changes,
is enough to provide good agreement with observations, since
it effectively amounts to increasing the power of the spectrum at scales
larger than the break point
$(\lambda > \lambda_{\rm eq})$ with respect to those smaller than it
$(\lambda < \lambda_{\rm eq})$.
Using Eq. (2) the relation
$\lambda_{\rm eq} \equiv (10h^{-1}M_{\rm pc})\Gamma^{-1}$ defines
the ``shape parameter" \cite{Efstathiou}
$\Gamma \equiv \Omega_oh(g_*/3.36)^{-1/2}$. The LSS
data require $\Gamma \simeq 0.25\pm 0.05$, while standard CDM models (with
the standard choices of $h = 0.5, \Omega_o = 1,~g_* = 3.36$) has $\Gamma =
0.05$. In fact, as we have explained, a larger $\lambda_{\rm eq}$, thus a
smaller $\Gamma$, would provide agreement with data. In order to lower the
value of $\Gamma$ with respect to that of the standard CDM model one needs to
either, 1): lower $h$, $h < 0.5$ \cite{Bartlett}
(what implies an older Universe and is very unlikely given the present determinations of $h$), or 2): increase
$g_*$ (namely increase the radiation content of the Universe at
$t_{\rm eq}$), or 3): lower $\Omega_o$ (i.e. take $\Omega_o < 1$), so that
we either live in an open Universe (open CDM models, OCDM) if
$\Lambda = 0$ or in a Universe with a cosmological constant that provides
$\Omega_{\rm vac} = 1-\Omega_o$ ($\Lambda$CDM models \cite{Turner}), or 4: a combination of all three above.
A way of obtaining the large amount of radiation needed for the
second possibility is through a heavy
neutrino decaying into relativistic particles, i.e. radiation,
with the right combination of mass and
lifetime, in so-called $\tau$CDM models \cite{Bardeen} \cite{Dodel}.
A massive neutrino matter
dominates the energy density of the Universe as soon as it becomes
non-relativistic, i.e. as soon as
$m_\nu \geq T$ (since $n_\nu \simeq n_\gamma$ and
$\rho_\nu = n_\nu m_\nu,~\rho_{\rm rad} \simeq n_\gamma T$), thus their decay
products radiation-dominate the Universe at decay. For $m_\nu < 1$ MeV standard neutrinos the
right mass-lifetime combination lie on a narrow strip around the previously
mentioned ``galaxy formation" bound \cite{Hut}.
Near this bound, at the boundary
between being irrelevant and harmful, unstable neutrinos could help in the
formation of structure in the Universe \cite{Bardeen}.
A heavier neutrino, of $m_\nu \simeq 1 - 10$ MeV, necessarily $\nu_\tau$,
decaying at or just before nucleosynthesis, $\tau = 0.1 - 100$ sec,
would also provide a solution \cite{Dodel}.
The $\nu_\tau$ decay modes involved here should all be into neutral
particles, $\nu_\tau \to 3~\nu'{\rm s}$ or $\nu_\tau \to \nu\phi$, with
$\phi$ a Majoron (a zero mass Goldstone boson) for example. All visible
modes, i.e. producing electrons or photons, are forbidden in the necessary
range.
Radiation-matter equality may also be delayed
by the existence of large lepton asymmetries in neutrinos, so that these may be more abundant than photons \cite{Sarkar},
$(n_\nu/n_\gamma)> 1$ (and dominate the entropy $s$ of the Universe,
$n_\nu /s = O(1)$). Relic neutrinos this abundant would be Fermi-degenerate, since their chemical potential $\mu_\nu$
would be larger than their temperature. Let us recall that, while charge neutrality imposes a lepton number
asymmetry in electrons as large as the baryon asymmetry in protons, i.e.
$(n_e-n_{\bar e})/n_\gamma\simeq 10^{-10}$, no such restrictive bound operates
on neutrinos.
We will return below to this very recently explored possiblity,
that we call LCDM (for CDM with large lepton asymmetry L).
All these modified CDM models seem to be able to fit present large scale structure data, however
they predict very different patterns of acoustic peaks in the CMBR
anisotropy power spectrum, and already present data on this anisotropy allow to constrain the models. All recent analizes perform a combined fit to LSS and CMBR anisotropy data.
CHDM \cite{Gawiser}, $\tau$CDM \cite{Jenkins}, and LCDM \cite{Sarkar} have been favorably compared with others in their ability
to fit LSS and CMBR data.
Gawiser and Silk \cite{Gawiser} claimed that CHDM with
5 eV total neutrino mass ($\Omega_\nu = 0.2$, $\Omega_m = 1$, $\Omega_\Lambda = 0$, $h = 0.5$) gives the best fit among all the 10 models they studied.
However, they mention that this model, as all others with $\Omega=\Omega_0=1$ (flat matter dominated Universe) may not account for the recent evidence for early galaxy formation. In fact, evidence has been found for the
existence of a large amount of bright galaxies rather early, at redshifts
$z\simeq 3$, and Somerville, Primack and Faber \cite{Somerville} concluded that
no $\Omega =\Omega_o= 1$ model with a realistic power spectrum
makes anywhere near enough of them. This can be understood by recalling that
in a matter-dominated universe $(\delta\rho/\rho$) grows as the scale
factor while in a curvature or $\Lambda$-dominated universe the growth of
$(\delta\rho/\rho)$ stops. Thus, in order to get to the same present level
of structure, the density contrast of perturbations $(\delta\rho/\rho)$ should be bigger at early times in the
later case (growth of $(\delta\rho/\rho)$ stops at some point in the past)
than in the former (in which the growth of $(\delta\rho/\rho)$ continues
until now). Based on these considerations, as well as on the Type Ia SN data,
which favor $\Lambda > 0$, Primack and Gross \cite{Primack} studied the
combined fit to LSS and CMBR data of a $\Lambda$CHDM model,namely flat models
with $\Omega_m < 1$ and $\Omega_\Lambda= 1- \Omega_m$ , and with some HDM in neutrinos. They found
the best fits had $\Omega_m$ = 0.5 (0.6) with ($\Omega_\nu/\Omega_m)$ =
0.1 (0.2) (namely $\Omega_\nu$ = 0.05 (0.12))
corresponding to $\sum_i m_{\nu_i} = 1.6 (4)$ eV (since they used h = 0.6). They found that the addition of HDM
does not change a $\Lambda$CDM model by much, contrary to the substantial
improvement this addition provided to standard CDM models. However, they also
concluded that $\Lambda$CDM and also $\Lambda$CHDM models provide a
relatively poor fit to the LSS data (to the power spectrum near the peak),
what is also mentioned in Ref. 33. The new large scale surveys
under way (2dF and SDSS) will be crucial in determining the viability
of these models.
Light neutrinos not being a necessary addition to the matter
composition of the Universe to explain the known data, the studies
just mentioned provide an upper bound (already mentioned in Section 3)
on the relative amount of neutrinos with respect to CDM: in all of them
$\Omega_\nu/\Omega_m \leq 0.2-0.3$.
Let us finally return to the possibilty of relic neutrinos with a large lepton
asymmetry. Adams and Sarkar \cite{Sarkar} found that a relic neutrino species with
chemical potential $\mu_\nu = 3.4~T_\nu$ added to a standard CDM model
(flat matter dominated universe) provides a good fit to the LSS and CMBR
data. The best bounds on large $\mu_\nu$ come from BBNS (which
becomes severely non-standard in the presence of large neutrino asymmetries)
and structure formation. Relic neutrinos with large
chemical potentials $\mu_\nu > T$, would form a Fermi degenerate background
with number density
$n_\nu = (12 \zeta(3))^{-1} \left (T_\nu / T_\gamma \right )^3
[\pi^2[(\mu_\nu / T)+(\mu_\nu / T)^3] = 0.0252 [9.87 (\mu_\nu / T) + (\mu_\nu / T)^3]$,
where $(T_\nu/T_\gamma)^3$ has the standard value of (4/11) for
$(\mu_\nu/T) < 12$. For $\mu/T = 3.4$, the density of neutrinos would
be $n_\nu = 1.8~n_\gamma = 756/{\rm cm}^3$. Parenthetically let us point
out that slightly larger chemical potentials,
still alowed by LLS and CMBR data, could make neutrinos with the
mass implied by SuperKamiokande data, in the case of hierarchial masses,
$m_\nu = \sqrt{\delta m^2} \simeq 0.07~{\rm eV}$, a relevant HDM component.
For example with $(\mu_\nu/T) = 4.6$, the relic neutrino density would be
$n_\nu = 3.6~n_\gamma$ and a neutrino of mass 0.7 eV would have
$\Omega_\nu h^2 = 0.01$.\cite{Gelmini} These neutrinos would still be
relativistic at the moment of radiation-matter equality, so they would have
on structure formation almost the same effect as the massless neutrinos
studied by Adams and Sarkar. The large neutrino asymmetries necessary in
these models could possibly be generated through neutrino oscillations after the elctroweak phase transition (this needs to be studied) and certainly with the Affleck-Dyne mechanism. \cite{Casas}
Many possiblities are still open, but the quality data necessary to
discriminate among models are coming, and a confirmation of one of
them may be possible within a few to ten years. Besides getting to know
$\Omega$, $\Omega_B$, $\Omega_\Lambda$, $\Omega_m$, H, $t_o$ etc., the relevant
cosmological parameters to a few \%, standard neutrinos with no lepton
asymmetry will be seen either as HDM for $m_\nu \geq 1$ eV (may be up to
0.3 with low $\Omega_m$)\cite{Hu1} or for lighter neutrinos, the number of
neutrino species will be determined with precision similar to the NS
bounds.\cite{Hu2} The case of large neutrino lepton asymmetry has yet to
be studied.
\section{Astrophysics.}
Neutrinos may have an important role in the evolution of some types of stars,
in particular in the explosion of Type II SN. These are stars in which the
Fe-core reaches the Chandrasekar limit and, thus,
collapses into a neutron star, trapping the emitted neutrinos for several
seconds within a region called ``neutrino-sphere", and exploding the mantle of the star. Actually the neutrino spheres of different neutrinos types are slightly different leading to different neutrino average energies. Most of the binding energy of the remaining neutron star
goes into neutrinos. Thus, the explosion mechanism is sensitive to non-standard neutrino properties. Some examples are the following.
A problem in Type II SN modelling is that
the shock wave which should explode the mantle stalles before getting to do it. This problem might be solved
if the $\nu_e$ arriving at the shock from the $\nu$-sphere are actually
$\nu_\mu$ or $\nu_\tau$ which oscillated resonantly into $\nu_e$
on their way.\cite{Fuller} These $\nu_e$ arriving at the shock would have the higher energy ofthe originally emitted $\nu_\mu$-$\nu_\tau$, 24-27 MeV, instead of the lower energy of 10-12 MeV with which $\nu_e$'s are emitted, leading to a more efficient energy transfer to the shock. This would require $\Delta m^2 = 10^2$ to $10^4$eV$^2$ and mixing angles $\sin^2(2 \theta) > 10^{-7}$ approximately. However, these resonant oscillations may prevent the
r-process (rapid neutron capture) synthesis of heavy elements, whose
preferred site is behind the shock wave of an exploding Type II SN. Because the
energy of $\nu_e$ in this case would be larger than that of
$\bar\nu_e$, 14-17 MeV, the environment would become proton rich, while r-process requires a
neutron rich medium.\cite{Qian} This would happen for
$\Delta m^2 = 10$ to $10^4$eV$^2$ and mixing angles $\sin^2(2 \theta) > 10^{-6}$ approximately. The matter enhanced oscillations of
$\nu_e$ and a sterile neutrino may instead help r-process, by eliminating
the $\nu_e$ which decimate neutrons (through the reaction $n\nu_e\to pe^-$) \cite{Fetter}, for $\Delta m^2 = 1$ to $10^2$eV$^2$ and mixing angles $\sin^2(2 \theta) > 10^{-3}$ approximately. These arguments involving r-process in Type II SN are to
be aken with a grain of salt, considering that there is another possible site for it, in
binary neutron star - neutron star mergers.
As the
last point, let us mention ``pulsar kicks". Pulsars, which are neutron stars
with large magnetic fields, are not confined to the disk of our galaxy, where
they must have been born, but move in all directions with large velocities
of 500 km/sec on average. A ``pulsar kick" which may impart these velocities may result from the existence of a
matter-enhanced resonant region between the more internal $\nu_\mu, \nu_\tau$ and the more external
$\nu_e$ neutrino spheres.\cite{Kusenko} Along the direction of the magnetic
field the resonance happens at larger densities, thus more internally in the
protostar.\cite{D'Olivo} The $\nu_e$ oscillated into $\nu_\mu$ or $\nu_\tau$ within the $\nu_e$ sphere (from which they could not escape otherwise) but outside the $\nu_\mu,\nu_\tau$ sphere can freely
escape, and they carry more energy away when the conversion happens the
farthest from the surface of the star (where temperatures are higher). Thus,
the deformation of the resonance region due to the presence of a large
magnetic field, has as a consequence the emission of hotter converted $\nu_e$
neutrinos in the direction of
the magnetic field and cooler ones in the opposite direction. A 1\% asymmetry
in the total momentum carried by the emitted neutrinos would be enough to
impart to the remaining neutron star the large velocities of pulsars.\cite{Kusenko} Several versions of this mechanism have been proposed.\cite{others}
\noindent{\it Acknowledgments}
I thank the organizers of this workshop for their invitation.
This work was supported in part by the U.S. Department of Energy under Grant
DE-FG03-91ER 40662 Task C.
|
\section{Introduction}
Determining the number of real solutions to a system of polynomial equations
is a challenging problem in symbolic and numeric
computation~\cite{G-VRRT,Sturmfels_monthly} with real world applications.
Related questions include when a problem of
enumerative geometry can have all solutions real~\cite{Sottile97c} and
when may a given physical system be controlled by real output
feedback~\cite{Byrnes,RSW,SADG}.
In May 1995, Boris Shapiro and Michael Shapiro
communicated to the author a remarkable conjecture connecting these three
lines of inquiry.
They conjectured a relation between topological invariants of the real and
of the complex points in an intersection of Schubert cells in a flag
manifold, if the cells are chosen according to a recipe they give.
When the intersection is
zero-dimensional, this asserts that all points are real.
Their conjecture is false---we give full description and present a
counterexample in Section 5.
However, there is considerable evidence for their conjecture if the Schubert
cells are in a Grassmann manifold.
It is this variant which is related to the lines of inquiry
above and which this paper is about.
Here is the simplest (but still very interesting and open) special case of
this conjecture:
Let $m,p>1$ be integers and let $X$ be a $p\times m$-matrix of
indeterminates.
Let $K(s)$ be the $m\times(m+p)$-matrix of polynomials in $s$
whose $i,j$th entry is
\begin{equation}\label{eq:K-matrix}
\binom{j-i}{i-1} s^{j-i}.
\end{equation}
Set
$$
\varphi_{m,p}(s;X)\ :=\ \det\zwei{K(s)}{I_p\quad X},
$$
where $I_p$ is the $p\times p$ identity matrix.
\begin{conj}[Shapiro-Shapiro]\label{conj:shapiroI}
For all integers $m,p>1$, the polynomial system
\begin{equation}\label{eq:conjectureI}
\varphi_{m,p}(1;X)\ =\ \varphi_{m,p}(2;X)\ =\ \cdots\ =\
\varphi_{m,p}(mp;X)\ =\ 0
\end{equation}
is zero-dimensional with
\begin{equation}\label{grassdeg} d_{m,p} \quad := \quad
\frac{1! \, 2! \, 3! \cdots (p\!- \!2) ! \, (p \!-\!1)! \cdot
(mp)!}{m!\, (m \! + \! 1)! \, (m \! + \, 2)!
\cdots(m \! + \! p \! - \! 1)!}
\end{equation}
solutions, and all of them are real.
\end{conj}
It is a Theorem of Schubert~\cite{Schubert_degree} that $d_{m,p}$ is a sharp
bound for the number of isolated solutions.
Conjecture~\ref{conj:shapiroI} has been verified for all $1<m\leq p$ with
$mp\leq 12$.
The case of $(m,p)=(3,4)$ $(d_{m,p}=462)$ is due to an heroic calculation
of Faug\`ere, Rouillier, and Zimmermann~\cite{FRZ} (see
Section~\ref{sec:computation} for a discussion).
\medskip
Conjecture~\ref{conj:shapiroI} is related to a question of
Fulton~\cite[\S7.2]{Fulton_introduction_intersection_new}:
``How many solutions to a problem in enumerative geometry
may be real, where that problem consists of counting figures of
some kind having a given position with respect to some given (fixed)
figures.''.
For 2-planes having a given position with respect to fixed
linear subspaces, the answer is that all may be
real~\cite{Sottile97a}.
This was also shown for the problem of 3264 plane conics tangent to five
given conics~\cite{RTV}.
More examples, including that of 3-planes in
${\mathbb C}^6$ meeting 9 given 3-planes nontrivially,
are found in~\cite{Sottile97c,Sottile97b}.
The result~\cite{FRZ} extends this to 3-planes in
${\mathbb C}^7$ meeting 12 given 4-planes nontrivially.
Only the simplest form of the conjecture of Shapiro and Shapiro
has appeared in print~\cite{HSS,RS98,Sottile97c}.
While more general forms have circulated informally, there is no
definitive source describing the
conjectures or the compelling evidence that has accumulated
(or a counterexample to the original conjecture).
The primary aim of this paper is to rectify this situation and make these
conjectures available to a wider audience.
\medskip
In Section 2, we describe a version of the conjecture
related to the pole placement problem of linear systems theory.
For this, the integers $1,2,\ldots,mp$ in the polynomial
system~(\ref{eq:conjectureI}) of Conjecture~\ref{conj:shapiroI} are
replaced by generic real numbers and all $d_{m,p}$ solutions are asserted to
be real.
We present evidence (computational and Theorems) in support of it.
Subsequent sections describe the conjecture in greater
generality---for enumerative problems arising from the Schubert calculus on
Grassmannians in Section~\ref{sec:LR} and a newer extension involving
totally
positive matrices~\cite{Ando} in Section~\ref{sec:tot_pos}.
We describe and give evidence for each extension
and show how the version of the conjecture in Section 2 implies more
general versions involving Pieri-type enumerative problems.
In Section~\ref{sec:counter}, we present a counterexample to
their original conjecture and discuss further questions.
A remark on the form of these conjectures is warranted.
Conjecture~\ref{conj:shapiroI} gives an infinite list of specific polynomial
systems, and conjectures that each has only real solutions.
The full conjectures are richer.
For each collection of {\it Schubert data}, Shapiro and Shapiro give a
continuous family of polynomial systems and conjecture that each of the
resulting systems of polynomials has only real solutions.
Conjecture~\ref{conj:shapiroI} concerns one specific polynomial system in
each family, for an infinite subset of Schubert data.
Results here were aided or are due to computations.
Further documentation including Maple V.5 and Singular
1.2.1~\cite{SINGULAR} scripts used are available on the web
page~\cite{Sottile_shap-www}.
\tableofcontents
\section{Linear equations in Pl\"ucker coordinates}\label{sec:hypersurface}
\subsection{Some enumerative geometry}
Consider the following problem in enumerative geometry:
How many $p$-planes meet $mp$ general $m$-planes in ${\mathbb C}^{m+p}$
nontrivially?
The set of $p$-planes in ${\mathbb C}^{m+p}$, $\mbox{\it Grass}(p,m+p)$, is
called the {\it Grassmannian of $p$-planes in ${\mathbb C}^{m+p}$}.
This complex manifold of dimension $mp$ is an algebraic subvariety
of the projective space ${\mathbb P}^{\binom{m+p}{p}-1} $.
To see this, represent a $p$-plane $X$ in ${\mathbb C}^{m+p}$ as the
row space of a $p\times(m+p)$-matrix, also written $X$.
The maximal minors of $X$ are its {\it Pl\"ucker coordinates} and
determine a point in ${\mathbb P}^{\binom{m+p}{p}-1}$.
This gives the Pl\"ucker embedding of $\mbox{\it Grass}(p,m+p)$.
If $X$ is generic, then its first $p$ columns are linearly
independent, so we may assume they form a $p\times p$-identity matrix.
The remaining $mp$ entries determine $X$ uniquely and give local coordinates
for $\mbox{\it Grass}(p,m+p)$, showing it has dimension $mp$.
Consider a $m$-plane $K$ to be the row space of a $m\times(m+p)$-matrix,
also written $K$.
Then $K\cap X$ is nontrivial if and only if
$$
\det\zwei{K}{X}\ =\ 0.
$$
Laplace expansion along $X$
gives a linear equation in the Pl\"ucker coordinates of $X$.
If $K_1,\ldots,K_{mp}$ are $m$-planes in general position, then the
conditions that $X$ meet each of the $K_i$ nontrivially are $mp$
linear equations in the Pl\"ucker coordinates of $X$,
and these are independent by Kleiman's Transversality
Theorem~\cite{Kleiman}.
Hence there are finitely many
$p$-planes $X$ which meet each $K_i$ nontrivially
and this number is the
degree of $\mbox{\it Grass}(p,m+p)$ in ${\mathbb P}^{\binom{m+p}{p}-1}$,
which Schubert~\cite{Schubert_degree} determined to be
$d_{m,p}$.
\subsection{The conjecture of Shapiro and Shapiro}\label{sec:shapiroII}
Shapiro and Shapiro gave a recipe for selecting real
$m$-planes $K_1,\ldots,K_{mp}$ and conjecture that when they are in
in general position, all $d_{m,p}$ $p$-planes meeting each $K_i$ are real.
The standard rational normal curve is the image of the map
$\gamma:{\mathbb R}\rightarrow {\mathbb R}^{m+p}$
given by
\begin{equation}\label{eq:afine_rat}
\gamma\ :\ s\ \longmapsto\ (1,s,s^2,\ldots,s^{m+p-1})
\end{equation}
Then the matrix $K(s)$ of the Introduction~(\ref{eq:K-matrix}) has
rows
$$
\gamma(s),\ \gamma'(s),\ \frac{\gamma''(s)}{2},\ \ldots,\
\frac{\gamma^{(m-1)}(s)}{(m-1)!},
$$
where we take derivatives with respect to the parameter $s$.
Let $X$ be a $p\times m$-matrix of
indeterminates.
Define
$$
\varphi_{m,p}(s;X)\ :=\ \det\zwei{K(s)}{I_p\ X}.
$$
\begin{conj}[Shapiro-Shapiro]\label{conj:shapiroII}
For all integers $m,p>1$ and almost all
distinct real numbers $s_1,\ldots,s_{mp}$,
the system of $mp$ equations
\begin{equation}\label{eq:shapiroII}
\varphi_{m,p}(s_1;X)\ =\ \varphi_{m,p}(s_2;X)\ =\ \cdots\ =\
\varphi_{m,p}(s_{mp};X)\ =\ 0
\end{equation}
is zero-dimensional with $d_{m,p}$ real solutions.
\end{conj}
Let $K(s)$ denote both the $m\times (m+p)$-matrix defined above and its row
space, an $m$-plane.
Conjecture~\ref{conj:shapiroII} asserts that the $m$-planes
$K(s_1),\ldots,K(s_{mp})$ are in general position, and any $p$-plane
meeting each $K(s_i)$ is real.
The systems are zero-dimensional~\cite{Brockett_Byrnes,EH83}
and there are generically no multiplicities.
Conjecture~\ref{conj:shapiroI} is the special case when $s_i=i$.
\begin{ex}\label{ex:22}
We establish Conjecture~\ref{conj:shapiroII} when $m=p=2$.
Then
$$
\varphi_{2,2}(s;X)\ =\
\det\left[\begin{array}{cccc}
1&s&s^2&s^3\\
0&1&2s&3s^2\\
1&0&x_{11}&x_{12}\\
0&1&x_{21}&x_{22}\end{array}\right]
$$
is
$$
s^4-2s^3x_{21}+s^2x_{22}-3s^2x_{11}+2s x_{12}+x_{11}x_{22}-x_{12}x_{21}.
$$
We show that if $s,t,u,v\in{\mathbb R}$ are distinct, then the system of
polynomial equations
\begin{equation}\label{eq:system22}
\varphi_{2,2}(s)\ =\ \varphi_{2,2}(t)\ =\ \varphi_{2,2}(u)\ =\
\varphi_{2,2}(v)\ =\ 0
\end{equation}
has all $d_{2,2}=2$ solutions real.
Our method will be to solve~(\ref{eq:system22}) by elimination.
Let $e_i$ be the $i$th elementary symmetric polynomial in $s,t,u,v$.
In the lexicographic term order with $x_{11}>x_{12}>x_{22}>x_{21}$
on the ring ${\mathbb Q}(s,t,u,v)[x_{11},x_{12},x_{22},x_{21}]$, the ideal
$\langle\varphi_{2,2}(s),\varphi_{2,2}(t),
\varphi_{2,2}(u),\varphi_{2,2}(v)\rangle$
has a Gr\"obner basis consisting of the following polynomials:
$$
2x_{21}-e_1,\ \
x_{22}-3x_{11}-e_2,\ \
2x_{12}+e_3,\ \ \mbox{ and }\ \
12x_{11}^2 + 4e_2 x_{11}+ e_1e_3 - 4e_4.
$$
Thus, for distinct $s,t,u,v$, the system~(\ref{eq:system22})
has 2 solutions and they are real if the discriminant
of the last equation,
$$
16e_2^2 - 48e_1e_3 + 192 e_4,
$$
is positive.
Expanding this discriminant in the parameters $s,t,u,v$, we obtain
$$
8\left(
(s-t)^2(u-v)^2 + (s-u)^2(t-v)^2 + (s-v)^2(t-u)^2
\right).
$$
Hence all solutions are real, establishing Conjecture~\ref{conj:shapiroII}
when $m=p=2$.
Theorem~\ref{thm:23} proves Conjecture~\ref{conj:shapiroII}
when $(m,p)=(2,3)$.
\end{ex}
\subsection{Pole placement problem}\label{sec:control}
Suppose we have a physical system (for example, a mechanical linkage) with
inputs
$u\in {\mathbb R}^m$ and outputs $y\in {\mathbb R}^p$ for which there are
internal states $x\in {\mathbb R}^n$ such that the system evolves
by the first order linear differential equation
\begin{equation}\label{linsystem}
\begin{array}{rcl}
\dot{x}&=&Ax + Bu,\\
y&=&Cx.
\end{array}
\end{equation}
(We assume $n$ is the minimal number of
internal states needed to obtain a first order equation.)
If the input is controlled by constant output feedback, $u=Xy$,
then we obtain
$$
\dot{x}\ =\ (A+BXC)x.
$$
The natural frequencies of this controlled system are the roots
$s_1,\ldots,s_n$ of
\begin{equation}\label{charpoly}
\varphi(s)\ :=\ \det(sI_n-A-BXC).
\end{equation}
The pole assignment problem asks the inverse question:
Given a system~(\ref{linsystem}) and a
polynomial $\varphi(s)$ of degree $n$, which feedback laws $X$
satisfy~(\ref{charpoly})?
A coprime factorization of the transfer function is two matrices
$N(s)$, $D(s)$ of polynomials with $\det(D(s))= \det(sI_n-A)$
and $N(s)D(s)^{-1}=C(sI_n-A)^{-1}B$.
This always exists.
A standard transformation~({\em cf.}~\cite[\S 2]{Byrnes})
shows that, up to a sign of $\pm 1$,
\begin{equation}\label{schubert_form}
\varphi(s)\ =\ \det\left[
\begin{array}{cc} N(s) & D(s)\\ I_p & X \end{array}\right],
\end{equation}
If we set $K(s)\ :=\ [N(s) \ D(s)]$, write $K(s)$ for the $m$-dimensional
row space of this matrix, and let $X$ be the
$p$-plane $[I_p \ X ]$,
then~(\ref{schubert_form}) is equivalent to
\begin{equation}\label{eq:control}
X\cap K(s_i)\ \neq \ \{0\}\quad\mbox{for}\quad i\ =\ 1,\ldots,n,
\end{equation}
where $s_1,\ldots,s_n$ are the roots of $\varphi(s)$.
If the $m$-planes $K(s_1),\ldots,K(s_n)$ are in general position, then
$mp\geq n$ is necessary for there to be any feedback laws.
These $m$-planes are not {\it a priori} in general position.
To see this,
let $K:{\mathbb P}^1\rightarrow\mbox{\it Grass}(m,m+p)$ be the
extension of the map given by
$s\mapsto K(s)$.
Then $K$ is a parameterized rational curve of degree $n$ in
$\mbox{\it Grass}(m,m+p)$.
The space of all such curves $K$ with $n$ distinguished points
$\{K(s_1),\ldots,K(s_n)\}$ has dimension~\cite{Stromme}
$$
mp + n(m+p) + n.
$$
The space of all $n$-tuples of $m$-planes has dimension $nmp$.
Therefore when
$$
n \ >\ mp/(mp-m-p-1),
$$
such $n$-tuples constitute a proper subvariety of
all $n$-tuples of $m$-planes.
However, the General Position Lemma~\cite{Byrnes} (see
also~\cite{EH83}) states that there is a
Zariski open subset of the data $A,B,C,\varphi$ such that the $m$-planes
$K(s_1),\ldots,K(s_n)$ are in general position in that the set of
$X$ satisfying (\ref{eq:control}) has dimension $mp-n$.
Since all rational curves
$K: {\mathbb P}^1 \rightarrow \mbox{\it Grass}(p,m+p)$ of degree $n$
with $K(\infty)=[0\ I_p]$
arise in this way~\cite{Hermann_Martin}, the polynomial
systems of
Conjecture~\ref{conj:shapiroII} are instances of the pole placement
problem.
Interestingly, these very systems figure prominently in a proof of the General
Position Lemma~\cite{Byrnes_80}.
\smallskip
An important question is whether a given real system may be controlled by
{\em real} feedback~\cite{Byrnes_real,RSW,RS98,SADG,Willems_Hesselink}:
If all roots of $\varphi(s)$ are real, are there any real
feedback laws $X$ satisfying~(\ref{schubert_form})?
Few specific examples
have been
computed~\cite{BS_homotopy,MWA,RS98,Willems_Hesselink}.
In~\cite{RS98} an attempt was made to gauge how likely it is for a real
system to be controllable by real feedback and how many of the feedback laws
are real---in the case of $(m,p)=(2,4)$ so that $d_{m,p}=14$.
In all, 600 different curves $K(s)$ were generated, and each of these were
combined with 25 polynomials $\varphi(s)$ having 8 real roots.
Only 7 of the resulting 15,000 systems had all feedback laws real.
This is in striking contrast to the systems given in
Conjecture~\ref{conj:shapiroII}, where all the feedback laws are conjectured
to be real.
\subsection{Computational evidence}\label{sec:computation}
Consider~(\ref{schubert_form}) as
a map $\mbox{\it Grass}(p,m+p)\rightarrow{\mathbb P}^{mp}$ in local
coordinates which associates a $p$-plane $X$ to a polynomial $\varphi$
(modulo scalars) of degree at most $mp$.
When $K(s)$ is the curve $K_{m,p}(s)$ of Conjecture~\ref{conj:shapiroII},
the inverse image of the polynomial 1 is the single real point $[0\ I_p]$.
Rosenthal suggested that the fibre
over a nearby polynomial may consist of $d_{m,p}$ real points.
Inspired by this, Rosenthal and Sottile~\cite{RS98}
tested and verified several thousand instances of
Conjecture~\ref{conj:shapiroII} when $(m,p)=(2,4)$.
Each was a specific choice of $m,p$ and $mp$ distinct real numbers
$s_1,\ldots,s_{mp}$ for which we showed all solutions
to~(\ref{eq:shapiroII}) are real.
Any verified instance implies
that all nearby instances in the space of parameters $s_1,\ldots,s_{mp}$
has all of its solutions real.
In light of the computations described in Section~\ref{sec:control}, we felt
this provided overwhelming evidence for the validity of
Conjecture~\ref{conj:shapiroII}.
Our method was to solve the
polynomial systems by elimination (see~\cite[\S 2]{CLO_using} for a
discussion of methods to solve systems of polynomial
equations).
We first choose distinct integral values of the parameters $s_i$ and generate
the resulting system of integral polynomial equations.
Since we are performing an exact symbolic computation, we necessarily work
with integral polynomials.
Next, we compute an eliminant, a univariate polynomial $g(x)$ with the
property that its roots are the set of $x$-coordinates of solutions to our
system.
When $g(x)$ has $d=d_{m,p}$ roots (Schubert's bound), there is a
lexicographic Gr\"obner basis satisfying
the Shape Lemma, since this system is zero-dimensional~\cite{EH83}.
It follows that the solutions are rational functions (quotients of integral
polynomials) of the roots of $g(x)$.
In some instances, the eliminant we calculated did not have $d$ roots.
For these we found a different eliminant with $d$ roots.
Lastly, we checked that these eliminants had only real roots.
Table~\ref{table:instances} gives the number of instances we know have
been checked.
By Lemma~\ref{lem:ordering}(ii), there is a bijection
between instances of $(m,p)=(a,b)$ and $(m,p)=(b,a)$.
Table~\ref{table:instances} also lists the running time to
compute a degree reverse lexicographic Gr\"obner basis for the systems of
Conjecture~\ref{conj:shapiroI}, and the size of that basis.
This used Singular-1.2.1~\cite{SINGULAR} on a K6-2-300 processor with 256M
running Linux.
The instances reported in the last 3 columns are not due
the the author.
A more complete account is found in~\cite{Sottile_shap-www}.
\begin{table}[htb]
\begin{tabular}{|c||c|c|c|c|c|c|c|}
\hline$m,p$&4,2&5,2&3,3&6,2&7,2&4,3&2,8\\\hline\hline
$d_{m,p}$&14&42&42&132&429&462&1430\\\hline
\# checked&$>12,000$&1000&550&55&2&2&1\\\hline
time (sec)&.04&1.42&1.50&78.6&8175&--&--\\\hline
size&1.4K&12.8K&18.6K&202K&4.58M&32M&--\\\hline
\end{tabular}\bigskip
\caption{Instances checked\label{table:instances}}
\end{table}
The computations of the last 2 columns stand out.
The first is the case $8,2$ (also one instance each of $7,2$ and $4,3$)
computed by Jan Verschelde~\cite{Verschelde} using his
implementation of the SAGBI homotopy algorithm in~\cite{HSS}.
Since the polynomial system of Conjecture~\ref{conj:shapiroII} was
ill-conditioned, he instead used the equivalent system of
Conjecture~\ref{conj:shapiroII}$'$ (in Section~\ref{sec:simpler} below),
where the $P_i(s)$ were the Chebyshev polynomials.
These numerical calculations give approximate
solutions whose condition numbers determine a neighborhood
containing a solution.
The solutions of this real system are stable under complex
conjugation, so it sufficed to check that each neighbourhood and its complex
conjugate were disjoint from all other neighborhoods.
This computation took approximately 25 hours on a 166MHz Pentium II
processor with 64M running Linux.
These algorithms are `embarrassingly parallelizable', and
in principle they can be used to check far larger polynomial systems.
The second is the case of $(m,p)=(3,4)$ of
Conjecture~\ref{conj:shapiroI}
(also all smaller cases with $m\leq p$), computed by Faug\`ere, Rouillier, and
Zimmermann~\cite{FRZ}.
They first used FGB~\cite{Faugere_FGB} to calculate a degree reverse
lexicographic Gr\"obner basis for the system~(\ref{eq:conjectureI}) for
$(m,p)=(3,4)$ with $s_i=i$.
This yielded a Gr\"obner basis of size 32M.
They then computed a rational univariate
representation~\cite{Rouillier_RUR} (a sophisticated substitute for an
eliminant) in two ways.
Once using a multi-modular implementation of the
FGLM~\cite{FGLM} algorithm and a second time using RS,
an improvement of the RealSolving
software~\cite{Rouillier_RS} under development.
The eliminant had degree 462 and size 3M, thus its general
coefficient had 2,000 digits.
Using an early implementation of Uspensky's algorithm, they
verified that all
of its zeroes were real, proving
Conjecture~\ref{conj:shapiroI} for $(m,p)=(3,4)$.
In the course of this calculation, they found it necessary
to rewrite their software.
\subsection{Equivalent systems}\label{sec:simpler}
The extension of the map~(\ref{eq:afine_rat}) to ${\mathbb P}^1$
$$
\gamma\ :\
[t,s]\ \longmapsto\
[t^{m+p-1},\,st^{m+p-2},\,\ldots,\,s^{m+p-2}t,\,s^{m+p-1}]
$$
is a parameterization of the standard real rational normal curve in
${\mathbb P}^{m+p-1}$
and $K(s)$ is the $m$-plane osculating this
curve at the point $\gamma[1,s]$.
In general, a parameterized real rational normal curve is a map
$\gamma:{\mathbb P}^1\rightarrow{\mathbb P}^{m+p-1}$ of the form
$$
[t,s]\ \longmapsto\ [P_1(s,t),\,P_2(s,t),\,\ldots,\,P_{m+p}(s,t)]
$$
where $P_1(s,t),\ldots,P_{m+p}(s,t)$ form a basis for the space of
real homogeneous polynomials in $s,t$ of degree $m+p-1$.
All parameterized real rational normal curves are conjugate by a real
projective transformation of ${\mathbb P}^{m+p-1}$.
We deduce
\medskip
\noindent{\bf Conjecture~\ref{conj:shapiroII} (Geometric form)}
{\it
For all integers $m,p>1$ and almost all choices of
$mp$ $m$-planes $K_1,\ldots,K_{mp}$ osculating
a real rational normal curve at distinct real points, there are exactly
$d_{m,p}$ $p$-planes $X$ satisfying
$$
X\cap K_i\ \neq\ \{0\}\quad\mbox{ for }\quad i=1,\ldots,mp
$$ and all of these $p$-planes $X$ are real.
}\medskip
Thus Conjecture~\ref{conj:shapiroII} is equivalent to
a conjecture concerning a much richer class of polynomial systems.
\medskip
\noindent{\bf Conjecture~\ref{conj:shapiroII}$'$. }
{\it
Suppose $m,p>1$ are integers
and $P_1(s),\ldots,P_{m+p}(s)$ are a basis of the space of real polynomials
of degree at most $m+p-1$.
Let $K(s)$ be the $m\times (m+p)$ matrix of polynomials whose $i,j$th entry
is $P^{(i-1)}_j(s)$.
Set
$$
\varphi(s;X)\ :=\ \det\zwei{K(s)}{I_p \ X}.
$$
Then, for almost all choices of distinct real numbers $s_1,\ldots,s_{mp}$,
the system
$$
\varphi(s_1;X)\ =\ \varphi(s_2;X)\ =\ \cdots\ =\ \varphi(s_{mp};X)\ =\ 0
$$
has exactly $d_{m,p}$ solutions, and all of them are real.
}\medskip
The polynomial matrix $K(s)$ of Conjecture~\ref{conj:shapiroII}$'$ differs
from that of Conjecture~\ref{conj:shapiroII} by right multiplication by an
invertible $(m+p)\times(m+p)$-matrix.
Thus the resulting polynomial systems differ primarily by choice of local
coordinates for the Grassmannian.
In linear systems theory, two physical systems are output
feedback-equivalent if their matrices of coprime factors $[N(s)\ D(s)]$
differ in this manner~\cite{RRH}.
We give an equivalent conjecture concerning a simpler system of polynomials
with 2 fewer equations and unknowns.
We may reparameterize the curve
$K(s)$ of Conjecture~\ref{conj:shapiroII}
and assume $s_{mp-1}=0$ and $s_{mp}=\infty$.
Observe that $K(0)=[I_p\ 0]$ and $K(\infty)=[0\ I_p]$.
The collection of all $p$-planes $X$ satisfying
\begin{equation}\label{eq:small}
X\cap[I_p\ 0]\neq\{0\} \quad\mbox{ and }\quad
X\cap[0\ I_p]\neq\{0\}
\end{equation}
is an irreducible rational variety of dimension $mp-2$.
Let ${\mathcal X}$ be the set of all $p\times(m+p)$-matrices $X$ whose
entries $x_{i,j}$ satisfy:
$$
\begin{array}{rcl}
x_{i,j}=1&\mbox{ if }&\mbox j=i<p\mbox{ \ or \ } (i,j)=(p,p+1)\\
x_{i,j}\ =\ 0&\mbox{ if }&
\left\{\begin{array}{ccl}
i=1&\mbox{ and }& j\geq m\\
1<i<p&\mbox{ and }& j<i\mbox{ or } j> i+m\\
i=p&\mbox{ and }& j\leq p\end{array}\right.
\end{array}
$$
The remaining $mp-2$ entries are unconstrained and give
coordinates for ${\mathcal X}$.
The row space of a matrix $X$ is a $p$-plane $X$
satisfying~(\ref{eq:small})
and almost all such $p$-planes arise in this fashion.
Thus ${\mathcal X}$ parameterizes a dense subset of the
subvariety of $p$-planes $X$ satisfying~(\ref{eq:small}).
For example, if $(m,p)=(4,3)$, then ${\mathcal X}$ is the set of all
matrices of the form:
$$
\left[\begin{array}{ccccccc}
1 &x_{12}&x_{13}&x_{14}& 0 & 0 & 0 \\
0 & 1 &x_{23}&x_{24}&x_{25}&x_{26}& 0 \\
0 & 0 & 0 & 1 &x_{35}&x_{36}&x_{37}
\end{array}\right].
$$
Since the $1,m$th entry of a matrix in ${\mathcal X}$ vanishes,
$$
\det\zwei{K(s)}{X}
$$
factors as $s\cdot \psi(s;X)$.
\medskip
\noindent{\bf Conjecture~\ref{conj:shapiroII}$''$. }
{\it
Let $m,p>1$ be integers.
Then, for almost all choices of non-zero real numbers $s_1,\ldots,s_{mp-2}$,
the system of equations
\begin{equation}\label{eq:simpler}
\psi(s_1;X)\ =\ \psi(s_2;X)\ =\ \cdots\ =\
\psi(s_{mp-2};X)\ =\ 0
\end{equation}
is zero-dimensional with $d_{m,p}$ solutions, and all of them are real.
}\medskip
The systems of Conjecture~\ref{conj:shapiroII} and the variations given here
are deficient:
They have fewer solutions than standard combinatorial bounds.
For example, the system~(\ref{eq:simpler}) consists of $mp-2$ equations of
degree $p$, thus its B\'ezout number is $p^{mp-2}$.
A better combinatorial bound is
the normalized volume of the Newton polytope ${\mathcal A}_{m,p}$ of the
polynomial $\psi$~\cite{Kouchnirenko}.
Table~\ref{table:bounds} compares these combinatorial bounds with $d_{m,p}$,
for some values of $m,p$.
The volumes of ${\mathcal A}_{m,p}$ were computed using
PHC~\cite{PHC}, a software package for performing general polyhedral
homotopy continuation.
Note the striking difference between the equivalent systems $m,p$ and $p,m$.
\begin{table}[htb]
\begin{tabular}{|r||c|c|c|c|c|c|c|c|c|c|c|c|}
\hline
$m,p$:& 2,2& 3,2& 4,2& 5,2& 6,2&7,2&8,2 & 2,3&3,3 & 4,3&2,4&3,4\\\hline\hline
$d_{m,p}$:
&2&5&14&42&132&429&1430&5&42&462&14&462\\\hline
vol$\,{\mathcal A}_{m,p}$:
&2&5&18&67&248&919&3426&5&130&3004&42&7156\\\hline
$p^{mp-2}$: \rule{0pt}{12pt}
&4&16&64&256&1024&4096&16384&81&2187&59,049&4096&1048576\\\hline
\end{tabular}\bigskip
\caption{Combinatorial bounds vs. $d_{m,p}$\label{table:bounds}}
\end{table}
\subsection{Conjecture~\ref{conj:shapiroII} for $m=2, p=3$}
\begin{thm}\label{thm:23}
Conjecture~\ref{conj:shapiroII} holds for $(m,p)=(2,3)$.
\end{thm}
L.~Gonzalez-Vega has also obtained this using resultants and Sturm-Habicht
sequences.
\medskip
\noindent{\bf Proof. }
We will prove the equivalent Conjecture~\ref{conj:shapiroII}$''$.
Let $X:=\{x_{12},x_{23},x_{24},x_{35}\}$ be indeterminates.
Set
$$
\psi(s;X)\ =\ \ \det \left[\begin{array}{ccccc}
1&s&s^2&s^3&s^4\\
0&1&2s&3s^2&4s^3\\
1&x_{12}&0&0&0\\
0&1&x_{23}&x_{24}&0\\
0&0&0&1&x_{35}\end{array}\right]
$$
We solve the system of polynomials
\begin{equation}\label{eq:param-sys}
\psi(s;X)\ =\ \psi(t;X)\ =\
\psi(u;X)\ =\ \psi(v;X)\ =\ 0
\end{equation}
by elimination.
The ideal $\langle \psi(s),\psi(t),\psi(u),\psi(v)\rangle$
in the ring ${\mathbb Q}(s,t,u,v)[x_{12},x_{23},x_{24},x_{35}]$
has degree $5=d_{2,3}$ and the lexicographic Gr\"obner basis with
$x_{12}<x_{23}<x_{24}<x_{35}$ contains the following univariate polynomial
$g$, which is the universal eliminant for this family of systems:
$$
x_{35}^5 - 4e_1 x_{35}^4 + (4e_1^2+6e_2) x_{35}^3
- (12e_1e_2+4e_3) x_{35}^2 + (9e_2^2+8e_1e_3-4e_4) x_{35}
-(12e_2e_3-8e_1e_4)
$$
Here $e_i$ is the $i$th the elementary symmetric polynomial in $s,t,u,v$.
We show that $g$ has 5 distinct real roots for every choice of distinct
parameters $s,t,u,v$.
The discriminant $\Delta$ of $g$ has degree 20
in the variables $s,t,u,v$ and 711 terms:
$$
\begin{array}{l}
9e_3^4e_2^2e_1^4 - 54e_3^4e_2^3e_1^2 + 81e_3^4e_2^4 - 32e_3^5e_1^5
+ 204e_3^5e_2e_1^3 - 324e_3^5e_2^2e_1 - 108e_3^6e_1^2 + 324e_3^6e_2
\\\ \rule{0pt}{13pt}
+ 81e_4^2e_2^4e_1^4 - 486e_4^2e_2^5e_1^2 + 729e_4^2e_2^6
- 54e_4e_3^2e_2^3e_1^4 + 324e_4e_3^2e_2^4e_1^2 - 486e_4e_3^2e_2^5
\\\ \rule{0pt}{13pt}
+ 204e_4e_3^3e_2e_1^5 - 1296e_4e_3^3e_2^2e_1^3
+ 2052e_4e_3^3e_2^3e_1 - 8e_4e_3^4e_1^4 + 738e_4e_3^4e_2e_1^2
- 2106e_4e_3^4e_2^2
\\\ \rule{0pt}{13pt}
- 108e_4e_3^5e_1
- 324e_4^2e_3e_2^2e_1^5 + 2052e_4^2e_3e_2^3e_1^3 - 3240e_4^2e_3e_2^4e_1
- 108e_4^2e_3^2e_1^6 + 738e_4^2e_3^2e_2e_1^4
\\\ \rule{0pt}{13pt}
- 2592e_4^2e_3^2e_2^2e_1^2 + 3834e_4^2e_3^2e_2^3 - 368e_4^2e_3^3e_1^3
+ 1800e_4^2e_3^3e_2e_1 - 27e_4^2e_3^4 + 324e_4^3e_2e_1^6
\\ \ \rule{0pt}{13pt}
- 2106e_4^3e_2^2e_1^4 + 3834e_4^3e_2^3e_1^2
- 972e_4^3e_2^4- 108e_4^3e_3e_1^5 + 1800e_4^3e_3e_2e_1^3
- 5544e_4^3e_3e_2^2e_1
\\\ \rule{0pt}{13pt}
- 634e_4^3e_3^2e_1^2 + 984e_4^3e_3^2e_2
- 27e_4^4e_1^4 + 984e_4^4e_2e_1^2 + 432e_4^4e_2^2
- 352e_4^4e_3e_1 - 64e_4^5.
\end{array}
$$
This vanishes when $g$ has a double root.
Thus the number of real roots of $g$ is constant on each connected
component (in ${\mathbb R}^4$) of the locus $\Delta\neq 0$.
We show there is only one connected component, and so the number of real
roots of $g$ (and thus the original system)
does not depend upon the choice of real parameters.
Since the roots of $g$ evaluated at $(s,t,u,v)=(1,2,3,4)$
are
$$
8, \ 8 \pm \sqrt{19}, \ 8 \pm \sqrt{11}
$$
it follows that there are always five real roots of $g$,
and thus the system~(\ref{eq:param-sys})
has $d_{2,3}=5$ real solutions whenever $s,t,u,v$ are real and
distinct.
We complete the proof.
For $w\in{\mathbb Z}_{\geq0}^{10}$, consider the polynomial:
\begin{equation}\label{eq:choice}
s^{w_1}t^{w_2}u^{w_3}v^{w_4}(s-t)^{w_5}(s-u)^{w_6}(s-v)^{w_7}
(t-u)^{w_8}(t-v)^{w_9}(u-v)^{w_{10}}
\end{equation}
Let $A_w$ be the primitive part of the symmetrization of this polynomial.
Thus $A_w$ is a sum of squares, none of which vanish on the locus where
$s,t,u,v$ are distinct.
Then $\Delta$ is
$$
\begin{array}{c}
\frac{1}{2}(7A_{2220222224}+
3A_{2222402204}+
6A_{4222022222}+
7A_{4220222222}+
2A_{4420022222}\\
\quad +2A_{2222440022}
+2A_{0222443022}
+A_{4420202222}
+2A_{4222420022}
+A_{4220022422}\\
\qquad\hspace{.02in}
+A_{0222442202}
+A_{2202024422}
+6A_{2222420024}
+10A_{4220022242}
+3A_{2222222222}).
\end{array}
$$
Note that the term $7A_{4220222222}$ does not vanish when a single
parameter is zero.
Similarly, the term $3A_{2222402204}$ does not vanish when $s=u$ and $t=v$
(but $u\neq t$).
Thus the locus where $\Delta=0$ has dimension 1 and so its complement is
connected.
\QED
We have a Maple program which performs the computations described and runs
in $\sim$15 seconds on a K6-2-300 processor.
A {\it positive semidefinite} polynomial is a real polynomial that takes only
nonnegative values.
In the proof we showed $\Delta$ is positive semidefinite by exhibiting it
as a sum of squares.
Not all positive semidefinite polynomials are sums of squares of
polynomials.
There exist positive semidefinite polynomials of degree $l$ in $k$
variables which are not sums of squares of polynomials if $\min(k,l)>2$ and
$(k,l)\neq(3,4)$~\cite{Hilbert_squares}.
For $\Delta$, $(k,l)=(4,20)$.
The form of the squares we used~(\ref{eq:choice}) for the discriminant
$\Delta$, while motivated by the observation that no two parameters
($0,s,t,u,v,\infty$) should coincide, is justified by the observation that
any real zero of $\Delta$ must also be a zero of all the squares, if
$\Delta$ is a sum of squares.
(See~\cite{CLR_symmetry} for other applications of this idea.)
Each of the polynomials $A_w$ is a sum of squares, the number given by the
orbit of the symmetric group on its index $w$.
Since 6 have trivial stabilizer, 7 are stabilized by a transposition, one by
the dihedral group $D_8$, and one is invariant, there are
$6\cdot 24 + 7\cdot 12 + 3 + 1 = 232$ squares in all.
This is not the best possible.
Choi, Lam, and Reznick~\cite{CLR} show, for degree $l$ homogeneous
polynomials in $k$ variables that are a sum of squares of polynomials, at
most
$$
\Lambda(k,l)\ :=\ \left\lfloor \frac{1}{2}
\left( \sqrt{1+8\binom{k+l-1}{l}} \ -\ 1\right) \right\rfloor
$$
squares are needed.
Note that $\Lambda(4,20)=59$.
\section{Schubert conditions on a Grassmannian}\label{sec:LR}
\subsection{The Schubert calculus on
$\mbox{\it Grass}(p,m+p)$}\label{ssec:LR}
The enumerative problems of Section~\ref{sec:hypersurface} are special cases
of more general problems given by Schubert conditions
on $\mbox{\it Grass}(p,m+p)$.
A {\it Schubert condition} on $\mbox{\it Grass}(p,m+p)$ is an increasing
sequence of integers
$$
\alpha\ :\ 1\leq \alpha_1<\alpha_2<\cdots<\alpha_p\leq m+p.
$$
Let $\binom{[m+p]}{p}$ be the set of all such sequences.
A {\it Schubert variety} $\Omega_\alpha{F\!_{\DOT}}$ is given by a Schubert
condition $\alpha$ and a complete flag
${K_{\DOT}}$ in ${\mathbb C}^{m+p}$, a sequence of subspaces
$$
{K_{\DOT}}\ :\ K_1\subset K_2\subset \cdots \subset K_{m+p}={\mathbb C}^{m+p}
$$
where $\dim K_i=i$.
Then the Schubert variety $\Omega_\alpha{F\!_{\DOT}}$ is the set of all
$p$-planes $X$ satisfying
\begin{equation}\label{eq:schubert_condition}
\dim X\cap K_{m+p+1-\alpha_i}\ \geq \ p+1-i
\end{equation}
for each $i=1,2,\ldots,p$.
This irreducible subvariety of
$\mbox{\it Grass}(p,m+p)$ has
codimension $|\alpha|:=\sum_i (\alpha_i-i)$.
A sequence
$\alpha^\DOT=\alpha^1,\ldots,\alpha^n$ with $\alpha^j\in\binom{[m+p]}{p}$
with $\sum_j|\alpha^j|=mp$
is {\it Schubert data} for $\mbox{\it Grass}(p,m+p)$.
Given Schubert data $\alpha^\DOT$
and flags $K^1_\DOT,\ldots,K^n_\DOT$ in general position,
there are finitely many (complex)
$p$-planes $X$ which lie in the intersection of the Schubert
varieties $\Omega_{\alpha^j}K^j_\DOT$ for $j=1,\ldots,n$.
The classical Schubert calculus~\cite{KlLa} gives
the following recipe for computing this number $d=d(m,p;\alpha^\DOT)$.
Let $h_1,\ldots,h_m$ be indeterminates with $degree(h_i) = i$.
For each integer sequence $\beta_1<\beta_2<\cdots<\beta_r$
define the following polynomial:
\begin{equation}\label{schur}
S_\beta\ :=\ \det ( h_{\beta_i-j})_{1\leq i,j\leq r}.
\end{equation}
Here $h_0 := 1$ and $h_i:=0$ if $i<0$ or $i>m$.
Let ${\mathcal I}$ be the ideal in
${\bf Q}[h_1,\ldots,h_m]$ generated by those
$S_\beta$ with $r=p+1$, $1<\beta_1$, and $\beta_{p+1}\leq m+p$.
The quotient ring
${\mathcal A}_{m,p}:= {\bf Q}[h_1,\ldots,h_m]/{\mathcal I}$
is isomorphic to the cohomology ring of $\mbox{\it Grass}(p,m+p)$.
It is Artinian with one-dimensional socle in degree $mp$.
In the socle we have the relation
$$
d \cdot (h_m)^p \, - \,
S_{\alpha^1} S_{\alpha^2} \cdots S_{\alpha^n} \quad \in \quad {\mathcal I}.
$$
We can compute the number $d$ by normal form reduction modulo
any Gr\"obner basis for ${\mathcal I}$.
If $\gamma$ is a rational normal curve, then the flag of subspaces
osculating $\gamma$ at a point is the {\it osculating flag} to $\gamma$ at
that point.
\begin{conj}\label{conj:shapiro-grass}
Let $m,p>1$ and $\alpha^\DOT$ be Schubert data for
$\mbox{\it Grass}(p,m+p)$.
For almost all choices of
flags $K^1_\DOT,\ldots,K^n_\DOT$ osculating a fixed rational
normal curve at real points, there
are exactly $d(m,p,\alpha^\DOT)$ $p$-planes $X$ in the intersection
of Schubert varieties
$$
\Omega_{\alpha^1}K^1_\DOT \cap \Omega_{\alpha^2}K^2_\DOT\cap
\cdots\cap \Omega_{\alpha^n}K^n_\DOT,
$$
and each of these $p$-planes is real.
\end{conj}
As with Conjecture~\ref{conj:shapiroII}, the intersection is
zero-dimensional if the points of osculation are distinct~\cite{EH83},
and there are no multiplicities for the
important class of Pieri Schubert data, (described below) which includes the
case of Conjecture~\ref{conj:shapiroII}.
If $\alpha_i=1+\alpha_{i-1}$, then
condition~(\ref{eq:schubert_condition}) for $i-1$
implies~(\ref{eq:schubert_condition}) for $i$.
Thus only those conditions~(\ref{eq:schubert_condition})
with $\alpha_i-\alpha_{i-1}>1$ (or $\alpha_1>1$) are {\it essential},
and so only the subspaces $K_{m+p+1-\alpha_i}$ corresponding to essential
conditions need be specified in a flag.
If $\alpha:=(1,2,\ldots,p-1,p+1)$, then only the last condition is
essential, thus the Schubert variety $\Omega_\alpha {K_{\DOT}}$ consists of
those $X$ with $\dim X\cap K_m\geq 1$.
This shows Conjecture~\ref{conj:shapiroII} is a special case of
Conjecture~\ref{conj:shapiro-grass}.
\subsection{Systems of polynomials}\label{sec:loc-coords}
A complete flag ${K_{\DOT}}$ is represented by a nonsingular matrix
also written ${K_{\DOT}}$:
The $i$-plane $K_i$ is the row space of $K_i$, the first $i$ rows of
${K_{\DOT}}$.
The condition that $\dim X\cap K_{m+p+1-\alpha_i}\geq p+1-i$ is given
by
$$
\mbox{$(m+p+1+i-\alpha_i)$-minors of }\
\zwei{K_{m+p+1-\alpha_i}}{X}\ =\ 0.
$$
The flag ${K_{\DOT}}(s)$ osculating the rational normal curve
$\gamma$ with the parameterization~(\ref{eq:afine_rat}) at $\gamma(s)$ is
represented by the $(m+p)\times(m+p)$-matrix whose $i,j$th entry is
$\binom{j-i}{i-1}s^{j-i}$.
\medskip
\noindent{\bf Conjecture~\ref{conj:shapiro-grass}$'$. }
{\it
Let $m,p>1$ and $\alpha^\DOT$ be Schubert data for
$\mbox{\it Grass}(p,m+p)$.
For almost all $n$-tuples of distinct real numbers $s_1,\ldots,s_n$,
the system of polynomials
$$
(m+p+1+i-\alpha_i^j)\mbox{-minors of }\
\zwei{K_{m+p+1-\alpha_i^j}(s_j)}{I_p\ \ X}\ =\ 0
$$
for $i=1,\ldots,p$ and $j=1,\ldots,n$ has
$d(m,p,\alpha^\DOT)$ solutions, and each is real.
}\medskip
For any Schubert conditions
$\alpha,\beta$ with $\alpha_i+\beta_{p+1-i}\leq m+p$
for $i=1,\ldots,p$, let ${\mathcal X}_{\alpha,\beta}$ be the collection of
all $p\times(m+p)$-matrices $X$ whose entries $x_{ij}$ satisfy
\begin{equation}\label{eq:loc_coords}
\begin{array}{rcl}
x_{i,\alpha_i}\ =\ 1&\quad&\mbox{for }i=1,\ldots,p\\
x_{i,j}\ =\ 0&&\mbox{if }j<\alpha_i\ \mbox{ or }\ j>m+p+1-\beta_{p+1-i}
\end{array}
\end{equation}
If $X\in{\mathcal X}_{\alpha,\beta}$, then the row space of $X$ is a
$p$-plane in the intersection
$\Omega_\alpha{K_{\DOT}}(\infty)\cap \Omega_\beta{K_{\DOT}}(0)$.
In this way, ${\mathcal X}_{\alpha,\beta}$ parameterizes a Zariski open
subset of the set of all such $p$-planes.
This parameterization can be used to obtain a system of equations simpler
than, but equivalent to, the system of
Conjecture~\ref{conj:shapiro-grass}$'$.
The map ${\mathcal X}_{\alpha,\beta}\rightarrow\mbox{Grass}(p,m+p)$ is not
injective.
For example, ${\mathcal X}_{123,134}$ consists of all $3\times 7$-matrices
of the form:
$$
\left[
\begin{array}{ccccccc}
1&x_{12}&x_{13}&x_{14}&0&0&0\\
0&1&x_{23}&x_{24}&x_{25}&0&0\\
0&0&1&x_{34}&x_{35}&x_{36}&x_{37}\end{array}\right]
$$
Let $r_1,r_2,r_3$ be the rows of such a matrix.
If $x_{36}=x_{37}=0$, then for each $a\in{\mathbb C}$, the matrix with rows
$r_1,r_2+ar_3,r_3$ is in ${\mathcal X}_{123,134}$, and these all have the
same row space.
Similarly, if $x_{25}=0$, then the same is true of the matrices with rows
$r_1+ar_2,r_2,r_3$.
Let ${\mathcal X}^\circ_{\alpha,\beta}\subset{\mathcal X}_{\alpha,\beta}$
be the set of those matrices whose entries further satisfy
\begin{equation}\label{eq:non-zero}
\mbox{
\begin{minipage}{4.5in}
For each $i=2,\ldots,p$, at least one $x_{ij}\neq 0$, for $j$ satisfying
$$
\beta_{p+1-i}\leq m+p+1-j < \beta_{p+2-i}.
$$
\end{minipage}
}
\end{equation}
The map
${\mathcal X}^\circ_{\alpha,\beta}\rightarrow\mbox{Grass}(p,m+p)$ is
injective.
For ${\mathcal H}_{123,134}$ this condition is that $x_{25}\neq 0$ and
$(x_{36},x_{37})\neq(0,0)$.
\subsection{Pieri Schubert conditions}
If $\alpha\in\binom{[m+p]}{p}$ has $\alpha_{p-1}=p-1$ and $\alpha_p=p+a$,
then the Schubert variety $\Omega_\alpha{K_{\DOT}}$ is
$$
\{X \mid X\cap K_{m+1-a}\neq \{0\}\}.
$$
We call such a Schubert condition a {\it Pieri condition} and
denote it by $J_a$.
{\it Pieri Schubert data} are Schubert data $\alpha^1,\ldots,\alpha^n$
were at most 2 of the conditions $\alpha^i$ are {\it not} Pieri
conditions.
These include the Schubert data of Conjecture~\ref{conj:shapiroII}.
\begin{prop}[Theorem 9.1 in~\cite{EH83}]\label{no-mult}
If $\alpha^\DOT$ are Pieri Schubert data and the flags
$K_\DOT^1,\ldots,K_\DOT^n$ osculate a rational normal curve at general
points, then the intersection of Schubert varieties
$$
\Omega_{\alpha^1}K^1_\DOT \cap \Omega_{\alpha^2}K^2_\DOT\cap
\cdots\cap \Omega_{\alpha^n}K^n_\DOT,
$$
is transverse.
In particular, there are no multiplicities.
\end{prop}
Here is the main theorem of this section.
\begin{thm}\label{thm:implies}
Let $a,b>1$ and suppose that Conjecture~\ref{conj:shapiroII} holds for this
$(m,p)=(a,b)$.
Then Conjecture~\ref{conj:shapiro-grass} holds for any Pieri Schubert
data for
$\mbox{\it Grass}(p,m+p)$ with $(p,m)\leq (a,b)$ or $(p,m)\leq (b,a)$,
in each coordinate.
\end{thm}
\begin{rem}\label{rem:stronger}
If the conclusion of Proposition~\ref{no-mult} held for all Schubert data,
then the proof we give of Theorem~\ref{thm:implies} would imply its
conclusion for all Schubert data as well.
\end{rem}
Pieri conditions are special because of Pieri's formula.
For $\alpha,\beta\in\binom{[m+p]}{p}$ and $a>0$, we write
$\alpha<_a\beta$ if $|\alpha|+a=|\beta|$ and
$$
\alpha_1\leq\beta_1<\alpha_2\leq\beta_2<\cdots<\alpha_p\leq \beta_p.
$$
\begin{prop}[Pieri's Formula]\label{prop:pieri}
Let $J_a:=1<2<\cdots<p{-}1<p{+}a\in\binom{[m+p]}{p}$.
\begin{enumerate}
\item[{\rm (i)}]
In the cohomology ring ring ${\mathcal A}_{m,p}$ of
$\mbox{\it Grass}(p,m+p)$, $S_{J_a}=h_a$ and
$$
S_\alpha \cdot S_{J_a}\ =\ \sum_{\alpha<_a\beta} S_\beta.
$$
\item[{\rm (ii)}]
If ${K_{\DOT}}(s)$ and ${K_{\DOT}}(t)$ are flags osculating a rational normal curve at
points $s$ and $t$, then
$$
\lim_{s\rightarrow t}\left(
\Omega_\alpha{K_{\DOT}}(t) \cap \Omega_{J_a}{K_{\DOT}}(s)\right)\ =\
\sum_{\alpha<_a\beta} \Omega_\beta{K_{\DOT}}(t).
$$
Here, the limit is taken as cycles.
By this we mean that the sum is the fundamental cycle of the
limit of the schemes $\Omega_\alpha{K_{\DOT}}(t) \cap \Omega_{J_a}{K_{\DOT}}(s)$
as $s$ approaches $t$ along the rational normal curve.
\item[{\rm (iii)}]
Suppose $\alpha^\DOT=\alpha^1,J_a,\alpha^2,\ldots,\alpha^n$ are Schubert
data.
Then
$$
d(m,p;\alpha^\DOT)\ =\ \sum_{\alpha<_a\beta}
d(m,p;\beta,\alpha^2,\ldots,\alpha^n).
$$
\end{enumerate}
\end{prop}
Statement (i) is the usual statement of Pieri's
formula~\cite{Fulton_tableaux,Hodge_Pedoe},
Statement (ii) is Theorem 8.1 of~\cite{EH83}, and Statement (iii)
is a direct consequence of (i).
\medskip
\noindent{\bf Proof.}
We deduce Theorem~\ref{thm:implies} from Lemma~\ref{lem:ordering} below,
which shows some simple dependencies between
Conjecture~\ref{conj:shapiro-grass} for different collections of Schubert
data.
Definition~(\ref{eq:schubert_condition}) implies that
$\Omega_\beta{K_{\DOT}}\subset \Omega_\alpha{K_{\DOT}}$ if and only if
$\alpha\leq \beta$ coordinatewise.
In fact, $\Omega_\beta{K_{\DOT}}\cap
\Omega_\alpha{K_{\DOT}}=\Omega_{\beta\vee\alpha}{K_{\DOT}}$, where $\beta\vee\alpha$
is the coordinatewise maximum of $\alpha$ and $\beta$.
We make some definitions needed for the statement of Lemma~\ref{lem:ordering}.
\begin{defn}\label{def:triple}
Let $m,p>1$ be integers.
\begin{enumerate}
\item
For $\alpha\in\binom{[m+p]}{p}$ define
$\alpha^\perp\in\binom{[m+p]}{m}$ to
be the increasing sequence obtained from the numbers
$\{1,2,\ldots,m+p\}\setminus\{\alpha_1,\ldots,\alpha_p\}$.
Given Schubert data $\alpha^\DOT$ for $\mbox{\it Grass}(p,m+p)$, set
$\alpha^{\DOT\perp}$ to be $(\alpha^1)^\perp,\ldots,(\alpha^n)^\perp$.
\item
Suppose $p>2$.
For $\alpha\in\binom{[m+p-1]}{p-1}$ define
$\alpha^+\in\binom{[m+p]}{p}$ to
be $1<1+\alpha_1<\cdots<1+\alpha_{p-1}$.
Given Schubert data $\alpha^\DOT$ for $\mbox{\it Grass}(p,m+p)$, set
$\alpha^{\DOT+}$ to be $(\alpha^1)^+,\ldots,(\alpha^n)^+$.
\item
Let $\preceq$ be the partial order on Pieri Schubert data where we
say that $\beta^\DOT$ covers
$\alpha^\DOT=\alpha^1,\ldots,\alpha^n$ if one of the following holds
$$
\begin{array}{cc}
\beta^\DOT\ =\ \beta,\alpha^3,\ldots,\alpha^n\ \mbox{with}\
\alpha^2=J_a\ \mbox{and}\ \alpha^1<_a\beta\\
\beta^\DOT\ =\ \alpha^1,\ldots,\alpha^{n-2},\beta\ \mbox{with}\
\alpha^{n-1}=J_a\ \mbox{and}\ \alpha^n<_a\beta.
\end{array}
$$
\end{enumerate}
\end{defn}
\begin{lem}\label{lem:ordering}
Let $m,p>1$ be integers.
\begin{enumerate}
\item[(i)]
If $\alpha^\DOT$ is Schubert data for $\mbox{\it Grass}(p,m+p)$, then
$\alpha^{\DOT\perp}$ is Schubert data for $\mbox{\it Grass}(m,m+p)$.
Moreover, Conjecture~\ref{conj:shapiro-grass} holds for
$m,p,\alpha^\DOT$ if and only if it holds for $p,m,\alpha^{\DOT\perp}$.
\item[(ii)]
Suppose $p>2$ and let $J_m:=1<2<\cdots<p-1<p+m$.
If $\alpha^\DOT$ is Schubert data for $\mbox{\it Grass}(p-1,m+p-1)$,
then $\beta^\DOT:=\alpha^{\DOT+},J_m$ is Schubert data for
$\mbox{\it Grass}(m,m+p)$.
Moreover, Conjecture~\ref{conj:shapiro-grass} holds for
$m,p-1,\alpha^\DOT$ if and only if it holds for $m,p,\beta^{\DOT}$.
\item[(iii)]
Let $\alpha^\DOT,\beta^\DOT$ be Pieri Schubert data for
$\mbox{\it Grass}(p,m+p)$ with $\alpha^\DOT\preceq\beta^\DOT$.
If Conjecture~\ref{conj:shapiro-grass} holds for
$\alpha^\DOT$ for $\mbox{Grass}(p,m+p)$, then it holds for
$\beta^\DOT$.
\end{enumerate}
\end{lem}
\noindent{\bf Proof of Theorem~\ref{thm:implies}. }
First note that Conjecture~\ref{conj:shapiro-grass} holds for Schubert data
$\alpha^\DOT$ for $\mbox{\it Grass}(p,m+p)$ if and only if it holds for any
rearrangement of the data $\alpha^\DOT$.
Suppose Conjecture~\ref{conj:shapiroII} holds for $\mbox{Grass}(b,a+b)$.
Let $\alpha^\DOT$ be Pieri Schubert data for $\mbox{\it Grass}(p,m+p)$ where
$(m,p)\leq(a,b)$ or $(m,p)\leq(b,a)$ coordinatewise.
Since $J_1^\perp=J_1$, Conjecture~\ref{conj:shapiroII} holds also for
$(m,p)=(b,a)$, by Lemma~\ref{lem:ordering}(i).
Thus we may assume that $(m,p)\leq(a,b)$.
By Lemma~\ref{lem:ordering}(ii), there exist Pieri Schubert data $\beta^\DOT$
for $\mbox{Grass}(b,a+b)$ such that Conjecture~\ref{conj:shapiro-grass}
holds for $\alpha^\DOT$ if and only if it holds for $\beta^\DOT$.
Finally, Theorem~\ref{thm:implies} follows from (iii) by noting that the
Schubert data of Conjecture~\ref{conj:shapiroII}, namely
$\alpha^1=\cdots=\alpha^{ab}=J_1$, is minimal among all
Pieri Schubert data for $\mbox{\it Grass}(b,a+b)$.
\QED
\noindent{\bf Proof of Lemma~\ref{lem:ordering}. }
For (i), fix a real inner inner product on ${\mathbb C}^{m+p}$.
Then the map $X\mapsto X^\perp$ gives an isomorphism between
$\mbox{\it Grass}(p,m+p)$ and
$\mbox{\it Grass}(m,m+p)$.
Given a flag ${K_{\DOT}}$ and an increasing sequence $\alpha$, let
${K_{\DOT}^\perp}$
be the flag of annihilators of the subspaces of ${K_{\DOT}}$.
Then we have
$$
X\in \Omega_\alpha{K_{\DOT}}\ \Longleftrightarrow\
X^\perp\in \Omega_{\alpha^\perp}{K_{\DOT}^\perp}.
$$
Furthermore, if ${K_{\DOT}}(s)$ is the flag of subspaces osculating a rational
normal curve $\gamma$ at a point $\gamma(s)$, then $(K_{m+p-1}(s))^\perp$ is
a rational normal curve with ${K_{\DOT}^\perp}(s)$ its osculating flag.
Thus Conjecture~\ref{conj:shapiro-grass} for Schubert data $\alpha^\DOT$ for
$\mbox{Grass}(p,m+p)$ is equivalent to
Conjecture~\ref{conj:shapiro-grass} for Schubert data $\alpha^{\DOT\perp}$
for $\mbox{Grass}(m,m+p)$.
\smallskip
For (ii), let $\gamma$ be the rational curve~(\ref{eq:afine_rat}) with
${K_{\DOT}}(s)$ as before.
Then $X\in \Omega_{J_m}{K_{\DOT}}(\infty)$ if and only if
$\langle\gamma(\infty)\rangle=K_1(\infty)\subset X$.
Consider the projection
$\pi:{\mathbb C}^{m+p}\twoheadrightarrow{\mathbb C}^{m+p-1}$ from the last
coordinate $\gamma(\infty)$.
If $X\in \Omega_{J_m}{K_{\DOT}}(\infty)$, then $X':=\pi X$ is a $(p{-}1)$-plane.
This induces an isomorphism
$\pi:\Omega_{J_m}{K_{\DOT}}(\infty)\stackrel{\sim}{\longrightarrow}
\mbox{\it Grass}(p{-}1,m{+}p{-}1)$.
The inverse map is given by $X'\mapsto K_1(\infty)+X'$.
The projection $\pi\circ\gamma$ is the standard rational normal curve
$\gamma'$ in ${\mathbb C}^{m+p-1}$.
Similarly, the flag ${K_{\DOT}}'(s)$ osculating $\gamma'$ at $\gamma'(s)$ is
$\pi {K_{\DOT}}(s)$.
Note that if $L$ is a linear subspace of ${\mathbb C}^{m+p}$
with $\gamma(\infty)\not\in L$, then $\dim X\cap L=\dim \pi X\cap \pi L$.
In particular, if $X\in \Omega_{J_m}{K_{\DOT}}(\infty)$, $s\neq \infty$, and
$\alpha\in\binom{[m+p-1]}{p-1}$, then
$\dim X'\cap K'_{(m+p-1)+1-\alpha_i}\geq (p-1)+1-i$ if and only if
$\dim X\cap K_{m+p+1-(1+\alpha_i)}\geq p+1-(i+1)$.
Thus we have
\begin{equation}\label{eq:equiv-proj}
X\in \Omega_{J_m}{K_{\DOT}}(\infty)\cap \Omega_{\alpha^+}{K_{\DOT}}(s)
\ \Longleftrightarrow\ X'\in \Omega_{\alpha}{K_{\DOT}}'(s).
\end{equation}
In fact, this induces an isomorphism of schemes.
This gives a strong equivalence between enumerative problems:
If $\alpha^1,\ldots,\alpha^n$ are in $\binom{[m+p-1]}{p-1}$ and
$s_1,\ldots,s_n$ any complex numbers, then
the map $\pi$ induces an isomorphism between the schemes
$$
\Omega_{J_m}{K_{\DOT}}(\infty)\cap\bigcap_{i=1}^n \Omega_{(\alpha^i)^+}{K_{\DOT}}(s_i)
\quad\mbox{ and }\quad
\bigcap_{i=1}^n \Omega_{\alpha^i}{K_{\DOT}}'(s_i).
$$
Part (ii) follows by noting that any real reparameterization of the rational
normal curve $\gamma$ induces an
isomorphism of polynomial systems, thus preserves real solutions.
Hence given $s_0,s_1,\ldots,s_n\in{\mathbb P}^1_{\mathbb R}$, there is
an equivalent system with $s_0=\infty$.
\medskip
It suffices to prove (iii) when $\beta^\DOT$ covers
$\alpha^\DOT$ in the partial order $\preceq$ defined on Pieri Schubert data.
Suppose Conjecture~\ref{conj:shapiro-grass} fails for $\beta^\DOT$
and $\beta^\DOT$ covers $\alpha^\DOT$ with $\alpha^1<_a\beta$ and
$\alpha^2=J_a$ as in Definition~\ref{def:triple} (iii).
Then there exist distinct real numbers $s_1,s_3,\ldots,s_n$
such that
\begin{equation}\label{eq:intersec}
\Omega_{\beta}{K_{\DOT}}(s_1) \cap
\Omega_{\alpha^3}{K_{\DOT}}(s_3) \cap \cdots \cap \Omega_{\alpha^n}{K_{\DOT}}(s_n)
\end{equation}
is transverse with some complex $p$-planes in the intersection.
We may assume without any loss that $s_1=0$.
Then there is an open subset ${\mathcal O}$ of the set of
$(n-1)$-tuples of real numbers $s_3,\ldots,s_n$ such
that~(\ref{eq:intersec}) is transverse and contains a complex $p$-plane
$X$.
By the dimensional transversality results of~\cite{EH83},
we may assume further that for $\beta'\in\binom{[m+p]}{p}$ and
$(s_2,\ldots,s_n)\in {\mathcal O}$, the
intersection
$$
\Omega_{\beta'}{K_{\DOT}}(0)\cap \bigcap_{i=3}^{n}\Omega_{\alpha^i}{K_{\DOT}}(s_i)
$$
has the expected dimension and is transverse if 0-dimensional.
This is empty if $|\beta'|>a+|\alpha^1|$, for dimension
reasons.
Thus
$$
\left(\sum_{\alpha<_a\beta'}\Omega_{\beta'}{K_{\DOT}}(0)\right)
\cap \bigcap_{i=3}^{n}\Omega_{\alpha^i}{K_{\DOT}}(s_i)
$$
is transverse for $(s_3,\ldots,s_n)\in {\mathcal O}$.
Fix $(s_3,\ldots,s_n)\in {\mathcal O}$.
By Proposition~\ref{prop:pieri}(i),
there is an $\epsilon>0$
such that for $|t|<\epsilon$
$$
\Omega_\alpha{K_{\DOT}}(0) \cap \Omega_{J_a}{K_{\DOT}}(t)
\cap \bigcap_{i=3}^{n}\Omega_{\alpha^i}{K_{\DOT}}(s_i)
$$
is transverse.
Here, when $t=0$, replace $\Omega_\alpha{K_{\DOT}}(0) \cap \Omega_{J_a}{K_{\DOT}}(t)$
by $\sum_{\alpha<_a\beta}\Omega_{\beta'}{K_{\DOT}}(0)$.
Since at $t=0$ not all points in the intersection are real, the same holds
for $0<t<\epsilon$.
But then Conjecture~\ref{conj:shapiro-grass} fails for
the Schubert data $\alpha^\DOT$
and completes the proof of Lemma~\ref{lem:ordering}.
\QED
\subsection{An infinite family}
We show that Conjecture~\ref{conj:shapiro-grass} holds for an infinite
family of nontrivial Schubert data.
\begin{thm}\label{infinite_family}
Conjecture~\ref{conj:shapiro-grass} holds for any $m$ with
$p=2$ and Pieri Schubert data where one condition is $J_{m-1}$.
\end{thm}
\noindent{\bf Proof. }
By Lemma~\ref{lem:ordering}(iii), it suffices to show this for
$\alpha^1=\cdots=\alpha^{m+1}=J_1$ and $\alpha^{m+1}=J_{m-1}$.
Geometrically, we are looking for the 2-planes which meet a 2-plane
and $m+1$ general $m$-planes nontrivially.
We first show there are $m$ such 2-planes.
Let $L=K_2(\infty)=[0\ I_2]$ and $M=K_m(0)=[I_m\ 0]$, and let
$N_i=K_m(s_i)$,
where $s_1,\ldots,s_m$ are distinct nonzero real numbers.
For each one-dimensional subspace $\lambda$ of $L$ and each
$1\leq i\leq m$, the composition
$$
M\ \hookrightarrow\ L\oplus M \simeq {\mathbb C}^{m+2}\
\twoheadrightarrow\
L\oplus M/(\lambda+N_i)\ \simeq {\mathbb C}
$$
defines a linear form $\psi_{i,\lambda}$ on $M$.
Each one-dimensional subspace $\mu$ of its kernel gives a 2-plane
$\lambda\oplus\mu$ containing $\lambda$ and meeting both $M$ and $N_i$
nontrivially.
Thus if $X$ is a 2-plane meeting $L,M$, and each $N_i$ nontrivially,
then $H\cap L=\lambda$ and $H\cap M=\mu$ are lines with $\mu$ in the kernel
of each form $\psi_{i,\lambda}$.
Hence the forms are dependent.
Similarly, if $\lambda$ is a line in $L$ such that the forms
$\psi_{i,\lambda}$ are dependent, then any line $\mu$ they
collectively annihilate gives a 2-plane $\lambda\oplus\mu$ meeting
$L,M$, and each $N_i$ nontrivially.
It follows that the number of such 2-planes is the degree of the determinant
of the forms $\psi_{i,\lambda}$, a polynomial in
$\lambda\in{\mathbb P}(L)\simeq{\mathbb P}^1$.
Since each form $\psi_{i,\lambda}$ is a linear function of
$\lambda$, the determinant has degree $m$, so there are $m$ 2-planes $X$
meeting $L,M$, and each $N_i$ nontrivially.
We compute this determinant and show it has only real roots.
Let $\lambda=\lambda(x)$ be the span of the vector
$$
(0,\ldots,0,1,(m+1)x).
$$
Let the rational normal curve $\gamma$ have the parameterization
$$
\gamma\ :\ s\ \longmapsto\ (1,-s,s^2,\ldots,(-1)^{m+1}s^{m+1}).
$$
Then $K_m(s)$, the osculating $m$-plane to $\gamma$ at $\gamma(s)$,
is the kernel of the matrix:
\begin{equation}\label{vf}
\left[\begin{array}{cccccccc}
s^m&ms^{m-1}&\ldots& {m\choose{j}}s^{m-j} &\ldots& ms &1&0\\
0 & s^m &\ldots&{m\choose{j-1}}s^{m-j+1}&\ldots&\binom{m}{2}s^2&ms&1
\rule{0pt}{14pt}
\end{array}
\right].
\end{equation}
If $R_j(s)$ is the linear form given by the $j$th row of this
matrix, then
$$
((m+1)x+ms_i)R_1(s_i)\ -\ R_2(s_i)
$$
vanishes on $\lambda(x)$ and its restriction to $M$ gives the form
$\psi_{i,\lambda(x)}$.
This restriction is represented by the vector $\Lambda(s_i,x)$
whose $j$th coordinate for $j=0,\ldots,m-1$ is:
$$
{\textstyle {m+1\choose j}}
\left((m{-}j{+}1)xs_i^{m-j}+ (m{-}j)s_i^{m-j+1}\right).
$$
We seek the determinant of the following matrix:
$$
\left[\begin{array}{c}
\Lambda(s_1,x)\\ \vdots\\\Lambda(s_m,x)\end{array}\right].
$$
This factors as $A\cdot B$, where $A$ is the
bidiagonal $m\times (m+1)$-matrix
$$
\left[
\begin{array}{cccccccccc}
m&(m+1)x&0\\
0&m^2-1&m(m+1)x&0\\
&\ \makebox[.1in][l]{\qquad$\ddots$}&
\makebox[.1in][l]{\ \qquad$\ddots$}&\\
&0&{m+1\choose j}(m{-}j)& {m+1\choose j}(m{-}j{+}1)x&&0\\
&&\makebox[.1in][l]{\qquad\quad$\ddots$}&
\makebox[.1in][l]{\qquad\quad\ $\ddots$}&\\
&&0&{m+1\choose 2}\quad&&{m+1\choose 2}2x\end{array}\right]
$$
and $B$ is the $(m+1)\times m$-matrix whose $i,j$th entry is
$s_j^{m+2-i}$.
Numbering the rows of $A$ and the columns of $B$ from 0 to $m$, we see
that
$$
\det(A(x)\cdot B)\ =\ \
\sum_{i=0}^m (-1)^i \det A_i(x) \det B_i,
$$
where $A_i$ is the matrix $A$ with its $i$th column removed and $B_i$
is the matrix $B$ with its $i$th row removed.
We find that
\begin{eqnarray*}
\det(A_i)&=& m!(m+1-i)x^{m-i}\prod_{j=1}^m{\textstyle \binom{m+1}{j}}\\
\det B_i&=& e_i(s_1,\ldots,s_m)s_1s_2\cdots s_m\cdot\prod_{j<k}(s_j-s_k)
\end{eqnarray*}
and so $\det(A\cdot B)$ is
$$
m! \prod_{j<k}(s_j-s_k)\prod_{j=1}^m s_j {\textstyle \binom{m+1}{j}}
\cdot \left(\sum_{i=0}^m (-1)^i(m-i+1)x^{m-i}e_i(s_1,\ldots,s_m)\right).
$$
Thus the coordinate $x$ of the line $\lambda$ satisfies the polynomial
$$
P_m(s_1,\ldots,s_m;x)\ :=\
\sum_{i=0}^m (-1)^i(m-i+1)x^{m-i}e_i(s_1,\ldots,s_m).
$$
Since we have
$e_i(s_1,\ldots,s_m)=e_i(s_1,\ldots,s_{m-1})+s_me_{i-1}(s_1,\ldots,s_{m-1})$,
we see that
$$
P_m(s_1,\ldots,s_m;x)= (x-s_m)P_{m-1}(s_1,\ldots,s_{m-1};x) +
x \prod_{i=1}^{m-1}(x-s_i).
$$
To complete the proof, we use induction to show that
$$
(*)\qquad\qquad \mbox{
\begin{minipage}{4.5in}
If $0<s_1<\cdots<s_m$, then the roots $r_1,\ldots,r_m$
of $P_m$ satisfy:
$$
0<r_1<s_1<r_2<s_2<\cdots<r_m<s_m.
$$
\end{minipage}
}\qquad
$$
This suffices, if we can assume $0<s_1<\cdots<s_m$.
But we may assume this:
Given a set of distinct real numbers
$s_1,\ldots,s_m,s_{m+1},s_{m+2}$, we may assume $s_{m+2}=\infty$ and
$s_{m+1}<s_1<\cdots<s_m$ and then apply the automorphism
$s \mapsto s-s_{m+1}$ of ${\mathbb P}^1({\mathbb R})$ which fixes
$\infty=s_{m+2}$.
The case $m=1$ of $(*)$ holds as $P_1(s_1;x)=2x-s_1$.
Suppose $P_{m-1}$ satisfies $(*)$.
Then the roots of $(x-s_m)P_{m-1}$ are $r_1<r_2<\cdots<r_{m-1}<s_m$
and those of $x\prod_{i=1}^{m-1}(x-s_i)$ are
$0<s_1<\cdots<s_{m-1}$.
Moreover the leading coefficients of both polynomials are positive.
The result follows by the Intermediate Value Theorem:
If $P(x)$ and $Q(x)$ are polynomials of degree $n$ with positive leading
coefficients and real interlaced roots $p_i$ of $P$ and $q_i$ of $Q$
$$
p_1<q_1<p_2<q_2<\cdots<p_n<q_n,
$$
then $P(x)+Q(x)$ has real roots $r_i$ satisfying
$p_i<r_i<q_i$, for $i=1,\ldots, n$.
\QED
\subsection{Computational evidence}
We have proven Conjecture~\ref{conj:shapiro-grass} in a number of cases
besides those of Theorem~\ref{infinite_family}.
We also have done many computations along the lines of those in
Section~\ref{sec:computation}.
To describe these, we use the following compact notation.
If a Schubert condition $\alpha$ is repeated $k$ times in some Schubert
data, we abbreviate that by $\alpha^k$.
Thus, the conditions of Conjecture~\ref{conj:shapiroII} are written
as $J_1^{mp}$.
\begin{thm}\label{thm:grass-proof}
Conjecture~\ref{conj:shapiro-grass} holds for the following
Schubert data.
\begin{enumerate}
\item[(i)] $(m,p)=(4,2)$, $\alpha^\DOT=J_2^4$.
Here, $d(4,2;J_2^4)=3$.
\item[(ii)] $(m,p)=(3,3)$, $\alpha^\DOT=J_2^4,J_1$.
Here, $d(3,3;J_2^4,J_1) = 3$.
\item[(iii)] $(m,p)=(3,3)$, $\alpha^\DOT=(135)^2,J_1^3$.
Here, $d(3,3;(135)^2,J_1^3) = 6$.
\item[(iv)] $(m,p)=(4,3)$, $\alpha^\DOT=135^4$.
Here, $d(4,3;135^4)=8$.
\end{enumerate}
\end{thm}
\noindent{\bf Proof. }
We consider a polynomial system with parameters, give a universal eliminant,
and show the eliminant has only real roots for distinct values of the
parameters.
We work in the local parameterization ${\mathcal X}_{\alpha^1,\alpha^2}$
of Section~\ref{sec:loc-coords}.
\noindent{(i) }
Let $(m,p)=(4,2)$ and $\alpha^\DOT=J_2^4$.
The equations are
$$
\mbox{maximal minors }
\left[\begin{array}{cccccc}
1&s&s^2&s^3&s^4&s^5\\
0&1&2s&3s^2&4s^3&5s^4\\
0&0&1&3s&6s^2&10s^3\\
1&x_{12}&x_{13}&0&0&0\\
0&0&0&1&x_{25}&x_{26}\end{array}\right]
\ =\ 0
$$
and the same equations with $t$ replacing $s$.
The ideal of these polynomials contains the following univariate polynomial
$g$ of degree $3=d(4,2,J_2^4)$.
$$
25x_{12}^3-25x_{12}^2(s+t) +x_{12}(19st+6s^2+6t^2)-3(s^2t+st^2)
$$
whose discriminant has primitive part
$$
9(s-t)^6 + 23s^2t^2(s-t)^2 + 9(s^6+t^6).
$$
Since $g(x_{12};1,2)$ has roots
$$ 1,\ 1\pm\frac{1}{5}\sqrt{7},$$
we have shown that $g$ always has real roots, when $s$ and $t$ are distinct.
\medskip
\noindent{(ii) }
Let $m=p=3$ and $\alpha^\DOT=J_2^4,J_1$.
Here, ${\mathcal X}_{J_2,J_2}$ consists of all matrics $X$ of the form
$$
\left[\begin{array}{cccccc}
1&x_{12}& 0 & 0 & 0 & 0 \\
0& 1 &x_{23}&x_{24}&x_{25}& 0 \\
0& 0 & 0 & 0 & 1 &x_{36}\end{array}\right]
$$
and our equations are
$$
\det\zwei{K_3(s)}{X}\ =\
\mbox{maximal minors }\zwei{K_2(t)}{X}\ =\ 0
$$
and the same equations with $u$ replacing $t$.
The ideal of these polynomials contains the following univariate polynomial
$g$, here $e_1=t+u$ and $e_2=tu$.
$$
\begin{array}{c}
x_{36}^3-x_{36}(3s+4e_1)
+x_{36}(4e_1^2+3e_2+10se_1)
-(6e_1e_2+8se_1^2+se_2)\qquad\qquad\qquad\qquad\\
\qquad\qquad\qquad\qquad =\
(x_{36}-2e_1)(x_{36}^2-2e_1x_{36}+3e_2)
\ -\ s(x_{36}^2-10e_1x_{36}+8e_1^2+e_2).\rule{0pt}{18pt}
\end{array}
$$
These last two polynomials have roots
$$
e_1\pm \sqrt{e_1^2-3e_2},\ 2e_1\quad\mbox{ and }\quad
\frac{5}{3}e_1\pm \frac{\sqrt{e_1^2-3e_2}}{3},
$$
which are interlaced.
For example, if $e_1>0$, then
$$
e_1- \sqrt{e_1^2-3e_2}<\frac{5}{3}e_1- \frac{\sqrt{e_1^2-3e_2}}{3}<
e_1+ \sqrt{e_1^2-3e_2}<\frac{5}{3}e_1+ \frac{\sqrt{e_1^2-3e_2}}{3}<
2e_1.
$$
When $s,t,u$ are distinct and different from 0, $g$ always has 3 real
roots, by the Intermediate Value Theorem.
We could also note that the discriminant of $g$
$$
\begin{array}{c}
s^2(t-u)^4+t^4(s-u)^2 + u^4(s-t)^2 + s^2t^2(s-t)^2 + s^2u^2(s-u)^2+
\qquad\qquad\qquad\qquad\\
\qquad\qquad (s-t)^2(s-u)^2(t-u)^2 \ +\
{\displaystyle \frac{7}{2}}
( s^4(t-u)^2 +t^2(s-u)^4 + u^2(s-t)^4 + t^2u^2(t-u)^2)
\rule{0pt}{18pt}
\end{array}
$$
is a sum of squares and $g(x_{36};1,2,3)$ has
(floating point) roots
$$
4.736,\ 7.756,\ 10.508.
$$
\noindent{(iii) }
Let $(m,p)=(3,3)$ and $\alpha^\DOT=(135)^2,J_1^3$.
Here, ${\mathcal X}_{135,135}$ consists of all matrics $X$ of the form
$$
\left[\begin{array}{ccccccc}
1&x_{12}& 0 & 0 & 0 & 0 \\
0& 0 & 1 &x_{24}& 0 & 0 \\
0& 0 & 0 & 0 & 1 &x_{36}\end{array}\right]
$$
and our equations are
$$
\det\zwei{K_3(s)}{X}\ =\ \det\zwei{K_3(t)}{X}\ =\
\det\zwei{K_3(u)}{X}\ =\ 0.
$$
We write the universal eliminant, $g(x_{36})$, in terms of the elementary
symmetric polynomials in $s,t,u$:
$$
\begin{array}{c}
9x_{36}^6
-48e_1x_{36}^5
+(64e_1^2+108e_2)x_{36}^4
-(288e_1e_2-198e_3)x_{36}^3\qquad\qquad\qquad\\
\qquad\qquad\qquad+(320e_2^2+540e_1e_3)x_{36}^2
-1200e_2e_3x_{36}+1125e_3^2.\rule{0pt}{18pt}
\end{array}
$$
Evaluating the parameters $(s,t,u)$ at $(1,2,3)$, we see that
$g(x_{36};1,2,3)$ has roots:
$$
1.491,\ 1.683,\ 3.210,\ 5.630,\ 9.213,\ 10.773.
$$
The discriminant of $g$ is a sum of squares.
The primitive part of the discriminant is
$$
e_3^4
(4e_2^2e_1^2-15e_3e_1^3-15e_2^3+63e_3e_2e_1-81e_3^2)
(256e_2^2e_1^2-768e_3e_1^3-768e_2^3+2592e_3e_2e_1-2187e_3^2)^2.
$$
The second factor is a sum of squares
$$
\frac{7}{2}(s-t)^2(s-u)^2(t-u)^2\ +\
\frac{1}{2}s^2((t-u)^4+t^2(s-u)^4+u^2(s-t)^4)
$$
Interestingly, the last (squared) factor is itself a sum of
squares:
$$
\begin{array}{cc}
112(s-t)^2(u^4+s^2t^2)+ 112(t-u)^2(s^4+t^2u^2)
+112(u-s)^2(t^4+s^2u^2)+\qquad\qquad\qquad\\
\qquad 16(s-t)^2(s-u)^2(t-u)^2
+309s^2t^2u^2
+16(s^4(t^2+u^2)+ t^4(s^2+u^2)+u^4(t^2+u^2)).
\rule{0pt}{18pt}
\end{array}
$$
\noindent{(iv) }
Let $(m,p)=(4,3)$ and $\alpha^\DOT=(135)^4$.
Here, ${\mathcal X}_{135,135}$ consists of all matrics $X$ of the form
$$
\left[\begin{array}{ccccccc}
1&x_{12}&x_{13}& 0 & 0 & 0 & 0 \\
0& 0 & 1 &x_{24}&x_{25}& 0 & 0 \\
0& 0 & 0 & 0 & 1 &x_{36}&x_{37}\end{array}\right]
$$
and our equations are
$$
\mbox{maximal minors }\zwei{K_3(s)}{X}\ =\
\mbox{maximal minors }\zwei{K_5(s)}{X}\ =\ 0
$$
and the same equations with $t$ replacing $s$.
In this case, the universal eliminant has 4 quadratic factors:
$$
\begin{array}{c}
(36x_{12}^2 -x_{12}( 12t+30s)+6st+5s^2), \ \
(36x_{12}^2 -x_{12}( 12s +30t) +6st + 5t^2), \\
(3x_{12}^2 -2x_{12}(s+t) + st), \ \ \mbox{ and }\ \
(36x_{12}^2 -30x_{12}(s+t)+ 5t^2+14st+5s^2).\rule{0pt}{15pt}
\end{array}
$$
When $s\neq t$ and neither is zero, we see that each has 2 real roots.
\QED
Observe that in all 4 cases, the discriminant was a sum of squares and the
eliminant has the correct number of real roots for distinct values of the
parameters.
Of particular note is that the system in (ii) was not symmetric in the
parameters and the Schubert data of (iv) was not Pieri Schubert data.
\medskip
Table~\ref{tab:pieri} gives the number of instances of
Conjecture~\ref{conj:shapiro-grass} we have
checked.
\begin{table}[htb]
\begin{tabular}{|r||c|c|c||c|c|c|c|c|}
\hline
$\alpha^\DOT$:&$(J_2)^5$&$(J_2)^6$&$(J_2)^7$
&$(J_2)^6$&$(J_3)^5$&$(135)^5$&$(135)^2,(J_1)^6$
\rule{0pt}{15pt}\\\hline\hline
$m,p$:& 5,2& 6,2& 7,2&5,3&4,3&5,3&6,3\\\hline
$d(m,p;\alpha^\DOT)$:&6&15&36&6&16&32&61\\\hline
\# checked&10,000&2821&504&10160&2002&400&294\\\hline
\end{tabular}\bigskip
\caption{General Schubert data tested\label{tab:pieri}}
\end{table}
\section{Total positivity}\label{sec:tot_pos}
Previous sections have dealt with Schubert conditions given by flags
osculating a real rational normal curve.
Recently, Shapiro and Shapiro have conjectured
that a generalization of this
choice involving totally positive real matrices would also give only real
solutions.
We describe that here, prove the first nontrivial instance, and present
some computational evidence in support of this generalization.
A real upper triangular matrix $g$ with 1's on its diagonal is {\em totally
positive} if every minor of $g$ is positive, except those minors which
vanish on all upper triangular matrices.
Let ${\mathcal T}{\mathcal P}$ be the set of all totally positive,
a multiplicative semigroup.
Define a partial order on real flags ${F\!_{\DOT}}$ by ${F\!_{\DOT}}<g{F\!_{\DOT}}$ if
$g\in{\mathcal T}{\mathcal P}$.
\begin{conj}\label{conj:tot-pos}
For any $m,p>1$, let $\alpha^\DOT$ be
Schubert data for $\mbox{\it Grass}(p,m+p)$.
If ${F\!_{\DOT}}^1<\cdots<{F\!_{\DOT}}^n$ are real flags, then
the Schubert varieties $\Omega_{\alpha^1}{F\!_{\DOT}}^1,\ldots,
\Omega_{\alpha^n}{F\!_{\DOT}}^n$
intersect transversally, with all points of intersection real.
\end{conj}
We will prove Conjecture~\ref{conj:tot-pos} in the first nontrivial case
of $m=p=2$.
First, we relate Conjecture~\ref{conj:tot-pos} to
Conjecture~\ref{conj:shapiro-grass}.
Let ${K_{\DOT}}(s)$ be the square matrix of size $(m+p)$ whose $i,j$th entry is
$\binom{j-i}{i-1}s^{j-i}$ ({\it cf.}~(\ref{eq:K-matrix})).
If $s>0$, then ${K_{\DOT}}(s)$ is totally positive
and for any $s,t$ we have ${K_{\DOT}}(s)\cdot{K_{\DOT}}(t)={K_{\DOT}}(s+t)$.
To see this, first recall that ${\mathcal T}{\mathcal P}$ is generated as a semigroup by
$\exp(E_{i,i+1})$, where $E_{i,i+1}$ is the elementary matrix whose only
non-zero entry is in position $i,i+1$~\cite{Loewner}.
These assertions follow from the observation that
$$
{K_{\DOT}}(s)\ =\ \exp(s N),
$$
where $N$ is the nilpotent matrix whose only non-zero entries are
$(1,2,\ldots,m+p-1)$ lying just above its main diagonal.
Theorem~\ref{thm:implies} holds in
this new setting.
For this, we alter the notion of Pieri Schubert data $\alpha^\DOT$
to Schubert data $\alpha^1,\ldots,\alpha^n$ where all except
possibly $\alpha^1$ and $\alpha^n$ are Pieri conditions.
\begin{thm}\label{thm:tot-pos}
Let $a,b>1$ and suppose that Conjecture~\ref{conj:tot-pos} holds for
$(m,p)=(a,b)$ and Schubert data $\alpha^\DOT=(J_1)^{mp}$.
Then Conjecture~\ref{conj:tot-pos} holds for any Pieri Schubert data for
$\mbox{\it Grass}(p,m+p)$ where $(m,p)\leq (a,b)$ or $(b,a)$
coordinatewise.
\end{thm}
\noindent{\bf Proof. }
The arguments used to prove Theorem~\ref{thm:implies} work
here with minor adjustments.
We first remark that total positivity, and hence our order $<$ on real
flags, is defined with respect to a choice of ordered basis for
${\mathbb R}^{m+p}$.
Suppose that $e_1,\ldots,e_{m+p}$ is the basis we used to define this order.
Then ${F\!_{\DOT}}<G_\DOT$ is and only if $G_\DOT<'{F\!_{\DOT}}$, where $<'$ is defined
with respect to the basis $e_1,-e_2,e_3,-e_4,\ldots$.
Similarly, if the basis $e_1,\ldots,e_{m+p}$ is orthonormal,
then ${F\!_{\DOT}}<G_\DOT$ if and only if
$F_\DOT^\perp<''G_\DOT^\perp$, where $<''$ is defined with respect to
the basis in reverse order $e_{m+p},\ldots,e_2,e_1$.
Thus
$$
F_\DOT^1<F_\DOT^2<\cdots<F_\DOT^n\
\Longleftrightarrow\
F_\DOT^n<'\cdots<'F_\DOT^2<'F_\DOT^1
$$
so that Conjecture~\ref{conj:tot-pos} holds for Schubert data $\alpha^\DOT$
if and only if it holds for the data in reverse order.
(This is the only rearrangment we used in the proof of
Lemma~\ref{lem:ordering}.)
Similarly, the analogue of Lemma~\ref{lem:ordering}(i) holds.
For the analogue of Lemma~\ref{lem:ordering}(ii),
permute the last two Schubert conditions, so that $\beta^\DOT$ is
still Pieri Schubert data, in our new, restricted definition.
Finally, in the proof of Lemma~\ref{lem:ordering}(iii), replace
$s_3,\ldots,s_n$ in defining the set ${\mathcal O}$ by fixing $F_\DOT^1$ to
be the standard flag represented by the matrix $I_{m+p}$ and
let ${\mathcal O}$ be the set of all
$$
F_\DOT^1<\cdots<F_\DOT^n
$$
where the appropriate transversality conditions hold.
Since ${\mathcal T}{\mathcal P}$ is open, it follows that there exists $\epsilon>0$ and
totally positive matrix $M$ (which stabilizes $F_\DOT^1$) such that
if $0<s<\epsilon$, then
$F_\DOT^1< M\cdot{K_{\DOT}}(s)\cdot F_\DOT^1< F_\DOT^2$.
Then the same arguments used to prove Theorem~\ref{thm:implies}
suffice.
In particular, the analog of Proposition~\ref{no-mult} also holds in this
setting.
\QED
Totally positive matrices have a useful description.
Let ${\mathcal U}$ be the group of real unipotent (upper triangular)
matrices.
Then ${\mathcal T}{\mathcal P}$ is a connected component of the complement of
a hypersurface $H$ ${\mathcal U}$ defined by the vanishing of all minors
consisting of the first $i$ rows and last $i$
columns~\cite{Shapiro_TP}.
This has a geometric description.
Associating a matrix to a flag as in Section~\ref{sec:loc-coords}, we may
identify ${\mathcal U}$ with a Zariski open subset of
the real flag manifold.
Then the hypersurface $H$ is the union of all positive
codimension Schubert varieties defined by the flag determined by the
identity matrix.
Given a matrix $M\in{\mathcal U}$, the translate
${\mathcal T}{\mathcal P}.M$ is a component of the complement of all Schubert varieties of
positive codimension defined by the flag given by $M$.
Similarly, given a totally positive matrix $M$, the set of upper triangular
matrices $N$ for which there exists a totally positive $g$ with $gN=M$ is
the component of this complement containing the
identity matrix.
Let $F_\DOT^1<\cdots<F_\DOT^n$ be real flags.
Using a real automorphism of the flag manifold, we may assume that
$F_\DOT^1={K_{\DOT}}(0)=I_{m+p}$.
Then $F_\DOT^2,\ldots,F_\DOT^n\in{\mathcal T}{\mathcal P}$, since they are all
translates of the identity by totally positive matrices.
Also, $F_\DOT^1,\ldots,F_\DOT^{n-1}$ are in the same component of the
complement
of all positive dimensional Schubert cells defined by $F_\DOT^n$.
If we now consider a real coordinate transformation fixing $F_\DOT^1$, but
with $F_\DOT^n$ becoming ${K_{\DOT}}(\infty)$, then this complement becomes
${\mathcal T}{\mathcal P}$, in these new coordinates.
Thus we may work in the local coordinates
${\mathcal X}:={\mathcal X}_{\alpha^1,\alpha^n}$.
We do this in our proof of the following theorem
and in subsequent calculations.
\begin{thm}\label{thm:tp22}
Conjecture~\ref{conj:tot-pos} holds for $m=p=2$ and Schubert data
$(J_1)^4$.
\end{thm}
\noindent{\bf Proof. }
Let $F,G\in {\mathcal T}{\mathcal P}$ be totally positive matrices
and set $H = G\cdot F$.
When $m=p=2$, ${\mathcal X}={\mathcal X}_{J_1,J_1}$ is the set of matrices
$$
\left[
\begin{array}{cccc}1&a&0&0\\0&0&1&b\end{array}\right].
$$
For a matrix $L$, let
$L_{ij}$ denote the $2\times 2$-minor of $L$ given by the first two rows
and columns $i$ and $j$.
Then the equations for a 2-plane in ${\mathcal X}$ to meet the flags given
by $F$ and $H$ are
\begin{eqnarray*}
f&:=& F_{24} - b F_{23} - a F_{14} + ab F_{13}\\
h&:=& H_{24} - b H_{23} - a H_{14} + ab H_{13}
\end{eqnarray*}
The lexicographic Gr\"obner basis for this (with $a<b$) is
\begin{eqnarray*}
H_{13}f-F_{13}h&=& J_{14} - b J_{24} -a J_{34}\\
(H_{14}-b H_{13})f-(F_{14}-b F_{13})h
&=&J_{13}-b(J_{23}+J_{14}) + b^2 J_{24},
\end{eqnarray*}
where $J_{ij}$ is the $ij$th minor of the matrix:
$$
\left[
\begin{array}{cccc}
F_{24}& F_{23}& F_{14}& F_{13}\\
H_{24}& H_{23}& H_{14}& H_{13}\end{array}\right].
$$
We may write the the discriminant of the quadratic
equation for $b$ as follows
$$
(J_{23}+J_{14})^2-4J_{13}J_{24}\ =
\ (L_{23}+L_{14})^2-4L_{13}L_{24},
$$
where $L$ is the matrix
$$
\left[
\begin{array}{cccc}
F_{13}& F_{14}& H_{13}& H_{14}\\
F_{23}& F_{24}& H_{23}& H_{24}\end{array}\right].
$$
Thus we will have two real roots for our original system if and only if
$$
\Lambda(B)\ :=\ L_{13}-B(L_{23}+L_{14}) + B^2 L_{24}\ =\ 0
$$
has 2 real solutions.
Painstaking calculations reveal that
$\Lambda(1)=-G_{12}G_{34}<0$.
Since $L_{24}=H_{13}H_{24}-H_{23}H_{14} = H_{12}H_{34}$
by the Pl\"ucker relations, we see that
$L_{24}>0$ and so $\Lambda(B)=0$ will have 2 real solutions.
\QED
Table~\ref{table:tpinstances} shows the number of instances of
Conjecture~\ref{conj:tot-pos} that we have verified.
\begin{table}[htb]
\begin{tabular}{|c||c|c|c|c|c|c|}
\hline $\alpha^\DOT$&$(J_1)^6$&$(J_2)^5$&$(135)^4$&$(J_1)^8$
&$(J_2)^6$&$(135)(136)(J_1)^5$\rule{0pt}{12pt}\\\hline\hline
$m,p$&3,2&5,2&4,3&4,2&6,2&4,3\\\hline
$d$&5&6&8&14&15&25\\\hline
\# checked&12000&4000&4000&1500&300&150\\\hline
\end{tabular}\bigskip
\caption{Instances checked\label{table:tpinstances}}
\end{table}
\section{Further remarks}\label{sec:counter}
We present a counterexample to the original conjecture of Shapiro and
Shapiro and close
with a discussion of further questions.
\subsection{A counterexample to the original conjecture}
The original conjecture of Shapiro and Shapiro concerned the $M$-property
for flag manifolds~\cite{Shapiro_M}.
An algebraic set $X$ defined over ${\mathbb R}$ has the $M$-property if
the sum of the ${\mathbb Z}/2{\mathbb Z}$-Betti numbers of $X({\mathbb R})$
and of $X({\mathbb C})$ are equal.
Shapiro and Shapiro conjectured that an intersection of Schubert cells in a
flag manifold has the $M$-property, if the cells are defined by flags
osculating the rational normal curve at real points.
When such an intersection is zero-dimensional all of
its points are real.
It is this consequence we have been studying.
While there is much evidence in support of this conjecture for zero
dimensional intersections in a Grassmannian
(Conjectures~\ref{conj:shapiroII},~\ref{conj:shapiro-grass},
and~\ref{conj:tot-pos}),
it does not hold for more general flag manifolds.
In fact, we give a counter example in the simplest enumerative problem in a
flag manifold that does not reduce to an enumerative problem in
a Grassmannian.
\begin{cex}\label{cex}
Consider the manifold ${\mathbb F}(2,3;5)$ consisting of partial
flags $X\subset Y$
in ${\mathbb C}^5$ with $\dim X=2$ and $\dim Y=3$.
This manifold has dimension 8;
the projection to $\mbox{\it Grass}(2,5)$ has fibre over a 2-plane $X$ equal
to ${\mathbb P}({\mathbb C}^5/X)\simeq{\mathbb P}^2$.
Given general 2-planes $a,b$, and $c$ and general 3-planes $A,B$, and $C$,
there are 4 flags $X\subset Y$ which satisfy
\begin{equation}\label{eq:flag-cond}
\begin{array}{c}
\mbox{$X$ meets $a,B$, and $C$ nontrivially}\\
\mbox{$\dim Y\cap A\geq 2$ and $Y$ meets $b$ and $c$ nontrivially}
\end{array}
\end{equation}
That this number is 4 may be verified using the Schubert calculus for a flag
manifold~\cite{Fulton_tableaux} or the equations we give below.
Let ${K_{\DOT}}(s)$ be the flag of subspaces osculating the standard rational
normal curve.
Set
$$
\begin{array}{ccc}
a\ :=\ K_2(4)&\quad& A\ :=K_3(0)\\
b\ :=\ K_2(1)& & B\ :=K_3(3)\\
c\ :=\ K_2(-5)& & C\ :=K_3(-1)
\end{array}
$$
We claim that of the 4 flags $X\subset Y$ satisfying~(\ref{eq:flag-cond})
for this choice of $a,b,c,A,B,C$, 2 are real and 2 are complex.
We outline the computation.
Choose local coordinates for ${\mathbb F}(2,3;5)$ as follows.
Let $Y$ be the row space of the $3\times 5$-matrix
$$
\left[\begin{array}{ccccc}
0&0& 1 &x_{14}&x_{15}\\
1&0&x_{23}&x_{24}&x_{25}\\
0&1&x_{33}&x_{34}&x_{35}\end{array}\right]
$$
and $X$ be the row space of its last 2 rows.
We seek the solutions to the following overdetermined system
of polynomials:
$$
\begin{array}{c}
\det\zwei{K_2(1)}{Y}\ =\ \det\zwei{K_2(-5)}{Y}\ =\
\det\zwei{K_3(3)}{X}\ =\ \det\zwei{K_3(-1)}{X}\ = \\ \rule{0pt}{25pt}
\mbox{maximal minors }\zwei{K_2(4)}{X}\ =\
\mbox{maximal minors }\zwei{K_3(0)}{Y}\ =\ 0
\end{array}
$$
These polynomials generate a zero-dimensional ideal
containing the following univariate polynomial, which is part of a
lexicographic Gr\"obner basis satisfying the Shape Lemma:
$$
27063 - 117556 x_{14} - 5952 x_{14}^2
-10416 x_{14}^3 + 32400 x_{14}^4.
$$
This has roots
$$
-.736 \pm 1.30 \sqrt{-1}, \quad .227, \quad 1.62.\
$$
Thus 2 of the flags are complex.
\end{cex}
\subsection{Further questions}
While Counterexample~\ref{cex} shows that we cannot guarantee all points of
intersection real when the Schubert varieties are given by flags osculating
a real rational normal curve, a number of questions remain (besides the
resolution of the conjectures of the previous sections).
There remains the original question of Fulton:
\medskip
\noindent{\bf Question 1: }
Given Schubert data for a flag manifold, do there exist real flags in general
position whose corresponding Schubert varieties have {\it only} real points
of intersection?
\medskip
In every case we know, this does happen.
For instance, if we change the 3-plane $B$ to $K_3(2)$ in
Counterexample~\ref{cex}, then all 4 solution flags are real.
There is also the following result, showing this holds in infinitely many
cases.
Let ${\mathbb F}(2,n-2;n)$ be the manifold of flags $X\subset Y$
in ${\mathbb C}^n$ where $\dim X=2$ and $\dim Y=n-2$.
A Grassmannian Schubert condition is a Schubert condition on the flag
$X\subset Y$ which only imposes conditions on one of the subspaces.
We likewise define Grassmannian Schubert data.
For example, Counterexample~\ref{cex} involves
Grassmannian Schubert data.
\begin{prop}[Theorem 13 of~\cite{Sottile97b}]\label{prop:MEGA}
Given any Grassmannian Schubert data for ${\mathbb F}(2,n-2;n)$,
there exist real flags whose corresponding Schubert varieties meet
transversally with all points of intersection real.
\end{prop}
The beauty of the conjectures of Shapiro and Shapiro is that they give a
simple algorithm for selecting the flags defining the Schubert varieties.
\medskip
\noindent{\bf Question 2: }
Can the choice of flags in Question 1 (or Proposition~\ref{prop:MEGA})
be made effective?
In particular, is there an algorithm for selecting these flags?
\medskip
While computing the examples described here,
we have made a number of observations which deserve further scrutiny.
These concern eliminant polynomials in the ideals defining the intersections
of Schubert varieties in the local coordinates we have been using.
Suppose we have Schubert data $\alpha^\DOT$, and have chosen local
coordinates either for the Grassmannian or are working in
${\mathcal X}_{\alpha^n,\alpha^{n-1}}$.
Conjecture~\ref{conj:shapiro-grass} or~\ref{conj:tot-pos} may be formulated
in terms of a parameterized system of polynomials with parameters either
$s_1,\ldots,s_n$ in the case of Conjecture~\ref{conj:shapiro-grass} or
$(n-1)$-tuples of totally positive matrices (or in terms of some
parameterization of ${\mathcal T}{\mathcal P}$~\cite{BFZ}).
For each of the coordinates, the ideal of this system contains a
universal eliminant, which is the minimal univariate polynomial in that
coordinate with coefficients rational functions in the parameters.
We ask the following questions about the eliminant.
\medskip
\noindent{\bf Question 3: }
Does the universal eliminant have degree equal to the generic number of
solutions?
That is, do generic solutions satisfy the shape lemma?
\medskip
\noindent{\bf Question 4: }
Let $\Delta$ be the discriminant of the polynomial system,
a polynomial in the parameters which vanishes when there are solutions with
multiplicities.
\begin{enumerate}
\item[a)]
Is the locus $\Delta\neq 0$ connected?
\item[b)]
In the case of Conjecture~\ref{conj:shapiro-grass}, where $\Delta$ is
a polynomial in the parameters $s_1,\ldots,s_n$, is $\Delta$ always a
sum of squares of polynomials?
\item[c)]
If so, are these polynomials monomials in the $s_i$ and their
differences $(s_i-s_j)$?
This would imply that the polynomial systems are always multiplicity-free
for distinct real values of the parameters, and hence the stronger
version of Theorem~\ref{thm:implies} mentioned in
Remark~\ref{rem:stronger}.
\end{enumerate}
\medskip
The discriminants we have computed for instances of the conjectures
for the Grassmannian (including the discriminant for system of
Theorem~\ref{thm:tp22}) are always non-negative when the parameters are
distinct.
For the case of Counterexample~\ref{cex}, we computed a
discriminant for a simpler, but equivalent system, in the spirit of
sections~\ref{sec:simpler} and~\ref{sec:loc-coords}.
This polynomial in parameters $s_1,s_2,t_1,t_2$ is symmetric in
the $s$'s and in the $t$'s separately (and in the transformation
$s_i\leftrightarrow t_i$) and has degree 24.
It has three factors, the first of degree 20 with 857 terms,
and the square
$$
(2s_1s_2+2t_1t_2-(s_1+s_2)(t_1+t_2))^2.
$$
While this factor will not prevent the discriminant from being a sum of
squares, this factor shows that there is
a choice of distinct parameters for which the discriminant vanishes.
Indeed, if we set $s_1=3,s_2=6,t_1=9$, and $t_2=5$, then this factor
vanishes, and the resulting system has a root of multiplicity 2.
This also explains why different values of the parameters in
Counterexample~\ref{cex} give different numbers of real and complex
solutions.
\medskip
\noindent{\bf Question 5: }
When the universal eliminant factors over ${\mathbb Z}$, it
reflects either some underlying geometry or some interesting arithmetic.
More generally, one might ask about the Galois group of these enumerative
problems~\cite{Harris_galois},
or the Galois group of the universal eliminant.
For instance, is it the full symmetric group?
That is not always the case, as the example
of Theorem~\ref{thm:grass-proof}(iv) shows.
\medskip
\noindent{\bf Question 6: }
In many cases with the substitution of $s_i=i$,
the eliminant factors over the integers.
This happens in Conjecture~\ref{conj:shapiroI}, Theorem~\ref{thm:23},
Theorem~\ref{thm:grass-proof}(i) and (iv), and in other cases.
Table~\ref{table:factor} lists the degrees of the factors in the case of
Conjecture~\ref{conj:shapiroI}.
\begin{table}[htb]
\begin{tabular}{|c||c|c|c|c|c|c|c|}
\hline$m,p$&3,2&4,2&5,2&6,2&7,2&3,3&3,4\\\hline\hline
$d_{m,p}$&5&14&42&132&429&42&462\\\hline
Factors&2,3&6,8&10,32&20,112&&6,36&16,30,416\\\hline
\end{tabular}\bigskip
\caption{Factorization of the eliminant\label{table:factor}}
\end{table}
Why does this choice of $s_i=i$ induce a factorization?
Is there any special geometry or interesting arithmetic here?
If 2 parameters are allowed to come together, then
the resulting ideal factors in a way respecting the
product of Schubert classes, by the Corollary to Theorem 1
in~\cite{EH87}.
From the Schubert calculus, we would expect factors of 9 and 5 for
$(m,p)=(2,4)$, 14 and 28 for $(m,p)=(2,5)$, and 21 and 21 for
$(m,p)=(3,3)$, but these do not appear in Table~\ref{table:factor}.
\section*{Acknowledgements}
I thank Boris Shapiro for sharing these conjectures
and Joachim Rosenthal with whom I found the first strong evidence for their
validity~\cite{RS98}.
Bernd Sturmfels helped me learn computer algebra and
suggested using discriminants.
This work also benefitted from discussions with
David Eisenbud, Ioannis Emiris, Jean-Charles Faug\`ere, Bill
Fulton, Birk Huber, Bernard Mourrain, Bruce Reznick,
Fabrice Rouillier, Jan Verschelde,
and Paul Zimmermann.
Lastly, this would not be possible were it not
for the authors of the following software packages:
Linux, Maple, Singular, Macaulay2, MuPAD, FGB, RealSolving, PHC,
and also my brothers Larry Sottile and Joe Sottile who helped me build the
computer which performed most of these experiments.
|
\section{Introduction}
The quasar Q2237+0305 at a redshift of $z = 1.609$ is lensed by a
relatively nearby galaxy at $z_L = 0.039$ (\cite{huc85}). It is a
quadruply-imaged case, and one of the best investigated lens systems,
both observationally and theoretically. For recent work, see
\cite{cha98,yon98} and \cite{bla98,med98}, Schmidt et al. (1998),
respectively.
Q2237+0305 was the first multiple quasar system in which microlensing
was detected (see, e.g. \cite{irw89}, \cite{cor91}, \cite{lew98}).
The analysis of well covered microlensing light curves of a quasar can
be used to uncover its size and structure (Wambsganss et al. 1990,
Wambsganss \& Paczy\'nski 1992). A number of groups are optically
monitoring this system to measure any microlensing effects. The
expected time delay between the four images is only of order a day
(\cite{rix92, wam94}) and hence unlikely to be determined from optical
light curves.
Recently, HST observations in the UV allowed the determination of
highly accurate relative positions of the four images (Blanton et
al. 1998). With ground-based spectrophotometry, an extended arc
comprising three of the four images was discovered (Mediavilla et
al. 1998).
Not very much is known yet about gravitationally lensed quasars in
X-rays. The double quasar Q0957+561 was seen with HEAO-1 and with
ROSAT, and dramatic differences in the flux of image B of up to a
factor of five were observed \cite{chartas95}. There is an X-ray
selected gravitationally lensed quasar, RX J0911.4+0551 which was
found in the ROSAT All-Sky Survey \cite{bade97}. With $z_Q=2.8$ and
an X-ray luminosity of $L_X=4.1 \times 10^{46}$ergs/s it is a very
X-ray bright quasar. The two bright images are separated by 0.8
arcsec. High-resolution optical/infrared imaging revealed four images
with a maximum distance of 3.1 arcsec (\cite{bur98}, see also
Munoz et al. 1999 or
http://cfa-www.harvard.edu/castles for HST/NICMOS data obtained by
the CASTLES collaboration).
Here we present the first X-ray detection of Q2237+0305, an analysis
of a ROSAT/HRI observation. The combined X-ray emission of the four
quasar images is clearly detected, though at a relatively low count
rate. Due to the coarse resolution of the ROSAT/HRI the individual
images are not resolved.
\section{Observations}
We observed the quasar Q2237+0305 with the ROSAT/HRI (\cite{tr83}) for
a total exposure time of $t_{\rm ex} = 53 869.7$ seconds. The
observations took place between November 20 and December 5, 1997. The
Standard Analysis Software System (SASS) determined an average
background rate of 0.0032 counts/sec/arcmin$^2$ which -- multiplied by
the ``exposure time" $t_{\rm ex}$ yields an average of 171.7
background counts per square arcmin in total.
To determine the count rate of Q2237+0305, we extracted the photons in
circles of different sizes and subtracted the background, whose count
rate was determined from empty regions of considerably larger size. A
circle with a radius of 15 arcseconds centered on the pixel with the
highest count (RA 22:40:30.21, Dec +03:21:28.7; J2000) resulted in a
total number of 361 counts. The average number of background photons
determined from ten ``empty" circles nearby resulted in 39.8
background photons. This leaves 321.2 source counts, which results in
a count rate of 6.0 counts per kilosecond.
A similar determination with a much larger extraction radius of 50
arcseconds (100 pixels) centered on position RA 22:40:30.0, Dec
+03:21:28.7 produced 800 counts. The average background for this size
is 466.8 counts, which results in 333.2 source counts and a count rate
of $(6.2 \pm 2.8)$ counts/ksec.
The complete HRI field of the exposure is shown in Figure
\ref{fig-hri}. The central source is Q2237+0305 (labelled ``1").
Table \ref{tab-sources} contains the positions and count rates of all
the sources in the field that are detected with a S/N of at least 4.0
in one of the ROSAT/HRI detection cells (squares with sizes ranging
from 12 to 120 arcsec side length). In the table we also list the
count rates and possible identifications of these X-ray detections.
Cross-checking with the databases SIMBAD and NED,
we
could find one other identification of an X-ray detection in our field
with a catalogued source (aside from the ``target" Q2237+0305): Source
No. 2 coincides with the G0 star BD +02 4540 (V magnitude: 9.6). The
G5 star HD 214787 (V magnitude: 8.3) is about 50 arcsec off the
position of detection No. 5, but this is a very unlikely match, even
considering the poor accuracy for large off-axis angles.
\begin{table*}[htb]
\begin{center}
\begin{tabular}{lllrrrrrrl}
& & & frame & pixel& world &coord & & counts per & nearby \\
No. & RA & Dec & x &y & x & y &S/N & 1000sec & counterpart \\
\\
1 & 22 40 30.05 & 3 21 28.94 & 304&299& 34.5& 14.5 & 14.4 & 6.2 & Q2237+0305\\
2 & 22 39 25.21 & 3 20 35.14 & 499&288& 1984.5&124.5 & 12.5 & 25.5 & BD +02 4540, V=9.6, \\
& & & & & & & & & G0-star (within fraction of arcsec) \\
3 & 22 40 6.87 & 3 18 10.74 & 373&259& 724.5&414.5 & 10.1 & 2.9 &\\
4 & 22 40 22.62 & 3 24 50.49 & 326&339&254.5&-385.5 & 5.2 & 1.1 &\\
5 & 22 40 24.74 & 3 07 32.64 & 319&131&184.5&1694.5 & 4.6 & 2.8 & HD 214787, V=8.3, \\
& & & & & & & & & G5-star (50 arcsec off)\\
6 & 22 40 50.78 & 3 23 8.02 & 242&318&-585.5&-175.5 & 7.3 & 1.7 &\\
7 & 22 40 53.3 & 3 38 34.29 & 236&505&-645.5&-2045.5 & 4.5 & 9.8 &\\
8 & 22 41 23.25 & 3 22 25.15 & 144&310&-1565.5&-95.5 & 3.6 & 1.1 &\\
\\
\end{tabular}
\end{center}
\caption{\label{tab-sources}
Positions (J2000.0), signal-to-noise ratio, counts per ksec
and comments on optical
identifications/nearby counterparts
for the eight sources detected in the field of quasar Q2237+0305.
The numbers correspond to the labels in Figure \ref{fig-hri}.
}
\end{table*}
Figure \ref{fig-zoom} depicts a higher-resolution map of
Q2237+0305.
The image appears slightly elliptical; but the same small
ellipticity is seen in other images as well and hence seems
to be an artifact of imperfect pointing
(the separation of the four quasar images is only of order one
arcsecond; this cannot explain the apparent extension).
\begin{figure}[htb]
\psfig{figure=2237_ROSAT_fig1.ps ,width=7.8cm,clip=}
\caption{\label{fig-hri} ROSAT/HRI image of the full field
of Q2237+0305 (center, labelled ``1"). The field size is about 1 degree.
Positions and count rates of the numbered X-ray detections are
given in Table \ref{tab-sources}. The data is smoothed with a Gaussian
of $\sigma=5$ arcsec.
}
\end{figure}
\begin{figure}[htb]
\psfig{figure=2237_ROSAT_fig2.ps ,width=7.8cm,clip=}
\caption{\label{fig-zoom} Zoom of the central part
of Fig. \ref{fig-hri} (here unsmoothed):
ROSAT/HRI X-ray image of Q2237+0305; the four images are unresolved.
}
\end{figure}
\section{Results and Discussion } \label{sec-results}
The average rate of $(6.2\pm2.8)$ counts/ksec
translates into an energy flux of
$(2.2\pm1.0) \times 10^{-13}$ erg/sec/cm$^2$ in the
interval 0.1-2.4 keV,
with the assumption of a hydrogen column density of $n_H = 5.5 \times
10^{20}$ cm$^{-2}$ (Dickey \& Lockman 1990) and a power law index
(photon) of
$\alpha = 1.5$. This flux can be converted into an X-ray luminosity
in the ROSAT energy window of $4.2 \times 10^{45}$erg/sec. The
intrinsic X-ray luminosity of the quasar must be lower, since due to
the gravitational lensing there is a magnification of at least a few,
possibly even a few hundred (\cite{ken88,rix92,wam94}).
As briefly mentioned in the introduction, it is of great interest to
study any variability of this quasar. The standard ROSAT/SASS analysis
did not find any indication for variability in Q2237+0305. With only
about 330 source photons spread over two weeks in real time, it is
difficult to determine any variability. In Figure \ref{fig-timing},
we display the observing intervals of this X-ray exposure for possible
comparison with observations in other wave bands at the same time.
The ``zero" on the time axis corresponds to November 20, 1997,
01:49:58 UT (or JD 2450772.576). It is clear from this Figure that
the ``coverage factor" is less than 5\%. So it makes no sense to
present a continuous light curve of Q2237+0305 over the observing
period.
\begin{figure}[htb]
\psfig{figure=2237_ROSAT_fig3.ps ,width=7.8cm,clip=}
\caption{\label{fig-timing}
Graphical ``log" of duration and epoch
of the observing intervals for the ROSAT pointing
at Q2237+0305.
}
\end{figure}
We can also ``bin" the data to compare the Q2237+0305 ``light curve"
with the light curve of the ``background" (or of the other seven
detected sources, cf. Table \ref{tab-sources}). Such a binned
artificial light curve is displayed in Figure \ref{fig-lightcurve} for
bin widths of 1000 seconds\footnote{For the binning
we basically put together all observing intervals back-to-back,
thus ignoring all the ``dead times". The data in these
53 bins of 1000sec each are in fact spread out over about
about two weeks (cf. Figure \ref{fig-timing})}.
The top panel is the X-ray light curve
for Q2237+0305, the bottom panel is the ``light curve"
of a background field with the same
average count rate (6.2 counts per ksec).
In order to find out whether the ``peaks" in the quasar light curve
could be due to enhanced background radiation, we chose
to compare the quasar light curve with a background light curve
normalized to the same average count rate.
Figure \ref{fig-lightcurve} shows that the fluctuations in the
flux of Q2237+0305
are uncorrelated with the variations in the background flux.
\begin{figure}[htb]
\psfig{figure=2237_ROSAT_fig4.ps ,width=7.8cm,clip=}
\caption{\label{fig-lightcurve} TOP:
``X-ray light curve" for Q2237+0305.
This is the light curve as obtained by
attaching all observing intervals next to each other
(cf. Figure \ref{fig-timing}).
Only eleven out of the 53 bins correspond to
real observing intervals with length
1000 seconds or more. The photons were obtained in a circle
with a diameter of 15 arcsecond.
The expected background contribution is 0.63 counts per 1000-sec-bin.
The average count rate is 6.2 counts per ksec.
The abscissa gives the bin number. The bins are
1000 second wide.
BOTTOM: X-ray light curve for a background field with the same
average count rate as the above quasar light curve.
The binning is identical for the top and bottom panels.
}
\end{figure}
There is no obvious variability visible at the top of Figure
\ref{fig-lightcurve} that exceeds any variability in the bottom panel
(which sets the ``noise" level). Neither is there any ``correlated"
variability obvious, which could be expected for a highly increased
background at certain phases of the observation (since the extraction
circle of the Q2237+0305 light curve is much smaller than that for the
background, this would be surprising).
We investigated the issue of variability more quantitatively. We
performed a KS-test with the real arriving times of the source photons
of Q2237+0305 (extraction radius 30 arcseconds), comparing them with
the background light curve of a circular region corresponding to the
same total number of photons and for the total background. Both were
consistent with no variability. Furthermore, we performed a
chi-square test with the binned data. Similarly, we found no
indication for variability.
Among the 53 bins, the bin with the highest count contains 15 photons,
and one bin is completely empty. The Poissonian probability for
finding 15 counts for an average of 6.2 is only about
$1.2 \times 10^{-3}$, the
Poissonian probability for finding 0 counts with the same average is
$2.0 \times 10^{-3}$.
Similarly, if one divides the bins in two sets, the first 26 bins
contain 150 counts; the standard variation for an average of 150
counts is $\sigma_{26{\rm ksec}} = 12.25$ counts. The second 26 bins
contain 194 counts, which is 2.4 $\sigma_{26{\rm ksec}}$ above the
counts of the first half. These two tests leave open the possibility
of source variability, but at very low significance.
Another issue is a possible contamination of the X-ray counts by the
(lensing) galaxy. We estimated the X-ray luminosity of this galaxy
both by using Dell'Antonio et al. (1994)'s correlation $L_X(spiral) =
2.0 \times 10^{29} L_B^{1.0}$ (where $L_X$ is in erg s$^{-1}$ and
$L_B$ is in solar luminosities), as well as by extrapolating from the
known X-ray count rate of M31 (West et al. 1997). Both these estimates
result in a possible contribution of the lensing galaxy of less than
one percent of our detection, which hence can be neglected.
\section{Conclusions and Outlook}\label{sec-dis}
The detection of the quadruply-imaged quasar Q2237+0305 in X-rays with
($6.2\pm2.8$) counts per ksec opens up the possibility of being able to monitor
this system with the next generation of X-ray telescopes (Chandra
X-ray Observatory, XMM, ASTRO-E). It would then be feasible to study
both the intrinsic variability of the quasar and microlens-induced
fluctuations. The Chandra X-ray Observatory with its on-axis
resolution of 0.5 arcsec and its effective area almost
twice as large as ROSAT's
will be able to detect and resolve the four images of
Q2237+0305. In addition to the possibility of determining
microlens-induced fluctuations (see Yonehara et al. 1998), such
observations could offer the opportunity of measuring relative time
delays in this system. Intrinsic X-ray variations of the lensed
quasar on time scales of less than a day would be required. On the
other hand, if one could follow a ``caustic crossing event" in X-rays
(which should appear in only one of the four images, according to the
very high magnifications expected due to the small source size), we
could also have the possibility of determining the size or even the
source profile of the X-ray emission region of the quasar.
\begin{acknowledgements}
It is a pleasure to thank Ingo Lehmann for providing help with his
MIDAS tools. This research has made use of the NASA/IPAC
Extragalactic Database (NED) which is operated by the Jet Propulsion
Laboratory, California Institute of Technology, under contract with
the National Aeronautics and Space Administration.
This research has also made use of the SIMBAD database,
operated at CDS, Strasbourg, France.
\end{acknowledgements}
|
\section{acknowledgment}
\begin{description}
\item The work of O. Z. was supported by International Science Education
Program (ISEP), grant No. QSU082068.
\end{description}
|
\section{Introduction}
\ifpreprintsty
\else
\begin{multicols}{2}
\fi
We have had a reasonably complete picture of the bacterial light-harvesting (LH)
system recently~\cite{WK,HS}.
Both the inner antenna, LH1, and the outer antenna, LH2, are assembled
from the same modules to form rings.
Each module consists of two short $\alpha$-helical polypeptides coordinate
one carotenoid and three bacteriochlorophylls (BChls).
The LH2 is composed of 9 units, for
{\it Rhodopseudomonas acidophila}~\cite{MPFHPCI}, resemble a cylinder,
with an inner diameter $36 \AA$ and an outer diameter $68 \AA$,
while the LH1 is composed of 16 units,
for {\it Rhodospirillum rubrum}~\cite{KBG}, in order
to accommodate the reaction center (RC).
The later has an outer diameter $116 \AA$
and a central diameter $68 \AA$. However, the
exact numbers of both complexes are variable~\cite{WK,KBG,BJMGF}.
Furthermore, the LH2 B850 BChl $a$ form a complete over-lapping ring
in a hydrophobic environment, which reduces the dielectric constant,
while the B800 BChl $a$ are well separated and are in a polar environment.
When a BChl molecule is excited by light, the energy
can reach equilibrium within about $10 ps$~\cite{TFG}.
A LH2 can function as a storage ring to store the excited
singlet state energy for about $1100 ps$. However, the energy will
transfer to other rings before decaying. The hopping of
energy continues from one ring to another one until a LH1, which contains
the RC, is finally reached. The total trip lasts
for about $ 5$ to $50 ps$~\cite{MPFHPCI,BFGRT,SGBAG}.
Apparently, there is a competition
between energy relaxation and energy transfer.
Historically, relatively few
physicists have tackled problems of photosynthesis.
Notably, Montroll used random walk concepts to model
energy transfer
amongst antenna rings on a lattice by
considering its first passage time~\cite{M}.
Later, Hemenger {\it et al.} proposed a more realistic model by
taking inhomogeneous transfer rates and trapping of RCs
into account~\cite{HPL}.
Interestingly, it is Pearlstein's work which is most often cited
in the literature~\cite{P}.
In the mean time,
almost all experimentalists try to find some
explanations for their spectral data.
However, due to lack of precise geometrical
information most efforts are in vain.
Progresses in physics are often made along the line structures -
energy - dynamics.
A goal of researches nowadays is to find the relation between structural
and spectral information obtained, expecting that
the function of photosynthesis will be explained
in terms of its structure, and further
drawing inferences from the model by applying methods of mathematical or
numerical analysis.
Recently Timpmann {\it et al.} used a rate equation model to describe
energy trapping and detrapping by the RC~\cite{TZFS}.
However, their antenna has no structure.
Sk{\'a}la {\it et al.} also carried out a series of investigation
by analyzing the spectrum of a more realistic LH1 model~\cite{SK,SJ,DS}.
However, their model is incompatible with the recent structural finding.
In this paper we established a two-parameter model
based on recent structural data.
\ifpreprintsty
\else
\end{multicols}
\fi
\section{model}
With the known periodical structure, shown
in Fig.\ref{LH1}, we can built, from chemical rate equation,
the following phenomenological model
of energy transfer,
\begin{eqnarray}
{{d E} \over {d t}} &=& {k^{'}} A_{1} - ( k^{''} + k_E ) E \; , \label{m1}\\
{{d A_1} \over {d t}}&=& k A_{16} - 2 k A_1 + k A_{2} - k^{'}
A_1 + k^{''} E \; ,\\
{{d A_n} \over {d t}}&=& k A_{n-1} - 2 k A_n + k A_{n+1} \; ,
\qquad n = 2...15 \; , \\
{{d A_{16}} \over {d t}}&=& k A_{15} - 2 k A_{16} + k A_{1} \; ,\label{m4}
\end{eqnarray}
in which $A_n$s denote the excited BChl dimer,
$E \equiv P^* B H$ is the excited state,
with B representing the
chlorophyll monomer within the RC, and $P^*$ is the excited special pair of
BChl molecules.
It is a set of 17 coupled linear differential equations.
The symmetry of this
system is broken due to $k^{'} \neq k^{''}$.
A similar model has been proposed by Sk{\'a}la {\it et al.} ~\cite{SJ}.
However, the RC and the antenna
ring are connected only at one site in the present
model, corresponding to the recent experimental observation.
\ifpreprintsty
\else
\begin{figure}[hbt]
\begin{center}
\epsfxsize=\columnwidth\epsfbox{lh1.eps}
\caption[1]{Schematic plot of LH 1 and definition of symbols used.}
\label{LH1}
\end{center}
\end{figure}
\fi
In the homogeneous case with the same transition rate amongst the units,
the characteristic polynomial of the above rate-constant-matrix
can always be expressed as
\begin{equation}
P_{16} = P_{16}^1 P_{16}^2 P_{16}^3 P_{16}^4 \;,
\end{equation}
with
\begin{eqnarray}
P_{16}^1 &=& s+2 k \;, \\
P_{16}^2 &=& s^2+4 k s+2 k^2 \;, \\
P_{16}^3 &=& s^4 + 8 k s^3 + 20 k^2 s^2 + 16 k^3 s + 2 k^4 \;, \\
P_{16}^4 &=& s^{10} + ( k_E + k^{''} + k^{'} + 18 k ) s^9 + \nonumber \\
& & ( k^{'} k_E + 18 k k_E + 18 k k^{''} +
16 k k^{'} + 134 k^2 ) s^8 + \nonumber \\
& & 2 (8 k^{'} k_E + 67 k k_E + 67 k k^{''} + 52 k k^{'} + 266 k^2 ) k s^7 + \nonumber \\
& &2 (52 k^{'} k_E + 266 k k_E + 266 k k^{''} +
176 k k^{'} + 605 k^2) k^2 s^6 + \nonumber \\
& & 2 ( 176 k^{'} k_E + 605 k k_E +
605 k k^{''} + 330 k k^{'} + 786 k^2 ) k^3 s^5 + \nonumber \\
& & 12 ( 55 k^{'}k_E+ 131 k k_E + 131 k k^{''} + 56 k k^{'}+ 91 k^2) k^4 s^4 + \nonumber \\
& & 4 ( 168 k^{'}k_E + 273 k k_E + 273 k k^{''}+ 84 k k^{'}+ 86 k^2 ) k^5 s^3
+ \nonumber \\
& &8 ( 42 k^{'}k_E +43 k k_E +43 k k^{''} + 8 k k^{'}+ 4 k^2 ) k^6 s^2 + \nonumber \\
& & 2 ( 32 k^{'}k_E +16 k k_E +16 k k^{''}+ k k^{'} ) k^7 s+2 k^8 k^{'}k_E \; ,
\end{eqnarray}
which is a consequence of the master equation used, and is independent
of the detail geometrical symmetry.
The mode controlling the decay to the RC is within $P_{16}^4$,
since $P_{16}^1$, $P_{16}^2$,
$P_{16}^3$ do not contain $k^{'}$, $k^{''}$ and $k_E$. However, all four parts
will be influenced by the change of $k$.
If one solves this set of differential equations
by applying the Laplace transformation method,
one finds the solution divides into four distinct groups of decay channels,
namely, $A_5$-$A_{13}$;
E-$A_1$-$A_9$; $A_3$-$A_7$-$A_{11}$-$A_{15}$;
$A_2$-$A_4$-$A_6$-$A_8$-$A_{10}$-$A_{12}$-$A_{14}$-$A_{16}$.
Because the matrix of rate constants is hermitian, all eigenvalues are negative.
Furthermore, no eigenvalues are degenerated, in contrast to
Sk{\'a}la's model which posses too high degree of symmetry~\cite{DS}.
Letting $k^{'} = k^{''}$ does not results in
additional factorizability although the symmetry of our model is restored.
At $k^{'} = k^{''} = 0$, $P_{16}$ becomes
\begin{equation}
s {( s + 2 k )}^2 ( s + 4 k ) ( s + k_E )
( s^2 + 4 k s + 2 k^2 )^2
( s^4 + 8 k s^3 + 20 k^2 s^2 + 16 k^3 s + 2 k^4 )^2 \;. \label{deg}
\end{equation}
It contains
a zero eigenvalue, which signals the existence of a steady-state solution,
as should be happened without the decay to the RC. Degeneracy of
eigenvalues is introduced as the transition to the RC is decreased.
\section{spectrometry comparison}
\ifpreprintsty
\else
\begin{multicols}{2}
\fi
We can verify our model against experiments:
The pump-probe spectroscopy measures the difference between two beams, with
\begin{equation}
\Delta D = \Delta \epsilon_A \sum_n A_n +
\Delta \epsilon_E E\;, \label{PP}
\end{equation}
being the signal measured. The symbol $\Delta \epsilon$\,s are the
differences in dielectrical
constants between pump and probe beams of the
corresponding pigments.
By choosing the pump and probe laser frequencies, we can selectively
detect the population changes of $\sum A_n$ or $E$.
Summing over Eq. (\ref{m1})-(\ref{m4})
we know that the decay of the total population should be
$d (\sum A_n ) / dt = - k^{'} A_1 + k^{''} E$.
The measured charge separation rate is
$k_{E} \approx 3.57 \times 10^{11} s^{-1}$ at room temperature, and
increases by $2$ to $4$
times from $300 K$ to $10 K$ depending on the species chosen~\cite{FMB,MBHMA}.
The ratio of the forward and backward transition to the RC
is know to be about $25 \%$~\cite{TZFS} for an open
RC, i.e., the RC BChl dimer (P) is reduced and the
iron quinone electron acceptor is oxidized; $40 \%$ for pre-reduced RC.
The back-trapping rate can, in principle, be estimated from
${k^{''} / k^{'}} = exp ({- \Delta G / k_B T})$,
with $\Delta G$ the free-energy gap between $A_1$ and $E$ is
estimated from their absorption peaks,
$k_B$ is the Boltzmann
constant, and $T$ is the absolute temperature.
However, the measured absorption peaks of the excited RC are
broad and imprecise\cite{PF}.
We do not know the absolute values of $k^{'}$ or $k^{''}$ experimentally
since it is difficult to tune
the laser frequency to distinguish $A_n$ from $E$.
Nor do we know the transition rate between $A_n$s because
transition between the same species cannot
be measured directly.
Furthermore, at room temperature, energy equilibrium within the
antenna interferes with the trapping process.
Therefore we have taken $k$ and $k^{'}$ as
parameters and fit the slow mode of fluorescence decay of excited population
observed, i.e. $200 ps$~\cite{TFG,BFGRT,SGBAG}.
Thus, the absolute value of the largest eigenvalue
should be about $3 / 200 ps = 1.5 \times 10^{10} s^{-1}$.
A computer code is written to scan all combinations of $k$ and $k^{'}$,
with $k^{''} = k^{'} / 5$, for the largest eigenvalue to be smaller than
$ - 1.5 \times 10^{10} s^{-1}$
between $ - 10^{8} s^{-1}$ to $ - 10^{15} s^{-1}$. Interestingly, we
find all possibility happened at $ k = k^{'}$ and $ k > 6.97 \times
10^{11} s^{-1}$
for $k^{''} = k^{'} / 5$.
Presumably, it is an extremum of $P_{16}^4$.
At the lowest $k$, we can match the required
$200 ps$ decay whose
curve is plotted at Fig.\ref{LH2}.
If $k^{''} = k^{'} / 4$, we obtained $k = 7.25 \times 10^{11} s^{-1}$.
That $k$ has to be equal to $k^{'}$
might sound peculiar in view of the geometrical distance
between $A_1$ and $RC$ is less than the distance between $RC$ and other
$A_n$s \cite{KBG}.
However, the species for donor and acceptors
are different at these two cases.
There are possibilities that the final hopping rate are still the same.
\ifpreprintsty
\else
\begin{figure}[hbt]
\begin{center}
\begin{minipage}[hbt]{8.5cm}
\epsfxsize=\columnwidth\epsfbox{lh22.eps}
\caption{Numerically calculated pump-probe
signal from Eq. (\protect{\ref{PP}})
at $k = 6.97 \times 10^{11}$,
$k_E = 3.57 \times 10^{11}$, $k^{'} = k$, $k^{''} = k^{'}/5$,
$\Delta \epsilon_A = 1$,
$\Delta \epsilon_E = 0$.
The initial condition is $A_5 = 0.2$,
$A_7 = 0.4$, $A_{10} = 0.3$, $A_{12} = 0.1$, while other sites are not excited
at $t = 0$.}
\label{LH2}
\end{minipage}
\end{center}
\end{figure}
\fi
The transfer of excitation energy requires
coupling between the emitting molecule and the ground state molecule.
At an intermolecular separation involved between
$10 \AA$ to $ 100 \AA$, long-range resonance
transfer of electronic excitation arises from coupling between the transition
dipoles of the donor and the acceptor, which is the
F{\"o}ster theory~\cite{F,W}.
Since the BChl $Q_y$ dipoles lie in the same plane, we have
\begin{equation}
k ( R ) \propto {1 \over {\tau_F}} ( {R_0 \over R })^6 \; ,
\end{equation}
in which $R_0$,
measures transfer efficiency, is the F{\"o}ster radius.
van Grondelle gave $R_0 = 90 \AA$ for
the BChl 875 to BChl 875 energy transfer
and a fluorescence life
time, $\tau_F$, about $3000 ps$ or
slightly higher~\cite{RVG,CSJ}.
If a putative
separation distance between interacting BChl $a$ dimers
$\approx 17.5 \AA$
is used~\cite{KBG} we obtain
an estimation of $k \approx 6.17 \times 10^{12} s^{-1}$. This number is
about an order of magnitude higher than the value obtained from our model.
However, the pairwise energy transfer is about $ 1 ps$ according to our
calculation~\cite{SGBAG}.
On the other hand, from the value of $k$ obtained here, by
fitting the $200 ps$ decay as well as the $\tau_F$,
we estimated the F{\"o}ster radius to be
$26.8 \AA$. This result is consistent within our model
since we assume only nearest neighbour transition.
Further, since we put the population at
the antenna at $t = 0$ for our calculation,the rising time is
infinitely short,
instead of having some instrumental limits as observed experimentally.
Although the light wave length is much larger than the ring size, the ring
still might receive energy in localized form by energy transfer from
other rings as the initial condition we used in Fig.\ref{LH2}.
Table I provides a list of all eigenvalues and corresponding amplitudes
obtained from our model.
From the table, we found that the largest eigenvalue mode
is important, not only for its large separation from the other
eigenvalues but also for its corresponding large amplitude.
\ifpreprintsty
\else
\begin{table}[bht]
\begin{center}
\begin{minipage}[bht]{7cm}
\begin{tabular}{ l l }
amplitude & eigenvalue \\
\hline
$ -0.0070 $& $-2.9707 \times 10^{12}$ \\
$ -0.0006 $& $-2.7443 \times 10^{12}$ \\
$ ~~0.0000 $& $-2.6819 \times 10^{12}$ \\
$ ~~0.0221 $& $-2.4614 \times 10^{12}$ \\
$ ~~0.0000 $& $-2.3797 \times 10^{12}$ \\
$ -0.0200 $& $-2.0143 \times 10^{12}$ \\
$ ~~0.0000 $& $-1.9275 \times 10^{12}$ \\
$ ~~0.0124 $& $-1.4844 \times 10^{12}$ \\
$ ~~0.0000 $& $-1.3940 \times 10^{12}$ \\
$ -0.0086 $& $-9.5864 \times 10^{11}$ \\
$ ~~0.0000 $& $-8.6054 \times 10^{11}$ \\
$ -0.0159 $& $-5.5606 \times 10^{11}$ \\
$ ~~0.0000 $& $-4.0829 \times 10^{11}$ \\
$ ~~0.0034 $& $-3.8570 \times 10^{11}$ \\
$ -0.0716 $& $-1.4875 \times 10^{11}$ \\
$ ~~0.0000 $& $-1.0611 \times 10^{11}$ \\
$ ~~1.0858 $& $-1.5107 \times 10^{10}$
\label{tab}
\end{tabular}
\caption{Eigenvalues for LH 1 for Fig.\protect{\ref{LH2}}.
In a time-resolved experiment the relaxation rates correspond to the
lifetimes observed from antenna fluorescence or bleaching kinetics.
}
\end{minipage}
\end{center}
\end{table}
\fi
We have also introduced inhomogeneity, due to geometrical distortion, into the
rate constant. However, even at large
distortion, the basic character of the spectrum is not altered considerably.
If the criteria for $k = k^{'}$ can be established, we can further reduce
the free parameters in our model.
\section{Conclusion}
In summary, a physicist's approach \cite{GLNV} of incoherent energy transfer
within an antenna ring is taken by considering
a two-parameter two-dimensional model.
This model differs from the one presented by Sk{\'a}la {\it et al.}.
The reality might be somewhere between these two models.
In our model, we numerically found $k$ has to be equal to $k^{'}$.
Furthermore, we are able to calculate analytically some of the eigenvalues
and demonstrate explicitly that there is a
mode for decaying to the RC and other three modes.
However, this result of mode separation
depends upon the exact number of unit involved in the
ring. Therefore should not be important.
Perhaps we should interpret the finding
as: $P_{16}^1$, $P_{16}^2$, $P_{16}^3$ are redundant,
since $P_{16}^4$ contains $k_E$ which should be important.
A ring of $16$ units is huge.
The only purpose for such a large antenna is to accommodate the $RC$.
Finally we remark that it is possible to extend a two-dimensional
random walk model of energy transfer into a quasi-three-dimensional one,
in view of the recent structural finding, with a recent result of random walk
on bundled structures by Cassi and Regina\cite{CR2}.
Furthermore, this theoretical result should be able to be verified
experimentally using its spectral dimension by measurements
involving diffusion processes such as time-resolved spectroscopy of
nearest-neighbours energy transfer.
Other light-harvesting models and mechanisms are under further investigation.
|
\section{Introduction}
The object Q2237+0305 comprises a source quasar at redshift $z=1.695$ that is gravitationally lensed by a foreground galaxy ($z=0.0394$) producing 4 images with separations of $\sim 1''$. Each of the 4 images are observed through the bulge of the galaxy, which has an optical depth in stars that is of order unity (eg. Kent \& Falco 1988; Schneider et al. 1988; Schmidt, Webster \& Lewis 1998). In addition, the proximity of the lensing galaxy means that the projected transverse velocity at the source is an order of magnitude higher than for a typical lens configuration. The combination of these considerations make Q2237+0305 the ideal object from which to study microlensing. Indeed, Q2237+0305 is the only object in which cosmological microlensing has been confirmed (Irwin et al. 1989; Corrigan et al. 1991). In 1988 a high magnification event (HME) was observed in image A. It had a measured rise time of $\sim 26$ days, and was followed by an event having a decline time of $\la 3$ months. The resulting light-curve has been interpreted as a double peaked event corresponding to the source having crossed two fold caustics inside a cusp (Witt \& Mao 1994), implying that the peak was caused by a single caustic crossing. The observation thus provides an estimate of the event crossing time, and a lower limit to the event amplitude. Attempts have been made on both of these fronts to interpret the HME in terms of the size of the magnified continuum region (eg. Wambsganss, Paczynski \& Schneider (1990); Rauch \& Blandford 1991; Jaroszynski, Wambsganss \& Paczynski 1992; Webster et al. 1991).
The amplitude of an HME is a function of source size - smaller sources produce events of larger amplitude. However, the event amplitude is also dependent on the width or strength of a caustic, a quantity that is the constant of proportionality in the near caustic approximation of Chang \& Refsdal (1979). Witt (1990) termed this constant the flux factor and found that it takes on a range of values in the complex caustic structures that are produced by high optical depth microlensing models. The mean value of the flux factor depends on the mass function. Thus the event amplitudes have a distribution that is model dependent. Wambsganss, Paczynski \& Schneider (1990) found that a microlensing model with a mean microlens mass of 0.225$M_{\odot}$ reproduces the observed luminosity variation when the source has a radius smaller than $\sim 2\times 10^{15}cm$. Several attempts have been made to discuss the observed HME amplitude in terms of physical models for the continuum region (eg. Rauch \& Blandford 1991; Jaroszynski, Wambsganss \& Paczynski 1992). The results of these calculations remain controversial.
In addition, Lewis et al. (1998) employ the observed difference in spectra obtained at two different epochs to infer a continuum region that is $\la 10^{15}\,cm$ (based on a blackbody accretion disk model).
A different approach is to use the observed event duration in combination with the transverse velocity to obtain an estimate of source size. Approximate values have been obtained assuming a galactic transverse velocity of $v_{t}\sim 600\,km\,sec^{-1}$ (providing the caustic velocity with respect to the source), and a perpendicular caustic crossing (eg. Irwin et al 1989; Webster et al. 1991). These analyses have yielded values for the continuum source size that are consistent with those obtained through consideration of the event amplitude.
In this work we present an analysis of the source size required to re-produce the observed event duration from model light-curves. This work follows results for the probability of the galactic transverse velocity that we have obtained previously (Wyithe, Webster \& Turner 1999b,c, hereafter WWTb, WWTc). In this analysis we consider the effect on the caustic crossing time of a stellar velocity dispersion and the orientation of caustics with respect to the source trajectory.
\section{The Microlensing model}
\label{model}
This work is based on an extensive suite of numerical microlensing simulations presented previously (WWTc); their salient features are discussed below.
We adopt the optical depth and shear ($\kappa$ and $\gamma$) parameters for the lensing galaxy of Q2237+0305 calculated by Schmidt, Webster \& Lewis (1998), and limit our attention to microlensing models in which all the point masses have identical mass. Four classes of models are produced, combining 0\% and 50\% smoothly distributed matter with source trajectory directions parallel to the A$-$B ($\gamma_{A},\gamma_{B}<0$) and C$-$D ($\gamma_{A},\gamma_{B}>0$) axes. We simulate mock data sets using the sampling rate and period of the available monitoring data (Irwin et al. 1989; Corrigan et al. 1991; $\O$stensen et al. 1995). Two assumptions are made for the photometric error: $\sigma_{SE}$=0.01 mag ($\sigma_{LE}$=0.02 mag) in images A/B and $\sigma_{SE}$=0.02 mag ($\sigma_{LE}$=0.04) in images C/D. We assume that microlensing is produced through the combination of a galactic transverse velocity with an isotropic Gaussian velocity dispersion ($\sigma_{*}\sim 165\,km\,sec^{-1}$ from Schmidt, Webster \& Lewis 1998). Where required, a cosmology having $\Omega=1$ with $H_{0}=75\,km\,sec^{-1}$ is assumed.
This paper builds on work presented in WWTc. Discussions of the microlensing models, the effective transverse velocity ($V_{eff}$), and the method for the determination of its probability $P_v(V_{gal}<V_{eff}|\langle m\rangle)$ where $\langle m\rangle$ is the mean microlens mass can be found in that paper. The galactic transverse velocity is obtained from the effective transverse velocity by comparing the microlensing rates using histograms of light curve derivatives (see WWTb).
\begin{figure}
\vspace*{75mm}
\special{psfile=fig1.epsi hscale=55 vscale=55 hoffset=-50 voffset=-120 angle=0}
\caption{\label{cross_geom}A schematic of the geometry of a caustic crossing a source.}
\end{figure}
\begin{figure*}
\vspace*{65mm}
\special{psfile=fig2.epsi hscale=95 vscale=95 hoffset=-245 voffset=210 angle=270}
\caption{\label{caust_prob}The probability distribution functions of normal caustic velocities. The thin, medium and thick lines correspond to transverse velocities of $v_{t}=0,2.5$ and $5\times$ the average projected speed of the stellar velocity dispersion respectively. The thin light line shows the Gaussian distribution of stellar speeds. Left: The applied shear is positive $(\gamma>0)$. Right: The applied shear is negative $(\gamma<0)$.}
\end{figure*}
\section{Source Size}
\subsection{Source Size determination}
The HME in image A observed during late 1988 is described by 5 observations (Irwin et al. 1989; Corrigan et al. 1991) that constrain a peak in the R-band light-curve. We assume that the length of the HME is twice the estimated rise time of 26 days (Corrigan et al. 1991). The likelihood for the true R-band continuum source size is determined by comparing the observed event duration to distributions of model event durations. The calculations of these distributions include consideration of the probability for effective transverse velocity (WWTc), the normal caustic velocities that arise from the proper motion of stars, and caustic orientations.
The velocity of the caustic network resulting from the stream motion of a starfield was calculated by Schramm et al. (1993). This result was generalised by Kundic, Witt \& Chang (1993) to compute the velocity of individual caustic points due to the proper motion. We use the expressions obtained to find the velocities of caustic points in the $y_{1}$ and $y_{2}$ directions. The vector addition of these components is in general in a direction other than normal to the caustic at that point. Therefore we also calculate the caustic gradient. The method of calculating both the caustic location and gradient are described below.
We implement the contouring algorithm described by Lewis et al. (1993). The image curves are followed in discrete steps in one direction. When an image line crosses a critical curve, the image magnification changes sign due to the opposite parity of critical images on either side of the critical curve. The location of the critical curve is then found by searching for the image position that produces a lens mapping Jacobian determinant of zero, and used to calculate the caustic velocity. The gradient of the caustic tangent is found by first computing the location of an image position a short distance along the critical curve in both directions. The gradient of the caustic tangent is computed from the corresponding caustic points.
In each of the models considered we found the distributions of caustic velocities and gradients by searching for caustics along an equal length of source track in two orthogonal directions. The source tracks were each $10\eta_{o}$ (corresponding to $\sim 70$ years for a transverse velocity of $600\,km\,sec^{-1}$) in length, where $\eta_o$ is the Einstein radius of a $1M_{\odot}$ star. We computed the properties of the caustics that intersect with 500 such source tracks in directions parallel ($\gamma>0$) and perpendicular ($\gamma<0$ ) to the shear. A separate model starfield was used for each track. This is equivalent to finding the caustic locations on a $500\times 500$ line grid covering a $10\eta_{o} \times 10\eta_{o}$ square of source plane, although the use of 1000 starfields makes the sample more statistically representative.
Fig.~\ref{cross_geom} is a schematic representation of a caustic crossing, all velocities in this figure are source plane velocities. The caustic velocity $\vec{v}_{caust}$ is the vector addition of the contribution to caustic velocity that results from stellar proper motion $(v_{prop\,1},v_{prop\,2}$), and the transverse velocity $(v_{tran\,1},v_{tran\,2})$,
where
\begin{equation}
(v_{tran\,1},v_{tran\,2})=\vec{v}_{tran}=\frac{D_S}{D_D}\vec{V}_{tran}.
\end{equation}
$V_{tran}$ is the galactic transverse velocity relative to the observer source line of sight, and $D_S$ and $D_D$ are the Angular diameter distances of the source and lens respectively.
The normal caustic velocity $\vec{v}_{\perp}$ is computed from $\vec{v}_{caust}$ and the relative orientation between the tangent to the caustic and the caustic velocity vector. The magnitude of the normal caustic velocity is:
\begin{equation}
v_{\perp}=|\vec{v}_{\perp}|=|\vec{v}_{caust}|\sin(\phi),
\end{equation}
where $\phi:0\leq\phi\leq\frac{\pi}{2}$ is the angle between the caustic tangent and velocity vector. A caustic with a velocity vector that makes a small angle with the caustic tangent sweeps out a smaller area per unit length per unit velocity than one having a velocity normal to its tangent. Therefore, calculations of the distribution of normal caustic velocities at a stationary source position, include an additional weighting of $\sin(\phi)$.
\begin{figure}
\vspace*{120mm}
\special{psfile=fig3a.epsi hscale=45 vscale=50 hoffset=-15 voffset=-52 angle=0}
\special{psfile=fig3b.epsi hscale=45 vscale=50 hoffset=-15 voffset=-162 angle=0}
\special{psfile=fig3c.epsi hscale=45 vscale=50 hoffset=-15 voffset=-310 angle=0}
\caption{\label{cross} Bottom: A sample light-curve for image A ($\gamma_{A}=+0.4$) of Q2237+0305. One half of a double peaked event is highlighted by a box. Centre: A top-hat intensity profile (dark line), and a Gaussian intensity profile (light line). Top: A plot of the the highlighted HME showing the details of the event shapes that result from the two source intensity profiles.}
\end{figure}
\begin{table*}
\caption{\label{tab_size} Table showing the upper and lower 99\%, 95\%, and 90\% limits as well as the mean of the distributions $p_{s}(S_{s}\,|\,0.1M_{\odot})$ and $p_{s}(S_{s}\,|\,1.0M_{\odot})$. The values are given for the two directions assumed for the transverse velocity as well as for simulations that contain 0\% and 50\% smoothly distributed matter. The limits shown correspond to the cases where the error was $\sigma_{SE}=0.01$ mags in images A/B, $\sigma_{SE}=0.02$ mags in images C/D, and where the error was $\sigma_{LE}=0.02$ mags in images A/B, $\sigma_{LE}=0.04$ mags in images C/D (in parentheses).}
0\% Smooth Matter
\begin{tabular}{|c|c|c|c|c|c|c|c|c|c|}
\hline
Mean &Trajectory &$S_{low}(99\%)$ &$S_{low}(95\%)$ &$S_{low}(90\%)$ & Mean &$S_{up}(90\%)$ &$S_{up}(95\%)$ &$S_{up}(99\%)$ \\
Mass &Direction &($\times 10^{14}cm$)&($\times 10^{14}cm$)&($\times 10^{14}cm$)&($\times 10^{14}cm$)&($\times 10^{14}cm$)&($\times 10^{14}cm$)&($\times 10^{14}cm$) \\ \hline\hline
0.1$M_{\odot}$&$\gamma_{A},\gamma_{B}>0$ & 0.244 (0.238) & 0.637 (0.617) & 1.01 (0.972) & 6.70 (6.52) & 12.1 (11.9) & 16.6 (16.2) & 29.2 (29.0) \\
1.0$M_{\odot}$&$\gamma_{A},\gamma_{B}>0$ & 0.394 (0.280 ) & 1.10 (0.746) & 1.80 (1.19) & 12.7 (8.48) & 24.1 (16.0) & 31.3 (21.7) & 49.2 (36.6) \\
0.1$M_{\odot}$&$\gamma_{C},\gamma_{D}>0$ & 0.240 (0.237) & 0.622 (0.614) & 0.982 (0.968) & 6.56 (6.50) & 11.9 (11.9) & 16.3 (16.2) & 29.1 (29.0) \\
1.0$M_{\odot}$&$\gamma_{C},\gamma_{D}>0$ & 0.370 (0.272) & 1.03 (0.723) & 1.67 (1.16) & 11.9 (8.11) & 22.6 (15.2) & 29.2 (20.5) & 46.0 (35.1) \\ \hline
\end{tabular}
\vspace{7 mm}
50\% Smooth Matter
\begin{tabular}{|c|c|c|c|c|c|c|c|c|c|}
\hline
Mean &Trajectory &$S_{low}(99\%)$ &$S_{low}(95\%)$ &$S_{low}(90\%)$ & Mean &$S_{up}(90\%)$ &$S_{up}(95\%)$ &$S_{up}(99\%)$ \\
Mass &Direction &($\times 10^{14}cm$)&($\times 10^{14}cm$)&($\times 10^{14}cm$)&($\times 10^{14}cm$)&($\times 10^{14}cm$)&($\times 10^{14}cm$)&($\times 10^{14}cm$) \\ \hline\hline
0.1$M_{\odot}$&$\gamma_{A},\gamma_{B}>0$ & 0.207 (0.187) & 0.556 (0.493) & 0.895 (0.795) & 5.95 (5.31) & 11.1 (9.75) & 14.6 (13.3) & 24.0 (23.2) \\
1.0$M_{\odot}$&$\gamma_{A},\gamma_{B}>0$ & 0.487 (0.274) & 1.45 (0.772) & 2.48 (1.26) & 16.4 (10.1) & 30.1 (19.4) & 38.1 (25.7) & 58.4 (40.4) \\
0.1$M_{\odot}$&$\gamma_{C},\gamma_{D}>0$ & 0.206 (0.188) & 0.553 (0.495) & 0.888 (0.796) & 5.73 (5.27) & 10.5 (9.57) & 13.1 (13.2) & 23.3 (23.0) \\
1.0$M_{\odot}$&$\gamma_{C},\gamma_{D}>0$ & 0.468 (0.272) & 1.36 (0.760) & 2.26 (1.23) & 14.5 (9.25) & 26.8 (17.7) & 33.7 (23.2) & 49.3 3(6.7) \\ \hline
\end{tabular}
\end{table*}
Figure \ref{caust_prob} shows probability distributions of normal caustic velocities $p_c(v_{\perp}|\langle m\rangle, v_{tran})$ where $v_{tran}=|\vec{v}_{tran}|$. The left and right hand diagrams correspond to the cases of positive and negative galactic shear respectively. Distributions are shown in each case for the normal caustic velocities due to the stellar proper motions alone (thick line), and in combination with a transverse velocity in the $y_{1}$ direction of 2.5 (medium line) and 5.0 (thin line). Also plotted in Fig.~\ref{caust_prob} is the distribution of stellar speeds (light line). All velocities are in units of the projected mean speed of the isotropic Gaussian stellar velocity dispersion.
These distributions demonstrate the dependence of the normal caustic velocity on the direction of transverse velocity; a transverse velocity along the direction of caustic clustering ($\gamma<0$) results in lower normal caustic velocities because the tangent to the caustic is more likely to lie along the source trajectory. Note however that the distribution must be independent of the shear direction if the transverse velocity is zero.
We assume a circular projected source profile so that the event timescale is not dependent on the source orientation. We define the source diameter $S_{s}$. For a small source ($S_{s}\ll\eta_{o}$) the crossing time of the caustic, and therefore the event timescale is
\begin{equation}
\label{size}
\Delta T\sim S_{s}/v_{\perp}.
\end{equation}
Thus, $S_{s}$ can be estimated by combining a measured timescale with statistical information about caustic orientations and velocities. Eqn \ref{size} demonstrates a systematic uncertainty in our determination of source size that is proportional to the uncertainty in the estimate of the event duration.
Eqn \ref{size} defines our measure of source size, however its interpretation is dependent on the intensity profile of the source. The lower panel in Fig.~\ref{cross} shows an example of an extended source light-curve for image A ($\gamma_{A}=+0.4$) computed using the method described in Wyithe \& Webster (1999). One half of a double peaked event is highlighted by a box and reproduced in the top panel of Fig.~\ref{cross} for two source intensity profiles: a Gaussian profile (light line) and a top-hat profile (dark line). Cross-sections of the intensity profiles are shown in the central panel of Fig.~\ref{cross} and comparison of these with the corresponding HMEs demonstrates that $S_{s}$ is approximately the source diameter for a top-hat source profile, but twice the half width $S_{s}$ for a Gaussian profile.
Probability distributions for normal caustic velocity $p_{c}(v_{\perp}\,|\,\langle m\rangle,v_{tran})$ are computed assuming various values for the galactic transverse velocity. By differentiating the cumulative distribution $P_v(V<V_{tran}|\langle m\rangle)$ (following a correction for the contribution of stellar proper motion to the value of effective transverse velocity (WWTb)), we also find the probability distribution for galactic transverse velocity,
\begin{equation}
p_{v}(v_{tran}=\frac{D_S}{D_D}V_{tran}\,|\,\langle m\rangle)=\frac{D_D}{D_S}p_v(V_{tran}|\langle m\rangle).
\end{equation}
The probability distribution for source size $S_s$ is
\begin{eqnarray}
&\hspace{-42mm}p_{s}(S_{s}=v_{\perp}\Delta T\,|\,\langle m\rangle)= \nonumber \\
&\hspace{5mm}\frac{1}{\Delta t}\,\int p_{c}(v_{\perp}\,|\,\langle m\rangle,v_{tran})\,p_{v}(v_{tran}\,|\,\langle m\rangle)\,dv_{tran} .
\end{eqnarray}
\subsection{\label{source_size}Source size limits}
\begin{figure*}
\vspace*{130mm}
\special{psfile=fig4.epsi hscale=95 vscale=95 hoffset=-245 voffset=390 angle=270}
\caption{\label{diff}The differential distributions of source size $p_s(S_{s}\,|\,0.1M_{\odot})$ (thin lines) and $p_s(S_{s}\,|\,1.0M_{\odot})$ (thick lines). The solid and dotted curves represent the resulting functions when the photometric error was assumed to be $\sigma_{SE}$=0.01 mag in images A/B and $\sigma_{SE}$=0.02 in images C/D, and $\sigma_{LE}$=0.02 mag in images A/B and $\sigma_{LE}$=0.04 in images C/D.}
\end{figure*}
\begin{figure*}
\vspace*{130mm}
\special{psfile=fig5.epsi hscale=95 vscale=95 hoffset=-245 voffset=390 angle=270}
\caption{\label{contour}Contours of percentage peak height of the function $p_{m,s}(S_{s},\langle m\rangle)$. The contours shown are the 3.6\%, 14\%, 26\% and 61\% levels. The solid and dotted curves represent the resulting functions when the photometric error was assumed to be $\sigma_{SE}$=0.01 mag in images A/B and $\sigma_{SE}$=0.02 in images C/D, and $\sigma_{LE}$=0.02 mag in images A/B and $\sigma_{LE}$=0.04 in images C/D. The light and dark lines correspond to the assumptions of logarithmic and flat priors for transverse velocity referred to in the text.}
\end{figure*}
Our results are presented in Tab.~\ref{tab_size}. Values are given for the upper and lower 99\%, 95\%, and 90\% limits as well as the mean of the distributions $p_{s}(S_{s}\,|\,0.1M_{\odot})$ and $p_{s}(S_{s}\,|\,1.0M_{\odot})$ for models having various combinations of assumptions for the transverse velocity directions and smooth matter fraction. Limits are shown assuming the small (SE) and large (LE, in parentheses) errors defined in Sec.~\ref{model}.
The differential distributions of source size $p_s(S_{s}\,|\,0.1M_{\odot})$ (thin lines) and $p_s(S_{s}\,|\,1.0M_{\odot})$ (thick lines) are plotted in Fig.~\ref{diff}. The solid and dotted lines correspond to the cases of small and large errors respectively. We find that the source size lies between the values of $2\times 10^{13}cm$ and $6\times 10^{15}cm$ (98\% level).
While the lower source size limit is not very sensitive to the microlens mass assumed, the upper limit is higher for larger assumptions of mean microlens mass. The trend is demonstrated in Fig.~\ref{diff}, and is due to the limit of effective transverse velocity being proportional to $\sqrt{\langle m\rangle}$. Lower limits for source size come from the lowest velocity caustics, which are produced by the stellar proper motions alone, hence the insensitivity to the microlens mass. However the upper limits result from the fastest moving caustics, and so depend on the microlens mass dependent upper limits to transverse velocity. The effect is magnified if a component of smooth matter is assumed. Due to the more sparsely distributed caustic network produced by such a model, a larger transverse velocity is measured (WWTc). Also the contribution of stellar proper motions is reduced. These factors result in an increased sensitivity of source size estimates to microlens mass if the model smooth matter content is higher. The source size distributions are highly asymmetric. However, due to the stability of the lower end of the distributions, the mode is consistent, between $\sim 1\times 10^{14}cm$ and $4\times 10^{14}cm$ for all models considered.
It is apparent from Fig.~\ref{diff} that the assumptions for photometric error and the direction of the transverse velocity are important factors, particularly for a larger assumed mean microlens mass. Like the assumptions for microlens mass, assumptions for the error and transverse velocity direction effect the measurement of transverse velocity, and hence source size. Upper limits are more sensitive than lower limits which are set primarily by the stellar velocity dispersion. These effects are also enhanced by the presence of a component of smooth matter.
The co-dependence of mean microlens mass and source size is now treated more rigorously.
\subsection{Simultaneous limits of mean microlens mass and source size}
In Sec~\ref{source_size} the normalised probability distributions, $p_s(S_{\rmn s} | \langle m \rangle)$, that the continuum source size is between $S_{\rmn s}$ and $S_{\rmn s} + {\rmn d}S_{\rmn s}$ given that the mean microlens mass is $\langle m \rangle$, were obtained. The probability that the source size lies in the above range and the mean microlens mass is between $\langle m \rangle$ and $\langle m \rangle + {\rmn d} \langle m \rangle$ is then given by
\begin{equation}
p_{m,s}(S_{\rmn s}, \langle m \rangle) = p_s(S_{\rmn s} | \langle m \rangle)
\, p_m (\langle m \rangle),
\end{equation}
where $p_m(\langle m \rangle)$ is the probability that the mean microlens mass is between $\langle m \rangle$ and $\langle m \rangle + {\rmn d} \langle m \rangle$. $p_m(\langle m \rangle)$ is a known function (given in WWTc), that is approximately independent of the source size assumed and was determined under the same assumptions for the Bayesian prior (flat and logarithmic) of galactic effective transverse velocity as the function $p_{v}(v_{tran}|\langle m\rangle)$. However unlike $p_{v}(v_{tran}|\langle m\rangle)$, $p_m (\langle m \rangle)$ is dependent on the prior assumed (see WWTc).
Figure~\ref{contour} shows contour plots of the two-dimensional distribution $p_{m,s}(S_s,\langle m\rangle)$ for the assumed source orientations, smooth matter fraction and photometric errors. All possible combinations of the two values of each parameter are included, so that eight models are shown. The contours are at 3.6, 14, 26, 61 per cent of the peak height, and so the extrema of the relevant contours represent $4\sigma$, $3 \sigma$, $2\sigma$, and $1\sigma$ limits on the single variables. The light and dark lines correspond to the assumptions of logarithmic and flat priors for effective transverse velocity. The most likely model is that $\langle m \rangle \simeq 0.05-0.2 M_{\odot}$ (dependent on the model systematics of trajectory direction, smooth matter content and photometric uncertainty) and $S_{\rmn s} \simeq 2\times10^{14}$ cm (approximately independent of model).
The most important benefit of combining the distributions for $S_{\rmn s}$ and $\langle m \rangle$ is that an upper limit can be placed on the mean microlens mass. The limits on $\langle m \rangle$ are subject to both random and systematic uncertainties; the measured mass increases with photometric errors and decreases with the fraction of smooth matter. Assuming that the chosen values for these parameters bracket the likely models, upper and lower limits for $\langle m \rangle$ can be found by using the most extreme values of the contour limits. This yields the 95 per cent result that $0.01 M_{\odot} \la \langle m \rangle \la 1 M_{\odot}$. The upper limit, which agrees with that obtained from simultaneous consideration of probabilities for $\langle m\rangle$ and $V_{tran}$ (WWTc) is lower than that obtained by Lewis \& Irwin (1996), although sufficiently high that it does not conflict with any popular models of galactic halos or bulges. Similarly, the contour maxima yield the result that there is a 95 per cent chance of $5 \times 10^{12}$ cm $\la S_{\rmn s} \la 10^{15}$ cm. At a given confidence level, both the upper and lower limits of source size, and the lower limit of mean microlens mass have smaller values in Fig.~\ref{contour} than those obtained from probabilities for single variables in previous sections and WWTc. This can be attributed to the asymmetry of the distribution in both dimensions.
\section{Conclusion}
The observed peak in the light-curve for image A of Q2237+0305 in 1988 appears to be one half of a double peaked event that is characteristic of a source having passed inside of a cusp. If this interpretation is correct then the observed peak is the result of the source having been crossed by a single caustic, and the observations provide a good record of the caustic crossing time. The source size is the product of the event length and the normal velocity of the caustic involved. We have placed limits on normal caustic velocities by combining measurements of the galactic transverse velocity with calculations of the distributions of caustic gradients and the component of the caustic velocity that results from stellar proper motions. We find that the most likely size of the magnified continuum region is $S_{s}\sim 2\times 10^{14}\,cm$, and that the upper and lower limits are $S_{s}\sim 6\times 10^{15}\,cm$ and $S_{s}\sim 2\times 10^{13}\,cm$ (99\% level) respectively. Our calculations are quite insensitive to the systematic considerations of trajectory direction, smooth matter content and assumption of Bayesian prior for effective transverse velocity (within the ranges considered). However, our conclusions are sensitive to the microlens mass assumed, as well as to the error assumed for the published photometric magnitudes. This sensitivity arises from the dependence of the determination of effective transverse velocity on these quantities.
Our conclusions are consistent with those of Wambsganss, Paczynski \& Schneider (1990) and Rauch \& Blandford (1991) who found that the event rise-time and peak height require a source smaller than $\sim2\times10^{15}\,cm$ (if stellar mass objects are assumed for the microlenses). The microlensing determinations from Q2237+0305 support length scales of $ct\sim 10^{14}\,t_{hrs}\,cm$ associated with X-ray and optical changes in continuum emission. Also, theoretical calculations of the typical scale-size of a continuum emitting accretion disc orbiting a supermassive black-hole (Rees 1984) suggest a size of $\sim 10^{15}\,cm$.
Simultaneous consideration of microlens mass and source size provides an upper limit to the mean microlens mass of $1M_{\odot}$ that is unavailable from consideration of the microlensing rate alone. We find that $0.1M_{\odot}<\langle m\rangle<1M_{\odot}$ (95\% level). Unlike the source size, our mass determinations are dependent on the prior assumed for effective transverse velocity (see Wyithe, Webster \& Turner 1999c).
It was noted in (Wyithe, Webster \& Turner 1999b) that the statistical determination of the effective transverse velocity will be aided when monitoring data from a longer period becomes available. The narrowing of the probability for the effective transverse velocity will in turn provide tighter limits on the mean microlens mass and source size. At a known transverse velocity, the component of uncertainty due to the range of event lengths that can be produced by a source will be reduced upon the observation of more HME durations. Thus the extent of the contours in Fig.~\ref{contour} will be reduced as more monitoring data becomes available.
\section*{Acknowledgements}
The authors would like to thank Chris Fluke for helpful discussions. This work was supported in part by NSF grant AST98-02802. JSBW acknowledges the support of an Australian Postgraduate award and a Melbourne University Overseas Research Experience Award.
|
Subsets and Splits